Previously Published Works

UC Irvine

A University of California author or department has made this article openly available. Thanks to
the Academic Senate’s Open Access Policy, a great many UC-authored scholarly publications
will now be freely available on this site.
Let us know how this access is important for you. We want to hear your story!
http://escholarship.org/reader_feedback.html

Peer Reviewed

Title:
Going to Pot? The Impact of Dispensary Closures on Crime

Journal Issue:
Journal of Urban Economics,

Author:
Chang, T
Jacobson, M

Publication Date:
04-01-2017

Series:
UC Irvine Previously Published Works

Also Available:
Papers

Permalink:
http://escholarship.org/uc/item/97r97669

DOI:
https://doi.org/10.1016/j.jue.2017.04.001

Local Identifier(s):
UCPMS ID: 2078608

Copyright Information:
All rights reserved unless otherwise indicated. Contact the author or original publisher for any
necessary permissions. eScholarship is not the copyright owner for deposited works. Learn more
at http://www.escholarship.org/help_copyright.html#reuse

eScholarship provides open access, scholarly publishing

services to the University of California and delivers a dynamic
research platform to scholars worldwide.
 Going to Pot?
∗
The Impact of Dispensary Closures on Crime
Tom Chang
USC Marshall School of Business

and

Mireille Jacobson
The Paul Merage School of Business, UCI and NBER
March 2017

Abstract
Jurisdictions that sanction medical or, more recently, recreational marijuana use
often allow retail sales at dispensaries. Dispensaries are controversial as many believe
they contribute to local crime. To assess this claim, we analyze the short-term mass
closing of hundreds of medical marijuana dispensaries in Los Angeles. Contrary to
popular wisdom, we find an immediate increase in crime around dispensaries ordered
to close relative to those allowed to remain open. The increase is specific to the type
of crime most plausibly deterred by bystanders, and is correlated with neighborhood
walkability. We find a similar pattern of results for temporary restaurant closures due
to health code violations. A likely common mechanism is that“eyes upon the street”
deter some types of crime.

∗
We thank the editor, Stuart Rosenthal, and two anonymous referees for their attention to this paper.
We also thank James Anderson, Victor Bennett, Kitt Carpenter, Jonathan Caulkins, Harry DeAngelo,
Matthew Freedman, Paul Heaton, Jim Hosek, Beau Kilmer, Jeff Kling, Jens Ludwig, John MacDonald,
John Matsuzaka, Kevin Murphy, Sendhil Mullainathan, Rosalie Pacula, Greg Ridgeway, seminar partic-
ipants at the University of Southern California and the University of Colorado-Denver and participants
of the 2011 NBER Summer Institute Crime Working Group and the 2013 Southern California Conference
in Applied Microeconomics for helpful comments. Special thanks to Jens Ludwig for the idea to look
at restaurant closures. We thank Ben Welsh and Doug Smith from The Los Angeles Times for providing
feedback on the data available through the Crime L.A. project and Aaron Kofner for geocoding these crime
data. The Los Angeles Times had no role in the data analysis. Jacobson acknowledges financial support
from NIDA 1R01DA032693-01. All mistakes are our own.

1
1 Introduction
One of the most dramatic shifts in public opinion in the U.S. over the past four and
a half decades has been a surge in support for marijuana legalization, both medical and,
increasingly, recreational. Currently 60% of adults in the U.S. favor broad-based marijuana
legalization, compared to only 12% in 1969 (Swift 2016), and nearly 90% think adults should
be allowed to use marijuana for prescribed medical purposes (CNN/ORC 2014). Despite
this support, 44% indicate that they would be somewhat or very concerned if a “store
that sold medical marijuana” opened in their area (The Pew Research Center 2010). In
particular, many maintain that these stores, usually called dispensaries, attract or, even,
cause crime (McDonald and Pelisek 2009; National Public Radio 2010; Reuteman 2010).
The idea that marijuana dispensaries attract crime has proved influential with policy-
makers. For example, an Oregon state senator argued that a law allowing cities to ban
dispensaries was important to “empower them to protect our children and families” (Zheng
2014). In Los Angeles, the setting for this study, the city council cited crime in its 2010 de-
cision to cap the number of dispensaries in the city.1 Yet, empirical evidence to support any
link (positive or negative) between marijuana dispensaries and crime is quite limited. State
difference-in-differences estimates find no relationship between medical marijuana laws and
crime rates (Morris et al. 2014). Since not all medical marijuana states have operational
dispensaries, however, these estimates do not speak directly to the impact of dispensaries
on crime. The density of dispensaries across 95 census tracts in Sacramento, CA is uncorre-
lated with either violent or property crime rates (Kepple and Freisthler 2012).2 Well-known
limitations of cross-sectional analyses and a general lack of statistical power in that study
suggest the importance of continued work on the topic.
How, in theory, might medical marijuana dispensaries affect crime? First, marijuana
use, which may be concentrated around dispensaries if some buyers consume onsite or
nearby, may be criminogenic. Similar effects have been cited for alcohol outlets, where
1
See the fifth paragraph of Ordinance 181069 http://clkrep.lacity.org/onlinedocs/2008/
08-0923_ord_181069.pdf
2
The Denver and Colorado Springs Police Departments each analyzed the number of crimes around
dispensaries and compared them to the numbers around banks, pharmacies, and other businesses (Ingold,
2010; Rodgers, 2010). Neither found that dispensaries attract crime, although recent work demonstrates
that dispensaries in Denver tend to be located in high crime neighborhoods (Boggess et al. 2014).

2
openings and availability in Los Angeles and other jurisdictions are associated with increases
in crime (Teh 2008; Scribner et al. 1995; Gorman et al., 1998; Scribner et al., 1999;
Gruenewald and Remer 2006; Gruenewald et al. 2006; Franklin et al. 2010; Grubesic
and Pridemore 2011). In contrast to alcohol, however, some work suggests marijuana may
not increase crime commission per se (Pacula and Kilmer, 2003) and may even inhibit
aggressive behavior (Myerscough and Taylor 1985; NAS 1994; Hoaken and Stewart 2003).3
Second, given the quasi-legal status of these stores and their products, dispensary cus-
tomers, employees or owners may resort to violence to resolve disputes (Miron 1999; Resig-
nato, 2000).4 If so, we might expect increases in crimes such as aggravated assault, which
increased for such reasons with the emergence of crack cocaine (Grogger and Willis 2000).
Third, crime could increase near dispensaries as individuals try to finance their pur-
chases through the proceeds of crime (Grogger and Willis 2000). If so, we would expect
theft or other property crimes to increase with dispensaries. Finally, marijuana users and
the dispensaries they frequent, which are a direct source of drugs and cash, may offer op-
portunities that attract criminals. Anecdotal evidence suggests that dispensaries have been
subject to break-ins and robberies (e.g., see McDonald and Pelisek, 2009). Thus, we would
expect an increase in robbery and burglary around dispensaries.5
While these channels seem plausible and have captured public attention, dispensaries
could, in principle, decrease crime. Dispensaries tend to have their own security systems
and often security guards to protect their assets and resolve disputes. Analyses of business
improvement districts find that private security can have large returns in terms of crime
3
The correlation between marijuana use and non-drug crime, although positive, is generally small (Ben-
nett et al. 2008) and largely inconclusive (Pedersen and Skardhamar 2010; Farrington 2010). Longitudinal
studies that find clearer positive relationships, such as Green et al. (2010), cannot rule out the role of third
factors that affect both the commission of non-drug crime and marijuana use (Caulkins et al. 2012).
4
We describe dispensaries as quasi-legal for several reasons. First, although medical marijuana use
is legal in California, large-scale production and sales are not. Second, while cooperative are allowed
under California law, Los Angeles and other localities can tightly regulate and, in some cases ban, their
operations. Finally, under federal law, it remains illegal to manufacture, distribute or possess marijuana.
Consequently, dispensaries have been targeted and raided by federal law enforcement.
5
In this case, dispensaries may affect the spatial distribution of crime rather than increase the overall
level. Such a change has clear negative implications for dispensary neighbors, but may not have broader
societal implications. The welfare impact of geographic redistribution of crime depends on such factors as
heterogenous effects (e.g., different costs across neighborhoods), multiplicative effects (if two crimes in one
area impose higher costs than one crime in each of two areas) and economies of scale (if two crimes in one
area is less costly than one crime in each of two areas).

3
reduction (Brooks, 2008; Cook and MacDonald 2011). Likewise, if police allocate more
patrols around dispensaries, they might reduce crime as in Di Tella and Schargrodsky
(2004). To the extent that dispensaries increase foot traffic through a neighborhood, they
might prevent crime by increasing “eyes on the street” (Jacobs 1961). In addition, by
legitimizing the marijuana trade, actors in this market may have legal channels to resolve
disputes. This last possibility is somewhat less plausible given the ambiguous legality of
many aspects of the medical marijuana market, such as large scale distribution.
Finally, if marijuana is a substitute for alcohol, as suggested by Anderson, Hansen and
Rees (2013) and Crost and Rees (2013), increased access to marijuana could reduce crime
since drinking increases arrests for both property crime (Carpenter 2007) and violent crime
(Carpenter and Dobkin 2015). Ultimately, given the range of theoretical predictions, the
impact of dispensaries on crime is an empirical question.
To evaluate the claim that dispensaries attract or otherwise contribute to crime, we
exploit a plausibly exogenous source of variation in dispensary activity – the temporary
shutdown of medical marijuana dispensaries in the City of Los Angeles. On June 7, 2010,
roughly 70% of the nearly 600 shops operating in the city of Los Angeles were ordered to
close (Hoeffel 2010a). The shutdown came after years of concern and indecision over how
to handle the burgeoning medical marijuana dispensary business in the city. In September
2007, the city adopted an “Interim Control Ordinance” (ICO), placing a temporary mora-
torium on new dispensaries and requiring existing dispensaries to register with the city by
November 13, 2007 (see Appendix Table 1 for a timeline).
Given the limited time that dispensaries had to submit a registration form along with
the required city business tax registration certificate, registration was quite ad hoc. How
the city would use the registrations was unclear and the market continued to grow for
several years despite the moratorium. In January 2010, final regulations, including closure
orders, were adopted. The new ordinance set the number of dispensaries in the city at
70. Dispensaries that had registered between September and November 2007 and had been
operating legally since that time were grandfathered, meaning that the number of legal
dispensaries in the city could exceed 70 in the short run.
Consistent with the seeming arbitrariness of the closure criteria, we find that dispen-

4
saries ordered to close and those allowed to remain open look similar on observable di-
mensions. In other words, closure orders were not correlated with observable dispensary
characteristics (including the level of or trend in crime around specific dispensaries) that
might have otherwise made them of specific interest to law enforcement. We leverage the
quasi-random nature of closure orders using a difference-in-differences framework and de-
tailed data on exact dispensary locations and crime reports by city block to compare daily
crime counts within varying radii (as small as 1/8 of a mile) around dispensaries ordered
to close and those allowed to remain open. If dispensaries attract crime, then crime should
decrease around dispensaries subject to closure relative to those allowed to remain open.6
Contrary to conventional wisdom, we find no evidence that closures decreased crime.
Instead, we find a significant relative increase in crime around closed dispensaries. Like
compliance with the closures orders themselves, which first was high, fell off with legal
challenges and collapsed after a December 2010 injunction (Hoeffel 2010b), the increase
in crime is temporary. Relative crime rates return to normal within four weeks. The
increase is also very local – the estimated crime effects decrease rapidly and monotonically
with distance around dispensaries. Bearing in mind that our analysis captures short-run
effects, these findings imply that closing medical marijuana dispensaries is unlikely to reduce
crime. Although there may be a myriad of reasons to regulate the number of marijuana
dispensaries, protection from crime is one that seems difficult to substantiate.
We perform several analyses to better understand how dispensary closures affect crime.
First, we analyze crime by categories. We find that the increase in crime is strongest and
most precise for the type of crime most plausibly deterred by the presence of bystanders
– property crime and theft from vehicles, specifically. Second, we analyze the interaction
between closures and neighborhood foot traffic. We proxy for foot traffic using Walk Scores,
a proprietary measure that scores each address based on the walking time to amenities,
population density, block length and the density of street intersections. We find that the
the magnitude of the crime effect varies in a non-linear way with Walk Scores. Specifically,
the magnitude of the closure effect varies negatively with walkability, except in the most
6
An alternative question, not explicitly evaluated here, is how dispensaries affect crime relative to
alternative business types (e.g., ice cream parlors, convenience stores or banks.) While we cannot speak to
this directly, our analyses of temporary restaurant closures can help shed light on this issue.

5
geographically isolated areas for which closures have no measurable effect on crime.
To shed further light on mechanisms, we explore the generalizability of the findings.
Specifically, we analyze the impact of temporary restaurant closures due to public health
code violations on crime in Los Angeles County. Despite the very different nature of these
businesses, the reason for and timing of their closures, and the identifying assumptions, we
find a nearly identical pattern of results. Crime increases in the local neighborhood around
closed restaurants, the increase is driven by property crime, the effect is concentrated
in areas without a high volume of foot traffic, and the effect disappears as soon as the
restaurant reopens.
The common pattern of results for dispensaries and restaurants suggests that business
closures in general exert a significant negative crime externality. By extension, businesses
offer very local protection against some types of crime. Given that police are unlikely
to systematically change their behavior in response to temporary restaurant closures, this
analysis further suggests that changes in policing cannot explain the common pattern of
results. Rather, a likely common mechanism may be “eyes upon the street” (Jacobs 1961),
meaning that the presence of individuals helps deter crime. While part of the canon of mod-
ern urban design and crime prevention, this theory is virtually unsupported by rigorous
empirical evidence. In addition, Jane Jacob’s original 1961 formulation of the hypothesis
makes clear that the impact of additional individuals on local crime is theoretically am-
biguous; crowds provide some form of natural policing but also more perpetrators of and
opportunities for crime. Our findings suggest that the first channel dominates, at least in
the case of medical marijuana dispensaries and restaurants in urban environments.
The rest of the paper is organized as follows. In section 2 we discuss the June 2010
closure of medical marijuana dispensaries in Los Angeles and describe our data. In section
3 we describe our analytic approach for the dispensary analysis. In section 4 we present our
main results. In section 5 we discuss spatial and temporal displacement. In section 6 we
presents the institutional details of and results from our analysis of temporary restaurant
closures due to health code violations in Los Angeles County. In section 7 we explore
potential mechanisms behind the shared pattern of findings. In section 8 we conclude.

6
2 Medical Marijuana Dispensaries in Los Angeles
In 1996, voters in California approved Proposition 215, the state’s medical marijuana law.
Marijuana dispensaries opened to serve the patients newly qualified to use the drug under
the law. Like the state as a whole, the City of Los Angeles saw rapid growth in dispensaries
after the 2004 passage of a bill (SB 420) that clarified several operational aspects of the
state’s medical marijuana law.7 At its peak, some estimates put the number of dispensaries
in the City of Los Angeles at over 800 (McDonald and Pelisek 2009).
Not all Los Angeles residents welcomed these stores. Many believed that dispensaries
attract crime and law enforcement fueled these concerns. In a July 2005 report, the LAPD
cited several felony narcotics arrests made at dispensaries and speculated that “crimes such
as theft, robbery and assault have occurred and will occur along with the sale of marijuana
from these locations.” As a result, they called for restricting dispensaries to commercial
areas in the city if not banning them outright.8 A later report by the Los Angeles Police
Commission argued that the increase in dispensaries within the city (from 4 to 98 between
July 2005 and November 2006) was tied to an increase in crime in reporting districts that
had received complaints about dispensaries.9 While these crime changes were not compared
to that around other businesses or areas, many found the argument persuasive.
In 2006, the City Attorney’s Office laid out options for regulating dispensaries – a land
use ordinance establishing zoning requirements, an interim moratorium until state law was
”further clarified” or an outright ban. Almost a year later, in September 2007, the city
adopted an “Interim Control Ordinance” (ICO) that temporarily banned new dispensaries
and required existing ones to register with the city by November 13, 2007. The ICO aimed
to pacify constituents concerned with the growth of dispensaries while the city drafted
permanent legislation.
While in principle the ICO should have stopped the growth in dispensaries, in practice
it had the opposite effect. Hundreds of dispensaries opened after the moratorium by filing
7
Among other things, SB 420 recognized a patient’s right to cultivate marijuana through nonprofit
collectives and cooperatives, i.e., dispensaries, a right that was later affirmed in People v. Urziceanu. See
http://caselaw.lp.findlaw.com/data2/californiastatecases/C045276.PDF
8
See http://clkrep.lacity.org/onlinedocs/2005/05-0872_rpt_atty_10-19-06.pdf
9
See http://californiapolicechiefs.org/site/uploads-calchiefs/2012/02/fact_sheet.pdf

7
applications for “hardship exemptions” allowed under the ICO (McDonald and Pelisek,
2009).10 The large number of applicants stemmed in part from the recognition that the city
would not prosecute dispensaries until their hardship applications had been reviewed and
that the city council was in no hurry to review applications. By June 2009, when the city
council first began to rule on the hardship exemption applications, over 500 applications
had been submitted (Hoeffel, 2009). On June 19, 2009, the city passed an ordinance
amending the ICO to eliminate the hardship exemption.11
Although intended as a stop-gap measure, the ICO remained in place for more than
a year and half. On January 26, 2010 the city council approved final legislation limiting
the number of dispensaries in the city to 70 but grandfathering in those that had reg-
istered and been operating legally since the ICO.12 Based on the 2007 registrations, 187
dispensaries were initially deemed eligible to apply for permits to remain operational. All
other dispensaries were to cease operation by June 7, 2010. On May 4, 2010, the city sent
“courtesy notices” to the 439 dispensaries that were being ordered to shut their doors.13
Several listed establishments were later identified as ancillary businesses (e.g., clinics offer-
ing medical marijuana recommendations or smoke shops selling paraphernalia).14 Our own
scrutiny of the city’s lists eliminated several duplicate listings, yielding 180 dispensaries
eligible to remain open and 417 dispensaries ordered to close.
Some dispensaries and patient advocates responded to the city’s notices by filing tem-
porary restraining orders to prevent the closures. Efforts to win temporary restraining
orders proved unsuccessful (Yoshino 2010; Kim 2010). That the appeals continued up to
3 days before June 7, 2010, however, suggests that dispensaries were not preparing weeks
in advance to close. While the city declined to detail how the law would be enforced,
it noted that it would “rely on reports from police, neighbors and building inspectors to
identify violators” (Hoeffel 2010a). This characterization suggests that special patrols and
10
Hardship exemption requests often cited delays beyond a dispensary’s control, such as in receiving
a city business tax registration certificate required for registering. However, many later applicants cited
failure to register because of fear imposed by federal authorities (Hoeffel 2009).
11
http://clkrep.lacity.org/onlinedocs/2009/09-0964_ord_180749.pdf
12
See http://clkrep.lacity.org/onlinedocs/2008/08-0923_ord_181069.pdf
13
For a sample letter, see http://blogs.laweekly.com/informer/2010/05/pot_shops_warned_to_
close.php
14
E.g., see http://www.latimes.com/local/la-me-closing-dispensaries-htmlstory.html.

8
enforcement units were not allocated, although we know of no data on this point.
Early reports on compliance with the law indicated that most of the dispensaries ordered
to close on June 7, 2010 did so.15 Within weeks, however, compliance seemed to break
down and legal challenges to the law mounted (Wei and Romero 2010; Guerrero, 2010).
In December 2010 an injunction was issued against the law and in January 2011 the city’s
dispensary closures were formally invalidated.16

2.1 Crime Data

To assess the relationship between medical marijuana dispensaries and crime, we analyze
incident level crime data provided by the Los Angeles Police Department (LAPD) and
the Los Angeles Sheriff’s Department (LASD) to The Los Angeles Times (LAT) as part
of its “Mapping L.A.” project.”17 The LAPD provides police services to neighborhoods
throughout the city while the LASD provides primary police services to all unincorporated
parts of Los Angeles County, to all Metropolitan Transportation Authority (MTA) stations
within the city and beyond, nine community college campuses throughout the county, as
well as to numerous contract cities in the county.18 The crime data include the date, time
and location of reported crimes at the block level.19
Crimes not reported to the LAPD or LASD are not in our dataset. Thus, some crime
committed in adjacent jurisdictions, such as the City of Santa Monica, is not captured
15
The LA City Attorney’s office estimated that only 20-30 stores defied the initial closure order (Rubin
and Hoeffel 2010).
16
Specifically, a Los Angeles County Superior Court Judge issued an injunction barring the city from
enforcing many aspects of the medical marijuana ordinance, including dispensary closures based on regis-
tration (or lack thereof) at the time of the moratorium (Hoeffel 2010a).
17
See http://maps.latimes.com/crime/ for details on the data. As noted in the FAQ, the LAPD and
LASD provide the raw data directly to the LAT.
18
According to the LAT, they capture data from 42 LASD contract cities: Agoura Hills, Artesia, Avalon,
Bellflower, Bradbury, Calabasas, Carson, Cerritos, Commerce, Compton, Cudahy, Diamond Bar, Duarte,
Hawaiian Gardens, Hidden Hills, Industry, La Canada Flintridge, La Habra Heights, La Mirada, La Puente,
Lakewood, Lancaster, Lawndale, Lomita, Lynwood, Malibu, Maywood, Norwalk, Palmdale, Paramount,
Pico Rivera, Rancho Palos Verdes, Rolling Hills, Rolling Hills Estates, Rosemead, San Dimas, Santa Clarita,
South El Monte, Temple City, Walnut, West Hollywood, Westlake Village. See http://maps.latimes.
com/about/#why-no-crime-reports
19
For mapping purposes, the LAT data repeat some crimes that occur at the boundaries of neighborhoods.
See http://maps.latimes.com/about/#double-crime-counting. We clean the data to eliminate any
multiple counts introduced for mapping purposes.

9
here.20 Nonetheless, the LAT data should contain the vast majority of crimes occurring
around the City and County of Los Angeles.21 In addition, the dataset has gone through
a rigorous vetting process by the Los Angeles Times, including the correction of numerous
data omissions and flaws.22 More importantly for our estimation, since any data omissions
are determined by geographic coverage not dispensary closures and occur both pre and
post closure, the missing data should not bias our findings.
The LAT crime data capture Part I offenses, defined as “serious crimes [that] occur
with regularity in all areas of the country, and are likely to be reported to police” (FBI
2010). We analyze total Part I crimes and subcategories of Part I crimes defined by either
the LAPD’s Crime Class Code Hierarchy or the FBI’s coding, which is used by the LASD
and in national level datasets such as the Uniform Crime Reports. The LAPD’s categories
differ somewhat from the FBI’s coding; specifically, the LAPD breaks out theft into general
theft and theft from vehicles because of the high share of crime in the theft from vehicles
category.23 Since the LAPD comprises 98% of our data in the dispensary analysis and
much of data for the restaurant analysis, we have adopted this as our main coding system.

2.2 Dispensary Data

We use information from the Los Angeles City Attorney’s Office (LACAO) on the exact
location of dispensaries subject to closure or allowed to remain open. Of the 597 dispen-
saries in our dataset, 417 were ordered to close and 180 were allowed to remain open.24
Figure 1 shows the geographic distribution of dispensaries by closure status.
We code dispensaries based on their closure order status, adopting an intent-to-treat
(ITT) approach to the analysis. As described above, the city initially reported that nearly
20
Some crime in these areas are included because the LASD had jurisdiction, e.g., at an MTA stop, or
because the LAPD was called.
21
Together the LAPD and LASD provide police services for about 63% of the 9.9 million LA County
residents – about 3.8 million within the city, 1.1 million in unincorporated parts of the county and another
1.4 million in contract cities (based on authors’ calculations).
22
For details see http://maps.latimes.com/about/#crime-data-sources
23
For a discussion of the LAPD’s categories for Part I offenses, see http://projects.latimes.com/
mapping-la/about/#what-crimes
24
The initial number of dispensaries cited by the LACAO was closer to 640 shops. However, careful
scrutiny of the official list of dispensaries ordered to close and allowed to remain open revealed many
duplicate listings as well as listing of shops that were subsequently deemed not to be dispensaries (these
were generally clinics offering medical marijuana recommendations or smoke shops selling paraphernalia).

10
all shops complied with their orders to close. However, a small number of dispensaries
that were supposed to close were later raided by the LAPD (see Rubin and Hoeffel 2010)
or reported to be operating by the LA Weekly, (see Wei and Romero 2010). While the
LACAO later indicated that the closure orders were not nearly as effective as they originally
claimed and that many dispensaries defied their orders, it is unclear whether the revised
statements apply to the short or long run. If indeed a substantial number of dispensaries
failed to close immediately after the June 7, 2010, the results may significantly understate
the true effect of dispensary closures on crime. In sensitivity checks, we replicate our
analysis but recode as open or drop entirely dispensaries that, according to reports from
the Los Angeles Times and the LA Weekly, defied the city’s orders to close (Rubin and
Hoeffel 2010, Wei and Romero 2010). Such revisions do not materially affect our results.

3 Empirical Strategy
We estimate the effect of dispensary closures on crime using a regression of the following
basic form:
Citd = αi + β ∗ 1(closedit ) + δt + it (1)

where Citd is the number of crimes within a distance d of a dispensary i on date t, αi

is a dispensary fixed effect, δt is a date fixed effect and closed is an interaction between
1(date ≥ June7), an indicator for dates after and including the June 7, 2010 closures,
and 1(closed), an indicator for dispensary closure status, as determined by city orders and
in some sensitivity checks by reports of defying these orders. The main post June 7 and
closure indicators are subsumed in fixed effects for date and dispensaries, respectively.
Given the (non-negative) count nature of the crime data, we estimate equation (1) using
a Poisson regression model. All standard errors allow for two-way clustering to account
for serial correlation of an arbitrary structure at the dispensary level as well as correlation
across dispensaries on a given day (Cameron, Gelbach and Miller, 2011). With robust
standard errors, the Poisson model is a quasi-maximum likelihood estimator that does
not impose the equality of mean and variance condition and leads to consistent standard
errors (Winkelmann and Zimmermann 1992). Allowing for overdispersion with a negative

11
binomial model, which we show in sensitivity checks, yields similar results.
We use the 10 days prior to and 10 days after (but not including) the closure date
(June 7) for the main analysis. We drop June 7 because of the possibility of enhanced
police presence to enforce closures, the general confusion over the meaning of the date (e.g.,
whether stores had to shutter on June 7 or the day after), and potential protests against
closing on the proscribed date. In addition, since there can be a lag between the commission
and reporting of a crime, crime reports for the closure date may be contaminated by crime
committed in the pre-closure period, which would attenuate the estimates. As we show in
sensitivity checks, the results are robust to the inclusion of the closure date.
We focus on a short (20 day) time window because many closures were temporary.
As the legality and enforceability of the measure came under question, many dispensaries
reopened or were replaced by other businesses. In addition, the short window increases
our confidence that our results are due to dispensary closures and not differences in longer
run crime trends around open and closed dispensaries. Analyses that extend the window
around June 7 but decompose the post-closure period into smaller time periods confirm
the immediate but temporary impact of closure orders on crime.
The identifying assumption for our analysis is that, in the absence of closures, crime
and the factors that impact crime in the immediate area around dispensaries subject to
closure would be similar (or at least not differentially different post-closure) to those in
the immediate area around dispensaries allowed to remain open. While the somewhat
arbitrary process the city took to determine closure status suggests this should be the case,
we present several pieces of evidence to support this claim.
First, in Table 1, we show that daily crime counts at 1 or 1/3 mile around dispensaries
ordered to close (col (1)) and allowed to remain open (col (2)) are virtually indistinguishable
in the pre-period.25 This remains true even when we narrow to 1/4 or 1/8 mile around
dispensaries (not shown), and for total Part I crime as well as for Part I property and
violent crime, and across subcategories of Part I crime.
Second, we compare crime trends around dispensaries ordered to close and those al-
25
We use 1/3 of a mile because that is the distance where we detect changes in crime across both
dispensaries and restaurants, i.e., this distance best balances the trade-off between local crime effects and
the loss of power from considering a very small area with few crimes on any given day.

12
lowed to remain open. As shown in Figure 2a, at 1 mile, average daily Part I crimes are
indistinguishable by closure status in both the pre and post period. A different pattern
emerges when we consider crime in the immediate area around a dispensary. Specifically, at
1/3 of a mile (Figure 2b), average Part I crime tracks closely in the pre-period but diverges
after June 7. This is suggestive of a closure effect on localized crime. The divergence in
Part I crime trends after June 7 can also be seen at 1/8 of a mile (Figure 2c).26
Third, we find that dispensaries are indistinguishable based on closure status across
a range of zip code characteristics from the 2010 census and 2011 ACS. Dispensaries are
located in zip codes of about 42,000 people or 15,500 households, irrespective of closure
status. Median household income, median age, housing occupancy rates and the share
foreign-born are also independent of closure status. We also consider Walk Scores, a walk-
ability measure that rates an address based on a weighted function of walking distance to
amenities in 9 different categories, such as grocery, restaurants, and entertainment. (Walk
Score 2011). Scores, which are on a scale of 0 to 100, are adjusted for pedestrian-friendliness,
such as block connectivity. Dispensaries in Los Angeles tend to be in very walkable areas
(scores of 70-89 are considered ‘Very Walkable’) and while we fail to reject zero difference
in Walk Scores across dispensary types at the 10% level (p-value 0.056), this difference is
quite small in magnitude – about 2.3 points (74.9 vs. 77.2) or less than 1/5 of a standard
deviation (13.7 points) in Walk Scores. A comparison of the distribution of Walk Scores
by closure status in Appendix Figure 2 shows that the mean difference is driven by a few
dispensaries in very low Walk Score areas that were ordered to close.27 Our results are not
sensitive to these outliers. In short, the descriptive statistics are consistent with de-facto
random closures, with open dispensaries serving as good controls for closed dispensaries.
Next we consider the possibility of spatial clustering of dispensary closures. If closure
status is geographically clustered, it could impact inference and lead to over-rejection of
the null of no effect of closures on crime (Barrios et al. 2012). In the last row of Table 1,
26
Consistent with the patterns at 1 mile, we see no obvious impact of dispensary closures on citywide
Part I crime counts or counts by region (see Appendix Figures 1a and 1b). This null effect is consistent
with either the change in crime around dispensaries being too small to show up in citywide crime counts
or a displacement of crime from areas farther to nearer to closed dispensaries.
27
Excluding the 31 dispensaries characterized as “Car-Dependent,” the difference in mean Walk Scores
for dispensaries ordered to close and those allowed to remain drops to 0.5 with values of 76.9 and 77.4
respectively.

13
however, we demonstrate that the likelihood that a closed dispensary’s nearest neighbor
is open is similar to the likelihood that an open dispensary’s nearest neighbor is open.
Both probabilities are about a third and are statistically indistinguishable from each other,
suggesting that closure status is not geographically clustered, and standard approaches to
inference (e.g., two-way clustering) should be valid in this setting.
A separate but related issue is the overlap of crime catchment areas. Because dis-
pensaries often locate close to one another, a given crime may be assigned to multiple
dispensaries, particularly as we widen the catchment area. This overlap will likely bias our
results towards zero, and potentially affect inference. To deal with this issue, below (in sec-
tion 5.1) we present analyses using only dispensaries without nearby neighbors. Consistent
with the predicted downward bias, the results are larger in magnitude but still statistically
significant when we reduce or eliminate overlap, despite a greatly reduced sample size.28
Finally, we run a series of placebo regressions that analyze crime 1, 2, or 3 months
prior to the closure orders taking effect (section 5.1). These regressions provide another
check on whether our results are driven by differential crime trends around dispensaries
that registered with the city in 2007 (and thus were eligible to remain open) and those
that failed to do so. As discussed below, the placebo checks provide further support for
the identification strategy.

4 Main Results
Our main Poisson regression model estimates of the impact of dispensary closures on total
Part I crimes are in Table 2. In col (1), we show the pre-closure mean of Part I crime at
1/8, 1/4, 1/3, 1/2, 1 and 2 miles around dispensaries ordered to close. For crime at each
of these distances, we show the results of the intent-to-treat (ITT) analysis (col (2)), the
analysis that recodes dispensaries that were known to have defied closure orders as open
(col (3)) and the analysis that drops known defiers (col (4)).
At distances of 1 to 2 miles, the estimated effects of closure on Part I crime are rather
precisely estimated zeros. At 1/2 of a mile the effects are larger but insignificantly different
from zero. At 1/3 of a mile, we detect increases in Part I crime of about 12 to 14% around
28
We thank Steve Raphael for suggesting this check for a related, unreleased RAND report.

14
dispensaries ordered to closed relative to those allowed to remain open.
The point estimates imply increases between 14 to 16% at 1/4 of a mile and 23 to 24%
at 1/8 of a mile around dispensaries ordered to close relative to those allowed to remain
open. These findings suggest that while dispensary closures affect (increase) total Part I
crime, they do so only in a very localized fashion (i.e., in the immediate vicinity of the
affected business). Columns 2 and 3 show a similar pattern of results when we recode
known defiers or drop them from our sample.29

4.1 Robustness Checks and Sensitivity Analyses

In Table 3 we present results from placebo regressions in which we repeat the ITT anal-
ysis but code the closure period as 1, 2 or 3 months prior to the actual June 7 closure
date. As in our main analysis, we use 10 days each of pre and post data and drop the
placebo closure date. Col (1) in Table 3 repeats our main ITT result from Table 2, while
cols. (2) - (4) show results of the ITT analysis using placebo closure dates. Unlike the
true closure period results, we find no clear pattern of results using placebo dates (e.g.,
monotonically increasing/decreasing with distance). The placebo estimates are sometimes
positive and sometimes negative and are never significant at short distances (less than 0.5
miles) around a dispensary. These findings suggest our main results are not driven by
systematic differences in crime trends by closure orders.
Next, we check whether the results are sensitive to model choice. First, we re-run the
regressions using a negative binomial model in place of a Poisson model to deal with over
dispersion. Second, we use a zero-inflated Poisson regression model to handle excess zeros
in the data. As shown in Appendix Table 2, the estimates from both models are quite
similar to our main Table 2 estimates.
In Appendix Table 3, we explore the impact of extending the study window. Specifically,
we present (ITT) results that lengthen the study period to 60 days but include 3 separate
indicators for closure days 1-10 (col 1), 11-20 (col 2), and 21-30 (col 3). We break up the
29
Since defiant dispensaries may differ systematically from other dispensaries, recoding them as open or
dropping them could introduce bias. In practice, since we know of only nine defiant dispensaries, we use
this exercise to demonstrate that the results are not driven by the treatment of defiant dispensaries. As
expected, the treatment of defiers has little effect on the magnitude or significance of the coefficients.

15
extended post-period into three parts because lengthening the post-closure period likely
introduces control days to the treatment period since, as documented in McDonald and
Pelisek (2009), some dispensaries reopened within a couple of weeks of closure.30 Col (1)
shows that increasing the pre-period generates results similar to our main specification with
slightly tighter confidence intervals: the estimated effect of the first 10 days of closure on
total Part I crime at 1/8 of a mile is almost 30% and is significant at the 1% level. At
1/4 and 1/3 of a mile, the first 10-day estimates are 12 and 9%, respectively, consistent
with a decreasing monotonic relationship between the distance around dispensaries and
the change in crime. Col (2) shows the effects of dispensary closures 11-20 days after the
event. We find effects that are both smaller in magnitude and only significantly different
from zero (at the 10% level) at 1/3 of a mile. Estimates for the 21-30 day closure period
in col (3) are much less precise and are inconsistent in sign. This analysis confirms that
pre-period trends are not driving our findings and that temporary dispensary closures had
an immediate and temporary impact on crime.
In Appendix Table 4, we test the sensitivity of the results to confusion over the closure
date and potential lags in crime reporting. Specifically, we drop June 6-8, 2010 from the
analysis. Because this significantly limits our sample, we show results using 9, 19 and 29
days on either side of the June 7, 2010 but excluding June 6-8. Those results are quite
similar and, in many cases, more precisely estimated than our main Table 2 results.
Finally we examine the effect of the multiple counting of crimes due to geographic
overlap in dispensary neighborhoods. Because closure status is not geographically clustered,
the main effect of this overlap is to mechanically bias our estimates towards zero, leading
to an underestimate of the magnitude of the closure effect. To see this, we would ideally
analyze dispensaries that have no neighbors within a wide radius, e.g., 1 mile. In practice,
less than 5% of dispensaries are so geographically isolated. Consequently, in Appendix
Table 5, we show sensitivity checks using the less restrictive requirements that dispensaries
have a nearest neighbor more than 1/3 mile or more than 1/2 mile away. Using these
restrictions leaves us with 158 dispensaries with a nearest neighbor more than 1/3 mile
30
In addition, lengthening the post-period could reduce our estimates if they capture a generic busi-
ness effect, as suggested by the restaurant analysis below, and new businesses open at the site of closed
dispensaries.

16
away and 79 dispensaries with a nearest neighbor more than 1/2 mile away.
Across both restricted samples, the magnitude of the change in Part I crime is con-
sistently larger than in the sample as a whole. The results for crime at 1/3 and 1/4 of
a mile are statistically significant, despite the greatly reduced sample size. Restricting to
dispensaries with a nearest neighbor more than 1/3 mile away, the estimates imply that
Part I crime within a radius of 1/4 mile was about 47% higher around dispensaries ordered
to close compared to those allowed to remain open, more than triple the main estimate in
Table 2. When we restrict to the 79 dispensaries with a nearest neighbor more than 1/2
mile away, the estimates imply that Part I crime within 1/4 mile is 93 percent higher around
dispensaries ordered to close compared to those allowed to remain open. While the results
in Appendix Table 5 follow the expected pattern of increasing in magnitude as we reduce
catchment overlap, the set of geographically isolated dispensaries may differ on other unac-
counted for dimensions. As such, we cannot use the difference in these coefficients relative
to the full sample to measure the average downward bias. Rather, these results provide
suggestive evidence that our main results underestimate the true effect sizes.

4.2 Results for Crime by Type

We next analyze categories of Part I crimes, which are divided by the FBI into property and
violent crimes. We estimate separate models for the following property crimes: burglary,
grand theft auto, and larceny theft. Larceny theft is separately broken out as thefts from
vehicles and other theft. Arson, a sub-category of Part I property crime is too rare to
analyze separately. For violent Part I crime, we analyze aggravated assault and robbery.
Murder and rape, which are included in total Part I violent crimes, are also too rare to
analyze separately (see Appendix Table 6 for pre-closure mean).31
Table 4 shows the impact of dispensary closures on crime by type using the preferred
ITT approach that codes closures according to order status. These results show that the
effect of dispensary closures loads on property crimes, specifically larceny, and, breaking
that out further, theft from vehicles. As with total crime, the effects are very local and
31
Appendix Table 6 makes apparent the difficulty in analyzing rare crimes. For example, even at 2 miles,
there were only an average of 0.04 murders per day around dispensaries ordered to close. At 1/8 of a mile,
the count is 0.0002.

17
monotonically decrease with catchment area radii. This monotonic decrease in the closure
estimates and confidence intervals can be seen clearly in Figures 3 and 4, which plot the
implied percent change in Part I crimes and theft from vehicles, respectively, along with
95 percent confidence intervals at distances from 1/8 to 2 miles. At distances of 1/2 mile
or greater we find no effect of closures on crime, and the small coefficients with relatively
tight confidence intervals means we can explicitly rule out even small increases in crime at
these larger distances. At 1/3 of a mile the models imply that property crimes increase by
12%, largely driven by increases in larceny and, specifically, theft from vehicles. Even more
locally, the estimated effects imply that thefts from vehicles increase by almost 30% at 1/4
of a mile and by 100% at 1/8 of a mile around dispensaries ordered to close relative to those
allowed to remain open. While the percent increase in crime near closed dispensaries is
large, proper interpretation of these effects must take into account the low number of crimes
around each dispensary on any given day. For example, combining the results of Tables
1 and 2, we see that closing a dispensary leads to just 0.0512 additional crimes (0.0399
additional property crimes) per day within a third of a mile of the closed dispensary.
Burglary is the one exception to the general monotonic pattern. Here we find a large,
negative and marginally significant (p-value=0.07) coefficient for closures at 1/8th of a
mile, positive and statistically insignificant coefficients at 1/4th, 1/3rd and 1/2 of a mile,
a small negative and statistically insignificant coefficient at 1 mile, and a small negative
statistically significant coefficient at 2 miles. While intriguing, this non-monotonic pattern
does not admit to an obvious explanation. In addition, unlike the results for total crime or
larceny, the burglary results do not hold up in robustness checks and are based on a very
small number of events, with an average of 0.0245 burglary per day at 1/8 of a mile. As
such, this result should be interpreted with caution.
As with our main results, we find that results for crime by type are insensitive to the
treatment of defiers (see Appendix Table 7, which drops defiers, and Appendix Table 8,
which recodes them as open) or the inclusion of the closure date (see Appendix Table 9).

18
5 More crime or displaced crime?
A crucial question in determining the social costs of crime associated with dispensary
closures is whether the changes represent an increase (or decrease) in total crime or a shift
of crime across either space or time. If crime is spatially displaced, then the increase in
crime near a closed dispensary may be offset by decreases in crime further away. Since our
main results show that closures lead to significant crime increases at distances of 1/4 to 1/3
of a mile around a dispensary, spatial displacement would imply corresponding decreases in
crime at distances of greater than 1/4 to 1/3 mile. To check for this type of displacement,
we examine the impact of closures on crime in concentric rings around each dispensary.32
Specifically, in Table 5 we analyze crime occurring between 1/4 and 1/3 of a mile, 1/3
and 1/2 of a mile, 1/2 to 1, 1/2 to 2 and 1 to 2 miles around dispensaries. At distances of 1/4
to 1/3 of a mile (a band fully contained within the radii where we find increases in crime)
the coefficient on closure is, with the exception of violent crimes, positive. The increase
within this band is not statistically distinguishable from zero, however. At 1/3 to 1/2 of a
mile, the property crime estimate is negative but close to zero, albeit with a wide confidence
interval. Since the overlap issue discussed previously should be exacerbated at larger radii,
the magnitude of the estimates within the larger rings could be more downward biased
than those at smaller distances. But given that these coefficients are never significant,
these results do not provide strong evidence for (or against) spatial displacement.
Analogous to spatial displacement, temporal displacement of crime would mean that the
changes in crime associated with closures are offset by changes in crime either before or after
the closure period. While the dispensary closure date was well known in advance, there are
no clear “re-opening” dates.33 As such if criminal activity exhibited a significant ex-ante
temporal elasticity, we would expect a decrease in crime around dispensaries scheduled to
32
An alternative approach to checking for displacement would aggregate our data to larger geographic
levels as in Freedman and Owens (2011) or sum results across areas as in Aliprantis and Hartley (2015).
We choose not to take this approach for several reasons. First, dispensaries may border neighborhoods or
police reporting districts, two potential levels of analysis. In this case, aggregation can mask displacement
as an increase in crime in an area assigned the dispensary appears bigger when measured relative to a
decrease in a neighboring area. In addition, the effects we observe here may be too small relative to the
city or region to statistically detect in aggregated data, even in the absence of any actual displacement.
33
Note that re-opening could be due to either the dispensary deciding to re-open (as many did) or the
space itself being taken over by another business.

19
close but prior to actual closures as criminals waited until June 7 to commit crimes.
We find little evidence of pre-closure differences in either the level or trend in daily
crime around dispensaries ordered to close relative to those allowed to remain open. Most
directly, since extending the pre-period window around June 7, 2010 yields similar results
(see Appendix Table 3), it is unlikely that a pre-period decline in crime in anticipation of
future crime commission can explain our results. In other words, criminals do not appear
to postpone (or move forward) crimes in anticipation of the mass closure of dispensaries.
Given the variation in pre-closure crime levels, we can generally rule out economically
significant temporal displacement in the period just prior to the June 7, 2010 closures.

6 Restaurant Closures and Crime

While the results above demonstrate that crime increased near dispensaries that closed
relative to those allowed to remain open, such an increase is consistent with multiple mech-
anisms (see Section 2). Here, we attempt to disentangle some of the possible mechanisms
by testing whether this closure effect is unique to dispensaries or reflects a more general
business closure phenomenon. We do this by performing a parallel analysis for temporary
restaurant closures due to public health code violations in Los Angeles County.

6.1 Background on Restaurant Closures

In Los Angeles County, the Department of Public Health (DPH) is charged, under the Cal-
ifornia Uniform Retail Food Facilities Law (CURFFL), with enforcing uniform statewide
health and sanitation standards for retail food facilities according to the“science-based
standards.” DPH inspects all facilities that provide food to the public (restaurants, bak-
eries and markets). Based on the guidelines outlined in the California Retail Food Code
(Cal-Code), DPH environmental health specialists grade restaurants on various health and
sanitation measures including improper holding temperatures, poor personal hygiene of
food employees, contaminated equipment and the presence of vermin and, depending on
the outcome, may order a temporary shutdown for remediation.
Based on a Food Official Inspection Report (FOIR), restaurants receive a numerical
score between 0-100. Restaurants that score 70 and above are given a grade card that

20
must be posted in an easily visible location (90-100 is an ”A”, 80-89 a ”B”, 70-79 a ”C”).
Restaurants that score less than 70 receive a numerical score card rather than a grade.
Restaurants that score less than 70 twice in any twelve month period are subject to closure
and the filing of a court case. Such closures are rare. More commonly, if the inspection
turns up a “major violation,” meaning a violation, such as vermin harborage or infesta-
tion, sewage disposal problems or food temperature problems, that poses an imminent
health hazard, the restaurant is subject to immediate closure without a permit suspen-
sion hearing.34 Restaurants closed for major violations remain closed until a subsequent
follow-up inspection confirms that the situation has been satisfactorily resolved. Follow-up
inspections generally take place within two-days but can take up to a week.35
Restaurants are inspected twice a year, although those that handle large quantities
of “risky foods” (e.g., meat) or consistently score low may be inspected three times a
year. The DPH may conduct an additional inspection in response to consumer complaints.
Individual inspectors work specific geographic areas determined by the local environmental
health office. They work with supervisors to set a schedule for restaurant inspections
in increments of one or more months. While inspection scheduling is not standardized,
inspections are, depending on the specific supervisor, scheduled weeks to months ahead of
time. As such, although the timing of inspections are not explicitly randomized, the process
makes it highly unlikely that the exact timing of inspections are correlated with trends in
crime in the immediate area around each restaurant. In addition, DPH officials have stated
that local conditions (including crime) have no bearing on the timing of inspections.

6.2 Restaurant Data

Data on restaurant closures are from the Environmental Health Division (EHD) of the Los
Angeles County DPH, the enforcement division in charge of inspecting retail food facilities.
The EHD data include the name and exact location of restaurants closed by the agency, the
date of closure, the reason for closure, and in most cases a reopen date. In total, we have
888 restaurant closures during our study period, February 1, 2010 to October 31, 2010.36
34
See: http://publichealth.lacounty.gov/eh/docs/RetailFoodInspectionGuide.pdf
35
This timing is based on conversations with LA County Department of Public Health officials.
36
The study period was determined by the original data made available to us by the LAPD.

21
Most closures are caused by “major violations,” with roughly two-thirds of the closures in
our sample due to vermin harborage or infestation. The next most common offense is a lack
of potable or hot water, which accounts for 12 percent of closures. Of the 888 closures, 766
or 86% of them have valid reopen dates. In all the cases we investigated, restaurants with
no-reopen dates were in fact open and operational. In multiple conversations with EHA,
we were unable to obtain any official reason for missing reopen dates. As described below,
we take three approaches to dealing with restaurants with missing restaurant reopen dates
– assigning the median closure period of 2 days, treating them as permanently closed or
dropping them from the sample. Our primary approach uses the median closure period
but, as shown below, the results are not sensitive to this choice.

6.3 Restaurant Analysis

We focus on the universe of Los Angeles County restaurants that were closed for heath
code violations between February 1, 2010 to October 31, 2010.37 Using the same basic
specification as in equation (1), we define 1(closed) as the period between a restaurant’s
closure and reopen date. Because we restrict the sample to restaurants with health code
violations, the identifying assumption for this analysis is that the timing of closures is
uncorrelated with crime in the area immediately around the affected restaurant.
Paralleling our dispensary analysis, we drop each first closure day in the analysis. In
addition to the concern that crimes reported on closure dates may have occurred prior
to that date, many restaurants will be closed for only part of the first closure day. In
other words, some restaurants ordered to close temporarily remain open for part of the
first closure day – both before and during the inspection. However, as with dispensaries,
the results are similar when we include the first closure day in the analysis (shown below).
Appendix Table 10 shows summary statistics for restaurants in the 10 days prior to
closure. Since all restaurants in our sample were subject to closure, there are no separate
time-invariant restaurant characteristics for closed and open restaurants. Rather, these
summary statistics show pre-closure characteristics of neighborhoods around restaurants
37
Since we will require 10 days of pre and post closure data, the restriction is actually restaurants closed
between February 10, 2010 and October 21, 2010.

22
subject to closure during our sample period. In general, the neighborhoods around restau-
rants do not look dramatically different from that around dispensaries (in Table 1). The
most noteworthy differences are that these neighborhoods are slightly more populous, with
larger families (i.e., fewer households, despite more people) and lower family incomes. And,
consistent with the fact that restaurant closures occur across the county, not just in the
city of Los Angeles, the average Walk Score is slightly lower (71.1) around restaurants than
either dispensaries ordered to close (74.9) or allowed to remain open (77.2).
While the inspection scheduling process makes it unlikely that inspections are corre-
lated with crime (since it would require that the DPH be able to predict crime at a very
disaggregated level), a related concern is that the probability of closure conditional on an
inspection is correlated with local crime conditions.38 If the probability of closure is affected
when crime in the immediate vicinity of a restaurant is rising – because, for example, the
inspector does a less rigorous review in order to minimize his exposure to crime – it could
bias our results. To assess these concerns, we run placebo regressions to test for differences
in crime within 1/4, 1/3, 1/2, 1 or 2 miles around restaurants in the days leading up to a
closure. In other words, we estimate a regression of the form in (1) but define a placebo
closed dummy equal to 1 for the same length of time as the actual closure for the days
prior to the closure event (see columns 1-3 in Appendix Table 11). As an alternate test, we
define a placebo closed indicator for the day prior to, or the 2 days prior to the closure date
(see columns 4-5 in Appendix Table 11). In all cases, we find no statistically significant
relationship between the placebo closures and crime. The point estimates are also small
in magnitude, with the exception of the 1 day dummy (column 4), which, representing the
shortest placebo time period, also has the largest standard errors. In short, we find no
evidence of systematic changes in crime in the days leading up to these restaurant closures.
38
The importance of plausibly exogenous restaurant closures status is made clear by earlier work doc-
umenting a the complex relationship between crime and the business activity. For example Greenbaum
and Tita (2004) find that surges in violence leads to less business formation and downsizing, while Sloan,
Caudill and Mixon Jr. (2015) find that criminal activity is positively correlated with restaurant openings.

23
6.4 Restaurant Results
In Table 6 we show restaurant results that (i) recode those with missing reopen dates as
having been closed for the median number of days closed across the sample, 2 days (col
(2)), (ii) treat those with missing reopen dates as closed through the entire post-period
(col (3)), or (iii) drop those restaurants with missing reopen dates (col (4)). As with the
dispensary analysis, we limit this analysis to the 10 days prior to and 10 days after any
restaurant’s closure.39 Since results at 1/8 of a mile generally do not converge, we show
results for crime at 2, 1, 1/2, 1/3 and 1/4 mile around restaurants. Pre-closure means for
Part I crime at each of these distances are provided in col (1).
Table 6 indicates that total Part I crime increases during temporary restaurant closures.
At 1/3 of a mile, total Part I crime increases by about 9 to 12% around closed restaurants
relative to open restaurants that were temporarily shut down within plus or minus 10 days.
The results are similar irrespective of the treatment of restaurants without re-open dates.
In addition, the results show a monotonic increase in the effect size as distance narrows up
until 1/4 of a mile, at which point the coefficient is small and statistically insignificant.40
Table 7 presents results for the breakdown of crime by type, where restaurants with
missing reopen dates are coded as closed for the median length of time in the data. As
with dispensaries, we find that the effects of closures are concentrated on property crimes,
specifically thefts from vehicles. The estimates imply an almost 30% increase in thefts
from vehicles at 1/4 of a mile – generally the smallest radii we can analyze for restaurants.
Again as with the dispensaries results, the effects quickly diminish with distance, becoming
not just insignificant but also small in magnitude at distances of 1 mile and greater. As
39
In addition to making this analysis as similar as possible to the dispensary closure analysis, the short
time window addresses a concern regarding clustering in inspections. Specifically, the use of a short window
around each closure helps ensure that identification is not affected by any gross correlations between the
timing of inspections and local crime and reduces the possibility that any results are due to differences in
medium or long-run crime trends. The focus on restaurants with similar closure dates also mechanically
reduces overlap in the catchment areas simply by reducing the number of restaurants examined on any
given day, which as previously discussed introduces a downward bias in our estimates.
40
The difference in the crime change-distance pattern for restaurants and dispensaries likely reflects
differences in catchment overlap and statistical power and not necessarily any difference in the magnitude
of the effect across establishment types. While we have more restaurants than dispensaries (888 restaurants
vs. 597 dispensaries), restaurants are generally closed for only a couple days and, more importantly, these
closures are spread out over 250 calendar days as opposed to just one period for dispensaries.

24
detailed in the appendix, these results are robust to several additional sensitivity checks:
lengthening the window of time around restaurant closures (Appendix Table 12), including
closure days (Appendix Table 13), coding restaurants with missing re-open dates as closed
for the full post-closure period (Appendix Tables 14) and dropping restaurants with missing
reopen dates (Appendix Table 15).
We next check for the displacement of crime either spatially or temporally in response
to temporary restaurant closures. As with dispensaries, we check for spatial displacement
by examining changes in crime in rings of various sizes around closed restaurants. Table 8
shows the crime changes occurring between 1/4 and 1/3 of a mile, 1/3 and 1/2 of a mile,
1/2 to 1 mile, 1/2 to 2 miles and 1 to 2 miles around closed restaurants. At 1/4 to 1/3 of
a mile, which is fully contained within the radii where we find increases in crime around
closed restaurants, the coefficient on closure is positive. The increase within this band
is significant only for total crimes. The point estimates then drop and are both small in
magnitude and not distinguishable from zero at 1/3 to 1/2 of a mile, suggesting that the
increase in crime is localized to distances of less than 1/3 of a mile.
To test for temporal displacement, we re-run our standard regression but supplement the
restaurant closure period indicator with dummies for both the re-open date and the re-open
date plus 1. We focus on the reopening period since restaurant closures are unexpected
and thus could not have caused pre-closure shifts in criminality. Rather, the temporary
restaurant closures could have led criminals to shift crime earlier in time to the closure
period. Such a shift would decrease crime after a reopening. Instead, as shown in Table 9,
we find significant increases in crime at 1/3 of a mile around restaurants during the closure
period but no compensating decrease in crime on either the re-open day or the day after.
The similarity in the broad pattern of results for restaurants and dispensaries despite the
differences in the nature of these businesses, the reason for and timing of their closures, and
the identifying assumptions of the analyses, provides additional evidence that the increase
in crime following dispensary closures is not spurious. Furthermore, it suggests that the
mechanism behind the decrease in crime is not dispensary-specific but indicative of a more
general effect of business closures on crime.

25
7 Modes and Mechanisms
The results presented above show that temporary dispensary closures increase crime in the
short-run and that temporary restaurant closures affect crime in a similar fashion. While
the increase in crime after both dispensary and restaurant closures may be unrelated, it
seems more likely that a common factor drives the shared pattern of results. Under this
assumption, we can rule out dispensary-specific mechanisms such as the substitution of
alcohol for marijuana or diminished access to formal dispute resolution channels in medical
marijuana markets, as the driving force behind the increase in crime. Below, we explore
the evidence for and against several possible common factors.

7.1 Walkability and the Role of “Eyes Upon the Street”

One potential common factor affecting crime may be a reduction in foot traffic. If dispen-
sary and restaurant closures reduce foot traffic, informal policing or “eyes upon the street”
(Jacobs 1961) may also be diminished and crime could increase. This hypothesis requires
that the impact of business closures on crime be mediated through customer foot traffic.
Such a connection seems intuitive since a closed business necessarily has fewer customers
than an open one. The ideal data to test this would include measures of foot traffic by
location. Given that such measures are unavailable, we use neighborhood characteristics
to proxy for the relative impact of business closures on foot traffic in an area.
To proxy for foot traffic by location, we collect “Walk Scores” from www.walkscore.com
by exact business address.41 Scores range from 0 to 100, and are based on walking paths
to amenities. Amenities within a 5 minute walk are given maximum points. More distant
amenities receive points based on a decay function, with zero points after a 30 min walk.
Pedestrian friendliness is comported into the measure based on population density, block
length and intersection density. While Walk Scores do not capture the presence of sidewalks,
street lights or speed limits, which likely improve the walking experience, they have been
shown to be a useful measure of walkability (Hirsch et al. 2013).
41
A complementary approach might be to use Dunn and Bradstreet data on the level of employment
and the composition (retail vs. wholesale) of establishments at the address and block level as in Rosenthal
and Ross (2010).

26
 Walk Score identifies four categories of addresses based on their scoring system: Car-
Dependent (0-49), Somewhat walkable (50-69), Very walkable (70-89) and Walker’s paradise
(90-100). Walkability is determined by the number and proximity of restaurants, bars,
coffee shops, grocery stores, and so on. An address with a high Walk Score has many
businesses and other features that generate foot traffic nearby whereas one with a low
Walk Score has few businesses nearby and relatively little foot-traffic.
How should the Walk Score interact with business closures to affect crime? Since a
business with a high Walk Score is located near many other businesses, its customers likely
represent a small share of local foot traffic. On the other hand, the closure of a business
in a low Walk Score area should have a proportionally large impact on total foot traffic.
As such, the eyes upon the street hypothesis (hereafter EUS) would predict that, all else
equal, the impact of business closures on crime should be negatively related to Walk Scores
(i.e., that a closure should increase crime more in low Walk Score areas.)
A more complete consideration of foot traffic must acknowledge that people are both
crime deterrents and crime targets. For very isolated, car dependent areas with little foot
traffic, a business closure could reduce crime in the area by removing the few existing crime
targets. As an extreme example, consider a business that is the only feature for 1/3 of a
mile (i.e., in an extremely car dependent area) and that its closure decreases the number
of people in the area from N to zero. Such a closure would substantially decrease foot
traffic. But, since there are virtually no remaining crime targets in the immediate area,
crime would likely decline despite the loss of crime-deterring eyes upon the street.42 In
this way, EUS predicts a non-monotonic relationship between business closures and Walk
Scores: business closures will have smaller (and in the case of isolated areas possibly even
negative) effects on crime in the most and least walkable areas and larger, positive effects
in moderately walkable areas.
In Table 10 we explore the interaction of business (dispensary or restaurant) closures
and walkability on crime. Panel A shows results for dispensary closures and Panel B for
restaurant closures. Column 1 shows the impact of closures on total crime within 1/3 of a
42
Along these lines Sandler (2012) finds that the eviction of residents from, and subsequent demolition
of Chicago public housing led to a decrease in crime in the area immediately surrounding the demolition.

27
mile for dispensaries or restaurants with Walk Scores above versus below 70, corresponding
to walkscore.com’s cutoff between ‘Very’ and ‘Somewhat’ walkable.43 We find a significant
positive closure effect on crime for both dispensaries (Panel A) and restaurants (Panel B)
with low Walk Scores, with effect sizes approximately double that found in the full sample
(i.e., compared to Tables 2 and 6). When we examine crime by type (columns 3-6), we
see that, as in the full sample, the interaction effect is driven by increases in property
crime, specifically larceny and theft from vehicles. In low Walk Score areas, dispensary or
restaurant closures have more than double the impact on property crime than they do in
high Walk Score areas.
In column 2, we further divide up businesses using separate closure dummies for the
Car-dependent, Somewhat walkable, Very walkable, and Walker’s paradise categories. Here
again we find that the closure effect is smaller in highly walkable areas (i.e., areas where a
single business closure has little impact on total foot traffic) and larger and positive in the
“somewhat walkable” areas. For “Car-dependent” areas, the sign of the coefficient flips and
becomes negative; it is also both small in magnitude and statistically indistinguishable from
zero. With the caveat that the coefficients for “car-dependent” and “somewhat walkable”
areas are only marginally statistically different (p-values of 0.074 and 0.110 for restaurants
and MMDs respectively), this pattern is consistent with a non-linear relationship between
closures and walkability as predicted by EUS.44
While the Walk Score findings suggest that our main results are driven by changes
in customer foot-traffic, interpreting the elasticities of these effects with respect to foot
traffic is difficult since we have no measure of a business’s customer base or, by extension,
43
Nearly identical results are obtained by dividing the sample into above and below median Walk Scores.
44
As an alternative proxy for foot traffic, we interacted the closure indicator with indicators for whether
the dispensary’s ZIP code was above or below median for the density of employees in all dispensary
ZIP codes based on the 2010 census ZIP Code Business Patterns data. The results of this analysis are
qualitatively consistent with out Walkscore results (see Appendix Table 16): closure effects are larger in
less dense areas, where the closure of a dispensary represents a larger proportional decrease in foot traffic.
That said, the estimates are somewhat difficult to directly compare to the Walkscore results. In particular,
while the density of employees in a ZIP code is a better measure of business activity, it suffers, as a
measure of foot traffic, from several shortcomings. First, it cannot distinguish between retail vs non-retail
establishments (e.g., business parks and factories). More significantly, ZIP codes capture relatively large
areas (over 3.6 (5.8) square miles for the median (mean) ZIP Code in either sample) while our business
closure estimates (and foot traffic more generally) are localized; we observe crime effects within a radius
of 1/3 of a mile or less, representing an area of approximately 0.35 square miles.

28
the change in the number of “eyes” associated with a closure. With this very strong
caveat in mind, we can nevertheless generate a simple back of the envelope calculation as
a face validity check on the magnitude of our coefficients with respect to EUS. Specifically
if we assume that a business has 50 customers per day, the results for the 1/3rd mile
catchment area suggest that it would take approximately 1,250-1,800 customers to deter
one property crime. If we further assume that each customer effectively contributes 15
minutes of monitoring time, then our estimates suggest that it takes roughly 300-450 eyes-
upon-the-street hours to deter one property crime.

7.2 Private Security and Public Policing

While our results are consistent with EUS, they may be consistent with several alternative
explanations. Perhaps the most plausible alternative is that businesses provide formal,
direct on-site security that deters crime. Studies of business improvement districts in Los
Angeles have demonstrated the deterrent effect of paid security services (Brooks, 2008;
Cook and MacDonald, 2011). Thus, closures may increase crime by removing security
services. But, while dispensaries typically employ many forms of security, most restaurants
have little more than security cameras, if anything at all. These cameras may not be
external, as they often are for dispensaries. And, assuming they are external, it is unclear
why restaurants would remove or disable cameras during temporary closure periods. As
such, the dismantling of private security seems unlikely to be the main driver of crime
effects that are common to both dispensaries and restaurants.
A second alternative explanation relates to changes in police presence, which have been
shown to affect similar categories of crime. Klick and Tabarrok (2005) find that a large
increase in police presence, combined with increases in closed-circuit surveillance cameras
in Washington D.C., led to a significant decrease in a combined category of theft from
vehicles and auto-thefts. Draca, Machin and Witt (2011) find that a 50% increase in police
presence in London led to a reductions in larceny theft of around 20% in the affected
neighborhoods. If closures decrease police presence, they could plausibly increase crime.
Like private security, changes in police service allocations seem unlikely to be a signifi-
cant driver of our results. For restaurants, it is hard to imagine that police would formally

29
or informally change their behavior in response to temporary restaurant closure. For dis-
pensaries, anecdotal reports suggest that, if anything, police presence may have increased
in the days following the closures in order to check for compliance. This increase would
be predicted to decrease rather than increase crime. Even if police presence did change in
response to temporary restaurant and dispensary closures, the change, based on existing
evidence from the literature, would have to be implausibly large to generate our findings.
For example Draca et al. (2011) find effects similar in magnitude to our findings in response
to a 50% increase in police presence.

8 Discussion and Conclusions

Analyzing medical marijuana dispensary closures in the City of Los Angeles, we find no
support for the idea that closing dispensaries reduces crime. Rather, temporary closures
deter some types of Part I crime. To understand the mechanism, we evaluate the impact of
temporary restaurant closures due to public health code violations. We find a nearly iden-
tical impact of these closures. Both temporary dispensary and restaurant closures increase
Part I property crime, specifically theft from vehicles, in a very localized area around closed
businesses. The magnitude of the closure effect is correlated with the relative impact of the
closure on area foot traffic. Specifically, the increase in crime due to closures (dispensary
or restaurant) is negatively correlated with neighborhood walkability in relatively walkable
areas but reverses sign or is nonexistent in areas classified as car dependent.
That the pattern of results is so similar across the temporary closure of two different
establishment types suggests a common factor may be at play. In other words, the main
findings may not capture a dispensary specific effect on crime but rather a more general
retail specific effect. Since temporary restaurant closures should have no meaningful effect
on policing patterns, police presence is unlikely to be this common factor. On the other
hand, the results on neighborhood walkability suggest that “eyes upon the street” may
account for the common pattern of results
Taken together, our results provide support for the hypothesis that retail establishments,
when operational, provide informal security through their customers. That is, to the extent
that businesses bring foot traffic or as Jane Jacobs once famously proclaimed, “eyes upon

30
the street” to a neighborhood, they may deter certain types of “dark alley” crimes. While
the idea that “eyes belonging to those we might call the natural proprietors of the street”
(Jacobs 1961) can provide public safety on city streets has spawned many studies on the
role of urban design and architecture on crime (Newman 1972; Hunter and Baumer 1982;
Glaeser and Sacerdote 2000; Foster and Giles-Corti 2008), albeit not all in support of the
hypothesis, credible empirical evidence on the impact of local activity on crime remains
quite limited. Our results are consistent with a somewhat nuanced view: increased foot-
traffic appears to decrease crime but only above a certain threshold of traffic.
Our findings have direct policy implications for regulating marijuana sales in the U.S.
They imply that dispensary closures, and potentially the closure of other types of retails
establishments, exert a significant negative externality in terms of neighborhood criminality.
A quick back of the envelope cost calculation using the change in larceny theft at 1/3
of a mile (from Table 4) and crime costs from McCollister et al. (2010) suggests that
an open dispensary provides over $30,000 per year in social benefit in terms of larcenies
prevented.45 This calculation ignores potential offsets in terms of quality-of-life issues,
such as loitering, graffiti, double parking and noise.46 In addition, the current study is
underpowered to detect any impact of closures on high-cost, low frequency crimes such
as robbery, aggravated assaults, homicide, rape or arson. Future research on the impact
of dispensaries on these low-frequency crimes and on quality-of-life issues are crucial for
understanding the full economic impact of these establishments.

References
Aliprantis, A. and D. Hartley, 2015. “Blowing it up and knocking it down: The local and
city-wide effects of demolishing high concentration public housing on crime.” Journal
of Urban Economics, 88: 67-81.
Anderson, M.D., B Hansen and D.I. Rees, 2013. “Medical Marijuana Laws, Traffic Fatali-
ties, and Alcohol Consumption,” Journal of Law and Economics, 56(2): 333-369.
45
A similar calculation using the point estimates for aggravated assault and robbery at 1/3 of a mile from
Table 4 suggest social benefit of approximately $350,000 per annum. Given the relatively large standard
errors on these estimates, this number should be mainly considered illustrative of the potential relative
costs associated with property vs violent crimes.
46
Indeed, these types of NIMBY issues may be the real cost of a dispensary to area residents.

31
Barrios, Thomas, Rebecca Diamond, Guido Imbens and Michal Kolesar, 2012. “Clus-
tering, Spatial Correlations, and Randomization Inference,” Journal of the American
Statistical Association, 107(498): 578-591.
Boggess, Lyndsay N., D.M. Perex, K. Cope, C Root and P.B. Stretesky, 2014. “Do Medical
Marijuana Centers Behave Like Locally Undesirable Land Uses? Implications for the
Geography of Health and Environmental Justice,” Urban Geography” 1-22.
Brooks, Leah , 2008. “Volunteering to be Taxed: Business Improvement Districts and
the Extra-Governmental Provision of Public Safety,” Journal of Public Economics,
92(1-2): 388-406.
Cameron, A. Colin, Jonah Gelbach and Douglas Miller, 2011. “Robust Inference with
Multi-way Clustering,” Journal of Business and Economic Statistics, 29(2): 238-249.
Carpenter, Christopher 2007. “Heavy Alcohol Use and Crime: Evidence from Underage
Drunk-Driving Laws,” Journal of Law and Economics, 50(3): 539-557.
Carpenter, Christopher and Carlos Dobkin, 2015. “The Minimum Legal Drinking Age and
Crime,” Review of Economics and Statistics, 97(2): 521-24.
Caulkins, Jonathan P., Angela Hawken, Beau Kilmer and Mark A. R. Kleiman, 2012.
Marijuana Legalization: What Everyone Needs to Know, Oxford UP, 2012.
CNN/ORC, 2014. CNN/ORC Poll, January 8, 2014. Available at http://i2.cdn.turner.
com/cnn/2014/images/01/08/cnn.orc.poll.marijuana.colorado.pdf
Cook, Philip J. and John MacDonald, 2011. “Public Safety Through Private Action: An
Economic Assessment of BIDS,” The Economic Journal, 121(May): 445-462.
Crost, Benjamin, and Daniel I. Rees. 2013. “The Minimum Legal Drinking Age and
Marijuana Use: New Estimates from the NLSY97,” Journal of Health Economics
32:47476.
Di Tella, Rafael and Ernesto Schargrodsky, 2004. “Panic on the Streets of London: Police,
Crime, and the July 2005 Terror Attacks,” American Economic Review, 94(1): 115-
133.
Draca, Mirko, Stephen Machin and Robert Witt , 2012. “Do Police Reduce Crime? Es-
timates Using the Allocation of Police Forces After a Terrorist Attack,” American
Economic Review, 101(5): 2157-2181.
Farrington, David P. 2010. “Commentary on Pedersen and Skardhamar (2010): Does
cannabis use predict non-drug offending?” Addiction, 105(1): 119-120.
FBI, 2010, “Uniform Crime Reports, Offense Definitions. Crime in the United States,
2009.” U.S. Department of Justice Federal Bureau of Investigation, Released Septem-
ber 2010.
Foster, Sarah and Billie Giles-Corti, 2008, ”The built environment, neighborhood crime
and constrained physical activity: An exploration of inconsistent findings,” Preventive
Medicine, 47(3): 241-251.

32
Franklin F. Abron, Thomas A. LaVeist, Daniel W. Webster and William K. Pan, 2010. “Al-
cohol Outlets and Violent Crime in Washington D.C.” Western Journal of Emergency
Medicine, 11(3): 283-290.
Freedman, Matthew and Emily G. Owens, 2011. “Low-income Housing Development and
Crime,” Journal of Urban Economics, 70: 115-131.
Glaeser, Edward L. and Bruce Sacerdote, 2000. “The Social Consequences Of Housing,”
Journal of Housing Economics, 9(1/2), 1-23.
Gorman, Dennis M. Paul W. Speer, Erich W. Labouvie et al. 1998. “Risk of Assaultive
Violence and Alcohol Availability in New Jersey,” American Journal of Public Health,
88(1): 97-100.
Green, Kerry M., Elaine E. Doherty Elizabeth Stuart and Margaret E. Ensminger, 2010.
“Does Heavy Adolescent Marijuana Use Lead to Criminal Involvement in Adulthood?
Evidence from a Multiwave Longitudinal Study of Urban African Americans,” Drug
and Alcohol Dependence,: 112(1-2): pp. 117-125.
Greenbaum, Robert T and George E. Tita, 2004. “The Impact of Violence Surges on
Neighbourhood Business Activity.” Urban Studies, 41(13): 2495-2514.
Grogger, J. and Michael Willis. 2000. “The Emergence of Crack Cocaine and the Rise in
Urban Crime Rates.” Review of Economics and Statistics, 82(4): 519-529.
Grubesic Tony H. and William A. Pridemore, 2011. “Alcohol Outlets and Clusters of
Violence,” International Journal of Health Geographics 10:30.
Gruenewald, Paul J. and Lillian Remer, 2006. “Changes in Outlet Densities Affect Violence
Rates,” Alcoholism Clinical & Experimental Research 30(7):pp118493.
Gruenewald, PJ, B Freisthler, L Remer, EA Lascala, AJ Treno and WR Ponicki. and
Michael Willis. 2010. “Ecological Associations of Alcohol Outlets with Underage
and Young Adult Injuries.” Alcoholism: Clinical and Experimental Research, 34(3):
519-527.
Grubesic Tony H. and William A. Pridemore, 2011. “Alcohol Outlets and Clusters of
Violence,” International Journal of Health Geographics 10:30.
Hirsch, JA, KA Moore, KR Evenson et al. 2013. “Walk Score and Transit Score and
Walking in the Multi-Ethnic Study of Atherosclerosis.” American Journal of Preven-
tive Medicine, 45(2), 158?166.
Hoeffel, John 2009. “LA’s Medical Pot Moratorium Led to a Boom Instead,” Los Angeles
Times, June 3, 2009.
Hoeffel, John 2010a. “Move it, Bud, Pot Outlets are Told,” Los Angeles Times, June 7,
2010.
Hoeffel, John 2010b. “Injunction Issued Against L.A.’s Medical Marijuana Law,” Los
Angeles Times, December 10, 2010.

33
Hoaken, Peter N.S. and Sherry H. Stewart. 2003. “Drugs of Abuse and the Elicitation of
Human Aggressive Behavior,” Addictive Behaviors, 28: 1533-1554.
Hunter, Albert and Terry L. Baumer, 1982. “Street Traffic, Social Integration, and Fear of
Crime,” Sociological Inquiry, 52(2), 22131.
Kepple, Nancy J. and Bridget Freisthler, 2012. “Exploring the Ecological Association
Between Crime and Medical Marijuana Dispensaries,” Journal of Studies on Alcohol
and Drugs 73(4): 523?530.
Kim, Victoria, 2010. “Judge Clears Way for L.A. to Begin Marijuana Dispensary Crack-
down,” Los Angeles Times, June 5, 2010.
Klick, Jonathan and Alexander Tabarrok, 2005. “Using Terror Alert Levels to Estimate
the Effect of Police on Crime,” Journal of Law and Economics, 48(1), 267-280.
Linthicum Kate and David Zahnise 2013. “L.A. Council Advances Pot Dispensary Mea-
sure,” Los Angeles Times, January 16, 2013.
McCollister, Kathryn, Michael T. French and Hai Fang, 2010. “The Cost of Crime to
Society: New Crime-Specific Estimates for Policy and Program Evaluation,” Drug
and Alcohol Dependence, 108(1-2), 98-109.
McDonald, Patrick Range and Christine Pelisek 2009. “L.A.’s Medical Weed Wars,” LA
Weekly, posted on November 23, 2009.
Marijuana Policy Project, 2008. State-by-State Medical Marijuana Laws: How to Remove
the Threat of Arrest.
Miron, Jeffrey A.1999. “Violence and the US Prohibitions of Drugs and Alcohol.“ American
Law and Economics Review, 1(1), 78.
Morris RG, TenEyck M, Barnes JC, Kovandzic TV. 2014 “The Effect of Medical Mari-
juana Laws on Crime: Evidence from State Panel Data, 1990-2006.” PLoS ONE 9(3):
e92816. doi:10.1371/journal.pone.0092816
Myerscough, Rodney and Stuart P. Taylor. 1985. “The Effects of Marijuana on Human
Physical Aggression.“ Journal of Personality and Social Psychology, 49(6): 1541-1546,
National Academy of Science (NAS), 1994. “Understanding and Preventing Violence,”
Volume 3: Social Influences, Reiss, Albert J. and Jeffrey A. Roth (eds), National
Research Council,
Newman, O., 1976. Defensible Space - People and Design in the Violent City,’ Macmillan,
New York.
Pacula, Rosalie Liccardo and Beau Kilmer, 2003. “Marijuana and Crime: Is there a Con-
nection Beyond Prohibition,” NBER Working Paper No. 10046.
Pedersen W., Skardhamar T. 2010. “Cannabis and Crime: Findings from a Longitudinal
Study.” Addiction 105(1): 10918.
Pfeiffer, Sacha and Lynn Jolicoeur, 2013., ”Will Local Bans on Medical Marijuana Dispen-
saries Hold Up Legally,” All Things Considered , WBUR March 11, 2013.

34
Resignato, A.J., 2000. “Violent Crime:A Function of Drug Use or Drug Enforcement?”Applied
Economics, 32(6), 681.
Reston, Maeve, 2013., ”L.A. mayoral candidates sound off on pot dispensaries,” Los Angeles
Times, January 26, 2013.
Rosenthal, Stuart S. and Amanda Ross. 2010. “Violent Crime, entrepreneurship, and
cities” Journal of Urban Economics, 67: 135-149.
Rubin, Joel, and John Hoeffel, 2010, “Clearing Away the Pot Stores, One Raid at a Time,”
Los Angeles Times, July 5, 2010.
Sandler, Danielle, 2012. “Is Public Housing a Public Bad? Externalities of Chicago’s Public
Housing Demolitions.” Unpublished working paper.
Scribner, Richard A., Deborah Cohen, Stephen Kaplan et al. 1999. “Alcohol Availability
and Homicide in New Orleans: Conceptual Considerations for Small Area Analysis of
the Effect of Alcohol Outlet Density.” Journal of Studies on Alcohol, 60: 310-316.
Scribner, Richard A, David P. MacKinnon and James H. Dwyer 1995. “The Risk of As-
saultive Violence and Alcohol Availability in Los Angeles County,” American Journal
of Public Health, 85(3): 335-340.
Sloan, CarlyWill, Steven B. Cuadill and Franklin G. Mixon Jr., 2016. “Entrepreneur-
ship and crime: The case of new restaurant location decisions” Journal of Business
Venturing Insights, 5: 19-26
Swift, Art, 2016. “Support for Legal Marijuana Use Up to 60% in U.S” GALLUP Social
Issues, October 19, 2016.
Teh, Bing-Ru, 2008. “Do Liquor Stores Increase Crime and Urban Decay? Evidence from
Los Angeles,” Unpublished Manuscript.
The Pew Research Center 2010. “Modest Rise in Percentage Favoring General Legaliza-
tion.” Research Report, April 2010.
Walk Score, 2011. Walk Score Methodology. Available at:
www2.walkscore.com/pdf/WalkScoreMethodology.pdf
Wei, Clarissa, and Dennis Romero, 2010, “L.A. Medical Marijuana Outlets Defy City,” LA
Weekly, June 17, 2010.
Winkelmann, Rainer and Klaus F. Zimmermann, 1992. “Robust Poisson Regression,”
Lecture Notes in Statistics Volume 78: 201-206.
Yoshino, Kimi, 2010 “Judge Denies Temporary Restraining Order to Halt Closure of 4
Medical Marijuana Dispensaries,” Los Angeles Times, May 26, 2010.
Zheng, Yuxing, 2014. “Medical Marijuana Dispensaries Ban: Bill Sent to House Commit-
tee, Faces Uncertain Fate,” The Oregonian, March 4, 2014.

35
Figure 1. Dispensary location by closure order status.

36
 Figure 2a. Mean Daily Part I Crime at 1 Mile by Closure Status
4.5
4
3.5
3
2.5 closed
2 open
1.5
1

Figure 2b. Mean Daily Part I Crime at 1/3 Mile by Closure Status
0.7
0.65
0.6
0.55
0.5
0.45
closed
0.4
0.35 open
0.3
0.25
0.2

Figure 2c. Mean Daily Part I Crime at 1/8 Mile by Closure Status
0.15
0.13
0.11
0.09
0.07 closed
0.05
open
0.03
0.01
-0.01

Figure 3. Dispensary Closure Impacts on Part I Crime by Distance

50

40
Percent Change in Crime

30

20

10

0
0 1/2 1 1 1/2 2
-10
Radius in Miles
percent change lower 95 upper 95

Figure 4. Dispensary Closure and TheK From Vehicles by Distance

200

150
Percent Change in Crime

100

50

0
0 1/2 1 1 1/2 2 2 1/2

-50
Radius in Miles
percent change lower 95 upper 95



 Table 1. Dispensary Summary Statistics Pre versus Post June 7, 2010

Dispensaries ordered to Dispensaries allowed to

clos, Pre-June 7 remain open, Pre-June 7 p-value
Daily Crimes
Part I Crimes < 1 Mile 3.12 3.24 0.122
Part I Property Crime <1 Mile 2.47 2.56 0.117
Part I Violent Crime < 1 Mile 0.657 0.679 0.466
Part I Crimes < 1/3 Mile 0.466 0.479 .563
Part I Property Crime <1/3 Mile 0.356 0.362 0.751
Part I Violent Crime < 1/3 Mile 0.104 0.106 0.813
Daily Part I subcategories < 1/3 mile
Aggravated assault 0.049 0.054 0.535
Auto Theft 0.061 0.054 0.276
Burglary 0.052 0.054 0.713
Homicide 0.0014 0.0017 0.835
Rape 0.0046 0.0044 0.957
Robbery 0.048 0.046 0.724
Larceny Theft 0.243 0.254 0.480
Theft 0.144 0.145 0.960
Theft from Vehicles 0.099 0.109 0.254
Zip Code Characteristics
Population 41960 41947 0.994
Households 15414 15669 0.609
Median Household Income 54621 54900 0.867
Median Age 35.5 35.7 0.480
Occupancy Rate 0.930 0.930 0.772
Share Foreign born 0.376 0.381 0.592
Other
Walkscore 74.9 77.2 0.056
Closest Neighbor Allowed Open 0.326 0.354 0.514

The p-value is for a two-sided test of differences in means for dispensaries ordered to close vs. allowed
to remain open. We compare crime counts in radii of 1 and 1/3 mile around dispensaries in the 10 days
prior to June 7, 2010. Zip Code characteristics are from the 2010 Census and the 2011 American
Community Survey. Walkscores are from walkscore.com and are matched to dispensaries by exact
address. Walkscore.com chategorizes its scores as followss: (1) 0-49 = Car Dependent; (2) 50-69=
Somewhat Walkable; (3) 70-89 = Very Walkable and (4) 90-100 = Walker's Paradise.
 Table 2. Effect of Dispensary closures on Total Part 1 crime

Treatment of Defiant Dispensaries

Pre-closure
Radius (miles) Mean Intent to Treat Recoded Dropped
1/8 0.097 0.206* 0.207* 0.213+
(0.1) (0.097) (0.115)

1/4 0.286 0.128+ 0.150* 0.137*

(0.068) (0.063) (0.068)

1/3 0.466 0.112* 0.131* 0.121*

(0.053) (0.057) (0.055)

1/2 0.938 0.0341 0.073 0.047

(0.042) (0.031) (0.043)

1 3.12 -0.004 0.004 -0.001

(0.026) (0.017) (0.026)

2 10.6 -0.011 -0.008 -0.010

(0.014) (0.01) (0.013)

N 4170 11940 11940 11760

Notes: + p<0.10, * p<0.05, ** p<0.01.

Pre-closure means are for crime around dispensaries ordered to close at each distance are
shown in col (1). All other cols show point estimates from Poisson regression models.
Standard errors, shown in parenthesis, allow for twoway clustering by dispensary and by
date. All regressions include date and dispensary fixed effects. We include 10 days of data
pre and post-closure but drop the actual closure date. To handle dispensaries known to be
defiant col (2) does nothing and estimates the intent to treat, col (3) recodes dispensaries
known to be defiant and col (4) drops known defiers.
 Table 3. Placebo Checks of Dispensary closures on Total Part 1 crime

Dispensaries
Period Relative to June 7th Closure Orders
Radius (miles) Actual -1 month -2 months - 3 months
1/8 0.206* -0.086 0.093 0.105
(0.1) (0.142) (0.106) (0.137)

1/4 0.128+ -0.034 0.078 -0.045

(0.068) (0.077) (0.067) (0.076)

1/3 0.112* -0.039 0.008 -0.037

(0.053) (0.043) (0.053) (0.047)

1/2 0.0341 -0.028 0.076+ -0.050+

(0.042) (0.020) (0.044) (0.026)

1 -0.004 0.004 0.014 -0.001

(0.026) (0.015) (0.024) (0.014)

2 -0.011 .0126+ 0.000 0.003

(0.014) (0.007) (0.011) (0.01)

N 11940 11940 11940 11940

Notes: + p<0.10, * p<0.05, ** p<0.01.

The table shows point estimates from Poisson regression models. Standard errors,
shown in parenthesis, allow for twoway clustering by dispensary and by date. All
regressions include date and dispensary fixed effects. Column (1) is from Table 2. It
shows our main results for the June 7th closure orders. Placebo results are in colums 2 -
4. Column 2 assumes a May 10 closure date; column 3 assumes an April 12 closure date
and column 4 a March 15 closure data. In all cases, we drop the data from the closure
day (actual or placebo) and use 10 days of data on either side of thise date.
 Table 4. Effect of Dispensary Closures by Crime Type

Radius Around Dispensaries (miles)

1/8 1/4 1/3 1/2 1 2
Property 0.209+ 0.107 0.112* 0.050 -0.005 -0.018
(0.113) (0.07) (0.05) (0.034) (0.022) (0.014)
Burglary -0.518+ 0.106 0.160 0.006 -0.013 -0.056*
(0.284) (0.199) (0.195) (0.142) (0.053) (0.022)
Auto Theft 0.617 -0.071 0.061 -0.072 -0.036 0.016
(0.509) (0.318) (0.2) (0.136) (0.065) (0.029)
Larceny 0.288+ 0.145 0.126* 0.091* 0.002 -0.017
(0.159) (0.09) (0.063) (0.052) (0.033) (0.023)
Thefts from Vehicle 0.712* 0.241+ 0.224+ 0.106 -0.014 -0.003
(0.318) (0.128) (0.123) (0.101) (0.044) (0.028)
Theft N/A 0.080 0.057 0.079 0.016 -0.030
(0.13) (0.078) (0.067) (0.035) (0.024)

Violent 0.188 0.177 0.099 -0.024 0.002 0.020

(0.274) (0.162) (0.147) (0.106) (0.062) (0.035)
Agg. Assault N/A 0.257 0.200 -0.130 -0.014 -0.000
(0.245) (0.241) (0.166) (0.083) (0.065)
Robbery N/A -0.001 -0.040 0.072 -0.004 0.030
(0.214) (0.194) (0.126) (0.088) (0.042)

N 11940 11940 11940 11940 11940 11940

Notes: + p<0.10, * p<0.05, ** p<0.01.

The table shows point estimates from Poisson regression models as well as standard errors, in
parenthesis, that allow for twoway clustering by dispensary and by date. All regressions include date and
dispensary fixed effects. Standard errors allow for twoway clustering by dispensary and by date.
Regressions are estimated using 10 days pre and post the June 7th closure orders. June 7th is not
included in the sample. Arson is included in total property crime and rape and murder are included in
total violent crime; we do not estimate separate count models for these 3 types of crimes because they
are too rare to allow for convergence. Aggravated assault and robbery do not convege at 1/8 mile.
 Table 5. Spatial Displacement of Crime due to Dispensary Closures

Crime Within Rings of… 1/4-1/3 1/3-1/2 1/2-1 1/2-2 1-2

All Part 1 0.080 -0.036 -0.019 -0.015 -0.014

(0.115) (0.065) (0.029) (0.014) (0.014)
Property 0.120 -0.010 -0.028 -0.025+ -0.023
(0.117) (0.06) (0.025) (0.014) (0.015)
Violent -0.085 -0.149 0.014 0.024 0.028
(0.188) (0.126) (0.085) (0.037) (0.048)

N 11940 11940 11940 11940 11940

Notes: + p<0.10, * p<0.05, ** p<0.01.

The table shows point estimates from Poisson regression models as well as standard errors, in
parenthesis, that allow for twoway clustering by dispensary and by date. All regressions include
date and dispensary fixed effects. Standard errors allow for twoway clustering by dispensary and
by date. Regressions are estimated using 10 days pre and post the June 7th closure orders. June
7th is not included in the sample.
 Table 6. Effect of restaurant closures on Total Part 1 crime

Restaurants
Treatment of Missing Re-Open Date
Pre-closure Use Median Assume Ongoing Drop Those with
Radius (miles) Mean Closure Period Closure Missings

1/4 0.234 0.061 0.013 0.003

(0.063) (0.053) (0.043)

1/3 0.331 0.110* 0.094* 0.094+

(0.045) (0.038) (0.049)

1/2 0.803 0.053* 0.052+ 0.050

(0.027) (0.027) (0.031)

1 2.75 0.000 0.013 -0.003

(0.018) (0.015) (0.02)

2 9.52 0.014 0.002 0.018

(0.01) (0.009) (0.011)

N 8880 17760 17760 14600

Notes: + p<0.10, * p<0.05, ** p<0.01.

Pre-closure means are for crime at each distance are shown in col (1). The table shows
point estimates from Poisson regression models. Standard errors, shown in parenthesis,
allow for twoway clustering by restaurnt and by date. All regressions include date and
business fixed effects. Standard errors allow for twoway clustering by business and by
date. We include 10 days pre and post-closure. To handle missing reopen dates, col (2)
codes them as closed through the rest of the sample period, col (3) uses the median
number of days closed and col (4) drops these cases. Poisson regression for restaurants
for distance of <1/8 mile do not converge.
 Table 7. Effect of Restaurant Closures by Crime Type

Radius Around Restaurants (miles)

1/4 1/3 1/2 1 2
Property 0.082 0.096+ 0.046 0.000 0.018
(0.07) (0.051) (0.028) (0.02) (0.011)
Burglary 0.021 -0.038 0.007 -0.031 0.039
(0.196) (0.139) (0.097) (0.05) (0.024)
Auto Theft -0.008 0.038 0.024 -0.011 0.007
(0.208) (0.156) (0.103) (0.042) (0.025)
Larceny 0.096 0.128* 0.067+ 0.012 0.016
(0.08) (0.064) (0.038) (0.026) (0.015)
Thefts from Vehicle 0.255* 0.227* 0.100 0.007 0.007
(0.124) (0.096) (0.066) (0.04) (0.022)
Theft 0.012 0.054 0.050 0.022 0.024
(0.105) (0.086) (0.055) (0.032) (0.018)

Violent 0.001 0.157 0.078 -0.002 0.004

(0.146) (0.101) (0.067) (0.039) (0.018)
Agg. Assault -0.209 0.181 0.016 -0.007 -0.026
(0.228) (0.166) (0.127) (0.068) (0.03)
Robbery N/A N/A 0.097 -0.005 0.026
(0.084) (0.05) (0.026)

N 17760 17760 17760 17760 17760

Notes: + p<0.10, * p<0.05, ** p<0.01.

The table shows point estimates from Poisson regression models. Standard errors, shown in
parenthesis, allow for twoway clustering by restaurant and by date. All regressions include
date and restaurant fixed effects.. The regressions us e10 days of datat pre and post-closure.
We use the median number of days closed to handle missing reopen dates. Poisson regression
for restaurants for distance of <1/8 mile do not converge. Robbery does not converage at 1/4
or 1/3 of a mile. While arson is included in property crime and rape and murder are included
in violent cime counts, these crimes are too few to separately estimate changes due to
business closures.
 Table 8. Spatial Displacement of Crime due to Restaurant Closures

Panel A: Spatial Displacement 1/4-1/3 1/3-1/2 1/2-1 1/2-2 1-2

All Part 1 0.172* 0.002 -0.021 0.011 0.020+

(0.086) (0.041) (0.023) (0.011) (0.012)
Property 0.109 0.005 -0.018 0.016 0.025+
(0.096) (0.043) (0.027) (0.012) (0.013)
Violent N/A 0.007 -0.038 -0.004 0.005
(0.103) (0.103) (0.021) (0.023)

N 17760 17760 17760 17760 17760

Notes: + p<0.10, * p<0.05, ** p<0.01.

Notes: The table shows point estimates from Poisson regression models. Standard errors, shown in parenthesis, allow for
twoway clustering by business and by date. All regressions include date and restaurant fixed effects. We include 10 days
pre and post-closure. We use the median number of days closed to handle restaurants with no reopen date.

Table 9. Temporal Displacement of Crime at 1/3 Mile around Restaurant Closures

All Property Thefts from Violent

All Larceny Vehicle All
closed 0.111* 0.096+ 0.128* 0.233* 0.161
(0.045) (0.052) (0.065) (0.1) (0.099)
re-open 0.034 -0.006 0.063 -0.005 0.157
(0.061) (0.081) (0.093) (0.176) (0.12)
re-open + 1 -0.021 0.002 -0.058 0.086 -0.113
(0.062) (0.083) (0.097) (0.153) (0.127)

N 17760 17760 17760 17760 17760

Notes: + p<0.10, * p<0.05, ** p<0.01.

Notes: The table shows point estimates from Poisson regression models. Standard errors, shown in parenthesis, allow for
twoway clustering b ybusiness and by date. For both datasets - restaurants and dispensaries - we include 10 days pre and
post-closure. We use the median number of days closed to handle restaurants with no reopen date. For dispensary
regressions, we exclude June 7 from the data. Re-open and re-open + 1 are dummies equal to one on the day a restaurant
is allowed to re-open and the day after a restaurant is allowed to re-open respectively.
 Table 10. Dispensary or Restaurant Closures and Walkscores: Crime at 1/3 of a Mile Around
Establishments
Panel A: Dispensary Closures
Part I Crime Property Violent
Total Theft from Total
Property Larceny Vehicle Violent
Closed (Low Walk Score) 0.194* 0.180+ 0.285* 0.367+ 0.284
(0.083) (0.094) (0.106) (0.203) (0.236)
Closed (High Walk Score) 0.083 0.091 0.085 0.176 0.022
(0.069) (0.058) (0.071) (0.146) (0.157)
Closed * Car-dependent -0.063
(0.197)
Closed * Somewhat Walkable 0.233*
(0.092)
Closed * Very Walkable 0.104
(0.077)
Closed * Walker's Paradise 0.042
(0.11)

N 11940 11940 11940 11940 11940 11940

Panel B: Restaurant Closures
Part I Crime Property Violent
Total Theft from Total
Property Larceny Vehicle Violent
Closed * Low Walk Score 0.187+ 0.180 0.347* 0.513* 0.194
(0.103) (0.114) (0.152) (0.212) (0.207)
Closed * High Walk Score 0.083 0.068 0.071 0.135 0.141
(0.051) (0.056) (0.069) (0.108) (0.155)

Closed * Car-dependent -0.155

(0.241)
Closed * Somewhat Walkable 0.244*
(0.114)
Closed * Very Walkable 0.084
(0.069)
Closed * Walker's Paradise 0.082
(0.073)

N 17760 17760 17760 17760 17760 17760

Notes: + p<0.10, * p<0.05, ** p<0.01.
Notes: The table shows point estimates from Poisson regression models. Standard errors, shown in
parenthesis, allow for twoway clustering by business and by date. For both datasets - restaurants and
dispensaries - we include 10 days pre and post-closure. We use the median number of days closed to handle
restaurants with no reopen date. For dispensary regressions, we exclude June 7 from the data. Low Walk
Scores are those that are defined as either "Car-dependent" or "Somewhat Walkable", while High Walk Scores
are those defined as "Very Walkable" and "Walker's Paradise". Dividing the sample into businesses above and
below the median Walk Score generate qualitatively similar results.
Online Appendix

Appendix Figure 1a. Trends in Part I Crimes

400

350

300

Total Part I Crime
150 200 250

100

50

18may2010 28may2010 07jun2010 17jun2010 27jun2010

Incident Date



Appendix Figure 1b. Trends in Part I Crimes by Region

Central LA Eastside Harbor

0 20 40 60 80

Northeast LA San Fernando Valley South Bay

Total Part I Crime
0 20 40 60 80

South LA Southeast Westside

0 20 40 60 80

30may2010 07jun2010 15jun2010 30may2010 07jun2010 15jun2010 30may2010 07jun2010 15jun2010

Incident Date
Graphs by Region

Notes: Data are from the LA Times Crime L.A. Project. The overall trends are restricted to the regions, as
defined here http://maps.latimes.com/neighborhoods/, listed in Figure 1b. Excluded are crimes committed
in the Angeles Forest, Antelope Valley, Northwest County, Pomona Valley, San Gabriel Valley, Santa
Monica Mountains, Verdugos. We excluded them based on proximity to city dispensaries.
 Appendix Figure 2. Distribution of Walk Scores by Dispensary Closure Orders

Closed

.04
.03
.02
.01
Density
0

Open
.04
.03
.02
.01
0

20 40 60 80 100
Walk Score

Notes: The figure above plots the distribution of Walk Scores by exact dispensary address for those
dispensaries ordered to close (top panel) and those dispensaries allowed to remain open.
 Appendix Table 1. Timeline of Events Impacting Medical Marijuana Dispensaries in Los Angeles
Date Law/Event Key Details
Details the explosion of medical marijuana dispensaries in the City of
Los Angeles, shows statistics to support the view that the dispensaries
14-Dec-06 LAPD fact sheet released
increase crime, and recommends a moratorium on new dispensaries and
regulations for existing dispensaries

Placed a temporary moratorium on the opening of new medical

14-Sep-07 ICO:L.A. Ordinance 179027 marijuana dispensaries in the City of Los Angeles. Allows for a
hardship exemption.

13-Nov-07 ICO registration deadline Deadline for dispensary registration under the ICO
ICO amended via L.A.
24-Jun-09 Eliminates hardship exemption
Ordinance 180749
Caps the number of dispensaries at 70. Allows dispensaries in excess of
70 to remain operational provided that they comply with the ICO and
abide by new requirements. Dispensaries must be geographically
L.A. Ordinance 181069 to
distributed across L.A. community plan areas in proportion to the
26-Jan-10 regulate medical marijuana
population; must be at least 1,000 feet from “sensitive use” buildings,
collectives passes
such as schools and parks; and must not be located on a lot “abutting,
across the street or alley from, or having a common corner with a
residentially zoned area.”

The city shuts down the more than 400 dispensaries that had not
L.A. Ordinance 181069, registered by November 13, 2007. Offenders face civil penalties of
7-Jun-10 Chapter IV, Article 5.1, takes $2,500 per day and may receive up to six months in jail . The remaining
effect dispensaries have 180 days to comply with the new zoning
requirements, which, in many cases, means moving.

City states that 128 of the remaining 169 dispensaries must shut down
because they had changes in management, which were precluded under
25-Aug-10 Villaraigosa memo
the ICO. City allows these dispensaries to remain open until the courts
can rule on the decision’s legality.

City Council adopts amendments that clarify and effectively eliminate

Koretz-Hahn and other
the “same ownership and management” requirements and extend the
24-Nov-10 amendments to L.A. Ordinance
timeline for full compliance for “qualifying” dispensaries. Mayor has
181069
until December 6, 2010, to decide on the amendments.

Los Angeles County Superior Court Judge Anthony J. Mohr grants an

10-Dec-10 Mohr injunction injunction that bars the city from enforcing key aspects of L.A.
Ordinance 181069, including closures based on the moratorium.

Amends L.A. Ordinance 181069 to cap the number of dispensaries at

L.A. Ordinance 181530 takes
25-Jan-11 100 among those continuously operating since September 14, 2007.
effect
Allocates permits by lottery.
SOURCES: Brown (2008), California Senate Bill 420 (2003), Compassionate Use Act of 1996, Council of the City of Los Angeles
(2007), Council of the City of Los Angeles (2009), Council of the City of Los Angeles (2010), Hoeffel (2010a), Hoeffel (2010b),
Hoeffel (2011d), Johnston and Lewis (2009), LACityClerk Connect (undated[b]), Lagmay (2010), and Los Angeles County
Department of Regional Planning (2009), Los Angeles Police Department, Narcotics Division (2006), and United States Department
of Justice (2009).
 Appendix Table 2. Alternative Models of the Effect of Dispensary Closures on Total Part 1 crime
Negative Binomial Zero-inflated Poisson
Treatment of Defiant Dispensaries Treatment of Defiant Dispensaries
Radius (miles) ITT Recoded Dropped Ongoing Recoded Dropped
1/8 0.206+ -- -- 0.205+ 0.217+ 0.216+
(0.109) (0.111) (0.127) (0.119)

1/4 0.127+ 0.150* 0.137* 0.130+ 0.156* 0.141*

(0.068) (0.063) (0.067) (0.07) (0.066) (0.070)

1/3 0.114* 0.134* 0.123* 0.116* 0.135* 0.123*

(0.055) (0.057) (0.056) (0.054) (0.057) (0.056)

1/2 0.036 0.073 0.049 0.038 0.071 0.049

(0.042) (0.046) (0.043) (0.041) (0.044) (0.042)

1 -0.005 0.004 -0.002 -- -- -0.001

(0.027) (0.026) (0.026) (0.026)

2 -0.011 -0.008 -0.010 -0.011 -0.009 -0.010645

(0.014) (0.013) (0.013) (0.013) (0.012) 0.0132143

N 11940 11940 11760 11940 11940 11760

Notes: + p<0.10, * p<0.05, ** p<0.01.

Models in the first three columns are negative binomial and in the last three are zero-inflated poisson
models. All regressions include date and business fixed effects. Standard errors allow for twoway
clustering by business and by date. For both datasets - restaurants and dispensaries - we include 10 days pre
and post-closure. To handle dispensaries known to be defiant cols (1) and (4) do nothing and estimate the
intent to treat, cols (2) and (5) recode dispensaries known to be defiant and cols (3) and (6) drop known
defiers. Negative Binomial Regressions do not converge at 1/8 mile if we recode or drop defiers (col(2) or
col(3)). Zero-inflated Poisson models don't converge at 1 mile in the intent to treat and the recoded
analysises (col(4) and col(5)).
 Appendix Table 3. Effect of lengthening the study window to 60 Days

Total Part I Crimes

Radius (miles) Days 1-10 Days 11-20 Days 21-30
1/8 0.263** 0 .048 0.146
(0.09) (0.114) (0.145)

1/4 0.115+ 0.057 -0.052

(0.059) (0.043) (0.054)

1/3 0.089* 0.054+ -0.005

(0.042) (0.03) (0.045)

1/2 0.024 0.014 0.004

(0.026) (0.028) (0.027)

1 -0.003 -0.012 -0.001

(0.018) (0.021) (0.012)

2 -0.003 -0.015 -0.010

(0.012) (0.014) (0.011)

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and dispensary fixed effects and use 60 days of
data -- 30 days on either side of June 7, 2010. The total number of
observations in each regression is 35820 = 60 days * 597 dispensaries.
Standard errors allow for twoway clustering by dispensary and by date.
Within a crime category, each row represents a separate regression. For each
crime, column (1) provides the coefficient on the first, column (2) the second
and column (3) the third and last 10 days in the full 30 day post closure
period.
 Appendix Table 4. Effect of Dropping June 6-8 in Estimation of Closues on Total Part 1 crime

Dispensaries
Length of Window
Pre-closure
Radius (miles) Mean +/- 9 days +/- 19 days +/- 29 days
1/8 0.097 0.176 0.204+ 0.230*
(0.119) (0.105) (0.1)

1/4 0.286 0.145+ 0.101+ 0.130*

(0.075) (0.06) (0.06)

1/3 0.466 0.144** 0.092** 0.105**

(0.053) (0.038) (0.037)

1/2 0.938 .043156 .030805 .0341048

(0.044) (0.027) (0.025)

1 3.12 -.0003429 .0043993 .0014057

(0.028) (0.021) (0.02)

2 10.6 -.0123588 -.0009354 .0000517

(0.015) (0.012) (0.012)

N 4170 10,746 22,686 34,626

Notes: + p<0.10, * p<0.05, ** p<0.01.

Pre-closure means are for crime around dispensaries ordered to close at each distance are shown in
col (1). All other cols show point estimates from Poisson regression models. Standard errors, shown
in parenthesis, allow for twoway clustering by dispensary and by date. All regressions include date
and dispensary fixed effects. In all regressions, we drop the day of as well as the day before and day
after the closure order took effect, i.e., June 6-Jun 8, 2010. In col (1), we include 9 days, in col (2)
19 and col (3) 29 days on either side of the closure orders. Thus, for the 597 dispensaries, col (1)
captures 18 days (N=18*597=10746), col (2) 38 days (N=22686 = 38* 597) and col (3) 58 days
(N=34626 = 58*597).
Appendix Table 5. Effect of limiting overlap - Total Part 1 crime
Restricting to dispensaries with nearest
neighbor more than…
Radius (miles) 1/3 mile away 1/2 mile away
1/8 0.293 0.412
(0.222) (0.589)

1/4 0.382* 0.660*

(0.178) (0.319)

1/3 0.325** 0.491+

(0.131) (0.26)

1/2 0.196* 0.342

(0.098) (0.227)

1 0.035 0.045
(0.049) (0.094)

2 -0.006 0.023
(0.024) (0.033)

N 3160 1580

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and dispensary fixed effects. Standard errors
allow for twoway clustering by business and by date. For both datasets -
restaurants and dispensaries - we include 10 days pre and post-closure.
Column (1) restricts to dispensaries with a nearest neighbor more than 1/3
of a mile away while column (2) restricts to dispensaires with a nearest
neighbor more than 1/2 mile away.
 Appendix Table 6. Pre-closure Summary Statistics for Dispensaries Ordered to Close

Radius Around Dispensaries (miles)

1/8 1/4 1/3 1/2 1 2
Property 0.072 0.219 0.357 0.734 2.47 8.35

Burglary 0.011 0.032 0.052 0.111 0.418 1.47

Auto Theft 0.011 0.039 0.061 0.120 0.396 1.30

Larceny 0.050 0.148 0.243 0.502 1.65 5.58

Thefts from Vehicle 0.016 0.059 0.099 0.219 0.762 2.63

Theft 0.034 0.088 0.144 0.284 0.891 2.95

Violent 0.025 0.066 0.104 0.204 0.657 2.23

Agg. Assault 0.011 0.032 0.049 0.095 0.314 1.01

Robbery 0.013 0.031 0.048 0.096 0.306 1.08

Rape 0.0014 0.0026 0.0046 0.0094 0.0283 0.1053

Homicide 0.0002 0.0007 0.0014 0.0034 0.0084 0.0367

N 4170 4170 4170 4170 4170 4170

This table shows mean crime counts by type in the 10 days prior to June 7, 2010 for dispensaries ordered
to close. Means are generally indistinguishable for those allowed to remain open.
 Appendix Table 7. Effect of Dispensary Closures by Crime Type - Dropping Defiers

Radius Around Dispensaries (miles)

1/8 1/4 1/3 1/2 1 2
All Part 1 0.213+ 0.137* 0.121* 0.047 -0.001 -0.010
(0.115) (0.068) (0.055) (0.043) (0.026) (0.013)

Property 0.206+ 0.110 0.116* 0.059+ -0.004 -0.017

(0.114) (0.069) (0.051) (0.034) (0.022) (0.014)
Burglary -0.518+ 0.102 0.117 0.013 -0.016 -0.057**
(0.284) (0.193) (0.195) (0.141) (0.054) (0.021)
Auto Theft 0.617 -0.072 0.069 -0.057 -0.031 0.022
(0.509) (0.324) (0.204) (0.139) (0.067) (0.031)
Larceny 0.281+ 0.152+ 0.126* 0.098+ 0.003 -0.018
(0.165) (0.087) (0.062) (0.051) (0.032) (0.022)
Thefts from Vehicle 0.714* 0.241+ 0.218+ 0.111 -0.014 -0.004
(0.323) (0.128) (0.123) (0.101) (0.044) (0.027)
Theft N/A 0.092 0.062 0.087 0.019 -0.030
(0.132) (0.076) (0.064) (0.034) (0.024)

Violent 0.229 0.212 0.130 0.007 0.002 0.022

(0.29) (0.17) (0.15) (0.108) (0.062) (0.035)
Agg. Assault N/A 0.279 0.235 -0.099 -0.012 -0.002
(0.351) (0.242) (0.167) (0.082) (0.064)
Robbery N/A 0.053 -0.007 0.105 0.014 0.034
(0.215) (0.197) (0.131) (0.088) (0.042)

N 11760 11760 11760 11760 11760 11760

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and dispensary fixed effects. Standard errors allow for twoway clustering by
dispensary and by date. Regressions are estimated using 10 days pre and post the June 7th closure orders.
June 7th is not included in the sample and dispensaries known to have defied closure orders are dropped.
Arson is included in total property crime and rape and murder are included in total violent crime; we do not
estimate separate count models for these 3 types of crimes because they are too rate to allow for convergence.
Theft, aggravated assault and robbery do not convege at 1/8 mile.
 Appendix Table 8. Effect of Dispensary Closures by Crime Type with Defiers Recoded

Radius Around Dispensaries (miles)

1/8 1/4 1/3 1/2 1 2
All Part 1 0.207+ 0.150* 0.131* 0.073 0.004 -0.008
(0.118) (0.064) (0.057) (0.046) (0.025) (0.012)

Property 0.180 0.112+ 0.116* 0.073* -0.002 -0.016

(0.111) (0.065) (0.051) (0.035) (0.022) (0.014)
Burglary -0.357 0.088 0.130 0.027 -0.021 -0.058**
(0.301) (0.173) (0.189) (0.134) (0.054) (0.019)
Auto Theft 0.479 -0.073 0.086 -0.019 -0.020 0.036
(0.474) (0.326) (0.209) (0.139) (0.07) (0.033)
Larceny 0.235 0.156* 0.117* 0.105* 0.004 -0.018
(0.165) (0.078) (0.056) (0.048) (0.031) (0.021)
Thefts from Vehicle 0.696** 0.235+ 0.192 0.116 -0.017 -0.006
(0.318) (0.124) (0.124) (0.098) (0.043) (0.025)
Theft -0.027 0.107 0.067 0.097* 0.023 -0.028
(0.207) (0.112) (0.067) (0.058) (0.032) (0.024)

Violent 0.308 0.277 0.196 0.079 0.034 0.025

(0.311) (0.179) (0.153) (0.115) (0.064) (0.033)
Agg. Assault N/A 0.257 0.285 -0.005 0.026 0.020
(0.329) (0.231) (0.163) (0.082) (0.057)
Robbery 0.043 0.164 0.068 0.176 0.052 0.043
(0.276) (0.219) (0.197) (0.145) (0.088) (0.041)

N 11940 11940 11940 11940 11940 11940

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and dispensary fixed effects. Standard errors allow for twoway clustering by
dispensary and by date. Regressions are estimated using 10 days pre and post the June 7th closure orders.
June 7th is not included in the sample. Dispensaries known to have defied closure orders are recoded as
open. Arson is included in total property crime and rape and murder are included in total violent crime; we
do not estimate separate count models for these 3 types of crimes because they are too rate to allow for
convergence. Aggravated assault do not convege at 1/8 mile.
 Appendix Table 9. Effect of Dispensary Closures by Crime Type including Closure Date

Radius Around Dispensaries (miles)

1/8 1/4 1/3 1/2 1 2
All Part 1 0.163 0.101 0.094+ 0.021 -0.007 -0.009
(0.109) (0.07) (0.054) (0.042) (0.025) (0.013)

Property 0.193+ 0.081 0.099+ 0.033 -0.012 -0.017

(0.109) (0.071) (0.052) (0.037) (0.023) (0.014)
Burglary -0.484+ 0.066 0.072 -0.008 -0.005 -0.043+
(0.285) (0.199) (0.194) (0.137) (0.052) (0.024)
Auto Theft 0.484 -0.118 0.019 -0.084 -0.037 0.017
(0.459) (0.302) (0.193) (0.132) (0.063) (0.029)
Larceny 0.273+ 0.128 0.125* 0.070 -0.009 -0.021
(0.156) (0.086) (0.063) (0.051) (0.033) (0.022)
Thefts from Vehicle 0.657* 0.217+ 0.240* 0.099 -0.013 -0.032
(0.303) (0.123) (0.116) (0.091) (0.041) (0.026)
Theft 0.027 0.066 0.045 0.048 -0.005 -0.028
(0.227) (0.125) (0.074) (0.068) (0.041) (0.024)

Violent 0.015 0.139 0.081 0.021 0.013 0.026

(0.294) (0.159) (0.135) (0.099) (0.059) (0.063)
Agg. Assault N/A 0.193 0.181 -0.106 0.018 0.020
(0.31) (0.221) (0.158) (0.082) (0.057)
Robbery N/A -0.018 0.054 0.059 -0.011 0.026
(0.2) (0.172) (0.114) (0.08) (0.04)

N 12537 12537 12537 12537 12537 12537

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and dispensary fixed effects. Standard errors allow for twoway clustering by
dispensary and by date. Regressions are estimated using 10 days pre and post the June 7th closure orders.
June 7th is not included in the sample. Dispensaries known to have defied closure orders are recoded as
open. Arson is included in total property crime and rape and murder are included in total violent crime; we
do not estimate separate count models for these 3 types of crimes because they are too rare to allow for
convergence. Aggravated assault and robbery do not convege at 1/8 mile.
 Appendix Table 10. Restaurant Summary Statistics

Restaurants, 10-days pre-closure

Daily Crimes
Part I Crimes < 1 Mile 2.75
Part I Property Crime <1 Mile 2.08
Part I Violent Crime < 1 Mile 0.674
Part I Crimes < 1/3 Mile 0.383
Part I Property Crime <1/3 Mile 0.285
Part I Violent Crime < 1/3 Mile 0.098
Daily Part I subcategories < 1/3 mile
Aggravated assault 0.042
Auto Theft 0.048
Burglary 0.043
Homicide 0.0015
Rape 0.0030
Robbery 0.051
Larceny Theft 0.194
Theft 0.119
Theft from Vehicles 0.074
Zip Code Characteristics
Population 44040
Households 13579
Median Household Income 50472
Median Age 34.6
Occupancy Rate 0.939
Share Foreign born 0.403
Other
Walkscore 71.1
Closest Neighbor Allowed Open N/A
 Appendix Table 11. Restaurant Closure Placebo Checks

Placebo based on Actual Closure Length Placebo Based on Fixed Periods

Radius Assume Use Median Drop Those with
(miles) Ongoing Closure Closure Period Missings 1 day prior 2 days prior
1/4 0.020 -0.003 0.013 -0.090 -0.006
(0.054) (0.052) (0.061) (0.09) (0.065)

1/3 -0.007 -0.043 -0.012 -0.076 -0.009

(0.043) (0.042) (0.05) (0.074) (0.05)

1/2 0.021 0.003 0.027 -0.007 0.020

(0.032) (0.031) (0.035) (0.051) (0.035)

1 -0.004 -0.013 -0.003 -0.030 0.006

(0.017) (0.015) (0.018) (0.023) (0.019)

2 -0.001 0.004 -0.006 0.001 -0.001

(0.009) (0.008) (0.009) (0.012) (0.01)

N 17760 17760 14600 17760 17760

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and business fixed effects. Standard errors allow for twoway clustering
by business and by date. Placebo closures indicators reverse the date of closures around the closure
date. That is if a restaurant was closed for 2 days (the median closure length), then the placebo
closure would be a dummy equal to one for the to days prior to the restaurant closure. Cols (2)-(4)
use actual closure periods but vary in how they treat missing reopen dates. Cols (5) and (6) use
fixed pre-closure periods.
 Appendix Table 12. Effect of lengthening the study window - Total Part 1 crime

Restaurant Closures
Radius (miles) +/-15 days +/-20 days +/-30 days
1/4 0.045 0.018 0.016
(0.057) (0.056) (0.054)

1/3 0.094* 0.043 0.073+

(0.042) (0.047) (0.041)

1/2 0.051+ 0.032 0.044+

(0.026) (0.025) (0.026)

1 0.001 -0.004 -0.005

(0.016) (0.015) (0.016)

2 0.012 0.007 0.006

(0.009) (0.008) (0.008)

N 26573 36696 52031

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and restaurant fixed effects. Standard errors allow for
twoway clustering by restaurant and by date. We include either 15, 20 or 30 days on
either side of the closure period (in contrast to the 10 on either side in the main tables).
To handle missing reopen dates, we use the median number of days closed. Poisson
regression for restaurants for distance of <1/8 mile do not converge.
Appendix Table 13. Effect of Restaurant Closures by Crime Type including Closure Date

Radius Around Restaurants

1/4 1/3 1/2 1 2
All 0.040 0.082* 0.037 -0.004 0.008
(0.051) (0.037) (0.024) (0.015) (0.008)

Property 0.066 0.098* 0.029 -0.000 0.013

(0.059) (0.043) (0.026) (0.018) (0.009)
Burglary 0.114 0.031 -0.016 -0.020 0.026
(0.162) (0.11) (0.077) (0.041) (0.019)
Auto Theft 0.109 0.149 0.040 -0.006 0.017
(0.165) (0.119) (0.076) (0.035) (0.02)
Larceny 0.046 0.093+ 0.044 0.009 0.009
(0.068) (0.052) (0.035) (0.022) (0.012)
Thefts from Vehicle 0.141 0.169* 0.039 0.015 0.013
(0.106) (0.078) (0.054) (0.033) (0.019)
Theft -0.005 0.047 0.055 0.008 0.005
(0.09) (0.073) (0.048) (0.027) (0.014)

Violent -0.047 0.043 0.075 -0.015 -0.007

(0.122) (0.085) (0.054) (0.03) (0.014)
Agg. Assault N/A 0.052 0.064 -0.017 -0.045*
(0.128) (0.101) (0.052) (0.024)
Robbery N/A 0.029 0.079 -0.016 0.029
(0.123) (0.073) (0.038) (0.022)

N 18648 18648 18648 18648 18648

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and restaurant fixed effects. Standard errors allow for
twoway clustering by restaurant and date. The regressions use 10 days of data pre and
post-closure. We also include the closure date even though the restaurant may have
closed for only part of the day. We use the median number of days closed to handle
missing reopen dates. Poisson regression for restaurants for distance of <1/8 mile do
not converge. Robbery does not converage at 1/4 or 1/3 of a mile. While arson is
included in property crime and rape and murder are included in violent cime counts,
these crimes are too few to separately estimate changes due to business closures.
 Appendix Table 14. Effect of Restaurant Closures by Crime Type with Missing
Reopen Dates Coded as Continued Closure

Radius Around Restaurants

1/4 1/3 1/2 1 2
All Part 1 0.061 0.110** 0.053** 0.000 0.014
(0.063) (0.045) (0.027) (0.018) 0.01

Property -0.003 0.063 0.036 0.012 0.007

(0.058) (0.042) (0.031) (0.017) (0.009)
Burglary -0.185 -0.089 -0.022 -0.029 0.002
(0.185) (0.132) (0.088) (0.043) (0.024)
Auto Theft -0.007 0.110 0.056 0.004 -0.007
(0.16) (0.114) (0.083) (0.04) (0.02)
Larceny 0.028 0.086 0.051 0.022 0.013
(0.068) (0.054) (0.037) (0.022) (0.013)
Thefts from Vehicle 0.092 0.126 0.054 0.025 0.009
(0.118) (0.091) (0.064) (0.035) (0.019)
Theft -0.026 0.036 0.044 0.021 0.016
(0.089) (0.072) (0.05) (0.028) (0.016)

Violent 0.100 0.201** 0.100* 0.014 -0.015

(0.115) (0.086) (0.06) (0.034) (0.014)
Agg. Assault 0.001 0.238* 0.146 0.026 -0.025
(0.178) (0.14) (0.107) (0.059) (0.025)
Robbery N/A N/A 0.047 -0.017 -0.015
0.072 0.042 0.02

N 17760 17760 17760 17760 17760

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and restaurant fixed effects. Standard errors allow for
twoway clustering by restaurant and by date. The regressions us 10 days of data pre
and post-closure. We treat restaurants with missing re-open dates as permanently
closed. Poisson regression for restaurants for distance of <1/8 mile do not converge.
Robbery does not converage at 1/4 or 1/3 of a mile. While arson is included in
property crime and rape and murder are included in violent cime counts, these crimes
are too few to separately estimate changes due to business closures. Restaurants with
missing reopen dates are coded as closed for the full post-closure observation period.
 Appendix Table 15. Effect of Restaurant Closures by Crime Type, Dropping
Restaurants with Missing Reopen Dates

Radius Around Restaurants

1/4 1/3 1/2 1 2
All Part 1 0.000 0.084** 0.047* -0.007 0.010
(0.054) (0.041) (0.027) (0.016) (0.008)

Property 0.008 0.081* 0.032 -0.002 0.015

(0.06) (0.047) (0.028) (0.018) (0.011)
Burglary -0.101 -0.055 -0.043 -0.044 0.017
(0.174) (0.128) (0.086) (0.046) (0.021)
Auto Theft 0.078 0.021* 0.046 -0.001 0.018
(0.168) (0.12) (0.081) (0.038) (0.023)
Larceny 0.014 0.076 0.056 0.014 0.014
(0.073) (0.058) (0.037) (0.023) (0.014)
Thefts from Vehicle 0.072 0.124 0.008 0.019 0.023
(0.114) (0.088) (0.06) (0.036) (0.022)
Theft -0.002 0.057 0.100** 0.015 0.006
(0.1) (0.081) (0.05) (0.03) (0.016)

Violent -0.027 0.096 0.106* -0.024 -0.000

(0.131) (0.097) (0.061) (0.036) (0.016)
Agg. Assault N/A N/A 0.089 -0.033 -0.028
(0.113) (0.058) (0.028)
Robbery N/A 0.101 0.121 -0.017 0.027
(0.132) (0.077) (0.043) (0.023)

N 15330 15330 15330 15330 15330

Notes: + p<0.10, * p<0.05, ** p<0.01.

All regressions include date and restaurant fixed effects. Standard errors allow for
twoway clustering by restaurant and by date. The regressions us 10 days of data pre
and post-closure. We drop restaurants that are missing reopen dates. Poisson regression
for restaurants for distance of <1/8 mile do not converge. Robbery does not converage
at 1/4 or 1/3 of a mile. While arson is included in property crime and rape and murder
are included in violent cime counts, these crimes are too few to separately estimate
changes due to business closures. Restaurants with missing reopen dates are excluded
from the analysis.
 Appendix Table 16. Dispensary or Restaurant Closures and Employee Density: Crime at 1/3 of a Mile Around Establishments

Panel A: Dispensary Closures

Part I Crime Total Property Larceny Theft from Vehicle Total Violent
Closed below median density) 0.128 0.21** 0.229 0.265** -.164
(0.078) (0.087) (0.125) (0.107) (0.275)

Closed (above median density) 0.094 0.049 0.064 0.183** .289

(0.08) (0.06) (0.079) (0.065) (0.202)

N 11900 11900 11900 11900 11900

Panel B: Restaurant Closures
Part I Crime Total Property Larceny Theft from Vehicle Total Violent
Closed below median density) 0.198* 0.195* 0.239+ 0.17 0.109
(0.087) (0.087) (0.137) (0.144) (0.181)

Closed (above median density) 0.048 0.050 0.088 0.261* 0.184

(0.05) (0.069) (0.077) (0.122) (0.123)

N 15356 15356 15356 15356 15356

Notes: + p<0.10, * p<0.05, ** p<0.01.

Notes: The table shows point estimates from Poisson regression models. Standard errors, shown in parenthesis, allow for twoway
clustering by business and by date. For both datasets - restaurants and dispensaries - we include 10 days pre and post-closure. We use
the median number of days closed to handle restaurants with no reopen date. For dispensary regressions, we exclude June 7 from the
data. To measure density, we use the number of employees per square mile in a dispensary or restaurant ZIP code.