Content uploaded by Vincenzo Bove
Author content
All content in this area was uploaded by Vincenzo Bove on Oct 11, 2017
Content may be subject to copyright.
Electronic copy available at: http://ssrn.com/abstract=2678967
Policeman on the Frontline or a
Soldier? The Effect of Police
Militarization on Crime
Vincenzo Bove∗
University of Warwick
Evelina Gavrilova†
Norwegian School of Economics
October 23, 2015
Abstract
Recently a considerable popular debate has been dedicated to the
issue of “police militarization”. We investigate whether the “1033
Program”, which allows local law enforcement agencies to acquire ex-
cess property of the US Department of Defense, affects crime rates.
To identify the causal effect of militarized policing on crime, we use
temporal variations in US military expenditure and between-counties
variations in the odds to receive a positive amount of military aid. We
find that (i) military aid reduces street-level crimes; (ii) the program
is cost-effective; and (iii) there is evidence in favor of a deterrence
mechanism.
JEL classification: K42, H49, H76
Keywords: police, crime, militarization, 1033 Program
∗v.bove@warwick.ac.uk
†evelina.gavrilova@nhh.no
1
Electronic copy available at: http://ssrn.com/abstract=2678967
The Effect of Police Militarization on Crime 2
1 Introduction
In August 2014 a series of unrests erupted in the city of Ferguson, Missouri,
following the fatal shooting of an unarmed 18-year-old African-American
male by a police officer. Two days after the shooting, Ferguson’s police
deployed military-grade weapons and armored tactical vehicles to quell the
riots and clear protesters in front of the police department. The reaction of
the police sparked a still combative debate about the use of force doctrine
in the US and the militarization of its police. In particular the existence of
the “1033 Program”, a federal initiative that since 1997 has transferred more
than $4.3 billion in surplus military equipment from the Defense Department
to domestic police agencies across the country, entered the national spotlight.
This article investigates the causal effect of an increase in the militarization of
US local police forces on their effectiveness in preventing and clearing crime.
Since Ferguson’s events, a wave of reports and images circulated around
the social media of police in small municipalities using high-level military
equipment, such as mine-resistant vehicles, grenade lunchers and assault ri-
fles, while news of police acting disproportionally made the headlines.1. Vir-
tually all the debate in the media revolved around the question of whether the
use of military tools escalates the risk of violence, undermines individual lib-
erties and causes mistrust between local police forces and their own citizens.
In particular, a recent study by American Civil Liberties Union (2014, p.3)
explores how the presence of military weapons and tactics has impacted polic-
ing culture, finding that it “encourages them to adopt a warrior mentality
and think of the people they are supposed to serve as enemies”.2Following
public pressure, in January 2015 Barack Obama announced, by executive
order (White House, 2015), a reduction in the scope of the program.
Somewhat surprisingly, however, the existing debate has focused on the
misuse of high-tech weaponries by paramilitary squads, often based on lim-
1See e.g., “Don’t shoot”, The Economist, 13/12/14.
2See also, among others, Hall & Coyne (2012), who explore historical attempts to
establish constraints to separate the military functions and policing functions in the US;
Hill & Beger (2009), who discuss how militarization affects democratic policing; and Balko
(2013) on the rise of paramilitary units in the US and their battlefield mentality.
The Effect of Police Militarization on Crime 3
ited factual evidence, without a discussion of the relative merits and draw-
backs of the 1033 Program, in particular with respect to crime prevention. To
what extent did the proliferation of military weapons within US local police
forces affect their effectiveness in countering crime? Is it really the case that
the acquisition of military-style equipment have “contributed to the protec-
tion of the public” and can still provide “effective and efficient contributions
to public safety” (White House, 2014, p.6)? The above questions have crucial
policy implications for security policies worldwide but have so far remained
not addressed.
As of yet, and despite its size and scope, there is surprisingly no research
on whether Program 1033 actually improved the performances of local police
departments in the US. We use newly released data by the US Department
of Defence on more than 176,000 transfers of equipment currently held by
8,000 local police agencies over the period 2006-2012. We explore whether
more advanced equipment have a tangible effect on the production function
for law enforcement, measured by crime and clearance rates. These two
variables allow us to disentangle deterrence effects produced by the display
of military equipment from an efficiency effect when the police forces know
how to use military equipment in order to resolve more crimes. We exploit
the level of disaggregation of military aid to review and assign each item to a
category, i.e., weapons, vehicles, gears and a residual non-military aid group.
This allows us to further unlock the black box of policing and explore whether
e.g., positive shocks in the amount of lethal vs. non-lethal resources allocated
to the police have discernible effects on crime and arrest rates. Finally, to
tease out competing mechanisms linking aid to crime, we look at whether aid
affects police labour inputs such as number of officers and employees, and
whether it affects police activities, such as the number of calls, the number
of offenders killed as well as the number of assaults and injuries on police
officers.
The main hurdle to identifying the causal effect of militarized policing on
crime stems from reverse causality and spurious correlation, which bias OLS
estimates: regions with higher crime rates can respond by acquiring more re-
sources, whereas regions with lower crime rates need to use fewer resources.
The Effect of Police Militarization on Crime 4
We exploit a strategy similar to the one used by Nunn & Qian (2014), where
we interact exogenous time variation induced by military spending and local
variation in a county’s likelihood of being an aid recipient. High military
spending, driven by international factors such as the war in Afghanistan,
causes the Department of Defence to accumulate excess reserves in high
spending years. The “1033 Program” allows the delivery of this excess prop-
erty to law enforcement agencies across the country. We interact this variable
with a county’s propensity to acquire military aid, measured as the fraction
of years that a county receives a positive quantity of equipment. Our identi-
fication rests on filtering out the excess component in the quantity received,
conditional on receiving aid, and recalls a difference-in-difference strategy,
where however the treatment variable is continuous.
We find that military aid reduces the crime rates but do not have any
effect on the arrest rates. In particular, more military aid leads to a decline
in robberies, assaults, larcenies and motor vehicles thefts, which are part of
the so-called “susceptible crimes” (a la Draca et al. , 2011), i.e., that are
more likely to be prevented by police visibility. By the most conservative es-
timate, a 10% increase in aid reduces total crimes by 4.9 units. Even though
the implied substantive effect is relatively small, the annual average value
of aid acquired by a county is around $51,000, indicating that this is a very
inexpensive crime-reducing tool. Our main results survive a variety of ro-
bustness checks among which population weighting, differential county crime
trends, alternative instruments such as US military fatalities and accounting
for county crime reporting rates.
Additionally, the second stage of our IV analysis resembles the reduced
form equation for evaluating the elasticity of crime with respect to police
grants such as the two programs analyzed in Evans & Owens (2007). We
find that our estimates are higher than the reduced form estimates for both
the programs in Evans & Owens (2007). Even more, when we evaluate the ef-
fect of aid on police activities we find an insignificant effect with a sufficient
F-statistic, implying that the effects we observe on crime are not passing
through an observable effect on police manpower. Additionally, we do not
find evidence of an effect of military aid on injuries and assaults on police
The Effect of Police Militarization on Crime 5
officers, as well as on the number of offenders killed. Taken together, our
results suggest that employing military equipment improves the capabilities
of law-enforcement to deter crime, perhaps through an unobservable police
effort channel. Our cost-benefit analysis shows that for a 10 percent of spend-
ing increase, around $5,100 per county per year, the crimes deterred amount
to a social benefit of roughly $105,000. From the three differentiated cate-
gories: weapons, gear and vehicles, it seems that vehicles lead to the highest
net benefit of roughly $78,000.
Whereas qualitative studies on the effects of the militarization of the po-
lice are at best sporadic, systematic quantitative analyses are nonexistent.
The empirical literature most closely related to our study is that on the
causal effect of an increase in the funds provided to the police force on crime.
Machin & Marie (2011) find a decrease in robbery rates following the in-
crease in targeted funds to police forces in England and Wales. Evans &
Owens (2007) study the effect of a program that provided grants to states
and localities to hire police officers, and find a decline in auto theft, bur-
glary, robbery, and aggravated assault following receipt of the grants. Owens
(2013) notes that the observed decrease in crime may be due to deterrence,
as arrest rates do not seem to be affected by the additional grants. Secon-
darily, the empirical literature on the increase in the size of the police force
hinges on the indirect availability of more resources. This literature has been
recently reviewed by Chalfin & McCrary (2014). To identify an effect several
studies have exploited temporary redeployment policies due to terror-related
events (Di Tella & Schargrodsky, 2004; Klick & Tabarrok, 2005; Draca et al.
, 2011). The former two find a decrease in auto thefts, whereas the latter
find evidence that more police reduce “susceptible crimes” such as robberies
and thefts. Yet, none of the above studies tackles the effect of an increase in
police equipment, in particular weapons, on crime, as opposed to an increase
in funding or manpower. The only exception is Mastrobuoni (2014), who
investigates whether differences in clearance rates across two police forces in
Milan can be attributed to the availability of advanced Information Tech-
nology strategies. He finds that this is indeed the case, and adopting IT
innovation doubles the productivity of policing.
The Effect of Police Militarization on Crime 6
We proceed as follows: Section 2 gives some insights into the 1033 Pro-
gram; Section 3 presents our data and Section 4 explains our identification
strategy. Section 5 describes our results and Section 6 offers concluding re-
marks.
2 The 1033 Program
In 1990, after several years of increasing crime levels, often associated with
drugs, the US Congress, through the National Defense Authorization Act,
authorized the transfer of excess property from the Department of Defense
to federal and state agencies, mainly for counter-drug activities. A few years
later, in 1997, the Congress made the program permanent and expanded its
scope by allowing law enforcement agencies to acquire military property to
generally assist in their arrest and apprehension tasks, although preferences
were still given to counter-drug and counter-terrorism requests. The program
was initially called the “1208 program” and from 1996, the “1033 Program”,
after Congress replaced Section 1208 with Section 1033. The program is
under the jurisdiction of the Defense Logistics Agency (DLA) and is overseen
by the Law Enforcement Support Office (LESO), located at DLA Disposition
Services Headquarter.
Law enforcement agencies follow a three-step procedure to acquire mili-
tary hardwares: they first need to get the approval by the State Coordinator
and LESO to participate in the program; then they place requests for specific
items and their justification, which is screened through the State Coordinator
and the LESO Staff; finally, law enforcement agencies that receive approvals
must cover all transportation and/or shipping costs.3Since the inception
of the 1033 Program, over 8,000 federal and state law enforcement agencies
have requested a variety of equipment, from assault rifles and grenades to
Mine Resistant Ambush Protected (MRAP) vehicles, helicopters and drones,
to non-lethal equipment, such as high-tech cameras, camouflage/deception
equipment or office supplies.
3For more information: http://www.dispositionservices.dla.mil/leso/pages/
1033programfaqs.aspx
The Effect of Police Militarization on Crime 7
It was not until media coverage of the Ferguson unrests however - when
an unarmed 18-year-old was killed by a police officer, and street protests were
confronted by heavily-armed police forces, nearly indistinguishable from sol-
diers - that the program drew media and public attention. In fact, since
Ferguson’s episode, there has been much debate about whether local author-
ities’ response is often disproportionate and why e.g., a police department
in a city of 20,000 residents looked like an invading army engaged in urban
warfare against street protesters.4As surplus military equipment is provided
free of charge to police departments with little accompanying training on the
safe and proper use or supervision, it can encourage the police to employ
disproportionally and improperly military weapons and tactics.
In August 2014 US President Barack Obama ordered a review of the
distribution of military hardware to state and local police agencies (Reuters,
23/08/14). Following this, the White House released a report on the use of
military equipment provided through federal programs, which stops short of
suggesting the elimination of the program, but rather aims at finding a middle
ground. Although, on one hand, military equipment has “contributed to the
protection of the public and to reduced operational risk to peace officers
[..] when police lack adequate training, make poor operational choices, or
improperly use equipment, these programs can facilitate excessive uses of
force and serve as a highly visible barrier between police and the communities
they secure. When officers misuse equipment, the partnership, problem-
solving and crime prevention collaboration with citizens that is at the heart
of effective policing can be eroded.” (White House, 2014, p.6). In January
2015 president Obama announced, by executive order (White House, 2015),
major policy changes in the 1033 Program and in May 2015 he banned federal
transfers of certain types of military-style gear to local police departments.5
Effectively, the 1033 Program was scaled down and in the following sections
4See Amanda Taub, Why America’s police forces look like invading
armies, Vox, Aug. 19, 2014, http://www.vox.com/2014/8/14/6003239/
police-militarization-in-ferguson
5Obama ordered in particular a ban on grenade launchers, tracked armored vehicles,
armed aircraft, guns and ammunition of .50 caliber or higher, and restrictions to the
transfer of wheeled armored vehicles, drones, helicopters, firearms and riot gear, to ensure
officers are trained in their use (Washington Post, 18/05/2015).
The Effect of Police Militarization on Crime 8
we look at how effective it was in helping law-enforcement combat crime
during the years in which the program was growing.
3 Data
To address the question of the effect of military equipment on the activity
of the local police force we construct several outcome variables. Our main
data sources are the Uniform Crime Reports (UCR) data at the county level
and the Law Enforcement Officers Killed and Assaulted (LEOKA) data at
the agency level. Crime in the United States is reported by law-enforcement
agencies to the FBI, where summaries of these reports are created and pub-
lished as yearly statistics. The Interuniversity Consortium for Political and
Social Research (ICPRS) aggregates the separate agencies into counties tak-
ing into account issues such as agencies spanning several counties, or agencies
not reporting for a certain period of time, or agencies closures and openings.
The UCR data allows to distinguish between several major crime cat-
egories such as homicide, assault, robberies, burglaries, larceny and motor
vehicle thefts. The LEOKA data allows to look for an impact of military
aid on several law-enforcement characteristics such as the numbers of officers
and employees at the agency, the ratio between them, the amount of calls,
injuries, assaults received by the police and the number of offenders killed.
Some of the last variables have been widely implicated in the debate on po-
lice militarization. Recently, Chalfin & McCrary (2013) raise concerns about
the measurement error in UCR police records. We have no reason to assume
that this measurement error is associated with the amount of military aid
received. Finally, we drop 8 percent of the crime data, as control variables
were not provided for some counties.
Data for the amount of military aid awarded to each county have been
only recently released by the US Department of Defense, Defense Logistics
Agency, Disposition Services and are now available in the public domain.
In Figure 1 we show the monetary size of the given aid for the years in
the sample period. As we can see, in the period between 2006 and 2012
alone, the value of military donations to law enforcement agencies went from
The Effect of Police Militarization on Crime 9
slightly less than $30 million to almost $400 million per year, thus expanding
significantly. It then reached $750 million in 2014.
—————— [Figure 1 in here] ——————
We use the Defense Logistics Agency’s federal supply category and class
name to identify the type of equipment. We then aggregate several categories
into four groups: weapons (e.g., explosives, guided missiles, guns), vehicles
(e.g., aircrafts, combat, assault, and tactical vehicles, including their com-
ponents), gears (e.g., communication devices, special clothing, night vision
equipment) and a residual category (e.g., office supplies, maps, furniture,
plumbing items). Table A.1 shows our classification in more details, with
the relative frequency of each category. We mainly use information on the
original acquisition value of items, which refers to the (original) amount the
military services paid for the property.6
We also put together information on the poverty rate, median income, un-
employment rate, the size of the population, and the shares of males, blacks,
and people between 15-19, 20-24, 25-29 and 30-34 years old. These covariates
capture both individual criminal decision-making analysis and heterogeneous
trends across counties. The data are taken from the US Census Bureau and
the US Department of Labor. Table 1 offers summary statistics for the main
variables of interest as well as for the control variables. The table shows
that although each county in our sample experiences approximately 2300
crimes per 100,000 population every year, most of them are burglaries and
larcenies, whereas homicides are a much rarer event. All military aid are
recorded as quantities and acquisition value per unit, and according to our
dataset, a county receives on average material worth $51000 per year. As one
would expect, the most expensive items obtained through the 1033 Program
are vehicles, such as aircraft, watercraft and armored vehicle, and the most
commonly requested items are gears, in particular clothing.
6We prefer not to use the quantities (e.g., 10 guns over 30 mm up to 75 mm), as in each
category, different classes of aid have very different values (e.g., helicopters vs tractors).
The same applies to the other categories, but as a robustness check, we do replace values
with quantities.
The Effect of Police Militarization on Crime 10
—————— [Table 1 in here] ——————
In addition, Figure 2 presents the scatter plot of military aid and crime
rate using our data source. Despite the simplicity of such plot, it reveals an
interesting pattern and indicates a quite strong positive association between
the crime rate and the total value of military aid acquired by counties. This
does not come across surprising though, as counties with higher crime rate
are more likely to request support by the federal government. We, therefore,
now turn to presenting the empirical strategy that would allow us to tease
out the effect of equipment on crime from the positive correlation observed
in Figure 2.
—————— [Figure 2 in here] ——————
4 Empirical strategy
We are interested in the coefficient βfrom the following model:
Yc,t =βEquipmentc,t−1+γ‘Xc,t +αc+ηs,t +c,t (1)
The outcome variable Yc,t can be the crime rate for county cin year t.
Equipmentc,t−1is the monetary size of the military equipment that has been
given to the county, i.e., the flow of military aid from the government to
county i. We use a linear-log model i.e., the log values of Equipmentc,t−1
and Yc,t in its original scale.7This specification is useful in the presence of
diminishing marginal returns and it is easy to interpret, as the estimated
coefficient implies that a 1% increase in aid affects crime by βunits.
We lag the values of aid by one year to allow time for the equipment to be
transferred and placed into use.8Xc,t is a vector of control variables described
7We use the transformation log(1 + Equipment). It is easy to check that log(1 +
Equipment)≈log(E quipment) for numbers around the minimum value of most equip-
ment.
8We do not use more than one lag as the federal government requires agencies that
receive 1033 equipment to use it “within one year of receipt, unless the condition of the
property renders it unusable, in which case, controlled property must be returned to the
DOD/DLA Disposition Services Site” (White House, 2014, p.7).
The Effect of Police Militarization on Crime 11
in section 3. County fixed effects αcand interacted state-year fixed effects
ηs,t absorb county-specific constant features such as geographical location and
year-state specific effects such as state-wide policies and legislation changes.
To verify whether an effect of aid on the crime rate might be driven
by increased police efficiency we try to capture clearance rates by using as
dependent variable arrest rates, while controlling for the crime rate on the
right-hand side of equation 1. It is also important to note that when the
dependent variable is one of the police outcomes from the LEOKA data,
the dependent variable is at the level of the reporting agency, whereas X
and Equipment remain defined at the (geographically larger) county level.
Furthermore, note that equation 1 with crime rates as dependent variables
resembles the reduced form in a two stage model estimating the elasticity
of crime with respect to police characteristics with a first stage equation 1
with dependent variable police characteristics. Therefore, an effect of aid on
crime could either be channeled through for e.g. police manpower or through
responses of criminals. We would find evidence for the former if we find a
significant effect of equipment on police and of equipment on crime. We
would find evidence for the latter if we find an effect only on crime rates.9
As Figure 2 suggests, there is an ex-ante positive correlation between the
crime rates a county experiences and the amount of military aid it requests.
This means that the estimates of our main coefficients of interest β, in equa-
tion 1, are most certainly contaminated by a positive spurious correlation,
which would lead to an upward bias in the estimate of β. In order to allevi-
ate this concern, we employ an IV estimation method by using US military
spending in the previous year. By law, the 1033 Program allows local police
forces to acquire excess property of the Department of Defense, and when
US military spending is high (due e.g., to an increase in the intensity of the
war in Afghanistan), this generates an accumulation of military hardware,
which increases the amount of military aid delivered to local law enforcement
agencies. Yet, as US military spending only varies over time we follow the
same procedure of Nunn & Qian (2014) and interact military spending with
9We do not consider estimating a 3 stage model because of aggregation issues.
The Effect of Police Militarization on Crime 12
a county’s tendency to receive military aid from the federal government.10
In this way we aim to capture the excess equipment received to the one re-
quested, in a manner that varies between counties but yet treats counties
that received aid in the same number of years in the same way.
The first stage is then:
Equipmentc,t =α+θ"U SM ilext−1× 1
7
2012
X
t=2006
Pct!#+δ‘Xc,t +α2
c+η2
s,t +υc,t
(2)
where USM ilext−1is US military spending in constant US$ and Pct is a
dummy for whether creceived any military aid in year t. The instrument
varies by county and over time, and allows us to control for year fixed effects.
As Nunn & Qian (2014) make clear, this strategy resembles a difference-in-
differences estimation strategy, where in the first-stage (and in the reduced-
form) we compare military equipment in counties that frequently receive aid
to counties that rarely receive aid, in years following high US military spend-
ing relative to years following lower military spending. The main difference
with a difference-in-differences strategy is that our treatment variable is con-
tinuous. Our identifying variation is along the intensive margin of receiving
military aid, that is, how much is received by a county, conditional on re-
ceipt. Yet, the extensive margin - whether or not a county receives aid - is
endogeneous. The county fixed effects absorb the constant component in the
endogeneous variation in the likelihood of requesting and receiving aid, that
is, when a high-crime county is likely often to request and receive weapons.
In order to nail down the exogeneous variation we perform a robustness check
where we set the extensive margin variation to 0 and we present estimates
on a subsample of counties which received aid.
Our identification strategy is based on the premise that, conditional on
other contextual variables, our instrument has an impact on the crime rate
only through the provision of military equipment. Note that the exclusion
10Nunn & Qian (2014) investigate the effect of US food aid on conflict by using exogenous
variations in US wheat production and in recipients’ tendency to receive a positive amount
of US food aid.
The Effect of Police Militarization on Crime 13
restriction is not violated if higher US military spending affects crime rate
through its national or regional influence on e.g., voluntary military recruit-
ment, as the inclusion of state-year fixed effects flexibly account for any
national or state-specific changes over time. Note also that neither our in-
strument nor the crime rate in US counties display monotonic trends, thus
ruling out the possibility of a spurious correlation.11
We report robust standard errors clustered at the state level to allow
for the variance to differ across states and for an unstructured covariance
between counties within the clusters. Finally, we transform all the non-
percentage variables (i.e, military aid, military spending, income and popu-
lation) into logs to simplify the interpretation of the coefficients, scale down
the variance and reduce the effect of outliers.
5 Results
The purpose of this section is threefold. First, we want to examine the impact
of military aid on crime and clearance rate within the context of the long-
standing crime regression literature. We account for endogeneity bias using
instrumental variable estimation and controlling for specific county and state-
year fixed effects. Second, we wish to establish the extent to which distinct
indexes of crime such as robbery or assault responds differently to increases
in each category of military aid such as weapons or vehicles. Third, we
provide a discussion of the degree to which our estimates can be interpreted
as providing evidence of deterrence and run a number of robustness tests.
11On one hand, although the US has experienced a general decline in crime rates in
recent years, there is a lot of heterogeneity across counties and some categories of crime
like burglaries or larcenies have hardly seen significant changes in the aggregate over time.
At the same time, both military spending and the number of casualties per year display a
inversely U-shape pattern over time (see Figure A.1). Moreover, the interaction between
US military spending (or casualties) and Pct gives lower weights to counties with arguably
less crime i.e., those who have requested/received aid less frequently. Thus, even if we
assumed that the relation is spurious, the resulting coefficient would be positive rather
than negative.
The Effect of Police Militarization on Crime 14
5.1 Does aid affect crime and arrest rates?
Our first question revolves around the existence of a causal effect of military
aid on crime and arrest rates. We start from panel A of Table 2, which
incorporates the baseline model.12 Before coming to our main results, note
that the instrumental variable is positively correlated to the value of military
aid received, implying that the OLS estimates in column 1 suffer from an
upward bias. In column 2, the first stage regression of the change in military
aid on the interaction between US military spending and the propensity to
receive aid gives a significant positive coefficient, as expected. This means
that an increase in the military expenditure of last year holding aid receipt
probability constant leads to a higher amount of aid received by the county.
The Kleinbergen-Paap F-Statistic is similar to the conventional F-statistic,
but takes into account the clustering of the standard errors. The values are
all above 20, and above conventional levels characterizing weak instruments.
As concerns our variables of main interest, remember that our model
specification allows for direct reading of the coefficients. Column 3 of Table
2 shows that military aid does reduce the total number of crimes: a 10%
increase in the total value of military aid leads to a decrease of 4.9 units in
the overall number of events (per 100,000 population).13
Reading across the first row of results, we find that this reduction goes
mainly through a decrease in robberies, assaults, larcenies and motor vehicle
thefts. The effect is very pronounced with street-level crime types, like larce-
nies (-17.7), vehicle thefts (-12) and assaults (-6.9), whereas it is insignificant
for burglaries, which suggests that military aid could have a deterrent effect
based on greater visibility. This “display” mechanism could deter crime by
increasing the subjective probability of arrest. Even though the effect of mil-
itary aid on the homicide rate is statistically discernible from zero at the 10%
level, the magnitude is quite negligible: a 10% increase in the total value of
aid leads to an approximately 0.02 unit decrease in the number of homicides.
Given that we identify mainly the effect along the intensive margin of
12We report the results for the control variables in Table A.2 in the Appendix.
13This corresponds to an elasticity of crime with respect to aid of approximately -0.02.
Running a log-log model yields virtually the same elasticity.
The Effect of Police Militarization on Crime 15
aid received, recall that a positive correlation does exist between the crime
rate and the value (or amount) of military aid received. This suggests a
positive bias to interpreting the effect of aid at the extensive margin, that is,
whether or not to award aid to a given county. Thus any significant negative
relation uncovered is a conservative estimate of what is likely to be stronger
relationship in the true population.
A hint of this is presented in panel B of table 2, where we estimate the
same model on a subsample of counties that received a positive quantity of
aid. The coefficients in the last 7 columns are almost all higher in absolute
terms than the ones in panel A. Here we observe a stronger effect on crime
rates - a decrease of 9 units instead of 4.9 and an increase in the observed
effect of homicides, robberies, assaults, burglaries and vehicle theft. In the
following estimates we prefer to use the specification that earns the more
conservative estimates.
—————— [Table 2 in here] ——————
Turning to arrest rates, and moving across the columns of Table 3, we
do not find support for an effect of military aid on the number of arrests.
Note that in this specification we include as independent variable the crime
rate in order to account for changes in the number of arrests that are driven
by changes in the crime rate. Military aid does not seem to improve the
performance of local police units when measuring with the arrest rate how
good police are at solving crimes, thus revealing that it rather helps deterring
individuals from participating in illegal activities in the first place.14
—————— [Table 3 in here] ——————
5.2 Which type of military aid is most constructive in
combating crime?
Our second question is whether specific categories of military aid have dis-
tinguishable effects on the crime rate. This is crucial in light of the recent
14Note that we use most of the control variables included in studies on arrest and
clearance rates (see e.g., Levitt, 1998).
The Effect of Police Militarization on Crime 16
heated debate on the excessive use of military weapons and tactics by the
police forces. We therefore restrict our attention to aggregate crime as well
as to the indexes which were found to be significantly affected by aid in Table
2. Results are reported in Table 4, where we detect a sort of hierarchy in the
marginal impact of aid on crime: the residual category, which we labelled
“others”, and which includes only non-lethal equipment, without military
attributes, has the biggest marginal effect on the reduction of crime. This
is most evident in the fifth panel of Table 4, where a 10% increase in this
category reduces the total number of crimes by more than six units per year.
The second class of items is vehicles, and the final one is gears. Three basic
implications emerge from this Table: firstly, weapons do not appear to work
as a deterrence tool and our instrument performs worse in predicting the
amount of weapons allocated than other components of aid. This might be
due to the controversy on the value-added of using battlefield weapons to
police urban areas. This result applies to all the four sub-categories of crime
we are interested in as well as to the overall crime level.
Secondly, in terms of the other components of military aid, two highly
visible tools, gears and vehicles, have strong and sizable effects on all the
types of crime. Although vehicles are easily detectable, note that gears in-
clude sophisticated electronic equipments, training aid and, in almost half
of the instances, clothing. This is consistent with early studies by e.g., Bell
(1982), who explores how wearing military-style uniforms influence citizens’
perception of the police’s authority and legitimacy, and reinforces the notion
that a main causal channel could be based on perceptual deterrence. Thirdly,
even though the residual category is still too aggregate and large to make
reasonable claims about which of its subcomponents are driving the effect,
the inclusion of diverse office equipment could entail that law-enforcement is
better able to allocate its time resources, resulting in more patrols or other
crime-deterring activities. Additionally, laboratories and IT hardwares might
ultimately lend support to previous findings by Mastrobuoni (2014) on how
IT adoption (or innovation) affects crime.15
15As aid without clear military attributes is not the focus of this article, however, we
leave this facet to future research.
The Effect of Police Militarization on Crime 17
—————————— [Table 4 in here] ——————————
5.3 Additional checks
We submit our findings to a round of additional checks. Firstly, we want to
look further into whether there are other possible channels linking military
aid to a reduction in crime other than “deterrence via display”. We therefore
explore in more details the different inputs of the production function for law
enforcement and estimate a number of additional models. One could look
at these models as being the first stage in a two stage framework, where the
main variable of interest is the elasticity of crime with respect to police char-
acteristics. Table 5 includes seven dependent variables: the number of police
officers, the number of employees, the ratio between officers and employees,
the number of calls, the number of injuries and assaults on police officers,
and the number of offenders killed by the police. As we can see, military aid
does not influence hiring decisions by inducing law enforcement agencies to
devote more resources to new police and employees hires; it neither affects
police activities measured by the number of calls nor it has a significant im-
pact on the number of police officers assaulted or injured in the line of duty;
and it does not affect the number of offenders killed by the police. Even
though we find a significant F-statistics, it seems that the deterrent effect
observed in Table 2 comes either through a response of the supply side of
the crime market, that is, criminals themselves respond rather than police,
or through unobservable police effort. Military aid therefore seems to lead
to a reduction in crime rates mainly through a deterrence mechanism.
Secondly, in Table 6, panel A, all results are weighted by the size of the
mean population to reflect crime as a population mean, and by and large, the
results carry over. In fact, the coefficients are now distinctly bigger, while
keeping the statistical significance. Thirdly, in panel B we keep a subsam-
ple of counties that contain law-enforcement that have fully complied with
reporting, that is with 100 percent coverage of reported crime. We however
drop almost half of the observations and find that robberies, assaults and
vehicle-theft survive this robustness check. Fourthly, panels C and D exploit
The Effect of Police Militarization on Crime 18
alternative, yet related, instruments, such as the amount of military spending
in % of the GDP and the total number of US fatalities in Afghanistan and
Iraq. The rationale behind the inclusion of the share of output devoted to
the military, also called the military burden, is that it measures the priority
given to defense rather than military power or the absolute level of military
expenditure (Smith, 2009). As we can see, the coefficients are significant
and in the same order of magnitude of those in Table 2. Using US military
casualties as an alternative instrument allows us to effectively capture the
intensity of military deployments and the severity of war, which in turn influ-
ence the procurement of military equipment. Again, previous findings about
aid and crime are strongly borne out by this new set of empirical results.
The coefficients are greater than in our conservative baseline models and the
F-statistics are above conventional levels. Fifthly, when we replace the value
of military aid by its quantity, in Table 7, panel E, our results are still sig-
nificant, thus lending additional support to previous results. Note that here
we do not explicitly comment on the substantive effects as each category
contains highly heterogeneous items. From these last checks, exploring var-
ious estimation techniques and specifications, we feel confident to conclude
that military aid reduces crime, and the effect is likely to be driven by a
deterrence mechanism. Finally, in the remaining panels F and G of Table
7, we account in different ways for preexisting trends in crime. In panel F
we limit the underlying sample to consisting of counties in which the mean
population size is lower than 250,000 inhabitants. In this way we exclude
urban counties, which are likely to have different crime trends than urban
counties. In panel G we include county-specific linear trends to the baseline
specification and we exclude state-time dummies. In this way we account
for differential crime trends across counties. On both exercises our baseline
results remain unaffected and, if anything, accounting for differential crime
trends lead to higher effect sizes.
The Effect of Police Militarization on Crime 19
5.4 Cost and Benefit Analysis
We perform a basic cost-benefit analysis by comparing estimates from our
baseline models in Table 2 with estimates of the social cost of particular
crimes. Heaton (2010) provides one of most recent review of academic re-
search on the cost of crime in the US, including accounting-based methods
(all the individual costs that individuals and society bear) and contingent
valuations (what individuals are willing to pay for crime reduction). He
summarizes the cost estimates from three studies of the cost of crime, two
using accounting-based methods (Cohen & Piquero, 2009; McCollister et al.
, 2010) and one using contingent valuation (Cohen et al. , 2004). He calcu-
lates that the average cost of a robbery is $67,277 (in 2007 US dollars), of a
serious assault is $87,238, of a larceny is $2,139 and of a motor-vehicle theft
is $9,079.
By our baseline and most conservative models (Table 2) a 10% increase
(around $5,100) in the value of aid reduces robberies by 0.5 units, assaults
by 0.7 units and vehicle thefts by 1.2 unit. This means that the benefit of a
10% increase in aid is roughly $105,000,16 making the donation of military
equipment a good investment.
When we differentiate by type of military equipment received, we find
that the net benefit from receiving military gear is roughly $ 75,000. For
vehicles it is $ 78,000 and for the catchall residual category it is $ 123,000.
It is not surprising that the latter effect, taken together with the content in
Table A.2, is so big given the heterogeneity of items within. We leave the
mechanism of this effect to further research.
6 Conclusions
In 1990 the US Congress enacted the National Defense Authorization Act,
later called “1033 Program”, which allowed local law enforcement agencies to
acquire excess property of the Department of Defense, including drones, mili-
160.5∗67277 + 0.7∗87238 + 1.2∗9079 compared to a cost of $5,100. We do not consider
homicides here, although the cost is clearly much larger than the cost of all the other types
of crime, and it is estimated to be around $8,649,216.
The Effect of Police Militarization on Crime 20
tary weapons and armored vehicles. The program came under severe scrutiny
in 2014, following a wave of public protests against the disproportionate use
of military tools by local police forces. By most reports, providing mili-
tary equipment free of charge encourages hyper-aggressive forms of domestic
policing, which can increase tension, mistrust and uncooperative behaviors
between local police departments and local communities. Yet, there are no
attempts to examine the tangible outcomes of issuing military equipment to
law enforcement agencies, in particular its effect on crime rates.
Using panel data for counties over the 2007-2012 period, we provide the
first quantitative evidence of the effect of the 1033 Program on police per-
formances. En route, we complement the economic literature on the deter-
minants of crime. Our identification strategy relies on exogenous variation
in timing and size of military spending to test whether the militarization of
local police forces improved their performances.
The results reported in this article provide supportive evidence for a de-
terrent effect of military hardware on crime rate via a deterrence mechanism.
In other words, when a criminal observes that the police has at their disposal
additional military vehicles or military gears, she is more likely to be deterred.
We run a number of additional models to isolate this mechanism from other
competing channels. Interestingly, although all the non-lethal categories of
aid are effective in preventing crime, this effect is not reflected in a shift in
observable police characteristics such as arrest rates, manpower or others,
hinting at an effect channel of unobservable police effort.
The lower end of this estimated impact implies that a 10% increase in the
value of military aid reduces the total number of crimes by 4.9 units. Despite
a small total effect, the program is quite cost-effective, and adding an extra
$5,100 in overall aid leads to a drop of roughly $105,000 in the social costs
of robberies, assaults and vehicle thefts combined. That said, taken together
our results do not directly provide evidence in favor or against the possibility
that military weapons contribute to overly aggressive approaches by police
units, which in turn can escalate to a standoff between urban communities
and the officers that police them. This is clearly a social cost that our analysis
cannot duly capture and it is an important avenue for future research.
The Effect of Police Militarization on Crime 21
References
American Civil Liberties Union, (ACLU). 2014. War Comes Home: The Excessive Mili-
tarization of American Policing. New York, NY: American Civil Liberties Union.
Balko, Radley. 2013. Rise of the warrior cop: The militarization of America’s police forces.
PublicAffairs.
Bell, Daniel J. 1982. Police uniforms, attitudes, and citizens. Journal of criminal justice,
10(1), 45–55.
Chalfin, Aaron, & McCrary, Justin. 2013. Are US Cities Under-Policed? Theory and
Evidence. NBER Working Paper,18815.
Chalfin, Aaron, & McCrary, Justin. 2014. Criminal Deterrence: A Review of the Litera-
ture. Tech. rept. Working paper May.
Cohen, Mark A, & Piquero, Alex R. 2009. New evidence on the monetary value of saving
a high risk youth. Journal of Quantitative Criminology,25(1), 25–49.
Cohen, Mark A, Rust, Roland T, Steen, Sara, & Tidd, Simon T. 2004. WILLINGNESS-
TO-PAY FOR CRIME CONTROL PROGRAMS. Criminology,42(1), 89–110.
Di Tella, Rafael, & Schargrodsky, Ernesto. 2004. Do police reduce crime? Estimates using
the allocation of police forces after a terrorist attack. The American Economic Review,
94(1), 115–133.
Draca, Mirko, Machin, Stephen, & Witt, Robert. 2011. Panic on the streets of london:
Police, crime, and the july 2005 terror attacks. The American Economic Review, 2157–
2181.
Evans, William N, & Owens, Emily G. 2007. COPS and Crime. Journal of Public Eco-
nomics,91(1), 181–201.
Hall, Abigail R, & Coyne, Christopher J. 2012. The Militarization of US Domestic Policing.
The Independent Review: A Journal of Political Economy, Forthcoming, 12–50.
Heaton, Paul. 2010. Hidden in Plain Sight: What Cost-of-Crime Research Can Tell Us
About Investing in Police. Santa Monica, Calif.: RAND Corporation.
Hill, Stephen, & Beger, Randall. 2009. A paramilitary policing juggernaut. Social justice,
25–40.
Klick, Jonathan, & Tabarrok, Alexander. 2005. Using Terror Alert Levels to Estimate the
Effect of Police on Crime*. Journal of Law and Economics,48(1), 267–279.
The Effect of Police Militarization on Crime 22
Levitt, Steven D. 1998. Why do increased arrest rates appear to reduce crime: Deterrence,
incapacitation, or measurement error? Economic Inquiry,36(3), 353.
Machin, Stephen, & Marie, Olivier. 2011. Crime and police resources: The street crime
initiative. Journal of the European Economic Association,9(4), 678–701.
Mastrobuoni, Giovanni. 2014. Crime is Terribly Revealing: Information Technology and
Police Productivity. Mimeo, University of Essex.
McCollister, Kathryn E, French, Michael T, & Fang, Hai. 2010. The cost of crime to
society: New crime-specific estimates for policy and program evaluation. Drug and
alcohol dependence,108(1), 98–109.
Nunn, Nathan, & Qian, Nancy. 2014. US food aid and civil conflict. The American
Economic Review,104(6), 1630–1666.
Owens, Emily Greene. 2013. COPS and Cuffs. Lessons from the Economics of Crime:
What Reduces Offending?, 17.
Smith, Ron P. 2009. Military economics: the interaction of power and money. Palgrave
Macmillan.
White House, (WH). 2014. Review: Federal Support for Local Law Enforce-
ment Equipment Acquisition. Executive Office of the President, Washing-
ton DC, https: // www. whitehouse. gov/ sites/ default/ files/ docs/ federal_
support_ for_ local_ law_ enforcement_ equipment_ acquisition. pdf .
White House, (WH). 2015. Executive Order 13688 - Federal Support for Local Law
Enforcement Equipment Acquisition. Law Enforcement Equipment Woking Group,
http: // www. gpo. gov/ fdsys/ pkg/ DCPD-201500033/ pdf/ DCPD-201500033. pdf .
The Effect of Police Militarization on Crime 23
Figure 1: Total Aid by Year in Monetary Terms
Figure 2: Military Aid (in log) and Crime (per 100,000)
Notes: Each point of the scatter presents average crime per county over the sample
period and the average monetary amount of aid received.
The Effect of Police Militarization on Crime 24
Table 1: Summary statistics
Variable Mean Std. Dev. Min. Max.
Crime Rate 2306.8 1441.6 0 12102.9
Murder Rate 3.1 5.9 0 118.2
Robbery Rate 37.4 62.4 0 795.7
Assault Rate 184.7 175.1 0 1989
Burglary Rate 543.3 385.7 0 5071.6
Larceny Rate 1414 906.2 0 7168
Vehicle Theft Rate 124.3 126.8 0 2006
Murder Arrest Rate 2.8 6 0 134.2
Robbery Arrest Rate 14.9 22.3 0 628.9
Assault Arrest Rate 99.7 100.6 0 4742
Burglary Arrest Rate 84.2 70.1 0 1097.8
Larceny Arrest Rate 295.1 261.7 0 7844.2
Vehicle Arrest Theft 19.5 24.4 0 608.5
Military Exp. IV 4.2 3.6 0 13.5
Total Aid (value) 51094.8 1229356 0 152824128
Total Aid (quantity) 62.8 1026.3 0 97348
Weapons Aid 1339.1 12292.7 0 641136
Weapons Quantity 10 211.2 0 21586
Vehicles Aid 31649 1202910.4 0 152279640.2
Vehicles Quantity 1.2 14.7 0 715
Gears Aid 9830.5 130130.3 0 8574024
Gears Quantity 28.4 674.4 0 78634
Others Aid 8276.3 100904.6 0 4099528
Others Quantity 23.2 433.2 0 26153
Percent Poverty 16.3 6.4 2.4 62
Median Income 43621.4 11197.7 17488 121250
Unemployment Rate 7.6 3.2 1.1 29.1
Population 95241.9 251349.1 385 4272223
Share Males 0.3 0.1 0 1
Share Blacks 0.1 0.1 0 0.6
Share Age 15-19 0.1 0 0 0.3
Share Age 20-24 0.1 0 0 0.4
Share Age 25-29 0.1 0 0 0.2
Share Age 30-34 0.1 0 0 0.2
N 17514
The Effect of Police Militarization on Crime 25
Table 2: The Effect of Military Aid on Crime
OLS First Stage Crime Homicide Robbery Assault Burglary Larceny Vehicle Theft
Panel A. Extensive and Intensive Margin
Military Exp. IV 17.154***
(2.754)
Lagged Total Aid 1.127 -49.296*** -0.155* -4.868*** -6.873*** -7.646 -17.725** -12.029***
(1.399) (15.373) (0.090) (1.124) (2.106) (5.030) (7.939) (2.989)
Constant 26500.637*** -21.290
(8191.361) (18.795)
Observations 17514 17514 17514 17514 17514 17514 17514 17514 17514
Kleibergen-Paap F-Statistic 38.795 38.795 38.795 38.795 38.795 38.795 38.795
Panel B. Intensive Margin only
Military Exp. IV 15.534***
(3.201)
Lagged Total Aid -1.617 -90.849*** -0.535** -7.729***-10.080** -14.265* -42.669*** -15.571***
(1.338) (22.610) (0.255) (2.231) (4.650) (8.065) (11.154) (4.211)
Observations 4268 4268 4268 4268 4268 4268 4268 4268 4268
Kleibergen-Paap F-Statistic 22.119 22.119 22.119 22.119 22.119 22.119 22.119
Notes: Control variables: median income, poverty rate, unemployment rate, population, share of males, blacks, and age 15-19, 20-24, 25-29, 30-34. All models
include county and interacted state-year fixed effects. The instrument is lagged military expenditure times the probability of receiving military aid. R-squared
of the model in the first column in panel A is 0.095, for panel B it is 0.221. Robust standard errors clustered at the state level reported in parenthesis.
Asterisks denote: ∗∗∗p < 0.01,∗ ∗ p < 0.05,∗p < 0.1.
The Effect of Police Militarization on Crime 26
Table 3: The Effect of Military Aid on Arrest Rates
Homicide Robbery Assault Burglary Larceny Vehicle Theft
Lagged Total Aid -0.125 0.164 0.777 -1.224 -1.135 0.283
(0.108) (0.287) (1.252) (1.199) (2.165) (0.420)
Murder Rate 0.472***
(0.042)
Robbery Rate 0.196***
(0.032)
Assault Rate 0.243***
(0.029)
Burglary Rate 0.060***
(0.007)
Larceny Rate 0.076***
(0.012)
Vehicle Theft Rate 0.106***
(0.013)
Observations 17514 17514 17514 17514 17514 17514
Kleibergen-Paap F-Statistic 38.814 38.671 38.919 38.825 38.759 39.750
Notes: Control variables: median income, poverty rate, unemployment rate, population, share of males, blacks, and
age 15-19, 20-24, 25-29, 30-34. All models include county and interacted state-year fixed effects. The instrument
is lagged military expenditure times the probability of receiving military aid. Robust standard errors clustered at
the state level reported in parenthesis. Asterisks denote: ∗∗∗p < 0.01,∗ ∗ p < 0.05,∗p < 0.1.
The Effect of Police Militarization on Crime 27
Table 4: The Effect of Categories of Military Aid on Crime
Crime Robbery Assault Larceny Vehicle Theft
Lagged Total Aid -49.296** -4.868*** -6.873** -17.725* -12.029***
(15.373) (1.124) (2.106) (7.939) (2.989)
Kleibergen-Paap F-Statistic 38.795 38.795 38.795 38.795 38.795
Lagged Weapons Aid 221.725 21.896 30.914 79.724 54.105
(139.062) (12.463) (19.069) (58.559) (30.066)
Kleibergen-Paap F-Statistic 3.351 3.351 3.351 3.351 3.351
Lagged Gears Aid -37.621*** -3.715*** -5.245*** -13.527* -9.180***
(10.781) (0.641) (1.535) (5.900) (1.645)
Kleibergen-Paap F-Statistic 154.215 154.215 154.215 154.215 154.215
Lagged Vehicles Aid -39.827*** -3.933*** -5.553** -14.320* -9.719***
(11.848) (0.819) (1.751) (6.248) (1.910)
Kleibergen-Paap F-Statistic 98.387 98.387 98.387 98.387 98.387
Lagged Others Aid -62.164*** -6.139*** -8.667** -22.352* -15.169***
(18.190) (1.111) (2.812) (9.700) (2.711)
Observations 17514 17514 17514 17514 17514
Kleibergen-Paap F-Statistic 74.605 74.605 74.605 74.605 74.605
Notes: Control variables: median income, poverty rate, unemployment rate, population, share of
males, blacks, and age 15-19, 20-24, 25-29, 30-34. All models include county and interacted state-year
fixed effects. The instrument is lagged military expenditure times the probability of receiving military
aid. Robust standard errors clustered at the state level reported in parenthesis. Asterisks denote:
∗∗∗p < 0.01,∗ ∗ p < 0.05,∗p < 0.1.
The Effect of Police Militarization on Crime 28
Table 5: The Effect of Military Aid on Police Activities
Officers to Civil Disorder Offenders
Officers Employees Employees Ratio Calls Injuries Assaults Killed
Lagged Total Aid -0.404 0.219 0.001 0.011 0.001 -0.000 0.000
(0.345) (0.445) (0.001) (0.012) (0.005) (0.000) (0.000)
Observations 112262 112262 103426 112262 112262 112262 15965
KP F-Statistic 11.170 11.170 10.207 11.170 11.170 11.170 8.440
Notes: Control variables: median income, poverty rate, unemployment rate, population, share of males, blacks, and age 15-
19, 20-24, 25-29, 30-34. All models include county and interacted state-year fixed effects. The instrument is lagged military
expenditure times the probability of receiving military aid. Robust standard errors clustered at the state level reported in
parenthesis. KP F-Statistic stand for Kleibergen-Paap F-Statistic. Asterisks denote: ∗∗∗p < 0.01,∗ ∗ p < 0.05,∗p < 0.1.
The Effect of Police Militarization on Crime 29
Table 6: Robustness Checks on the Effect of Categories of Military Aid on Crime
First Stage Crime Homicide Robbery Assault Burglary Larceny Vehicle Theft
Panel A. Population Weights
Military Exp. IV 22.652***
(3.609)
Lagged Total Aid -69.830*** -0.203*** -8.877*** -6.307*** -11.775** -25.085*** -17.584***
(16.435) (0.057) (1.726) (1.462) (5.453) (8.209) (3.469)
Constant 7.621
(80.926)
KP F-Statistic 39.405 39.405 39.405 39.405 39.405 39.405 39.405
Panel B. Full Coverage
Military Exp. IV 18.893***
(3.576)
Lagged Total Aid -27.405 -0.103 -3.171*** -6.011** -2.357 -7.610 -8.152*
(23.421) (0.128) (1.033) (2.642) (5.951) (14.292) (4.282)
Constant -36.678
(25.123)
KP F-Statistic 27.908 27.908 27.908 27.908 27.908 27.908 27.908
Panel C. US Military Spending in % of GDP as IV
Burden IV 3.767***
(0.650)
Lagged Total Aid -42.962*** -0.150 -3.562*** -4.723** -3.577 -23.426** -7.524***
(16.569) (0.121) (0.930) (1.911) (4.704) (9.731) (2.801)
Constant 18.315
(19.165)
KP F-Statistic 33.544 33.544 33.544 33.544 33.544 33.544 33.544
Panel D. US Military Fatalities in Afghanistan and Iraq as IV
Casualties IV -2.847***
(0.594)
Lagged Total Aid -74.350*** -0.144 -6.689*** -11.185***-11.262* -28.922*** -16.149***
(22.076) (0.142) (1.700) (3.167) (6.323) (11.017) (4.542)
Constant 43.761**
(17.189)
KP F-Statistic 22.999 22.999 22.999 22.999 22.999 22.999 22.999
Notes: Control variables: median income, poverty rate, unemployment rate, population, share of males, blacks, and age 15-19, 20-24,
25-29, 30-34. All models include county and interacted state-year fixed effects. In Panels A and B the instrument is lagged military
expenditure times the probability of receiving military aid. In Panels C and D we replace military expenditure with military burden
or the number of US fatalities, respectively. Robust standard errors clustered at the state level reported in parenthesis. KP F-Statistic
stand for Kleibergen-Paap F-Statistic. Asterisks denote: ∗ ∗ ∗p < 0.01,∗ ∗ p < 0.05,∗p < 0.1. The number of observations for the
estimates in Panel B is 9524. All other panels have the same number of observations: 17514.
The Effect of Police Militarization on Crime 30
Table 7: Robustness Checks on the Effect of Categories of Military Aid on Crime
First Stage Crime Homicide Robbery Assault Burglary Larceny Vehicle Theft
Panel E. Quantities instead of Monetary Equivalent of Aid
Military Exp. IV 730.392***
(152.215)
Lagged Total Quantity -1.158*** -0.004 -0.114*** -0.161** -0.180 -0.416** -0.283***
(0.403) (0.002) (0.027) (0.064) (0.125) (0.198) (0.061)
Constant -3679.880***
(1035.496)
Observations 17514 17514 17514 17514 17514 17514 17514 17514
KP F-Statistic 23.026 23.026 23.026 23.026 23.026 23.026 23.026
Panel F. Population <250000
Military Exp. IV 16.366***
(3.266)
Lagged Total Aid -22.095 -0.142 -1.954*** -5.922** -3.141 -5.184 -5.751**
(14.399) (0.113) (0.620) (2.510) (5.443) (8.277) (2.242)
Constant -15.584
(17.127)
KP F-Statistic 25.106 25.106 25.106 25.106 25.106 25.106 25.106
Panel G. Including County Linear Trends
Military Exp. IV 9.853***
(3.062)
Lagged Total Aid -102.927** -0.360** -8.218***-12.530** -7.714 -46.314** -27.791***
(44.437) (0.171) (2.908) (5.716) (11.070) (20.958) (9.388)
Constant 38.813
(23.547)
KP F-Statistic 10.358 10.358 10.358 10.358 10.358 10.358 10.358
Notes: Control variables: median income, poverty rate, unemployment rate, population, share of males, blacks, and age 15-19, 20-24, 25-29,
30-34. All models include county and interacted state-year fixed effects. In Panel E the instrument is lagged military expenditure times
the probability of receiving military aid. In Panel G we include county linear time trends and we exclude the interacted state-year fixed
effects. Robust standard errors clustered at the state level reported in parenthesis. KP F-Statistic stand for Kleibergen-Paap F-Statistic.
Asterisks denote: ∗ ∗ ∗p < 0.01,∗ ∗ p < 0.05,∗p < 0.1. The number of observations in Panel F is 16074. All other panels have the same
number of observations: 17514.
The Effect of Police Militarization on Crime 31
Appendix A
The Effect of Police Militarization on Crime 32
Table A.1: Federal supply categories
Weapons Freq. Percent Cum.
AMMUNITION AND EXPLOSIVES 528 0.54 0.54
GUIDED MISSLES 6 0.01 0.54
WEAPONS 97,772 99.46 100
Total 98,306 100
Vehicles Freq. Percent Cum.
AIRCRAFT COMPONENTS/ACCESSORIES 856 4.91 4.91
AIRCRAFT LAUNCH/LAND/GROUND HANDLE 201 1.15 6.06
AIRCRAFT/AIRFRAME STRUCTURE COMPTS 750 4.3 10.36
ENGINE ACCESSORIES 358 2.05 12.41
ENGINES AND TURBINES AND COMPONENT 443 2.54 14.95
MECHANICAL POWER TRANSMISSION EQPT 139 0.8 15.74
MOTOR VEHICLES, CYCLES, TRAILERS 10,425 59.74 75.48
SHIP AND MARINE EQUIPMENT 134 0.77 76.25
SHIPS, SMALL CRAFT, PONTOON, DOCKS 152 0.87 77.12
TIRES AND TUBES 456 2.61 79.74
TRACTORS 366 2.1 81.83
VEHICULAR EQUIPMENT COMPONENTS 3,170 18.17 100
Total 17,450 100
Gears Freq. Percent Cum.
CLOTHING/INDIVIDUAL EQPT, INSIGNIA 30,993 48.36 48.36
COMM/DETECT/COHERENT RADIATION 12,694 19.81 68.17
ELECTRICAL/ELECTRONIC EQPT COMPNTS 1,373 2.14 70.31
FIRE CONTROL EQPT. 12,013 18.75 89.06
FIRE/RESCUE/SAFETY; ENVIRO PROTECT 3,434 5.36 94.42
PHOTOGRAPHIC EQPT 1,468 2.29 96.71
RECREATIONAL/ATHLETIC EQPT 1,713 2.67 99.38
TRAINING AIDS AND DEVICES 398 0.62 100
Total 64,086 100
The Effect of Police Militarization on Crime 33
Others Freq. Percent Cum.
ADP EQPT/SOFTWARE/SUPPLIES AND EQPT 4,656 8.95 8.95
AGRICULTURAL MACHINERY AND EQPT 477 0.92 9.86
ALARM, SIGNAL, SECURITY DETECTION 265 0.51 10.37
BEARINGS 84 0.16 10.53
BOOKS, MAPS, OTHER PUBLICATIONS 20 0.04 10.57
BRUSHES, PAINTS, SEALERS, ADHESIVES 102 0.2 10.77
CHEMICALS AND CHEMICAL PRODUCTS 51 0.1 10.87
CLEANING EQPT AND SUPPLIES 414 0.8 11.66
CONSTRUCT/MINE/EXCAVATE/HIGHWY EQPT 742 1.43 13.09
CONSTRUCTION AND BUILDING MATERIAL 215 0.41 13.5
CONTAINERS/PACKAGING/PACKING SUPPL 2,377 4.57 18.07
ELECTRIC WIRE, POWER DISTRIB EQPT 3,007 5.78 23.84
FIBER OPTIC 18 0.03 23.88
FOOD PREPARATION/SERVING EQPT 1,084 2.08 25.96
FUELS, LUBRICANTS, OILS, WAXES 80 0.15 26.12
FURNACE/STEAM/DRYING; NUCL REACTOR 46 0.09 26.2
FURNITURE 2,174 4.18 30.38
HAND TOOLS 9,433 18.12 48.5
HARDWARE AND ABRASIVES 1,687 3.24 51.74
HOUSEHOLD/COMMERC FURNISH/APPLIANCE 1,433 2.75 54.5
INSTRUMENTS AND LABORATORY EQPT 2,322 4.46 58.96
LIGHTING FIXTURES, LAMPS 3,220 6.19 65.15
LUMBER, MILLWORK, PLYWOOD, VENEER 37 0.07 65.22
MAINT/REPAIR SHOP EQPT 1,637 3.15 68.36
MATERIALS HANDLING EQPT 1,303 2.5 70.86
MEASURING TOOLS 156 0.3 71.16
MEDICAL/DENTAL/VETERINARY EQPT/SUPP 6,462 12.41 83.58
METAL BARS, SHEETS, SHAPES 480 0.92 84.5
METALWORKING MACHINERY 919 1.77 86.27
MISCELLANEOUS 283 0.54 86.81
MUSICAL INST/PHONOGRAPH/HOME RADIO 265 0.51 87.32
The Effect of Police Militarization on Crime 34
NONMETALLIC FABRICATED MATERIALS 237 0.46 87.78
OFFICE MACH/TEXT PROCESS/VISIB REC 178 0.34 88.12
OFFICE SUPPLIES AND DEVICES 634 1.22 89.34
ORES, MINERALS AND PRIMARY PRODUCTS 2 0 89.34
PIPE, TUBING, HOSE, AND FITTINGS 199 0.38 89.72
PLUMBING, HEATING, WASTE DISPOSAL 393 0.76 90.48
PREFAB STRUCTURES/SCAFFOLDING 432 0.83 91.31
PUMPS AND COMPRESSORS 572 1.1 92.41
REFRIG, AIR CONDIT/CIRCULAT EQPT 902 1.73 94.14
ROPE, CABLE, CHAIN, FITTINGS 768 1.48 95.61
SERVICE AND TRADE EQPT 303 0.58 96.2
SPECIAL INDUSTRY MACHINERY 269 0.52 96.71
SUBSISTENCE 33 0.06 96.78
TEXTILE/LEATHER/FUR; TENT; FLAG 1,443 2.77 99.55
TOILETRIES 61 0.12 99.67
VALVES 23 0.04 99.71
WATER PURIFICATION/SEWAGE TREATMENT 56 0.11 99.82
WOODWORKING MACHINERY AND EQPT 95 0.18 100
Total 52,050 100
The Effect of Police Militarization on Crime 35
Table A.2: The Effect of Military Aid on Crime
OLS First Stage Crime Homicide Robbery Assault Burglary Larceny Vehicle Theft
Military Exp. IV 17.154***
(2.754)
Lagged Total Aid 1.127 -49.296*** -0.155* -4.868*** -6.873*** -7.646 -17.725** -12.029***
(1.399) (15.373) (0.090) (1.124) (2.106) (5.030) (7.939) (2.989)
Percent Poverty -4.522 -0.046*** -5.724 0.045 -0.547** -0.799 -0.428 -3.034 -0.960*
(4.018) (0.017) (4.082) (0.034) (0.227) (0.847) (1.209) (2.177) (0.577)
Log Median Income 391.175*** -1.219* 277.033** 3.695** 12.213** 41.113* 7.783 149.840** 62.389***
(129.410) (0.652) (117.431) (1.535) (5.848) (24.961) (30.783) (69.312) (14.994)
Unemployment Rate -4.712 -0.036 -8.204 -0.028 -0.565* -2.054* -0.773 -0.981 -3.804***
(6.932) (0.039) (6.710) (0.062) (0.310) (1.225) (1.813) (4.221) (0.972)
Log Population -2717.370*** -3.271* -2726.147*** 0.039 -88.941*** -204.142*** -688.173*** -1405.422*** -339.508***
(707.534) (1.702) (701.668) (3.883) (17.616) (63.781) (214.550) (357.163) (100.403)
Share Males -1668.337 -8.786 -1724.922 5.474 -187.848*** 188.446 -517.860 -630.953 -582.182***
(1418.587) (5.992) (1395.498) (10.451) (59.464) (206.920) (497.757) (895.251) (190.190)
Share Blacks -1441.323 -2.850 -1487.895 12.889 85.273 621.058** 143.619 -2068.722*** -282.012
(1534.553) (5.421) (1520.110) (14.729) (132.714) (266.945) (439.708) (753.125) (271.501)
Share Age 15-19 -2504.878 -9.803 -2977.527 -11.183 -71.419 -67.509 -492.013 -1545.682 -789.719
(4915.545) (10.972) (4668.099) (18.423) (185.562) (960.865) (1120.580) (1735.259) (948.210)
Share Age 20-24 4766.030 12.817 5609.379 -6.907 162.335 -223.204 1157.741 3622.200* 897.215**
(3678.968) (10.371) (3557.857) (20.979) (106.987) (477.634) (946.619) (2112.749) (440.623)
Share Age 25-29 3226.783 11.052 3400.424* -4.487 15.301 198.881 -348.357 3053.505** 485.581
(2000.568) (7.230) (1944.485) (20.001) (81.209) (276.205) (661.377) (1242.685) (300.273)
Share Age 30-34 -884.952 -11.430 -1325.736 -14.355 -348.213* -500.294 106.665 -706.225 136.686
(2939.058) (11.466) (2954.187) (24.493) (205.917) (511.239) (837.819) (1707.247) (479.149)
Constant 26500.637*** -21.290
(8191.361) (18.795)
Observations 17514 17514 17514 17514 17514 17514 17514 17514 17514
Kleibergen-Paap F-Statistic 38.795 38.795 38.795 38.795 38.795 38.795 38.795
Notes: All models include county and interacted state-year fixed effects. The instrument is lagged military expenditure times the probability
of receiving military aid. Robust standard errors clustered at the state level reported in parenthesis. Asterisks denote: ∗ ∗ ∗p < 0.01,∗ ∗ p <
0.05,∗p < 0.1.
The Effect of Police Militarization on Crime 36
Figure A.1: Trends in US Military Spending and US Military Casualties
