ArticlePDF Available


Content may be subject to copyright.
Police Violence and the Underreporting of Crime
Daniel W. Gingerich
Virginia Oliveros
University of Virginia, Charlottesville,
Tulane University, New Orleans, LA,
Daniel W. Gingerich, Department of
Politics, University of Virginia,
Charlottesville, VA, USA.
This paper examines the relationship between police vio-
lence and the reporting of crime. Utilizing original data
from a large-scale household survey conducted in Costa
Rica from October 2013 to April 2014 (n=4,200), we find
that the observation of police violence significantly reduces
citizenswillingness to report crime. The implications of
this finding are explored using a game-theoretic model of
crime, crime reporting, and police misconduct. The model
reveals that although the prospect of police violence against
criminals may generate a degree of deterrence for criminal
behavior, permissiveness toward police violence also raises
expectations about the likelihood of police abuse against
law-abiding citizens. Consistent with our empirics, this
reduces citizenspropensity to report crime, thereby foster-
ing a climate of impunity for criminal activity.
A requirement for the prosperity and well-being of any society is that it contains policing and judi-
cial institutions willing and able to protect the property and physical security of its citizens.
Throughout the developing world, many governments are failing to meet this minimal obligation.
This appears to be especially true for a growing number of areas in Latin America and the Carib-
bean. Overall, the region is the most violent in the world. According to recent estimates, the homi-
cide rate in this region is 21.8 per 100,000 inhabitants. With 9% of the world population, the
region registers 33% of the world homicides; followed by Africa (31%) and Asia (28%); and trailed
by large margins by Europe (5%) and North America (3%) (Jaitman, 2015; based on UNODC
data). Moreover, the region is the only one in which homicides have been increasing since 2005
(Jaitman, 2015). And homicides are just a small part of the story. Although reliable data are diffi-
cult to find, robberies have also increased significantly in the last decade and, on average, 6 out of
10 of those robberies are violent (PNUD 2013, cited in Jaitman, 2015). According to recent
Received: 14 December 2016
Revised: 28 July 2017
Accepted: 20 September 2017
DOI: 10.1111/ecpo.12102
©2017 John Wiley & Sons Ltd Econ Polit. 2018;30:78105.
victimization surveys, nearly one in five citizens in Latin American and the Caribbean is a victim
of some type of crime in a given year (LAPOP 2012). Of these, less than half claim to have
reported the crime they experienced to authorities.
A consideration of the extant evidence on the social and economic consequences of crime lays
bare the enormous costs of violence in the region. For the case of rural Brazil, Koppensteiner and
Manacorda (2013) show that exposure to violence during pregnancy leads to an increased fre-
quency of low birthweight births. Exploiting a natural experiment in Peru, Ag
uero (2013) finds
that instances of domestic violence against women in that country have produced a variety of neg-
ative short-term health outcomes in children. A recent study based upon microdata in Brazilian
cities estimates that increasing the sense of security of homeowners by one standard deviation (on
a survey measure) would increase average home values by US$757, or by about US$13.6 billion
total in their sample (Vetter, Beltr~
ao, & Massena, 2013). A similar study conducted in Bogot"
Colombia, finds that households in the highest socioeconomic stratum are willing to pay up to 7%
of their house values to avoid a one standard deviation increase in the homicide rate (Gaviria,
Medina, Morales, & N"
nez, 2010). Such attempts at crime insulation naturally generate important
distributional effects: recent work on Argentina finds that the poor are unable to use private sector
remedies to mitigate their exposure to crime, so a disproportional amount of the economic suffer-
ing during crime waves is borne by them (Di Tella, Galiani, & Schargrodsky, 2010).
Latin Americans seem to be fully aware of the seriousness of the situation. Indeed, crime and
public security is identified as the most important problem in the region as a whole (followed by
unemployment and economic issues), and in 13 out of 18 countries (Argentina, Bolivia, Chile, the
Dominican Republic, El Salvador, Guatemala, Honduras, Mexico, Panama, Paraguay, Peru, Uru-
guay, and Venezuela) (Latinobar"
ometro, 2015). However, in spite of the very high premium placed
on security in the region, instances of direct citizen collaboration with police to reduce crime are
rare. The most basic, and in many respects most fundamental, form of collaboration with police
would be if crime victims consistently reported their experiences to the police. Unfortunately,
many victims and observers of crime are reticent to report illegal activities to the police. As a con-
sequence, crimes reported to police are a fraction of those that are actually experienced by citizens.
Comparing data obtained from victimization surveys and from official records provides a sense
of scope of the problem of crime non-reporting. Soares (2004b) compares data from these two
sources for three types of crimes: thefts, burglaries, and contact crimes (such as robberies, sexual
incidents, and threats/assaults) and the differences are striking. According to official records, vic-
timization rates were 2.1% for thefts, 0.7% for burglaries, and 0.3 for contact crimes. According to
the victimization survey, however, the rates were 25.1% for thefts, 6.7% for burglaries, and 7.7 for
contact crimes a remarkable contrast.
In Costa Rica, according to data from the National Household survey, only 29.7% of the crimes
committed in 2014 were reported (INEC and PNUD 2015). Our own data from a survey we con-
ducted in Costa Rica shows that 24% (994/4,200) of the respondents or members of their family suf-
fered a theft or attempted theft taking place in the street, their house, or car during the previous year,
but only 43% reported the incident to the police. The number of victims of robbery in the home was
9%; of those only 42% reported the crime to the police. In terms of witnessing criminal activities,
38% of our respondents reported having witnessed drug trafficking, but only 24% of those individuals
reported it to the police. In sum, we are facing a puzzling situation. Crime and violence are among the
The sources in Soares (2004b) are United Nations Survey of Crime Trends and Operations of Criminal Justice Systems
(UNCS) for the official records, and the International Crime Victimization Survey (ICVS) for self-reported victimization.
The dataset contains data from 46 countries, with measurements taken in the 1990s.
most serious problems Latin American countries are facing today, and citizens seem to be fully aware
of this fact. Yet, the willingness of citizens to themselves collaborate with state authorities by report-
ing criminal activities does not match the level of concern with the issue.
In this paper, we explore the conditions that affect such collaboration. More specifically, we
study the impact of one potentially important inhibitor of crime reporting: the observation of police
brutality. We address the relationship between police brutality and crime reporting at both an
empirical and theoretical-level.
Empirically, we examine the link between direct observation of police violence and propensity
to report crime through the analysis of original survey data from a large-scale household survey
conducted in the Gran
Area Metropolitana (GAM) of Costa Rica from October 2013 to April 2014
(n=4,200). This survey is unique in its simultaneous focus on citizensprior observations of
police violence and its extensive battery of questions on willingness to report crime. Holding con-
stant individual demographics, past experiences, social networks, and neighborhood characteristics,
the data show that citizens who directly witness police violence are far less likely to report a wide
variety of crimes both when they are victims of crime and when they are witnesses of crime.
Given that the institutional features of Costa Rica (discussed below) would tend to weigh against
such a finding, we surmise that this relationship is likely to hold for other countries in the region.
Subsequent to establishing the empirical link between police violence and the non-reporting of
crime using the Costa Rica data, the paper develops a novel theoretical framework that traces out
the implications of our findings for attempts to combat crime. The framework, which consists of a
game-theoretic model of crime, crime reporting, and police misconduct, homes in on the question
of how institutional arrangements that give police wide latitude to act violently against suspected
criminals are likely to affect the incidence of crime in the long run. Our model incorporates the
possibility that such latitude may act as a deterrent to criminal activity, yet it nevertheless reveals
that at the limit high levels of latitude are likely to encourage crime. The key mechanism, fol-
lowing the empirical results, is the non-reporting of crime. Institutionalized permissiveness toward
police violence raises expectations about the likelihood of police abuse against law-abiding citi-
zens. This inhibits crime reporting by victims, thereby fostering a climate of impunity.
Given the widespread consensus about the seriousness of the problem of crime in Latin America,
why do so many victims and witnesses of crime fail to report it when it occurs? Most of what we
know today about the determinants of crime reporting is based on research conducted in the devel-
oped world. With few exceptions, relatively little has been done to study this issue in middle income
and poor countries (cf. Estienne & Morabito, 2016). However, the available evidence suggests that
the determinants of crime reporting across developed and developing countries might not be the same.
National context seems to be crucial in explaining reporting since huge disparities exist in crime
reporting rates across the globe (Estienne & Morabito, 2016; Soares, 2004a,b). For instance, accord-
ing to the International Crime Victims Survey (ICVS) data, the self-reported rate of victimization for
theft in Denmark is 11%; while the rate of theft reported to police is 8%, so around 28% of thefts go
unreported. In contrast, Georgia has a rate of self-reporting of thefts of 18.9% and a rate of police-
reported thefts of 5.5%, so the rate of unreported thefts is approximately 71%.
One can find even
more extreme cases. According to a comparison of victimization surveys and administrative records
Calculations based on data reported in Van Wilsem (2004).
in Peru and El Salvador, the rate of non-reporting for theft in those countries is estimated to be a
remarkable 99.1% and 99.3%, respectively (UNDP, 2013, pp. 3-4).
Following most of the literature, we understand the decision to report as a result of a cost-bene-
fit analysis conducted by crime victims: individuals report crimes because they expect to receive
some benefit from doing so. Such benefits could be material (reporting in order to be able to claim
compensation from an insurance company) or psychological (reporting because it fulfills ones per-
ceived civic duty or generates satisfaction from seeing the perpetrator be punished). When individ-
uals decide to report a crime, they consider possible outcomes from reporting and evaluate the
costs and benefits of those different outcomes. In terms of the former, the cost of reporting a crime
depends mainly on the access to police and the judicial system. In terms of the benefits, these basi-
cally depend on the efficiency and trustworthiness of the police and other institutions in charge of
enforcing the law (Soares, 2004a). For instance, a citizen is more likely to report a crime if the
police station is closer to her home (cost is low) and she believes that the crime is likely to be
solved (the potential material or psychological benefit is high).
Costs and benefits of reporting, in turn, are affected by individual and context-level characteristics
that have an effect on rates of reporting across individuals, across crimes, and across countries. The
literature so far has identified four different types of determinants of crime reporting: the specific
characteristics of the crime, victim characteristics, national context, and police-related variables. First,
the characteristics of the crime are key. In this regard, the most important factor is the seriousness of
the crime. Whether the crime was only attempted or actually consummated, the degree of material or
financial loss, the use of a weapon or violence, and the extent of injuries (if any), are all factors that
affect the likelihood that a crime is reported to the authorities (Bowles, Reyes, & Garoupa, 2009; Esti-
enne & Morabito, 2016; Gottfredson & Hindelang, 1979; Goudriaan, Lynch, & Nieuwbeerta, 2004;
Skogan, 1984; Tarling & Morris, 2010). As noted by Estienne and Morabito (2016), the more serious
the crime, the higher the benefits of reporting, such as the benefit of seeing the perpetrator punished
and the belief that the crime will be successfully pursued within the system.
Second, victim characteristics can also explain differences in reporting. For instance, women
(Tarling & Morris, 2010) and older victims (Bosick, Rennison, Gover, & Dodge, 2012; Skogan,
1984; Tarling & Morris, 2010) are more likely to report crimes. Third, national contextual variables
have also been associated with crime reporting. The most significant of these variables is economic
development (usually measured as income per capita). Since development generally increases the
benefits of reporting and reduces its cost, richer countries tend to have significantly higher levels of
reporting (Soares, 2004a,b). Greater development lowers costs of reporting because it leads to more
police presence, more police stations, better transportation to access these police stations, higher
levels of urbanization, and so on. It increases the benefits because wealthier countries typically have
more professional and less corrupt police departments and judicial systems.
This leads to the fourth main group of determinants of crime reporting: citizen perceptions of the
police. A contingent of scholars examining diverse populations around the world has presented evi-
dence showing that perceived police competence, effectiveness, and fairness have a positive effect on
the likelihood of crime reporting. Utilizing ICVS data for 16 advanced industrial democracies,
Goudriaan et al. (2004) find that the perceived competence of the police plays an important role in
the reporting of property crimes. Xie, Pogarsky, Lynch, and McDowall (2006) examine data from the
National Crime Victimization Survey in the United States and find that the professionalism of police
in handling crime reports, in particular, the effort they dedicate to investigating the crime, positively
affects the likelihood of subsequent crime reporting by victims. In a study of in-person interviews of
citizens conducted in Trinidad and Tobago, Kochel, Parks, and Mastrofski (2013) report that deci-
sions to report crime appear to be driven by perceptions of the legitimacy of the police.
Just as police professionalism appears to encourage crime reporting, existing evidence suggests
that malfeasance by the police may reduce it. Along these lines, survey work conducted in Ghana
by Justice Tankebe suggests that the perceptions of police efficacy play an important role in crime
reporting, and that knowledge of incidents of police malfeasance erodes the said perceptions
(Tankebe, 2009, 2010). Similarly, in a cross-national study of 33 countries, Soares (2004b) finds
that perceptions of corruption are a significant factor in explaining non-reporting for theft, burglar-
ies, and contact crimes.
Studies on the specific impact of police violence remain relatively scarce, although this appears
to be a topic of increasing scholarly concern. Most work on the issue provides evidence consistent
with the notion that police violence may dampen collaboration with the police, including crime
reporting. In interviews conducted with high-risk youth in Philadelphia, Carr, Napolitano, and
Keating (2007) report that crime non-reporting among their subjects appeared to have been driven
by adverse interactions with the police, including the excessive use of force. In a recent study of
residents of Lagos, Nigeria, Akinlabi (2016) finds that both the experience and perception of police
abuse and brutality are associated with cynicism toward and non-compliance with the law.
Event studies examining the impact of widely publicized incidents of police violence in the
United States also suggest that brutality may erode collaboration with the police. For instance,
Weitzer (2002) utilizes panel surveys to examine attitudes toward the police before and after high-
profile incidents of police violence in Los Angeles and New York, finding that such incidents
erode confidence in the police. Desmond, Papachristos, and Kirk (2016) employ 911 call data by
city blocks and an interrupted time series design to examine how the public dissemination of infor-
mation about the beating of an unarmed Black man by police affected crime reporting in the city
of Milwaukee. They find a significant reduction in 911 calls from Black neighborhoods in the
weeks following the dissemination of the incident.
Our paper builds upon and extends this literature in three important ways. Firstly, our paper
provides the first examination of police misconduct on crime reporting in Latin America, as well
as the first large-scale study of this topic in a developing country context that draws upon the tools
of modern program evaluation. Given Latin Americas status as the worlds most violent region
and its exceptional challenges in encouraging crime reporting, in-depth examinations of the factors
inhibiting such reporting within the region are sorely needed. Moreover, the large scale of the sur-
vey we conducted in Costa Rica permits us to match observers of police violence to non-observers
in a fine-grained and highly exact way, thereby buttressing the credibility of our conclusion that
police violence inhibits the willingness to report crime. A second contribution of the paper is the
fact that our empirical analysis is based upon an assessment of willingness to report across a wide
variety of different types of crime. Given the extensive evidence suggesting that the nature of the
crime plays an important role in crime reporting, the fact that we gauge the inclination to report
across various types of crimes imbues our analysis with a degree of robustness that many studies
lack. Finally, our paper extends the existing literature by developing the first general theoretical
model of the interplay between institutional permissiveness toward police violence, the reporting
of crime, and decisions by potential criminals to engage in criminal activity.
We begin our investigation with an empirical analysis of how personal observation of incidents of
police violence affect individualswillingness to report crime, focusing on the Central American
country of Costa Rica. The country is a particularly apt unit of analysis, since in many ways it is a
least likelycase for a finding that police violence erodes citizen collaboration with police. Crime
and police brutality are low by regional standards and democratic institutions and the rule of law
are strong. According to the estimates of Jaitman (2015, based on UNODC data) mentioned above,
the homicide rate in Costa Rica is 8.5 per 100,000, whereas the regional average is 20 per
100,000 inhabitants.
In terms of the frequency of the most commonly perpetrated street crimes,
Costa Rica is more representative of the region, with 14.3% of Costa Ricans claiming to have been
victims of burglary or theft, and 11.3 of contact crimes (robberies, sexual incidents, and assaults)
(Soares & y Naritomi, 2010). Yet, these numbers are only slightly worse than what one finds in
industrialized nations such as the United States, where 8.4% have been victims of burglary, 9.5 of
thefts, and 9.6 of contact crimes (ibid.).
In terms of its institutions, Costa Rica is among the best performers in the region. An uninter-
rupted democracy since 1949, the country has one of the oldest and strongest democratic systems
in Latin America. According to the Freedom House measure of democracy, Costa Rica scores 90/
100, with only Uruguay (98/100) and Chile (95/100) getting better scores. This institutional
strength appears to generally apply to the countrys police forces as well. A recent study on police
abuse in Latin America found that the rate of victimization at the hands of the police was the
fourth lowest in the region, with 2.8% of Costa Ricans experiencing police abuse in a given year
(Cruz, 2009). Given these characteristics of the Costa Rican case, there is reason to believe that a
finding of a strong link between witnessing police violence and the non-reporting of crime in this
country would extend to other countries in the region where concerns about police abuse are likely
to be even more salient.
To assess the effect of police violence on crime reporting, we conducted a face-to-face household
survey during October 2013April 2014 in the Gran
Area Metropolitana (GAM) of Costa Rica.
The GAM is the principal urban center in the country and it includes 30 cantons in the provinces
of San Jos"
e, Heredia, Cartago, and Alajuela. With approximately 2.6 million residents, it contains
60% of the total population. The survey targeted residents of this area 18 years or older. The sam-
ple size of the survey was 4,200 respondents. To sample the population, we used a two-stage strat-
ified random sampling design (with fixed proportions defined for gender and age groupings).
Additional details about the survey methodology are provided in the Appendix.
By comparison, the highest rates in the region are found in Honduras (90.4) and Venezuela (53.7), while the lowest are in
Cuba (4.2) and Chile (3.1). The UNODC data cited by Jaitman (2015) is from 2012. A different local source reports a
homicide rate in Costa Rica of 8.7 per 100,000 inhabitants for that same year, and a rate of 10.0 per 100,000 for the year
2014 (Poder Judicial de Costa Rica 2016).
Our characterization of Costa Rica as a hard case for our theory is based on the assumption that in a low crime/good insti-
tutions environment, citizens who witness police brutality would be more inclined to believe that it was a rare occurrence
and therefore less likely to change their priors. In contrast, it could be argued (and we thank one of the anonymous review-
ers for calling our attention to this) that precisely in a low crime/good institutions environment, where citizens priors about
police violence are good,it might be easier to move those priors when witnessing an instance of police violence. In that
case, Costa Rica would be an easy case, instead of a hard one as we argue. In our empirical analysis (see section 3.3), we
find that the negative effect of police violence on crime reporting is present among respondents in both low and high crime
areas, suggesting that our findings may be indeed be relevant to other countries with more/less crime than Costa Rica. In
the end, whether our results travel to other settings with more crime and worse institutions is an empirical question that
requires future research for a definitive answer.
Our independent variable of interest is an indicator variable denoting a citizens direct observa-
tion of police violence. This variable takes on a value of 1 for individuals who reported that over
the previous 12 months they had observed police violence being committed against citizens, 0
otherwise. Approximately 11% of the respondents in our sample, 468 in total, had observed police
violence according to this standard.
Our dependent variables consist of a series of responses to
seven questions about willingness to report crime. These questions present scenarios where the
respondent is the victim of a crime as well as scenarios where the respondent witnesses crimes
being committed. The incidents described in the questions were selected to reflect the most com-
mon types of crime that citizens in Costa Rica grapple with in their daily lives: robbery of valu-
ables on the street, theft of valuables from ones home, drug sales, gang activity, and gun shots.
Responses were coded on a four-point scale (highly likely, likely, unlikely, highly unlikely) indi-
cating the respondents stated likelihood of reporting the incident in question to the authorities.
Table 1 presents the wording for these questions.
In our statistical analysis, we condition on an extensive set of covariates that may affect an
individuals likelihood of observing police violence and that could also be related to her inclination
to collaborate with authorities. In particular, we condition on four types of different covariates:
individual demographic variables, indicators of relevant past experiences, indicators of social net-
works that may be relevant to crime and crime reporting, and variables capturing neighborhood
In terms of individual demographics, we include the respondentssex,age,levelofeducation
(primary or less, secondary incomplete, secondary complete, some technical education, some
university education), citizenship status, an indicator variable for head of household, and several
indicators of material wealth (ownership of a car, laptop, tablet, internet connection in the
home). Our experience variables include an indicator variable denoting direct contact with a
TABLE 1 Questions on willingness to report crime
I am going to present to you a list of hypothetical situations. They deal with different types of criminal activities.
For each situation, please tell me how likely it would be that you would report the incident in question to the
police (either as a victim or a witness)
Ql. Imagine that you are walking on the street and someone snatches your purse or wallet. How likely is it that
you would report this incident to the authorities?
Q2. Consider the same situation, but now the person who snatches your wallet threatens you with a weapon (a
gun, knife, or any other). How likely is it that you would report this incident to the authorities?
Q3. Now imagine that thieves break into your house to rob it. How likely is it that you would report this incident
to the authorities?
Q4. How inclined would you be to call the police to report a crime or suspicious activities in your neighborhood?
Q5. More specifically, if said crime consisted of someone selling drugs in the street, how inclined would you be
to call the police?
Q6. And if the crime consisted of gangs or groups disturbing the peace, how inclined would you be to call the
Q7. And if you were in your home and heard gun shots, how inclined would you be to call the police?
Policing in Costa Rica is divided among a number of units, with the three units most relevant to the everyday experiences
of Costa Ricans being the Fuerza P"
ublica (a national police force), the Transit Police, and the municipal police. As our
question about police violence did not specify the particular unit of the perpetrator, the observed police violence reported by
respondents should not be attributed to any specific police force.
police officer in the previous 12 months and an indicator variable denoting whether or not the
respondent had been the victim of a crime in the previous 12 months.
The network variables
include an indicator variable denoting whether or not the respondent personally knows a police
officer and an indicator variable denoting whether or not the respondent knows someone
indicted or convicted of a crime. Our neighborhood indicators are recorded at the level of the
district. In Costa Rica, these are postal code areas and they are the smallest administrative units
for which census data are available. Our district-level variables, measured in the 2011 census,
include the percentage of households for which basic necessities are being unmet, population
density, percentage of individuals with higher education, percentage of households with a laptop,
percentage of households with a car, percentage of households with internet access, and percent-
age of homes deemed to be in poor condition.
Identification and estimation
Our estimand of interest is the average treatment effect for the treated (ATT), the average impact
of witnessing police violence for those who observed it. In estimating this quantity, we employ the
so-called exogeneity assumption (cf. Imbens, 2004). The assumption requires that the analyst be
able to observe and condition on all factors that contribute both to the likelihood that an individual
observes police violence and to her willingness to report crime. In our study, exogeneity implies
that conditional on an individuals demographic and socioeconomic background, personal experi-
ences, social networks, and the characteristics of her neighborhood, whether or not she actually
observes police violence in a given year is effectively random.
Although we acknowledge that exogeneity is a strict (and unverifiable) assumption that is
difficult to satisfy in any observational setting, including ours, we would submit that the rich
covariate set that we bring to bear in the analysis makes it a plausible basis for our analysis.
The demographic characteristics utilized capture personal aspects of respondents potentially
associated with the likelihood of witnessing police violence and which very likely would affect
respondentscomfort level in reaching out to authorities after observing or being a victim of
crime. The experience variables measure past instances of police contact or opportunities for
said contact, thereby likely affecting the observation of police violence. The network variables
represent aspects of familial and/or community relations which are likely to color views of
police and the judicial system, and, through that channel, affect reporting behavior. Finally, the
district variables capture aspects of a respondentsneighborhoodlikelytobeassociatedwith
police activity and presence, elements crucial to the witnessing of police violence. While it is
certainly possible that there remain unmeasured features of respondents correlated with both the
likelihood of witnessing police violence and the propensity to report crime, it seems highly
likely that these unmeasured features would be fairly strongly correlated with the features we do
observe and condition on. Thus, we would expect that any violation of exogeneity if present
is likely to be fairly minor.
Given the exogeneity assumption, estimating the ATT requires the implementation of a covari-
ate conditioning strategy. We condition on the covariates using (one-to-one) nearest neighbor
matching based on an estimated propensity score. This algorithm contains two steps. In the first, a
An individual was classified as a crime victim if she or a family member had experienced one of the following: attempted
robbery in the street, home, or of ones car, robbery of valuables in ones home, robbery in the street, robbery of ones car
or of valuables contained therein.
These data come from the website of Costa Ricas National Institute for Statistics and the Census (INEC),
logistic regression model is estimated that captures the propensity to observe police violence given
the respondent and district characteristics captured by the set of covariates. In the second step, each
respondent who observed police violence is matched to the single respondent not having observed
violence who is most similar to the first in terms of the estimated probability of having observed
violence based on the covariates. After creating a matched subset of the data in this way, we calcu-
late the average differences in willingness to report crime across the matched pairs.
More specifi-
cally, given that we have four categories for each outcome variable, we calculate the proportion of
responses in each category for the respondents who observed police violence, then subtract from
this quantity the proportion of responses in the corresponding category among the matched respon-
dents who did not observe police violence.
We conduct the analysis in this way for two reasons. Presuming the exogeneity assumption is
satisfied, the use of matching permits one to unbiasedly estimate the average treatment effect of
interest without having to correctly specify the functional form relationship between the outcome
and the covariates. Moreover, analyzing outcome categories separately similarly obviates concerns
about functional form misspecification while also providing greater transparency in the data analy-
sis (allowing the analyst to detect differences in the degree to which each category of response is
affected by police violence). In any case, we supplement our matching analysis with estimations
from regression-based approaches, generating findings that are highly similar in terms of the sub-
stantive conclusions that they support.
Main results
Table 2 presents descriptive statistics on the unconditional relationships between respondentspre-
vious observations of police violence and their willingness to report crime. Across every type of
crime, individuals who had observed an instance of police violence in the previous year were less
inclined to report crime than individuals who had not observed an instance of violence. Thus, a
first look at the data is indeed consistent with the view that aggressive and violent policing may
lead citizens to be more reluctant to collaborate with authorities on matters of crime enforcement.
Of course, individuals who have observed police violence may be fundamentally different from
those who have not in terms of personal characteristics, experiences, and environment in ways that
could be associated with a disinclination to report crime. For this reason, we base our inferences
about the impact of observing police violence upon our matching analysis that holds constant the
potentially confounding influence of the respondentscharacteristics and those of their
Matching can only be considered successful in holding constant the influence of covariates if it cre-
ates balance in the distribution of covariates across the groups of primary interest in a study. To verify
that our matching algorithm was indeed successful in creating balance in covariate values between the
individuals who observed police violence and the matched individuals who did not, we calculated
Our matching analysis utilizes the package Matching (Sekhon, 2015), written for the R statistical environment. Since miss-
ingness in our sample due to question non-response is very low (around 3%), we conduct complete case analyses through-
out. Also, since we analyze each outcome variable separately and only remove observations with missingness pertinent to
the particular outcome being studied, there are very slight variations in the set of observations utilized in the complete case
analyses across outcomes.
Unfortunately, we did not have enough observations by district to conduct matching within districts. (In our sample,
approximately one third of respondents are in districts where the number of respondents is less than 30.) To deal with dis-
trict effects, we incorporated district characteristics directly as covariates with which to match our respondents on (see
Table 3).
standardized differences in covariate means between these groups in the full sample (pre-matching) as
well as our matched sample. The results of this exercise are presented in Table 3.
An often employed rule-of-thumb is that a standardized difference of greater than 0.25 indicates
serious imbalance in a covariate. According to this standard, there were several serious imbalances
in the full sample. In particular, men were more likely than women to observe police violence,
younger individuals were more likely to observe violence than older individuals, individuals who
personally knew a police officer were more likely to observe violence than those who did not, indi-
viduals who personally knew someone prosecuted by the justice system were more likely to observe
violence than those who did not, individuals who had had direct contact with the police were more
likely to observe violence than those without direct contact, and those who had been victims of
crime were more likely to observe violence than non-crime victims. (The interested reader is referred
to Table A3 in the Appendix, which displays the results from the propensity score estimating equa-
tion and corroborates the present discussion about imbalance on the aforementioned characteristics
in the full sample). Fortunately, the nearest neighbor matching algorithm was able to select a
matched sample among the respondents not observing police violence that had very similar back-
ground characteristics to those respondents who did observe violence. As the table shows, after
employing the matching algorithm, all covariate means were balanced across the two groups.
TABLE 2 Descriptive statistics
How likely is it that you would call the police in the case of [event]?
Observed police violence Did not observe police violence
H. Likely Likely Unlikely H. Unlikely H. Likely Likely Unlikely H.Unlikely
Robbery w/o weapon 0.21 0.22 0.29 0.28 0.25 0.30 0.25 0.19
(0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01)
Robbery w/ weapon 0.34 0.25 0.23 0.18 0.37 0.30 0.21 0.12
(0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01) (0.01)
Robbery of home 0.54 0.27 0.12 0.07 0.56 0.30 0.09 0.05
(0.02) (0.02) (0.01) (0.01) (0.01) (0.01) (0.00) (0.00)
Suspicious activity 0.33 0.28 0.24 0.15 0.35 0.42 0.16 0.08
(0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01) (0.00)
Drug sales 0.27 0.24 0.23 0.26 0.29 0.35 0.21 0.15
(0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01)
Gang activity 0.32 0.29 0.21 0.18 0.31 0.41 0.17 0.10
(0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01) (0.01)
Gun shots 0.30 0.28 0.20 0.22 0.32 0.36 0.18 0.14
(0.02) (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01)
n=468 n=3,722
Note: Standard errors in parentheses.
Shown in the table are standardized differences calculated using a complete case sample where the dependent variable
was the response to Q1 in Table 1. The results obtained using complete case samples based on the other dependent vari-
ables are nearly identical (see footnote 8). These are available upon request. Simple differences in covariate means are dis-
played in Table A2 of the Appendix.
Figure 1 presents the results of our matching analysis. The dot plots contained in the figure dis-
play the point estimates (circles) and 95% confidence intervals (bolded lines) of the ATT associ-
ated with witnessing police violence for each response category for each hypothetical scenario
presented in Table 1. In other words, these plots show how witnessing police violence affects the
probability of each response. There is a strong degree of consistency in our findings. For all of the
scenarios of experiencing and witnessing crime, observing police violence made respondents more
reticent to report crime. Generally speaking, likely to reportand highly unlikely to reportwere
the categories most strongly affected by police violence. Observing police violence had a negative
and statistically significant impact on the probability of being likely to reportfor all types of
crime, save for robbery with a weapon and a home robbery. Similarly, observing police violence
had a positive and statistically significant impact on the probability of being highly unlikely to
reportfor all types of crime, again save for robbery with a weapon and a home robbery (for
which the effects were at the threshold of statistical significance).
TABLE 3 Standardized differences in covariate means (before and after matching)
Covariate Before matching After matching
Male 0.30 0.04
Age !0.46 !0.02
Education: Primary school or less !0.09 0.04
Secondary school incomplete 0.15 0.00
Secondary school complete 0.08 !0.01
Some technical education !0.06 0.02
Costa Rican national 0.10 !0.04
Head of household 0.01 !0.01
Material wealth Laptop 0.05 0.01
Tablet 0.14 !0.01
Car !0.02 !0.04
Internet connection in home 0.05 !0.01
Knows police officer 0.29 !0.00
Knows someone prosecuted 0.49 0.01
Direct contact with police 0.48 !0.04
Crime victim 0.26 !0.03
District characteristics Unmet basic necessities (% households) 0.07 0.08
Population density 0.05 !0.01
Higher education (% population) !0.12 !0.09
Laptop (% households) !0.11 !0.09
Car (% households) !0.10 !0.08
Internet connection (% households) !0.09 !0.08
Homes in poor condition (% households) 0.06 0.06
The relative inelasticity of reporting to observed police violence in the case of a home robbery could very well be due to an
insurance effect,that is, citizens whose homes are insured can obtain compensation for the valuables stolen only if they for-
mally file a report with authorities. Thus, these individuals would be likely to report such a robbery irrespective of their expo-
sure to police violence. The high level of stated willingness to report for this crime is consistent with such an interpretation.
The magnitude of the effects were substantial. For instance, in the case of robbery of ones valu-
ables in the street (without a weapon present), having previously observed police violence reduced
the probability of the likely to reportresponse by 8 percentage points and increased the probability
of the highly unlikely to reportresponse by 9 percentage points. In the case of observing suspicious
activity, the estimated effect was a reduction of the probability of the likely to reportresponse of 12
percentage points and an increase in the probability of the highly unlikely to reportresponse of 8
percentage points. For witnessing drug sales, the effect was a reduction of the probability of the
likely to reportresponse of 9 percentage points and an increase of the probability of the highly
FIGURE 1 Impact of police violence on willingness to report crime.
Notes: The ATT for the observation of police violence for each response category for each crime reporting outcome
is arrayed on the x-axes in the panels in the figure. The estimated ATT is the difference in the probability of a
given response category for treated units as compared to matched control units. Point estimates are denoted by black
dots. Ninety-five percent confidence intervals are denoted by horizontal black lines.
unlikely to reportresponse of 8 percentage points. For the observation of gang activity, the estimated
effect was a reduction of the probability of the likely to reportresponse of 10 percentage points and
an increase in the probability of the highly unlikely to reportresponse of 8 percentage points.
Finally, in the case of hearing gun shots, having previously observed police violence reduced the
probability of the likely to reportresponse by 9 percentage points and increased the probability of
the highly unlikely to reportresponse by 10 percentage points.
How robust are these findings? To address this question, we began by evaluating the sensitivity of
our causal effect estimates to the precise specification of our covariate set. In particular, we re-esti-
mated our ATT estimates for one particular outcome robbery in the street without a weapon using
a large number of randomly selected subsets of the full covariate set employed in the analysis above.
The algorithm had three steps. First, we randomly selected a subset of covariates from the full
set, with each covariate assigned a probability of 0.5 for inclusion in the subset. Second, using the
randomly selected subset, we estimated the ATT using the nearest neighbor matching procedure.
Finally, we repeated this process 10,000 times and graphed the resulting histograms (one for each
category of the dependent variable). Figure 2 presents the histograms.
The key point to notice about the histograms is that the sign of the effect of police violence on
reporting behavior is highly insensitive to the specification of the covariate subset. For the out-
comes highly likely to report,”“likely to report,and highly unlikely to report,none of the
ATTs estimated using the randomly selected covariate subsets differed in sign from those esti-
mated with the full covariate set. This implies that our findings enjoy a strong degree of covariate
specification robustness, at least relative to the total set of covariates included in this paper. The
fact that the sign of the estimated effect of police violence on reporting behavior is so strongly
invariant to covariate specification gives us added confidence in our finding that witnessing police
violence reduces the inclination to report crime.
We also examined the robustness of our findings to different strategies for modeling the behav-
ior of the dependent variable. Our first step in this regard was to collapse the information con-
tained in our seven reporting outcomes into a single index representing the respondentsreticence
to report crime. To do this in a principled fashion, we utilized an item response theory approach
(cf. Johnson & Albert, 1999). First, using the Graded Response Model for ordinal polytomous
data, we estimated the relationship between the seven reporting outcomes and respondentslatent
propensity against reporting crime. The results of this estimation are presented in Table A4 in the
Appendix. As expected, all reporting outcomes were related to latent propensity in the same way:
all had positive discrimination parameters and all of these were statistically significant by any rea-
sonable standard. Second, we created our index by estimating factor scores from the fitted model.
With the unidimensional index in hand, we then utilized both nearest neighbor matching and ordi-
nary least squares to examine the relationship between the observation of police violence and a
respondents latent propensity against reporting crime.
The results are presented in Tables A5A6 in the Appendix. In both analyses, the observation of
police violence was found to be a powerful and highly statistically significant deterrent to reporting
crime. In the matching analysis, the estimated ATT was 0.25, with a standard error of 0.06
(p<.001). To give a sense of scale, this estimated treatment effect is equal to 0.29 standard devia-
tions of the dependent variable among the control group. The coefficient on police violence using
OLS was quite similar: it was estimated as 0.23, with a standard error of 0.04 (p<.001). Thus, treat-
ing the seven reporting outcomes as observable manifestations of an underlying reticence to report
crime only strengthens our conclusion that observing police violence dampens crime reporting.
Factor scores were estimated using the Empirical Bayes Method.
Finally, we conducted a set of analyses in which we examined each of the seven reporting out-
comes separately, but imposed greater structure than in the analysis displayed in Figure 1. We did
so by modeling the outcomes using both ordinary least squares and ordered probit. The results of
this exercise are presented in Tables A7A8 in the Appendix. The results are entirely consistent
with the findings presented thus far. For every reporting outcome examined using either regression
specification, the coefficient on the observation of police violence is positive (indicating that vio-
lence increases respondentsreticence to report) and statistically significant by any reasonable stan-
dard. Indeed, the p-values on these coefficients were less than .001 for every outcome, save for
reporting a robbery in the home, and even for this outcome for which the baseline predilection
to report is relatively high the p-value was less than .01.
FIGURE 2 Robustness of ATT estimates to specification of covariate set.
Notes: Displayed are the histograms of the ATT estimates for the observation of police violence. Each ATT estimate
is based on the application of nearest neighbor matching to a randomly selected subset of the full covariate set,
where each covariate in the full set has a probability of 0.5 of inclusion in any given subset.
Another issue we considered in our analysis concerns the external validity of our findings. Although
this is an issue which can only be definitively addressed by replicating our study in other countries, one
can get a rough sense of the potential of the findings presented here to extend to other settings by exam-
ining the degree of context specificity of the effect of police violence on crime reporting within our sam-
ple from Costa Rica. Since a key aspect in which Costa Rica differs from its neighbors is the level of
crime, we performed a set of supplementary analyses specifically examining how levels of crime in a
respondents neighborhood shape the relationship between the observation of police violence and crime
reporting. We did this in three steps. First, we calculated the crime victimization rate in each district
contained in our sample by using the responses of our survey respondents to a series of questions about
crime (see footnote 5 for the definition of crime victimization). Second, we divided our sample into two
groups: a group of respondents living in low crime districts (districts with a victimization rate at or
below the median for districts in the sample) and a group of respondents living in high crime districts
(districts with a victimization rate above the median).
Third, we estimated the ATT for observing
police violence using nearest neighbor matching within the two subgroups.
Figures A1A2 in the Appendix display the results of this exercise. The figures show that the
negative effect of police violence on crime reporting is clearly present among respondents in both
the low crime and high crime areas. This suggests that our findings may have relevance for devel-
oped countries where crime rates are generally lower than those (on average) in Costa Rica as well
as other Latin American countries where crime rates tend to be higher. Indeed, if there is any dif-
ference in the effects of police violence across the two groups, it would appear that they are a bit
pronounced in the high crime areas. As such, for countries in the region with higher levels of
crime than Costa Rica, the dampening effect of police violence on crime reporting may very well
be even larger than we have estimated here.
The findings presented thus far leave open the question as to precisely why witnessing police vio-
lence erodes the inclination of individuals to report crime. More specifically, they undercover the
consequence of observing police brutality but are silent about the cognitive mechanisms by which
this consequence emerges.
There are three potential mechanisms that merit evaluation. Firstly, following the literature on pro-
cedural justice, the observation of police violence may erode reporting because witnesses to violence
no longer consider the police to be a legitimate institution that deserves respect (cf. Sunshine & Tyler,
2003). Secondly, the observation of police violence may erode reporting because witnesses to vio-
lence come to view the police as being ineffective in carrying out their duties (Tankebe, 2009).
Finally, and particularly relevant to the context of Latin America, the observation of police violence
may erode reporting because witnesses to violence come to view the police as potentially dangerous
and unbound by the strictures of the law, such that future interactions with the police including
crime reporting could entail a risk of violence or mistreatment by police for those contemplating
such an action. Instances of this type of mistreatment by the police are not hard to find in newspapers
around the world. For instance, in February 2014 a woman in Argentina went to report domestic vio-
lence to a local police station and ended up being beaten by the police.
More recently, an Australian
There were 1,780 respondents in the low crime group (179 observed police violence, 1,595 did not) and 2,420 in the
high crime group (289 observed police violence, 2,127 did not).
ın, Feb 12, 2014. Accessed at: comisaria 0
Hyx8 jJjPXe.html.
woman was fatally shot by a police officer in Minneapolis after she called 911 to report what she
thought was a sexual assault occurring near her house.
In order to assess the relevance of these mechanisms, we reran our nearest neighbor matching
analysis utilizing as our dependent variables a battery of seven attitudinal questions about the
respondents perceptions of Costa Ricas police. To tap into views about legitimacy, respondents
were queried about the extent to which the police inspire confidence. To tap into views about
police efficacy, respondents were queried about the extent to which they view the police as being
well trained and whether or not they respond quickly to the calls of citizens. Finally, in order to
tap into views about the potential risk of abuse or mistreatment at the hands of police, respondents
were queried about the extent to which they view the police as respecting the law, treating people
with respect, engaging in corrupt activity, and being involved in crime. Figure 3 presents the
results of the analysis.
For all questions, observation of police violence led to significantly more negative views of the
police force. In particular, observing police violence decreased the extent to which the police
inspired confidence among respondents, decreased the perception that police were well trained,
decreased the perceived responsiveness of police, decreased the perception that the police respect
the law, decreased the perception that police are respectful in their interactions with citizens,
increased the perception that the police are corrupt, and increased the perception that the police are
involved in criminal activities. In general, the effects of violence on attitudes were very large,
shifting category probabilities by more than 20 percentage points in several cases.
The overall negative impact of observed violence notwithstanding, the largest effects were
recorded for views that reflected the potential riskiness of engaging police officers. In particular,
the attitudinal measures most strongly affected by the observation of police violence were beliefs
about the degree to which police respect the law and the degree to which they treat people with
respect. In the aggregate, observing police violence increased the probability that a respondent
would strongly disagree with the statement that police respect the law by 25 percentage points; it
decreased the probability of agreement with that same statement by 19 percentage points. For
expectations regarding respectful treatment, observing police violence also increased the probability
that a respondent would strongly disagree with the statement that police treat people with respect
by 25 percentage points and it similarly decreased the probability of agreement with that statement
by 19 percentage points. Thus, although the evidence suggests that legitimacy, efficacy, and the
perceived riskiness of engaging police are all effected by witnessing police violence, it appears the
latter category riskiness was the one most strongly impacted by such an experience. In what
follows, we explore how the operation of such a mechanism may affect the relationship between
policing strategies and dynamics of crime.
In this section of the paper, we trace out some of the logical implications of our findings in order
to better understand the relationship between institutional permissiveness toward violent policing
and the prevalence of crime. We do so through the use of a simple game-theoretic model that
The Washington Post, July 14, 2017. Accessed at: mix/wp/2017/07/17/
bride-to-be-called-911-for-help-and-was-fatally-shot-by-a-minneapolis-police- officer/?hpid=hp hp-top-table-main minneapolis-
1150am%3Ahomepage%2Fstory&utm term=.c41686515e10.
concentrates on what we view as the three central facets of the relationship between police vio-
lence and crime reporting: the initial decision to engage in criminal activity, the decision of victims
to report crime, and police treatment of suspected criminals and civilians. In line with the argu-
ments of some proponents of aggressive policing strategies, we incorporate into the model the pro-
spect that police violence against criminals may generate a degree of deterrence for criminal
behavior. At the same time, we also incorporate into our model the empirically derived insight that
institutional tolerance for violent policing may augment the perceived riskiness of engaging the
FIGURE 3 Impact of police violence on attitudes toward police force.
Notes: The ATT for the observation of police violence for each response category for each attitudinal outcome is
arrayed on the x-axes in the panels in the figure. The estimated ATT is the difference in the probability of a given
response category for treated units as compared to matched control units. Point estimates are denoted by black dots.
Ninety-five percent confidence intervals are denoted by horizontal black lines.
police. Reflecting the inherent uncertainties of crime and crime reporting, both insights are woven
into an incomplete information framework in which police officers come in good and bad types,
with neither potential criminals nor crime victims knowing a priori what type of officer they will
Players, timing, and actions
There are three actors in our model: a (potential) criminal (C), a law-abiding citizen (L), and a
police officer (P).
At the beginning of the game, Nature chooses Ps type, denoted by v2{0, 1}. If v=1, then the
officer is prone toward violence. If v=0, then the officer is not prone toward violence. The offi-
cers type is private information: it is known to him but unknown to the criminal or law-abiding
The probability that Pis violence prone is equal to q(r)2[0,1], where qis an increasing
function of rwith q(0) =0 and q(+)=1. The parameter ris the central policy variable in the
model. It represents the permissiveness within police departments, the judiciary, and other state
institutions toward violence committed by officers in the course of their duties. Permissiveness in
this sense may be manifested in promotion policies within police departments that reward officers
involved in police shootings or in patterns of prosecutorial and judicial decision-making that effec-
tively shield officers from punishment for unnecessary use of force (Brinks, 2003, 2008).
The fact that qis increasing in rreflects a selection mechanism for police officers. Formal and
informal policies that tolerate or encourage violence among those in uniform make a career in law
enforcement relatively attractive to individuals with an innate propensity toward violence. As a
consequence, the likelihood that any particular officer will be violence prone will be higher in such
polities than in those where the unnecessary use of force is actively discouraged. This dynamic is
widely understood in the polities in which it operates. In our model, this broad societal understand-
ing is reflected by the fact that qis known by all actors in the model.
In the second stage of the game, Cchooses an action s2{0, 1}, where s=1 denotes that C
steals from Land s=0 denotes that Crefrains from stealing from L. If Cchooses not to steal,
then the status quo (SQ) outcome obtains and utilities are disbursed accordingly.
If Cdoes steal from L, then Lsubsequently must decide whether or not to report the crime to
the police officer, r2{0,1}, where r=1 indicates that Lreports the crime to Pand r=0 indi-
cates that Ldoes not report the crime to P. If Csteals from Lbut Ldoes not report it, then the
crime without punishment (CWP) outcome obtains and utilities are disbursed accordingly.
Penters the game only upon receiving a report of crime from L. If Preceives a report, he is
able to locate and detain Cwith certainty. Thus, Pcan choose to act in one of three different
ways, with his action denoted by a2{1, 2, 3}. First, Pcan act professionally, arresting Cwithout
unnecessary violence and subsequently submitting Cto the judicial system to receive the official
sanction for his crime (a=1). In this case, the service with professionalism (SWP) outcome
obtains. Alternatively, Pcan act with deliberate and unnecessary aggression against Cduring his
arrest, subjecting him to physical violence in addition to the official sanction the judicial system
will mete out for his crime (a=2). This is the targeted violence outcome (TV). Finally, Pcan act
with generalized abusiveness, arresting Cwith unnecessary violence and simultaneously subjecting
Lto hostility and mistreatment during the investigative process (a=3). This is the indiscriminate
violence outcome (IV). The game tree for this model is displayed in Figure 4.
The distinction between targeted vs. indiscriminate violence, while subtle, is crucial to the logic
of the model. The prospect of the indiscriminate violence outcome raises the possibility that
citizens not involved in crime may themselves suffer abuse should they come into contact with
violence-prone police officers. As such, it reflects the notion that institutional tolerance for police
brutality generates negative spillovers that extend well beyond the set of proven or suspected crim-
inals for whom such violence is ostensibly targeted. As will be discussed more fully in what fol-
lows, this implies that law-abiding citizens may in certain circumstances be apprehensive about
reporting crime, even if they have no particular qualms about violence being visited upon those
who have committed crime.
We adopt a quantal response framework, an approach that treats agent choice as stochastic and
which generates smooth comparative statics (Goeree, Holt, & Palfrey, 2016; McKelvey & Pal-
frey, 1995, 1998). Let p
2Πrepresent a decision to be taken by a generic agent jat informa-
tion set h. According to the framework, the utility jassociates with choosing p
=kis equal to
, where e
is a random error representing an idiosyncratic shock to
preferences that is known only to player j. Adopting the nomenclature of Signorino (1999), we
refer to the quantity u
=k) as the direct (true) utility to jof choosing option kat informa-
tion set hand u
=k) as the indirect utility of the same. Agent choice is rational in the sense
that agents choose the option that provides them with the highest direct utility among the options
available at a given information set. However, since each agents direct utility is unobservable to
FIGURE 4 Game tree for the police violence and crime reporting game.
Note: Information sets denoted by dashed lines.
other agents, any agents choices are ex ante probabilistic from the vantage point of all other
players. In particular, when the preference shocks are independently and identically distributed
according to a log Weibull distribution, then the probability that agent jchooses p
=kis writ-
Prðph¼kÞ¼ expðkujðph¼kÞÞ
Pl2Pexpðkujðph¼lÞÞ :(1)
In the random utility framework, the ex ante probability that an agent chooses a particular
action is a function of the indirect utility of that action relative to the indirect utilities associated
with the other actions available to her. The parameter k0, often referred to as a responsiveness
parameter, captures the importance of the relative value of indirect utilities in agent choice. If
k=0, then the probability of each choice is equal, irrespective of the corresponding indirect
utilities. If k=+, then agents always choose the action which gives them the greatest indirect
We consider an elementary preference structure for the indirect utility functions. Police offi-
cers who are prone to violence prefer apprehending suspected criminals with violence than
doing so peacefully. Moreover, the amount of utility they are able to derive from violent polic-
ing is an increasing function of the institutional permissiveness parameter, r. Violence-prone
officers most highly value the indiscriminate violence outcome, as this state of affairs allows
them to maximize their coercive behavior by assaulting suspected criminals and abusing citi-
zens. The preference structure for non-violence-prone officers is the opposite. These individuals
most highly favor the service with professionalism outcome, as it affords them the opportunity
to enforce justice and serve their community with respect. Least favored for these individuals
is the indiscriminate violence outcome, followed by targeted violence against suspected
This discussion leads to the following expressions for Ps indirect utility function:
where h>0 parameterizes the distaste of non-violence-prone officers to discharging their
duties with unnecessary violence against criminals and gis the utility gain (loss) that
violence-prone (non-violence prone) officers associate with abusing citizens. Note that,
according to the above, institutional permissiveness toward police violence only affects the util-
ity of violence-prone police officers; non-violence prone officers find unnecessary violence
against suspected criminals and citizens distasteful irrespective of the level of institutional
All else equal, law-abiding citizens receive a civic, material, and/or psychological return from
reporting crime to authorities when they are victimized. However, the utility associated with report-
ing crime is inherently connected to how they are treated by police in the investigative process. If
a law-abiding citizen is treated well in the process, she receives positive utility equal to l>0
from reporting crime. If she is instead mistreated by police, then the utility from reporting crime is
negative. More specifically, the magnitude of the disutility she receives from police abuse is
increasing in rsince the same types of institutional shields that permit officers to engage in exces-
sive violence against suspected criminals also limit protection against police mistreatment more
This suggests the following indirect utility function for L:
where I(.) is an indicator function equal to 1 if its argument is true, 0 otherwise.
Since the indirect utility to Lfrom choosing to report is a function of the anticipated actions of
Pand these, in turn, are inherently probabilistic, the expression u
(r=1) must be recast as an
expected utility. Applying the expectations operator gives:
The indirect utility function for the criminal follows standard cost-benefit models of crime (cf.
Becker, 1968). If Cabstains from engaging in theft, then he receives utility based on the wage,
->0, that he commands in the (licit) private sector. If he does engage in theft, his utility depends
on the subsequent actions of Land P. If Ldoes not report the theft, then Cgets away with his
crime and receives the value of the resulting loot, s>ϖ. If Ldoes report the crime, then Cs util-
ity depends on the action taken by P. If Pacts with professionalism, then Creceives a return equal
to the official sanction for his crime, !c<0. If Pchooses either of the two actions that entail vio-
lence against C, then Creceives the official sanction as well as the disutility associated with physi-
cal violence, which is an increasing function of institutional permissiveness toward police brutality,
!br. All told, the criminals indirect utility function is:
Here again indirect utility is a function of the anticipated and probabilistic actions of other
actors, namely Land P. Applying the expectations operator to u
(s=1) gives the expected
uCðs¼1Þ¼sPr Iðr¼0Þ!cPr Iðr¼1\a¼1Þ!ðcþbrÞPr Iðr¼1\a6¼1Þ;(7)
Pr Iðr¼1\a¼1Þ¼Prðr¼1Þ½Prða¼1jv¼1ÞqðrÞþPrða¼1jv¼0Þð1!qðrÞÞ'
Pr Iðr¼1\a6¼1Þ¼Prðr¼1Þ½Prða6¼1jv¼1ÞqðrÞþPrða6¼1jv¼0Þð1!qðrÞÞ' (8)
Equilibrium impunity
We consider here how tolerance for police violence against suspected criminals affects the incen-
tives for crime reporting and, ipso facto, overall citizen security. To this end, we derive the equi-
librium probability of the crime without punishment outcome, which represents impunity for
criminal activities, and then proceed to describe how the likelihood of this outcome changes with
greater permissiveness toward police violence.
The probability of crime without punishment is simply the product of the probability that C
chooses to steal and the probability that Lchooses not to report the crime:
Given the assumed distribution of the random choice errors, the probability that Lfails to report
is equal to:
Prðr¼0Þ¼ 1
1þexpðkuLðr¼1ÞÞ ;(10)
uLðr¼1Þ¼l½ð1!2 Prða¼3ÞÞ ! rPrða¼3Þ':(11)
Intuitively, these expressions reveal that the probability of crime reporting is inversely related
to Ls assessment of the likelihood of being mistreated by the police.
The probability that Cchooses to steal is equal to:
Prðs¼1Þ¼ expðkuCðs¼1ÞÞ
expðkuCðs¼1ÞÞ þ expðk-Þ;(12)
The above demonstrates that the probability of crime is inversely related to Cs assessment of
the likelihood that Lwould report crime. It is also inversely related to Cs assessment of the like-
lihood that Pwould act with targeted or generalized violence, since these outcomes impose special
hardship on C.
In order to complete the characterization of the probability of crime without punishment, one
must derive the probabilities that Pacts with professionalism or generalized violence. Note that for
any action k2{1, 2, 3}taken by P, one can write
In words, the probability that Ptakes a particular action is equal to the sum, across Ps two
potential types, of the probability of the action given a specific type multiplied by the probability
that Pis of that type. The relevant conditional probabilities, in turn, are equal to:
Prða¼1jv¼0Þ¼ 1
Prða¼1jv¼1Þ¼ 1
Prða¼3jv¼0Þ¼ expð!kðhþgÞÞ
Prða¼3jv¼1Þ¼ expðkðrþgÞÞ
1þexpðkrÞþexpðkðrþgÞÞ :
These expressions demonstrate four important points: (1) a non-violence-prone officer is more
likely to act professionally (not use unnecessary violence at all) than a violence-prone officer; (2) a
non-violence-prone officer is less likely to abuse a law-abiding citizen than a violence-prone offi-
cer; (3) the probability that a violence-prone officer acts professionally is decreasing in institutional
tolerance for violent policing; and (4) the probability that a violence-prone officer abuses a law-
abiding citizen is increasing in institutional tolerance for violent policing.
Taken together, equations 915 provide a complete characterization of the equilibrium likeli-
hood of crime without punishment in terms of the fixed parameters of our model. Using this char-
acterization, we are able to establish the following proposition linking institutional tolerance for
police violence and impunity for criminal activity.
Proposition 1: The relationship between crime without punishment and tolerance for
police violence is potentially non-monotonic. At very low levels of initial permissiveness
toward violent policing, an increase in the same may for certain combinations of para-
meter values lead to a decrease in the probability of crime without punishment. This is
because the reduction in the attractiveness of crime due to the prospect of police violence
may overwhelm the effect of the growing non-reporting of crime. However, as institutional
permissiveness toward violent policing becomes sufficiently great, the likelihood of non-
reporting of crime reaches a point such that the attractiveness of crime increases even as
the cost associated with police violence becomes more extreme. At the limit, extremely high
levels of permissiveness toward police violence generate a high equilibrium probability of
crime without punishment (greater than 1/2) in which the criminal chooses to steal based
on the returns to theft vs. the private sector wage and the law-abiding citizen never reports
the crime (see Appendix).
Figure 5 displays a figure providing the key intuitions underlying Proposition 1. The figure dis-
plays the equilibrium probability of crime (i.e., the probability that Cchooses to steal), the
FIGURE 5 The impact of increasing permissiveness of police violence.
Notes: The upper left-hand side panel depicts the relationship between permissiveness and the probability of crime.
The upper right-hand side panel depicts the relationship between permissiveness and the probability of crime
reporting. The lower panel depicts the relationship between permissiveness and the probability of the crime without
punishment outcome.Graphs shown are for q=1!exp(!r), k=1, h=0.5, g=0.7, l=0.5, -¼0:2, s=0.8,
c=0.9, and b=3. The parameter rvaries from 0 to 20, with r=3.12 and r’’ =2.65.
equilibrium probability of crime reporting, and the equilibrium probability of the crime without
punishment outcome as the value of rvaries. The total effect of permissiveness toward police vio-
lence on the probability of crime without punishment can be decomposed into two distinct effects:
the effect of permissiveness on crime and the effect of permissiveness on reporting of crime. Echo-
ing our empirical findings from Costa Rica, the effect of violent policing on reporting is unambig-
uous: it monotonically reduces the likelihood that the law-abiding citizen will report crime if she is
victimized (see upper left-hand side panel of Figure 5). This is so for two reasons. First, an
increase in permissiveness produces a perverse incentive effect : violence-prone officers get greater
utility from abusing citizens the greater is permissiveness, so contingent on the officer being vio-
lence prone, this increases the probability of abuse and reduces the attractiveness of reporting. Sec-
ond, an increase in permissiveness generates a negative selection effect. Greater permissiveness
makes it more likely that the police officer will be violence-prone in the first place: since violence-
prone officers are more inclined to abuse citizens than non-violence-prone officers, this, too,
reduces the attractiveness of reporting.
The subtlety of the result emerges from the strategic calculus of the criminal. The impact of an
increase in permissiveness toward police violence on the utility the criminal receives from engag-
ing in crime consists of two separate effects that pull in different directions: the sanction severity
effect and the non-reporting effect.
The most obvious of the two, and the one that undergirds most arguments in favor of more vio-
lent policing, is sanction severity. This effect captures the increasing disutility from crime the crim-
inal receives due to the increasing magnitude of physical violence that can be perpetrated by the
police officer. Sanction severity grows with permissiveness because more permissiveness implies
greater suffering if the police officer chooses to use violence (holding the likelihood of violence
constant) and because the likelihood of violence also increases with permissiveness.
The second effect is the non-reporting effect. As described above, increasing permissiveness
toward police violence decreases the probability of crime reporting by the law-abiding citizen.
From the vantage point of the criminal, this leads to an increase in the likelihood of capturing
the benefits from crime and a decrease in the likelihood of assuming its potential costs. As
such, the non-reporting effect, driven by the citizens fears of engaging with the police officer,
increases the attractiveness of crime.
The fact that there are these two countervailing effects of permissiveness toward police violence
on the utility of crime is the reason for the non-monotonicity exhibited in the upper left-hand side
and bottom panel of Figure 5. At low levels of permissiveness, the sanction severity effect may
dominate and further increases in permissiveness may reduce the probability of crime at least up
to a point. However, at high levels of permissiveness, the non-reporting effect comes to dominate,
and further increases serve to stimulate crime, thereby making the crime without punishment out-
come more likely.
The reason the non-reporting effect dominates sanction severity at high levels of permissiveness is
because the sanction severity effect is inherently dependent on the probability of reporting. Intu-
itively, if the law-abiding citizen will surely not report the theft to the police officer, then the prospect
of even extreme violence perpetrated by the police officer is irrelevant since the crime will never
come to be investigated in the first place. Since the probability of crime reporting approaches zero as
permissiveness grows very large, the probability of crime without punishment eventually reaches a
steady state in which the criminal ignores completely the sanction associated with police violence and
decides to engage in crime based on the relative value of theft vs. the licit wage. If the difference
between these is large, or, alternatively, if the rationality parameter kis large, then the probability of
crime without punishment approaches 1 as permissiveness reaches an extreme level.
The novelty of our framework rests with the emphasis it places on crime reporting and, by
extension, citizen collaboration with police more generally, as a fundamental lynchpin underlying
any plausible attempt to reduce crime. Although it is true that giving police officers wide latitude
to impose physical hardship on suspected criminals might deter some set of potential criminals, the
policies that make such latitude possible are likely to spill over into police interactions with non-
criminals, thereby heightening the risk law-abiding members of the public associate with engaging
with the police. Criminals, being rational and forward looking, will naturally take into considera-
tion the reluctance of law abiding citizens to report crime at the moment when they decide whether
or not to engage in crime. As a consequence, giving officers relatively free reign to abuse sus-
pected criminals, rather than quelling outbreaks of crime, is actually more likely to promote a situ-
ation of equilibrium impunity for crime.
In this paper, we have examined the manner in which police violence affects citizenswillingness to
report crime and therefore collaborate with authorities in reducing crime. Our analysis proceeded in
two steps. First, we empirically assessed the relationship between the observation of police violence
and crime non-reporting through the analysis of a large-scale household survey we conducted in
urban Costa Rica. According to the empirics, citizens who witness police violence are far less likely
to indicate a willingness to report crime both crimes for which they are the victim as well as crimes
for which they are a witness. Subsequent analysis of attitudes suggested that a significant driver of
the reticence to report was the perception that engaging the police would be a potentially risky endea-
vor. We interpret these findings as fairly solid evidence in favor of the proposition that police vio-
lence erodes critical linkages with communities afflicted by crime, making effective policing that
much more difficult. After presenting the empirics, we proceeded to develop a general theoretical
framework that outlined logically the ways in which expectations of violence may shape crime report-
ing and, by extension, the prevalence of crime. The model revealed that although the prospect of
police violence against criminals may generate a degree of deterrence for criminal behavior, the larger
consequence of having a violent police force will be an increase in law-abiding citizensreticence to
report crime. This non-reporting effect can diminish the risks associated with crime and increase the
expected returns to such a degree that high expectations of police violence may generate effective
impunity for crime what we dub crime without punishment.
Herein lies the fundamental irony at the heart of our paper. Apologists for police violence often
justify their position based on the notion that some evils must be tolerated in order to achieve a
greater societal good. But our empirics and theoretical analysis suggests that there is no such good
to be had from permissiveness toward police brutality. To the contrary, in the long run bad means
also produce bad outcomes.
The logical flaw here resides in a failure to properly comprehend the link between security and
community engagement. Effective policing, at its heart, is about gathering pertinent information and
being able to separate the wheat from the chaff in communities where some actors are dedicated crim-
inals, some are law-abiding citizens, and some go back and forth between these worlds. Doing so
requires carefully cultivating mutual respect and trust between members of the community and the
forces of law and order. When the police engage in acts of violence that exceed the strictures of the
law, this erodes community trust and marks the forces of law and order as a potential threat to both
criminals and law-abiding citizens alike. Information channels from communities to the police cease
to function, the latter rendered incapable of properly fulfilling their duties to the public.
Any improvement in this state of affairs requires a change within state institutions in the level
of permissiveness toward police violence. Unfortunately, these changes take time. While we think
that the ultimate goal must be to restore the trust between citizens and the forces of law and order
which requires the aforementioned change in permissiveness toward police brutality we recog-
nize that some short-term measures might help to mitigate the incidence of non-reporting due to
fear or distrust of police. In this regard, the use of innovative reporting platforms that allow the
anonymous reporting of crime, such as Disque-Den"
uncia (established in 1995 in Rio de Janeiro,
Brazil) and (established in 2013 in Medell"
ın, Colombia), strike us as
particularly promising.
Of course, we are fully cognizant that police violence is only one obstacle to establishing such
trust. Combatting corruption and promoting professionalism are also crucial components of any
strategy to cultivate greater collaboration between citizens and their police forces. However, in an
age of growing calls for an iron fist in order to combat crime, we find it necessary to point out that
this dark road will likely lead to disappointment.
The authors would like to thank Ana Corbacho and Mauricio Ruiz Vega for their help in develop-
ing and executing the survey in Costa Rica. We also thank Daniel Brinks, Roberto Guerrero
an, Ben Ross Schneider, Phil Keefer, Horacio Larreguy, Alberto Simpser, and seminar par-
ticipants at Harvard University and the University of Virginia for helpful comments and
Virginia Oliveros
uero, J. (2013). Causal Estimates of the Intangible Costs of Violence against Women in Latin America and the
Caribbean.IDB Working Paper Series No. IDB-WP-41, Inter-American Development Bank, Research Depart-
Akinlabi, O.M., (2016). Do the police really protect and serve the public? Police deviance and public cynicism
towards the law in Nigeria. Criminology and Criminal Justice,17(2), 158174.
Becker, G. S. (1968). Crime and punishment: An economic perspective. Journal of Political Economy,76,
Bosick, S. J., Rennison, C. M., Gover, A. R., & Dodge, M. (2012). Reporting violence to the police: Predictors
through the life course. Journal of Criminal Justice,40(6), 441451.
Bowles, R., Reyes, M. G., & Garoupa, N. (2009). Crime reporting decisions and the costs of crime. European Jour-
nal on Criminal Policy and Research,15(4), 365377.
Brinks, D. M. (2003). Informal institutions and the rule of law: The judicial response to state killings in Buenos
Aires and S˜ao Paulo in the 1990s. Comparative Politics,36(1), 119.
Brinks, D. M. (2008). The judicial response to police killings in latin America. New York: Cambridge University
Carr, P. J., Napolitano, L., & Keating, J. (2007). We never call the cops and here is why: A qualitative examination
of legal cynicism in three Philadelphia neighborhoods. Criminology,45(2), 445480.
Cruz, J. M. (2009). Police Abuse in Latin America. AmericasBarometer Insights 11, 15.
Desmond, M., Papachristos, A. V., & Kirk, D. S. (2016). Police violence and citizen crime reporting in the black
community. American Sociological Review,81(5), 857876.
Di Tella, R., Galiani, S., & Schargrodsky, E. (2010). Crime distribution and victim behavior during a crime wave.
In R. Di Tella, S. Edwards & E. Schargrodsky (Eds.), The economics of crime: Lessons for and from Latin
America (pp. 175204). Chicago: University of Chicago Press.
Estienne, E., & Morabito, M. (2016). Understanding differences in crime reporting practices: A comparative
approach. International Journal of Comparative and Applied Criminal Justice,40(2), 123143.
Gaviria, A., Medina, C., Morales, L., & N"
nez, J. (2010). The cost of avoiding crime: The case of Bogot"
a. In R. Di
Tella, S. Edwards & E. Schargrodsky (Eds.), The economics of crime: Lessons for and from Latin America (pp.
101132). Chicago: University of Chicago Press.
Goeree, J., Holt, C., & Palfrey, T. (2016). Quantal Response Equilibrium: A Stochastic Theory of Games. Princeton:
Princeton University Press.
Gottfredson, M. R., & Hindelang, M. J. (1979). A study of the behavior of law. American Sociological Review,
44(1), 318.
Goudriaan, H., Lynch, J. P., & Nieuwbeerta, P. (2004). Reporting to the police in western nations: A theoretical
analysis of the effects of social context. Justice Quarterly,21(4), 933969.
Imbens, G. W. (2004). Nonparametric estimation of average treatment effects under exogeneity: A review. Review of
Economics and Statistics,86(1), 429.
Instituto Nacional de Estadisticas y Censos and PNUD. (2015). (Marzo). La Victimizacion en Costa Rica al a~
Jaitman, L. (2015) Introduccion: Los costos del crimen en el bienestar. In L. Jaitman (ed.), Los costos del crimen y
la violencia en el bienestar en Am!erica Latina y el Caribe.Banco Interamericano de Desarrollo. Washington,
Johnson, V. E., & Albert, J. H. (1999). Ordinal Data Modeling. New York: Springer-Verlag.
Kochel, T. R., Parks, R., & Mastrofski, S. D. (2013). Examining police effectiveness as a precursor to legitimacy
and cooperation with police.Justice Quarterly 30(5): 895925.
Koppensteiner, M. F., & Manacorda, M. (2013). The Effect of Violence on Birth Outcomes: Evidence from Homi-
cides in Rural Brazil.IDB Working Paper Series No. IDB-WP-416, Inter-American Development Bank,
Research Department.
ometro (2015). Informe 2015.Corporaci"
on Latinobar"
ometro. Santiago de Chile.
McKelvey, R. D., & Palfrey, T. R. (1995). Quantal response equilibria for normal form games. Games and Eco-
nomic Behavior,10(1), 638.
McKelvey, R. D., & Palfrey, T. R. (1998). Quantal response equilibria for extensive form games. Experimental Eco-
nomics,1(1), 941.
Poder Judicial de Costa Rica, Direccion de Planificacion, Seccion de Estadistica. (2016). Compendio de Indicadores
Judiciales 2010-2014).
Sekhon, J. S. (2015). Matching: Multivariate and Propensity Score Matching Estimator for-Causal Inference (v 4.9-2).
Package for the R statistical environment.
Signorino, C. (1999). Strategic interaction and the statistical analysis of conflict. American Political Science Review,
93(2), 279297.
Skogan, W. G. (1984). Reporting crimes to the police: The status of world research. Journal of Research in Crime
and Delinquency,21(2), 113137.
Soares, R. R. (2004a). Crime reporting as a measure of institutional development. Economic Development and Cul-
tural Change,52(4), 851871.
Soares, R. R. (2004b). Development, crime, and punishment: Accounting for the international differences in crime
rates.Journal of Development Economics 73(1): 155184.
Soares, R. R., & y Naritomi, J. (2010). Understanding high crime rates in Latin America: The role of social and
policy factors.In R. Di Tella, S. Edwards & E. Schargrodsky (Eds), The economics of crime: Lessons for and
from Latin America (pp. 1955). Chicago: University of Chicago Press.
Sunshine, J., & Tyler, T. R. (2003). The role of procedural justice and legitimacy in shaping public support for
policing. Law & society review,37(3), 513548.
Tankebe, J. (2009). Public cooperation with the police in Ghana: Does procedural fairness matter? Criminology,
47(4), 12651293.
Tankebe, J. (2010). Public confidence in the police testing the effects of public experiences of police corruption in
Ghana. British Journal of Criminology,50(2), 296319.
Tarling, R., & Morris, K. (2010). Reporting crime to the police. British Journal of Criminology,50(3), 474490.
UNDP, 2013, Citizen security with a human face: Evidence and proposals for Latin America. (UNDP, New York,
Van Wilsem, J. (2004). Criminal victimization in cross-national perspective an analysis of rates of theft, violence
and vandalism across 27 countries.European Journal of Criminology 1(1): 89109.
Vetter, D. M., Beltr˜ao, K. I., & Massena, R. M. (2013). The Impact of the Sense of Security from Crime on Resi-
dential Property Values in Brazilian Metropolitan Areas (No. IDB-WP-415). Inter-American Development Bank,
Research Department.
Weitzer, R. (2002). Incidents of police misconduct and public opinion. Journal of Criminal Justice,30(5), 397408.
Xie, M., Pogarsky, G., Lynch, J. P., & McDowall, D. (2006). Prior police contact and subsequent victim reporting:
Results from the NCVS. Justice Quarterly,23(4), 481501.
Additional Supporting Information may be found online in the supporting information tab for this
How to cite this article: Gingerich DW, Oliveros V. Police Violence and the
Underreporting of Crime. Economics & Politics. 2018;30:78105.
... 2 For exceptions, see Birkbeck et al. (1993), who compare the United States and Venezuela; Bennett and Wiegand (1994) Gingerich and Oliveros (2018) on Costa Rica. Also see Estienne and Morabito (2016), who use ICVS data to look at determinants of reporting robbery and assault in 23 developing and transition countries. ...
... 5 Anticipating this threat, victims or witnesses may be reluctant to report, thus failing both to cooperate with the state and to hold the state accountable for its actions. This is particularly true for those who may have directly witnessed police violence (Gingerich & Oliveros 2018). Further, fear of retaliation may also be greater in countries where the police generate additional income by selling information about criminal investigations (Taylor 2011). ...
Full-text available
What factors affect citizens’ engagement with the state? We explore this question through a study of victims’ and bystanders’ willingness to report crimes to the police, using data from survey experiments conducted in Russia and Georgia. We find that citizens’ willingness to report in both countries is strongly influenced by the nature of the crime, but not generally by instruments that the state might use to encourage greater reporting. Our results recommend scepticism about the ability of governments to easily engineer citizens’ engagement with the state.
... Such demands have tended to follow violent crime spikes linked to gang activity and narcotics trafficking. Although recent evidence suggests that Mano Dura policies have not been particularly effective in reducing crime (Gingerich and Oliveros 2018;Bullock 2018), they have imposed huge costs for inmates and for the population at large. 4 Thus, understanding why citizens in LAC come to support or reject such policies is of the utmost importance. ...
Anti-crime policy is often unresponsive to reductions in crime. To address why, we provide a model and empirical test of how citizens’ anti-crime policy preferences respond to information. Our model shows that preferences for anti-crime policy hinge on expectations about the crime rate: punitive policies are preferred in high crime contexts, whereas social policies are preferred in low crime contexts. We evaluate these expectations through an information experiment embedded in the 2017 Latin American Public Opinion Project survey conducted in Panama. As expected by our theory, a high crime message induced stronger preferences in favour of punitive policies. Unanticipated by our theory, but in line with cursory evidence and survey results, we find that a low crime message did not induce stronger preferences in favour of social policies. These findings are consistent with policy ratcheting: punitive policies increase during periods of high crime and remain in place during periods of low crime.
... The low reporting of crime offers further clarity on the nature of crime trends in Ghana. The underreporting of crime is common in instances of rape, homicides, and domestic violence (Feldman et al. 2017;Myers 1980;Gingerich and Oliveros 2018;Loftin et al. 2003). A 2011/2013 Afrobarometer survey found that 75% of victims of crime in Ghana did not report the crime to the police (Wambua 2015). ...
Full-text available
This study examines crime rate trends in Ghana in the context of research into the international crime drop. Ghanaian police-recorded crime data are imperfect, but crime rates appear to have increased to the early 2000s then to have declined significantly. However, the national trend for all crime types masks significant variation by crime type and region. The national trends appear, broadly speaking, consistent with long-term crime increases and decreases in high-income countries, but with a lag that may reflect Ghana’s economic development trajectory. This may be consistent with the routine activity perspective if both crime opportunities and economic development increased in parallel, followed by a crime decline that reflected increased security measures or other change. The study raises many questions that require further research, and some suggestions to that end are outlined.
... For example, in the case of predictive policing, prior cases of inferior, discriminatory or outright illegal policing ("dirty policing") are visible in criminological data ("dirty data") (Richardson, Schultz and Crawford 2019, p. 192), e. g. due to underreporting of sexual crimes (Taylor and Gassner 2010, p. 241 ff.), racist crimes (Kushnick 1999, ¶1.7) and police violence (Loftin, McDowall and Xie 2017;Gingerich and Oliveros 2018). Following the legal precepts of non-discrimination in policing and counteracting such biases is made all the harder by the fact that there is often little awareness and acknowledgement of these underlying problems and the discriminatory structures which are (at least partially) responsible therefor. ...
Digitisation, automation and datafication permeate policing and justice more and more each year -- from predictive policing methods through recidivism prediction to automated biometric identification at the border. The sociotechnical issues surrounding the use of such systems raise questions and reveal problems, both old and new. Our article reviews contemporary issues surrounding automation in policing and the legal system, finds common issues and themes in various different examples, introduces the distinction between human "retail bias" and algorithmic "wholesale bias", and argues for shifting the viewpoint on the debate to focus on both workers' rights and organisational responsibility as well as fundamental rights and the right to an effective remedy.
... For example, in the case of predictive policing, prior cases of inferior, discriminatory, or outright illegal policing ("dirty policing") are visible in criminological data ("dirty data") (Richardson et al., 2019, p. 192), e.g., due to underreporting of sexual crimes (Taylor and Gassner, 2010, p. 241 ff.), racist crimes (Kushnick, 1999, ¶1.7), and police violence (Loftin et al., 2017;Gingerich and Oliveros, 2018). Following the legal precepts of non-discrimination in policing and counteracting such biases is made all the harder by the fact that there is often little awareness and acknowledgement of these underlying problems and the discriminatory structures which are (at least partially) responsible therefor. ...
Full-text available
Digitisation, automation, and datafication permeate policing and justice more and more each year—from predictive policing methods through recidivism prediction to automated biometric identification at the border. The sociotechnical issues surrounding the use of such systems raise questions and reveal problems, both old and new. Our article reviews contemporary issues surrounding automation in policing and the legal system, finds common issues and themes in various different examples, introduces the distinction between human “retail bias” and algorithmic “wholesale bias”, and argues for shifting the viewpoint on the debate to focus on both workers' rights and organisational responsibility as well as fundamental rights and the right to an effective remedy.
... Nevertheless, when hotspot neighborhoods are over-patrolled, questions arise from a justice perspective, as patrols can become profiling operations (Green, 1999;Rosenbaum, 2006). Moreover, hotspot patrols can turn into aggressive policing, which doesn't reduce crime, as shown by Gingerich and Oliveros (2018) and Bullock and Garland (2018). ...
Full-text available
Crime is a major problem in Latin America and the Caribbean. With 9 percent of the world's population, the region accounts for 33 percent of global homicides. Using new, extensive survey data, we endeavor to identify what anti-crime policies citizens in the region demand from their governments. We also analyze who is demanding what and why. We find that harsher penalties appear to be the preferred weapon in the anti-crime arsenal but people are willing to spend public moneys not only for punishment, but also for anti-poverty and detection policies. Citizens recognize that allocating resources to the police is better than subsidizing private security for citizens. Nevertheless, most oppose raising taxes to fund the police, a reluctance that might stem from mistrust in governments' ability to manage these resources. Mistrust, misinformation, and impatience combine to create flawed anti-crime policy. Educating citizens both about crime and about the fiscal consequences of their policy preferences may help move the region's public opinion toward a better policy equilibrium. Governments should also invest in their capability to design and deliver evidence-based solutions for fighting crime, and work to increase trust levels among citizens.
... La mayoría de los estudios sobre denuncias se basan en la presunción de que la decisión de denunciar es racional (Bowles, Garcia Reyes y Garoupa, 2009;Carr, Napolitano y Keating, 2007;Kirk y Matsuda, 2011;Kury, Teske y Würger, 1999;Skogan, 1984). Como tal, es el resultado de la evaluación de los costos y beneficios de acudir a la policía (Gingerich y Oliveros, 2018;Greenberg y Beach, 2004;Skogan, 2009;Soares, 2004). Desde esta perspectiva, ha sido recurrente encontrar que las variables más importantes son aquellas vinculadas con las características del crimen. ...
Full-text available
Perú no solo tiene una de las mayores tasas de victimización en Latinoamérica (24%), sino también la menor en delitos denunciados (15%) (Latinobarómetro, 2016). En este trabajo, el objetivo fue identificar las características del crimen y las características individuales que predicen la decisión de denunciar cuatro delitos patrimoniales distintos (robo de hogares, de auto, de autopartes y de moto). Para ello, se emplearon siete años de una encuesta de hogares a nivel nacional en Perú (2010-2016). Las estimaciones se realizaron mediante regresión logística multinivel de efectos mixtos a fin de controlar las características del ambiente que también influyen en la decisión de denuncia. El uso de arma de parte del delincuente es el factor que eleva en mayor proporción la probabilidad de denunciar cualquiera de los cuatro delitos evaluados. La victimización reiterada es también un predictor importante, aunque para menos delitos. La confianza en la policía no está asociada a la decisión de denunciar. Este estudio es el primero que analiza cuantitativamente los factores que afectan la decisión de denunciar en el Perú. Sus resultados son útiles para un mejor entendimiento de la baja tasa de denuncias en el país.
Comprised of three papers, this dissertation examines how changes within the urban environment affect crime and the criminal justice system. Collectively, these three papers address policy-relevant issues for urban communities that include police-community relations, the caretaking role of the police, firearm-related violence, blighted and vacant land, the opioid epidemic, and access to treatment for opioid use disorder. The first paper assesses the place-based effects of outpatient methadone maintenance treatment facilities on crime in Philadelphia. The analysis in this paper uses spatial and temporal variation in OMMT facility presence during an eleven-year period (2007-2017) to determine these effects. Within a 200-meter radius, the presence of an OMMT facility caused a significant decrease in property and total crime but a significant increase in drug and violent crime. There were no significant effects on crime outside of the 200-meter radius. The second paper evaluates the effect of randomized vacant lot remediation on the frequency of shootings that result in serious injury or death. The analysis in this paper uses data from a 2013 randomized controlled trial was conducted in Philadelphia involving 541 vacant lots. Some vacant lots were remediated, other vacant lots received a lesser level of remediation, and the remainder received no remediation. Over a 60-month period, remediating lots significantly reduced shooting incidents; there was no evidence that the interventions displaced shootings into adjacent areas. The third paper tests whether a high-profile death-in-police-custody incident in Baltimore affected community reliance on the police, as measured through citizen calls requesting police assistance for non-criminal caretaking matters. This paper presents a Negative Community-Police Relationship Index Score that operationalizes the risk of a negative community-police relationship for a given location within Baltimore based on factors such as the percentage of residents living in poverty and the percentage of vacant residential housing units. Even in high-risk sections, the death-in-police-custody incident did not significantly affect community reliance on the police for non-criminal caretaking matters.
Full-text available
Importance: Domestic violence (DV) has become a growing public health concern during the COVID-19 pandemic because individuals may be sheltering in place with abusers and facing mounting economic and health-related stresses. Objective: To analyze associations of the 2020 COVID-19 stay-at-home (SH) order with DV police reporting and resource availability, including differences by community area racial/ethnic composition. Design, setting, and participants: This longitudinal cohort study assessed DV police reports (January-June 2020) obtained from the Chicago, Illinois, Police Department and DV resource availability (March and August 2020) obtained from the NowPow community resource database, both for 77 community areas in Chicago. Data were analyzed July through December 2020. Exposures: The COVID-19 SH order effective March 21, 2020. Main outcomes and measures: Monthly rates of DV police reports and DV resource availability per 100 000 persons. Results: Of 77 community areas in Chicago, 28 (36.4%) were majority Black, 19 (24.7%) majority Hispanic/Latinx, 18 (23.4%) majority White, and 12 (15.6%) a different or no majority race/ethnicity, representing an estimated population of 2 718 555 individuals. For each community area, the SH order was associated with a decrease in the rate of DV police reports by 21.8 (95% CI, -30.48 to -13.07) crimes per 100 000 persons per month relative to the same months in 2019. Compared with White majority community areas, Black majority areas had a decrease in the rate of DV police reports by 40.8 (95% CI, -62.93 to -18.75) crimes per 100 000 persons per month relative to the same months in 2019. The SH order was also associated with a decrease in DV resource availability at a rate of 5.1 (95% CI, -7.55 to -2.67) resources per 100 000 persons, with the largest decreases for mental health (-4.3 [95% CI, -5.97 to -2.66] resources per 100 000 persons) and personal safety (-2.4 [95% CI, -4.40 to -0.41] resources per 100 000 persons). The Black majority south side of Chicago had a larger decrease in resource availability (-6.7 [95% CI, -12.92 to -0.46] resources per 100 000 persons) than the White majority north side. Conclusions and relevance: In this longitudinal cohort study, the rate of DV police reports decreased after the SH order was implemented in Chicago. This decrease was largely observed in Black majority communities, whereas there was no significant change in White majority communities. These findings may reflect decreased DV incidence but may also reflect an exacerbation of underreporting. In addition, DV resource availability decreased disproportionately on the predominantly Black south side of Chicago.
Full-text available
Using qualitative legal scholarship, combined with literature analysis from post-conflict peacebuilding and police studies, this article provides a normative and theoretical lens through which police and other actors can view and carry out reform efforts. It explores whether and how the concept of guarantees of non-repetition could contribute to or reframe discussions in order to prevent future violence and facilitate lasting institutional changes. The article examines the development of a broader approach to security sector reform and explores guarantees of non-repetition and the conceptual confusion it has encountered. It teases out the main aspects of guarantees of non-repetition, including its human rights elements, such as due diligence obligations. Finally, it addresses how guarantees of non-repetition provide a normative institutional policy framework that offers the possibility to shift the rhetoric to focus on State obligations that are context-driven. As a result, guarantees of non-repetition could prove useful when addressing police reform.
Full-text available
This study empirically examines the everyday problem of corrupt policing and other related abuses in Nigeria, and how these deviant behaviours engender public cynicism towards the law. In any democratic society, police officers are expected to be accountable for their actions and inactions. But the perennial problem in Nigeria is that the police are not accountable to anyone. The history of Nigeria policing is littered with accounts of deviance, malevolent attitudes towards the public and failures of the police organization in detecting or disciplining errant officers. Using a sample of 462 participants from a cross-sectional survey, this study examines whether actual or vicarious experiences of police deviance are likely to predict public cynicism towards the law. This current study corroborates previous assertions that the relationship between the police and the public in Nigeria is poor and that police deviance engenders cynicism towards the law. Implications for policymaking and law-abiding behaviour are discussed.
Full-text available
Does an informal institution grant police wide latitude to use lethal force in Argentina and Brazil? Evidence from extensive fieldwork and a sample of police homicide prosecutions in Buenos Aires and São Paulo indicate that an informal institution is at work in cases involving the killing of a victim perceived as a violent criminal but not in more routine cases of excessive use of force. In the latter cases, the problem is more properly characterized as a failure of the system to gather the requisite information to support a prosecution.
High-profile cases of police violence—disproportionately experienced by black men—may present a serious threat to public safety if they lower citizen crime reporting. Using an interrupted time series design, this study analyzes how one of Milwaukee’s most publicized cases of police violence against an unarmed black man, the beating of Frank Jude, affected police-related 911 calls. Controlling for crime, prior call patterns, and several neighborhood characteristics, we find that residents of Milwaukee’s neighborhoods, especially residents of black neighborhoods, were far less likely to report crime after Jude’s beating was broadcast. The effect lasted for over a year and resulted in a total net loss of approximately 22,200 calls for service. Other local and national cases of police violence against unarmed black men also had a significant impact on citizen crime reporting in Milwaukee. Police misconduct can powerfully suppress one of the most basic forms of civic engagement: calling 911 for matters of personal and public safety.
High-profile cases of police violence—disproportionately experienced by black men—may present a serious threat to public safety if they lower citizen crime reporting. Using an interrupted time series design, this study analyzes how one of Milwaukee’s most publicized cases of police violence against an unarmed black man, the beating of Frank Jude, affected police-related 911 calls. Controlling for crime, prior call patterns, and several neighborhood characteristics, we find that residents of Milwaukee’s neighborhoods, especially residents of black neighborhoods, were far less likely to report crime after Jude’s beating was broadcast. The effect lasted for over a year and resulted in a total net loss of approximately 22,200 calls for service. Other local and national cases of police violence against unarmed black men also had a significant impact on citizen crime reporting in Milwaukee. Police misconduct can powerfully suppress one of the most basic forms of civic engagement: calling 911 for matters of personal and public safety.
This book examines the effect of social inequality, political influence, and institutional design on the effectiveness of legal systems in Argentina, Brazil, and Uruguay. It demonstrates the inequality existent in these systems, as well as the occasional successes. Its focus is on the criminal prosecution of violent police officers, but it draws implications for democracy, the rule of law, court functioning, and police violence. The book describes judicial, prosecutorial, and police structures and operation, as well as the nature of and response to lethal police violence in each location.
Despite a large body of literature detailing crime reporting practices, scant research examines the correlates of the crime reporting decision in developing nations with newer democracies, newer economies, or developing economies. Using a sample of 23 nations from the 2000 International Crime Victimization Survey (ICVS), this study tests the generalizability of correlates of robbery and assault reports to the police in a non-US sample to determine whether there are national differences in reporting practices. Based on this analysis emphasizing the developing world and countries in transition, an expanded model, integrating incident, demographic, police-related, and national variables, is developed that will enhance our understanding of differences in reporting practices in developing and developed nations. Results identified both similarities and differences in crime reporting practices between developed and developing nations. Although the findings reinforce the importance of incident and demographic characteristics on reporting crime to the police, they also suggest that national variables, reflecting economic and social context, should not be excluded from studies examining reporting practices.
Violence has a striking gender pattern. Men are more likely to be attacked by a stranger, while women experience violence mostly from their partners. This paper estimates the costs of violence against women in terms of intangible outcomes, such as women’s reproductive health, labor supply, and the welfare of their children. The study uses a sample of nearly 83,000 women in seven countries from all income groups and all sub-regions in Latin American and the Caribbean. The sample, consisting of 26. 3 million women between the ages of 15 and 49, strengthens the external validity of the results. The results show that physical violence against women is strongly associated with their marital status because it increases the divorce or separation rate. Violence is negatively linked with women’s health. The study shows that domestic violence additionally creates a negative externality by affecting important short-term health outcomes for children whose mothers suffered from violence. To obtain the child health outcomes, the study employs a natural experiment in Peru to establish that these effects appear to be causal. Finally, the paper presents evidence indicating that women’s education and age buffer the negative effect of violence against women on their children’s health outcomes.
Using a hedonic residential rent model for Brazil’s metropolitan areas calibrated with microdata from Brazil’s annual household survey, this study estimates that increasing the sense of security in the home by one standard deviation would increase average home values by R$1,513 (US$757), or about US$13. 6 billion if applied to all 18. 0 million households in the study area. The principal components analysis of sense of security and crime victimization variables indicates that higher-income households feel more secure from crime in the home, even though theft and robbery victimization rise with household income and rent level. Higher levels of home protection measures by higher-income households partially explain this result.
In The Behavior of Law, Black (1976) sets forth a theory of law that he argues explains variations in law across societies and among individuals within societies. Black argues that law can be conceived of as a quantitative variable, measured by the number and scope of prohibitions, obligations and other standards to which people are subject. Law varies, according to Black, with other aspects of social life, including stratification, morphology, culture, organization, and social control. Many of Black's principal propositions regarding the quantity of law are tested in this paper with National Crime Survey data on the victim's decision to report a crime to the police. An alternative model that views the quantity of law as depending largely on the gravity of the infraction against legal norms is posed and tested against Black's theory. The data are generally inconsistent with the propositions derived from The Behavior of Law and strongly suggest that a theory attempting to explain the criminal law cannot ignore the gravity of the infraction against legal norms.