Brayne 2017

725865
2017
ASRXXX10.1177/0003122417725865American Sociological ReviewBrayne
American Sociological Review
Big Data Surveillance: The 2017, Vol. 82(5) 977–1008

© American Sociological
Association 2017
Case of Policing DOI: 10.1177/0003122417725865
https://doi.org/10.1177/0003122417725865
journals.sagepub.com/home/asr
Sarah Braynea
Abstract
This article examines the intersection of two structural developments: the growth of
surveillance and the rise of “big data.” Drawing on observations and interviews conducted
within the Los Angeles Police Department, I offer an empirical account of how the adoption
of big data analytics does—and does not—transform police surveillance practices. I argue
that the adoption of big data analytics facilitates amplifications of prior surveillance practices
and fundamental transformations in surveillance activities. First, discretionary assessments
of risk are supplemented and quantified using risk scores. Second, data are used for
predictive, rather than reactive or explanatory, purposes. Third, the proliferation of automatic
alert systems makes it possible to systematically surveil an unprecedentedly large number
of people. Fourth, the threshold for inclusion in law enforcement databases is lower, now
including individuals who have not had direct police contact. Fifth, previously separate data
systems are merged, facilitating the spread of surveillance into a wide range of institutions.
Based on these findings, I develop a theoretical model of big data surveillance that can be
applied to institutional domains beyond the criminal justice system. Finally, I highlight the
social consequences of big data surveillance for law and social inequality.
Keywords
police, big data, inequality, crime, law
In the past decade, two major structural devel- subject of contentious debate in policy, media,
opments intersected: the proliferation of sur- legal, regulatory, and academic circles. How-
veillance in everyday life and the rise of “big ever, discourse on the topic is largely specula-
data.” Emblematic of the expansion of sur- tive, focusing on the possibilities, good and
veillance (Lyon 2003; Marx 2016; Rule 2007) bad, of new forms of data-based surveillance.
is the rapid growth of the criminal justice The technological capacities for surveillance
system since 1972 (Carson 2015; Garland far outpace empirical research on the new
2001; Wakefield and Uggen 2010; Western data landscape. Consequently, we actually
2006). At the same time, facilitated by the know very little about how big data is used in
mass digitization of information, there has
been a rise in the computational analysis of
massive and diverse datasets, known as “big a
The University of Texas at Austin
data.” Big data analytics have been taken up
Corresponding Author:
in a wide range of fields, including finance, Sarah Brayne, Assistant Professor, Department of
health, social science, sports, marketing, Sociology, The University of Texas at Austin, 305
security, and criminal justice. The use of big E. 23rd St., A1700, Austin, TX 78712-1086
data in police surveillance activities is the Email: [email protected]
978 American Sociological Review 82(5)
surveillance activities and to what conse- into relational systems and include data origi-
quence. nally collected in other, non-criminal-justice
This article provides a case study of a large institutions. These shifts made possible by big
urban police department—the Los Angeles data have implications for inequality, law, and
Police Department (LAPD)—to investigate organizational practice in a range of institu-
the relationship between big data analytics and tional domains.
surveillance. In particular, it asks whether and
how the adoption of big data analytics trans-
forms police surveillance practices. Moreover, The Intensification of
it investigates implications of new surveillance Surveillance
practices not only for policing, but also for law, Surveillance is ubiquitous in modern societies
social inequality, and research on big data sur- (Giddens 1990; Lyon 1994, 2003, 2006, 2015;
veillance in other institutions. On the one hand, Marx 1974, 2002, 2016; Rule 1974). The
big data analytics may be a rationalizing force, practice of surveillance involves the collec-
with potential to reduce bias, increase effi- tion, recording, and classification of informa-
ciency, and improve prediction accuracy. On tion about people, processes, and institutions
the other hand, use of predictive analytics has (Foucault 1977; Haggerty and Ericson 2000;
the potential to technologically reify bias and Lyon 2003; Marx 2016). A wide range of
deepen existing patterns of inequality. scholars highlight the growing pervasiveness
To shed light on the social dimensions of of surveillance, referring to the emergence of
surveillance in the age of big data, I draw on “mass surveillance” (Rule 1974) and “surveil-
original interview and observational data col- lance societies” (Lyon 1994). Although it has
lected during fieldwork over the course of two received more attention in recent decades,
and a half years with the LAPD. As an agency surveillance as a practice is not new. Processes
at the forefront of data analytics, the Depart- of surveillance can be traced back to at least
ment serves as a strategic site for understand- the sixteenth century, during the appearance of
ing the interplay between technology, law, and the nation-state (Marx 2016), through the
social relations. I provide one of the first on- transatlantic slave trade (Browne 2015),
the-ground accounts of how the growing suite bureaucratization, rationalization, and modern
of big data systems and predictive analytics management in the nineteenth and twentieth
are used for surveillance within an organiza- centuries (Braverman 1974; Rule 1974; Weber
tion, unpacking how in some cases, the adop- 1978), and as an axiomatic accompaniment to
tion of big data analytics is associated with risk management practices in the twentieth
mere amplifications in prior practices, but that century (Ericson and Haggerty 1997). The
in others, it is associated with fundamental attacks on September 11th, 2001, further
transformations in surveillance activities. I stimulated and legitimated the expansion of
argue there are five key ways in which the surveillance. Widely viewed as a case of infor-
adoption of big data analytics is associated mation sharing failure in the intelligence com-
with shifts in practice to varying degrees: (1) munity, 9/11 encouraged an alignment of
discretionary assessments of risk are supple- actors with vested interests in enhancing sur-
mented and quantified using risk scores; (2) veillance operations (Ball and Webster 2007;
data are increasingly used for predictive, Lyon 2015), spurred an infusion of tax dollars
rather than reactive or explanatory, purposes; for development of new surveillance sensors
(3) the proliferation of automated alerts makes and data mining programs to produce strategic
it possible to systematically surveil an unprec- intelligence (Gandy 2002), and accelerated
edentedly large number of people; (4) datasets the convergence of previously separate sur-
now include information on individuals who veillance systems into a “surveillant assem-
have not had any direct police contact; and (5) blage” (Deleuze and Guattari 1987; Haggerty
previously separate data systems are merged and Ericson 2000).
Brayne 979
Surveillance scholars have documented a remote, low visibility or invisible, involun-

quantitative increase in surveillance, arguing tary, automated, preemptive, and embedded
that surveillance is one of the major institu- into routine activity (Marx 2002, 2016). In
tional dimensions of modern societies (Ball fact, new forms of systematic surveillance
and Webster 2007; Giddens 1990; Lyon 1994, have become quotidian to the point that it is
2003; Marx 1988, 2016). Although surveil- now an unavoidable feature of everyday life
lance is growing in all areas of society, its (Ball and Webster 2007; Lyon 2003; Marx
penetration is unevenly distributed (Fiske 2016). Consider the extent to which surveil-
1998). Some individuals, groups, areas, and lance is a requisite of participating in today’s
institutions are surveilled more than others, world (Ball and Webster 2007): to use a bank,
and different populations are surveilled for send an e-mail, obtain medical care, make a
different purposes (Lyon 2003). On the one phone call, travel on a highway, or conduct an
hand, there is a deepening of surveillance of Internet search, individuals leave digital
“at-risk” groups, such as parolees and indi- traces that are recorded and saved (see also
viduals on public assistance (Gilliom 2001; Foucault 1977; Haggerty and Ericson 2000;
Gustafson 2011; Soss, Fording, and Schram Lyon 2003).
2011), who can increasingly be tracked across Technologically mediated surveillance has
institutional boundaries. On the other hand, become routine organizational practice in a
emerging “dragnet” surveillance practices— wide range of public and private domains. It
meaning those that collect data on everyone, is a tool of general governance (Lyon 2003),
rather than merely individuals under suspi- a basic prerogative of individuals in all sorts
cion—result in increased monitoring of of private and public institutions, and a tool to
groups “previously exempt from routine sur- accomplish “goals that meet particular inter-
veillance” (Haggerty and Ericson 2000:606; ests” (Ericson and Haggerty 2006:22). Sur-
see also Angwin 2014; Lyon 2015). Surveil- veillance scholars have accordingly extended
lance is therefore now both wider and deeper: the core concerns of surveillance from polic-
it includes a broader swath of people and can ing and military contexts to other institutions,
follow any single individual across a greater including finance, commerce, labor, health,
range of institutional settings. education, insurance, immigration, and activ-
Surveillance is increasingly technologi- ism (Lyon 2003; Marx 2016; Rule 2007).
cally mediated, and emergent technologies According to David Lyon (2015:68–69), sur-
make it possible at an unprecedented scale veillance in each of these institutions, “cannot
(Ericson and Haggerty 1997; Lyon 1994; be understood without a sense of how the
Marx 2016). With the development of com- quest for ‘big data’ approaches are becoming
puting, mass surveillance emerged alongside increasingly central.”
mass communication (Rule 1974). The mass
digitization of information enables much of
what Marx terms the “new surveillance”: the Rise of Big Data
“scrutiny of individuals, groups, and contexts Big data is an emerging modality of surveil-
using technical means to extract or create lance. A wide range of organizations—from
information” (Marx 2016:20). Although the finance to healthcare to law enforcement—
goals of traditional and new surveillance are have adopted big data analytics as a means to
similar, the means are different. Whereas tra- increase efficiency, improve prediction, and
ditional surveillance is inductive, involving reduce bias (Christin 2016). Despite its take-
the “close observation, especially of a sus- up, big data remains an ambiguous term
pected person” (Oxford American Dictionary whose precise definition can vary across
of Current English 1999), and relying on the fields and institutional contexts. Drawing on
unaided senses, new surveillance is more previous definitions (e.g., Laney 2001; Lazer
likely to be applied categorically, deductive, and Radford 2017; Mayer-Schönberger and
Cukier 2013), the working definition of big perspective (DiMaggio and Powell 1983;
data used in this research is that it is a data Meyer and Rowan 1977) questions the assump-
environment characterized by four features: it tion that organizational structures stem from
is vast, fast, disparate, and digital. First, big rational processes or technical imperatives
data analytics involve the analysis of large (Scott 2004). Instead, it highlights the role of
amounts of information, often measured in culture, suggesting organizations operate in
petabytes and involving tens of millions of technically ambiguous fields in which they
observations. Second, big data typically adopt big data analytics not because of empiri-
involves high frequency observations and fast cal evidence that it actually improves effi-
data processing. Third, big data is disparate— ciency, but in response to wider beliefs of what
it comes from a wide range of institutional organizations should be doing with big data
sensors and involves the merging of previ- (Willis, Mastrofski, and Weisburd 2007; see
ously separate data sources. Fourth, big data also Kling 1991 on computerization). In other
is digital. The mass digitization of records words, using big data may confer legitimacy. If
facilitates the merging and sharing of records other institutions are marshalling big data and
across institutions, makes storage and pro- algorithmic predictions for decision-making—
cessing easier, and makes data more efficient rather than relying on human discretion—there
to analyze and search remotely. These four may be institutional pressure to conform.
characteristics are not limited to any one Big data makes possible new forms of clas-
institutional context, and they enable the use sification and prediction using machine learn-
of advanced analytics—such as predictive ing algorithms. Applications of big data
algorithms or network analysis—and complex analytics range from spam and fraud detection
data display—such as topical-, temporal-, or to credit scoring, insurance pricing, employ-
geo-analysis. For purposes of sociological ment decisions, and predictive policing.
research on the topic, this definition shifts the Although much of the appeal of algorithms
focus from features of the data itself to the lies in their replacement of human decision-
social processes that give rise to big data col- making with automated decisions, this study
lection and analysis (i.e., the data environ- highlights the ways in which humans remain
ment). For example, instead of focusing on integral to the analytic process.
the “variety” of big data (one of the “3 Vs” in In the big data environment, individuals
Laney’s [2001] definition), the focus here is contribute to a growing trove of data as they go
on the disparate institutional data sources big about their daily lives (Ball and Webster 2007;
data is gathered from. Garland 2001). Every time people make a pur-
There are numerous theories for why such a chase using a credit card, drive through a toll-
wide range of institutions adopted big data sur- booth, or click on an advertisement online,
veillance as an organizational practice, most of they leave a digital trace (Rule 2007). The
which fit into one of two theoretical perspec- adoption of digital information and communi-
tives: the technical/rational perspective and the cations technologies transformed previously
institutional perspective. Both perspectives are paper files and face-to-face data collection
premised on the notion that organizations are (Marx 1998), making it possible for records
self-interested (Scott 1987), but actors within “initially introduced with limited intentions” to
organizations may adopt big data analytics in be “developed, refined and expanded to deal
response to different pressures. According to with new problems and situations” (Innes
the technical perspective, big data is a means 2001:8). Function creep—the tendency of data
by which organizational actors improve effi- initially collected for one purpose to be used
ciency through improving prediction, filling for another often unintended or unanticipated
analytic gaps, and more effectively allocating purpose (Innes 2001)—is a fundamental com-
scarce resources. By contrast, the institutional ponent of the big data surveillant landscape.
Brayne 981
Big Data Surveillance: In the 1970s, the dominant police patrol

The Case of Policing model was reactive (Reiss 1971), involving
random patrols, rapid responses to 911 calls,
This article provides a case study of policing, and reactive investigations (Sherman 2013).
one of the many organizational contexts in However, practitioners and researchers
which the use of big data surveillance has became increasingly aware that these strate-
grown. More generally, criminal justice sur- gies had little effect on crime, catalyzing a
veillance has increased dramatically in the shift from reactive to more proactive, evidence-
United States in the past four decades. It has based forms of policing, such as hot spots
expanded at all levels, including incarceration policing (Braga and Weisburd 2010; Sher-
(Travis, Western, and Redburn 2014), parole man, Gartin, and Buerger 1989). In 1994,
and probation (Bonczar and Herberman 2014), CompStat—a management model linking
and policing (Carson 2015). For example, the crime and enforcement statistics—was estab-
Violent Crime Control and Law Enforcement lished in New York City (Weisburd et al.
Act of 1994 provided funds to hire 100,000 2003). CompStat quickly spread to other cit-
new police officers, and the Homeland Secu- ies, including Los Angeles in 2002, as a
rity Act of 2002 committed over 17 billion managerial model for identifying crime pat-
dollars for state and local governments to fund terns, quantifying and incentivizing police
local law enforcement agencies (Roush 2012). activity, and directing police resources. The
More recently, federal funds have been tar- attacks on 9/11 spurred the development of
geted at improving and expanding law “intelligence-led policing” (Ratcliffe 2008).
enforcement’s use of technology. For exam- Viewing local law enforcement agencies as
ple, the Smart Policing Initiative—a consor- actors on the front lines of the domestic war
tium of the Bureau of Justice Assistance, local against terror (Waxman 2009), federal agen-
police departments, and researchers—pro- cies provided considerable funding to local
vides federal funds to more than 30 local law law enforcement agencies to collect, analyze,
enforcement agencies (including the LAPD) share, and deploy a wide range of new data.
to support new data-driven practices. In 2008, William Bratton, then-Chief of the
The use of data for decision-making in LAPD (and former Commissioner of the New
criminal justice is not new. In 1928, Ernest York City Police Department) began working
Burgess of the Chicago School designed an with federal agencies to assess the viability of
actuarial model that predicted the probability a more predictive approach to policing.
of parolees’ reoffending (Harcourt 2006). In Today, predictive analytics are used for a
the courts, quantification was embedded into wide range of law enforcement–related activ-
legal practices in the 1970s and 1980s through ities, including algorithms predicting when
sentencing guidelines (Espeland and Vannebo and where future crimes are most likely to
2007). In the past three decades, the criminal occur (Perry et al. 2013), network models
justice system experienced a shift toward predicting individuals most likely to be
“actuarial justice” (Feeley and Simon 1992), involved in gun violence (Papachristos,
in which actors use criteria derived from risk Hureau, and Braga 2013), and risk models
management (Lyon 2003) to estimate proba- identifying law enforcement officers most
bilities of criminal risk (Ericson and Haggerty likely to engage in at-risk behavior (U.S.
1997). That said, although actuarial methods Department of Justice 2001 [2015]).
have existed in corrections and the courts for What explains the proliferation of big-
almost a century (Feeley and Simon 1992; data-driven decision-making in organizations
Harcourt 2006; Lyon 2003), data-driven generally, and law enforcement specifically?
decision-making has become systematically Much like in other institutional domains, it
incorporated into law enforcement practices has the potential to improve both efficiency
only in recent decades. and accountability. It may improve the
prediction and preemption of behaviors by focuses on the experiences and outcomes of

helping law enforcement deploy resources individuals under surveillance, rather than the
more efficiently, ultimately helping prevent surveilling agents themselves. Therefore,
and intercept crimes, thus reducing crime offering an organizational perspective may
rates. Data-driven policing also holds poten- generate new insight about this mediating
tial as an accountability mechanism and level of analysis. Fifth, although there is a
response to criticisms organizations are fac- strong body of work demonstrating that mark-
ing over discriminatory practices. For exam- ing someone in the criminal justice system is
ple, in response to police violence, nationwide consequential for life outcomes and patterns
movements such as Black Lives Matter have of inequality (Becker 1963; Brayne 2014;
brought racial tensions to the forefront of Kohler-Hausmann 2013; Pager 2007; Rios
demands for police reform. Data-driven polic- 2011), we know relatively little about whether
ing is being offered as a partial antidote to and how the marking process has changed in
racially discriminatory practices in police the age of big data. Consequently, there is a
departments across the country (e.g., see dearth of theoretically informed empirical
White House Police Data Initiative 2015). research on the relationship between surveil-
However, although part of the appeal of lance, big data, and the social consequences of
big data lies in its promise of less discretion- the intersection of the two forces.
ary and more objective decision-making (see Many of these gaps in the policing context
Porter’s [1995] work on mechanical objectiv- can be attributed to practical constraints—it is
ity; see also Espeland and Vannebo 2007; difficult for researchers to secure the degree
Hacking 1990), new analytic platforms and of access to police departments necessary to
techniques are deployed in preexisting organ- obtain in-depth qualitative data on day-to-day
izational contexts (Barley 1986, 1996; Kling police practices. Although classic police eth-
1991) and embody the purposes of their crea- nographies exist (e.g., Bittner 1967; Manning
tors (boyd and Crawford 2012; Gitelman and Van Maanen 1978; Wilson 1968), there
2013; Kitchin 2014). Therefore, it remains an have been only a handful of in-depth studies
open empirical question to what extent the within police departments since data analytics
adoption of advanced analytics will reduce became an integral part of police operations
organizational inefficiencies and inequalities, (for early exceptions, see Ericson and Hag-
or serve to entrench power dynamics within gerty 1997; Manning 2011; Moskos 2008;
organizations. The present study sheds light Skogan 2006; Willis et al. 2007). Although
on these questions and helps us understand these studies offer important insight into the
the changing relationship between quantifica- use of CompStat and crime mapping, they
tion, prediction, and inequality. predate algorithmic policing. Consequently,
we still know little about how big data polic-
ing is exercised in practice.
Limitations to Existing This article has three aims. First, it draws
Literature on unique, original data to analyze how a law
There are five key limitations to the existing enforcement organization conducts big data
literature related to big data surveillance. First, surveillance. Second, it forwards an original
most work on actuarialism was written before theoretical framework for understanding the
big data analytics took hold. Second, although changes—and continuities—in surveillance
there is strong theoretical work in surveillance practices associated with the adoption of big
studies, how big data surveillance plays out on data analytics that can be applied to other
the ground remains largely an open empirical institutional domains. Finally, it highlights
question. Third, the majority of sociological implications of big data surveillance for law
research on criminal justice surveillance and social inequality.
Brayne 983
Fieldwork Additionally, I conducted observations on

ride-alongs in patrol cars and a helicopter to
Over the course of two and a half years, I study how officers deploy data in the field. I
conducted a qualitative case study of the Los also shadowed analysts as they worked with
Angeles Police Department (LAPD). The data, observing them responding to queries
LAPD is the third-largest local law enforce- from detectives and supervisors and proac-
ment agency in the United States, employing tively analyzing data for patrol, investiga-
9,947 sworn officers and 2,947 civilian staff tions, and crime analysis.
(Los Angeles Police Department 2017). The To supplement my research within the
Department covers an area of almost 500 LAPD, I interviewed individuals within the
square miles and a population of almost four L.A. County Sheriff’s Department (LASD), as
million people. It consists of four bureaus— it is an integral part of the broader ecosystem
Central, South, Valley, and West—which are of public services in the region. In addition, I
divided into a total of 21 geographic areas. conducted interviews at the Joint Regional
There are also two specialized bureaus, Intelligence Center (JRIC), the “fusion center”
Detective and Special Operations. in Southern California. Fusion centers are
I conducted interviews and observations multiagency, multidisciplinary surveillance
with 75 individuals, and conducted between organizations by state or local agencies that
one and five follow-up interviews with a sub- received considerable federal funding from
sample of 31 individuals to follow how cer- the Department of Homeland Security and the
tain technologies were disseminated and Department of Justice (Monahan and Palmer
information was shared throughout the 2009). JRIC is one of 78 federally funded
Department. Interviewees included sworn fusion centers established across the country
officers of various ranks and civilian employ- in the wake of 9/11. Individuals at JRIC con-
ees working in patrol, investigation, and crime duct data collection, aggregation, and surveil-
analysis. I was able to gain analytic leverage lance in conjunction with other fusion centers
by exploiting different adoption temporalities and agencies, including, but not limited to, the
within and between divisions during my field- Department of Homeland Security (DHS), the
work. Not all divisions adopted big data sur- Federal Bureau of Investigations (FBI), the
veillance at the same time. Rather, there was Central Intelligence Agency (CIA), and Immi-
considerable variation in whether and when gration and Customs Enforcement (ICE). I
different big data technologies were adopted also conducted observations at surveillance
in the area and specialized divisions.1 For industry conferences and interviewed indi-
example, I was able to conduct interviews and viduals working at technology companies that
observations in divisions that were not using design analytic platforms used by the LAPD,
predictive policing and other big data surveil- including Palantir and PredPol, and individu-
lant technologies at the beginning of my field- als working in federal agencies in Washing-
work, but were by the end. This variation ton, DC, to understand how data on criminal
enabled me to observe actual changes in and noncriminal activity are shared across
practice, but also to talk to respondents in agencies. I supplemented my fieldwork with
patrol, investigation, and analysis roles about archival research of law enforcement and mil-
how they interpreted their work changing in itary training manuals and surveillance indus-
light of big data analytics. I also interviewed try literature. Triangulating across various
individuals in specialized divisions—includ- sources of data provided the analytic leverage
ing Robbery-Homicide, Information Technol- necessary to better understand how law
ogy, Records and Identification, Fugitive enforcement uses big data in theory, how they
Warrants, Juvenile, Risk Management, and use it in practice, and how they interpret and
Air Support—and at the Real-Time Crime make meaning out of its changing role in daily
Analysis Center (see Figure 1). operations.
Figure 1. Situation Room at the Real-Time Crime Analysis Center (RACR)

Source: Author’s photo.
Site Selection
consent decree3 with the LAPD from 2001 to
I selected the LAPD as a strategic site for 2009 that mandated, among other things, the
studying big data surveillance because it is an creation and oversight of a new data-driven
agency at the forefront of data analytics. The employee risk management system, TEAMS
LAPD invests heavily in its data collection, II. The legacy of the decree extends beyond
analysis, and deployment capacities, and offers employee risk management; it led to more
international training sessions on how law information sharing and data-driven decision-
enforcement can better harness big data. There- making within the organization in general.
fore, practices within the Department may The second factor is the influence of state
forecast broader trends that may shape other legislative decisions concerning offender
law enforcement agencies in the coming years. management. In the wake of Brown v. Plata4
In addition to being one of the largest law and the associated order to dramatically
enforcement agencies in North America, there reduce the prison population, the California
are additional, contextual reasons why the Legislature passed AB 109, a bill that shifted
LAPD is on the leading edge of data analyt- the responsibility of supervising released
ics. The first factor relates to external pres- non-violent, non-serious, non-sex offenders
sures for transparency and accountability. The from state to local law enforcement and
LAPD was involved in a number of high- county probation officers. It also outsourced
profile scandals in the 1990s, including the compliance checks to local law enforcement
Rampart Scandal2 and the now infamous agencies, including the LAPD and LASD. As
Rodney King beating, which led to investiga- a result, local law enforcement agencies were
tions exposing an expansive web of corrup- responsible for approximately 500 additional
tion, training deficiencies, and civil rights individuals released into L.A. County each
violations within the Department. In response, month. Therefore, they needed a means by
the Department of Justice entered into a which to efficiently stratify the post-release
Brayne 985
community supervision population according officers’ discretionary assessments of risk

to risk, necessitating risk modeling and inter- with quantified risk scores. Second, there is an
agency data integration efforts across the increase in the use of data analytics for predic-
region. tive—rather than reactive or explanatory—
A third relevant factor to the LAPD’s use of purposes. Third, there is a proliferation in
big data is the availability and adoption of new alert-based systems, which facilitates the pas-
data integration technologies. In 2011, the sive, systematic surveillance of a larger num-
LAPD began using a platform designed by ber of individuals than is possible with
Palantir Technologies. Palantir was founded in traditional query-based systems. Fourth, the
2004 and has quickly grown into one of the threshold for inclusion in law enforcement
premier platforms for compiling and analyz- databases is lower, now including individuals
ing massive and disparate data by law enforce- who have not had direct police contact.
ment and intelligence agencies. Originally Finally, previously separate data systems are
intended for use in national defense, Palantir merged into relational systems, making it pos-
was initially partially funded by In-Q-Tel, the sible for the police to use data originally col-
CIA’s venture capital firm. Palantir now has lected in other, non–criminal justice contexts.
government and commercial customers, I offer an original conceptual framework
including the CIA, FBI, ICE, LAPD, NYPD, for understanding the continuities and changes
NSA, DHS, and J.P. Morgan. JRIC (the South- associated with the adoption of big data ana-
ern California fusion center) started using lytics within the police organization. Figure 2
Palantir in 2009, with the LAPD following depicts this framework and illustrates the
shortly after. The use of Palantir has expanded migration of traditional police practices
rapidly through the Department, with regular toward big data surveillance. Each line repre-
training sessions and more divisions signing sents a continuum of surveillance practices,
on each year. It has also spread throughout the from traditional to big data surveillance. The
greater L.A. region: in 2014, Palantir won five shifts in practice do not represent discrete
the Request for Proposals to implement the either/or categories, but rather are better
statewide AB 109 administration program, understood as continuous gradations of vary-
which involves data integration and monitoring degrees between the extreme values of
ing of the post-release community supervision traditional and big data surveillance. The
population. length of the black lines represents the degree
of transformation in surveillance practices
associated with the use of big data. For exam-
Changes Associated with ple, the first two shifts—from discretionary to
Adoption of Big Data quantified risk assessment, and explanatory
Analytics to predictive analytics—are not particularly
transformative; rather, they represent quanti-
To what extent does the adoption of big data fied recapitulations of traditional surveillance
analytics change police surveillance? Based practices. By contrast, the last two shifts—the
on my fieldwork, I argue that in some cases, inclusion of data on individuals with no direct
the adoption of big data analytics is associated police contact, and from institutions typically
with mere amplifications in prior surveillance not associated with crime control—represent
practices, but in others, it is associated with fundamental transformations in surveillance
fundamental transformations in surveillance activities. The shift from query-based systems
activities and daily operations. I empirically to automated alerts is a moderate shift, repre-
demonstrate five key ways in which the adop- senting, in part, an elaboration of existing
tion of big data analytics is associated with practices, and in part, a new surveillance
shifts in surveillance practices to varying strategy. I will analyze each of these shifts in
degrees. First, law enforcement supplements the following sections.
Figure 2. Migration of Traditional Police Practices toward Big Data Surveillance
The shift from traditional to big data sur- and the adoption of big data analytics facili-
veillance is associated with a migration of tated and accelerated this shift.
law enforcement operations toward intelli-
gence activities. The basic distinction between
The Quantification of Individual Risk
law enforcement and intelligence is as fol-
lows: law enforcement typically becomes The first shift in police practice is the quantifi-
involved once a criminal incident has cation of civilians according to risk. Quantified
occurred. Legally, the police cannot under- knowledge is supplementing officers’ experi-
take a search and gather personal information ential knowledge through the implementation
until there is probable cause. Intelligence, by of a new point system: Operation LASER (Los
contrast, is fundamentally predictive. Intelli- Angeles’ Strategic Extraction and Restoration
gence activities involve gathering data; iden- program). The program began in 2011 and was
tifying suspicious patterns, locations, activity, funded through the Smart Policing Initiative, a
and individuals; and preemptively interven- national initiative encouraging local police
ing based on the intelligence acquired. Before departments and researchers to use evidence-
discussing the findings, one caveat is worth based, data-driven tactics. The strategy
noting: the migration of law enforcement includes place-based and offender-based
toward intelligence was in its nascency before models. The offender-based strategy was
the use of big data, in part due to Supreme implemented in a low-income, historically
Court decisions dismantling certain criminal high-crime division in South Bureau. It is pre-
protections. Technically, the Fourth Amend- mised on the idea that a small percentage of
ment makes unreasonable searches and sei- high-impact players are disproportionately
zures illegal in the absence of probable cause. responsible for most violent crime. Therefore,
However, in practice, decisions such as Terry identifying and focusing police resources on
v. Ohio and Whren v. United States made it the “hottest” individuals should be an efficient
easier for law enforcement to circumvent the means of reducing crime.5
barrier of probable cause, ultimately contrib- The strategy begins by plotting crimes in
uting to the proliferation of pretext stops. In the division. The Crime Intelligence Detail
other words, the Supreme Court’s disman- (CID), which is composed of three sworn
tling of probable cause catalyzed the migra- officers and a civilian crime analyst, identi-
tion of law enforcement toward intelligence, fies a problem crime, which in this division is
Brayne 987
often armed robbery. Next, the CID shifts that citation can, the sum of all information
their unit of analysis from crimes to individu- can build out what is needed.” In addition to
als. They gather intelligence daily from inputting the content of the cards, a captain
patrols, the Parole Compliance Unit, field explained there is an incentive to simply “get
interview (FI) cards (police contact cards), them in the system” as entities that future data
traffic citations, release from custody forms, points can be linked to.
crime and arrest reports, and criminal histo- Because point values are largely based on
ries to generate a list of “chronic offenders,” police contact, an important question emerges:
who are each assigned a point value and given What are grounds for police contact? Merely
a numerical rank according to that value. being identified as a chronic offender does
Individuals are assigned five points for a vio- not constitute reasonable suspicion or proba-
lent criminal history, five points for known ble cause. However, used in conjunction with
gang affiliation, five points for prior arrests Palantir, FIs represent a proliferation of data
with a handgun, and five points if they are on from police–civilian interactions that law
parole or probation. One officer explained: enforcement does not need a warrant to col-
lect. When I asked an officer to provide
We said ok, we need to decide who’s the examples of why he stops people with high
worst of the worst . . . we need something to point values, he replied:
pull them apart. So this was the important
one, and this is really what gives the impor- Yesterday this individual might have got
tance of FI-ing someone [filling out a field stopped because he jaywalked. Today he
interview card] on a daily basis instead of mighta got stopped because he didn’t use
just saying, okay, I saw that guy hanging his turn signal or whatever the case might
out, I’m gonna give him two weeks and I’ll be. So that’s two points . . . you could con-
go FI him again. It’s one point for every duct an investigation or if something seems
police contact.6 out of place you have your consensual
stops.7 So a pedestrian stop, this individual’s
As illustrated in Figure 3, FI cards include walking, “Hey, can I talk to you for a
personal information such as name, address, moment?” “Yeah what’s up?” You know,
physical characteristics, vehicle information, and then you just start filling out your card
gang affiliations, and criminal history. On the as he answers questions or whatever. And
back of the card, there is space for officers to what it was telling us is who is out on the
include information on persons with the sub- street, you know, who’s out there not neces-
ject and additional intelligence. sarily maybe committing a crime but who’s
FIs are key intelligence tools for law active on the streets. You put the activity of
enforcement and were one of the first data . . . being in a street with maybe their violent
sources integrated into Palantir. When entered background and one and one might create
into the system, every FI is tagged with the the next crime that’s gonna occur.
time, date, and geo-coordinates. Officers are
trained to pull out an FI card and “get a The point system is path dependent; it gener-
shake” as soon as they interact with someone ates a feedback loop by which FIs are both
in the field. One supervisor described how he causes and consequences of high point val-
uses it “to tag all the personal information I ues. An individual having a high point value
can get . . . these things come into play later is predictive of future police contact, and that
on in ways you could never even imagine.” police contact further increases the individu-
Similarly, a software engineer explained how al’s point value.
little pieces of data that might seem unsuspi- The CID also creates work-ups, referred to
cious at the time of collection can eventually as “Chronic Violent Crime Offender Bulle-
be pulled together to create useful intelli- tins.” Individuals on these bulletins are not
gence: “It’s a law enforcement system where necessarily “wanted” nor do they have
Figure 3. Field Interview (FI) Cards

Source: LAPD.
outstanding warrants for their arrest. Rather, it case and specific criminal networks, but by
is an “information only” bulletin that includes creating a list and disseminating bulletins, the
physical descriptors and oddities, gang affilia- point system and associated bulletins broad-
tion, criminal history, parole/probation status, ens previously particularized police familiar-
vehicles, frequented areas, and law enforce- ity of individuals on the street.
ment contacts. The goal of these bulletins is to Ideally, one officer explained, they could
give officers what they refer to as “situational put one officer on every individual on the list
awareness.” Officers previously had to rely and “odds are [you’re] probably going to find
exclusively on their direct knowledge of a them committing another crime.” However,
Brayne 989
the police operate in an organizational context Shift from Reactive to Predictive

with resource constraints; respondents fre- Analytics
quently referenced budget cuts and personnel
shortages. Therefore, instead of having one Historically, policing was mostly reactive.
officer on every chronic offender, officers Patrol officers used to spend much of their
engage in what I term “stratified surveillance”: time “chasing the radio.” In the early 1980s,
differentially surveilling individuals according faced with evidence that reactive strategies
to their risk score. An officer explained: were ineffective at reducing crime, there was
a paradigm shift toward more proactive, problem-
[We] utilize undercover operations, or oriented policing strategies, including hot
undercover units and . . . then sit our surveil- spots policing. Predictive policing is an exten-
lance on some of the higher point offenders sion of hot spots policing, made possible by
and just watch them on a daily basis. . . . And the temporal density of big data (i.e., high-
you start building either, you know, there’s frequency observations). In 2012, the LAPD
two ways of looking at it. Either kind of began using software designed by PredPol, a
conducting your investigation to see if predictive policing company. PredPol uses a
maybe there was a crime that had just been proprietary algorithm8 predicated on the near-
committed. Or, “We know who you are, you repeat model, which suggests once a crime
know, I just called you Johnny, I’ve never occurs in a location, the immediate surround-
really met you before, but I know who you ing area is at increased risk for subsequent
are now,” so maybe it’s put in his mind, “Oh, crime. PredPol uses three types of inputs—
they’re on to me, they know who I am.” past type, place, and time of crime—to iden-
tify areas where future crime is most likely to
This excerpt sheds light on the multiple purposes occur. Predictive policing is expanding rap-
of stratified surveillance, including ongoing idly within the Department; as of March 2015,
intelligence gathering and deterrence through it had disseminated to 10 divisions.9
signaling to individuals on the street that they are Officers receive printouts at the beginning
being tracked by law enforcement. of their shift that show 500 by 500 square-
Why did the police turn to the point system foot boxes overlaying small areas of division
in this division? In the words of one officer, maps. Patrol officers are encouraged to spend
time in predictive boxes, a strategy referred to
The code of federal regulations. They say as “risk-based deployment.” Deployment is
you shouldn’t create a—you can’t target based on available time, such as when offic-
individuals especially for any race or I for- ers are not responding to calls or “booking a
get how you say that. But then we didn’t body.” Officers record their self-reported
want to make it look like we’re creating a minutes in the predictive boxes on their in-car
gang depository of just gang affiliates or computers. Although “data drives deploy-
gang associates. . . . We were just trying to ment,” what the police do once in the predic-
cover and make sure everything is right on tive box, and how long they stay there,
the front end. remains within their discretion.
One supervisor explained that by relying
Other respondents echoed this sentiment, on data, rather than human interpretation of
explaining the strategy was adopted, in part, crime patterns, it helps him deploy his
as a legal compliance mechanism. resources more efficiently:10
The point system is a form of quantified
policing, but it is not dramatically different There’s an emotional element to it, and you
from its discretionary predecessor. As indi- think right now with crime being this low, a
cated by the low degree of transformation in cluster could be three or four crimes. Clus-
Figure 2, it is largely a quantified recapitula- ters used to be 10, 12 crimes. Now three or
tion of traditional surveillance practices. four and they jump on it, you know. So,
there could be overreaction. Because, in other ways, I assumed cars’ locations would
there’s, you know, I mean it’s a human be tracked automatically. When I asked the
doing it. And they cannot sort out what’s officer why he manually placed himself at the
noise. scene, he explained that although every police
unit was equipped with an automatic vehicle
Officers were quick to emphasize the contin- locator (AVL) that pings the vehicle’s location
ued importance of their own expertise. When every five seconds, they were not turned on
discussing predictive policing, most patrol because of resistance from LAPD union
officers said some version of the following representatives.11
statement made by a sergeant on a ride-along:
“I already know where the crime’s at.” Part of
Shift from Query-Based to Alert-
this sentiment may stem from officers’ concern
Based Systems
that the use of algorithms represents a form of
deskilling, devaluing their local and experien- The shift from query-based to alert-based sys-
tial knowledge and threatening their profes- tems represents, in part, an extension of exist-
sional autonomy. In that vein, one captain ing practices and, in part, a fundamental
described a typical exchange with his officers: transformation in surveillance activities. By
“query-based systems,” I mean databases to
They’re like, “You know what, I know which users submit requests for information in
where the crime’s occurring.” . . . And I the form of a search. A familiar example of a
show them the forecast and they say, “Okay, query is when a police officer runs a license
so [at intersection], I know there are crimes, plate during a traffic stop. In alert-based sys-
I could have told you that. I’ve been work- tems, by contrast, users receive real-time noti-
ing here 10 years! There’s always crime fications (alerts) when certain variables or
there.” I go, “Okay, you’re working here 10 configurations of variables are present in the
years on that car, why is there still crime data. The shift from query-based to alert-based
there if you’re so knowledgeable?” systems—which is made possible by high fre-
quency data collection—has implications for
Despite some within-department conflict over the relational structure of surveillance.
their efficacy, PredPol outputs still informed Consider the following example: all war-
where some officers drove during their rants in L.A. County can be translated into
uncommitted time. For example, when driv- object representations spatially, temporally,
ing back from booking an individual at the and topically in Palantir. Through tagging,
station, a sergeant I was with decided to drive users can add every known association that
to an area not known to him for being high- warrant has to people, vehicles, addresses,
crime, because he thought the location of the phone numbers, documents, incidents, cita-
PredPol box was odd and he wanted to see tions, calls for service, ALPR readings, FIs,
what was going on there. In other words, pre- and the like. Officers and analysts can then set
dictive policing outputs sometimes—but not up alerts by putting a geo-fence around an area
always—acted as a substitute for localized and requesting an alert every time a new war-
experiential knowledge. rant is issued within the area. Warrants are but
A related but distinct reason why officers one example; users can request alerts for any
contest predictive policing is because they data points related to the entity they are inter-
believe it places officers themselves under ested in (e.g., calls for service, involvement in
greater surveillance. For example, when we or witnesses to a traffic accident, ALPR [auto-
arrived at a crime scene on my first ride-along, matic license plate reader] readings, FIs, and
I was surprised to see an officer manually type so on). Using a mechanism in Palantir similar
our location on his laptop. Considering how to an RSS feed, officers can be automatically
technologically advanced the Department was notified of warrants or events involving
Brayne 991
specific individuals (or matching descriptions plate, type of weapon, cause of death, motive,
of individuals), addresses, or cars directly on type of crime, and M.O., such as “what kind
their cell phone. Prior to automated alerts, law of bindings were used . . . or was there torture
enforcement would know individuals’ real- involved? What type of trauma has occurred?
time location only if they were conducting 1:1 Was there, you know, was there some type of
surveillance, received a tip, or encountered symbolic activity?” Although the matching
them in person. process is automated, decisions about what
Real-time notifications can be useful in parameters the system matches on remain
operational planning. An interviewee who within the discretion of individuals at ViCAP,
worked at the fusion center described how if a unit of the FBI.
he is about to conduct a search of a house, he That said, the use of alerts represents not just
can draw a fence around the house and receive a scaling up of existing police practices, but also
notifications about risks such as whether a a fundamental transformation in how patrol
known gang associate lived in the home, if officers and investigators generate case knowl-
there was a gun registered in the house next edge. Under the traditional surveillance model,
door, or if there was a warrant for assault with alerts about hot incidents and suspects are sent
a deadly weapon issued down the street. out from dispatch centers. However, by exploit-
Alerts can also be used to break down infor- ing variation in divisions that did and did not
mation silos within the Department. LAPD’s use place- and person-based predictive polic-
jurisdiction is almost 500 square miles. There- ing—and divisions that started using big data
fore, individual detectives may not able to during the course of my fieldwork—I was able
connect crime series that occur across different to observe the automation of alerts and relative
divisions. One captain explained: lack of human intermediation in broadcasting
out these alerts or conducting data grazing.
Let’s say I have something going on with It is worth noting that alert-based systems
the medical marijuana clinics where they’re are supplementing, rather than replacing,
getting robbed. Okay? And it happens all query-based systems. Searches are still criti-
over, right? But I’m a detective here in cal features of law enforcement information
[division], I can put in an alert to Palantir systems. In fact, one of the transformative
that says anything that has to do with medi- features of big data systems is that queries
cal marijuana plus robbery plus male, black, themselves are becoming data. One detective
six foot. explained:
He continued, “I like throwing the net out I queried the system a certain way and then
there, you know? Throw it out there, let it another person queried the system a certain
work on it while you’re doing your other stuff, way . . . we were looking for something very
you know?” Relatedly, an interviewee in similar in our query, and so even though the
Robbery-Homicide Division described a pilot data may not have connected the two, the
project in which automated data grazing can queries were similar. Yeah, so then it will be
flag potential crime series that span jurisdic- able to say hey, listen, there’s an analyst in
tional boundaries and are therefore difficult San Francisco PD that ran a very similar
for any one person to identify. He said, “You query as to yours and so you guys might be
could get an alert that would say, you know looking for the same thing.
what, your case is pretty similar to this case
over in Miami.” If the case reaches a “merit A different detective explained how when he
score” (i.e., a threshold at which a certain con- searches someone’s name in one national sys-
figuration of variables is present), the system tem, he can see the number of times that name
flags the cases as similar. The system matches has been queried by other people. When I
on fields such as suspect description, license asked why he would want to know how many
times someone’s name has been queried, he surveillance network,” individuals do not
replied that “if you aren’t doing anything need to have direct law enforcement contact;
wrong,” the cops are not going to be looking they simply need to have a link to the central
you up very many times over the course of person of interest. Once individual relation-
your life. He continued: “Just because you ships are inputted and social networks are
haven’t been arrested doesn’t mean you built into the system, individuals can be
haven’t been caught.” In other words, in aud- “autotracked,” meaning officers can receive
itable big data systems, queries can serve as real-time alerts if individuals in the network
quantified proxies for suspiciousness. come into contact with the police or other
government agencies again.
The Automatic License Plate Reader
Lower Database Inclusion Thresholds
(ALPR) is another example of a low-threshold
The last two shifts in practice represent the “trigger mechanism” (Tracy and Morgan
most fundamental transformations in surveil- 2000) that results in more widespread inclu-
lance activities. Law enforcement databases sion in a database. ALPRs are dragnet surveil-
have long included information on individuals lance tools; they take readings on everyone,
who have been arrested or convicted of crimes. not merely those under suspicion. Cameras
More recently, they also include information mounted on police cars and static ALPRs at
on people who have been stopped, as evi- intersections take two photos of every car that
denced by the proliferation of stop-and-frisk passes through their line of vision—one of
databases. However, as new data sensors and the license plate and one of the car—and
analytic platforms are incorporated into law records the time, date, and GPS coordinates
enforcement operations, the police increas- (see Figure 5). Law enforcement–collected
ingly utilize data on individuals who have not ALPR data can be supplemented with pri-
had any police contact at all. Quotidian activi- vately collected ALPRs, such as those used
ties are being codified by law enforcement by repossession agents. ALPR data give the
organizations. One way this is occurring is police a map of the distribution of vehicles
through network analysis. Figure 4 is a de- throughout the city and, in some cases, may
identified mockup I asked an employee at enable law enforcement to see an individual’s
Palantir to create based on a real network typical travel patterns. For example, an ana-
diagram I obtained from an officer in the lyst used ALPR data to see that a person of
LAPD. The person of interest, “Guy Cross,” is interest was frequently parked near a particu-
an individual with a high point value. An lar intersection at night, explaining to me that
LAPD officer explained, “with the [Palantir] this intersection is likely near that person’s
system . . . I click on him and then [a] web residence or “honeycomb” (hideout).
would spread out and show me the phones that There are several ways to use ALPR data.
he’s associated with and the cars.” One is to compare them against “heat lists” of
Radiating out from “Guy Cross,” who has outstanding warrants or stolen cars. Another
direct police contact, are all the entities he is strategy is to place a geo-fence around a loca-
related to, including people, cars, addresses, tion of interest in order to track cars near the
and phone numbers. Each line indicates how location. For example, after a series of copper
they are connected, such as by being a sib- wire thefts in the city, the police found the car
ling, lover, cohabiter, co-worker, co-arrestee, involved by drawing a radius in Palantir
or listed on a vehicle registration. The net- around the three places the wire was stolen
work diagram illustrates only one degree of from, setting up time bounds around the time
separation, but networks can expand outward they knew the thefts occurred at each site, and
to as many degrees of separation as users querying the system for any license plates
have information and can tie in with other captured by ALPRs in all three locations dur-
networks. To be in what I call the “secondary ing those time periods.
Brayne 993
Figure 4. Network in Palantir

Source: Palantir Technologies.
However, the most common use of ALPRs inclusion of ALPR data in the Palantir plat-
is simply to store data for potential use during form offer clear examples in which individu-
a future investigation. For example, one ser- als with no criminal justice contact are
geant described a “body dump” (the disposal included in law enforcement databases.
of a dead body) that occurred in a remote
location near a tourist attraction where there
Institutional Data Systems Are
was an ALPR. By searching ALPR readings
Integrated
within the time frame that police determined
the body was disposed, they captured three Finally, the proliferation of digitized records
plates—one from Utah, one from New Mex- makes it possible to merge data from previ-
ico, and one from Compton. The sergeant ously separate institutional sources into an
explained that assuming the Compton car was integrated, structural system in which dispa-
most likely to be involved, they ran the plate, rate data points are displayed and searchable
saw the name it was registered under, searched in relation to one another, and individuals can
the name in CalGang (gang database), saw be cross-referenced across databases. This
that the individual was affiliated with a gang integration facilitates one of the most trans-
currently at war with the victim’s gang, and formative features of the big data landscape:
used that information to establish probable the creep of criminal justice surveillance into
cause to obtain a search warrant, go to the other, non–criminal justice institutions. Func-
address, find the car, search the car for trace tion creep—the phenomenon of data origi-
evidence, and arrest the suspect. nally collected for one purpose being used for
Although LAPD and Palantir employees another—contributes to a substantial increase
frequently told me that to be “in the system,” in the data police have access to. Indeed, law
a person needed to have had criminal justice enforcement is following an institutional data
contact, the use of network diagrams and the imperative (Fourcade and Healy 2017),
Figure 5. Plotted ALPR Readings

Source: Palantir Technologies.
securing routine access to a wide range of screen [Figure 6] where you log on there’s
data on everyday activities from non-police another icon about another database that’s
databases. Before Palantir, officers and ana- been added . . . they now have been working
lysts conducted predominantly one-off with Palantir to develop a database of all the
searches in “siloed” systems: one to look up a foreclosure properties . . . they just went out
rap sheet, another to search a license plate, and found some public data on foreclosures,
another to search for traffic citations, and so dragged it in, and now they’re mapping it
on. The Palantir platform integrates disparate where it would be relative to our crime data
data sources and makes it possible to quickly and stuff.
search across databases.
Expressing his faith in the Department’s The Palantir platform allows users to
investment in the platform, one captain told organize and visualize structured and unstruc-
me, “We’ve dumped hundreds of thousands tured data content (e.g., e-mails, PDFs, and
into that [Palantir]. . . . They’re gonna take photos) through “tagging,” the process of
over the world. . . . I promise you they’re labeling and linking objects and entities to
gonna take over the world.” During my field- identify emerging relationships. By tagging
work, there were more than 1,300 trained Pal- objects and entities—including, but not lim-
antir users in the region. New data sources are ited to, persons, phone numbers, addresses,
incorporated regularly, including information documents such as law enforcement reports
collected by the Department, external data col- or tips and leads, and calls for service—and
lected by other government agencies, and pri- displaying the data spatially, temporally, or
vately collected data the Department purchases. topically, users can see data points in context
Remarking on the growth, one captain said: and make new connections.
Another important interagency data inte-
I’m so happy with how big Palantir got. . . . gration effort is the initiative to create an
I mean it’s just every time I see the entry Enterprise Master Person Index (EMPI) in
Brayne 995
Figure 6. Palantir Homepage

Source: LAPD.
L.A. County. L.A. EMPI would create a single university camera feeds; rebate data such as
view of a client across all government systems address information from contact lens rebates;
and agencies; all of an individual’s interac- and call data from pizza chains, including
tions with law enforcement, social services, names, addresses, and phone numbers from
health services, mental health services, and Papa Johns and Pizza Hut. In some instances,
child and family services would be merged it is simply easier for law enforcement to pur-
onto one unique ID. Although interviewees chase privately collected data than to rely on
working in the county’s information technol- in-house data because there are fewer consti-
ogy office stated the explicit motivation tutional protections, reporting requirements,
behind the initiative was to improve service and appellate checks on private sector surveil-
delivery, such initiatives effectively serve the lance and data collection (Pasquale 2014).
latent function of extending the governance Moreover, respondents explained, privately
and social control capacities of the criminal collected data is sometimes more up-to-date.
justice system into other institutions. It is worth noting that such data integration
I encountered several other examples of is not seamless. Merging data from different
law enforcement using external data originally sources and creating interoperable systems is
collected for non–criminal justice purposes, part of the invisible labor that makes big data
including data from repossession and collec- analytics possible. One civilian employee
tions agencies; social media, foreclosure, and lamented, “You always forget about the data
electronic toll pass data; and address and guy . . . [the] guy that does all the dirty work
usage information from utility bills. Respond- is usually forgotten. I’m that guy.” Moreover,
ents also indicated they were working on efforts at acquiring new data were not received
integrating hospital, pay parking lot, and evenly throughout the Department. A vocal
minority of interviewees explained they did LAPD such as consent decree mandates, but
not believe leadership was fully thinking also broader isomorphic shifts (DiMaggio
through the implications of collecting such a and Powell 1983) toward use of predictive
wide range of new data. A civilian employee analytics across organizational fields. In ana-
complained about the seduction of new tech- lyzing how the LAPD uses big data in their
nology, saying: “We tend to just say, ‘Let’s surveillance activities, I argue it is both con-
just go for the sexy tool,’ right? . . . We just tinuous and transformative: the adoption of
never think about to what end.” He added, advanced analytics facilitates amplifications
of existing surveillance practices, but also
Maybe we shouldn’t collect this informa- fundamentally changes daily operations. I
tion. Maybe we shouldn’t add consumer describe five key shifts in practice associated
information. Maybe we shouldn’t get every- with adoption of big data analytics, each of
body’s Twitter feed in. . . . All we’re doing which falls on different points on the contin-
right now is, “Let’s just collect more and uum between law enforcement and intelli-
more and more data and something good gence activities. Whereas the person-based
will just happen.” And that’s I think that’s point system and place-based predictive algo-
kind of wishful thinking. rithms are largely quantified recapitulations
of “traditional” (Marx 2016) surveillance, the
Law enforcement’s adoption of data and inter-institutional integration of data and pro-
analytic tools without a specific technical pur- liferation of dragnet surveillance practices—
pose also surfaced during my time at surveil- including the use of data on individuals with
lance industry conferences. When I first no direct police contact and data gathered
observed software representatives interact from institutions typically not associated with
with potential law enforcement customers, I crime control—represent fundamental trans-
assumed law enforcement would tell software formations in the very nature of surveillance.
representatives their needs and ask how the Big data and associated new technological
products could help them achieve their opera- tools permit unprecedentedly broad and deep
tional goals. However, the inverse pattern was surveillance. By broad, I mean surveillance
more frequently the case: software representa- capable of passively tracking a large number
tives demonstrated the use of their platform in of people. Information that would previously
a non-law enforcement—usually military— have been unknown to law enforcement
context, and then asked local law enforcement because it was too labor intensive to retrieve
whether they would be interested in a similar is more readily available, and individuals
application in their local context. In other previously unknown to law enforcement are
words, instead of filling analytic gaps or tech- now part of the corpus through dragnet sur-
nical voids identified by law enforcement, veillance and data collection by non–criminal
software representatives helped create new justice organizations. By deep, I mean able to
kinds of institutional demand. track one individual more intensively over
time, including across different institutional
settings. The intended and unintended social
Discussion: Big Data as consequences of new surveillance practices
Social have implications for social inequality, law,
This article draws on unique data to offer an and future research on big data surveillance in
on-the-ground account of big data surveil- other fields.
lance. Providing a case study of the Los
Angeles Police Department (LAPD), it offers
Implications for Social Inequality
insight into the reasons why the use of big
data analytics spread throughout the organi- The role of the criminal justice system in the
zation, including factors particular to the reproduction of inequality has received
Brayne 997
considerable attention in the literature (for a On the other hand, this research highlights
review, see Laub 2014). However, the impact how data-driven surveillance practices may
of the use of big data surveillance on inequal- be implicated in the reproduction of inequal-
ity remains an open empirical question. The ity in at least three ways: by deepening the
use of new surveillant technologies could surveillance of individuals already under sus-
either reduce or reinforce existing inequali- picion; widening the criminal justice dragnet
ties. By contributing new insights into how unequally; and leading people to avoid “sur-
big data plays out on the ground in policing, veilling” institutions that are fundamental to
this research helps adjudicate between the social integration. First, mathematized police
two possibilities. practices serve to place individuals already
On the one hand, big data analytics may be under suspicion under new and deeper forms
a means by which to ameliorate persistent of surveillance, while appearing to be objec-
inequalities in policing. Data can be mar- tive, or, in the words of one captain, “just
shaled to replace unparticularized suspicion of math.” Despite the stated intent of the point
racial minorities and human exaggeration of system to avoid legally contestable bias in
patterns with less biased predictions of risk. police practices, it hides both intentional and
Social psychological research demonstrates unintentional bias in policing and creates a
that humans are “cognitive misers” (Fiske and self-perpetuating cycle: if individuals have a
Taylor 1991) who rely on shortcuts—such as high point value, they are under heightened
the conflation of blackness and criminality surveillance and therefore have a greater like-
(Quillian and Pager 2001)—to understand the lihood of being stopped, further increasing
world. Because stereotypes have the most cog- their point value. Such practices hinder the
nitive utility in the face of incomplete informa- ability of individuals already in the criminal
tion, if big data can be utilized to provide more justice system from being further drawn into
complete information, it may lead officers to the surveillance net, while obscuring the role
rely less on stereotypes about race and class. In of enforcement in shaping risk scores. More
that sense, the use of big data may serve to over, individuals living in low-income,
reduce hyper-surveillance of minority neigh- minority areas have a higher probability of
borhoods and the consequent erosion of com- their “risk” being quantified than those in
munity trust (Sampson and Bartusch 1998). more advantaged neighborhoods where the
Big data may also be used to “police the police are not conducting point-driven sur-
police.” Digital trails are susceptible to over- veillance. Importantly, this quantified modal-
sight. Therefore, aggregating data on police ity of social control has consequences that
practices may shed light on systematic patterns reach beyond individuals with high point
and institutional practices previously dis- values. Field interview cards record informa-
missed as individual-level bias, ultimately pro- tion not only about the individual in question,
viding an opportunity to increase transparency but also information on people the individual
and accountability. However, transparency and is with. The exponential capture of personal
accountability do not flow automatically from data beyond the primary individuals involved
big data policing. Data-based surveillance is in the police encounter is a strategic means of
less visible than traditional street policing channeling more individuals into the system,
methods (Joh 2016) and is embedded in power thus facilitating future tracking.
structures. The outcomes of struggles between Whereas the point system is consequential
law enforcement, civilians, and information for racial and class inequality, if not imple-
technology companies—who increasingly mented effectively,12 place-based algorithms
own the storage platforms and proprietary may exacerbate neighborhood inequalities.
algorithms used in data analysis—will play a Historical crime data are incomplete; esti-
role in determining whether big data policing mates of unreported crime range from less
will ameliorate or exacerbate inequalities. than 17 percent to over 68 percent, depending
on the offense (Langton et al. 2012). More deployed based on department crime statistics
over, crime data are not missing at random. (i.e., to higher crime areas), raising similar
Therefore, there is systematic bias in the train- questions to those posed earlier about unequal
ing data: crimes that take place in public places enforcement and reporting practices along
are more visible to police and therefore more lines of race, class, and neighborhood. In that
likely to be recorded (Duster 1997); individu- sense, ALPR datasets are investigatory tools
als and groups who do not trust the police are for law enforcement, but they are dispropor-
less likely to report crimes (Sampson and Bar- tionately “populated by the movements of
tusch 1998); and police focus their attention particular groups” (Renan 2016:1059). Simi-
and resources on black communities at a dis- larly, the ability to build out secondary sur-
proportionately high rate relative to drug use veillance networks in Palantir has implications
and crime rates (Beckett et al. 2005). These for inequality, as minority individuals and
social dynamics inform the historical crime individuals in poor neighborhoods have a
data that are fed into the predictive policing higher probability of being in the primary
algorithm. However, once they are inputted as (and thus secondary) surveillance net than do
data, the predictions appear impartial; human people in neighborhoods where the police are
judgment is hidden in the black box (Pasquale not conducting point-driven or other data-
2014) under a patina of objectivity. intensive forms of policing.
Unchecked predictions may lead to an How are unequal mechanisms for inclusion
algorithmic form of confirmation bias, and in the surveillance net consequential for social
subsequently, a misallocation of resources. inequality? Recall the operative theory from a
They may justify the over-policing of minor- detective that if people are not doing anything
ity communities and potentially take away wrong, the police should not be looking them
resources from individuals and areas invisible up many times over the course of their lives.
to data collection sensors or subject to sys- However, queries are not raw data (Gitelman
tematic underreporting. Put differently, the 2013); rather, they are, in part, a product of
mechanisms for inclusion in criminal justice enforcement practices. Empirical research
databases determine the surveillance patterns consistently demonstrates that stop-and-query
themselves. Predictive models are performa- patterns are unequally distributed by race,
tive, creating a feedback loop in which they class, and neighborhood (Epp, Maynard-
not only predict events such as crime or Moody, and Haider-Markel 2014). Quantified
police contact, but also contribute to their practices may thus serve to exacerbate inequal-
future occurrence.13 ities in stop patterns, create arrest statistics
Second, new digitized surveillance prac- needed to justify stereotypes, and ultimately
tices broaden the scope of people law enforce- lead to self-fulfilling statistical prophecies
ment can track. This can be understood as a (Merton 1948). Moreover, as police contact is
new form of “net widening” (Cohen 1985), the entry point into the criminal justice system,
effectively widening the criminal justice the digital feedback loops associated with pre-
dragnet, and doing so unequally. Consider dictive policing may ultimately justify the
ALPRs, one of the primary means of tracking growth and perpetuation of the carceral state.
people without police contact. Even though One might argue that if you have nothing
ALPRs are dragnet surveillance tools that to hide, being included in police databases is
collect information on everyone, rather than nothing to fear. However, once individuals
merely those under suspicion, the likelihood are in the primary or secondary surveillance
of being inputted into the system is not ran- net, they can become intelligence targets and
domly distributed. Crime and enforcement linked to future data points. By virtue of
patterns lead to unequal data capture across being in the system, individuals are more
individuals, groups, and the city. ALPRs are likely—correctly or incorrectly—to be
Brayne 999
identified as suspicious. Consider how the that involvement with the criminal justice
quantification of previous stops in the point system is highly stratified, the negative con-
system serves as justification for future stops, sequences of system avoidance—for future
or how the detective suggested a man was health outcomes, financial self-sufficiency,
suspicious because his name had been que- acquisition of human capital, and upward
ried multiple times. Using a series of data economic mobility—will be similarly dispro-
points to reconstruct an individual’s inten- portionately distributed, thus exacerbating
tions and behaviors, whether incriminating or any preexisting inequalities for an expanding
exculpatory, rests on the assumption of an group of already disadvantaged individuals.
infallible state and of actors who run searches This research builds on work on labeling
without error or prejudice. Much like in DNA theory, extending the relationship between the
databases (Duster 2005; Hindmarsh and stigma of criminal justice contact and ine-
Prainsack 2010; Lynch et al. 2008), in order quality into the digital age (Becker 1963;
to be a hit, one has to be in the database in the Brayne 2014; Goffman 2014; Goffman 1963;
first place. Unequal rates of database inclu- Kohler-Hausmann 2013; Lyon 2006; Pager
sion can have real consequences—African 2007; Rios 2011; Stuart 2016; Wakefield and
Americans are seven times more likely than Wildeman 2013; Western and Pettit 2005).
whites to be wrongly convicted of murder The integration of records may effectively
(Gross, Possley, and Stephens 2017). There- extend the mark of a criminal record (Pager
fore, analyzing the feeder mechanisms by 2007)—or merely the mark of criminal jus-
which individuals are channeled into criminal tice contact—into other institutions. This
justice databases helps us better understand creep of data across institutional contexts can
how inequalities produced by differential sur- lead to “cascading disadvantages” (Pasquale
veillance may be magnified as individuals are 2014:218; see also Gandy 2009). As individu-
processed through the criminal justice als leave more digital traces, a “new economy
system. of moral judgement” (Fourcade and Healy
Third, integrating external, non-police 2017:24) becomes possible. Building on
data into the law enforcement corpus has Weber’s concept of class situation, Fourcade
unanticipated consequences. Although inte- and Healy (2013) argue that institutions now
grated systems create new opportunities for use actuarial techniques to track, sort, and
service delivery, they also make surveillance categorize individuals into “classification sit-
possible across formerly discrete institutional uations” with different rewards and punish-
boundaries. By using other institutions’ data, ments attached. These classification situations
criminal justice surveillance practices may differentially shape life chances (see also
have a chilling effect, deterring people from Bowker and Star 2000). For example, classi-
using such institutions and thereby subverting fying individuals as low or high risk for
their original mandates. For example, indi- crime, terrorist activity, loan default, or medi-
viduals wary of criminal justice surveillance cal conditions structures not only if and how
may avoid interacting with important institu- they will be surveilled, but also their life
tions where they would leave a digital trace. chances more generally. This research begins
Previous research demonstrates that individu- to account for how the marking process may
als involved in the criminal justice system be changing in the age of digitized policing,
(i.e., who have been stopped by police, and how the big data environment creates
arrested, convicted, or incarcerated) engage potentially farther-reaching digitized collat-
in “system avoidance,” systematically avoid- eral consequences of involvement in the
ing surveilling institutions such as medical, criminal justice system.
financial, educational, and labor market insti- In summary, the burden of new surveillance
tutions that keep formal records (i.e., put practices is not borne equally, nor is the error
them “in the system”) (Brayne 2014). Given they produce (Guzik 2009). That said, this
research does not necessarily suggest the police sometimes suspicionless (Renan 2016). In
intentionally use big data maliciously. Rather, terms of ALPRs, for example, “what begins
as Barocas and Selbst (2016) argue, discrimina- as more generalized collection can morph
tion may be, at least in part, an artifact of the into something quite different when the gov-
data collection and analysis process itself. ernment runs individuated searches in its
Algorithmic decision procedures can “repro- datasets” (Renan 2016:1053). Therefore, it is
duce existing patterns of discrimination, inherit an open question whether cumulative surveil-
the prejudice of prior decision makers, or sim- lance should require different legal frame-
ply reflect the widespread biases that persist in works, such as administrative law, “from
society” (Barocas and Selbst 2016:674). Under- those that govern each isolated step” (Renan
standing each step of data collection and analy- 2016:1058), namely criminal procedure.
sis is crucial for understanding how data Third, when big data—such as predictive
systems—despite being thought of as objec- policing forecasts—are combined with small
tive, quantified, and unbiased—may inherit the data—such as traditional individualized sus-
bias of their creators and users. As an institution picion based on particularized facts about a
historically implicated in the reproduction of suspect—it effectively makes it easier for law
inequality, understanding the intended and enforcement to meet the reasonable suspicion
unintended consequences of machine-learned standard in practice. In the words of one cap-
decisions and new surveillant technologies in tain, “Some officer somewhere if this [predic-
the criminal justice system is of paramount tive policing] gets big enough is going to say,
importance. ‘okay, everybody in the box is open season,’
you know? And that’s not the case.” There-
fore, legal scholars such as Ferguson (2015:
Implications for Law
336) suggest the courts should “require a
Technological tools for surveillance are far higher level of detail and correlation using the
outpacing legal and regulatory responses to insights and capabilities of big data.”
the new surveillant landscape. Therefore, the Fourth, dragnet surveillance tools such as
findings from this project have important ALPRs represent a proliferation of pre-
implications for law. First, current privacy warrant surveillance and make everyday mass
laws—such as the Privacy Act of 1974—are surveillance possible at an unprecedented
anachronistic because they largely concern scale. Pre-crime data can be mined for links
controls at the point of data collection. With once criminal suspicion comes into play.
the increased capacity to store vast amounts Once in a database, a suspect can repeatedly
of data for significant periods of time, privacy be surveilled; law enforcement can retroac-
laws now must also account for function tively search ALPR data and identify indi-
creep, protecting individuals from potential viduals, vehicles, times, and places, rather
future secondary uses of their data. Relatedly, than starting to gather information on them
law enforcement routinely purchases pri- only once they come under suspicion. The
vately collected data, blurring the lines retroactive nature of policing in an era of
between public and private and highlighting dragnet data collection means information is
the importance of revisiting third-party doc- routinely accumulated and files are lying in
trine in the digital age.14 wait. In that sense, individuals lead incrimi-
Second, use of big data for predictive ana- nating lives—daily activities, now codified as
lytics challenges the traditional paradigm of data, can be marshaled as evidence ex post
Fourth Amendment law, which is transac- facto. The proliferation of pre-warrant sur-
tional: it focuses on one-off interactions veillance tools also creates new opportunities
between law enforcement and a suspect. for parallel construction, the process of build-
However, police surveillance is increasingly ing a separate evidentiary base for a criminal
programmatic: it is ongoing, cumulative, and investigation to conceal how the investigation
Brayne 1001
began, if it involved warrantless surveillance technologies and advanced analytics. Think-

or other inadmissible evidence. ing beyond policing, future research may
Finally, previous practical constraints, consider how big data surveillance practices
which placed natural limits on the scope of identified in this study may operate in similar
surveillance, are less relevant in light of new or different ways across fields.
dragnet tools. One new analytic technique or Drawing from work in surveillance studies
surveillant technology on its own may not be helps us systematically compare use of the new-
consequential, but the combined power of est modality of surveillance—big data—across
using, for example, the person-based point domains. Informed by Lyon’s (2003) theory of
system in conjunction with ALPR data in surveillance as social sorting and Marx’s (2016)
conjunction with network diagrams in Pal- discussion of surveillance means, goals, and
antir grants authorities a level of insight into data attributes, I offer three concrete questions
an individual’s life that historically would for future research that could help us better
have constituted a Fourth Amendment search understand the use of big data for surveillance
and thus required a warrant. However, across institutional domains: Why was big data
because no one of those surveillance practices surveillance adopted (goals)? How is big data
falls outside the parameters of the law in iso- surveillance conducted (means)? What inter-
lation, neither does their combination.15 In ventions are made based on big data surveil-
that sense, intelligence is essentially pre- lance, and to what consequence (ends)? Table 1
warrant surveillance. Detectives and prosecu- summarizes these questions.
tors rarely find a “smoking gun,” a member of First, why was big data surveillance
Palantir’s legal counsel explained, but they adopted? What institutional goals was it
can now build up a sequence of events that intended to achieve? Lyon (2003:1) argues
they were previously unable to. By “integrat- that the organizational imperative for surveil-
ing data into a single ontology,” he continued, lance is one of social sorting: “Surveillance
users can draw connections between actors today sorts people into categories, assigning
and depict a coherent scheme. Hunches that worth or risk, in ways that have real effects on
would be insufficient grounds for obtaining a their life-chances . . . it is a vital means of
warrant can be retroactively backed up using sorting populations for discriminatory treat-
existing data, and queries can be justified in ment” (see also Ericson and Haggerty 1997;
hindsight after data confirm officer suspi- Rule 2007). He identifies different categories
cions. Instead of needing to justify to a judge of surveillance, each of which have different
why they require a warrant, law enforcement mandates and classificatory goals. Actors in
can first take advantage of the surveillance the criminal justice system, for example,
opportunities new technologies provide. engage in “categorical suspicion” (Lyon
2003), collecting information to classify indi-
viduals according to risk and to identify
Implications for Research in Other
threats to law and order. The purpose of sur-
Fields
veillance in other institutions, however, may
Big data is being utilized for surveillance not be categorical suspicion but rather “cate-
practices in a wide range of institutional gorical seduction”—classifying customers for
domains beyond policing, including but not targeted marketing, financial services, or
limited to health, finance, credit, marketing, credit (Lyon 2003; see also Gandy [1993] on
insurance, education, immigration, defense, the “panoptic sort”)—or “categorical care”—
and activism. Although this study focused on surveillance in health and welfare organiza-
law enforcement, big data surveillance is not tions aimed at improving services through
something over which the LAPD has exclu- better coordination of personal data (Ball and
sive domain. Rather, this case reflects broader Webster 2007). Analyzing changes and conti-
institutional shifts toward the use of emergent nuities in the structure of relationships
Table 1. Framework for Analyzing Big Data Surveillance across Institutional Contexts
Goals Means Ends
Relationship Shifts in
between Surveillance
Individual Practices
Types of Institutional and Associated with Institutional Consequences
Surveillance Field Institution Big Data Interventions for Inequality
Categorical Criminal Classifying 1) Discretionary Marking, Stigma, spillover
Suspicion justice, individuals to quantified apprehension, into other
intelligence according to risk assessment social control institutions
risk; potential 2) Explanatory
as criminals/ to predictive
terrorists analytics
Categorical Finance, Classifying 3) Query-based Different Upward or
Seduction marketing, individuals to alert-based products, downward
credit according to systems perks, access economic
their value to 4) Moderate to to credit, mobility;
companies; low inclusion opportunities, reproducing
potential as thresholds constraints current
customers 5) Disparate to patterns
Categorical Medical care, Classifying integrated Personalized May reduce
Care public individuals data medicine, inequality
assistance according to welfarist except when
their need; service intersects with
potential as delivery suspicion or
clients seduction
between agents of surveillance and those who institutional environments being used in
are surveilled (Marx 2016) may help us more healthcare or finance?
fully understand the social process of big data The goals and means of big data surveil-
surveillance and its consequences for social lance inform the final question posed for
stratification. future research: what are the ends of big data
Whereas surveillance goals have not fun- surveillance? What do institutional actors do
damentally changed much over the past cen- based on insights gleaned from big data sur-
tury, surveillance means have transformed veillance, and with what consequence? Sur-
considerably. By providing a detailed analysis veillance involves extracting information
of how the means of surveillance have from different flows (Deleuze and Guattari
changed in the age of big data, this study may 1987; Haggerty and Ericson 2000; Marx
inform future research on big data surveil- 2016). Distinct information flows are then
lance in other fields. Are the five shifts in reassembled into a “data double”—a digital
practice identified in this study—discretion- approximation of individuals based on the
ary to quantified risk assessments, explana- electronic traces they leave (Poster 1990:97)—
tory to predictive analytics, query-based to which is used to decide on differential treat-
alert-based systems, moderate to low inclu- ment. Digital scores and ranks can be
sion thresholds, and disparate to integrated understood as a form of capital (Fourcade and
databases—occurring in other institutional Healy 2017), used to determine who the
contexts? For example, to what extent are police stop, who credit bureaus determine as
data collected beyond their proximate credit-worthy, and who public assistance
Brayne 1003
agencies deem eligible for benefits. Simply have ambiguous implications for inequality,
put, one’s surveillance profile structures the may be particularly fruitful sites for future
types of communications, opportunities, con- sociological inquiry.
straints, and care one receives. Big data sur- Finally, future research may examine the
veillance could therefore have stratifying political economy underpinning the procure-
effects if individuals in positions of structural ment of analytic software that organizations
disadvantage are more likely to be subject to use for big data surveillance. Examining the
harmful forms of surveillance, and those in genealogies of surveillance technologies, for
positions of structural advantage are more example, reveals that many of the resources
likely to be targeted by advantageous surveil- for developing big data analytics come from
lance and classification schemes (Fourcade federal funds. In the law enforcement context,
and Healy 2017). Therefore, categorical sus- those grants quickly become subsumed into
picion, seduction, and care may have very police organizations’ operating budgets.
different implications for social inequality, Therefore, departments have an incentive to
depending on what institutional actors do continue using big data—or appear to be
based on the intelligence acquired through using it—even if it is not an effective means
big data surveillance. of solving the organization’s first-order prob-
This article demonstrates that in the digital lems, such as reducing crime.
age, individuals leave data traces hundreds of Understanding the implications of big data
times throughout the day, each of which con- surveillance is more complex than simply
tributes to the corpus of big data that a growing knowing who is surveilled more or less.
number of institutions use for decision- Instead, we need to understand who is sur-
making. Institutional actors making decisions veilled by whom, in what way, and for what
based on big data may assume that data doubles purpose. How surveillance structures life
are more accurate, or unbiased, representations chances may differ according to the goals,
of a person’s profile than are those gleaned means, and ends involved. Although surveil-
from “small” data, such as personal observa- lance is a generalizable organizational imper-
tions. However, this perspective obscures the ative, big data is changing the means of
social side of big data surveillance. Systematic surveillance. Accordingly, this article helps us
bias—whether intentional or unintentional— better understand how big data surveillance is
exists in training data used for machine learn- conducted, and calls for systematic research
ing algorithms, and it may be an artifact of on the relationship between the goals, means,
human discretion or the data mining process and ends of big data surveillance across insti-
itself. Moreover, the implications of false posi- tutional domains.
tives and false negatives associated with big
data surveillance vary widely across domains.
The stakes for being wrongly arrested for a Conclusions
crime you did not commit are very different Through a case study of the Los Angeles
from receiving a movie recommendation not to Police Department, this article analyzed the
your taste. Furthermore, categories of surveil- role of big data in surveillance practices. By
lance are not mutually exclusive in the age of socially situating big data, I examined why it
big data, as information can be shared across was adopted, how it is used, and what the
previously separate institutional boundaries. implications of its use are. Focusing on the
For example, electronic medical records were interplay between surveillance practices, law,
originally created to improve prescription drug and technology offers new insights into social
and care coordination, but they are increas- control and inequality. I argued that big data
ingly used to police the illicit use and sale of participates in and reflects existing social
prescription drugs. The places where catego- structures. Far from eliminating human dis-
ries of surveillance intersect, and therefore cretion and bias, big data represents a new
form of capital that is both a social product Social Media Collective at Microsoft Research, and the
and a social resource. What data law enforce- ASR editors and anonymous reviewers. Finally, I wish to
thank anonymous individuals within the Los Angeles
ment collects, their methods for analyzing Police Department for making this research possible.
and interpreting it, and the way it informs
their practice are all part of a fundamentally
social process. Characterizing predictive Funding
models as “just math,” and fetishizing com- This research was funded by the Horowitz Foundation
for Social Policy. Additional support was provided by
putation as an objective process, obscures the
grant, 5 R24 HD042849, awarded to the Population
social side of algorithmic decision-making. Research Center at The University of Texas at Austin by
Individuals’ interpretation of data occurs in the Eunice Kennedy Shriver National Institute of Child
preexisting institutional, legal, and social set- Health and Human Development.
tings, and it is through that interpretive pro-
cess that power dynamics come into play. Notes
Use of big data has the potential to amelio-
1. For example, one division in the valley began using
rate discriminatory practices, but these find- PredPol (predictive policing) in 2012, and nine
ings suggest implementation is of paramount other divisions followed suit between then and
importance. As organizational theory and lit- March 2015.
erature from science and technology studies 2. More than 25 officers in Rampart Division’s special
operations anti-gang unit, C.R.A.S.H., were investi-
suggests, when new technology is overlaid
gated or charged, and over 100 criminal cases were
onto an old organizational structure, long- overturned due to police misconduct.
standing problems shape themselves to the 3. A consent decree is a binding court order memori-
contours of the new technology, and new unin- alizing an agreement between parties in exchange
tended consequences are generated. The pro- for an end to a civil litigation or a withdrawal of a
criminal charge.
cess of transforming individual actions into
4. Brown v. Plata is a 2011 U.S. Supreme Court deci-
“objective” data raises fundamentally socio- sion holding that the overcrowding of California
logical questions that this research only begins prisons and lack of access to adequate healthcare
to address. In many ways, it transposes classic violated prisoners’ Eighth Amendment constitu-
concerns from the sociology of quantification tional rights.
5. To date, one study has evaluated the efficacy of
about simplification, decontextualization, and
Operation LASER (Uchida and Swatt 2015). It
the privileging of measurable complex social found a reduction in crime in reporting districts
phenomena onto the big data landscape. that adopted both person-based and location-based
Surveillance is always ambiguous; it is approaches. The program has not yet been subject to
implicated in both social inclusion and exclu- external evaluation; the authors are the President of
and Senior Research Associate at Justice and Secu-
sion, and it creates both opportunities and
rity Strategies, who designed Operation LASER.
constraints. The way in which surveillance 6. Block quotes are drawn from audiotaped, tran-
helps achieve organizational goals and struc- scribed interviews.
ture life chances may differ according to the 7. Consensual stops may be conducted at any time
individuals and institutions involved. Exam- when the police lack the “specific and articulable
facts” (Terry v. Ohio 392 U.S. at 21) that justify
ining the means of big data surveillance
detention or arrest.
across institutional domains is an open and 8. PredPol’s algorithm was published by Mohler and
timely line of inquiry, because once a new colleagues in 2015.
technology is disseminated in an institutional 9. In line with other law enforcement agencies, the
setting, it is difficult to scale back. LAPD is a hierarchical organization and employ-
ees are usually subject to tight managerial con-
trol. However, there was more between-division
variation in the use of algorithms than I originally
Acknowledgments expected. Big data technologies were not adopted
I wish to thank Devah Pager for her invaluable support and in the 21 area divisions and the specialized divi-
guidance. I am also grateful for the many helpful comments sions at the same time, largely due to the autonomy
I received from Paul DiMaggio, Janet Vertesi, Kim Lane granted to captains in each division during the early
Scheppele, Maria Abascal, David Pedulla, Becky Pettit, the stages of algorithmic policing. Entrepreneurial
Brayne 1005
captains who were early adopters used algorithmic Bonczar, Thomas P., and Erinn J. Herberman. 2014.
techniques largely of their own volition, but as pre- “Probation and Parole in the United States, 2013.”
dictive policing was piloted in more divisions, one Washington, DC: Bureau of Justice Statistics.
captain explained to me that he was starting to feel Bowker, Geoffrey C., and Susan Leigh Star. 2000. Sort-
pressure to use it in his division, because he did not ing Things Out: Classification and Its Consequences.
want to be the last to sign on. Cambridge, MA: MIT Press.
10. In a randomized controlled field trial, Mohler and boyd, danah, and Kate Crawford. 2012. “Critical Ques-
colleagues (2015) found PredPol’s algorithm out- tions for Big Data: Provocations for a Cultural, Tech-
performs crime analysts predicting crime, and nological, and Scholarly Phenomenon.” Information,
that police patrols using algorithmic forecast- Communication & Society 15(5):662–79.
ing led to significant reductions in crime volume. Braga, Anthony A., and David L. Weisburd. 2010. Polic-
The algorithm has not yet been subject to external ing Problem Places: Crime Hot Spots and Effective
evaluation; the authors include co-founders and Prevention. New York: Oxford University Press.
stockholders of PredPol. Braverman, Harry. 1974. Labor and Monopoly Capital:
11. After protracted negotiations, AVLs were turned on The Degradation of Work in the Twentieth Century.
in Central Bureau in March 2015. New York: Monthly Review Press.
12. Place-based algorithms are most effective (and least Brayne, Sarah. 2014. “Surveillance and System Avoid-
biased) when predicting crimes with high reporting ance: Criminal Justice Contact and Institutional
rates, such as motor vehicle theft. Attachment.” American Sociological Review 79(3):
13. For related work on performativity in a different field— 367–91.
finance—see MacKenzie, Muniesa, and Siu (2007). Brown v. Plata. 2011. 563 U.S. 493.
14. According to United States v. Miller (1939) and Browne, Simone. 2015. Dark Matters: On the Surveil-
Smith v. Maryland (1979), the third-party doctrine lance of Blackness. Durham, NC: Duke University
maintains that “when an individual voluntarily Press.
shares information with third parties, like telephone Carson, E. Ann. 2015. “Prisoners in 2014.” Washington,
companies, banks, or even other individuals, the DC: Bureau of Justice Statistics.
government can acquire that information from the Christin, Angèle. 2016. “From Daguerreotypes to Algo-
third-party absent a warrant” (Executive Office of rithms: Machines, Expertise, and Three Forms of
the President 2014). Objectivity.” ACM Computers & Society 46(1):
15. See Justice Sotomayor’s concurring opinion in 27–32.
United States v. Jones (2012) and Joh (2016). Cohen, Stanley. 1985. Visions of Social Control: Crime,
Punishment and Classification. Malden, MA: Polity
Press.
References Deleuze, Gilles, and Felix Guattari. 1987. A Thousand
Angwin, Julia. 2014. Dragnet Nation: A Quest for Pri- Plateaus: Capitalism and Schizophrenia. Minneapo-
vacy, Security, and Freedom in a World of Relentless lis: The University of Minnesota Press.
Surveillance. New York: Times Books. DiMaggio, Paul J., and Walter W. Powell. 1983. “The
Ball, Kirstie, and Frank Webster, eds. 2007. The Intensi- Iron Cage Revisited: Institutional Isomorphism and
fication of Surveillance: Crime, Terrorism & Warfare Collective Rationality in Organizational Fields.”
in the Information Age. London, UK: Pluto Press. American Sociological Review 48(2):147–60.
Barley, Stephen R. 1986. “Technology as an Occasion for Duster, Troy. 1997. “Pattern, Purpose and Race in the
Structuring: Evidence from Observations of CT Scan- Drug War.” Pp. 206–287 in Crack in America:
ners and the Social Order of Radiology Departments.” Demon Drugs and Social Justice, edited by C. Rein-
Administrative Science Quarterly 31(1):78–108. arman and H. G. Levine. Berkeley: University of
Barley, Stephen R. 1996. “Technicians in the Workplace: California Press.
Ethnographic Evidence for Bringing Work into Orga- Duster, Troy. 2005. “Race and Reification in Science.”
nization Studies.” Administrative Science Quarterly Science 307:1050–51.
41(3):404–441. Epp, Charles R., Steven Maynard-Moody, and Donald P.
Barocas, Solon, and Andrew D. Selbst. 2016. “Big Data’s Haider-Markel. 2014. Pulled Over: How Police Stops
Disparate Impact.” California Law Review 104:671–732. Define Race and Citizenship. Chicago: University of
Becker, Howard S. 1963. Outsiders: Studies in the Soci- Chicago Press.
ology of Deviance. New York: Free Press. Ericson, Richard V., and Kevin D. Haggerty. 1997. Policing
Beckett, Katherine, Kris Nyrop, Lori Pfingst, and Melissa the Risk Society. Toronto: University of Toronto Press.
Bowen. 2005. “Drug Use, Drug Possession Arrests, Ericson, Richard V., and Kevin D. Haggerty. 2006. The
and the Question of Race: Lessons from Seattle.” New Politics of Surveillance and Visibility. Toronto:
Social Problems (52)3:419–41. University of Toronto Press.
Bittner, Egon. 1967. “The Police on Skid-Row: A Study Espeland, Wendy N., and Berit I. Vannebo. 2007.
of Peace Keeping.” American Sociological Review “Accountability, Quantification and Law.” Annual
32(5):699–715. Review of Law and Society 3:21–43.
Executive Office of the President. 2014. “Big Data: Seiz- Hacking, Ian. 1990. The Taming of Chance. Cambridge,
ing Opportunities, Preserving Values.” Washington, UK: Cambridge University Press.
DC: The White House. Haggerty, Kevin D., and Richard V. Ericson. 2000. “The
Feeley, Malcolm M., and Jonathan Simon. 1992. “The Surveillant Assemblage.” British Journal of Sociol-
New Penology: Notes on the Emerging Strategy ogy 51(4):605–622.
of Corrections and Its Implications.” Criminology Harcourt, Bernard E. 2006. Against Prediction: Profiling,
30(4):449–74. Policing, and Punishing in an Actuarial Age. Chi-
Ferguson, Andrew G. 2015. “Big Data and Predictive cago: University of Chicago Press.
Reasonable Suspicion.” University of Pennsylvania Hindmarsh, Richard, and Barbara Prainsack, eds. 2010.
Law Review 63(2):327–410. Genetic Suspects: Global Governance of Forensic
Fiske, John. 1998. “Surveilling the City: Whiteness, the DNA Profiling and Databasing. Cambridge, UK:
Black Man and Democratic Totalitarianism.” Theory, Cambridge University Press.
Culture and Society 15(2):67–88. Innes, Martin. 2001. “Control Creep.” Sociological
Fiske, Susan T., and Shelley E. Taylor. 1991. Social Cog- Research Online 6:1–10.
nition, 2nd ed. New York: McGraw-Hill. Joh, Elizabeth E. 2016. “The New Surveillance Discre-
Foucault, Michel. 1977. Discipline and Punish: The Birth tion: Automated Suspicion, Big Data, and Policing.”
of the Prison. New York: Random House. Harvard Law and Policy Review 10(1):15–42.
Fourcade, Marion, and Kieran Healy. 2013. “Classifica- Kitchin, Rob. 2014. The Data Revolution: Big Data,
tion Situations: Life-Chances in the Neoliberal Era.” Open Data, Data Infrastructures & Their Conse-
Accounting, Organizations and Society 38:559–72. quences. London, UK: SAGE.
Fourcade, Marion, and Kieran Healy. 2017. “Seeing like Kling, Rob. 1991. “Computerization and Social Trans-
a Market.” Socioeconomic Review 15(1):9–29. formations.” Science, Technology & Human Values
Gandy, Oscar H. 1993. The Panoptic Sort: A Political 16(3):342–67.
Economy of Personal Information. Boulder, CO: Kohler-Hausmann, Issa. 2013. “Misdemeanor Justice:
Westview Press. Control without Conviction.” American Journal of
Gandy, Oscar H. 2002. “Data Mining and Surveillance Sociology 119(2):351–93.
in the Post-9.11 Environment.” Presentation to the Laney, Doug. 2001. “3D Data Management: Controlling
Political Economy Section, International Association Data Volume, Velocity, and Variety.” Stamford, CT:
for Media and Communication Research. META Group.
Gandy, Oscar H. 2009. Coming to Terms with Chance: Langton, Lynn, Marcus Berzofsky, Christopher Krebs,
Engaging Rational Discrimination and Cumulative and Hope Smiley-McDonald. 2012. “Victimizations
Disadvantage. Farnham, UK: Ashgate Publishing. Not Reported to the Police, 2006-2010.” Washington,
Garland, David. 2001. The Culture of Control: Crime and DC: Bureau of Justice Statistics.
Social Order in Contemporary Society. New York: Laub, John H. 2014. “Understanding Inequality and the
Oxford University Press. Justice System Response: Charting a New Way For-
Giddens, Anthony. 1990. The Consequences of Moder- ward.” New York: William T. Grant Foundation.
nity. Stanford, CA: Stanford University Press. Lazer, David, and Jason Radford. 2017. “Data ex
Gilliom, John. 2001. Overseers of the Poor: Surveillance, Machina: Introduction to Big Data.” Annual Review
Resistance, and the Limits of Privacy. Chicago: Uni- of Sociology 43:19–39.
versity of Chicago Press. Los Angeles Police Department. 2017. “Sworn Per-
Gitelman, Lisa, ed. 2013. Raw Data Is an Oxymoron. sonnel by Rank, Gender, and Ethnicity (SPRGE)
Cambridge, MA: MIT Press. Report” (http://www.lapdonline.org/sworn_and_
Goffman, Alice. 2014. On the Run: Fugitive Life in an civilian_report).
American City. Chicago: University of Chicago Press. Lynch, Michael, Simon A. Cole, Ruth McNally, and
Goffman, Erving. 1963. Stigma: Notes on the Manage- Kathleen Jordan. 2008. Truth Machine: The Conten-
ment of Spoiled Identity. Englewood Cliffs, NJ: Pren- tious History of DNA Fingerprinting. Chicago: Uni-
tice-Hall. versity of Chicago Press.
Gross, Samuel R., Maurice Possley, and Kalara Stephens. Lyon, David. 1994. The Electronic Eye: The Rise of Sur-
2017. “Race and Wrongful Convictions in the United veillance Society. Minneapolis: University of Min-
States.” National Registry of Exonerations, Newkirk nesota Press.
Center for Science and Society, University of Cali- Lyon, David, ed. 2003. Surveillance as Social Sorting:
fornia-Irvine. Privacy, Risk, and Digital Discrimination. New York:
Gustafson, Kaaryn S. 2011. Cheating Welfare: Public Routledge.
Assistance and the Criminalization of Poverty. New Lyon, David, ed. 2006. Theorizing Surveillance: The
York: NYU Press. Panopticon and Beyond. New York: Polity.
Guzik, Keith. 2009. “Discrimination by Design: Predic- Lyon, David. 2015. Surveillance after Snowden. New
tive Data Mining as Security Practice in the United York: Polity.
States’ ‘War on Terrorism.’” Surveillance and Society MacKenzie, Donald, Fabian Muniesa, and Lucia Siu, eds.
7(1):1–17. 2007. Do Economists Make Markets? On the Perfor-
Brayne 1007
mativity of Economics. Princeton, NJ: Princeton Uni- Enforcement Operations.” RAND Safety and Justice
versity Press. Program. Santa Monica, CA. Retrieved July 6, 2017
Manning, Peter K. 2011. The Technology of Policing: (http://www.rand.org/content/dam/rand/pubs/research_
Crime Mapping, Information Technology, and the reports/RR200/RR233/RAND_RR233.pdf).
Rationality of Crime Control. New York: NYU Press. Porter, Theodore M. 1995. Trust in Numbers: The Pursuit
Manning, Peter K., and John Van Maanen, eds. 1978. of Objectivity in Science and Public Life. Princeton,
Policing: A View from the Street. New York: Random NJ: Princeton University Press.
House. Poster, Mark. 1990. The Mode of Information. Chicago:
Marx, Gary T. 1974. “Thoughts on a Neglected Category University of Chicago Press.
of Social Movement Participant: The Agent Provoca- Quillian, Lincoln, and Devah Pager. 2001. “Black Neigh-
teur and the Informant.” American Journal of Sociol- bors, Higher Crime? The Role of Racial Stereotypes
ogy 80(2):402–442. in Evaluations of Neighborhood Crime.” American
Marx, Gary T. 1988. Undercover: Police Surveillance in Journal of Sociology 107(3):717–67.
America. Berkeley: University of California Press. Ratcliffe, Jerry. 2008. Intelligence-Led Policing.
Marx, Gary T. 1998. “Ethics for the New Surveillance.” Cullompton, UK: Willan Publishing.
The Information Society 14(3):171–85. Reiss, Albert J. 1971. The Police and the Public. New
Marx, Gary T. 2002. “What’s New About the ‘New Sur- Haven, CT: Yale University Press.
veillance’? Classifying for Change and Continuity.” Renan, Daphna. 2016. “The Fourth Amendment as
Surveillance and Society 1(1):9–29. Administrative Governance.” Stanford Law Review
Marx, Gary T. 2016. Windows into the Soul: Surveillance 68(5):1039–1129.
and Society in an Age of High Technology. Chicago: Rios, Victor M. 2011. Punished: Policing the Lives of
University of Chicago Press. Black and Latino Boys. New York: NYU Press.
Mayer-Schönberger, Viktor, and Kenneth Cukier. 2013. Roush, Craig R. 2012. “Quis Custodiet Ipsos Custodes?
Big Data: A Revolution That Will Transform How We Limits on Widespread Surveillance and Intelligence
Live, Work, and Think. New York: Houghton Mifflin Gathering by Local Law Enforcement after 9/11.”
Harcourt. Marquette Law Review 96(1):315–76.
Merton, Robert K. 1948. “The Self-Fulfilling Prophecy.” Rule, James B. 1974. Private Lives and Public Surveil-
The Antioch Review 8(2):193–210. lance: Social Control in the Computer Age. New
Meyer, John W., and Brian Rowan. 1977. “Institutional- York: Schocken Books.
ized Organizations: Formal Structure as Myth and Rule, James B. 2007. Privacy in Peril: How We Are
Ceremony.” American Journal of Sociology 83(2): Sacrificing a Fundamental Right in Exchange for
340–63. Security and Convenience. New York: Oxford Uni-
Mohler, George O., Martin B. Short, Sean Malinowski, versity Press.
Mark Johnson, George E. Tita, Andrea L. Bertozzi, Sampson, Robert J., and Dawn J. Bartusch. 1998. “Legal
and P. Jeffrey Brantingham. 2015. “Randomized Con- Cynicism and (Subcultural?) Tolerance of Deviance:
trolled Field Trials of Predictive Policing.” Journal of The Neighborhood Context of Racial Differences.”
the American Statistical Association 110:1399–1411. Law and Society Review 32:777–804.
Monahan, Torin, and Neal A. Palmer. 2009. “The Emerg- Scott, Richard W. 1987. Organizations: Rational, Natu-
ing Politics of DHS Fusion Centers.” Security Dia- ral, and Open Systems. Englewood Cliffs, NJ: Pren-
logue 40(6):617–36. tice-Hall.
Moskos, Peter. 2008. Cop in the Hood: My Year Policing Scott, Richard W. 2004. “Reflections on a Half-Century
Baltimore’s Eastern District. Princeton, NJ: Princeton of Organizational Sociology.” Annual Review of Soci-
University Press. ology 30:1–21.
Oxford American Dictionary of Current English. 1999. Sherman, Lawrence W. 2013. “Targeting, Testing and
Oxford: Oxford University Press. Tracking Police Services: The Rise of Evidence-
Pager, Devah. 2007. Marked: Race, Crime, and Finding Based Policing, 1975–2025.” Crime and Justice
Work in an Era of Mass Incarceration. Chicago: Uni- 42(1):377–451.
versity of Chicago Press. Sherman, Lawrence W., Patrick R. Gartin, and Michael
Papachristos, Andrew V., David M. Hureau, and Anthony E. Buerger. 1989. “Hot Spots of Predatory Crime:
A. Braga. 2013. “The Corner and the Crew: The Routine Activities and the Criminology of Place.”
Influence of Geography and Social Networks on Criminology 27(1):27–55.
Gang Violence.” American Sociological Review Skogan, Wesley G. 2006. Police and Community in Chi-
78(3):417–47. cago: A Tale of Three Cities. New York: Oxford Uni-
Pasquale, Frank. 2014. The Black Box Society: The Secret versity Press.
Algorithms That Control Money and Information. Smith v. Maryland. 1979. 442 U.S. 735.
Cambridge, MA: Harvard University Press. Soss, Joe, Richard C. Fording, and Sanford F. Schram.
Perry, Walter L., Brian McInnis, Carter C. Price, Susan 2011. Disciplining the Poor: Neoliberal Paternalism
C. Smith, and John S. Hollywood. 2013. “Predictive and the Persistent Power of Race. Chicago: Univer-
Policing: The Role of Crime Forecasting in Law sity of Chicago Press.
Stuart, Forrest. 2016. Down, Out & Under Arrest: Polic- Weisburd, David, Stephen D. Mastrofski, Ann Marie
ing and Everyday Life in Skid Row. Chicago: Univer- McNally, Rosann Greenspan, and James J. Willis.
sity of Chicago Press. 2003. “Reforming to Preserve: COMPSTAT and Stra-
Terry v. Ohio. 1968. 392 U.S. 1. tegic Problem-Solving in American Policing.” Crimi-
Tracy, Paul E., and Vincent Morgan. 2000. “Big Brother nology and Public Policy 2:421–56.
and His Science Kit: DNA Databases for 21st Cen- Western, Bruce. 2006. Punishment and Inequality in
tury Crime Control.” Journal of Criminal Law and America. New York: Russell Sage Foundation.
Criminology 90(2):635–90. Western, Bruce, and Becky Pettit. 2005. “Black-White
Travis, Jeremy, Bruce Western, and Steve Redburn, eds. Wage Inequality, Employment Rates, and Incarcer-
2014. Growth of Incarceration in the United States: ation.” American Journal of Sociology 111(2):553–
Exploring Causes and Consequences. Washington, 78.
DC: National Academy of Science. White House Police Data Initiative. 2015. “Launching
Uchida, Craig D., and Marc L. Swatt. 2015. “Operation the Police Data Initiative” (https://obamawhitehouse
LASER and the Effectiveness of Hotspot Patrol: A .archives.gov/blog/2015/05/18/launching-police-
Panel Analysis.” Police Quarterly 16(3):287–304. data-initiative).
United States v. Jones. 2012. 132 S. Ct. 945, 565 U.S. Whren v. United States. 1996. 517 U.S. 806.
United States v. Miller. 1939. 307 U.S. 174. Willis, James J., Stephen D. Mastrofski, and David
U.S. Department of Justice. 2001 [2015]. “L.A. Consent Weisburd. 2007. “Making Sense of COMPSTAT: A
Decree.” Washington, DC: U.S. Department of Justice. Theory-Based Analysis of Organizational Change in
Wakefield, Sara, and Christopher Uggen. 2010. “Incar- Three Police Departments.” Law and Society Review
ceration and Stratification.” Annual Review of Sociol- 41(1):147–88.
ogy 36:387–406. Wilson, James Q. 1968. Varieties of Police Behavior: The
Wakefield, Sara, and Christopher Wildeman. 2013. Chil- Management of Law and Order in Eight Communi-
dren of the Prison Boom: Mass Incarceration and the ties. Cambridge, MA: Harvard University Press.
Future of American Inequality. New York: Oxford
University Press.
Waxman, Matthew C. 2009. “Police and National Secu- Sarah Brayne is an Assistant Professor of Sociology and
rity: American Local Law Enforcement and Counter- Faculty Research Associate in the Population Research
Terrorism after 9/11.” Journal of National Security Center at the University of Texas at Austin. Using quali-
Law and Policy 3:377–407. tative and quantitative methods, her research examines
Weber, Max. 1978. Economy and Society: An Outline the use of “big data” within the criminal justice system as
of Interpretive Sociology. Berkeley: University of well as the consequences of surveillance for law and
California Press. social inequality.

Brayne 2017

Uploaded by

Copyright:

Available Formats

Brayne 2017

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brayne 2017

Uploaded by

Copyright:

Available Formats

725865

American Sociological Review

Big Data Surveillance: The 2017, Vol. 82(5) 977–1008

Surveillance scholars have documented a remote, low visibility or invisible, involun-

Big Data Surveillance: In the 1970s, the dominant police patrol

prediction and preemption of behaviors by focuses on the experiences and outcomes of

Fieldwork Additionally, I conducted observations on

Figure 1. Situation Room at the Real-Time Crime Analysis Center (RACR)

community supervision population according officers’ discretionary assessments of risk

Figure 2. Migration of Traditional Police Practices toward Big Data Surveillance

Figure 3. Field Interview (FI) Cards

the police operate in an organizational context Shift from Reactive to Predictive

Figure 4. Network in Palantir

Figure 5. Plotted ALPR Readings

Figure 6. Palantir Homepage

began, if it involved warrantless surveillance technologies and advanced analytics. Think-

Goals Means Ends

You might also like

Brayne 2017

Uploaded by

Copyright:

Available Formats

Brayne 2017

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Brayne 2017

Uploaded by

Copyright:

Available Formats

725865

American Sociological Review

Big Data Surveillance: The 2017, Vol. 82(5) 977­–1008

Surveillance scholars have documented a remote, low visibility or invisible, involun-

Big Data Surveillance: In the 1970s, the dominant police patrol

prediction and preemption of behaviors by focuses on the experiences and outcomes of

Fieldwork Additionally, I conducted observations on

Figure 1. Situation Room at the Real-Time Crime Analysis Center (RACR)

community supervision population according officers’ discretionary assessments of risk

Figure 2. Migration of Traditional Police Practices toward Big Data Surveillance

Figure 3. Field Interview (FI) Cards

the police operate in an organizational context Shift from Reactive to Predictive

Figure 4. Network in Palantir

Figure 5. Plotted ALPR Readings

Figure 6. Palantir Homepage

began, if it involved warrantless surveillance technologies and advanced analytics. Think-

Goals Means Ends

You might also like

Big Data Surveillance: The 2017, Vol. 82(5) 977–1008