ArticlePDF Available

Search and Seizure, Racial Profiling, and Traffic Stops: A Disparate Impact Framework

Authors:

Abstract and Figures

In response to nationwide attention to the issue of racial profiling, numerous law enforcement agencies have reexamined their policies and collected data on the racial demographics of motorists stopped and searched by police. This article advocates a disparate impact framework for understanding the relationship between race and searches and seizures. Using data on the Washington State Patrol, analysis indicates that disparities in the proportions of racial minorities searched by the Patrol are likely not the result of intentional or purposeful discrimination. Additionally, factors such as age, sex, time of day, and the number of violations that motivated the stop affect the likelihood of a search.
Content may be subject to copyright.
LAW & POLICY, Vol. 31, No. 1, January 2009 ISSN 0265–8240
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Blackwell Publishing LtdOxford, UKLAPOLaw & Policy0265-82401467-9930© 2008 The Author Journal compilation © 2008 Baldy Center for Law and Social PolicyXXX
Original Articles
Pickerill, Mosher & Pratt Search and Seizure, Racial Profiling, and Traffic StopsLaw & Policy xxx
Search and Seizure, Racial Profiling, and
Traffic Stops: A Disparate Impact Framework
J. MITCHELL PICKERILL, CLAYTON MOSHER, and TRAVIS PRATT
In response to nationwide attention to the issue of racial profiling, numerous law
enforcement agencies have reexamined their policies and collected data on the
racial demographics of motorists stopped and searched by police. This article
advocates a “disparate impact” framework for understanding the relationship
between race and searches and seizures. Using data on the Washington State
Patrol, analysis indicates that disparities in the proportions of racial minorities
searched by the Patrol are likely not the result of intentional or purposeful
discrimination. Additionally, factors such as age, sex, time of day, and the
number of violations that motivated the stop affect the likelihood of a search.
In response to high-profile cases of racial profiling, the question of whether
police officers discriminate against racial minorities has been a major
concern for many law enforcement agencies, public watchdog groups, and
academic researchers (Fredrickson and Siljander 2002; Tomaskovic-Devey,
Mason, and Zingraff 2004). As a result of nationwide attention to the issue,
numerous law enforcement agencies have reexamined their policies and
have collected data on the racial demographics of motorists stopped by
police. As a result of these efforts, it is the “stop decision” that has received
the bulk of the empirical attention thus far (see, e.g., Batton and Kadleck
2004; Engel and Calnon 2004; Harris 1999; Meeks 2000; Novak 2004; Rojek,
Rosenfeld, and Decker 2004; Smith and Petrocelli 2001; Zingraff, Smith, and
Tomaskovic-Devey 2000).
While addressing potential racial disparities in the decision to stop a
motorist has important legal and public policy implications, it is equally
important to note that scholars have long recognized that a considerable
The authors would like to acknowledge and thank Bernard Harcourt, Chuck Epp, Bert Kritzer,
Lori Fridell, and the anonymous reviewers for their comments and suggestions on various
versions of the manuscript along the way, as well as Nick Lovrich and Mike Gaffney, who were
part of a research team that worked on a larger and multifaceted project involving the
Washington State Patrol. The authors also thank the Washington State Patrol for providing the
raw data and for allowing the research team unhindered access to research materials.
Address correspondence to: J. Mitchell Pickerill, 801 Johnson Tower, Department of Political
Science, Washington State University, Pullman, WA 99164-4880, USA. Telephone: 509-335-4544;
E-mail: mitchp@wsu.edu.
2
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
amount of police officer discretion is exercised
after
the stop has taken
place (see, e.g., Brown 1981; Davis 1971, 1975; Goldstein 1960; Klinger
1994; Lipskey 1968; Pepinsky 1984; Sherman 1992; Smith, Visher, and Davidson
1984; Visher 1983; Walker 1999; Wilson 1973). Although the body of
literature on the influence of race on poststop officer decision making—at
least from a “racial profiling” framework—is still developing, one issue that
has emerged from these data is the extent of racial disparities among
motorists who are subjected to searches (see, e.g., Becker 2004).
Accordingly, scholars and commentators have offered differing views on
the propriety of racial profiling, generally, and interpretations of the dis-
proportions of minorities who are searched during traffic stops specifically.
In the process, two views have been advanced for understanding the
relationship between race/ethnicity (hereinafter, we use “race” to mean race
and/or ethnicity)
1
and searches. First, the constitutional-civil libertarian
view argues that disparities in search rates of White and minority motorists
are morally unacceptable and constitutionally impermissible under the
Equal Protection Clause of the Fourteenth Amendment (see, e.g., Harris
2002). In contrast, a second approach, emanating from law and economics,
argues that the instances of searches are to be evaluated according to “hit
rates.” These scholars argue that the goal of searches is to achieve efficient
policing by maximizing the discovery and seizure of contraband (see, e.g.,
Packer 1968) and, because rates of criminal offenses may vary by race
(Elliott et al. 1996), the rates of searches may also vary by race. According
to this perspective, disproportionate searches of racial minorities are
justified if hit rates are equal across races or if some other measure of
“efficient policing” is achieved.
Although proponents of these two approaches disagree with one another
as to whether racial disparities in search rates are morally or legally acceptable,
their analyses suffer from similar shortcomings. In particular, both the civil
libertarian and the economics approaches assume intentional or purposeful
discrimination by police offers. Although data from around the country
suggest that racial minorities are often searched at a disproportionately
higher rate than Whites, there is little empirical evidence that those disparities
are the result of malice or purposeful profiling. We instead begin with the
assumption—based on considerable empirical evidence—that the factors
influencing police officers’ decisions are complex (Black 1980; Goldstein
1977; Reiss 1984; Wilson 1973). In the present research context, therefore, it
is implausible that race is the sole factor that causes police officers to search
motorists. To be sure, the recognition of the “multivariate” nature of police
officers’ decisions to search has yet to be met by concurrent multivariate
empirical models of such behavior.
Indeed, incidents of searches are complex events that cannot be understood
through simple univariate frequency statistics and/or bivariate relationships.
The Supreme Court itself often adopts a contextual, or “totality of the
circumstances,” approach to analyze the reasonableness of an individual
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK
3
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
search under the Fourth Amendment, and the Court allows race to be used
as one factor among many in order to pursue a compelling state interest
under Equal Protection doctrine when a governmental policy has a “disparate
impact” on minorities. It is these approaches by which police officers believe
their searches will ultimately be judged—and from a legal perspective, that
belief is correct. Moreover, social science research indicates that a host of
social and contextual factors affect police-citizen interactions (Brandl et al.
1994; Cao, Frank, and Cullen 1996; Dunham and Alpert 1988; Frank et al.
1996; Jacob 1971; Reisig and Correia 1997; Reisig and Parks 2000; Terrill
and Reisig 2003). Thus, we adopt a “Disparate Impact” framework for under-
standing the relationship between race and searches during traffic stops—
one that is consistent with constitutional doctrine as well as social science
research. As such, our approach begins by assessing whether the police
officers purposefully choose to search more racial minority motorists than
White motorists. We then assess which factors in addition to, or in conjunction
with, race affect the likelihood that a motorist will be searched.
To do so, we analyze data collected by the Washington State Patrol
(WSP) recording every traffic stop by the Patrol in the state of Washington
from March 2002 through October 2002. Our analysis contributes to the
broader debate over racial profiling in three important respects. First, our
analysis leverages the difference between low- and high-discretion searches
in order to determine how law enforcement officers exercise discretion.
Second, we go beyond the usual racial classification of White/non-White to
include more specific racial classifications of both drivers and police
officers. Third, and consistent with the “Disparate Impact” approach, we
also demonstrate the need to carefully incorporate additional socio-
demographic, contextual, and geographic factors surrounding individual
contacts between police and citizens into research on racial profiling and
biased policing. The inclusion of such factors, in turn, highlights important
methodological shortcomings of the most common approaches used in
analyzing racial profiling and searches: the relative absence of multivariate
modeling approaches to the study of this dimension of police behavior. In
the process, our primary research objective is to develop a framework for
evaluating law enforcement agencies in a manner consistent using a con-
stitutional, or “Disparate Impact,” framework.
I. THEORETICAL PERSPECTIVES ON RACE AND SEARCH AND SEIZURE
A. EFFICIENT POLICING VS. CIVIL LIBERTARIAN FRAMEWORKS
The law and economics approach posits that the propriety of a police
agency’s search behavior should be evaluated according to “efficient policing”
standards (Knowles, Persico, and Todd 2001; see also Borooah 2001; Persico
2002). Because the underlying purpose of searches is to maximize the
4
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
discovery and seizure of contraband—especially illegal drugs—the legitimacy
of a law enforcement agency’s search policies must be evaluated according
to how well they achieve their purpose. The proportion of searches yielding
contraband is referred to as a “hit rate.” According to scholars who
advocate this approach, when hit rates among minorities and Whites are
relatively equivalent to one another, searches are not the result of racial
discrimination, but rather efficient policing. When hit rates are lower for
racial minorities than for Whites, the evidence supports a claim of intentional
racial discrimination; and, conversely, when hit rates for racial minorities
are higher among racial minorities than Whites, the evidence supports a
claim of reverse discrimination (Knowles, Persico, and Todd 2001).
Somewhat consistent with the law and economics perspective, Harcourt
(2004) has advocated yet another approach for understanding and analyzing
the relationship between race and searches. Harcourt argues that the use of
race in policing is not that different than the use of race in other contexts,
such as education and employment, where under the doctrinal test of strict
scrutiny, race may be considered as part of a narrowly tailored policy
designed to achieve a compelling state interest:
Racial profiling for purposes of police searches is a narrowly tailored policing
technique that promotes the traditional law enforcement interest in fighting
crime if, first, racial profiling reduces the amount of profiled crime, while
second, maintaining or increasing the efficient allocation of police resources,
without, third, producing a ratchet effect on the profiled population. A ratchet
effect occurs when racial profiling produces a supervised population that is
disproportionate to the distribution of offending by racial group. (Ibid., 4–5)
Rather than focusing on hit rates, Harcourt instead argues that assessments
of police efficiency should be based on the impact racial profiling has on
overall crime, on the profiled population, and on other social costs.
Others are critical of these efficient policing models. Civil libertarians, for
example, argue that hit rates cannot (and do not) provide conclusive evidence
that racially biased policing is not occurring. Many researchers and com-
mentators instead argue that disparities in search rates between minorities
and Whites provide
prima facie
evidence of discriminatory police behavior
(Harris 2002). Others argue even further that equal hit rates are not particularly
meaningful because there is no evidence to suggest that the underlying
“offending rates” among races are equal (Alschuler 2002; Gross and Barnes
2002; Rudovsky 2001; cf. Wilbanks 1987). For some, the disparities in
search rates are thus clear indicators of unconstitutional discrimination by
police under the Equal Protection Clause of the Fourteenth Amendment
(Gross and Barnes 2002; Maclin 2001; Rudovsky 2001).
These debates raise a number of important questions regarding racial
profiling and search and seizure. Most notably, although they differ on the
normative implications of disparities in search rates among different racial
groups, both of the dominant theoretical approaches share certain empirical
assumptions and limitations. First, they assume that disparities in search
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK
5
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
rates among different races are the result of intentional or purposeful
decisions to search different races at different rates. Second, the empirical
analyses tend to be limited to univariate descriptive statistics (e.g., frequencies
and proportions) or, at best, to bivariate analyses. Lastly, both of these
approaches focus on ends rather than means. Civil libertarians conclude
that disparities in search rates of minorities are never justified regardless of
the means used, while the efficient policing models suggest that the ends
justify the means.
2
As a consequence of these limitations, most studies of racial profiling and
search and seizures ignore factors
other
than race that might influence
police officers’ decisions to search in the first place. In other words,
although an increasing amount of data has been collected and analyzed,
most theory and analysis shed very little light on
when
and
why
searches
occur, nor do these approaches help us to evaluate whether agencies as
institutions are unconstitutionally discriminating under current Supreme
Court doctrine. We contend that before drawing conclusions about dis-
crimination in searches, much broader (than the law and economics
approach), and at the same time more in-depth (than the civil libertarian
approach), analyses are needed to determine whether disparities in search
rates of different racial groups result from intentional discrimination and
what factors in addition to race affect the likelihood that a motorist will be
searched.
B. VARIATION IN SEARCH TYPES
There are good reasons to expand the racial profiling debate to these areas,
one of which is that not all searches involve the same degree of discretion
by police officers. Based upon Supreme Court precedent and law enforcement
policies, there are at least six different categories of searches permissible in
different circumstances: (1) a search incident to arrest after a suspect has
been arrested; (2) an impound or inventory search after a vehicle has been
confiscated; (3) a warrant search pursuant to an existing search or arrest
warrant; (4) a K9 search when an officer requests a K9 unit be dispatched
to the scene so that a drug-sniffing dog can check the vehicle for evidence
of illegal drugs; (5) a consent search when an officer asks a motorist or
a passenger for permission to search the person or the vehicle; and (6) a
“Terry,” or pat down search, for when an officer believes a suspect might
be dangerous, and the officer conducts a search for his or her own pro-
tection. These different categories of searches have largely been carved out
by judges, but they reflect the reality that searches are conducted for different
reasons under different circumstances.
The decision to conduct the first three types of searches involves a
relatively lower degree of discretion compared to the latter three.
3
Thus,
empirical analyses of searches can explore the variation among the different
types of searches conducted to learn more about how officers exercise
6
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
discretion. Differences between lower-discretion searches and higher-discretion
searches, for example, can also indicate whether officers intentionally choose
to search minorities. If racial minorities are searched at approximately the
same rates as Whites in low-discretion searches, but are disproportionately
searched compared to Whites in high-discretion searches, we could infer that
police officers are intentionally selecting minorities for searches when their
discretion to conduct the search is high.
C. EQUAL PROTECTION DOCTRINE
In addition to variations in search types, judicial doctrines do, in fact, allow
the use of race in governmental decision making in areas such as education
and employment affirmative action programs (Harcourt 2004). A “narrowly
tailored” policy under the Equal Protection Clause of the Fourteenth
Amendment can be one in which race is simply one factor among many
used to make a decision that achieves a compelling state interest. Such
policies would be consistent with the Court’s holding in
Regents of University
of California v Bakke
(1978), and more recently in
Grutter v Bollinger
(2003)
and
Gratz v Bollinger
(2003), involving the use of race in the admissions
practices in higher education as a means of achieving diversity.
Where a government policy that is racially neutral on its face appears to
have a “disproportionate impact” on racial minorities (as is potentially the
case in the racial profiling debate), the Court looks at the context in which
the policy was formulated, passed, initiated, and implemented to determine
whether there was a “bad purpose” that led to the disproportionate impact
(e.g.,
Yick Wo v Hopkins
(1886);
Gomillian v Lightfoot
(1960);
Washington v
Davis
(1976);
Arlington Heights v Metropolitan Housing Development Corp
.
(1977)). Only if the Court finds bad purpose will it apply the strict scrutiny
level of review, making it less likely that the policy will pass constitutional
muster. Of course, when searching for a person suspected of committing a
crime, it is generally accepted that police may consider race if an eyewitness’s
description of a crime suspect includes race among the distinctive character-
istics of the suspect, and in other circumstances where race along with other
factors may increase the probability of catching a criminal offender (see,
e.g., Heuman and Cassack 2003; Schauer 2003; Gross and Livingston 2002;
Rudovsky 2001).
D. THE FOURTH AMENDMENT AND THE “TOTALITY OF THE
CIRCUMSTANCES”
The notion that race may be used as one factor among many in pursuit of a
compelling state interest under the Equal Protection Clause is also consistent
with the Court’s Fourth Amendment jurisprudence, under which probable
cause for a warrant and the reasonableness of a search are to be determined
by the “totality of the circumstances” (see, e.g.,
Illinois v Gates
(1983);
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK
7
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Massachusetts v Upton
(1984); and
Ohio v Robinette
(1996)).
4
Indeed, Segal
(1984) has shown empirically that the presence or absence of specific
contextual factors can influence the votes of Supreme Court justices in
determining the constitutionality of searches in specific cases. Segal found
that factors such as the prior justification of the search (e.g., probable cause
and warrant requirements), the nature and extent of the intrusion (e.g.,
whether the search took place in a home, business, or car, or whether it was
a full search versus a more limited search), and a host of mitigating factors
(e.g., whether evidence was seized in plain view) affect the likelihood that
Supreme Court justices would vote to uphold or overturn a search. Because
courts consider the contextual factors that surround race under the Fourth
Amendment, and a narrowly tailored policy under the Fourteenth Amendment
might involve the use of race as one factor among many, social science
research on the subject should do the same. Accordingly, it is likely that
police officers act under the belief that this is how judges will ultimately
evaluate the legitimacy of their searches.
E. THE NATURE OF POLICE-CITIZEN INTERACTIONS
In addition to the doctrinal basis for understanding race as one among
many factors that might increase the likelihood of a search, there is a
wealth of social science research that lends support to the prudence of such
an approach. In particular, studies of criminal justice processing, in general,
and of police–citizen interactions, in particular, are virtually unanimous in
their conclusion that context “matters” with regard to the role of race
(Pratt 1998; Walker, Spohn, and Delone 1996). To be sure, race may certainly
be correlated with police officer decision making (see, e.g., Weitzer 1996),
yet it is important to recognize how race is intertwined with a host of other
contextual factors that also influence officers’ decisions.
For example, studies have highlighted factors such as citizen and officer
demographic characteristics—including race—that may interact with
“demeanor” to influence outcomes such as arrest and use of force (Correia,
Reisig, and Lovrich 1996; Mastrofski, Reisig, and McCluskey 2002; Reisig et al.
2004; cf. Klinger 1994, 1996). In particular, if a suspect (or in the present
case, a motorist) is disrespectful to the officer, such demeanor may increase
the probability of more coercive actions on the part of the police. Others
have noted how neighborhood conditions (Grinc 1994; Reisig and Giacomazzi
1998; Webb and Marshall 1995) and citizens’ expectations (Frank et al. 1996;
Reisig and Chandek 2001) can influence police–citizen interactions. To the
extent that such conditions and expectations are related to race—and there
is plenty of evidence to suggest that they are (Wilson 1987)—these factors
may help to explain certain racial disparities in policing outcomes. Scholars
have even noted how larger social-structural conditions such as overall
racial distributions, economic disparities, and power differentials (Black
1976; Chamlin 1989a, 1989b; Liska 1987; Liska and Chamlin 1984; see also
8
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Murty, Roebuck, and Smith 1990) can be critically important to citizens’
perceptions of the legitimacy of police behavior and to patterns of officer
decision making. Thus, it is not merely race that matters; more important is
an understanding of how race is tied to the broader context in which police
make decisions.
II. “DISPARATE IMPACT” FRAMEWORK
We argue that our theoretical understanding of the relationship between
race and searches needs to move beyond
assumptions
that police officers
intentionally or purposefully profile (for either discriminatory or efficient
policing purposes). As such,
empirical analyses
of searches need to get
beyond univariate and bivariate statistics to instead view this phenomenon
for what it is: a multivariate research question. Even so, quantitative data
and statistical analysis alone cannot prove or disprove “bad purpose” by
police officers or get at other important but idiosyncratic explanatory
variables such as citizen or officer demeanor. This approach can, however,
provide a better understanding of how multiple factors—or the “totality of
the circumstances”—influence the likelihood that a search will occur once a
motorist has been stopped. Thus, it is inappropriate to indict or condemn
an entire law enforcement agency for racial profiling or biased policing
based solely on univariate and bivariate statistical disparities. Borrowing
the Court’s phrase in its Equal Protection jurisprudence, we thus characterize
this research approach as a Disparate Impact framework for understanding
and analyzing the relationship between race and search and seizure.
A Disparate Impact framework incorporates the need for both generality
and particularity when evaluating police officers’ decisions to conduct
searches. On the one hand, it is necessary for law enforcement to identify
factors that, in general, make it more probable that individuals are engaged
in criminal conduct and concealing contraband (Schauer 2003). For example,
police officers may reasonably conclude that younger men are more likely
to be carrying illegal narcotics than older women. Thus, age and gender
may be two factors that affect the probability of criminal conduct and may
therefore reasonably figure into a police officer’s calculus when deciding
whether to conduct a search. On the other hand, there will be substantial
variation in the combination of factors present in any particular case,
making it necessary to consider the specific contexts in which searches take
place. Moreover, our framework demands an evaluation of whether officers
within the police agency use their discretion to purposely target minorities.
Ours is an aggregate-level approach because our concern is at the institu-
tional level (i.e., the law enforcement agency). Thus, the Equal Protection
approach is appropriate from a legal vantage point. Of course, the test used
at the individual level would generally be the Fourth Amendment’s totality
of the circumstances test. While it would be possible to analyze a sample of
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK
9
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
individual cases under Fourth Amendment doctrines and then aggregate
those cases to attempt to make generalizations about the agency (see, e.g.,
Gould and Mastrofski 2004), such an approach can be problematic for
several reasons. For example, this approach would require a large and
representative sample of stops and searches. Such data are not readily
available to litigants who are challenging the practices of an agency since
these types of data are prohibitively expensive for most law enforcement
agencies to collect and maintain. Indeed, searches are infrequent events,
and conducting enough observational analyses (the most dominant method
of data collection in racial profiling research; Batton and Kadleck 2004)
would require large amounts of research hours and funding. Even so,
there still exists the problem of confidence that the sample of searches
is representative and that the observers are not biased. On the other hand,
many police agencies now collect and have readily available data on
police–citizen contacts. It will be these data to which most litigants turn
when asserting violations of Equal Protection by a law enforcement agency
(any individual litigant is free to challenge his or her arrest, indictment, or
conviction on a Fourth Amendment basis). What is important is that these
data are found to be reliable, and that they are analyzed properly to best
inform the courts on issues of importance under Equal Protection doctrine.
In other words, in
most
instances where law enforcement agencies are the
subjects of lawsuits for institutional discrimination, the best and most readily
available data for analyzing institutional practices will be official data on
police–citizen contacts.
To these ends, the current analysis uses official aggregate-level data to
focus on two questions. First, we examine the proposition that the likelihood
of a search will depend upon a multitude of factors, including a variety of
social characteristics of the driver and of the police officer, the context and
nature of the traffic stop, and geographic considerations. Second, and con-
sistent with the framework for analyzing why searches occur, we examine
whether the “race effect” varies according to the context of the search—in
particular, low- versus high-discretion search contexts.
III. DATA AND METHODS
A. THE WASHINGTON STATE PATROL DATA
The WSP has been collecting traffic stop data since 2000.
5
Washington
State troopers are required to fill out a data sheet for every contact they
have with a motorist, and the information from that record is then entered
into the WSP Traffic Stops Database. The data used for the present analysis
cover every stop made by members of the WSP from March 2002 through
October 2002, for a total of approximately 677,514 observations.
6
We are
mindful of concerns over using this type of official data. At this juncture,
10
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
however, this method of data collection is the only realistic way of compiling
large quantitative datasets on police contacts. As we noted above, this
method of data collection is also the most likely to be used in Equal Protection
challenges in court.
7
In addition to the demonstrated reliability of the WSP data (at least with
respect to the race codes), there are numerous other advantages of using the
WSP data over other similar datasets on racial profiling involving search
and seizures during traffic stops. Specifically, the WSP data contain more
detailed information on the race of both the driver and the police officer
(the WSP data go beyond simple White/non-White or White, Black, and
Hispanic categories), more detailed categories for the type of search that is
conducted, and a host of other contextual and geographic variables not
included in most other datasets—each of which are discussed below.
B. DEPENDENT VARIABLE
Searches are not truly dichotomous events, and thus we resist the tempta-
tion to create a simple dummy variable (“search” and “no search”) as our
dependent variable. Police offices conduct different types of searches that
involve varying degrees of discretion on their part. For example, the WSP
data divides searches into seven categories (including “no search”). We have
used those codes to create three theoretically relevant categories that serve
our goal of analyzing how the police use their discretion with respect to
race: No Search, Low-Discretion Search, and High-Discretion Search.
8
No
Search
is coded the same as the “no search” category in the WSP data. The
Low-Discretion Search
category includes those searches that, at least
theoretically, troopers are obliged to conduct or have a relatively lower
degree of discretion in choosing whether to conduct. This category includes
“Search Incident to Arrest,” “Impound Search,” and “Warrant Search,” as
coded in the WSP database. The
High-Discretion Search
category includes
those searches that are conducted entirely at the officer’s discretion. We
include “Consent Search,” “K9 Search,” and “Terry (Pat Down) Search” in
this last category. Thus, the search variable we use for the bulk of our
analyses here is made up of three unordered nominal values (the numeric
codes for which are: No Search = 0, Low-Discretion Search = 1, High-
Discretion Search = 2).
C. INDEPENDENT VARIABLES
The full set of covariates include variables drawn from driver characteristics,
the nature of the contact, officer characteristics, and the geographic context.
Driver Characteristics
include age (measured in years), the gender of the
driver (coded as female = 1 and male = 0), and a series of dummy variables for
different races (Black, Native American, Asian, and Hispanic, with White
as the reference category).
9
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK
11
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Nature of Contact
variables include three contextual variables:
10
the
number of violations (measured as the number of violations recorded by
the trooper), the driver receives as a result of the stop), a variable indicating
whether the stop occurred during daylight hours (coded such that 7 p.m. to
7 a.m. constituted nondaylight stops),
11
and a dummy variable for whether
or not the stop occurred on an interstate highway (coded as 1 = yes,
0 = no).
12
Officer Characteristics
include the gender of the officer (coded as 1 = female,
0 = male), officer experience (coded as number of months the officer had
served on the WSP at the time of the stop), and a series of dummy variables
for the race of the officer (Black, Asian, Hispanic, and Native American,
with White as the reference category).
Geographical Context
variables include a series of dummy variables for
each of eight geographical units of the WSP known as “Districts,” coded as
a series of dummy variables for each District (seven dummy variables are
identified by name in Table 6 below—the Spokane District is the reference
category). The WSP Districts are eight administrative units covering specific
and unique geographical areas of the state.
D. STATISTICAL ANALYSIS PLAN
We begin our analysis by examining the proportions of five different racial
groups who are subjected to searches. We also compare the proportion of
searches by race for low- and high-discretion searches. These analyses
represent an initial step toward assessing whether racial disparities exist in
the rate of searches. Next, consistent with the law and economics per-
spective on the conditions under which racial disproportionality in searches
may be permissible, we compute “hit rates” for each racial group. We then
conduct multivariate analysis using a multinomial logit model to examine
whether racial differences in the likelihood of a search would “hold up”
when controlling for a host of individual (both motorist and officer),
situational, and contextual characteristics of a stop. Finally, given the rather
large sample size for the present analysis, where even small relationships may
end up being statistically significant, to assess the magnitude of these relation-
ships we compute predicted probabilities of each racial group being searched
in a single geographical District, letting Driver Characteristics and the
Nature of Contact vary.
IV. RESULTS
Table 1 reports the frequencies of our dependent variable. There are a total
of 677,514 observations in the data. Of those, 654,121, or 96.5% of all
stops, did not result in a search, and 23,393, or 3.5% of all stops, resulted in
a search. Of the searches, 18,062 (2.7%) were low-discretion searches while
12
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
5,331 (0.8%) were high-discretion searches. Table 1 also reports the frequencies
of searches for those stops identified by troopers as self-initiated contacts.
13
Although one might expect the rate of searches (especially high-discretion
searches) in self-initiated contacts to differ from the rate of searches in all
contacts, in fact the rate of searches in self-initiated contacts is nearly
identical to the rate in all contacts. Table 1 also shows that of 513,815 self-
initiated contacts, the overall search rate is just under 3.5%, nearly identical
to the overall rate of searches in all contacts as reported in Table 1.
Furthermore, the rate of high-discretion searches in self-initiated contacts is
only 0.5% compared to 0.8% in all contacts. This finding dispels the
argument that WSP troopers target certain motorists for searches prior to
the actual contact.
Table 2 is a cross tabulation of the search categories by race for March–
October of 2002. This table reports rates of searches for all motorists who
had contact with the WSP during the specified period based on the reported race
of the driver. The findings here are consistent with those from other studies
that there are, in fact, some statistical disparities in the rates of searches
for different racial groups. Only about 3% of Whites and 2.5% of Asians are
searched after being pulled over, while about 6.7% of Hispanics, 7.6% of Blacks
and 15% of Native Americans who are stopped are searched.
Table 1. Frequencies of Searches
All Observations Self-Initiated Contacts
Low-Discretion Search 18,062 (2.7%) 15,083 (2.9%)
High-Discretion Search 5,331 (0.8%) 2,744 (0.5%)
No Search 654,121 (96.5%) 494,664 (96.5%)
Total 677,514 (100%) 513,815 (99.9%)
Low-discretion searches include search incident to arrest, impound search, and warrant search.
High-discretion search includes consent searches, K9 searches and Terry searches. Percentages
may not add to 100% due to rounding.
Table 2. Search by Race (All Observations)*
No Search
Low-Discretion
Search
High-Discretion
Search Total
White 552,578 (97.0%) 14,647 (2.6%) 2,427 (0.4%) 569,652 (100%)
Black 21,469 (92.4%) 1,541 (6.6%) 235 (1.0%) 23,245 (100%)
Native Am. 3,307 (84.9%) 505 (12.9%) 83 (2.1%) 3,895 (100%)
Asian 20,073 (97.5%) 450 (2.2%) 67 (0.3%) 20,590 (100%)
Hispanic 43,557 (93.4%) 2,658 (5.7%) 438 (0.9%) 46,653 (100%)
Total 640,984 (96.5%) 19,801 (3.0%) 3,250 (0.5%) 664,035 (100%)
*Not including observations coded as East Indian, Pacific Islander and Other.
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK
13
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
While these disparities are clear, they are not necessarily indications of
discrimination, biased policing, or “racial profiling.” Before concluding that
the WSP engages in racial profiling based solely on these statistical disparities,
further inquiry into the relationship between race and searches is necessary.
Consistent with the law and economics approach, we begin this inquiry by
noting that there are greater differences in search rates among the racial
groups for low-discretion searches than for high-discretion searches. About
3% of all contacts result in a low-discretion search, and about 0.5% in a
high-discretion search. Table 2 shows that about 2.6% of Whites and 2.2%
of Asians are subject to low-discretion searches, 5.7% of Hispanics, 6.6% of
Blacks and 12.9% of Native Americans are subject to low-discretion
searches. On the other hand, approximately 0.4% of Whites, 0.3% of
Asians, 0.9% of Hispanics, 1% of Blacks and 2.1% of Native Americans are
subject to high-discretion searches.
14
We also calculated hit rates for each racial group, reported in Table 3.
15
The hit rates for low-discretion searches are: Overall (22.8%), White
(24.9%), Black (18.4%), Native American (22.0%), Asian (10.7%), Hispanic
(16.5%). The hit rates for high-discretion searches were: Overall (22.6%)
White (24.1%), Black (22.1%), Native American (18.1%), Asian (22.4%),
Hispanic (17.6%).
Finally, we report the results of multivariate analysis in which we analyze
the influence of race on the likelihood of searches while controlling for
other variables. Because the dependent variable is a nominal variable with
three unordered categories, the appropriate analytical model is a multinomial
logit. Table 4 presents the results of a multinomial logit of all searches
conducted by WSP troopers in the state of Washington from March 2002
to October 2002. In this model, we use the independent variables specified
above to test the effects of specific driver characteristics (gender, age, and
race), the nature of the stop (number of violations, daylight versus night
stops, and whether the stop took place on an interstate highway), and
officer characteristics (gender, race, and experience), and we control for
geographical location of the stop at the District level. We present the results
Table 3. Hit Rates
Low-Discretion Searches High-Discretion Searches
No Contraband Contraband Hit Rate No Contraband Contraband Hit Rate
White 11,004 3,643 24.9% 1,843 584 24.1%
Black 1,258 283 18.4% 183 52 22.1%
Hispanic 2,219 439 16.5% 361 77 17.6%
Nat. Am. 394 111 22.0% 68 15 18.1%
Asian 402 48 10.7% 52 15 22.4%
Total 15,277 4,524 22.8% 2,507 743 22.6%
14
LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
of both a baseline model and a multiplicative model. The multiplicative
model includes interaction terms for age and race and for sex and race
using the product of the two variables.
16
These multiplicative models
enable direct statistical assessment of whether the race variables operate
differentially based on the sex and age of the driver.
Table 4. Multinomial Logit Coefficients for Predicting Low- and High-Discretion
Searches (0 = No Search, 1 = Low-Discretion Search, 2 = High-Discretion Search),
March 2002–October 2002 (N = 535,405)
Variables
Baseline Model Multiplicative Model
Low-Discretion
Search
High-Discretion
Search
Low-Discretion
Search
High-Discretion
Search
Driver Characteristics:
Age .023 (.001)*** .047 (.002)*** .023 (.001)*** .047 (.002)***
Female .214 (.021)*** .612 (.051)*** .150 (.023)*** .570 (.056)***
Black .714 (.033)*** .685 (.077)*** .615 (.101)*** .543 (.242)**
Hispanic .584 (.027)*** .585 (.061)*** 1.086 (.085)*** .878 (.202)***
Native American 1.736 (.060)*** 1.833 (.125)*** 1.776 (.174)*** 1.09 (.355)**
Asian .266 (.056)*** .400 (.136)** .166 (.161) .515 (.457)
Age*Black .006 (.003) .004 (.008)
Age*Hispanic .016 (.003)*** .009 (.007)
Age*Native Am. .003 (.004) .024 (.010)*
Age*Asian .001 (.005) .031 (.017)
Female*Black .378 (.087)*** .177 (.191)
Female*Hispanic .381 (.082)*** .783 (.256)**
Female*Native Am. .128 (.123) .090 (.275)
Female*Asian .564 (.145)*** .594 (.408)
Nature of Contact:
Number of Violations .671 (.005)*** .444 (.013)*** .673 (.005)*** .444 (.013)***
Daylight 1.633 (.017)*** .567 (.04)*** 1.63 (.017)*** .567 (.039)***
Interstate .003 (.019) .423 (.043)*** .005 (.019) .422 (.043)***
Officer Characteristics:
Female Officer .149 (.030)*** .095 (.071) .152 (.030)*** .096 (.071)
Black Officer .053 (.057) .996 (.079)*** .051 (.057) .994 (.079)***
Hispanic Officer .322 (.065)*** .446 (.105)*** .321 (.065)*** .445 (.105)***
Native Am. Officer .154 (.059)** .555 (.185)** .154 (.059)** .554 (.185)**
Asian Officer .814 (.074)*** .774 (.177)*** .813 (.074)*** .774 (.177)***
Officer Exp. (months) .000 (.000) .001 (.000)*** .000 (.000) .001 (.000)***
Geographical District:
Tacoma .144 (.035)*** .309 (.076)*** .143 (.035)*** .312 (.076)***
King .094 (.034)** .284 (.073)*** .095 (.034)** .288 (.073)***
Yakima .388 (.041)*** .552 (.089)*** .383 (.041)*** .547 (.089)***
Vancouver .215 (.036)*** .149 (.0768)* .214 (.036)*** .152 (.078)*
Wenatchee .055 (.038) .293 (.083)*** .057 (.038)* .291 (.082)***
Marysville .045 (.034) .318 (.074)*** .044 (.034) .319 (.074)***
Bremerton .287 (.036)*** .302 (.075)*** .284 (.036)*** .298 (.075)***
Constant 3.608 (.040)*** 4.17 (.090)*** 3.55 (.042)*** 4.18 (.093)***
Initial Log Likelihood = 89,314.813
Final Log Likelihood = 75,607.855
Pseudo R2 = .154
N = 535,405
Initial Log Likelihood = 89,314.813
Final Log Likelihood = 75,548.835
Pseudo R2 = .154
N = 535,405
*.05, **.01, ***.001
Baseline Category for the Dependent Variable = No Search
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 15
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
The two strongest predictors of either a low-discretion or a high-discretion
search are the Native American and daylight variables. At the statewide
level, Native American and nighttime drivers are most likely to be searched—
these variables have the largest coefficients and are both statistically
significant. The number of violations variable is also a statistically
significant predictor of searches. The more violations involved in a traffic
stop, the more likely it is that a search will occur. High-discretion searches
are more likely to take place on interstate highways, while there is no statistical
difference in the probability of a low-discretion search on different types of
roads. Although these variables are hardly exhaustive of the different
contextual factors involved in the nature of individual traffic stops, the
importance of these variables suggests that contextual factors are a crucial
explanation for why searches occur.
These analyses also show that race is clearly an important factor influencing
the likelihood of a search. In addition to the finding on Native American
drivers being the most likely to be searched, searches are more likely to
occur with Black and Hispanic drivers relative to Whites, and less likely to
occur to Asians than to Whites. So while we can show the importance of
the other contextual variables in the model, race still appears to be an
important factor. Nevertheless, the influence of race does not seem to
depend upon whether the trooper conducts a low-discretion or high-discretion
search. The coefficients for Black, Hispanic, and Native American drivers
remain positive and at about the same magnitude for both categories of
searches. This suggests that where officers have the most discretion in
choosing to conduct a search, they do not act any differently toward
particular racial groups than when they act with no (or with little) discretion.17
This, in turn, suggests that while there appears to be systematic disparities
in the probability that Black, Hispanic, and Native American drivers will be
searched compared to Whites, those disparities do not seem to be a result
of the racial animus or purposeful discrimination on the part of the officers.
Other driver characteristics also have significant effects on the probability
of a search. Younger drivers are more likely to be searched than older
drivers, and women are less likely to be searched than men. According to
this analysis, women are always less likely than men to be searched, but
they are even less likely to experience a high-discretion search compared to
a low-discretion search. In fact, the coefficients for the age and female
variables change more from low-discretion to high-discretion searches than
those for race across the different categories of searches.
Likewise, some of the officer characteristics have statistically significant
effects on the likelihood of searches. Asian and Hispanic officers are less
likely, and Native American officers more likely, to conduct low- and
high-discretion searches than White officers. The positive coefficient for
female officer suggests that female officers are more likely to conduct
low-discretion searches than men (although this effect was nonsignificant for
high-discretion searches). Officers’ experience, as measured by their number
16 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
of months on the WSP, indicates that more experienced officers are less
likely to conduct high-discretion searches relative to their more inexperi-
enced counterparts (yet this effect was not extended to low-discretion
searches).18
Turning to the multiplicative model, we see that a number of interaction
terms are significantly related to both low- and high-discretion searches.
For example, the age interaction terms indicate that younger Hispanic
motorists are significantly more likely to be subjected to a low-discretion
search, and that older Native American motorists are significantly more
likely to be subjected to a high-discretion search. With regard to the race–
gender interactions, we find that Hispanic males are significantly more likely
to be subjected to both low- and high-discretion searches, and that Asian
females are significantly less likely to be subjected to low-discretion searches.
Also in the multiplicative model we see that the direction and magnitude
of the coefficients for the “main effects” of the variables included in the
interaction terms are close to, if not exactly the same as, those of the
baseline model, with a couple of notable exceptions. First, the magnitude of
the Hispanic coefficients in both the low-discretion and high-discretion
search categories increases somewhat from the baseline to the multiplicative
model, indicating what has been referred to as a “suppression” effect
(Sharpe and Roberts 1997), where the main effect of a variable gets stronger
with the inclusion of additional covariates.19 Second, the main effect of the
Native American coefficient in the high-discretion search category—one of
the most pronounced racial disparities revealed in our initial analyses—is
mediated by over 40% from the baseline to the multiplicative model.
Because the magnitude of the multinomial logit coefficients cannot easily
be interpreted, we calculated predicted probabilities for low-discretion and
high-discretion searches. Tables 5 and 6 present predicted probabilities of
searches in King County (Seattle) allowing for different values of two
independent variables in the Driver Characteristics category (age and
female), and two independent variables in the Nature of Contact cate-
gory (number of violations and daylight), holding others constant at their
median or modal values.20
Tables 5 and 6 demonstrate how the likelihood of a search—particularly
high-discretion searches—is extremely low for all racial groups when the
age of the driver increases for male and female drivers. So for example, the
predicted probability that a White male driver, pulled over in the King
County District by a White male officer with 117 months of experience, in
the daytime, on an interstate will be subjected to a high-discretion search is
.006 if the driver is eighteen years old, .003 if the driver is thirty-four years
old, and .001 if the driver is fifty years old. For a Black male driver under
the same circumstances, the predicted probability is .012 if the driver is
eighteen years old, .006 if the driver is thirty-four years old, and .003 if the
driver is fifty years old. For a Hispanic male driver under the same circum-
stances, the predicted probability is .011 if the driver is eighteen years old,
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 17
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
.005 if the driver is thirty-four years old, and .002 if the driver is fifty years
old. For a Native American male driver under the same circumstances, the
predicted probability is .035 if the driver is eighteen years old, .017 if the
driver is thirty-four years old, and .008 if the driver is fifty years old.
Table 5. Predicted Probabilities of Searches in King County District (Letting Driver
Characteristics Vary)*
Age
No Search Low-Discretion Search High-Discretion Search
18 34 50 18 34 50 18 34 50
Male
White .982 .989 .993 .012 .008 .005 .006 .003 .001
Black .965 .978 .986 .023 .016 .011 .012 .006 .003
Hispanic .969 .981 .988 .021 .014 .010 .011 .005 .002
Native Am. .904 .940 .961 .061 .043 .030 .035 .017 .008
Asian .987 .992 .995 .009 .006 .004 .004 .002 .001
Female
White .987 .992 .995 .009 .006 .004 .003 .002 .001
Black .974 .984 .990 .020 .013 .009 .006 .003 .001
Hispanic .977 .986 .991 .017 .012 .008 .006 .003 .001
Native Am. .930 .955 .971 .051 .035 .025 .020 .009 .005
Asian .990 .994 .996 .007 .005 .003 .002 .001 .001
*Predicted probabilities were calculated for stops involving a male police officer with 117
months of experience, in the daytime, on an interstate with one violation.
Table 6. Predicted Probabilities of Searches in King District (Letting Race and
Nature of Contact Vary)*
# of
Violations
No Search
Low-Discretionary
Search
High-Discretionary
Search
1 2 4 124 124
Daylight
White .991 .980 .933 .008 .016 .057 .001 .004 .010
Black .975 .960 .871 .016 .031 .108 .006 .009 .019
Hispanic .981 .965 .886 .014 .027 .096 .005 .008 .018
Native Am. .940 .894 .707 .043 .081 .244 .017 .026 .049
Asian .992 .985 .949 .006 .012 .044 .002 .003 .007
Night
White .970 .944 .824 .025 .048 .160 .005 .008 .016
Black .941 .893 .700 .050 .093 .277 .010 .014 .027
Hispanic .947 .905 .723 .044 .082 .252 .009 .013 .025
Native Am. .848 .747 .449 .125 .215 .496 .027 .038 .055
Asian .977 .958 .860 .019 .037 .128 .003 .005 .011
*Predicted probabilities were calculated for stops involving a thirty-four-year-old male driver,
male police officer with 117 months of experience, on an interstate.
18 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
On the other hand, the predicted probability that a White female driver,
pulled over in the King County District by a White male officer with 117
months of experience, with one violation, in the daytime, on an interstate
will be subjected to a high-discretion search is .003 if the driver is eighteen
years old, .002 if the driver is thirty-four years old, and .001 if the driver is
fifty years old. For a Black or Hispanic female driver under the same cir-
cumstances, the predicted probability is .006 if the driver is eighteen years
old, .003 if the driver is thirty-four years old, and .001 if the driver is fifty
years old. For a Native American female driver under the same circum-
stances, the predicted probability is .020 if the driver is eighteen years old,
.009 if the driver is thirty-four years old, and .005 if the driver is fifty years
old. We note that for all racial groups and ages, the predicted probabilities
of high-discretion searches are considerably lower than the predicted proba-
bilities of low-discretion searches.
The predicted probability that a White male driver, pulled over by a
White male officer with 117 months of experience, in the daylight, on an
interstate will be subjected to a high-discretion search is .001 with one
violation, .004 with two violations, and .010 with four violations. At night
the probability of a search under the same conditions is .005 with one
violation, .008 with two violations, and .016 with four violations. For a
Black male driver under the same circumstances, the predicted probability
of a high-discretion search in the daylight is .006 with one violation, .009
with two violations, and .019 with four violations, while at night the predicted
probability of a high-discretion search is .010 with one violation, .014 with
two violations, and .027 with four violations. For Hispanic male drivers
under the same circumstances, the predicted probability of a high-discretion
search in the daylight is .005 with one violation, .008 with two violations,
and .018 with four violations, while at night the predicted probability of a
high-discretion search is .009 with one violation, .013 with two violations,
and .025 with four violations. For a Native American male driver under the
same circumstances, the predicted probability of a high-discretion search in
the daylight is .017 with one violation, .026 with two violations, and .049
with four violations, while at night the predicted probability of a high-
discretion search is .027 with one violation, .038 with two violations, and
.055 with four violations. For an Asian male driver under the same
circumstances, the predicted probability of a high-discretion search in the
daylight is .002 with one violation, .003 with two violations, and .007 with
fourr violations, while at night the predicted probability of a high-discretion
search is .003 with one violation, .005 with two violations, and .011 with
four violations.
As we discuss in more detail below, the main value of analyzing these
predicted probabilities is to emphasize the way in which multiple factors
influence the likelihood of a search for motorists of different races. Table 5
clearly shows that while there remains some small disparities in the probability
of searches based on race, the probability is also highly dependent on at
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 19
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
least two other driver characteristics: age and sex. And Table 6 indicates
that the probability is at least in part dependent upon the context of the
search—in this case the number of violations involved in the stop and
the time of day. The implication is clearly that race does not appear to be
the sole determinative factor in the probability a search will occur after a
driver has been stopped, but that it works with other variables to determine
the probability of a search after a traffic stop.
V. DISCUSSION AND CONCLUSIONS
At the bivariate level, racial minorities are disproportionately subjected to
searches in traffic stops conducted by the WSP on Washington highways.
Civil libertarians would be inclined to immediately declare that inappropriate
racial profiling, or biased policing, is occurring in Washington. Law and
economics scholars would agree that racial minorities are being targeted by
the WSP and purposefully searched at higher rates than Whites, yet they
would conclude that, with a couple of exceptions, the hit rates are relatively
even across races—especially for high-discretion searches. Indeed, the hit
rates do suggest that the WSP achieve fairly efficient policing. The hit rates
are thus more useful than the civil libertarian approach for determining
whether the agency is achieving the important goal of searching those who
are guilty of carrying contraband. Nevertheless, neither approach would go
much beyond the basic proportions of those searched or contraband seized.
Our goal here has been to move toward a more accurate and richer explanation
of when and why searches occur.
When we control for other factors that influence whether or not searches
are conducted after motorists are contacted by the WSP, we find that race
still has an impact on the likelihood of a search. Even so, numerous other
factors are also related to the likelihood of a search by the WSP. The age
and the gender of the driver are also important predictors of searches, and
of the three variables we included to help contextualize individual contacts,
each appears to have an important effect on the likelihood of a search.
Most importantly, the time of day a motorist is contacted by the WSP is
one of the best predictors of a search being conducted for both low-
discretion and high-discretion searches. Additionally, the likelihood of a
search increases with the number of violations. Although there are most
certainly a host of other contextual factors that are related to searches but
are not included in the WSP data (or other datasets around the country for
that matter), it is clear that the context in which the traffic stop takes place
has an important bearing on the likelihood of a search being conducted.
Another important finding is that while Blacks, Hispanics, and Native
Americans are more likely to be searched than Whites, our analysis suggests
that this is not necessarily a result of officers’ discretion. First, as we
showed in Table 1, the proportion of searches for all cases is virtually the
20 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
same as the proportion of searches for self-initiated contacts. In fact,
high-discretion searches occur at a slightly lower rate for self-initiated
contacts than for all contacts. Further, the influence of particular racial
categories on the likelihood of a search is roughly the same for low-discretion
and high-discretion categories of searches. While we acknowledge that these
are not precise measurements of the amount of discretion officers exercise
when conducting different types of searches, it is generally true that the
types of searches in the former category usually involve less officer dis-
cretion than those in the latter. Additionally, we also acknowledge that it is
possible that there may be discrimination in both types of searches, and
undoubtedly further research should be designed to determine different
ways in which discrimination may be manifested in low-discretion as well as
high-discretion searches. In any event, our discussion and analysis represent
significant theoretical and analytical steps toward moving beyond treating
all searches the same, and these and other questions raised by our study can
help to identify important issues for future research.
Looking to the predicted probabilities in Tables 5 and 6, it is clear that
age, sex, and contextual factors are as good as—and perhaps better than—
race as predictors of searches. Moreover, the predicted probabilities
indicate that the influence of race on the likelihood of a search matters
most for low-discretion searches of men when multiple violations are
involved. On the other hand, there is comparatively little variation in the
likelihoods that drivers of different races will be subjected to high-discretion
searches—especially for older drivers with only one violation in the daylight.
There are at least two implications of this finding. First, it indicates that
researchers can use variations in search types to look for evidence of bad
intent or bad purpose. If a police agency is indeed purposely profiling based
on race, we would expect to find higher probabilities that minorities, as
opposed to Whites, will be searched when officers have the most discretion.
Such is not the case with the WSP. Second, and consistent with arguments
made by scholars such as Schauer (2003), we can identify factors in addition
to race that police officers reasonably believe increase the probability that a
motorist is concealing contraband. In turn, these factors would be expected
to increase the likelihood of a search and, more importantly, the likelihood
of particular categories of searches. Our analysis shows that research on
racial profiling and biased policing should seek to identify whether there
are particular categories of searches (and factors associated with those
categories) where racial disparities are particularly acute compared to
categories of searches with relatively small or no such disparities. The next
step, of course, will be to probe why the disparities exist in those particular
categories.
For example, the real issue for the WSP is probably not whether officers
purposely target minorities for consent or other high-discretion searches,
but rather why is it that young Blacks, Hispanics, and Native Americans
appear more likely to be the subject of low-discretion searches when multiple
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 21
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
violations are present. There are two plausible explanations. First, it is
possible that officers make decisions based on race prior to the search. That
is, perhaps officers’ decisions to arrest, to seek a warrant, or to impound a
car are predicated on race, and the search therefore follows. Alternatively, a
second explanation may be that the underlying offending rates of certain
racial groups are higher than those for other groups. For instance, it could
be speculated that based on alcoholism-related problems on Indian
Reservations, Native Americans engage in driving while intoxicated more
than other groups. Thus, if they are subject to arrest more frequently for
DUIs based upon higher offending rates, Native Americans would also be
subjected to a higher proportion of searches incident to arrest. Additionally,
during the period of this study, motorists who were driving with a suspended
license were automatically arrested and searched incident to that arrest in
accordance with official WSP policy.21 It is possible that the proportion of
suspended licenses of racial minorities differs from that of Whites, perhaps
as a result of socio-economic status. If this is the case, minorities would be
arrested and searched at higher rates than Whites. Although examining
these potential research questions goes beyond the scope of this article, we
have shown how an appropriately crafted analysis of searches can help to
inform the inquiry.
Although the contextual and other control variables in our analysis do
not eliminate the influence of race (in a statistical sense), some of those
variables seem to be highly significant relative to race. Thus, future data
collection and analysis designed to shed light on racial profiling and biased
policing matters should be carefully designed to control for as many of
these types of contextual variables as possible—including many that were
not included in our analyses here (see Schauer 2003; Heumann and Cassak
2003). For example, we have shown that the age and sex of the driver, the
time of day, the number of violations, and the road or highway where the
car is stopped all influence the likelihood of a search, regardless of race.
Other studies have shown that the demeanor of a citizen in a contact with a
police officer affect outcomes (see, e.g., Correia, Reisig, and Lovrich 1996;
Mastrofski, Reisig, and McCluskey 2002; Reisig et al. 2004; cf. Klinger 1994,
1995). Neighborhood conditions are also likely to influence interactions
between police and citizens they encounter (Grinc 1994; Reisig and Chandek
2001). As we have shown, it is probably a combination of factors taken
together that results in a search or no search. An elderly woman with one
violation pulled over in a nice neighborhood with a clean car is probably
not going to be searched, whereas a nineteen-year-old man pulled over
with multiple violations in a neighborhood known for drug deals will prob-
ably be more likely to be searched.
In order for researchers to better understand the multitude of factors that
influence searches, and in particular how race is related to other factors, it
would be useful for law enforcement agencies to collect more detailed data.
Researchers and agencies should work together to find ways of recording
22 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
and coding officers’ perceptions of the demeanor of the driver and any
passengers during traffic stops. It is plausible that perceptions of a “bad”
demeanor are related to race, some record of an officers’ perception of
demeanor could be made. It would also be helpful to know more informa-
tion about the neighborhood or specific geographic area where the stop
took place and the condition of the car, inside and out. The socio-demographic
background of those stopped may also be related to searches and related to
race as well. Was there anything in plain view in the car that may have led
to a search? Were there smells in the car (e.g., marijuana smoke)? It would
also be helpful to know which motorists were asked for consent to search
but who said “no.” And we believe that all agencies should distinguish
between the types of searches conducted in their data collection.
As currently configured, data on searches from around the country may
not include all the contextual variables that would be ideal for the sake of
analysis, but it is true that in individual cases judges look at the totality of
circumstances, and when evaluating an agency’s practices and policies,
courts allow that race can be one factor among many if it is narrowly
tailored to achieve a compelling state interest—that of reducing crime. It is
thus reasonable to expect police agencies and individual officers to act
accordingly. Under the Fourth Amendment, searches are evaluated by the
totality of the circumstances. Under Equal Protection analysis, the first
question to be answered is whether racial minorities are intentionally
searched at higher rates than Whites. Our analysis of low-discretion versus
high-discretion searches may not be a flawless indicator (in that it is possible
there is some discrimination in both), but it is a novel effort to get at the
issue of intentional searches, and it suggests this is not the case with the
WSP. Even if there is no evidence of intentional discrimination, when
government actions result in a disproportionate impact on particular racial
groups, statistical disparities must be justified by narrowly tailored policies
designed to achieve a compelling state interest, and there must be no evidence
of bad (racially motivated) purpose. Given the influence of other important
variables, such as age, gender, location, time of day, and seriousness of the
violation involved on the likelihood of a search, as well as our comparison
of race across low- and high-discretion searches, it would seem difficult to
establish a bad purpose here.
The fact that hit rates are relatively even across races in the WSP data
furthers the constitutional analysis in two ways. First, the hit rates support
a finding of no bad purpose behind the WSP’s search practices. Second,
they also suggest that the WSP is indeed achieving a compelling state
interest. By incorporating hit rate analysis into a disparate impact
approach, we shed light on both the means and the ends involved in
searches and seizures.
Our point here is not to advocate a particular legal argument on behalf
of the agency in question, but rather to demonstrate how scholarly empirical
analyses can be designed to complement the applicable constitutional
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 23
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
standards and perhaps even to shed light on the usefulness and feasibility of
those standards. We also do not mean to suggest that our analyses indicate
that WSP troopers never conduct an illegal or improper search, or even
that no trooper ever engages in racial discrimination when deciding to
conduct a search. It is quite plausible that improper searches are indeed
conducted and that there are so-called “bad apples” in the WSP who escape
detection in a very large dataset that represents a majority of traffic stops
conducted by professional and unbiased troopers.
We also recognize that due to differences between the type of law
enforcement conducted by state patrol agencies and local law enforcement,
our findings are not necessarily generalizable to other law enforcement
agencies. The indication that there is no bad purpose underlying the search
rates of different racial groups by the WSP are consistent with those of
Latour and Dedman (2003) who found that, in the state of Massachusetts,
the state police “[stood] out for its fairness, its even-handed toughness” and
those of Smith et al. (2003) who found that, in comparison to local law
enforcement agencies in North Carolina, the North Carolina State Highway
Patrol (NCSHP) were perceived by citizens to be less likely to engage in
biased policing. While there are some important limitations to the generaliz-
ability of our study across law enforcement agencies, our findings do point
to the importance of agencies’ data collection efforts and, significantly, the
importance of evaluating and modifying data collection policies in law
enforcement agencies at all levels.
Lastly, we believe that while large quantitative datasets and rigorous
statistical analyses can help establish systematic trends and should be a core
part of any study of biased policing and racial profiling, such data and
analyses have some inherent limitations. Most importantly, it is simply
impossible to capture every detail within the context of every traffic stop in
a quantitative dataset, and these data cannot “get inside the head” of police
officers who have to make difficult decisions in the heat of the moment.
This research is immensely complex in scope and dimension, and it will be
necessary to analyze searches from multiple perspectives and at various
data points before scholars and researchers are able to gain a full under-
standing of the role race plays in the decision to search. This is an area in
which researchers and scholars need to proceed with caution. On the one
hand, agencies that engage in systematic racial discrimination need to be
exposed. On the other hand, jumping to conclusions about an entire law
enforcement agency based solely on frequency and bivariate analyses may
be premature and have detrimental consequences—especially with regard to
citizens’ levels of trust and satisfaction with the police. By refining both our
theoretical and analytical approaches, scholars cannot only further our
knowledge of these important issues, but they can help balance and mitigate
the pitfalls of current research strategies. And clearly, law enforcement
agencies should focus on data collection efforts, refining their coding
schemes to provide a broader range of variables that may explain searches.
24 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
j. mitchell pickerill is an Associate Professor of Political Science at Washington
State University. He received his Ph.D. from the University of Wisconsin-Madison and
J.D. from Indiana University-Bloomington. He is the author of Congress and Constitutional
Deliberation (Duke University Press) and articles in journals such as Law & Policy,
Georgetown Law Journal, Perspectives on Politics, and Publius.
clayton mosher is an Associate Professor of Sociology at Washington State University,
Vancouver. He received his Ph.D. from the University of Toronto. His research inter-
ests include crime and deviance, drug policy, and racial issues in the criminal justice
system. His research has appeared in journals such as Social Forces, Criminal Justice
Policy Review, Social Science Quarterly, and the Idaho Law Review.
travis c. pratt is an Associate Professor, School of Criminology and Criminal Justice,
Arizona State University. He received his Ph.D. from the University of Cincinnati. His
research focuses on correctional policy and criminological theory, with particular
attention to structural theories of crime. His recent work has appeared in Criminology,
the Journal of Research in Crime and Delinquency, and Justice Quarterly.
NOTES
1. We understand and appreciate the important distinction between race and
ethnicity. Any study of “racial profiling” and “biased policing” must be about
discrimination based on race and/or ethnicity. For the sake of style and convention,
we will use “race” to include differences in ethnicity as well as race throughout
this article.
2. From a pure policing policy perspective, we recognize that it is often accepted
an ends justifies the means mentality; that is, we expect law enforcement to
identify factors that will help them find and arrest guilty persons and to minimize
harassing innocent ones. And thus profiling based on these factors may be per-
fectly acceptable on policy grounds. Thus, hit rates may be very useful tools for
these limited purposes.
3. We recognize that there may have been more discretion at an earlier decision-
making point that led to these “low-discretion” searches. However, the decision
to conduct these types of searches is generally justified by a preexisting con-
dition, such as an arrest or an impound, and therefore the officer’s decision to
search actually involves less discretion.
4. As a result of Whren v United States (1996), the Supreme Court has essentially
bifurcated Equal Protection and Fourth Amendment analysis when claims of
racial discrimination are made related to searches and seizures. Our point here
is that regardless of which constitutional provision is being interpreted, the
Court’s doctrines in both areas are premised on the notion that multiple factors
may influence how government actors exercise discretion.
5. The Washington State Legislature passed legislation that requires periodic
reporting to the state Legislature of progress toward the elimination of racial
profiling and encourages data collection and independent analysis for all law
enforcement agencies in the state. While only a handful of city and county law
enforcement agencies have collected and analyzed such data, the legislature
required the WSP to do so.
6. Although the WSP data go back to 2000, we do not analyze any data prior to
March 1, 2002, because the codes for searches and seizures were modified
in February 2002. The old codes did not differentiate between types of
searches.
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 25
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
7. Moreover, we have conducted a reliability audit to determine how accurately the
WSP codes the race of the driver. This audit entailed randomly sampling 1,000
citizens who had been stopped by the WSP and gathering digital drivers license
photos from the Washington State Department of Licensing (WDOL). From
this original list, 812 photos were available—those missing were largely
comprised of out-of-state residents whose photos were not online with the
WDOL. We then had two independent members of the audit team rate each of
the 812 photos for their race/ethnicity using a White/non-White classification
scheme so that the audit members’ evaluations could be compared with the
racial designation listed by the WSP troopers for each case. The results indicated
a high degree of interrater reliability—at least with respect to the designation of
White versus non-White drivers—where the classifications of both members of
the audit team were inconsistent with the troopers’ designations of race in only
4.2% (n = 34) of the cases.
8. We recognize that the decision to stop a motorist precedes the decision by an
officer to search the motorist. Ideally then, this process could be conceptualized
as a two-stage model wherein the stop would be the first dependent variable and
the search the second dependent variable. At this point, data limitations prevent
us from including the first stage. While, as mentioned above, much of the recent
research on racial profiling has focused on bias at the level of who is contacted,
our analyses comparing the racial distribution of individuals contacted by the
WSP to census data, racial differences in involvement in accidents, and daylight
versus nondaylight stops indicated that the WSP is not engaged in racial
profiling at the level of which citizens they contact (Mosher et al. 2008). Thus,
the proportion of those contacted appears to reflect the driving population on
Washington State highways.
9. The WSP data also include race categories for East Indian, Pacific Islander, and
“Other.” Due to extremely low numbers of observations in these categories, we
exclude them for the sake of our analysis here.
10. There are many contextual factors that might lead to a search that we cannot
control for here. For example, the smell of alcohol, marijuana, or other intoxicating
substances is a common factor that might lead to a search. As with most datasets
collected by law enforcement agencies in the effort to study racial profiling, the
WSP data are somewhat limited in the contextual variables included. While this
is understandable from a practical standpoint, it requires us at this stage to use
what contextual variables are available and/or to use specific variables that
might serve as proxies for other factors. For example, one of the three variables
we use in our “nature of the stop” category is the number of violations. We
believe that because older and less valuable vehicles are more likely to result
in a multiple violation stop, the number of violations is a crude proxy for
socio-economic status. Thus, where racial minorities are also lower on the socio-
economic scale, they are more likely to drive older cars with more equipment
failures, which would be reflected by more violations during a traffic stop. Thus,
if the number of violations variable has a positive influence on the likelihood of a
search, then individuals of lower socio-economic status, who are disproportionately
non-White, may be more likely to be searched. Despite these limitations, the WSP
data allow for us to control for several contextual variables that are not included
in most of the other data being collected around the country, and so while our
measures here are not perfect indicators of the nature of the stop, they represent
a significant advancement in the effort to control for these types of factors.
11. While we realize that there are monthly/seasonal differences in the number of
daylight hours, there were not substantial differences in the proportion of stops
over the various months included in the data set. This coding thus assumes that
differences will cancel each other out.
26 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
12. We recognize there may be some limitations of operationalizing the nature of the
contact using these three variables. However, these variables do capture where
and why the stop took place, and represent the best effort to date to try and
control for the nature of the stop when assessing the likelihood of a search. As
we note elsewhere in this article, there are both practical and theoretical limitations
to collecting quantitative data to measure contextual matters, but that some
attempt to control for the context of the stop should be made.
13. Non-self-initiated contacts include those cases where officers are called to the
scene due to an accident, vehicle break-down, etc.
14. It is worth noting here that high-discretion searches do not include cases in
which troopers use their discretion to ask a driver for permission to search but
are denied. These data contain no information about the frequencies of those
occurrences or about the drivers who refuse.
15. There are some inconsistencies in the way in which officers coded whether they
had found and seized contraband, raising potential validity problems for this
variable. However, there is no indication that these inconsistencies are the result
of a systematic problem, but rather they appear to be the result of random data
entry mistakes.
16. Based on the existent literature on race, we also hypothesized that the nature of
police–citizen interactions, and specifically the likelihood of a search during a
traffic stop, might vary depending on different combinations of race of driver
and race of officer. We attempted to incorporate racial interactions between
drivers and police officers into the multivariate analyses by creating a series of
interaction terms between the race of the driver and the race of the officer in a
multiplicative model. However, the large number of interaction terms and low
numbers of observations in some of the interaction categories resulted in
significant collinearity. Nonetheless, we computed Pearson correlation
coefficients for race of driver and race of officer. While some of the coefficients
were statistically significant, the magnitudes of the all coefficients were small.
The strongest relationships based on the magnitudes of the correlation
coefficients for all observations in which a search was conducted and statistically
significant at the .05 level (two-tailed) were: White driver, Black officer (.037);
White driver, Asian officer (.023); Black driver, Black officer (.021); Black
driver, Asian officer (.021) Asian driver, Black officer (.020); White driver,
White officer (.019); Asian driver, White officer (.014); Black driver, Hispanic
officer (.013); Black driver, White officer (.013). Three categories resulted in
correlation coefficients over .01 but were not statistically significant at the .05
level (two-tailed), these were: Native American driver, White officer (.012);
Native American driver, Native American officer (.011); and Native American
driver, Black officer (.011). Correlation coefficients for all other racial pairs of
drivers and officers were below .01 and/or not statistically significant at the .05
level (two-tailed).
17. It is, of course, possible that there is discrimination in both types of searches. As
we indicate in the discussion section, further research should be designed to flesh
out this possibility.
18. Although we included the geographical District variables mainly as control
variables, there are substantial variations among the sign and magnitude of the
coefficients for the different District variables, indicating that the likelihood of
a search does vary based upon geographical location.
19. It is also possible that the inclusion of the interaction term resulted in collinearity
between certain covariates that caused the changes in the magnitude of the
Hispanic parameter estimates—a problem common to multivariate models that
include interaction terms (Jaccard, Turrisi, and Wan 1990). As an explicit test for
multicollinearity we computed variance inflation factors (VIFs) for the variables
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 27
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
included in this model. None of the VIFs exceeded a value of 10, a standard
cut-off point for establishing when harmful collinearity is present. Given the
potential limitations of VIF values (Maddala 1992)—in particular, the restricted
relevance of the VIF to individual coefficients of interest with no applicability to
“sets” of regressors (Fox 1991; Fox and Monette 1992)—we also examined the
condition index values for the multiplicative model. The condition index of
14.358 was well below the threshold of 30 specified by Belsley, Kuh, and Welsch
(1980). Thus, we are confident that the change in the parameter estimates discussed
here represent a real phenomenon and is not an artifact of multicollinearity.
20. Because the multiplicative model did not indicate substantial differences in the
influence of race on the likelihood of searches (again, with the exception of the
Native American variable on high-discretion searches) when the specified inter-
action terms were included, we calculate the predicted probabilities here based
on the baseline model.
21. In 2005, the Washington Supreme Court held in Pulfrey v State of Washington
that under Washington statutory law police officers have discretion to arrest for
a suspended license, and therefore the WSP could not, as a matter of policy,
require arrests in all cases.
REFERENCES
Alschuler, Alvert W. (2002) “Racial Profiling and the Constitution,” University of
Chicago Legal Forum 2002: 163–270.
Batton, Candace, and Colleen Kadleck (2004) “Theoretical and Methodological
Issues in Racial Profiling Research” Police Quarterly 7: 30–64.
Becker, Stan (2004) “Assessing the Use of Profiling in Searches by Law Enforcement
Personnel,” Journal of Criminal Justice 23: 183 93.
Belsley, David A., Edwin Kuh, and Roy E. Welsch (1980) Regression Diagnostics.
New York: Wiley.
Black, Donald (1976) The Behavior of Law. San Diego, CA: Academic Press.
Black, Donald (1980) The Manners and Customs of the Police. New York: Academic
Press.
Borooah, Vani K. (2001) “Racial Bias in Police Stops and Searches: An Economic
Analysis,” European Journal of Political Economy 17: 17 37.
Brandl, S. G., James Frank, R. E. Worden, and Timothy S. Bynum (1994) “Global
and Specific Attitudes Toward the Police: Disentangling the Relationship,” Justice
Quarterly 11: 119–34.
Brown, Michael K. (1981) Working the Street: Police Discretion and the Dilemmas of
Control. New York: Russell Sage Foundation.
Cao, Lin, James Frank, and Francis T. Cullen (1996) “Race, Community Context
and Confidence in the Police,” American Journal of Police 15: 3–22.
Chamlin, Mitchell B. (1989a) “Conflict Theory and Police Killings,” Deviant Behavior
10: 353–368.
Chamlin, Mitchell B. (1989b) “A Macro Social Analysis of Change in Police Force
Size, 1972–1982: Controlling for Static and Dynamic Influences,” The Sociological
Quarterly 30: 615–24.
Correia, Mark, Michael D. Reisig, and Nicholas P. Lovrich (1996) “Public
Perceptions of State Police: An Analysis of Individual and Contextual Variables,”
Journal of Criminal Justice 24: 17–28.
Davis, Kenneth Culp (1971) Discretionary Justice: A Preliminary Inquiry. Urbana:
Univ. of Illinois Press.
Davis, Kenneth Culp (1975) Police Discretion. St. Paul, MN: West.
28 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Dunham, Roger G., and Geoffrey P. Alpert (1988) “Neighborhood Differences in
Attitudes Toward Policing: Evidence for a Mixed–Strategy Model of Policing in a
Multi-Ethnic Setting,” Journal of Criminal Law and Criminology 79: 504–23.
Elliott, Delbert S., William Julius Wilson, David Huizinga, Robert J. Sampson,
Amanda Elliott, and Bruce Rankin (1996) “The Effects of Neighborhood
Disadvantage on Adolescent Development,” Journal of Research in Crime and
Delinquency 33: 389–426.
Engel, Robin S., and Jennifer M. Calnon (2004) “Comparing Benchmark Methodolo-
gies for Police-Citizen Contacts: Traffic Stop Data Collection for the Pennsylvania
State Police,” Police Quarterly 7: 97–125.
Fox, John (1991) Regression Diagnostics. Newbury Park, CA: Sage.
Fox, John, and Georges Monette (1992) “Generalized Collinearity Diagnostics,”
Journal of the American Statistical Association 87: 178–83.
Frank, James, S. G. Brandl, Francis T. Cullen, and Amy Stichman (1996) “Reassessing
the Impact of Race on Citizens’ Attitudes Toward the Police: A Research Note,”
Justice Quarterly 13: 321–34.
Fredrickson, D. D., and R. P. Siljander (2002) Racial Profiling. Springfield, IL:
Charles Thomas Publisher, LTD.
Goldstein, Herman (1977) Policing a Free Society. Cambridge, MA: Ballinger.
Goldstein, Joseph (1960) “Police Discretion not to Invoke the Criminal Process: Low
Visibility Decisions and the Administration of Justice,” Yale Law Journal 69: 543
88.
Gould, Jon B., and Stephen Mastrofski (2004) “Suspect Searches: Assessing Behavior
under the U.S. Constitution,” Criminology and Public Policy 3: 315–62.
Grinc, R. (1994) “Angels in the Marble: Problems in Stimulating Community
Involvement in Community Policing,” Crime and Delinquency 40: 437–68.
Gross, Samuel R., and Katherine Y. Barnes (2002) “Road Work: Racial Profiling
and Drug Interdiction on the Highway,” Michigan Law Review 101: 651–754.
Gross, Samuel R., and Debra Livingston (2002) “Racial Profiling Under Attack,”
Columbia Law Review 102: 1413–38.
Harcourt, Bernard E. (2004) “Rethinking Racial Profiling: A Critique of the Comics,
Civil Liberties, and Constitutional Literature and of Criminal Profiling More
Generally,” The University of Chicago Law Review 71: 1275 –381.
Harris, David A. (1999) “The Stories, the Statistics, and the Law: Why ‘Driving
While Black’ Matters,” Minnesota Law Review 84: 265–326.
Harris, David A. (2002) Profiles in Injustice: Why Racial Profiling Cannot Work. New
York: New Press.
Heumann, Milton, and Lance Cassak (2003) Good Cop, Bad Cop: Racial Profiling
and Competing Views of Justice. New York: Peter Lang.
Jaccard, James, Robert Turrisi, and Choi K. Wan (1990). Interaction Effects in
Multiple Regression. Thousand Oaks, CA: Sage.
Jacob, H. (1971) “Black and White Perceptions of Justice in the City,” Law & Society
Review 5: 69–89.
Klinger, David A. (1994) “Demeanor or Crime? Why ‘Hostile’ Citizens Are More
Likely to Be Arrested,” Criminology 32: 475–93.
Klinger, David A. (1996) “More on Demeanor and Arrest in Dade County,”
Criminology 34: 61–79.
Knowles, John, Nicola Persico, and Petra Todd (2001) “Racial Bias in Motor Vehicle
Searches: Theory and Evidence,” Journal of Political Economy 109: 203–29.
Latour, Francie, and Bill Dedman (2003) “Troopers Fair, Tough in Traffic Encounters,”
Boston Globe, 20 July: A1.
Lipskey, Michael (1968) Street-Level Bureaucracy. New York: Russell Sage Foundation.
Liska, Allen E. (1987) “A Critical Examination of Macro Perspectives on Crime
Control,” Annual Review of Sociology 13: 67 88.
Pickerill, Mosher, and Pratt A DISPARATE IMPACT FRAMEWORK 29
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Liska, Allen E., and Mitchell B. Chamlin (1984) “Social Structure and Crime Control
among Macrosocial Units,” American Journal of Sociology 90: 383–95.
Maclin, Tracey (2001) “The Fourth Amendment on the Freeway,” Rutgers Race &
the Law Review 3: 117–89.
Maddala, G. S. (1992). Introduction to Econometrics. New York: Macmillan.
Mastrofski, Stephen D., Michael D. Reisig, and John McCluskey (2002) “Police
Disrespect Toward the Public: An Encounter-Based Analysis,” Criminology 39:
519–52.
Meeks, Kenneth (2000) Driving While Black. New York: Broadway Books.
Mosher, Clay, J. Mitchell Pickerill, Travis Pratt, and Nicholas Lovrich (2008) “The
Importance of Context in Understanding Biased Policing: State Patrol Traffic
Citations in Washington State,” Police Practice and Research: An International
Journal 9: 43–57.
Murty, Komanduri, Julian Roebuck, and James Smith (1990) “The Image of the
Police in Black Atlanta Communities,” Journal of Police Science and Administration
17: 250–57.
Novak, Kenneth J. (2004) “Disparity and Racial Profiling in Traffic Enforcement,”
Police Quarterly 7: 65–96.
Packer, Herbert (1968) The Limits of the Criminal Sanction. Stanford, CA: Stanford
Univ. Press.
Pepinsky, Harold E. (1984) “Better Living Through Police Discretion,” Law and
Contemporary Problems 47: 249–67.
Persico, Nicola (2002) “Racial Profiling, Fairness, and Effectiveness of Policing,” The
American Economic Review 92: 1472–97.
Pratt, Travis C. (1998) “Race and Sentencing: A Meta-Analysis of Conflicting Empirical
Research Results,” Journal of Criminal Justice 26: 51324.
Reiss, Albert J. (1984) “Consequences of Compliance and Deterrence Models of Law
Enforcement for the Exercise of Discretion,” Law and Contemporary Problems 47:
88–89.
Reisig, Michael D., and Mark E. Correia (1997) “Public Evaluations of Police
Performance: An Analysis Across Three Levels of Policing,” Policing: An International
Journal of Police Strategies and Management 20: 311–25.
Reisig, Michael D., and Andrew L. Giacomazzi (1998) “Citizen Perceptions of
Community Policing: Are Attitudes Toward Police Important?” Policing: An Inter-
national Journal of Police Strategies and Management 21: 547–61.
Reisig, Michael D., and Roger B. Parks (2000) “Experience, Quality of Life, and
Neighborhood Context: A Hierarchical Analysis of Satisfaction with Police,”
Justice Quarterly 17: 608–30.
Reisig, Michael D., and Meghan Stroshine Chandek (2001) “The Effect of Expect-
ancy Disconfirmation on Outcome Satisfaction in Police-Citizen Encounters,”
Policing: An International Journal of Police Strategies and Management 24: 88–99.
Reisig, Michael D., John D. McCluskey, Stephen D. Mastrofski, and William Terrill
(2004) “Suspect Disrespect Toward the Police,” Justice Quarterly 21: 241–68.
Rojek, J., Richard Rosenfeld, and Scott Decker (2004) “The Influence of Driver’s
Race on Traffic Stops in Missouri,” Police Quarterly 7: 126–47.
Rudovsky, David (2001) “Law Enforcement by Stereotypes and Serendipity: Racial
Profiling and Stops and Searches Without Cause,” University of Pennsylvania Journal
of Constitutional Law 3: 298–366.
Schauer, Frederick (2003) Profiles, Probabilities and Stereoypes. Cambridge, MA:
Belknap Press of Harvard Univ. Press.
Segal, Jeffrey A. (1984) “Predicting Supreme Court Cases Probabilistically: The
Search and Seizure Cases: 1962–1981,” American Political Science Review 78: 891–900.
Sharpe, N. Radke, and R. A. Roberts (1997) “The Relationship Among Sums of Squares,
Correlation Coefficients, and Suppression,” The American Statistician 51: 4648.
30 LAW & POLICY January 2009
© 2008 The Authors
Journal compilation © 2008 Baldy Center for Law and Social Policy
Sherman, Lawrence W. (1992) Policing Domestic Violence. New York: Free Press.
Smith, Douglas A., Christy A. Visher, and Laura A. Davidson (1984) “Equity and
Discretionary Justice: The Influence of Race on Police Arrest Decisions,” Journal
of Criminal Law and Criminology 75: 234–49.
Smith, Michael, and Matthew Petrocelli (2001) “Racial Profiling?: A Multivariate
Analysis of Police Traffic Stop Data,” Police Quarterly 4: 4–27.
Smith, W., D. Tomaskovic-Devey, M. T. Zingraff, M. Mason, P. Y. Warren, and C.
P. Wright (2003) The North Carolina Highway Traffic Study. Washington, DC:
National Institute of Justice.
Terrill, William, and Michael D. Reisig (2003) “Neighborhood Context and Police
Force,” Journal of Research in Crime and Delinquency 40: 291– 321.
Tomaskovic-Devey, D., M. Mason, and M. Zingraff (2004) “Looking for the Driving
While Black Phenomena: Conceptualizing Racial Bias Processes and Their
Associated Distributions,” Police Quarterly 7: 3–29.
Visher, Christy A. (1983) “Gender, Police Arrest Decisions, and Notions of Chivalry,”
Criminology 21: 5–28.
Walker, Samuel (1999) The Police in America, 3rd ed. Boston: McGraw-Hill College.
Walker, Samuel, Cassia Spohn, and Miriam DeLone (1996) The Color of Justice:
Race, Ethnicity, and Crime in America. Belmont, CA: Wadsworth.
Webb, Vince, and C. Marshall (1995) “The Relative Importance of Race and Ethnicity
on Citizen Attitudes Toward the Police,” American Journal of Police 14: 4566.
Weitzer, Ronald (1996) “Racial Discrimination in the Criminal Justice System: Findings
and Problems in the Literature,” Journal of Criminal Justice 24: 30922.
Wilbanks, William (1987) The Myth of a Racist Criminal Justice System. Monterey,
CA: Brooks-Cole Publishing Company.
Wilson, James Q. (1973) Varieties of Police Behavior. New York: Atheneum.
Wilson, William Julius (1987) The Truly Disadvantaged. Chicago: Univ. of Chicago
Press.
Zingraff, Matthew, William Smith, and Donald Tomaskovic-Devey (2000) “North
Carolina Highway Traffic and Patrol Study: ‘Driving While Black’,” Criminologist
25: 1–4.
CASES CITED
Arlington Heights v Metropolitan Housing Development Corp., 429 US 252 (1977).
Regents of University of California v Bakke, 438 US 265 (1978).
Gomillian v Lightfoot, 364 US 339 (1960).
Gratz v Bollinger, 539 US 244 (2003).
Grutter v Bollinger, 539 US 306 (2003).
Illinois v Gates, 462 US 213 (1983).
Massachusetts v Upton, 466 US 727(1984).
Ohio v Robinette, 519 US 33 (1996).
Pulfrey v State of Washington. 154 Wash 2d 517, 111 P3d 1162 (2005).
Washington v Davis, 426 US 229 (1976).
Yick Wo v Hopkins, 118 US 356 (1886).
... First, stops for moving violations are more likely to result in a consent search (Fallik & Novak, 2012;Smith & Petrocelli, 2001). Second, not only are male drivers more likely to be involved in a consent search, but that the odds are even greater for Black male drivers (Briggs & Keimig, 2017;Close & Mason, 2007;Fallik & Novak, 2012;Pickerill et al., 2009;Schafer et al., 2006). The final factor is both the age and gender of the driver because prior research shows that younger male drivers have a greater chance of being involved in a consent search (Briggs & Keimig, 2017;Fallik & Novak, 2012: Pickerill et al., 2009Rosenfeld et al., 2012;Schafer et al., 2006;Tillyer et al., 2012). ...
... Second, not only are male drivers more likely to be involved in a consent search, but that the odds are even greater for Black male drivers (Briggs & Keimig, 2017;Close & Mason, 2007;Fallik & Novak, 2012;Pickerill et al., 2009;Schafer et al., 2006). The final factor is both the age and gender of the driver because prior research shows that younger male drivers have a greater chance of being involved in a consent search (Briggs & Keimig, 2017;Fallik & Novak, 2012: Pickerill et al., 2009Rosenfeld et al., 2012;Schafer et al., 2006;Tillyer et al., 2012). Swencionis and Goff (2017) identify five risk factors related to bias that may impact an officer's daily activities, and two of these could apply to the use of consent searches. ...
... Black male drivers are more likely than White male drivers to receive a request for a consent search. The current findings are consistent with the research on consent searches involving male drivers (Briggs & Keimig, 2017;Close & Mason, 2007;Fallik & Novak, 2012;Pickerill et al., 2009;Schafer et al., 2006). ...
Article
This study sought to understand the issue of racial profiling in police requests to consent search the driver. The social conditioning model was applied as a theoretical explanation of the officer based on the citizen's race, gender, and age. The propensity score matching (PSM) results show that Black drivers (vs. White drivers), Black male drivers (vs. White Male drivers), and young Black male drivers (vs. young white Male drivers) are all more likely to have the officer request to consent search the driver. Similar results were found when considering the reason for the stop is a moving violation. Overall, the results show evidence of racial profiling for Black drivers, Black male drivers, and young Black male drivers.
... For instance, Smith and Petrocelli (2001) showed Caucasian drivers were more likely to consent to a search. Other studies found no difference in consent search requests between Caucasian and minority drivers (Fallik and Novak, 2012;Pickerill et al., 2009;Ridgeway, 2006;Tillyer et al., 2012). Some studies have also shown other conditional factors may impact the chances of a consent search taking place. ...
... Male drivers, younger drivers (i.e. 19 to 29), young African-American male drivers, the time of day, the presence of multiple violations, location in a predominately minority neighborhood, officer deployment patterns, moving violations and driving an out-of-state vehicle were also related to conducting a consent search (Barnum and Perfetti, 2010;Briggs and Keimig, 2016;Fallik and Novak, 2012;Pickerill et al., 2009;Roh and Robinson, 2009;Schafer et al., 2006;Smith and Petrocelli, 2001;Tillyer et al., 2012). ...
... While the breadth of racial-profiling literature is extensive, studies using theoretical explanations for police officer decision making and advanced analyses of decision-making processes comprises only a small portion of the racial-profiling literature. Pickerill et al. (2009) applied the disparate impact framework to the issue of racial profiling involving highdiscretion stops (i.e. consent search, K9 search, and Terry (Pat Down) search) by the Washington State Patrol. ...
Article
Purpose The purpose of this paper is to apply focal concerns theory as a theoretical explanation for police officer decision making during a traffic stop that results in a consent search. The study uses coefficients testing to better examine the issue of racial profiling through the use of a race-specific model. Design/methodology/approach The data for this study come from traffic stops conducted by the Louisville Police Department between January 1 and December 31, 2002. Findings The results show that the three components of focal concerns theory can explain police officer decision making for consent searches. Yet, the components of focal concerns theory play a greater role in stops of Caucasian male drivers. Research limitations/implications The data for this study are cross-sectional and self-reported from police officers. Practical implications This paper shows the utility of applying focal concerns theory as a theoretical explanation for police officer decision making on consent searches and how the effects of focal concerns vary depending on driver race. Social implications The findings based on focal concerns theory can provide an opportunity for police officers or departments to explain what factors impact the decision making during consent searches. Originality/value This is the first study (to the researchers’ knowledge) that examines the racial effects of focal concerns on traffic stop consent searchers using coefficients testing.
... Some studies support this claim (Baumgartner et al., 2017a(Baumgartner et al., , 2017bC. Regoeczi and Kent, 2014;Geiger-Oneto and Phillips, 2003;Helfers, 2016;Lundman and Kaufman, 2003;Miller, 2008;Novak and Chamlin, 2012;Roh and Robinson, 2009;Ryan, 2016;Vito et al., 2017;Warren et al., 2006;Withrow, 2004), some not (Lange et al., 2005;Pickerill et al., 2009;Ritter, 2017;Tillyer and Engel, 2013), and some others got mixed results (Meehan and Ponder, 2002;Novak and Chamlin, 2012), i.e. racial profiling only existed in some specific scenarios. These inconsistent conclusions imply that racial profiling in traffic law enforcement is a complex issue and might be highly location-dependent. ...
... Another common observation is the so-called weekly or monthly effects (Auerbach, 2017), i.e. police officers are more likely to issue tickets at some specific time points, such as the end of the month. In addition, driver gender, driver age, and many other factors have been shown to possibly influence the law enforcement outcome (Pickerill et al., 2009;Quintanar, 2017;Ryan, 2016). Debates on these controversies have continued to increase the distrust between the public and police (Horowitz, 2007). ...
Article
Full-text available
Law enforcement is critical for improving traffic safety. However, disputes on the equity in law enforcement have continuously exacerbated the distrust between the public and the law enforcement agencies in the United States in the past decades. This study explores this issue by identifying factors influencing outcomes of traffic stops - the most common scenarios where people need to deal with law enforcement agencies. To exclude possible confounding factors, this study specifically focuses on speeding traffic stops leading to tickets or warnings in Burlington, Vermont from 2012 to 2017. The Euclidean distance-based autologistic regression model is adopted due to the presence of spatial correlations of traffic stops. It is found that with the increasing speeding severity, a speeding traffic stop is more likely to lead to a ticket. Speeding of 20 mph over the speed limit significantly influences the penalty type. Young drivers, male drivers and minority drivers are found to be more likely to be issued tickets, which suggests the possible presence of some inherent biases against these groups. In addition, time of day and month are also found to influence the likelihood of receiving speeding tickets. These findings are expected to help both the public and law enforcement agencies to better understand the characteristics of law enforcement and take appropriate measures to eliminate possible biases.
... While tactics have been updated based on new best practices and research, the traditional nature of the proactive traffic stop is one of the most consistent areas of law enforcement (Wu & Lum, 2019). Various aspects of traffic stops have received a systematic review in research, including racial profiling of traffic stops (Grogger & Ridgeway, 2006;Higgins et al., 2012;O'Reilly, 2002;Jones, 2003;Pickerill et al., 2009;Roh & Robinson, 2009;Vito et al., 2017), use of force incidents (Jefferies et al., 2011;Rojek et al., 2010), the Ferguson Effect (Wolfe & Nix, 2016), and law enforcement decision-making in regard to traffic stops (Alpert et al., 2005;Andersen & Gustafsberg, 2016;Ishoy & Dabney, 2018). However, the lack of empirical information on traffic stops resulting in officer fatalities is troubling. ...
Article
Full-text available
Traffic stops are a staple of law enforcement patrol and provide regular interaction with the community. Previous research has examined many aspects of traffic stop incidents, particularly when officers mortally wound civilians. However, accounts of peace officers feloniously killed during traffic stop incidents have received much less empirical review. The goal of this study was to establish a profile of fatal peace officer traffic stops and felony traffic stop encounters utilizing content analysis of federal, state, and local opensource data. Demographic and incident level characteristics of law enforcement officers feloniously killed during the course of traffic stops revealed an average tenure of 9.59 years, were alone in their patrol vehicle at the time of the fatal incident, and were killed during the ante phase of the stop. Firearms were overwhelmingly used in the incidents, most suspects worked alone, and the majority of incidents involved a non-felony traffic stop. Our study contributes to an already growing body of literature on traffic stop fatalities by being one of the first to establish a profile of United States peace officers feloniously killed during traffic stops.
... The increased level of contact presents greater opportunity for discovery of arrestable offenses. Additionally, the strengthening of implicit associations between Black individuals and crime increases the likelihood for arrest in highly discretionary situations, such as low-level offenses (e.g., traffic violations, loitering, trespassing) (Pickerill, Mosher, & Pratt, 2009) and automatic processing, such as decisions to use deadly force during identification of the presence of a weapon (Correll, Park, Judd, & Wittenbrink, 2002). ...
Chapter
Historically, police have struggled to build trust and legitimacy in communities of color where the tumultuous relationship between the police and community have created contentious encounters, some ending in police use of force. Events in recent years, such as the shooting of Michael Brown in Ferguson, Missouri, among numerous other highly publicized incidents of police use of deadly force against unarmed Black men, have renewed the national conversation about disparate minority contact with police, policing practices, and policing culture. The purpose of this chapter is three-fold. First, we set out to explore how attitudes toward the police are formed and why people of color have historically low levels of trust in police. Second, we examine the effect of Ferguson and similar incidents of police use of deadly force, as well as the rise of activist groups, like Black Lives Matter, on public trust in the police and changes in policing practices and policies. Last, we explore how the rise in media attention and public outcry following police use of deadly force cases have changed police officer perceptions of policing and institutional policing practices. This chapter sheds light on the systemic issues contributing to racial disparities in attitudes toward the police and proposes several new areas of inquiry for future research.
Article
The overwhelming majority of research on officer-initiated contacts with civilians is drawn from traffic stops, while relatively little is known about officer decision-making during non-vehicular, street stops. The current study fills this gap by examining intrusive detentions, investigatory actions and enforcement activities undertaken by the police during street stops. Using data from a racially/ethnic diverse metropolitan area, analyses examine encounter-related variables, civilian and officer-related demographic characteristics, and contextual correlates of actions undertaken during these incidents. Conclusions drawn from this study provide specific insight into the patterns and practices of street stop encounters and offer a contribution to our broader understanding of police-civilian encounters.
Article
Purpose The purpose of this paper is to draw a better understanding of the potential impact of daylight in officer decision making. In order to this, the authors test the veil of darkness hypothesis, which theorizes that racial bias in traffic stops can be tested by controlling for the impact of daylight, while operating under the assumption that driver patterns remain constant across race. Design/methodology/approach Publicly available traffic-stop records from the Louisville Metro Police Department for January 2010–2019. The analysis includes both propensity score matching to examine the impact of daylight in similarly situated stops and coefficients testing to analyze how VOD may vary in citation-specific models. Findings The results show that using PSM following the VOD hypothesis does show evidence of racial bias, with Black drivers more likely to be stopped. Moreover, the effects of daylight significantly varied across citation-specific models. Research limitations/implications The data are self-reported from the officer and do not contain information on the vehicle make or model. Practical implications This paper shows that utilizing PSM and coefficients testing provides for a better analysis following the VOD hypothesis and does a better job of understanding the impact of daylight and the officer decision-making on traffic stops. Social implications Based on the quality of the data, the findings show that the use of VOD allows for the performance of more rigorous analyses of traffic stop data – giving police departments a better way to examine if racial profiling is evident. Originality/value This is the first study (to the researchers' knowledge) that applies the statistical analyses of PSM to the confines of the veil of darkness hypothesis.
Article
Objectives Investigate the potential relationship between officer characteristics and poststop outcomes during police–citizen encounters. Methods Data on police-initiated contacts were drawn from a large, racially/ethnically mixed urban environment. Cross-classified, multilevel models were estimated to identify correlates of five poststop outcomes. Results Officer gender, race/ethnicity, and length of service exerted a direct and/or moderating effect on the likelihood of a traffic citation, criminal citation, and a search. Conclusions Officer characteristics are a key, yet understudied, component to understanding the complexities of police officer decision-making in police–citizen encounters. The results indicate a noticeable component of the variation in poststop outcomes rests at the officer level thereby substantiating the continued exploration of these relationships. Moreover, the moderating effect of officer characteristics on the relationship between minority citizens and poststop outcomes presents new potential directions for expanding the social conditioning model and offers evidence regarding the in-group versus out-group perspectives within the context of police–citizen interactions. Limitations include omitted variable bias and inexact measurement of some variables.
Article
Full-text available
The gap between citizen perceptions and the realities of police work is most pronounced among detective work: Little, for example, is known about how detectives use their investigative discretion. To overcome this issue within the context of race/ethnicity, detectives reported the amount of time they worked assigned cases. These data were paired with case file information containing the complainant and suspect’s racial and ethnic identity. Dyads of complainant and suspect racial/ethnic arrangements were explored to see whether there was a difference in the likelihood that a case would be worked and for how long it was worked. The results were mixed: There was no difference in the likelihood that a case would be worked or how long it was worked across differing complainant racial/ethnic identities. Cases with a minority suspect, however, were more likely to be worked and for longer periods of time. The implications for these findings are discussed.
Article
Police checking for illegal drugs are much more likely to search the vehicles of African-American motorists than those of white motorists. This paper develops a model of police and motorist behavior that suggests an empirical test for distinguishing whether this disparity is due to racial prejudice or to the police's objective to maximize arrests. When applied to vehicle search data from Maryland, our test results are consistent with the hypothesis of no racial prejudice against African-American motorists. However, if police have utility only for searches yielding large drug finds, then our analysis would suggest bias against white drivers. The model's prediction regarding nonrace characteristics is also largely supported by the data.
Book
"The Truly Disadvantagedshould spur critical thinking in many quarters about the causes and possible remedies for inner city poverty. As policy makers grapple with the problems of an enlarged underclass they—as well as community leaders and all concerned Americans of all races—would be advised to examine Mr. Wilson's incisive analysis."—Robert Greenstein,New York Times Book Review "'Must reading' for civil-rights leaders, leaders of advocacy organizations for the poor, and for elected officials in our major urban centers."—Bernard C. Watson,Journal of Negro Education "Required reading for anyone, presidential candidate or private citizen, who really wants to address the growing plight of the black urban underclass."—David J. Garrow,Washington Post Book World Selected by the editors of theNew York Times Book Reviewas one of the sixteen best books of 1987. Winner of the 1988 C. Wright Mills Award of the Society for the Study of Social Problems.