Hate Crime Research
| Published date | 01 August 2014 |
| Author | Amber D. Spry,Donald P. Green |
| Date | 01 August 2014 |
| DOI | 10.1177/1043986214536662 |
| Subject Matter | Articles |
research-article2014
Article
Journal of Contemporary Criminal Justice
2014, Vol. 30(3) 228 –246
Hate Crime Research:
© 2014 SAGE Publications
Reprints and permissions:
Design and Measurement
sagepub.com/journalsPermissions.nav
DOI: 10.1177/1043986214536662
ccj.sagepub.com
Strategies for Improving
Causal Inference
Donald P. Green1 and Amber D. Spry1
Abstract
The credibility revolution in the social sciences has placed new emphasis on research
designs that provide strong evidence of cause and effect. The next generation of hate
crime research must move in this design-based direction. This essay reviews recent
examples of experiments and quasi-experiments in criminology, political science, and
economics that provide useful design templates for hate crime researchers. At the
same time, we caution that advances in design must also be accompanied by advances
in measurement if researchers are to gauge the long-term effects of interventions
designed to reduce the risk of hate crime.
Keywords
hate crime, causal inference, research design
During the 1980s and 1990s, the terms hate crime and bias crime took root in popular
discourse, media coverage, and public law (Jackson, 2005; Jacobs & Potter, 1998;
Messner, McHugh, & Felson, 2004). Although usage varied, the key features of this
neologism were embodied in the Federal Hate Crime Statistics Act of 1990, which
defined hate crime as
crimes that manifest evidence of prejudice based on race, gender and gender identity,
religion, disability, sexual orientation, or ethnicity, including where appropriate the
crimes of murder, non-negligent manslaughter; forcible rape; aggravated assault, simple
assault, intimidation; arson; and destruction, damage or vandalism of property. (Hate
Crimes Statistics Act, 1990)
1Columbia University, New York, NY, USA
Corresponding Author:
Donald P. Green, Department of Political Science, Columbia University, 420 W. 118th St., New York,
NY, 10027 USA.
Email: dpg2110@columbia.edu
Green and Spry
229
By this definition, hate crime is a behavioral manifestation of prejudice, where the
behavior in question is conduct, such as assault, that would otherwise be unlawful.
Although this definition of hate crime was and would remain controversial, with vary-
ing opinions about which target groups should be covered (Boyd, Berk, & Hamner,
1996; Craig, 1999; Jenness & Broad, 1997; Wang, 1994), whether the definition
should include forms of political expression (Lawrence, 2009), and whether motiva-
tions can be measured reliably (Berk, 1990; McDevitt et al., 2000), the advent of hate
crime as a public policy issue attracted the attention of scholars from a wide variety of
disciplines. The 1990s saw the rapid proliferation of academic books and articles
across an array of disciplines: law, sociology, criminology, and political science.
Perhaps because hate crime research straddled disciplinary boundaries, scholarship
subsided as hate crime faded from front-page news and policy debates. Unlike preju-
dice, a core topic in social psychology, or ethnic conflict, a core topic in political sci-
ence, hate crime never achieved the status of a topic on which courses were routinely
taught. Ironically, by the turn of the century, hate crime had been upstaged by growing
scholarly interest in kindred topics such as genocidal violence (Fearon & Laitin, 2003;
Kaufmann, 2006; Kiernan, 2007; Madley, 2005; Peterson, 2002; Verwimp, 2005) and
the automatic activation of prejudices (Dasgupta et al., 2000; Fazio, Jackson, Dunton,
& Williams, 1995; Greenwald et al., 2009; Gregg, Seibt, & Banaji, 2006). Hate crime
research was also methodologically vulnerable. At a time when the social sciences
were experiencing a “credibility revolution” (Angrist & Pischke, 2010) that placed
new emphasis on the experimental or quasi-experimental research designs that could
convincingly demonstrate cause-and-effect, hate crime research was predominantly
the study of correlations. Some research sought to look at over-time aggregate rela-
tionships between hate crime and macroeconomic conditions (Green, Glaser, & Rich,
1998; McLaren, 1999) or demographic transformations (Esses, Jackson, & Armstrong,
1998; Grattet, 2009; Green, Strolovitch, & Wong, 1998; Olzak, 1989), but for the most
part empirical research focused on cross-sectional correlations (Green, Strolovitch,
et al., 2001; Messner et al., 2004; Nolan & Akiyama, 1999; Stotzer, 2010; Waldner &
Berg, 2008).
Looking back at the rise and decline of hate crime research, one might reasonably
ask whether the study of prejudice-motivated crime warrants continuing scholarly
attention. In our view, the answer is yes, not only because of the substantive impor-
tance of the topic but also because this is one of the few literatures in which theories
of prejudice are tested using behavioral outcomes outside the laboratory and outside
the United States. But if this literature is to move forward, it must do so in step with
methodological advances in social science. Specifically, researchers must place a pre-
mium on research designs that can convincingly identify causal effects. As we suggest
below, this requirement puts new emphasis not only on experimental designs but also
on the development of new measurement techniques for assessing outcomes.
To preview our argument, we begin by defining what we mean by experimental,
quasi-experimental, and observational research designs. We next suggest a variety of
untapped research opportunities for field experimentation and the investigation of
naturally occurring experiments, drawing examples from other substantive domains.
230
Journal of Contemporary Criminal Justice 30(3)
At the same time, we acknowledge an important limitation of experiments that assess
the effects of interventions on hate crime rates: These rates reflect the incidence, as
opposed to risk of hate crime. If victimized groups relocate to more tolerant areas in
the wake of an anti-hate crime intervention, it is possible for an intervention that truly
reduces the risk of hate crime to nevertheless raise the incidence of hate crime in
treated areas (Bowling, 1994; Ferraro, 1995; LaGrange, Ferraro, & Supancic, 1992;
Weisburd et al., 2006). Interventions that actually work may appear to be counterpro-
ductive. For the hate crime literature to overcome this problem, researchers must
develop alternative outcome measures that gauge the risk of victimization. We sketch
out some possible ways of doing so and conclude by envisioning the next generation
of hate crime research.
Experimental and Quasi-Experimental Designs
In the social sciences, the term experiment refers to studies in which the units of obser-
vation are assigned by some known random process (e.g., a coin flip) to treatment and
control conditions. Random assignment helps ensure that treatment and control groups
differ systematically only insofar as one group receives the treatment and the other
does not.1 For example, imagine a study in which the objective is to estimate the aver-
age effect of an advertising campaign that uses roadside billboards to increase public
awareness of hate crime laws. If the unit of analysis were the municipality and the
outcomes were gauged by a survey of public awareness conducted a year after the
launch of the advertising campaign, an experimental study would randomly assign
each municipality to advertising or no advertising with known probability.
An observational study, by contrast, does not employ random assignment; instead,
an unknown process determines whether units are treated or not. In the context of the
billboards example, an observational study would compare municipalities that hap-
pened to deploy public information billboards with those that did not. Although
observational studies are easier to conduct than experiments, the lack of control over
the assignment process introduces uncertainty when researchers draw causal infer-
ences from the results. If the data were to reveal differences in public awareness
between treated and untreated municipalities, does this difference indicate the causal
effect of the treatment, or was the deployment of billboards a marker for unmeasured
differences in municipalities that are correlated with public awareness? Although
researchers may strive to measure these unobserved confounders and control for
them statistically, the very fact that they are unobserved implies that one can never
entirely rule out threats to causal inference. This fundamental uncertainty undercuts
the scientific value of much observational research (Gerber, Green, & Kaplan,
2004).
Somewhere along the continuum between experimental and observational research
lies the quasi-experiment (Cook & Campbell, 1979) or natural experiment (Dunning,
2012). This intermediate category encompasses research designs that focus on natu-
rally occurring assignments that are plausibly characterized as random. For example,
if a federal government grant were to make funds available to cities of more than
Green and Spry
231
100,000 residents for deploying public information billboards, one could arguably
compare cities with just under 100,000 residents to cities with just over 100,000 on the
grounds that this arbitrary population threshold partitions these cities in a near random
way.2 Like an observational study, a quasi-experiment does not use a random proce-
dure to determine...
To continue reading
Request your trial