![Free access]({"type":"image" , "src":"/pb-assets/3rdPartyLogos/FREE_ACCESS_ICON-1465051266437.png"}){kind=icon}
![CrossMark Logo]({"type":"image" , "src":"/templates/jsp/images/CROSSMARK_Color_horizontal.svg"})
ABSTRACT
?
![MathJax Logo]({"type":"image","src":"/templates/jsp/_style2/_tandf/pb2/images/math-jax.gif"})
?ABSTRACT
This paper explores perceptions of the killings of African-Americans by police officers. We show how characteristics of the victim, officer and surrounding environment, as well as political cues, shape such perceptions. In the first study, we employ a conjoint survey experiment, wherein subjects are exposed to descriptions of hypothetical police killings. Focusing on subjects who score high on the Symbolic Racism Scale (SRS), we identify what leads such subjects to view shootings as more justified. We replicate and extend these effects in a second study in which subjects read fictitious newspaper articles. We find that exposing high SRS subjects to primes related to Black Lives Matter can decrease their belief in shootings' justifiability.
1. Introduction
Shortly after noon on August 9, 2014, in Ferguson Missouri, Darren Wilson, a white police officer, shot and killed Michael Brown, an African-American teenager. Almost as soon as the news was reported, a debate began in the media and among the public about the role race had played in the shooting, and whether details of the incident somehow made it justified. This debate was hardly unfamiliar. Only the previous summer, George Zimmerman, a “neighborhood watch” patrol man in Florida, had been acquitted of murdering Trayvon Martin, an unarmed African-American teenager. While the list of African-Americans brutalized or otherwise murdered by police stretches back longer from slavery through Jim Crow to more recent victims including Amadou Diallo and Rodney King, the debate over the issue has become newly intense. The summer of Brown's killing inaugurated a morbid parade of death. Eric Garner, an African-American man selling individual cigarettes was choked to death by police officers; Tamir Rice, a 12-year-old African-American boy, was killed when officers said they mistook a toy gun for a real gun, an incident also caught on videotape; Freddie Gray, 25, died in police custody after being arrested for possession of a switchblade; Sandra Bland, 28, died in police custody in Waller Country, Texas. Widely disseminated video footage of the first two incidents helped propel them to the national spotlight. Members of the public and Black communities in particular have protested, elected officials have issued policy proposals, and a social movement, called “Black Lives Matter” (BLM) by the public but organized under the label “Movement for Black Lives,” has attracted domestic and worldwide attention.
Yet even as police killings of African-Americans have unsettled American politics, we know little about the causal determinants behind mass public perceptions of such incidents. To be sure, we have observational data on perceptions of race and the police, with attitudes often splintering sharply along racial lines. Fifty percent of whites believe African-Americans are treated fairly by the police; 84% of African-Americans say otherwise (Pew 2016). There were also sharp divisions along racial lines in response to the Garner and Brown cases (Pew 2014). Yet the available survey evidence provides little insight into the precise reasons why some police shootings are regarded asjustified – and others are not, especially by white Americans. What available evidence there is, is often from the the perspective of police officers, in studies that examine what, and why, they are more likely to shoot at African-Americans, comes largely from the perspective of the police officer – not the average citizen (Correll et al. 2002; Greenwald, Oakes, and Hoffman 2003).
Such work has tended to focus on the role that stereotypes play in police decisions and how they structure everyday interactions (Allport 1954; Steele 2010). Stereotypes linking blacks to crime are especially well-documented. Research indicates that mere exposure to the faces of black people can lead one to think about crime; conversely, thinking about crime can itself lead one to think of black Americans (Eberhardt et al. 2004). Priming exercises can also result in increased public support for harsh punitive policies among white respondents (Hetey and Eberhardt 2014). In experiments that place subjects in the role of a police officer via video game, white players were quicker to shoot African-American targets, with the degree of racial bias related to the extent to which players relied on stereotypes (Correll et al. 2002). Work in similar simulated environments corroborates this bias (Greenwald, Oakes, and Hoffman 2003).
In the latter two studies, researchers presented subjects with environments meant to mimic those in which police officers make decisions. In this paper, we broaden the scope of the simulated environment. We ask: Given the limited information at their disposal, how do citizens make judgments about police killings? Do their judgments rely on stereotypes alone? How does racism affect their judgments? What specific attributes best explain their beliefs about whether specific killings seem justified? To help answer these questions, we mount two studies. In Study 1, we employ a conjoint survey experiment, wherein subjects are presented with descriptions of hypothetical situations in which police shoot and kill individuals. While we understand the sensitivity of this work, we are interested in what happens in terms of evaluations of the justification of such shootings when we vary the characteristics assigned to the victim, perpetrator and context.
Conjoint experiments have recently come to the fore of survey research in political science (Hainmueller, Hopkins, and Yamamoto 2014). The virtue of conjoint survey experiments is straightforward: Many experiments present respondents with treatments that consist of multiple constituent parts. Disentangling what, specifically, about the treatment provoked the observed effects is a difficult task. In contrast, a conjoint experiment allows researchers to isolate the causal effects of specific attributes. By randomizing the information vended about each attribute, and compelling subjects to evaluate paired scenarios, data emerges about what precisely motivates outcomes.
In Study 1, we presented subjects with paired profiles of hypothetical police shootings. We randomized information about each shooting, including the victim's age, the location where it occurred, and the background of the police officer. Subjects were then asked to evaluate the perceived justifiability of each killing. Such a request, to determine whether a shooting was justifiable based on a very limited information, may seem artificial, but the public routinely makes decisions about whether police shootings are justifiable with less information. Following the procedure outlined by Hainmueller, Hopkins, and Yamamoto (2014), we isolate the effects of each level on beliefs about justifiability. Moreover, as we collected information about subjects' racial attitudes and their attitudes toward the police weeks before administering this experiment, we can understand how such attitudes intersect with beliefs about justifiability.
Across all subjects, we found that a victim's described criminal background, his or her age, his or her reported recent drug use, and the weapon possibly in his or her possession all played powerful roles in beliefs about justifiability. Thus, how the media reports on these issues when describing a shooting has an effect on perceived justification. The picture becomes more complicated when we separate subjects based on their scores on the Symbolic Racism Scale (SRS). SRS is premised on the idea that, as the most hostile forms and violent forms of anti-black racism have become socially taboo, racist attitudes have persisted by other means. In place of what is often called “old fashioned racism” whites have expressed their anti-black attitudes by focusing on Blacks' supposed deficiencies in upholding “traditional” American values such as individualism and hard work (Kinder and Sanders 1996); Henry and Sears 2002; Kinder and Sears 1981. There are, of course, other ways to explore the impact of race and racism on political attitudes. Some researchers focus on group conflict and threat (Bobo, Kugel, and Smith 1997; Bobo and Hutchings 1996; Quillian 1995); while others explore the manifestation of color-blind racism (Bonilla-Silva and Forman 2000). Still others have found the reemergence in some domains of old-fashioned racism (Tesler 2012). We are using SRS because of its dominance in the field, making our findings easily accessible and replicable to other researchers. We find that those who score high on the SRS scale react more sharply to the location of a shooting.
In Study 2, we tested whether our observations about justifiability can be replicated with a different treatment deployed outside of a conjoint setting, and what, if any, effects social and political primes have on beliefs about justifiability. Eight weeks after Study 1, we recontacted the same subject pool and randomly vended them a news article about a hypothetical police shooting. Based on the conjoint results, each article described one ofthe two hypothetical victims. One victim was described as displaying all the characteristics that our conjoint results indicated would lead a respondent to perceive her death as being unjustified; the other victim was described as displaying all the characteristics that our results indicated would lead to the death being perceived as more justified. In some of the articles, we embedded primes related to and Donald J. Trump. In the BLM prime, a movement supporter describes the victim as a beloved family and community contributor; in the Trump prime, the then-presidential candidate urged people not to blame the police.
Study 2 both corroborates and enriches our findings from Study 1. The death of the victim who was described with seemingly unjustifiable characteristics was regarded as unjustified, while the death of the victim described with seemingly justifiable characteristics was regarded as justified. Additionally, the political primes had differential effects based on respondents' SRS levels. High SRS subjects exposed to a BLM prime viewed a police shooting of a hypothetical undeserving victim as less justified than all other subjects. A similar pattern is observed when we isolate responses by ideology and partisanship. We speculate that this pattern is owed to a seminal finding of the research into new kinds of racism, wherein individuals whose racial attitudes are most conservative are in fact more supportive of black Americans when blacks are described as meeting traditional American values – because those with conservative racial attitudes are surprised that black Americans can confirm to their supposed standards (Sniderman et al. 1991). It seems the facts confront them with the humanity of black Americans.
Our results speak to how individuals, and especially how high SRS individuals, make use of stereotypes. In Study 1, high SRS individuals were uniquely responsive to the location of the shooting, suggesting that they are more willing to extrapolate from environmental evidence than their low SRS counterparts. It seems that high SRS individuals are looking for a reason to believe the shooting by a police officer is justified. And while all subjects were responsive to the individual characteristics offered in ourscenarios – the age of the victim, his use of drugs, his criminalhistory – high SRS subjects were alone in the effect that the“evidence” offered by the shooting location increased their belief in the justifiable nature of the shooting. Indeed, the difference between locations in our study had no effect on low SRS subjects. High SRS subjects make inferences about police shootings not only based on characteristics of the shooting victim himself, but characteristics of the victim's environment. In Study 2, we show that the same high SRS subjects regard the shooting of hypothetical unjustified victims as less justified when they are described as meeting traditional American values.
2. Theoretical motivation
In U.S. courts of law, jurors are given detailed instructions on how to assess a defendant's guilt. In the court of public opinion, no such instructions exist. Instead, individuals make use of the limited information available to them, as well as their own biases, to answer the question at hand. Their judgments often reflect their acceptance of racist stereotypes – which by definition are formed on the basis of incomplete information (Allport 1954) about particular groups. To resort to stereotypes is to make a judgment quickly and imperfectly (Tajfel 1969), failing to account for the ways in which individual members of a group do not conform to all external characteristics and narratives of that group. Stereotypes surrounding African-Americans abound. As Allport (1954) recognized, stereotypes can begin with a grain of truth but quickly take on a life of their own. Consider the stereotype which holds African-Americans in possession of superior athletic ability. It may be the case that African-Americans make up a disproportionate share of players in the National Basketball Association. Individuals take this limited information and seek to confirm the stereotype. In Stone et al. (1999), survey participants were randomly told that some basketball players on a radio broadcast were black, and others were white. Players randomly described as black were subsequently rated as more athletic than players randomly described as white.
This form of confirmation bias is hardly limited to athletics. A robust literature shows that, given the racist stereotype that blacks are more violent, blacks will be considered more violent in ostensibly ambiguous situations. Duncan (1976) randomly assigned subjects to watch an interaction in which people physically shove each other. When the shove was performed by a black person, it was perceived as more violent than when it was performed by a white person. Eberhardt et al. (2004) shows that criminal defendants rated as more “stereotypically black” than others were more likely to be sentenced to death. In Hetey and Eberhardt (2014), respondents were asked to sign a petition calling for stricter punitive laws. Those who had been shown pictures that made it seem that blacks comprised an even larger share of the prison population than they do became more likely to sign. Once activated, racist stereotypes of blacks as more violent led to calls for more punitive policies.
These consequences extend to interactions between individuals and law enforcement. Several studies have used simulations to evaluate how police officers react to race in potentially violent settings. In Correll et al. (2002), subjects played a video game in which they had to decide whether to shoot a target. Players shot targets with guns more quickly if the target was black. When players were instructed to fire as fast as possible, players shot unarmed black targets more often than unarmed white targets. Greenwald, Oakes, and Hoffman (2003) use a similar video game to show that subjects have greater trouble distinguishing violent objects from non-violent objects when in the hands of a black target.
While stereotypes may be an inescapable component of the human categorization process (Tajfel 1981), reliance on stereotypes leads to prejudice and racism (Devine 1989). From the perspective of a member of the stereotyped group, stereotypes are particularly pernicious. Steele (2010) describes “stereotype threat,” wherein members of a stereotyped group anticipate that they will be stereotyped and respond accordingly. This threat looms over their lives and compels them to change their behavior. Steele quotes journalist Brent Staples on his experience of walking as a black man in Chicago:
I became an expert in the language of fear. Couples locked arms or reached for each other's hand when they saw me
I whistled popular tunes from the Beatles and Vivaldi's Four Seasons. The tension drained from people's bodies when they heard me. A few even smiled as they past me in the dark.
For Staples, “whistling Vivaldi” was a low-cost way to signal to passer-bys that he was harmless. He was black, but contrary to stereotypes, he was not violent; the disjunction between expectation and reality disarmed Staples' neighbors.
Our task in this paper is more morbid. We are trying to identify the details of police shootings that are likely to make people think of the shootings as more justifiable. We are spurred on by the fact that in many of the high-profile shootings, seemingly small details about the victim have played central roles in the subsequent public debate. For example, in the aftermath of the Michael Brown killing, commentator noted the marijuana that toxicology reports indicated Brown had consumed. Also noted was the shoplifting Brown had engaged in prior to his encounter. Both of these factors were used to justify the shooting and killing of Michael Brown in media reports. We outline additional examples of much-discussed details used to justify shootings in the design section.
3. Empirical expectations
While some whites likely react to police shootings with barely disguised racial animus toward blacks, this form of out-and-out or old-fashioned racism is likely not as prevalent as “modern” (McConahay 1986) or “symbolic” (Kinder and Sears 1981) racism. For those who score high on the SRS, blacks are regarded negatively for being perceived as failing to work as hard as other minority groups have done (Henry and Sears 2002). Yet harboring high levels of symbolic racism does not necessarily mean that one always expresses racially hostile attitudes. In their investigation of new forms of racism, Sniderman et al. (1991) found that conservatives are more likely than liberals to believe that the government should offer to aid to laid-off black workers. However, their research suggests that this generosity is contingent upon the laid-off black worker being described as a “dependable worker.” By exhibiting a strong work ethic, the black worker is demonstrating her fealty to the American value of hard work; the authors assert that conservatives support the worker in part because she confounded their expectations. The so-called racial novelty can push people, particularly whites or those who score high on SRS, toward supporting blacks more so than they would otherwise (Porter and Wood 2016).
We believe the anti-black attitudes measured by SRS are not just individual assessments but are similarly shaped by perceptions of group threat. As Ash (1993, 308) writes, “there is reason to identify symbolic racism with group-based, emotion-linked appraisals.” Thus our measures of SRS help us to understand how race and racism assessed through an individual and collective lens, influences whether an individual will believe the killing of another individual by the police is justified.
With all this in mind, so as to be precise about our expectations, we offer the following hypotheses:
Hypothesis 1:The extent to which one relies on stereotypes when forming perceptions of police killings will fluctuate dependent on one's level of symbolic racism.
Hypothesis 2: Among those who exhibit high levels of symbolic racism, victims of police killings who appear to come closer to meeting what has been represented as “traditional” American values will be regarded as more sympathetic, and thus their shootings will be viewed as less justified, among those who exhibit high levels of symbolic racism.
We use Study 1 to evaluate Hypothesis 1. Hypothesis 2 is evaluated in Study 2.
4. Study 1: experimental design
We administered Study 1 over two waves in July 2016. The study was conducted over Mechanical Turk, a popular, low-cost means of subject acquisition (Berinsky, Huber, and Lenz 2012). As described in the supporting information, we cross-validated the Mechanical Turk sample against face-to-face government survey data; our sample compares favorably. To run a two-wave panel on Mechanical Turk, we followed Christenson and Glick (2013). In wave one, we asked standard demographic questions, as well as the SRS scale of Henry and Sears (2002). Following a two-week washout period, we recontacted subjects and administered the conjoint experiment. In creating a lag between waves 1 and 2, we were simply adhering to the standard of Hainmueller, Hopkins, and Yamamoto (2015) study of immigration attitudes.
4.1. Wave 1
We recruited 1700 subjects for Wave 1 in July 2016, paying each subject $.50, a standard level of incentive pay on the Mechanical Turk platform. Wave 1 consisted of a standard demographic battery, including questions about subjects' age, race, party identification, household income and gender, and the SRS. Table 1 provides the demographic composition for all three waves (Study 1 comprised waves one and two, while Study 2 comprised wave 3). Study 2 is disaggregated by the condition. The top row (in italics) reports the number of respondents. Subsequent rows report the distribution of demographic characteristics by wave. We have evaluated white and people of color subjects simultaneously. However, while our sample includes whites, African-Americans, Latino and other people of color, there are not enough respondents outside of whites to rigorously analyze each racial and ethnic group separately.
Table 1. Experimental balance, by wave and prime (for third wave.) first row indicates count of respondents. subsequent entries are column percentages.
4.2. Wave 2
At the beginning of the second wave, subjects were then told that they would be evaluating descriptions of videos, with each video showing the police killing a person. We explained to subjects that they would not actually be able to see the videos but wanted them to make the best decision they could about the justifiable nature of the killing with the limited information we presented to them. The full text read:
In this study, we are going to describe several videos. Each video shows a policeman killing a person. We know that you are used to actually watching videos, but we want you to make the best decision you can with the limited information we give you.
By describing the profiles as related to videos of police killing individuals, we aimed to increase the verisimilitude of the study, given that many of the more high-profile police killings involved videos of the incident.1
In a conjoint experiment, a respondent is compelled to evaluate a series of paired profiles. Each profile consists of attributes, under which exist a set of levels. The forcedresponse – with subjects choosing between the two profiles – is repeated multiple times. In our case, each profile contained attributes about a hypothetical police killing. We selected attributes that might be driving attitudes about police killings. Each profile contained attributes about the gender of the person killed, the age of the person killed, if the person killed had a criminal record, if the person killed was in possession of a weapon and if so what type, the policeman's disciplinary record, the location of the shooting, if the person killed had taken an illegal substance, the race of the person killed and the race of the police officer.2
The attributes remain the same across each profile. What randomly varies is the “level” of each attribute, within each profile. For example, a level of the “age” attribute was 21; a level of the “criminal record of person killed attribute ” was “no criminal record.” Table 4 displays all attributes and levels. Upon telling subjects that they would see description of videos of police killings, we described each level. Subjects then were confronted with five sets of two profiles. Upon seeing each set of two profiles, subjects were asked how justified they believed the killing described in the first profile was, and how justified they believed the killing described in the second profile was. To assess justifiability, subjects answered a seven-point scale, ranging from “entirely justified” to “entirely unjustified.”3 To construct the conjoint experiment, we relied on software by Hainmueller, Hopkins, and Yamamoto (2014). Figure 1 displays an example of what a paired profile would look like to respondents. Table 2 displays all attributes and levels.
Published online:
08 November 2018![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0001_ob.jpg"})
Figure 1. Indicative conjoint options facing each respondent.
Table 2. Conjoint survey design.
The attributes and levels were not chosen accidentally. Each attribute could, on its own, warrant an investigation. For example, the relationship between stereotypes and gender has a robust literature (e.g. Cadinu et al. 2005; Carr and Steele 2010). So too does the relationship between stereotypes and race (e.g. Nelson, Sanbonmatsu, and McClerking 2007; Steele 2010). By using a conjoint, we are able to precisely distill the exact effect that each attribute has on attitudes. Our choice of attributes was owed to two factors. First, as our ambition was to unpack the determinants of attitudes toward police killings, we believed that our attributes should include details that featured prominently in media discussions of such killings. The role of media coverage in shaping racial attitudes is well-documented (e.g. Iyengar 1991; Entman and Rojecki 2000). Each level we tested had featured prominently in discussions about at least one of the more prominent shootings. Tamir Rice's youth was highlighted by multiple media accounts (Blow 2015). The public dwelled on Michael Brown's supposed criminal activity before the shooting (Roller 2014). The race of the police officers was featured prominently in coverage of the Freddie Gray case, as photographs of the officers led news accounts, highlighting that police officers from different races were involved in this incident (Stolberg and Oppel 2015). The shooting of Laquan MacDonald was followed by stories about the drugs in his system, and the knife he allegedlywielded – but did not actually wield – at officers (Gorner 2015).
We also tested attributes that might signal whether the victim was or was not conforming to race-related stereotypes about crime. This was tested most directly by including the attribute about the victim's supposed criminal history, and it was tested more by including the reference to the location. Previous work has found that perceptions of the police are often tied up with the location of the respondent, with residents of middle-class neighborhoods perceiving the police differently than residents of lower-class neighborhoods (Weitzer 1999). With that in mind, describing a shooting as occurring in a middle-class neighborhood might spur respondents to regard that shooting as more justified. The victim's age, often mentioned in media reporting of police killings, might also relate to perceptions of innocence, given the relationship between age, crime and jury outcomes (e.g. Farrington 1986; Anwar, Bayer, and Hjalmarsson 2014). In sum, the attributes we tested both increased our ability to shed empirical light on real-life incidents as people might have heard about them through media and helped us speak to a variety of attributes that scholarship suggests could have played a role in affecting attitudes.
In this study, a conjoint design also offered several methodological advantages. Consider what kind of standard survey experiment could have been designed to answer these questions. Before administering the survey, we would have had to make predictions informed by the literature about the specific characteristics (or “levels”) that drive the public's attitudes about police killings. Perhaps people believe shootings of young children are especially unjustified; perhaps they think reported marijuana use by the victim makes a shooting more justifiable. We could have constructed fictitious news articles or vignettes around those predictions and judged responses accordingly. But just as our treatments would have been byproducts of informed guesswork, so too would our understanding of their effects in relationship to other possible characteristics. To understand one level's effect in relationship to others, we would have had to conduct multiple follow-up studies. A level of precision that would have taken multiple rounds with standard survey experiments is achieved with the conjoint design in one fell swoop.
4.3. Study 1 results
Exactly 1352 subjects completed Waves 1 and 2. We are estimating the Average Marginal Component Effect (AMCE) of each attribute, which is best thought of as the difference in probability that one attribute, compared to another, will lead to a shooting being perceived as justified. As each level is randomly assigned within each profile, each displayed level is orthogonal to one another, allowing us to make inferences about the effects of any level on our outcome variable. Hainmueller, Hopkins, and Yamamoto (2015) helpfully compare the inferences drawn from a conjoint survey to the inferences researchers make when reviewing results from survey questions vended in random order. To make our estimates, we run regressions in which a benchmark level is regressed against all other levels within the attribute. For additional details on the procedure, consult Hainmueller, Hopkins, and Yamamoto (2014).
Figure 2 displays the effects of each attribute on the probability that respondents will regard a shooting as justified. To understand how to interpret the figure, turn to the results for the “age” attribute at the top of the plot. The result for a hypothetical 12-year-old victim,.16, indicates that just being shown a profile of a shooting in which the victim was aged 12 years reduces the likelihood that the shooting will be regarded as justified by 16%, in comparison to the baseline level (in this case, 21). Similarly, the result for the “gun” level, within the “victim” attribute, shows that being shown a profile in which the victim is described as having a gun increases the likelihood that the shooting will be regarded as justified by 34%. In this case, that number is in comparison to the baseline level of having no gun.
Published online:
08 November 2018Figure 2. Overall conjoint experiment results. Points to the right of the dashed horizontal indicate shootings with this characteristic were more likely to be deemed justified, relative to the baseline levels indicated by the square. Within each attribute, the square indicates the reference level.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0002_oc.jpg"})
Figure 2. Overall conjoint experiment results. Points to the right of the dashed horizontal indicate shootings with this characteristic were more likely to be deemed justified, relative to the baseline levels indicated by the square. Within each attribute, the square indicates the reference level.
Across the sample, we see respondents making inferences about the shooting's justifiability based on individual traits of the victim. Shootings with young victims are less likely to be regarded as justified; shootings in which victims are said to have guns or knives are likely to lead respondents to think of the shooting as more justified. Criminal records matter too – shootings in which the victim is reported to have an extensive criminal record are 16% more likely to be regarded as justified than shootings in which the victim has no criminal record. Equally as interesting as what we do observe is what we do not observe. The race of the police officer did not matter in our analysis. The number of complaints a police officer had received mattered somewhat, but hardly so in comparison to the criminal record of the victim. Respondents seem conditioned to believe the police irrespective of conditions. This is an especially important finding because it raises the question of what individuals need to know about a police officer involved in a shooting to move the burden of proof of innocence from the victim to the officer. The finding also suggests that without controlling for trust in the police, it appears that most individuals are willing to give the police the benefit of doubt when it comes to police officers killing civilians.
In contrast, when we vary the purported gender of the victim, we observe significant effects, suggesting that shootings with male victims are regarded as more justifiable than shootings with female/transgender victims, but only slightly. Finally, we saw effects on race of the victim, but in the opposite direction anticipated; shootings with black victims were seen as slightly less justifiable than shootings with white or Latino victims. The negative effect is small and could be explained by some combination of social desirability bias, wherein subjects know the socially “correct” answer, and an awareness caused by actual shootings which led subjects to think that shootings with black victims are less justified. It is possible that the constant stream of news about police shooting, and possibly protests over police shootings, had conditioned subjects to be especially skeptical of such shootings of black victims.
4.4. Symbolic racism
While the first analysis tells us how varying characteristics impacts the general trend in justifiability, we are also interested in how individuals who hold greater anti-black attitudes respond to the varying characteristics. We disaggregated SRS scores into quartiles, regarding subjects in the bottom quartile as scoring low on the SRS scale and those in the top quartile as high on the SRS scale. Figure 3 displays the effects for each level on the probability that respondents will regard a shooting as justified. For each attribute, the benchmark level in the regression is noted by a square. Hollow circles denote effects for those who score low on SRS, while solid circles denote effects for those who score high.
Published online:
08 November 2018Figure 3. Subgroup conjoint experiment – effect of conjoint attributes for respondents of high symbolic racism (indicated by the solid points) and low symbolic racism (hollow points). Within each attribute, the square indicates the reference level, shared between the model types.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0003_ob.jpg"})
Figure 3. Subgroup conjoint experiment – effect of conjoint attributes for respondents of high symbolic racism (indicated by the solid points) and low symbolic racism (hollow points). Within each attribute, the square indicates the reference level, shared between the model types.
The circles largely follow one another, with most differences between high and low SRS respondents registering as statistically significant but substantively small. The reader should pay special attention to the marked difference on the question of shooting location. High SRS respondents become 10% more likely to view a shooting as justified merely on the basis of its location, with high SRS respondents more likely to think of a shooting as justified if it occurred in a “known gang zone.” We also observe effects on the race of the victim and the race of the police officer. On the race of the victim, high SRS respondents were more likely to view a shooting as justified if the victim were Black or Latino/a. On the race of the officer, we observe effects, but in the opposite direction; high SRS subjects regarded shootings by Black or Latino/a officers as less justifiable.
These results suggest that high SRS subjects make use of stereotypes differently than their low SRS counterparts. That high SRS subjects were responsive to the location of the shooting suggests that they are more willing to generalize from environmental evidence. While all subjects were responsive to the individual characteristics offered in our scenarios – the age of the victim, whether she was reported to use drugs, her reported criminal history – SRS subjects were alone in the effect that the “evidence” offered by the location shooting had on them. Indeed, the difference between locations in our study had no effect on low SRS subjects. High SRS subjects make inferences about police shootings not only based on characteristics of the shooting victim, but characteristics of the surrounding environment. To be sure, both those with low and high SRS scores relied on stereotypes; no one is absolved from relying on stereotypes to make judgments about police shootings. That said, as anticipated by Hypothesis 1, those who exhibit high levels of SRS are willing to rely on non-individual characteristics like environmental cues. (To see how responses varied with subjects’ level of trust in the police, consult the appendix.)
5. Study 2
To determine whether the beliefs about justifiability we observed in Study 1 would translate to another experimental context and to evaluate whether justifiability beliefs were shaped by social movement activity and electoral politics, we administered Study 2. For this experiment, we created newspaper articles, modeled on actual media coverage, about hypothetical police shootings.
We constructed profiles of two victims. One victim was described as having characteristics that respondents indicated made a police shooting seemingly justified. We call that our Justified Hypothetical Victim. Our second hypothetical victim was described as having characteristics that respondents indicated made a police shooting was less justified. We call this our Unjustified Hypothetical Victim. In addition, in some of the articles we randomly included primes relating to the BLM movement and the presidential candidacy of Donald J. Trump. The experiment was administered three weeks before the 2016 presidential election and two months after the completion of Study 1.
Here are the six experimental conditions:
Unjustified Hypothetical Victim/No Prime: The article relayed a story about the police shooting of a 12-year-old girl with no prior convictions, who did not appear to be on any substances, that occurred in a middle-class neighborhood, by a police officer who had previously received civilian complaints.
Unjustified Hypothetical Victim/Black Lives Matter Prime: The article relayed a story about a police shooting of the same victim as described above. The article also contained a hypothetical statement of mourning from the local Black Lives Matter chapter.
Unjustified Hypothetical Victim/Trump Prime: The article again relayed a story of the same victim. This time, the article contained a hypothetical quote from Donald J. Trump about the shooting, in which he urged his listeners not to blame police.
Justified Hypothetical Victim/No Prime: The article relayed a story about the police shooting of a 30-year-old man with an extensive criminal record, who witnesses said appeared high on PCP, with a gun in a known gang area, by a police officer with no known record of complaints.
Justified Hypothetical Victim/Black Lives Matter Prime: The article relayed a story about the police shooting of the same victim described above. The article also contained a hypothetical statement of mourning from the local BLM chapter.
Justified Hypothetical Victim/Trump Prime: The article relayed a story about the police shooting of the same victim, while also containing a hypothetical quote from Donald J. Trump, in which he urged his listeners not to blame police.
The test of the resultant articles is provided in Figure 4 on page 22.
Published online:
08 November 2018Figure 4. Text scheme for study 2 treatments. Italicized text indicates article characteristics which have were randomized for either a high (the first entry) or low justifiability (the second entry). The second dimension of random assignment – whether respondents saw a BLM or Trump prime – are captured in the paragraphs inside brackets. Across both randomization types, each respondent saw either a description of a low or high justifiability shooting, and a Trump or a BLM prime.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0004_ob.jpg"})
Figure 4. Text scheme for study 2 treatments. Italicized text indicates article characteristics which have were randomized for either a high (the first entry) or low justifiability (the second entry). The second dimension of random assignment – whether respondents saw a BLM or Trump prime – are captured in the paragraphs inside brackets. Across both randomization types, each respondent saw either a description of a low or high justifiability shooting, and a Trump or a BLM prime.
Given the well-known relationship between media coverage and racial attitudes (e.g. Iyengar 1991; Entman and Rojecki 2000), we based our treatments on articles that appeared in U.S. media. We modeled our treatments on an article about the Michael Brown shooting (O'Neil 2014) and an article about the Alton Sterling shooting (Stole 2016); the latter article featured a family member describing Sterling in ways similar to the description we used in the BLM prime. We edited the articles somewhat; the variation in the final paragraph of each treatment is owed to concerns about differential length. The longer version of the paragraph was used in those conditions without additional primes. The study was administered on the same subjects who participated in Study 1, recontacting them eight weeks later.
After reading their article, subjects evaluated the justifiability of the victim's death, using the Study 1 scale. They were then asked how important they believed the issue of “police shootings of black people” is. We presented subjects with feeling thermometers of Donald Trump and the BLM movement. As a manipulation check, we asked subjects how old the victim they saw was, and as an attention check, we asked what time the press conference would occur. The latter was an especially aggressive attention check, given that the information was conveyed at the end of the article.
5.1. Study 2 results
Exactly 826 subjects from our original sample were recruited and successfully completed Study 2. Subjects were significantly more likely to view a killing as justified if the victim was described with the high deserving attributes identified from the conjoint data. On our 1–7 scale, higher values indicate greater belief in justifiability. Subjects in the Justified Hypothetical Victim/No Prime condition reported a mean response of 4.54, or in between “Neither Justified and Justified ” and “Somewhat Justified.” Meanwhile, subjects in the Unjustified Hypothetical Victim/No Prime reported a mean response of 2.13, close to the “Unjustified” choice. This 2.4 difference is significant by conventional standards (p<.000). The Justified Hypothetical Victim's shooting was regarded as more justified when paired with either the Trump or BLM prime. The same was true of the Unjustified Hypothetical Victim's shooting – regardless of the prime associated with it, people were comparatively skeptical that the shooting was justified. The statement of mourning voiced by BLM did not change subjects' attitudes in one direction or another, nor did Trump's defense of the police.
An OLS model of the results further affirms this finding. In column 1 in Table 3, we estimate an OLS model of the perceived justifiability of the shooting, with the Justified Hypothetical Victim/No Prime condition omitted. The coefficients for the Justified Hypothetical Victim are large and significant, regardless of the prime. (The coefficients for this victim across conditions are not statistically distinct from one another.) The coefficients for the Unjustified Hypothetical Victim, meanwhile, are small and insignificant across conditions. In column 2 in Table 3, we restrict the model to those who passed the hardest manipulation check, for which correct answers could be obtained by reading to the end of the treatment. Our results remain robust to this restriction.
Table 3. Overall study 2 estimates of treatment effects on perceived justifiability of shooting. Column 1 includes all respondents. Columns 2 only includes respondents who passed manipulation check. intercept omitted for simplicity.
Next, we consider how subjects' pre-treatment characteristics interacted with their susceptibility to the BLM and Trump primes. In Figure 5, we show how the primes mattered for the Hypothetical Justified and Unjustifiable and victims by SRS, by confidence in police, by partisanship, by ideology and by level of education. The plots each reflect separate OLS models with the unjustified/no prime condition as the excluded condition. First, consider SRS related to responses to Hypothetical Justified and Unjustified victims. Of special interest are the responses of high SRS subjects to the Hypothetical Unjustifiable Victim. Such subjects are motivated by the Trump prime to view this victim's shootings as more justifiable – but the BLM prime pushes them significantly in the other direction. The BLM prime makes high SRS subjects view the shooting of a hypothetical unjustifiable victim as significantly less justified than a Trump prime. For the Hypothetical Justified Victim, high SRS subjects who are exposed to the BLM prime effectively come to view the shooting as more justified than the version with the Trump prime (though the difference for the Hypothetical Justified Victim is not significant). For no other condition, and for no other level of SRS, did seeing the BLM prime deflate subjects' belief in the justifiability of the shooting. It only did so for high SRS subjects.4
Published online:
08 November 2018Figure 5. Treatment effects as a function of Trump (light bars) or and BLM (dark bark) frames. Each bar shows the mean difference compared to demographically similar respondents (indexed on the y-axis) exposed to no frame. Differences are measured on a seven point scale. Significant differences to control (p<.05) are indicated by a black border. Redundantly, significant differences are also labeled with asterisks (* p<.05, ** p<.01, *** p<.001). Column facets show the effect of changing the shooting's circumstance either to encourage respondents to justify the shooting – “justified hypothetical victim” – or to believe the shooting was not justified – “unjustified hypothetical victim.” Row facets interact different respondent characteristics with the treatments. The regression models which provide these estimates are described in Table 4 in the appendix.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0005_ob.jpg"})
Figure 5. Treatment effects as a function of Trump (light bars) or and BLM (dark bark) frames. Each bar shows the mean difference compared to demographically similar respondents (indexed on the y-axis) exposed to no frame. Differences are measured on a seven point scale. Significant differences to control (p<.05) are indicated by a black border. Redundantly, significant differences are also labeled with asterisks (* p<.05, ** p<.01, *** p<.001). Column facets show the effect of changing the shooting's circumstance either to encourage respondents to justify the shooting – “justified hypothetical victim” – or to believe the shooting was not justified – “unjustified hypothetical victim.” Row facets interact different respondent characteristics with the treatments. The regression models which provide these estimates are described in Table 4 in the appendix.
What first might appear to be an especially surprising result makes more sense when we consider the content of the BLM prime. Subjects in the BLM prime were exposed to one specific example of BLM rhetoric. In his prime, (based on how Sterling's father described him (Stole 2016), BLM mentioned the victim's family and made clear that the victim had been loved. Had the victim not been shot, he or she would be “contributing to their community.” For high SRS subjects, this rhetoric may have been unusually effective in turning the victim into a person deserving of sympathy. Recall Sniderman et al. (1991) finding that conservatives are more likely than liberals to be sympathetic toward a laid-off black worker, contingent on upon the worker being described as a “dependable worker.” A similar dynamic may be at play in high SRS subjects' responses to the BLM prime when attached to the Unjustified Hypothetical Victim. The victim described – a beloved family member, a community contributor – arguably conforms to “traditional” American values of hard work and nuclear family dynamics. High SRS subjects may find this combination novel and re-evaluate the victim as one who has met traditional values. They thus object more to her death than they would otherwise. There are limits to this finding; a similar response to the BLM prime was not observed when the prime was used in relationship to the Justified Hypothetical Victim. We should not interpret this finding as evidence of high SRS subjects' broad sympathy for BLM. A different example of BLM rhetoric may have produced much different results. Further research on the relationship between attitudes toward BLM and SRS is needed.
6. Discussion
We live in a time when increased attention is being paid to the shooting of civilians, in particular black Americans, by the police. Individuals are provided limited information and are being asked to draw conclusions on these shootings. We believe that researchers must explore what factors shape how individuals view these shootings. What causes people to view such shootings as justifiable? And can such evaluations be affected by politicians and social movements? To evaluate the causal determinants of mass attitudes about police shootings, we employed a two-wave conjoint survey experiment in which we varied attributes of hypothetical police killings. Across the sample, we observed subjects drawing inferences about the justifiability of such shootings based on their knowledge of the individuals involved in the shooting. Shootings in which the victim had an extensive criminal record, had a gun, or was reported to be using psychedelic drugs were especially likely to be viewed as justified. Conversely, shootings with young victims were more likely to be viewed as unjustified.
In addition, we find individuals who exhibit high SRS levels heavily weighing environmental evidence. This suggests that SRS levels condition what kind of stereotypes individuals use when evaluating police shootings. Those who demonstrate high SRS need only environmental evidence to judge the shooting as more justifiable, while those who exhibit low SRS dwell more on individual-level information. Our results indicate that SRS also relates to the kinds of evidence subjects use when evaluating racially charged and racist events. This lends credence to our first hypothesis, which anticipated that SRS levels would affect subjects' beliefs about justifiability.
We corroborate these initial findings about justifiability by administering another experiment. For Study 2, we showed subjects fictitious newspaper articles about police shootings, in which we varied shooting descriptions by hypothetical justifiability and included BLM or Trump primes. We find that the results from Study 1 hold. Across conditions, the shooting of the Justified Hypothetical Victim was regarded as much more justified.
Our results over both studies speak to how individuals, especially high SRS individuals, make use of stereotypes, in particular racist stereotypes. In Study 1, high SRS individuals were uniquely responsive to the location of the shooting, suggesting that they are more willing to extrapolate from environmental evidence than their low SRS counterparts. It may be that high SRS individuals are looking for a reason to believe the shooting by a police officer is justified. While all subjects were responsive to the individual characteristics offered in our scenarios, high SRS subjects were alone in the effect that the “evidence” offered by the location shooting had on their assessment of justifiability. Indeed, the difference between locations in our study had no effect on low SRS subjects. High SRS subjects make inferences about police shootings not only based on characteristics of the shooting victim himself, but characteristics of the victim's environment.
In Study 2, we show that the same high SRS subjects are less likely to view the shooting of a hypothetical unjustified victims as justified when they are described as meeting traditional American values. Study 2 demonstrates that what we observed from the conjoint holds in a more traditional survey experiment environment. In addition, in a variation of prior research into new racism, we find that those with high SRS view the shooting of a Unjustified Hypothetical Victim as less justified than those who have lower SRS scores when a BLM prime is included. The same is true when we compare conservatives to liberals and Republicans to other partisans. When BLM describes victims as meeting “traditional” American values, they may be able to allow conservatives and Republicans to see the humanity of black Americans. This lends support to our second hypothesis, which posited that there might be a relationship between how high SRS subjects perceived shootings and the extent to which victims were reported to conform to what are thought to be traditional American values.
Our studies are not without their limitations. First, our findings may have been contingent on the precise time of each study's administration. Study 1 was fielded in mid-July 2016, shortly after the high-profile shootings of Alton Sterling and Philando Castile. The tragic fact is that police killings have long been prominent in mass media and salient in the public mind. It is possible that, were police killings not as common as they are, respondents' attitudes would have been less well-developed than we observed. If, for example, no police killings had been widely reported around the time of the study, more inchoate attitudes might have resulted in less cogent responses. Yet, sadly, that is not the world we live in. Study 2 was fielded in early October 2016, when the presidential campaign was nearing its end. If we had run the study after the election of President Trump, it stands to reason that the treatments related to him might have yielded different effects.
We were also limited by our decision to frame the treatments in Study 1 as stemming from videos, without actually showing subjects the videos. As discussed above inNote 1, we did so in order to balance experimental realism with experimenter control. Similarly, in Study 2, we were limited by existing news articles and descriptions of victims, limiting our ability to separate the effects of a BLM prime from a traditional values prime. Our evidence only suggests that people are willing to set aside some concerns they may have about BLM, as they are moved by traditional value primes. Finally, while our use of a conjoint design permits us to test a large number of potential determinants, we did not test all possible factors. There may be more at work beyond how the victim, the police and the location of the shooting are described. Future research should pursue these lines of inquiry.
6.1. Appendix study 1: results and trust in police
In Study 1, because we believed that subjects' trust in police might shape their responses, we asked them a question from Gallup's battery of questions about trust in institutions. Subjects were asked: “As an institution, how much confidence do you have in the police in the United States?” and allowed answers on a five-point scale. Next, we adopt the same procedure to that we used when evaluating the relationship between Study 1 results and symbolic racism. We anticipated that those with higher levels of trust in the police will be less likely to see most killings as unjustifiable. Figure 6 displays the effects for each level on the probability that respondents will regard a shooting as justified, separating the bottom quartile of trust in police respondents from the top. Again, squares denote the benchmark level for each regression, solid points represent effects for those who score high and hollow points represent effects for those who score low. Once again, the effects for low scores tend to lag behind the effects for high scores. Subjects who reported high trust in the police have a lower evidentiary threshold for believing a shooting was justified. The difference in the effects on marijuana is especially pronounced. Respondents with low trust in the police effectively disregarded marijuana consumption. In contrast, high trust respondents were nearly 10% more likely to regard a shooting as justified if the victim had consumed marijuana. In addition, though the difference is smaller than that observed in the symbolic racism section, those who are very trusting of police take away more from the location of the shooting than those with low trust in police. Also worth noting are the effects on victim; those who trust police are more likely to view shootings as justified if the victim is Latino. One possible explanation for this is that, while subjects believe that race matters, social desirability effectively constrains their responses to a black victim. Speculatively, they may transfer their remaining racial anxiety to Latino victims.
Published online:
08 November 2018Figure 6. Subgroup conjoint experiment—effect of conjoint attributes for respondents with high confidence in the police (indicated by the solid points) and low confidence in the police (hollow points). Within each attribute, the square indicates the reference level, shared between the model types.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0006_ob.jpg"})
Figure 6. Subgroup conjoint experiment—effect of conjoint attributes for respondents with high confidence in the police (indicated by the solid points) and low confidence in the police (hollow points). Within each attribute, the square indicates the reference level, shared between the model types.
| Study 2 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Wave 3 | ||||||||||
| Study 1 | Hypothetical Justified Victim | Hypothetical Unjustified Victim | ||||||||
| Wave 1 | Wave 2 | None | BLM | Trump | None | BLM | Trump | Balance | ||
| n | 1723 | 1352 | 146 | 153 | 155 | 130 | 143 | 143 | ||
| 18–34 | 60 | 57 | 53 | 50 | 48 | 51 | 46 | 42 |  | |
| Age | 35−54 | 32 | 34 | 37 | 36 | 39 | 40 | 42 | 42 |  |
| 55+ | 8 | 9 | 10 | 14 | 13 | 10 | 12 | 16 | ||
| HSD or Less | 12 | 11 | 12 | 16 | 9 | 12 | 9 | 15 |  | |
| Education | Some College | 38 | 37 | 31 | 31 | 29 | 36 | 34 | 43 |  |
| BA+ | 50 | 52 | 57 | 54 | 62 | 52 | 57 | 43 | ||
| Full-Time Employee | 63 | 64 | 68 | 59 | 61 | 67 | 63 | 69 |  | |
| Employment | Self Employed | 19 | 19 | 15 | 18 | 20 | 20 | 19 | 14 |  |
| Unemployed | 8 | 8 | 8 | 10 | 9 | 6 | 6 | 8 | ||
| Not Working | 10 | 9 | 9 | 13 | 10 | 8 | 11 | 9 | ||
| Gender | Female | 45 | 45 | 47 | 46 | 46 | 48 | 46 | 49 |  |
| Male | 55 | 55 | 53 | 54 | 54 | 52 | 54 | 51 |  | |
| Income |  | 43 | 43 | 42 | 53 | 40 | 39 | 32 | 45 |  |
 | 40 | 39 | 39 | 30 | 44 | 37 | 50 | 41 |  | |
 | 18 | 18 | 19 | 17 | 16 | 25 | 18 | 14 | ||
| Conservative | 24 | 24 | 19 | 24 | 25 | 33 | 26 | 27 |  | |
| Ideology | Moderate | 19 | 19 | 23 | 16 | 17 | 18 | 21 | 14 |  |
| Liberal | 56 | 57 | 59 | 60 | 58 | 48 | 53 | 59 | ||
| Republican | 24 | 24 | 20 | 21 | 26 | 27 | 23 | 27 |  | |
| Partisanship | Independent | 25 | 24 | 28 | 23 | 21 | 26 | 26 | 23 |  |
| Democrat | 52 | 53 | 52 | 56 | 53 | 47 | 51 | 50 | ||
| White | 78 | 78 | 85 | 77 | 80 | 81 | 82 | 80 |  | |
| Race | Black | 6 | 6 | 5 | 8 | 6 | 6 | 5 | 7 |  |
| Hispanic | 6 | 5 | 5 | 8 | 7 | 5 | 6 | 3 | ||
| Other | 10 | 10 | 6 | 7 | 8 | 8 | 7 | 9 | ||
| North Central | 20 | 21 | 20 | 21 | 20 | 14 | 27 | 17 |  | |
| Region | Northeast | 14 | 13 | 11 | 14 | 15 | 13 | 17 | 16 |  |
| South | 31 | 30 | 29 | 32 | 24 | 30 | 31 | 34 | ||
| West | 35 | 35 | 38 | 32 | 40 | 42 | 24 | 33 | ||
| Attribute | Levels |
|---|---|
| Age of Person Killed | 12, 15, 18, 21, 24, 27, 30 |
| Gender of Person Killed | Male, Female, Transgender |
| Criminal Record of Person Killed | No Criminal Record, Short Criminal Record, Extensive Criminal Record |
| Weapon in Possession of Person Killed | No weapon, gun, knife |
| Policeman's Disciplinary Record | Spotless record, Some complaints, many complaints |
| Location of Shooting | Middle-class neighborhood, known gang area |
| Substances Taken By Person Killed | No substances taken, Marijuana, alcohol, PCP |
| Race of Person Killed | White, Latino, Black |
| Race of Policeman | White, Latino, Black |
| All respondents | Passed manipulation check | |
|---|---|---|
| Unjustified Victim No Prime | 2.41*** | 2.64*** |
| (. 15) | (.16) | |
| Justified Victim BLM Prime | 2.46** | 2.59*** |
| (.15) | (.16) | |
| Unjustified Victim BLM Prime | −.13 | .05 |
| (.16) | (.18) | |
| Justified Victim Trump Prime | 2.39*** | 2.57*** |
| (.15) | (.17) | |
| Unjustified Victim Trump Prime | .04 | .09 |
| (.15) | (.17) | |
| n | 826. | 602. |
 | .48 | .53 |
Adjusted  | .38 | .52 |
| Variable interacted with condition | |||||||
|---|---|---|---|---|---|---|---|
| Education | Ideology | Partisanship | Conf. Police | Symb.Racism | No Interactions | ||
| High Deserving Victim, No Prime | −.32 (.44) | −.04 (.2) | −.24 (.21) | .59 (.58) | −.16 (.15) | −.13 (.15) | |
| High Deserving Victim, Trump Prime | .52 (.47) | .05 (.2) | −.04 (.21) | −.54 (.57) | 0 (.15) | −.02 (.14) | |
| Low Deserving Victim, BLM Prime | −3.35*** (.49) | −2.6*** (.21) | −2.75*** (.22) | −2.8*** (.61) | −2.67*** (.15) | −2.69*** (.15) | |
| Low Deserving Victim, No Prime | −2.48*** (.43) | −2.57*** (.21) | −2.58*** (.22) | −1.55** (.57) | −2.58*** (.15) | −2.56*** (.15) | |
| Low Deserving Victim, Trump Prime | −2.92*** (.41) | −2.41*** (.2) | −2.66*** (.22) | −2.53*** (.58) | −2.39*** (.15) | −2.39*** (.15) | |
| Some College | −.07 (.35) | −.14 (.14) | |||||
| BA + | −.2 (.32) | −.16 (.14) | |||||
| High Deserving Victim, No Prime x Some College | .2 (.53) | ||||||
| High Deserving Victim, Trump Prime x Some College | −.64 (.56) | ||||||
| Low Deserving Victim, BLM Prime x Some College | .8 (.57) | ||||||
| Low Deserving VictimNo Prime x Some College | .09 (.51) | ||||||
| Low Deserving Victim, Trump Prime x Some College | .42 (.49) | ||||||
| High Deserving Victim, No Prime x BA + | .21 (.49) | ||||||
| High Deserving Victim, Trump Prime x BA + | −.55 (.52) | ||||||
| Low Deserving Victim, BLM Prime x BA + | .78 (.54) | ||||||
| Low Deserving Victim, No Prime x BA + | −.01 (.48) | ||||||
| Low Deserving Victim, Trump Prime x BA + | .82 (.47) | ||||||
| Moderate | .31 (.31) | .31* (.13) | |||||
| Conservative | 1*** (.27) | .44* (.2) | |||||
| High Deserving Victim, No Prime x Moderate | −.11 (.41) | ||||||
| High Deserving Victim, Trump Prime x Moderate | −.04 (.43) | ||||||
| Low Deserving Victim, BLM Prime x Moderate | .34 (.43) | ||||||
| Low Deserving Victim, No Prime x Moderate | .57 (.43) | ||||||
| Low Deserving Victim, Trump Prime x Moderate | .47 (.45) | ||||||
| High Deserving Victim, No Prime x Conservative | −.3 (.4) | ||||||
| High Deserving Victim, Trump Prime x Conservative | −.33 (.38) | ||||||
| Low Deserving Victim, BLM Prime x Conservative | −.6 (.39) | ||||||
| Low Deserving Victim, No Prime x Conservative | −.3 (.36) | ||||||
| Low Deserving Victim, Trump Prime x Conservative | −.24 (.37) | ||||||
| Independent | −.17 (.27) | −.15 (.13) | |||||
| Republican | .92** (.29) | −.2 (.2) | |||||
| High Deserving Victim, No Prime x Independent | .66 (.38) | ||||||
| High Deserving Victim, Trump Prime x Independent | .27 (.39) | ||||||
| Low Deserving Victim, BLM Prime x Independent | .72 (.39) | ||||||
| Low Deserving VictimNo Prime x Independent | .63 (.38) | ||||||
| Low Desert, Trump Prime x Independent | .99* (.39) | ||||||
| High Deserving Victim, No Prime x Republican | −.4 (.41) | ||||||
| High Deserving Victim, Trump Prime x Republican | −.36 (.39) | ||||||
| Low Desert, BLM Prime x Republican | −.45 (.41) | ||||||
| Low Desert, No Prime x Republican | −.38 (.39) | ||||||
| Low Desert, Trump Prime x Republican | −.04 (.39) | ||||||
| Confidence in Police | .34** (.13) | .14** (.05) | |||||
| High Deserving Victim, No Prime x Confidence in Police | −.22 (.18) | ||||||
| High Deserving Victim, Trump Prime x Confidence in Police | .17 (.17) | ||||||
| Low Desert, BLM Prime x Confidence in Police | .05 (.18) | ||||||
| Low Desert, No Prime x Confidence in Police | −.28 (.17) | ||||||
| Low Desert, Trump Prime x Confidence in Police | .06 (.17) | ||||||
| Symbolic Racism | .63*** (.11) | .45*** (.05) | |||||
| High Deserving Victim, No Prime x Symbolic Racism | −.06 (.16) | ||||||
| High Deserving Victim, Trump Prime x Symbolic Racism | −.09 (.15) | ||||||
| Low Desert, BLM Prime x Symbolic Racism | −.35* (.16) | ||||||
| Low Desert, No Prime x Symbolic Racism | −.08 (.15) | ||||||
| Low Desert, Trump Prime x Symbolic Racism | −.02 (.14) | ||||||
| r.squared | .49 | .52 | .52 | .52 | .58 | .58 | |
| adj.r.squared | .48 | .51 | .51 | .51 | .58 | .58 | |
| statistic | 4.49 | 45.59 | 44.61 | 7.21 | 9.31 | 78.47 | |
| BIC | 2497.45 | 2453.73 | 2461.97 | 2423.8 | 2323.18 | 2318.99 | |
| High justifiability victim, no prime | −.12 (.15) |
| High justifiability victim, Trump prime | .01 (.14) |
| Low justifiability victim, BLM prime | −2.67 |
| Low justifiability victim, no prime | −2.55 |
| Low justifiability victim, Trump prime | −2.37 |
| Symbolic racism scale | 2.16 |
| Confidence in police | .99 |
| SRS × Conf. in police | −.71 (.61) |
| Observations | 725 |
| R2 | .59 |
| Adjusted R2 | .58 |
Note: 
p<0.1;
p<0.05;
p<0.01
| Perceived justifiability of shooting | |
|---|---|
| Condition: High justifiability, no prime | .15 (.36) |
| Condition: High justifiability, Trump prime | −.48 (.31) |
| Condition: Low justifiability, BLM prime | −2.76 |
| Condition: Low justifiability, no prime | −2.38 |
| Condition: Low justifiability, Trump prime | −2.19 |
| Symbolic Racism Scale (SRS) | .20 (.24) |
| Race: White | .29 (.24) |
| Condition: High justifiability, no prime * SRS | .26 (.44) |
| Condition: High justifiability, Trump prime * SRS | .41 (.30) |
| Condition: Low justifiability, BLM prime * SRS | −.14 (.38) |
| Condition: Low justifiability, no prime * SRS | .13 (.37) |
| Condition: Low justifiability, Trump prime * SRS | .53 (.32) |
| Condition: High justifiability, no prime * White | −.40 (.39) |
| Condition: High justifiability, Trump prime * White | .59 |
| Condition: Low justifiability, BLM prime * White | .08 (.38) |
| Condition: Low justifiability, no prime * White | −.26 (.35) |
| Condition: Low justifiability, Trump prime * White | −.25 (.36) |
| SRS * White | .54 |
| Condition: High justifiability, no prime * White * SRS | −.41 (.47) |
| Condition: High justifiability, Trump prime * White * SRS | −.65 |
| Condition: Low justifiability, BLM prime * White * SRS | −.31 (.42) |
| Condition: Low justifiability, no prime * White * SRS | −.29 (.40) |
| Condition: Low justifiability, Trump prime * SRS | −.70 |
| Observations | 724 |
 | .59 |
Adjusted  | .58 |
| Residual std. error | 1.13 (df = 700) |
| F-statistic | 44.49 |
Note: 
.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Although we did not have actual videos, the aim in claiming otherwise was to increase the chance that respondents would take the hypotheticals seriously and engage with them as if they reflected an underlying incident. Otherwise, we feared that participants would generally disregard all differences we presented to them. There is always a trade-off between experimental realism and experimenter control. We tried to balance those two objectives, using descriptions of videos to enhance realism but not actually showing videos to maintain control. It remains possible that responses may have been confounded by people thinking about real-world incidents in response to our experimental treatments but that would also have been a risk had we shown fictionalized videos.
2 The reader will notice that we have made a decision not to identify those killed by the police as victims in the survey, as describing them as victims might have affected results. This was a difficult decision to make.
3 This is a novel survey item we constructed for this study. It is strongly related, in the expected direction, to scores on symbolic racism, ideology, and confidence in the police. Different responses on this perceived justifiability scale related to significant differences on these related scales. These mean differences by scale are depicted in the appendix, in Figure 1.
4 A reviewer speculated SRS and confidence in the police were actually tapping the same underlying dimension, and that their effect on perceived justifiability might be interactive. The two variables are in fact only weakly correlated (.28) and their interaction does not significantly predict perceived justifiability. A regression estimate is included in the appendix, in Table 5.
References
- Allport, Gordon. 1954. The Nature of Prejudice.
New York : Basic Books. [Google Scholar] - Anwar, Shamena, Patrick Bayer, and Randi. Hjalmarsson. 2014. “The Role of Age in Jury Selection and Trial Outcomes.” Journal of Law, Economics and Organization 57 (4): 1001–1030. doi: 10.1086/675257 [Crossref], [Google Scholar]
- Berinsky, Adam J., Gregory A. Huber, and Gabriel S. Lenz. 2012. “Evaluating Online Labor Markets for Experimental Research: Amazon.com's Mechanical Turk.” Political Analysis 20 (3): 351–368. doi: 10.1093/pan/mpr057 [Crossref], [Web of Science ®], [Google Scholar]
- Blow, Charles. 2015. “Tamir Rice and the Value of Life.” The New York Times, January 11. [Google Scholar]
- Bobo, Lawrence, and Vincent L. Hutchings. 1996. “Perceptions of Racial Group Composition.” American Sociological Review 61: 951–972. doi: 10.2307/2096302 [Crossref], [Web of Science ®], [Google Scholar]
- Bobo, Lawrence, James R. Kugel, and Ryan A. Smith. 1997. “Laissez-faire Racism: The Crystallization of a Kinder, Gentler, Antiblack Ideology.” In Racial Attitudes in the 1990s: Continuity and Change, edited by Steven A. Tuch and Jack K. Martin. New York, NY: Russell Sage Foundation. [Google Scholar]
- Bonilla-Silva, Eduardo, and Tyrone A. Forman. 2000. “I Am Not A Racist, But….” Discourse and Society 11 (1): 50–85. doi: 10.1177/0957926500011001003 [Crossref], [Web of Science ®], [Google Scholar]
- Cadinu, Mara, Anne Maass, Alessandra Rosabianca, and Jeff Kiesner. 2005. “Why Do Women Underperform Under Stereotype Threat?: Evidence for the Role of Negative Thinking.” Psychological Science 16 (7): 572–578. doi: 10.1111/j.0956-7976.2005.01577.x [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Carr, Priyanka B., and Claude Steele. 2010. “Stereotype Threat Affects Financial Decision Making.” Psychological Science 21 (10): 1411–1416. doi: 10.1177/0956797610384146 [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Christenson, Dino P., and David M. Glick. 2013. “Crowdsourcing Panel Studies and Real-Time Experiments in MTurk.” The Political Methodologist 20 (2): 27–32. [Google Scholar]
- Correll, Joshua, Bernadette Park, Charles M. Judd, and Bernd Wittenbrink. 2002. “The Police Officer's Dilemma: Using Ethnicity to Disambiguate Potentially Threatening Individuals.” Journal of Personality and Social Psychology 83 (6): 1314–1329. doi: 10.1037/0022-3514.83.6.1314 [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Devine, Patricia G.. 1989. “Stereotypes and Prejudice: Their Automatic and Controlled Components.” Journal of Personality and Social Psychology 56 (1): 5–18. doi: 10.1037/0022-3514.56.1.5 [Crossref], [Web of Science ®], [Google Scholar]
- Duncan, Birt L.. 1976. “Differential Social Perception and Attribution of Intergroup Violence: Testing the Lower Limits of Stereotyping of Blacks.” Journal of Personality and Social Psychology 34 (4): 590–598. doi: 10.1037/0022-3514.34.4.590 [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Eberhardt, Jennifer L., Valerie Purdie, Phillip Goff, and Paul Davies. 2004. “Seeing Black: Race, Crime, and Visual Processing.” Journal of Personality and Social Psychology 87 (6): 876–893. doi: 10.1037/0022-3514.87.6.876 [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Entman, Robert, and Andrew. Rojecki. 2000. The Black Image in the White Mind. Chicago, IL: University of Chicago Press. [Google Scholar]
- Farrington, David P.. 1986. “Age and Crime.” Crime and Justice 7: 189–250. doi: 10.1086/449114 [Crossref], [Web of Science ®], [Google Scholar]
- Gorner, Jeremy. 2015. “PCP Found in Body of Teen Shot 16 Times by Chicago Cop”. April 15. [Google Scholar]
- Greenwald, Anthony G., Mark A. Oakes, and Hunter G. Hoffman. 2003. “Targets of Discrimination: Effects of Race on Responses to Weapons Holders.” Journal of Experimental Social Psychology 39: 399–405. doi: 10.1016/S0022-1031(03)00020-9 [Crossref], [Web of Science ®], [Google Scholar]
- Hainmueller, Jens, Daniel Hopkins, and Teppei. Yamamoto. 2014. “Causal Inference in Conjoint Analysis: Understanding Multi-Dimensional Choices via Stated Preference Experiments.” Political Analysis 22 (1): 1–30. doi: 10.1093/pan/mpt024 [Crossref], [Web of Science ®], [Google Scholar]
- Hainmueller, Jens, Daniel Hopkins, and Teppei Yamamoto. 2015. “The Hidden American Immigration Consensus: A Conjoint Analysis of Attitudes toward Immigrants.” American Journal of Political Science59 (3): 529–548. doi: 10.1111/ajps.12138 [Crossref], [Web of Science ®], [Google Scholar]
- Henry, P. J., and David O. Sears. 2002. “The Symbolic Racism 2000 Scale.” Political Psychology 3 (2): 253–283. doi: 10.1111/0162-895X.00281 [Crossref], [Google Scholar]
- Hetey, Rebecca, and Jennifer L. Eberhardt. 2014. “Racial Disparities in Incarceration Increase Acceptance of Punitive Policies.” Psychological Science 25 (10): 1949–1954. doi: 10.1177/0956797614540307 [Crossref], [PubMed], [Web of Science ®], [Google Scholar]
- Iyengar, Shanto.. 1991. Is Anyone Responsible? How Television Frames Political Issues. Chicago, IL: University of Chicago Press. [Crossref], [Google Scholar]
- Kinder, Donald R., and Lynn Sanders. 1996. Divided by Color: Racial Politics and Democratic Ideals. Chicago, IL: University of Chicago Press. [Google Scholar]
- Kinder, D. R., and D. O. Sears. 1981. “Prejudice and politics: Symbolic racism versus racial threats to the good life.” Journal of Personality and Social Psychology 40 (3): 414–431. doi: 10.1037/0022-3514.40.3.414 [Crossref], [Web of Science ®], [Google Scholar]
- McConahay, John B.. 1986. “Modern Racism, Ambivalence, and the Modern Racism Scale.” In Prejudice, Discrimination, and Racism, edited by John F. Dovidio, and Samuel L. Gaertner. San Diego, CA: Academic Press.. [Google Scholar]
- Nelson, Thomas, Kira Sanbonmatsu, and Harwood K McClerking. 2007. “Playing a Different Race Card: Examining the Limits of Elite Influence on Perceptions of Racism.” Journal of Politics 69 (2): 416–429. doi: 10.1111/j.1468-2508.2007.00540.x [Crossref], [Web of Science ®], [Google Scholar]
- O'Neil, Bridjes. 2014. “STL County Police Chief Claims Struggle With Gun Lead to Fatal Shooting of Michael Brown.” St. Louis Dispatch, August 10. http://www.stlamerican.com/news/local_news/article_86cdc228-20c9-11e4-984c-001a4bcf887a.html [Google Scholar]
- Pew. 2014. “Sharp Racial Divisions in Reactions to Brown, Garner Decisions.” Pew Research Center, Dec 8. http://www.people-press.org/2014/12/08/sharp-racial-divisions-in-reactions-to-brown-garner-decisions/. [Google Scholar]
- Pew. 2016. “On Views of Race and Inequality, Blacks and Whites Are Worlds Apart.” 2016. Pew Research Center, June 27. http://www.pewsocialtrends.org/2016/06/27/on-views-of-race-and-inequality-blacks-and-whites-are-worlds-apart/. [Crossref], [Google Scholar]
- Porter, Ethan, and Thomas J. Wood. 2016. “Race, Interracial Families, and Political Advertising in the Obama Era: Experimental Evidence.” Political Communication 33 (3): 481–502. doi: 10.1080/10584609.2015.1106625 [Taylor & Francis Online], [Web of Science ®], [Google Scholar]
- Quillian, Lincoln. 1995. “Prejudice as a Response to Perceived Group Threat.” American Sociological Review 60 (4): 586–611. doi: 10.2307/2096296 [Crossref], [Web of Science ®], [Google Scholar]
- Roller, Emma. 2014. "The Character Assassination of Michael Brown." The Atlantic, Aug 19. https://www.theatlantic.com/politics/archive/2014/08/the-character-assassination-of-michael-brown/455124/. [Google Scholar]
- Sniderman, P. M., T. Piazza, P. E. Tetlock, and A. Kendrick. 1991. “The New Racism.” American Journal of Political Science 35 (2): 423–447. doi: 10.2307/2111369 [Crossref], [Web of Science ®], [Google Scholar]
- Steele, Claude. 2010. Whistling Vivaldi: How Stereotypes Affect Us and What We Can Do. New York, NY: W.W Norton and Company. [Google Scholar]
- Stolberg, Sheryl, and Richard Oppel. 2015. “Suspects in Freddie Gray Case.” The New York Times, May 9. [Google Scholar]
- Stole, Bryn. 2016. “Alton Sterling's Son Says Father 'Was Good Man,' Calls For Peaceful Protests.” The Advocate, July 13. https://www.theadvocate.com/baton_rouge/news/alton_sterling/article_d28eafa4-48f1-11e6-93d3-e333aa7efd80.html. [Crossref], [Google Scholar]
- Stone, Jeff, Mike Sjomeling, Christian Lynch, and John Darley. 1999. “Stereotype Threat Effects on Black and White Athletic Performance.” Journal of Personality and Social Psychology 77 (6): 1213–1227. doi: 10.1037/0022-3514.77.6.1213 [Crossref], [Web of Science ®], [Google Scholar]
- Tajfel, Henri. 1969. “Cognitive Aspects of Prejudice.” Journal of Social Issues 25 (4): 79–97. doi: 10.1111/j.1540-4560.1969.tb00620.x [Crossref], [Web of Science ®], [Google Scholar]
- Tajfel, Henri. 1981. Human Groups and Social Categories. New York, NY: Cambridge University Press. [Google Scholar]
- Tesler, Michael. 2012. “The Return of Old-Fashioned Racism to White Americans' Partisan Preferences in the Early Obama Era.” Journal of Politics 75 (1): 110–123. doi: 10.1017/S0022381612000904 [Crossref], [Web of Science ®], [Google Scholar]
- Weitzer, Ronald. 1999. “Citizens' Perceptions of Police Misconduct: Race and Neighborhood Context.” Justice Quarterly 16 (4): 819–846. doi: 10.1080/07418829900094381 [Taylor & Francis Online], [Google Scholar]
Table A1. Regression models which provide the estimates for the differences in Figure 5.
Table A2. Regression estimates for study two testing for the possibility that confidence in police and the symbolic racism scale have an interactive effect on perceived justifiability of a shooting. for simplicity, the intercept is omitted.
Table A3. Racial differences in study 2.
Published online:
08 November 2018Figure A1. Relationship between our dependent variable (perceived justifiability of some police shooting) and three related attitudinal scales. The x-axis categorizes respondents by their modal evaluation of each described shooting. The y-axis shows that group of respondents' deviation from some related scale – specifically, ideology, symbolic racism, and confidence in the police. Each related variable is rescaled to be mean centered and measured by standard deviations.
![]({"type":"image","src":"/na101/home/literatum/publisher/tandf/journals/content/rpgi20/2021/rpgi20.v009.i01/21565503.2018.1528162/20210522/images/medium/rpgi_a_1528162_f0007_ob.jpg"})
Figure A1. Relationship between our dependent variable (perceived justifiability of some police shooting) and three related attitudinal scales. The x-axis categorizes respondents by their modal evaluation of each described shooting. The y-axis shows that group of respondents' deviation from some related scale – specifically, ideology, symbolic racism, and confidence in the police. Each related variable is rescaled to be mean centered and measured by standard deviations.