Abstract
The stereotype that Blacks are violent is pervasive in the United States. Yet little research has examined whether this stereotype is linked to violent behavior from members of different racial groups. We examined how state-level violent crime rates among White and Black Americans predicted the strength of the Black-violence stereotype using a sample of 348,111 individuals from the Project Implicit website. State-level implicit and explicit stereotypes were predicted by crime rates. States where Black people committed higher rates of violent crime showed a stronger Black-violence stereotype, whereas states where White people committed higher rates of violent crime showed a weaker Black-violence stereotype. These patterns were stronger for explicit stereotypes than implicit stereotypes. We discuss the implications of these findings for the development and maintenance of stereotypes.
How do people form the stereotype that Black people are violent? Much research has focused on cultural transmission of stereotypes (e.g., Brauer, Judd, & Micah, 2004; Thompson, Judd, & Park, 2000) and how stereotypes—once in place—resist change (Hunzaker, 2014; Lyons & Kashima, 2001). However, these findings do not address why the stereotype that Black people are violent might vary from place to place. We propose that a portion of this stereotype might originate from environmental cues. If more White (Black) individuals commit violent crimes in an area, the Black-violence stereotype might be weaker (stronger) among people who live in that area. Thus, geographic variation in this stereotype might be partially explained by the rates of violent crime perpetrated by White and Black Americans in those geographic regions. In the current study, we combine multiple data sources on implicit and explicit bias and rates of violent crime to answer this question.
Determinants of the Black-Violence Stereotype
Disentangling the factors that determine the stereotype that Black people are violent is difficult, partially because individuals are often unaware of or unwilling to explicitly report racial stereotypes (McCrae & Bodenhausen, 2000; Nisbett & Wilson, 1977). Researchers have increasingly relied on implicit measures to assess the strength of such stereotypes (for a review, see Fazio & Olson, 2003). However, it is often unclear exactly what these measures assess. Greenwald and Banaji (1995) assert that implicit measures reflect “traces of past experience” (p. 8). However, it is not fully known what past experiences are reflected in these measures. Arkes and Tetlock (2004) propose that—in addition to personal experiences—implicit measures reflect (1) shared cultural stereotypes, (2) emotions that do not reflect prejudice, and (3) rational behavior influenced by base rate differences among groups.
We focus on this last possibility, that the stereotype that Black people are violent might be influenced by exposure to base rate differences in violent crime among White and Black Americans. We assume that information about these crime rates in a general sense is transmitted through media (e.g., television broadcasts, newspapers), witnessing crimes, and word of mouth. This information influences the degree to which Black people are associated with violence on both an explicit and implicit level. In areas where Black people commit higher rates of violent crime, the Black-violence stereotype should be stronger. Conversely, in areas where other racial groups (e.g., White people) commit higher rates of violent crime, the Black-violence stereotype should be weaker.
There is some evidence that stereotypes vary based on geography at the between-country level. Dunham, Baron, and Banaji (2006) had American and Japanese citizens complete a measure of evaluative associations—the implicit association test (IAT; Greenwald, McGhee, & Schwartz, 1998). In this task, American and Japanese citizens categorized valenced words and American and Japanese faces. Based on research on in-group preferences, Dunham and colleagues predicted and found that American citizens were faster to categorize American names paired with positive words and Japanese names paired with negative words, whereas Japanese students showed the opposite pattern (for a similar pattern, see Greenwald et al., 1998). These results provide some evidence that implicit measures such as the IAT “tap into the associations a person has been exposed to in his or her environment” (Karpinski & Hilton, 2001, p. 776). However, the degree to which geographic variation in stereotypes is directly associated with geographic variation in behavior among group members has not been explored.
Although culturally transmitted beliefs no doubt form a part of stereotypes, actual events (e.g., hearing about a violent crime) should influence them as well. Experimental research on stereotype formation consistently shows that stereotypes are formed about novel groups when base rates of some behavior vary between groups (Biernat & Crandall, 1996; Manis, Nelson, & Shedler, 1988; Manis, Paskewitz, & Scotler, 1986; Maris & Hoorens, 2012). For example, Biernat and Crandall (1996) gave information about assertive behaviors that individual students from two fictitious schools had engaged in. Behaviors were manipulated such that students from one school were similar to or more assertive than individuals from the other school. Even when the amount of students who were assertive was only moderately higher in one school, students still formed the stereotype that students from that school were assertive. Outside the laboratory, receiving information about the relative frequency that certain Black (or White) people commit violent crimes within a region may have a similar effect, creating location-based differences in the Black-violence stereotype.
Measuring the Black-Violence Stereotype
To test the hypothesis that geographic differences in crime rates influence the stereotype that Black people are violent, we obtained IAT data on the association between Black people and weapons from the Project Implicit website (https://implicit.harvard.edu). In the weapons IAT, participants see pictures of Black and White people as well as images of harmless objects (e.g., cell phones) and weapons (e.g., guns). Pictures are presented one at a time and participants are instructed to categorize them quickly and accurately by pressing one of the two keys. The task is difficult because participants must categorize different types of stimuli with the same key. Participants who are faster to categorize weapons when paired with Black people and harmless objects when paired with White people (than vice versa) are said to have a greater implicit association between Black people and weapons.
We relied on the weapons IAT as a proxy for the strength of the Black-violence stereotype for several reasons. First, weapons are used in the majority of violent crimes that result in serious injury (57%; Perkins, 2003). Most homicides are also committed with firearms (71%; Planty & Truman, 2013). Thus, we assume that the stereotype that Black people are violent is closely tied to the stereotype that Black people use weapons to commit violence. Second, individuals who take the weapons IAT also report the degree to which they explicitly associate Black people with weapons. This allowed us to test whether crime rates might have similar or different effects on stereotypes at the implicit or explicit level. Third, individuals taking the IAT report their state of residence, making a state-level analysis possible. Finally, because the weapons IAT has been available since 2006, there were ample data to estimate state-level scores of implicit and explicit bias.
The Current Study
In the current study, we examined state-level associations between violent crime and stereotypes that Black people are violent. We hypothesized a congruency effect to be manifested at a broader geographic level (i.e., states). Specifically, in states where Black people commit higher rates of violent crime, we expected that the stereotype that Black people were violent would be stronger. Likewise, in states where White people commit higher rates of violent crime, we expected that the stereotype that Black people were violent would be weaker. To this end, we drew on a large sample of participants who completed implicit and explicit measures of the association between Black people and weapons. We link these attitudinal measures with existing databases of violent crime provided by government agencies and organizations.
Method
Participants
To maximize statistical power, we drew on data from all participants who completed the weapons IAT from the Project Implicit website. Data were available between 2006 and 2015 (N = 863,164). Following protocol used by Westgate, Riskind, and Nosek (2015), we removed participants who made more than 30% errors overall on the weapons IAT or who made 40% or more errors on one block and had a response time of 400 ms or less on more than 10% of trials. We also removed participants who did not have implicit or explicit measures recorded or did not report their state. The final sample was 348,111 individuals (M age = 26.7 years old, standard deviation [SD] = 11.6; 43.1% female).
In terms of race, 67.1% of participants reported their race as White, 6.1% as Black, 4.3% as multiracial (Other), 4.0% as Other, 2.3% as East Asian, 1.4% as South Asian, 1.2% as multiracial (Black and White), 0.8% as Native American, and 0.6% as Pacific Islander; 12.3% did not report their race. Most participants (75.2%) had received at least some college education. Participants were more likely to identify as liberal (35.5%) than conservative (28.8%); 33.6% identified as moderate. Participants reported their state of residence. 1 Responses were averaged to yield one estimate of implicit and explicit race weapons bias for each state.
Measures
The degree to which participants explicitly associated White people and Black people with weapons was assessed using a 7-point scale ranging from 1 (strongly with White people) to 7 (strongly with Black people). The degree to which participants implicitly associated White people and Black people with weapons was assessed using the weapons IAT. IAT scores were computed using the D algorithm recommended by Greenwald, Nosek, and Banaji (2003). Higher scores on this measure correspond to stronger associations between Black people and weapons. Implicit and explicit associations were moderately correlated at the state level (r = .44, p = .001). Although there was variation between states, in every state, the implicit and explicit associations between Black people with weapons were stronger than the associations between White people with weapons.
We obtained several unique proxies for race-specific crime-related weapon use: rates of murder, aggravated assault, and illegal weapon possession for Black and White perpetrators. State race-specific arrest rates (per 100,000) for 2006–2014 for Black and White perpetrators were obtained via the Federal Bureau of Investigation (FBI) Uniform Crime Reports. Because the rates were often highly correlated, these rates were averaged to create a state-level standardized composite of violent crime for Black (α = .74 [.62, .87]) and White (α = .93 [.90, .96]) perpetrators. One issue with these measures of violent crime is that they are self-reported by police departments. If there is any racial bias in arrests or even only the reporting of those arrests within a department, these rates may not provide an accurate proxy for actual crime-related weapon use by race. To address this, we also collected data on victims who died from weapon use (e.g., gun discharge, assault by sharp or blunt objects). Data were obtained from the Center for Disease Control (CDC) from 2006 to 2015 for each state, which included death rates broken down by the race of the victim. Deaths were chosen as a proxy for crime rates because weapon-related deaths occur overwhelmingly within race (Harrell, 2007). From 1980 to 2008, Black offenders killed 93% of Black victims, whereas White offenders killed 84% of White victims (Cooper & Smith, 2011). In addition, as these data were obtained independently from police departments, they are less likely to be influenced by biased reporting. Together, these two sources of arrests and victim reports comprised our measures of violent crime rates.
To account for geographic variation in other variables unrelated to the current study, we also collected information on each state’s male-to-female sex ratio, median age, median income, proportion of White non-Hispanic residents, proportion of adults with a bachelor’s degree, and conservatism. These variables were chosen to rule out potential confounds that might bias the relationship between weapon use and implicit/explicit associations and were selected because they have historically shown to be important predictors of geographic social indicators (e.g., Brethel-Haurwitz & Marsh, 2014; Chopik & Motyl, 2017; Park & Peterson, 2010; Rentfrow, Gosling, & Potter, 2008). We were inclusive in our choice of control variables; all variables were entered in as covariates in each analysis. However, the pattern of relationships between the predictors and implicit and explicit associations remains, even if the covariates are not controlled for. State-level estimates of the variables were taken from the 2010 Census, with the exception of the measure of education and conservatism. The former was only available from the 2000 Census, whereas the latter was calculated by averaging the proportion of the popular vote won by the Republican Party presidential candidate in 2008, 2012, and 2016 (α = .98 [.97, .99]). State-level popular vote totals were obtained from the National Archives and Records Administration.
Results
Mean scores for implicit and explicit associations between Black people and weapons were computed for the residents in each state. Means, SDs, sample sizes, and rankings for each state are presented in Table 1. States with the strongest implicit associations between Black people and weapons were Connecticut, Indiana, and Iowa; states with the weakest were Mississippi, Georgia, and Louisiana. The states with the strongest explicit associations between Black people and weapons were Illinois, New Jersey, and Michigan; states with the weakest were Montana, Wyoming, and Oregon. State-level variation is presented in Figure 1 for implicit associations (top panel) and for explicit associations (bottom panel). We also present the means and SDs for all variables (Table 2) and the correlations between all variables (Table 3). Data and scripts for all analyses in this article can be found at https://osf.io/pnxgf/.
Sample Sizes and Descriptive Statistics.
Note. Lower ranking means stronger associations between Black people and weapons. IAT = implicit association test; M = mean; SD = standard deviation.
Geographic variation in weapon implicit association test scores (A) and explicit associations between Blacks and weapons (B).
Descriptive Statistics for Variables.
Note. Crime rate is only reported for 50 states because Florida does not provide crime rate data to the Uniform Crime Report. Crime rates are the standardized composite of murder rates, assault rates, and weapons violations rates. GOP = Republican Party; B = Black; W = White; IAT = implicit association test; M = mean; SD = standard deviation.
Correlation Table.
Note. Correlation coefficients are presented above the diagonal. The maximum 95% confidence interval for a correlation of N = 50 is ±.278. Sample sizes are reported below the diagonal. Bolded correlations are significant at p < .05. GOP = Republican Party; B = Black; W = White; IAT = implicit association test; M = mean; SD = standard deviation.
Are Weapon Associations Related To Race-Specific Crime Rates?
We hypothesized that for states in which White people committed more crime with weapons, the association between Black people and weapons would be weaker. In contrast, for states in which Black people committed more crime with weapons, the association between Black people and weapons would be stronger. To test these hypotheses, we regressed state-level implicit and explicit weapon associations separately onto the state-level proxies for crime-related weapon use while controlling for each state’s male-to-female sex ratio, median age, median income, proportion of White non-Hispanic residents, proportion of adults with a bachelor’s degree, and proportion of citizens who voted Republican (e.g., White and Black death rates were regressed onto implicit and explicit associations while controlling for our covariates). 2
As seen in the left panel of Table 4, our hypotheses were strongly supported for explicit associations. In states where more Black people were arrested for violent crime or were killed with weapons more often, explicit associations between Black people and weapons were stronger. Mirroring these results, in states where more White people were arrested for these crimes or were killed with weapons, explicit associations between Black people and weapons were weaker. This pattern held regardless of whether the proxies for weapon use came from police departments (i.e., crime rates) or the CDC (i.e., death rates).
Regressions Predicting Explicit and Implicit Weapon Associations From State-Level Variables.
Note. Regression analyses control for each state’s male to female sex ratio, median age, proportion of White, non-Hispanic residents, median income, proportion of residents with a bachelor’s degree, and proportion of citizens who voted Republican. Regressions were conducted separately for each association and pair of rates. SE = standard error; CI = confidence interval; B = Black; W = White.
The results for implicit associations were more mixed (see right panel of Table 4). In states where Black people were killed with weapons more often, implicit associations between Black people and weapons were stronger. In states where White people were killed with weapons more often, implicit associations between Black people and weapons were weaker. These effects were not significant for the index of violent crime, although the point estimates were in the hypothesized direction.
Discussion
The current study linked the implicit and explicit stereotype that Black people are violent with geographic variation in race-based violent crime. In states where Black people were arrested for violent crime at higher rates, the explicit association between Black people and weapons was stronger. The opposite was true for the explicit association between White people and weapons. We replicated these findings using a proxy less likely to suffer from potential police bias—deaths caused by weapons as indexed by the CDC. This pattern held for implicit associations, albeit not for crime rates as indexed by the FBI.
These results may seem surprising given that individuals are often reluctant to explicitly state socially undesirable views (McCrae & Bodenahausen, 2000; Nisbett & Wilson, 1977). One possibility is that making an explicit judgment that Blacks are associated with weapons is less fraught with social consequences than admitting negative evaluations of Black people more generally. Another possibility for these results is that individuals may simply be more willing to explicitly express such beliefs in areas where violent crime rates are higher among Black people. This argument suggests that base rate racial differences in violent crime only make people more willing to share their preexisting views instead of actually changing their views. Future research should focus on disentangling these two possibilities.
The race weapon congruency effect suggests that between-state variation in the stereotype that Blacks are violent can be reliably explained by between-state variation in violent crime. This finding challenges the idea that the Black-violence stereotype is completely divorced from reality and not sensitive to actual rates of Black people committing crime. Rather, “as long as there is a differential crime rate between racial groups, a perfectly rational decision maker may manifest different behaviors—implicit and explicit—toward members of different races” (Arkes & Tetlock, 2004, p. 270). However, it was always the case that Black people were more strongly associated with weapons, even if violent crime rates were higher for White people than Black people. Thus, the stereotype that Black people are violent reflects more than just an awareness of differential crime rates among Whites and Blacks. The Black-violence stereotype likely originates from multiple sources. For example, cultural transmission of the belief that Blacks are violent is another probable component of the stereotype. This would explain the consistent bias toward Black people being associated with violence, even among states where White people commit higher rates of violent crime.
One unexpected result was that we did not find an association between race-based violent crime and the Black-violence stereotype when predicting implicit associations from crime rates. One explanation for this pattern is that implicit associations (vs. explicit associations) about weapons might be less affected by local crime rates. This would argue against the hypothesis raised by Arkes and Tetlock (2004) that base rate differences equally influence both implicit and explicit associations and behaviors toward members from different groups. We find such a conclusion premature because Black/White death rates did predict the implicit Black-violence stereotype.
Another explanation for these weaker results between implicit Black-violence stereotypes and crime rates is that IAT may have too much measurement error to reliably detect relationships in small samples (i.e., in a sample of 50 states). Part of this issue may be that there is less between-state variability in IAT scores (SD = .02) compared to the explicit association measure (SD = .11). Another issue is that the IAT has poor test–retest reliability when compared to explicit measures (.40 < rs < .60; Egloff, Schwerdtfeger, & Schmukle, 2005). Egloff et al. suggest this may be because the IAT is sensitive to both trait- and state-level variance in associations, making it difficult to ascertain the effect of crime rates averaged over time on implicit associations measured at one point in time. Future research could test this experimentally by measuring whether the implicit Black-violence stereotype (measured by the IAT) is more impacted by a manipulation of the accessibility of Black or White violent crime, relative to the explicit Black-violence stereotype (self-reported).
It is worth noting that the between-state variability in implicit and explicit stereotypes was far outweighed by the within-state variability of those stereotypes. While this might seem to cut against the importance of these results, the relationship between crime rates and the Black-violence stereotype might be even stronger if the level of analysis was more local (e.g., at the county or city level). For example, crime rates for the entire state of Illinois are much lower than crime rates in Cook County, which contains the city of Chicago. Within-state variability might be better captured if these more local crime rates are taken into account. Indeed, Hehman, Flake, and Calanchini (in press) recently took such an approach. They linked racial stereotypes with lethal force used by police officers against Black people within core-based statistical areas (CBSAs), a more focal level of analysis. They found that in CBSAs with stronger Black-weapons stereotypes, there were a higher number of instances of lethal force used against Black people by police (provided by the Guardian, an independent news agency). The fact that the Black-violence stereotype and its related correlates are manifested at different levels of analysis (CBSAs and states) and different data sources (the FBI, CDC, and news agencies) further strengthens our confidence in our study’s findings.
Strengths and Limitations
The current study used a data set with a large number of respondents, examined both implicit and explicit associations between Black people and weapons, and drew on multiple data sources from disparate government agencies to yield a comprehensive assessment of violent crime. We found much consistency across these indicators; state-level Black-violence stereotypes were associated with state-level violent crimes perpetrated by Black and White individuals.
Nevertheless, the study had some limitations worth acknowledging. Although we used the most reputable sources available for our proxies of violent crime, they remain imperfect measures of violent crime rates. We tried to alleviate this concern by drawing on two sources—one source from the FBI that relies on police department self-reports and the other from the CDC. In the future, greater logging of violent crime rates should be undertaken by organizations, so that more accurate violent crime rates can be used. Ideally, such organizations would be independent of law enforcement agencies to prevent concerns over biased reporting by police.
As previously stated, a more local unit of analysis might be useful for examining associations between violent crime and racial stereotypes. However, there are major limitations to examining low frequency events at such narrow levels of analysis. For example, when Hehman and colleagues (in press) studied police lethal force at the CBSA level, the majority of CBSAs were excluded from the analysis because there were no incidents of lethal force. Further, some CBSAs included only one lethal force incident. The problem is that in many CBSAs where there is a high Black-violence stereotype there may be no lethal force incidents. Excluding these CBSAs may lead to a skewed interpretation of whether stereotypes are present. The same critique holds true for violent crime, which is also a rare event. This is why we chose a broader regional level (states) as our unit of analysis.
We view this issue of unit of analysis as a broader challenge in the field of geographical psychology, which often relies on aggregated information at broader geographic levels (e.g., Bach, Defever, Chopik, & Konrath, 2017; Chopik, O’Brien, & Konrath, 2017; Rentfrow, 2014; Rentfrow et al., 2013). The absence of local data sources and low-frequency events not only limits researchers’ ability to examine associations at more local levels of analysis but also prevents the implementation of other research designs. For example, we argue that higher rates of violent crime perpetrated by White and Black people should influence the Black-violence stereotype. Although our data show a clear relationship between the two, we cannot demonstrate causality because our data come from the same 10-year period. One solution to this issue would be to establish temporal precedence by tracking how changes in rates of violent crime influence the Black-violence stereotype over time. However, this approach runs into the same issue. Because violent crime is such a rare event, it is difficult to obtain precise state-level estimates when using smaller time periods.
One solution is to collect data at both the individual and broader regional level, which can also help prevent researchers from committing the ecological fallacy (i.e., incorrectly inferring individual-level associations from region-level associations; Robinson, 1950). Integrating geographic studies and individual-level experiments is also a way to further establish a causal link between an individual’s environment and their psychological characteristics (Oishi, Talhelm, & Lee, 2015). Given past research showing that individuals easily learn stereotypes when given information about members of novel groups (Biernat & Crandall, 1996; Manis et al., 1986, 1988; Maris & Hoorens, 2012), we have some confidence that our prediction regarding the direction of causality (i.e., exposure to crime rates facilitate stereotype formation) is correct. Information (from the media, experience, or word of mouth) about the race of a perpetrator of violent crime should have a similar impact on one’s stereotype that Black people are violent. The alternative explanation that the Black-violence stereotype in a geographic area increases the rates of Black individuals committing violent crime is less likely and straightforward considering the experimental evidence on stereotype formation. Thus, experimental data at the individual level and correlational data at the regional level all suggest that information such as crime rates can contribute to the development and maintenance of stereotypes about a group.
Conclusion
The question of whether stereotypes accurately reflect differences between racial groups is a highly debated area of research in psychology (Bian & Cimpian, 2017; Jussim, 2012). We found that geographic variability in the strength of the Black-violence stereotype can partially be explained by violent crime rates perpetrated by White and Black Americans in these different regions. These findings provide evidence that this stereotype and similar stereotypes at least partially reflect real-world base rate differences between groups. However, Black people were always more strongly associated with weapons than White people even when Whites committed more violent crime, suggesting a bias that base rate differences alone cannot explain. Thus, the stereotype that Blacks are violent likely reflects a complex association influenced by group differences in base rates of violence as well as shared cultural beliefs.
Supplemental Material
Supplemental Material, SPPS753522_suppl_mat - Geographic Variation in the Black-Violence Stereotype
Supplemental Material, SPPS753522_suppl_mat for Geographic Variation in the Black-Violence Stereotype by David J. Johnson and William J. Chopik in Social Psychological and Personality Science
Footnotes
Authors’ Note
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental Material
The supplemental material is available in the online version of the article.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
