Is the evidence from racial bias shooting task studies a smoking gun? Results from a meta-analysis

Journal of Experimental Social Psychology 61 (2015) 120–130 Contents lists available at ScienceDirect Journal of Experimental Social Psychology journal homepage: www.elsevier.com/locate/jesp Is the evidence from racial bias shooting task studies a smoking gun? Results from a meta-analysis☆ Yara Mekawi ⁎, Konrad Bresin University of Illinois at Urbana-Champaign, United States H I G H L I G H T S • Laboratory shooter tasks have yielded mixed results regarding racial shooter biases. • This study reports a meta-analysis of racial shooter biases. • Shooter biases were significant for shooting threshold and reaction time. • State level gun laws and proportion of non-Whites moderated shooter biases. • Implications for training of police officers and gun owners are discussed. a r t i c l e i n f o a b s t r a c t Article history: The longstanding issue of extrajudicial police shootings of racial and ethnic minority members has received Received 4 September 2014 unprecedented interest from the general public in the past year. To better understand this issue, researchers Revised 20 August 2015 have examined racial shooter biases in the laboratory for more than a decade; however, shooter biases have Accepted 21 August 2015 been operationalized in multiple ways in previous studies with mixed results within and across measures. We Available online 23 August 2015 meta-analyzed 42 studies, investigating five operationalizations of shooter biases (reaction time with/without Keywords: a gun, false alarms, shooting sensitivity, and shooting threshold) and relevant moderators (e.g., racial prejudice, Shooting task state level gun laws). Our results indicated that relative to White targets, participants were quicker to shoot Shooting threshold armed Black targets (dav = −.13, 95% CI [−.19, −.06]), slower to not shoot unarmed Black targets (dav = .11, Racial shooter biases 95% CI [.05, .18), and more likely to have a liberal shooting threshold for Black targets (dav = −.19, 95% CI Meta-analysis [−.37, −.01]). In addition, we found that in states with permissive (vs. restrictive) gun laws, the false alarm rate for shooting Black targets was higher and the shooting threshold for shooting Black targets was lower than for White targets. These results help provide critical insight into the psychology of race-based shooter decisions, which may have practical implications for intervention (e.g., training police officers) and prevention of the loss of life of racial and ethnic minorities. © 2015 Published by Elsevier Inc. Although police brutality toward racial and ethnic minorities George Zimmerman for the murder of Trayvon Martin (Black Lives has been a pervasive problem in the United States for decades Matter, 2015; Mays, Johnson, Coles, Gellene, & Cochran, 2013). In (Binder & Scharf, 1982; Department of Justice, 2001; National Center particular, the past year involved unprecedented media coverage of for Injury Prevention and Control, 2014), these issues have been extrajudicial shootings of racial and ethnic minorities (most often a particularly salient topic of public debate since the acquittal of Black men) by police (e.g., Michael Brown, Tamir Rice, Walter Scott). In response to similar cases that have occurred in the past decade, researchers have used a novel laboratory task to examine this phenomenon in the hopes that such research will eventually lead ☆ YM conceptualized the study. YM and KB searched for studies, coded studies, conducted data analysis, and wrote the manuscript. Both authors made revisions to the to interventions that will prevent the unjustified loss of life of manuscript and approved the final version of the manuscript for submission. The people of color. In a typical first-person shooter task (Correll, authors would like to thank Dr. Joshua Correll, Dr. E. Ashby Plant, Dr. Jack Glaser and Dr. Park, Judd, & Wittenbrink, 2002), participants are shown images Sang Hee Park for their cooperation and immediate willingness to retrieve datasets for of Black and White targets holding either a gun or a neutral object inclusion in these analyses. (e.g., cell phone) and given less than one second to respond whether ⁎ Corresponding author at: Psychology Department, University of Illinois Urbana- Champaign, 603 East Daniel Street, Champaign, IL 61820, United States. or not they should “shoot” or “not shoot” the target. Over a dozen E-mail address: yaramekawi@gmail.com (Y. Mekawi). studies have investigated shooter biases using targets from various http://dx.doi.org/10.1016/j.jesp.2015.08.002 0022-1031/© 2015 Published by Elsevier Inc.  Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 121 racial groups, using different stimuli, and identifying various individual One moderator that has been identified in shooting task studies is differences in shooting decisions. This shooter bias has been operation- quantity of interaction with racial out-groups. Some studies found a alized in a number of ways (e.g., reaction time, shooting threshold), negative association between intergroup contact and relative reac- with mixed results both within and across studies and operation- tion time (Correll et al., 2002; Correll et al., 2007), suggesting that alizations. Given the importance of the topic and the mixed results in more contact is associated with a quicker decision to shoot Black tar- the literature, we conducted a meta-analysis of shooting task studies gets. This is consistent with criminology research, which finds a pos- to synthesize the research about how target race affects shooting itive association between proportion of Black individuals in a decisions. neighborhood and Whites' fear of crime (Chiricos, Hogan, & Gertz, A cursory reading of the literature would suggest that the evidence 1997). Others have found a positive association between intergroup for a racial shooter bias is quite strong, as all published papers on the contact and shooting threshold for White targets, suggesting that the topic have significant results suggesting a bias against Black (versus more contact with out-group members, the more conservative they White) targets. Upon further analysis, however, it becomes clear were in their shooting threshold for their own group (Kenworthy that researchers do not operationalize shooter biases in the same way et al., 2011). However, Kenworthy et al. (2011) did not find a corre- across studies and null and significant results exist for each definition. lation between shooting threshold for Black targets and contact with For example, some studies have found that participants are quicker Blacks. Thus, it is unclear whether the effect is unreliable, or whether to shoot armed Black (versus White) targets and slower to not it differs based on the shooter bias outcome analyzed. To clarify the shoot unarmed Black (versus White) targets (Correll et al., 2002; role of inter-group contact in moderating shooter biases, we used Correll, Park, Judd, & Wittenbrink, 2007; Park, Glaser, & Knowles, percentage of non-Whites in the community where the data was col- 2008; Sadler, Correll, Park, & Judd, 2012), while others have not found lected as a proxy for inter-group contact. reaction time differences (Harmer, 2012; Taylor, 2011). More relevant Second, prior research has found that permissive (vs. restrictive) to shootings of unarmed individuals, some studies have focused on gun laws (e.g., less government regulation) are associated with trials where participants shoot when no gun is present (i.e., false more extrajudicial shootings, with Black victims disproportionately alarms). Some of these studies have shown that participants are more affected (Miller, Hemenway, & Azrael, 2007; Price, Thompson, & likely to shoot unarmed Black (versus White) targets (e.g., Plant Dake, 2004). The culture of gun ownership may affect how careful & Peruche, 2005) while others have not (e.g., Sadler et al., 2012). people are in making shooting decisions in a way that affects However, one limitation to this type of operationalization is that racial and ethnic minorities more than Whites. People living in it ignores correct trials. Participants who make a large amount of states with more permissive gun laws are more likely to own and errors may have a tendency to respond to shoot (i.e., bias to use guns and therefore may be less inhibited in their shooting respond). decisions. Thus, we examined whether the strictness of gun laws in Hence, to understand racial biases in shooting decisions, some the states where participants were recruited moderated racial shoot- researchers have used signal detection theory to estimate two parame- er biases. ters representing decision-making processes. The first parameter, called Finally, endorsement of prejudicial attitudes was frequently d′ or sensitivity, represents the strength of the signal relative to noise, assessed in shooting task studies. It is possible that to the extent that with larger values indicating a stronger signal (i.e., more sensitive). an individual is prejudiced toward Blacks, he or she is more likely to Research has not consistently found differences based on target race have a shooter bias toward Black (versus White) targets. Despite how when looking at shooting sensitivity (e.g., Correll et al., 2002). Another intuitive it may seem that prejudicial attitudes play a role in shooter parameter, c, denotes the threshold for responding. A value of zero biases, results have been largely inconsistent (e.g., Correll et al., 2002, signifies an observer who evenly balances responding (i.e., shoot, 2007). Given that studies varied in other ways, we also examined a don't shoot). Participants who have a negative value have a bias toward number of methodological differences between studies (e.g., response shooting (i.e., liberal threshold), and participants with a positive time window, sample type). value have a bias toward not shooting (i.e., conservative threshold). This meta-analysis was conducted to answer the following Some previous studies have found that participants have a more questions: liberal shooting threshold (i.e., bias to shoot) for Black (versus White) (1) What are the overall effects of shooter biases for all targets (e.g., Correll et al., 2002), whereas others have not (e.g., Taylor, operationalizations of shooter biases (reaction time with/without gun, 2011). error rate, shooting sensitivity, shooting threshold)? Given the many operationalizations of shooter biases and inconsis- (2) Does racial heterogeneity of community moderate shooter tencies in which operationalizations are reported across studies, the biases? results from this paradigm are difficult to interpret. Of particular (3) Does the strictness of gun laws moderate shooter biases? concern is the possibility that researcher degrees of freedom in how to (4) How strong is the association between prejudicial attitudes operationalize shooter bias increases the likelihood of false positive and shooter biases? research findings (Simmons, Nelson, & Simonsohn, 2011). For instance, if a researcher analyzes all operationalizations of shooter bias, there 1. Method is a higher chance of a Type I error than if they only analyzed one. This, combined with selective reporting of operationalizations 1.1. Study inclusion (i.e., those that have significant results), could make the results from this paradigm appear more robust than they really are. It should Studies were identified by searching the words “shooting bias” be noted that it is possible that some operationalizations may have and “shooter bias” using PsychINFO and Google Scholar. We also true effect, whereas others might not. Therefore, we sought to deter- examined papers that had cited Correll et al.'s (2002) original study mine the effect size for each operationalization of shooter bias across examining shooter biases. Initially, papers were included if they were studies. deemed relevant based on a review of the title and brief review of Inconsistent results across studies might also be explained by the abstract. Papers were excluded at this stage if they clearly did variables that differ between the studies. To better explain variance in not involve a racial shooter bias task. This search identified 35 shooter biases across studies, we were also interested in investigating papers. the role of three relevant variables that may moderate racial shooter The 35 papers were reviewed closely for inclusion in the biases: racial heterogeneity of community, strictness of gun laws, and meta-analysis. To be considered for the meta-analysis, papers had prejudicial attitudes. to include a) some variant of the Correll et al. (2002) shooting task, 122 Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 b) a comparison of White male targets to Black male targets1, and (i.e., slower to identify a Black target without a gun). Similarly, a positive c) sufficient information to calculate effect sizes for at least one effect size for false alarms would indicate a bias against Black relative to operationalization of shooter bias was included in the paper. We White targets (i.e., more false alarms for Black relative to White targets). contacted authors when relevant data were missing, and 90% There is no clear prediction for the direction of the shooting sensitivity responded with the requested data. We also unsuccessfully attempted effect. Finally, a negative effect size for shooting threshold would to elicit unpublished data. However, we did include unpublished indicate a lower threshold for shooting Black (versus White) targets. dissertations as an attempt to detect and counteract the possibility To summarize, positive effect sizes would be expected for reaction of publication bias (Ferguson & Brannick, 2012). The final analysis time to no gun trials and false alarms, whereas negative effect sizes included 16 papers with a total of 42 different studies or samples would be expected for reaction time to gun trials and shooting (N = 3427). threshold. We coded multiple demographic variables to characterize the studies' samples (i.e., age, % of women, % of White participants, % of 1.2. Coding Asian participants, % of Black participants, % of Latino/a participants). Table 1 displays the racial demographics of participants across the M's and SD's for reaction time with/without a gun, false alarms, studies. We also coded sample type into three groups: undergraduates, shooting sensitivity, and shooting threshold were coded and used to community members, and police officers/cadets. In terms of methodol- calculate Cohen's dav (Lakens, 2013). We calculated the variance of dav ogy, we coded the response time window for reaction times, as it has using the equation provided by Dunlap, Cortina, Vaslow, and Burke been suggested that shorter times lead to smaller reaction time effects (1996), which takes into account the correlation between dependent and larger false alarm effects (Correll et al., 2002; Plant, Peruche, & measures. Given that no study reported the necessary correlations, we Butz, 2005). We also coded whether the study used Correll et al.'s used the correlations based on our data (Mekawi, Bresin, & Hunter, in (2002) materials, which depict full body images with an environmental press; r = .7 for RT, r = .8 for false alarms, r = .7 for d′, and r = .2 background, or stimuli that depicted faces with objects (e.g., gun or soda for c). A small number of studies used between-subject manipulations can) superimposed (e.g., Plant & Peruche, 2005). to examine their effects on the shooter bias (e.g., reading an article We also coded the correlation between shooter bias and individual about a Black or White person committing a crime; Correll et al., differences measures that were reported in five or more studies (in 2007). Unfortunately, there were too few studies to combine them order to have enough studies to pool effect sizes), namely awareness into meaningful categories for calculating effect sizes to compare of discrimination against Blacks (Wittenbrink, Judd, & Park, 1997), against the normative effect. This, combined with our interested in the personal endorsement of stereotypes, awareness of cultural stereotypes shooter bias, in the absence of other factors, led us to just code the (e.g., Correll et al., 2002), implicit and explicit motivation to control control condition for these studies. prejudice (Dunton & Fazio, 1997), and contact with Blacks (e.g., Correll All effect sizes were calculated by subtracting White target trials et al., 2002). In all but one study (Kenworthy et al. 2011, who used a from Black target trials. The implication of this is that different difference score of c) the shooter bias was parameterized as a difference operationalizations of shooter bias should have different signs score of reaction time, with larger values indicating a larger shooter (i.e., positive or negative effect). For reaction time to gun trials, a racial bias. bias would be indicated by a negative effect size, indicating faster To code the strictness of gun laws, we used the 2013 Brady decision making for Black targets with guns relative to White targets Campaign State Scorecard (Brady Campaign to Prevent Gun Violence, with guns. For reaction time to no gun trials, a positive effect size 2013). In this report, each state is given a score based on multiple would be indicative of a bias against Black relative to White targets categories of gun laws (e.g., background checks, firearms in public places, gun owner accountability) with scores ranging from 0 (the most permissive gun laws) to 100 (the most restrictive). For example, in the sub-category of “dealer regulation”, a state that requires firearms 1 For completeness, we also examined out-groups other than Blacks that have been in- dealers to be licensed would likely receive a score of “6” whereas a state cluded in studies using the shooing task. The results for White women compared with that does not require licensure would likely receive a score of “0”. As White men as targets (k = 4) suggested a bias against shooting women. Compared with another example, in the sub-category of background checks, states targets who are men, participants were slower to shoot armed women (dav = .25, 95% that establish categories of persons deemed ineligible to purchase or CI [.17, .32]), slower to not shoot unarmed women (dav = .99, 95% CI [.50, 1.48]), made less errors for unarmed targets (dav = −.26, 95% CI [−.40, −.12]), and had a higher shooting possess firearms (e.g., history of serious mental illness) would receive threshold (dav = .40, 95% CI [.24, .67]). However, the effect for shooting sensitivity was not a score of “5,” whereas states without these categories would likely significantly different from zero (dav = .03, 95% CI [−.14, .21]). receive a score of “0”. The range of our sample (17–75.5) was similar For Asian relative to White targets (k = 4), the effect was not significantly different from to that of the United States as a whole (6–89), though somewhat zero for reaction time on gun trials (dav = −.55, 95% CI [−1.20, .10]), reaction time on no more constricted. Only two studies did not specifically indicate the gun trials (dav = −.07, 95% CI [−.12, .27]), or shooting threshold (dav = .25, 95% CI [−.15, .65]). There was a significant effect for false alarms (dav = .35, 95% CI [.05, .65]) and sensi- city where the data was collected, and thus the first authors of those tivity (dav = −.13, 95% CI [−.20, −.06]), indicating that relative to White targets, partici- papers were contacted to confirm the data collection site. Two studies pants are more likely to make errors and have less sensitivity for Asian targets. used samples of police officers from multiple states; thus these two For Latino targets (k = 2), the two studies showed significant effects for all three studies were not used in analyses using state level variables. operationalizations of shooter bias reported. Participants were faster to shoot armed Lati- no targets (dav = −.66, 95% CI [−1.20, −.11]) and unarmed Latino targets (dav = −.47, 95% CI [−.82, −.12]) relative to White targets. Moreover, in comparison with White tar- gets, participants had more sensitivity for Latino targets (dav = .35, 95% CI [.24, .46]). Only two studies examined Muslim/Middle Eastern targets with the same Table 1 operationalization of shooter bias (i.e., shooting threshold). The results showed a lower Study characteristics for studies included in the meta-analysis. shooting threshold for Muslim or Middle Eastern targets relative to White targets (dav = −.66, 95% CI [−1.20, −.11]). M SD Range Overall comparison of groups is difficult due to missing data for all operationalizations of % Women (k = 39) 48.89 22.89 39–100 shooter biases. While no clear pattern emerges for all out-groups, participants seem to % White (k = 39) 67.64 22.73 7.25–100 have a relatively lower threshold for groups stereotyped as being dangerous (men, Mus- % Black (k = 34) 9.74 18.08 0–100 lim or Middle Easterners, but not Asians) though other operationalizations of shooter % Asian (k = 36) 7.71 11.07 0–100 biases do not follow this pattern (e.g., participants were less sensitive to shooting Asians % Latino/a (k = 20) 10.03 12.28 0–49.6 but more sensitive to shooting Latinos). Further research is needed to clarify how general- izable various shooter biases are to other out-groups. Note. k = number of studies reporting relevant information.  Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 123 The percentage of non-Whites was coded from the US and/or 2.1.3. False alarms Canada census data for the most recent census.2 This allowed us to The effect size for false alarm differences was not significant (k = 28. use the data from all studies, regardless of whether researchers assessed dav = −.01, 95% CI [−.11, .09]). This suggests that across all studies, the participants' frequency of intergroup contact. This was calculated based rate of false alarms for shooting unarmed targets was not different for on the city where the data collection occurred. The average percentage Black and White targets. There was significant heterogeneity in this of non-Whites (M = 24.43%, SD = 12.46) was slightly lower than effect, Q(27) = 553.18, p b .001. The forest plot for this effect is shown the national average (36.3%). Table 2 displays the values of the main in Fig. 2. moderators for each study. 2.1.4. Shooting sensitivity (d′) 1.3. Data analysis The effect for shooting sensitivity was not significant (k = 30, dav = .07, 95% CI [−.01, .15]). This suggests that across all studies, participants' Data were analyzed using the Metafor package in R (R Development shooting sensitivity was not different for Black and White targets. Again, Core Team, 2010; Viechtbauer, 2010). We used a mixed effects model there was significant heterogeneity in this effect, Q(29) = 162.74, with restricted maximum likelihood estimation for parameters. Study p b .001. The top panel of Fig. 3 shows the forest plot of this effect. effects were weighted by sample size. For individual effect sizes and parameters, we report 95% confidence intervals. To test for heterogeneity 2.1.5. Shooting threshold (c) in effect sizes, we calculated the Q statistic, which tests the null hypoth- The effect for shooting threshold was significant and in the expected esis of no heterogeneity. First, we tested the average effect for each negative direction (k = 29, dav = −.19, 95% CI [−.37, −.01]), suggesting operationalization of shooter bias. Second, for effects with significant that across all studies, participants had a lower shooting threshold heterogeneity, we conducted our primary moderator analyses by (i.e., bias to shoot) for Black targets relative to White targets. As with including gun law score and percentage of non-Whites in separate the other effects, there was significant heterogeneity Q(28) = 404.30, analyses. Third, we explored whether study characteristics or methodo- p b .001. The bottom in Fig. 3 shows the forest plot for this effect. logical considerations moderated effect size. Fourth, we looked at the These data lead us to conclude that there are no overall effects of results for studies correlating the shooter bias to self-report measures. shooting sensitivity or false alarm rate based on target race, but that Finally, we checked for publication bias using the tandem procedure there are small negative effects for reaction time for reaction time to proposed by Ferguson and Brannick (2012). This procedure consists of armed targets and shooting threshold and a small positive effect for calculating the fail-safe-N (i.e., the number of missing studies to make reaction time to unarmed targets. Relative to White targets, participants the effect not significant), the rank order correlation of funnel plot were quicker to shoot armed Black targets, slower to not shoot unarmed asymmetry, and the trim-and-fill procedure. Black targets, and more likely to have a liberal shooting threshold for Black targets. 2. Results 2.2. Main moderators 2.1. Overall effects 2.2.1. Gun laws We found that the strictness of gun laws significantly moderated the 2.1.1. Reaction time for gun trials effect size for two operationalizations: false alarms and shooting thresh- The effect size for reaction time for gun trials was significantly old. For false alarms, the relation was negative (b = −.007, CI [−.010, different from zero and in the expected negative direction (k = 32, −.004], k = 28), and explained 45% of the variance across studies. To dav = −.13, 95% CI [−.19, −.06]). This suggests that across all studies, interpret this effect, we calculated estimated effect sizes for − 1 SD participants were faster to shoot armed Black targets relative to armed and + 1 SD from the mean. In states with more permissive gun laws White targets. There was significant heterogeneity in this effect, (−1 SD) there was a small positive effect (dav = .20, 95% CI [.08, .32]; Q(31) = 159.72, p b .001. The top panel in Fig. 1 displays the forest more false alarms for Black targets), whereas in states with relatively plot for this effect. restrictive gun laws, the effect was negative (dav = −.17, 95% CI [−.06, −.57]; less false alarms for Black targets), suggesting greater 2.1.2. Reaction time for no gun trials bias in states with more permissive gun laws. For shooting threshold, The effect size for reaction time for no gun trials was also significant the effect was positive (b = .01, CI [.005, .017], k = 28), indicating and in the expected positive direction (k = 32, dav = .11, 95% CI [.05, that in states with restrictive (versus permissive) gun laws, the effect .18]). This suggests that across all studies, participants were slower to size was small and non-significant (dav = .07, 95% CI [−.13, .28]) not shoot unarmed Black targets relative to unarmed White targets. compared with states with more permissive gun laws (dav = −.54, Again, there was significant heterogeneity in this effect, Q(31) = 95% CI [−.77, −.30]), where the effect was medium to large and 150.57, p b .001. The bottom panel of Fig. 1 displays the forest plot for negative (i.e., lower threshold for Black targets). This explained 32% of this effect. the heterogeneity in effect sizes. Gun law strictness did not moderate the effects of shooting sensitivity (k = 30. p = .169), reaction time for 2 Based on a reviewer suggestion, we compared studies conducted in Canada to those gun trials (k = 32, p = .108), or reaction time for no gun trials (k = conducted in the United States. In these analyses, we found we found evidence of moder- ation for all operationalizations aside from shooting sensitivity (p = .906). For both reac- 32, p = .354). These data suggest that in states with more permissive tion time types the effects from studies conducted in Canada were small and not (versus restrictive) gun laws, the false alarm rate for shooting Black significant (Gun trials: dav =.04 [-.06, .15], k = 9; No gun trials: dav = −.03 [−.15, .08], (versus White) targets is higher, and the shooting threshold for k = 9); however, the effect for studies conducted in the United States were significant shooting Black (versus White) targets is lower. Consistent with our and in the expected direction (Gun trials: dav = −.19 [−.26, −.13], k = 23; no gun trials: predictions, stricter gun laws were associated with less shooter biases dav = .17 [.10, .24] , k = 23). For false alarms, there was no effect for the studies in conduct- ed in United States (dav = 08 [−.03, .19], k = 20) and there was a significant positive effect toward Black targets relative to White targets. studies conducted in Canada (dav = −.20 [−.33, −.03], k = 8), which is evidence of a bias To rule out the hypothesis that this effect could be explained by to make more errors for White versus Black targets. Finally, for shooting threshold there state-level political orientation, we adjusted for political orientation of was a complete reversal. For studies conducted in the United States there was a significant the state. We operationalized state-level political orientation by using negative effect (dav = −.42 [−.60, −.23], k = 19) and for studies conducted in Canada, there was a significant positive effect (dav = .25 [−.01, .50], k = 10), suggesting a bias to- the percentage of the popular vote in the 2012 presidential election ward lower thresholds for Black versus White targets in the United States and a bias for for the liberal candidate (i.e., Barack Obama) for studies in the United higher thresholds for Black versus White targets in Canada. States and popular vote by province for the 2006 Prime Minister 124 Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 Table 2 Study characteristics for key moderator variables. Authors (year) N Age Gun laws % non-White RW Stim type Sample type Akinola and Mendes (2011) 78 41.2 74.5 33.4 850 Correll Police Officers Correll et al. (2002) S1 40 28.5 12 850 Correll Undergraduates Correll et al. (2002) S2 44 28.5 12 630 Correll Undergraduates Correll et al. (2002) S3 48 28.5 12 850 Correll Undergraduates Correll et al. (2002) S4 BP 25 28.5 31.1 850 Correll Community Correll et al. (2002) S4 WP 21 28.5 31.1 850 Correll Community Correll et al. (2007) S1 NO⁎ 113 38.4 850 Correll Police Officers Correll et al. (2007) S1 DO 124 37.9 28.5 31.1 850 Correll Police Officers Correll et al. (2007) S1 Com 127 35.5 28.5 31.1 850 Correll Community Correll et al. (2007) S2 DO 31 35.6 28.5 31.1 630 Correll Police Officers Correll et al. (2007) S2 Com 45 36.8 28.5 31.1 630 Correll Community Correll, Wittenbrink, Park, Judd, and Goyle (2011) 55 18.87 59 55 630 Correll Undergraduates Harmer (2012) 89 20.16 75.5 16.2 630 Correll Undergraduates Hunsinger (2011) S1 214 20.21 74.5 21.1 700 Correll Undergraduates Husinger (2010) S2 180 20.03 74.5 21.1 700 Correll Undergraduates Hunsinger (2011) S3 72 20.31 74.5 21.1 700 Correll Undergraduates Mekawi et al. (in press) 290 19.21 32.2 59 1000 Plant Undergraduates Miller, Zielaskowski, and Plant (2012) S2 50 19.5 17 42.6 630 Plant Undergraduates Musolino (2012) 123 23 75.5 16 630 Correll Undergraduates Park et al. (2008) 58 81.27 40.5 None Correll Undergraduates Park and Glaser (2011) CG 27 81.27 40.5 None Correll Undergraduates Plant and Peruche (2005) 50 17 39.9 630 Plant Police Officers Plant et al. (2011) Study 2 122 17 42.6 630 Plant Undergraduates Plant et al. (2005) S1 ET 125 19 17 42.6 630 Plant Undergraduates Plant et al. (2005) S2 CG 60 19 17 42.6 630 Plant Undergraduates Plant et al. (2005) S3 ET 61 18.64 17 42.6 630 Plant Undergraduates Sadler et al. (2012) S1 69 28.5 12 850 Correll Undergraduates Sadler et al. (2012) S2⁎ 224 850 Correll Police Officers Taylor (2011) S1 PR 47 27.72 75.5 16 630 Correll Police Officers Taylor (2011) S1 PO 49 38.63 75.5 16 630 Correll Police Officers Taylor (2011) S1 UG 50 20.84 75.5 16 630 Correll Undergraduates Taylor (2011) S1b 74 21.82 75.5 16 630 Correll Undergraduates Taylor (2011) S2 PR 50 26.94 75.5 16 630 Correll Police Officers Taylor (2011) S2 PO 30 39.90 75.5 16 630 Correll Police Officers Taylor (2011) S2 UG 50 20.29 75.5 16 630 Correll Undergraduates Note. RS = response window, Stim type = stimulus type, S = study, BP = Black participants, WP = White participants, Com = community, DO = Denver officers, NO = national officers, CP = control participants, ET = early trials, PO = police officers, PR = police recruits, UG = undergraduates, Correll = full body images on backgrounds based on Correll et al. (2002), Plant = images of faces with objects superimposed based on Plant and Peruche (2005), Gun laws = Brady Scorecard score, where larger (relative to smaller) numbers indicate more restrictive gun laws; % non-White = % of non-White in the city where the study was conducted ⁎ because officers were from multiple states, gun laws and % non-White could not be calculated. elections in Canada. When this was included in the model, the effects for significant effect with lower thresholds for Black versus White targets false alarms (b = −.008 [−.01, −.006], k = 19) and shooting threshold (dav = −.38, 95% CI [−.62, −.14]), suggesting that more racially diverse (b = .012 [.007, .018], k = 27) were still significant and largely cities had more shooter bias. The proportion of non-Whites only unchanged in magnitude. Surprisingly, in both cases, percentage of the marginally moderated the false alarm rate (b = .006, CI [−.001, .013], popular vote for liberal candidates had the opposite relation with effect k = 28), which explained only 6% of the variance. This effect was sizes compared with the gun law effects. For false alarms, higher positive, indicating that proportion of non-Whites was related to a percentage of liberal voting was related to more errors for Black versus larger bias (e.g., −1 SD: dav = −.09, 95% CI [−.24, .04]; +1 SD: dav = White targets, b = .007 [−.000, .015], k = 19. For shooting threshold, .07, 95% CI [−.06, .21]). The proportion of non-Whites did not moderate higher percentage of liberal voting was related to lower shooting the effects of shooting sensitivity (k = 29, p = .509), reaction time for thresholds for Black versus White targets, b = −.023 [−.042, −.004], gun trials (k = 31, p = .481), or reaction time for no gun trials (k = k = 27. Although these results are counterintuitive, there is likely an 31, p = .367). The results suggested that data collected in a city with a issue of restriction of range, given that the liberal candidate won the greater proportion of non-Whites had larger shooter biases in terms of electoral votes for the United States elections (for studies conducted shooting threshold (lower shooting threshold for Black versus White in the United States: M = 52%, min = 50%, max = 61%; for studies targets) and false alarm rate (more errors for Black versus White collected in Canada: M = min = max = 39%). It is possible that with targets). more variability in liberal voting a more intuitive association would be found. More importantly, these results suggest that state-level political 2.3. Methodological moderators orientation does not account for the association between gun laws and effect sizes of shooter biases. 2.3.1. Response time window Given the large number of studies using 630 milliseconds as the 2.2.2. Proportion of non-Whites response window (k = 22), we contrasted this with all other windows We found that the proportion of non-Whites in the city where the (ranging from 700 milliseconds to no time limit). We found that sample was recruited moderated the effect for shooting threshold response window moderated the effect size for reaction time for both (b = −.015, CI [−.028, −.002], k = 28), which explained 14% of the gun (b = .10, CI [.05, .16]) and no gun trials (b = −.12, CI [−.17, heterogeneity. The negative slope indicates that in cities with a greater −.06]). For gun trials, the effect size was significant when the window proportion of non-Whites, there was a larger shooter bias against Black was larger than 630 milliseconds (k = 15, dav = −.24, 95% CI [−.32, versus White targets. For example, cities one SD below the mean had no −.16]), but not when the window was 630 milliseconds (k = 16, significant difference between Black and White targets, (dav = .03, 95% dav = −.02, 95% CI [−.10, .06]). Similarly, for no gun trials, the effect CI [−.23, .30]); however, in cities one SD above the mean there was a was significant for longer windows (k = 16, dav = .24, 95% CI [.16,  Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 125 Fig. 1. Forest plot for reaction time to gun trials (top panel) and no gun trials (bottom panel). For gun trials, a negative effect was expected (faster to shoot armed Black versus White targets). For no gun trials, a positive effect was expected (slower to not shoot unarmed Black versus White Targets). Note that S = study, BP = Black participants, WP = White participants, Com = community, DO = Denver officers, NO = national officers, CP = control participants, ET = early trials, PO = police officers, PR = police recruits, and UG = undergraduates. .31]), but not for the 630 millisecond window (k = 15, dav = −.01, 95% White targets as has been proposed in the literature (e.g., Correll et al., CI [−.04, .09]). The effect of response time window was not significant 2002). for false alarm rate (k = 28, p = .915), shooting sensitivity (k = 30, p = .410), or shooting threshold (k = 28, p = .122). These results 2.3.2. Stimulus type make sense given that shorter response windows reduce variability in We found that in terms of reaction time, the stimulus type only reaction times and diminish the ability to detect significant reaction marginally moderated the effect size for gun trials (b = −.06, CI time effects. However, it does not appear that shorter response [−.13, .01], k = 32) and no gun trials (b = .06, CI [−.01, .13], k = 32). windows differentially increase the number of errors for Black versus Correll et al.'s (2002) task depicts full body images of Black and White 126 Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 Fig. 2. Forest plot for false alarms. A positive effect was expected (more errors for shooting unarmed Black versus White targets). Note that S = study, BP = Black participants, WP = White participants, CP = control participants, ET = early trials, PO = police officers, PR = police recruits, and UG = undergraduates. men with an environmental background, holding either a gun or a type was also not significant for any operationalization of shooter bias neutral object (e.g., cell phone). Plant and Peruche's (2005) task depicts (p's ranged from .125–.942). Thus, it does not appear that police officers, Black and White faces with a gun or neutral object superimposed. Given community members, and undergraduates have different shooter that many of the studies using Correll et al.'s (2002) task had a shorter biases in general. response time window, we adjusted for response window and in both cases the effect of stimulus was no longer significant (p N .74). More- over, the effect of response window remained significant. In terms of 2.3.4. Comparison of moderators false alarm rates, the effect of stimulus type significantly moderated Given that for some operationalizations of shooter bias, there were the effect (b = −.18, CI [−.27, −.08], k = 28). The effect size for studies multiple significant moderators, we conducted follow-up analyses using Plant and Peruche's (2005) materials was positive (k = 8, dav = entering all moderators that were significant when entered individually .25, CI [.09, .42]), suggesting that there were more false alarms for into one model. As noted above, for both reaction time measures, Black versus White targets, whereas for studies using Correll et al. response window was the only significant moderator when both (2002)'s materials the effect size was negative (k = 20, dav = −.11 stimulus type and response window were entered into the model. For CI [−.21, −.01]). This effect was still significant when adjusting for shooting sensitivity, there was only one significant moderator therefore response time window. Stimulus type significantly moderated the effect no further analyses were needed. for shooting sensitivity (b = .09, CI [.01, .17], k = 30). For studies using For false alarms, when all three significant moderators were added Plant and Peruche's (2005) materials, there was no effect for shooting to the model, neither percentage of non-White (b = −.001, 95% CI sensitivity (dav = .06, CI [−.21, .08], k = 9) whereas for studies using [−.008, .005]), nor stimulus type (b = −.09, 95% CI [−.22, .03]) were Correll et al.'s (2002) materials, there was a significant effect (dav = still significant. However, strictness of gun laws was still significantly .12, CI [.03, .21], k = 21), suggesting greater sensitivity for shooting related to effect size (b = −.005, 95% CI [−.009, −.001]. For states Black vs. White targets in these studies. Finally, stimulus type did not with permissive gun laws, the effect was positive (dav = .20, 95% CI moderate the effect of shooting threshold (k = 29, p = .346). In sum, [.05, .35]) suggesting more false alarms for Black (versus White targets), there does not appear to be an overall moderating effect of stimulus whereas for states with strict gun laws the effect was not significantly type across different operationalizations of shooter biases. The effect different from zero (dav = −.08, 95% CI [−.24, .07]). These results sizes of shooting threshold biases (and to some extent reaction time) possibly suggest that the association between percentage of non- were, however, the most reliable across stimuli, possibly suggesting White/stimulus type and effect size were due to shared variance with that they should be the main focus of investigations. gun laws (percentage of non-Whites: (r = −.42, 95% CI [−.65, −.11]; stimulus type: r = −.41, 95% CI [−.65, −.09]). However it is also 2.3.3. Sample type possible that the more comprehensive model reduced statistical To examine whether the sample type moderated the effect size we power. Similar to the results for false alarms, the results for shooting used two effect coded variables: one that contrasted undergraduates threshold showed that gun laws (b = .010, 95% CI [.003, .017]), but with police officers and recruits, the other contrasting community not percentage of non-Whites (b = −.003, 95% CI [−.017, .011]), was members with police officers and recruits. For all operationalizations related to effect size. In states with permissive gun laws (−1 SD), the of shooter bias, adding these two variables to the model did not signifi- effect size was medium in size and in the direction of lower thresholds cantly improve the fit (Q[2] ranged from .30–4.31, p's ranged from for Black versus White targets (dav = −.49, 95% CI [−.81, −.16]). In .115–.859). Based on suggestions from a reviewer, we also conducted states, with strict gun laws (+ 1 SD) the effect was not significantly a follow-up analysis where undergraduates and community members different from zero (dav = .07, 95% CI [−.14, .28]). Again these results were compared with police officers. In this case, the effect of sample could indicate shared variance or changes in statistical power.  Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 127 Fig. 3. Forest plots for shooting sensitivity (top panel) and shooting threshold (bottom panel). There was no directional prediction for shooting sensitivity. For shooting threshold, a negative effect was expected (lower threshold for deciding to shoot Black versus White targets). Note that S = Study, BP = Black participants, WP = White participants, Com = community, DO = Denver officers, NO = national officers, CP = control participants, ET = early trials, PO = police officers, PR = police recruits, and UG = undergraduates. 2.3.5. Correlation results implicit (r = −.02, 95% CI [−.11, .06], k = 5) nor explicit motivation We coded the correlation between shooter bias (in all cases but one to control prejudice (r = .03, 95% CI [−.10, .17], k = 5), had a significant operationalized as a difference in reaction time) and six different relation with shooter bias. Finally, consistent with the results for individual difference measures. The relation between shooter bias and percentage of non-Whites in the community, there was a significant awareness of discrimination against Blacks was small and not signifi- positive relation between shooter bias and contact with out-groups cantly different from zero (r = .03, 95% CI [−.06, .14], k = 5). Personal (r = .14, 95% CI [.04, .23], k = 6), suggesting that people who had endorsement of stereotypes (r = .08, 95% CI [.01, .15], k = 6) but not more contact with Blacks had a larger shooter bias. None of these effects knowledge of cultural stereotypes (r = .06, 95% CI [−.01, .13], k = 7) had significant heterogeneity (p's ranged from .13–.62), aside from had a small significant relation with shooter bias, such that greater explicit control of prejudice, which had marginal heterogeneity, endorsement of stereotypes was related to more shooter bias. Neither Q(4) = 9.17 p = .05. Given that these effects are based on a small 128 Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 number of studies (k's range from 5 to 7), these results should be One theory posited by Correll, Hudson, Guillermo, and Ma (2014) to interpreted with caution. With that in mind, the results seem to suggest explain why in their studies, police officers, in contrast to community that prejudicial attitudes, at best, have very small relations with shooter members and college students, do not show shooting threshold biases, biases. is because police officers might develop more cognitive control from training to overcome these biases. They are, however, still susceptible 2.3.6. Publication bias to reaction time biases due to implicit or explicit knowledge or endorse- We assessed for publication bias in four ways. First, we calculated ment of stereotypes about Blacks. More studies specifically designed the fail-safe-N using three different methods for the operationalizations to determine what factors predict different shooter biases are needed. of shooter bias that were significant (i.e., gun trials, no gun trials, and It is clear that more work on the construct validity of different shooting threshold). All methods (Rosenthal, Orwin, and Rosenberg) operationalizations of shooter biases is needed. suggested that there was no effect of publication bias (N's ranged from We found that studies conducted in states with more permissive gun 392 to 956) on the gun trials effect. Similarly, there was no effect of laws (i.e., less laws regulating gun usage) had bigger effects for false publication bias on the no gun trials effect (N's ranged from 360 to alarm rates and shooting threshold biases. This effect was still present 635) or shooting threshold (N's ranged from 514 to 758). Second, we when adjusting for general liberalness of the state as indexed by state calculated the rank order correlation of funnel plot asymmetry and voting behavior, suggesting that the effect is somewhat specific to gun found no evidence for publication bias for the gun trials effect (τ = laws. Our results do not allow us to draw conclusions about what char- .02, p = .884), no gun trials effect (τ = .03, p = .809), or shooting acteristics of states with permissive gun laws explain the association threshold (τ = .04, p = .749). Third, we used the trim-and-fill proce- with individual racial shooter biases; however, we offer two possibilities dure, which suggested that there were no missing studies. Finally, that may be tested in future research. First, research shows that support given that we included many unpublished dissertations (41% of our for permissive gun control is related to conservative political ideology studies), we compared the effect size of published and unpublished (Branscombe, Weir, & Crosby, 1991). Other research has found that studies (i.e., dissertation studies). We found that the effects were larger political conservatism is related to prejudicial attitudes against Blacks for published studies for the included operationalizations of shooter (e.g., Reyna, Henry, Korkfmacher, & Tucker, 2006; Sidanius, Pratto, & biases. For gun trials, the effect size for published studies was −.20 Bobo, 1996). Therefore, to the extent that racial shooter biases are a (CI: [−.32, −.08], k = 20) and for unpublished studies was −.01 (CI: form of prejudice, political conservatism (and the underlying need to [−.10, .08], k = 12). For no gun trials, the effect size for published manage uncertainty and threat; Jost, Glaser, Kruglanski, & Sulloway, studies was .16 (CI: [.02, .29], k = 20) and for unpublished studies it 2003), may explain the gun control findings. was .04 (CI: [−.06, .15], k = 12). For shooting threshold, the effect A second possibility is that the gun culture in states with permis- size for published studies was −.44 (CI: [−.76, −.12], k = 16) and sive gun laws affects the wiliness to shoot possible perpetrators of a for unpublished studies it was .11 (CI: [−.11, .35], k = 13). Thus, crime in a way that affects Black targets more than White targets. On while there does not seem to be any publication bias across studies the surface, this makes sense because states with more permissive used in the meta-analysis, our results suggest that studies with larger gun laws also have more extrajudicial shootings of racial and ethnic effect sizes are more likely to be published versus unpublished minorities (Price et al., 2004). Moreover, Coa, Cullen, Barton, and (i.e., dissertations). Blevins (2002) found that willingness to shoot a perpetrator of a crime was positively related to childhood socialization to guns, fear 3. Discussion of crime, and recent increases in the number of Black residents in the neighborhood. This possibly suggests that people with more ex- In summarizing a decade of research on racial shooter biases, we posure to guns (e.g., people from states with permissive gun laws) found that across all studies, there were significant effects for reaction are more willing to use guns against perpetrators of crime, who time and shooting threshold biases. Compared with White targets, they assume, due to stereotypes, to be racial and ethnic minorities. participants were quicker to shoot armed Black targets, slower to not Thus, it is possible there are cumulative effects of permissive gun shoot unarmed Black targets, and were more likely to have a liberal laws, willingness to shoot, and race-related fear of crime that, to- shooting threshold for Black targets. They were not, however, more gether, contribute to racial shooter biases. Alternatively, individuals likely to be sensitive to or have a higher false alarm rate for Black (versus who are more willing to shoot may be influencing policy in such a White) targets. way that may increase the likelihood of permissive gun laws The significance of reaction time and shooting threshold biases – but (e.g., voting for political candidates in line with their views on gun not false alarm rates or shooting sensitivity – presents an avenue for laws). Though plausible, without further research these hypotheses theorizing about what drives different shooter biases. Given that people should be considered tenuous. respond to stereotype-consistent information more quickly than The results of our racial proportion moderator analysis showed stereotype-inconsistent information (Blair & Banaji, 1996; Devine, that studies conducted in cities with a higher proportion of non- 1989), it is possible that reaction time biases are a result of participants' Whites had a larger effect for shooting threshold. The correlational awareness or endorsement of stereotypes about Blacks and criminality/ results also showed that contact with Blacks was related to larger dangerousness. Specifically, participants may be faster to shoot Black shooter biases. These findings stand in contrast to intergroup contact (versus White) targets with a gun because it fits with a racial stereotype. theory (Pettigrew & Tropp, 2006), which suggests that contact with This interpretation is partially supported by the correlation between members of an out-group should reduce prejudicial behavior. For in- reaction time biases and endorsement of stereotypes. However, this tergroup contact to be effective, however, specific conditions are re- relation was quite small, suggesting other factors may be at play. quired (e.g., common goals, equal status; Allport, 1954/1988; The real world implications of reaction time biases may not be as Pettigrew, 1998) that could not be assessed in our analysis. Our find- devastating as with the shooting threshold bias because reaction time ing is in line with other work showing that changes in racial/ethnic biases are calculated based on correct decisions. For example, if a person makeup of a neighborhood increases fear of crime, gun ownership, takes a few extra milliseconds to decide not to shoot an unarmed Black and willingness to shoot perpetrators (Coa et al., 2002; Young, person, the consequences are low. Conversely, if that person has a 1985). Thus, in the absence of the necessary conditions, more contact liberal shooting threshold and is generally more likely to make the with non-Whites might make Whites more afraid of crime and thus decision to shoot Black targets, then the outcome for the Black person more willing to shoot Blacks, whom they might perceive to be perpe- has the potential to be lethal. It is therefore important to understand trators of crime. When both proportion of non-Whites and gun laws what factors may put people at risk to having different types of biases. were in the same model, it appeared that the results for gun laws  Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 129 were more robust. Therefore, it is possible that the overlap between References racial makeup of a city and state level gun laws explain these results. Akinola, M., & Mendes, W.B. (June 2011). Stress exacerbates shooting errors among police Reduced statistical power associated with more complex models, officers. IACM 24TH Annual Conference paper. however, should also be considered as an explanation. Allport, G.W. (1954/1988). The nature of prejudice Addison-Wesley, Reading, Mass. Our results also allow us to make some recommendations about Binder, A., & Scharf, P. (1982). Deadly force in law enforcement. Journal of Research in Crime and Delinquency, 28, 1–23. http://dx.doi.org/10.1177/001112878202800101. conducting and reporting of racial shooter bias studies. First, the ef- Black Lives Matter (2015). About Us Retrieved from: http://blacklivesmatter.com/about/ fects are generally small, so researchers may need large samples to on August 13, 2015. detect effects. For example, according to G*Power, to have .80 Bliar, I.V., & Banaji, M.R. (1996). Automatic and controlled processes in stereotype power to detect a within-subjects effect size of 0.13, 467 participants priming. Journal of Personality and Social Psychology, 70, 1142–1163http://dx.doi. org/10.1037/0022-3514.70.6.1142 are required. This is much larger than the average 85 participants in Brady Campaign to Prevent Gun Violence. (2013). [[Legislative initiatives]. Retrieved the studies included in this meta-analysis. Second, researchers January 24, 2014], from http://www.bradycampaign.org/lesislation should at least report results for the two reaction time measures Branscombe, N. R., Weir, J. A., & Crosby, P. (1991). A three‐factor scale of attitudes toward guns. Aggressive Behavior, 17(5), 261–273. and shooting threshold, as they are the most reliable. For full trans- Chiricos, T., Hogan, M., & Gertz, M. (1997). Racial composition of neighborhood and fear of parency and if space allows, it would be ideal for researchers to re- crime. Criminology, 35, 107–132. http://dx.doi.org/10.1111/j.1745-9125.1997. port all operationalizations of shooter biases. Such standardization tb00872.x. in reporting across studies should help reduce the likelihood of re- Coa, L., Cullen, F.T., Barton, S.M., & Blevins, K.R. (2002). Willingness to shoot: Public attitudes toward defensive gun use. American Journal of Criminal Justice, 27, 87–109. searchers only publishing operationalizations that have significant http://dx.doi.org/10.1007/BF02898972. results. Third, if researchers are interested in focusing on reaction Correll, J., Hudson, S.M., Guillermo, S., & Ma, D.S. (2014). The police officer's dilemma: A time biases, they should uses response windows longer than 630 decade of research on racial bias in the decision to shoot. Social and Personality Psychology Compass, 8, 201–213. http://dx.doi.org/10.1111/spc3.12099. milliseconds, as the effects are essentially zero with shorter Correll, J., Park, B., Judd, C.M., & Wittenbrink, B. (2002). The police officer's dilemma: timeframes. Researchers more interested in shooting threshold Using ethnicity to disambiguate potentially threatening individuals. Journal of could use any response window according to our results. Finally, Personality and Social Psychology, 83, 1314–1329. http://dx.doi.org/10.1037/0022- 3514.83.6.1314. our results do not suggest that stimulus type or sample type have Correll, J., Park, B., Judd, C.M., & Wittenbrink, B. (2007). The influence of stereotypes on particularly important effects on the results. decisions to shoot. European Journal of Social Psychology, 37, 1102–1117. http://dx. This meta-analysis contributes to the literature by quantitatively doi.org/10.1002/ejsp.450. Correll, J., Wittenbrink, B., Park, B., Judd, C.M., & Goyle, A. (2011). Dangerous enough: examining the effect sizes of shooter biases and examining modera- Moderating racial bias with contextual threat cues. Journal of Experimental Social tors that may further explain these effects. Two important limita- Psychology, 47, 184–189. http://dx.doi.org/10.1016/j.jesp.2010.08.017. tions, however, should be considered. First, the proportion of non- Department of Justice (2001). Policing and homicide, 1976–98: Justifiable homicide by police, police officers murdered by felons. Whites in the region where the participants were recruited does R Development Core Team (2010). R: A language and environment for statistical computing. not give us information about the quality or quantity of inter-group Vienna, Austria: R Foundation for Statistical Computing. contact, which may be important mediating variables. The fact that Devine, P.E. (1989). Journal of Personality and Social Psychology, 56, 5–18. http://dx.doi. participants live in a city with more non-Whites does not necessarily org/10.1037/0022-3514.56.1.5. Dunlap, W.P., Cortina, J.M., Vaslow, J.B., & Burke, M.J. (1996). Meta-analysis of imply that they have more contact with them. Second, given that gun experiments with matched groups or repeated measures designs. Psychological law strictness is a policy-level variable, it is difficult to infer how it Methods, 1, 170–177. http://dx.doi.org/10.1037/1082-989X.1.2.170. reflects individual attitudes, particularly based on the data available Dunton, B.C., & Fazio, R.H. (1997). An individual difference measure of motivation to control prejudiced reactions. Personality and Social Psychology Bulletin, 23, 316–326. to us. http://dx.doi.org/10.1177/0146167297233009. This meta-analysis highlights a number of stimulating future Ferguson, C.J., & Brannick, M.T. (2012). Publication bias in psychological science: Prevalence, directions and recommendations for this area of study. First, future methods for identifying and controlling, and implications for the use of meta-analyses. Psychological Methods, 17, 120–129. http://dx.doi.org/10.1037/a0024445. work should be focused on understanding what the shooter biases Harmer, S. (2012). The role of multiple racial stereotypes in simulated shooting decisions. really measure and what psychological processes affect them Carleton University: Master's Thesis. (e.g., What is the nomological network for shooting threshold?). Harmer, S. (2012). The Role of Multiple Racial Stereotypes in Simulated Shooting Decisions. Carleton University Ottawa: Doctoral dissertation. Second, researchers should seek to understand the psychological fac- Hunsinger, M. (2011). Threat on the mind: The impact of incidental fear on race bias in tors that explain how community level variables are associated with rapid decision-making. Order No. AAI3427537, Dissertation Abstracts International. individual level behavior. For example, researchers should investi- Section B: The Sciences and Engineering, 594. Jost, J. T., Glaser, J., Kruglanski, A. W., & Sulloway, F. J. (2003). Political conservatism as mo- gate both the quantity and quality of contact to better understand tivated social cognition. Psychological Bulletin, 129(3), 339–375. the racial heterogeneity findings in the context of intergroup theory. Kenworthy, J. B., Barden, M. A., Diamond, S., & del Carmen, A. (2011). Ingroup identifica- Researchers should also specifically assess attitudes about guns tion as a moderator of racial bias in a shoot–no shoot decision task. Group Processes & (e.g., willingness to shoot perpetrators of crime) to better under- Intergroup Relations, 14(3), 311–318. Lakens, D. (2013). Calculating and reporting effect sizes to facilitate cumulative science: A stand the role of gun laws in shooting decisions. Third, the inconsis- practical primer for t-tests and ANOVAs. Frontiers in Cognitive Science, 4, 863. tency in self-reported prejudice correlations suggests that it may be Mays, V.M., Johnson, D., Coles, C.N., Gellene, D., & Cochran, S.D. (2013). Using the science fruitful to investigate other racial biases that may be relevant to ra- of psychology to target perpetrators of racism and race-based discrimination for intervention efforts: Preventing another Trayvon Martin tragedy. Journal for Social cially prejudiced behavior (e.g., dehumanization). Action in Counseling and Psychology, 5(1), 11. Given the potentially high translational value of research on Mekawi, Y., Bresin, K., & Hunter, C.D. (2015). White fear, dehumanization and low empa- shooter biases, continued work in this area should be a priority to re- thy: A lethal combination for shooting biases. Cultural Diversity and Ethnic Minority Psychology (in press). searchers interested in social justice across various domains Miller, M., Hemenway, D., & Azrael, D. (2007). State-level homicide victimization rates in (e.g., psychology, political science, law). Understanding the factors the US in relation to survey measures of household firearm ownership, 2001–2003. that contribute to shooter biases in the laboratory may provide crit- Social Science & Medicine, 64, 656–664. http://dx.doi.org/10.1016/j.socscimed.2006. 09.024. ical insight into targets of change for interventions designed for gun- Miller, S.L., Zielaskowski, K., & Plant, E.A. (2012). The basis of shooter biases: Beyond owners and law enforcement officers to reduce the frequency of un- cultural stereotypes. Personality and Social Psychology Bulletin, 38, 1358–1366. justified shootings of racial and ethnic minorities. Given the impact http://dx.doi.org/10.1177/0146167212450516. Musolino, E. (2012). The influence of target gender on shooting behaviour: An examination of of these shootings not only on families of victims, but on entire com- the role of automatic bias and controlled processes. Master's Thesis Carleton University. munities' sense of safety and wellbeing (Mays et al., 2013; Sha’Kema National Center for Injury Prevention and Control, Centers for Disease Control and & Thomas, 2015; Thomas & Blackmon 2015), research identifying ef- Prevention (2014). WISQARS: Fatal injury reports, national and regional, 1999– fective interventions is needed to maintain the basic human rights of 2010. Retrieved from http://webappa.cdc.gov/sasweb/ncipc/mortrate10_us.html Park, S.H., & Glaser, J. (2011). Implicit motivation to control prejudice and exposure to racial and ethnic minorities, keep communities safe, and increase the counterstereotypic instances reduce spontaneous discriminatory behavior. Korean effectiveness of policing. Journal of Social and Personality Psychology, 25, 107–120. 130 Y. Mekawi, K. Bresin / Journal of Experimental Social Psychology 61 (2015) 120–130 Park, S.H., Glaser, J., & Knowles, E.D. (2008). Implicit motivation to control prejudice mod- Sadler, M.S., Correll, J., Park, B., & Judd, C.M. (2012). The world is not black and white: Ra- erates the effect of cognitive depletion on unintended discrimination. Social cial bias in the decision to shoot in a multiethnic context. Journal of Social Issues, 68, Cognition, 26, 401–419. http://dx.doi.org/10.1521/soco.2008.26.4.401. 286–313. http://dx.doi.org/10.1111/j.1540-4560.2012.01749.x. Pettigrew, T.F. (1998). Intergroup contact theory. Annual Review of Psychology, 49, 65–85. Sha’Kema, M. B., & Thomas, A. J. (2015). African Americans and Trayvon Martin: Black Ra- http://dx.doi.org/10.1146/annurev.psych.49.1.65. cial Identity Profiles and Emotional Responding. Journal of African American Studies, Pettigrew, T.F., & Tropp, L.R. (2006). A meta-analytic test of intergroup contact theory. 19(3), 279–297. Journal of Personality and Social Psychology, 90, 751–783. http://dx.doi.org/10.1037/ Sidanius, J., Pratto, F., & Bobo, L. (1996). Racism, conservatism, affirmative action, and in- 0022-3514.90.5.751. tellectual sophistication: A matter of principled conservatism or group dominance? Plant, E.A., & Peruche, B.M. (2005). The consequences of race for police officers' responses Journal of Personality and Social Psychology, 70(3), 476–490. to criminal suspects. Psychological Science, 16, 180–183. http://dx.doi.org/10.1111/j. Simmons, J.P., Nelson, L.D., & Simonsohn, U. (2011). False-positive psychology 0956-7976.2005.00800.x. undisclosed flexibility in data collection and analysis allows presenting anything as Plant, E. A., Goplen, J., & Kunstman, J. W. (2011). Selective responses to threat: The roles of significant. Psychological Science, 22, 1359–1366. race and gender in decisions to shoot. Personality and Social Psychology Bulletin, 37(9), Taylor, A. (2011). The influence of target race on split-second shooting decisions in 1274–1281. http://dx.doi.org/10.1177/0146167211408617. simulated scenarios: A Canadian perspective. Order No. AAINR83227, Dissertation Plant, E.A., Peruche, B.M., & Butz, D.A. (2005). Eliminating automatic racial bias: Making Abstracts International. Section B: The Sciences and Engineering. race non-diagnostic for responses to criminal suspects. Journal of Experimental Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal Social Psychology, 41, 141–156. http://dx.doi.org/10.1016/j.jesp.2004.07.004. of Statistical Software, 36, 1–48. Price, J.H., Thompson, A.J., & Dake, J.A. (2004). Factors associated with state variations in Wittenbrink, B., Judd, C.M., & Park, B. (1997). Evidence for racial prejudice at the implicit homicide, suicide, and unintentional firearm deaths. Journal of Community Health, level and its relationship with questionnaire measures. Journal of Personality and 29, 271–283. http://dx.doi.org/10.1023/B:JOHE.0000025326.89365.5c. Social Psychology, 72, 262–274http://dx.doi.org/10.1037/0022-3514.72.2.262 Reyna, C., Henry, P. J., Korfmacher, W., & Tucker, A. (2006). Examining the principles in Young, R. L. (1985). Perceptions of crime, racial attitudes, and firearms ownership. Social principled conservatism: the role of responsibility stereotypes as cues for Forces, 64, 473–486. deservingness in racial policy decisions. Journal of personality and social psychology, 90(1), 109–128.