Experimental sleep loss, racial bias, and the decision criterion to shoot in the Police Officer’s Dilemma task

Violent behavior, police brutality, and racial discrimination are currently at the forefront of society’s attention, and they should be. We investigated whether mild sleep loss—as typical for many adults throughout the work week—could aggravate the socio-emotional-cognitive processes contributing to violence and discrimination. In a sample of 40 healthy young adults, we either experimentally restricted participants’ sleep for four nights (6.2 h/night) or let participants obtain normal sleep (7.7 h/night)—and then had them complete the Police Officer’s Dilemma Task. In this computerized task, the participant must rapidly decide to shoot or not shoot at White and Black men who either are or are not holding a gun. Results showed significant racial biases, including more and quicker shooting of Black targets compared to White targets. Furthermore, signal detection analyses demonstrated that mild sleep restriction changed participants’ decision criterion, increasing the tendency to shoot, even when controlling for psychomotor vigilance, fluid intelligence, and self-reported desirability to behave in a socially acceptable manner. The increased tendency to shoot was also observed in participants who reported believing that they had adapted to the sleep loss. Future experimental research using trained police officers will help establish the generalizability of these laboratory effects. Importantly, sleep loss is modifiable via organization-level changes (e.g., shift scheduling, light entrainment) and individual-level interventions (e.g., sleep hygiene education, incentives for behavioral change), suggesting that if sleep loss is corrected, it could save lives—including Black lives.

The Friday test session 22 concluded with the Police Officer's Dilemma Task 23,24 . This task was developed to simulate the real-world challenge of police officers needing to make quick decisions in response to perceptually ambiguous situations (i.e., threatening or non-threatening). Participants were shown scenes that were either empty (no person) or contained a target (person). The target person was either unarmed or armed with a gun and either White or Black, and participants had to quickly decide whether to shoot or not shoot the target. We analyzed accuracy, reaction times, timeouts (failures to respond within the time limit), and signal-detectiontheory metrics by conducting a series of 2 × 2 analyses of variance (ANOVAs) that included target race (White, Black) and sleep condition (normal, restricted).
The results revealed significant evidence for racial biases. On armed trials, participants were more likely to "shoot" Black targets than White targets (Fig. 1A), F(1, 38) = 5.37, p = 0.026, η p 2 = 0.12, and with quicker response times (Fig. 1B), F(1, 38) = 4.13, p = 0.049, η p 2 = 0.10. On unarmed trials, there were no effects for mean response times (ps > 0.10), but there were far more failures to respond within time limits (timeouts) for Black targets than White targets (Fig. 1C), F(1, 38) = 19.63, p < 0.001, η p 2 = 0.34. This result suggests that participants experienced greater difficulty in quickly identifying the Black targets as non-threatening. Similarly, there were significantly fewer "don't shoot" responses to unarmed Black targets than unarmed White targets (Fig. 1D), F(1, 38) = 11.45, p = 0.002, η p 2 = 0.23. False alarms were rare and showed nonsignificant effects. We conducted follow-up analyses to determine whether any of the significant effects interacted with gender. All gender main effects and interactions were nonsignificant (ps > 0.05), indicating that men and women demonstrated similar levels of racial biases in the Table 1. Sample characteristics and task performance. Data are plotted as mean ± standard deviation, with measures separated by Monday (Session 1) and Friday (Session 2) data collection. The conditions did not differ pre-experimentally (Session 1), but showed strong differences in sleep duration, daytime sleepiness, mood disturbances, and vigilance by the end of the week. An effect of sleep condition emerged when conducting analyses based on signal detection theory. We distinguished sensitivity to armed targets (d′) and the decision criterion (c) for shooting a target. Similar to previous reports 24 , there were no significant effects on d′ ( Fig. 1E; ps > 0.  www.nature.com/scientificreports/ to White targets. The interaction remained statistically significant even when controlling for participants' psychomotor vigilance, fluid intelligence, self-reported desirability to behave in a socially acceptable manner, and self-reported belief that they had adapted to the sleep loss, F(1, 34) = 6.07, p = 0.019, η p 2 = 0.15. The effects of sleep restriction and target race on decision criterion were not moderated by gender, F(1, 36) = 0.74, p = 0.40, η p 2 = 0.02 (and there were no interactions; all Fs < 1). Women and men showed identically lenient c scores after sleep restriction [Ms = − 0.158 and − 0.151, respectively, F(1, 18) = 0.004, p = 0.95, η p 2 < 0.01] and similar c scores after normal sleep (Ms = − 0.056 and 0.048, respectively, F(1, 18) = 1.84, p = 0.19, η p 2 = 0.09). Interestingly, when limiting the sample to only White participants, the sleep restriction effects on decision criterion were exacerbated. For example, the effect size increased from η p 2 = 0.11 in the overall sample to η p 2 = 0.37 in only the White participants, F(1, 18) = 10.50, p = 0.005. The sleep by race interaction also increased in effect size from η p 2 = 0.17 in the overall sample to η p 2 = 0.39 in the White participants, F(1, 18) = 11.64, p = 0.003.

Discussion
To promote social justice, society must address violent behavior, police brutality, and racial discrimination. The current data suggest that some of these behaviors may be exacerbated by sleep loss, even when the sleep restriction is relatively mild (6.2 h/night), even when sleep restricted individuals believe they have "adapted" to the short sleep, and even in individuals with inclinations toward behaving in a socially desirable manner. The current experimental findings provide causal support for an earlier correlational study in newly recruited police officers 21 . In Ma et al. 's 21 study, newly recruited officers with self-reported total sleep time − 1 SD below the mean (≤ 5.25 h) showed greater racial bias in the shooter task than officers with self-reported total sleep time + 1 SD above the mean (≥ 8.05 h). Both their study and the current results implicated sleep loss in more "itchy trigger finger" type responding. A distinction in the studies, however, was that Ma et al. 21 found that short sleepers were more likely to show a lenient decision criterion toward Black targets, whereas we found sleep restriction affected the decision criterion primarily for White targets. In another recent study 25 , total sleep deprivation ("all nighter") in college student participants led to more errors on the Police Officer's Dilemma Task, particularly more "shooting" of unarmed targets (but not in a racially biased manner). These effects were not mitigated by caffeine consumption. The differences across these data sets and our results could be due to study differences in experimental/correlational design, sample size, sleep measurement, type of sleep loss (acute, chronic), sample composition (e.g., young adults tend to be more open and liberal), or other factors that might become evident with further studies.
Our findings also join a growing literature that implicates short and poor-quality sleep in stereotyping and bias toward out-groups 13,16-18 , perhaps in a circadian-sensitive manner 19,20 . Sleep deprivation might lead to increased manifestations of discrimination by altering frontoparietal attentional processes 9,10 , amygdala/emotional reactivity processes 8,14 , and/or threat perception processes 15,26 . The first potential mechanism-altered frontoparietal attentional processes-relates to Ghumman and Barnes' 13 idea that resources are needed to maintain control over prejudices, and if such resources are depleted by sleep loss, then biases and discriminatory behaviors will manifest. This idea is logical, but we did not observe direct evidence for it in the current study. When we controlled for performance on psychomotor vigilance and fluid intelligence tasks, we continued to observe that sleep restriction affected the decision criterion to shoot.
Distinguishing the amygdala/emotional reactivity and threat perception accounts is trickier because these views predict fairly similar outcomes. For example, sleep loss may make one react more emotionally to negative stimuli, manifesting in this paradigm as "shoot" responses on armed target trials. However, if threat perception is elevated after sleep loss, then "shoot" responses should increase across both armed and unarmed target trials. The collective data on decision criterion, therefore, favored the altered threat perception account. Nonetheless, we cannot dismiss the possibility that sleep loss also affects the decision criterion to shoot via amygdala/emotional reactivity, inhibitory inefficiency, impulsivity, altered ethical decision making, or general tendencies toward aggression [26][27][28][29][30][31][32] , accounts which are not necessarily mutually exclusive with the altered threat perception account.
The purpose of the current study was to investigate the causal effects of sleep loss on socio-emotional-cognitive processes that could contribute to racial discrimination. Another emerging literature that connects sleep and racial discrimination falls under the umbrella of health disparities. Health disparities refer to group differences in disease rates/outcomes that are systematic, avoidable, and unfair 33 . For example, the Centers for Disease Control and Prevention 34 report that Black or African American individuals have considerably higher rates of diabetes, hypertension, heart disease, and HIV infection than White and Asian American individuals (disparities are also observed for Hispanic/Latinx individuals as well as American Indian/Alaska Native individuals) 35,36 .
One of the leading models for explaining health disparities points to upstream determinants influencing downstream determinants. Upstream determinants refer to fundamental social/environmental causes, including differences in socioeconomic status, discrimination, and other societal inequities. Downstream determinants refer to the individual's experiences and can include allostatic load, self-determination, and health behaviors (note that models also indicate that gene-environment interactions feed back to upstream factors) 37,38 . Pertinent to the present work, racial/ethnic differences in sleep health are recognized as markers of health disparities (i.e., sleep disparities) [39][40][41][42][43][44][45][46][47] . Furthermore, sleep disparities are emerging as an important contributor to race/ethnic disparities in cardiovascular [48][49][50][51][52] , metabolic 52,53 , and cognitive 54,55 health outcomes. Clearly, sleep health is important in all communities, even when isolated from matters of violence and aggressive race-based discrimination.
One of the limitations of the current study was the sample size, including a limited number of participants from non-White groups. On the one hand, this limitation does not undermine the primary purpose of understanding how majority groups (White participants) display racial biases and altered behaviors under mild sleep restriction (note that police officers in the U.S. are disproportionately White relative to Census estimates 56 ). On the other hand, the limited variability in participant demographics hindered us from examining whether effects Scientific Reports | (2020) 10:20581 | https://doi.org/10.1038/s41598-020-77522-z www.nature.com/scientificreports/ were driven by in-group/out-group determinants 57,58 versus general stereotyping of Black men 59 . Increasing diversity in participant enrollment is a longstanding problem in both psychological and medical research [60][61][62] .
Overcoming barriers to participant representation requires additional time and monetary resources for recruitment, collaboration with community leaders, and cultural tailoring 62,63 . It also requires increasing representation at the level of scientists, editors, and funding agencies 64,65 . In other words, academia as a whole must acknowledge and work to address its own systematic biases.

Conclusions and future directions
Police work is one of the most sleep-depriving professions 5,6 . It is also one of the few professions that involves carrying firearms and encountering potentially hostile situations. It remains an open, empirical question whether the gun-safety training offered to police officers shields them from the deleterious effects of sleep loss. Though that is possible, we note the following empirical evidence: (1) shielding was not observed in a correlational study on sleep and shooter bias in newly recruited police officers 21 ; (2) night shift patrol officers were less likely to prevent deadly outcomes in video-based use-of-force simulators than day shift patrol officers 66 ; (3) police officers sleep even worse than the general public 5,6 ; and (4) workers in other fields that require extensive, safety-focused training (e.g., medicine) still show susceptibility to sleep loss and shift-work fatigue [67][68][69][70] . Therefore, sleep loss deserves closer consideration by public officials and law enforcement as an insidious culprit underlying people's sometimes faulty judgments and impulsive behaviors.
To eliminate discriminatory behaviors in society-including the shooting of unarmed Black individualsthere will be no substitute for eliminating underlying racism within individuals and systems. To achieve this end, education, policy changes, and individual efforts will be key. But, we also contend that there is room for reducing sleep debt in society 3,71,72 . Workplace sleep education programs 73 , behavioral change programs 74 , targeted memory reactivation interventions 75 , culturally tailored and personalized programs [76][77][78] , and schedules allowing for circadian alignment 79 have all been shown to be effective. Based on the current results, as well as the literature on health disparities, it appears that correcting sleep loss in society may play a significant role in saving lives, including Black lives.

Methods
Participants. Forty-six adult participants were recruited through flyers and social media advertisements in spring 2018. The sample size was determined a priori to detect a large condition effect size, as often observed in sleep restriction studies 16,80 . They received monetary compensation for participating. Advertisements referred to the study as examining Sleep and Social Cognition to prevent expectation effects that would have emerged if the study were advertised as testing racial biases. Exclusion criteria included being under age 18 or over the age of 30, being a habitual short sleeper (< 7 h per night), and having clinically-diagnosed sleep disorders. Two participants did not complete all study sessions. Because there were only four African American participants we were unable to examine in-group versus out-group effects in this study; therefore, we excluded those data from the Police Officer's Dilemma Task analyses (their data were included in other measures collected as part of this larger sleep study 22 ). The study was approved by the Baylor University Institutional Review Board and all research was performed in accordance with relevant guidelines/regulations. All participants signed informed consents prior to the study and were debriefed upon study completion.
Overview of design. The 5-day study consisted of two 1.5 h in-lab experimental sessions -one on Monday and one on Friday. For the four nights between each session, participants were instructed to maintain a wake time of 7:30 am and were randomly assigned to maintain a bedtime of either 10:30 pm (normal sleep condition) or 1:30 am (restricted sleep condition). To promote ecological validity, participants slept in their home environment (home-based and lab-based sleep restriction protocols produce similar effect sizes on cognitive functioning 80 ). Across the week, we measured sleep using wristband actigraphy, which is designed to detect ambient light and movement (accelerometer-based).
Equipment. The actigraphy device was the Actiwatch Spectrum Plus (Phillips Respironics Actiwach-2), which has been validated for distinguishing sleep/wake state against the gold standard of polysomnography 81,82 . The device was set to factory recommended settings that defined epoch length as 15 s, wake onset as 40 activity counts, and sleep as periods of rest with 10 min without movement. Participants also maintained a sleep diary 83 89 . At the end of Session 1, participants received the actigraphy device and sleep diary, and they were randomly assigned to the normal sleep and restricted sleep conditions. Assignment order was determined using the blocked random assignment method (blocks of 2, 4) using the sealedenvelope.com online tool. To minimize the risk of experimenter bias, the researchers who interacted with the participants were masked to each participant's experimental condition. We accomplished masking by having a research staff member who would not be interacting with participants generate the random assignments and seal participants' condition assignments in envelopes. Only after participants completed Session 1, did a research staff member hand the participant a sealed envelope with their condition with the instruction to open the envelope after leaving Session 1. Participants were explicitly instructed to not mention their assigned sleep condition to the experimenter as they were leaving or when they returned for Session 2. Participants were instructed not to nap during the week.  23,24 . The participant's objective is to shoot (Q key) the armed targets and to not shoot (P key) the unarmed targets. Armed targets are holding either a revolver or a pistol, whereas unarmed targets are holding an aluminum can, a camera, a cell phone, or a wallet. In the procedure, empty scenes continuously flash in 500-1000 ms intervals, and after 1-4 empty scenes, a target scene will appear. After 16 practice trials, participants completed 100 experimental trials (25 White armed trials, 25 White unarmed trials, 25 Black armed trials, and 25 Black unarmed trials). To simulate the speeded nature of real world dilemmas, participants had to respond within 630 ms, otherwise the trial resulted in a timeout 90 . Response times were analyzed on correct trials. Accuracy data produced estimates for correct hits (shooting an armed target), misses (not shooting an armed target), false alarms (shooting an unarmed target), and correct rejections (not shooting an unarmed target). Following previous methods 24 , we subjected these accuracy data to signal detection analyses to produce estimates of d′, which reflects sensitivity, or the ability to discriminate armed from unarmed targets (which typically does not differ across White and Black targets 24 ). We also generated estimates of decision criterion, or c, which reflects the threshold one sets for making a shoot response. Positive c values indicate a conservative threshold, such that one favors a don't-shoot response; negative c values indicate a lenient threshold akin to having an "itchy trigger finger. " Signal detection estimates (d′ and c) were calculated separately for White targets and Black targets.
Statistical analyses. Statistical analyses were completed using SPSS 26 (IBM) and graphically displayed using Prism 8.4 (Graphpad). D prime was calculated as d′ = zH = ZFA and criterion was calculated as c = − 0.5 X (zFA + zH). For both calculations, zFA refers to the z-score of false alarms and zH refers to the z-score for hits (with minimum and maximum set to 1/2n and 1-1/2n, following 24 ). We conducted a series of 2 × 2 mixed analyses of variance (ANOVA) in which condition (normal, restricted) varied between subjects and target race (White, Black) varied within subjects. Alpha was set to 0.05 and all tests were two-tailed. Effect sizes were estimated using Cohen's d and partial eta squared. License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.