On March 5, 2019, officers in Harlem, New York, responded to a call stating that there was a man with a gun on the fourth floor of an apartment building (Santia and Strahan 2019). When they approached the man, Michael Cordero, 34, in the fourth-floor hallway, they told Cordero to remove his hands from his pockets. According to the Chief of the NYPD’s Force Investigation Division, Kevin Maloney, Cordero told officers, “What do you mean, ‘take my hands out of my pockets?’ I’ve got a gun. What the f— do you mean?” (Santia and Strahan 2019). Cordero furtively removed his hands from his pockets, put his right hand back into his pocket, and removed a black object that he pointed at officers while assuming a shooting stance. One officer fired three rounds at Cordero, hitting him non-fatally in the hip. The object in Cordero’s hand was a wallet.

The Washington Post maintains a national database of officer-involved shootings (Police Shootings Database 2021). According to that database, since January 2015 approximately 1000 people have died each year in officer-involved shootings. Of the 1021 individuals shot and killed by police in 2020, 118 people (12%) were either unarmed, holding a toy weapon, or holding some other object that was unidentifiable as a weapon (e.g., a rock, a metal object/pole). A similar proportion of “mistake-of-fact” shootings is seen in the data for the years 2017–2019 (Police Shootings Database 2021) and is consistent with the 15% mistake-of-fact shootings found by Fachner and Carter (2015) in a review of deadly-force incidents in Philadelphia, Pennsylvania.

Mistake-of-Fact Shootings

Mistake-of-fact shootings—such as the shooting of Michael Cordero—are most commonly the result of an officer making a threat-perception failure (Fachner and Carter 2015). This type of error occurs when “an officer perceives that a suspect is armed due to the misidentification of a non-threatening object (e.g., a cell phone) or movement (e.g., tugging at the waistband)” (Fachner and Carter 2015, p. 30). Mistakes in threat perception may occur when expectations are set before the officer arrives at the scene. Along with prior knowledge about, or experience with, a suspect (Aveni 2003), dispatch information received en route to an incident can prime officers before they even reach the scene (Johnson et al. 2018; Mitchell and Flin 2007) and impacts their decision to use deadly force (Taylor 2020).

Upon arriving at the scene of an incident, situational factors can further affect an officer’s perception. These include low lighting, poor visibility due to distance between the officer and suspect, and/or the deceptive or threatening in-situ behavior of a suspect—such as rapid and unexpected movements (Aveni 2003). Improper identification of an object held by the suspect is a common mistaken fact. In the absence of confirmatory cues (e.g., a gunshot, a muzzle flash), officers may find it difficult to perceive whether a suspect is holding a weapon (e.g., a gun or a knife), a non-weapon (e.g., a cell phone or a wallet), or even nothing at all (e.g., Swaine 2015). In other words, officers’ object-identification decisions often involve a degree of uncertainty. In the interest of maximizing their own safety, officers might deal with this uncertainty by adopting a liberal response bias: tending to identify unknown objects as weapons—rather than as non-weapons—and responding accordingly.

Factors contributing to officer-involved shootings and mistake-of-fact shootings have been of interest to researchers in psychology, criminology, and criminal justice. For example, research in laboratory settings has examined the effects of suspect race (Correll et al. 2002; Plant and Peruche 2005), suspect attire and demeanor (James et al. 2018), officer experience (Ho 1997), and the content of dispatch information (Mitchell and Flin 2007; Taylor 2020) on police officers’ shooting behavior. This work was primarily conducted either by briefly presenting static images on a screen for rapid identification or by having participants respond to video simulations. Relatively little research has examined mistake-of-fact shootings from the perspective of perceptual–cognitive expertise, an area of research commonly applied to human performance in sport (Mann et al. 2007).

Perceptual–Cognitive Expertise

In the domain of sport research, perceptual–cognitive expertise is “the ability to identify and acquire environmental information for integration with existing knowledge such that appropriate responses can be selected and executed” (Mann et al. 2007, p. 457). Therefore, one thing differentiating experts from novices in their field is the ability to better attend to task-relevant perceptual cues while ignoring irrelevant cues. In interceptive sports such as soccer, studies have consistently found that expert players are better at making predictive judgements of an opponent’s actions than less experienced individuals (Mann et al. 2007). The ability to anticipate an opponent’s actions is particularly important when the time it takes to carry out a response to an event is greater than the time it takes the event itself to occur. For example, tennis players must start moving in the fractions of a second after the ball has left the opponent’s racket to successfully return a hit (Huys et al. 2009; Williams et al. 2002). Similarly, baseball batters must decide to begin their swing or refrain from swinging at a pitch within 200 ms of the ball being released to have a chance at successfully hitting the ball (Morris-Binelli et al. 2017). In such situations, waiting to obtain more information to facilitate an accurate response could mean that the response occurs too late to be effective.

Studies on perceptual–cognitive expertise in sport have consistently found that experts accurately anticipate and respond to their opponent faster than a novice in an adversarial situation (Mann et al. 2007). Experts may respond before their opponent even completes their movement because they are able to extract pertinent information at an earlier point in their decision-making process. An example of this would be an expert tennis player predicting whether a serve will arrive on their forehand or backhand side prior to racket/ball contact, while a novice tennis player may not be able to predict the serve direction until after the ball has left the racket. Waiting until the serve motion is complete would delay the novice’s response, leaving them unable to position themselves to return the serve. Not only do experts perceive and respond faster, but the result of their decision-making when predicting a scenario outcome is also typically significantly more accurate than their less-skilled counterparts (Loffing and Cañal-Bruland 2017).

Adversarial Situations in Law Enforcement

Shared situational aspects and parallels in the cognitive functions required by both athletes and law enforcement officers allow the work conducted on sport expertise to be applied to the study of law enforcement. Clearly, the potential negative outcomes in sport (e.g., losing a game or series) cannot be compared to negative outcomes of police–citizen interactions (i.e., serious injury or death of citizens and officers). Nevertheless, the study of skilled performance in dynamic, time-constrained, and uncertain situations in the sport context has much to offer in terms of a principled way of understanding skilled performance in law enforcement.

Both athletes and law enforcement personnel enter adversarial situations in which they must anticipate what their opponent—either an athlete on the opposing team or a suspect who may present a threat—will do next. Both scenarios are competitive in nature such that the opposing sides prefer differing outcomes that cannot both be achieved. “Athletes must perceive, process, and react to information physically” (Steel et al. 2015, p. 82), a task that law enforcement officers must constantly do as well. To accurately predict what an opponent will do next, athletes and law enforcement officers must be able to rapidly identify the motions made by their opponents, infer their intent (Steel et al. 2015), anticipate their opponents’ next move, and decide the best way to respond to achieve their preferred situational outcome. Each adversarial meeting is highly dynamic in nature, time-constrained, and contains an aspect of uncertainty. When stakes are high, competitors often attempt to succeed by misleading their opponent (e.g., via deception).

Deception

Deception is a way for individuals to increase their likelihood of outperforming their opponent by minimizing the amount of valid information they provide their opponent about their true intentions (Schmidt and Lee 2005). Deception is the concealment of information or the act of providing “misleading cues about [one’s] current intentions regarding their own future actions” (Güldenpenning et al. 2017, p. 1). The dependence on reliable visual cues in adversarial contexts is integral to performance. Although the presence of deception correlates with a diminished ability to accurately determine the intent of an opponent, studies with boxers (Ripoll et al. 1995) and badminton players (Park et al. 2019) have shown that the experience level of an individual moderates this effect, with experienced athletes experiencing a less severe decrement in performance compared to novice athletes.

Considering the underlying similarities between adversarial situations in sport and law-enforcement, it is reasonable to expect that similar aspects of expertise may be found in law enforcement officers as are seen in professional athletes. Like the pitcher trying to strike out a batter or the soccer player trying to get the penalty kick past the goalkeeper, suspects and law enforcement officers “face off” in an adversarial context where the suspect—much like the pitcher and the kicker—may try to convince the law enforcement officer that they are less of a threat than they truly are. In this case, the law enforcement officer—akin to the batter or the goalkeeper—must assess the suspect and determine how to proceed to achieve their desired outcome.

Perceptual–Cognitive Expertise in Law Enforcement

The study of perceptual–cognitive expertise has been applied to mistake-of-fact shootings in law enforcement using the temporal-occlusion method; a predominant method used in sport research to examine an athlete’s ability to anticipate their opponent’s actions. Suss and Raushel (2019) presented college students with temporally-occluded videos of actors drawing either a weapon or a non-weapon from a concealed location on their bodies. At the point of occlusion, participants identified the object as either a weapon or a non-weapon. In each clip, the actor—who faced the camera—drew either a replica revolver or a wallet from a concealed location behind their back or on their front (i.e., “appendix carry” for the revolver or front pocket for the wallet). The video stimuli were then temporally occluded (i.e., replaced by a black screen) at five points in the draw motion. Suss and Raushel applied a signal-detection-theory approach to their data analysis which—while uncommon in the anticipation literature (but see Cañal-Bruland and Schmidt 2009)—showed, as expected, that participants were more sensitive to the presence of a weapon at later occlusion points in the videos and found a significant main effect of increased sensitivity for front versus back draws. Scott and Suss (2019) used similar, temporally-occluded video stimuli in which the actors started facing away from the camera and then spun around to face the camera as they drew the object. Scott and Suss found that at early occlusion points, participants adopted a conservative response bias (i.e., they tended to identify the object as a non-weapon). However, at later occlusion points—when more information was available—participants adopted a more liberal response bias (i.e., they tended to identify the object as a weapon).

Although the use of video stimuli was a good first step to assess behavior in the specific context of use-of-force scenarios, there are limitations associated with their use. The finding that participants were more accurate in their response at later occlusion points is confounded by the fact that the object in the suspect’s hand was also visible at this point in the videos (Scott and Suss 2019; Suss and Raushel 2019). The participants in these studies were college students and the findings may not generalize to the actual population of interest: law enforcement officers. Additionally, the temporally-occluded videos were not free from potentially subjectively biasing features—such as suspect race, demeanor and facial expressions, age, gender, clothing (Correll et al. 2002, 2007; James et al. 2018)—that may affect an officer’s decision in a shoot/don’t-shoot task (Correll et al. 2002, 2007). The limited number of actors (i.e., 3 males and 1 female between the ages of 23 and 28) drawing their concealed object in a few ways means it is highly unlikely that their stimuli were fully representative of the “police suspect” population. This means that their results are possibly tainted by the lack of stimulus sampling (Wells and Windschitl 1999) and the generalizability of their findings is inherently limited.

Use of Biological Motion to Investigate Expertise

In both sport and law enforcement domains, decisions often need to be made before time-constrained situations can be fully analyzed. Therefore, there is a strong emphasis on the ability to anticipate an opponent’s next move based on kinematics to achieve a reliable prediction on how one’s opponent is about to behave (Wright et al. 2011). Biological movements—meaning any movements of a living organism—are easily recognized and identified, even in the initial stages before the motion is fully completed (Alaerts et al. 2011). Detailed models of human bodies are not required to study biological motion; stick-figure-like images are effectively able to depict patterns of biological motion (Johansson 1973).

The study of biological motion began in the 1970s and focused on motion pattern recognition, finding that people can easily recognize common patterns of biological motion (Johansson 1973). Across domains, general findings in biological-motion research show that “individuals are able to use relative (temporal and spatial) information from a person’s movement to recognize factors, including gender, age, deception, [and] emotion” (Steel et al. 2015, p. 78) to infer details about that person’s actions and intentions. Loffing and Cañal-Bruland (2017) reiterated that expert sport performers consistently display their superiority in performance over their less skilled counterparts due to their ability to “predict their opponents’ action intentions based on the opponents’ kinematics” (p. 8).

Due to humans’ established ability to recognize deception, emotion, identity, and intention from biological motion alone, we wondered whether law enforcement officers can discern a suspect’s immediate future actions in adversarial scenarios based on kinematic information alone. Even though the specific details of each suspect encounter (e.g., the location, lighting, the type and color of the weapon/object, the suspect’s characteristics) vary from case to case, one of the few commonalities between each adversarial law enforcement encounter is the fact that the suspect must perform physical motions when retrieving a weapon/object from a concealed location on their person. The only studies that we are aware of that assessed law enforcement anticipation ability partly on the basis of suspect biological motion (Scott and Suss 2019; Suss and Raushel 2019) were limited by the presence of other social contextual factors. Sport researchers have found a way to exclude social factors and focus only on kinematics through the use of point-light displays.

Point-Light Displays

Point-light displays are created by attaching reflective markers to a person’s major joints (i.e., shoulders, elbows, wrists, hips, knees, feet). When the person performs a physical action, the reflective markers are tracked in three-dimensional (i.e., xyz) space. The resulting point-light display depicts the underlying biological motion of a movement through the configuration of dynamic light points, typically against a black background. Point-light display techniques have been used in the study of biological motion to eliminate the external social factors (e.g., gender or race) that may bias one’s response to a stimulus (Davidson and Edgar 2012; Steel et al. 2015).

Point-light displays have been used to study perceptual skill and expertise differences in sport such as in the prediction of shuttle trajectory in badminton (Wright et al. 2011), hit trajectory in tennis (Williams and Ericsson 2005), and shot-on-goal trajectory in handball (Loffing and Hagemann 2014). These studies consistently found that participants were able to adequately recognize the biological motion of an opponent and accurately make anticipatory decisions—with expert athletes showing a performance advantage in their respective domains—regarding the intent of the figure in the video (Loffing and Cañal-Bruland 2017). Wright et al. (2011) showed badminton players of varying skill levels regular videos and point-light displays of a badminton player hitting a shuttle toward a receiving player and asked them to indicate where on the court they believed the shuttle would land. Using this procedure, they found that the use of point-light displays—compared to regular videos—did not inhibit participants’ ability to recognize the player’s motion (Wright et al. 2011).

The Current Study

Research regarding law enforcement use-of-force decision-making has yet to investigate whether officers are able to effectively use a suspect’s biological motion as a reliable cue in potential use-of-force scenarios; specifically, in their ability to identify an unknown object that a suspect is concealing on their person, based on underlying biological motion. To the best of our knowledge, point-light displays have not been used to assess perceptual–cognitive expertise in a law enforcement context. Therefore, the current study comprised two major phases: (1) creating realistic representations of a suspect drawing an unknown object from a concealed location and (2) presenting these stimuli to law-enforcement officers with varying amounts of law-enforcement experience.

This experiment was designed to determine whether officers are able to use biological motion as a cue to determine whether a potential suspect in front of them is holding a weapon or a non-weapon. Additionally, the behavioral patterns of the suspect were also considered to reflect times in which suspects may attempt to deceive officers by behaving threateningly while holding a non-weapon or behave compliantly while holding a weapon (i.e., surrendering). This resulted in a 2 (object type: weapon vs. non-weapon) × 2 (intent: threatening vs. non-threatening) within-subjects design. Based on the work of Suss and Raushel (2019) and Scott and Suss (2019), anticipation performance was assessed using signal-detection-theory measures of sensitivity and bias.

Sensitivity hypothesis: Officers, regardless of experience, will be sensitive to the presence of a weapon when the object is drawn with a threatening intent due to the consistent “threat” level between the object and intent.

Bias hypothesis: Officers, regardless of experience, will be liberally biased in their response (i.e., more likely to indicate that the object is a weapon) when the object is drawn threateningly due to the greater perceived danger from a threatening draw.

Experience hypothesis: Officers with more law enforcement experience (i.e., more years working in law enforcement) will be more sensitive to object type than less experienced officers. We make this prediction based on the likelihood that, compared to less experienced officers, more experienced officers will have encountered more unknown-object cases in their operational duties.

Method

Stimuli Creation

To create point-light displays that would present ecologically valid biological motion, we first sought information on how suspects—who might be armed—behave in face-offs with police.

Case Reviews

To ensure that the actor in our stimuli simulated actions that officers encounter on duty, we reviewed police use-of-force case files from a local law enforcement agency. In this case, the use-of-force is defined as “the amount of effort required by police to compel compliance by an unwilling subject” (International Association of Chiefs of Police 2001, p. 1). The amount of force used by officers can range from less-lethal techniques (e.g., verbalized commands with no physical contact, soft grabbing of subject to restrain them) to lethal force (e.g., use of a lethal weapon such as firearms) (National Institute of Justice 2009). We focused on cases that included video footage (i.e., body-camera and/or dash-camera recordings). In total, we reviewed approximately 1500 case files to find those that most closely reflected the same situation pattern that the stimuli were intended to emulate. Of those case files, 22 involved an officer or officers facing a suspect with an unknown object in their possession that the officer(s) could not see clearly (e.g., because the object was concealed behind the suspect or the suspect was too far away to be able to accurately determine what they held). Objects concealed by suspects included guns, knives, pens, wallets, cellphones, and video game controllers. Nineteen of those 22 cases included video footage from body-worn cameras.

In the body-camera footage, we discerned two intents: one in which suspects moved to attack or to appear threatening to the officers, and one in which suspects moved to comply with officer commands or appeared non-threatening to the officers. When suspects moved to attack or to appear threatening to the officers, they held (i.e., aimed) the weapon—or object in their hand—at shoulder or hip height. When suspects moved to comply with officer commands or appeared non-threatening to the officers, they held their hands up above their heads or out at their sides. We therefore created four experimental conditions to represent the four combinations of intent (i.e., threatening, non-threatening) and final hand position (i.e., high, low). This resulted in two types of threatening draws (i.e., object aimed at the camera from the actor’s shoulder or hip) and two types of non-threatening draws (i.e., actor in “surrender” position with both hands raised above their head or hands held out to the sides at hip height). When crossed with four objects (i.e., weapons: handgun, knife; non-weapons: cell phone, wallet), this yielded a total of 16 experimental conditions.

We initially recruited an experienced law-enforcement firearms instructor to perform the 16 motion/object combinations during stimuli creation. However, when reviewing the videos from their pilot filming session, it was clear that—despite the instructor drawing the objects from concealed locations, just as a suspect would—their movement was stylized and corresponded closely with how police officers are trained to draw and present their unconcealed handguns from external holsters. As such, the instructor’s movements did not reflect the suspects’ movement observed in the case file footage. Therefore, we recruited a new actor with no prior firearms or law-enforcement experience to perform the 16 motion/object combinations used for the study. This resulted in movements that corresponded closely with those observed in the case file footage.

Stimuli Filming

The right-handed, male actor was instructed to stand in a neutral position with his feet about shoulder width apart, knees slightly bent, and arms relaxed at his side. When conducting a threatening draw, the actor reached behind his back with both hands to remove the object and then rapidly brought his right arm around to the front of his body with the object in his right hand while his left arm returned to his side to help him maintain balance. When conducting a non-threatening draw, the actor reached behind his back with both hands to remove the object and then brought both arms to the same end position, mirroring one another.

The actor’s motion was recorded using a 12-camera Cortex Motion Analysis (2013, version 5.0) motion-capture system in a completely enclosed room. A 500 mm wand was used to calibrate the system; the system measured the average length of the rod to be 499.98 mm (SD = 0.60 mm). Prior to starting the motion capture, the actor signed a consent form, a media release form, and was given a safety briefing regarding the use of simulated weapons. Note that the replica handgun used was inert (i.e., not capable of live-firing ammunition) and the replica rubber knife had a flexible, dulled blade (i.e., not capable of causing bodily harm). Pictures of the objects are available on the project’s Open Science Framework (OSF) page (https://osf.io/uw5jk/?view_only=e762e22bf0c642768d82de4f635514ae). The actor wore an OptiTrack Motion Capture Suit Classic (figure on OSF). Thirty-one reflective markers were placed on bony landmarks at joints on the actor’s body (figure on OSF). The motion-capture system recorded the x, y, and z coordinates of each marker at a rate of 120 Hz.

The actor used only one object at a time and each object was concealed on his body as follows: (a) the cellphone was tucked into a pocket centered on the actor’s right gluteal muscle; (b) the wallet was inserted into a pocket centered on the actor’s right gluteal muscle; (c) the knife was tucked into the waistband of the motion capture bodysuit, centered on the actor’s sacrum; and (d) the gun was tucked into the waistband of the motion capture bodysuit, centered on the actor’s sacrum. The actor began each trial with his hands by his side. To draw the object (i.e., remove it from its concealed location), the actor reached behind his back to the concealment location with both hands, removed the object with his right hand, and either pointed it at a standing, life-size anatomical model placed 4 m in front of him or held the object in a surrender stance. Threatening draws ended with the actor pointing the object at either shoulder height or hip height, with their left hand at their side. Non-threatening draws ended with the actor putting both hands either straight up in the air or held out to the sides at hip height. See Fig. 1 for conditional end positions. The actor practiced drawing each object from the concealed location in both a threatening and non-threatening manner a minimum of 10 times prior to recording the motion capture.

Fig. 1
figure 1

Sample screen captures from point-light display stimuli. Note. The images display the front of the actor; the actor held the object—a cellphone in all cases here—in their right hand (indicated with a circle in e–h). Images a–d represent what participants actually saw during the study. Images e–h have had lines added to better display the human figure that can be difficult to discern without the figure being in motion. Each of the four motions displayed is as follows: a and e object drawn with threatening intent pointed at shoulder height; b and f object drawn with threatening intent pointed at hip height; c and g object drawn with non-threatening intent with hands at shoulder height; and d and h object drawn with non-threatening intent with hands at hip height

Full size image

For each trial, the actor stood in the center of the motion-capture stage with one of the objects concealed on his person and was instructed on which type of draw motion to perform (i.e., threatening or non-threatening; ending hand location). The actor faced a projector screen on a wall beyond the motion-capture stage; the screen was used to display a visual “start” signal. This was accomplished via slide transitions in a Microsoft PowerPoint presentation. When the actor indicated he was ready to start, a blank white slide was projected; this instantly transitioned to a blank, black slide after a period that varied between 1 and 3 s. This transition was the signal for the actor to start the draw motion. After finishing the draw motion, the actor held himself in the end position for at least 2 s before being told he could return to the start position. Fifty trials of each object/intent/end-position combination (see Fig. 1 for description of all conditions) were recorded in blocks, resulting in a total of 800 trials. Each trial was saved as a separate coordinates file by the Cortex software.

Stimuli Editing

After motion-capture recording was complete, the raw motion-capture output was converted into video stimuli using MATLAB (Version 7.10.0[R2010a]). See the OSF project page for the MATLAB code. The spatial location of the recorded markers located in pairs at each of the elbows, shoulders, wrists, hips, knees, and ankles was averaged to create a skeleton template that more accurately captured the motion at the joint—rather than on the surface of the body—and prevented markers from disappearing if they were ever blocked from the cameras. The final point-light-display videos displayed a white point-light figure with 13 visible points of light against a black background (see Fig. 1a–d). MATLAB was also used to obtain a data file containing the total video time (i.e., the duration of the original motion capture data), the total movement time (i.e., the duration of the draw motion), the movement start time (i.e., the file time that the right elbow moved a distance greater than 1 cm), and the movement end time (i.e., the file time that the right elbow motion ended). This temporal output was used to determine which of the 50 trials in each condition would be included in the study.

Selection of Video Stimuli

From the 50 trials in each of the 16 conditions, 11 videos were selected to be used in the data-collection portion of the study: one video for the training portion and ten videos to be used in the experimental trials. Videos were initially selected based on temporal parameters—as output by MATLAB—and then assessed for acting and visual quality. Using the temporal data, the video with the median movement-duration time for each object and intent condition was selected as the training video. Using the median as a reference, the five next-fastest and five next-slowest videos were selected for the experimental trials. We assessed those 10 videos for their recording quality. Videos were excluded if any marker in the point-light display disappeared at any point. When a video was excluded from amongst the five videos that were faster (or slower) than the median, then the sixth video that was faster (or slower) than the median was assessed as a replacement. See Table 1 for video duration times. Of the selected videos, the average duration for a threatening draw was 0.76 s (SD = 0.10 s) and the average duration for a non-threatening draw was 1.33 s (SD = 0.17 s).

Table 1 Video durations for raw point-light display stimuli by condition and hand position
Full size table

Video Editing

The 176 selected videos (16 practice videos plus 160 experimental videos) were imported into Blender computer graphics software (Community 2018) and edited as follows: (a) a white fixation cross on a black background was added for 0.5 s before the point-light display appears; (b) a static pose of the actor—as a point-light array—was displayed for 0.5 s before the motion start time to afford observers an opportunity to orient to the point-light video; and (c) the video ended two video frames (i.e., 0.067 s) after the motion ended. The videos were produced as.mp4 files (resolution = 1920 × 1080 pixels, frame rate = 30 Hz, video codec = H.264, keyframe interval = 18, output quality = high).

Experimental Task

Participants

Participants were invited to partake in the study via word of mouth, email communication, and a call for participants shared by the Force Science Institute (https://www.forcescience.com/) as well as online forums such as LinkedIn and Twitter. The online study was accessed by 358 individuals, 129 of which completed at least one experimental trial (i.e., 36.0% of those who accessed study) and 51 of whom completed the demographic and post-task survey. Only these observations were included in the reported analyses.

Within the subset of the sample that completed the demographics questions (n = 51), there were 47 male participants, 3 female participants, and 1 who chose not to specify. A majority of the participants (n = 46) were from the United States (1 Arizona, 3 California, 1 Colorado, 1 Georgia, 2 Idaho, 1 Iowa, 4 Kansas, 24 Minnesota, 1 Missouri, 1 New York, 1 Ohio, 1 Oregon, 1 Pennsylvania, 1 South Carolina, 3 Texas), two from Canada, and three from Australia. Participants’ law enforcement experience is reported in Table 2. Officers reported working for a range of agency types (20 Municipal, 12 Sheriff, 5 State, 4 Federal, and 2 Other). All participants reported receiving firearms training as part of their law enforcement experience. Seven participants had previous military experience (3 Army, 3 Marine Corps, 1 Navy; M = 5.0 years, SD = 1.2); only one of these participants served in a combat role.

Table 2 Summary of participants’ law enforcement experience
Full size table

Stimuli Presentation and Response Collection

We considered presenting the stimuli on a large projection screen (e.g., using a police judgement-and decision-making simulator) and collecting responses using a replica firearm. However, had we done this, we would have conflated perception (i.e., is the object a weapon or a non-weapon) with decision-making (e.g., shoot vs. don’t shoot). Furthermore, this would have introduced other factors, such as participants’ proficiency in handling weapons (e.g., speed of drawing from the holster) and their capacity to inhibit responses. As our goal was to focus on participants’ perceptual abilities, we elected to present the stimuli on a computer screen and collect responses (i.e., weapon vs. non-weapon) using a keyboard. Not only does this method align with common perceptual psychology research practices, but it also allowed us to continue with our research given the COVID restrictions on in-person research that were put in place 1 month prior to the start of our planned data-collection period.

The experimental task was presented to the participants using LabVanced Online Experiment software (Finger et al. 2017). See the OSF project page for the link to the online study. Participants were presented with on-screen instructions and were guided through 16 unique practice trials prior to the experimental trials. The practice trials appeared identical to the experimental trials and the stimuli used during them were not included in the experimental trial stimuli. During each trial, participants saw a fixation cross and were instructed to place a finger on the “A” and “L” keys of the keyboard, each indicating either a “weapon” or “non-weapon” response (counterbalanced between participants). Although some researchers have asked participants to make a shoot/don’t-shoot decision (e.g., Correll et al. 2002, 2007), we intentionally asked participants to indicate whether the object being drawn was a weapon or a non-weapon. This type of decision—about whether the motion was genuine (i.e., weapon) or deceptive (i.e., non-weapon)—is consistent with studies of deception in sport (e.g., Cañal-Bruland and Schmidt 2009; Jackson et al. 2006; Sebanz and Shiffrar 2009). When the trial began, participants watched a point-light display video before responding during a 500 ms response window. The response window was emphasized by displaying a green (HEX #00be3c) screen with the text “ANSWER NOW” in large black letters in the center. Keystrokes made prior to the end of the video (i.e., before the green “ANSWER NOW” screen appeared) were not recorded and early responses were not penalized; participants were still required to respond at the completion of the video. If participants did not respond within the 500 ms response window, the green screen changed to an orange screen; this signified that the response was late, but that participants should still respond. Responses were still recorded; however, responding after the 500 ms response window caused participants to experience a 10-s time penalty before they were able to proceed to the next trial. During the practice trials, participants still experienced a time penalty and were instructed to respond faster in subsequent trials. Participants controlled the rate of progress through the experiment by pressing the spacebar to initiate each trial. After every 40 trials, the program displayed a screen encouraging participants to take a break of several minutes. Trial presentation order was randomized by the software for each participant and participants never saw the same clip repeated, unless they responded before the response window. No feedback regarding response accuracy was given at any point during the experiment.

Post-Task Survey

Following the completion of the object-identification task, participants completed a demographics and post-task survey, presented via the same LabVanced Experiment Software (see the OSF project page for the link to the online study with a post-task survey at the end). Participants reported any video playback errors experienced, described the strategies they used in the decision-making process, and rated how important each point in the point-light display was in their decision-making process, using a scale that ranged from 1 (extremely unimportant) to 7 (extremely important). Participants also reported their demographic information, law enforcement experience, and any military training and service.

Procedure

The research was approved by the Institutional Review Board at Wichita State University (IRB: #4521). Participants clicked on a link to access the study on Labvanced.com and completed it on their own computers. After reading an explanation of the study, the officers gave their informed consent to participate. Officers were instructed to turn off their phone and remove it from their reach to minimize possible distractions during the task. They then completed the experimental task and the post-task survey.

Data Analysis Plan

All data analyses were conducted using R statistical computing software (R Core Team 2021). Copies of the data and R script are available on the OSF Project page. Multi-level probit regression (DeCarlo 1998) was used to conduct a signal detection analysis of the data. This analysis produced estimates of officers’ sensitivity and response bias. The analysis also assessed how these estimates changed as a function of the actor’s intent and statistically controlled for dependence across repeated observations from each officer. This analytic approach has greater statistical power (DeCarlo 1998; Wright et al. 2009) than traditional approaches to signal detection analysis in which data are aggregated over repeated trials (e.g., Suss and Raushel 2019). This analytic approach required correct responses (e.g., “weapon” when a gun or knife was drawn) to be coded as 1’s and incorrect responses (e.g., “weapon” when a cell phone or wallet was drawn) to be coded as 0’s. The type of object that the actor held in the video was coded so that weapons were represented by 0.5 and non-weapons were represented by − 0.5. Officer experience was measured in years and was mean-centered (M = 14.3 years) prior to analysis.

Results

Of the 129 people whose data was included in analyses, only 55 (42.6% of included respondents) completed all 160 experimental trials. The analytic approach we took, multi-level modeling, handles missing data by reducing the influence that incomplete observations have on the results, eliminating the need for omission or interpolation of missing values (Finch et al. 2019). Therefore, no data were removed prior to analysis.

Response Time

Most participants responded within the 500-ms response window (M = 267.9 ms, SD = 371.5 ms); however, 1.25% of responses took longer than 500 ms. The inclusion of these trials did not change the pattern of results; therefore, all responses were included in the final analyses.

Response time variance was assessed using a linear mixed-effects model. The model’s fixed effect structure included the main effects of intent and object type. The model also included the intercept random effect to account for individual differences in overall response time between participants. Tukey’s HSD post hoc tests indicated that response time varied significantly between draw intents (B = 0.16, SE = 0.03, z = 5.9, p < .001). On average, officers responded 1.17 ms faster on trials where the draw was threatening, as compared to non-threatening draws. There was no significant difference in response time based on the object the actor was holding (B = 0.02, SE = 0.03, z = 0.81, p = .42).

Generalized Linear Mixed Effects Model

A multi-level probit regression was used to conduct a signal-detection analysis (SDT). This analysis affords the same information as a traditional approach (e.g., Macmillan and Creelman 2004) while also modeling individual differences in bias and sensitivity that occur across participants. Specifically, d′ is captured by the post hoc comparison between levels of the signal present/absent predictor (here, “object”) and c is captured by the model intercept. Interaction and main effect terms represent the influence that key predictors have on sensitivity (d′) and bias (c), respectively. Readers interested in additional information about the approach are directed to DeCarlo’s (1998) thorough description of the analysis.

A multi-level probit regression models individual differences using a random effect structure. Because the preparation was novel, we did not know which aspects of the SDT analysis would be affected by individual differences. Therefore, we created and compared two random effect structures using BICs (see Table 3). The best-fitting model allowed the object predictor, representing d′, and the intercept, representing c, to vary across participants. Once the random effect structure was established, we built a fixed effect structure informed by the research hypotheses. It included the object predictor, which modeled overall d′, as well as the object × intent and object × experience interactions. These additional terms allowed us to determine the impact that intent (sensitivity hypothesis) and expertise (bias hypothesis) had on discrimination ability. Bias was captured by the model intercept, and the main effects of stance and expertise allowed us to assess the impact these variables had on participants’ decision-making bias.

Table 3 BIC comparisons of random effect structures
Full size table

The interaction between Object Type and Intent tested how draw type affects officers’ sensitivity to a weapon (Sensitivity Hypothesis). A Tukey’s HSD post hoc test failed to provide support for this hypothesis (Bdiff = −0.17, SE = 0.10, z = −1.70, p = .09). Participants demonstrated equivalent sensitivity, regardless of whether actors moved in a threatening (d′ = −0.07, 95% CI = [−0.05, 0.25]) or a non-threatening (d′ = 0.10, 95% CI = [−0.24, 0.10]) manner (see Fig. 2).

Fig. 2
figure 2

Officer sensitivity is unaffected by draw intent. Note. Error bars indicate ± 1SE. A d′ value of 0 represents no difference in sensitivity between signal and non-signal conditions. Negative d′ values signify more false alarms than hits and vice versa for positive d′ values

Full size image

The main effect of intent tested whether draw type affected participants’ response bias (bias hypothesis). A Tukey’s HSD post hoc test provided support for this hypothesis (Bdiff = −2.53, SE = 0.07, z = −34.36, p < .001). Figure 3 shows that participants’ bias was significantly more liberal when the actors’ draw was threatening (c = − 1.15, 95% CI = [− 0.98, −1.33]) than when it was conducted in a non-threatening manner (c = 1.38, 95% CI = [1.56, 1.20]).

Fig. 3
figure 3

Officer bias in object identification during non-threatening and threatening draws. Note. Error bars indicate ± 1SE. Positive c values represent a conservative response bias (i.e., the tendency to identify an object as a non-weapon). Negative c values represent a liberal response bias (i.e., the tendency to identify an object as a weapon)

Full size image

Finally, the Object Type × Officer Experience interaction tested the effect of officers’ experience on sensitivity to a weapon (Experience Hypothesis). Figure 4 shows the non-significant interaction, indicating that officers’ ability to discriminate weapons from non-weapons does not change as they gather experience in the field (Bdiff =  −0.12, SE = 0.01, z = 0.07, p = .94).

Fig. 4
figure 4

Officer sensitivity as a function of their years of law enforcement experience. Note. Error ribbon indicates ± 1SE

Full size image

Post-Task Strategy Elicitation

Recall that after completing the experimental task, participants were asked what strategy they used, if any, when making their decision. The post-task survey responses from the 51 participants who completed it serve to supplement our understanding of how the participants treated the stimuli in the study. Content analysis indicated the most common strategies included watching the motion of the hands (N = 20 participants, 39.2%), the speed of the motion—faster motions eliciting a “weapon” response—(N = 13 participants, 25.5%), and whether the motion ended pointing or aiming at the participant (i.e., indicating threatening intent; N = 13 participants, 25.5%). Recall that in Table 1, the average duration of videos with threatening draws was shorter than those with non-threatening draws. The tendency for participants to give importance to the intent/speed of the draw also provides qualitative support for the Bias Hypothesis (i.e., that officers are more liberally biased in their identification of unknown objects when the potential suspect behaves threateningly). Participants were asked to rate each point (e.g., right elbow) on the point-light display in terms of its importance in making their weapon/non-weapon determination; see Fig. 5 and Table 4 for a summary of responses.

Fig. 5
figure 5

Labeled figure of point-light display seen by officers in post-task survey

Full size image
Table 4 Participant self-report of point-light display aspect importance
Full size table

Participants also indicated how frequently they felt they were guessing when giving their response and when they felt most unsure. On average, participants indicated guessing on 31.1% of trials (range: 1–90%, SD = 23.2%). The most common reasons why participants reported feeling more unsure included the motion being too fast, being unable to see the object, ambiguous motions around the hip area for threatening and non-threatening draws, and feeling that the use of point-light displays (cf. regular videos) did not present sufficient information.

Discussion

This study was conducted in two major phases: (1) creating realistic representations of a suspect drawing an unknown object from a concealed location and (2) presenting those stimuli to participants during experimental trials. The experiment used a 2 (Object Type) × 2 (Intent) within-subjects design to address the question: are officers able to use biological motion as a cue when determining whether a potential suspect in front of them is drawing a weapon or a non-weapon?

The Sensitivity Hypothesis predicted that officers would be better able to identify weapons on threatening draws than non-threatening draws. The 95% confidence intervals both included zero and overlapped on the interaction between object type and intent, indicating that the results do not support the hypothesis. This leads us to conclude that officers were not sensitive to the presence of a weapon in both conditions. Furthermore, when graphing this interaction (see Fig. 2), we see that the d′ values for sensitivity are nearly zero for both threatening and non-threatening draw conditions. Not only do these combined findings suggest that officers are equally sensitive to weapons in threatening and non-threatening conditions, but it indicates that officers are generally insensitive to the presence of a weapon overall.

The bias hypothesis predicted that officers would be more liberally biased in their responses when the draw was threatening compared to non-threatening (i.e., officers would be more likely to indicate that the object was a weapon for threatening draws compared to non-threatening draws). This hypothesis was supported at a 95% confidence interval. The graph of this interaction (see Fig. 3) indicates little to no overlap in the c values for bias between threatening and non-threatening draws. In other words, officers were as consistently liberal in their responses to threatening draws as they were conservative in their responses to non-threatening draws.

Despite officers being instructed that they would see both weapons and non-weapons drawn in both threatening and non-threatening manners, many officers reported using qualities of the threatening draws (i.e., the speed of the motion and whether the object ended pointing/aiming at the camera) as indicators that the object was a weapon. Based on the hypotheses, it was not anticipated that participants would use intent as the main decision criteria. However, based on officers’ reports—in the post-task survey—that draw intent was a primary factor used in the object identification process, it is logical that officers were equally non-sensitive to weapons in both threatening and non-threatening conditions and liberally biased for threatening draws while being conservatively biased for non-threatening draws. There were more misses and false alarms than initially anticipated due to half of the threatening draws being weapons and the other half non-weapons. It seems that officers placed greater importance on the suspect’s perceived intent (i.e., how threatening the draw appeared) rather than looking for cues of deception. This is similar to the finding of James et al. (2018) that officers were significantly more likely to escalate situations to a deadly outcome when presented with suspects who had a confrontational demeanor compared to suspects who had a friendly demeanor.

It is known that officers will occasionally encounter suicidal individuals who will behave threateningly towards the officer to elicit a potentially fatal response from the officer (Swaine 2015). Unfortunately, our findings give further support for why this suspect behavior is occasionally effective. These findings indicate that if an individual behaves in a threatening manner—regardless of whether they are armed or not—this would likely lead officers to feel sufficiently threatened that they need to respond with lethal force, resulting in suicide-by-cop situations.

The experience hypothesis predicted that officers with more law enforcement experience would be more sensitive to the object type compared to less experienced officers (i.e., the more experience the officers have, the better they will be at discriminating between weapons and non-weapons being drawn). The finding that performance was not affected by officer experience indicates a lack of support for this hypothesis and calls into question how expertise should be measured for law enforcement officers and if perceptual–cognitive expertise is even present in the identification of unknown objects. Contrary to the findings in sport research (Mann et al. 2007), compared to less experienced officers, more experienced officers did not demonstrate perceptual–cognitive expertise. These findings, or lack thereof, could be due to how we decided to define and measure law enforcement experience; we chose to measure in “number of years spent as a law enforcement officer” when a more appropriate measure may have been based on the quality and content of their law enforcement work. Future research should consider more precise ways of defining the aspect of skilled performance under investigation. For example, it may be helpful to ask participants how many times they have responded to a situation similar to that presented by the study, and how much time they have spent on patrol versus undercover work versus desk duty. Another approach is to recruit skilled participants from units that encounter a relatively high proportion of citizens who carry concealed weapons (e.g., undercover units).

Object Identification Versus a Shoot/Don’t-Shoot Decision

This task was presented to participants as an unknown-object identification task rather than as a shoot/don’t-shoot task that has been used in social psychology literature. For example, in Correll’s studies of the effect of race on decision-making (Correll et al. 2002, 2007), participants were instructed to respond based on the object they saw in the images. If the object was a weapon, participants were instructed to respond “shoot”; if the object was a non-weapon they were told to respond “don’t shoot.” In the current study, we only required participants to judge if the object was a weapon or non-weapon, rather than taking the additional step to decide whether a “shoot” response was justified.

We elected to focus participants on object identification to be more consistent with the literature on deception in sport (e.g., Cañal-Bruland and Schmidt 2009; Jackson et al. 2006; Sebanz and Shiffrar 2009) and to avoid the potential conflict between object type and the likelihood that the officer would choose to shoot the suspect. The case reviews and subsequent interviews with officers indicated that decisions to shoot or use force are based on factors over and above the mere presence of a weapon. For example, some officers may adopt a general mindset that using their firearm is a last resort, regardless of the object the suspect possesses. Officers with this mindset—which is consistent with a conservative response bias—may elect to delay their decision when it is not clear whether a suspect is armed or not, and subsequently take on more risk for themselves, until they can be sure that the suspect is armed or not. Waiting to acquire this information would delay their decision to use force until they can be certain that a suspect is armed and has a threatening intent. Had we framed the task as a shoot/don’t-shoot task, the potential dissociation between an object being identified as a weapon and the decision to respond by shooting could have affected responses. Therefore, we elected to focus participants on identifying the object, rather than on deciding whether a “shoot” response was justified.

A Note on Social Climate

It should be noted that the data were collected between April 1 and June 30 of 2021. Within the previous year, there was an escalation in attention placed on the police by the general public following the deaths of Breonna Taylor in March 2020 and George Floyd Jr. in May 2020. The “Ferguson effect”—also known as the “YouTube effect” and the “viral video effect”—is a term used to describe the “de-policing process” in which officers are distrustful of citizens and more fearful of scandal (Nix and Pickett 2017). Officers perceive media coverage of policing to have a large (primarily negative) effect on civilian attitudes towards police, enough so to impact crime rates (Nix and Pickett 2017). The increase in hostile media has resulted in distrust between the police and civilians that impacts the ways officers interact with citizens. The increased media attention on police in 2020 may have further pushed a social shift and subsequently impacted the way participating officers responded to the stimuli in the study.

Limitations

Most notably, the incomplete nature of the data is this study’s greatest limitation. The restrictions that the COVID-19 pandemic placed on in-person research and the desire to gather data from officers in multiple cities and states caused the study to be moved to a completely remote, online format. Because the researchers could not monitor participants while they participated, there was minimal control over the data-collection process (as is normal for remote online studies). As expected, and for varying reasons (e.g., connectivity issues, participant fatigue, boredom, etc.), not all participants who accessed the online study completed it. Those who abandoned the study during the experimental trials not only left all remaining experimental trials incomplete, but they were never presented with the demographic questions at the end of the study, therefore leaving the demographic questionnaire blank.

All gathered data were included in the analysis, regardless of completion. The models built to include officer experience are based on the 51 participants who completed all 160 experimental trials and the demographic questionnaire but also include the incomplete experimental trials from those with missing demographic information. We were advised by a law enforcement trainer to place the demographic questionnaire at the end of the study (cf. before the experimental trials). The reason for this was so that participants could start engaging in the experimental task sooner and not be confronted immediately with requests for demographic information, which could be seen by officers as intrusive. However, the low completion rate for the experimental portion of our study (like many online studies) makes a compelling argument for placing demographic questions before experimental trials. Although the model used was able to handle the missing demographic data, placing the demographic questionnaire before the experimental trials would have prevented the need to account for missing data (but may have negatively affected completion rates).

Another limitation—one that several participants emailed the researchers about after completing the study—was that the point-light-display stimuli contained very little information. Although we succeeded in capturing the underlying biological motion—which was our goal and is consistent with previous biological-motion research—it could be argued that in real show-of-force and/or use-of-force scenarios, officers see more than just a figure depicted by white dots against a black background. This is an issue of ecological validity. Therefore, it is possible that presenting participants with a figure that better represents a human body (e.g., showing the shape of the arms, torso, and legs), rather than points that may be connected to form a stick figure, could lead to greater object sensitivity as well as increased officer confidence in their responses.

Other limitations to external validity, such as response modality (Roca and Williams 2016), are expressed in sport research as well. The mechanics of pressing a key on a keyboard are an extremely different process than drawing and firing an actual weapon (or withholding a response, which was not measured here). Based on the fact that simply holding a gun results in more liberal response bias (i.e., a tendency to identify an object as a weapon rather than a non-weapon; Witt and Brockmole 2012; Witt et al. 2020), a change in response modality could lead to a shift in participants’ behavior. Future research would benefit from assessing sensitivity and bias differences between officers based on differing response modalities.

Additionally, only one actor was used to create the stimuli. Though the actor was monitored to ensure they behaved consistently between trials, it is possible that the results could have been different with a different actor. We previously noted that Suss and Raushel (2019) and Scott and Suss (2019) used a limited number of actors whose physical characteristics did not generalize well to the population they were depicting. Although we used only one actor to create all the point-light stimuli for the current study, the use of point-light displays—rather than regular videos—removed potentially biasing information that would have been present had we used a limited number of actors to create regular videos (see Wells and Windschitl 1999).

The choice to use a single actor was based on our experience creating similar stimuli for video-based studies (Suss and Raushel 2019; Scott and Suss 2019). In those studies, multiple non-police-trained actors were used to develop the visual stimuli. We learned that training each naïve actor to consistently perform realistic draw motions is an extremely time-consuming process that results in few viable trials from each actor. Therefore, for the current study, we decided to devote our time to recording a single actor with the goal of obtaining a large number of high-quality (i.e., representative) trials. Although our primary concern was with the quality and ecological validity of the actor’s movements, this decision was also influenced by other constraints placed on the stimuli development process (i.e., cost of the equipment used, available lab time with the motion capture software, etc.).

Initially, when filming the pilot videos, a police firearms instructor (i.e., an experienced officer with advanced training who trains law-enforcement officers to handle and shoot firearms) volunteered to be the actor. We thought that the instructor’s familiarity with weapons and the arcana of violence would an advantage in creating realistic stimuli. However, we found that the instructor’s movements did not reflect those seen in the body camera footage of actual suspects previously reviewed and therefore would likely result in stimuli that did not truly reflect suspect behaviors. From a subject-matter expert who observed the filming, we learned that law enforcement officers are trained to handle firearms (e.g., draw from an unconcealed hip holster) using specific (i.e., choreographed) movement patterns that can differ from ways in which suspects—who do not have law-enforcement training—instinctively draw weapons from concealed locations (e.g., tucked into the waistband at the small of back). Despite our substantial efforts and feedback (e.g., playing back recordings of their movement, reviewing case file footage), the instructor was unable to move in ways they were unaccustomed to (i.e., consistent with those seen in the case file footage). In light of this, we opted to try filming the stimuli with an actor who did not have formal law enforcement training. We found that they were better able to mimic the movement patterns observed in the case file footage. Arguably, by giving our naïve actor practice and feedback, we were able to get them to fluidly perform the movement patterns observed in the case file footage. That is not to say that researchers should avoid engaging police officers as actors; the lesson here is that just because a police officer (e.g., a firearms instructor) may have extensive experience with firearms does not necessarily mean they will be able to produce suitable (i.e., representative) movements. It is possible that police officers with substantial undercover experience (i.e., carrying concealed weapons) may be better able to produce representative movement patterns than those without such experience.

Where possible, future research should consider presenting stimuli of multiple actors. When creating stimuli, researchers might consider intentionally varying factors such as the actors’ body proportions (e.g., height and weight) and their familiarity with weapons (i.e., someone who regularly concealed carries a weapon as well as someone with little-to-no familiarity carrying a weapon). This would encompass the experience range that potential suspects may have as well as the speed at which they move. Additionally, having stimuli from multiple actors would enable researchers to investigate whether exposure to numerous trials from one actor provides an advantage when observing the movements of another actor(s).

Future Research

Officer experience was a significant component of the hypotheses. It was assumed that there may be skills that officers learn and develop after leaving the academy (e.g., through patrol experience, speaking to seasoned officers, etc.). Most participants had significant law-enforcement experience (M = 14.02 years, SD = 8.77 years), with few officers-in-training participating. Making a greater effort to include recruit officers in future studies would allow more comparisons to be made between officers who have little-to-no patrol/field experience (i.e., fresh out of the academy) and those who have been on duty for at least 10 years.

Temporal manipulation is another factor that could shed more light on the effect that movement speed has on participant’s decision-making. Due to the effect of draw type, and subsequent draw speed, on officer’s responses, altering the speed of the movements during post-motion capture editing could yield more information regarding whether officers are more focused on how quickly a suspect is moving or how aggressively they are moving. This would mitigate the issue created by the confounding speed difference between the two draw intent conditions. The average speed of threatening draws was 0.76 s compared to 1.33 s for non-threatening draws. By speeding up the non-threatening draws to the threatening draw speed, slowing down the threatening draws to the non-threatening draw speed, or altering both intents to a median duration of 1.05 s, it could be possible to compare the influence of motion speed and draw intent on an officer’s sensitivity and bias to the presence of a weapon. This temporal manipulation may be applied to the current stimuli and could be achieved by adding a command to the current MATLAB code that would re-output the point-light display videos in the appropriate duration.

Similar to temporal occlusion (Scott and Suss 2019), spatial occlusion has been used in sport research to systematically investigate expertise differences in the ability of athletes to use visual information (Mann et al. 2007). Spatial occlusion systematically removes visual information from the display to further assess the importance and relevance of each aspect of the visual display. For example, as in Fig. 5, the right shoulder (i.e., point 2) could be removed from the point-light figure. This spatial occlusion would allow us to compare participant performance when missing a self-reportedly vital aspect of the figure to previous performance when the entire point-light display was visible. Hodges et al. (2005) found that participants could use partial body images (e.g., only a foot or only a leg) as effectively as full-body display to recognize a kicking motion in soccer. This effect could be tested with point-light displays by occluding some points using a spatial-occlusion paradigm. This would aid in identifying which parts of the body have the greatest influence on an officer’s response and could mimic cases in which the entirety of the suspect is not visible to the officer.

An eventual application of this work is to develop law-enforcement training that improves officers’ anticipation ability. One training approach that was found to be effective in sport presented cricket players with temporally-occluded point-light displays of bowlers’ motions (Brenton et al. 2019). After responding (i.e., played a batting stroke), players viewed an unoccluded version of the bowling motion and received feedback about the type of bowl. Critically, one group was also required to mimic the bowling motion they just observed by actually bowling a ball. That “motor practice” group performed better on a transfer test than the group that did not perform the motor practice. This training approach, which was based on common-coding theory (Prinz 1997), addressed the fact that not all cricket players are bowlers—and therefore, those who are not bowlers do not possess motor experience of the very bowling motions they are trying to anticipate when batting. Law-enforcement officers face a similar situation: their firearms training typically involves drawing an unconcealed handgun from a hip holster without any deceptive intent. However, the suspects they face are likely to draw an object—either a weapon or a non-weapon—from a concealed location and possibly with deceptive intent. Given that law-enforcement officers’ training does not typically require them to practice drawing weapons and non-weapons from concealed locations with both deceptive and non-deceptive intent, they are not afforded the opportunity to develop motor experience of the very motions they may need to anticipate. Therefore, training that provides the opportunity to develop such motor expertise, together with practice anticipating deceptive and non-deceptive motions, may improve their anticipation ability.

Conclusion

To the best of our knowledge, this is the first study that attempts to fully isolate the biological motion aspect of unknown-object identification. It was found that officers heavily weight the perceived intent of the suspect’s behavior when determining the threat potentially posed by an unknown object in the suspect’s possession. This information could be used to inform future use-of-force training procedure development for law enforcement officers. Because officers can’t necessarily avoid all ambiguous situations, further research regarding ways to disambiguate use-of-force situations could result in greater safety for the officers as well as the community.