Page 2 of 3
Performing under pressure: Gaze control, decision making and shooting performance
of elite and rookie police officers
2.3. The Scenario
The scripted scenario was designed to simulate situations common in law enforcement where officers are faced with a rapidly unfolding event in which critical cues are visible for only a short period of time and where they have to make a decision under extreme time pressure to fire their handgun or inhibit the shot. The scenario was about one minute in length and consisted of an entrance phase of seven seconds, a middle phase of about 45 s and a final attack phase of seven seconds. The assailant entered from a side door and approached a receptionist who was seated at a desk 7 m from the officer who stood on duty guarding the entrance to a government office (see Fig. 1). Upon reaching the desk he turned his back to the officer and complained to the receptionist that he had been unjustly jailed for the past three days. He requested a meeting with an official so that he could get his passport back. The receptionist was polite, but not helpful, leading the man to become increasingly agitated. During the final seven seconds the scenario escalated with the assailant slamming his hands on the table and angrily raising his voice. Suddenly, with two seconds remaining he pulled a handgun (or cell phone) from under his coat with his right hand, executed a rapid reverse pivot and shot at the officer, or brandished the cell phone in such a way he appeared to shoot. There was no hesitation in his movements as he aimed and fired from a distance of about 5 m. The scenario was designed so there was a very brief window of time at the very beginning and at the end when critical cues were available that could be fixated only at that time. The assailant’s coat was open during the first three seconds as he entered the room and it was only during the final two seconds that the gun or cell phone was visible prior to firing. These points of time were therefore critically important and in many respects simulated conditions found on the street when events unfold rapidly and critical information is available for only a brief period of time. Four and five seconds were added, respectively, after each of these critical moments to also allow for an analysis of fixations while the assailant stood with the weapon hidden, resulting in an equal seven seconds at the beginning and seven second at the end of the scenario that was analyzed. Prior to the study, both the assailant and receptionist worked with an acting coach who trained them to maintain the same timing and mannerisms throughout. The man who played the assailant was a police lieutenant who had extensive handgun and role playing experience. The receptionist worked in law enforcement and had taken part previously in scenarios for research purposes.
The officer and assailant were fitted with protective gear and a standard Glock handgun loaded with one Simunition shell by a safety officer. All had used a similar weapon in training. The eye tracker was fitted and calibrated to a nine-point grid, followed by a second calibration to locations on the receptionist’s desk. Before the first trial the assailant was located out of sight in a side room. Prior to beginning, each officer was read the following instructions in order to ensure the same information was delivered:
‘‘Your task today is to provide security for a government entrance. Behind you are the doors you must guard. We have intelligence information that an armed encounter will happen on your location today. You are to handle the threat using only your handgun. Recognizing the first shot is the most important, you are limited to only one shot – your magazine is loaded with one round. You may not step outside this designated area; you may not move forward of the red cones’’.
The two conditions were then performed consisting of five gun trials and two cell trials. Trial one was always a gun trial, while the cell trials were randomly assigned from trials two to seven. A recorder kept track of shot accuracy (hit, miss) and the location of shots (head, upper body/arms, lower body/legs, missed). The data as shown in Fig. 1 were monitored throughout using a firewire connection thus ensuring accuracy of the calibration. When necessary re-calibration was carried out by having the officer fixate objects on the desk. Data collection took approximately 45 min.
2.5. Officer and assailant motor phases
Motor phases were coded for the first seven seconds and final seven seconds using the Quiet Eye Solutions v2 (QES v2) software, a software package that synchronizes the gaze, motor and audio data. During the first seven seconds, two officer phases were identified (prepare, unholster). The prepare phase occurred from the first frame as the assailant entered the room until the officer’s hand moved to his or her gun. The unholster phase continued from the hand on the gun until 7000 ms had elapsed. During the final seven seconds, four officer phases were determined (assess, draw, hold, aim/fire) and three assailant phases (confront, pivot, aim/fire). The officer’s assess phase began 7000 ms before the end of the trial until the first frame showing the officer’s gun being drawn upward from the holster; the draw occurred from the upward movement of the hand to maximum flexion of the elbow; the hold phase occurred from maximum flexion of the elbow until the frame showing extension of the gun to- ward the assailant; the aim/fire phase occurred from the frame showing extension of the gun arm to- ward the assailant until the shot was fired (audio of shot), or during the cell trials when the gun slowed and was maximally extended. The assailant’s confront phase began 7000 ms before the end of the trial and lasted until the frame showing the first movement of the shooting hand/arm upward to draw the gun or cell phone from under the coat; the pivot phase occurred from the first movement of the shooting arm/elbow until the gun or cell was first visible; the phase occurred from the gun/cell first being visible until the a shot was fired, or the hand with the cell phone was maximally extended. The trigger pull was estimated from previous studies as occurring a constant 100 ms prior to the shots being recorded by the audio (Bumgarner, Lewinski, Hudson, & Sapp, 2006; Tobin & Fackler, 1997).
2.6. Coding officer fixations
The officer’s fixations were coded during the first seven seconds and during the final seven seconds using the QES v2 software. Since the goal during the first seven seconds was to determine how soon the E and R fixated potential weapon locations and for how long, the fixations were ordered sequentially from first to last in the trial using a sort function of QES v2. During the final seven seconds fixations were sorted in reverse order (last to first). By ordering the fixations from last to first it was possible to determine the sequence and duration of locations fixated prior to the shot and their importance in performance. This procedure has been used previously in studies where a culminating action occurred at the end of the trial, such as stepping over a barrier (Patla & Vickers, 1997), stopping a shot in ice hockey goaltending (Panchuk & Vickers, 2009), or firing a gun (Vickers & Williams, 2007). Fixations and saccades were coded using definitions from previous studies (Vickers, 1996, 2007; Vickers & Williams, 2007). A fixation occurred when the gaze was held stable on a location for a minimum of 100 ms within two degrees of visual angle (the width of the cursor in Fig. 1). Four fixation locations were identified: assailant weapon/cell, assailant non-weapon, officer weapon and off the assailant. Weapon/cell sub-locations included the assailant’s gun, hand/elbow, arm, inside coat, chest, right pocket, belt). Non-weapon sub-locations included the assailant’s non-gun hand/arm, back, head, face, shoulders, legs, non-gun pocket. Fixations on the officer’s gun included the Glock pistol and shooting hand. Off the assailant occurred when a fixation was more than two degrees off the assailant’s body, or on the receptionist, or any other location in the room. Saccades were coded when the eyes moved rap- idly from one fixated location to another with a minimum duration of 66.66 ms (2 frames). Code-re- code reliability was established using two independent coders. Intra-class correlations in excess of .90 were established for the all phases and quiet eye duration using procedures from Thomas and Nelson (2001).
2.7. Statistical analysis
Shooting accuracy (%) was determined in the gun trials using a one way (expertise, E, R) factorial ANOVA. Percent of officers who successfully inhibited the shots during the cell condition was analyzed using a one way (expertise, E, R) factorial ANOVA. Overall shooting performance took into account measures of shooting accuracy, shot speed and decision making and was analyzed using Chi square procedures. Officer and assailant motor phase durations and onsets were analyzed using an Expertise x Phase x Condition ANOVA, with repeated measures on the last two factors. Percent of fixations during the first seven seconds was determined using an Expertise (E, R) x Fix Sequence (Fix 1– 10) x Location (assailant weapon/cell, assailant non-weapon, off the assailant) ANOVA, with repeated measures on the last two factors. During the final seven seconds, percent fixations was analyzed using an Expertise (E, R) x Fix Sequence (Fix 1–6) x Location (assailant weapon/cell, assailant non-weapon, off-assailant, officer weapon) ANOVA, with repeated measures on the last two factors. Fixation durations were analyzed using an Expertise (E, R) x Fixation Sequence (1–6) and Performance (high, low) x Fixation Sequence (1–6) ANOVA, with repeated measures on the last factor. QE duration was determined using an Expertise (E, R) x Location (assailant weapon/cell, officer weapon) factorial AN- OVA. Greenhouse-Geisser epsilon was used to control for violations of sphericity in the repeated measures designs and adjusted p-values are reported where necessary. The effect sizes were calculated using partial eta squared (g2p ). Significance level was set at p < .05 for all tests.
3.1. Shooting accuracy
A significant difference was found due to expertise, F(1, 22) = 8.23, p < .009, g2p = .27. The E hit the assailant on 74.54% ± 5.44 of shots compared to 53.85% ± 4.74 for the R.
3.2. Decision making
Significant difference were found due to expertise, F(1, 22) = 5.22, p < .03, g2p = .14. During the cell condition, 18.18% ± 12.19% of the E officers (2 of 11) fired at the assailant compared to 61.54% ± 14.04% of the R (8 of 13).
3.3. Firing speed
The E fired before the assailant on 92.50% of trials compared to 42.22% for the R. They were also faster than the assailant by an average of 179.05 ms (36.84 ± 6.06) and the R were slower than the E by an average of 13.26 ms. Four shots ended in a draw (2 E; 2 R). The E hit the assailant in the upper torso (62.07%), the arms/hands (31.03%) and legs (6.90%). Respective percentages for the R were upper torso (48.39%), arms/hands (35.48%), legs (6.45%) and head (9.67%).
3.4. Overall performance
Overall performance (high, low) was determined by combining measures of shooting accuracy, shot speed and decision making. High performance trials occurred when all three of the following criteria were met per trial: on gun trials the shot was accurate; on gun trials accurate shots occurred be- fore the assailant’s shot, and on cell trials the shot was inhibited. Low performance trials occurred when the shot missed, the shot occurred after the assailant’s shot and on cell trials a shot was fired. Failing to draw during a gun trial was a low performance characteristic (4 R trials). The E group re- corded a significantly more high performance trials than the R, X2 (1, N = 255) = 6.63, p < .0009, / = .51. On 75.00% of trials the E officers met the criteria of high performance compared to 52.86% of trials for the R.
3.5. Percent fixations during first seven seconds
During the first seven seconds three trials were analyzed per officer (the first hit, the first miss, and the first cell trial). No significant differences were found in fixation frequency or duration due to condition, but significant differences were found in location due to expertise, F(1, 270) = 9.19, p < .008, g2p = .38, location, F(2, 270) = 25.14, p < .0001, g2p = .63, and the interaction of Expertise x Location, F(2, 270) = 5.60, p < .009, g2p = .16. Fig. 2 shows that the E fixated more locations where a weapon could be concealed (M = 50.29% ± 3.53) than the R (M = 30.62% ± 2.89), while the R looked more to non- weapon locations (M = 51.13 ± 3.35; E M = 42.11 ± 3.53), or off the assailant (M = 18.07, SE ± 2.43; M + 7.59 ± 2.26). Frequency distributions were determined for the sub-locations. The E looked inside the assailant’s coat more than the rookies during the brief time this was possible (E=9.44; R = 3.18). E and R were similar in percent fixations to the assailant’s face (E = 11.165; R = 13.43), which was visible only as the assailant entered the room. They were also similar in the percent of fixations on the receptionist (E = 5.15%, R = 5.65%).
3.6. Officer’s phase duration (ms) during first seven seconds
Of interest was how soon the officer’s moved their hand to unholster their gun as this signaled the officer’s awareness of an escalating situation. Significant differences were found for phase, F(1, 50) = 4.46, p < .04, g2p = .08, and the interaction of Expertise x Phase, F(1, 50) = 49.41, p < .0001, g2p = .50. The E officers unholstered within 1774.75 ms ± 3056 ms of the assailant entering the room, while the R did not unholster until 6275 ms ± 1529 ms had elapsed.
3.7. Assailant’s phase durations during final seven seconds
Since all officers faced the same assailant, it was important to determine if his movement durations were similar against the E and R. No significant differences were found due to expertise, p > .05, or the interactions of Expertise x Condition, p > .05 or Expertise x Condition x Phase, p > .05, but significant differences were found for phase, F(1, 122) = 43.78, p < .0001, g2p = .26, and the interaction of expertise by phase, F(1, 122) = 13.76, p < .0003, g2p = .10. Table 1 shows that against the E, the assailant’s pivot and aim/fire phase durations were, respectively, M = 995.21 ms ± 459.21 ms and M = 579.81 ms ± 100.93 ms. Corresponding values against the rookies were M = 919.56 ms ± 483.13 ms and M = 887.16 ms ± 313.89 ms. This meant that the assailant’s pivot phase was similar against E and R, but he slowed his aim/fire phase against the R. Review of the video data showed this occurred on trials when the R were having great difficulty and the assailant slowed his shot to give them more time to aim and fire.
3.8. Officer’s phase durations and onsets during final seven seconds
No significant differences were found in phase durations due to condition p > .05, or the interactions of Expertise x Condition p > .05, or Expertise x Condition x Phase p > .05, but significant differences were found for expertise, F(1, 244) = 10.42, p < .002, g2p = .08, phase, F(2, 244) = 6.88, p < .001, g2p = .08, and the interactions of expertise by phase, F(2, 244) = 11.09, p < .0001, g2p = .01. Table 2 shows that the E draw, hold and aim/fire durations were M = 180.34 ms ± 88.79 ms, Table 1 1016.02 ms ±
1976.83 ms and 611.43 ms ± 418.64 ms, respectively. Corresponding means for the R were 226.70 ms ± 179.62 ms, 96.67 ms ± 632.88 ms and 567.63 ms ± 287.49 ms. Contrast of means was significant for the hold phase only, F(1) = 39.61, p < .0001, the R having shorter durations.
Significant differences were found in the officer’s phase onsets due to expertise, F(1, 372) = 19.31, p < .0001, g2p =.13, phase, F(3, 372) = 63.79, p < .0001, g2p =.34 and the interaction of expertise by phase, F(1, 372) = 15.186, p < .0001, g2p =.11. The R were later drawing, holding, aiming and firing than the E (see Fig. 3). Onsets (with SE) for the E were M = 4628.45 ms ± 333.68 ms, 4808.89 ms ± 342.19 ms, 5826.75 ms ± 217.58 ms and 6866.61 ms ± 16.03 ms, respectively. Corresponding means for the R were 6037.61 ms ± 97.95 ms, 6263.33 ms ± 99.19 ms, 6360.47 ms ± 40.09 ms and 6928.13 ms ± 15.79 ms.
3.9. Fixation analysis: Final seven seconds
A total of 15 fixations were found per officer during the final seven seconds. Onset, offset and duration were analyzed separately using an Expertise (E, R) Condition (gun, cell) factorial ANOVA and no significant differences were found until the final six fixations which occurred during the final two seconds when the events shown in Fig. 1 (A-C) occurred. Image A shows the assailant’s raised elbow which signaled the beginning of the attack, image B the moment the gun or cell first became visible and image C the moment the shots were fired by the officer and assailant.
Percent of fixations during the final two seconds differed due to expertise, F(1, 240) = 5.00, p < .04, g2p = .24, location, F(3, 240) = 54.58, p < .0001, g2p = .77, and interactions of Location x Fixation Sequence, F(15, 240) = 18.87, p < .0001, g2p = 54. and Location x Expertise x Fixation Sequence, F(15, 240) = 3.811, p < .0001, g2p = .19. Each fixation location was then analyzed separately in order to isolate expertise differences by location during the final sequence of six fixations.
Percent fixations on the assailant’s weapon/cell (Fig. 4A) differed due to expertise, F(1, 96) = 9.91, p < .002, g2p = .09, fix sequence, F(5, 96) = 5.93, p < .0001, g2p = .24, and the interaction of Expertise Fix Sequence, F(5, 96) = 3.56, p < .005, g2p = .16. Across the six fixations, the E increased fixations to the assailants weapon or cell from 18% to 71%, compared to 18% to 34% for the R. Contrast of means indicated E and R differed during fixation five, F(1) = 9.03, p < .003, and fixation six, F(1) = 13.52, p < .0004.
Fixations on non-weapon locations (Fig. 4B) differed due to fix sequence, F(5, 96) = 24.45, p < .0001, g2p = .56, and the interaction of Expertise x Fix Sequence, F(5, 96) = 2.66, p < .03, g2p = .12. The E decreased fixations to non-weapon locations from 78% to 7% compared to 62% to 16% for the R. Means comparisons showed the E and R differed during fixation one, F(1) = 3.82, p < .05, and fixation five, F(1) = 6.50, p < .01.
Fixations on the officer’s own weapon (Fig. 4C) differed due to fix sequence, F(5, 96) = 28.23, p < .0001, g2p = .60, and the interaction of Expertise Fix Sequence, F(5, 96) = 3.66, p < .005, g2p =.16. The R increased the percent of fixations to their own weapon to 39% on fixation 6 compared to 20% for the E. Means contrast indicated E and R differed during fixation six, F(1) = 19.11, p < .0001.
Fixations off the assailant (Fig. 4D) differed due to expertise, F(1, 96) = 11.83, p < .0009, g2p =.11. During all six fixations more R fixations (M = 13.21 ± 14.09) were off target more than the E (M = 4.76 ± 10.35).
3.10. Expertise and performance differences in fixation duration
The final six fixation durations were analyzed and significant differences found for fixation sequence, F(5, 715) = 3.15, p < .008, g2p = .02; the effect for expertise neared significance, F(1, 715) = 3.29, p < .069, g2p = .01. Fig. 5 shows the E maintained a mean duration of 275–350 ms across fixations, while the rookies took longer during fixations one and two (475–425 ms) and then had shorter durations during fixations five and six (250 ms). Contrast of means showed E and R differed in fixation duration on the officer’s weapon, F(1) = 20.38, p < .0004.
The analysis of performance (high, low) fix sequence (fix 1–6) found significant differences due to fixation sequence, F(5, 715) = 4.48, p < .0005, g2p = .03, and the interaction of Fixation Sequence x Performance, F(5, 715) = 2.42, p < .03, g2p = 02. Fixations one and two occurred as the assailant drew and executed the pivot, fixations four and five occurred as the gun or cell first became visible and fixation 6 as the shots were fired or inhibited. Fig. 6 shows that during the low performance trials, fixations one and two were longer and were followed by shorter fixations four and five, while the opposite occurred during the high performance trials. The results show that, irrespective of expertise, if too much time was taken to fixate locations as the assailant began his pivot (f1 and f2) this was then followed by very
brief fixations on the gun or cell when these first became visible. Contrast of means was significant for fixation one, F(1) = 6.03, p < .01, and fixation five, F(1) = 4.69, p < .03.
3.11. Percent of final saccades by location
Interconnecting fixations five and six was a final saccade. Significant differences were found for location, F(2, 34) = 9.86, p < .0004, g2p =.37, and the interaction of Expertise x Location, F(2, 34) = 11.78, p < .0001, g2p =.41 (see Fig. 7). E percent saccades on the assailant weapon, non-weapon and, officer’s own weapon were, respectively, M = 44.07% ± 9.90%, 23.89% ± 8.8% and 32.04% ± 11.24%,. Corresponding percentages for the R were 4.00% ± 2.66%, 12.00% ± 6.80% and 84.00% ± 6.50%. Overall, 84% of the R groups final saccades were to their own weapon compared to 32% for the E.