Error Management and Enemy Detection Biases

Viewing novel faces entails automatic encoding of communicative signals and social categories [9], [10]. Because these signals and categories may be differentially associated with the inclination and potential to do harm, error management strategies that minimize one's exposure to threats are likely be relatively automatic as well. Three such signals of threat seem especially likely to signal danger. First, angry facial expressions signal an enhanced likelihood of aggression and can be difficult to ignore [11]. The safer error to make is to incorrectly presume that an angry person intends harm than to incorrectly assume the person is friendly; it is less costly to mistakenly avoid a benign individual than to mistakenly approach a dangerous one. Second, because ingroup members facilitate a wide range of functionally critical goals for one another (e.g., [12]) and intergroup interactions have historically been associated with relatively higher levels of danger [13], [14], biases should, and do, exist to generally see ingroup members more positively (e.g., [15]) and outgroup members negatively, especially when threat-related goals are temporarily active [16], [17]. Third, male faces should also be particularly likely to incur such a bias because human males, relative to females, are more likely to hurt others [18], [7].

A number of studies support the often combinatorial role these three factors play in threat detection processes. An emerging literature on intergroup processing suggests that cues to threat enhance processing of outgroup members (from groups stereotypically associated with harm). These cues include angry expressions and masculinity as well as threatening concepts and objects [16], [19], [20]. For example, Correll and colleagues have shown that, when placed in situations that require discriminating between targets that possess or do not possess weapons, people are more likely to “shoot” unarmed Black males than unarmed White males [21]. Even professional police officers exhibit such biases [22]. Outgroup members can themselves also enhance processing of threat-related stimuli. For instance, viewing Black male faces leads perceivers to more rapidly identify images of guns, even when these images are strongly degraded [19], [23]. Such findings fit within an error management framework. However, existing research has tended to presume underlying motivation, or measure it at the level of individual differences [24], but not directly manipulate threat-relevant goals. We might expect that because primed goals can influence automatic evaluations [25], in the context of self-protection or revenge goals, such biases may be further amplified.

Accuracy

Whereas error management theory generates clear predictions about biases, it is mute regarding accuracy. Recent research reveals, however, that the activation of important goals can enhance the accuracy of cognitive processing. For instance, thinking about survival enhances performance on a memory task [26], and perceivers focused specifically on self-protection more efficiently encode outgroup male faces—they gain additional memory “bang” for their attentional “buck” [6]. We might thus also anticipate perceivers concerned with self-protection to exhibit enhanced accuracy in the detection and use of cues identifying individuals as enemies and friends.

In contrast, a functional perspective does not predict that revenge should increase accuracy. The purpose of revenge is retaliatory and retributive, and this can often be accomplished with broadly biased responses against whole groups rather than focused responses to the specific individual who elicited the desire for revenge. Furthermore, the anger experienced when we desire revenge may actually work in opposition to self-protective motivations, as anger has been shown to suppress fear in mildly dangerous situations [27]. Revenge thus serves as a nice counterpoint to self-protection, because while both may introduce biases that promote ingroup/outgroup discriminations, they should produce quite different effects when it comes to accuracy.

Standard methods of signal detection analysis can be used to estimate both bias and sensitivity, or signal discriminability. Importantly, these effects can be independent of one another (i.e., we can become more sensitive to a discrimination but continue to show a response bias as well). In line with the above reasoning, we hypothesized (1) that activating a goal of self-protection should place the cognitive system on alert, increasing sensitivity relative to revenge motivations (the self-protective vigilance hypothesis) and (2) that there would also be enhanced anti-male, anti-angry, and anti-outgroup biases in both motivation conditions (the intergroup bias hypothesis).

Overview of the Present Experiment

We asked participants to make decisions concerning the “friend” or “enemy” status of briefly presented target individuals who were identified with perfectly diagnostic enemy/friend insignia but who differed in terms of perfectly nondiagnostic ingroup/outgroup status, facial expression, and sex. Prior to the decision task, participants visualized scenarios designed to activate one of two motivational systems especially relevant to the lives of soldiers in combat—self-protection and revenge-mindedness.

Ethics Statement

This research was approved by the IRB at Arizona State University, and all participants read and signed statements of informed consent.

Participants

249 (86 women) students who identified themselves as not being of Arabic or Middle Eastern background participated in exchange for course credit.

Materials

Stimuli consisted of black and white photographs of faces, sized at 2×1 1/3 in., that varied across the following dimensions: Ethnicity (Arab or Western European, indicated by the presence or absence of Arabic headdress), Sex (Male or Female), and Expression (slightly Angry or Neutral). Eight faces of each of the eight types were constructed, though any individual participant was randomly assigned to see only half of these faces. Expressions were created by digitally manipulating neutrally expressive faces to lower the inner brow, tighten the mouth, and flare the nostrils. A separate group of 24 participants identified these faces as angry with 100% accuracy, and no one reported suspicion that the expressions were not naturally posed.

The task-relevant stimuli were the letters “E” and “F”, constructed from pieces of the face stimuli so the facial photos could serve as post-stimulus masks. The resulting stimuli were approximately 1/3 in. square. We should emphasize here that the signal that was being detected was separate and distinct from the images of the faces. It was an “E” or an “F” that was temporally sandwiched between two displays of the face image, so qualities of the face images, like the headgear on Arabic targets or the greater variability of the hairstyles for the Western targets, should have little effect on the signal's discriminability.

Procedure

Participants were informed that they were participating in a simulation of circumstances faced by U.S. soldiers in Iraq and given a simple task: identify whether faces are “enemies” or “friends” based on briefly presented insignias. Participants were told that equal numbers of enemies (and friends) were male and female, angry versus neutrally expressive, and Arab versus Western. These characteristics were explicitly descried as non-diagnostic of enemy status, and great efforts were taken to ensure that participants understood this.

Motivational states were activated by having participants listen to one of three narrated, guided visualization passages: (1) self-protection (participants imagined being ambushed while in a hostile foreign city), (2) revenge-mindedness (participants imagined a friend had been ambushed by a foreign combatant, after which the combatant becomes vulnerable to retribution), or (3) a control scenario (participants imagined visiting a foreign marketplace, which featured the same foreign peoples but involved no threat). Participants were asked to try to experience the emotions described in the scenario, because these emotions “should facilitate performance on the subsequent tasks.”

Following this manipulation, participants engaged in three blocks (32 trials each) of a decision task. In each trial, participants were asked to focus on a fixation point (‘+’) appearing in the center of the screen for 1000 ms. A face then appeared (for 500 ms in block one, 200 ms in blocks two and three), followed by the letter ‘E’ or ‘F’ (for 50 ms in block one, 30 ms in block two, and 10 ms in block three), which appeared in four different places within the contours of the faces to prevent participants from focusing on a single location. Finally, the original face reappeared again as a mask, giving participants 1.5 seconds to identify the face as enemy or friend based on the letter that had appeared by pressing the ‘A’ key or the ‘5’ key on the numeric keypad, labeled with an ‘E’ or ‘F’ accordingly. Each face's status as an enemy or friend was counterbalanced, but remained constant across the three blocks for any given participant.

Bias

At the most general level, there was an overall bias to identify faces as “friends” (grand mean = −0.07, F_1,56 = 9.56, p = .003, η²_p = .145). Moreover, although participants were explicitly told that ethnicity, facial expression, and sex were not diagnostic of enemy status, all three cues elicited significant biases. Relative to their counterparts, angry faces were more likely to be called “enemy”, F_1,56 = 74.8, p<.001, η²_p = .578, and female faces were more likely to be called “friend”, F_1,56 = 5.56, p = .022, η²_p = .090, as were Western faces, F_1,56 = 34.3, p<.001, η²_p = .380.

More importantly, there was an interaction of motivation condition, target race, and target expression, F_2,112 = 4.29, p = .016, η²_p = .071, supporting the intergroup bias hypothesis (see Figure 1b). Whereas control participants were biased to identify angry faces as enemies regardless of whether they were Arabic or Western European (main effect within control: F_1,31 = 7.95, p = .01; no interaction with race), the bias against angry Westerners vanished in both the self-protection and revenge-mindedness conditions (both Fs<1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Figure 1a shows bias as a function of target gender and ethnicity, collapsed across facial expression.

While there is a bias to see Arabic men as “enemies”—and this bias becomes more pronounced in both the self-protection and revenge conditions—there is an even greater bias to call ingroup members “friends”, which entails missing more ingroup enemies. Figure 1b depicts bias for faces showing slight anger (collapsed across target gender), and clearly shows that self-protection and revenge wipe out the bias to call angry ingroup members “enemies”: In the control condition there is a strong bias to call any angry face an enemy, but both self-protection and revenge conditions completely eliminate this bias for ingroup faces. Figure 1c shows participant accuracy in discriminating between enemies and friends.

https://doi.org/10.1371/journal.pone.0023929.g001

Signal discriminability

Analyses of d′ within each motivation condition revealed two main effects. First, participants were significantly better at discriminating the insignias of Western faces as opposed to Arabic faces (F_1,56 = 4.72, p = .034, η²_p = .078). More importantly, supporting the self-protection vigilance hypothesis, a main effect of motivation (F_2,112 = 3.95, p = .022, η²_p = .066) revealed that self-protection concern enhanced accuracy whereas revenge-mindedness decreased it, relative to control (see Figure 1c). In contrast to revenge-mindedness, self-protection motivation enhanced participants' accuracy in detecting enemies while minimizing the false alarms in which they mistakenly labeled friends as enemies, F_1,56 = 4.72, p = .034, η²_p = .078.

Conclusion

We found that people were generally inclined to perceive angry, outgroup, and male faces as enemies. A self-protection motivation biased perceivers to view angry outgroup individuals as enemies and ingroup members as friends, but actually increased identification accuracy. Revenge motivation increased the tendency to label outgroup individuals as enemies as well, but it generally decreased the overall accuracy of judgments. We can therefore conclude that self-protection appears to sharpen the senses at the same time that it biases decisions, while revenge-mindedness inclines the perceiver to be biased against outgroup members and in favor of ingroup members, but it does so at the cost of accuracy. We can find little precedent for this effect in the scientific literature, but the idea that revenge dulls decision making is quite consistent with anecdotal accounts of retributive violence in combat situations as well as in counter-insurgency operations.

These findings underscore the utility of investigating how threat-detection can be altered by motivating circumstances. Merely having participants imagine themselves in dangerous situations led to functionally adaptive alterations in signal detection that could have profound consequences under more engaging circumstances in the field. Although these effects may generally reflect traditionally functionally adaptive responses, they may also be quite problematic from the perspective of policy-makers and employers. However, the methods employed here could identify individuals capable of highly accurate decisions under pressure as well as those most prone to dangerous biases, thereby positively informing selection and training for signal detection tasks that occur in emotionally-charged environments. More broadly, the present results suggest that fundamental motivational systems adaptively modulate the detection of social signals, and research programs exploring social perception in this way may lay the groundwork for a more developed model of human information processing in real world circumstances.