The Survival Processing Advantage of Face: The Memorization of the (Un)Trustworthy Face Contributes More to Survival Adaptation

Abstract

Researchers have found that compared with other existing conditions (e.g., pleasantness), information relevant to survival produced a higher rate of retrieval; this effect is known as the survival processing advantage (SPA). Previous experiments have examined that the advantage of memory can be extended to some different types of visual pictorial material, such as pictures and short video clips, but there were some arguments for whether face stimulus could be seen as a boundary condition of SPA. The current work explores whether there is a mnemonic advantage to different trustworthiness of face for human adaptation. In two experiments, we manipulated the facial trustworthiness (untrustworthy, neutral, and trustworthy), which is believed to provide information regarding survival decisions. Participants were asked to predict their avoidance or approach response tendency, when encountering strangers (represented by three classified faces of trustworthiness) in a survival scenario and the control scenario. The final surprise memory tests revealed that it was better to recognize both the trustworthy faces and untrustworthy faces, when the task was related to survival. Experiment 1 demonstrated the existence of a SPA in the bipolarity of facial untrustworthiness and trustworthiness. In Experiment 2, we replicated the SPA of trustworthy and untrustworthy face recognitions using a matched design, where we found this kind of memory benefits only in recognition tasks but not in source memory tasks. These results extend the generality of SPAs to face domain.

Evolution theorists believe that compared with other mechanisms, the human memory system should process information related to survival or reproduction better than other information because of its benefit to evolutionary fitness (Buss, 2014; McBride, Thomas, & Zimmerman, 2013). Nairne and colleagues, who were the forerunners in conducting experiments, found that compared with other existing conditions (e.g., pleasantness), the coding information related to survival can significantly increase the retrieval rate of pertinent information. This effect is known as “survival processing advantage” (SPA, Nairne, Pandeirada, & Thompson, 2008; Nairne, Thompson, & Pandeirada, 2007).

The Argument for the Face as a Boundary Condition

Since then, experiments on SPA have identified in different types of visual pictorial material, such as pictures and short video clips (Fernandes, Pandeirada, Soares, & Nairne, 2017; Otgaar, Smeets, & van Bergen, 2010). However, the advantage of memory cannot be extended to face stimulus. Savine, Scullin, and Roediger (2011) employed either computer-generated faces (Experiment 1) or photographs of real faces (Experiments 2–5) as stimuli; unexpectedly, the results suggested that the memory of faces in the survival condition did not show significant enhancement compared with nonsurvival conditions. As a consequence, some scholars consider that the face is a boundary condition of SPA (Gelin, Bonin, Méot, & Bugaiska, 2017; Maguinness & Newell, 2014; Nairne, Cogdill, & Lehman, 2017).

However, other scholars claim that face processing offers no SPA is questionable (Kazanas & Altarriba, 2015; Sandry, Trafimow, Marks, & Rice, 2013). Theorists propose that some stimuli are likely privileged in our cognitive system because of their fitness relevance (Wilson, Darling, & Sykes, 2011), facial evaluation and recognition have great significance in environmental adaptation and fitness (Maguinness & Newell, 2014; Oosterhof & Todorov, 2008). In particular, accurately distinguishing whether a person is a potential source of danger is extremely crucial for social behaviors regarding whether to avoid or approach this person (Young, Slepian, & Sacco, 2015). Evaluation based on facial trustworthiness is considered as an adaptive phenomenon (Engell, Haxby, & Todorov, 2007). Moreover, people not only evaluate the reliability of cues by using facial characteristics (Oosterhof & Todorov, 2008) but also encode the information into memory (Klapper, Dotsch, van Rooij, & Wigboldus, 2016).

The Role of Facial Trustworthiness in SPA

By retrieving the previous literatures, we speculate that the face as a SPA boundary condition may largely lie in facial trustworthiness. In Savine et al.’s (2011) study, they used neutral faces as stimuli and did not manipulate the strength of facial trustworthiness. Although adopting a neutral stimulus is a tradition in SPA research that can be traced to the initial studies conducted by Nairne and colleagues (2007), which can be beneficial to explain the universality and stability of survival processing. However, it fails to generalize all types of stimuli.

On the one hand, within the framework of the ultimate mechanism, regarding biological mechanisms, compared with certain types of stimuli (e.g., threats from snakes), humans cannot receive benefits from neutral stimuli when confronting natural pressures (Wilson et al., 2011); neutral faces likely fail to provide sufficient information regarding survival decisions (Pandeirada, Fernandes, Vasconcelos, & Nairne, 2017). In contrast, considering that facial trustworthiness provides implicit information about trust (Oosterhof & Todorov, 2009), bipolar facial trustworthiness can provide obvious information regarding survival decisions. One seems trustworthy or untrustworthy, which affects our responses in terms of approach or avoidance behavior (Otgaar et al., 2010). More specifically, a trustworthy face may indicate a likelihood of cooperation and social affinity, whereas an untrustworthy face may imply danger and a need for avoidance (Slepian, Young, Rule, Weisbuch, & Ambady, 2012; Suzuki & Suga, 2010).

On the other hand, within the framework of the proximate mechanism, the reduction of adaption-related information is likely associated with the lack of attention to keep this type of information (Nairne & Pandeirada, 2016). Sandry et al. (2013) suggest that the reason why our memory system can flexibly adapt to different survival conditions is that the mnemonic sensitivity of the processed information related to fitness is hierarchical. Certain environment can possibly activate people’s selective attention on information about adapting stimuli and lead to different attention preferences. Since facial trustworthiness can provide implicit information about trust (Oosterhof & Todorov, 2009), bipolar facial trustworthiness (e.g., more trustworthy/untrustworthy faces) is more related to the survival scenarios than neutral faces. Neutral faces cannot provide clearly defined information about trustworthiness, which can result in creating more cognitive confusion and involuntary attention and processing among participants (Coren & Russell, 1992). Obviously, bipolar facial trustworthiness is more likely to capture people’s attention, and neutral faces are relatively less likely to attract people’s attention. Indeed, previous researches have shown that under survival-related pressures, the accuracy of discriminating between the faces of friends and foes increases (Xin, Yang, & Liu, 2017). Under salient conditions, the encoded information of trustworthiness and untrustworthiness were both identified more accurately than under spontaneous conditions, and “the cheaters” and “the cooperators” were encoded into memory better than the neutral faces (Felisberti & Pavey, 2010). Overall, the evidence from previous studies supports the idea that face could have a SPA.

The Current Experiments

We expected to explore a possible survival memory benefit for stimuli of direct relevance to trustworthiness. To accomplish this end, several changes were made to the typical survival processing paradigm in current experiments. In a survival experiment, participants are asked to imagine themselves stranded in the grasslands of a foreign land, without any basic survival materials. Participants are told that over the next few months, they will need to find steady supplies of food and water and to protect themselves from predators (Nairne et al., 2007), or they are in charge of contributing meat to feed their tribe (Nairne, Pandeirada, Gregory, & Van Arsdall, 2009). Next, randomly selected stimuli (e.g., pictures) are presented, and participants are asked to rate the relevance of each stimulus to the imagined scenario. In a later surprise memory test, participants typically recognized the pictures in fitness-relevant scenario better than those in matching encoding scenario that was no fitness-relevant (e.g., moving to a foreign land).

First, we manipulate the parameters of trustworthiness with regard to the targeted faces; specifically, we employ neutral expression male faces with different levels of trustworthiness as stimuli (trustworthy faces, neutral faces, and untrustworthy faces). Evolutionary psychologists generally believed that the majority of humans’ cognitive “sculpting” occurred during the Pleistocene era, during which the human species lived largely as foragers and the division of labor typically found in hunter-gatherer societies—men hunt and women gather. Men generally outperformed women on tasks thought to be related to hunting skills (for an overview, see Nairne et al., 2009). Consequently, men have more capability to help our ancestors escape from predator and even injure them. Memory is thought to be geared toward retaining information related to specific adaptive problems faced in hunting-and-gathering environments. Thus, the more the male faces are likely relevant to survival, the more mnemonic sensitivity there is.

Second, we replace the task design for assessing the relevance of stimuli to a survival scenario with judging strangers’ intentions in a survival scenario. By drawing on the experimental design of Bell, Buchner, Erdfelder, et al.’s (2012) face research, we use a 7-point scale, where −3 = avoid and +3 = approach, to reflect participants’ expectancy of others’ intentions. Previous researches on mnemonic advantages of human life show that for ancestral hunter-gatherers who immersed in a rich biotic environment, both animals and human beings are species who deserve careful treatments (Nairne, Vanarsdall, & Cogdill, 2017; Silva, Medeiros, & Albuquerque, 2017). Human beings had to judge whether stranger a friend or a potential foe was when they met each other. Therefore, it seems that it is more appropriate to judge others’ intentions rather than relevance.

Finally, the most important change was that we applied statistical indicators appropriate to the categorical tasks in the results analysis. Previous studies primarily used the hit rate as the dependent variable, but it cannot produce separate accurate recognition ratios for different categories, making it inappropriate for a categorical study. We used the “unbiased hit rate” of Hu as an alternative method (Wagner, 1993), just as Suzuki and Suga’s (2010) face memory study did. By multiplying the probability of detection by the frequency of hits, we addressed the potential influence of intentionality within decision-making and the bias of memory task results. This improvement of the simple hit rate method was better for categorical tasks than the percentage of correct responses (Chaby, Hupont, Avril, Luhernedu, & Chetouani, 2017).

In current experiments, specifically, participants in Experiment 1 were asked to predict their own response tendency, when encountering strangers (represented by three classified faces of trustworthiness) either to ensure their survival or to move to a new home; in Experiment 2, participants were asked to assess their response tendency to select strangers as memberships, either to guard the tribe’s survival or to win a hunting game. The final surprise memory test focused on identifying whether the faces have been encountered before (recognition in Experiments 1, 2) and in which context (source memory in Experiment 2).

As just reviewed, trustworthy and untrustworthy faces can provide significant information about adaptation. In the present case, it was predicted that (un)trustworthy faces in survival condition could provide a mnemonic advantage compared with those in control condition.

In addition, scholars believe that the survival processing is initiated by the combination of item-specific and relational coding, which may induce a SPA to occur in source memory besides SPA occurs in recognition. The term source memory refers to the ability to remember the context in which information was learned (Bell, Buchner, Erdfelder, et al., 2012). Research indicates that survival processing likely leads to solid memories in a survival scenario (Fernandes et al., 2017; Kroneisen & Bell, 2018; Pandeirada et al., 2017), and source memory and face memorization are independent of each other (Bell, Buchner, Erdfelder, et al., 2012; Kroneisen & Bell, 2018). To examine whether facial memorization in survival scenario is more accurately identified than that in hunting contest scenario, we added a test of source memory in Experiment 2.

Experiment 1

In Experiment 1, participants viewed stranger faces with different levels of trustworthiness, and a face could be untrustworthy, neutral, or trustworthy. Their task was to rate how they would react to strangers (using a 7-point scale).

Processing condition was manipulated within subject. Participants randomly received instructions stating that judging the intentions of others was vital to survival in the grassland of a foreign land or judging other’s intention was necessary to move to a new home in a foreign land. After a series of rating trials, and after a dictator test, participants performed an old/new recognition task for the faces. Finally, participants were asked to evaluate facial trustworthiness for all faces (using a 7-point scale).

Facial trustworthiness could provide implicit information about trust and such judgments were made rapidly (Oosterhof & Todorov, 2008, 2009). These heuristic clues from faces are crucial for survival (Young et al., 2015). If memory can solve survival or adaptive problems, the faces that provide the obvious clues of trustworthiness should be left more memory traces that make retrieval more probable. We expect to observe a SPA for (un)trustworthy face.

Method

Participants

Sixty-four Chinese undergraduate students (M_age = 19.34 years, SD = 0.895, 26 female; age range 18–22) were recruited, as most of these participants were nonpsychology major. They participated in exchange for partial credit toward a course requirement. We conducted a power analysis using G*Power (Version 3.1.9.2) and found the sample size (N = 64) had sufficient power (1 − β = .80) to detect a medium effect size (f = 0.38) at a significance level of α = .05. All participants had normal visual acuity or corrected visual acuity (>1.0). To the best of our knowledge, participants had not previously participated in face recognition experiments and they were tested individually. The stimuli were presented, and the responses were collected by computer.

In our study, written informed consent was given prior to participation. This study was approved by the ethics committee of Northeast Normal University.

Materials and Design

Ninety-six male faces selected from Hou’s trustworthiness database (see Supplemental Appendix; Hou, 2017) were used as experimental stimuli, including 32 faces each for the trustworthy, neutral, and untrustworthy stimuli. Of the 96 faces, 48 were used as learning stimuli. Of these, 24 faces (8 trustworthy, 8 neutral, and 8 untrustworthy faces) were presented in the survival scenario, and the remaining 24 faces (8 trustworthy, 8 neutral, and 8 untrustworthy faces) were presented separately in the control scenario. The other 48 of the 96 faces were used as distractors in the recognition test.

The experiment employed a 2 (rating scenario: survival vs. moving) × 3 (facial trustworthiness: untrustworthy vs. neutral vs. trustworthy) within-subjects design. The dependent variable was the Hu value, which was calculated by multiplying the raw hit rate by the positive predictive value.

Procedure

The experiment comprised of three stages: learning stage, distraction stage, and recognition stage.

Learning stage

Upon arrival, informed written consent was obtained from each participant. Then the participants were informed that they would rate each face based on the stranger’s intentions in the following two scenarios:

• The survival scenario read the following instructions: “In this task, we would like you to imagine that you are stranded in grassland of a foreign land, without any basic survival materials. Over the next few months, you’ll need to find steady supplies of good and water and protect yourself from predators. However, you can hardly survive on your own. Imagine that you encounter another individual in the grassland.”

• Moving scenario read the following instructions: “In this task, we would like you to imagine that you are planning to move to a new home in a foreign land. Over the next few months, you’ll need to find and purchase a new home and transport all your belongings. However, you can hardly achieve them on your own. Imagine that you encounter another individual in the foreign land.”

In both scenarios, the instructions continued as follows: “We are going to show you a set of faces of whom the person may be. You are required to consider other’s intention in this situation and predict your own response. You can choose to avoid or approach to this person—it’s all up to you to decide.”

In this stage, the participant was asked to rate 24 faces in a block, including 8 trustworthy faces, 8 neutral faces, and 8 untrustworthy faces. After a fixation cross appeared for 500 ms, a face was presented in the center of the screen for 5 s, and the participants were asked to evaluate their behavioral response to this person on a 7-point scale, on which −3 = “avoid” and +3 = “approach.” The rating scale appeared under each face. Next, a black screen appeared for 500 ms. After finishing all of the facial judgments for this condition, the participants rated another set of 24 faces in the other scenario (e.g., moving). The order of scenarios (survival vs. moving) was counterbalanced by the program script. In addition, the order of faces in each condition was randomized for each participant. No mention was made of the subsequent recognition test.

Distraction stage

After finishing the two rating tasks, the participants completed a 2-min “plus–minus” task as a distraction task.

Recognition stage

Then, the participants performed an old/new recognition test in which they were shown a set of faces and asked to decide whether they had seen the faces earlier; subsequently, they were asked to evaluate the trustworthiness of these faces. The results of the two operations were combined to obtain an unbiased hit rate calculation for late-stage data analysis. Each recognition task trial consisted of a fixation cross (300 ms) followed by a random face presented in the center of the screen without a time limit until a response was provided by pressing a key indicating “old” or “new” (1 or 2 on the keyboard, respectively). After this interface, the same face was presented in the center of the screen once again and remained until the participants pressed a number key to judge its facial trustworthiness (on a scale from 1 to 7), after which the screen turned black (300 ms).

The recognition decisions and trustworthiness evaluations of faces were self-paced; 48 of the 96 faces had been seen previously, and the other 48 were new and served as distractors. All the faces were displayed in a random order. The entire experiment lasted approximately 30 min. The specific experimental process is shown in Figure 1.

Figure 1.

Graphical representation of Experiment 1 procedure.

Results and Discussion

Manipulation Check

First, we analyzed the rating differences of facial trustworthiness during the recognition test to examine the manipulation of the stimuli traits. The results of a repeated measures multivariate analysis of variance (MANOVA) suggested that the main effect of facial trustworthiness was significant, F(2, 126) = 40.80, p < .001, η² = .393); the main effect of the rating scenario and the interactions between the rating scenario and facial trustworthiness were both not significant (Fs < 1).

However, MANOVA designs have decreased power as the number of tests increases and may eventually become ineffective (von Ende, 1993), the Bonferroni correction by Holm (1979) conducted the pairwise comparisons for significant MANOVA relationships. The results of Bonferroni correction indicated that the trustworthy faces were rated the highest (M = 3.65, SD = 0.111), significantly higher than neutral faces (p < .01) and untrustworthy faces (p < .001) at a significance level of α = .017; meanwhile, the neutral faces were rated in the middle (M = 3.46, SD = 0.114) but significantly higher than the untrustworthy faces (p < .001) at a significance level of α = .017. The untrustworthy faces were rated the lowest (M = 3.07, SD = 0.118). The results indicated that the manipulation of facial trustworthiness based on the classification generated in the pilot study was valid for the current sample.

Evaluation Results

Second, we investigated the response tendency to facial trustworthiness types under the two conditions during the learning stage. The results of the repeated measures MANOVA indicated that only the main effect of facial trustworthiness was significant, F(2, 126) = 95.23, p < .001, η² = .602, the main effect of the rating scenario was not significant (F < 1), and the interaction effect was not statistically significant, F(2, 126) = 2.77, p > .05. The results of Bonferroni correction suggested that the degree of approaching trustworthy faces was the highest (M = 3.84, SD = 0.122), significantly higher than the degree of approaching neutral faces (M = 3.40, SD = 0.109) or untrustworthy faces (M = 2.66, SD = 0.111, ps < .001) at a significance level of α = .017. The degree of approaching neutral faces was in the middle, significantly higher than the assessment level of untrustworthy faces (p < .001) at a significance level of α = .017. In general, the two polarities of facial trustworthiness can induce different responses: The participants tended to approach trustworthy faces and avoid untrustworthy faces. This finding was relatively consistent with previous studies on facial trustworthiness (Engell et al., 2007; Todorov & Oosterhof, 2011).

The scores of a scenario rating during the rating phase were also not associated with later recognition, survival, r(64) = −.015, ns; control, r(64) = −.089, ns. Furthermore, the three assessment levels of trustworthiness in the various scenarios also showed no significant correlation with the facial recognition scores (ps > .05).

This result is consistent with the conclusion that no significant relationship exists between rating scores and recognition scores found in previous survival processing studies (Nairne et al., 2017; Nairne & Pandeirada, 2016; Röer, Bell, & Buchner, 2013; Savine et al., 2011).

Facial Recognition Results

In light of the “old” decision in the new/old decision task during the recognition stage, we further divided the facial trustworthiness rating results into three categories based on the evaluation results: trustworthy (more than 4), neutral (4), and untrustworthy (less than 4). The original participants’ responses across decision types are presented in Table 1.

Table 1.

Raw Response Rates in the Recognition Test (M ± SD).

Face Stimuli		Rating Scenarios	Old Classification Faces			New Classification Faces
			Trustworthy	Neutral	Untrustworthy
Old faces	Trustworthy	Survival	1.16 (1.472)	1.47 (1.221)	1.48 (1.718)	3.89 (1.861)
		Moving	0.91 (1.165)	1.45 (1.321)	1.58 (1.762)	4.06 (1.542)
	Neutral	Survival	0.92 (1.313)	1.39 (1.410)	2.09 (1.966)	3.59 (1.571)
		Moving	1.00 (1.309)	1.20 (1.224)	1.73 (1.525)	4.06 (1.582)
	Untrustworthy	Survival	0.69 (1.367)	0.89 (1.492)	2.58 (1.688)	3.84 (1.482)
		Moving	0.66 (1.185)	0.78 (1.015)	2.17 (1.609)	4.39 (1.590)
New faces		Survival	1.43 (2.432)	1.54 (1.794)	3.02 (2.992)	35.58 (5.673)
		Moving	1.46 (2.176)	2.11 (2.017)	3.24 (3.191)

In addition, we computed the unbiased hit rate according to Wagner’s (1993) formula for the facial recognition tasks in the two scenarios, as indicated in Table 2.

Table 2.

Mean Unbiased Hit Rates of Facial Trustworthiness in Two Rating Scenarios (M ± SD).

Rating Scenarios	Old Classification Faces			New Classification Faces
	Trustworthy	Neutral	Untrustworthy
Survival	.06 (.090)	.07 (.077)	.12 (.093)	.65 (.088)
Moving	.04 (.062)	.05 (.063)	.09 (.069)	.62 (.074)

We examined the difference in Hu’s unbiased hit rate across the various levels of facial trustworthiness under the two rating scenario. To reduce the chance of Type I errors, the degrees of freedom for all repeated measures MANOVAs were adjusted using the Greenhouse and Geisser method (Greenhouse & Geisser, 1959).

The results showed that the main effect of the rating scenario condition was significant, ∊ = 1.000, F(1, 63) = 12.51, p < .01, η² = .166. The accurate recognition ratio of survival scenario (M_survival = 0.08, SD = 0.006) was significantly higher than control scenario (M_moving = 0.06, SD = 0.005, p < .05), the SPA of face was observed. The main effect of facial trustworthiness was significant, ∊ = 0.975, F(1.95, 122.85) = 13.14, p < .001, η² = .173. Except for these results, there was no significant interaction, F < 1.

The current research focused on whether such SPA could be observed on each classified faces of trustworthiness. We further divided the data into three groups according to facial trustworthiness conditions and conducted respectively three one-way repeated measures analysis of variance of the recognition scores for rating scenarios. The results showed that for trustworthy faces, the accuracy of the recognition tasks within the survival scenario (M_{survival_trust} = 0.06, SD = 0.090) was significantly higher than in the moving scenario, M_{moving_trust} = 0.04, SD = 0.062; ∊ = 1.000, F(1, 63) = 5.47, p < .05, η² = .080. It indicated that there was a significant SPA of trustworthy faces. Similarly, the SPA was also found with regard to untrustworthy faces, the accuracy of the recognition tasks within the survival scenario (M_{survival_untrust} = 0.12, SD = 0.093) was significantly higher than in the moving scenario, M_{moving_untrust} = 0.09, SD = 0.069; ∊ = 1.000, F(1, 63) = 5.78, p < .05, η² = .084. However, the SPA was not found in neutral faces: The difference in the accuracy of recognition tasks between the survival and control conditions was not significant, M_{survival_neutral} = 0.07, SD = 0.077; M_{moving_neutral} = 0.05, SD = 0.063; ∊ = 1.000, F(1, 63) = 1.85, p > .05). These results were consistent with those found in previous studies on survival processing (Savine et al., 2011).

In addition, we were primarily interested in the accuracy of facial recognition based on trustworthiness within the survival scenario condition. The results showed that the difference in each condition was significant, ∊ = 0.950, F(1.90, 119.67) = 8.37, p < .001, η² = .117. The results of Bonferroni correction indicated that the recognition of untrustworthy faces was the most accurate and was significantly more accurate than the unbiased hit rate of trustworthy (p = .005) and of neutral faces (p = .001) at a significance level of α = .017. These results suggest that a significant negative bias exists in survival contexts, which is consistent with previous studies showing that negative and threatening information is easier to recognize and remember than neutral or positive information (Kensinger, 2007).

Experiment 2

Although Experiment 1 showed that the face could serve to judge strangers’ intentions and produce a SPA, it may fail to match arousal, novelty, and prior media exposure across the survival and control conditions. To reveal the underlying mechanisms in the SPA, Kazanas and Altarriba (2015) recommended that studies adopt matched scenario designs. The matched-scenario design is widely used in survival processing studies, usually including survival hunting and hunting contests, survival gathering and gathering contests, and other scenarios (Nairne et al., 2009; Savine et al., 2011). Thus, our Experiment 2 was designed differently from Experiment 1. The matched scenario design required that participants act in the same way; the difference was that one scenario was relevant to evolutionary fitness, whereas the other scenario was not. In addition, to examine whether a SPA regard to survival scenario, Experiment 2 also added source memory test in final surprise memory test.

Method

Participants

Sixty-two Chinese undergraduate students (M_age = 19.71 years, SD = 1.030, 20 female; age range 18–22) were enrolled in this study, as most of these participants were nonpsychology major. They enrolled in exchange for partial credit toward a course requirement. This sample size allowed us to detect a medium effect size (f = 0.40) by using G*Power (Version 3.1.9.2; Faul et al., 2007), assuming a power of .80 and a α level of .05. All participants had normal visual acuity or corrected visual acuity (>1.0) .To the best of our knowledge, participants had not previously participated in face recognition experiments and they were tested individually. The stimuli were presented, and the responses were collected by computer.

In our study, written informed consent was given prior to participation. This study was approved by the ethics committee of Northeast Normal University.

Materials and Design

Except for the processing scenarios and additional source memory test, which were once again manipulated between participants, the design was identical to Experiment 1.

Procedure

After completing the informed written consent, participants randomly received one of two rate scenarios: survival hunting or hunting contest.

• The instructions for the survival hunting scenario read as follows: “In this task, please imagine that you are living long ago in the foreign land. You are in charge of picking a small group that will be in charge of contributing meat to feed your tribe. You and your group will need to hunt big game, trap small animals, or even fish in a nearby lake or river. Hunters often have to travel great distances, pursue animals through unfamiliar terrain, and successfully return home. Whatever the conditions are, you and your group must hunt successfully to feed your tribe. We are going to show you a set of faces of whom the person may be. You are required to consider other’s intention in this situation and predict your own response because it is highly important to your success when hunting prey. You can choose to avoid or approach to this person—it’s all up to you to decide.”

• The hunting contest scenario instructions read as follows: “In this task, please imagine that you have been invited to participate in a hunting contest. You are in charge of picking a small group that will be in charge of contributing captured game to the team effort. You and your group will need to hunt big game, trap small animals, or even fish in a nearby lake or river. Members of the team often have to travel great distances, pursue animals through unfamiliar terrain, and successfully return to the contest center. Whatever the conditions are, you and your group must hunt successfully to help your team win the contest. We are going to show you a set of faces of whom the person may be. You are required to consider other’s intention in this situation and predict your own response because it is highly important to your success when hunting prey. You can choose to avoid or approach to this person—it’s all up to you to decide.”

The rating trials and dictator test mimicked those of Experiment 1.

However, during the memory test, participants were instructed to perform both a recognition test and a source memory test. For each trial, after a fixation cross appeared for 300 ms, then, a random face appeared on the screen with no time limit. The participants were asked to judge whether they had seen the face before by pressing a key for “old” or “new” (1 or 2, respectively). The participants used a 7-point rating scale to rate each face in terms of trustworthiness, also with no time limit. If an “old” keyboard response was provided in the new-old task, then a source memory test interface would appear after the trustworthiness judgment task and require the participant to press “F” or “J” to select the condition (survival hunting or hunting contest, respectively) in which the face had previously appeared. This interface would not disappear until the participant responded. If the participant gave a “new” response in the new-old task, then the source memory test interface would not appear. A black screen appeared for 300 ms after the participant completed all of the tasks. The experimental procedure is shown in Figure 2.

Figure 2.

Graphical representation of Experiment 2 procedure.

Results and Discussion

Evaluation Results

We examined the participants’ tendencies to respond during the learning stage. The repeated measures MANOVA results showed that only the main effect of facial trustworthiness was significant, F(2, 122) = 58.84, p < .001, η² = .491, and the results of a Bonferroni correction indicated that the evaluation of faces as trustworthy was the highest (M = 3.63, SD = 0.110), significantly higher than the evaluation of faces as neutral (M = 3.26, SD = 0.102) or untrustworthy (M = 2.59, SD = 0.120; ps < 0.001) at a significance level of α = .017. The evaluation of neutral faces was significantly higher than that of untrustworthy faces (p < .001) at a significance level of α = .017. This finding shows that the participants tended to approach trustworthy faces and avoid untrustworthy faces. The main effect of the rating scenario and the interaction effect were not significant (Fs < 1).

The score of the scenario rating during the rating phase was also not associated with later recognition, survival, r(62) = .008, ns; control, r(62) = .037, ns Furthermore, the three assessment levels of trustworthiness in the various scenarios also showed no significant correlation with facial recognition scores (ps > .05).

The evaluation results of Experiment 2 are consistent with those of Experiment 1, indicating that we should not attribute the difference in the recognition test to the deep processing of information and that other explanations must be considered.

Facial Recognition Results

We examined the difference in Hu with regard to faces varying in trustworthiness under two rating scenario and the raw response rates of the participants in the recognition test, as shown in Table 3.

Table 3.

Raw Response Rates in the Recognition Test (M ± SD).

Face Stimuli		Rating Scenarios	Old Classification Faces			New Classification Faces
			Trustworthy	Neutral	Untrustworthy
Old faces	Trustworthy	Survival hunting	1.26 (1.546)	1.39 (1.407)	1.87 (1.815)	3.48 (1.627)
		Hunting contest	1.00 (1.414)	1.31 (1.362)	1.66 (1.557)	4.03 (1.515)
	Neutral	Survival hunting	0.86 (1.353)	1.24 (1.399)	2.16 (1.748)	3.74 (1.679)
		Hunting contest	0.79 (1.230)	1.18 (1.208)	1.89 (1.538)	4.15 (1.608)
	Untrustworthy	Survival hunting	0.50 (1.083)	0.87 (1.180)	2.77 (1.787)	3.86 (1.782)
		Hunting contest	0.36 (1.073)	0.98 (1.443)	2.08 (1.571)	4.58 (1.825)
New faces		Survival hunting	1.16 (2.107)	1.75 (2.079)	2.85 (2.863)	33.61 (6.781)
		Hunting contest	1.56 (2.240)	2.56 (2.981)	4.74 (4.163)

The results showed that the main effect of the rating scenario was significant, ∊ = 1.000, F(1, 61) = 22.12, p < .001, η² = .267, indicating a SPA (M_survival = 0.09, SD = 0.008; M_hunting = 0.05, SD = 0.005, p < .001), therefore, it replicated the findings of Experiment 1. The main effect of facial trustworthiness was significant, ∊ = .992, F(1.98, 121.01) = 10.47, p < .001, η² = .146. There was no significant interaction, ∊ = .916, F(1.83, 111.72) = 2.48, p > .05, η² = .039.

Just as the previous Experiment 1 did, we further examined whether such SPA could be observed on each classified faces of trustworthiness by dividing the data into three groups. The results showed that for trustworthy faces, the accuracy of the recognition tasks within the survival hunting scenario (M_{survival_trust} = 0.08, SD = 0.105) was significantly higher than in the hunting contest scenario, M_{hunting_trust} = 0.05, SD = 0.072; ∊ = 1.000, F(1, 61) = 6.32, p < .05, η² = .094, demonstrating a SPA of trustworthy faces. Regarding untrustworthy faces, the recognition accurate recognition ratio in the survival scenario (M_{survival_untrust} = 0.13, SD = 0.109) was significantly higher than in the contest scenario, M_{hunting_untrust} = 0.07, SD = 0.075, ∊ = 1.000, F(1, 61) = 13.04, p < .01, η² = .176, a SPA of untrustworthy faces was also detected. These results were consistent with our prediction, which suggested that faces with a bipolarity of trustworthiness (untrustworthiness and trustworthiness) can help people fit their environment. However, there was no SPA found in neutral faces, M_{survival_neutral} = 0.06, SD = 0.074; M_{hunting_neutral} = 0.04, SD = 0.050; ∊ = 1.000, F(1, 61) = 2.5, p > .05.

In addition, we also explored the difference in the accuracy of the recognition of facial trustworthiness in the survival hunting scenario. Significant differences were found among the three types of faces varying in trustworthiness, ∊ = .996, F(1.99, 121.54) = 9.47, p < .001, η² = .134. The results of Bonferroni correction indicated that the memorization of untrustworthy faces was the best and was significantly better than the unbiased hit rate of the memorization of trustworthy (p = .010) or neutral faces (p = .001) at a significance level of α = .017. No significant differences were found between trustworthy and neutral faces (p > .05). Experiment 2 also found the existence of a significant negative bias in face memorization. Many previous studies have suggested that threat-related information receives more attention (Buchner, Rothermund, Wentura, & Mehl, 2004) and that it is associated with better memorization than with positive information (Kensinger, Garoff-Eaton, & Schacter, 2007; Ochsner, 2000); our findings are consistent with these results.

The participants’ unbiased hit rates of facial recognition tasks in the two rating scenarios are shown in Table 4.

Table 4.

Mean Unbiased Hit Rates of Facial Trustworthiness in Two Rating Scenarios (M ± SD).

Rating Scenarios	Old Classification Faces			New Classification Faces
	Trustworthy	Neutral	Untrustworthy
Survival hunting	.08 (.105)	.06 (.074)	.13 (.109)	.64 (.082)
Hunting contest	.05 (.072)	.04 (.050)	.07 (.075)	.58 (.098)

Source Memory Results

The repeated measures MANOVA results showed that there was no any statistically significant main effect or interaction effect, Fs < 1. Thus, no advantage existed for source memory in facial survival processing; this result was consistent with the finding that people cannot obtain an advantage in selection by using their memories of survival processing in source-identification tasks (Nairne, Pandeirada, Vanarsdall, & Blunt, 2015; Savine et al., 2011). The raw recognition performances are shown in Table 5.

Table 5.

Source Memory of Different Trustworthiness Types in Two Rating Scenarios (M ± SD).

Face Traits	Survival Hunting	Hunting Contest
Trustworthy	3.95 (1.995)	4.10 (1.743)
Neutral	4.44 (1.997)	4.18 (1.675)
Untrustworthy	4.26 (1.855)	4.10 (1.808)

Overall, in recognition test, we found an asymmetry between the memorization of trustworthy faces and untrustworthy faces across survival hunting scenarios. Specifically, a negative bias existed toward memorizing untrustworthy faces; people tended to recognize untrustworthy faces more accurately than trustworthy faces. Unfortunately, in source memory test, there was no advantage of source memory was found in the SPA. It indicates that face memorization and source memory appear separate from each other, which is consistent with the findings of Savine et al. (2011).

General Discussion

Across the life span, facial recognition seems to employ unique cognitive abilities. The occurrence of this ability in perception might be beneficial to our survival (Oosterhof & Todorov, 2008; Maguinness & Newell, 2014). Although previous research has failed to obtain evidence for such SPA for facial memorization, scholars suppose the occurrence of a boundary condition is not surprising because all traits could show boundary conditions, regardless of whether they arise from exaptations or adaptations (Nairne & Pandeirada, 2016).

In the current experiments, which used faces categorized as untrustworthy, neutral, and trustworthy, during encoding, participants have to directly evaluate their avoidance or approach response tendency to the strangers. In two experiments, the accurate recognition of both trustworthy and untrustworthy faces is higher in the scenarios related to survival than those in the control scenarios, which means that the SPA of the trustworthy face and the SPA of the untrustworthy face are observed. After Nairne and colleagues (2012) have shown the existence of a spatial SPA by replacing vocabulary stimuli with pictorial stimuli, our research expands the boundary conditions of the SPA by using facial stimuli (Nairne, Vanarsdall, Pandeirada, & Blunt, 2012). A great difference seemed to exist between facial recognition and other memory processes.

It is worth noting that, for these two kinds of SPAs of facial trustworthiness, untrustworthy faces are recognized more accurately than trustworthy faces which reveal a negative bias phenomenon. Evolutionary psychologists have noted negative and positive events are usually asymmetrical in human memory; untrustworthy faces are more accurately recognized than trustworthy faces because negative events can help humans avoid potential threats. This negative bias is adaptive (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001).

As mentioned earlier, our ancestral hunter-gatherers who were immersed in a rich environment, both animals and human beings were species that deserved careful treatment. Given that untrustworthy conspecifics pose considerable risks (e.g., exploitation, attacker), it is sensible that apparently untrustworthy faces would be linked with avoidance. Conversely, trustworthy faces should be connected with approach, as they offer desirable social outcomes (Slepian, Young, & Harmon-Jones, 2017). Therefore, these faces with obvious clues of trustworthiness capture more attention, which seems to be consistent with the hierarchical view of survival processing (Sandy et al., 2013).

However, we propose that survival processing preference caused by facial trustworthiness might be attributed to self-protective motivation. Self-protection motivation is a fundamental motivation for human cognition and behavioral responses. It can help people to detect and avoid potential threats from the natural setting and from others (Neuberg, Kenrick, & Schaller, 2011). Moreover, the self-protection mechanism does not mean only that people react with avoidance. Although overemphasizing negative information (e.g., untrustworthy faces) can help people avoid potential threats, it is not helpful in social exchanges (Bell, Buchner, Kroneisen, & Giang, 2012). The face of a potential cooperator that provides underlying information about trustworthiness is one of the factors that trigger the self-protective mechanism (Young et al., 2015). Thus, individuals willingly seek people who at least seem trustworthy (and who may become reliable sources of protection) for cooperation to avoid potential damage. Thus, self-protective motivation enables people to respond appropriately in a survival situation, such as paying more attention to those who seem trustworthy and untrustworthy.

In addition, there is no evidence that SPA of face exists in source memory. This current result suggests that source memory and facial recognition exist independently. But some researchers have found the SPA of source memory in certain scenarios (e.g., infectious diseases that induce immune system performance; Scofield, Buchanan, & Kostic, 2018; potential mates, Pandeirada et al., 2017). Controversy remains concerning the SPA of source memory.

Limitations and Implications

However, some limitations of the present study should be noted. First, we employed male faces as experimental stimuli based on the view that the male labor force was responsible for hunting in hunter-gatherer societies, we proposed that men could offer more threat and assistance than female did. Although we observed the SPA of face, however, some scholars suggested that the idea of sex-based cognitive specializations was still controversial (Nairne et al., 2009). Therefore, it is necessary to use female faces as experimental stimuli to further measure the SPA of face. Furthermore, our study primarily recruited Chinese undergraduate students, who are typically collectivism samples, whether the conclusion of the present study could be applied to other cultures needs further exploration.

In addition, our research on the SPA was carried out in a way that imaging the scenarios in the laboratory. Future research needs to verify this effect in a more realistic scenario, which may help to improve the external validity of the experiment.