Studies have shown that people display a preferential processing of negative information (e.g., negative facial expressions, immoral behaviors) than the corresponding positive one (Baumeister et al., 2001; Rozin and Royzman, 2001; Dyck et al., 2009). This phenomenon is known as the “negativity bias" and has been investigated within different domains, such as impressions formation, decision-making, social interaction, moral judgment, etc (Baumeister et al., 2001; Rozin and Royzman, 2001). Such a bias is also very common in face related studies: negative faces are preferentially processed; in terms of contextual influences on facial processing, negative information is always more influential than other information. However, presently, there are limited studies regarding the existence of a positivity bias. Therefore, this study aimed to explore the positivity bias in face processing.

The information processing bias toward negative stimuli may manifest in attention, memory, or perception. Studies show that negative faces draw more attention or are remembered better than are positive or neutral faces (Hansen and Hansen, 1988; Öhman et al., 2001; Anderson et al., 2011; Tsukiura et al., 2013). For example, Hansen and Hansen (1988) reported that the speed of identifying an angry face from a crowd of smiling faces was faster than vice versa. Tsukiura et al. (2013) found that the faces with an untrustworthy impression were remembered more accurately than those with a neutral or trustworthy impression were. Furthermore, studies show that the processing of human faces is affected not only by facial movements but also by context information (e.g., person-related information). The negativity bias exists in the context effect is conceptualized as stronger influence that negative information had on faces than did positive information (Abdel Rahman, 2011; Anderson et al., 2011; Baker et al., 2013; Suess et al., 2015; Luo et al., 2016). For instance, Anderson et al. (2011) explored the impact of gossip on the processing of neutral faces, with the paradigm of binocular rivalry. The result indicated that the neutral faces paired with negative gossip dominated in visual awareness significantly longer than did faces paired with other gossip. However, no difference was found for the faces paired with positive and neutral gossip. Baker et al. (2013) investigated the influence of moral behaviors on face recognition. They firstly presented vignettes related to moral behaviors (either immoral, morally neutral, or altruistic) with neutral faces, and then asked participants to identify the target faces within a set of faces with different levels of trustworthiness. Results showed that faces paired with immoral vignettes were recognized as less trustworthy than the actual faces, while, there was no difference in the altruistic or neutral condition, namely, only immoral behaviors influenced facial recognition memory. Crucially, by means of event-related potentials (ERP), the negative personal knowledge bias on the faces was also found in the early perceptual processing period (Abdel Rahman, 2011; Wieser et al., 2014; Suess et al., 2015). Suess et al. (2015) reported that the neutral expressions of unfamiliar faces paired with negative biographical information were perceived as more negative than the faces paired with relatively neutral information, indexed by larger early posterior negativity (EPN), but the effect was not apparent for positive biographical information. EPN was taken as the earliest ERP component reflecting valenced personal information influence on facial perception (Wieser et al., 2014; Luo et al., 2016).

However, it should be noted that the negative information involved in the previous studies mentioned above is mostly the one carrying threat, such as angry faces or evil behaviors (Öhman et al., 2001; Abdel Rahman, 2011; Feldmann-Wüstefeld et al., 2011). For security and survival, people prioritize paying attention to the potential threat in the environment and display the “negativity bias” in information processing. From the perspective of evolution, the threat-related negativity bias is reasonable and of high adaptive value: the consequences of dangerous stimuli are often much more dramatic than those of ignoring or reacting slowly to neutral or even appetitive stimuli. But what if there is no threat in the surroundings? Specifically, we were interested in whether negativity bias still exists when the negatively valenced information is not threatening or adverse (we define such information as “non-threatening information”).

The current study assumes that negative non-threatening information (e.g., unattractive faces, low ability information) does not pose a threat to our survival and security; thus, it would not make people go on alert. The corresponding positive one (e.g., attractive faces, high ability information), however, carries desirable, beneficial information. In such cases, the positive information may have a more powerful influence than the negative one. Some studies have shown indirect evidence for this assumption. Research showed that in aesthetic processing, compared with non-attractive faces, the attractive faces elicited an EPN (Werheid et al., 2007). The EPN is closely related to personal selective attention in the early phase (Schupp et al., 2007; Frühholz et al., 2011). A recent study also found that attractive faces dominated in visual awareness significantly longer than average and unattractive ones (Mo et al., 2016). However, none of them has directly explored the positivity bias or summarized the attribute of the negative information. Accordingly, combining behavioral assessment and ERP technology, we planned to test the “positivity bias” effect that non-threatening information may have on face perception.

We chose ability as a representative of non-threatening information in the present study. Ability is appropriate because low ability information carries no threat. At the same time, ability (or competence) is one of the universal dimensions of social cognition, playing an important role in person perception and evaluation (Skowronski and Carlston, 1987; Fiske et al., 2007; Freddi et al., 2014). Based on this, we further used a similar paradigm as those studies displaying the negativity bias, a minimal affective learning task (Abdel Rahman, 2011; Suess et al., 2015; Luo et al., 2016), for a direct comparison. In the current experimental tasks, the neutral expression faces were paired with high ability, low ability or control sentences, to test whether valenced ability information could bias the perception of facial expressions. If the expression ratings of the faces paired with high ability sentences show greater changes between pre- and post-learning than other faces (Experiment 1: Behavioral Assessment), and the EPN of the faces paired with high ability information is more pronounced (Experiment 2: ERP Data), we can conclude that there does exist a “positivity bias” evident in the effect of ability information on face perception.

The present study investigated the influence of non-threatening information (i.e., ability information, whose negative dimension involves no threat) on human face processing. Specifically, neutral faces and valenced ability sentences (high ability, low ability, and control) were paired to identify whether the valenced ability information could bias the perception of neutral facial expressions. The effect was expected to be more evident after exposure to high ability sentences, displaying a positivity bias. Both behavioral and ERP data supported our hypothesis, i.e., compared with low ability behaviors, high ability behaviors induced stronger effect on people’s positivity/negativity ratings of faces.

In Experiment 1, the behavioral data showed a significant change in the expression ratings for the faces associated with both high ability and low ability (the high ability group were rated as more positive, while the low ability group were rated as more negative), but the change in the former was greater than those in the latter. The behavioral results in Experiment 2 were in accordance with the results of Experiment 1; greater change happened in the high ability group than in the low ability group after learning. In conclusion, the behavioral results suggested that the high ability information has a stronger effect on facial evaluation than the low ability information. The results of the ERP analyses in Experiment 2 showed that the three experimental conditions (high ability, low ability, and control) all elicited the obvious component of N170 and EPN in the time window 130–180 ms and 250–300 ms. For N170, we could not find any effect, which indicates that it may be unaffected by emotional faces (Eimer et al., 2003; Schacht and Sommer, 2009; Klein et al., 2015). For EPN, high ability elicited larger amplitude than other conditions, but no differences emerged between the low ability condition and control condition. Numerous studies have shown that EPN is related with more attention and enhanced perceptual encoding of emotional stimuli (Schupp et al., 2003, 2007; Abdel Rahman, 2011). Moreover, EPN is deemed as the earliest component reflecting the effect of affective information on facial perception (Wieser et al., 2014; Luo et al., 2016). In the present study, the EPN effect was more pronounced for faces paired with high ability sentences compared to those faces paired with low ability sentences. This suggests that the high ability information plays a more important role in expression perception and induces a larger bias in facial perception compared to low ability information. In other words, consistent with our hypothesis, the ability trait did exhibit a “positivity bias” influence in facial perceptual processing.

The valenced ability information effect found in our results provide new evidence that affective person-related information influences face processing (Wieser et al., 2014; Klein et al., 2015; Suess et al., 2015; Luo et al., 2016). This might have implications for social communications. Before meeting someone, we may have had some knowledge of them (e.g., ability information). Such available information may influence our inferences to their mental state or intentions, which further regulate our own behaviors and attitudes toward them. Further, the “positivity bias” results were different from those found in previous studies (Abdel Rahman, 2011; Suess et al., 2015; Luo et al., 2016). For example, Suess et al. (2015) paired morality-relevant actions with neutral faces in the learning phase. The results showed that a significant change in expression evaluation value occurred when the faces were paired with negative moral actions; by contrast, little change occurred when those faces were paired with positive moral actions. These results imply that in the process of perceptual impression formation, negative moral information is more influential; that is, there exists a negativity bias for moral information. Negative moral information involved in these studies contains threat; negative ability information in current study, however, is not threatening. Thus, combining our results with prior research on the positive attention bias (Werheid et al., 2007; Mo et al., 2016), we suggest the assertion that positive information may take precedence over negative one when people process non-threatening information whose negative aspect do not embody threat.

Comparing the present study with previous research results, we conclude that people tend to exhibit a “negativity bias” when they process information whose negative aspect carries survival threat; when it comes to information whose negative aspect carries no threat, people may tend to display a “positivity bias.” Such a processing style may be explained by human tendency to avoid harmful stimuli and approach beneficial stimuli, and it reflects the flexibility of humans’ cognitive processing (Smith et al., 2006; Rothermund et al., 2008). Out of basic survival needs, people focus on potential threats in their surroundings. The vigilance for dangerous signals can protect people from being hurt and is of great adaptive value for survival. However, in circumstances where there are no dangers, or where the potential threat is less than the potential benefits, the requirements for benefits are predominant. Thus, people focus more on favorable or potentially beneficial information. In this context, being sensitive to positive signals is of more adaptive value.

To summarize, our results add to the evidence on semantic context effects in face processing and suggest that not only threatening information (like morality information) but also non-threatening one (like ability information) can shape expression perception. What’s more, the “positivity bias” phenomenon has significant value for further understanding negativity bias in the facial perception, demonstrating that the “negativity bias” in face processing is not universal, but may be varied with the type of study stimuli. However, our study still has some limitations. Firstly, if an experiment testifying the negativity bias of threat-related information was added for a comparison, the results would be more conclusive. Secondly, we cannot ensure that low ability contains no threatening information absolutely in all cases, and further study should be conducted to explore this topic. Finally, further systematic studies can be carried out to strengthen our conclusion and enrich the effect, including experiments with other domains (like attention, memory) and other “non-threatening information (like social status information).”

This study was approved by the Human Research Ethics Committee of South China Normal University. Informed written consent was obtained from participants before the experiment.

SZ: study design, data collection, data analysis, paper writing. YX: study design, paper writing. JX: data analysis, paper revising. YY: data collection, paper revising. TL: data collection, paper revising. LM: study design, paper writing.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.