Generalized tendency to make extreme trait judgements from faces

Generalized tendency to make extreme trait judgements from faces. Atsunobu Suzuki, Saori Tsukamoto and Yusuke Takahashi. Royal Society Open Science, November 23 2022. https://doi.org/10.1098/rsos.220172

Abstract: People differ in their tendency to infer others' personalities and abilities from their faces. An extreme form of such face-based trait inference (FBTI) is problematic because of its unwarranted impact on real-world decision making. Evolutionary perspectives on FBTI suggest that its inter-individual variation would be trait-specific: e.g. those who make extreme face-based inferences about trustworthiness may not necessarily do so about dominance. However, there are several psychological variables that can increase the FBTI extremity across traits. Here, we show that there is a generalized individual tendency to make extreme FBTI across traits, in support of the latter view. We found that the degrees of extremity of face-based inferences about seven traits had high cross-trait correlations, constituting a general factor. This generalized FBTI extremity had good test–retest reliability and was neither an artefact of extreme nor socially desirable response biases. Moreover, it was positively associated with facial emotion recognition ability and tendencies to believe physiognomy and endorse stereotypes. Our results demonstrate that there are individuals who have a temporally stable disposition to draw extreme conclusions about various traits of others from facial appearance as well as their psychological characteristics.

4. General discussion

4.1. Summary of main findings

Studies 1 and 2 consistently showed that those who make extreme face-based judgements on a certain trait also tend to make extreme judgements on other traits, providing evidence for generalized FBTI extremity. The results are in contrast with the evolutionary perspectives on FBTI [22–24] that its inter-individual variation would be trait-specific. Instead, they are more compatible with psychology theory-based predictions that there are several variables that could enhance the FBTI extremity across traits. In partial accordance with our hypotheses, generalized FBTI extremity was positively associated with physiognomic belief, facial emotion recognition ability and stereotype endorsement, but not with cognitive miserliness. In short, we demonstrate that there are individuals who have a temporally stable disposition to draw extreme conclusions about various traits of others from facial appearance as well as their psychological characteristics. These findings and continued research into this issue will contribute to a better understanding of the nature of the excessive impact of face on social decision making, sometimes called face-ism [6], and to an identification of a prime target population for its intervention [15,16].

4.2. Generalized face-based trait inference extremity

Our analyses ensured that the generalized FBTI extremity reflects true individual differences in social cognition and is not merely a by-product of methodological artefacts. First, the well-established trustworthiness-by-dominance model of facial impressions [17,25,26] was reproduced from the present data, dismissing the possibility that our task format prompted participants to judge faces solely on a single dimension. Second, high cross-trait correlations of the FBTI extremity were demonstrated even after controlling for non-specific response biases such as extreme response style [51] and socially desirable responding [52]. Third, we showed that the generalized FBTI extremity had good test–retest reliability over a 2-month interval. Electronic supplementary material, text S2 also includes a complementary study we conducted to determine whether the temporal stability of the generalized FBTI extremity was explained by response biases. While this study exhibited a somewhat lower test–retest correlation of generalized FBTI extremity (r = 0.699, 95% CI [0.586, 0.786]) than Study 1, the value changed only slightly after controlling for extreme response style and socially desirable responding (r = 0.681, 95% CI [0.563, 0.772]). These results indicate that the generalized FBTI extremity is a temporally stable disposition.

Concerns may be raised if our attempts to rule out a response-bias account are sufficient. First, the extreme response bias when rating others' faces may not be well captured by the bias when rating self-descriptive verbal statements (i.e. Extreme Response Style Measure [51]). To address this issue, extreme response bias was quantified using data from the face-based trait-rating task. Taking trustworthiness as an example, 10 faces that were not classified as either stereotypically trustworthy- or untrustworthy-looking (i.e. whose mean trustworthiness rating ranked between 11 and 20 out of 30 faces) were selected separately from male and female faces. Then, for each participant, the ratings were recoded such that higher values reflected a greater tendency to choose extreme categories on the scale (i.e. ratings of 6, 5, 4, 3, 2 and 1 were recoded into 3, 2, 1, 1, 2 and 3, respectively), and the recoded ratings were averaged across the selected 20 faces. Bias scores for other trait judgements were defined in the same way, and the averages of trait-specific scores were treated as an overall index of extreme response bias in the face-based trait-rating task. Even when controlling for this additional measure of extreme responses, high cross-trait correlations of the FBTI extremity were reproduced. Moreover, while the association of generalized FBTI extremity with physiognomic belief was not significant, its associations with facial emotion recognition ability and stereotype endorsement were replicated. Second, it may be argued that, since the face-based trait-rating task had participants rate all traits at once for each face, lazy response behaviours, such as frequently choosing the same values on the rating scale, and/or strategic response behaviours, such as trying to provide a consistent (plausible) pattern of ratings, could have inflated cross-trait correlations of FBTI extremity. Regarding this point, we would like to note that in the face-based trait-rating task, desirable traits appeared almost evenly on the left and right sides (figure 1a)—in other words, there were reverse items for each face, and the order of the trait dimensions was randomized across trials. Therefore, it is unlikely that a lazy response leads to high cross-trait correlations. In addition, this counterbalanced and randomized setting would require significant effort when viewing the scale to provide a consistent pattern of ratings. Thus, we used survey completion times as viable measures of participant effort [58] and examined their correlations with the variables of interest in the present research. We found that the generalized FBTI extremity did not correlate positively with completion times, whereas it negatively correlated with squared completion times. Albeit a post hoc speculation, it is possible that participants finishing the survey at a moderate pace made the strongest effort, which might have elevated cross-trait correlations of FBTI extremity. This interpretation could explain the negative correlation between the squared completion times and generalized FBTI extremity. However, controlling for completion times (including their squares) did not change our main findings. Based on these additional analyses (see electronic supplementary material, text S3 for details), we believe that the generalized FBTI extremity would not be an artefact of the response bias.

It is also worthwhile to discuss the distinctions between explicit rating tasks akin to the one used in this study and those tasks measuring FBTI in a more indirect fashion. A seminal work proposing the trustworthiness-by-dominance model of facial impressions was based on the data of explicit trait ratings of faces [17]. Subsequent influential studies that replicated and extended the two-dimensional model ([23,25,34]; for a review, see [26]) and recent pursuits of perceiver effects on FBTI [3–5,21,59] also used explicit rating tasks. Therefore, our results indicating the presence of individuals who explicitly form extreme facial impressions across traits would be a valuable addition to the literature on FBTI, which relies heavily on data from explicit rating tasks. On the other hand, facial impressions have also been assessed by more indirect measures, such as the latency and dynamics of responses related to trait inference [60–62] and decision making in simulated social settings [53,63,64]. In general, explicit and indirect measures assess related but distinct constructs [65], and one measure is not necessarily superior to the other. For example, perceived trustworthiness of a given face is often measured using trust games in which participants are asked to decide how much money they are willing to entrust to the owner of the face [53,63,64]. Trust games may appear to provide a better indicator of trustworthiness impressions than explicit rating tasks given that money has an objective unit of measurement (e.g. British Pound), whereas the meaning of verbal anchors on a rating scale (e.g. very trustworthy) is subjective and may vary across participants. However, the relationship between monetary decisions in trust games and trustworthiness impressions of game partners is actually not straightforward and depends on participants, because those decisions are influenced by various factors, such as the participant's general trust, preference for fairness and betrayal aversion [66]. In addition, the use of indirect measures is usually recommended when explicit measures are likely to be biased by people's tendency to hide socially undesired responses [65]. Importantly, such social desirability bias was found to have a nearly null correlation with generalized FBTI extremity in the present study. Therefore, we suggest that the use of an explicit rating task does not significantly undermine our findings. Moreover, a fruitful direction for further research is to investigate individual differences in FBTI using indirect measures, since even their basic characteristics (e.g. how large and temporally stable the individual differences are) remain unknown.

4.3. Correlates of generalized face-based trait inference extremity

The positive association between the physiognomic belief and the generalized FBTI extremity means that those who believe that various traits can be inferred from faces tend toward extreme FBTI, which is in line with previous work [29]. This association was more clearly shown with the Physiognomic Belief Scale [29] than with the other questionnaire [53], perhaps because the former contains more items than the latter (14 versus 3) and, consequently, has higher scale reliability (Cronbach's α = 0.918 versus 0.817), and because the items of the former directly refer to the trait dimensions in the face-based trait-rating task, while those of the latter do not. In any case, the correlations were small, ranging from 0.104 to 0.171. The weak correlations may be reasonable considering that the questionnaires we used assessed the strength of the physiognomic belief without referring to any stereotypical facial cues to trait inference. It would thus be interesting in future research to measure the strength of explicit verbalizable beliefs about commonly held face–trait relations (e.g. ‘A kind-hearted person has big, round eyes’ [9]) and examine its links with the FBTI extremity.

It is noteworthy that we successfully demonstrated in an adult population that those who were adept at recognizing facial expressions tended toward extreme FBTI. While such a relationship is consistent with an emotion overgeneralization account of FBTI [17,22,30,31], a recent developmental study showed that emotion comprehension skills (including facial emotion recognition ability) covaried with the extremity of face-based trustworthiness inference only in 5-year-old children and not in 7-year olds [20]. The authors opined that covariation was absent in older children because their emotion comprehension skills were more developed and thus less variable. In the present study, the facial emotion-rating task scores were more variable than the facial emotion identification task scores, which approached a ceiling effect, and the former scores had a larger correlation with generalized FBTI extremity than the latter. These results indicate that if individual differences in facial emotion recognition are quantified using sensitive tests [35,36], their association with FBTI can be detected even in adulthood.

Our research partially supports the tendency toward extreme FBTI among people who endorse stereotypes. Specifically, the category-based trait-rating task scores moderately and positively correlated with generalized FBTI extremity, while the Acceptance of Stereotyping Questionnaire [39] scores hardly did. In addition, the two measures, which were assumed to reflect the same construct (i.e. stereotype endorsement), were almost entirely uncorrelated with each other. A critical difference between the two indices is that whereas the rating task explicitly deals with stereotypes about social categories that are closely connected with FBTI (i.e. smiling people [17,33], babies [37] and women [38]), the questionnaire does not specify any social categories. Thus, if participants who self-identified as White (i.e. the majority of the study participants) answered the questionnaire with stereotypes about racial outgroups in mind [67], it is natural that the answers showed little correlation with the FBTI extremity, considering the exclusive use of Caucasian faces (i.e. ingroup faces) in the face-based trait-rating task.

The fact that the hypothesized positive correlations between generalized FBTI extremity and cognitive miserliness were not observed may be taken as evidence of the cognitive impenetrability of FBTI [68]. However, a caveat is that the face is the only information available to perform the face-based trait-rating task. This could obscure possible facilitative effects of cognitive miserliness on FBTI, since even those participants who doubted the validity of their intuitive impression of the face—our measures of cognitive miserliness are supposed to assess such a tendency (not) to rely on intuition—had no choice but to make a face-based inference. Thus, it is important to examine whether cognitive miserliness boosts reliance on facial appearance in trait inference when more diagnostic but cognitively taxing cues (e.g. a record of previous cooperative and cheating behaviours [69]) are available.

4.4. Limitations

This study has some other limitations worthy of mention. First, a direct and more extensive investigation into the causal mechanisms of generalized FBTI extremity is important in future research. Our major contribution is the demonstration of high cross-trait correlations of the FBTI extremity, or the existence of a generalized FBTI extremity, providing new directions for and constraints on further studies and theorizing about FBTI. As an initial attempt to explain the generalized FBTI extremity, we proposed and examined several variables that can potentially foster it. However, those variables were a small subset of possible causal factors, and the present study provided only partial and correlational evidence for our hypotheses. Second, different researchers have adopted different methods to score FBTI performance [3,20,21,29,59]. For example, when agreement scores [21] (see the Introduction for a definition) were computed using the present data, they had high correlations (above 0.7) with extremity scores. However, the scatter plots between the two types of scores showed that there were larger variations in extremity scores when the agreement scores were higher (see electronic supplementary material, text S4 for details). In other words, individuals with higher agreement scores can have both low and high extremity scores. Thus, FBTI extremity is not only related to agreement scores but also contains distinct perceiver characteristics. Future investigations need to explore a set of quantifications that can efficiently and comprehensively capture individual differences in FBTI. Third, our surveys started with a block of socio-demographic questions, since collecting certain socio-demographic data (i.e. current residence, primary language and age) was necessary to confirm participants' eligibility to participate in the surveys. These questions might have directed participants' attention to their age, sex and ethnicity, which in turn could have influenced how they approached the subsequent social cognitive tasks. However, the participants' age and sex were controlled for in the statistical analyses presented here. In addition, the results remained virtually unchanged when we analysed the data of only White participants (i.e. a large majority of participants). Therefore, we suggest that even if participants' age, sex and ethnicity had been activated by socio-demographic questions, they should have exerted little impact on our findings. Finally, our studies only targeted native English speakers in the USA, given the availability of well-normed materials in English, leaving cross-cultural similarities and differences unexplored [25].