This study recruited 170 Japanese participants (88 men, 82 women; mean [M] age = 45.8 years, range = 20–69 years, standard deviation [SD] = 13.9) from an internet research company (Cross Marketing, Tokyo, Japan). A medium-effect-size power analysis (ρ = 0.30, power of the test = 0.95) indicated that 134 participants were needed to detect significant effects.

Cross Marketing sent emails to potential candidates and invited them to participate in the study. We invited participants of varying ages to take part in this study to obtain a wide range of responses and in turn, more generalizable results. The emails explained the purpose of the survey, confidentiality and security of personal information, voluntary nature of participating in the study, freedom to withdraw from the study, and reward for participation. It also contained a link to the survey website. The potential participants were informed that the study was designed to investigate the relationship between personality traits and evaluations of the participants’ impressions of other peoples’ faces. They were given a contact address for any inquiries about the handling of personal information and were informed that if they experienced any negative effects while participating in the survey, they should close their browser immediately. They were also told that withdrawal from the study would not incur any disadvantages. The participants were required to read the study description and then click a button to access the survey. This process was considered to constitute the provision of informed consent. This survey was approved by the institutional review board of Otemon Gakuin University.

To create the composite facial stimuli used in the face judgment task, we took photographs of 31 young adult Japanese men (M age = 20.5 years, SD = 2.0) and 28 young adult Japanese women (M age = 19.6 years, SD = 1.7). The subjects in the photographs had a neutral expression and were looking directly at the camera under normal lighting and a gray background. They then completed the Japanese version of the Dark Triad Dirty Dozen scale (DTDD-J) (Tamura et al., 2015). The distance between the camera and the targets was held constant at 2 m. Before the photo shoot, the targets removed their facial jewelry and makeup and secured their hair off their foreheads with hairpins.

The DTDD-J comprises a 12-item questionnaire that measures the DT personality traits on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree): Machiavellianism (e.g., “I tend to manipulate others to get my way”), psychopathy (e.g., “I tend to be unconcerned with the morality of my actions”), and narcissism (e.g., “I tend to want others to admire me”). The scores of the four items that assessed each DT personality trait were calculated by summing the scores of the items measuring Machiavellianism (M = 10.67, SD = 3.72), psychopathy (M = 10.91, SD = 2.97), and narcissism (M = 12.33, SD = 3.77).

We then created prototype facial stimuli for the high and low DT personality traits by morphing five individuals who scored highest or lowest on each DTDD-J subfactor for each sex. Specifically, when we created the prototype male face with high Machiavellianism, we merged the facial photographs of five targets who scored highest on Machiavellianism. The composite face stimuli were created by averaging shape, color, brightness, and texture information for the face image using FantaMorph™ software (version 5; Abrosoft, Lincoln, NE, USA). We repeated the same process to create a total of 12 composite face stimuli, including six for male targets and six for female targets with high and low DT personality traits, respectively. Finally, we masked the hairstyles and clothing from the faces to conceal these features. Figure 1 shows the prototype faces for each DT personality trait used in the current study.

The participants completed the face judgment task online following the procedure in the experimental method described by Holtzman (2011). The online data collection was conducted over 2 days (October 17–18, 2023). The face judgment task comprised 12 trials, and participants were presented with two facial photographs (on the left and right of the screen) during each trial, one of which was the high DT prototype face (e.g., Machiavellianism, psychopathy, or narcissism) and the other corresponded to the low DT prototype face. The face stimuli were presented alongside a rating scale and a definition of each of the DT personality traits. For Machiavellianism, the definition was: “manipulative for personal gain, scheming, or conspiring.” For psychopathy, the definition was: “reckless, antagonistic, assertive with others, and angry at others.” For narcissism, the definition was: “arrogant, vain, pompous, self-absorbed, and assertive.” These definitions for each DT personality trait corresponded to earlier studies (Holtzman, 2011; Shiramizu et al., 2019; Alper et al., 2021). The participants completed a questionnaire using an 11-point Likert scale to indicate confidence in their ratings (−5 = confident that the person on the left matches the defined trait to +5 = confident that the person on the right matches the defined trait). The locations for the composite facial stimuli (i.e., left vs. right) were counterbalanced, and the order of trials was randomized for each participant. Following the face judgment task, the participants completed the DTDD-J to assess their levels of DT personality traits. In the current study, each DTDD-J subfactor demonstrated adequate internal consistency (Cronbach’s α > 0.63).

To calculate the degree of the accurate inference for DT personality traits from facial photographs, we first recoded the items for which the correct response was the leftmost key on the face judgment task. For example, −5 (+5) was recoded as +5 (−5). If the correct response corresponded to the right high DT personality trait option, no recoding was necessary. This recoding procedure allowed us to interpret that a higher score on the face judgment task corresponded to higher accuracy in discriminating DT personality traits from facial photographs.

We calculated the M, SD, skewness, and kurtosis for the scores in the face judgment task and each DTDD-J subfactor. We then conducted a normality test (i.e., the Kolmogorov–Smirnov test) to examine the normality of data distribution. In addition, we compared whether there were differences in the accuracy of the ratings for each target sex and DT personality trait. Finally, the correlation coefficients between the face judgment task scores and the participants’ DTDD-J scores according to sex were calculated.

We calculated the M, SD, skewness, and kurtosis for the face judgment task scores and each DTDD-J subfactor score (Table 1). The mean face judgment task score for Machiavellianism among male targets was −0.31 (SD = 1.54), whereas those for psychopathy and narcissism were − 0.40 (SD = 1.52) and − 0.33 (SD = 1.46), respectively. The mean face judgment task score for Machiavellianism among female targets was 0.78 (SD = 1.57), whereas those for psychopathy and narcissism were 0.55 (SD = 1.28) and 0.26 (SD = 1.07), respectively. The mean scores for each DTDD-J subfactor were: Machiavellianism = 2.35 (SD = 0.78), psychopathy = 2.75 (SD = 0.71), and narcissism = 2.43 (SD = 0.81).

The results of the Kolmogorov–Smirnov test indicated that the face judgment task scores were not distributed normally (ps < 0.001). A series of Kolmogorov–Smirnov tests also showed non-normal distributions for each DTDD-J subscale score (ps < 0.004). Thus, all subsequent analyses were conducted using nonparametric tests.

We tested whether the accuracy of inferring DT personality traits was significantly different from chance (i.e., 0) using the Wilcoxon (W) signed-rank sum test. For female targets, a significant difference was found for all three DT personality traits (Machiavellianism: W = 7,093, p < 0.001; psychopathy: W = 3,858, p = 0.005; narcissism: W = 5,747, p < 0.001). The accuracy of all three DT personality trait inferences for female targets was significantly above chance. For male targets, a significant difference was found for all three DT personality traits (Machiavellianism: W = 2,658, p = 0.005; psychopathy: W = 2,755, p < 0.001; narcissism: W = 2,767, p = 0.004). Therefore, the accuracy of DT personality trait inferences for male targets was significantly below chance.

Next, to explore whether the face judgment task scores differed for each sex according to the targets or DT personality traits, we conducted W signed-rank sum and Friedman tests. The results of the Wilcoxon signed-rank sum test revealed that the accuracy of inferring Machiavellianism (Z = 5.50, p < 0.001), psychopathy (Z = 5.30, p < 0.001), and narcissism (Z = 3.97, p < 0.001) was significantly higher for female targets than male targets. The Friedman test revealed significant differences between Machiavellianism and narcissism (Z = 4.20, p < 0.001) and psychopathy and narcissism (Z = 2.06, p = 0.04) in female targets for the accuracy of inference for each DT personality trait. The results showed that accurate inferences for DT personality traits in the female target faces were higher than those in male target faces. Furthermore, both Machiavellianism and psychopathy were more easily recognizable than narcissism in female targets.

We calculated Spearman’s rank correlation coefficients to examine the associations between the accuracy of inferring DT personality traits from faces and participants’ DT personality traits. Tables 2, 3 present the correlation coefficients between face judgment task scores and participants’ DTDD-J scores according to sex.

The results indicated that there was a positive relationship between the accurate inference for Machiavellianism in male targets and three DTDD-J subfactor scores among men (Machiavellianism: ρ = 0.37, p < 0.001, 95% confidence interval [CI] = 0.17–0.54; psychopathy: ρ = 0.25, p = 0.02, 95% CI = 0.05–0.44; narcissism: ρ = 0.22, p = 0.04, 95% CI = 0.01–0.41; Table 2). However, we observed a negative association between the accurate inference for Machiavellianism in female targets and scores for Machiavellianism (ρ = −0.21, p = 0.05, 95% CI = −0.40–0.00). For female participants, there was a negative relationship between the accurate inference for Machiavellianism in male targets and scores for psychopathy and narcissism (psychopathy: ρ = −0.28, p = 0.01, 95% CI = −0.47–0.07; narcissism: ρ = −0.28, p = 0.01, 95% CI = −0.47–0.06). In addition, there was a negative relationship between the accurate inference for narcissism in male targets and psychopathy and narcissism scores (psychopathy: ρ = −0.24, p = 0.03, 95% CI = −0.44–0.03; narcissism: ρ = −0.24, p = 0.03, 95% CI = −0.43−0.02; Table 3). The above association between participants’ DT personality traits and the accuracy of inferring DT personality traits from facial information remained significant after controlling for participant age.

The face judgment task results revealed intriguing patterns in the accuracy of inferring DT personality traits from facial photographs. Notably, participants demonstrated high accuracy in identifying Machiavellianism, psychopathy, and narcissism among female targets, which is consistent with previous findings showing that DT personality traits can be deduced from facial information (Holtzman, 2011; Shiramizu et al., 2019; Alper et al., 2021). Furthermore, our results are consistent with previous research on facial cues and the Big Five personality traits, such as extraversion and conscientiousness, in female targets (Penton-Voak et al., 2006; Little and Perret, 2007; Alper et al., 2021). This suggests that female facial cues communicate information related to both favorable and unfavorable personality dimensions.

However, we found that the accuracy of inferring DT personality traits from facial photographs among men was below chance, which is inconsistent with previous findings (Holtzman, 2011; Shiramizu et al., 2019; Alper et al., 2021). This suggests that individuals found it difficult to identify DT personality traits using male facial information. To understand this unexpected result, it is important to consider the impact of individual differences in each target. In a previous study, Holtzman (2011) created a prototype face for each DT personality trait by synthesizing the faces of 10 individuals. In our study, the prototype face was synthesized from five individuals. Therefore, it is possible that in our study, the impact of each model’s facial features on the assessment of personality traits was greater than that in the previous study. As a result, the accuracy of inferring DT personality traits from facial photographs among men was below chance. It is also possible that the results of the face judgment task were influenced by a preference for physical or psychological features, such as facial features, of individuals with high DT personality traits. Previous research has shown that individuals with high DT personality traits are more likely to be rated as physically attractive by others. Holtzman and Strube (2011) hypothesized that DT personality traits evolved to exhibit higher levels of physical attractiveness. Subsequently, Holtzman and Strube (2013) confirmed this hypothesis when they found a positive association between the degree of narcissism of a photographed target and others’ ratings of the target’s physical attractiveness. In addition, Marcinkowska et al. (2015) found that women who are motivated to acquire short-term mates have a preference for highly narcissistic male faces. When men perceive the world as dangerous, they may prefer to interact with individuals who exhibit high levels of psychopathy over individuals with low levels of psychopathy (Brown et al., 2017). This suggests that psychopathy is perceived as a desirable personality trait in threatening environments. Collectively, these findings suggest that physical features, such as the face, of individuals with high DT personality traits are perceived favorably by others. This favorable perception of the physical features of individuals with high DT personality traits may explain the weaker link between socially undesirable personality traits and facial features among men observed in this study.

Our analysis revealed sex differences in the accuracy of inferring DT personality traits from facial photographs. Machiavellianism, psychopathy, and narcissism were more accurately inferred from female targets than from male targets. These results are consistent with previous findings that the accuracy of identifying DT or Big Five personality traits from facial photographs is higher for women than men (Little and Perret, 2007; Holtzman, 2011). The observed sex differences in accuracy suggest potential variations in the expressiveness of facial features related to DT personality traits between men and women. The difference in judgments between male and female faces may be because female faces contain more cues to their actual personality than male faces. It is also possible that the sex differences in correctly identifying DT personality traits using facial information were influenced by differences in the ratings for attractiveness, masculinity, and femininity for male and female faces. Further research is needed to determine the impact of impression ratings for faces of each sex.

When we examined the relationship between the accuracy of inferring DT personality traits and observers’ DT personality traits, we found a positive association between the accuracy of inference for Machiavellianism in male targets and male participants’ own DT personality traits. These results suggest that males who are high in each of the DT personality traits are more sensitive to others’ Machiavellian cues, which leads to more accurate inferences. In a previous study, Lyons et al. (2013) found that male individuals with high psychopathy could detect other people’s lies easily. Furthermore, Turi et al. (2022) examined the association between DT personality traits and deception detection, with a focus on the role of high DT traits in deception detection given the tendency of deceptive individuals to engage in unethical behavior. Our results showed a positive relationship between the accuracy of inferring Machiavellianism in male targets and male participants’ own DT personality traits, which expands earlier findings. In addition, individuals with high DT personality traits are characterized by deceptive and manipulative behaviors toward others to achieve their goals (Paulhus and Williams, 2002). To successfully deceive or manipulate others, individuals with high DT personality traits would need to accurately infer others’ personalities based on limited information. The positive relationship between the accuracy of inferring Machiavellianism in male targets and participants’ own DT personality traits may be a factor that contributes to the development of cunning social strategies.

In contrast, there was a negative association between the accuracy of inferring DT personality traits for male targets and female participants’ psychopathy and narcissism. These results indicate that female participants with higher levels of psychopathy and narcissism exhibited a poorer ability to accurately discern DT personality traits in male targets. On the basis of previous findings, we speculate that women with higher DT tendencies are less able to detect the lies of others (Lyons et al., 2013). However, a more detailed examination of the relationship between the accuracy of identifying DT personality traits from other people’s facial information and observers’ DT personality traits, especially in regard to sex differences, is necessary.