Sixty-two adult undergraduate psychology students from the University of Vienna participated in the experiment (25 of them male, M_age = 23.37, SD = 5.19). The experimenter explained the procedure to the participants and obtained written informed consent prior to the experiment in accordance with the ethical guidelines of the University of Vienna. Participants had normal or corrected-to-normal vision and received course credits for their participation in the experiment.

We produced 12 high-quality gray scale photographs of scenes containing pairs of female university students showing neutral face expressions. Scenes were pre-rated as expressing neutral emotional valence (Strick et al., 2008) and pre-selected so that in half of the scenes both models’ facial attractiveness differed less than 0.5 rating points (equal attractive scenes), and the other half differed more than 1.2 ratings points (different attractive scenes) on a seven point rating scale (separate pre-study). We photographed the models in front of a gray background, and placed two models side by side with a lateral separation of one meter from the midpoint of the torso in the scene (Kaisler and Leder, 2016). The models were photographed under conditions that controlled for distance, angle, and luminosity. In all gaze conditions, model’s gaze direction and head angle were matched, resulting in three gaze conditions: DIR+LAT, DIR+AV, and LAT+AV (Figures 1A–C). For isolated views, i.e., one-face-scenes, we took separate photographs of the models showing direct gaze in front view (0°) in the middle of a scene with gray background to obtain baseline measures (Figure 1D). The side (left–right) of the looking and looked at face was counterbalanced, resulting in a total of 96 stimuli (12 scenes with three gaze conditions and two left–right versions plus 24 one-face scenes).

Stimuli were presented on a 19″ LCD display at 1280 × 1024 pixel resolution and viewed from a distance of approximately 50–60 cm. Stimulus presentation, trial randomization, and data collection were conducted with E-Prime (V.2, Psychology Software Tools). Participants rated all 24 faces separately in two rating blocks (randomized for left- and right face and scene), once for attractiveness and once for trustworthiness. The order of the rating blocks was counterbalanced across participants and the stimuli randomized for trials and side. The instructions above the picture indicated whether the left or right face had to be rated. The original wording stated: “How attractive/trustworthy is the left/right person?” Participants were instructed to use a computer mouse to select a number on a seven point Likert scale presented below each scene, with “1” being “very unattractive” or “not at all trustworthy” and “7” being “very attractive” or “very trustworthy.” We instructed participants to respond as spontaneously as possible and encouraged them to use all the numbers available on the scale, as they deemed appropriate. The exposure duration of the picture was unlimited and the rating scale remained on screen until a response was given. Each participant thus rated each of the 24 faces presented in 12 scenes twice in a single session lasting approximately 40 min. After the rating blocks, participants filled in a questionnaire, which contained questions about the familiarity of the faces (data not shown). Because familiarity strongly affects the quality of face processing (see Burton et al., 2015), we excluded 10 participants who indicated that they knew any of the presented faces from data analyses.

In order to show the influence of facial attractiveness on person judgments of others, we analyzed gaze effects of equal and different attractive scenes. We calculated mean values of direct (DIR), looking (LAT), and averted gaze (AV) across three gaze conditions (Table 1). Reliability analyses indicated internal consistency of participant’s attractiveness ratings, Cronbach’s α = 0.96, and trustworthiness ratings, Cronbach’s α = 0.90 in scenes of equal attractiveness levels. Similar for scenes with different attractiveness levels, consistency between attractiveness ratings, Cronbach’s α = 0.93, and trustworthiness ratings, Cronbach’s α = 0.89. We conducted a 3x3x2 repeated measurement ANOVA (95% confidence interval) with gaze directions (DIR, LAT, AV) and scene type (equal attractive scenes: average; different attractive scenes: highly and less attractive faces) as within-subject factor and gender (male, female), as between-subject factors.

We found significant differences between the gaze direction and scene type for attractiveness judgments (Figure 2A): gaze, F(2,61) = 13.15, MSE = 2.23, p = 0.001, ηp2 = 0.18, scene type, F(2,61) = 208.15, MSE = 84.17, p = 0.001, ηp2 = 0.78, and an interaction between gaze and scene type, F(4,61) = 12.79, MSE = 2.09, p = 0.001, ηp2 = 0.18. We found a similar pattern for trustworthiness judgments (Figure 2B): gaze, F(2,61) = 14.79, MSE = 5.46, p = 0.001, ηp2 = 0.20, scene type, F(2,61) = 45.11, MSE = 15.58, p = 0.001, ηp2 = 0.43, as well as an interaction between gaze and scene type, F(4,61) = 10.73, MSE = 2.06, p = 0.001, ηp2 = 0.15. We did not find gender effects or interactions with gender. Therefore, further analyses were conducted separately for equal and different attractive scenes.

We analyzed participants’ judgments of three gaze conditions in which each face and scene were rated twice, for attractiveness and trustworthiness. For equal attractiveness scenes, a repeated measurement ANOVA (95% CI) of attractiveness judgments with gaze directions (left and right face) as within-subject factor, showed a preference for the looking faces (45°) in the social approach (DIR+LAT), for direct gaze faces (0°) in the social avoidance (DIR+AV), and for the looked-at face in the ambiguous gaze condition (LAT+AV, Table 1). However, regarding social evaluations, participants considered direct-gaze faces (0°) as more trustworthy in the social approach and avoidance condition. In scenes with different facial attractiveness levels, participants rated highly attractive faces as more attractive and trustworthy, independent of the gaze direction and head angle.

These findings emphasize the esthetic advantage of three-quarter views of faces when judging facial attractiveness (e.g., Langton and Bruce, 2000), hence participants considered faces looking toward (45°) the second face as being more attractive than direct-gaze (0°) faces. However, trustworthiness judgments were unaffected by three-quarter views and confirmed the direct gaze bias for social evaluations. Further, we found a preference for highly attractive faces; increased levels of facial attractiveness might have overruled subtle gaze effects. Taken together, social in contrast to esthetic evaluations seem to be more sensitive to directions of gaze, independent of the head orientation.

We analyzed direct-gaze faces (0°) in social approach (DIR+LAT) and avoidance conditions (DIR+AV) to reveal whether being looked-at by another person positively increases participant’s ratings. A two-paired t-test (95% CI) showed a preference for direct-gaze faces that were looked-at by the second face in equal attractiveness scenes, as indicated by increased attractiveness ratings, t(1,62) = −3.29, p = 0.001. We found no preference for faces that were looked at by another face for trustworthiness ratings, t(1,62) = 0.72, p = 0.461. We did not find an effect in scenes with different levels of facial attractiveness-highly attractive faces looking toward the second, average attractive face and vice versa. These findings show that the subtle gaze effects of being looked-at, as shown by Kaisler and Leder (2016), might also be relevant in social situations with two people of average facial attractiveness. In order to focus on the interaction of head and gaze cues, we conducted Experiment 2 in which we omitted the perception of the variation of gaze direction by blurring scenes.

Seventy undergraduate psychology students from the University of Vienna participated in Experiment 2 (50 female, M_age = 21.7 years, SD = 2.6). The experimenter explained the procedure to the participants and obtained written informed consent prior to the experiment in accordance with the ethical guidelines of the University of Vienna. The participants had normal or corrected-to-normal vision. Participants received course credits for their participation.

We used the same two-face scenes as in Experiment 1, but for the rating blocks we blurred the scenes with a Gaussian blur filter (σ = 8) in Photoshop expect for one clear ellipse (225 × 260 pixels) that showed only the face to be rated in order to direct attention toward it (Figures 1E,F). The gray background in the pictures matched the surrounding screen color when presented with E-Prime (V.2, Psychology Software Tools), thus effectively disappearing during stimulus presentation.

The design comprised a phase to familiarize participants with the scenes that varied in gaze direction of both faces. Afterward, as in Experiment 1, participants rated all 24 faces separately in two rating blocks, once for attractiveness and once for trustworthiness, 48 faces in total.

Participants’ ratings were excluded for analysis when forced-direction judgments in the familiarization phase contained more than 10% errors, which accounted for elimination of 16 participants’ ratings (4.38% of all participants). From the remaining 54 participants’ judgments, we sampled across participants for attractiveness and trustworthiness judgments for each gaze direction and condition (M and SD, Table 2). Reliability analyses indicated internal consistency of participant’s attractiveness ratings, Cronbach’s α = 0.70, and trustworthiness ratings, Cronbach’s α = 0.89 in scenes of equal attractiveness levels. Similar for scenes with different attractiveness levels, consistency between attractiveness ratings, Cronbach’s α = 0.78, and trustworthiness ratings, Cronbach’s α = 0.86. As described in Experiment 1, we analyzed participants’ judgments of three gaze conditions (Table 2).

In equal attractiveness scenes, an ANOVA of attractiveness judgments (95% CI) revealed no preference for a direct or averted gaze in the social approach (DIR+LAT) and ambiguous condition (LAT+AV), but a preference for averted gaze faces looking away in the social avoidance condition (DIR+AV). Regarding social evaluations, participants considered direct-gaze faces (0°) as more trustworthy in the social approach (DIR+LAT) and showed no preference in the other conditions. In scenes of different levels of facial attractiveness, participants rated highly attractive faces as more attractive and trustworthy, independent of the direction of gaze and head orientation.

The comparison of the results of Experiment 1 with those of Experiment 2 reveals that the presence and direction of gaze plays an important role in scenes with two average attractive faces. It seems that gaze, and therefore the presence of a second person, influences effects of head angles: for example, the three-quarter views advantage vanished when gaze was blurred. Regarding trustworthiness judgments, the direct gaze bias might capture more attention than other cues, even when the gaze direction of the second person is blurred. However, in the social avoidance situation the direct gaze bias disappeared, possibly because the attention was drawn to another point outside of the scene. Regarding facial attractiveness, the ability of highly attractive faces to capture attention and elicit pleasure was not influenced by a second person or her direction of gaze.

As described in Experiment 1, we analyzed direct-gaze faces (0°) in social approach (DIR+LAT) and avoidance conditions (DIR+AV) to reveal whether being looked-at by another person positively increases participants’ ratings. A two-paired t-test (95% CI) showed no preference for direct-gaze faces that were looked-at by the second face in equal and different attractiveness scenes, nor for highly attractive faces. As expected, subtle gaze effects of being looked-at, as shown by Kaisler and Leder (2016) may only occur when both faces and the gaze direction are clearly visible.