Revisiting the Watching Eyes Effect: How Emotional Expressions, Sex, and Age of Watching Eyes Influence Stereotypical Statement Endorsement

1.1 Examining the Eye Images to Understand the Watching Eyes Effect

The eye stimulus used in the WEE experiment often depicts a pair of stylized eyes (Dear et al., 2019; Haley & Fessler, 2005; Oda et al., 2015; Rotella et al., 2021). These stylized eyes resembled the eye of Horus (Haley & Fessler, 2005). According to ancient Egyptian beliefs, the eye of Horus serves as protection to ward off danger (Darnell, 1997). However, Horus was sometimes associated with anger because of anger episodes occurring during a constant conflict with his nemesis, Seth (Balegh, 2014; Griffiths, 1958). In ancient Egypt, red was also associated with power, anger, and violence (Singer, 2010). Therefore, there is a belief that, when the eye of Horus turns red, it is a sign of anger and danger (Hussein, 2010). This also suggests that the configuration of the eye of Horus (i.e., spatial positioning of the eye region) is predisposed to expressing anger.

To date, no studies have examined if observers can perceive emotional expressions from such stylized eyes in the WEE experiment. We do not know if the response differences in the WEE are related to the perceived emotion of the eye images shown. Moreover, recent studies have adopted human eyes when investigating the WEE (Kelsey et al., 2018; Rotella et al., 2021; Shinohara & Yamamoto, 2018; Van Doesum et al., 2021), and begun questioning how features of the eye images affect the response differences in the WEE (Manesi et al., 2016; Pauwels et al., 2017; Sueur et al., 2023). We briefly describe three features that can be conveyed by the eyes: emotional expressions, age, and sex. These features, in conjunction with recent findings, may help us better understand the response differences seen in the WEE.

Human eyes convey emotional expressions. We move the eyes and the surrounding regions to show emotions, like widening the eyes to show fear or surprise (Ekman et al., 1990; Gunnery & Ruben, 2016; Hess et al., 2009; Lau, 2021; Lau & Huckauf, 2021; Lee et al., 2013; Matsumoto, 1989; Roy-Charland et al., 2015; Schützwohl & Reisenzein, 2012), or by displaying crow’s feet when we make genuine smiles (Ekman et al., 1990), or by enhancing/altering eye expressions using only the eyebrows (Tipples et al., 2002). Eye sex can be estimated by looking at the eye size (Midelfart, 1996) or the skin color around the eye region (Nestor & Tarr, 2008). We may also guess how old a person is by looking at the number of wrinkles in the eye region (Ganel, 2015; Ganel & Goodale, 2021). From these, more abstract characteristics of a person, such as attractiveness, trustworthiness, and intelligence, can be read just by looking at variations in the eyes (Kret & De Dreu, 2019; Kret et al., 2015; Lau et al., 2022; Todorov & Duchaine, 2008; Todorov et al., 2009; Van Breen et al., 2018; Willis & Todorov, 2006).

Recent studies using human eyes have unveiled some key components of why responses differ in the WEE. For instance, gaze direction refers to where the eyes are looking. Averted gazes (i.e., looking away) do not elicit response differences in the WEE compared to direct gazes (i.e., looking into the camera/observer; Manesi et al., 2016). However, several studies that used forward-gazing eyes also failed to replicate response the response differences in the WEE (Carbon & Hesslinger, 2011; Kawamura & Kusumi, 2017; Matsugasaki et al., 2015; Rotella et al., 2021; Sparks & Barclay, 2015). This means that gaze direction is not the only contributing factor. Since the eyes convey several information simultaneously, it is possible that these eye features act in combination to elicit response differences.

1.2 Known Eye Features that Influence the Watching Eyes Effect

Recent studies varied the emotional expression, age, and sex of the eye images to investigate the WEE. Emotional expressions of the eye images may shape response differences seen in the WEE. Pauwels and colleagues (2017) showed that the response differences were greater for angry eyes than for happy eyes. Specifically, participants donated and cooperated more when viewing angry eyes than happy eyes. One limitation of the study was using only young female eye images. According to the literature, how strong a facial expression is perceived depends on both the observer’s and the actor’s sex (Hall & Matsumoto, 2004; Johnsen et al., 1995; Matsumoto, 1989; Sacco & Hugenberg, 2009). Specific to anger and fear, anger expressions are often associated with men, and fear expressions are often associated with women (Becker et al., 2007; Hess et al., 2009). This means that specific emotional expressions and the sex of the person expressing the emotions are highly intertwined. In Pauwels and colleagues’ study (Pauwels et al., 2017), female eye images were enough to demonstrate response differences in angry eyes but not in happy eyes. However, angry expressions are associated more with men, than with women (Becker et al., 2007; Hess et al., 2009). If response differences observed in the WEE are due to the perceived emotional expression from the eye images, then emotional expressions from different sexes should affect the magnitude/direction of the response differences.

Studies that use eye images from different sexes do not show response differences (Manesi et al., 2016; Nettle et al., 2013; Panagopoulos, 2014; Pauwels et al., 2017; Shinohara & Yamamoto, 2018; Sueur et al., 2023). However, the lack of response differences in the WEE occurs between neutral male and neutral female eyes. As one study has shown, eye expressions can influence the WEE (Pauwels et al., 2017). Furthermore, emotional expressions are tightly coupled to the sex of the person displaying the emotions (Hall & Matsumoto, 2004; Johnsen et al., 1995; Matsumoto, 1989; Sacco & Hugenberg, 2009). Thus, when varying eye expressions in the WEE experiment, it is also important to vary the eye sex to evaluate whether emotional expressions influence a change in responses or behaviors in the WEE paradigm. If the response differences in the WEE are due to the perceived emotional expressions, then emotional expressions from male and female eyes should be used because of the convoluted nature between emotional expressions and sex.

There are also hints that age is important to investigate in the WEE. One study showed a decrease in antisocial behaviors when viewing a child’s eyes, in contrast to older eyes (Sueur et al., 2023). Specifically, participants obeyed traffic laws more after seeing a child’s eyes, compared to an adult male’s eyes or an adult female’s eyes. This finding is not surprising since we usually consider that children are the world’s future. Consequently, we do our best to ensure their survivability (Clark et al., 2020; Fulcher & Coyle, 2011). What remains unanswered are questions concerning older ages. From the literature, we commonly assume that older people hold powerful and senior positions in large organizations (Cox & Harquail, 1991; Shen & Cannella, 2002) and that they, like our parents or grandparents, have more experience and are wiser than us (Kogan & Shelton, 1962). This can also shape our respect towards the elders, even if performance does not differ across age (Jacobs et al., 1990; Yue & Ng, 1999). Since younger people often act differently and harbor different expectations about those from other age groups (Finkelstein et al., 2013; Hummert et al., 1994), it makes sense that a person’s behavior would change in the vicinity of someone else from another age group. Hence, eyes of different ages could affect how much responses change in the WEE.

1.3 Searching for Comparable Images that Contrast Eyes with Varied Expressions

The between-subject design is almost always used in past literature to evaluate the WEE (Bateson et al., 2013; Ekström, 2012; Pfattheicher & Keller, 2015). Specifically, one group of participants is shown only a pair of neutral eyes, and the other group is presented with only a neutral image, like flowers (Francey & Bergmüller, 2012; Oda & Ichihashi, 2016; Shinohara & Yamamoto, 2018). With this design, some studies did not replicate response differences (Dancer & Burn, 2019; Fenzl & Brudermann, 2021; Rotella et al., 2021). One study even found a reversed effect, where responses differed more for the flower images than for the neutral eye images (Raihani & Bshary, 2012).

Experimentally speaking, eyes and flowers have very different visual features (e.g., color, shape, size, etc.). Flowers can look more complex than eyes, depending on how complicated the petals are positioned. However, we do not detect them the same way as human eyes (McCarthy et al., 1997). Even when the flower is made to look like a pair of eyes, it neither represents human eyes, nor elicits response differences seen in the WEE (Carbon & Hesslinger, 2011). In contrast, eyes are more readily recognized since we are predisposed to recognize them almost immediately (Jessen & Grossmann, 2014; Simion et al., 2011; Simion & Giorgio, 2015). Eyes also convey social signals that are important for communication (Bavelas et al., 2002; Broz et al., 2012; Burgoon & Le Poire, 1999; Kleinke, 1986). If we varied eye features like emotional expressions, age, and sex, to investigate the WEE, then flower images would not be suitable comparisons because they differ from eyes. Instead, eye image variations should be compared to the relevant control eye image, accounting for the age and sex of the eyes.

Eyes are highly arousing stimuli (Fawcett et al., 2016; Gobel et al., 2015; Helminen et al., 2011; Jarick & Bencic, 2019). Merely becoming aware of a pair of eyes is enough to raise arousal levels (Conty et al., 2010). If arousal drives response differences in the WEE, then it makes sense to compare high-arousing eye images to low-arousing images like flowers or no image. Typically, the WEE is investigated using a between-subjects design. This design cannot account for participants’ unique arousal differences since they only see an eye image instead of the whole set of images. However, arousal differences within each participant can be measured by adopting a within-subject design (Charness et al., 2012). There is only one study adopting this design to investigate the WEE (Bateson et al., 2006). Importantly, the study found an increase in cooperative behavior whenever eye images were presented compared to flower images. Since within-subject experiments also show response differences in the WEE, we can evaluate whether arousal is important by adopting a within-subject design in our study. Specifically, we would expect responses to differ between neutral eye images and the control flower images.

To better understand the role of arousal, it would be important to first show or replicate an absence of an effect when comparing neutral eyes to control flower images, following past studies (Bateson et al., 2013; Ekström, 2012; Pfattheicher & Keller, 2015). If responses differ between neutral eyes and control flower images, this suggests that arousal plays a role in the WEE. However, if there is an absence of an effect between neutral eyes and control flower images, then this suggests that the current comparison method (between eyes and flowers) cannot evaluate the effect. Then, the WEE should be further evaluated by comparing only the different eye images to determine if specific eye images (or features) induce a change in responses.

1.4 Study Goal

The following was an exploratory study. Dear and colleagues (2019) recommended investigating antisocial behaviors from their meta-analysis. We adopted their recommendation and used the extent to endorse stereotype statements as a kind of antisocial behavior (Crosby et al., 1980; Devine, 1989; Genthner & Taylor, 1973; Park et al., 2003; Zhang, 2010). Antisocial behavior was quantified by the extent participants would endorse 36 stereotypical statements about race, religion, and gender. Participants rated the extent they would endorse a given stereotype statement presented on the screen.

We first explored whether arousal between neutral eyes and flower images played a role in inducing response changes in the WEE paradigm. We hypothesized that stereotype endorsement ratings would differ between neutral eyes and control flower images. If stereotype endorsement did not differ between flower images and neutral eyes, then this would suggest that arousal between neutral eyes and flower images does not play a huge role in the WEE effect. Hence, neutral eye images would be used as control images for comparing other eye images with different characteristics.

We then explored whether the underlying mechanism behind response differences in the WEE paradigm was driven by individual, varying eye features (emotional expressions, sex, and age) or a combination of them.

We hypothesized that the stereotypical statement endorsement would be larger for emotionally expressive eyes compared to non-expressive eyes (i.e., neutral). Based on our review, emotional expressions and sex are closely related. There is a tendency to associate anger with males and fear with females (Becker et al., 2007; Hall & Matsumoto, 2004; Hess et al., 2009; Johnsen et al., 1995; Matsumoto, 1989; Sacco & Hugenberg, 2009). Thus, when varying emotional expressions, one cannot ignore how sex also influences perceived emotions. Since previous studies investigating emotional expressions in the WEE used only anger and happy expressions (Pauwels et al., 2017), we also included fear because this expression is quickly recognized (Hoffmann et al., 2010; Misailidi & Bonoti, 2008).

We hypothesized that stereotype statement endorsement would differ between old-eye images and young-eye images. This is because a child’s eyes elicited the largest behavioral changes than adult eyes (Sueur et al., 2023). Furthermore, the literature also showed that people typically behave differently around those from very different age groups (Finkelstein et al., 2013; Hummert et al., 1994). We compared young adult eyes to old adult eyes since no studies have made such comparisons.

In the current study, we adopted a within-subject design to investigate whether emotional expressions, sex, and age of eye images could influence how much participants endorse 36 stereotypical statements across race, religion, and gender. From past research, forward-gaze elicits response differences in the WEE but not averted-gaze (Manesi et al., 2016). We presented colored forward-gazing human eyes. We first compared neutral eye expressions to control flower images, imitating past literature (Fenzl & Brudermann, 2021; Manesi et al., 2016), before comparing neutral eyes to eyes of other expressions. The eye expressions used were happy, angry, fear, and neutral. Eye sexes from male and female eyes were used. Eye ages from young and old adults were used. All eye images (emotional expression, sex, and age) were counterbalanced.

2.1 Participants

Participants were recruited in accordance with the Declaration of Helsinki. The study was approved by the ethics committee of Ulm University within the context of the DFG-funded project, grant number 425867974. Power analysis was conducted before recruitment using G*Power 3.1 (Faul et al., 2009). About 170 participants were needed for the repeated measures F-test to achieve power (1 – β) = 0.99, α = 0.05, and small effect size f = 0.10. We expected a small effect size because many studies failed to replicate the watching eyes effect. Therefore, if there is indeed an effect, it would likely be a small one. From this calculation, we recruited 232 respondents between May 7 and Jun 14, 2022. All participants provided informed consent prior to the study.

Those who failed the two comprehension check questions were excluded from the analysis (shown in Design below). From the data, 23 (9.91%) failed the first comprehension check, and 44 (18.97%) failed the second comprehension check. Thus, 188 participants (mean age = 25.68, SD = 7.18) successfully completed the experiment. There were more female (N = 103, 57.79%) than male (N = 84, 44.68%) and diverse (N = 1, 0.53%) respondents. The sampling method was of convenient means (i.e., friends, colleagues, social media). Those from Ulm University were compensated with course credits. Others volunteered for the experiment knowing that there was no monetary incentive at the end of the experiment. Thus, participants’ age and gender were not controlled in the current study.

2.2 Stimuli

Eye images used in the experiment were derived from full faces of the FACES database (Ebner et al., 2010) and consent to use the images was given in writing by the database administrators. The database features Caucasian faces across various ages and sex. All images from the database are colored. The original face images have the following dimensions: 2,835 × 3,543 pixels, 300 pixels per inch, and 8 bits depth. Eight models were selected from the database (two young males, two old males, two young females, and two old females). The following IDs were selected for old (ID = 004, 146, 012, 060) and young (ID = 031, 057, 152, 182) faces. The average perceived age of these images from the FACES database (Ebner et al., 2010) was 66.87 years (SD = 3.36) for old faces and 26.44 years (SD = 2.21) for young faces. The stimulus selection criteria are further described in Appendix A1

Each model expressed angry, fear, happy, and neutral expressions. We cropped the eyes and eye region such that all eye stimuli used in the experiment had the following dimensions: 1,356 × 418 pixels, 300 pixels per in. (Figure 1a). The image manipulation was done using GIMP (version 2.10). There were four flower images which acted as control images for comparisons with neutral eyes (Figure 1b). The control images had the same dimensions as the eye stimuli. In total, we used eight models × four expressions + four flowers = 36 images for the experiment.

Figure 1

The figure shows (a) example eye images, (b) control flower images used in the experiment, and (c) the symbols depicting the eight eye images used in the study. The illustration for (a) depicts an old adult male expressing angry, happy, fear, and neutral expressions. Eyes were derived from full faces in the FACES database (Ebner et al., 2010). The symbols depicted in this figure apply to Figures 3 and 4 in the Results section.

Stereotype statements used in the experiment were in the German language. Statements were taken from reputable sources such as news websites and the German-based market and consumer data company, Statista. Example news websites include Zeit.de, which is central leaning and does not hold strong political biases (Dallmann et al., 2015). Stereotype statements were used to poll the German public’s opinions on various socioeconomic and political issues between 2000 and 2020. We selected 36 statements to match the number of images used in the experiment. These statements were divided into three stereotype categories: race, gender, and religion. An example stereotypical statement concerning gender is “Meiner Meinung nach sind Frauen hormongesteuert” (In my opinion, women are hormone-driven). The full list of the statements used is found in Appendix A2

Half the stereotypical statements were phrased to convey positive meaning. This was to prevent participants from always responding using the same rating if all the statements were negatively phrased. For example, “Meiner Meinung nach sind Frauen schlecht in Mathe” (In my opinion, women are bad at Math) was changed to “Meiner Meinung nach sind Frauen gut in Mathe” (In my opinion, women are good at Math). The conversion was done by a native German speaker. We also ensured that statements across the three categories began with “Meiner Meinung/Erfahrung nach” (In my opinion/experience) whenever possible.

2.3 Design

The experiment was within-subject design involving 36 trials that corresponded to the images used. The experiment was administered online using the EFS Survey by Questback GmbH (2021). The experiment was conducted in German. All 36 images and 36 stereotype statements were fully randomized to ensure that no specific image could be attributed to a specific statement. We presented each trial on a single page. Participants always answered the question: “Wie wahrscheinlich würden Sie diese Aussage machen?” (How likely would you make this statement?) in each trial.

As the experiment was conducted in German, the question “Wie wahrscheinlich würden Sie diese Aussage machen?” was translated literally to “How likely would you make this statement?”. However, terms such as “likely” or “likelihood” could carry meaning related to probability or statistics. To avoid such confusion, we used the phrase “how much the participant endorses the stereotype statement” in this manuscript. The replaced phrase is to broadly refer to whether participants agree with the stereotype statements based on their own views.

A stimulus appeared only at the top of the page, center-aligned to the page. The stereotypical statement was presented below the stimulus, left-aligned to the page. The question was left-aligned to the page and was always centered between the stimulus and the statement. Figure 2a shows two example screenshots of a trial and the randomization between eye stimuli and stereotypical statements.

Figure 2

An example of two individual screenshots of a trial across two participants is depicted in (a). The screenshots show an eye image of an old angry male above two different stereotype statements. The experimental sequence is illustrated in (b). All images and stereotype statements were fully randomized across participants.

There were four instruction pages. On the first page, participants learned about the 9-point Likert scale used in the task. The scale indicated how much the participant would endorse the stereotype statement shown. The value 1 indicated the least and the value 9 indicated the most. A comprehension check question was implemented here. Participants had to click on “1” to indicate “the least.” The second instruction page described the duration and number of trials. The third instruction page was an example of how each trial would look like, and where participants can indicate their responses. The fourth page was feedback on their responses.

Participants then entered brief demographic information about themselves (i.e., gender, age, ethnicity, etc.) across two pages. The second comprehension check question was implemented here. Participants must click on the corresponding number on one of the items. The experiment ended, and a thankyou screen was shown together with a brief explanation of what the experiment was about.

Respondents who completed the survey on mobile devices were encouraged to use landscape orientation. There was no backward button and returning to previous pages to change responses was not allowed. All questions must be answered during the experiment. Otherwise, the message “Bitte beantworten Sie diese Frage” (Please answer this question) appeared. Participants were neither asked whether they saw the eye images nor to report the eye emotions during the study.

There were 36 trials in the experiment, corresponding to 36 images and stereotype statements used. Each trial was presented on a separate page. In total, there were 44 pages in the experiment (Figure 2b). The entire experiment took 10–20 min to complete.

2.4 Procedure

Participants first saw the welcome screen and were encouraged to orient their personal devices to landscape mode so that the images would display properly during the experiment. Informed consent was provided by clicking on “Ich bin einverstanden” (I agree) and on the “weiter” icon at the bottom right of the page before starting the experiment.

The instructions concerning the Likert scale, the task, and how the trials would appear were shown. Participants navigated through the four instructional pages before commencing with the experimental trials. In each of the 36 trials, an image appeared on top of the page. Participants rated how likely they would say the stereotypical statement on the 9-point Likert scale. At the end of the experiment, basic demographic information (i.e., gender, age, ethnicity, etc.) across two pages were surveyed. The experiment ended with a thankyou screen and a brief explanation of what the experiment was about.

2.5 Analysis

The analysis was conducted in SPSS Version 28.0. Those who failed the comprehension check questions were excluded from the analysis (2.1 Participants). Since we rephrased half the negative stereotypes to prevent participants from responding with the same response, we re-coded the responses of the negative stereotypes so that the arithmetic values did not neutralize themselves when we averaged participants’ responses. This also eases data interpretation as higher ratings indicate higher stereotype endorsement.

We evaluated whether variations (i.e., emotional expressions, sex, and age of the watching eyes) affected stereotype statement endorsement. The dependent variable (DV) was the level of endorsement. The independent variables (IVs) were emotional expressions (angry, fearful, happy, and neutral), eye sex (male and female), and eye age (young and old).

We first compared endorsement between neutral eyes and flower images by running a paired-samples t-test on the DV. Since we found no differences between neutral eyes and flower images (Figure 3), we used neutral eyes as the control image and ran the remaining analyses by comparing only the eye images.

Figure 3

Bar graph comparing mean ratings between neutral eyes and control flower images Symbols indicate the averaged ratings for each eye model. Error bars illustrate standard error of means (SEM). n.s. – non-significant.

We also checked whether participants could have responded to the three stereotype categories differently by running a one-way repeated measures ANOVA with the factor: stereotype categories (race, gender, and religion). The results showed that it was unlikely that participants responded more biased toward a specific stereotype category (Appendix A3). Therefore, subsequent analyses did not involve comparing responses across different stereotype categories.

We then ran a 4 × 2 × 2 repeated measures ANOVA to test whether emotional expressions, sex, and age influenced stereotype endorsement in the WEE paradigm. The main effects for each factor, the two-way interactions (eye sex × emotional expressions, eye age × emotional expressions) and the three-way interaction (eye sex × eye age × emotional expressions) were examined. Post-hoc comparisons were conducted for significant main and interaction effects. A conservative approach to account for inflated α values was adopted via Bonferroni-correction (Abdi, 2007). For the three-way interaction, we compared emotional expressions across both ages and sexes (e.g., happy eyes: old male vs old female). We also compared each emotional expression to neutral eyes for every age and sex (e.g., old male: happy eyes vs neutral eyes). Finally, we included a comparison between young female happy eyes and young female angry eyes to confirm the findings of previous literature (Pauwels et al., 2017). Therefore, this resulted in a Bonferroni-corrected α value of 0.05 ÷ (4 expressions × 2 age × 2 sex + 3 expression pairs × 2 age × 2 sex + 1 extra pair) = 0.0017.

To streamline the readability of this manuscript, we present the results section in the following order: we first present the comparison between neutral eyes and control flowers. We then summarize the main effects of each factor (emotional expression, age, and sex) and the two-way interactions. Finally, we show results concerning the three-way interactions. We described the post-hoc comparisons for each factor with significant main effects. Since the analyses of the two-way interactions are like the three-way interactions, the post hoc results for the two-way interactions are reported in Appendix A4.

Statistical assumption checks were conducted for the repeated measures ANOVA. We checked whether the sphericity assumption was violated using Mauchly’s W (Mauchly, 1940). Greenhouse–Geisser correction was applied when comparisons violated the sphericity assumption (Greenhouse & Geisser, 1959). The results of the statistical checks were reported in Appendix A5.

3.1 No Difference Between Neutral Eyes and Flower Images

The extent to endorse stereotypical statements did not differ between neutral eyes (M = 5.09, SD = 0.98) and flower images (M = 4.97, SD = 1.38), t(187) = 0.98, p = 0.330, and Cohen’s d = 0.07 (Figure 3). Therefore, subsequent analyses compared only the eye images.

3.2 Higher Stereotype Endorsement in Happy Eyes and Younger Eyes, But Lower Stereotype Endorsement in Fearful Eyes

The main effects of the factors, emotional expression, age, sex, and the two-way interactions are reported in Table 1. Details of the post-hoc comparisons for the two-way interactions are found in Appendix A4

The main effect of emotional expressions showed that they affected stereotype endorsement (Figure 4a). Post-hoc comparisons using Bonferroni-corrected α = 0.008 revealed that the extent to endorse stereotypical statements was higher for angry eyes (M = 5.34, SD = 0.91), than fearful eyes (M = 4.18, SD = 1.14), t(187) = 10.46, p < 0.001, and Cohen’s d = 0.76. The extent was also less in angry eyes than in happy eyes (M = 5.62, SD = 1.03), t(187) = −2.87, p = 0.005, and Cohen’s d = −0.21. There was no difference between angry eyes and neutral eyes (M = 5.09, SD = 1.28), t(187) = 2.66, p > 0.008, and Cohen’s d = 0.19.

Figure 4

Bar graphs for the factors (a) eye emotional expressions, and (b) age. Symbols in (a) indicate mean responses for each eye model/identity. Error bars illustrate the standard error of means (SEM). *** p < 0.001, ** p < 0.010, * p < 0.050, n.s. – non-significant.

For fearful eyes, participants showed the least stereotypical statement endorsement. That is, ratings were significantly lower than for happy watching eyes, t(187) = −11.32, p < 0.001, Cohen’s d = −0.83, and neutral eyes, t(187) = −8.35, p < 0.001, and Cohen’s d = −0.61. For happy eyes, participants were more inclined to endorse stereotypical statements than for neutral eyes, t(187) = 4.85, p < 0.001, and Cohen’s d = 0.35.

There was a significant main effect of eye age on stereotypical statement endorsement, F(1, 187) = 4.20, p = 0.040, and η p 2 = 0.02 (Figure 4b). The extent was lower for old eyes (M = 4.99, SD = 0.67) than for young eyes (M = 5.13, SD = 0.64). There was no significant main effect of eye sex on making stereotype statements, F(1, 187) = 1.14, p = 0.290, and η p 2 = 0.006.

In summary, we found the effects of emotional expressions and age of the watching eyes on the extent to endorse stereotypical statements. Happy eyes, along with young eyes, influenced the most stereotypical statement endorsement. Fearful eyes influenced the least stereotype endorsement.

3.3 Higher Stereotype Endorsement for Old Angry Male Eyes, Happy Eyes, and Old Fearful Female Eyes, Compared to Each Respective Expression

The goal of the three-way interactions was to further explore specific eye features used in the study that contributed to the main effects and two-way interactions. From the analysis, we found a significant three-way interaction between emotional expressions, age, and sex, F(3, 561) = 11.23, p < 0.001, and η p 2 = 0.06 (Figure 5). Post-hoc comparisons using Bonferroni-corrected α = 0.0018 are reported in Tables 2 and 3. Table 2 depicts the post-hoc statistics of each emotional expression between the factors of age and sex. Table 3 shows the post-hoc statistics of each emotional expression compared to neutral eyes for each age and sex.

Figure 5

The three-way interaction between the factors: emotional expressions, age, and sex. Symbols indicate age and sex. *** p < 0.001, n.s. – non-significant.

We then evaluated whether our data supported an existing literature showing higher response tendencies for angry young female eyes than for happy young female eyes (Pauwels et al., 2017). Our data showed similar response patterns. In our study, how much the participant endorses the stereotypical statements was higher in angry female eyes than in happy female eyes, t(187) = −4.03, p = < 0.001, and Cohen’s d = −0.29.

Summarizing the results from the three-way interaction, the data revealed that the extent to endorse stereotypical statements was dependent on the kinds of emotional expressions, the age, and the sex of the eyes. Old angry eyes elicited a larger endorsement for stereotypical statements in male eyes. Fearful eyes elicited a stronger endorsement for stereotypical statements in old female eyes. Across both sexes, happy eyes elicited a larger extent to endorse stereotypical statements for young compared to old eyes.

4.1 The Role of Arousal in the Watching Eyes Effect Paradigm

Our hypothesis that the WEE was not associated with arousal was supported. We replicated recent studies showing no rating differences between neutral eyes and control flowers (Carbon & Hesslinger, 2011; Fenzl & Brudermann, 2021; Raihani & Bshary, 2012) even when adopting a within-subject design. This provided evidence that arousal was unlikely an important factor behind the WEE, supporting past studies. For example, Carbon and Hesslinger (2011) postulated that if the WEE was associated with arousal, then the effect should be observable when showing other arousing non-eye images, but they found no such evidence. Fenzl and Brudermann (2021) also found no evidence of WEE because participants’ self-reported arousal ratings did not differ when viewing the eye image or the control flowers image.

Arousal from eye images is not as arousing as looking at real pairs of eyes. Although the eyes are highly arousing stimuli (Fawcett et al., 2016; Gobel et al., 2015; Helminen et al., 2011; Jarick & Bencic, 2019), we are bad at judging how aroused we feel when looking at static eye images partially occluded faces (Adolphs et al., 1999; Lau, 2021; Lau et al., 2022; Sato & Yoshikawa, 2007; Wang et al., 2018). We also do not experience the same physiological arousal when making eye contact with a person on a video recording, compared to making eye contact with a live person (Lyyra et al., 2018). This means that a better way of inducing arousal, such as making eye contact, is required to determine whether arousal plays a role in the WEE.

What remains unclear is also the psychophysical data about the observers’ arousal levels. Studies show a drastic increase in arousal when monitoring skin conductance and heart rate as observers make eye contact with another person (Helminen et al., 2011; Jarick & Bencic, 2019; Nichols & Champness, 1971). However, the arousal levels in healthy participants are lower when viewing images (Wang et al., 2018). No studies have measured participants’ internal arousal levels during a WEE experiment. Thus, there is still a need to rule out whether participants’ own arousal in the experiment influences their responses.

4.2 Happy Eyes Trigger a Sense of Acceptance

Our hypothesis that stereotype endorsement would differ between emotionally expressive eye images and neutral eye images was also supported. In our study, participants gave high endorsements when viewing happy young eyes.

Positive valence, such as feeling happy or seeing happy eyes, may trigger a sense of acceptance from social norms. Affective valence, that is, how positive or negative we feel, influences the way we interact with objects or people. Everyday objects elicit some form of valence (Lebrecht et al., 2012). We are more likely to approach positive valence stimuli and avoid negative valence stimuli (Alexopoulos & Ric, 2007). Nettle and colleagues (Nettle et al., 2013) suggested that an alternative explanation for the WEE could be that people are just doing things that conform to the social norms. If watching eyes make people act more closely to the social norms, then the extent to endorse stereotypes should increase if the norm reinforces such stereotypes. In our study, we observed high endorsement scores for happy young eyes. This could be interpreted as happy eyes inducing a sense of approval, and whichever response the participants made were seen as acceptable by the social norm.

The absence of an approval/acceptance effect for old happy eyes could be attributed to own-age face biases. In face perception, own-age faces are remembered and recognized more accurately (Anastasi & Rhodes, 2005; Harrison & Hole, 2009; Rodin, 1987). This means that younger participants are more likely to recognize expressions in younger faces. Since our participants were mostly young adults, they would be better at recognizing young happy eyes than old happy eyes. Thus, participants would not have felt a sense of approval for endorsing stereotype statements when viewing old happy eyes if participants did not strongly perceive positive valence from those eyes.

4.3 Fearful Eyes Promote Avoidance from the Content

We found that stereotypical statement endorsement was lower across almost all fearful eyes, despite randomizing the presentation of the statements (Figure 4). It was possible that the fearful watching eyes signaled warning/danger against endorsing stereotypes (Jack et al., 2014; Mathews et al., 2003; Morris et al., 2002). Fear is a powerful emotion because it acts as a defense mechanism for protection (LeDoux, 2012). We perceive whether a person is feeling afraid because fear is conveyed through facial expressions and body language (De Gelder, 2006). Fear can already be recognized from raised eyebrows and widened eyes (Dadds et al., 2006; Kohler et al., 2004; Matsumoto, 1989; Vernon & Berenbaum, 2002).

It is common for observers to react to another person exhibiting fear (Marsh & Ambady, 2007). Such reactions for the observers, called social fear, can be as strong as the person exhibiting the fear (Olsson et al., 2007). In fact, observers avoid fear-inducing stimuli when the stimuli cause fear in another person (Debiec & Olsson, 2017). Since messages from a fearful person appear more reliable and believable than those from a neutral person (Reed & DeScioli, 2017), this indicates that the contents experienced by a fearful person serve as an indicator of avoidance.

Fear as an indicator of avoidance could explain why participants gave the least endorsement when viewing fearful eyes. Specifically, participants did not endorse statements presented together with fearful watching eyes due to fear avoidance (Debiec & Olsson, 2017; Haaker et al., 2017; Olsson et al., 2007). Participants might have distanced themselves from the stereotypical statements when viewing fearful eyes. Therefore, we speculate that fearful-watching eyes may be used in future studies to reduce or deter stereotype endorsements.

4.4 Final Note on Ethical Concerns

It is important to point out the potential ethical concerns surrounding the study. We showed that stereotype endorsement was higher when displaying angry old male eyes and young happy eyes. We also prompted future research to use stimuli with similar features to investigate stereotype endorsement in the WEE paradigm. However, this can be ethically challenging when an experiment promotes antisocial/maladaptive behaviors like reinforcing stereotypes. Although there is a need to understand what evokes the response differences in the WEE, we caution researchers to carefully measure the risks and benefits of promoting certain behaviors when examining the WEE.

4.5 Future Directions

The current study may benefit from further improvements. It is unknown whether participants knew about the nature of the study. We do not know where participants align themselves when endorsing stereotypical statements. Future research could investigate if participants’ prior expectations of the study goal could bias the results or formulate a way to measure the ground truth to evaluate stereotype endorsement. It is also unclear whether emotionally expressive eyes can influence prosocial behaviors. Specifically, we do not know if expressive eyes affect donation behaviors (Haley & Fessler, 2005). Future studies could vary the eye expression, along with the eye’s age and sex, to better understand how variations of the eye stimuli affect donation amount, and/or donation frequency. In addition, ethnicity could play a role in stereotype endorsement and facial expression perception. Future studies should also investigate how ethnicity, prior stereotype biases, and emotion perception interact with eye images of varying expressions.

There are also challenges in interpreting some results. For instance, why did angry old male eyes lead to higher endorsement scores? One reason could be the daily experiences of growing up and living in a patriarchal-dominating society, in which the sight of an angry old man could conjure uneasiness and evoke behaviors to appease the angry old man. However, one might look at the results differently and ask: why were there no endorsement changes in old female eyes? Could old females be perceived as less strict? Since these interpretations are all very speculative, it would be beneficial for future studies to replicate the current findings before delving deeper into the relationship between eye age and stereotype endorsements, and further investigating how eye age influences responses in the WEE experiments.

Many factors beyond varying the eye images can also influence how responses differ in the WEE paradigm. For instance, self-referential processing when seeing a pair of eyes causes us to be more self-aware and cautious of our actions (Hietanen & Hietanen, 2017). Consequently, this motivates our need to seek positive approval of our actions (Conty et al., 2016) such as by acting more charitable (Cañigueral & Hamilton, 2019). Previous studies have also shown that a pair of eyes influences responses because of the observer’s sense of public self-awareness (Pfattheicher & Keller, 2015) and the number of people in the room where the experiment took place (Oda & Ichihashi, 2016). We also did not survey our participants whether they felt watched or observed by the eye images. Future studies may consider including some or all these measures to better understand how responses change in the WEE paradigm.

Going beyond self-report, future WEE experiments should also include psychophysiological data to gain insight into the observer’s potential changes of short-term phasic arousal and where the observer is looking. Examples of such data include the participant’s eye movements, pupillary changes, heart rate, and skin-conductance level. Such data is needed because looking at the eyes is arousing, more so when eye contact with another person is involved (Myllyneva & Hietanen, 2015; Nichols & Champness, 1971). However, no studies have measured the observer’s physiological arousal levels, aside from the self-reported arousal levels (i.e., arousal ratings). Furthermore, it is unknown how often observers look at the eyes, or do they simply ignore them. These additional parameters allow us to dissociate the role of the observer’s arousal in the WEE. Replication of the data in a lab enabling collection of more behavioral data such as time spent on each page, gaze spent on the eyes and the statements, would be helpful to get a richer picture of the processes underlying the response differences.