Perception of an Ambiguous Figure: Subconscious Social Group Bias or Anchor Effect?


 Older participants who are briefly presented with the ‘my wife and my mother-in-law’ ambiguous figure estimate its age to be higher than young participants do. This finding is thought to be the result of a subconscious social group bias that influences participants’ perception of the figure. Because people are better able to recognize similarly aged individuals, young participants are expected to perceive the ambiguous figure as a young woman, while older participants are more likely to recognize an older lady. We replicate the difference in age estimates, but find no relationship between participants’ age and their perception of the ambiguous figure. This leads us to conclude that the positive relationship between participants’ age and their age estimates of the ambiguous ‘my wife and my mother-in-law’ figure is better explained by the own-age anchor effect, which holds that people use their own age as a yard stick to determine the age of the figure, regardless of whether the young woman or the older lady is perceived. Assimilation of others’ characteristics to one’s own is particularly likely to occur in uncertain circumstances that provide little information to base judgments on, such as estimating the age of a briefly presented ambiguous figure.


Introduction
Ambiguous gures are gures that contain two or more different images and therefore can be perceived in distinct ways. The classic duck-rabbit ambiguous gure [1], for instance, contains an image of a rabbit facing right and an image of a duck facing left. People see only one of these images at a time.
Ambiguous gures create amusing visual effects and are therefore often found in popular media to attract and entertain readers. However, they are also used in psychological experiments to study the role of both conscious and subconscious processes in their perception [2,3,4,5].
Nicholls, Churches, and Loetscher [6] used an ambiguous gure to investigate whether high-level social processes subconsciously affect face perception. Following Bar [7], they hypothesized that early visual brain regions pass on a partially analyzed version of a face to the prefrontal cortex where it is subject to social expectations that may in uence the percept that is formed in the temporal cortex and any subsequent behavior that is directed at it. To test this hypothesis, Nicholls et al. presented their participants with the ambiguous 'my wife/mother-in-law' gure ( Fig. 1). This gure, which was originally introduced as a psychological tool by Boring [8], can be perceived as either a young woman facing away from the observer, or as an older lady facing more toward the observer. After a presentation time of 500 ms, which was deemed too short for conscious processes to be of in uence, participants were asked to estimate the age of the woman they saw. Nicholls et al. found that older participants estimated the woman in the gure to be signi cantly older than young people did. On average, participants who were older than 30 estimated the woman to be 6.32 years older than did participants who were 30 years or younger. Nicholls et al. also established a reliable positive relationship between the estimated age and the participants' own age. Across 393 participants, the correlation between the two variables measured .24.
The ndings by Nicholls et al. suggest that the way a face is processed is subconsciously affected by the perceiver's age. But how does one's age in uence face perception exactly? Nicholls and colleagues interpret their nding in terms of a social group bias. The in-group vs. out-group distinction is a common one in the social sciences. One's in-group consists of people who one shares similar characteristics with.
These characteristics are not shared with members of the out-group. Studies have found that people exhibit a recognition bias toward in-group members. That is, people are better able to recognize individuals from the same ethnicity [9], gender [10], or age [11,12]. For example, Wright and Stroud [13] showed that an eyewitness is more accurate at identifying the suspect of a crime if they belong to the same age group. The increased recognition accuracy for similarly aged faces, also known as the own-age bias in face perception, is believed to be due to increased contact and familiarity with people of one's own age group [14,15,16], as well as more in depth processing of the faces of the members of one's age ingroup [17; 18, 19, 20]. According to a meta-analysis on the own-age bias in face recognition [21], the way participants study and encode faces does not in uence the size of the group difference, suggesting that the bias presents automatically, which is in line with the unconscious age effects on face perception observed by Nicholls and colleagues.
If superior recognition of faces from one's own age group is at the basis of the nding by Nicholls et al., this would imply that an unconscious social group bias lead participants to recognize the woman in the ambiguous 'my wife/mother-in-law' gure who most closely matches their age group, resulting in a higher age estimate by older than by young participants. That is, participants would perceive the ambiguous gure differently, depending on their age. Young participants would be likely to see the young woman facing away from the observer, while older participants would be more inclined to perceive the older lady facing the observer.
A second interpretation does not require the gure that is seen by the participants to differ systematically between age groups. What Nicholls et al. observed might be explained alternatively by the own-age anchor effect. According to Tversky and Kahneman [22], anchor effects occur when people's estimates assimilate toward a salient value (the anchor) causing their estimates to be inaccurate. In the case of the own-age anchor effect, people use their own age as an anchor, causing age estimates to be biased toward their own age. An early study on the own-age anchor effect conducted by Mintz [23], for example, demonstrated that children estimate the age of the cartoon character Peter Pan to be similar to their own age. The observation that older participants estimate the ambiguous 'my wife/mother-in-law' gure to be older than younger participants do, could thus be the result of assimilation of the age estimates to participants' own age, regardless of the woman that is perceived. That is, both when viewing the young woman and the old lady, participants would make an age estimate that is anchored to their own age, resulting in a mean estimate difference between age groups, as well as a positive correlation between participant age and estimated age.
People's tendency to assimilate others' characteristics to their own has received less attention than the own-age social group bias, but empirical evidence supporting the own-age anchor effect is increasing [24,25,26,27,28]. The basis for the effect is not yet well understood. One reason why one's own age is selected as a yard stick might be because people usually choose what is highly accessible in memory as their standard [29]. In that case, the own-age anchor effect might have a similar origin as the own-age bias in face recognition: both biases could result from increased familiarity with people of one's own age [30]. Alternatively, the effect may be the result of a social cost reduction strategy, whereby participants overestimate the age of younger faces and underestimate the age of older faces in order to err on the safe side [28]. The own-age anchor effect is also reminiscent of egocentricity biases, whereby people consider themselves rather than others as a reference point [31], making them judge others to be more similar to themselves than the other way around [32] or project their own properties onto others [33].
Anchoring estimates to salient or normative values might be particularly adaptive in highly uncertain situations, where there is insu cient other information to base one's judgment on [30]. The brief presentation of an ambiguous gure might therefore constitute a situation in which the own-age anchor effect is likely to present.
The own-age social group bias is different from the own-age anchor effect in that the former is usually considered to be a perceptual bias, while the latter tends to be characterized as a cognitive bias.
Extensive experience with faces of our own age group is believed to facilitate their perceptual processing according to the own-age social group bias, while in the case of the own-age anchor effect it is believed to skew our judgments. The own-age social group bias thus predicts that young and older participants are respectively more inclined to see the young woman and the older lady in the ambiguous 'my wife and my mother-in-law' gure, and that this difference in perception is what is driving the higher age estimates by older than by young participants. Were the relationship between own age and age estimates due to the own-age anchor effect, the ambiguous gure would not need to be perceived systematically different by young and older participants. Participants would assimilate the age of the perceived women to their own, regardless of whether the young woman or older lady is perceived.
We present a study that aims to examine whether the ndings by Nicholls et al. can be best explained by the own-age social group bias or the own-age anchor effect. In order to do this, we will replicate the original study as closely as possible, but in addition ask participants to indicate whether they saw the young woman or the older lady. If participants' age were to affect their perception of the ambiguous 'my wife and my mother-in-law' gure, this would constitute evidence that older participants' higher age estimates of the ambiguous gure result from an own-age social group bias. If not, the own-age anchor effect better explains the difference in the estimates of the ambiguous gure's age.

Participants
An a priori power analysis was conducted using the R package pwr [34] to determine the required sample size to test the mean difference in age estimation between two independent groups using a one-tailed test with an α of .05, and assuming the original effect size (Cohen's d = . 39). Results showed that a sample of 232 participants, comprised of two equally sized groups of n = 116, was required to achieve a power of .90. In order to compensate for the potential exclusion of participants, a total of 260 participants was recruited. An analysis of participants' eligibility for inclusion (see below for criteria) resulted in the exclusion of 14 participants, reducing the nal sample to N = 246.
All participants were adult US citizens (f = 135, m = 111) recruited using Proli c Academic and compensated with $0.30. In order to make sure that the young and older group had a similar number of participants, the survey was run twice; once for people of 30 years or younger, and once for people older than 30. The mean age of the young group was 23.73 years (SD = 3.28, n = 124), while the mean age of the old group was 44.98 years (SD = 12.35, n = 122). All participants answered demographic questions regarding their age, sex and nationality The experiment took on average 83.55 seconds to complete.

Materials and Procedure
The experiment was set up to approximate the original study by Nicholls et al. as closely as possible.
There are three noteworthy deviations from the original (see below for details). After estimating the age of the woman in the 'my wife/mother-in-law' gure, participants were presented with two additional gures highlighting the young woman and older lady to have them indicate which of the two images they perceived. Participants were also asked whether they had seen the ambiguous gure prior to the experiment and invited to estimate the age of a computer generated face. The latter two changes were respectively included to explore the effect of familiarity with the ambiguous gure on age estimation and to establish the generalizability of the original ndings.
The experiment was implemented in the software Qualtrics version May 2020 (Qualtrics, Provo, UT). After providing informed consent and indicating their age, sex, and nationality, participants were shown a screen which asked them to pay close attention to the gure on the next screen, as it would only be shown for a short period of time. After clicking to the next page, a copy of the original 'my wife and my mother-in-law' gure was shown for 500 ms (Fig. 1). Subsequently, participants' eligibility was tested using two questions. First, participants were asked to indicate if they saw a person, an animal, or neither.
Participants who responded 'person', were then asked if that person was male or female. When a participant answered one of these questions incorrectly, the experiment was terminated immediately. Otherwise, participants indicated whether they had seen the gure prior to the experiment, after which they were asked to estimate the person's age in whole numbers.
A text was then shown that stated that the gure consisted of two different women. The participants were told that on the next screen, two gures would be shown, each highlighting one of these two women (Fig. 2). The order in which the two percepts were highlighted in the gure was counterbalanced across all participants. That is, for half of the participants, the young woman was depicted in the left panel (panel A), while for the other half, the young woman was depicted in the right panel (panel B). Participants were asked to indicate which of the two women they had seen. Based on a study by Georgiades and Harris [35], critical features were removed or highlighted in order to enable participants to clearly discriminate between the two percepts. From panels A and B in Fig. 2, for instance, the eye of the older lady, and the nose of the young woman were removed, respectively. These modi ed gures were originally presented by Shakhnazarova in the newspaper The Sun [36].
After completing this part of the experiment, participants were told once again to pay attention to the gure on the next page, as it would only be presented for a short period of time. This time, a computergenerated face was shown for 500 ms. The face was synthesized using the FaceGen Modeller Software (Singular Inversions, 1998) to represent an average thirty year-old Caucasian female. The participants were once more asked to estimate the age of the face in whole numbers, after which they were thanked for participating.

Results
All analyses were performed with the statistical software R version 3.6.1 [37] using the packages dplyr [38], ggthemes [39], ggplot2 [40], lsr [41], and sjstats [42]. An α = .05 was employed in all analyses. The data and the R script are available on osf.io/y3bqa/. Here, we only report the results of the pre-registered analyses pertaining to the own-age social group bias and the own-age anchor effect. The results of the pre-registered exploratory analyses looking into the effect of familiarity with the ambiguous gure and the generalizability of the ndings to a computer generated face are supplied as Supplementary Information.

Replication of Original Findings
In order to replicate the original ndings of Nicholls et al. [6], a one-tailed independent samples t-test was performed to examine the difference in mean age estimates of the ambiguous gure between the age groups. Older participants were found to estimate the ambiguous gure signi cantly older (M = 41.04, SD = 17.51) than young participants (M = 33.50, SD = 14.56); t(234.76) = 3.67; p < .001, Cohen's d = 0.47. On average, older participants thus estimate the person in the 'my wife/mother-in-law' gure to be 7.54 years older than young participants do. A one-tailed Pearson correlation was computed between participants' own-age and their age estimates of the ambiguous gure. Results indicated that there was a signi cant positive association between participants' age and their age estimates; r(244) = .22, p < 0.001. Similar results were obtained in the original experiment of Nicholls et al., who found a mean age difference of 6.32 years and a correlation of .24.
In the original paper, the initial analysis was repeated with the 10% youngest and 10% oldest participants.
In the case of the current experiment, this would cause some 21-year old and some 57-year old participants to be excluded from the analysis on an arbitrary basis. Therefore, 8.94% of the youngest and oldest participants were selected for the analysis. This resulted in a sample of n = 44 with 22 participants in each age group. The youngest participants had a mean age of 19.14 years (SD = .83) and the oldest participants had a mean age of 65.77 (SD = 5.45). A one-tailed independent samples t-test showed the 96, demonstrating that the results are not a consequence of the arbitrary age split, and that the difference in the mean age estimates was 6.24 years greater when comparing the oldest with the youngest participants instead of the original age groups.

Social Group Bias or Anchor effect?
Participants were assigned to two different percept groups based on whether they reported seeing the young woman or the older lady. The older lady was perceived by the majority of the participants. Only 114 out of 246 participants (46.34%) reported perceiving the young woman. Another one-tailed independent samples t-test was performed to compare the mean difference in the estimated age of the ambiguous gure by the young percept group (M = 32.89, SD = 13.24) and by the old percept group (M = 41.00, SD = 18.09). This difference was found to be signi cant: t(237.81) = 4.05, p < 0.001, Cohen's d = .51, indicating that the older lady was perceived to be signi cantly older than the young woman. This difference is to be expected if participants can reliably indicate which of the two clearly differently aged women they perceived in the ambiguous picture.
A chi-square test was performed to determine whether the frequency of the reported percepts differed between the age groups. The relationship was not signi cant (χ 2 (1) = .01, p = .91), suggesting that participants' age does not systematically in uence how they perceived the gure. Therefore, the relationship between participants' own age and their age estimates of the ambiguous gure does not appear to be a consequence of the own-age social group bias.
To determine whether the relationship between own age and estimated age was independent of the woman that was perceived in the ambiguous gure, a one-tailed independent samples t-test was conducted for each percept group independently. A signi cant difference in age estimates was found A one-tailed Pearson correlation between participants' own age and the age they estimated the ambiguous gure to be was calculated for each of the percept groups. Figure 3 summarizes the results. It shows a signi cant positive association between participants' own-age and their age estimates within both percept groups; r(112) = .34, p < 0.001 in the young percept group (black), and r(130) = .17, p = 0.03 in the old percept group (gray). That is, participants provided higher age estimates for the ambiguous gure the older they were, regardless of whether they perceived the young woman or the older lady.
Taken together, these results provide support for the hypothesis that the positive relationship between participants' own age and their age estimates of the 'my wife/mother-in-law' ambiguous gure are best explained by the own-age anchor effect.

Discussion
One's age has been shown to affect how old one estimates the 'my wife and my mother-in-law' ambiguous gure to be [6]. In this study, we replicated the nding that older participants estimate the woman in the gure to be signi cantly older than young participants do. Contrary to what was originally suggested, we take this age difference to re ect the own-age anchor effect, not a social group bias toward processing similarly aged faces.
If the difference in age estimation were due to the social group bias, we would expect a propensity among older participants to report seeing the older lady and a propensity among young participants to report seeing the young woman. We did not nd this to be the case. The proportion of times the wife and the mother-in-law were reported was comparable in the two age groups. This makes it unlikely that a recognition bias toward age in-group members is driving the relationship between participants' own age and their age estimates of the 'my wife and my mother-in-law' ambiguous gure. After conducting the study, we learned about a similar result that was already obtained by Botwinick, Robin, and Brinsley in 1959 [43]. They presented male participants with the 'my wife and my mother-in-law' ambiguous gure and asked them to report what they saw within 90 seconds. Botwinick et al. found that of all participants who reported either of the percepts, 76% of the young participants and 94% of the older participants reported seeing the young woman, which contradicts the predicted pattern were a social group bias in play.
Previous research has shown that the interpretation of an ambiguous gure is dependent on the xation point and the critical features of the imagine that are attended to [35,44]. Fixating a particular point of an ambiguous gure might make on process features that are critical for eliciting one interpretation of the gure, while xating on other critical features will result in the other interpretation. The absence of an age difference in the reported percept seems to suggest that there were no systematic age differences in the manner in which the ambiguous gure was processed (as opposed to the systematic age differences in gaze that have been established for emotional face perception [45]) or that these were overridden by differences in the various applications on which the online survey were taken and the corresponding differences in the size of the retinal image.
Our ndings are more in line with the own-age anchor effect, according to which people tend to assimilate toward their own age, when estimating someone else's age [24,25,26,27,28]. The own-age anchor effect does not require participants' perception of the ambiguous gure to be in uenced by their own age. It predicts a positive association between participants' own age and their age estimates, irrespective of the woman they see. In other words, there should be a difference in age estimation between younger and older participants even if they perceive the same woman. This is what we found.
A nal result that appears to be more in line with the own-age anchor effect than with the social group bias, is the observation that the age estimates for the ambiguous gure are rather low, even in older participants reporting seeing the older lady. The older lady in the ambiguous gure is supposedly intended to be older than the average age of 44.71 years reported by this group of participants. It strengthens our opinion that participants' age estimates are driven by a cognitive bias to use one's own age as a yardstick, rather than high-level social processes subconsciously affecting face perception. The brief presentation of an ambiguous gure might just constitute the uncertain circumstances in which the own-age anchor effect is likely to occur. When people have an age judgment to make, but little information is available to base that judgment on, it not illogical for them to anchor it on a normative value such as their own age, which is salient and likely to be representative for the majority of people they interact with.
While we interpret the observation that older participants estimate the 'my wife and my mother-in-law' ambiguous gure to be older than young participants do in terms of the own-age anchor effect, we nevertheless believe that social in uences might play a role as well. Speci cally, people might only assimilate other characteristics to their own, if these others are considered part of the in-group. Sörqvist and colleagues [26] showed that women demonstrate assimilation toward their own age when estimating the ages of male targets, but not vice versa. The authors speculated that this difference may result from men's tendency to categorize others as an in-group or out-group member based on gender, while women do not. The ndings by Sörqvist et al. would then be the result of women having larger age in-groups than men. That is, women's age in-groups may include men, whereas men's age in-groups do not include women. In the context of ambiguous gures, this could be further investigated by comparing the extent to which the own-age anchor effect presents in men and women's age judgments of the 'my wife and my mother-in-law' ambiguous gure, and a comparable ambiguous gure featuring differently aged men, such as the 'my husband/father-in-law' ambiguous gure [46]. The current study pertained to age estimates by US citizens. Future research could examine the own-age anchor effect in populations with different cultural backgrounds to investigate the generalizability of the effect. Individuals from individualistic and collectivist cultures might perform differently, for instance. Speci cally, in individualistic cultures emphasis is placed on the individual, therefore it makes sense to expect individuals to regard themselves as the yard stick when judging others. However, individuals from collectivist cultures are taught to prioritize the group and their relationship with group members. This might make them less likely to consider themselves as a reference. In other words, individuals from collectivist cultures might be less susceptible to the own-age anchor effect than individuals from individualistic cultures. In general, in future work both the mechanisms underlying the own-age anchor effect, as well as its rami cations should be further investigated. The effect has several practical implications, for instance in the context of eyewitness testimonies and the provision of age-restricted services.

Data Availability
The data that support the ndings of this study are openly available on the Open Science Framework at osf.io/y3bqa/. The study was pre-registered (osf.io/xqc35). The data used in this article are licensed under a Creative Commons Attribution 4.0 International License (CC-BY), which permits use, sharing, adaptation, distribution, and reproduction in any medium or format, as long as you give appropriate credit to the original authors and the source, provide a link to the Creative Commons license, and indicate if changes were made. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. Figure 1 My Wife and My Mother-In-Law, by the cartoonist W. E. Hill, 1915. This media le is in the public domain in the United States. This applies to U.S. works where the copyright has expired, often because its rst publication occurred prior to January 1, 1923.