Facial femininity and perceptions of eating disorders: A reverse-correlation study

Valerie Douglas; Benjamin Balas; Kathryn Gordon

doi:10.1371/journal.pone.0255766

. 2021 Aug 6;16(8):e0255766. doi: 10.1371/journal.pone.0255766

Facial femininity and perceptions of eating disorders: A reverse-correlation study

Valerie Douglas ¹, Benjamin Balas ^1,^*, Kathryn Gordon ²

Editor: Kamila Czepczor-Bernat³

PMCID: PMC8345843 PMID: 34358270

Abstract

Eating disorders are prevalent in college students but college students are not accurate in identifying the presence of eating disorders (ED) especially when race is involved. Much has been researched about diagnostic ability in vignette form, but little outside of this. For example, it is not known how facial features, such as perceived femininity, may affect observers’ beliefs about the likelihood of disordered eating depending on race. In the present study, we examined how biases regarding facial appearance and disordered eating may differ depending on the race of face images. Using a technique called reverse correlation, we estimated the image templates associated with perceived likelihood of disordered eating using both White and Black Faces. Specifically, we recruited 28 college students who categorized White and Black faces according to perceived likelihood of an eating disorder diagnosis in the presence of image noise. Subsequently, we asked Amazon Mechanical Turk participants to categorize the resulting race-specific face templates according to perceived ED likelihood and femininity. The templates corresponding to a high likelihood of an ED diagnosis were distinguished from low-likelihood images by this second independent participant sample at above-chance levels. For Black faces, the templates corresponding to a high likelihood of an ED diagnosis were also selected as more feminine than low-likelihood templates at an above-chance level, whereas there was no such effect found for White faces. These results suggest that stereotyped beliefs about both femininity and the likelihood of disordered eating may interact with perceptual processes.

Introduction

There are three main eating disorder (ED) diagnoses: anorexia nervosa, bulimia nervosa, and binge eating disorder [1]. Subclinical ED symptoms that do not fully cross the clinical threshold for a disorder can also be problematic [2] One particularly vulnerable group for the development of disordered eating symptoms is college-aged individuals, with an increased rate from 1995 to 2008 [3]. This is troubling considering that college students do not reliably recognize the symptoms of EDs, the need for treatment when presented with ED vignettes [4], and do not consistently identify what constitutes disordered eating [5].

This ability for college students to recognize EDs or the seriousness of EDs is further complicated by interference from stereotypes of EDs such as racial stereotypes [4, 6]. Besides affecting untrained college students, research has also found that clinicians tend to underestimate the severity of ED symptoms among African-American girls [7]. College students, who have not had training on diversity in mental illness, may be prone to a racial bias as well [6], although this effect was not replicated in a more recent study [4]. One aspect of racial stereotypes that has not been investigated in regard to EDs is how someone’s facial appearance may influence the perceived likelihood of an ED separate from a vignette. That is, what beliefs do people have about what someone with ED looks like? Do these beliefs differ by race? While it may seem odd to speculate on stereotyped beliefs that observers may have about facial appearance and eating disorders, there is a large (and growing) literature describing how individuals make judgments regarding the personality characteristics and expected behavior of others based on facial appearance. Naïve observers make consistent judgments about what characteristics make faces look more trustworthy or dominant [8], more competent in leadership roles [9], and make consistent judgments regarding the mental health of unfamiliar individuals [10]. Many of these judgments are not valid: Perceived trustworthiness from face images does not reflect the likelihood that an individual will behave in a trustworthy fashion, for example [11]. For our purposes, however, it is observers’ beliefs about correlations between facial appearance and the likelihood of disordered eating that are of interest: What biases do people have about the appearance of people with eating disorders and do these biases vary by the race of the people under consideration? Given that EDs are more likely to be diagnosed in females than in males [1], we hypothesized that observers would believe that faces that looked more feminine would also be more likely to have an ED diagnosis. Further, we hypothesized that this tendency might be affected by the race of the face under consideration due to additional observer biases affecting how femininity is perceived across racial categories.

We chose to test these hypotheses using a technique called reverse correlation. Reverse correlation (RC) is a powerful technique for revealing the internal representations that support perceptual judgments. Originally applied to low-level psychophysical research [12, 13], Reverse correlation designs have since been widely used to characterize the internal templates used in face categorization tasks. All image-based reverse correlation experiments share some common properties: Observers are typically presented with a stimulus image (sometimes called the base image) multiple times with some form of noise applied to it on each presentation so that each presentation is unique in terms of the combination of the stable original image and a noise pattern that is generated randomly on each trial. Examples of noise that may be added to the original stimulus images include the addition of white noise to each pixel in the image, or the application of “Bubbles” [14], which appear as randomly placed windows in a field of black pixels through which the underlying base image is visible. Presented with the combination of a base image and randomly generated noise, observers categorize each stimulus: Labeling the emotion expressed by a face, for example. These responses are used to analyze the noise patterns applied during the experiment according to how they were categorized by participants. Noise patterns that led observers to assign images to a particular category (“Happy”) are grouped together and separated from noise patterns that led observers to assign images to different categories (“Sad”). By averaging together all of the noise patterns that led to a particular judgment (e.g. “Happy”), the experimenter can obtain an image that possibly reflects how some properties of the noise correlated with perceived membership in that category. That is, on average, what noise do we have to add to a neutral face so that it looks happy, or like a particular person [15]? Applying that aggregate noise pattern to the base image used in the experiment results in a classification image (or CI), which reflects the average stimulus appearance that led to each category judgment. This classification image can be interpreted as a sort of perceptual template or internal prototype of the category under investigation.

This technique can reveal what parts of a face support different tasks like gender categorization or identification [16], or it can be used to examine other aspects of face categorization. For example, by varying the base images to which noise is applied (e.g. a familiar face vs. a stranger), we can determine how the visual information used to make categorization judgments may vary from stimulus to stimulus [17]. Similarly, we may also vary properties of the observer to see how variation across participants changes the visual information used for categorization judgments. Dotsch et al. [18], for example, found that prejudice towards Moroccan individuals affected the classification image that resulted from a race categorization task such that the Moroccan classification image looked angrier when generated by individuals with higher levels of prejudice. Whether or not these classification images truly reflect perceptual mechanisms supporting these inferences or instead illustrate how both perceptual and cognitive mechanisms contribute to the criteria used to make decisions about face images is challenging to determine unequivocally. Presently, we treat these images as a useful means of demonstrating what aspects of appearance are linked to categorization judgments as a way of examining how face race may affect the expectations or stereotypes naïve observers have about the appearance of individuals with eating disorders.

An important point to acknowledge is that reverse-correlation as it is being applied here does not establish any link between physiognomy and disordered eating behavior. That is, the templates we obtain from this technique do not reflect real correlations between facial appearance and behavior, personality characteristics, or other individual differences. A classification image estimate of an observer’s “Happy” face template does not imply that all happy faces actually do look this way, for example, but instead reflects the template that this observer used to categorize faces as such. While one could conceivably attempt to measure correlations between facial appearance and disordered eating in a large sample of real individuals, such physiognomic research is ethically questionable (see [19] for a discussion of this point) and can reinforce stereotyping and prejudicial beliefs about what individuals with specific disorders or behaviors look like. By contrast, our use of reverse correlation in the current study is designed to reveal what biases naïve observers may have with regard to disordered eating. We do not endorse such biases as valid. Instead, we argue that identifying and characterizing such biases is an important step towards understanding how real individuals with and without disordered eating may face differential treatment (clinical and otherwise) based on race-specific biases that observers maintain about facial appearance and behavior. For example, if such biases indeed affect the way observers estimate the likelihood that an individual has an eating disorder, this could lead individuals with disordered eating to be discouraged from seeking treatment either by family, friends, or health-care professionals. To the extent that such biases interact with perceived race, this may in turn increase disparities in referrals and treatment for individuals with disordered eating. We applied this technique to determine how race category (White vs. Black) affected the templates used to assess the likelihood of an eating disorder diagnosis based on facial appearance. In particular, we examined whether or not perceived femininity was differentially linked to perceived likelihood of having an eating disorder diagnosis as a function of race.

Experiment 1 –Do faces that look likely to have an eating disorder look more feminine to naïve observers?

In our first experiment, we recruited two samples of participants: (1) A group of observers who completed a reverse correlation task using images of White and Black female faces. (2) A group of workers recruited via the Amazon Mechanical Turk who provided forced-choice judgments using the classification images created by the first group of observers.

Methods

Participants

We recruited 28 participants (13 female) between the ages of 18–23 to take part in this experiment. All participants were recruited from the North Dakota State University (NDSU) Undergraduate Psychology Study pool and received course credit for volunteering. All participants were naïve to the purpose of the experiment, reported having either normal or corrected-to-normal vision, and gave written informed consent prior to the experimental session. Given our interest in understanding the perceptual biases maintained by the general public with regard to race and disordered eating, we did not include individuals in this sample who had clinical experience with individuals with an eating disorder diagnosis. This is an interesting question in its own right, as clinicians may maintain their own set of different perceptual biases regarding facial appearance and disordered eating behavior that would be important to understand in the context of practitioner-patient relationships. At present, however, we opted to examine perceptual biases in an inexperienced sample to more broadly characterize what perceptual templates naïve observers may use to make judgments about disordered eating. Finally, we note that our sample was predominantly White individuals (26 White participants, 2 Black participants), which limits our ability to comment on how observer race may affect our results. Due to the small number of Black participants, we do not include observer race as a covariate in any of our analyses and refrain here from offering comparisons between the responses of White participants and Black participants due to the low statistical power for doing so. As such, our results should be interpreted primarily as reflecting the biases of White observers, with unknown generality of these results as a function of observer race.

Stimuli

We used two 400x400 pixel grayscale “Average” female faces (White and Black) as base images for our task (Fig 1). These images were created using commercial software (Face Gen, Singular Inversions) that includes the capability to manipulate 3D models of faces of multiple races, including an average face for each race.

Because we used FaceGen faces as stimuli in both experiments reported here, it is important to discuss how the underlying stimuli (average faces for race and gender categories) were created within the software. Briefly, the FaceGen model is based on high-resolution 3D scans of ~270 real human faces including men and women from varying racial categories. These scans include both 3D shape information describing the geometry of the head and face as well as 2D albedo information describing how pigmentation varies across the surface of the skin. These measurements have been used to develop parametric models of face shape and face pigmentation (a “face space”) that account for the variance in those head models. That is, given this particular sample of faces described in terms of their 3D shape and 2D surface properties, this compact dimensional model of the data based on the faces makes it possible to assign a small number of spatial coordinates to each face such that faces can be meaningfully characterized in terms of their distribution within this space. Random FaceGen faces can then be sampled by gender, race, or other categories by randomly sampling from the distributions that correspond to those categories. Average faces (such as the faces described above) can be sampled from each category by identifying the point that lies at the center of the distribution of faces belonging to that category. The images created by this procedure thus are limited by the characteristics of that sample, but nonetheless are based on real people and their appearance variability. As such, these average faces reflect differences in shape and surface pigmentation that correspond to category differences in face appearance as reported by the individuals who contributed to the scanned dataset. While this procedure certainly is limited in scope by the nature of the training corpus, we suggest that these capture sufficiently meaningful and representative appearance differences to be useful for our purposes.

Procedure

Participants completed a 2AFC categorization judgment using the images described above. On each trial, participants were presented with either two versions of the “Average White” image or two versions of the “Average Black” image, presented to the left and right of the center of the screen. Critically, these two images were created by adding two different noise patterns to the target face, resulting in appearance differences between the two stimuli. Participants indicated on each trial which face looked more likely to depict someone who had an eating disorder diagnosis. We chose to ask this fairly broad question about eating disorder likelihood rather than asking participants to categorize faces according to a more specific diagnosis (e.g. anorexia nervosa or bulimia) for two reasons. First, given that our sample was comprised of naïve observers, we anticipated that many of our participants might not be familiar with these more specific terms and we wished to avoid providing descriptive information prior to completing the task for fear of influencing the criteria they might use to respond to our stimuli. Second, we felt that phrasing the question in broad terms might encourage participants to think more generally about the mental health of the faces they were presented with, which we think is potentially more representative of a layperson’s understanding of the topic.

These noise patterns were created online using custom routines in Matlab–one pattern was made by generating random 1/f fractal noise, and the second image was made by “negating” the first: Light pixels in one image corresponded to dark pixels in the other, and vice-versa. This maximizes the potential for the two images presented on each trial to look different from one another (Fig 1B). Participants used the “z” and “m” keys on the keyboard to choose the left or right face, and both their reaction time and the noise patterns applied to the target face on each trial were recorded.

Participants completed 500 trials in a single session (250 trials per race), with race category pseudo-randomized across trials for each participant. We presented the task on a 1200x800 pixel LCD monitor, and all stimulus display and response collection routines were controlled by custom software written using the Matlab Psychophysics toolbox [20, 21].

All procedures described above (and in Experiment 2) were approved by the North Dakota State University Institutional Review Board (#SM11168). Written informed consent was obtained from all participants.

Results

We computed each participant’s CI corresponding to the average appearance of White faces and Black faces that were selected as looking more likely to have an eating disorder diagnosis (we will call these “+ED” images). We calculated these by averaging together the noise patterns added to the faces participants selected on each trial and adding that average noise pattern to the original target face. We also carried out the same process for the unselected images, yielding the average appearance of White and Black faces that looked less likely (“-ED” images) to have an eating disorder diagnosis (Fig 2).

Fig 2 — These represent the average perceptual template for faces that looked likely to have an eating disorder diagnosis (+ED) and those that looked less likely to have such a diagnosis (-ED), calculated by averaging together the individual templates estimated for each observer.

In some reverse correlation studies, researchers determine which regions in the image significantly correlate with a categorization decision [22]. Our goal, however, was to determine how classification images varied by race category according to their apparent femininity. To do so, we presented all of the individual classification images generated by our participants to a new group of online observers via the Mechanical Turk. Specifically, we paired each +ED image with its corresponding–ED image and asked separate groups of online observers to choose the face that looked: (1) More likely to have an eating disorder, and (2) More feminine. The first judgment serves a purely confirmatory purpose: Do the classification images created by our observers based on their perceptual templates for disordered eating look more likely to have an eating disorder diagnosis to independent raters? The second judgment addresses our critical question: How does femininity co-vary with image features related to the perception of disordered eating, and does this relationship vary by race? Again, we emphasize that these relationships do not reflect valid correlations between facial appearance and behavior, but instead reflect the biases maintained by our observers. We also note that we did not collect demographic information about the participants who completed this task using the AMT platform. We expect that this sample of participants differs from the predominantly White college students who completed our first task, but beyond that we cannot make strong statements about the demographics of this group. As such, we again do not make any attempt to include observer race, age, or nationality as a covariate in our analyses, opting instead to simply report the aggregate statistics for the 2AFC judgments used here. In terms of the generality of our results, above-chance classification of the templates generated by our college-student sample by an independent sample of participants recruited from the much broader population of AMT workers to some extent demonstrates that the perceptual judgments made by our more specific sample are fairly robust. At the very least, the different populations contributing to these tasks do not offer an easy account of the data that invalidates our main conclusions.

We collected a total of 560 independent judgments for each question for both White faces and Black faces. While we cannot conclusively rule out the possibility that some AMT participants may have used multiple accounts to make many responses, we treat these judgments as independent categorizations made to each stimulus by individual AMT workers. For both White faces and Black faces, the “+ED” face was chosen as looking more likely to have an eating disorder at above-chance levels (White faces, 56%, p = 0.003; Black faces, 58%, p < 0.001; two-tailed binomial test). The proportion of “+ED” faces selected did not differ as a function of race (Chi-squared = 0.23, df = 1, p = 0.63). In terms of femininity, however, we observed different outcomes as a function of race. For Black faces, the “+ED” face was significantly more likely to be chosen as the more feminine looking face (66%, p <0.001, two-tailed binomial test), but we observed no significant preference for White faces (49.9%, p = 0.97, two-tailed binomial test). Critically, these two proportions did differ as a function of race (Chi-squared = 14.48, df = 1, p < 0.001). All of these analyses were carried out using the JASP statistics application [23].

The classification images created by in-lab participants were thus reliably interpreted by independent observers as looking more likely to have an eating disorder across both White and Black faces, but race affected how perceived femininity covaried with the appearance characteristics associated with disordered eating by our first group of observers. In Experiment 2, we further examined this interaction between race, perceived femininity, and perceived likelihood of eating disorder by directly manipulating the appearance of White and Black faces using the FaceGen model’s estimates of facial femininity.

Image statistics analysis

One concern regarding the perception of the images generated by the reverse correlation is that both the foreground and the background of the images created for Experiment 1 were subject to the addition of noise. This means that while we assume that differences in how the resulting face images are perceived by independent observers result from varying facial appearance, it is possible that characteristics of the background could systematically bias perception as well. This is especially relevant to our judgments of femininity. Female faces differ from male faces in terms of skin luminance and higher contrast between facial features [24], and an image background that is systematically brighter or darker could induce a change in the perceived lightness or contrast of the face pattern through simultaneous contrast effects. To determine if this could be the case in the reverse correlation images our participants generated, we analyzed a set of image statistics extracted from the foreground (face) and background of these images.

To segment foreground pixels from background pixels in each reverse correlation image, we created binary masks for the original White and Black faces used in Experiment 1. We created these masks by selecting the entire background in each original image, inverting that selection so that the entire face was selected instead, and then deleting the face pattern from the image. This yielded images with zero values where the face pattern had been and uniform intensity values everywhere else. Next, we imported these masks into MATLAB along with the reverse correlation images. The masks were used to select either the foreground or background pixels of each image for further analysis. We chose to calculate the mean intensity of the foreground and background of each image, which would allow us to determine if there were systematic differences in how light or dark the background of the reverse correlation images were across categories. Specifically, we characterized the foreground and background luminance by calculating the average pixel intensity (grayscale value) across all pixels in each region. This yields a single foreground and background value per image, which can then be compared across conditions to look for systematic differences.

We analyzed the mean luminance values from the foreground and background separately by testing the hypothesis that the difference between these values for ED+ and ED- images was non-zero. Because these values were constrained to be redundant by our use of positive and negative noise patterns on each trial, we divided the difference between ED+ and ED- values in half before using a single-sample t-test to compare the resulting difference scores to zero. This procedure was carried out using the JASP statistics application [23]. This analysis revealed that neither White faces (t (27) = -1.55, p = 0.132) nor Black faces (t (27) = -2.21, p = 0.035) had systematically darker backgrounds for ED+ face images compared to ED- face images after correcting for multiple comparisons via the Bonferroni correction (reducing the critical alpha value to 0.025). It is evident that there is a trend favoring darker ED+ backgrounds for Black faces, but our analysis suggests that this is not an overwhelmingly large effect (Table 1).

Table 1. Descriptive statistics for background intensity analysis.

race	response	Mean	SD	N
White	ED+	128.6	25.85	28
	ED-	143.7	26.08	28
Black	ED+	137.0	25.34	28
	ED-	156.1	21.67	28

Open in a new tab

The foregrounds of both White and Black face images, however, did differ in luminance across ED+ and ED- images. Specifically, both White faces (t (27) = 3.56, p = 0.001) and Black faces (t (27) = 2.95, p = 0.006) tended to have higher intensity values in ED+ images than ED- images (Table 2).

Table 2. Descriptive statistics for foreground intensity analysis.

race	response	Mean	SD	N
White	ED+	141.1	9.777	28
	ED-	128.3	10.825	28
Black	ED+	122.6	16.566	28
	ED-	104.5	17.500	28

Open in a new tab

This outcome demonstrates that there is a clear effect of image type on the intensity of image foregrounds, but not image backgrounds. Thus, the differences in how femininity and eating disorder likelihood are perceived in these images are primarily due to the appearance of the face itself and not induced by the surrounding region. There are of course a great many other image statistics one can examine besides mean intensity, but the relationship between background and foreground structure are less well understood. The effects of surround contrast on the perception of a central target are complex, for example, and depend on the relative orientation of image structure in the surround and the target (which is not controlled or manipulated here) and also on the relationship between the surround and target contrast [25]. We therefore have chosen to only include this relatively simple image analysis here, and have provided code for carrying out this analysis and the segmentation code to support more complex statistical extraction via the Open Science Framework (https://osf.io/kbmz7/).

Experiment 2 –Do feminized faces look more likely to have an ED?

In our second experiment, we asked naïve observers to determine whether a novel face or a feminized version of that same face looked more likely to have an eating disorder diagnosis. Based on the outcome of Experiment 1, we hypothesized that feminized Black faces would be more likely to be selected as having an eating disorder than feminized White faces.

Methods

Participants

In Experiment 2, we collected data from 400 Amazon Mechanical Turk workers. We elected to use this platform for this task primarily because we only had need of a small number of judgments in this experiment but wished to obtain these judgments from a large pool of participants. As described in Experiment 1, we did not collect demographic data regarding these participants and thus analyze our data only in the aggregate without covariate analyses of observer race or gender. Again, we anticipate that this participant sample is likely more heterogeneous than our original sample of predominantly White college students in Experiment 1, but beyond that do not make strong statements about the generality of the main results.

Stimuli

We used FaceGen to create novel full-color images of White and Black faces via random sampling of the FaceGen model space. Specifically, we drew samples that were constrained to be female, White or Black (according to the distribution of faces used to create the FaceGen database), and slightly more average-looking than FaceGen’s parameters for “typical” appearance. We generated 10 images per race category using these parameters. Next, we manipulated the femininity of each of these faces using FaceGen’s slider-bar interface for continuously varying the appearance of a target face, yielding a total of 40 images (10 original White Faces, 10 original Black faces, and the feminized counterpart of each). Again, this interface facilitates sampling from the dimensional model established by the original laser-scanned faces used to create the FaceGen software. Increasing the femininity of faces within the model corresponds to drawing samples that are further from the center of the distribution of male faces and closer to the distribution of female faces, which can be computed either across all races or with regard to a specific race category. In Fig 3, we display examples of each stimulus type. As we described previously, the parametric space underlying this interface is limited by the characteristics of the sample used to build the FaceGen appearance model. Nonetheless, varying faces according to FaceGen’s sense of femininity does correspond to meaningful appearance variation across male and female faces both in terms of 3D shape and 2D surface properties. As in Experiment 2, we suggest that this makes it a useful choice for our purposes despite the constraints on its generality.

Fig 3 — In each case, an original face was used as the basis for creating a feminized version of the same individual using FaceGen’s model space.

Procedure

We presented image pairs comprised of an original face and its feminized counterpart to Mechanical Turk workers, and asked workers to choose which of the two faces looked most likely to have an eating disorder diagnosis. Participants had unlimited time to make their judgment, and we collected a total of 20 judgments per image pair (a total of 400 judgments).

Results

We found that for both White and Black faces, observers tended to choose feminized faces more than original faces as looking more likely to have an eating disorder diagnosis (White faces = 62.5%; Black faces = 71.5%, two-tailed binomial tests against chance each with p < 0.01). This difference reached marginal significance (Chi-squared = 3.65, df = 1, p = 0 .056), suggesting that there is at most a weak effect of race on the outcome of this judgment.

Discussion

The current study investigated the nature of observers’ perceptual biases concerning the appearance of White and Black faces and the likelihood that individuals have an eating disorder diagnosis. Although there is vignette-based evidence that race affects the rates of being diagnosed with an eating disorder (e.g., [6, 8]), no research has investigated how one’s facial appearance may affect the likelihood of being diagnosed with an eating disorder or being perceived as being likely to have disordered eating. Using reverse correlation methods, we determined sets of Black or White faces that were categorized according to the likelihood of having an eating disorder diagnosis or not and then separately confirmed these judgments with an independent group of observers. In addition, we assessed how perceived femininity of the faces affected the diagnosis of an eating disorder. We found that for Black faces, perceived femininity was associated with the perceived likelihood of having an eating disorder diagnosis. However, no such link was found for female White faces. In our second experiment, directly manipulating femininity to influence eating disorder categorization did not lead to differences in the outcome for Black and White faces, though we did observe a trend favoring a larger impact of feminization on the rate at which feminized Black faces were categorized as more likely to have an eating disorder diagnosis. Our data across the two experiments thus differ, potentially because our second experiment involved manipulations of feminine appearance that were not the same in magnitude as the differences in perceived femininity in ED+ and ED- templates. More careful work to examine the magnitude of femininity differences between faces perceived as more or less likely to have an eating disorder diagnosis may help provide a better sense of how these perceptual variables are related to one another.

Due to eating disorder having a higher prevalence rate for females than males by a factor of between 3:1 and 5:1 depending on the specific eating disorder [1], we predicted that it would be likely that, in general, femininity of the face would be a relevant characteristic for observers’ judgments. However, we did not expect the importance of facial femininity in the determination of an eating disorder diagnosis to vary by race. Perhaps due to the stereotyped belief that White women are most likely to have an eating disorder, femininity may not be the most important marker of this stereotype among White faces. For Black faces, on the other hand, since their race violates the stereotype of who is more likely to have an eating disorder, assessors may tend to focus more on the relevant characteristic they have that goes with the stereotype—being female. As being female becomes the characteristic of focus for Black faces, an increase in facial femininity leads to a stronger association with eating disorder. Critically, there is no evidence that this is at all a valid correlation that should be used to guide diagnosis. Instead, our work suggests that this may be a perceptual bias that affects judgments related to eating disorder diagnosis and should perhaps be taken into consideration when training care providers who make referrals regarding disordered eating or determine interventional strategies for individuals with eating or body image concerns. Future research should focus on assessing other relevant facial features that may be of stronger focus in White faces, as well as determine if facial femininity is also the most important characteristic in faces of other races. Facial attractiveness, perceived health based on facial appearance, and a wide range of other variables that observers estimate from face images (whether these are valid judgments or not) may also make contributions to how expert and naïve observers alike form first impressions of individuals based on facial appearance. In particular, given the role of skin lightness in our observers’ classification images, our work suggests that further examination of how pigmentation cues that observers interpret in terms of physical health and attractiveness may be an important extension of this study. More complex descriptors of skin pigmentation are known to contribute to perceived age, health, and attractiveness [26, 27], and consideration of these features may help elucidate what critical information observers use to make judgments regarding the likelihood of disordered eating behavior in unfamiliar individuals. We also note that eating disorders occur in men as well as women, and applying our approach to an examination of perceptual biases that affect judgments of eating disorder likelihood in male faces could further establish how first impressions of variables like masculinity/femininity [28], attractiveness, and physical health may be further used to make biased inferences about other behaviors.

In general, these results indicate that facial appearance may affect decision-making related to the estimated likelihood of health-related behaviors in others, in this case, disordered eating. The existence of race-based perceptual biases that influence these judgments may lead to disparities in the identification and treatment of eating disorders. Minimizing the impact of these perceptual biases is an important direction for future work, as is continued investigation of other perceptual biases that may affect other category judgments that determine how physical and mental health needs are addressed.

Acknowledgments

Thanks to Dan Gu and Ganesh Padmanabhan for technical support.

Data Availability

All data and materials used for the research described here are publicly available via the Open Science Framework (https://osf.io/kbmz7/).

Funding Statement

This research was supported by NIH (National Institute of Health) grant NIGMS P30 GM1148748. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. One of the authors (Balas) received summer salary from the grant listed in response to part (a) above.

References

1.American Psychiatric Association. Diagnostic and statistical manual of mental disorders (5th ed.). 2013. Washington, DC: Author. [Google Scholar]
2.Hoyt W. D., & Ross S. D. Clinical and subclinical eating disorders in counseling center clients. Journal of College Student Psychotherapy. 2003; 17(4), 39–54. doi: 10.1300/J035v17n04_06 [DOI] [Google Scholar]
3.White S., Reynolds-Malear J. B., & Cordero E. Disordered eating and the use of unhealthy weight control methods in college students: 1995, 2002, and 2008. Eating Disorders. 2011. 19(4), 323–324. doi: 10.1080/10640266.2011.584805 [DOI] [PubMed] [Google Scholar]
4.Sala M., Reyes-Rodriguez M. L., Bulik C. M., & Bardone-Cone A. Race, ethnicity, and eating disorder recognition by peers. Eating Disorders. 2013; 21, 423–436. doi: 10.1080/10640266.2013.827540 [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Czepczor K., Koscicka K., & Brytek-Matera A. Students’ awareness of and attitudes towards eating disorders: Preliminary study. Archives of Psychiatry and Psychotherapy. 2016;1, 48–54. [Google Scholar]
6.Gordon K. H., Perez M., & Joiner T. E. The impact of racial stereotypes on eating disorder recognition. The International Journal of Eating Disorders. 2001; 32(2), 219–224. doi: 10.1002/eat.10070 [DOI] [PubMed] [Google Scholar]
7.Gordon K. H., Brattole M. M., Wingate L. R., Joiner T. E. The impact on client race on clinician detection of eating disorders. Behavior Therapy. 2006. 37(4), 319–325. doi: 10.1016/j.beth.2005.12.002 [DOI] [PubMed] [Google Scholar]
8.Oosterhof N. N., & Todorov A. The functional basis of face evaluation. Proceedings of the National Academy of Sciences of the United States of America. 2008; 105(32), 11087–11092. 10.1073/pnas.0805664105 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Na J., Kim S., Oh H., Choi I., & O’Toole A. Competence Judgments Based on Facial Appearance Are Better Predictors of American Elections Than of Korean Elections. Psychological Science. 2015; 26(7), 1107–1113. doi: 10.1177/0956797615576489 [DOI] [PubMed] [Google Scholar]
10.Slepian M. L., Bogart K. R., & Ambady N. Thin-slice judgments in the clinical context. Annual review of clinical psychology, 10, 131–153. doi: 10.1146/annurev-clinpsy-090413-123522 [DOI] [PubMed] [Google Scholar]
11.Rule N. O., Krendl A. C., Ivcevic Z., & Ambady N. Accuracy and consensus in judgments of trustworthiness from faces: behavioral and neural correlates. Journal of Personality and Social Psychology. 2013; 104(3), 409–426. doi: 10.1037/a0031050 [DOI] [PubMed] [Google Scholar]
12.Ahumada A. J. Perceptual classification images from Vernier acuity masked noise. Perception, 1996; 26, 1831–1840. [Google Scholar]
13.Gosselin F., & Schyns P. G. Superstitious perceptions reveal properties of internal representations. Psychological Science. 2003; 14, 505. doi: 10.1111/1467-9280.03452 [DOI] [PubMed] [Google Scholar]
14.Gosselin F., & Schyns P. G. Bubbles: A technique to reveal the use of information in recognition tasks. Vision Research. 2001; 41, 2261–2271. doi: 10.1016/s0042-6989(01)00097-9 [DOI] [PubMed] [Google Scholar]
15.Mangini M.C., & Biederman I. Making the ineffable explicit: Estimating the information employed for face classifications. Cognitive Science. 2004; 28, 209–226. [Google Scholar]
16.Schyns P. G. Bonnar, & L. Gosselin F. Show me the features! Understanding recognition from the use of visual information. Psychological Science. 2002; 13, 402–409. doi: 10.1111/1467-9280.00472 [DOI] [PubMed] [Google Scholar]
17.Gunaydin G, & DeLong J.E. Reverse Correlating Love: Highly Passionate Women Idealize Their Partner’s Facial Appearance. PLoS ONE. 2015; 10(3): e0121094. doi: 10.1371/journal.pone.0121094 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Dotsch R., Wigboldus D. H. J., Langner O., & Van Knippenberg A. Ethnic out-group faces are biased in the prejudiced mind. Psychological Science. 2008; 19, 978–980. doi: 10.1111/j.1467-9280.2008.02186.x [DOI] [PubMed] [Google Scholar]
19.Birhane A. & Guest O. Towards decolonising computational sciences. arXiv. 2020; arXiv:2009.14258. [Google Scholar]
20.Brainard D. H. The psychophysics toolbox. Spatial Vision. 1997; 10(4), 433–436. [PubMed] [Google Scholar]
21.Pelli D. G. The VideoToolbox software for visual psychophysics: Transforming numbers into movies. Spatial Vision. 1997; 10(4), 437–442. [PubMed] [Google Scholar]
22.Chauvin A., Worsley K. J., Schyns P. G., Arguin M., & Gosselin F. Accurate statistical tests for smooth classification images. Journal of Vision. 2005; 5, 1–1. doi: 10.1167/5.1.1 [DOI] [PubMed] [Google Scholar]
23.JASP Team. JASP (Version 0.14.1) 2020. [Computer software].
24.Russell R. Sex, beauty, and the relative luminance of facial features. Perception. 2003; 32(9), 1093–1107. doi: 10.1068/p5101 [DOI] [PubMed] [Google Scholar]
25.Yu C, Klein SA, Levi DM. Cross- and iso- oriented surrounds modulate the contrast response function: the effect of surround contrast. J Vis. 2003;3(8):527–540. doi: 10.1167/3.8.1 [DOI] [PubMed] [Google Scholar]
26.Fink B., Grammer K. & Matts P.J. Visible skin color distribution plays a role in the perception of age, attractiveness, and health in female faces. Evol Hum Behav. 2006; 27, 433–442. doi: 10.1016/j.evolhumbehav.2006.08.007 [DOI] [Google Scholar]
27.Matts P.J., Fink B., Grammer K. & Burquest M. Color homogeneity and visual perception of age, health, and attractiveness of female facial skin. Journal of the American Academy of Dermatology. 2007; 57, 977–984. doi: 10.1016/j.jaad.2007.07.040 [DOI] [PubMed] [Google Scholar]
28.Mitteroecker P., Windhager S., Møller G.B., Schaefer K. The morphometrics of “masculinity” in human faces. PLoS ONE. 2015; 10, e0118374. doi: 10.1371/journal.pone.0118374 [DOI] [PMC free article] [PubMed] [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0255766.r001

Decision Letter 0

Darren Rhodes

26 Aug 2020

PONE-D-20-16005

Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study

PLOS ONE

Dear Dr. Balas,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

May I first place on record my apologies for the delay in the review process. I have taken the decision to proceed with the comprehensive single review below, and will take on a second reviewer in the next round. I have invited an incredible amount of reviewers but in this time of civil upheaval, academics have rightly not been able to find the time over the summer to take on the review. Once again, my apologies.

The work you present is very interesting and a novel analysis applied to facial femininity and the perception of eating disorders. As such, there is the potential to warrant publication in PLOS ONE. I hope your review can consider the points raised below and we can move forward. May I also ask that you consider making your data and analysis code/script publicly available in order for this work to become transparent and open.

Please submit your revised manuscript by Oct 10 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

We look forward to receiving your revised manuscript.

Kind regards,

Darren Rhodes

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Thank you for including your ethics statement:

"This study was approved by the NDSU IRB (#SM11168). Written informed consent was obtained from all participants.".

i) Please amend your current ethics statement to include the full name of the ethics committee/institutional review board(s) that approved your specific study.

ii) Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”).

For additional information about PLOS ONE ethical requirements for human subjects research, please refer to http://journals.plos.org/plosone/s/submission-guidelines#loc-human-subjects-research.

3. Thank you for stating the following in the Acknowledgments Section of your manuscript:

"This research was supported in part by NIGMS P30 GM1148748, which supports the

Center for Visual and Cognitive Neuroscience at NDSU."

We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form.

Please provide an amended statement that declares *all* the funding or sources of support (whether external or internal to your organization) received during this study, as detailed online in our guide for authors at http://journals.plos.org/plosone/s/submit-now. Please also include the statement “There was no additional external funding received for this study.” in your updated Funding Statement.

Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows:

"The authors received no specific funding for this work."

4. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide.

5. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: https://www.youtube.com/watch?v=_xcclfuvtxQ

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: N/A

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: 1. Summary

The research report “Facial Femininity and Perception of Eating Disorders: A Reverse-Correlation Study” applied the approach of the Reverse Correlation (which I find novel and rare) to examine, whether some concrete facial cues are associated with the perception of Eating Disorders (ED). The research focused on the problem of raising ED prevalence in contemporary US College population. It used artificial stimuli from FaceGen database and two distinct groups of raters; US college students and Amazon Mechanical Turk (AMT) workers. Specifically, the research used female facial representations of two racial origins (faces with black and white complexion) and either set of faces with artificial white noise or set of faces manipulated in sexual dimorphism. Authors report a racially-specific association between sexual dimorphism (facial femininity) and the perception of EDs. When noise was not applied to the images, both black and white female facial representations with feminised features were significantly more frequently perceived as suffering from Eating Disorders. However, when noise was applied on non-distorted (non-feminised) female faces, only black female faces that were perceived as ED-positive were at the same time identified as more feminine. The noise patterns that caused non-feminised (unmanipulated) white female faces to be perceived as ED-positive were not significantly associated with noise patterns that made their appearance more feminine. This result was based on two independent datasets of raters (College students and AMT workers). The authors also present an analysis of the effect of an image background and foreground darkness and root-mean-scale contrast on the perception of ED and femininity. The perception might have been affected by the difference in contrast and foreground/background luminance. The authors conclude that the perception of eating disorders is stereotypically associated with femininity only in black female faces. In white female faces, perceived femininity is not a cue to an ED.

The authors have brought convincing arguments that EDs are an issue in current US society. They also cited references to the studies showing that both students and clinicians are inaccurate in diagnosing the EDs in female faces. This accuracy may be further affected by stereotypes based on the sex typicality and ethnicity of patients. People stereotypically think that women of European origin possess eating disorders more frequently. The authors utilised two independent sets of raters (college students and AMT workers). At above-chance levels, the AMT workers and college students agreed on the diagnosis of ED in the stimuli faces. The authors also showed an important difference between white and black faces concerning the association between ED and PERCEIVED femininity. The goal to determine how race category (White vs Black) and perceived/manipulated femininity affect the likelihood of being perceived as a patient with ED-diagnosis has been fulfilled.

2. Primary objections

Nonetheless, author should address the following objections:

(a) In my opinion, authors should explain the evolutionary relevance and adaptive consequences of the fast and accurate perception of cues of an eating disorder in a female face. Any perceiver should struggle to maximise his or her fitness. Accordingly, it is adaptive to assess women’s impaired health and fertility accurately. Eating disorders have been shown to multiply the risk of death and impair fertility [1,2]. Therefore, I would appreciate a review section on the effects of eating disorders on perceived health and attractiveness. Perceived health seems to predict actual health partially [3,4]. Perceived attractiveness and healthiness are also essential cues in mate choice, mate success, and status negotiation [5,6]. People should perceive markers of eating disorders as cues of impaired health and fertility, low social status and unwanted social and mating partner.

Moreover, the noise the authors applied on the faces can be interpreted as an alteration of the texture of facial skin, which seems crucial for the perception of female youthfulness and fertility [7,8]. I think a paragraph on evolutionary and mate-choice consequences of impaired health would highlight the crucial importance of improving the knowledge of perception of eating disorders. It would also "anchor" the research topic in the paradigm of evolutionary psychology.

(b) I do not understand why the authors use the method of adding a "noise" into artificial FaceGen faces to study “how does femininity co-vary with physiognomic cues related to the perception of disordered eating, and does this relationship vary by race?”. I would prefer to use natural faces of ED-positive and ED-negative individuals, get them rated on perceived femininity and presence of eating disorders. Then, I would analyse skin texture and facial shape to identify which cues in facial surface and shape make their appearance more ED-like. Following this, I would manipulate the faces to make them more ED-like and get those manipulated faces rated again. By the proposed attitude, I could identify cross-ethnical differences in various cues’ effect. If I ever wondered, which cues make people perceive someone ED-positive, I would use the methodology described above (texture and shape analysis, natural stimuli). And even if I used the method of "random noise", I would also try to identify, which patterns of noise drive the perception of ED-positivity and femininity! The face texture is of key importance for the perception of youthfulness, attractiveness and healthiness (see above). Nonetheless, authors do not identify which features of the texture drive the perception of ED and femininity. The most prominent "take-home message" is that "some pattern of noise is perceived as more feminine. Unlike in White women, in Black women, this noise is also perceived as associated with eating disorders". Still, the manuscript presents a good starting point concerning studying the perception of facial cues affecting the perception of Eating disorders. With regard to the PLOS ONE policy ("PLOS ONE uses peer review to determine whether a paper is technically rigorous and meets the scientific and ethical standards for inclusion in the published scientific record") the manuscript is in good order. However, authors apply a rare method, withouth further explaination (and justification). Authors might have been completely aware of the benefits of their method to a degree that these benefits are not mentioned (e.g. that it is easier than to get facial photos of people possessing ED-diagnosis). Nonetheless, such method might have been a result of a decision like "lets' use the RC method on the topic of EDs: the method of RC is novel and rare, the topic of ED seems important".

(c) The section “Image statistics analysis” (p. 11-14), is, in my opinion, excessive. The authors attempted to determine the effect of differences of background and foreground darkness and also the effect of root-mean-square contrast of both background and foreground on the perceived attributes. The section seems to me to be based upon request or advice of other (previous) reviewers. Background and foreground may differ in contrast, darkness or both. One reason for face to be perceived feminine (and ED-positive) is to possess a texture (a pattern of foreground noise) that makes features more female-like (while keeping darkness and root-mean-scale of contrast constant in background and foreground). The other reason that some face would be perceived more feminine (and ED-positive) is that the noise in the foreground area was lighter than the noise in the background (or there were differences in RMS contrast between foreground and background). Therefore, I find it unnecessary to run such analyses. Authors should make the section shorter or transfer it into supplementary materials. It seems too long compared with the rest of the paper. It only tells: ‘The noise of the background/foreground of more feminine/ED-positive stimuli may be per se different from the noise in the background/foreground of less feminine/ED-negative stimuli'. It is, in my opinion, an inevitable property of the set of “noises” (and human perception). Authors may e.g. sum the results of these analyses into a table. If they decide to keep the section here, they should add (i) the definition of the “darkness” authors apply (e.g. a reverse scale to the CIELab Lightness?); (ii) the description of the variables (mean ± SD, range). Also, the name of the used statistical programme and a brief description of the analysis should be included in the manuscript (and the analysis code should be provided on-line). Last but not least, interested readers could run much more analyses, and I predict their courage would enable them to run all the analyses they wish (as soon as they will obtain the data), there is no need to tell them so explicitly.

3. Secondary objections

The minor issues are as follows (in order of appearance in the text):

Page 2 (abstract, last four lines): Feminisation is of importance for ED diagnosis; however, when the noise is applied, perceived femininity is unimportant for ED diagnosis in white faces.

On page 3 of the manuscript, authors state that “Subclinical eating disorders symptoms that do not fully cross the clinical threshold for a disorder can also be problematic”. They should cite a relevant source for this claim. I miss a definition of subclinical eating disorders symptoms. As an evolutionary psychologist, I would expect them to exist and affect the perceived traits.

The college students should present future clinicians? Or, is there, due to difficulties with ED-diagnosing, a need for attentive students who recognise there is “something wrong” with a schoolmate?

The introduction, as a whole, seems to propose a “benchmark” study, which utilises the method of reverse correlation in an ecologically relevant context. However, the authors do not discuss (and justify) the psychological context in enough detail (see above).

Experiment 1

What does (on page 7) so-called “face space” stands for? I am aware of face space based upon facial landmarks and geometric morphometric (see, e.g.[9,10]). However, I would highly appreciate if the authors could bring a more detailed description of characteristics of FaceGen face space (e.g. a reference to previous studies that used FaceGen database). Also, the term “average faces” may describe a face that was attained by blending (or morphing) together with a set of faces [11,12]. Does “averageness” in FaceGen follow a similar method to create “average face”? In my opinion, there is no objection against using FaceGen stimuli; however, I cannot decide, whether the FaceGen faces “reflect systematic real-world category differences in face appearance” and whether “these capture sufficiently meaningful and representative appearance differences”. Are those claims based on some geometric morphometrics or other shape analyses? Otherwise, the authors should mention that using FaceGen faces may present a limitation (with regard to actual face diversity in the current US population).

Page 9-11: There is a full description of the analysis, and I think that authors use all the tests properly. Nonetheless, I miss the name of statistical software you use (e.g. “We did the test W using the function xyz in package zyx for R software (CITATION-OF-THE-PACKAGE)”).

Page 10 (and 15): To my knowledge, it is possible to obtain more detailed information about AMT workers who completed the tasks. Authors should describe their country of origin, sex and average age.

The data has been UNAVAILABLE from the link https://osf.io/kbmz7/ (even though I possess an OSF account, and I requested the permission). Therefore, I could not observe the data, calculate the missing variable descriptions (e.g. mean +/- SD and range of darkness and RMS contrast), neither run the binomial, chi-quadrat tests and ANOVAs on my own. This needs to be fixed before the article release.

Experiment 2 (page 15-17)

Authors should define the term “feminised” face in more detail. It may point either to (a) a result of artificial manipulation of faces based on scale connecting male and female facial configurations [9]; (b) a concept of perceived sex typicality, that may also be affected by non-shape cues [13]. Based upon information available, I predict that the authors followed the criteria and scale of femininity defined implicitly within the FaceGen database. However, I miss a definition of this scale: As mentioned previously, a “face space” is a rather diverse conceptual framework, which can be based upon 2D or 3D facial shape, skin contrast, colouration, and the allometric component of sexual dimorphism (and combination of those, like in FaceGen, most obviously). It is important to define the process of “feminisation” in more detail to allow other researchers to reply and extend the research based on other face databases. The statement that “varying faces according to FaceGen’s sense of femininity does correspond to meaningful appearance variation across male and female faces both in terms of 3D shape and 2D surface properties” is not sufficient. Authors should provide either (i) reference to the analysis of the 3D and 2D FaceGen faces’ properties which proves this; (ii) report of thier own analysis to support this claim.

Sometimes, authors do not follow formal rules for reporting the p-values: In the section “Image statistics analysis” (page 12), p = 0.035 is in neither case a “trend”. According to p = 0.035, you are allowed to “reject H0”.

In section “Experiment 2” (page 16-17) I miss a (two-tailed binomial) test to compare the proportions of trials in which feminised faces were chosen as looking more likely to have an ED diagnosis.

In the Discussion (page 17-18) authors should discuss in more detail the fact that artificially feminised faces of both White and Black race were perceived more likely as having an ED diagnosis. When noise applied, only in black faces was there a significant association between perceived femininity and ED-diagnosis. Perceived femininity (based upon noise) and manipulated femininity tell a different story (the latter being not affected by the race of the face).

The language and wordings are easily-accessible.

Finally, I would like to thank you for that the authors brought me an option to read the current version of the report on the study. I hope that the authors have been able to understand all my remarks and questions and that they are able to cope with them to present their research report to the scientific community.

References:

1. Stewart DE, Robinson GE, Goldbloom DS, Wright C. Infertility and eating disorders. Am J Obstet Gynecol. 1990;163: 1196–1199. doi:10.1016/0002-9378(90)90688-4

2. Smink FRE, Van Hoeken D, Hoek HW. Epidemiology of eating disorders: Incidence, prevalence and mortality rates. Curr Psychiatry Rep. 2012;14: 406–414. doi:10.1007/s11920-012-0282-y

3. Rhodes G, Zebrowitz LA, Clark A, Kalick SM, Hightower A, McKay R. Do facial averageness and symmetry signal health? Evol Hum Behav. 2001;22: 31–46. doi:10.1016/S1090-5138(00)00060-X

4. Nedelec JL, Beaver KM. Physical attractiveness as a phenotypic marker of health: An assessment using a nationally representative sample of American adults. Evol Hum Behav. 2014;35: 456–463. doi:10.1016/j.evolhumbehav.2014.06.004

5. Little AC, Jones BC, DeBruine LM. Facial attractiveness: evolutionary based research. Philos Trans R Soc B Biol Sci. 2011;366: 1638–1659. doi:10.1098/rstb.2010.0404

6. Henderson AJ, Holzleitner IJ, Talamas SN, Perrett DI. Perception of health from facial cues. Philos Trans R Soc B Biol Sci. 2016;371. doi:10.1098/rstb.2015.0380

7. Matts PJ, Fink B, Grammer K, Burquest M. Color homogeneity and visual perception of age, health, and attractiveness of female facial skin. J Am Acad Dermatol. 2007;57: 977–984. doi:10.1016/j.jaad.2007.07.040

8. Fink B, Grammer K, Matts PJ. Visible skin color distribution plays a role in the perception of age, attractiveness, and health in female faces. Evol Hum Behav. 2006;27: 433–442. doi:10.1016/j.evolhumbehav.2006.08.007

9. Mitteroecker P, Windhager S, Møller GB, Schaefer K. The morphometrics of “masculinity” in human faces. PLoS One. 2015;10: e0118374. doi:10.1371/journal.pone.0118374

10. Kleisner K, Pokorný Š, Saribay SA. Toward a new approach to cross-cultural distinctiveness and typicality of human faces: The cross-group typicality/ distinctiveness metric. Front Psychol. 2019;10: 1–13. doi:10.3389/fpsyg.2019.00124

11. Langlois JH, Roggman LA. Attractive Faces Are Only Average. Psychol Sci. 1990;1: 115–121. doi:https://doi.org/10.1111/j.1467-9280.1990.tb00079.x

12. Perrett DI, May KA, Yoshikawa S. Facial shape and judgements of female attractiveness. Nature. 1994;368: 239–242. doi:10.1038/368239a0

13. Komori M, Kawamura S, Ishihara S. Multiple mechanisms in the perception of face gender: Effect of sex-irrelevant features. J Exp Psychol Hum Percept Perform. 2011;37: 626–633. doi:10.1037/a0020369

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Vojtěch Fiala

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2021 Aug 6;16(8):e0255766. doi: 10.1371/journal.pone.0255766.r002

Author response to Decision Letter 0

8 Jan 2021

Please see the "Response to Reviews" document that we have uploaded along with our other manuscript files.

Attachment

Submitted filename: ReverseCorr_DouglasBalasGordon_ResponsePLOS.docx

Click here for additional data file.^{(16.2KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0255766.r003

Decision Letter 1

Avanti Dey

27 Apr 2021

PONE-D-20-16005R1

Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study

PLOS ONE

Dear Dr. Balas,

==============================

The manuscript has been further evaluated following your initial revisions, and the reviewers' comments are available below.

The reviewers note that you have largely addressed their previous concerns with sufficient explanation and revision. However, there are still a few minor points which they request that you address, including further explanation in the Discussion regarding the relationship between the current and previous results. They also encourage you to more fully explain terms and abbreviations prior to usage, as well as include all references and maintain a consistent citation style.

Could you please revise the manuscript to carefully address the concerns raised?

==============================

Please submit your revised manuscript by Jun 10 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Avanti Dey, PhD

Staff Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: All comments have been addressed

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Reviewer #1: 1.Summary

Commentary on the revised version of the research report "Facial Femininity and Perception of Eating Disorders: A Reverse-Correlation Study". In the current version, the author did not substantially change the layout of the manuscript; therefore, the research report's description would be very much the same. Briefly, the authors used reverse correlation and FaceGen stimuli database to study whether facial femininity affects the probability that the face would be classified as belonging to a person who suffers from Eating Disorders (ED+ when yes, ED- when not). Authors also studied whether such ascribed characteristic is affected by the race (White or Black) of the face (stimuli). They found that when noise is applied, only faces of the Black race perceived as ED+ are also perceived as more feminine (by an independent group of on-line raters). Therefore, the authors demonstrated that race affects the way raters utilize facial cues when assessing eating disorders. In the second study, however, when they used faces manipulated in facial femininity, both the Black and White race stimuli that were feminized, were more likely classified ad ED+.

The authors highlight the importance of their study as a "cornerstone" of a promising research field. The field would study whether facial appearance may affect the probability ("likelihood") of a person being perceived (and eventually mistreated) as likely or unlikely "victim" of an eating disorder, based solely on their appearance. They show that being of certain origin may affect the probability of being perceived as suffering from a certain illness (ED, in this very case). Authors also took the first step in identifying the traits that serve in such "folk classification" (sex-typicality).

2. Review of revisions

The authors took all my suggestions and objections into consideration and either extended/shortened the manuscript to address them, or replied why I might have been wrong. I appreciate their patience. Therefore, I recommend the paper for acceptance because, after the revisions, I can more clearly see its potential impact. The authors also state their willingness to proceed on this topic.

They added or extended the following sections within the manuscript:

They re-wrote the abstract to emphasize that they studied "perceptual biases regarding facial appearance" and if they affect the diagnosis based on the race of the stimuli face. They also extended the introduction to stress that they studied (racially-specific) bias itself and did not identify the alleged "kernel of truth" in such a classification.

Throughout the description of methods, the authors also extended the section on the FaceGen database. I appreciate that. The extended description substantially improved my image on the utility of the database.

The authors also edited the section on the image analysis. While they kept the section on the effect of image background/foreground colouration, they shortened the section on the effect of root mean square contrast.

In experiment 2, the authors added another short description of the stimuli processing within FaceGen. Authors also added the ethics statement and changed the term "Eating disorder" to ED (and of course, they explained the abbreviation).

Authors solved the technical issue of data-inaccessibility (regardless of whether it was a failure of mine or the OSF itself). They also cited the statistic software they use so that I could check the data and analyses. However, in my opinion, they only provide the JASP files for the Image statistical analysis. For the rest of the analyses, I could find only the raw MTurk data and classification images.

Coming to the Response to Reviews section, authors replied in detail to my notions and suggestions, bringing further justification to their changes (or their decisions not to change the text according to the requests).

I accept their reply on my primary objections: I would like to thank them for explaining the study's actual goal and demonstrating that I was wrong in my suggestions. They explain what it is not a good idea to link the paper to evolutionary psychology. The revised version is much clearer regarding studying perceptual stereotypes based on facial traits and race (see also above).

I also accept that it would be extraordinarily complicated and ethically questionable to use photos of actual patients who suffer from eating disorders. In light of the manuscript's update, my suggestions on the use of different stimuli and study design are no more relevant. Also, the authors are right that reverse correlation (RC) is not a rare technique. My notion regarded just its usage within the field of face perception. Now, it is much clearer that both the goals and the whole "paradigm" of the current research report are much further from evolutionary psychology that I've been thinking. RC, therefore, seems a useful approach within the study of perceptual biases.

They also claim that my suggestion to identify which features of the classification image affect the classification is partly resolved by the Image statistical analyses. Otherwise, they argue that capturing the aspects of texture that led to stimuli classification is above the article's scope of the article and exceeds the capacity and abilities of image analysis.

3. Suggestions:

Therefore, I have only several minor objections, which could, in my opinion, be solved easily:

Page 2: "...their femininity may affect observers' beliefs about their eating behaviour" I would add: "only under specific circumstances (manipulation of stimuli)". -> The abstract has been modified upon my request. My request, however, did not address the difference in results between the first and second experiment—my apologies.

Page 5/References: Authors should add the reference to the following citation: Birhane & Guest, 2020 (currently missing in References).

Page 8: "That is, given this particular sample of faces described in terms of their 3D shape and 2D surface properties, what is a compact dimensional model of the data that captures the correlational structure of the data? The resulting model makes it possible to assign spatial coordinates to each face such that groups of faces can be characterized in terms of their distribution within this space." The second sentence, in my opinion, do not answer the first question. Consider writing something like: "That is, given this particular sample of faces described in terms of their 3D shape and 2D surface properties, the compact dimensional model of the data based on the faces makes it possible to assign spatial coordinates to each face such that faces can be characterized in terms of their distribution within this space." Of course, if I miss the important message of the sentences, keep the original version.

Page 12: "...covaried with the appearance characteristics associated with disordered eating by our first group or observers." Did you mean "first group of observers"?

Page 12/References: Authors should add the reference to the following citation: JASP, 2018 (currently missing in References).

Page 20/References: Authors should add the reference to the following citations: Fink et al., 2006; Matts et al., 2007; Mitteroecker et al., 2015 (currently missing in References).

Throughout the text, authors should unify the citation of the American Psychological Association data (either 2015 or 2013).

Reviewer #2: The authors did an excellent job in addressing reviewer comments. I only have a few minor additional comments.

Abstract

Please briefly define what the reverse correlation technique is in the abstract

Instead of saying “above-change levels” please include the relevant statistics

Introduction

It is good that you explained the reverse correlation technique, but I would do it before you introduce the term. Perhaps refrain from introducing until it is explained.

I found the explanation of the reverse correlation technique to be confusing

Please elaborate why peer-identification of EDs is important and biases may be problematic – e.g., that peers could refer individuals to treatment,

Methods:

I would minimize the use of acronyms – e.g., say MTurk instead of AMT workers, reverse correlation instead of RC

Discussion:

Please discuss in more detail the prevalence in men vs. women with EDs (i.e., what percentage of EDs is in men vs. women? Does differ by the ED?)

Reviewer #3: Thank you for the opportunity to review this manuscript. This is an interesting topic that can be considered by readers. The Introduction section was very well prepared and comprehensively provides what is already known on this subject. The results are very interesting. However, after reviewing the manuscript, I have some minor comments:

1. I would recommend expanding the Discussion section as to how some of the results of the study could be implemented rather than a repeat of what was already discussed.

2. References are missing in several places in the discussion.

3. How these results compare to previous results is not discussed.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Vojtěch Fiala

Reviewer #2: No

Reviewer #3: No

PLoS One. 2021 Aug 6;16(8):e0255766. doi: 10.1371/journal.pone.0255766.r004

Author response to Decision Letter 1

4 May 2021

These responses are also included in the "Response to Reviews" document uploaded along with our revised manuscript files.

Response to Reviews

We would like to thank all three of our reviewers for their thoughtful comments on the latest draft of our manuscript. We were pleased to see that they were very positive about this version of the text and we hope that these revisions have adequately addressed the issues they raised here. Below we have reproduced their narrative comments about the current draft, with our responses in boldface text.

Journal Requirements:

We have reviewed the reference list and incorporated the new citations suggested by our reviewers (noted below in response to their suggestions to do so). Otherwise, we have also corrected reference formatting in a few instances to be consistent with PLoS’ instructions to authors.

Reviewer's Responses to Questions

We have omitted the Reviewers’ responses to these questions in this document because there were no additional comments accompanying their answers. Below, we reproduce their comments to the authors in full, with point-by-point responses to each suggestion that indicate how we have edited the text to incorporate their feedback.

6. Review Comments to the Author

Reviewer #1: 1.Summary

2. Review of revisions

They added or extended the following sections within the manuscript:

Thank you - we’re glad to hear that this provided useful context.

We apologize - these files should be available in our OSF repository at present.

We appreciate the reviewer’s thoughtful commentary about these issues and are glad to hear that our revisions helped clarify the key goals of our study.

3. Suggestions:

Therefore, I have only several minor objections, which could, in my opinion, be solved easily:

Thank you for these additional comments - below we describe how we have edited the text to address the remaining suggestions.

We have revised this as suggested by the reviewer.

Page 5/References: Authors should add the reference to the following citation: Birhane & Guest, 2020 (currently missing in References).

Thank you. We have included this in the revised references.

Thank you. We have adopted this suggestion in the revised text.

Page 12: "...covaried with the appearance characteristics associated with disordered eating by our first group or observers." Did you mean "first group of observers"?

We did. Thank you! This has been corrected in the revised text.

Page 12/References: Authors should add the reference to the following citation: JASP, 2018 (currently missing in References).

We have added this to the references.

Page 20/References: Authors should add the reference to the following citations: Fink et al., 2006; Matts et al., 2007; Mitteroecker et al., 2015 (currently missing in References).

We have added these to the references as well.

Throughout the text, authors should unify the citation of the American Psychological Association data (either 2015 or 2013).

We have corrected these inconsistencies.

Reviewer #2: The authors did an excellent job in addressing reviewer comments. I only have a few minor additional comments.

Abstract

Please briefly define what the reverse correlation technique is in the abstract

We have included a short description of the technique in the revised abstract.

Instead of saying “above-change levels” please include the relevant statistics

Our understanding is that specific statistical results are usually reserved for the main text of the paper, while the abstract is meant to serve as a more concise description of key outcomes. As such, we have edited the abstract to more clearly communicate this result, but only report the statistical tests we used and their outcomes in the relevant results section.

Introduction

It is good that you explained the reverse correlation technique, but I would do it before you introduce the term. Perhaps refrain from introducing until it is explained.

We have edited the text to minimize references to the technique before we explain its use.

I found the explanation of the reverse correlation technique to be confusing

We apologize for the lack of clarity. We have attempted to re-work this part of the manuscript to make it more clear how reverse correlation paradigms work and why it is useful in this case.

Please elaborate why peer-identification of EDs is important and biases may be problematic – e.g., that peers could refer individuals to treatment,

We have attempted to expand upon this point in the revised draft.

Methods:

I would minimize the use of acronyms – e.g., say MTurk instead of AMT workers, reverse correlation instead of RC

We have adopted this suggestion throughout the text.

Discussion:

Please discuss in more detail the prevalence in men vs. women with EDs (i.e., what percentage of EDs is in men vs. women? Does differ by the ED?)

This does vary by eating disorder - we have briefly included some of these statistics in the revised text.

1. I would recommend expanding the Discussion section as to how some of the results of the study could be implemented rather than a repeat of what was already discussed.

Thank you for this comment, we have attempted to briefly offer some speculations along these lines in the discussion.

2. References are missing in several places in the discussion.

We have included these in the revised text.

3. How these results compare to previous results is not discussed.

We have attempted to clarify how these results relate to previous findings, though to some extent this is a relatively new avenue of inquiry regarding how facial appearance and the perception of eating disorder diagnoses may be correlated.

Attachment

Submitted filename: ReverseCorr_DouglasBalasGordon_Response_r2.pdf

Click here for additional data file.^{(72.6KB, pdf)}

PLoS One. doi: 10.1371/journal.pone.0255766.r005

Decision Letter 2

Kamila Czepczor-Bernat

8 Jul 2021

PONE-D-20-16005R2

Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study

PLOS ONE

Dear Dr. Benjamin Balas,

Please submit your revised manuscript by August 31, 2021. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

We look forward to receiving your revised manuscript.

Kind regards,

Kamila Czepczor-Bernat

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #1: All comments have been addressed

Reviewer #4: All comments have been addressed

Reviewer #5: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #1: Yes

Reviewer #4: Partly

Reviewer #5: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #4: Yes

Reviewer #5: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #1: Yes

Reviewer #4: Yes

Reviewer #5: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #1: Yes

Reviewer #4: Yes

Reviewer #5: Yes

**********

6. Review Comments to the Author

Reviewer #1: The research report studied whether the diagnosis (perception) of eating disorder based solely on facial appearance is associated with facial femininity and whether the association is affected by race of the stimuli (Black or White). It used the reverse correlation technique.

In the previous (first) round of revisions, I already appreciated that the authors

(i) had improved and clarified any potentially misleading sections (e.g., explained that they did not study whether ED-diagnosing is anyhow accurate, described that they did not use a perspective of evolutionary psychology, and made clear that for serious reasons, they cannot use real ED-diagnosed individuals);

(ii) had extended section on the description of the reverse correlation technique;

(iii) had extended description on the FaceGen database, too.

After the second round of revisions, I have to confirm that none of those improvements was anyhow affected (worsened) by new modifications. Moreover, I appreciate that the section on the reverse correlation technique, which was good already, has been improved. I will remember that the paper contains such a section, and I would eventually refer to that when asked about the RC technique.

Furthermore, the authors added all missing references into the list and took my suggestions concerning several sentences, which were, in my opinion, misleading or hard to follow. The authors also uploaded potentially missing data to the online supplementary material. It allows following a "flow" of statistical analyses. Nonetheless, I have to declare that I am not an advanced user of the programs and tools that the authors used - JASP, Matlab, and MS Office Access. I am more familiar with R software.

Given that the paper focuses on two topics of increasing importance (eating disorders and racial stereotyping), concisely but sufficiently describes methods, reports and discuss the results correctly, and has been revised with regard to previously identified misleading or incorrect sections, I recommend it to final acceptance with no further revisions required. Apparent discrepancies in used font and poorly placed paragraphs within references would be, in my opinion, resolved by the final typeset in the journal.

Reviewer #4: Basically, the current research seems to be academically meaningful. However, it seems that there are parts that need correction and completion of the research.

First, what is the operational definition of perceptual bias in this study? I wonder how the authors can prove that what they observe in this study corresponds to perceptual bias.

Second, it is hard to understand what it means in reality for ordinary people to look at other people's faces and guess that they have eating disorders. In terms of diagnosis, clinically diagnosing eating disorders has no significant relationship with facial features. In particular, there are many forms of eating disorders, such as anorexia and bulimia, and it seems necessary to carefully explain how this potentially affected the research results.

Third, when participants are asked which of the two is more likely to have eating disorders by looking at the experimental stimulation faces, it is difficult to understand what criteria they made. How can we check if the participants understood the meaning of the task properly? Even if independent observers agreed with the results of in-lab participants, this does not address the question of what criteria they used in the evaluation process. Basically, it is a different task for the ordinary people to perform the task of choosing the more feminine of the two facial stimuli and the task of choosing the one with the potential for eating disorders.

Fourth, it seems difficult to interpret this study as providing evidence that “facial appearance may affect decisionmaking related to the estimated likelihood of health-related behaviors in others, in this case, disordered eating.” If another group of researchers constructed a research design to confirm the evidence that facial appearance can affect decisionmaking related to the estimated likelihood of eating disorders, I wonder if they would construct the same design as this study. The findings may not be a study of decision-making related to the estimated likelihood of eating disorders but simply reflecting stereotypes shared by ordinary people.

In conclusion, although this is an interesting study, it seems necessary to refine the purpose and meaning of the study by considering the limitations of the study.

Reviewer #5: Referee report “Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study” PONE-D-20-16005R2

Note: I use singular author throughout to refer to author/author(s)

Summary

This paper reports on two experiments (with three separate groups of participants involved) that elicited people to make choices regarding subjective evaluations of the appearance of faces, following a prompt. One experiment was a reverse correlation task (choosing between noise-adjusted face pairs according to a prompt), the other asked participants to characterize experimenter-manipulated face images (choosing between original and adjusted face according to a prompt). The faces and prompts varied on three dimensions: eating disorders (ED), gender, and race (white or black).

The author variously frames the main finding (or key hypothesis to be evaluated) as:

• “We hypothesized that observers would believe that faces that looked more feminine would also be more likely to have an ED diagnosis. Further, we hypothesized that this tendency might be affected by the race of the face under consideration due to additional observer biases affecting how femininity is perceived across racial categories.”

• “In particular, we examined whether or not perceived femininity was differentially linked to perceived likelihood of having an eating disorder diagnosis as a function of race.”

• “How does femininity co-vary with image features related to the perception of disordered eating, and does this relationship vary by race?”

The results are then described as:

• “We found that for Black faces, perceived femininity was associated with the perceived likelihood of having an eating disorder diagnosis. In our second experiment, directly manipulating femininity to influence eating disorder categorization did not lead to differences in the outcome for Black and White faces, though we did observe a trend favoring a larger impact of feminization on the rate at which feminized Black faces were categorized as more likely to have an eating disorder diagnosis.”

• “For Black faces, on the other hand, since their race violates the stereotype of who is more likely to have an eating disorder, assessors may tend to focus more on the relevant characteristic they have that goes with the stereotype - being female. As being female becomes the characteristic of focus for Black faces, an increase in facial femininity leads to a stronger association with eating disorder.”

The article is motivated by the observation that a distinct population, college students, on average neither recognizes ED symptoms, nor the need for ED treatment, nor is able to “consistently identify what constitutes disordered eating”. Moreover, college students (?) likely have racial stereotypes of EDs.

Overall Evaluation

The paper is well-written, the methods are appropriate, the analysis is competent. Authors offer a clear and concise presentation. The findings are somewhat mixed: at the end this reader was not sure whether to come away thinking, “race definitely matters in mediating people’s association with ED and femininity” or to come away thinking “probably race does not matter much in mediating people’s association with ED and femininity.”

A significant missing element in the protocol/data is an identifier of race for the experiment respondents: while Black v. white face might condition the correlation between femininity and ED, presumably the race-identification of the participants (as Black or white, or other ascribed or self-ascribed ethnic identities) would matter significantly. Would a population of Black persons likely have a correlation between femininity and ED similar to that of a population of white persons, for images of Black and white faces? This strikes me as unlikely, and yet the identities of the participants are not mentioned (only their gender is noted but not used as a covariate in the analysis). Perhaps there were no Black students in the college sample? The Amazon MechTurk tasks in Experiment 1 and Experiment 2 were done by respondents (unclear how many), and their identity status appears not to be a covariate in the analysis.

A related area of ambiguity (for this reader) was who was expected to be the population about which the research could be generalized. In first paragraph, this appears to be college students, but then the second paragraph mentions clinicians having stereotypes. Then bottom of p. 3 becomes more general, mentioning “people.” On p. 6 and 7 the relevant population is called “naïve observers.” Then in methods, the Experiment 1 RC task was done by 28 college students from North Dakota (no mention of how identity backgrounds might matter except to mention that clinicians were excluded). The two Amazon MechTurk samples were presumably not college students. So it is unclear who constitutes the population of interest. Amazon MechTurk are hardly a random sample of a well-defined population, and author should perhaps consult the expanding literature on how to generalize from Amazon MechTurk samples.

Presentation of results (some questions and suggestions for improvement)

There are essentially three results in the paper.

1. “The [28 pairs of] classification images created by in-lab participants were thus reliably interpreted by independent observers as looking more likely to have an eating disorder across both White and Black faces, but race affected how perceived femininity covaried with the appearance characteristics associated with disordered eating by our first group of observers.”

a. Unclear here whether 560 participants each scored one pair? Or whether 40 Amazon MechTurk participants each scored the 28 pairs? The paper says, “We collected a total of 560 independent judgments for each question for both.” This makes it sound like 560 different participants? Is there no information about these participants?

b. White and Black classification images +ED were correlated with the evaluation of separate raters that +ED were somewhat more likely to have an eating disorder. Black +ED faces were more likely to be perceived as more feminine than –ED pair.

2. The differences in reverse correlation +ED were not due to differences in background or face luminance (and nice validation exercise).

a. There appears to be a mistake in the text: on p. 15 “(t (27) = -2.21, p = 0.035)” so normally this indicates rejecting the null hypothesis, but the text states “nor Black face had systematically darker backgrounds for ED+ face images compared to ED- face images”

b. One might wonder though whether “mottling” of face is/was relevant for ED choice?

3. 400 Amazon Mechanical Turk workers each examined one image pair and rated it. The raters preponderately found that the femininized faces were more likely to have an eating order diagnosis. Black and white faces were only marginally different from each other.

Smaller questions

Fig. 2 – The average classification images – is this average of the averages- just to be clear? Or is it one of the 28 individual CI?

How many of Amazon MechTurk who did task may have been repeat workers? That is, could they not have multiple accounts and so the same person did many responses?

Comment

More modesty about what RC does? “the experimenter can obtain an image that reflects what properties of the noise were correlated with perceived membership in that category.” Could be “the experimenter can obtain an image that might possibly reflect some patterns of the noise correlated with perceived membership in that category.” (the possibly is because some of the correlations might be spurious as there are very large number of patterns (i.e., quite multidimensional) and simple patterns might be correlated when the “true” pattern that generates correlation is more complex.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Vojtěch Fiala

Reviewer #4: No

Reviewer #5: No

PLoS One. 2021 Aug 6;16(8):e0255766. doi: 10.1371/journal.pone.0255766.r006

Author response to Decision Letter 2

10 Jul 2021

Response to reviews:

We would like to thank the editor and the reviewers for their further comments on our revised manuscript. We are pleased to see that two of our reviewers indicated that the manuscript was acceptable for publication in its current form and we have incorporated the additional comments made by all of the reviewers into this current draft. Below, we respond to each of these comments individually to highlight how we have edited the text to address these remaining issues. The original review comments are in quotes and our responses follow throughout.

Reviewer's Responses to Questions

"Reviewer #1: The research report studied whether the diagnosis (perception) of eating disorder based solely on facial appearance is associated with facial femininity and whether the association is affected by race of the stimuli (Black or White). It used the reverse correlation technique.

In the previous (first) round of revisions, I already appreciated that the authors

(ii) had extended section on the description of the reverse correlation technique;

(iii) had extended description on the FaceGen database, too.

Thank you very much for these comments! We are very glad to see that our revisions addressed the comments this reviewer made in previous rounds of review.

Reviewer #4: Basically, the current research seems to be academically meaningful. However, it seems that there are parts that need correction and completion of the research.

"First, what is the operational definition of perceptual bias in this study? I wonder how the authors can prove that what they observe in this study corresponds to perceptual bias."

Our operational definition of perceptual bias in the current study is observers’ tendency to link specific aspects of facial appearance to category labels: In this study, we are specifically interested in the link between perceived femininity (one aspect of facial appearance that is determined in part by pigmentation and contrast relationships in the face) and the categorization of faces according to the likelihood of an eating disorder diagnosis. As to whether or not we can truly “prove” that our results are the result of a perceptual bias, we would argue that proving causal mechanisms for behavioral results is generally extremely difficult. What we offer here (and take more care in the revised text to clarify) is evidence that observers appear to have different expectations about the covariation between feminine appearance and eating disorder likelihood as a function of stimulus race. Whether the causal mechanisms leading to this outcome are uniquely perceptual or cognitive, our task reveals that these expectations affect performance in a perceptual task. In the revised text, we have attempted to be more careful about our conclusions in this regard which we hope adequately addresses the reviewer’s point.

"Second, it is hard to understand what it means in reality for ordinary people to look at other people's faces and guess that they have eating disorders. In terms of diagnosis, clinically diagnosing eating disorders has no significant relationship with facial features. In particular, there are many forms of eating disorders, such as anorexia and bulimia, and it seems necessary to carefully explain how this potentially affected the research results."

As we have described in the revised introduction, observers readily estimate a number of personality variables from face images despite many of these judgments having low validity. Across a large literature, observers have been demonstrated to make reliable judgments about trustworthiness, dominance, competence, and other aspects of mental health from face images, which we think indicates that this is indeed an aspect of face recognition that is relevant to real-world behavior and leads to quantifiable outcomes in a range of tasks that can be used to make inferences about the basis for those judgments. The reviewer is correct that there are not known links between facial appearance and eating disorder symptomology (which we are careful to point out in the text, contrasting our study with what has been dubbed “modern physiognomy” by experts in the ethics of face recognition), but this does not mean observers do not have expectations about correlations between appearance and behavior. We agree with the reviewer that the range of eating disorders could potentially be interesting to consider, but we chose to ask observers to make a judgment about disordered eating considered broadly to avoid the need to explain differences between, say, anorexia nervosa and binge eating disorder that might influence participants’ choices in our task. In the revised text we have edited the text to speak to both of these issues and make clear that it is not evidence of a valid link between facial appearance and disordered eating that we are pursuing, but rather observers’ potential beliefs about the same.

"Third, when participants are asked which of the two is more likely to have eating disorders by looking at the experimental stimulation faces, it is difficult to understand what criteria they made. How can we check if the participants understood the meaning of the task properly? Even if independent observers agreed with the results of in-lab participants, this does not address the question of what criteria they used in the evaluation process. Basically, it is a different task for the ordinary people to perform the task of choosing the more feminine of the two facial stimuli and the task of choosing the one with the potential for eating disorders."

We respectfully disagree with the reviewer on this point. The reverse correlation technique specifically allows us to examine what aspects of facial appearance the observers were using to make this decision: The individual classification images show us what image templates correlate with the ED+ and ED- judgments per observer, and the independent ratings made by our additional participants allow us to examine how high-level variables like femininity also correlate with these judgments. Further, we suggest that the agreement between our two samples of participants regarding the likelihood of an eating disorder diagnosis in ED+ and ED- classification images is indeed a useful demonstration that both participant groups understood the task and understood it in similar ways. We also see no evidence in our data (or from post-testing debriefing) that participants did not understand the task properly: Observers’ classification images were categorized by independent observers at above-chance levels, suggesting that the image features used by both groups to assess the likelihood of an eating disorder were similar, and we also observed differences in specific image statistics across ED+ and ED- categories. Observers thus made reliable image-based judgments that led to measurable effects across our target categories. This 2AFC judgment is thus not any more suspect than other 2AFC judgments about facial appearance, especially when one considers the other examples of social categorization task we now describe in the revised introduction. While we agree with the reviewer that estimating the likelihood of an ED diagnosis is “a different task” than categorizing gender, it is also a different task that still leads to consistent behavior within and between individuals and did not lead our participants to express confusion with task requirements.

"Fourth, it seems difficult to interpret this study as providing evidence that “facial appearance may affect decisionmaking related to the estimated likelihood of health-related behaviors in others, in this case, disordered eating.” If another group of researchers constructed a research design to confirm the evidence that facial appearance can affect decisionmaking related to the estimated likelihood of eating disorders, I wonder if they would construct the same design as this study. The findings may not be a study of decision-making related to the estimated likelihood of eating disorders but simply reflecting stereotypes shared by ordinary people."

Again, we respectfully disagree with the reviewer on this point. First, while we do not know if a hypothetical group of other researchers would come up with the same study design as we have used here, we also do not think this is a meaningful way to judge whether our work provides useful data or not. Different approaches to examine the same issue are valuable and we think this technique offers interesting insights that other designs would not. Indeed, as we describe in the introduction, reverse correlation tasks like this one have been used to establish similar links between facial appearance and various categorization judgments in many different cases. Our study is thus an application of an existing technique to a new problem domain, which we think does indeed address whether observers have expectations about how facial appearance may be related to health behaviors. We apologize, but we also do not understand distinction the reviewer makes between stereotypes shared by ordinary people (specifically with regard to what people who exhibit certain behaviors look like) and a link between facial appearance and decision-making. By the latter, we are simply referring to the outcomes we observed in our study: Observers made reliable decisions about behavior (eating disorders) based on what faces look like, which we think could reasonably be called a stereotype. We have assumed in preparing the revised text that the reviewer may have objected to our particular choice of words, and so we have edited the text to link our original wording to the language of stereotypes suggested by the reviewer. We hope this adequately addresses this point.

In conclusion, although this is an interesting study, it seems necessary to refine the purpose and meaning of the study by considering the limitations of the study.

Reviewer #5: Referee report “Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study” PONE-D-20-16005R2

Note: I use singular author throughout to refer to author/author(s)

Summary

The author variously frames the main finding (or key hypothesis to be evaluated) as:

• “In particular, we examined whether or not perceived femininity was differentially linked to perceived likelihood of having an eating disorder diagnosis as a function of race.”

• “How does femininity co-vary with image features related to the perception of disordered eating, and does this relationship vary by race?”

The results are then described as:

"Thank you for this summary – this indeed largely captures the design and results of our study."

"Overall Evaluation

"A significant missing element in the protocol/data is an identifier of race for the experiment respondents: while Black v. white face might condition the correlation between femininity and ED, presumably the race-identification of the participants (as Black or white, or other ascribed or self-ascribed ethnic identities) would matter significantly. Would a population of Black persons likely have a correlation between femininity and ED similar to that of a population of white persons, for images of Black and white faces? This strikes me as unlikely, and yet the identities of the participants are not mentioned (only their gender is noted but not used as a covariate in the analysis). Perhaps there were no Black students in the college sample? The Amazon MechTurk tasks in Experiment 1 and Experiment 2 were done by respondents (unclear how many), and their identity status appears not to be a covariate in the analysis."

We apologize for this omission – the sample of students who completed the classification image experiment was predominantly comprised of white students, which we include in the revised draft. We agree that including participant race as a covariate would be an excellent analysis, recruiting a sufficient sample of Black students is typically challenging at our institution given the demographics of the student population. With regard to our Mechanical Turk experiments, we chose not to collect race information from these participants. In the revised text we make this explicit, clarify the number of participants in each task, and discuss how observer race could affect these judgments.

"A related area of ambiguity (for this reader) was who was expected to be the population about which the research could be generalized. In first paragraph, this appears to be college students, but then the second paragraph mentions clinicians having stereotypes. Then bottom of p. 3 becomes more general, mentioning “people.” On p. 6 and 7 the relevant population is called “naïve observers.” Then in methods, the Experiment 1 RC task was done by 28 college students from North Dakota (no mention of how identity backgrounds might matter except to mention that clinicians were excluded). The two Amazon MechTurk samples were presumably not college students. So it is unclear who constitutes the population of interest. Amazon MechTurk are hardly a random sample of a well-defined population, and author should perhaps consult the expanding literature on how to generalize from Amazon MechTurk samples."

We apologize for the confusion. First, our brief mention of clinicians was not intended to be a claim about the generality of our results, but rather a broader statement about why understanding the nature of these biases could be relevant to real-world contexts. Reviewer #4, for example, doubted whether or not estimating the likelihood of an eating disorder diagnosis from appearance was even a reasonable task to imagine, and we suggest that the clinical context is an important way to respond to this question. Second, Experiment 1 was indeed completed using a sample of college students, with no particular aspects of “identity background” constrained or measured. While we agree that the use of AMT in our subsequent tasks does represent a different (and likely more variable) population, this is increasingly common across a range of cognitive tasks and we think does not pose any major challenges to our main conclusions. We do agree that it is important to highlight that the judgments made in Experiment 1 were made by a specific participant group (college students in the US) and this may not be representative of the performance we might see in a different or broader group. The revised text includes a discussion of this point. We also discuss what conclusions we can draw from our AMT sample with regard to generality, trying to emphasize a conservative interpretation of our results.

Presentation of results (some questions and suggestions for improvement)

There are essentially three results in the paper.

"a. Unclear here whether 560 participants each scored one pair? Or whether 40 Amazon MechTurk participants each scored the 28 pairs? The paper says, “We collected a total of 560 independent judgments for each question for both.” This makes it sound like 560 different participants? Is there no information about these participants?"

Yes, we asked for 560 independent judgments and did not collect demographic data. While this means we cannot include covariates in our analysis, our results simply collapse across whatever variability is in the data to describe the main results as simply as possible.

". White and Black classification images +ED were correlated with the evaluation of separate raters that +ED were somewhat more likely to have an eating disorder. Black +ED faces were more likely to be perceived as more feminine than –ED pair.

2. The differences in reverse correlation +ED were not due to differences in background or face luminance (and nice validation exercise).

We would like to be cautious in terms of our interpretation of this difference given the multiple comparisons we included here. Applying a Bonferroni correction our data requires an alpha of 0.025 rather than 0.05, which means this result does not reach significance. We have clarified this in the revised text.

"b. One might wonder though whether “mottling” of face is/was relevant for ED choice?"

This is an interesting issue, but difficult for us to address meaningfully using the current data.

Smaller questions

"Fig. 2 – The average classification images – is this average of the averages- just to be clear? Or is it one of the 28 individual CI?"

Yes, this is an average of averages.

"How many of Amazon MechTurk who did task may have been repeat workers? That is, could they not have multiple accounts and so the same person did many responses?"

We do not know a clear way to determine whether or not someone created multiple accounts to carry out different judgments. We suggest this is another reason to be skeptical of our ability to collect meaningful demographic data about our AMT workers and is part of the reason we chose to collect and analyze this data in the aggregate.

Comment

We have added the word “possibly” to address this important point.

________________________________________

Once again we would like to thank our reviewers for their continued commentary on this 3rd revision of our manuscript. We hope that our revisions here have addressed their concerns and appreciate their constructive comments. We feel that the manuscript has been further strengthened by their thoughtful questions and observations.

Attachment

Submitted filename: Response to reviews.docx

Click here for additional data file.^{(144.5KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0255766.r007

Decision Letter 3

Kamila Czepczor-Bernat

26 Jul 2021

Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study

PONE-D-20-16005R3

Dear Dr. Benjamin Balas,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Kamila Czepczor-Bernat

Guest Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #4: All comments have been addressed

Reviewer #5: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #4: Yes

Reviewer #5: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #4: Yes

Reviewer #5: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #4: Yes

Reviewer #5: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #4: Yes

Reviewer #5: Yes

**********

6. Review Comments to the Author

Reviewer #4: Lastly, I would like to advise you to separate the conceptual differences between expectations and bias more carefully and reflect them in the manuscript.

Reviewer #5: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #4: No

Reviewer #5: No

PLoS One. doi: 10.1371/journal.pone.0255766.r008

Acceptance letter

Kamila Czepczor-Bernat

30 Jul 2021

PONE-D-20-16005R3

Facial Femininity and Perceptions of Eating Disorders: A Reverse-Correlation Study

Dear Dr. Balas:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Kamila Czepczor-Bernat

Guest Editor

PLOS ONE

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Attachment

Submitted filename: ReverseCorr_DouglasBalasGordon_ResponsePLOS.docx

Click here for additional data file.^{(16.2KB, docx)}

Attachment

Submitted filename: ReverseCorr_DouglasBalasGordon_Response_r2.pdf

Click here for additional data file.^{(72.6KB, pdf)}

Attachment

Submitted filename: Response to reviews.docx

Click here for additional data file.^{(144.5KB, docx)}

Data Availability Statement

All data and materials used for the research described here are publicly available via the Open Science Framework (https://osf.io/kbmz7/).

[pone.0255766.ref001] 1.American Psychiatric Association. Diagnostic and statistical manual of mental disorders (5th ed.). 2013. Washington, DC: Author. [Google Scholar]

[pone.0255766.ref002] 2.Hoyt W. D., & Ross S. D. Clinical and subclinical eating disorders in counseling center clients. Journal of College Student Psychotherapy. 2003; 17(4), 39–54. doi: 10.1300/J035v17n04_06 [DOI] [Google Scholar]

[pone.0255766.ref003] 3.White S., Reynolds-Malear J. B., & Cordero E. Disordered eating and the use of unhealthy weight control methods in college students: 1995, 2002, and 2008. Eating Disorders. 2011. 19(4), 323–324. doi: 10.1080/10640266.2011.584805 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref004] 4.Sala M., Reyes-Rodriguez M. L., Bulik C. M., & Bardone-Cone A. Race, ethnicity, and eating disorder recognition by peers. Eating Disorders. 2013; 21, 423–436. doi: 10.1080/10640266.2013.827540 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0255766.ref005] 5.Czepczor K., Koscicka K., & Brytek-Matera A. Students’ awareness of and attitudes towards eating disorders: Preliminary study. Archives of Psychiatry and Psychotherapy. 2016;1, 48–54. [Google Scholar]

[pone.0255766.ref006] 6.Gordon K. H., Perez M., & Joiner T. E. The impact of racial stereotypes on eating disorder recognition. The International Journal of Eating Disorders. 2001; 32(2), 219–224. doi: 10.1002/eat.10070 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref007] 7.Gordon K. H., Brattole M. M., Wingate L. R., Joiner T. E. The impact on client race on clinician detection of eating disorders. Behavior Therapy. 2006. 37(4), 319–325. doi: 10.1016/j.beth.2005.12.002 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref008] 8.Oosterhof N. N., & Todorov A. The functional basis of face evaluation. Proceedings of the National Academy of Sciences of the United States of America. 2008; 105(32), 11087–11092. 10.1073/pnas.0805664105 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0255766.ref009] 9.Na J., Kim S., Oh H., Choi I., & O’Toole A. Competence Judgments Based on Facial Appearance Are Better Predictors of American Elections Than of Korean Elections. Psychological Science. 2015; 26(7), 1107–1113. doi: 10.1177/0956797615576489 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref010] 10.Slepian M. L., Bogart K. R., & Ambady N. Thin-slice judgments in the clinical context. Annual review of clinical psychology, 10, 131–153. doi: 10.1146/annurev-clinpsy-090413-123522 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref011] 11.Rule N. O., Krendl A. C., Ivcevic Z., & Ambady N. Accuracy and consensus in judgments of trustworthiness from faces: behavioral and neural correlates. Journal of Personality and Social Psychology. 2013; 104(3), 409–426. doi: 10.1037/a0031050 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref012] 12.Ahumada A. J. Perceptual classification images from Vernier acuity masked noise. Perception, 1996; 26, 1831–1840. [Google Scholar]

[pone.0255766.ref013] 13.Gosselin F., & Schyns P. G. Superstitious perceptions reveal properties of internal representations. Psychological Science. 2003; 14, 505. doi: 10.1111/1467-9280.03452 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref014] 14.Gosselin F., & Schyns P. G. Bubbles: A technique to reveal the use of information in recognition tasks. Vision Research. 2001; 41, 2261–2271. doi: 10.1016/s0042-6989(01)00097-9 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref015] 15.Mangini M.C., & Biederman I. Making the ineffable explicit: Estimating the information employed for face classifications. Cognitive Science. 2004; 28, 209–226. [Google Scholar]

[pone.0255766.ref016] 16.Schyns P. G. Bonnar, & L. Gosselin F. Show me the features! Understanding recognition from the use of visual information. Psychological Science. 2002; 13, 402–409. doi: 10.1111/1467-9280.00472 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref017] 17.Gunaydin G, & DeLong J.E. Reverse Correlating Love: Highly Passionate Women Idealize Their Partner’s Facial Appearance. PLoS ONE. 2015; 10(3): e0121094. doi: 10.1371/journal.pone.0121094 [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0255766.ref018] 18.Dotsch R., Wigboldus D. H. J., Langner O., & Van Knippenberg A. Ethnic out-group faces are biased in the prejudiced mind. Psychological Science. 2008; 19, 978–980. doi: 10.1111/j.1467-9280.2008.02186.x [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref019] 19.Birhane A. & Guest O. Towards decolonising computational sciences. arXiv. 2020; arXiv:2009.14258. [Google Scholar]

[pone.0255766.ref020] 20.Brainard D. H. The psychophysics toolbox. Spatial Vision. 1997; 10(4), 433–436. [PubMed] [Google Scholar]

[pone.0255766.ref021] 21.Pelli D. G. The VideoToolbox software for visual psychophysics: Transforming numbers into movies. Spatial Vision. 1997; 10(4), 437–442. [PubMed] [Google Scholar]

[pone.0255766.ref022] 22.Chauvin A., Worsley K. J., Schyns P. G., Arguin M., & Gosselin F. Accurate statistical tests for smooth classification images. Journal of Vision. 2005; 5, 1–1. doi: 10.1167/5.1.1 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref023] 23.JASP Team. JASP (Version 0.14.1) 2020. [Computer software].

[pone.0255766.ref024] 24.Russell R. Sex, beauty, and the relative luminance of facial features. Perception. 2003; 32(9), 1093–1107. doi: 10.1068/p5101 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref025] 25.Yu C, Klein SA, Levi DM. Cross- and iso- oriented surrounds modulate the contrast response function: the effect of surround contrast. J Vis. 2003;3(8):527–540. doi: 10.1167/3.8.1 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref026] 26.Fink B., Grammer K. & Matts P.J. Visible skin color distribution plays a role in the perception of age, attractiveness, and health in female faces. Evol Hum Behav. 2006; 27, 433–442. doi: 10.1016/j.evolhumbehav.2006.08.007 [DOI] [Google Scholar]

[pone.0255766.ref027] 27.Matts P.J., Fink B., Grammer K. & Burquest M. Color homogeneity and visual perception of age, health, and attractiveness of female facial skin. Journal of the American Academy of Dermatology. 2007; 57, 977–984. doi: 10.1016/j.jaad.2007.07.040 [DOI] [PubMed] [Google Scholar]

[pone.0255766.ref028] 28.Mitteroecker P., Windhager S., Møller G.B., Schaefer K. The morphometrics of “masculinity” in human faces. PLoS ONE. 2015; 10, e0118374. doi: 10.1371/journal.pone.0118374 [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Facial femininity and perceptions of eating disorders: A reverse-correlation study

Valerie Douglas

Benjamin Balas

Kathryn Gordon

Roles

Abstract

Introduction

Experiment 1 –Do faces that look likely to have an eating disorder look more feminine to naïve observers?

Methods

Participants

Stimuli

Fig 1. Examples of the stimuli used in our experiment.

Procedure

Results

Fig 2. Average classification images for White (top row) and Black (bottom row) faces.

Image statistics analysis

Table 1. Descriptive statistics for background intensity analysis.

Table 2. Descriptive statistics for foreground intensity analysis.

Experiment 2 –Do feminized faces look more likely to have an ED?

Methods

Participants

Stimuli

Fig 3. Examples of original (right column) and feminized faces (left column) used in Experiment 2.

Procedure

Results

Discussion

Acknowledgments

Data Availability

Funding Statement

References

Decision Letter 0

Darren Rhodes

Roles

Author response to Decision Letter 0

Decision Letter 1

Avanti Dey

Roles

Author response to Decision Letter 1

Decision Letter 2

Kamila Czepczor-Bernat

Roles

Author response to Decision Letter 2

Decision Letter 3

Kamila Czepczor-Bernat

Roles

Acceptance letter

Kamila Czepczor-Bernat

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases