Skeletal structure and makeup effects in facial attractiveness judgments

Abstract

This study examined whether attractiveness derived from skeletal indices (skeletal attractiveness) influences facial-attractiveness judgments after makeup application. Three-dimensional facial models were constructed from female faces in the Chicago Face Database (CFD). Based on the anteroposterior positions of the midface and mandible, faces were classified into nine skeletal types and then grouped into high- and low-skeletal-attractiveness categories. A single digital makeup condition, selected through a pretest, was uniformly applied to all faces. Attractiveness ratings were collected online and analyzed using ordinal logistic mixed-effects models with random intercepts for raters and facial identities. Makeup significantly increased attractiveness ratings overall, and faces in the high-skeletal-attractiveness group were rated as more attractive than those in the low group. By contrast, the Makeup $\times$ Skeletal Attractiveness Group interaction was not statistically significant, indicating that the makeup-related increase did not clearly differ between the two skeletal groups. Supplementary analyses incorporating target-face perceived age from the CFD norming data showed that faces perceived as older received lower ratings and that makeup-related gains were smaller for older-perceived faces; however, the main conclusions regarding skeletal attractiveness and makeup remained unchanged. These findings suggest that facial-attractiveness judgments are shaped by both surface-based cosmetic cues and three-dimensional structural cues, with skeletal structure functioning primarily as a baseline determinant rather than as a strong moderator of makeup effects.

Keywords

3D face model digital makeup perceptual rating cumulative link mixed model facial attractiveness

Introduction

Background

Facial appearance is an important cue that guides social judgments in interpersonal contexts, including inferences about others’ personality traits and abilities (Jaeger et al., 2022, 2025). In particular, first impressions based on faces are formed rapidly and often automatically, and can exert lasting effects on subsequent social judgments and behavior (Freeman & Johnson, 2016; Todorov et al., 2015). Judgments of facial attractiveness, for example, have been linked to real-world social outcomes such as interpersonal preference, trustworthiness judgments, and employment opportunities (Jaeger et al., 2025). Clarifying how deliberate modifications of facial appearance alter evaluations by others is therefore an important issue in face-perception research (Kramer et al., 2024).

Within this context, makeup can be understood as a form of appearance modification used strategically to shape impressions in accordance with situational demands. Women have been reported to adjust the amount and carefulness of their makeup depending on contextual circumstances (Biesiadecka et al., 2023), suggesting that makeup is not merely decorative but also a practice aimed at guiding social evaluation in desirable directions. More broadly, research on aesthetic facial interventions has likewise shown that people modify their appearance in order to create more favorable first impressions (Jaeger et al., 2025). Recent studies have also examined user experience and satisfaction with beauty-related applications using augmented reality (AR; Voicu et al., 2023). Against this background, the importance of a methodology for objectively evaluating the effects of appearance modification is likely to increase further.

Previous research on the effects of makeup has consistently shown that, on average, makeup enhances facial attractiveness. For example, professionally applied makeup significantly increases attractiveness ratings for female faces (Jones & Kramer, 2016), and attractiveness-enhancing effects of makeup have also been observed for male faces (Batres & Robinson, 2022). Moreover, effect sizes vary depending on how makeup is applied: professional makeup may enhance impressions such as attractiveness, femininity, and status more strongly than self-applied makeup (Batres et al., 2021). At the same time, more controlled approaches have begun to examine makeup effects using standardized or digitally manipulated cosmetic conditions, thereby reducing variation arising from application technique or product choice (Batres et al., 2023). Thus, it is necessary to examine not only whether makeup is present, but also under what conditions, with what degree of stimulus control, and to what extent its effects emerge.

At the same time, existing studies suggest that although makeup has an average positive effect, the magnitude of that effect varies across faces even under the same makeup manipulation. For instance, it has been reported that faces that are already highly attractive without makeup may show smaller gains from makeup (Jones & Kramer, 2016), indicating substantial individual variation in makeup effects. However, the factors underlying this variation remain insufficiently understood. Accordingly, the present study goes beyond treating makeup-related increases in attractiveness as an average effect and instead investigates the underlying factors that may generate such individual differences.

Three-Dimensional Facial Structure and Attractiveness

Research seeking to explain the mechanisms through which makeup enhances attractiveness has so far focused primarily on surface-level visual information. Specifically, light makeup that does not undermine naturalness tends to increase attractiveness (Tagai et al., 2017); skin evenness is associated with attractiveness (Batres et al., 2019; Coetzee et al., 2012); and makeup may alter impressions by increasing facial contrast (Russell, 2009). Taken together, these findings suggest that makeup influences attractiveness judgments mainly by modifying surface cues such as skin appearance and feature contrast.

However, facial attractiveness is not determined by surface information alone. The relationship between facial shape and attractiveness has traditionally been examined using ratios derived from facial length and width. For example, attractiveness may be higher when the face approximates a vertically balanced three-part proportion (Villavisanis et al., 2022). By contrast, associations with the golden ratio, ideal proportions, or ideal angles appear to be limited, and some indices have been reported to explain attractiveness poorly (Zwahlen et al., 2022). One possible reason is that ratio-based indices derived from two-dimensional images do not directly capture depth information or underlying skeletal morphology. In contrast, more recent work has shown that three-dimensional skeletal classifications focusing on the anteroposterior positions of the midface and mandible are significantly related to facial attractiveness (Kunz et al., 2025; Schwarz, 1961). In particular, skeletal types characterized by a protrusive or average midface combined with a retrusive or upright mandible have been reported to be evaluated as more attractive, suggesting that facial morphology including depth is an important underlying determinant of attractiveness judgments (Kunz et al., 2025).

This has important implications for makeup research. Although makeup is an operation that primarily changes surface information, skeletal structure constitutes a relatively stable source of information grounded in the three-dimensional form of the face (Kunz et al., 2025). Conceptually, then, judgments of facial attractiveness may be formed through the integration of both shape-based cues, such as skeletal structure, and surface-based cues, such as makeup. If skeletal structure determines the baseline level of attractiveness, skeletal differences may affect attractiveness levels even after makeup has been applied. Moreover, if skeletal structure also influences how makeup works, then it may contribute to individual differences in attractiveness gains produced by makeup.

Research Gap

Taken together, prior research has shown, first, that makeup increases attractiveness on average; second, that the magnitude of this effect varies across faces; and third, that skeletal differences derived from three-dimensional facial structure are associated with facial attractiveness. However, these findings have largely accumulated in separate lines of research, and studies that examine skeletal differences and makeup effects in an integrated manner under the same stimulus conditions and evaluation procedure remain limited. In particular, much of the research on makeup effects has focused on comparing bare and made-up faces (Batres et al., 2021; Jones & Kramer, 2016). As a result, it remains unclear how individual differences in makeup effects are related to underlying factors rooted in three-dimensional facial structure.

In addition, existing makeup studies have often used stimuli based on makeup manually applied by humans, in which makeup amount, technical skill, product choice, and naturalness can vary simultaneously. Consequently, not only the presence or absence of makeup but also differences attributable to the person applying it or to the specific style can become confounded, making it difficult to isolate the factors that generate individual variation in makeup effects. Indeed, professional makeup has been shown not only to enhance attractiveness more strongly than self-applied makeup, but also to appear heavier and less natural (Batres et al., 2021). Although digitally controlled makeup manipulations have been used to reduce such variation in surface appearance (Batres et al., 2023), these studies have not directly examined whether the effect of a standardized cosmetic manipulation depends on three-dimensional structural differences in the face. Therefore, to test whether skeletal differences moderate makeup effects, a controlled design is required in which the same makeup condition is applied reproducibly across all stimuli while skeletal structure is operationally defined within the same stimulus set.

A further issue concerns the perceptual status of digitally reconstructed three-dimensional faces. Faces that are highly human-like but still perceptibly artificial may evoke a sense of uncanniness or perceptual mismatch, a phenomenon discussed in the literature on the uncanny valley (Olivera-La Rosa, 2018; Wang et al., 2015). This issue is relevant to the present study because applying makeup to digitally reconstructed faces may alter not only surface attractiveness cues but also the perceived congruence between human-like facial structure and artificial rendering. Accordingly, the use of standardized digital stimuli provides experimental control, but it also requires caution when interpreting makeup effects as perceptual responses to natural human faces.

Moreover, recent research on facial appearance interventions suggests that effects relevant to first impressions are often small, yet still socially meaningful. For example, aesthetic interventions may significantly increase attractiveness while producing comparatively modest effect sizes and little spillover to other impression dimensions (Jaeger et al., 2025). Accordingly, the theoretical value of the present study lies not in presuming a large interaction between skeletal structure and makeup, but in disentangling whether skeletal structure functions primarily as a baseline determinant or whether it also alters the increment produced by makeup itself.

Research Aims and Hypotheses

Against this background, the aim of the present study was to examine whether attractiveness based on skeletal indices (hereafter, skeletal attractiveness) influences judgments of facial attractiveness after makeup application. Specifically, three-dimensional facial models were constructed from a standardized facial image database, and skeletal indices were calculated from the anteroposterior positions of the midface and mandible in order to define high- and low-skeletal-attractiveness groups. The same digital makeup condition was then applied to all facial models, and attractiveness judgments were collected through an online rating task. This design made it possible to examine the main effects of skeletal differences and makeup, as well as their interaction, within a single common procedure. In supplementary analyses, we also examined whether these conclusions were robust after accounting for face-level perceived age and Ethnicity category. The basic idea of the study was to conceptualize skeletal structure as the “base” and makeup as an “added layer,” and to clarify their relative contributions.

The study addressed the following three research questions (RQs):

RQ1: Does makeup consistently increase attractiveness ratings regardless of whether a face belongs to the high- or low-skeletal-attractiveness group? This question was derived from previous findings indicating that makeup increases facial attractiveness on average (Batres et al., 2021; Batres & Robinson, 2022; Jones & Kramer, 2016).

RQ2: Are faces in the high-skeletal-attractiveness group rated as more attractive than faces in the low group, independently of whether makeup is present? This question was based on reports that skeletal types defined by the anteroposterior positions of the midface and mandible are related to facial attractiveness (Kunz et al., 2025; Schwarz, 1961).

RQ3: Does skeletal attractiveness moderate the amount of attractiveness enhancement produced by makeup? In other words, are makeup effects different for high- and low-skeletal-attractiveness faces, such that skeletal differences partially explain individual differences in post-makeup attractiveness gains? Because prior research does not provide sufficient evidence regarding the size of this interaction, this question was treated as a directional but exploratory hypothesis.

The significance of the present study lies in linking makeup research and skeletal-structure research within a common perceptual-evaluation framework and in examining the relationship between surface information and three-dimensional shape information in attractiveness judgments. In doing so, the study not only advances the theoretical understanding of first impressions from faces, but may also provide a basis for evaluating personalized beauty technologies in the future.

Methods

Stimuli

Three-dimensional facial models were initially constructed from facial images of 308 women included in the Chicago Face Database (CFD; Ma et al., 2015). The CFD provides high-resolution face images obtained under standardized photographic conditions, together with subjective and objective norming data, making it well-suited for reproducible stimulus construction. To avoid possible influences of facial muscle movement on the apparent skeletal structure and on attractiveness judgments, only neutral-expression images were used. To obtain three-dimensional shape information from the original images, each face image was converted into a three-dimensional facial model using the Headshot2 function in Character Creator 4 (CC4). Hair, which could interfere with the visibility of skeletal structure or makeup, was hidden.

After excluding two faces for which the landmarks required for skeletal classification could not be extracted from the mesh, 306 models were available for skeletal classification. From these classified models, a final rating-stimulus set of 116 facial identities was selected using stratified sampling so that the CFD Ethnicity-category composition was matched as closely as possible between the high- and low-skeletal-attractiveness groups. For each of these identities, both a no-makeup version and a makeup-applied version were created, yielding 232 stimulus images in total.

Five anatomical landmarks were used for skeletal classification: Porion (P), Orbitale (O), Nasion (N), Subnasale (Sn), and Pogonion (Pog). Each landmark was identified on the three-dimensional mesh on the basis of either a prespecified corresponding vertex or a candidate region. The resulting skeletal indices should be interpreted as operational proxies for structural variation in perceptual research rather than as substitutes for clinical cephalometric measurements. Because the present study reconstructed three-dimensional models from standardized face images rather than from radiographic or scan-based clinical records, the classification should be understood as a perceptually motivated approximation of craniofacial structure, not as a clinical diagnostic measure. A schematic illustration of the coordinate system and the landmark locations on the left facial profile is shown in Figure 1, and the candidate regions used for landmark identification are shown in Figure 2. Region A corresponds to Orbitale, region B to Nasion, and region C to Pogonion. Based on the coordinates of these landmarks, indices of midfacial position and mandibular position were calculated, and each facial model was classified into one of the nine skeletal types following the framework of Kunz et al. (2025). Specifically, the four skeletal types reported by Kunz et al. (2025) as highly attractive were assigned to the high group, whereas the remaining five types were assigned to the low group. The correspondence between skeletal types and the high- and low-attractiveness groups is shown in Table 1. In addition, we obtained the CFD norming variable AgeRated for each stimulus identity. This variable represents independent raters’ estimates of each target’s approximate age and was therefore treated as a face-level perceived-age covariate, rather than as exact chronological age. AgeRated was available for all 116 facial identities in the present rating data set ( $M = 29.38$ , $S D = 6.19$ , range $=$ 19.71–51.70 years). Hereafter, we refer to this variable as target-face perceived age.

Figure 1.

Schematic illustration of the coordinate system and anatomical landmarks on the left facial profile. P $=$ Porion, O $=$ Orbitale, N $=$ Nasion, Sn $=$ Subnasale, and Pog $=$ Pogonion. Underlying source image: Chicago Face Database. The author-generated three-dimensional rendering was derived from CFD material and is used under a Creative Commons Attribution 4.0 (CC BY 4.0) license with permission from the University of Chicago, Center for Decision Research. Credit: Ma, Correll, & Wittenbrink (2015). The Chicago Face Database: A Free Stimulus Set of Faces and Norming Data. Behavior Research Methods, 47, 1122–1135.

Figure 2.

Candidate regions on the 3D mesh used for landmark identification. Region A corresponds to Orbitale, region B to Nasion, and region C to Pogonion.

Table 1.

Nine skeletal types and their assignment to high- and low-attractiveness groups.

Midface Category	Mandibular Category	Group
Anteface	Forward-slanting	Low
Anteface	Straight	High
Anteface	Backward-slanting	High
Average face	Forward-slanting	Low
Average face	Straight	High
Average face	Backward-slanting	High
Retroface	Forward-slanting	Low
Retroface	Straight	Low
Retroface	Backward-slanting	Low

Makeup manipulation was implemented using full-makeup presets included in the SkinGen function of CC4. The preset used in the main experiment was selected through a pretest conducted in advance. The pretest was conducted from October 13 to 16, 2025. A total of 79 respondents participated, and 55 valid responses were retained after data screening. For the pretest, 16 facial models were selected from the Black and White categories of the CFD by crossing the Ethnicity category and skeletal-attractiveness group, with four models sampled from each of the four resulting conditions. Each model was rendered with three full-makeup presets, yielding 48 pretest images. Each participant evaluated eight facial models. For each model, the three preset images were presented side by side, and participants answered all three forced-choice questions: which image appeared most natural, which appeared most even, and which appeared most clear. Thus, participants were not randomly assigned to only one criterion; rather, each participant answered all three criteria for the facial identities assigned to them. The presentation of facial identities was randomized and balanced using Qualtrics.

The three pretest criteria were naturalness, evenness, and clarity (facial contrast). Naturalness was defined as the degree to which harmony with the overall face was maintained without excessive emphasis (Tagai et al., 2017). Evenness was defined as low variability in skin tone and the absence of conspicuous unevenness or visible boundaries (Batres et al., 2019; Coetzee et al., 2012). Clarity was defined as the degree to which the eyes, eyebrows, lips, and facial contour were perceived as visually distinct (Russell, 2009). Across the three full-makeup presets, a composite score was calculated from the selection rates for these three criteria, and the highest-scoring preset was adopted as the single makeup condition used in the main experiment.

For each of the 116 facial identities in the rating-stimulus set, two stimulus images were created: a no-makeup version and a makeup-applied version. However, to prevent the same participant from seeing both versions of the same identity, only one of the two versions was assigned to each participant. Stimulus images were generated using the built-in preview function in CC4. Each facial model was rotated 33 degrees to the right relative to the frontal camera position and cropped so that only the head appeared in the center of the image. The output images were saved in JPEG format at a resolution of $1, 088 \times 1, 248$ pixels. In the makeup-applied condition, the same full-makeup preset selected in the pretest was uniformly applied to all facial models, with no individual adjustment of makeup intensity or color. This procedure was adopted in order to prioritize objectivity and reproducibility of the makeup condition. Representative stimulus images are shown in Figure 3.

Figure 3.

Representative stimulus images used in the attractiveness-rating task (left: no makeup; right: makeup applied). Underlying source image: Chicago Face Database. The author-generated three-dimensional renderings were derived from CFD material and are used under a Creative Commons Attribution 4.0 (CC BY 4.0) license with permission from the University of Chicago, Center for Decision Research. Credit: Ma, Correll, & Wittenbrink (2015). The Chicago Face Database: A Free Stimulus Set of Faces and Norming Data. Behavior Research Methods, 47, 1122–1135.

Participants

Attractiveness ratings were collected online. Participants were recruited from two sources: registered users of the crowdsourcing service CrowdWorks and noncrowdsourced volunteers. The noncrowdsourced volunteers were recruited through courses at several universities, personal networks, and in-person requests in public settings. All participants completed the same Qualtrics-based online questionnaire. At the beginning of the study, participants were informed about the purpose of the research and the anonymous handling of the response data, and only those who consented proceeded to the rating task. Participants reported their gender and age group at the beginning of the survey, and after completing the attractiveness ratings, they reported their makeup frequency and level of interest in makeup. To assess whether participants responded attentively, an attention-check item was embedded in the questionnaire, and data from respondents who failed to follow the instructions were excluded from the analyses. After exclusion of respondents who failed the attention-check item, the final sample comprised 1,197 valid respondents. The sample size was determined by the number of valid responses obtained during the planned data-collection period. Because no a priori power analysis was conducted, the precision of the nonsignificant interaction effect is evaluated using the reported odds ratio and 95% confidence interval (CI). Exact chronological age was not collected; instead, participants reported their age group. The largest age category was 40–49 years (30.7%), and 75.2% of valid respondents were between 30 and 59 years of age.

Among participant attributes, gender, age group, makeup frequency, and interest in makeup were treated as candidate covariates to be included in the statistical models when appropriate. The category structure of these variables and their dummy coding are shown in Table 2. The full distribution of rater attributes is reported in the Results section.

Table 2.

Candidate predictors, covariates, and coding used in the regression analyses.

Variable	Reference Category	Model Variable(s)
Makeup condition (stimulus-side)	No makeup	Makeup applied ( $1 = y e s, 0 = n o$ )
Skeletal attractiveness group (stimulus-side)	Low	High ( $1 = h i g h, 0 = l o w$ )
AgeRated (stimulus-side; supplementary)	—	AgeRated $_{z}$
Ethnicity category (stimulus-side; supplementary)	Asian American	Black, Latina, White
Gender (rater-side)	Female	Male ( $1 = m a l e, 0 = f e m a l e$ )
Age group (rater-side)	Under 20	20–29, 30–39, 40–49, 50–59, 60–69, 70+
Makeup frequency (rater-side)	Never	A few times per year, 1–3 days/month,
		about once/week, 2–4 days/week, 5+ days/week
Interest in makeup (rater-side)	Not at all interested	Not very interested, Neutral,
		Somewhat interested, Very interested
Makeup $\times$ Skeletal attractiveness group	—	Product of the makeup-applied and high-group dummies
Makeup $\times$ AgeRated	—	Product of the makeup-applied dummy and AgeRated $_{z}$

Note. For multilevel categorical variables, dummy variables were defined for all nonreference categories, coded 1 when applicable, and 0 otherwise. AgeRated refers to independent raters’ estimates of each target’s approximate age in the CFD norming data and was treated as target-face perceived age, not exact chronological age. CFD = Chicago Face Database.

Procedure

The attractiveness-rating task was conducted on the survey platform Qualtrics. Each participant evaluated one face image per trial and completed a total of eight attractiveness-rating trials. On each trial, a single facial image was presented, and participants rated its attractiveness on a six-point scale ranging from 1 (“not at all attractive”) to 6 (“very attractive”). The instructions asked participants to base their judgments on immediate first impressions rather than on deliberate reflection.

The order of stimulus presentation was randomized across participants. In addition, the same identity was never presented more than once to a given participant, such that for each facial model, either the no-makeup version or the makeup-applied version was assigned, but not both. This procedure was intended to minimize order effects, learning effects, and comparative judgments arising from repeated exposure to the same face.

Statistical Analysis

Dependent Variable and Model Selection

The dependent variable was the attractiveness rating obtained on the six-point scale. Because the ratings were ordinal and the data had a cross-classified repeated-measures structure in which each rater evaluated multiple images and each image was evaluated by multiple raters, an ordinal logistic mixed-effects model with a cumulative logit link was employed. The analysis was conducted using the clmm function in the R package ordinal (Christensen, 2022). The exponentiated regression coefficients, $\exp (β)$ , were interpreted as cumulative odds ratios for being at or above a given rating category.

For model selection, makeup condition, skeletal attractiveness group, and their interaction were included as mandatory fixed effects. In addition, rater attributes (gender, age group, makeup frequency, and interest in makeup) were treated as candidate covariates. Multiple candidate models were constructed based on combinations of these predictors, and the optimal model was selected on the basis of Akaike’s information criterion (AIC; Akaike, 1973). The main analysis reported below is based on the model with the smallest AIC among the candidate models. To address the possible influence of target-face perceived age, we also conducted supplementary models including standardized AgeRated (AgeRated $_{z}$ ) as a face-level covariate. We first added AgeRated $_{z}$ to the main model as an additional covariate and then fitted a further model including the Makeup Condition $\times$ AgeRated $_{z}$ interaction. These models were used to assess whether the main conclusions regarding makeup condition, skeletal-attractiveness group, and their interaction were robust after accounting for target-face perceived age.

In addition, as a robustness check for the Ethnicity-category composition of the stimulus set, we fitted an extended model that included the Ethnicity category, the Makeup Condition $\times$ Skeletal Attractiveness Group $\times$ Ethnicity Category interaction, and the corresponding lower-order terms. Asian American was used as the reference category for Ethnicity category.

Fixed Effects

The primary fixed effects in the main analysis were the stimulus-side factors of makeup condition, skeletal attractiveness group, and their interaction. These terms allowed us to examine simultaneously (1) the average effect of makeup, (2) the average difference between the skeletal-attractiveness groups, and (3) whether the size of the makeup effect varied as a function of the skeletal attractiveness group. In addition, when appropriate, the rater-side factors of gender, age group, makeup frequency, and interest in makeup were included. In supplementary analyses, AgeRated $_{z}$ and Ethnicity category were added as face-level covariates or moderators as described above. The levels of the explanatory variables and their coding are shown in Table 2.

Random Effects

Random intercepts were included for raters and for facial identities. The former captured between-rater differences in general rating leniency versus severity, whereas the latter captured between-identity differences in the baseline attractiveness level. This specification made it possible to estimate the effects of makeup condition, skeletal attractiveness group, and their interaction while accounting for the dependence structure within raters and within facial identities.

Results

Descriptive Statistics

After excluding two faces for which the landmarks required for skeletal classification could not be extracted from the mesh, 306 facial models were available for skeletal classification. The skeletal-classification procedure yielded 247 faces in the high-skeletal-attractiveness group and 59 faces in the low group. A cross-tabulation of the skeletal classification results for these 306 models is shown in Table 3. As shown in Table 3, the most common combination was Average face $\times$ Straight ( $n = 173$ ), followed by Anteface $\times$ Backward-slanting ( $n = 36$ ), and Retroface $\times$ Straight ( $n = 20$ ). From these classified models, the final rating-stimulus set comprised 116 facial identities: 59 in the high-skeletal-attractiveness group and 57 in the low group. Because both no-makeup and makeup-applied images were created for each identity, the rating-stimulus set contained 232 images in total.

Table 3.

Cross-tabulation of skeletal classification results for the 306 classifiable facial models.

	Backward-Slanting	Straight	Forward-Slanting	Total
Anteface	36	15	0	51
Average face	23	173	23	219
Retroface	0	20	16	36
Total	59	208	39	306

Next, an overview of the attractiveness-rating data is provided. The characteristics of the raters are shown in Table 4. The largest age group among valid respondents was 40–49 years (30.7%). In terms of makeup frequency, the most common category was Never (37.4%), followed by 5+ days per week (30.7%). For interest in makeup, the most frequent response was Somewhat interested (34.9%).

Table 4.

Characteristics of raters.

Characteristic	Category	$n$	%
Age group	Under 20	41	3.4
	20–29	161	13.5
	30–39	285	23.8
	40–49	367	30.7
	50–59	248	20.7
	60–69	82	6.9
	70 and above	13	1.1
Makeup frequency	Never	448	37.4
	A few times per year	73	6.1
	1–3 days per month	54	4.5
	About once per week	72	6.0
	2–4 days per week	182	15.2
	5+ days per week	368	30.7
Interest in makeup	Not at all interested	126	10.5
	Not very interested	314	26.2
	Neutral	184	15.4
	Somewhat interested	418	34.9
	Very interested	155	12.9

Table 5 presents the mean, standard deviation, median, and interquartile range of attractiveness ratings by the skeletal-attractiveness group (high vs. low) and makeup condition (no makeup vs. makeup applied). In the low group, the mean attractiveness rating was approximately 0.10 points higher for the makeup-applied condition than for the no-makeup condition. In the high group, the corresponding difference was approximately 0.15 points. In addition, regardless of makeup condition, the high-skeletal-attractiveness group showed higher mean attractiveness ratings than the low group. Figure 4 shows the proportional distribution of the six-point attractiveness ratings. Across all conditions, midrange ratings (ratings 2–4) accounted for approximately 80% of responses, whereas extreme ratings such as 1 and 6 each accounted for less than 10% in every condition. In particular, the proportion of rating 6 was small, ranging from 0.7% to 2.1%.

Figure 4.

Proportional distribution of attractiveness ratings by skeletal attractiveness group and makeup condition.

Table 5.

Mean attractiveness ratings by skeletal attractiveness group and makeup condition.

Skeletal Attractiveness Group	Makeup Condition	Mean $\pm S D$	Median [interquartile range]
Low	No makeup	2.888 $\pm$ 0.457	2.871 [2.529–3.232]
Low	Makeup applied	2.986 $\pm$ 0.513	2.966 [2.621–3.379]
High	No makeup	3.157 $\pm$ 0.530	3.143 [2.740–3.586]
High	Makeup applied	3.310 $\pm$ 0.463	3.351 [3.017–3.609]

Main Analysis

To examine the effects of makeup and skeletal attractiveness on attractiveness ratings, we conducted an ordinal logistic mixed-effects regression analysis. Comparison of the candidate models using AIC showed that the model including makeup condition, skeletal attractiveness group, gender, age group, and the Makeup $\times$ Skeletal Attractiveness Group interaction as fixed effects, together with random intercepts for raters and facial identities, yielded the smallest AIC (AIC $= 25,264.23$ ). This model was therefore adopted as the main model, and its estimates are reported below. The fixed-effect estimates are shown in Table 6.

Table 6.

Fixed-effect estimates from the ordinal logistic mixed-effects model.

Predictor	$β$	$S E$	$z$	$p$	Sig.
Makeup condition	0.32	0.06	5.73	$< .001$	***
Skeletal attractiveness group	0.61	0.21	2.91	$.004$	**
Gender (male)	$- 0.48$	0.10	$- 4.64$	$< .001$	***
Age group (20–29)	0.11	0.30	0.36	$.718$
Age group (30–39)	0.50	0.29	1.74	$.082$
Age group (40–49)	0.46	0.28	1.62	$.105$
Age group (50–59)	0.18	0.29	0.61	$.544$
Age group (60–69)	0.67	0.33	2.01	$.044$	*
Age group (70 and above)	0.15	0.55	0.28	$.781$
Makeup $\times$ Skeletal attractiveness group	0.13	0.08	1.62	$.104$

Note. *** $p < .001$ , ** $p < .01$ , * $p < .05$ .

The coefficient for makeup condition was positive and significant ( $β = 0.32$ , $S E = 0.06$ , $z = 5.73$ , $p < .001$ ), indicating that, in the makeup-applied condition, the cumulative odds of being placed in a higher attractiveness category increased ( $O R = 1.37$ , 95% CI [1.23, 1.53]). The coefficient for the skeletal-attractiveness group was likewise positive and significant ( $β = 0.61$ , $S E = 0.21$ , $z = 2.91$ , $p = .004$ ), indicating that faces in the high group had higher cumulative odds of receiving more attractive ratings than faces in the low group ( $O R = 1.85$ , 95% CI [1.22, 2.80]). By contrast, the Makeup $\times$ Skeletal Attractiveness Group interaction was positive but not statistically significant ( $β = 0.13$ , $S E = 0.08$ , $z = 1.62$ , $p = .104$ , $O R = 1.13$ , 95% CI [0.97, 1.32]). Thus, the present data did not provide reliable evidence that the magnitude of the makeup-related attractiveness increase differed clearly between the two skeletal-attractiveness groups, although small interaction effects cannot be ruled out.

Among the rater-side covariates, the coefficient for the 60–69 age group was positive and significant relative to the under-20 reference group ( $β = 0.67$ , $S E = 0.33$ , $z = 2.01$ , $p = .044$ , $O R = 1.94$ , 95% CI [1.02, 3.72]). Because the rater age group was included as a covariate rather than as a primary factor of interest, this effect is interpreted descriptively as indicating possible differences in evaluative baselines across rater age groups.

The variance of the random intercepts was 2.44 ( $S D = 1.56$ ) at the rater level and 1.20 ( $S D = 1.09$ ) at the facial-identity level, with larger variance at the rater level than at the facial-identity level. This suggests that, in attractiveness judgments, between-rater differences in overall evaluation tendency were relatively large, in addition to differences among facial identities.

Supplementary Analysis

Supplementary analyses were conducted to examine the robustness of the main results. First, to address the possible influence of target-face perceived age, we fitted models including AgeRated $_{z}$ as a face-level covariate. Adding AgeRated $_{z}$ improved model fit relative to the main model (AIC $=$ 25,252.73 vs. 25,264.23). In this model, AgeRated $_{z}$ showed a significant negative effect ( $β = - 0.36$ , $S E = 0.10$ , $z = - 3.78$ , $p < .001$ , $O R = 0.70$ , 95% CI [0.58, 0.84]), indicating that faces perceived as older by the CFD norming raters were less likely to receive higher attractiveness ratings.

Importantly, after adding AgeRated $_{z}$ , the main effects of makeup and skeletal-attractiveness group remained positive and significant, and the Makeup $\times$ Skeletal Attractiveness Group interaction remained nonsignificant ( $β = 0.13$ , $S E = 0.08$ , $z = 1.63$ , $p = .103$ , $O R = 1.14$ , 95% CI [0.98, 1.32]). Thus, accounting for target-face perceived age did not alter the main conclusion that makeup and the skeletal-attractiveness group had independent main effects, whereas their interaction was not reliable.

We further fitted a model including the Makeup Condition $\times$ AgeRated $_{z}$ interaction. This model showed the lowest AIC among the supplementary models considered (AIC = 25,242.64). The Makeup Condition $\times$ AgeRated $_{z}$ interaction was negative and significant ( $β = - 0.14$ , $S E = 0.04$ , $z = - 3.48$ , $p < .001$ , $O R = 0.87$ , 95% CI [0.80, 0.94]), suggesting that the makeup-related increase in attractiveness ratings was smaller for faces perceived as older. However, the Makeup $\times$ Skeletal Attractiveness Group interaction remained nonsignificant. This pattern indicates that target-face perceived age was relevant to attractiveness judgments and to the size of makeup-related gains, but it did not explain away the primary pattern concerning the skeletal-attractiveness group.

Second, as a robustness check for the Ethnicity-category composition of the stimulus set, we fitted an extended model that included the Ethnicity category, the Makeup Condition $\times$ Skeletal Attractiveness Group $\times$ Ethnicity Category interaction, and the corresponding lower-order terms. The Ethnicity category was entered as three dummy variables, with Asian American used as the reference category. In this extended model, the main effects of makeup and the skeletal-attractiveness group remained positive and significant. The interaction between the skeletal-attractiveness group and Ethnicity category and the three-way interaction among makeup condition, skeletal-attractiveness group, and Ethnicity category were not significant. Therefore, the interpretation obtained in the main analysis was not materially changed after accounting for the Ethnicity category.

Discussion

Summary of the Main Findings

The present study examined whether attractiveness derived from skeletal indices (skeletal attractiveness) influences judgments of facial attractiveness after makeup application. Analysis using an ordinal logistic mixed-effects model yielded three main findings. First, makeup consistently increased attractiveness ratings regardless of whether a face belonged to the high- or low-skeletal-attractiveness group. Second, faces in the high-skeletal-attractiveness group were more likely to receive higher attractiveness ratings than faces in the low group, independently of whether makeup was present. Third, although the interaction between makeup condition and skeletal-attractiveness group was positive in direction, it was not statistically significant, and thus no clear conclusion could be drawn that the magnitude of attractiveness enhancement produced by makeup differed as a function of skeletal attractiveness. Taken together, these results support RQ1 and RQ2, whereas support for RQ3 was limited. Supplementary analyses further showed that these conclusions were robust after accounting for target-face perceived age (AgeRated) and Ethnicity category. AgeRated itself was negatively associated with attractiveness ratings, and the Makeup Condition $\times$ AgeRated interaction suggested that makeup-related gains were smaller for faces perceived as older.

These findings suggest that the effects of makeup are better understood not as a uniform operation that works in the same way for all faces, but rather as an average additive effect operating on top of baseline differences rooted in three-dimensional facial structure. Importantly, the contribution of the present study does not depend on finding a statistically significant interaction. To our knowledge, this study is among the first to examine surface-based cosmetic cues and three-dimensional structural cues within a single perceptual-evaluation framework using common stimuli, a common rating procedure, and a single statistical model. In this sense, the present study makes a novel contribution by empirically linking two lines of research that have largely progressed separately: makeup research, which has focused primarily on surface appearance, and facial-shape research, which has focused primarily on structural morphology. More specifically, the pattern observed here indicates that faces in the high-skeletal-attractiveness group started from a higher baseline level of attractiveness, upon which the average enhancing effect of makeup was added. By contrast, no clear evidence was obtained that skeletal characteristics substantially altered the attractiveness increment produced by makeup itself.

Theoretical Implications

The first theoretical implication of the present study is that judgments of facial attractiveness should be understood as a multilayered perceptual process that depends not only on surface-derived cues but also on underlying factors derived from three-dimensional shape. More importantly, the present study provides an early empirical demonstration that these two classes of cues—surface-based cosmetic cues and three-dimensional structural cues—can be examined within a single perceptual framework rather than as separate explanatory domains. Previous studies have shown that makeup may enhance attractiveness by changing surface-level visual cues such as naturalness, skin evenness, and facial contrast (Batres et al., 2019; Coetzee et al., 2012; Russell, 2009; Tagai et al., 2017). However, such accounts are limited in their ability to explain the individual variation observed in responses to the same makeup manipulation. By showing that faces in the high-skeletal-attractiveness group were rated more highly under a controlled digital makeup manipulation, the present study provides grounds for positioning three-dimensional facial structure as a source of baseline differences in attractiveness judgments alongside surface-level appearance cues.

Second, the present findings support a two-layer view of the relationship between skeletal structure and makeup, in which skeletal structure functions as a “base” and makeup as an “added layer.” The idea that three-dimensional structure contributes to facial attractiveness is consistent with findings from orthodontics and facial-profile research, as well as with reports that skeletal types defined by the anteroposterior positions of the midface and mandible are associated with differences in attractiveness (Kunz et al., 2025; Schwarz, 1961). At the same time, because the interaction between skeletal attractiveness and makeup was not significant in the present study, it is more appropriate to interpret skeletal differences as primarily determining a face’s baseline attractiveness level, while makeup functions as an operation that provides a relatively consistent enhancement on top of that baseline, rather than strongly amplifying or attenuating the makeup effect itself.

The supplementary analysis of AgeRated extends this interpretation. Faces perceived as older by the CFD norming raters tended to receive lower attractiveness ratings, indicating that target-face perceived age also contributed to baseline differences in attractiveness. At the same time, the negative Makeup Condition $\times$ AgeRated interaction suggests that makeup-related gains were smaller for faces perceived as older. Thus, perceived age may operate both as a baseline factor and as a contextual factor that modulates the size of cosmetic enhancement. Importantly, however, accounting for AgeRated did not alter the main conclusion that the skeletal-attractiveness group and makeup condition had independent main effects, whereas their interaction was not reliable.

Third, the present results are consistent with recent work on facial appearance interventions showing that such interventions may significantly affect attractiveness without necessarily producing equally large changes in other impression dimensions, and that their effect sizes may sometimes be smaller than expected (Jaeger et al., 2025). In the present study as well, the average effect of makeup and the main effect of skeletal structure were confirmed, whereas the interaction between them was not clearly established. This pattern underscores the importance of distinguishing not only whether an effect exists, but also at what level it operates: as a surface-based enhancement, a structural baseline difference, or a moderation of one cue by another.

Methodological Implications

The first methodological contribution of the present study lies in reconstructing three-dimensional facial models from a standardized facial image database, classifying stimuli operationally on the basis of skeletal indices, and evaluating attractiveness under a controlled digital-makeup condition. Previous makeup research has often relied on stimuli created through manually applied makeup, in which multiple factors, such as application skill, intensity, naturalness, and product choice, may vary simultaneously (Batres et al., 2021). Although digitally controlled makeup manipulations have also been used to reduce such variation in surface appearance (Batres et al., 2023), the present study extends this approach by combining a standardized cosmetic manipulation with an operational classification of three-dimensional facial structure. Applying the same full-makeup preset uniformly to all stimuli prioritized objectivity and reproducibility, making it possible to separate the average effect of makeup from the contribution of skeletal differences.

Second, the study is methodologically meaningful in its use of a mixed-effects model that treated both stimuli and raters as random factors, thereby improving statistical validity in face research. Recent work on facial stimuli and aesthetic interventions has pointed out that analyses based on averaging across stimuli can inflate false-positive rates, especially when the number of stimuli is small relative to the number of raters (Jaeger et al., 2025). In the present study, attractiveness ratings were treated as ordinal data, and random intercepts for both raters and facial models were introduced in order to account for the cross-classified structure of the data. This modeling strategy strengthens the basis for making inferences about makeup and skeletal effects beyond the specific set of raters and facial identities included in the present data set.

Third, the present study also offers methodological implications for the evaluation of beauty-support technologies using BeautyTech and AR. Prior work has examined user experience and satisfaction with such technologies (Voicu et al., 2023), but frameworks for measuring social impressions themselves under controlled conditions remain insufficiently developed. A design such as the present one, combining a standardized stimulus set, operationally defined skeletal indices, reproducible digital-makeup conditions, and mixed-effects modeling, may provide a foundation for the objective evaluation of future beauty-support technologies. At the same time, such evaluations should also consider whether digitally rendered faces are perceived as natural human faces or as artificial stimuli that may evoke perceptual mismatch.

Limitations and Future Directions

The present study has several limitations. First, there are limitations concerning the validity of stimulus generation and skeletal-index estimation. In this study, three-dimensional facial models were reconstructed from frontal face images, and skeletal indices were calculated from landmarks identified on the resulting mesh. However, this procedure may contain errors related to three-dimensional reconstruction and landmark identification. Future work should therefore examine the measurement validity and reproducibility of the skeletal indices more rigorously, for example, by using three-dimensional scan data or multiview images and by assessing interrater or intrarater agreement in landmark identification. In this respect, the present skeletal indices should be interpreted as perceptually oriented operational proxies for facial-structure cues, rather than as clinical cephalometric substitutes or diagnostic measures.

A related limitation concerns the artificiality of the reconstructed facial stimuli. Digitally reconstructed faces provide strong experimental control, but they may also be perceived as less natural than photographs of real faces. The literature on the uncanny valley suggests that highly human-like but still artificial stimuli can evoke feelings of uncanniness or perceptual mismatch (Olivera-La Rosa, 2018; Wang et al., 2015). In the present study, applying makeup to reconstructed faces may have influenced not only surface-level attractiveness cues but also the perceived congruence between human-like facial structure and digital rendering. Because the main rating task did not include direct measures of perceived naturalness, artificiality, or uncanniness, future research should include such manipulation checks to clarify whether perceptual mismatch moderates the effect of digital makeup on attractiveness judgments.

Second, the makeup condition used in this study involved applying the same preset uniformly to all stimuli in order to maximize objectivity and reproducibility. Although this design is well-suited to testing average effects, it intentionally excludes effects based on the “fit” between makeup and facial structure, such as contour correction or shading adjustments tailored to particular skeletal types. Indeed, professional makeup has been reported to enhance attractiveness more strongly than self-applied makeup while also appearing heavier and less natural, suggesting that the effectiveness of makeup cannot be separated from how it is applied and at what intensity. The supplementary AgeRated analysis also suggests that makeup-related gains may vary with target-face perceived age. Future research should therefore reconsider the conditions under which makeup interactions may become detectable, for example, by varying makeup intensity, comparing multiple presets, or introducing correction conditions tailored to skeletal characteristics, perceived age, or other face-level attributes.

Third, the generalizability of the stimulus set and participant sample is limited. The present study was based on female facial images from the CFD and therefore cannot be directly generalized to male faces, to faces from a wider range of age groups, or to stimuli drawn from different cultural contexts. In addition, the raters were primarily Japanese-speaking participants recruited online, and it cannot be ruled out that cultural context affects both attractiveness judgments and the size of makeup effects. Research on aesthetic interventions has likewise suggested that differences in the national or cultural backgrounds of targets and raters may influence outcomes (Jaeger et al., 2025). Moreover, AgeRated was based on CFD norming ratings rather than on the exact chronological age of the target models. It should therefore be interpreted as target-face perceived age, which may itself covary with attractiveness and other facial impressions. Future work should therefore extend the design to male stimuli and to stimulus sets including faces from other cultural backgrounds, distinguish chronological age from perceived age, and recruit more diverse rater samples.

Finally, the interaction between the high- and low-skeletal-attractiveness groups and makeup was not significant in the present study. This does not mean that skeletal structure is entirely unrelated to makeup effects. The CI for the Makeup $\times$ Skeletal Attractiveness Group interaction included values compatible with very small or modest positive effects, and therefore smaller moderation effects cannot be ruled out. It remains possible that the true effect size is small, that the stimulus conditions were too homogeneous, or that the available data did not contain sufficient variability for precise estimation. Recent face research has likewise shown that effects of appearance interventions are often relatively modest, and that smaller estimates than previously reported may emerge under rigorous stimulus control and appropriate statistical modeling (Jaeger et al., 2025). Future studies should therefore employ more diverse makeup conditions, richer three-dimensional shape information, and larger and more heterogeneous stimulus sets in order to test more rigorously whether skeletal structure contributes not only to baseline differences in attractiveness but also to the attractiveness increment produced by makeup.

Footnotes

Acknowledgments

The authors thank Professor Gaetan Rappo for helpful comments on an earlier draft of this manuscript.

ORCID iD

Akihiro Kawase

Ethics Approval and Informed Consent Statements

According to the institutional guidelines of Doshisha University (Faculty/Graduate School policy on human-subject research), this study was deemed exempt from formal ethics committee review because it involved an anonymous online rating survey and collected no personally identifying or sensitive personal information. Participants provided informed consent electronically before beginning the survey. The study used face images obtained under the data-use conditions of the Chicago Face Database (CFD), and no new identifiable personal images were collected from participants.

Author Contribution(s)

Akihiro Kawase: Conceptualization; Formal analysis; Supervision; Visualization; Writing – original draft; Writing – review & editing.

Rikuto Yamamoto: Data curation; Formal analysis; Investigation; Visualization; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the KOSÉ Cosmetology Research Foundation (34th Cosmetology Research Grant, award listing No. 845). The funder had no role in the study design; data collection, analysis, or interpretation; manuscript preparation; or the decision to submit the manuscript for publication.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The analysis code, derived facial-shape measures, and anonymized rating data are available at Zenodo: . The original face photographs were obtained from the Chicago Face Database (CFD) and are subject to its data-use agreement; therefore, the original images, rendered stimulus images, 3D meshes, raw landmark coordinates, and identifiers linking the derived measures to the original CFD model IDs are not redistributed. The repository includes the scripts and documentation necessary to reproduce the reported analyses from the shared derived data.

References

Akaike

(1973). Information theory and an extension of the maximum likelihood principle. Proceedings of the 2nd International symposium on information theory (pp. 267–281). Budapest: Akadémiai Kiadó.

Batres

Porcheron

Courréges

Russell

(2021). Professional versus self-applied makeup: Do makeup artists add value? Perception, 50(8), 709–719. https://doi.org/10.1177/03010066211029218

Batres

Porcheron

Latreille

Roche

Morizot

Russell

(2019). Cosmetics increase skin evenness: Evidence from perceptual and physical measures. Skin Research and Technology, 25(5), 672–676. https://doi.org/10.1111/srt.12700

Batres

Robinson

(2022). Makeup increases attractiveness in male faces. PLOS ONE, 17, e0275662. 10.1371/journal.pone.0275662

Batres

Russell

Workowski

(2023). Makeup applied to facial features increases apparent skin evenness. Vision Research, 202, 108144. 10.1016/j.visres.2022.108144

Biesiadecka

Szymkow

Baryla

(2023). To enhance, or not to enhance: The situational context shapes women’s intentions on amount and diligence of makeup application. Evolutionary Psychology, 21, 14747049231219283. 10.1177/14747049231219283

Christensen

R. H. B.

(2022). Cumulative link models for ordinal regression with the R package ordinal. Retrieved, March 1, 2026, from https://cran.r-project.org/web/packages/ordinal/vignettes/clm_article.pdf

Coetzee

Faerber

S. J.

Greeff

J. M.

Lefevre

C. E.

D. E.

Perrett

D. I.

(2012). African perceptions of female attractiveness. PLOS ONE, 7, e48116. 10.1371/journal.pone.0048116

Freeman

J. B.

Johnson

K. L.

(2016). More than meets the eye: Split-second social perception. Trends in Cognitive Sciences, 20, 362–374. 10.1016/j.tics.2016.03.003

10.

Jaeger

Bucker

van der Meulen

van Vugt

(2025). Face value: The effect of facial aesthetic treatment on first impressions and partner preferences. Perception, 54(9), 651–673. https://doi.org/10.1177/03010066251337353

11.

Jaeger

Evans

A. M.

Stel

van Beest

(2022). Understanding the role of faces in person perception: Increased reliance on facial appearance when judging sociability. Journal of Experimental Social Psychology, 100, 104288. https://doi.org/10.1016/j.jesp.2022.104288

12.

Jones

A. L.

Kramer

R. S. S.

(2016). Facial cosmetics and attractiveness: Comparing the effect sizes of professionally-applied cosmetics and identity. PLOS ONE, 11, e0164218. 10.1371/journal.pone.0164218

13.

Kramer

R. S. S.

Koca

Mireku

M. O.

Oriet

(2024). Anchoring has little effect when forming first impressions of facial attractiveness. Perception, 53(11-12), 787–802. https://doi.org/10.1177/03010066241284956

14.

Kunz

Nordblom

N. F.

Goetz

Fenger

Stellzig-Eisenhauer

(2025). Which of the nine facial profiles according to A. M. Schwarz is the most (un)attractive? Journal of Orofacial Orthopedics/Fortschritte der Kieferorthopädie. 10.1007/s00056-025-00584-z

15.

D. S.

Correll

Wittenbrink

(2015). The Chicago Face Database: A free stimulus set of faces and norming data. Behavior Research Methods, 47, 1122–1135. 10.3758/s13428-014-0532-5

16.

Olivera-La Rosa

(2018). Wrong outside, wrong inside: A social functionalist approach to the uncanny feeling. New Ideas in Psychology, 50, 38–47. 10.1016/j.newideapsych.2018.03.004

17.

Russell

(2009). A sex difference in facial contrast and its exaggeration by cosmetics. Perception, 38(8), 1211–1219. https://doi.org/10.1068/p6331

18.

Schwarz

A. M.

(1961). Lehrgang der Gebissregelung. Untersuchungsgang (Diagnostik). Wien: Urban & Schwarzenberg.

19.

Tagai

Shimakura

Isobe

Nittono

(2017). The light-makeup advantage in facial processing: Evidence from event-related potentials. PLOS ONE, 12, e0172489. 10.1371/journal.pone.0172489

20.

Todorov

Olivola

C. Y.

Dotsch

Mende-Siedlecki

(2015). Social attributions from faces: Determinants, consequences, accuracy, and functional significance. Annual Review of Psychology, 66, 519–545. 10.1146/annurev-psych-113011-143831

21.

Villavisanis

D. F.

Workman

C. I.

Cho

D. Y.

Zapatero

Z. D.

Wagner

C. S.

Blum

J. D.

Bartlett

S. P.

Swanson

J. W.

Chatterjee

Taylor

J. A.

(2022). Associations of facial proportionality, attractiveness, and character traits. The Journal of Craniofacial Surgery, 33, 1431–1435. 10.1097/scs.0000000000008662

22.

Voicu

M. C.

Sîrghi

Toth

D. M. M.

(2023). Consumers’ experience and satisfaction using augmented reality apps in e-shopping: New empirical evidence. Applied Sciences, 13, 9596. 10.3390/app13179596

23.

Wang

Lilienfeld

S. O.

Rochat

(2015). The uncanny valley: Existence and explanations. Review of General Psychology, 19, 393–407. 10.1037/gpr0000056

24.

Zwahlen

R. A.

Tang

A. T. H.

Leung

W. K.

Tan

S. K.

(2022). Does 3-dimensional facial attractiveness relate to golden ratio, neoclassical canons, “ideal” ratios and “ideal” angles? Maxillofacial Plastic and Reconstructive Surgery, 44, 28. 10.1186/s40902-022-00358-2