The Faces of Success

Article
Journal of Marketing
1-19
The Faces of Success: Beauty and Ugliness ª American Marketing Association 2020
Article reuse guidelines:
Premiums in e-Commerce Platforms sagepub.com/journals-permissions

DOI: 10.1177/0022242920914861
journals.sagepub.com/home/jmx
Ling Peng , Geng Cui, Yuho Chung, and Wanyi Zheng
Abstract
Given the positive bias toward attractive people in society, online sellers are justifiably apprehensive about perceptions of their
profile pictures. Although the existing literature emphasizes the “beauty premium” and the “ugliness penalty,” the current studies
of seller profile pictures on customer-to-customer e-commerce platforms find a U-shaped relationship between facial attrac-
tiveness and product sales (i.e., both beauty and ugliness premiums and, thus, a “plainness penalty”). By analyzing two large data
sets, the authors find that both attractive and unattractive people sell significantly more than plain-looking people. Two online
experiments reveal that attractive sellers enjoy greater source credibility due to perceived sociability and competence, whereas
unattractive sellers are considered more believable on the basis of their perceived competence. While a beauty premium is
apparent for appearance-relevant products, an ugliness premium is more pronounced for expertise-relevant products and for
female consumers evaluating male sellers. These findings highlight the influence of facial appearance as a key vehicle for impression
formation in online platforms and its complex effects in e-commerce and marketing.
Keywords
attractiveness, beauty premium, e-commerce, social selling, ugliness premium
Online supplement: https://doi.org/10.1177/0022242920914861
The role of attractiveness in social judgments and the beauty sellers exhibit a wide range of attractiveness, their profile pic-
premium have been well documented in various social settings tures, as an integral part of seller identity, serve as a key vehicle
such as dating, hiring, selling, and advertising, especially when for impression formation and evoke feelings that affect buyer
the task or product is related to appearance (Argo, Dahl, and decisions (Forman, Ghose, and Wiesenfeld 2008). Most peo-
Morales 2008; Eagly et al. 1991; Langlois et al. 2000). A few ple, however, are not endowed with perfect facial symmetry
studies have found opposite results when a product is not rel- and proportions. In light of the increasing popularity of social
evant to appearance, but they have not provided coherent selling, how one’s attractiveness or lack thereof affects the
explanations for these findings (Kamins 1990; Trampe et al. sales of various products is of much concern among online
2010). Moreover, most researchers have compared attractive sellers and of great interest to marketing researchers and
models or endorsers with those who are less attractive, largely practitioners.
ignoring people who are unattractive altogether. Recent studies Drawing from the literature on impression formation, the
indicate a potential ugliness premium: unattractive people are match-up hypothesis, and evolutionary psychology, we argue
perceived as more intelligent and earn significantly more than that both attractive and unattractive online sellers command
their attractive counterparts (e.g., Gheorghiu, Callan, and Sky- more attention and source credibility than plain-looking sellers,
lark 2017; Kanazawa and Still 2018), which suggests that the resulting in a U-shaped effect of attractiveness on sales. In
effect of attractiveness is nonlinear. Thus, researchers have yet
to identify the precise underlying mechanisms and contexts for
the beauty premium or that for the ugliness premium, if it Ling Peng is Associate Professor, Department of Marketing and International
exists. Business, Lingnan University, Hong Kong (email: lingpeng@ln.edu.hk). Geng
Unlike conventional marketing that relies on celebrities or Cui is Professor of Marketing, Department of Marketing and International
salespeople promoting a specific product, customer-to- Business, Lingnan University, Hong Kong (email: gcui@ln.edu.hk). Yuho
Chung is Visiting Assistant Professor, Department of Marketing and
customer (C2C) e-commerce involves large numbers of ordi- International Business, Lingnan University, Hong Kong (email: yuhochung@
nary people as sellers pitching a variety of products, making ln.edu.hk). Wanyi Zheng is a doctoral student, Faculty of Business and
seller credibility a critical issue (Luca 2017). While online Economics, University of Hong Kong, Hong Kong (email: zwymkt@hku.hk).
2 Journal of Marketing XX(X)
contrast to previous studies, we go beyond consumer attitudes et al. 2000). According to evolutionary psychology (e.g.,
toward advertisements and products and focus on trait infer- Magro 1999), an attractive face indicates good health and pros-
ences to explore the underlying mechanisms of beauty and pect for survival and reproduction. Beauty is also correlated
ugliness premiums and their effect on source credibility and with perceived intelligence and social skills (Eagly et al. 1991;
purchase intention. We find that while attractive faces fare Hamermesh 2011). Attractive solicitors can obtain twice as
better in sociability than both plain-looking and unattractive much in donations as their unattractive counterparts (Reingen
people, they are not considered more competent than unattrac- and Kernan 1993), and a good-looking salesperson enhances
tive people, who are perceived as more competent than plain customer evaluation of a product simply by touching it (Argo,
people, resulting in a plainness penalty. These relations are Dahl, and Morales 2008). Although attractiveness is valued in
moderated by product relevance (appearance vs. expertise) and both men and women, men are more responsive to the physical
a cross-gender effect for women looking at male sellers. attractiveness of women (Li and Kenrick 2006). Meanwhile,
The remainder of this article is organized as follows. We studies have found that attractiveness sometimes fails to work,
first provide a succinct review of the relevant literature and for instance, when helping children in need or selling an embar-
present a conceptual framework for the effect of facial attrac- rassing product (Fisher and Ma 2014; Wan and Wyer 2015).
tiveness on consumers. We extract the geometric features of Several recent studies show that unattractive faces are asso-
facial images and adopt a machine learning approach to score ciated with certain positive outcomes. Gheorghiu, Callan, and
large samples of online seller portraits. Next, we investigate Skylark (2017) find that students rate unattractive professors as
beauty and ugliness premiums using a multimethod approach better scientists than attractive professors. A study of Nobel
involving large data sets from two e-commerce platforms and laureates reinforces the pervasive stereotype that scientists
two online experiments to assess the potential mediators and sacrifice physical appeal for intellectual pursuits (Fidrmuc,
moderators. Finally, we discuss the key findings and implica-
Paphawasit, and Tunalı 2017). Kanazawa and Still (2018) indi-
tions for e-commerce and internet marketing.
cate that very unattractive executives earn significantly more
than their attractive counterparts, although the study does not
consider perceptions of competence. These findings support
Relevant Literature the popular belief that unattractive people exert greater effort
Impression Formation and Face Perception to compensate for their disadvantaged appearance; however,
these studies fall short of offering plausible explanations for
Faces are known to bias decisions (Wheeler and Petty 2001). the ugliness premium.
We form first impressions of others and make judgments about
their social traits almost instantaneously on the basis of face
perceptions (Samper, Yang, and Daniels 2018; Todorov et al.
2005; Willis and Todorov 2006). The neural mechanism under- Online Profile Pictures
lying trait impressions of faces involves the amygdala, a sub- Online forums and social media have aggravated people’s con-
cortical brain region crucial in coding the value of stimuli (e.g., cern with appearance and greatly affected social and consump-
Engell, Haxby, and Todorov 2007). In functional magnetic tion behaviors (Grabe, Ward, and Hyde 2008). The advantages
resonance imaging (fMRI) studies, the amygdala has been of anonymity and lack of immediate social censoring may
observed to be more sensitive to unusual rather than to neutral make such biases more prevalent online (Guan et al. 2015).
stimuli, suggesting that our response to both attractive and Online transaction platforms (e.g., Uber, Airbnb) typically
unattractive faces may be stronger than to plain-looking ones require sellers to upload real photos as their profile pictures
(e.g., Said, Baron, and Todorov 2008). and to display them in prominent positions. These profile pic-
In addition, the amygdala response to facial attractiveness
tures provide impression-bearing information that affects
triggers rapid automatic inferences about people’s dispositions,
source credibility and behavioral outcomes (Forman, Ghose,
which in turn affects subsequent information processing and
and Wiesenfeld 2008; Luca 2017).
decisions (Engell, Haxby, and Todorov 2007). Greater atten-
Studies of the attractiveness effect have mostly used a small
tion to an eye-catching face makes it more likely that people
number of pictures in experimental settings rather than asses-
process additional information associated with the face, which
sing real-world situations, leaving the robustness and general-
may weaken but not change the nature of the relation between
izability of their findings open to question (Langlois et al.
inferences from faces and decisions (Todorov et al. 2005; Vuil-
2000). It is not clear from the literature whether social stereo-
leumier 2000). Thus, advertisers find it effective to use either
attractive or unattractive models to present certain products types based on attractiveness extend to the C2C e-commerce
(Guihaire 2018). context. Researchers usually adopt a linear model or compare
only two levels of attractiveness (i.e., attractive vs. less attrac-
tive), neglecting any potential nonlinear effect. Thus, C2C
Beauty and Ugliness Premiums e-commerce platforms involving ordinary people provide an
Studies in many fields have concluded that beauty has a pre- excellent setting to explore the effect of beauty and ugliness
mium and ugliness is penalized (Eagly et al. 1991; Langlois premiums and their underlying mechanisms.
Peng et al. 3
Product Relevance/
Cross-Gender Effect
Sociability/
Competence
Product Sales/
Online Profile Pictures: Source
Purchase
Attractive–Unattractive Credibility
Intention
Figure 1. Conceptual framework.
Research Framework and desirable trait inferences, which in turn leads to a greater
likelihood of a sale.
To explore the potential nonlinear effect of facial attractiveness
on product sales, we focus on the profile pictures of ordinary
H1: Holding other things constant, there is a U-shaped rela-
people in C2C platforms who display a wide range of attrac-
tionship between product sales (or purchase intention) and
tiveness (i.e., attractive, plain-looking, and unattractive). To
facial attractiveness of online sellers in that both attractive
investigate the mechanism underlying the beauty and ugliness
and unattractive people perform better than plain-looking
premiums, we conduct online experiments to assess the effect
people.
of seller attractiveness on perceptions of sociability and com-
petence, which in turn affect source credibility and purchase
However, gaining more attention cannot solely justify the
intention. With these objectives in mind, we present our con-
advantages of attractive and unattractive faces over plain faces.
ceptual framework in Figure 1 and elaborate the hypotheses in
In line with the implicit personality theory, trait inference is the
the ensuing sections.
key mechanism underlying the effect of attractiveness (Eagly
et al. 1991). In addition to the primary messages such as prod-
Mechanism for Beauty and Ugliness Premiums uct quality and price, source credibility is a key factor that
Faces have a special advantage over other stimuli in terms of influences consumer decisions (Goldsmith, Lafferty, and New-
visual processing and the attention-orienting mechanism (Vuil- ell 2000). Consumers use nonverbal cues (e.g., face attractive-
leumier 2000). On e-commerce platforms that involve infor- ness) to infer the perceived trustworthiness and expertise of a
mation overload, unusual faces (i.e., both attractive and source (i.e., two determinants of source credibility), which in
unattractive) have high arousal values compared with plain- turn affects their perceptions of the products (Ohanian 1990).
looking faces, and thus their messages are more likely to pass Previous research has suggested that a salesperson’s attrac-
through the attention gate rather than being ignored. Recent tiveness does not directly affect sales performance but, rather,
studies using fMRI have found that the amygdala, the part of influences some aspects of the customer’s impression of the
the brain responsible for visual attention and processing, exhi- salesperson such as sociality or competence (Ahearne, Gruen,
bits nonlinear responses to human faces as both attractive and and Jarvis 1999; Debevec, Madden, and Kernan 1986). In
unattractive faces elicit quicker and stronger responses than online platforms, the pictorial and aesthetic features of profile
plain-looking ones (Martı́n-Loeches et al. 2014; Said, Baron, pictures have a profound influence on consumers’ assess-
and Todorov 2008; Winston et al. 2007). ments of source credibility (Carusi 2008). Meanwhile, attrac-
Moreover, people instantaneously assign a set of tiveness has been found to be moderately correlated with
personality-like traits and judgments to faces, particularly perceived sociability, less so with competence, and almost
along the dimensions of warmth and competence (Fiske, not at all with honesty (Eagly et al. 1991; Grabe, Ward, and
Cuddy, and Glick 2007). Research suggests that good- Hyde 2008). Thus, it is plausible that beauty and ugliness
looking people are regarded as more sociable, likable, intelli- premiums operate under different mechanisms in terms of
gent, and persuasive (Hamermesh 2011). Unattractive people sociability and competence.
may obtain positive judgments derived from inferences of com- Beauty, as an endowment, has many benefits. Because of
petence (Gheorghiu, Callan, and Skylark 2017). Thus, while beauty’s halo effect, attractive faces lead to a higher level of
the beauty premium will be apparent in online platforms, we arousal and inferences of sociability and competence (Langlois
also expect that unattractive sellers elicit positive perceptions et al. 2000). Strong empirical evidence suggests that attractive
in certain contexts, which we elaborate in the next section. We individuals are perceived to possess more socially desirable
propose that compared with plain-looking faces, both attractive traits and exhibit greater persuasive power in selling products
and unattractive sellers command greater consumer attention with which they are associated (Eagly et al. 1991; Ohanian
1990). Thus, because attractive individuals are perceived as on social interactions (Agthe, Spörrle, and Maner 2011),
more likable and competent, they are considered more credible whereas the ugliness premium plays a role in assessing profes-
than plain-looking and unattractive ones. sional competence (Gheorghiu, Callan, and Skylark 2017;
Kanazawa and Still 2018). Therefore, the context of evaluation
H2: Compared with plain-looking faces, attractive faces influences the effect of beauty and ugliness premiums on trait
enhance (a) perceived sociability and (b) competence, perceptions and outcomes.
which in turn affect (c) source credibility and (d) purchase Product relevance is well grounded in the existing literature
intention. on attractiveness in marketing and serves as a key moderator on
how the attractiveness of an endorser or salesperson affects
For unattractive faces, attention alone may not be sufficient
their performance (Trampe et al. 2010). According to the
to induce a positive effect. In light of the overwhelming beauty
match-up hypothesis, endorsers of various degrees of attrac-
premium for attractive people and the ugliness penalty in socia-
tiveness are more effective when their perceived ability and
bility, unattractive people have an advantage only over plain-
credibility are relevant for presenting and interpreting the prod-
looking people in perceived competence for several reasons.
ucts (Kamins 1990). Following this logic, we expect that the
Compensatory adaptation is a widely held perception that unat-
advantages of attractive and unattractive faces are greater when
tractive people often work harder to compensate for their dis-
they are aligned with a product relevant to the positive traits
advantaged appearance, leading to a perception of greater
derived from their appearances. Whereas attractive people are
competence than those with plain-looking faces (Kock 2003).
more effective in presenting appearance-relevant products that
There is an ingrained perception that whereas attractive people
enhance sociability, unattractive faces bring an advantage over
obtain everything more easily, particularly in settings that
plain-looking faces when they are associated with a product
require social skills, unattractive people must exert greater
related to technical or professional expertise (e.g., Bower and
effort to compensate for their disadvantaged appearance and
Landreth 2001; Kang and Herr 2006).
often shift to areas that do not demand social skills, such as
scientific pursuits (Fidrmuc, Paphawasit, and Tunalı 2017). H4: Product relevance moderates the mediating effect of
Moreover, the “ugly Einstein” effect suggests that the sociability (competence) between beauty (ugliness) pre-
stereotypical expert may be an impartial truth seeker with lim- mium and source credibility. (a) The mediating effect of
ited personal appeal (Crane and Patterson 2012; Gheorghiu, sociability is stronger for attractive people selling
Callan, and Sylark 2017). The stereotypical belief that attrac- appearance-relevant products, whereas (b) the mediating
tiveness and intelligence are negatively associated is also pre- effect of competence is stronger for unattractive people sell-
valent, particularly for women (Heilman et al. 2004). This ing expertise-relevant products.
argument is used to explain the “dumb blonde” stereotype, in
which attractive women rely on their looks to advance rather
than intelligence (Ruffle and Shtudiner 2015). Not surprisingly, Gender Differences
much anecdotal evidence exists on the perceived creativity and
Gender greatly influences perceptions based on appearance,
extraordinary characteristics of unattractive people (Guihaire
and gender bias can be regarded as a subset of attractiveness
2018; Kaplan 1978). Our research extends these notions and
bias (Agthe, Spörrle, and Maner 2011). Unlike dating or hiring,
proposes that the ugliness premium operates through perceived
online shopping does not involve social decisions or represent a
competence, which in turn enhances source credibility and
competitive environment, so the negative vigilance toward an
purchase intention.
attractive person of the same sex found in previous studies is
H3: Compared with plain-looking faces, unattractive faces unlikely (e.g., Maner et al. 2009). Due to the opposites attract
enhance (a) perceived competence, which in turn affects (b) principle, studies have found that people are more subject to the
source credibility and (c) purchase intention. influence of attractiveness in the opposite sex (Kaplan 1978; Li
and Kenrick 2006). Thus, we expect that beauty and ugliness
premiums are stronger in a cross-gender context than in a same-
Product Relevance gender one.
Studies in labor economics and human resource management Moreover, attractiveness affects men and women differ-
suggest that based on perceived fit, people of various degrees of ently. Studies of evolutionary psychology indicate that men
attractiveness may self-select or be selected into occupations place greater importance on female attractiveness, so male
and positions that are “suitable” for their appearance (Biddle consumers are more likely to award a beauty premium to attrac-
and Hamermesh 1994; Heilman et al. 2004). Attractive people tive female sellers (Agthe, Spörrle, and Maner 2011). How-
are perceived as more fitting for positions in which a pleasing ever, although women may value attractiveness in men, they
appearance and sociability are appreciated, whereas unattrac- are more likely to prioritize other considerations and place
tive people are regarded as more competent in professions for greater emphasis on competence and favor status and intelli-
which technical or professional expertise matters (Gheorghiu, gence in men because these qualities indicate the ability to
Callan, and Skylark 2017; Lee et al. 2018). Likewise, the acquire resources and provide security (Sprecher, Sullivan, and
beauty premium has been found to accrue in situations centered Hatfield 1994). Thus, the ugliness premium in competence is
Peng et al. 5
likely to be stronger for female consumers looking at male square test on the distribution of ratings between male and
sellers. female raters also revealed no significant differences.
Second, we used image processing techniques to retrieve
H5: Compared with a same-gender setting, (a) the mediating key pictorial features. Substantial evidence from computing
effect of sociability for attractive female sellers is stronger and aesthetics research suggests that symmetry and propor-
for male buyers and (b) the mediating effect of competence tional facial features (e.g., distance between the eyes, cheek
for unattractive male sellers is stronger for female buyers. width, size of nose and forehead) are good predictors of facial
attractiveness (Gunes and Piccardi 2006). We used a set of 68
In the following sections, we use field data from two transac-
facial landmarks to extract these features and compute various
tional sites to provide empirical evidence for the U-shaped
facial ratios and proportions. For example, the aesthetic stan-
relationship between seller attractiveness and product sales
dard of the golden ratio can be obtained by comparing the
(Study 1). As social traits and source credibility are not directly
distance between the eyes and mouth to the distance between
observable online, we examine the different mechanisms
the mouth and chin.
underlying the beauty and ugliness premiums and the moder-
Third, we applied several machine learning methods (linear
ating effects of product relevance and gender in two online
regression, Bayesian ridge regression, Gaussian regression,
shopping experiments (Study 2).
support vector machine regression, random forest regression,
and convolutional neural networks) to learn the relationship
between facial geometrics and the attractiveness scores from
Study 1: Profile Pictures in Online Platforms the human raters. We used 80% of the portrait data as the
training set for model fitting and the remaining 20% for valida-
Research Settings tion and model selection. Random forest regression achieved
Airbnb and 5miles are the research settings for our field studies. the best performance in terms of the Pearson correlation, the
These C2C e-commerce platforms provide information on sell- mean absolute errors, as well as the computational cost. We
ers, including their photos, which serve as a means of identity thus applied random forest regression to predict the facial
verification and narrow the social distance between buyers and attractiveness of all the profile pictures in the Airbnb and
sellers (Luca 2017). Airbnb is a sharing economy platform in 5miles data sets as follows:
which travelers are matched with hosts who have properties for
r n ð X; D n Þ ¼ E Y ½ r n ð X; Y m ; D n Þ; ð1Þ
rent. We examine how an Airbnb host’s facial attractiveness
affects their listing’s occupancy rate. 5miles is a mobile app where rn (X, Ym, Dn) refers to the randomized base regression
that connects buyers with sellers of different products, and it tree. Y1, Y2, . . . , Ym are identically and independently distrib-
enables us to assess the effect of a seller’s facial attractiveness uted outputs of the randomized variables. EY denotes the
on the likelihood of a sale. expectation with respect to the random parameters, conditional
on the data set Dn and X. The predicted scores are highly
correlated with those of the raters (r ¼ .71). In addition to the
Assessing Facial Attractiveness with Machine Learning
cross-validation using the training data set, we adopt two other
Determining the facial attractiveness of profile pictures on procedures1 to assess the accuracy of machine learning. The
Airbnb and 5miles is a challenging task, as these sites have validation tests suggest that the algorithm works well for faces
over 1 million hosts and 100,000 sellers, respectively. Because from different genders, age, and ethnicity, and that the results
standards of facial attractiveness are universal across cultures, from human raters are highly consistent with those from ran-
ethnic groups, sexual orientations, and ages, facial attractive- dom forest regression.
ness is a quantifiable trait that can be assessed by both people
and computer algorithms (Langlois et al. 2000; Magro 1999).
We apply a machine learning method to process a large quan- Controls for Potential Confounds
tity of profile pictures with a high level of accuracy, making
cumbersome human coding of all the portraits unnecessary. To rule out potential confounding factors, we control for sev-
First, we retrieved a random sample of 32,386 profile pic- eral pictorial features including photographic quality, face
tures from Yelp and recruited ten male and ten female raters proximity, and smiling expressions. Profile pictures vary in
between 19 and 25 years old. Each image was randomly resolution, brightness, and quality and range from headshots
assigned to five raters (two men and three women or three men with high facial prominence (close-up) to full-body shots with
and men women), who scored it on a five-point scale from 1 low facial prominence (distant). We used the vision libraries
(“very unattractive”) to 5 (“very attractive”). The final attrac- available in OpenCV to derive the hue, saturation, and value
tiveness score is the average of the five ratings. We randomly
divided the raters into two groups and consistently obtained 1
Web Appendix 1 presents the procedure and results about inviting MTurkers
correlations of .87 to .96 for the average ratings between to code a random sample of profile photos. Web Appendix 2 reports the details
groups. The insignificant t-statistic confirms that the raters from using human coders to score 2,750 host pictures required by SIMEX
used similar criteria to assess facial attractiveness. A chi- approach to assess measurement errors.
(HSV) color spectrum for each picture and then aggregated effect of pictorial features including facial attractiveness. Xhl
these measures into a single index of photographic quality represents a set of listing characteristics and review character-
using principal component analysis. We measured face prox- istics. Yh denotes a set of host characteristics. The random
imity as the ratio of the area of a face to the whole picture, intercept, uh, is a host-specific error component that accounts
ranging from 0 to 1. The higher (lower) the facial proximity for unobserved heterogeneity across hosts, and ehl is a listing
ratio, the more (less) prominent the face is in the picture. In error component that varies between listing l and host h.
addition, a smiling face may be equally effective as the attrac-
tiveness of the seller because it can evoke a sense of familiarity
and increase the positive evaluations of viewers (Scharlemann Results. As Table 2 shows, the coefficients for pictorial features
et al. 2001). We used a random forest regression model to are as expected across all specifications. The presence of a
predict the likelihood of smiling for each profile picture in the profile picture has a positive effect, resulting in an approxi-
main sample.2 mately 4.1% increase in occupancy rate (Spec. 1). Better photo-
graphic quality (Spec. 2) and smiling expression (Spec. 3) are
positively related to occupancy rate. The results of Spec. 3
Study 1a: Sharing Economy Platform (Airbnb) show that a one-unit increase in a host’s facial attractiveness
Data collection. We collected all publicly available data for can increase the occupancy rate by approximately 1.3%, sug-
Airbnb listings in Los Angeles through June 15, 2017. We then gesting that the beauty premium is prevalent. We introduced
appended the annual occupancy data from AirDNA (a major the quadratic term in Spec. 4 and use a three-step procedure to
supplier of data on worldwide Airbnb listings) covering the test the U-shaped relationship between facial attractiveness and
same period. We combined the data from these sources and occupancy rate (Lind and Mehlum 2010). First, the results of
excluded properties without complete information (e.g., the Spec. 4 in Table 2 show the joint significance and expected
ones less than one year in operation). We downloaded host signs of the direct effect (b ¼ .911, p < .01) and the squared
profile pictures and used image processing techniques to term effect (b ¼ .150, p < .01). Second, as shown in Figure 2,
extract the pictorial features. The final sample consists of Panel A, the slope of the lower end is significantly negative
17,935 Airbnb properties from 10,979 hosts. Of these listings, (.341, p < .01), and the slope of the higher end is significantly
17,749 have a profile picture and 9,953 use a single portrait. positive (.368, p < .01). Third, the turning point (3.04, p < .01)
We controlled for (1) host characteristics (e.g., identity verifi- is significant and located well within the data range. Thus,
cation, reputation), (2) listing characteristics (e.g., accommo- these results support H1. To account for potential social inter-
dation type, price, postal code, age of listing, quality of the actions, we controlled for rental type (room vs. whole unit) by
listing photos), and (3) review characteristics (e.g., number of assuming that guests expect to meet their hosts face-to-face
reviews, property ratings by reviewers, review sentiment). when renting a room in the unit. While shared apartment/house
These three groups of variables are critical to rule out potential has a significant negative effect, the interaction between shared
confounds. For instance, postal codes are frequently used to unit and facial attractiveness is found to be insignificant, and
control for socioeconomic differences such as housing quality thus no concern of social pressure from the expectation of
across geographic areas, which may affect the outcome vari- meeting an attractive host is present.
able. This is also true for review volume and sentiments. The
dependent variable is the annual occupancy rate, which is a
proxy for sales performance. Table 1 provides the variable Correction of measurement errors3. In Specs. 5 and 6, we used the
definitions and summary statistics. simulation extrapolation method (SIMEX) to account for mea-
surement errors in the machine learning approach. SIMEX is a
Model specification. We used a hierarchical framework to assess data-driven approach to correcting measurement errors and
the effect of host attractiveness on occupancy rate. Approxi- requires relatively fewer assumptions and information than
mately 24% of the hosts own more than one listing, and thus the alternative methods (Yang et al. 2018). We followed its diag-
unit of observation is a listing. We estimated the model in a nostic procedure to assess the measurement error using the
stepwise fashion. The baseline econometric model is as known attractiveness scores rated by human coders in a random
follows: sample of host pictures from 2,750 listings. Compared with the
Occupancy hl ¼ b 0 þb 1 ð pictorial featuresÞ þ X hl b 2 naive model, the parameter estimates of facial attractiveness
ð2Þ using the SIMEX corrected model are larger in magnitude,
þ Y h b 3 þ u h þ e hl ; suggesting that the naive model may underestimate the effects.
where Occupancyhl is a measure of the annual occupancy rate The other variables, however, change little in the presence of
of listing l owned by host h. The parameter of interest is b1, the measurement error.
2
We searched for a “smiling human face” and “neutral human face” in a
Google image search. After extracting facial geometrics, we used the
3
random forest regression to predict the likelihood of smiling in each of the Web Appendix 2 reports the details for the correction of measurement errors
profile pictures. Web Appendix 1 presents the details. using simulation extrapolation.
Peng et al. 7
Table 1. Summary Statistics of Airbnb Data (Study 1a).
Variable Definition N M SD Min Max
Pictorial Characteristics
Presence of picture Presence of profile picture 17,935 .990 .101 0 1
Human portrait Presence of human portrait 17,749 .711 .454 0 1
Photographic quality Aggregated measure of HSV (hue, saturation, value) and picture 17,749 .303 .154 0 .962
resolution
Facial attractiveness Face attractiveness score determined by the machine learning 9,953 3.05 .433 1.91 4.26
approach
Smiling expression Likelihood of smiling expression determined by the machine 9,953 .645 .257 0 1
learning approach
Face proximity (%) Ratio of the area of a face to the whole picture 9,953 .111 .090 .001 .820
Host Characteristics
Superhost Binary indicator of whether the host has a superhost badge 10,979 .206 .404 0 1
representing a good reputation
Verified home email Binary indicator of whether the account is verified by home email 10,979 .972 .166 0 1
Verified work email Binary indicator of whether the account is verified by work email 10,979 .136 .342 0 1
Verified government-issued ID Binary indicator of whether the account is verified by government- 10,979 .445 .497 0 1
issued ID
Verified phone number Binary indicator of whether the account is verified by phone number 10,979 .996 .060 0 1
Verified selfie with ID Binary indicator of whether the account is verified by selfie with ID 10,979 .023 .149 0 1
Linked Facebook account Binary indicator of whether the account is linked to Facebook 10,979 .275 .446 0 1
account
Linked Google account Binary indicator of whether the account is linked to Google account 10,979 .065 .247 0 1
Linked LinkedIn account Binary indicator of whether the account is linked to LinkedIn 10,979 .029 .168 0 1
Listing Characteristics
Response rate % of new inquiries and reservation requests the host responded to 17,935 .939 .158 0 1
within 24 hours in the past 30 days
Average daily rate The average rate paid for rooms booked 17,935 148.76 164.386 6.670 4,290
Annual occupancy rate % of total available days in the year with a confirmed booking 17,935 .590 .256 .032 1
Apartment Binary indicator of whether the listing is an apartment 17,935 .576 .494 0 1
House Binary indicator of whether the listing is a house 17,935 .291 .454 0 1
Shared apartment/house Binary indicator of whether the apartment/house is a shared unit 17,935 .350 .477 0 1
# of listing photos Number of property photos shown 17,935 18.765 13.886 1 255
Quality of main listing photo Aggregated measure of HSV (hue, saturation, value) and picture 17,935 .506 .261 .049 .950
resolution
Listing postal code A series of dummy variables indicating the listing location, where X ¼ 1 (X 2 (all postal codes in Los Angeles)) if
the listing is located in the ZIP code tabulation areas X; 0 otherwise.
Joined year (age of listing) The year when the listing joined Airbnb
Review Characteristics
Property rating Average star rating by reviewers 17,935 4.663 .435 1 5
Ln (# of reviews) Log of the total number of reviews received 17,935 2.424 1.375 0 6.084
Review subjectivity Average subjectivity of customer review 17,935 .625 .039 .17 .95
Review polarity Average polarity of customer review 17,935 .411 .065 .344 1
Review readability Average readability of customer review 17,935 67.92 7.57 .92 100
Notes: The number of reviews is incremented by one before the log transformation. We assess the review polarity from 1 (negative) to 1 (positive), the
subjectivity score from 0 (objective) to 1 (subjective), and readability using Flesch reading ease scale from 0 to 100.
Study 1b: E-Commerce Platform (5miles) profile pictures. We controlled for seller characteristics
(e.g., trust level, star rating, gender, identity verifications),
Data collection. To validate the findings of Study 1a, we tested
and product characteristics (e.g., product category, number
the model using data from 5miles. We tracked a random sam-
of product photos, length of product description, price). We
ple of product listings on a daily basis for 60 days (January 31
used a topic modeling approach4 (guided latent Dirichlet allo-
to March 31, 2019) in three product categories—beauty prod-
cation) to classify the products on the basis of the degree of
ucts (11,842 items), electronics (7,171 items), and bags (7,215
relevance to either appearance or expertise. Drawing on the
items)—resulting in a sample of 26,228 items from 11,115
sellers. Approximately 46% of the products received at least
one offer during the observation period. We used the same 4
Web Appendix 2 provides details of the topic modeling approach with guided
method as in Study 1a to extract facial features from seller latent Dirichlet allocation.
Table 2. Estimation Results for Occupancy Rate (Study 1a).
Spec. 6
Spec. 1 Spec. 2 Spec. 3 Spec. 4 Spec. 5 (SIMEX) (SIMEX)
Presence of picture .041** – – – – –
(.020)
Photographic quality – .059*** .062*** .059*** .052*** .044**
(.014) (.018) (.018) (.016) (.017)
Human portrait – .025*** – – – –
(.005)
Smiling expression – – .024** .044*** .024*** .064***
(.011) (.011) (.009) (.009)
Face proximity – – .221*** .228*** .193*** .211***
(.032) (.032) (.028) (.029)
Facial attractiveness – – .013** .911*** .014** 1.899***
(.007) (.082) (.006) (.150)
Facial attractiveness2 – – – .150*** – .310***
(.013) (.024)
Listing Characteristics
Response rate .151*** .149*** .162*** .161*** .169*** .167***
(.014) (.015) (.020) (.020) (.018) (.015)
Average daily rate 2.26e-04*** 2.24e-04*** 2.59e-04*** 2.62e-04*** 2.85e-04*** 2.81e-04***
(2.25e-05) (2.26e-05) (3.90e-05) (3.73e-05) (3.62e-05) (3.38e-05)
Apartment .024*** .024*** .014* .012* .019** .019**
(.006) (.006) (.008) (.008) (.007) (.007)
House .022*** .022*** .020** .019** .015* .014*
(.007) (.007) (.009) (.009) (.008) (.008)
# of listing photos 1.61e-04 1.86e-04 1.95e-04 1.90e-04 1.71e-04 1.23e-04
(1.56e-04) (1.56e-04) (2.26e-04) (2.24e-04) (2.20e-04) (2.20e-04)
Quality of main listing photo .003 .002 .003 .003 .005 .002
(.006) (.006) (.008) (.008) (.008) (.009)
Shared apartment/house .068*** .068*** .072*** .072*** .076*** .076***
(.005) (.005) (.007) (.007) (.006) (.006)
Postal code fixed effect Yes Yes Yes Yes Yes Yes
Joined year fixed effect Yes Yes Yes Yes Yes Yes
Host Characteristics
Superhost .019*** .018*** .018*** .017** .012** .011*
(.005) (.005) (.007) (.007) (.005) (.006)
Host identity variables Yes Yes Yes Yes Yes Yes
Review Characteristics
Property rating .015*** .015*** .014* .014** .022*** .025***
(.005) (.005) (.007) (.007) (.007) (.007)
ln (# of reviews) .063*** .063*** .058*** .057*** .063*** .062***
(.002) (.002) (.002) (.002) (.002) (.002)
Avg. review polarity .060* .062** .031 .034 .020 .027
(.031) (.031) (.041) (.040) (.047) (.043)
Avg. review subjectivity .081 .076 .115 .128* .170** .192***
(.054) (.054) (.071) (.071) (.072) (.071)
Avg. review readability 1.14e-03*** 1.17e-03*** 1.26e-03*** 1.25e-03*** 1.56e-03*** 1.55e-03***
(2.27e-04) (2.29e-04) (3.11e-04) (3.08e-04) (3.31e-04) (3.64e-04)
# of observations 17,935 17,749 9,953 9,953 9,953 9,953
Notes: Host identity/verification information is also included in estimation but not reported for brevity. Heteroskedasticity consistent robust standard errors are
reported in parentheses. *p < .10; **p < .05; ***p < .01.
topic dominance in product descriptions, we classify 5,977 Model specification. We specify the utility that affects the sale of
listings as expertise-relevant, 8,841 listings as appearance- product j as follows:
relevant, and the other listings for which neither appearance
U ijt ¼ b 0 þ b1 PD ij þ X jt b 2 þ S i b3 þ e ijt ; ð3Þ
nor expertise topics are dominant serve as the baseline group.
Table 3 provides the variable definitions and summary where PDij is the picture decision of seller i who lists product j,
statistics. X jt represents a vector of product characteristics, and S i
Peng et al. 9
1.0 A: Airbnb
at risk. We estimated the model using a binary choice model
with time fixed effects, as it is equivalent to a piecewise expo-
nential hazard model when the data are observed at discrete
.9
time points. We thus adopted the probit specification and Equa-
tion 4 as a discrete time duration model.
Occupancy Rate
.8 .78 Estimated
slope at xh
Estimated slope at xl .72 Pr y1 ¼ 1 jPD ij ; X jt ; S i
.7 ð5Þ
Turning point = 3.04
¼ F b0 þ b1 PD ij þ X jt b2 þ S i b3 þ m i þ k t t 0 ;
.62
.6 .60 where y1 ¼ 1 if Uijt > 0, and kt t0 represents a set of temporal
.56
dummy variables.
xl xh
.5
1 2 3 4 5
Facial Attractiveness
Results. The Kaplan–Meier survival curves in Figure 3 show
that at any point in time, sellers with profile pictures of them-
.8 B:
selves are more likely to receive offers from buyers sooner (i.e.,
the lowest survival rate) than those with nonhuman pictures or
.7 without profile pictures. The survival curves for the three
.62
groups show that plain-looking sellers are associated with a
higher survival rate, suggesting that their listings (compared
Sales Probability
.6
with either attractive or unattractive sellers) have a longer sales
cycle. Again, the results of Spec. 1 and Spec. 2 in Table 4
.50
.5 suggest that the mere presence of a profile picture (b ¼ .255,
.44 p < .01) and a human portrait (b ¼ .118, p < .01) are positively
.41
.4 .38
related to sales performance. After controlling for smiling
expressions in Spec. 3, the coefficient of facial attractiveness
xl xh
remains significantly positive (b ¼ .086, p < .01). We include
.3
1 2 3 4 5
the quadratic term of facial attractiveness in Spec. 4, and the
Facial Attractiveness result is consistent with that of Study 1a, in that both attractive
and unattractive sellers are more likely to receive offers sooner
Figure 2. Relationship between facial attractiveness and sales than plain-looking sellers (slope: b ¼ 1.001, quadratic term:
performance. b ¼ .177; p < .01). Thus, H1 is again supported. In Spec. 5, we
Notes: This curve is drawn at the average level for all other variables. introduce the interaction between facial attractiveness and
product relevance. Compared with less attractive sellers, attrac-
represents seller characteristics. To accommodate unobserved tive sellers perform better for appearance-relevant products
seller heterogeneity, we split the error term (eijt ¼ mi þ Eijt) into (b ¼ .082, p < .10) but worse for expertise-relevant products
mi * N (0, s2), which is specific to seller i, and Eijt, which is (b ¼ .138, p < .05). These results provide support for H4.
unique for each listing.
Most sellers have only one item to sell, and this item may be
requested by multiple buyers at different times. While a sale is Robustness Checks5
made to one of the offers, we cannot observe which offer
received a sale. Thus, the time to receipt of the first offer is We tested the robustness of results in a number of ways. We
one of the most important outcomes that can be attributed to obtained the variance inflation factors for all the covariates in
facial appearance, among other factors. Given this dynamic Study 1a (see Table W2-6 in Web Appendix 2). They are all
process, we model the time-to-offer using a discrete-time pro- below the conventional threshold of 4, indicating that multi-
portional hazard model: collinearity does not appear to be a concern. We then explore
the potential problem arising from outliers. For example, we
Prð d ijt T< d ijt þ DtjT> d ijt Þ excluded observations within the top 5% of the average daily
h d ijt ; PD ij ; X jt ; S i ¼ lim
Dt!0 Dt rate (Spec. 2). We also excluded listings in the top 5% of
the distribution of occupancy (Spec. 3). The reestimated
¼ h0 d ijt expf b 0 þ b1 PD ij þ X jt b 2 þ S i b3 þ m i g; ð4Þ
results remain robust in terms of sign, magnitude, and statisti-
where h(dijt, PDij, Xjt, Si) is the hazard rate for product listing j cal significance.
receiving an offer in time period t given that it has not received
an offer before time t, and T is a stochastic representation of the 5
Web Appendix 2 reports the detailed results of robustness checks including
time duration. h0(dijt) is the baseline hazard rate capturing the multicollinearity, outliers, alternative DVs, and alternative U-shaped
likelihood of receiving an offer. The hazard rate depends on specifications for Study 1a and 1b. It also provides details for the propensity
both the independent variables and the length of time a listing is score matching method to address potential selection bias.
Table 3. Summary Statistics of 5miles Data (Study 1b).
Variable Definition N M SD Min Max
Presence of picture Presence of profile picture 26,228 .853 .354 0 1
Human portrait Presence of human portrait 22,371 .453 .498 0 1
Photographic quality Aggregated measure of HSV and picture resolution 22,371 .295 .162 0 .982
Facial attractiveness Face attractiveness score determined by the machine learning approach 8,184 3.08 .425 1.99 4.35
Smiling expression Likelihood of smiling expression determined by the machine learning 8,184 .543 .254 .020 1
approach
Face proximity (%) Ratio of the area of a face to the whole picture 8,184 .188 .142 .001 .949
Seller Characteristics
Female Binary indicator of the seller gender: female ¼ 1, otherwise ¼ 0 11,115 .523 .498 0 1
Trust level of seller Seller’s trust level determined by the platform 11,115 2.48 2.19 0 11
Seller star rating Average star rating by reviewers 11,115 .707 .431 0 1
Log (# of seller followers) Log number of followers the seller has 11,115 2.81 1.30 0 7.89
Verified email Binary indicator of whether the account is verified by email 11,115 .736 .441 0 1
Verified phone number Binary indicator of whether the account is verified by phone number 11,115 .953 .211 0 1
Linked Facebook account Binary indicator of whether the account is linked to Facebook account 11,115 .395 .489 0 1
Product Characteristics
# of product photos Number of product photos shown 26,228 2.99 2.23 0 12
Log length of listing Log of the total number of words in the product description 26,228 2.46 1.15 .69 7.06
description
Price of the product Listing price 26,228 107.36 236.06 1 7,000
Offer made by buyers Binary indicator of whether an offer is received 26,228 .460 .498 0 1
Notes: The number of seller followers and length of listing descriptions are incremented by one before the log-transformation.
For Study 1a, the use of linear regression may be inappropri- We also adopted the propensity score matching approach to
ate if the dependent variable is not normally distributed. The examine the sample with and without profile pictures and find
residuals of the model fit are approximately normal, suggesting them to be comparable in terms of products, seller, and review
that the possible violation of nonnormality is not likely. We characteristics.
also took the log-transformation of the occupancy rate and
rerun the model in Spec. 4 and the results remain consistent.
In addition, the use of a percentage as a dependent variable
Study 2: Online Experiments
(i.e., occupancy rate) in ordinary least squares regression may To investigate the mechanism underlying the beauty and ugli-
cause predictions that are nonsensical (below 0 or above 1). We ness premium, we first conduct online experiments to examine
thus rerun the model using beta regression, which is appropri- the mediating roles of perceived sociability and competence in
ate for a response variable that is restricted to the interval (0, 1), the relationship between seller attractiveness and source cred-
and find that the parameter estimates remain robust. To explore ibility and purchase intention.
the alternative specifications of the U-shaped relationship. We
used the inverse form rather than the quadratic form to specify Study 2a: Main Effects and Mechanisms
the relationship between facial attractiveness and sales. The
Stimuli. We selected seller photos from Chicago Face Database,
results of the parameter estimates are robust, as both the direct
which provides high-resolution, standardized photographs of
and inverse terms are significant. In addition, including the
male and female faces. Extensive norming data are available
cubic term of facial attractiveness does not improve model fit,
for each individual photo including physical attributes as well
thus further supporting the U-shaped relationship. For Study
as subjective ratings by independent judges (e.g., attractive-
1b, in addition to the duration to receiving an offer, we used the
ness, trustworthy, feminine/masculine). The manipulation of
seller’s offer (i.e., a sale dummy) as an alternative dependent attractiveness, while successful, may influence the perception
variable and find the parameter estimates to be consistent with of seller trustworthiness. Following previous research (e.g.,
the duration survival model (Figure 2, Panel B). Kamins 1990; Till and Busler 2000), we avoided this problem
Finally, sellers’ uploading portraits with varying degrees of by choosing sellers who vary in attractiveness yet are of equiv-
attractiveness may affect the accuracy of the parameter esti- alent trustworthiness. To control for facial expressions and
mates. We examined the distribution of attractiveness scores gender, we chose three male and three female photos with
for both data sets and find them to be normally distributed. For attractive, plain-looking, and unattractive faces, all with neutral
Airbnb data, we found an insignificant correlation between expressions. Except for the photos, the scenario for the shop-
hosts’ facial attractiveness and property ratings (r ¼ .0016). ping task was identical across conditions.
Peng et al. 11
familiar”) and perceived trustworthiness (“The seller is some-

A: Profile Picture
1 one I feel I can trust/never tries to mislead me/is always honest
in his/her dealing with others”; a ¼ .89; Sirdeshmukh, Singh,
0.8 and Sabol 2002).
Survival Probability
0.6 Manipulation check. The participants rated the attractive

sellers (Mattractive ¼ 3.55) as significantly more attractive than
0.4
the plain-looking (Mplain ¼ 3.03; p < .01) and unattractive
0.2 (Munattractive ¼ 2.57; p < .01) sellers. All pairwise comparisons
between conditions are significant at the .01 level, and there is
0 no significant difference in attractiveness between male and
0 10 20 30 40 50 60
Time (days)
female sellers within the same condition. The differences in
perceived trustworthiness turn out to be insignificant among the
No proﬁle picture Nonhuman picture Human portrait
three groups (Mattractive ¼ 3.37, Mplain ¼ 3.43, Munattractive ¼
B: Seller Attractiveness 3.54; F(2, 347) ¼ .84, p ¼ .43).
1
Visual attention. To examine whether unattractive and attractive
0.8 faces on the first page receive more attention from participants,
we recorded the browsing time between when a seller picture
Survival Probability
0.6 is completely loaded and when the “next” button is clicked.

We find that the participants take more time (in seconds) to
0.4
browse the pages of either attractive or unattractive faces than
0.2
those of plain-looking faces (Mattractive ¼ 34.29, Mplain ¼ 26.75,
Munattractive ¼ 33.03; F(2, 347) ¼ 3.22, p < .05). Given that
0 everything except for the picture is identical across the groups,
0 10 20 30 40 50 60
this finding confirms the U-shaped relationship between attrac-
Time (days)
tiveness and attention.
Unaracve seller Plain-looking seller Aracve seller
Purchase intention and source credibility. Consistent with the find-

Figure 3. Kaplan–Meier survival curves (Study 1b). ings from field studies, seller attractiveness has a U-shaped
relationship with purchase intention (F(2, 347) ¼ 4.18, p <
.05), in support of H1. Changing from unattractive to plain-
looking decreases purchase intention (Munattractive ¼ 3.86 vs.
Procedure and measures. We randomly assigned 350 participants Mplain ¼ 3.65; F(1, 229) ¼ 3.44, p < .10). Beyond that point,
(187 men; Mage ¼ 36.76 years, SD ¼ 12.83) recruited from however, additional attractiveness increases the purchase
consumer panelists on Amazon’s Mechanical Turk (MTurk) to intention (Mplain ¼ 3.65 vs. Mattractive ¼ 3.95; F(1, 238) ¼
one of the three (attractive, plain-looking, unattractive) 9.44, p < .01). There is no difference in purchase intention
between-subject conditions. They were first instructed to read between unattractive and attractive conditions (F < 1). As for
the materials describing a hypothetical shopping task for a source credibility, we observe a significant difference among
digital camera and then asked to investigate the seller and their the three conditions (F(2, 347) ¼ 5.97, p < .01). Both attractive
sellers (Mattractive ¼ 4.07 vs. Mplain ¼ 3.74; F(1, 238) ¼ 12.87,
product carefully. They had to click the “next” button to go to
p < .01) and unattractive sellers (M unattractive ¼ 3.91 vs.
the questions. Then they were asked to first indicate their pur-
Mplain ¼ 3.74; F(1, 227) ¼ 2.87, p < .10) are perceived as more
chase intention on a scale from 1 (“I definitely would not buy”)
credible than plain-looking sellers. There is no significant dif-
to 5 (“I definitely would buy”). Next, they assessed the seller’s
ference in perceived credibility between unattractive and
credibility on a four-item scale (“To what extent do you think
attractive faces (p > .10). Source credibility is highly corre-
the source is credible/reliable/trustworthy/an expert?” Chaiken
lated with purchase intention (r ¼ .80).
and Maheswaran 1994). The responses were averaged to form a
composite score of source credibility (a ¼ .91). The partici- Perceived sociability and competence. There is a significant dif-
pants then rated the perceived sociability (“The seller is easy to ference in perceived sociability (F(2, 347) ¼ 9.04, p < .01) and
like/a fun person to be around/like a good friend/a very nice competence (F(2, 347) ¼ 3.81, p < .05) among the three con-
person”; a ¼ .92; MacInnis and Park 1991) and competence ditions. Attractive sellers are perceived as more sociable than
(“The seller is competent/intelligent/capable/skillful”; a ¼ .91; plain-looking ones (Mattractive ¼ 3.63 vs. Mplain ¼ 3.34; F(1,
Wang et al. 2017) of the seller. All these measures use a five- 238) ¼ 7.08, p < .01) and unattractive ones (Mattractive ¼ 3.63
point scale from 1 (“strongly disagree”) to 5 (“strongly agree”). vs. Munattractive ¼ 3.15; F(1, 227) ¼ 18.03, p < .01). The results
To rule out potential confounds, we also measured face famil- reveal no significant difference in sociability between plain-
iarity (1 ¼ “does not look familiar at all,” and 5 ¼ “looks very looking and unattractive sellers (p ¼ .105). Perceived
Table 4. Estimation Results from Duration Model (Study 1b).
Spec. 1 Spec. 2 Spec. 3 Spec. 4 Spec. 5
Hazard Hazard Hazard Hazard Hazard

Estimate Ratio Estimate Ratio Estimate Ratio Estimate Ratio Estimate Ratio
Presence of picture .255*** 1.29 — — — —
(.018)
Photographic quality — .254*** 1.29 .218*** 1.24 .212*** 1.24 .215*** 1.24
(.036) (.063) (.063) (.063)
Human portrait — .118*** 1.13 — — —
(.012)
Smiling expression — — .120*** 1.13 .148*** 1.16 .122*** 1.13
(.037) (.037) (.037)
Face proximity (%) — — .177*** 1.19 .181*** 1.20 .173*** 1.19
(.067) (.066) (.067)
Facial attractiveness — — .086*** 1.09 1.001*** .37 .072*** 1.08
(.020) (.252) (.022)
Facial attractiveness2 — — — .177*** 1.19 —
(.041)
Facial attractiveness ER — — — — .138** .87
(.061)
Facial attractiveness AR — — — — .082* 1.09
(.044)
Seller Characteristics
Female .002 1.00 .016 .98 .044** .96 .042** .96 .043** .96
(.011) (.012) (.019) (.019) (.019)
Trust level of seller .071*** 1.07 .063*** 1.07 .065*** 1.07 .063*** 1.07 .064*** 1.07
(.005) (.005) (.008) (.008) (.008)
Seller star rating .023*** 1.02 .023*** 1.02 .023*** 1.02 .023*** 1.02 .023*** 1.02
(.001) (.001) (.001) (.001) (.001)
ln (# of seller followers) .011 1.01 .021*** 1.02 .021* 1.02 .022* 1.02 .022* 1.02
(.007) (.007) (.012) (.012) (.012)
Seller identity variables (Included in estimation)
Product Characteristics
Price of the product .015*** .99 .014*** .99 .014*** .99 .014*** .99 .014*** .99
(.000) (.000) (.000) (.000) (.000)
# of product photos .016*** 1.02 .017*** 1.02 .023*** 1.02 .023*** 1.02 .023*** 1.02
(.003) (.003) (.005) (.005) (.005)
Log length of listing .013** 1.01 .015*** 1.02 .023** 1.02 .022** 1.02 .022** 1.02
description
(.005) (.005) (.009) (.009) (.009)
Product categories (Included in estimation)
# of observations 26,228 22,371 8,184 8,184 8,184
Log likelihood at 40,756.18 36,820.86 13,531.48 13,522.19 13,523.17
convergence
*p < .10.
**p < .05.
***p < .01.
Notes: Seller verification information and product categories are included in estimation but not reported for brevity. Heteroskedasticity consistent robust
standard errors are reported in parentheses.
competence was significantly higher for unattractive and Mediation. We took a bias-corrected bootstrapping approach
attractive faces than for plain-looking faces (Munattractive ¼ with 5,000 samples to simultaneously test sociability and com-
3.94 vs. M plain ¼ 3.69; F(1, 229) ¼ 6.73, p < .05; petence as mediators, generating a 95% confidence interval
M attractive ¼ 3.87 vs. M plain ¼ 3.69; F(1, 238) ¼ 3.88, around the following paths: (1) from attractive faces to socia-
p < .10). There is no difference between the unattractive and bility to source credibility to purchase intention and (2) from
attractive conditions (F < 1). Perceived sociability/competence unattractive faces to competence to source credibility to pur-
were positively correlated with source credibility (r ¼ .46/.58, chase intention. The path coefficients from serial multiple
p < .01) and purchase intention (r ¼ .45/.57, p < .01). mediated models are presented in Figure 4, Panel A. It is worth
Peng et al. 13
A: Digital Camera
purchase intention (Mpicture ¼ 3.84 vs. Mno picture ¼ 3.60; F(1,
477) ¼ 5.05, p < .05). The results from mediation analysis on
.002 the sunscreen setting (shown in Figure 4, Panel B) are largely
.29***
Sociability
.06
consistent with those on the digital camera setting. In particu-
.17***
lar, the indirect effect of attractive faces on purchase intention
Attractiveness
.18*
.20**
via sociability/competence and source credibility is significant
Source .74*** Purchase
−.19* .08 Credibility Intention and positive (b ¼ .051/.054, SE ¼ .022/.031, 95% BCI ¼ [.010,
Unattractiveness .49*** .093]/p ¼ .076). The indirect effect of unattractive faces on
.26*** .15*** purchase intention via competence and source credibility is
Competence
also significant (b ¼ .066, SE ¼ .035, p ¼ .056). Thus, we
.06
found consistent beauty and ugliness premiums and the med-
iating mechanisms via sociability and competence for both
B: Sunscreen digital camera and sunscreen.
.02
Other potential mediators. Following the recommendations of
Sociability
.37*** .48*** Zhao, Lynch, and Chen (2010), we examined potential media-
Attractiveness .08
.30*** tors simultaneously alongside sociability and competence. We
.20*
Source .47*** Purchase performed serial mediation analyses on visual attention and test
−.19 −.06 Credibility Intention
whether it is a potential mediator driving the results. Although
Unattractiveness .59***
attractive and unattractive faces attract greater attention
.24**
Competence
.11**
(battractive ¼ 7.54, SE ¼ 3.17, p < .05; bunattractive ¼ 6.28,
.10 SE ¼ 3.19, p < .05), visual attention does not significantly
affect source credibility (b ¼ .001, SE ¼ .001, p > .10), which
influences purchase intention. The 95% BCI [.015, .03] of its
Figure 4. Mediation path diagram (Study 2a). indirect effect also includes zero. These results confirm our
*p < .10. conjecture that attention is only the starting point for percep-
**p < .05. tions but not sufficient to induce a positive effect on the out-
***p < .01.
comes. The potential mediating effects of trustworthiness and
face familiarity are also found to be insignificant.6 While
noting that the direct effect of attractive faces on sociability is
source credibility is an inference of expertise and trust based
much greater than that on competence (b ¼ .29 vs. b ¼ .18). on the perception of all available cues (with attractiveness
The indirect effect of attractive faces on purchase intention via being just one of them), visual-based trustworthiness is the
sociability/competence and source credibility is significant and trustworthiness judgment based on an online profile photo
positive (b ¼ .037/.065, SE ¼ .017/.034, 95% bootstrap confi- (Ert, Fleischer, and Magen 2016). Thus, it is independent of
dence interval [BCI] ¼ [.003, .071]/p ¼ .06). The indirect a purchase context. In contrast, source credibility is more
effect of unattractive faces on purchase intention through com- context-specific, especially relevant for evaluating products for
petence and source credibility is also significant (b ¼ .092, SE purchase. That explains why visual-based trustworthiness does
¼ .036, 95% BCI ¼ [.022, .164]). These results support H2 and not play a significant mediating role between facial attractive-
H3. We conducted a test of the alternative causal chain by ness and source credibility.
reordering the mediators and testing the following pathways:
(1) from attractive faces to source credibility to sociability to
purchase intention and (2) from unattractive faces to source
credibility to competence to purchase intention. However, the Study 2b: Product Relevance and Cross-Gender Effects
confidence intervals for these alternative mediation model con- Participants and design. We recruited 556 participants (306 men;
tain zero (sociability: b ¼ .010, SE ¼ .007, 95% BCI ¼ [.004, M age ¼ 37.15, SD ¼ 10.57) from MTurk and randomly
.025]; competence: b ¼ .014, SE ¼ .010, 95% BCI ¼ [.006, assigned them to a 3 (unattractive, plain-looking, and attractive
.034]). Thus, we concluded that the causal chain occurs only in faces) 2 (product relevance: appearance vs. expertise) 2
the predicted directions. (seller gender: male vs. female) between-subjects conditions.
The experiment simulates online shopping for a cookbook. To
rule out potential confounds from the difference between prod-
A replication study. We recruited 479 participants from MTurk
ucts in terms of features, prices, and so on, we followed the
and randomly assigned them to one of four between-subject
practice of using one product positioned to be different in its
conditions (no picture, attractive, plain-looking, and unattrac-
relevance, as it is possible that a product may be relevant to
tive). Except for the shopping task for a sunscreen, everything
else is identical to the original study. The presence of a picture
is found to have a positive effect on source credibility (Mpicture 6
Web Appendix 3 reports the mediation analysis and results for other potential
¼ 3.84 vs. Mno picture ¼ 3.63; F(1, 477) ¼ 5.35, p < .05) and mediators such as trustworthiness and face familiarity.
appearance or expertise to varying degrees (Bower and Land- A: ER Condition

reth 2001; Trampe et al. 2010). Thus, unlike some studies that
only used product type as a measure of product relevance (e.g., .05
Trampe et al. 2010), we manipulated product relevance by Sociability

−.01
.13
inserting a positioning message: “This cookbook contains Attractiveness .07**
.09
many beauty secrets in its recipes that will give you a healthy .10
Source .72*** Purchase
and radiant appearance” in the appearance-relevant (AR) con- −.13 .13 Credibility Intention
dition and “this cookbook can help you spend less time prepar- Unattractiveness .68***
ing nutritious meals and provide better cooking through .30*** .06
Competence
science” in the expertise-relevant (ER) condition. Participants .08
went through the same procedure as described in Study 2a. We
also asked questions regarding the manipulation check of prod-
uct relevance: “This book would improve the appearance of an B: AR Condition
unsatisfactory physical feature” and “this product would
−.02
improve the efficiency of cooking through scientific methods.”
Sociability
Participants responded using a five-point scale (1 ¼ “does not .53***
.06
describe at all,” 5 ¼ “describes completely”). At the end of the Attractiveness .34***

.12**
.05
study, we collected the genders of the participants to examine Source .82*** Purchase
−.07 −.083 Credibility Intention
the cross-gender effect.
Unattractiveness .52***
.12
−.08
Manipulation check. Participants viewed attractive sellers as sig- Competence
nificantly more attractive than plain-looking and unattractive .08
sellers (Mattractive ¼ 3.36 vs. Mplain ¼ 2.71 vs. Munattractive ¼

2.24; F(2, 553) ¼ 58.2, p < .01). Those in the AR condition C: Moderated Mediation
believed that the cookbook could help improve appearance Appearance
more than those in the ER condition (MAR ¼ 2.65 vs. MER ¼ Relevant
−.05 .31***
2.15; F(1, 554) ¼ 24.34, p < .01). In addition, participants in −.20* −.08
the ER condition believed that the cookbook could improve the .03
Sociability
efficiency of cooking more than those in the AR condition .17*
.03
(MAR ¼ 3.09 vs. MER ¼ 3.75; F(1, 554) ¼ 49.10, p < .01). Attractiveness .21***
.13***
.11 .78***
Source Purchase
−.07 .04 Credibility Intention
Purchase intention and source credibility. Consistent with previous Unattractiveness .58***
.31***
studies, seller attractiveness has a U-shaped relationship with Competence
−.02
purchase intention (F(2, 553) ¼ 5.12, p < .01) and source .08
credibility (F(2, 553) ¼ 6.77, p < .01). Moving from unattrac-

tive to plain-looking sellers decreases purchase intention
(Munattractive ¼ 3.91 vs. Mplain ¼ 3.72; F(1, 366) ¼ 4.42, p < Figure 5. Moderated mediation path diagram (Study 2b).
.05) and source credibility (Munattractive ¼ 3.84 vs. Mplain ¼ *p < .10.
**p < .05.
3.69; F(1, 366) ¼ 3.02, p < .10). Beyond that, however, addi-
***p < .01.
tional attractiveness increases purchase intention (Mplain ¼
3.72 vs. Mattractive ¼ 3.98; F(1, 371) ¼ 9.81, p < .01) and source
credibility (Mplain ¼ 3.69 vs. Mattractive ¼ 3.99; F(1, 371) ¼ positive (b ¼ .147, SE ¼ .057, 95% BCI ¼ [.034, .261]). H4a
13.71, p < .01). Source credibility is highly correlated with and H4b are largely supported.
purchase intention (r ¼ .74). Second, a moderated mediation analysis yields similar
results (Figure 5, Panel C). In particular, the AR product mod-
Moderated mediation for product relevance. First, we performed erates the sensitivity to sociability (b ¼ .31, SE ¼ .12, p < .01),
separate mediation analyses for the AR and ER conditions and sociability is positively related to source credibility (b ¼
(Figures 5, Panels A and B), simultaneously testing perceived .13, SE ¼ .04, p < .01). The conditional indirect effects show
sociability and competence as mediators. For the AR condition, that perceived sociability matters more in the AR condition
the indirect effect of attractive sellers on purchase intention via (b ¼ .049, SE ¼ .021, 95% BCI ¼ [.008, .089]) than in the
sociability and source credibility is significant and positive ER condition (b ¼ .017, SE ¼ .012, 95% BCI ¼ [.007, .041]).
(b ¼ .053, SE ¼ .028, p ¼ .052) whereas the path via compe- We also found that the ER product moderates the sensitivity to
tence is not significant (b ¼ .02, SE ¼ .055, p ¼ .72). For the competence (b ¼ .20, SE ¼ .12, p < .10) and that competence
ER condition, the effect of unattractive faces on purchase inten- is positively related to source credibility (b ¼ .58, SE ¼ .04,
tion via competence and source credibility is significant and p < .01). The conditional indirect effects show that perceived
Peng et al. 15
4.0 4.5
Competence
3.44 3.91
Sociability
3.5 4.0
3.71
3.61
3.01 3.71
3.0 2.88 3.13 3.5 3.63
3.59
2.91
2.84
2.5 3.0
Unattractive Plain-looking Attractive Unattractive Plain-looking Attractive
AR condition ER condition
Figure 6. The interaction between attractiveness and product relevance (Study 2b).
competence matters more in the ER condition (b ¼ .140, SE ¼ indirect effect (95% BCI ¼ [.089, .307]). Thus, H5b regarding a
.045, 95% BCI ¼ [.052, .228]) than in the AR condition (b ¼ stronger ugliness premium in the FBMS setting is supported.
.054, SE ¼ .048, 95% BCI ¼ [.043, .228]).
Finally, by examining their relative values across AR versus
ER conditions, we further assessed how perceived sociability Discussion
and competence together influence source credibility, which in Conclusions
turn affects purchase intention (Figure 6). For both conditions,
attractiveness increases perceived sociability. When a seller is Unlike previous studies of attractiveness that focus on social
attractive, perceived sociability is significantly higher in the selections in experimental settings, our field studies examine
AR condition than in the ER condition (MAR ¼ 3.44 vs. the effect of facial attractiveness among large numbers of sell-
MER ¼ 3.13; F(1, 186) ¼ 6.42, p < .05). When a seller is ers and buyers in an e-commerce context, in which profile
unattractive, perceived competence is significantly lower in the pictures serve as a primary vehicle for impression formation
AR condition than in the ER condition (MAR ¼ 3.71 vs. MER ¼ and trait inference. Although the literature has documented a
3.91; F(1, 181) ¼ 3.11, p < .10). These results confirm that beauty premium in a variety of settings and occasionally found
product relevance affects the attractiveness–purchase relation- an ugliness premium, our analyses of tens of thousands of seller
ship by influencing perceived sociability and competence, profile pictures from two websites provide converging evi-
respectively. dence of a U-shaped relationship between facial attractiveness
and sales. As for the underlying mechanisms, our experimental
results support previous findings of a beauty premium and of an
Cross-gender effect. We created two dummy variables to test the ugliness penalty when evaluating sellers’ sociability. We also
moderating effect of cross-gender: MBFS takes a value of 1 if a find an ugliness premium in perceived competence for unattrac-
male buyer faces a female seller, FBMS takes a value of 1 if a tive sellers over plain-looking people. Thus, whereas attractive
female buyer faces a male seller, and both take 0 for pairs of the faces signal sociability and competence, unattractive faces elicit
same gender. To test H5a, we conducted a moderated mediation an enhanced perception of competence over sellers with plain
analysis (from seller attractiveness to sociability to source cred- looks, even slightly more so than the attractive people. Thus,
ibility to purchase intention, with MBFS as the moderator) with contrary to the notion of the curse of ugliness, our findings
5,000 bootstrapped samples. For attractive sellers, there is no indicate that plain-looking faces are caught in the middle without
evidence of moderated mediation for the MBFS group from any real advantage, as they are considered less sociable than
sociability to source credibility to purchase intention (b ¼ .16, attractive people and less competent than unattractive people.
SE ¼ .11, p ¼ .139). The conditional indirect effect also suggests As such, when consumers make online purchases, sellers’ faces
that perceived sociability does not matter more in the MBFS serve an important discriminating function to encode sellers’
condition than in the other conditions (p > .10). Thus, H5a characters, sometimes in unexpected ways.
regarding a stronger beauty premium in the MBFS setting is not In addition, the effects of attractiveness and inferred traits
supported. For unattractive faces, a similar moderated mediation are mediated by source credibility and are subject to the influ-
analysis (from seller attractiveness to competence to source ence of important contextual variables—that is, product rele-
credibility to purchase intention with FBMS as the moderator) vance (to appearance or expertise) and gender. Our results
suggests that unattractive men moderate the sensitivity of female reveal that the mediating role of sociability on the relationship
buyers to perceived competence (b ¼ .30, SE ¼ .11, p < .01), between attractive sellers and source credibility is significantly
and perceived competence is positively related to source cred- stronger for products relevant to appearance. In contrast, the
ibility (b ¼ .59, SE ¼ .04, p < .01), which in turn affects mediating effect of competence is more associated with prod-
purchase intention (b ¼ .78, SE ¼ .04, p < .01). ucts for which expertise is more important than appearance.
A bootstrapping test with 5,000 resamples indicates a significant Finally, we find a greater ugliness premium for unattractive
male sellers in perceived competence awarded by female con- from the 5miles study show that both attractive and unattractive
sumers. However, male respondents do not reciprocate a sellers are more likely to make a sale than their plain-looking
greater beauty premium on attractive female sellers, perhaps counterparts (predicted probability: 44% for attractive, 38% for
because online purchases do not involve social selection like plain-looking, and 41% for unattractive; Figure 2, Panel B).
dating or hiring. It is not uncommon for attractive women to be Our experimental results suggest that while the beauty pre-
viewed negatively for certain products or professions (Heilman mium of female sellers does not hold true for male buyers, the
et al. 2004; Ruffle and Shtudiner 2015; Samper, Yang, and ugliness premium only applies to unattractive men seen by
Daniels 2018) or to draw suspicion for their appearance in female buyers, revealing the inequality in the cross-gender
online forums (Lo, Hsieh, and Chiu 2013). effect of beauty and ugliness premiums.
While the marketing literature is not short of studies empha-
sizing the effect of attractiveness in sales and customer service
Implications encounters (Keh et al., 2013; McColl and Truong 2013), our
The role of attractiveness in human interactions is complex. nuanced findings of the curvilinear relationship between attrac-
Although most studies indicate a prevailing beauty premium, tiveness and performance and the underlying mechanisms are
there are many exceptions and counterexamples (e.g., Eagly particularly relevant for today’s social selling on e-commerce
et al. 1991). Our findings of a U-shaped relationship and the platforms. First, like candidates in political campaigns who
different mechanisms and contexts underlying the beauty and often enhance their images (Mattes et al. 2010), aspiring entre-
ugliness premiums highlight the complex relations between preneurs in social selling and C2C e-commerce should be
facial attractiveness and outcomes in C2C e-commerce and, mindful of their self-presentation; attractive appearances help
to some extent, reconcile the previous disparate findings. Pre- create a favorable impression and gain the trust of shoppers. A
vious studies of the beauty premium have mainly considered professional photographer can produce a quality portrait to
mass media or interpersonal and face-to-face situations. enhance attractiveness, and sellers can pretest the effect of a
Although social pressure is of lesser concern in C2C portrait on their perceived sociability and competence using
e-commerce, attractive individuals retain the beauty premium services such as photofeelfer.com. As consumers often choose
in sociability and competence, whereas their plain-looking between many sellers pitching similar products online, sellers
counterparts suffer a penalty. Meanwhile, we find consistent with different degrees of attractiveness must be cognizant of
evidence that even unattractive individuals have an edge in their source of credibility, that is, sociability and/or expertise,
perceived competence over plain-looking people. More impor- as well as the type of products they are selling. A small per-
tantly, we shed light on the different mechanisms and condi- ceptual difference based on appearance or credibility can have
tions for the beauty and ugliness premiums, that is, social trait a nonnegligible effect.
inferences, product relevance, and gender interactions. While Although e-commerce platform operators have no control
the marketing and advertising literatures have emphasized the over how people take pictures, they should provide guidance
halo effect of beauty, our findings suggest that the effect of and suggestions and encourage sellers to provide attractive
attractiveness is more complicated and subject to the influence portraits of themselves. In addition to a good-quality photo-
of these factors, which researchers and practitioners must con- graph (i.e., in brightness and pixels), taking a photo from a
sider when assessing the effect of seller attractiveness on con- particular angle may enhance attractiveness to avoid the plain-
sumer responses. ness penalty. While attractive sellers enjoy an advantage, espe-
Our findings provide meaningful implications for both cially for appearance-related products, people without perfect
online sellers and platform operators who want to leverage facial symmetry and proportions should not shy away from
seller profile pictures to enhance business performance. Post- displaying their true appearance. Emphasizing expertise in
ing a photo of oneself instead of an avatar or landscape makes a technical products can enhance their credibility and perfor-
difference. Having said that, loading a profile picture is not a mance. Thus, on e-commerce platforms, both attractive and
task to be taken lightly. Similar to the beauty and ugliness unattractive sellers can increase their performance by enhan-
premiums in earnings found by studies of labor market (e.g., cing their perceived sociability or competence, especially when
Biddle and Hamermesh 1994; Kanazawa and Still 2018), our they are matched with products associated with the particular
results indicate that one’s attractiveness level has a tremendous strengths derived from appearances. Because a product may be
effect on sales performance in C2C e-commerce platforms. relevant to both appearance and expertise in varying degrees,
Figure 2, Panel A, suggests that the beauty premium over our treatment of product relevance goes beyond mere product
plainness in the annual occupancy rate on Airbnb is, on aver- type and is based on product positioning with additional infor-
age, 6% (62% vs. 56%) and as high as 22% (i.e., 78% vs. 56%) mation. For online marketers, this means that given the posi-
for perfect faces. Thus, everything being equal, good looks sell tioning of a product (as relevant to appearance or expertise),
more. Meanwhile, the ugliness premium over plain-looking they may select attractive or unattractive sellers as promoters
hosts is approximately 4%, on average, (60% vs. 56%) and and achieve similar results. Conversely, sellers with attractive
up to 16% (72% vs. 56%) for the most unattractive hosts. Thus, or unattractive faces may find themselves better off presenting
such premiums are much higher for the extreme cases, whether a product depending on its relevance to appearance or
it is extremely attractive or unattractive. Likewise, findings expertise.
Peng et al. 17
With respect to the cross-gender interactions, existing stud- attractiveness and sales and the disadvantages for plain-looking
ies in marketing have pointed to the potential positive effect of people. This research focuses on facial geometrics to assess
mismatched gender in service counters (e.g., McColl and attractiveness. Characteristics of attractiveness other than faces
Truong 2013) as well as its precarious pitfalls in other cases could be examined, such as expressions and head tilt, which
(Wan and Wyer 2015). Our findings of the inequality in the can affect perceptions of attractiveness. Although they are
cross-gender effect of attractiveness and ugliness premiums beyond the scope of this research, extrafacial features such as
suggest that attractive female sellers do not have an advantage clothing and body posture, biometric data such as skin tone,
over their less attractive counterparts in appealing to male buy- color, race, gender, and enhancement by cosmetics and acces-
ers, who may not succumb to the female beauty in online sories may affect social attributions and provide rich data and
purchase given the reduced social pressure. However, female broad avenues for future studies. For instance, attractiveness
buyers tend to consider unattractive men as more competent enhanced by cosmetics or perceived expertise from eyeglasses
than the Average Joes, perhaps perpetuating the stereotype of can augment or alter social perceptions.
the tech-savvy nerd. Social sellers and e-marketing managers Greater insight is needed regarding how other dimensions,
may heed such complex cross-gender interactions when such as cultural or dispositional variables, may moderate the
attempting to leverage the effect of seller appearances in online relationship between seller attractiveness and consumer reac-
selling. Altogether, these implications regarding the relevance tions. Online social interactions, such as messaging and chat,
of product and cross-gender effect of beauty and ugliness pre- and offline face-to-face meetings between sellers and buyers
miums are not limited to profile pictures of online sellers and may influence the effect of seller attractiveness. Software is
may be pertinent to advertising and marketing aesthetics in now commonly used to enhance self-presentation, but exces-
general. Thus, researchers should consider a broader range of sive manipulation in portraits may be deceptive, raise suspi-
attractiveness, traits inferred from appearance, and its complex cion, and lead to consumer dissonance and discontent. Thus,
interactions with product relevance and gender. how consumers perceive and react to enhanced portraits and
Recent trends in collaborative consumption increase the facial features warrants investigation. Finally, more tests are
already large number of selection decisions facing consumers; necessary to validate the mechanisms through which sociabil-
this could further contribute to information overload and poten- ity and competence judgments are derived from facial cues and
tially increase reliance on the physical and facial appearances carried over to decision making. Innovative methods such as
of sellers. Although poor-quality pictures may dampen con- neuroscience and fMRI scans may help to reveal how these
sumer confidence, attempts by sellers to make themselves evaluation processes influence consumer perceptions and pur-
appear more attractive may backfire if they appear otherwise chase decisions.
incompetent or suspicious (Lo, Hsieh, and Chiu 2013; Samper,
Yang, and Daniels 2018). Although consumers may consider Author Contributions
the attractiveness of sellers in their decision making, they Ling Peng, Geng Cui, and Yuho Chung share equal authorship.
should not allow a seller’s appearance to cloud their judgment
of source credibility and product quality. Due diligence in con-
firming the veracity of sellers and product information is nec- Associate Editor
essary, as platforms often provide indicators of sellers’ Raj Venkatesan
reputations, track records, and social media connections.
Methodologically, this study is the first to explore the effect Declaration of Conflicting Interests
of facial attractiveness using large data sets of real profile The author(s) declared no potential conflicts of interest with respect to
pictures in online transaction platforms. This further validates the research, authorship, and/or publication of this article.
the generalizability of studies based on laboratory methods
using a limited number of facial stimuli. The use of online field Funding
data and actual sales overcomes the limitations of perceptual
The author(s) disclosed receipt of the following financial support for
measures and strengthens the validity of our findings. The the research, authorship, and/or publication of this article: The authors
machine learning approach to assessing facial attractiveness acknowledge the financial support of Lingnan University, Hong Kong
proves to be reliable and robust and provides a useful tool for for this research (FRG 102016/DB18B1).
future studies using large data sets for facial recognition and
deep learning in online settings. ORCID iD
Ling Peng https://orcid.org/0000-0002-6034-4580
Limitations and Directions for Future Research
People make individual choices when uploading a profile pic- References
ture and selecting the type of products they sell. Future Agthe, Maria, Matthias Spörrle, and Jon K. Maner (2011), “Does
researchers could collect more data from other e-commerce Being Attractive Always Help? Positive and Negative Effects of
sites to address potential self-selection bias and to validate our Attractiveness on Social Decision Making,” Personality and Social
findings, particularly the U-shaped relationship between facial Psychology Bulletin, 37 (8), 1042–54.
Ahearne, Michael, Thomas W. Gruen, and Cheryl Burke Jarvis Credibility on Consumer Reaction to Advertisements and Brands,”
(1999), “If Looks Could Sell: Moderation and Mediation of the Journal of Advertising, 29 (3), 43–54.
Attractiveness Effect on Salesperson Performance,” International Grabe, Shelly, Monique L. Ward, and Janel S. Hyde (2008), “The Role
Journal of Research in Marketing, 16 (1), 269–84. of the Media in Body Image Concerns Among Women: A Meta-
Argo, Jennifer J., Darren W. Dahl, and Andrea C. Morales (2008), Analysis of Experimental and Correlational Studies,” Psychologi-
“Positive Consumer Contagion: Responses to Attractive Others in cal Bulletin, 134 (3), 460–76.
a Retail Context,” Journal of Marketing Research, 45 (6), Guan, Shu-Sha A., Kaveri Subrahmanyam, Kevin Linares, and Roy
690–701. Cheng (2015), “Beauty in the Eye of the Beholder? Attractiveness
Biddle, Jeff and Daniel Hamermesh (1994), “Beauty and the Labor in a Virtual World,” Cyberpsychology: Journal of Psychosocial
Market,” American Economic Review, 84 (5), 1174–94. Research on Cyberspace, 9 (2), 1–19.
Bower, Amanda B. and Stacy Landreth (2001), “Is Beauty Best? Guihaire, Edouard (2018), “Ugly Models, an Agency of Extraordinary
Highly Versus Normally Attractive Models in Advertising,” Jour- Character,” The Jakarta Post (February 16), http://www.thejakarta
nal of Advertising, 30 (1), 1–12. post.com/life/2018/02/16/ugly-models-an-agency-of-extraordi
Carusi, Annamaria (2008), “Scientific Visualisations and Aesthetic nary-characters.html.
Grounds for Trust,” Ethics and Information Technology, 10 (4), Gunes, Hatice and Massimo Piccardi (2006), “Assessing Facial
243–54. Beauty Through Proportion Analysis by Image Processing and
Chaiken, Shelly and Durairaj Maheswaran (1994), “Heuristic Process- Supervised Learning,” International Journal of Human-
ing Can Bias Systematic Processing: Effects of Source Credibility, Computer Studies, 64 (12), 1184–99.
Argument Ambiguity, and Task Importance on Attitude Hamermesh, Daniel S. (2011), Beauty Pays: Why Attractive People
Judgment,” Journal of Personality and Social Psychology, 66 Are More Successful. Princeton: Princeton University Press.
(3), 460–73. Heilman, Madeline E., Aaron S. Wallen, Daniella Fuchs, and Melinda
Crane, Tim and Sarah Patterson (2012), History of the Mind-Body
M. Tamkins (2004), “Penalties for Success: Reactions to Women
Problem. New York: Routledge.
Who Succeed at Male Gender-Typed Tasks,” Journal of Applied
Debevec, Kathleen, Thomas J. Madden, and Jerome B. Kernan (1986),
Psychology, 89 (3), 416–27.
“Physical Attractiveness, Message Evaluation, and Compliance: A
Kamins, Michael A. (1990), “An Investigation into the ‘Match-Up’
Structural Examination,” Psychological Reports, 58 (2), 503–08.
Hypothesis in Celebrity Advertising: When Beauty May Be Only
Eagly, Alice H., Richard D. Ashmore, Mona G. Makhijani, and Laura
Skin Deep,” Journal of Advertising, 19 (1), 4–13.
C. Longo (1991), “What Is Beautiful Is Good, But . . . : A Meta-
Kanazawa, Satoshi and Mary C. Still (2018), “Is There Really a
Analytic Review of Research on the Physical Attractiveness
Beauty Premium or an Ugliness Penalty on Earnings?” Journal
Stereotype,” Psychological Bulletin, 110 (1), 109–28.
of Business and Psychology, 33 (2), 249–62.
Engell, Andrew D., James V. Haxby, and Alexander Todorov (2007),
Kang, Yong-Soon and Paul M. Herr (2006), “Beauty and the
“Implicit Trustworthiness Decisions: Automatic Coding of Face
Beholder: Toward an Integrative Model of Communication Source
Properties in Human Amygdala,” Journal of Cognitive Neu-
Effects,” Journal of Consumer Research, 33 (6), 123–30.
roscience, 19 (9), 1508–19.
Kaplan, Robert M. (1978), “Is Beauty Talent? Sex Interaction in the
Ert, Eyal, Aliza Fleischer, and Nathan Magen (2016), “Trust and
Attractiveness Halo Effect,” Sex Roles, 4 (2), 195–204.
Reputation in the Sharing Economy: The Role of Personal Photos
Keh, Hean Tat, Run Ren, Sally Rao Hill, and Xuan Li (2013), “The
in Airbnb”, Tourism Management, 55, 62–73.
Fidrmuc, Jan, Boontarika Paphawasit, and Çiğdem Börke Tunalı Beautiful, the Cheerful, and the Helpful: The Effects of Service
(2017), “Nobel Beauty,” Working Paper, Rimini Center for Eco- Employee Attributes on Customer Satisfaction,” Psychology and
nomic Analysis. Marketing, 30 (3), 211–26.
Fisher, Robert J. and Yu Ma (2014), “The Price of Being Beautiful: Kock, Ned (2003), Compensatory Adaptation: Understanding How
Negative Effects of Attractiveness on Empathy for Children in Obstacles Can Lead to Success. Conshohocken, PA: Infinity
Need,” Journal of Consumer Research, 41 (2), 436–50. Publishing.
Fiske, Susan T., Amy J.C. Cuddy, and Peter Glick (2007), “Universal Langlois, Judith H., Lisa Kalakanis, Adam J. Rubenstein, Andrea
Dimensions of Social Cognition: Warmth and Competence,” Larson, Monica Hallam, and Monica Smoot (2000), “Maxims or
Trends in Cognitive Science, 11 (2), 77–83. Myths of Beauty? A Meta-Analytic and Theoretical Review,” Psy-
Forman, Chris, Anindya Ghose, and Batia Wiesenfeld (2008), chological Bulletin, 126 (3), 390–423.
“Examining the Relationship Between Reviews and Sales: The Lee, Margaret, Marko Pitesa, Madan M. Pillutla, and Stefan Thau
Role of Reviewer Identity Disclosure in Electronic Markets,” (2018), “Perceived Entitlement Causes Discrimination Against
Information Systems Research, 19 (3), 291–313. Attractive Job Candidates in the Domain of Relatively Less Desir-
Gheorghiu, Ana I., Mitchell J. Callan, and William Skylark (2017), able Jobs,” Journal of Personality and Social Psychology, 114 (3),
“Facial Appearance Affects Science Communication,” Proceed- 422–42.
ings of the National Academy of Sciences of the United States of Li, Norman P. and Douglas T. Kenrick (2006), “Sex Similarities and
America (PNAS), 114 (23), 5970–75. Differences in Preferences for Short-Term Mates: What, Whether,
Goldsmith, Ronald E., Barbara A. Lafferty, and Stephen J. Newell and Why,” Journal of Personality and Social Psychology, 90 (3),
(2000), “The Impact of Corporate Credibility and Celebrity 468–89.
Peng et al. 19
Lind, Jo T. and Halvor Mehlum (2010), “With or Without U? The Scharlemann, Jörn P.W., Catherine C. Eckel, Alex Kacelnik, and
Appropriate Test for a U-Shaped Relationship,” Oxford Bulletin of Rick K. Wilson (2001), “The Value of a Smile: Game Theory
Economics and Statistics, 72 (1), 109–18. with a Human Face,” Journal of Economic Psychology, 22 (5),
Lo, Shao-Kang, Ai-Yun Hsieh, and Yu-Ping Chiu (2013), 617–40.
“Contradictory Deceptive Behavior in Online Dating,” Computers Sirdeshmukh, Deepak, Jagdip Singh, and Barry Sabol (2002),
in Human Behavior, 29 (4), 1755–62. “Consumer Trust, Value, and Loyalty in Relational Exchanges,”
Luca, Michael (2017), “Designing Online Marketplaces: Trust and Journal of Marketing, 66 (1), 15–37.
Reputation Mechanisms,” Innovation Policy and the Economy, Sprecher, Susan, Quintin Sullivan, and Elaine Hatfield (1994), “Mate
17, 77–93. Selection Preferences: Gender Differences Examined in a Natural
MacInnis, Deborah J. and C. Whan Park (1991), “The Differential Sample,” Journal of Personality and Social Psychology, 66 (6),
Role of Characteristics of Music on High- and Low- Involvement 1074–80.
Consumers’ Processing of Ads,” Journal of Consumer Research, Till, Brian D. and Michael Busler (2000), “The Match-Up Hypothesis:
18 (2), 161–73. Physical Attractiveness, Expertise, and the Role of Fit on Brand
Magro, Albert M. (1999), “Evolutionary-Derived Anatomical Char- Attitude, Purchase Intent and Brand Beliefs,” Journal of Advertis-
acteristics and Universal Attractiveness,” Perceptual and Motor ing, 29 (3), 1–13.
Skills, 88 (1), 147–66. Todorov, Alexander, Anesu N. Mandisodza, Amir Goren, and Crystal
Maner, Jon K., Saul L. Miller, D. Aaron Rouby, and Matthew T. C. Hall (2005), “Inferences of Competence from Faces Predict
Gailliot (2009), “Intrasexual Vigilance: The Implicit Cognition Election Outcomes,” Science, 308 (5728), 1623–26.
of Romantic Rivalry,” Journal of Personality and Social Psychol- Trampe, Debra, Diederik A. Staple, Frans W. Siero, and Henriëtte
Mulder (2010), “Beauty as a Tool: The Effect of Model Attrac-
ogy, 97 (1), 74–87.
tiveness, Product Relevance, and Elaboration Likelihood on
Martı́n-Loeches, Manuel, Juan Antonio Hernández-Tamames, Adrian
Advertising Effectiveness,” Psychology & Marketing, 27 (2),
Martı́n, and Mabel Urrutia (2014), “Beauty and Ugliness in the
1101–21.
Bodies and Faces of Others: An fMRI Study of Person Esthetic
Vuilleumier, Patrik (2000), “Faces Call for Attention: Evidence from
Judgement,” Neuroscience, 277, 486–97.
Patients with Visual Extinction,” Neuropsychologia, 38 (5),
Mattes, Kyle, Michael Spezio, Hackjin Kim, Alexander Todorov,
693–700.
Ralph Adolphs, and R. Michael Alvarez (2010), “Predicting Elec-
Wan, Lisa C. and Robert S. Wyer (2015), “Consumer Reactions to
tion Outcomes from Positive and Negative Trait Assessments of
Attractive Service Providers: Approach or Avoid?” Journal of
Candidate Images,” Political Psychology, 31 (1), 41–58.
Consumer Research, 42 (4), 578–95.
McColl, Rod and Yann Truong (2013), “The Effects of Facial Attrac-
Wang, Ze, Huifang Mao, Yexin Jessica Li, and Fan Liu (2017), “Smile
tiveness and Gender on Customer Evaluations During a Web-Video
Big or Not? Effects of Smile Intensity on Perceptions of Warmth
Sales Encounter,” Journal of Personal Selling & Sales Management,
and Competence,” Journal of Consumer Research, 43 (5),
33 (1), 117–28. 787–805.
Ohanian, Roobina (1990), “Construction and Validation of a Scale to Wheeler, S. Christian and Richard E. Petty (2001), “The Effects of
Measure Celebrity Endorsers’ Perceived Expertise, Trustworthi- Stereotype Activation on Behavior: A Review of Possible Mechan-
ness, and Attractiveness,” Journal of Advertising, 19 (3), 39–52. isms,” Psychological Bulletin, 127 (6), 797–826.
Reingen, Peter H. and Jerome B. Kernan (1993), “Social Perception Willis, Janine and Alexander Todorov (2006), “First Impressions:
and Interpersonal Influence: Some Consequences of the Physical Making Up Your Mind After a 100-Ms Exposure to a Face,” Psy-
Attractiveness Stereotype in a Personal Selling Setting,” Journal of chological Science, 17 (7), 592–98.
Consumer Psychology, 2 (1), 25–38. Winston, Joel S., John O’Doherty, James M. Kilner, David I. Perrett,
Ruffle, Bradley J. and Ze’ev Shtudiner (2015), “Are Good-Looking and Raymond J. Dolan (2007), “Brain Systems for Assessing
People More Employable?” Management Science, 61 (8), 1760–76. Facial Attractiveness,” Neuropsychologia, 45 (1), 195–206.
Said, Christopher P., Sean G. Baron, and Alexander Todorov (2008), Yang, Mochen, Gediminas Adomavicius, Gordon Burtch, and Yuqing
“Nonlinear Amygdala Response to Face Trustworthiness: Contri- Ren (2018), “Mind the Gap: Accounting for Measurement Error
butions of High and Low Spatial Frequency Information,” Journal and Misclassification in Variables Generated Via Data Mining,”
of Cognitive Neuroscience, 21 (3), 519–28. Information Systems Research, 29 (1), 4–24.
Samper, Adriana, Linyun W. Yang, and Michelle E. Daniels (2018), Zhao, Xinshu, John G. Lynch Jr., and Qimei Chen (2010),
“Beauty, Effort, and Misrepresentation: How Beauty Work Affects “Reconsidering Baron and Kenny: Myths and Truths About
Judgments of Moral Character and Consumer Preference,” Journal Mediation Analysis,” Journal of Consumer Research, 37 (2),
of Consumer Research, 45 (1), 126–47. 197–206.

The Faces of Success

Uploaded by

Copyright:

Available Formats

The Faces of Success

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Faces of Success

Uploaded by

Copyright:

Available Formats

Article

Premiums in e-Commerce Platforms sagepub.com/journals-permissions

Ling Peng , Geng Cui, Yuho Chung, and Wanyi Zheng

Figure 1. Conceptual framework.

Table 1. Summary Statistics of Airbnb Data (Study 1a).

Variable Definition N M SD Min Max

Table 2. Estimation Results for Occupancy Rate (Study 1a).

Table 3. Summary Statistics of 5miles Data (Study 1b).

Variable Definition N M SD Min Max

familiar”) and perceived trustworthiness (“The seller is some-

0.6 Manipulation check. The participants rated the attractive

0.6 is completely loaded and when the “next” button is clicked.

Purchase intention and source credibility. Consistent with the find-

Table 4. Estimation Results from Duration Model (Study 1b).

Spec. 1 Spec. 2 Spec. 3 Spec. 4 Spec. 5

Hazard Hazard Hazard Hazard Hazard

appearance or expertise to varying degrees (Bower and Land- A: ER Condition

Trampe et al. 2010), we manipulated product relevance by Sociability

describe at all,” 5 ¼ “describes completely”). At the end of the Attractiveness .34***

nificantly more attractive than plain-looking and unattractive .08

sellers (Mattractive ¼ 3.36 vs. Mplain ¼ 2.71 vs. Munattractive ¼

credibility (F(2, 553) ¼ 6.77, p < .01). Moving from unattrac-

You might also like