Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Running Head: INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING
BEHAVIOR
The Influence of Objective Facial Features on Trusting Behavior
Jack Colvin
Department of Psychology and Neuroscience
University of Colorado at Boulder
Date for Defense: March 31st, 2021
Defense Committee:
Thesis Advisor: Leaf Van Boven, Psychology and Neuroscience
Honors Council Representative: Lewis Harvey, Psychology and Neuroscience
Outside Reader: Lawrence Williams, Marketing
Additional Member: Eric Pedersen, Psychology and Neuroscience
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 2
Abstract
First impressions have been repeatedly studied in psychology and yet while most people pay
attention to how one acts during the initial interaction, many judgements are made from the
visual appearance first. This study investigates how 11 unchangeable facial characteristics
influence trusting behavior without any prior interaction between the stimulus faces (N = 132)
and the trust subjects. It was predicted that larger eyes, as well as wider faces would be more
often trusted by unacquainted participants. Related facial features such as higher cheekbones and
longer foreheads, both of which would contribute to the length of a face, were hypothesized to
reduce the trusting behavior by the same participants. Further exploration into additional facial
characteristics such as the shape of the nose, shape of the chin, and fullness of the lips were
included in the study. Once the measurements were completed, the stimuli faces were analyzed
against pre-existing data. The results showed no significant relationships between the objective
facial characteristics and the trusting behavior, in contrast to some of the findings of previous
literature. This could be reflective of the difference between objective measurements and
subjective ratings and how each of those are able to be applied in social decisions.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 3
Introduction
A quote by the philosopher Aristotle claims “Man is by nature a social animal; an
individual who is unsocial naturally and not accidentally is either beneath our notice or more
than human. Society is something that precedes the individual. Anyone who either cannot lead
the common life or is so self-sufficient as not to need to, and therefore does not partake of
society, is either a beast or a god,” (Aristotle, 1905). While Aristotle lived well over 2
millennium ago, human nature has not separated from the social world. In fact, countless social
interactions occur every day, all around the globe and these interactions can range from
situations as simple as ordering food off a menu or as complex as selecting a single applicant for
a job from thousands of interviews. When this interaction between two or more people is
positive, it promotes gain and advancement for one or both parties. This human phenomenon is
defined in psychology as prosocial behavior, the voluntary behavior of one individual intended to
benefit another (Davidov, Vaish, Knafo‐Noam, & Hastings, 2016; Eisenberg, Fabes, & Spinrad,
2006). However, in both of the scenarios provided, the prosocial behavior is only completed
because individuals place trust in another subject to fulfill the obligations that are expected,
whether that be delivering an insignificant burger or agreeing to a social contract such as a job.
Because of the incredible extent that our daily lives require other people, it is important to
understand why we choose to place trust in one individual over another.
A study conducted in 2019 has shown that children as young as 3 years old begin to
recognize characteristics of personality traits from a person’s facial appearance, though these
trait inferences are not yet implemented in social interaction. However, by the age of 5, children
are shown to alter their behavior based on how trustworthy they deem that individual
(Charlesworth, Hudson, Cogsdill, Spelke, & Banaji, 2019). For example, after the researchers
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 4
concluded that children could consistently identify which face they deemed more trustworthy,
they conducted another experiment in which children 5 years old specifically gave gifts to faces
that they considered more trustworthy. Coincidentally, characters in films, especially animated
movies where the body shapes and proportions can be manipulated, are typically portrayed
differently based on the role of the character. Disney movies capitalize on this ability and
routinely display the villain as a slender character with a longer face and small eyes, for example,
Maleficent from Sleeping Beauty, Hades from Hercules, Jafar from Aladdin, or even Scar from
the Lion King. This technique is able to help children immediately identify who the villain is in
the show without having as much ability to comprehend the actions and language used. The final
example even shows that the facial characterizations that children are able to identify as negative
can extend to anthropomorphic faces in animals as well. With how young of an audience that
these movies target, it goes to show that humans begin making judgements on faces very early in
life and use these judgements to characterize people. However, it is difficult to be able to
determine whether these characterizations are due to a natural tendency, or if the technique that
these animators are causing the common perception of the audience.
Interestingly, physiognomy, which is the study of faces, has developed multiple
conflicting positions on the reliability of personality inferences from facial characteristics
(Hassin & Trope, 2000; Todorov, Said, Engell, & Oosterhof, 2008). Tests of inferring
personality traits from looking only at facial constructions have been consistently studied for the
past century (Anderson, 1921) and results have also consistently maintained the fact that people
are able to perceive specific personality traits from faces across settings (Secord, & Bevan, 1956;
Berry & Brownlow, 1987). However, despite the fact that separate people are able to find a
consensus on which faces are more trustworthy, attractive, deceitful, kind, etc., these personality
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 5
traits are not reflective of the actual personality of the person as observed from their behavior
(Cohen, 2016; Alley, & Hildebrandt, 1988). Beginning in the 1990s, studies began to make the
claim that some traits identified through facial judgements were reflective of that individual’s
personality (Berry & Brownlow, 1989). The kernel of truth theory originally stated that
overgeneralized stereotypes may not be fully accurate representations for each individual in the
group, but there may be a small part of the stereotype that still applies to each varying person.
After beginning to see small effects in the study of facial characteristics, researchers began to use
the same theory in physiognomy (Penton-Voak, Pound, Little, & Perrett, 2006). It is with this
theory in mind that research on judgements of facial structures and characteristics has continued.
There are a few aspects of face judgements that are of interest for this study. First, it is
important to acknowledge how judgements are made and how quickly they are able to influence
perceptions of an individual. Previous studies by Alexander Todorov have confirmed that
individuals are capable of making personality trait judgements on faces that are only seen for as
little as 33 ms and these judgements do not increase in accuracy past 167 ms of exposure time
(Todorov, Pakrashi, & Oosterhof, 2009). Furthermore, the same study concluded that subjects
that are primed for making judgements on trustworthiness were capable of making those
judgements in less than the 33 ms threshold. In addition, Jones, Schild, & Jones (2020)
conducted a study testing the validity between manual facial evaluation and computer automated
facial evaluation. This study explored how manually placing markers on a face compared to
computer-automated marker placement and if one or both would be able to accurately measure
the face characteristics that were commonly studied. The results found that both computer-
generated points and manually placed points are correlated and both methods are reliable to use
in studies. Finally, another study analyzed whether personality trait judgements were determined
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 6
from aspects of appearance by an external viewer or if salient structures of the face alone such as
the geometry of the face were able to determine the trait judgements (Rojas, Masip, Todorov, &
Vitria, 2011). The findings stated that a holistic approach that analyzed all appearance
information was better equipped to accurately make personality trait judgements, but the
structural information of the face such as objective measures also produced enough information
to make trait judgements as well.
Another aspect of facial judgements to consider is which traits are able to be recognized
in facial characteristics and how those traits may influence selection decisions. Since some
prosocial behavior requires a sacrifice by the acting individual, in the case of the trust experiment
used in this study the sacrifice would be money, people want to avoid having wasted that
sacrifice without any result. To do this, individuals require trust that the money will not be lost.
Therefore, trustworthiness is an important factor in selection criteria. The physiognomic
attributes of trustworthiness as determined by external judgement are often correlated to sexual
dimorphism, facial width to height ratio, and baby-faced aspects such as large heads, large eyes,
and round chins (Berry, & Brownlow, 1989; Hassin, & Trope, 2000; Jones, Schild, & Jones,
2020; Stirrat, & Perrett, 2012). Stirrat and Perrett (2012) concluded that wider male faces are
more trustworthy for in-group subjects, but curiously, when faced with competition from an out-
group, the same authors in a study 2 years earlier found that males with wider faces were rated as
more deceitful and were judged to be especially untrustworthy when rated by the opposite sex
(Stirrat, & Perrett, 2010). Facial attractiveness has also been extensively studied in physiognomy
and has been shown to be highly correlated with trustworthiness (Xiao, Zheng, Zhang, Xin,
Chen, & Li, 2016). Thus, the ratings of trustworthiness in attractive individuals were increased
when judged by members of the opposite sex. The physiognomic attributes for attractive faces as
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 7
determined by Milutinovic, Zelic, & Nedeljkovic (2014) are generally smaller facial features
such as in the nose, face size, and lips. In addition, sexually dimorphic faces and symmetry
between both sides of the face are often stated to be perceived as objectively attractive.
Finally, the effect of attention has found significant results in previous studies but
remains an important area to study further. Previous research has concluded that attention,
whether internally motivated or caused by external forces such as an experiment cueing
mechanism, increases an individual’s perception of importance for the cued target (Mrkva & Van
Boven, 2017). This effect remains even when controlled for possible primacy and recency
effects. The research on the effect of attention was also established in a prosocial trust game
scenario (Grant, Pedersen, Ramos, & Van Boven, 2019) which contributed the attention and trust
data for the current study’s analysis. In an attempt to answer some of the questions posedby the
previous literature, this study analyzes the effect that particular features of stimuli faces have on
the decision to trust one picture over another along with the interaction that facial characteristics
might have with attention.
For the current study, there are 11 facial metrics that were hypothesized on. These
measures are Facial Width to Height Ratio, Upper Head Length, Face Roundedness, Face
Heartshapeness, Cheekbone Prominence, Cheekbone Height, Eye Shape, Eye Size, Distance
Between Pupils, Lip Fullness, and Nose Shape. According to several studies (Berry, &
Brownlow, 1989; Bull, & Gibson-Robinson, 1981; Manesi, Van Lange, & Pollet, 2016;
Rezlescu, Duchaine, Olivola, & Chater, 2012; Todorov, Said, Engell, & Oosterhof, 2008) that
make the claim that baby-faced aspects and individuals with larger heads and wider faces are
more trustworthy, the first hypothesis states that there will be a positive correlation between
trusting behavior and Facial Width to Height Ratio, Face Roundedness, Eye Size, Eye Shape,
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 8
and the Distance Between Center of Pupils, all of which are associated with the previously
identified trustworthy characteristics. Second, this study is also expanding the measures by
expecting additional significant relationships between trusting behaviors and Upper Head
Length, Cheekbone Height, and Face Heartshapeness. Because of the studies identifying facial
width to be associated with trustworthy faces, Upper Head Length and Cheekbone Height, which
are more related to face height, are predicted to result in negative correlations with trusting
decisions. Plus, Face Heartshapeness analyzes how sloped the bottom portion of the face is and
how quickly the face narrows, thus, since this variable contrasts with facial width, it is also
predicted that this variable will have a negative relationship as well. The third and final set of
metrics analyzed in this study are Cheekbone Prominence, Lip fullness, and Nose shape. These
characteristics are used as exploratory factors to further the investigation of face structure on
trusting behaviors.
To summarize, it is hypothesized the (1) Facial width to height ratio, Face roundedness,
Eye shape, Eye size, and Distance between pupils will each be positively correlated with trusting
behavior. (2) Upper head length, Cheekbone height, and Face heartshapeness will each be
negatively correlated with trusting behavior. (3) Lip fullness, Nose shape, and Cheekbone
prominence, which are used increase the base of knowledge for physiognomy, will not be
significantly correlated with trusting behavior. In order to test this, each of the characteristics
were measured for every face available. This study then analyzed the trusting behavior data that
was previously obtained (Grant et al., 2019), as well as the measurements that were taken, in
order to look for any correlations between the variables. The design of the study as well as all of
the hypotheses, measurement data, and analyses plan were pre-registered online (osf.io/qhkby).
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 9
This pre-registration was completed before any analyses were conducted and all the information
is open for viewing.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 10
Methods
Faces
This study utilized a total of 144 stimuli faces that were taken of previous students at the
University of Colorado Boulder. The pictures were voluntarily taken in a research lab in
exchange for general psychology course credit. The ethnic distribution of the faces used were a
majority Caucasian but with multiple faces each representing Asian, Black, Latino, and Middle
Eastern ethnicities. Each of these stimuli pictures were taken before the start of the trust
experiment. All the volunteers for the photos were asked to not make any expressions, however,
a closed lip smile was allowed. All the pictures were taken in the same indoor location, so the
white wall background and standard overhead lighting all remained consistent. All pictures were
taken facing directly forward, aligned with the face and with no head tilt of the participants. The
pictures displayed the head, hair (no restrictions were placed on the hairstyle), and tops of the
participants shoulders. Due to the photos being taken with varying cameras and varying qualities,
all the pictures were digitally resized to be 500 pixels in width by 375 pixels in height for this
study. During the trusting behavior experiment, all the pictures were displayed on the same lab-
provided, desktop computers, so each participant also viewed the pictures with the same
conditions. An example of the participant’s view is display in Figure 3.
Previous Trust Experiment
The previously conducted experiment that these stimuli faces were used for was designed
to test if attention influences the perception of a target face and how this interaction can be
measured in an economic game (Grant et al., 2019). The experiment was conducted on a sample
of student participants at the University of Colorado Boulder. These participants were invited
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 11
into the lab with credit compensation to complete the roughly 30-minute computer experiment.
The students were given two credits in order to give a maximum time length of one full hour,
thus ensuring the participants finish the experiment. Since all participants were from the
introductory psychology course, the trust experiment participants primarily consisted of
Caucasian, first-year, undergraduate students about ages 18-20.
Trust Behavior Procedure
Before being allowed to enter the lab, subjects were deceived into believing they were
given money to spend during the experiment. Participants were then guided to begin the
experiment and given instructions by the lab-provided computers. After receiving the
instructions and verifying comprehension, the participants engaged in a three practice trials to
orient them and ensure understanding. After this, 25 recorded trials were completed as the
experiment. Each trial began with a stream of letters flashing on either the right or left side of the
computer screen (see fig. 2) The participants were directed to which side of the screen the letters
would be randomly assigned. In each stream of letters, the participant attempted to count the
number of X’s that appeared, in order to assure their attention is directed to that particular side.
Midway through the stream of letters, two randomized facial pictures appeared on either side of
the screen (see fig. 3) The pictures were used to have the participant decide who they are willing
to give money to. During each trial, the stimuli pictures were matched in gender so as to avoid
sex biases as a reason for giving money to one face over another. After the stream of letters had
finished, the participants were asked to choose one of the two people on the screen with the
theory that the participant will choose the person on the same side that their attention has been
drawn to.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 12
This previous data from the Grant et al. experiment had to first be formatted to
accommodate the new measurements in the current study. All extraneous variables (participant
and trial, demographic information of the participants, subjective ratings, etc.) were nullified and
filtered out of the dataset. The only key points of data that were retained were the trusting
decision and the cued or not cued condition for each time the face was chosen. From this point,
the proportion that each face was chosen to be trusted out of the total times that the face was
shown was calculated and separated into 4 separate conditions: every time the face was shown in
a trial, only the times the face was cued in each trial that it was shown, only the times the face
was not cued when it was shown, and the difference between cued and not cued conditions. The
difference category was included to incorporate the effect of attention when running the analyses
against the facial characteristics.
Current Study Measures
A total of 11 different facial metrics were used for analysis and computed in the following ways:
Eye Shape was measured by eye height divided by eye width. Eye Size was measured by eye
height divided by face length. Distance between center of eye pupils was taken directly from the
measurement. Facial width to height ratio was measured by the face at the widest point divided
by face height. Upper head length was measured by forehead length divided by face length. Face
Roundedness was measured by the face width at mouth divided by face length. Face
Heartshapeness was measured by face width at cheeks divided by face width at mouth.
Cheekbone Prominence was measured by [face width at cheek minus face width at mouth]
divided by face length. Cheekbone Height was measured by [average mid-cheek to chin for right
and left sides of the face] divided by face length. Lip Fullness was measured by lip thickness
divided by face length. Nose shape was measured by nose width divided by nose length. These
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 13
facial metrics were calculated from a total of 15 different measurements (see figure 1) that were
found with the GNU Image Manipulation Program (GIMP; The GIMP Development Team,
2020). Each measurement was recorded by the number of pixels between the two points.
Because each face is different and no single point can be referenced across all faces, specific
instructions for each measurement were also created and continually referenced to thoroughly
eliminate the possibility of poor measures. The measurements and instructions are explained
below:
Face Length was measured from the center base of chin to center of hairline. Forehead
Height was measured from the same center of hairline point to the top Y-coordinate of the
observed eyebrows. Facial Width was measured in three locations, across the upper cheekbones
and below the eyes, across the middle cheekbones and across the width of the nose as, and across
the mouth. The first two width measurements can be observed on each face and measured
accordingly, however for the measurement across the mouth of the face, additional instructions
were created. The faces used for this experiment were taken without consistent facial
expressions, therefore while some faces have a blank expression, others are slightly smiling, and
their lips do not follow a straight line. Thus, the Face Width at Mouth was measured across the
Y-coordinate of the center of their lips, regardless of the shape of their mouth. Eye Height was
measured from the base of the bottom eyelid to the top eyelid along the outside of their pupil
which was generally the largest point of the eye. Eye Width was measured from the corner of the
outside of the eyeball to the observable inside point of the eyeball, the tear duct was not included
for the measurements. Both the Eye Height measurement and the Eye Width measurement were
measured on both eyes and averaged to find a single number. Distance between Pupils was
measured by the observable center of each pupil. Lip Thickness was measured from the center of
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 14
the top lip to the center of the bottom lip. Nose Width was measured across the widest point of
the nose, along the outside edge of each nostril. Nose Length was measured from the center of
the observable bottom of the nose to the observable top point in the nose, in the depression
between the middle points in Eye Height. Lastly, Left and Right Chin to Middle Cheek was
measured from the same center base point used in Face Length to the Y-coordinate of the bottom
of the nose along each side of the face, generally found below the ear lobes. Each of these
measurements were then used to compute the face metrics used in the analysis.
During the measurements, a few data issues were identified and resolved. These issues
consisted of 3 faces being labeled as the wrong ethnicity and since this issue was discovered after
the previous experiment had been conducted, the faces were eliminated from the current study’s
analysis. One male face was found to be included twice but since the faces were randomly
included in the experiment trials, this did not impact the results of the analysis. In addition, one
white female face was not included in the previous study and to remain consistent in the dataset,
this face was also removed from this study’s analyses. Furthermore, there were 10 faces that
could have caused potential problems in the data. These problems consisted of hair that covered
certain parts of their face or the stimuli faces were wearing glasses which may have changed the
eye shape and size. One face featured parted lips and showed teeth in the smile, thus interfering
with the lip thickness measurement. Each of these faces were excluded from the dataset.
The total amount of stimuli faces that were used for analysis is 132 and all data
manipulation and analyses were completed in RStudio (http://www.rstudio.com/). The data from
the attention study that was recorded is the trusting decision and the cued or not cued condition
when the face was chosen to be trusted. Linear regression analyses were conducted on each of
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 15
the 11 facial metrics individually and all together against the proportion of trusting behavior for
all faces, as well as the proportion of trusting behavior after the attention manipulation.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 16
Results
Upon inspection of the facial characteristics, each of the measures appeared to be
normally distributed, thus a linear model seemed most appropriate for the data. Using the
Shapiro-Wilk’s test, it was determined the 2 of the 11 facial characteristics were significantly
non-normal, these measures were the distance between pupils, which was hypothesized to be
positively correlated with trusting behavior, and nose shape, which was hypothesized to not be
correlated with trusting behavior. A separate linear regression was run for each of the 11 facial
characteristics individually, against the proportion that the face was trusted out of all the times it
was shown, regardless of if the face was cued. These data show the basic relationship between
facial characteristics as measured with objective measures and trusting behavior which was taken
from how often participants of the previous study would choose to explicitly trust one face over
another face in the experiment. However, each of the characteristics resulted with non-significant
relationships: Facial Width to Height Ratio (b= –0.25, t(130) = -0.72, p = 0.47), Upper head
Length (b= –0.28, t(130) = -0.58, p = 0.56), Face Roundedness (b= –0.12, t(130) = -0.38, p =
0.71), Face Heartshapeness (b= –0.26, t(130) = -0.81, p = 0.42), Cheekbone Prominence (b= –
0.54, t(130) = -0.93, p = 0.35), Cheekbone Height (b= –0.21, t(130) = -0.37, p = 0.71.), Eye
Shape (b= 0.29, t(130) = 1.23, p = 0.22), Eye Size (b= 1.79, t(130) = 1.09, p = 0.28), Distance
Between Pupils (b= 0.0005, t(130) = 0.21, p = 0.83), Lip Fullness (b= 0.39, t(130) = 0.47, p =
0.64), Nose Shape (b= –0.11, t(130) = -0.79, p = 0.43). These results show that facial metrics are
not indicative of one face being trusted over another in this particular study (see figures 4-14).
Next, the same linear model analyses were run using the difference between the
proportion of trusting in the not cued face condition and proportion of trusting in the cued face
condition, as a method to include the effect of attention. For each facial characteristic, we again
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 17
found no significant relationships between any variables: Facial Width to Height Ratio (b= –
0.18, t(130) = -0.55, p = 0.58), Upper head Length (b= –0.42, t(130) = -0.94, p = 0.34), Face
Roundedness (b= –0.13, t(130) = -0.45, p = 0.66), Face Heartshapeness (b= 0.11, t(130) = 0.39, p
= 0.70), Cheekbone Prominence (b= 0.18, t(130) = 0.33, p = 0.74), Cheekbone Height (b= –
0.30, t(130) = -0.59, p = 0.55), Eye Shape (b= -0.13, t(130) = -0.61, p = 0.54), Eye Size (b= -
0.34, t(130) = -0.23, p = 0.82), Distance Between Pupils (b= 0.001, t(130) = 0.53, p = 0.60), Lip
Fullness (b= 0.68, t(130) = 0.90, p = 0.37), Nose Shape (b= –0.05, t(130) = -0.365, p = 0.72).
These results further indicate that facial characteristics do not share a relationship with trusting
behavior even after accounting for the effect of attention.
After acknowledging the initial results, additional analyses were run on multiple
combinations of related characteristics to determine if trusting behaviors were not identifiable
from a single objective feature, but instead could be found in the corresponding features that all
promoted a trustworthy face. First, a multiple regression analysis was run on all the
characteristics simultaneously against the proportion of trusted decisions in all faces (see table 1)
as well as the difference of not cued from cued faces (see table 2). Neither function produced any
significant relationships. A simple correlation test was run on each of the facial characteristics in
order to view how related the various facial measures were. The face characteristics were then
converted into z scores and using data from the previous correlation, the most related scores were
then averaged together. Two groups of highly correlated features were identified: Cheekbone
Height, Facial Width to Height Ratio, and Face Roundedness were averaged to create a face
shape and Eye Shape and Eye size were averaged to create a single eye measure. Finally, 6
additional linear regression analyses were run on the averaged features. The proportion of trusted
decisions on all faces was run against average face shape (b= -0.01, t(130) = -0.55, p = 0.58), and
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 18
average eye shape (b= 0.02, t(130) = 1.18, p = 0.24), as well as the proportion of trusted
decisions for the difference between cued faces and not cued faces was run on average face
shape (b= -0.01, t(130) = -0.60, p = 0.55) and average eye shape (b= -0.01, t(130) = -0.43, p =
0.67). A multivariate linear regression including independent variable interaction was ran on the
proportion of all faces (see table 3) against the averaged face shape (b= <-0.01, t(130) = -0.24, p
= 0.81), the averaged eye shape (b= 0.01, t(130) = 1.63, p = 0.29), as well as the interaction
between independent variables (b= <-0.01, t(130) = -0.15, p = 0.88). Another was also ran on the
cued difference (see table 4) against the averaged face shape (b= -0.01, t(130) = -0.81, p = 0.42),
averaged eye shape (b= -0.01, t(130) = -0.58, p = 0.56), and the interaction (b= 0.01, t(130) =
0.71, p = 0.48). There were still no significant interactions in any of the analyses conducted.
Discussion
Although this study did not find any significant relationships or interactions between the
face characteristics and trusting behaviors regardless of cueing, it nonetheless identified very
important results. The study of physiognomy has developed controversial findings, especially in
the past 3 decades, over how accurate initial judgements of personality traits are based solely off
of facial appearance. The results posited here are not significant but because of the inconsistent
findings in the past, these results are still able to support some of previous research that also
found no correlations between facial characteristics and personality traits (Cohen, 1973; Alley
1978). Furthermore, there are specific exploratory variables that now have a base of research to
extend from. There is also the possibility that this study committed a type 2 error and the null
hypotheses were actually false though the results showed for the null hypothesis to be accepted.
Therefore, this study would first require replication to ensure there was not a false negative
reported.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 19
Implications
This study analyzed the trusting behaviors of university students when presented with a
prosocial situation. The analyses were conducted on 4 separate trust conditions: the proportion of
trusted decisions regardless of an external cueing mechanism, the proportion of trusted decisions
when the stimulus face is cued, the proportion of trusted decisions when the stimulus face is not
cued, and the proportion of trusted decisions when the effect of cueing is taken into account.
None of the relationships for the any of the facial characteristics were significant. Therefore, the
main hypotheses that Facial width to height ratio, Face roundedness, Eye size, Eye shape, and
the Distance between pupils would all be positively correlated with trusting behavior was not
supported in any of the trusted proportion conditions. This conflicts with multiple of the previous
studies because the characteristics identified to have positive effects with trustworthiness were
all specifically mentioned and accounted for with the exception of facial width which had
contrasting effects based on whether the interaction involved competition with an outgroup.
Nonetheless, the results here are consistent with the theory that facial characteristics do not have
correlational relationships with personality traits of a subject and also provoke additional
skepticism for the experiments that found significant results. The secondary hypotheses on the
relationships between Cheekbone height, Upper head Length, Face heartshapeness and the
trusting decisions of the previous data were also based on the previous studies as each of the
metrics were related to facial height and correspondingly will have a lessened facial width.
Therefore, it was predicted that these measures would have a negative correlation with trusting
decisions, though this hypothesis is also unsupported. Finally, there were 3 facial characteristics
that were used as exploratory variables and the relationship between them could not be
hypothesized. These variables, Cheekbone prominence, Lip fullness, and Nose shape were also
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 20
uncorrelated with trusting behavior for any of the conditions. However, due to the fact that these
variables lack previous research, more analyses and data must be collected before no effect can
be confirmed.
Limitations
The lack of significant relationships in the analyses prompted an abundance of
consideration of different techniques or designs that could have been implemented. One of the
main limitations that should be acknowledged is the difference between trait descriptions and
behavioral actions. Each of the previous studies that were mentioned before described how facial
features were identified to be correlated with subjective ratings of personality traits such as
trustworthiness or attractiveness. For example, the objective facial features corresponded with
external ratings of trustworthiness, though these ratings of trustworthiness do not necessarily
predict that other individuals will trust those faces. These previous studies found a significant
effect, but because this study analyzed the facial features correlation to actual behaviors instead
of subjective ratings, it may be the reason for conflicting results. Therefore, if this study was
redesigned to analyze the relationships between objective facial features and participant’s
subjective ratings of the facial stimuli, it may produce different results. In addition, if the
distinction between subjective ratings and active trusting behaviors is not made clear in past
studies, it could be one of the potential explanations for the controversial findings from the past
few decades.
Furthermore, by using the actual trusting behaviors in this study, other social factors need
to be acknowledged. For example, because each of the participants in the attention study were
deceived into believing that they would be interacting with the person in the stimuli pictures, the
participants may have altered their judgements based on how they see that stimulus person. With
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 21
traits such as attractiveness, individuals can either specifically choose one person over another
regardless of feelings of trustworthiness or attention cueing or the individuals may avoid a
person they deem to be attractive because the participant may be intimidated by the
attractiveness. In either case, the results would then be disturbed by other effects interfering and
thus the objective measures would not reflect the choice being made. Situations like this could
potentially create a false positive result and should be kept in mind for future studies.
In addition, there are several lines of research on prosocial behavior that inspect the actor
in a prosocial situation and not the target of the behavior. Each of the studies cited here
(Cañigueral, & Hamilton, 2019; Dawans, Fischbacher, Kirschbaum, Fehr, & Heinrichs, 2012;
Guo, Sun, Cai, Zhang, & Song, 2019; Guo, Sun, & Li, 2018; Jones, 2008) focus on the role of
the actor in prosocial situations and how traits such as shyness, intelligence, stress reactivity, and
other factors all impact the decision to engage in prosocial behavior. Therefore, the participant’s
personality traits may impact the trusting behavior as much as the objective face measures do.
Another issue with the participant’s trusting behaviors is that they are not able to mimic the
decision that each participant would make outside of the study. Thus, because the trust
behaviors were observed in a lab setting under a forced option situation, the results are not
representative of real-world actions and decisions.
There were also a few operational limitations to the objective face analysis that could
have created a false negative result. First, the stimuli pictures were not systematically captured.
Researchers in the original study instructed picture stimuli participants to smile without showing
teeth or not smile at all, however, since these people had the option to choose to take the picture
differently or even smile differently when they choose to smile, there are notable differences
between pictures. One of the studies in facial features analyzes the effect of eyes being open,
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 22
closed, or looking away (Manesi, Van Lange, & Pollet, 2016). These findings are important
because while people are asked to smile, they each squint or close their eyes in different
amounts, thus possibly leading to a misleading characteristic during measurement. As the
attention experiment also notes, expressions on the stimuli picture’s faces and the emotional
ratings of those pictures can influence giving behavior. Therefore, if social situations can impact
the emotional salience of each photo then the slight difference in expressions between the faces
may alter the trusting decision. Another issue with the faces used is that many of the pictures
were not taken with the same qualities and so each photo had to be manually adjusted to attempt
to decrease the amount of photograph discrepancy. While it is important to have differing and
distinct features between stimuli faces, having a standardized photograph pool to use as the
recipients of trust may contribute to observing an effect. Finally, the data on trusting behaviors
itself was taken from another study that had a different research questions and different
experiment goals in mind. Because of this, although the simple results are able to translate onto
this current study, there may be errors in study design as it relates to the current research
question.
Future Directions
As mentioned multiple times already, this area of study is controversial and has held
multiple conflicting views in the past century. This fact alone requires that more research is
conducted to test and retest the results that have already been determined. For future studies in
objective facial measures, there are multiple improvements to be made. First, using a new
stimulus set that is either computer generated faces, or controlled for expressions, facial hair,
photographic qualities, and equal ethnic group representation would be able to increase the
validity of the results. Additionally, this study can be divided into two separate future lines of
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 23
research with one targeting objective facial metric’s influence in applied behavioral observations
and the other focusing on how the same objective facial features are individually rated on
personality traits without the behavioral or social factors, thereby refining the research question
and implementation of quality experiments. Furthermore, the research gathered on
trustworthiness and attractiveness shows that the two are correlated, even though the two
different traits are shown to be related to different features. The correlation may be due to the
halo effect in which seeing a stimulus face as generally good may influence the ratings of
trustworthiness as well as attractiveness despite the two actually being correlated. Therefore,
additional exploration into perceived attractiveness and trustworthiness and how each trait is
related to facial characteristics would be important.
Conclusion
This study resulted in no significant relationships between the facial characteristics and
trusting decisions, however, there is still a lot of data that can be taken from these results. Having
no significant relationships conflicts with several contending studies that propose these same
measures are correlated with personality traits such as trustworthiness and attractiveness.
Because of this, it is important to replicate the findings of the previous research as well, as the
results of this study, in order to start bridging the gap between this controversial debate.
Physiognomy is an interesting topic and there is clear evidence of humans making automatic and
highly consistent judgements on faces; thus, it is imperative that these judgements are able to be
reasoned and supported, or otherwise argued and discarded as useless first impressions.
Regardless, there is a reason why humans have developed the ability to recognize and judge
faces with such refined skill and that is a research area that is waiting to be explored.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 24
References
Alley, T. R., & Hildebrandt, K. A. (1988). Determinants and consequences of facial aesthetics
(pp. 101-140). TR Alley (Ed.), Social and applied aspects of perceiving faces.
Anderson, L. D. (1921). Estimating Intelligence by Means of Printed Photographs. Journal of
Applied Psychology, 5(2), 152–155. https://doi.org/10.1037/h0075189
Aristotle. (1905). Aristotle's Politics. Oxford :Clarendon Press,
Atkinson, M. A., Simpson, A. A., & Cole, G. G. (2018). Visual attention and action: How
cueing, direct mapping, and social interactions drive orienting. Psychonomic Bulletin &
Review, 25(5), 1585-1605. doi:10.3758/s13423-017-1354-0
Berry, D. S., & Brownlow, S. (1989). Were the physiognomists right? Personality correlates of
facial babyishness. Personality and Social Psychology Bulletin, 15(2), 266-279.
Bull, R., & Gibson-Robinson, E. (1981). The Influences of Eye-Gaze, Style of Dress, and
Locality on the Amounts of Money Donated to a Charity. Human Relations, 34(10), 895–
905. https://doi.org/10.1177/001872678103401005
Cañigueral, R., & Hamilton, Antonia F. de C. (2019). Being watched: Effects of an audience on
eye gaze and prosocial behaviour. Acta Psychologica, 195, 50-63.
doi:10.1016/j.actpsy.2019.02.002
Charlesworth, T. E. S., Hudson, S. T. J., Cogsdill, E. J., Spelke, E. S., & Banaji, M. R. (2019).
Children use targets’ facial appearance to guide and predict social
behavior. Developmental Psychology, 55(7), 1400-1413.
doi:http://dx.doi.org.colorado.idm.oclc.org/10.1037/dev0000734
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 25
Cohen, R. (2016). Patterns of personality judgment. Elsevier.
Colvin, J. A., Grant, M. D., Van Boven, L., & Pedersen, E. J. (2021, January 28). Honors Thesis.
Retrieved from: osf.io/qhkby
Dawans, B. v., Fischbacher, U., Kirschbaum, C., Fehr, E., & Heinrichs, M. (2012). The social
dimension of stress reactivity: Acute stress increases prosocial behavior in
humans. Psychological Science, 23(6), 651-660. doi:10.1177/0956797611431576
Grant, D. M. (2019). The Effect of Attention and Emotional Salience on Prosocial Behavior
(Undergraduate Honors Thesis). University of Colorado, Boulder, CO
Grant, D. M., Van Boven, L., Pederson, E., & Ramos, J. (2019). How Attention Influences
Prosocial Behavior. University of Colorado Boulder.
Guo, Q., Sun, P., Cai, M., Zhang, X., & Song, K. (2019). Why are smarter individuals more
prosocial? A study on the mediating roles of empathy and moral identity. Intelligence, 75,
1-8. doi:10.1016/j.intell.2019.02.006
Hassin, R., & Trope, Y. (2000). Facing faces: Studies on the cognitive aspects of
physiognomy. Journal of Personality and Social Psychology, 78(5), 837-852.
doi:10.1037/0022-3514.78.5.837
Jones, A. L., Schild, C., & Jones, B. C. (2020). Facial metrics generated from manually and
automatically placed image landmarks are highly correlated. Evolution and Human
Behavior, doi:10.1016/j.evolhumbehav.2020.09.002
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 26
Jones, G. (2008). Are smarter groups more cooperative? evidence from prisoner's dilemma
experiments, 1959–2003. Journal of Economic Behavior and Organization, 68(3), 489-
497. doi:10.1016/j.jebo.2008.06.010
Manesi, Z., Van Lange, P. A. M., & Pollet, T. V. (2016). Eyes Wide Open: Only Eyes That Pay
Attention Promote Prosocial Behavior. Evolutionary Psychology. Retrieved from:
https://doi.org/10.1177/1474704916640780
Milutinovic, J., Zelic, K., & Nedeljkovic, N. (2014). Evaluation of facial beauty using
anthropometric proportions. Thescientificworld, 2014, 428250-8.
doi:10.1155/2014/428250
Mrkva, K., & Van Boven, L. (2017). Attentional accounting: Voluntary spatial attention
increases budget category prioritization. Journal of Experimental Psychology: General,
146(9), 1296.
Pashler, H., & Harris, C. R. (2001). Spontaneous allocation of visual attention: Dominant role of
uniqueness. Psychonomic Bulletin & Review, 8(4), 747-752. doi:10.3758/BF03196213
Penton-Voak, I. S., Pound, N., Little, A. C., & Perrett, D. I. (2006). Personality judgments from
natural and composite facial images: More evidence for a “kernel of truth” in social
perception. Social cognition, 24(5), 607-640.
Rezlescu, C., Duchaine, B., Olivola, C. Y., & Chater, N. (2012). Unfakeable facial
configurations affect strategic choices in trust games with or without information about
past behavior. PloS One, 7(3), e34293. doi:10.1371/journal.pone.0034293
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 27
Rojas, M., Masip, D., Todorov, A., & Vitria, J. (2011). Automatic prediction of facial trait
judgments: Appearance vs. structural models. PloS One, 6(8), e23323-e23323.
doi:10.1371/journal.pone.0023323
RStudio Team (2020). RStudio: Integrated Development for R. RStudio, PBC, Boston, MA URL
http://www.rstudio.com/
Secord, P. F., & Bevan, W. (1956). Personalities in faces: III. A cross-cultural comparison of
impressions of physiognomy and personality in faces. The Journal of Social
Psychology, 43(2), 283-288.
Smillie, L. D., Lawn, E. C. R., Zhao, K., Perry, R., & Laham, S. M. (2019). Prosociality and
morality through the lens of personality psychology. Australian Journal of Psychology,
71(1), 50-58. doi:10.1111/ajpy.12229
Smith, F. G., Debruine, L. M., Jones, B. C., Krupp, D. B., Welling, L. L. M., & Conway, C. A.
(2009). Attractiveness qualifies the effect of observation on trusting behavior in an
economic game. Evolution and Human Behavior, 30(6), 393-397.
doi:10.1016/j.evolhumbehav.2009.06.003
Stirrat, M., & Perrett, D. I. (2010). Valid facial cues to cooperation and trust: Male facial width
and trustworthiness. Psychological Science, 21(3), 349-354.
doi:10.1177/0956797610362647
Stirrat, M., & Perrett, D. I. (2012). Face structure predicts cooperation: Men with wider faces are
more generous to their in-group when out-group competition is salient. Psychological
science, 23(7), 718-722.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 28
The GIMP Development Team. (2020). GIMP 2.10.20. Retrieved from: https://www.gimp.org
Todorov, A., Pakrashi, M., & Oosterhof, N. N. (2009). evaluating faces on trustworthiness after
minimal time exposure. Social Cognition, 27(6), 813-833.
doi:10.1521/soco.2009.27.6.813
Todorov, A., Said, C. P., Engell, A. D., & Oosterhof, N. N. (2008). Understanding evaluation of
faces on social dimensions. Trends in Cognitive Sciences, 12(12), 455-460.
doi:10.1016/j.tics.2008.10.001
Wilson, B. J. (2003). The role of attentional processes in children's prosocial behavior with
peers: Attention shifting and emotion. Development and Psychopathology, 15(2), 313-
329. doi:10.1017/S0954579403000178
Xiao, F., Zheng, Z., Zhang, H., Xin, Z., Chen, Y., & Li, Y. (2016). Who Are You More Likely to
Help? The Effects of Expected Outcomes and Regulatory Focus on Prosocial
Performance. PLoS ONE, 11(11), e0165717. Retrieved from: https://link-gale-
com.colorado.idm.oclc.org/apps/doc/A471838782/PPNU?u=coloboulder&sid=PPNU&xi
d=b27fa5d7
Appendix
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 29
Figure 1: Example of measurements taken for each face (the face shown here is an example from
the Chicago Face Database and was not used in the study). Each line represents a different
measurement. In cases such as the forehead height, nose height, or lip height, the example line is
offset to show a separate measurement, however the measurements used would be in the same
position as the facial height.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 30
Figure 2: Example of what is seen by experiment subject for the stream of letters.
Figure 3: Screen showing instructions for faces displayed on the screen for subjects.
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 31
Figure 4: Correlation between the Proportion of Trusting Behavior regardless of cueing and Facial Width
to Height Ratio
Figure 5: Correlation between the Proportion of Trusting Behavior regardless of cueing and Face
Roundedness
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 32
Figure 6: Correlation between the Proportion of Trusting Behavior regardless of cueing and Eye Shape
Figure 7: Correlation between the Proportion of Trusting Behavior regardless of cueing and Eye Size
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 33
Figure 8: Correlation between the Proportion of Trusting Behavior regardless of cueing and the Distance
between the Center of the Pupils
Figure 9: Correlation between the Proportion of Trusting Behavior regardless of cueing and Face
Heartshapeness
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 34
Figure 10: Correlation between the Proportion of Trusting Behavior regardless of cueing and Upper Head
Length
Figure 11: Correlation between the Proportion of Trusting Behavior regardless of cueing and Cheekbone
Height
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 35
Figure 12: Correlation between the Proportion of Trusting Behavior regardless of cueing and Cheekbone
Prominence
Figure 13: Correlation between the Proportion of Trusting Behavior regardless of cueing and Lip Fullness
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 36
Figure 14: Correlation between the Proportion of Trusting Behavior regardless of cueing and Nose Shape
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 37
Estimate Standard
Error t- Value Pr(>ltl)
Intercept 2.74 5.47 0.50 0.617
Facial Width to
Height Ratio
0.48 1.05 0.46 0.649
Upper Head Length -0.19 0.63 -0.31 0.757
Face Roundedness -0.95 1.41 -0.68 0.501
Face Heartshapeness -1.95 4.89 0.40 0.691
Cheekbone Prominence 1.64 8.52 0.19 0.848
Cheekbone Height 0.03 1.06 0.03 0.976
Eye Shape 0.42 0.77 0.54 0.590
Eye Size -1.31 5.61 -0.23 0.816
Distance Between Pupils 0.00 0.00 0.25 0.805
Lip Fullness 0.76 1.08 0.70 0.485
Nose Shape -0.13 0.19 -0.70 0.487
Table 1: Multiple Linear Regression for All Characteristics and All Faces
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 38
Estimate Standard
Error t- Value Pr(>ltl)
Intercept 0.81 4.94 0.16 0.870
Facial Width to
Height Ratio
-0.70 0.95 -0.74 0.462
Upper Head Length -0.85 0.57 -1.50 0.136
Face Roundedness 0.90 1.28 0.70 0.483
Face Heartshapeness 0.06 4.42 0.01 0.990
Cheekbone Prominence 0.84 7.71 0.11 0.913
Cheekbone Height -1.29 0.96 -1.35 0.180
Eye Shape -0.79 0.70 -1.13 0.262
Eye Size 3.66 5.07 0.72 0.472
Distance Between Pupils -0.00 0.00 -0.06 0.953
Lip Fullness 0.79 0.97 0.81 0.421
Nose Shape -0.08 0.17 -0.50 0.620
Table 2: Multiple Linear Regression for All Characteristics and the Difference Proportion
Between Cued Faces and Not Cued Faces
INFLUENCE OF OBJECTIVE FACIAL FEATURES ON TRUSTING BEHAVIOR 39
Estimate Standard
Error t- Value Pr(>ltl)
Intercept 0.50 0.01 38.12 <.001
Average Face Shape <-0.01 0.01 -0.23 0.813
Average Eye Shape 0.01 0.01 1.06 0.290
Avg Face : Avg Eye <-0.01 0.02 -0.15 0.880
Table 3: Multivariate Linear Regression with Interactions for Average Face Shape and Average
Eye Shape with proportion of all faces
Estimate Standard
Error t- Value Pr(>ltl)
Intercept 0.04 0.01 3.08 0.003
Average Face Shape -0.01 0.01 -0.81 0.422
Average Eye Shape -0.01 0.01 -0.58 0.563
Avg Face : Avg Eye 0.01 0.01 0.71 0.481
Table 4: Multivariate Linear Regression with Interactions for Average Face Shape and Average
Eye Shape with proportion of cued difference.