497969
research-article2013
PSSXXX10.1177/0956797613497969Walker, VulHierarchical Encoding of Individuals
Research Report
Hierarchical Encoding Makes Individuals
in a Group Seem More Attractive
Psychological Science
2014, Vol. 25(1) 230–235
© The Author(s) 2013
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0956797613497969
pss.sagepub.com
Drew Walker and Edward Vul
University of California, San Diego
Abstract
In the research reported here, we found evidence of the cheerleader effect—people seem more attractive in a group
than in isolation. We propose that this effect arises via an interplay of three cognitive phenomena: (a) The visual
system automatically computes ensemble representations of faces presented in a group, (b) individual members of the
group are biased toward this ensemble average, and (c) average faces are attractive. Taken together, these phenomena
suggest that individual faces will seem more attractive when presented in a group because they will appear more
similar to the average group face, which is more attractive than group members’ individual faces. We tested this
hypothesis in five experiments in which subjects rated the attractiveness of faces presented either alone or in a group
with the same gender. Our results were consistent with the cheerleader effect.
Keywords
visual perception, face perception
Received 1/18/13; Revision accepted 6/19/13
In the seventh episode of the fourth season of How I Met
Your Mother, the character Barney Stinson postulates the
cheerleader effect: that people seem more attractive in a
group than when considered individually (Rashid &
Fryman, 2008). As proposed, this effect is not simply that
a member of the cheerleading squad, for instance, is
more attractive than a person sitting alone in the bleachers (which could be due to factors such as objective
attractiveness, altered demeanor, or social signaling), but
rather that any given cheerleader will seem more attractive when seen as part of the squad than in isolation.
We propose that the cheerleader effect occurs at a
perceptual level, arising from the interplay between
ensemble coding in the visual system and properties of
average faces. The visual system automatically computes
summary representations of ensembles of objects, such
as the average size of an array of dots (Ariely, 2001;
Chong & Treisman, 2003), the average orientation of an
array of gratings (Parks, Lund, Angelucci, Solomon, &
Morgan, 2001), and even the average emotional expression of a group of faces (Haberman & Whitney, 2009).
Not only does the summary that is formed influence
observers’ perception of the group as a whole, but it also
biases their percepts of individual items to be more like
the group average (Brady & Alvarez, 2011). Thus, we
expected individual faces seen in a group to appear to be
more similar to the average of the group than when seen
alone. Moreover, the average of a number of faces tends
to be perceived as more attractive than the individual
faces it comprises (Langlois & Roggman, 1990). Thus, the
bias of individual elements toward the ensemble average,
when applied to faces, will yield a perception of individual faces as being more attractive than they would
otherwise be perceived to be. In other words, the biasing
effect of ensemble coding should produce a cheerleader
effect. We tested this prediction in five experiments.
Subjects
Subjects were undergraduate students from the University
of California, San Diego, and received partial course
credit. There were 25 subjects in Experiment 1 (4 men, 21
women), 18 in Experiment 2 (6 men, 12 women), 20 in
Corresponding Author:
Drew Walker, University of California, San Diego, Department of
Psychology, 9500 Gilman Dr., La Jolla, CA 92093-0109
E-mail: dehoffma@ucsd.edu
Hierarchical Encoding of Individuals
Experiment 3 (3 men, 17 women), 37 in Experiment 4 (13
men, 24 women), and 39 in Experiment 5 (10 men, 29
women).
231
Subjects rated the attractiveness of female faces in
Experiment 1 and male faces in Experiment 2. Faces were
presented in a group photograph and in isolated portraits
cropped from the group photos.
On group trials, the three faces in the image were
rated individually in a random order. Subjects saw the
group photo for 1 s, after which an arrow appeared for 1
s below one of the faces (randomly chosen). Then the
group image disappeared, and subjects made a rating.
The group photo then reappeared for 1s, and the next
face was cued for 1 s. This process repeated once more
so that all three faces in the image were rated. On portrait
trials, the cropped single-person image appeared for 2 s,
disappeared, and then subjects made their rating.
Method
Results
For each experiment, we found 100 group photographs
and cropped them to frame the faces of three people of
the same gender. We then cropped each individual face
to create three portrait images from each group photo. In
both experiments, subjects rated the 300 unique faces
twice, once in the group photo and once in an isolated
portrait. Ratings were made by moving a mouse to set a
marker on a continuous scale from unattractive to attractive (the rating scale and example stimuli are shown in
Fig. 1). The order of images and whether a face appeared
first in a group or as a portrait was random.
In our analysis, we aimed to measure the cheerleader
effect, the advantage in perceived attractiveness granted
a face when it is seen in a group rather than alone, while
factoring out the variation in how individual subjects
used our rating scale and variations in how attractive
they found the different faces to be. To factor out individual differences in rating-scale use, we converted the
raw rating given by a subject for each image in each
condition (group and portrait) into a within-subjects z
score by subtracting the mean rating and dividing the
result by the standard deviation of the 600 ratings made
by the subject. To factor out the effect of the attractiveness of specific faces, we then subtracted each subject’s
standardized rating of a face presented as a portrait from
his or her standardized rating of that same face presented
in a group. The resulting difference in z scores corresponded to the number of standard deviations higher
that a given image was rated in a group than when isolated in a portrait. Using these difference scores, we
assessed the average cheerleader-effect size (z-score difference) for each subject, as well as the average effect
size across subjects (Fig. 2).
Although there was considerable between-subjects
variation in effect sizes, subjects on average rated female
faces in a group as being 5.5% of a standard deviation
more attractive than those same faces in isolation
(Experiment 1), t(24) = 2.53, p = .018. This cheerleader
effect also held (with surprising consistency in effect
size) for male faces: There was an average advantage of
5.6% of a standard deviation for faces in a group
(Experiment 2), t(17) = 2.52, p = .022.
Experiments 1 and 2
Unattractive …………………………. Attractive
Unattractive …………………………. Attractive
Fig. 1. Rating scale and example stimuli used in Experiments 1 and
3. Subjects rated the attractiveness of 300 faces twice, once in a group
photo (top; the arrow indicated which face was to be rated) and once
in an isolated portrait (bottom). Attractiveness was rated using a mouse
to set a marker along a continuous scale. Stimuli were presented in
color in the actual experiments.
Experiment 3
In Experiments 1 and 2, each face in the group condition
was presented uncued three times for 1 s each (a total of
3 s) and presented cued for 1 s, which suggests that any
one face was on average attended for 2 s total. Thus,
average time spent attending to any one face in the group
condition was equivalent to the 2-s presentation in the
portrait condition. However, any one trial of the group
Walker, Vul
232
0.3
0.0
0.1
0.2
Female Faces
−0.2 −0.1
Effect Size
(z-Scored Difference Between
Portrait and Group Ratings)
Experiment 1 (N = 25)
Subject (Ordered by Effect Strength)
Pooled
0.3
0.0
0.1
0.2
Male Faces
−0.2 −0.1
Effect Size
(z-Scored Difference Between
Portrait and Group Ratings)
Experiment 2 (N = 18)
Subject (Ordered by Effect Strength)
Pooled
−0.2 −0.1 0.0 0.1 0.2 0.3
Effect Size
(z-Scored Difference Between
Portrait and Group Ratings)
Experiment 3 (N = 20)
Female Faces
Subject (Ordered by Effect Strength)
Pooled
Fig. 2. Results for Experiments 1, 2, and 3: standardized size of the cheerleader effect for ratings of faces,
separately for individual subjects (left panels) and pooled across subjects (right panels). To calculate effect
sizes, we converted the raw rating given by a subject for each image in each condition (group and portrait)
into a within-subjects z score by subtracting the mean rating and dividing the result by the standard deviation of the 600 ratings made by the subject. For each subject, we then subtracted this z score for the faces
in the portrait condition from the z score for the faces in the group condition. This difference yielded each
subject’s effect size: the number of standard deviations higher that a given face was rated when seen in
a group than when seen isolated in a portrait. Error bars for individual data show ±1 SEM, and error bars
for pooled data show 95% confidence intervals.
condition consisted of an uncued group of three faces for
1 s, and a cued face for 1 s, which meant that 1.33 s was
spent attending to that face. In this sense, the expected
time spent attending to a face in one group trial was
shorter than in a portrait trial. It is plausible that this difference drove the effect in Experiments 1 and 2 because
faces shown for shorter durations are rated as more
attractive than faces shown for longer durations (Willis &
Todorov, 2006). Although Willis and Todorov found an
advantage of shorter duration only for presentations
briefer than 500 ms (and ours were all longer than 1 s),
we wanted to replicate our results from Experiments 1
Hierarchical Encoding of Individuals
and 2 by equating the presentation time of one portrait
trial to one presentation in the group trial. We did so in
Experiment 3 by presenting the portrait images for just
1.33 s (otherwise, the design, stimuli, and method of data
analysis were the same as in Experiment 1). With this
modified timing, we replicated the cheerleader effect
from Experiments 1 and 2: When the presentation duration of portrait images was shortened, faces were rated
6.8% of a standard deviation more attractive when presented in a group than when presented alone, t(19) =
2.50, p = .022 (Fig. 2).
Experiment 4
In Experiments 1 through 3, all of the faces had originally
been photographed together in a real-life social context.
Perhaps group images were rated as more attractive than
single images not because of ensemble coding of the
group but because the coherent context disambiguated
facial expressions or other image idiosyncrasies (just as
videos of an individual are rated as more flattering than
the static photos that comprise them; Post, Haberman,
Iwaki, & Whitney, 2012). In Experiment 4, we sought to
rule out this class of explanations by presenting an array
composed of multiple portrait faces that had been photographed separately. In addition to this control, we also
tested for effects of group size: Increasing group size
should yield a more precise average face that should not
only be rated as more attractive (Langlois & Roggman,
1990) but should also exert a greater bias on the perceived attractiveness of individual faces (given a probabilistic combination of individuals and the ensemble; Brady
& Alvarez, 2011).
Method
We randomly chose 77 unique faces from the stimuli
used in Experiment 1. Each was presented once in each
of four conditions: alone and as part of a group of 4, 9,
and 16 other faces. The flanker faces in the group conditions were randomly chosen from the 223 remaining
faces used in Experiment 1 (target faces were never used
as flankers). Faces were presented in a square grid (1 × 1
for 1 face, 2 × 2 for 4 faces, 3 × 3 for 9, and 4 × 4 for 16;
Fig. 3a). Each grid appeared for 2 s, and then a box
appeared around the target face for 1 s. The faces then
disappeared, and subjects made a rating as in the previous experiments. In the portrait condition, the face was
presented alone in the center of the computer screen for
2 s before subjects made a rating.
Results
As in the previous experiments, we z-scored ratings
within a given subject to factor out between-subjects
233
variation in scale usage. To factor out variability in the
actual attractiveness of a given face, we subtracted the
average (across subjects and conditions) standardized
rating given to each face from each rating of that face.
This gave us a measure of the effect of each presentation
condition. Figure 3b shows the average standardized ratings in each condition across all subjects. There was a
significant effect of group size on attractiveness ratings,
F(3, 144) = 11.74, p < .001: Consistent with a cheerleader
effect, results showed that faces were rated as less attractive when presented alone than when presented in a
group of 4, t(36) = 3.23, a group of 9, t(36) = 4.25, or a
group of 16, t(36) = 4.0. However, attractiveness ratings
were not different for faces rated in groups of 4, 9, or 16.
These results suggest that it is not the coherent context of
group photos but rather the presence of additional faces
that drives the cheerleader effect.
Experiment 5
The influence of group membership on individual members may be greater when there is more uncertainty
about the individual elements in a scene; this is because
the average is less sensitive to the increased uncertainty
than the individual elements are. Following this logic, we
blurred the faces in Experiment 5 to see whether the
cheerleader effect would be increased when uncertainty
was increased.
Method
We randomly selected 50 group images from those used
in Experiment 1 and blurred them by convolving them
with a Gaussian filter with a standard deviation of 4 pixels. Subjects rated the three faces in each of those 50
images (150 unique faces) four times each: in unblurred
group and portrait conditions and in blurred group and
portrait conditions (example stimuli are shown in Fig.
4a). Other than the addition of the blurring factor, methods were identical to those used in Experiment 1.
Results
As in Experiment 4, we isolated the effect of condition by
z-scoring ratings within subjects and subtracting the
across-subjects average ratings for each face. Figure 4b
shows the average standardized ratings in each condition. As in our other experiments, faces were rated as
more attractive when seen in groups than when seen
alone, F(1, 152) = 9.0, p < .01, and subjects rated blurred
images as more attractive than unblurred images, F(1,
152) = 17.91, p < .001. However, although the cheerleader effect was bigger in the blurred condition than in
the unblurred condition (7.3% vs. 5.9% of a standard
deviation), the interaction between image clarity and
Walker, Vul
234
a
b
0.00
–0.05
–0.10
Average Standardized
Attractiveness Rating
0.05
Experiment 4 (N = 37)
1
4
9
16
Group Size
Fig. 3. Example stimuli (a) and results (b) from Experiment 4. Subjects rated the attractiveness of 77
faces four times: alone and in a group of 4, 9, and 16 other faces. Each group appeared for 2 s, and then
a box appeared around the target face for 1 s. In the portrait condition, the face was presented alone in
the center of the computer screen for 2 s. The faces then disappeared, and subjects made a rating as in
the previous experiments. Stimuli were presented in color in the actual experiment. The graph shows the
average standardized attractiveness ratings for each group size. To calculate attractiveness ratings, we first
obtained within-subjects z scores as in the previous experiments. For each subject, we then subtracted
the average (across subjects and conditions) z score given to each face from each rating of that face. Error
bars show 95% confidence intervals.
presentation condition was not significant F(1, 152) =
0.106, p = .75.
General Discussion
In the five experiments reported here, we found evidence
consistent with the cheerleader effect: Both female faces
(Experiment 1) and male faces (Experiment 2) in a group
appeared more attractive than those same faces seen
alone.1 This effect seems robust to presentation timing
(Experiment 3), to whether groups are created from natural photos or are synthetically created (Experiment 4),
and to image manipulations such as blurring (Experiment
5). We propose that this effect arises from the fact that the
visual system represents objects as an ensemble (Ariely,
2001), individual objects are biased toward the ensemble
average (Brady & Alvarez, 2011), and average faces are
perceived to be more attractive than faces in isolation
(Langlois & Roggman, 1990). Together, these phenomena
should cause faces in a group to appear more like the
Hierarchical Encoding of Individuals
235
a
in groups were rated as more attractive than the same
people alone. Thus, having a few wingmen—or wingwomen—may indeed be a good dating strategy, particularly if their facial features complement, and average out,
one’s unattractive idiosyncrasies.
Author Contributions
b
Both authors contributed to the design of the experiments.
The experiment was programmed by D. Walker. E. Vul and
D. Walker analyzed and interpreted the data. D. Walker and
E. Vul drafted the manuscript. Both authors approved the final
version of the manuscript for submission.
Portrait Condition, Blur
Group Condition, Blur
Portrait Condition, No Blur
Group Condition, No Blur
Declaration of Conflicting Interests
Average Standardized
Attractiveness Rating
Experiment 5 (N = 39)
0.10
The authors declared that they had no conflicts of interest with
respect to their authorship or the publication of this article.
Note
0.05
–0.05
1. Critically, the same face when seen in a group of different faces is rated as more attractive than when seen alone.
However, Post et al. (2012) found no such effect for a face presented in an arrays of the same face.
–0.10
References
0.00
Blur
No Blur
Image Clarity
Fig. 4. Example stimuli (a) and results (b) from Experiment 5. Subjects
rated the attractiveness of 50 faces four times each: in blurred group
and portrait conditions (shown here) and in unblurred group and portrait conditions. Stimuli were presented and ratings were made as in
Experiment 1. Stimuli were presented in color in the actual experiment.
The graph shows the average standardized attractiveness ratings as a
function of image clarity and presentation condition (group or portrait).
Attractiveness ratings were calculated as in Experiment 4. Error bars
show 95% confidence intervals.
group average than when presented alone, and that
group average should tend to be more attractive than the
individual faces, on average. However, some of our
results should give readers pause in accepting our interpretation: We predicted that increasing group size
(Experiment 4) or decreasing image quality (Experiment
5) should increase the bias of individuals to the group
average and would thus increase the cheerleader effect,
but we found no evidence of these effects. Despite this
caveat about our interpretation, the cheerleader effect
was robust: Across a wide range of settings, people
Ariely, D. (2001). Seeing sets: Representation by statistical properties. Psychological Science, 12, 157–162.
Brady, T. F., & Alvarez, G. A. (2011). Hierarchical encoding in
visual working memory: Ensemble statistics bias memory
for individual items. Psychological Science, 22, 384–392.
Chong, S. C., & Treisman, A. (2003). Representation of statistical
properties. Vision Research, 43, 393–404.
Haberman, J., & Whitney, D. (2009). Seeing the mean: Ensemble
coding for sets of faces. Journal of Experimental Psychology:
Human Perception and Performance, 35, 718–734.
Langlois, J. H., & Roggman, L. A. (1990). Attractive faces are
only average. Psychological Science, 1, 115–121.
Parks, L., Lund, J., Angelucci, A., Solomon, J. A., & Morgan,
M. (2001). Compulsory averaging of crowded orientation
signals in human vision. Nature Neuroscience, 4, 739–
744.
Post, R. B., Haberman, J., Iwaki, L., & Whitney, D. (2012). The
frozen face effect: Why static photographs may not do you
justice. Frontiers in Psychology, 3, 22. Retrieved from http://
www.frontiersin.org/Journal/10.3389/syg.2012.00022/full
Rashid, R. (Writer), & Fryman, P. (Director). (2008). Not a father’s
day [Television series episode]. In C. Bays & C. Thomas
(Creators), How I met your mother. New York, NY: CBS.
Willis, J., & Todorov, A. (2006). First impressions: Making up
your mind after a 100-ms exposure to a face. Psychological
Science, 17, 592–598.
497969
research-article2013
PSSXXX10.1177/0956797613497969Walker, VulHierarchical Encoding of Individuals
Research Report
Hierarchical Encoding Makes Individuals
in a Group Seem More Attractive
Psychological Science
2014, Vol. 25(1) 230–235
© The Author(s) 2013
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0956797613497969
pss.sagepub.com
Drew Walker and Edward Vul
University of California, San Diego
Abstract
In the research reported here, we found evidence of the cheerleader effect—people seem more attractive in a group
than in isolation. We propose that this effect arises via an interplay of three cognitive phenomena: (a) The visual
system automatically computes ensemble representations of faces presented in a group, (b) individual members of the
group are biased toward this ensemble average, and (c) average faces are attractive. Taken together, these phenomena
suggest that individual faces will seem more attractive when presented in a group because they will appear more
similar to the average group face, which is more attractive than group members’ individual faces. We tested this
hypothesis in five experiments in which subjects rated the attractiveness of faces presented either alone or in a group
with the same gender. Our results were consistent with the cheerleader effect.
Keywords
visual perception, face perception
Received 1/18/13; Revision accepted 6/19/13
In the seventh episode of the fourth season of How I Met
Your Mother, the character Barney Stinson postulates the
cheerleader effect: that people seem more attractive in a
group than when considered individually (Rashid &
Fryman, 2008). As proposed, this effect is not simply that
a member of the cheerleading squad, for instance, is
more attractive than a person sitting alone in the bleachers (which could be due to factors such as objective
attractiveness, altered demeanor, or social signaling), but
rather that any given cheerleader will seem more attractive when seen as part of the squad than in isolation.
We propose that the cheerleader effect occurs at a
perceptual level, arising from the interplay between
ensemble coding in the visual system and properties of
average faces. The visual system automatically computes
summary representations of ensembles of objects, such
as the average size of an array of dots (Ariely, 2001;
Chong & Treisman, 2003), the average orientation of an
array of gratings (Parks, Lund, Angelucci, Solomon, &
Morgan, 2001), and even the average emotional expression of a group of faces (Haberman & Whitney, 2009).
Not only does the summary that is formed influence
observers’ perception of the group as a whole, but it also
biases their percepts of individual items to be more like
the group average (Brady & Alvarez, 2011). Thus, we
expected individual faces seen in a group to appear to be
more similar to the average of the group than when seen
alone. Moreover, the average of a number of faces tends
to be perceived as more attractive than the individual
faces it comprises (Langlois & Roggman, 1990). Thus, the
bias of individual elements toward the ensemble average,
when applied to faces, will yield a perception of individual faces as being more attractive than they would
otherwise be perceived to be. In other words, the biasing
effect of ensemble coding should produce a cheerleader
effect. We tested this prediction in five experiments.
Subjects
Subjects were undergraduate students from the University
of California, San Diego, and received partial course
credit. There were 25 subjects in Experiment 1 (4 men, 21
women), 18 in Experiment 2 (6 men, 12 women), 20 in
Corresponding Author:
Drew Walker, University of California, San Diego, Department of
Psychology, 9500 Gilman Dr., La Jolla, CA 92093-0109
E-mail: dehoffma@ucsd.edu
Hierarchical Encoding of Individuals
Experiment 3 (3 men, 17 women), 37 in Experiment 4 (13
men, 24 women), and 39 in Experiment 5 (10 men, 29
women).
231
Subjects rated the attractiveness of female faces in
Experiment 1 and male faces in Experiment 2. Faces were
presented in a group photograph and in isolated portraits
cropped from the group photos.
On group trials, the three faces in the image were
rated individually in a random order. Subjects saw the
group photo for 1 s, after which an arrow appeared for 1
s below one of the faces (randomly chosen). Then the
group image disappeared, and subjects made a rating.
The group photo then reappeared for 1s, and the next
face was cued for 1 s. This process repeated once more
so that all three faces in the image were rated. On portrait
trials, the cropped single-person image appeared for 2 s,
disappeared, and then subjects made their rating.
Method
Results
For each experiment, we found 100 group photographs
and cropped them to frame the faces of three people of
the same gender. We then cropped each individual face
to create three portrait images from each group photo. In
both experiments, subjects rated the 300 unique faces
twice, once in the group photo and once in an isolated
portrait. Ratings were made by moving a mouse to set a
marker on a continuous scale from unattractive to attractive (the rating scale and example stimuli are shown in
Fig. 1). The order of images and whether a face appeared
first in a group or as a portrait was random.
In our analysis, we aimed to measure the cheerleader
effect, the advantage in perceived attractiveness granted
a face when it is seen in a group rather than alone, while
factoring out the variation in how individual subjects
used our rating scale and variations in how attractive
they found the different faces to be. To factor out individual differences in rating-scale use, we converted the
raw rating given by a subject for each image in each
condition (group and portrait) into a within-subjects z
score by subtracting the mean rating and dividing the
result by the standard deviation of the 600 ratings made
by the subject. To factor out the effect of the attractiveness of specific faces, we then subtracted each subject’s
standardized rating of a face presented as a portrait from
his or her standardized rating of that same face presented
in a group. The resulting difference in z scores corresponded to the number of standard deviations higher
that a given image was rated in a group than when isolated in a portrait. Using these difference scores, we
assessed the average cheerleader-effect size (z-score difference) for each subject, as well as the average effect
size across subjects (Fig. 2).
Although there was considerable between-subjects
variation in effect sizes, subjects on average rated female
faces in a group as being 5.5% of a standard deviation
more attractive than those same faces in isolation
(Experiment 1), t(24) = 2.53, p = .018. This cheerleader
effect also held (with surprising consistency in effect
size) for male faces: There was an average advantage of
5.6% of a standard deviation for faces in a group
(Experiment 2), t(17) = 2.52, p = .022.
Experiments 1 and 2
Unattractive …………………………. Attractive
Unattractive …………………………. Attractive
Fig. 1. Rating scale and example stimuli used in Experiments 1 and
3. Subjects rated the attractiveness of 300 faces twice, once in a group
photo (top; the arrow indicated which face was to be rated) and once
in an isolated portrait (bottom). Attractiveness was rated using a mouse
to set a marker along a continuous scale. Stimuli were presented in
color in the actual experiments.
Experiment 3
In Experiments 1 and 2, each face in the group condition
was presented uncued three times for 1 s each (a total of
3 s) and presented cued for 1 s, which suggests that any
one face was on average attended for 2 s total. Thus,
average time spent attending to any one face in the group
condition was equivalent to the 2-s presentation in the
portrait condition. However, any one trial of the group
Walker, Vul
232
0.3
0.0
0.1
0.2
Female Faces
−0.2 −0.1
Effect Size
(z-Scored Difference Between
Portrait and Group Ratings)
Experiment 1 (N = 25)
Subject (Ordered by Effect Strength)
Pooled
0.3
0.0
0.1
0.2
Male Faces
−0.2 −0.1
Effect Size
(z-Scored Difference Between
Portrait and Group Ratings)
Experiment 2 (N = 18)
Subject (Ordered by Effect Strength)
Pooled
−0.2 −0.1 0.0 0.1 0.2 0.3
Effect Size
(z-Scored Difference Between
Portrait and Group Ratings)
Experiment 3 (N = 20)
Female Faces
Subject (Ordered by Effect Strength)
Pooled
Fig. 2. Results for Experiments 1, 2, and 3: standardized size of the cheerleader effect for ratings of faces,
separately for individual subjects (left panels) and pooled across subjects (right panels). To calculate effect
sizes, we converted the raw rating given by a subject for each image in each condition (group and portrait)
into a within-subjects z score by subtracting the mean rating and dividing the result by the standard deviation of the 600 ratings made by the subject. For each subject, we then subtracted this z score for the faces
in the portrait condition from the z score for the faces in the group condition. This difference yielded each
subject’s effect size: the number of standard deviations higher that a given face was rated when seen in
a group than when seen isolated in a portrait. Error bars for individual data show ±1 SEM, and error bars
for pooled data show 95% confidence intervals.
condition consisted of an uncued group of three faces for
1 s, and a cued face for 1 s, which meant that 1.33 s was
spent attending to that face. In this sense, the expected
time spent attending to a face in one group trial was
shorter than in a portrait trial. It is plausible that this difference drove the effect in Experiments 1 and 2 because
faces shown for shorter durations are rated as more
attractive than faces shown for longer durations (Willis &
Todorov, 2006). Although Willis and Todorov found an
advantage of shorter duration only for presentations
briefer than 500 ms (and ours were all longer than 1 s),
we wanted to replicate our results from Experiments 1
Hierarchical Encoding of Individuals
and 2 by equating the presentation time of one portrait
trial to one presentation in the group trial. We did so in
Experiment 3 by presenting the portrait images for just
1.33 s (otherwise, the design, stimuli, and method of data
analysis were the same as in Experiment 1). With this
modified timing, we replicated the cheerleader effect
from Experiments 1 and 2: When the presentation duration of portrait images was shortened, faces were rated
6.8% of a standard deviation more attractive when presented in a group than when presented alone, t(19) =
2.50, p = .022 (Fig. 2).
Experiment 4
In Experiments 1 through 3, all of the faces had originally
been photographed together in a real-life social context.
Perhaps group images were rated as more attractive than
single images not because of ensemble coding of the
group but because the coherent context disambiguated
facial expressions or other image idiosyncrasies (just as
videos of an individual are rated as more flattering than
the static photos that comprise them; Post, Haberman,
Iwaki, & Whitney, 2012). In Experiment 4, we sought to
rule out this class of explanations by presenting an array
composed of multiple portrait faces that had been photographed separately. In addition to this control, we also
tested for effects of group size: Increasing group size
should yield a more precise average face that should not
only be rated as more attractive (Langlois & Roggman,
1990) but should also exert a greater bias on the perceived attractiveness of individual faces (given a probabilistic combination of individuals and the ensemble; Brady
& Alvarez, 2011).
Method
We randomly chose 77 unique faces from the stimuli
used in Experiment 1. Each was presented once in each
of four conditions: alone and as part of a group of 4, 9,
and 16 other faces. The flanker faces in the group conditions were randomly chosen from the 223 remaining
faces used in Experiment 1 (target faces were never used
as flankers). Faces were presented in a square grid (1 × 1
for 1 face, 2 × 2 for 4 faces, 3 × 3 for 9, and 4 × 4 for 16;
Fig. 3a). Each grid appeared for 2 s, and then a box
appeared around the target face for 1 s. The faces then
disappeared, and subjects made a rating as in the previous experiments. In the portrait condition, the face was
presented alone in the center of the computer screen for
2 s before subjects made a rating.
Results
As in the previous experiments, we z-scored ratings
within a given subject to factor out between-subjects
233
variation in scale usage. To factor out variability in the
actual attractiveness of a given face, we subtracted the
average (across subjects and conditions) standardized
rating given to each face from each rating of that face.
This gave us a measure of the effect of each presentation
condition. Figure 3b shows the average standardized ratings in each condition across all subjects. There was a
significant effect of group size on attractiveness ratings,
F(3, 144) = 11.74, p < .001: Consistent with a cheerleader
effect, results showed that faces were rated as less attractive when presented alone than when presented in a
group of 4, t(36) = 3.23, a group of 9, t(36) = 4.25, or a
group of 16, t(36) = 4.0. However, attractiveness ratings
were not different for faces rated in groups of 4, 9, or 16.
These results suggest that it is not the coherent context of
group photos but rather the presence of additional faces
that drives the cheerleader effect.
Experiment 5
The influence of group membership on individual members may be greater when there is more uncertainty
about the individual elements in a scene; this is because
the average is less sensitive to the increased uncertainty
than the individual elements are. Following this logic, we
blurred the faces in Experiment 5 to see whether the
cheerleader effect would be increased when uncertainty
was increased.
Method
We randomly selected 50 group images from those used
in Experiment 1 and blurred them by convolving them
with a Gaussian filter with a standard deviation of 4 pixels. Subjects rated the three faces in each of those 50
images (150 unique faces) four times each: in unblurred
group and portrait conditions and in blurred group and
portrait conditions (example stimuli are shown in Fig.
4a). Other than the addition of the blurring factor, methods were identical to those used in Experiment 1.
Results
As in Experiment 4, we isolated the effect of condition by
z-scoring ratings within subjects and subtracting the
across-subjects average ratings for each face. Figure 4b
shows the average standardized ratings in each condition. As in our other experiments, faces were rated as
more attractive when seen in groups than when seen
alone, F(1, 152) = 9.0, p < .01, and subjects rated blurred
images as more attractive than unblurred images, F(1,
152) = 17.91, p < .001. However, although the cheerleader effect was bigger in the blurred condition than in
the unblurred condition (7.3% vs. 5.9% of a standard
deviation), the interaction between image clarity and
Walker, Vul
234
a
b
0.00
–0.05
–0.10
Average Standardized
Attractiveness Rating
0.05
Experiment 4 (N = 37)
1
4
9
16
Group Size
Fig. 3. Example stimuli (a) and results (b) from Experiment 4. Subjects rated the attractiveness of 77
faces four times: alone and in a group of 4, 9, and 16 other faces. Each group appeared for 2 s, and then
a box appeared around the target face for 1 s. In the portrait condition, the face was presented alone in
the center of the computer screen for 2 s. The faces then disappeared, and subjects made a rating as in
the previous experiments. Stimuli were presented in color in the actual experiment. The graph shows the
average standardized attractiveness ratings for each group size. To calculate attractiveness ratings, we first
obtained within-subjects z scores as in the previous experiments. For each subject, we then subtracted
the average (across subjects and conditions) z score given to each face from each rating of that face. Error
bars show 95% confidence intervals.
presentation condition was not significant F(1, 152) =
0.106, p = .75.
General Discussion
In the five experiments reported here, we found evidence
consistent with the cheerleader effect: Both female faces
(Experiment 1) and male faces (Experiment 2) in a group
appeared more attractive than those same faces seen
alone.1 This effect seems robust to presentation timing
(Experiment 3), to whether groups are created from natural photos or are synthetically created (Experiment 4),
and to image manipulations such as blurring (Experiment
5). We propose that this effect arises from the fact that the
visual system represents objects as an ensemble (Ariely,
2001), individual objects are biased toward the ensemble
average (Brady & Alvarez, 2011), and average faces are
perceived to be more attractive than faces in isolation
(Langlois & Roggman, 1990). Together, these phenomena
should cause faces in a group to appear more like the
Hierarchical Encoding of Individuals
235
a
in groups were rated as more attractive than the same
people alone. Thus, having a few wingmen—or wingwomen—may indeed be a good dating strategy, particularly if their facial features complement, and average out,
one’s unattractive idiosyncrasies.
Author Contributions
b
Both authors contributed to the design of the experiments.
The experiment was programmed by D. Walker. E. Vul and
D. Walker analyzed and interpreted the data. D. Walker and
E. Vul drafted the manuscript. Both authors approved the final
version of the manuscript for submission.
Portrait Condition, Blur
Group Condition, Blur
Portrait Condition, No Blur
Group Condition, No Blur
Declaration of Conflicting Interests
Average Standardized
Attractiveness Rating
Experiment 5 (N = 39)
0.10
The authors declared that they had no conflicts of interest with
respect to their authorship or the publication of this article.
Note
0.05
–0.05
1. Critically, the same face when seen in a group of different faces is rated as more attractive than when seen alone.
However, Post et al. (2012) found no such effect for a face presented in an arrays of the same face.
–0.10
References
0.00
Blur
No Blur
Image Clarity
Fig. 4. Example stimuli (a) and results (b) from Experiment 5. Subjects
rated the attractiveness of 50 faces four times each: in blurred group
and portrait conditions (shown here) and in unblurred group and portrait conditions. Stimuli were presented and ratings were made as in
Experiment 1. Stimuli were presented in color in the actual experiment.
The graph shows the average standardized attractiveness ratings as a
function of image clarity and presentation condition (group or portrait).
Attractiveness ratings were calculated as in Experiment 4. Error bars
show 95% confidence intervals.
group average than when presented alone, and that
group average should tend to be more attractive than the
individual faces, on average. However, some of our
results should give readers pause in accepting our interpretation: We predicted that increasing group size
(Experiment 4) or decreasing image quality (Experiment
5) should increase the bias of individuals to the group
average and would thus increase the cheerleader effect,
but we found no evidence of these effects. Despite this
caveat about our interpretation, the cheerleader effect
was robust: Across a wide range of settings, people
Ariely, D. (2001). Seeing sets: Representation by statistical properties. Psychological Science, 12, 157–162.
Brady, T. F., & Alvarez, G. A. (2011). Hierarchical encoding in
visual working memory: Ensemble statistics bias memory
for individual items. Psychological Science, 22, 384–392.
Chong, S. C., & Treisman, A. (2003). Representation of statistical
properties. Vision Research, 43, 393–404.
Haberman, J., & Whitney, D. (2009). Seeing the mean: Ensemble
coding for sets of faces. Journal of Experimental Psychology:
Human Perception and Performance, 35, 718–734.
Langlois, J. H., & Roggman, L. A. (1990). Attractive faces are
only average. Psychological Science, 1, 115–121.
Parks, L., Lund, J., Angelucci, A., Solomon, J. A., & Morgan,
M. (2001). Compulsory averaging of crowded orientation
signals in human vision. Nature Neuroscience, 4, 739–
744.
Post, R. B., Haberman, J., Iwaki, L., & Whitney, D. (2012). The
frozen face effect: Why static photographs may not do you
justice. Frontiers in Psychology, 3, 22. Retrieved from http://
www.frontiersin.org/Journal/10.3389/syg.2012.00022/full
Rashid, R. (Writer), & Fryman, P. (Director). (2008). Not a father’s
day [Television series episode]. In C. Bays & C. Thomas
(Creators), How I met your mother. New York, NY: CBS.
Willis, J., & Todorov, A. (2006). First impressions: Making up
your mind after a 100-ms exposure to a face. Psychological
Science, 17, 592–598.
Writing tips for this assignment
After providing feedback on many drafts, Josh, Eric and I have noticed a few common issues. Please see below for general
pointers for improving your paper:
•
Cut down on the summary/description portion (this part should really only be about 1/3 of paper) and increase the
amount of critical response (should be about 2/3) of paper.
•
The summary/description should focus on (1) the theoretical motivation for the study (why did they do the
experiment in the first place?), (2) what did the experimenters do to test this (emphasizing the experimental details
that are relevant to the theoretical question. You do not need to go into extensive detail about the methods, it is
better to only describe the essential parts, not the irrelevant details that are not critical to experiment or to the
conclusions in the paper, making sure to include what was they manipulated and measured. (3 Include the relevant
results they found (in words, not by reciting the statistics). Often irrelevant findings are reported, you should not
take time on these. 3) discuss how the authors interpreted those results. DEMONSTRATING WHAT IS
RELEVANT TO THE THEORETICAL CLAIM IS A CRUCIAL PART OF THE ASSIGNMENT
•
The criticism should be your thoughts on the research. The prompt is more of a guide than a checklist. For all of
your opinions, make sure to justify them with reasons. This is where you would include details about article that are
beyond the gist description. For example, if you wanted to comment on the number of stimuli used, you would
include that detail about the stimuli as a means of justifying your opinion on it. Otherwise, you don’t need to include
details like that. Additionally, since we know you are commenting on a peer reviewed paper in a high impact
journal we assume that things, such as the writing, were generally done well. You don't need to comment on the
writing style unless you found all or part of it particularly well or poorly written. In either case, you should explain
the reasons why you felt that.
•
Your criticisms don't all have to be negative. Maybe you found everything clear and well done (these papers have
all been reviewed by professional scientists who are experts in the topic and they have been heavily edited prior to
publication). If that is the case, mention something about what you liked or found particularly compelling. Make
sure to explain your reasons. Additionally, you can comment on where you think the research should go next and
maybe a bit about how that could be tested. Science is an incremental process. Research often generates follow
up questions. Researchers often use the same experimental paradigm and only change one little thing in order to
test their new questions. Doing that makes it easier to come up with a way to test things and also makes it easier
for other scientists to interpret what everything should mean because it is kind of like using a template.
•
All else being equal: SHORTER IS BETTER. If two papers have the same critical content, the shorter on would
receive the higher grade.
•
AND try to stay in the green box!
Critical Article Response Paper Guidelines
Page Limit: 2-4 pages, double-spaced
The purpose of these response papers is to hone your ability to critically evaluate primary source material in cognitive science.
The format of the response paper is generally open, but the paper needs to include some summary of the article and some
analysis of the methods and results. You may use the following basic structure and questions to guide your writing, but do not
simply list bullet-points with answers. A critical response paper is an argument itself, taking a position ultimately on whether a
particular article is contributing to the particular field. Show both that you understand the points made in the article and that you
have thought critically about them.
FEEDBACK: You are highly encouraged to email a draft of your paper in .doc or .docx format to your assigned TA by the draft
deadline of Friday May 3rd at 12:30 pm to receive feedback. You will not get credit directly for turning in a draft, but the feedback
should help improve your final paper grade. The sooner in the quarter that you send your paper, the more detailed feedback we
will be able to provide. We will NOT provide feedback past the draft deadline, so start your paper early! Please put “101A PAPER
DRAFT” as the subject line of your email.
If you choose paper:
•
#1 or #2 send your draft to Drew (dehoffma@ucsd.edu)
•
#3 or #4 send you draft to Eric (e1morgan@eng.ucsd.edu)
•
#5 or #6 send you draft to Josh (jdd001@ucsd.edu)
SUMMARY QUESTIONS:
1) What was the purpose of this article?
a. What was the main question(s) the researchers were trying to answer?
b. Were there any important prior findings that this research is following up on or going against?
2) What are the hypotheses?
a. What did the researchers expect to find, in general, and why?
3) What did the researchers do? (VERY BRIEFLY – EXPLAIN SIMPLY IN YOUR OWN WORDS)
a. What kind of task did the subjects do?
b. What was the main manipulation the researchers used in the study?
c. What were the different conditions?
d. What did the researchers expect to find, specifically? How did they expect conditions to differ?
4) What did the researchers find?
a. What were the main results?
b. What comparisons did they make? Which ones were statistically different and which weren’t?
c. What do the important figure(s) in the paper mean?
d. What do the results mean? If two conditions were different, why?
5) What are the conclusions?
a. What are the answers to the questions the researchers were trying to ask (see question 1, above) and how do the
results support this?
b. What did we learn from this study?
CRITICAL REVIEW QUESTIONS:
1) What assumptions do the researchers bring to the present study?
a. Are these assumptions valid? Are they well supported?
b. Do the results or interpretation of results rely on these assumptions?
2) Is the manipulation a valid (good) test of the hypothesis?
a. Does the manipulation really test the phenomenon that the authors claim to be testing?
b. Can the effects be explained by a cause unrelated to the phenomenon that the authors claim to be testing?
4) Are there issues with the methods that might affect the outcome of the experiment? (e.g., were the subjects aware of the
manipulation? Does this matter?)
5) Are the conclusions that the authors draw justified by their results?
a. Do they engage in speculation that is unsupported by the data?
b. Can the results be generalized to other experimental settings/populations? If not, do the authors admit that?
6) Is the paper well-organized and easy to follow?
a. Is the paper well written, does it flow properly?
b. Are conditions and phenomena explained clearly and used consistently throughout the paper?
c. Are ideas and questions posed in the introduction then addressed in the discussion?
7) What are possible future directions?
a. Are there any follow-up experiments that would strengthen the case the authors are trying to make? Any
experiments that would contradict it?
b. What are other possible future directions for this line of research? Are there any other populations or languages that
could be tested (for a good theoretical reason)? Could this methodology, analysis or logic be used to study something
else?
Purchase answer to see full
attachment