Contents lists available at ScienceDirectComputers in Huma.docx

Contents lists available at ScienceDirect
Computers in Human Behavior
journal homepage: www.elsevier.com/locate/comphumbeh
Full length article
Deindividuation effects on normative and informational social
influence
within computer-mediated-communication
Serena Coppolino Perfumia,b,∗ , Franco Bagnolic, Corrado
Caudekd, Andrea Guazzinie
a Department of Sociology, Stockholm University, S-106 91,
Stockholm, Sweden
bDepartment of Educational Sciences and Psychology,
University of Florence, 50135, Florence, Italy
c Department of Physics and Astronomy and Center for the
Study of Complex Dynamics (CSDC), University of Florence,
50019 Sesto Fiorentino, also INFN sec, Florence,
Italy
dDepartment of Neuroscience, Psychology, Drug Research and
Children's Health (NEUROFARBA) – sect. Psychology,
University of Florence, 50135, Florence, Italy
e Department of Educational Sciences and Psychology and
Center for the Study of Complex Dynamics (CSDC), University
of Florence, 50135, Florence, Italy
A R T I C L E I N F O

Keywords:
Social influence
Conformity
Computer-mediated-communication
Anonymity
Deindividuation
A B S T R A C T
Research on social influence shows that different patterns take
place when this phenomenon happens within
computer-mediated-communication (CMC), if compared to face-
to-face interaction. Informational social influ-
ence can still easily take place also by means of CMC, however
normative influence seems to be more affected by
the environmental characteristics. Different authors have
theorized that deindividuation nullifies the effects of
normative influence, but the Social Identity Model of
Deindividuation Effects theorizes that users will conform
even when deindividuated, but only if social identity is made
salient.
The two typologies of social influence have never been studied
in comparison, therefore in our work, we
decided to create an online experiment to observe how the same
variables affect them, and in particular how
deindividuation works in both cases. The 181 experimental
subjects that took part, performed 3 tasks: one
aiming to elicit normative influence, and two semantic tasks
created to test informational influence. Entropy has
been used as a mathematical assessment of information
availability.
Our results show that normative influence becomes almost
ineffective within CMC (1.4% of conformity) when
subjects are deindividuated.

Informational influence is generally more effective than
normative influence within CMC (15–29% of con-
formity), but similarly to normative influence, it is inhibited by
deindividuation.
1. Introduction
With the diffusion of social networking platforms, the social
and
information seeking-related human behaviors have been affected
by the
“new” environment. Information seeking increasingly takes
place on
social media platforms, relying on what a users' contacts and
followed
pages share (Zubiaga, Liakata, Procter, Hoi, & Tolmie, 2016).
Because of this filtering and selection, the users' knowledge-
building
process could be severely biased and polarized.
For example, a study shows that 72% of participants (college
stu-
dents) trusted links sent by friends, even if they contained
phishing
attempts (Jagatic, Johnson, Jakobsson, & Menczer, 2007).
The recent debate on fake news, highlighted the potential link
be-
tween the increase in their spread, and the structure of social
networks
as well as their embedded algorithms, which turned these
environments
into “echo chambers”, in which users are selectively exposed to

information, and tend to filter the information in order to
reinforce
their positions (confirmation bias), rather than to find
alternatives (Del
Vicario et al., 2016).
These factors highlight the importance of studying the effects of
social influence within computer-mediated-communication, in
order to
understand which environmental factors can enhance its effects.
Social norms exist also in online environments, but the users'
per-
ception of them can be different according to the platform, to
anon-
ymity and the social ties among contacts. Therefore, compliance
to
social norms can emerge in different ways, than those
observable in
face-to-face interaction.
Also, information-seeking behavior can be affected by online
en-
vironments: on one side we observe its interrelation with social
norms,
especially when it takes place on social media platforms, and
users
gather information on the basis of what they read on their
personal
newsfeed. However, we also observe how users can rely on
opinions
https://doi.org/10.1016/j.chb.2018.11.017
Received 29 March 2018; Received in revised form 9 October
2018; Accepted 7 November 2018

∗ Corresponding author. Department of Sociology, Stockholm
University, S-106 91, Stockholm, Sweden.
E-mail address: [email protected] (S. Coppolino Perfumi).
Computers in Human Behavior 92 (2019) 230–237
Available online 13 November 2018
0747-5632/ © 2018 Elsevier Ltd. All rights reserved.
T
http://www.sciencedirect.com/science/journal/07475632
https://www.elsevier.com/locate/comphumbeh
mailto:[email protected]
http://crossmark.crossref.org/dialog/?doi=10.1016/j.chb.2018.11
.017&domain=pdf
expressed by unknown actors, as it happens on platforms like
TripAdvisor.
The present study, using online experiments, aims to separate
norms-oriented social influence from information-oriented
social in-
fluence, in order to observe which elements and environmental
factors
have an effect on both typologies and which are peculiar for
each.
1.1. Theoretical framework
A major understanding on the functioning of social influence
came

about thanks to the pioneering works of Sherif (1937) and then
Asch
(1951, 1955, 1956). The authors studied how the physical
presence of
other people can lead experimental subjects to conform their
judgment
to the one of the others. They used two different types of tasks:
while in
Asch conformity experiments, guessing the correct answer could
be
straightforward (Asch, 1955, 1956; Asch & Guetzkow, 1951),
Sherif
used the autokinetic effect, so a more ambiguous task, to test
the effects
of social influence (Sherif, 1937). From these experiments, two
typol-
ogies of social influence have been identified, called
“normative” when
people conform in order to satisfy a need to belong and comply
to social
norms, as observed in Asch's experiments, and “informational”
when
the subjects lack on information in order to perform a task, as
observed
in the autokinetic experiment (Deutsch & Gerard, 1955).
According to
this theorization proposed by Deutsch and Gerard (1955), we
can say
that we are able to observe normative social influence in Asch's
con-
formity experiments, because the task is relatively easy and the
sub-
jects, when interviewed after taking part to the experiment
stated that
they were able to spot the correct answer, but conform in order
not to

break the social norms and be group outsiders. Instead, given
that the
task presented in the autokinetic experiment is more ambiguous,
as it is
based on a visual illusion, in this case we can say that subjects
conform
because they are unsure on how to proceed.
While, as observed in these classical studies, to elicit
conformity in
face-to-face situations, the physical presence of other people
and being
exposed to their judgment can be enough, things go differently
when
people interact online, especially for normative social influence.
Indeed, it is still unclear which elements can have the power to
lead
people to conform during computer-mediated-communication.
Deindividuation, namely the diminished perception of one's per-
sonal traits (Zimbardo, 1969), has been identified as a potential
key
element in the discourse on normative influence.
The original deindividuation model was proposed by Zimbardo
in
1969, and the author identified a series of variables that
according to
him can lead to a deindividuation state. The variables
considered by
Zimbardo are for example anonymity, arousal, sensory overload,
novel
or unstructured situations, involvement in the act, and the use of
al-
tering substances (Zimbardo, 1969). Several other authors

suggest that
if people interact while being in a deindividuation state,
normative
social influence can disappear (Deutsch & Gerard, 1955;
Latané, 1981;
Lott & Lott, 1965; Short, Williams, & Christie, 1976). This
happens
because there is not the possibility to identify the interlocutors,
due to a
lack of actual or perceived proximity, and consequently,
deindividua-
tion should lighten the pressure to act according to social norms
(Latané, 1981).
Furthermore, a study which tested antinormative behavior by
counterposing deindividuation to the presence of an explicit
aggressive
social norm, showed that subjects were actually more aggressive
when
deindividuated, rather than when exposed to the explicit norm,
so in
this case, deindividuation resulted to be more powerful in
leading to
antinormative behavior (Mann, Newton, & Innes, 1982).
A significant advancement in explaining the functioning of
norma-
tive social influence in online environments is represented by
the
contribution provided by the Social Identity Model of
Deindividuation
Effects (SIDE Model), that takes the concept of deindividuation
and
expands it, explaining its link and implications on social
influence in
online environments (Spears, Postmes, Lea, & Wolbert, 2002).

The authors theorize that deindividuation is indeed likely to
occur
in online environments, but it can become a powerful tool to
trigger
conformity: given that while deindividuated, subjects have a
dimin-
ished perception of their personal traits, if the group the
subjects are
interacting with is made salient, then the subjects will be more
likely to
conform (Spears, Postmes, & Lea, 2018).
This happens because combining a lack of relevance of one's
per-
sonality with an enhancement of the importance of the
interlocutors,
will lead the subjects to identify at the group level, and
consequently to
comply to the social norms. The experimental results seem to
confirm
the predictions presented by the SIDE Model (Lee, 2004;
Postmes,
Spears, Sakhel, & De Groot, 2001), but it is not clear what
happens
when users are deindividuated but the group saliency is not
enhanced.
On the matter of informational influence during computer-medi-
ated-communication instead, studies have focused on different
aspects.
As aforementioned, a visible example of informational influence
in
online environments is represented by users making choices on
the

basis of reviews or ratings provided by other unknown users
while
using platforms such as Tripadvisor, Uber or Airbnb (Liu &
Zhang,
2010), but other examples show that it can take place easily also
in
other ways.
A study conducted by Rosander and Eriksson (2012), shows that
users facing a general knowledge quiz in which they were
exposed to
histograms showing the distribution of the answers provided by
other
unknown users, conformed in high percentages (52%).
While many studies on online consumers behavior focused on
fac-
tors such as the perceived importance of feedback (Liu &
Zhang, 2010)
on informational influence, or on the conjunct effect of
informational
and normative influence on behavior when subjects interact
without
personal contact (LaTour & Manrai, 1989), no study tried to
isolate it,
and point out the environmental factors that could be able to
enhance
or diminish the compliance of users in this case. Furthermore,
no study
tested the effects of deindividuation on informational influence.
In order to test and fulfill the predictions developed based on
the
literature, we developed an experimental framework aiming to
study
separately the two typologies of social influence during

computer-
mediated-communication.
On one side, we reduced group saliency to test how
deindividuation
works on both typologies of social influence and controlled the
possible
interactions between some psychological dimensions and the
operative
variables.
On the other side, we calculated the items entropy to test if task
ambiguity increases informational-based compliance. The
environ-
mental factors that we decided to manipulate and study in
relation to
both typologies of social influence are anonymity and physical
isola-
tion, as their combination can trigger deindividuation.
1.2. Overview and predictions
To test online normative influence, we replicated Asch's
conformity
experiment (Asch, 1955, 1956; Asch & Guetzkow, 1951) on a
web-
based platform, while to test online informational influence we
created
two linguistic tasks of increasing ambiguity, designed adopting
the
same structure of the “classical” Asch's items. Task ambiguity
was
measured by calculating the items' entropy, and in this way, we
were
able to assess the subjects' lack of information. The diversity of
the

tasks, allowed us to measure the interaction between anonymity,
phy-
sical isolation, and degree of ambiguity, in relation to the
behavior of
the experimental subjects. Considering the literature, we could
for-
mulate the following predictions:
• H1) Diminished effectiveness of normative influence due to
the
combination of a deindividuation state given by anonymity and
physical isolation, and minimum levels of group saliency, as
theo-
rized by several authors (Deutsch & Gerard, 1955; Latané,
1981;
Lott & Lott, 1965; Short et al., 1976) and hypothesized by the
SIDE
S. Coppolino Perfumi et al. Computers in Human Behavior 92
(2019) 230–237
231
Model (Postmes et al., 2001).
• H2) There is no specific evidence to build on, on the potential
re-
lationship between deindividuation and informational influence
(if
separated by normative influence), but we expect it to have the
same inhibitory effect it has on normative influence (Lee,
2007). The
effect of the anonymity and physical isolation variables alone
will

also be controlled.
• H3) We expect a positive correlation between conformity and
task
ambiguity, given that with more ambiguous items the subjects
will
possess less information on how to handle the task, and might
rely
on other people's judgment (Cialdini & Trost, 1998; Rosander &
Eriksson, 2012).
We also controlled the interaction of personality and
psychological
traits on conformity. In order to make sure that the analyzed
effects
were relatable to the manipulated features and not to particular
psy-
chological traits, we measured the psychological dimensions
that ac-
cording to literature, result related to some extent to
conformity. Only a
few studies analyzed the relation between conformity and
personality
traits, suggesting some interesting connections between social
con-
formity and Emotional Stability, Agreeableness and Closeness
(DeYoung, Peterson, & Higgins, 2002). So we expect that:
• H4) Factors as Neuroticism, Surgency (a trait linked to
Extraversion)
and Closeness will have an inhibitory effect on conformity
• H5) Agreeableness will increase the tendency to yield to
majority
pressure.

However, it is necessary to consider the contextual
peculiarities,
illustrated by both the deindividuation explanation provided by
lit-
erature (Latané, 1981; Postmes et al., 2001; Tsikerdekis, 2013),
and the
theoretical framework supporting the idea that real and virtual
iden-
tities are not consistent (Kim & Sherman, 2007), that highlight
the lack
of saliency of personality traits in anonymity conditions, which
may
predict a:
• H6) weak general effect of personality traits, especially if
measured
with scales calibrated to assess “real life” traits.
Finally, since the experiment was conducted both in group and
single (i.e., physical isolation) conditions, according to the
existing
literature that illustrates how the mere presence of other people
can
affect an individual's performance (Markus, 1978), we expect:
• H7) Physical isolation and group conditions to produce
significantly
different behavioral outcomes.
2. Method
In order to analyze the variables and dimensions of interests,
the
experiment was structured as follows. To analyze the anonymity
effect
on conformity, we manipulated anonymity levels making the

subjects
perform the experiment in either full or partial anonymity (i.e.,
anon-
ymity vs nonymity). In the full anonymity condition, the
participants
were distinguished from the other group members by a number
re-
presenting their response order, while in the nonymity condition
they
had to provide their name and surname and could see the
others'. To
test the physical isolation variable, we made the subjects
perform the
experiment alone (physical isolation) or with other experimental
sub-
jects in the same room (group condition). In the group
condition, the
subjects were not interacting with each other but with other
agents: the
group of confederates in the platform was composed by
programmed
bots that in some trials provided the correct answer, and in
some other
the wrong one. In order to induce normative influence, we
adapted
Asch's original line-judgment task for an online support and
adminis-
tered it as first task (Asch, 1956). We also maintained the
original
pattern in making the confederates provide wrong and correct
answers.
Adopting the structure of the classic Asch's experiment, we
designed
two brand new tasks, respectively labeled “cultural” and
“appercep-

tive”, in order to manipulate ambiguity both between tasks and
among
the single items. The cultural task consisted in a target word
(primer)
associated with three possible answer options more or less
semantically
related (targets). The apperceptive task, instead, consisted in
three
different combinations of real and invented words (i.e.,
condition A:
real primer word vs invented words as answer option; condition
B:
invented primer word vs real words as answer option; condition
C:
invented prime word vs invented words as answer option). In
order to
measure the informational influence effects, we first estimated
the
items' entropy, defined as an inverse function of the probability
to
observe a certain association between the prime and the target.
The
entropy of each item, measured by means of a preliminary
survey ad-
ministered to an ad hoc sample, represents a quantitative
estimation of
the “lack degree” of information contained by each item. A
study on the
voting tendencies related to conformity, hypothesized this
factor to be
inversely related to entropy, since the more predictable the
behavior is
(i.e., low entropy), the higher is the tendency to conform
(Coleman,
2004). Nevertheless, such result describes the behavior of a
subject

under a direct majority pressure. In our study we exposed the
experi-
mental subjects to a constant majority pressure always towards
a more
entropic answer. In this way, the cultural and apperceptive
tasks, in-
vestigate the relation between entropy of the choice, and the
informa-
tional influence dynamics.
2.1. Sampling and participants
The research was conducted in accordance with the guidelines
for
the ethical treatment of human participants of the Italian
Psychological
Association (AIP). The participants were recruited with a
snowball
sampling strategy. Most of them were undergraduate students
from an
Italian university. All participants gave their consent to
participate and
had the possibility to withdraw from the experiment at any time.
The
participants were 181 (76.8% identifying as female) and all of
them
were over 18 years of age (age: M=22.11, S D=4.44). All the
par-
ticipants filled out the survey and none of them withdrew during
the
experiment. In order to obtain a robust approximation of the
optimal
sample size, disregarding the debate about the standard sample
size
estimation for GLMM (Bolker et al., 2009), we conducted a
power

analysis by reducing the hypotheses to the case of two samples'
mean
comparison under a 2-sided equality hypothesis (eqs. (1)–(3))
(Chow,
Shao, Wang, & Lokhnygina, 2017). The results are reported in
Table 1.
⎜ ⎟= ⎛
⎝
+ ⎞
⎠
⎛
⎝
+
−
⎞
⎠
− −
n
K
σ
Z Z
μ μ
1 1b
β
a b

1 1σ2
(1)
with
− = − + − −− −( ) ( )β ϕ Z Z ϕ Z Z1 α α1 2 1 2 (2)
and
Table 1
Sample size estimation using the variable Conformity as
dependent measure, to
compare 2 means from 2 samples with 2 sided equality
hypothesis, requiring a
Power (1− β) of 80%, and a Type I Error confidence level (α) of
5%.
Dimension Mean test
(SD)
Control mean
(SD)
K Na/Nb Sample size
Required Available
Anonymity 18%
(11%)
15% (7%) 1.06 86 88
Physical Isolation 18%
(10%)

14% (7%) 0.5 106 120
(2019) 230–237
232
=
−
+
Z
μ μ
σ
A B
n n
1 1
a b (3)
where, =K nn
a
b
, σ is the standard deviation, Φ is the standard Normal
distribution function, −ϕ 1 is the standard Normal quantile
function, α is
Type I error, and β is Type II error, meaning 1− β is power.
This
analysis revealed that approximately 180 participants would be
needed
to achieve 80% power (1− β) at a 0.05 α level (α=0.05).

The exclusion criteria regarded any type of psychiatric
diagnosis
and a lack of fluency in the Italian language, since the cultural
and
apperceptive tasks were of semantic nature. Out of 181 subjects,
61
participants performed the experiment in the group condition
(groups
of six, seven or eight people), while 120 performed the
experiment in
the physical isolation condition (Table 2).
The participants were also balanced according to the anonymity
condition and 93 performed the experiment in partial anonymity
(i.e.,
“nonymity”), while 88 in full anonymity (Table 3).
Since the recruitment method consisted in a snowball sampling,
we
have not been able to balance the subjects according to their
genders
and as consequence, the majority of them identified as females
(76.8%,
versus 23.2% identifying as males). This factor has been
controlled
during the data analysis.
2.2. Materials and apparatus
At first, we administered a series of scales in order to determine
psychological traits and states. The scales have been chosen
according
to the dimension they aim to measure and its relation to social
influ-
ence. Studies have investigated the link between conformity and

Big-
Five traits, showing relations between some traits and
conformity
(DeYoung et al., 2002). Anxiety has been identified as a
potential
predictor for conformity, while self-esteem and self-efficacy
predict the
opposite tendency, namely nonconformity (Deutsch & Gerard,
1955).
Finally, according to the literature, a high sense of community
results to
be positively related to conformity (McMillan & Chavis, 1986).
For
these reasons, we chose scales that measure the aforementioned
di-
mensions:
• Five Factor Adjective Short Test (5-FasT) (Giannini,
Pannocchia,
Grotto, & Gori, 2012), a short version of the Big Five aiming to
asses
personality traits. It comprises 26 dichotomous items (true-
false).
All the subscales present a good reliability (Neuroticism=0.78;
Surgency=0.73; Agreeableness= 0.71; Closeness= 0.71; Con-
scientiousness= 0.70)
• The State-Trait Anxiety Inventory for Adults (Spielberger &
Gorsuch,
1983), a self-reporting 20-item measure on state and trait
anxiety.
The items are on a 4-point Likert scale whose range goes from 1
(not
at all) to 4 (very much so). The scale appears to have an
excellent
test-retest reliability (r=0.88) (Grös, Antony, Simms, &

McCabe,
2007).
• The Multidimensional Sense Of Community Scale, a 26-item
scale on
which each item is on a 4-point Likert scale (4-strongly agree to
1-
strongly disagree). The scale results to have good reliability and
good construct validity (Cronbach Alpha's from 0.61 to 0.80)
(Prezza, Pacilli, Barbaranelli, & Zampatti, 2009)
• The Rosenberg's Self-Esteem Scale, a 10-item scale on which
each
item is on a 4-point Likert scale (4-strongly agree to 1-strongly
disagree). The scale has an excellent internal consistency
(coeffi-
cient of reproducibility of .92), and stability (0.85 and 0.88 on a
2
weeks test-retest) (Rosenberg, 1965).
• The General Self-Efficacy Scale (Sibilia, Schwarzer, &
Jerusalem,
1995), a 10-item scale with items on a 4-point Likert scale (1-
not at
all true, 4-exactly true). The scale has a good reliability with
Cronbach Alphas' ranging from 0.76 to 0.90 (Schwarzer &
Jerusalem, 2010).
For what concerns the experiment, besides resizing Asch's
visual
task (Asch, 1956) for online supports, we created the cultural
and ap-
perceptive tasks, of semantic nature: examples of cultural and
apper-
ceptive tasks items are in Fig. 1.

Within the two tasks, we calculated the item's entropy, in order
to
mathematically assess the ambiguity of the stimuli. We
presented the
cultural items to a sample of 71 subjects and the apperceptive to
79
subjects, collected their answers and calculated frequencies and
per-
centage. On the basis of the latter, we proceeded to calculate the
en-
tropy for items i, using an equation (4) with pkj =(Σni=1 rki )/n,
and “n”
indicating the respondents to item k.
∑= −
=
E p logpk
j
j
k
j
k
1
3
(4)
Finally, according to the median, we divided the items in high
and
low entropy (Fig. 1). For what concerns the cultural and

apperceptive
items, the correct answer was the most chosen during the pre-
test, so,
when the majority gave a unanimous incorrect answer, they
picked the
least chosen option. However, differently from Asch's task, in
some
cases we randomized the majority's choices in order to make the
in-
teraction more believable. The experiment was composed by 20
Asch-
task items, 45 cultural items and 45 apperceptive items, for a
total of
110. The experiment was performed on an online software
graphically
based on the Crutchfield apparatus (Crutchfield, 1955),
designed by us
on Google Scripts (Fig. 2).
The interface was designed to allow interaction between the ex-
perimental subject and six other confederates, for a total of
seven ac-
tors: the experimental subject was always placed in sixth
position (Asch
& Guetzkow, 1951), and the interface simulated the responses
of six
other non-existing subjects. It also provided the possibility to
record the
subjects' response times and control anonymity, displaying only
num-
bers associated with each group member in the full anonymity
condi-
tion, and asking to provide name and surname, and showing
fictional
names and surnames in the nonymity condition. The
experimental

subjects could see the answers of the other fake group members
beside
their name or identification, and the stimulus appeared only
when their
turn came. After the experiment, we administered a
questionnaire in-
vestigating the subjects' experience, using questions based on
Asch's
post-experimental interview (Asch, 1956).
2.3. Procedure
The experiment was presented as a study on visual and semantic
perception, in order to avoid biases. The group-condition
experiment
took place in a computer room, where groups of 6, 7 or 8
subjects,
performed the experiment on distantly placed computers. The
physical
isolation-condition experiment, instead, took place in a
laboratory,
where the participants were alone with a maximum of three
Table 2
Physical Isolation versus group conditions.
Condition Frequency Percentage
Physical Isolation [PI(1)] 120 66.3
Group Condition [PI(0)] 61 33.7
Total 181 100
Table 3
Anonymity versus Nonymity conditions.
Condition Frequency Percentage

Anonymity [FA (1)] 88 48.6
Nonymity [FA (0)] 93 51.4
Total 181 100
(2019) 230–237
233
experimenters. Every participant was given an ID code that
needed to
be reported in all the three experimental phases. The first phase
con-
sisted in the filling of the scales that took approximately 15min.
When
completed, the participants could start the experiment, which
took
approximately 50min to be completed. The first task was
Asch's, the
second the cultural and the third the apperceptive, and each
phase was
introduced by means of an informational page with instructions.
The
last phase consisted in the filling of the post-experimental ques-
tionnaire, and this phase lasted 10min circa. When finished, the
sub-
jects were informed on the real purposes of the study and were
told not
to divulge details on the experiment, in order to avoid potential
biases
from the other experimental subjects.
3. Results

Fig. 3 shows the different percentage of conformity in each
task. In
Asch's task, the one used to test normative influence 1,4% of
the sub-
jects conformed to the majority when it gave a clearly incorrect
answer.
Conformity percentages grow significantly in the cultural task,
with
15,2% of subjects conforming and the highest rate is registered
in the
apperceptive task, with 29,8% of conformity.
Both the cultural and the apperceptive tasks were used to test
in-
formational influence and more insights on the effects of this
type of
influence can be obtained by observing the results concerning
entropy.
Conformity increased significantly with higher entropy, thus
with more
ambiguous items (Table 4).
Since the tasks have always been presented in the same order
(Asch
first, then cultural and finally apperceptive), we conducted
some ana-
lysis in order to verify if any eventual learning mechanisms
could have
occurred and invalidated the trustworthiness of conformity data.
The
only interaction appeared between conformity and entropy but
once
controlled the entropy effect, no significant learning mechanism
ap-

peared, besides a slight negative effect of time on the cultural
task. To
analyze the relationship between conformity, physical
condition,
anonymity and personality traits, we used Generalized Linear
Mixed
Models, the size effect of which results to be 77%. From the
model,
emerged that conformity takes place differently whether
subjects are
physically isolated, anonymous or in both conditions happening
at the
same time (deindividuated). Full anonymity and physical
isolation
analyzed singularly have a positive relationship with
conformity, but if
these two variables interact (creating deindividuation), the
relationship
becomes negative (Table 4). This analysis also provided results
re-
garding the effects of personality traits, in particular,
Neuroticism,
Surgency (i.e., Extraversion), Agreeableness, Closeness, Self-
Efficacy
and State and Trait Anxiety.
The factors that result to be positively related to conformity are
Closeness, Self-Efficacy and State Anxiety. The traits that are
negatively
related to conformity, are Neuroticism, Surgency,
Agreeableness.
4. General discussion and conclusions
The results of this study could help to explain the dynamics that
can

occur in online environments, where the different available
platforms
allow the users to interact under different levels of anonymity,
and with
known and unknown people. We found an almost non-existent
effect of
normative influence when social identity is not strengthened,
with only
1.4% of the subjects conforming to Asch's task.
In our experiment, group saliency was minimal due to
anonymity,
the impossibility to communicate with the other members, and
the
absence of any type of information exchange (except fictional
name and
Fig. 1. Example of cultural and apperceptive items. In
figure are shown three different examples of the stimuli
adopted in the experiment. In the first row there are two
examples of cultural items: in the first rectangle the primer
is associated with three options, among which one is more
semantically related than the others (low entropy), the
second example present three untied options (high entropy).
In the second row we can find two types of apperceptive
stimuli with invented words both for the primer and the
answer options.
Fig. 2. Screenshot representing the interface on which the
subjects performed the experiment in the nonymity condition.
(2019) 230–237
234

surnames in the nonymity condition) concerning the group
members.
Furthermore, the subject did not engage in any type of
cooperative task
before the experiment, a method often used to enhance group
saliency
(Postmes et al., 2001).
Thus, we confirm the existing literature on deindividuation
(Postmes et al., 2001), showing that deindividuation alone is an
in-
hibitory factor for normative influence in online environments.
On the other side, when the focus is on obtaining information
and
the subjects' knowledge on a topic lacks because the task is
particularly
difficult or ambiguous, even unknown users can be considered a
reli-
able source, even when deprived of cues about their actual level
of
knowledge. In fact, from our analysis, emerged that the
strongest pre-
dictor of conformity is task ambiguity: entropy resulted to have
a sig-
nificant positive effect on conformity. In the case of the present
study,
entropy was modulated both within and in-between tasks, and
we
registered a 15.2% of conformity in the cultural task, and a
29.8% in
the apperceptive, the most ambiguous task.

These results confirm other studies (Rosander & Eriksson,
2012)
that show the effectiveness of informational influence also in
online
environments. However, new evidence emerged from the present
study,
showing that two contextual characteristics can actually affect
in a
complex way the effects of informational influence: full
anonymity,
physical isolation, as well as their interaction (i.e.,
deindividuation).
Anonymity and physical isolation taken separately have a
positive ef-
fect on conformity, confuting the “mere presence-effect”
hypothesis, at
least in this case (Markus, 1978), but if combined, thus creating
a
deindividuation state, they actually reduce conformity. In this
way, we
can say that deindividuation has an inhibitory effect not only on
nor-
mative influence, as theorized by the SIDE Model (Postmes et
al., 2001),
but also on informational influence within CMC. These results
provide
us interesting insights on the environmental and psychological
elements
that can affect information-seeking behavior in online
environments.
The large amount of information available on the Internet,
combined
with online social dynamics often lead users not to verify the
credibility
of sources, and the present study provides new insights that
show that if

users are deindividuated, their tendency to trust unknown
sources of
information is minor. This result has two potential implications,
a so-
cially-related one and an exposure-related one. The first one is
related
to the fact that such result suggests that in order to trust random
in-
formation, the underlying social dynamics, namely, the
perceived im-
portance and/or trust towards who is supporting such
information is
crucial.
As the deindividuation perspective presented by the SIDE
Model
suggests, if there is no social identification with the group
members, the
effects of social influence will reduce and according to these
results, this
could happen also when the push towards conformity is not
strictly
related to a compliance with social norms, but rather to a need
for
information.
Future research could deepen this result, for example by
focusing on
the relationship between the spread of misinformation in social
net-
works and informational influence, deepening how social
dynamics
underlie this process, to what extent they influence information
Fig. 3. Percentages of conformity in Asch, Cultural and
Apperceptive tasks and Entropy's quadratic plot.

Table 4
Generalized Linear Mixed Model. Model's Size Effects: 66%.
∗ ∗ ∗ =p < 0.001,
∗ ∗ =p < 0.01, ∗ =p < 0.05. The variables included in the model
are en-
tropy, anonymity, physical isolation, Neuroticism, Surgency,
Agreeableness,
Closeness, Self- Efficacy and state anxiety.
GLMM Best Model
Model precision Akaike∗ F Df-1 (2)
81.5% 9396.12 67.67∗ ∗ ∗ 12 (9116)
Parameter Fixed effect (F) Coefficient St. Error Student t
Entropy 672, 98∗ ∗ ∗ 8, 714 0,34 25, 94∗ ∗ ∗
Full anonymity 23, 11∗ ∗ ∗ 2, 416 0,46 5, 31∗ ∗ ∗
Physical isolation 10, 71∗ ∗ ∗ 0, 474 0,09 5, 78∗ ∗ ∗
Neuroticism 7, 38∗ ∗ −0, 027 0,01 −2, 72∗ ∗
Surgency 7, 07∗ ∗ −0, 032 0,01 −2, 66∗ ∗
Agreeableness 23, 18∗ ∗ ∗ −0, 042 0,01 −4, 81∗ ∗ ∗
Closeness 6, 79∗ ∗ 0, 022 0,01 2, 61∗ ∗
Self-efficacy 24, 09∗ ∗ ∗ 0, 046 0,01 4, 91∗ ∗ ∗
STAI-State 9, 97∗ ∗ ∗ 0, 017 0,01 3, 16∗ ∗ ∗

FA (1)∗ PI(1) 24, 94∗ ∗ ∗ −0, 574 0,12 −4, 99∗ ∗ ∗
(2019) 230–237
235
acceptance, and whether other contextual factors can affect this
pro-
cess, since this phenomenon is having a strong political and
social
impact.
The second implication is related to the subjects' feeling of
exposure:
if they perceive that there is no way to identify them, as they
are both
anonymous and physically isolated, they are more prone to
disregard
the opinions they are exposed to.
Future research could investigate, for example, whether this
hap-
pens because subjects try to provide their own judgment,
because they
engage in explicit non-conformist behavior, or because they do
not put
too much effort in completing the task.
Finally, for what concerns the effects of personality traits, the
ones
which resulted to have an inhibitory effect on conformity are
Neuroticism, Surgency (i.e., Extraversion) and Agreeableness,
in line

with the existing literature (DeYoung et al., 2002), while
subjects with
higher scores in Closeness, Self-Efficacy and State Anxiety
conformed
more.
These results however predict a small portion of the general
ten-
dency to conform, so further studies are necessary to understand
the
entity of the impact of personality traits on conformity and its
pre-
dictability.
In line with the theoretical framework, the previous result could
support the literature stressing how personality changes when
users are
online (Kim & Sherman, 2007).
Within such a background, any type of personality assessment
re-
ferring to real-life personality traits could explain only a small
portion
of online behavior variance, and not fit with the purpose. Future
re-
search could develop new models of web-personality assessment
tools
in order to measure the impact of “online personality” on social
influ-
ence and conformity.
Furthermore, the study presented here has some limitations that
could be controlled in further research on the topic.
As mentioned while describing the sample, we have not been
able to

balance the subjects according to genders and we have an over-
representation of people identifying as females. The more dated
lit-
erature that explored the gender differences in conformist
behaviors
registered higher conformity in the females (Baumeister &
Sommer,
1997), while more recent studies found no differences
(Rosander &
Eriksson, 2012). This could be due by the increasing push
towards
gender equality which resulted in a less strict adherence to the
tradi-
tional division between gender roles that especially western
societies
(those in which the aforementioned studies were conducted)
have ex-
perienced throughout the years.
Another limitation regards the diversity of the pool of
participants.
For linguistic reasons related to the semantic nature of two of
the
three tasks, the participants had to be fluent in Italian, and this
resulted
in having mostly Italians taking part to the experiment, who, in
the
nonymity condition, interacted with bots to which were given
Italian-
sounding names and surnames.
We believe that these results can be generalized to other
contexts
and similar countries, but we must consider that cultural
differences

shaping the behavior in different ways may appear if the study
is re-
plicated elsewhere.
First and foremost, according to the literature, the perception
to-
wards conformity is different in individualistic and
collectivistic cul-
tures, where in the former it is a negatively connoted behavior,
while in
the latter it is generally seen more positively (Bond & Smith,
1996),
therefore, with a broader pool of participants, different patterns
might
emerge.
In addition, according to the context, the level of contact with
people having different backgrounds, and the potential
prejudices or
negative attitudes towards some social groups that the
experimental
subjects might present, there could be different levels of
identification
with the group members, if more information that indicates
diversity is
given to the participants. This factor could be interesting to
control and
analyze in further studies.
In the same way, at a broader level, the multiculturalism,
general
openness, political and social situation of the context could also
affect
the subjects' behavior in relation to the building of in-group and
out-
group perception towards the group members.

References
Asch, S. E. (1955). Opinions and social pressure. Readings
about the social animal, 193,
17–26.
Asch, S. E. (1956). Studies of independence and conformity: I.
A minority of one against a
unanimous majority. Psychological Monographs: General and
Applied, 70(9), 1.
Asch, S. E., & Guetzkow, H. (1951). Effects of group pressure
upon the modification and
distortion of judgments. Groups, Leadership, and Men, 222–
236.
Baumeister, R. F., & Sommer, K. L. (1997). What do men want?
Gender differences and two
spheres of belongingness: Comment on Cross and Madson
(1997).
Bolker, B. M., Brooks, M. E., Clark, C. J., Geange, S. W.,
Poulsen, J. R., Stevens, M. H. H.,
et al. (2009). Generalized linear mixed models: A practical
guide for ecology and
evolution. Trends in Ecology & Evolution, 24(3), 127–135.
Bond, R., & Smith, P. B. (1996). Culture and conformity: A
meta-analysis of studies using
Asch's (1952b, 1956) line judgment task. Psychological
Bulletin, 119(1), 111.
Chow, S. C., Shao, J., Wang, H., & Lokhnygina, Y. (2017).
Sample size calculations in
clinical research. Chapman and Hall/CRC.

Cialdini, R. B., & Trost, M. R. (1998). Social influence: Social
norms, conformity, and
compliance. In (4th ed.). D. T. Gilbert, S. T. Fiske, & G.
Lindzey (Vol. Eds.), The
handbook of social psychology: Vol. 2, (pp. 151–192). Boston:
McGraw-Hill.
Coleman, S. (2004). The effect of social conformity on
collective voting behavior. Political
Analysis, 12(1), 76–96.
Crutchfield, R. S. (1955). Conformity and character. American
Psychologist, 10(5), 191.
Del Vicario, M., Bessi, A., Zollo, F., Petroni, F., Scala, A.,
Caldarelli, G., & Quattrociocchi,
W. (2016). The spreading of misinformation online. Proceedings
of the National
Academy of Sciences, 113(3), 554–559.
Deutsch, M., & Gerard, H. B. (1955). A study of normative and
informational social in-
fluences upon individual judgment. Journal of Abnormal and
Social Psychology, 51(3),
629.
DeYoung, C. G., Peterson, J. B., & Higgins, D. M. (2002).
Higher-order factors of the Big
Five predict conformity: Are there neuroses of health?
Personality and Individual
Differences, 33(4), 533–552.
Giannini, M., Pannocchia, L., Grotto, L. P., & Gori, A. (2012).
A measure for counseling:
The five factor adjective short test (5-fast). Couns G Ital Ric

Appl, 3, 384.
Grös, D. F., Antony, M. M., Simms, L. J., & McCabe, R. E.
(2007). Psychometric properties
of the state-trait inventory for cognitive and somatic anxiety
(STICSA): Comparison to
the state-trait anxiety inventory (STAI). Psychological
Assessment, 19(4), 369.
Jagatic, T. N., Johnson, N. A., Jakobsson, M., & Menczer, F.
(2007). Social phishing.
Communications of the ACM, 50(10), 94–100.
Kim, H. S., & Sherman, D. K. (2007). “Express yourself”:
Culture and the effect of self-
expression on choice. Journal of Personality and Social
Psychology, 92(1), 1.
Latané, B. (1981). The psychology of social impact. American
Psychologist, 36(4), 343.
LaTour, S. A., & Manrai, A. K. (1989). Interactive impact of
informational and normative
influence on donations. Journal of Marketing Research, 327–
335.
Lee, E. J. (2004). Effects of visual representation on social
influence in computer‐me-
diated communication: Experimental tests of the social identity
model of deindivi-
duation effects. Human Communication Research, 30(2), 234–
259.
Lee, E. J. (2007). Deindividuation effects on group polarization
in computer‐mediated
communication: The role of group identification,

public‐self‐awareness, and per-
ceived argument quality. Journal of Communication, 57(2),
385–403.
Liu, R. R., & Zhang, W. (2010). Informational influence of
online customer feedback: An
empirical study. The Journal of Database Marketing &
Customer Strategy Management,
17(2), 120–131.
Lott, A. J., & Lott, B. E. (1965). Group cohesiveness as
interpersonal attraction: A review
of relationships with antecedent and consequent variables.
Psychological Bulletin,
64(4), 259.
Mann, L., Newton, J. W., & Innes, J. M. (1982). A test between
deindividuation and
emergent norm theories of crowd aggression. Journal of
Personality and Social
Psychology, 42(2), 260.
Markus, H. (1978). The effect of mere presence on social
facilitation: An unobtrusive test.
Journal of Experimental Social Psychology, 14(4), 389–397.
McMillan, D. W., & Chavis, D. M. (1986). Sense of community:
A definition and theory.
Journal of Community Psychology, 14(1), 6–23.
Postmes, T., Spears, R., Sakhel, K., & De Groot, D. (2001).
Social influence in computer-
mediated communication: The effects of anonymity on group
behavior. Personality
and Social Psychology Bulletin, 27(10), 1243–1254.

Prezza, M., Pacilli, M. G., Barbaranelli, C., & Zampatti, E.
(2009). The MTSOCS: A mul-
tidimensional sense of community scale for local communities.
Journal of Community
Psychology, 37(3), 305–326.
Rosander, M., & Eriksson, O. (2012). Conformity on the
Internet–The role of task diffi-
culty and gender differences. Computers in Human Behavior,
28(5), 1587–1595.
Rosenberg, M. (1965). Rosenberg self-esteem scale (RSE).
Acceptance and commitment
therapy. Measures Package, 61, 52.
Schwarzer, R., & Jerusalem, M. (2010). The general self-
efficacy scale (GSE). Anxiety,
Stress & Coping, 12, 329–345.
Sherif, M. (1937). An experimental approach to the study of
attitudes. Sociometry, 1(1/2),
90–98.
Short, J., Williams, E., & Christie, B. (1976). The social
psychology of telecommunications.
Sibilia, L., Schwarzer, R., & Jerusalem, M. (1995). Italian
adaptation of the general self-
(2019) 230–237
236
http://refhub.elsevier.com/S0747-5632(18)30552-1/sref1

efficacy scale. Resource document. Ralf Schwarzer web site.
Spears, R., Postmes, T., Lea, M., & Watt, S. (2018). A SIDE
view of social influence. In K.
D. Williams, & J. P. Forgas (Eds.). Social influence: Direct and
indirect processes (pp.
331–350). Philadelphia/Hove, UK: Psychology Press.
Spears, R., Postmes, T., Lea, M., & Wolbert, A. (2002). When
are net effects gross pro-
ducts? Communication. Journal of Social Issues, 58(1), 91–107.
Spielberger, C. D., & Gorsuch, R. L. (1983). State-trait anxiety
inventory for adults: Sampler
set: Manual, Tekst booklet and scoring key. Consulting
Psychologists Press.
Tsikerdekis, M. (2013). The effects of perceived anonymity and
anonymity states on
conformity and groupthink in online communities: AW ikipedia
study. Journal of the
American Society for Information Science and Technology,
64(5), 1001–1015.
Zimbardo, P. G. (1969). The human choice: Individuation,
reason, and order versus
deindividuation, impulse, and chaos. Nebraska symposium on
motivation. University of
Nebraska press.
Zubiaga, A., Liakata, M., Procter, R., Hoi, G. W. S., & Tolmie,

P. (2016). Analysing how
people orient to and spread rumours in social media by looking
at conversational
threads. PLoS One, 11(3), e0150989.
(2019) 230–237
237
http://refhub.elsevier.com/S0747-5632(18)30552-
1/sref40Deindividuation effects on normative and informational
social influence within computer-mediated-
communicationIntroductionTheoretical frameworkOverview and
predictionsMethodSampling and participantsMaterials and
apparatusProcedureResultsGeneral discussion and
conclusionsReferences

Contents lists available at ScienceDirect
Journal of Experimental Social Psychology
journal homepage: www.elsevier.com/locate/jesp
The strategic moral self: Self-presentation shapes moral
dilemma judgments
Sarah C. Roma,⁎, Paul Conwayb
a Department of Psychology, University of Cologne, Germany
b Florida State University, Department of Psychology, United
States
A R T I C L E I N F O
Keywords:
Moral dilemmas
Social judgment
Social perception
Self-perception
Meta-perception
A B S T R A C T
Research has focused on the cognitive and affective processes
underpinning dilemma judgments where causing
harm maximizes outcomes. Yet, recent work indicates that lay
perceivers infer the processes behind others'
judgments, raising two new questions: whether decision-makers
accurately anticipate the inferences perceivers
draw from their judgments (i.e., meta-insight), and, whether
decision-makers strategically modify judgments to
present themselves favorably. Across seven studies, a) people

correctly anticipated how their dilemma judg-
ments would influence perceivers' ratings of their warmth and
competence, though self-ratings differed (Studies
1–3), b) people strategically shifted public (but not private)
dilemma judgments to present themselves as warm
or competent depending on which traits the situation favored
(Studies 4–6), and, c) self-presentation strategies
augmented perceptions of the weaker trait implied by their
judgment (Study 7). These results suggest that moral
dilemma judgments arise out of more than just basic cognitive
and affective processes; complex social con-
siderations causally contribute to dilemma decision-making.
During the Second World War, Alan Turing and his team
cracked the
Enigma Code encrypting German war communications. Soon,
British High
Command discovered an impending attack on Coventry—but
taking
countermeasures would reveal the decryption (Winterbotham,
1974).
Thus, they faced a moral dilemma: allow the deadly raid to
proceed and
continue intercepting German communications, or deploy
lifesaving
countermeasures and blind themselves to future attack.
Ultimately, the
Allies allowed the attack to proceed. Lives were lost, but some
analysts
suggest this decision expedited the war's conclusion (Copeland,
2014). The
moral judgment literature suggests that such decisions reflect a
tension
between basic affective processes rejecting harm and cognitive
evaluations
of outcomes allowing harm (Green, Nystrom, Engell, Darley, &

Cohen,
2004). But is it possible that self-presentation also factored in?
The British
High Command may have considered how their allies would
react upon
learning they threw away a tool for victory to prevent one
deadly, but
relatively modest, raid.
Moral dilemmas typically entail considering whether to accept
harm
to prevent even greater catastrophe. Philosophers originally
developed
such dilemmas to illustrate a distinction between killing
someone as the
means of saving others versus as a side effect of doing so (Foot,
1967),
but subsequent theorists have largely described them as
illustrating a
conflict between deontological and utilitarian philosophy (e.g.,
Greene,
Sommerville, Nystrom, Darley, & Cohen, 2001). The dual
process model
suggests that affective reactions to harm underlie decisions to
reject
harm, whereas cognitive evaluations of outcomes underlie
decisions to
accept harm to maximize outcomes (Greene et al., 2004). Other
the-
orists have described these as processes in terms of basic
cognitive ar-
chitecture for decision-making (Crockett, 2013; Cushman,
2013), or
heuristic adherence to moral rules (Sunstein, 2005). Notably, all
such

existing models focus on relatively basic, non-social processing.
Yet, Haidt (2001) argued that moral judgments are intrinsically
social, and communicate important information about the
speaker. In-
deed, recent work indicates that lay perceivers view decision-
makers
who reject harm (upholding deontology) as warmer, more moral,
more
trustworthy, more empathic, and more emotional than decision-
makers
who accept harm (upholding utilitarianism), whom perceivers
view as
more competent and logical, with consequences for hiring
decisions
(Everett, Pizarro, & Crockett, 2016; Kreps & Monin, 2014;
Rom,
Weiss, & Conway, 2016; Uhlmann, Zhu, and Tannenbaum,
2013).1
Moreover, social pressure can influence dilemma judgments
(Bostyn &
Roets, 2016; Kundu & Cummins, 2012; Lucas & Livingstone,
2014).
Such findings raise the question of whether people have meta-
insight
http://dx.doi.org/10.1016/j.jesp.2017.08.003
Received 4 April 2017; Received in revised form 8 August
2017; Accepted 17 August 2017
⁎ Corresponding author at: Department of Psychology,
University of Cologne, Richard-Strauss-Str. 2, 50931, Cologne,
Germany.
E-mail addresses: [email protected] (S.C. Rom),
[email protected] (P. Conway).

1 Deontological dilemma judgments appear to convey both
warmth and morality (Rom et al., 2016). Although these
constructs can be disentangled (e.g., Brambilla et al., 2011), in
the
present context they happen to covary substantially. It may be
that different aspects of deontological decisions influence these
perceptions (e.g., whether they accord with moral rules;
whether they suggest emotional processing), but these aspects
overlap in the current paradigm. We focus primarily on
perceptions of warmth, which roughly corresponds to the
affective
processing postulated by the dual process model, and relegated
findings regarding morality the supplement. Future work should
disentangle warmth trait perceptions from moral
character evaluations.
Journal of Experimental Social Psychology 74 (2018) 24–37
Available online 30 August 2017
0022-1031/ © 2017 Elsevier Inc. All rights reserved.
T
https://www.sciencedirect.com/science/journal/00221031
https://www.elsevier.com/locate/jesp
https://doi.org/10.1016/j.jesp.2017.08.003
http://crossmark.crossref.org/dialog/?doi=10.1016/j.jesp.2017.0
8.003&domain=pdf
into how their dilemma judgments make them appear in the eyes

of
others, and whether decision-makers strategically adjust
dilemma
judgments to create desired social impressions. If so, this would
provide
the first evidence to our knowledge that higher-order processes
causally
influence judgments, suggesting dilemma decisions do not
merely re-
flect the operation of basic affective and cognitive processes.
1. Moral dilemma judgments: basic vs. social processes
Moral dilemmas originated as philosophical thought
experiments,
including the famous trolley dilemma where decision-makers
could
redirect a runaway trolley so it kills one person instead of five
(Foot,
1967). According to Greene et al. (2001), refusing to cause
harm to save
others qualifies as a ‘characteristically deontological’ decision,
because
in deontological ethics the morality of action primarily hinges
on its
intrinsic nature (Kant, 1785/1959). Conversely, causing harm by
re-
directing the trolley saves five people, thereby qualifying as a
‘char-
acteristically utilitarian’ decision, because in utilitarian ethics
the
morality of an action primarily hinges on its outcomes (Mill,
1861/
1998).2 Note that utilitarian philosophy technically entails
impartial
maximization of the greater good, which represents a subset of

the
broader concept of consequentialism, which advocates for
outcome-
focused decision-making more generally. We do not wish to
imply that
making a judgment consistent with utilitarianism renders one a
utili-
tarian—it need not (e.g., Kahane, 2015)—but rather we use the
term
‘utilitarian’ in the simpler senses that such judgments a)
objectively
maximize overall outcomes, b) appear to often entail ordinary
cost-
benefit reasoning, and c) utilitarian/consequentialist
philosophers
generally approve of such judgments (see Amit & Greene,
2012).
Although dilemmas originated in philosophy, research in psy-
chology, neuroscience, and experimental philosophy has aimed
to
clarify the psychological mechanisms driving dilemma
judgments. Most
prominent among these is the dual process model, which
postulates that
basic affective and cognitive processes drive dilemma
judgments
(Greene et al., 2001). Other theorists have argued judgments
reflect
decision-making systems focused on immediate action versus
long-
range goals (Crockett, 2013; Cushman, 2013), heuristic
adherence to
moral rules (Sunstein, 2005), or the application of innate moral
grammar (Mikhail, 2007a, 2007b). We do not aim to adjudicate
be-

tween these various claims, nor do we dispute the contribution
of such
processes. Rather, we simply note that these models focus on
basic,
nonsocial processes.
Research has largely ignored the possibility that higher-order
sophis-
ticated social processes might causally contribute to dilemma
judgments.
Yet, morality appears intrinsically social (Haidt, 2001), and
most real-
world moral judgments involve publicly communicating with
others (e.g.,
Hofmann, Wisneski, Brandt, & Skitka, 2014). We expect the
same is true of
dilemma judgments. Although the best-known dilemmas are
hypothetical
(such as the trolley dilemma), many real-world decisions entail
causing
harm to improve overall outcomes (e.g., launching airstrikes in
Syria to
prevent ISIS from gaining momentum, punishing naughty
children to
improve future behavior, imposing fines to prevent speeding).
As decisions
in such cases align with either deontological or utilitarian
ethical positions,
they correspond to real world moral dilemmas. Moreover, lay
decision-
makers employ verbal arguments that align with deontological
and
utilitarian ethical positions (Kreps & Monin, 2014). Hence,
social con-
sideration of dilemma judgments is not restricted to responses

to hy-
pothetical scenarios, but forms an ordinary part of
communication about
common moral situations.
Kreps and Monin (2014) examined deontological and utilitarian
arguments in speeches by Presidents Clinton and Bush, among
other
politicians. Lay perceivers viewed speakers as moralizing more
when
they framed arguments in terms of deontology rather than
utilitar-
ianism. These findings align with work on hypothetical dilemma
deci-
sions: perceivers rated and treated decision-makers who rejected
harm
(upholding deontology) as more trustworthy than decision-
makers who
accept harm (upholding utilitarianism, Everett et al., 2016), as
well as
more moral, more empathic, and less pragmatic than harm-
accepting
decision-makers (Uhlmann et al., 2013). Likewise, Rom et al.
(2016)
found that lay people appear to intuit the dual process model:
they
rated targets who rejected harm as relatively warm, and inferred
that
such judgments were driven by emotion. Conversely, perceivers
rated
targets who accepted harm as relatively competent, and inferred
that
such judgments were driven by cognitive deliberation.3
Moreover,
perceivers preferred harm-rejecting decision-makers for social
roles

prioritizing warmth, such as social partners or their child's
doctor, but
preferred harm-accepting decision-makers for roles prioritizing
com-
petence, such as hospital administration (Everett et al., 2016;
Rom
et al., 2016). Hence, decision-makers face a warmth/competence
tra-
deoff when presenting their decision to others. The current work
ex-
amines whether decision-makers are aware of this trade-off, and
whe-
ther they strategically adjust their decisions to present
themselves
favorably.
2. Meta-perceptions regarding dilemma judgments
We propose that lay perceivers hold fairly accurate meta-
perceptions
into how others will view them based on their dilemma
decision. People
care deeply about their moral reputation (Aquino & Reed, 2002;
Everett
et al., 2016; Krebs, 2011) and the moral reputations of others
(Brambilla,
Rusconi, Sacchi, & Cherubini, 2011; Goodwin, Piazza, & Rozin,
2014).
Clearly, the research described above on perceptions of
decision-makers
indicate that dilemma decisions can affect moral reputation,
suggesting
that people should be attuned to what messages their judgments
convey.
Moreover, past work suggests that people can be reasonably
accurate

when gauging how others perceive them. For example,
narcissists appear
aware that others view them less positively than they view
themselves
(Carlson & Furr, 2009; Carlson, Vazire, & Furr, 2011). Self-
and social-rat-
ings particularly converge when the underlying traits entail
public beha-
viors (e.g., loquaciousness signals extraversion) rather than
inner states
(e.g., neurotic feelings, Vazire, 2010). Sharing one's dilemma
judgment
entails a clear public behavior, suggesting relative accuracy in
meta-per-
ceptions.
2 Following Greene et al. (2001), we use the term
‘characteristically’ deontological/
utilitarian, because there are many variants of each theory that
do not all agree. None-
theless, this terminology is widely employed currently, and so
we follow in this termi-
nological tradition despite its limitations. Note that we are not
arguing that making a
given dilemma decision implies that decision-makers ascribe to
abstract philosophical
commitments. Rather, we argue simply that ‘utilitarian’
judgments qualify as such be-
cause they tend to maximize outcomes, regardless of decision-
makers' philosophical
commitments. Just as one need not be Italian to cook an Italian
meal, accepting outcome-
maximizing harm on a dilemma does not make one a utilitarian.
Hence, these terms
reflect only to the content of judgments, rather than the
qualities of judges (see

Amit & Greene, 2012).
3 If the dual-process model is correct, responses to classic
moral dilemmas do not reflect
the degree to which decision-makers experience affective
reactions or engage in cognition
in an absolute sense. If classic moral dilemmas place affect and
cognition in conflict, and
ultimately judges may only choose one option, then judgments
reflect the relative strength
of each process. For example, accepting harm that maximizes
outcomes may occur either
due to strong cognition coupled with strong but slightly weaker
affect, or weak cognition
coupled with weaker affect. Hence, a judgment to accept
causing harm does not reveal
whether the judge experienced strong or weak affect—only that
cognition outweighed
whatever degree of affect they experienced. Nor does such a
judgment guarantee that the
judge engaged in strong cognition—only that whatever
cognition they engaged in out-
weighed their affective experience. Some people may engage in
extensive affect and
cognition, whereas others engage in little of either. In order to
estimate each processes
independently, it is necessary to use a technique such as process
dissociation (see
Conway & Gawronski, 2013). However, in the current work we
are not interested in the
actual processes underlying dilemma judgments so much as lay
perceptions of these
processes. To that end, lay people, like many researchers,
equate harm avoidance judg-
ments with strong affect and harm acceptance judgments with
strong cognition. This

intuition is effective as a rough heuristic, so long as researchers
recognize that it does not
accurately describe moral dilemma processing.
S.C. Rom, P. Conway Journal of Experimental Social
Psychology 74 (2018) 24–37
25
However, other research casts doubt on the possibility of
accurate
dilemma meta-perceptions in dilemma research. Besides public
expression,
dilemma judgments entail intrapsychic aspects such as
emotional reac-
tions, perceptions of conflict, and so on (e.g., Andersen & Ross,
1984;
Kruger & Gilovich, 2004; Pronin, 2008; Winkielman &
Schwarz, 2001).
Decision-makers hold privileged knowledge of their experience
of these
inner states. People often fail to consider that others have
access to less
information than they do (Chambers, Epley, Savitsky, &
Windschitl, 2008).
Whereas egocentric perspectives come to mind easily, adjusting
away from
egocentricity is difficult (Epley, Keysar, Van Boven, &
Gilovich, 2004).
Thus, meta-perceptions are often biased by self-understanding
(Chambers
et al., 2008; Kaplan, Santuzzi, & Ruscher, 2009; Kenny &
DePaulo, 1993).
Moreover, people are motivated to view themselves positively

in the moral
domain (Epley & Dunning, 2000) much like non-moral domains
(e.g.,
Dunning & McElwee, 1995), and can rationalize either dilemma
decision
in self-flattering ways (Uhlmann, Pizarro, Tannenbaum, &
Ditto, 2009;
Liu & Ditto, 2013). Thus, people may well judge themselves as
high in both
warmth and competence regardless of their dilemma decision—
and may
expect others to agree with this flattering self-assessment.
If decision-makers erroneously base meta-perceptions on self-
per-
ceptions, meta-perception ratings should converge with self-
ratings and
diverge from ratings of others following the same judgment—
that is,
people may believe they come across as both warm and
competent
regardless of their dilemma decision, whereas they view others'
deci-
sions as reflecting a warmth/competence trade-off. Conversely,
if
people have accurate meta-insight into how others perceive
them,
meta-perception ratings should converge with other ratings and
diverge
from self-ratings—that is, people may privately believe they are
warm
and competent regardless of dilemma decision, yet expect others
to rate
them according to the same warmth/competence tradeoff
implied by
others' judgments. We contrasted these predictions empirically.

3. Strategic self-presentation in dilemma judgments
If people evince accurate meta-insight into what their dilemma
deci-
sion conveys, this raises the possibility that they strategically
adjust such
decisions to present themselves favorably. There are potential
upsides and
downsides to selecting each dilemma judgment, as the precise
cause of
others' dilemma decisions appear ambiguous. Upholding
utilitarianism by
accepting outcome-maximizing harm amounts to bloodying
one's hands
for the sake of the community. Such bold and brutal action may
convey
either competent leadership (Lucas & Galinsky, 2015) or a
callous dis-
regard for causing harm—as in psychopathy (Bartels & Pizarro,
2011) or
low empathy (Gleichgerrcht & Young, 2013). Conversely,
rejecting harm
(upholding deontology) may convey either a warm concern for
others
and/or principled respect for life and/or trustworthiness (Everett
et al.,
2016; Kreps & Monin, 2014; Rom et al., 2016), or suggest
incompetent
paralysis when the situation demands bold action (Gawronski,
Conway,
Armstrong, Friesdorf, & Hütter, 2015; Gold, Pulford, &
Colman, 2015).
Hence, in some circumstances it may be preferable to risk
appearing in-
competent in order to convey warmth, trustworthiness, and

respect for
life; in other situations, it may be preferable to risk appearing
cold and
callous in order to convey decisive competence and leadership.
People care deeply about presenting themselves favorably. They
tailor their public images in various domains to the perceived
values
and preferences of important others (Leary, 1995; Leary &
Kowalski,
1990; Reis & Gruzen, 1976; von Baeyer, Sherk, & Zanna,
1981). People
change social roles over time, and social roles carry
expectations re-
garding how individuals who occupy those roles ought to
behave
(Sarbin & Allen, 1968). Hence, people often flexibly present
themselves
to conform to different social role expectations (Leary, 1989;
Leary,
Robertson, Barnes, & Miller, 1986). Indeed, Everett et al.
(2016) argued
that deontological dilemma judgments may operate as a
reputation-
management mechanism to present oneself as a trustworthy
social in-
teraction partner by demonstrating respect for others autonomy
and
wishes (see also Bostyn & Roets, 2016).
Accordingly, previous work demonstrates that social situations
influ-
ence dilemma responses. In a modification of the Asch
conformity para-
digm, Kundu and Cummins (2012) asked participants whether
they would

accept or reject outcome-maximizing harm after a series of
confederates
gave a particular answer. They found evidence for conformity
pressure:
participants were more likely to give answers consistent with
those of the
confederates. Bostyn and Roets (2016) conducted a similar
study, and
argued that conformity pressure was stronger for harm rejection
(up-
holding deontology) than harm acceptance (upholding
utilitarianism).
However, Lucas and Livingstone (2014) found that participants
who so-
cially connected with others before completing dilemmas after
were more
willing to accept harm (upholding utilitarianism). It may be that
resolving
dilemmas in front of strangers motivated participants to skew
towards
deontological answers so as to avoid appearing immoral—after
all, re-
search suggests that moral traits appear especially important
when
forming first impressions (Brambilla et al., 2011; Goodwin et
al., 2014),
and that warmth may also be important when forming first
impressions
(Fiske, Cuddy, & Glick, 2006). Conversely, when participants
have an op-
portunity to establish warmth or morality through social
interactions, they
may have felt free to demonstrate other qualities, such as
competence.
These findings suggest that context may shift whether accepting
or re-

jecting harm seems to be the optimal answer. If participants
strategically
adjust dilemma judgments, their perception of expectations
should vary
depending on whether the circumstances appear to prioritize
warmth over
competence, and their public (but not private) dilemma answers
should
track such expectations.
4. Overview
Across seven studies, we investigated whether people hold
accurate
meta-perceptions regarding how others view them based on their
di-
lemma judgments, and whether they strategically modify such
judg-
ments to present themselves favorably. First, we examined
whether
people have accurate meta-insight into the warmth and
competence
ratings others infer from their dilemma judgments by comparing
warmth and competence ratings of others, the self, and meta-
percep-
tions of the self (Studies 1–3). Second, we tested whether
people shift
public (but not private) dilemma judgments depending on
whether
warmth or competence is favored in a given situation (Studies
4–6).
Third, we investigated whether people can use communication
strate-
gies to offset the weaker trait implied by their judgment—
whether
people who accept harm can come across as warm, and people

who
reject harm can come across as competent (Study 7). Across all
studies,
we disclose all measures, manipulations, and exclusions, as well
as the
method of determining the final sample size. In none of the
studies data
collection was continued after data analysis.
5. Study 1
Study 1 examined the accuracy of participants' meta-perceptions
(i.e., meta-accuracy, Anderson, Ames, & Gosling, 2008)
following moral
dilemma judgments. We randomly assigned participants to one
of three
conditions: participants either made a dilemma judgment
themselves
(self and meta-perception condition) or read about another
persons'
dilemma judgment (other condition). Then, participants in the
self-
condition rated their own warmth and competence, those in the
other
condition rated the others' warmth and competence, and those in
the
meta-perception condition rated how they believed others would
view
their warmth and competence. Hence, the design was a 3 (target:
self
vs. other vs. meta-perceptions) × 2 (decision: harm rejection vs.
ac-
ceptance) × 2 (personality dimension: warmth vs. competence)
quasi-
experimental design (as participants were free to make either
dilemma

judgment themselves) with the first two factors between-
subjects and
the third within-subjects.
Given that people tend to view themselves positively in the
moral
domain (Epley & Dunning, 2000), and have access to internal
Psychology 74 (2018) 24–37
26
perceptions of conflict between response options, we expected
partici-
pants in the self-condition would rate themselves high on both
warmth
and competence, regardless of their dilemma decision. We
expected
participants in the other condition to replicate the patterns
demon-
strated by Rom et al. (2016): they should rate targets who
rejected
causing harm as warmer but less competent than targets who
accepted
causing outcome-maximizing harm. Most importantly, we
predicted
that participants' meta-perception condition would exhibit meta-
accu-
racy, by anticipating that others would rate them using the same
warmth/competence tradeoff (depending on dilemma decision)
as
participants in the other condition, rather than the uniformly
high

warmth and competence ratings participants privately make
about
themselves.
5.1. Method
5.1.1. Participants
We recruited 200 American participants (134 males, 66 females,
Mage = 30.63, SD = 8.92) via Mechanical Turk, who received
$0.25,
aiming for ~50 per between-subjects condition, although actual
responses
varied substantially (nself_harm_rejection = 14;
nself_harm_acceptance = 30;
nother_harm_rejection = 54; nother_harm_acceptance = 46;
nmeta_harm_rejection = 14;
nmeta_harm_acceptance = 42). First, we randomly assigned
participants to ei-
ther learn about Brad's ostensible judgment or to make a
judgment
themselves. Next, we randomly assigned half of participants in
the self-
dilemma-judgment condition to rate themselves on warmth and
compe-
tence, and half to rate themselves as they expected others would
(meta-
perceptions).4 We excluded no one. Although we did not
conduct a priori
power analyses, we felt confident that this design provided
reasonable
power based on past work (Rom et al., 2016). Indeed, a post hoc
power
analysis using GPower (Faul, Erdfelder, Lang, & Buchner,
2007) for a
fixed-effects between-within design where ηp

2 = .10, N = 200, α = .05,
and the correlation between repeated measures was r = .33
suggested that
we had ~99% power to detect the obtained interaction.
5.1.2. Procedure
All participants read the widely-employed crying baby dilemma
(e.g., Conway & Gawronski, 2013), where the actor must decide
whe-
ther to smother a baby to prevent its cries from alerting
murderous
soldiers hunting for other townspeople in hiding. Participants in
the self
and meta-perception conditions then selected either yes, this
action is
appropriate or no, this action is not appropriate (following
Greene
et al., 2001). Participants in the other condition viewed a photo
of a
university student named Brad, then learned that Brad had
selected
either one or the other of these responses (following Rom et al.,
2016).
Then, participants completed measures of warmth and
competence
using items adapted from Fiske, Cuddy, Glick, and Xu (2002).
Depending on condition, participants either rated themselves,
Brad,
or indicated how they thought others would rate them following
their
decision (meta-perception). Specifically, those in the meta-
perception
condition read:

Now take a moment to imagine that another person saw the
judgment
you made.
Based on that information, what would they think about you?
From
their perspective how well do you think they would say each
trait
describes you? THEY would think you are…
Participants indicated how well four warmth traits (warm, good-
natured, tolerant, sincere) and five competence traits
(competent, con-
fident, independent, competitive, intelligent) described the
target on 7-
point scales anchored at 1 (not at all) and 7 (very much). We
averaged
judgments into composites of warmth (α = .91) and competence
(α = .87), which were modestly correlated (r = .33). Item order
was
randomized for each participant. For exploratory reasons, we
also in-
cluded the single item moral, consistent with Rom et al. (2016).
Some researchers have argued that morality and warmth are dis-
tinguishable constructs (Brambilla et al., 2011; Goodwin et al.,
2014).
We find these arguments persuasive—used car salesmen that
evince
warm sociability should not be trusted, whereas a cold and dis-
passionate judge who sentences criminals may nonetheless
appear
moral. Nonetheless, it may be that these constructs align more
in some

contexts than others. Hence, we empirically examined how well
these
constructs dissociated in the current studies using five
strategies.
First, we noted that the item moral consistently correlated
highly
with the warmth composite measure, ~r = .75, consistent with
Rom
et al. (2016). Second, we noted that the item moral varied
across con-
ditions in the same manner as the warmth composite on all
studies (see
Supplementary analysis). Now, it remains possible that these
findings
simply reflect the fact that some items in the warmth
composite—such
as sincerity—assess perceptions morality instead of warmth.
Therefore,
third, we conducted factor analyses (principle axis factoring
with ob-
limin rotation) for all studies assessing warmth and morality
(see Table
S1 in supplementary material). In each case, all warmth items
loaded
together with the item moral onto a single factor, whereas all
compe-
tence items loaded onto a separate factor. A couple of items
occasion-
ally loaded well on both factors—confident, tolerant,
competent, and in-
telligent—but these dual loadings each occurred only once, and
did not
replicate across the other studies. Fourth, we conducted follow-
up
analyses for each study using only the single items warmth and

com-
petent instead of the composite measures; findings were very
similar
(find an example for Study 1 in the supplementary material).
Fifth, we
conducted follow-up analyses for each study using an
alternative
warmth score based on two items (warm, good-natured), and an
alter-
native morality score based on three items (sincere, tolerant,
morality),5
as well as a combined warmth/morality score including all
warmth
items plus the item morality. In each case, the pattern of
findings re-
mained very similar to the patterns presented below.
These findings suggest that in the context of dilemma
perceptions,
participants may find it difficult to disentangle warmth and
morality.
After all, perceivers may find it ambiguous whether a given
deontolo-
gical judgment reflects affective processing or adherence to
moral rules.
Alternatively, it may be that the particular items presented in
this scale
underestimate the difference between these constructs. Either
way, the
current paradigm was not designed to distinguish between
warmth and
morality. Indeed, these analyses suggest it may even be
warranted to
include the item moral in the warmth composite measure.
Nonetheless,

in recognition of the important theoretical distinction between
warmth
and morality (Brambilla et al., 2011; Goodwin et al., 2014) and
to re-
main consistent with Rom et al. (2016), we decided to treat the
item
morality as a separate construct. Given that the current focus
was on
contrasting perceptions of warmth and competence, and the
similarity
between the patterns of warmth and morality, we decided to
relegate
the morality findings to the supplementary material.
5.1.3. Results
We submitted ratings to a 3 (target: self vs. other vs. meta-
percep-
tions) × 2 (decision: harm rejection vs. acceptance) × 2
(personality
dimension: warmth vs. competence) repeated measures ANOVA
with
the first two factors between and the last factor within subjects
(see
Fig. 1). We conducted Levene's tests to examine homogeneity of
var-
iance assumptions. This assumption was not violated for
warmth, F
(5194) = 1.88, p = .100, but was violated for competence,
F(5194)
= 3.15, p = .009. Therefore, to supplement the main analysis in
the
text, we also conducted non-parametric Kruskal-Wallis and
Mann-
Whitney tests (see Supplement), which are more robust to
violations of

4 We acknowledge that this two-stage random assignment is
suboptimal because it led
to uneven cell sizes, which is one reason we increased the
sample size in Study 2. 5 We thank an anonymous reviewer for
this suggestion.
Psychology 74 (2018) 24–37
27
homogeneity of variance (Tomarken & Serlin, 1986; Kruskal &
Wallis,
1952; Mann & Whitney, 1947). The results of these tests largely
corro-
borated the conclusions of the main analyses presented here.
There was a main effect of target: participants gave higher
ratings
overall in the self (M = 5.18, SD = 1.12) than other (M = 4.64,
SD = .90), or meta-perception conditions (M = 4.28, SD = 1.04),
F(2,
194) = 8.47, p < .001, ηp
2 = .08. There was also a main effect of
decision: participants rated targets who rejected harm,
upholding
deontology, higher overall (M = 4.86, SD = 1.10), than targets
who
accepted harm, upholding utilitarianism (M = 4.51, SD = 1.03),
F(2,
194) = 8.32, p = .004, ηp

2 = .04. There was no main effect of per-
sonality dimension, F(2, 194) = 1.75, p = .18, ηp
2 = .01. These main
effects were qualified by a significant two-way interaction
between
target decision and personality measure, F(1, 194) = 45.65, p <
.001,
ηp
2 = .19, and a marginal interaction between target and
personality
measure, F(2, 194) = 3.03, p = .050, ηp
2 = .03, 95%, whereas the in-
teraction between target and decision was not significant, F(2,
194)
= 1.55, p = .214, ηp
2 = .02. Moreover, the three-way interaction was
significant, F(2, 194) = 11.14, p < .001, ηp
2 = .10.
We decomposed these interactions by examining post-hoc tests
within each condition. As predicted, participants in the self-
condition
rated themselves equally high on warmth when they rejected
(M = 5.69, SD = 1.24) or accepted (M = 5.12, SD = 1.40)
causing
harm, F(1194) = 2.70, p = .102, ηp
2 = .01, and equally competent
when they rejected (M = 5.50, SD = 1.17) versus accepted
causing
harm (M = 4.85, SD = 1.94), F(1194) = 3.60, p = .059, ηp

2 = .03.
However, participants in the other-condition replicated the
predicted
warmth/competence tradeoff found previously: Participants
rated Brad
higher on warmth when he rejected (M = 5.00, SD = 1.19), than
when
he accepted causing outcome-maximizing harm (M = 4.03, SD =
.99),
F(1, 194) = 15.57, p < .001, ηp
2 = .07. Conversely, they rated Brad
as higher in competence when he accepted (M = 5.16, SD =
1.16),
rather than rejected causing outcome-maximizing harm (M =
3.36,
SD = 1.31), F(1, 194) = 11.67, p < .001, ηp
2 = .06.
Crucially, participants in the meta-perception-condition evinced
the
same warmth/competence tradeoff as participants in the other-
condi-
tion: When participants rejected harm they inferred others
would per-
ceive them as warmer (M = 5.16, SD = 1.59) than when they
accepted
causing outcome-maximizing harm (M = 3.36, SD = 1.31), F(1,
194)
= 22.95, p < .001, ηp
2 = .10. In contrast, when they accepted such
harm, they inferred that others would perceive them as equally
com-

petent (M = 4.89, SD = 1.10) than when they rejected such harm
(M = 4.38, SD = 1.46), F(1, 194) = 2.32, p = .129, ηp
2 = .01.
5.1.4. Discussion
These findings suggest that participants have accurate meta-
insight
regarding the inferences others will draw about their personality
from
their dilemma judgments. Privately, participants rated
themselves
equally high on warmth and competence regardless of their
dilemma
decision. However, in the meta-perception condition they
expected
others to rate them similar to how they rated others: just as
participants
viewed targets who rejected causing harm as warmer and less
compe-
tent than targets who accepted causing harm, they expected that
others
would rate them as warmer (though not significantly less
competent)
when they rejected vs. accepted causing harm themselves. To
our
knowledge, this is the first evidence that participants are aware
of the
impression their dilemma judgments convey to others.
However, our quasi-experimental design suffered from the
limita-
tion of nonrandom assignment: participants in the self and meta-
per-
ception conditions freely choose which dilemma decision to

make.
Hence, it remains possible that our meta-perception results
reflect the
general psychology of people who made a specific decision,
rather than
inferences regarding that decision per se. Even though this
interpreta-
tion seems unlikely give the null effect in the private self-rating
con-
dition, we aimed to resolve this confound in Study 2.
6. Study 2
Study 2 replicated the meta-perception condition from Study 1,
together with a communication error condition where
participants ima-
gined that others erroneously learned they made the dilemma
judgment
opposite to the one they truly made. This design allowed us to
test
whether meta-perceptions in Study 1 would hold for decisions
that
participants personally disagreed with. We expected that
warmth and
competence meta-perceptions would track the decision others
believed
participants made (harm rejection: higher warmth than
competence,
harm acceptance: higher competence than warmth), rather than
the
decision participants actually made.
6.1. Method
6.1.1. Participants
To increase confidence in the effects and address the uneven

cell
sizes in Study 1, we decided to approximately double the
sample size
and employ more traditional randomization procedures. We
recruited
397 American participants via Mechanical Turk, who received
$0.25.
We excluded 24 participants who completed less than 50% of
depen-
dent measures, leaving a final sample of 373 (244 males, 123
females, 6
unreported, Mage = 30.49, SD = 9.89. Participants were
randomly as-
signed to the correct versus error condition, though of course
they se-
lected which dilemma judgment to make in this quasi-
experimental
design (ncorrect_harm_rejection = 32;
ncorrect_harm_acceptance = 157;
3
4
5
6
Self Other Meta Self Other Meta
W
a
rm
th

a
n
d
C
o
m
p
e
te
n
ce
R
a
tin
g
s Warmth Competence
Harm rejection Harm acceptance
Fig. 1. Participants' self, target, and meta-perception warmth
and competence ratings when they or the target rejected
causing harm to maximize outcomes (upholding deontology),
or accepted such harm (upholding utilitarianism), Study 1.
Error bars reflect standard errors.
Psychology 74 (2018) 24–37
28

nerror_harm_rejection = 45; nerror_harm_acceptance = 139). In
both conditions,
many more people accepted than rejected harm, but due to the
com-
munication error these ratios appear to flip. A Levene's test of
equality
of error variances revealed that this time, the homogeneity of
variance
assumption was violated neither for warmth, F (3369) = 2.50,
p = .059, nor for competence, F(3369) = 1.31, p = .271. GPower
suggested we had ~99% post-hoc power to detect the obtained
inter-
action with this sample size.
6.1.2. Procedure
Each participant read the crying baby dilemma from Study 1,
and
selected one of the two dilemma responses. Then we randomly
assigned
them to the correct communication or communication error
condition.
Participants in the correct communication condition imagined
that
others correctly learned which dilemma decision they made, as
in Study
1. Participants in the communication error condition imagined
that
others erroneously learned they made the dilemma decision
opposite to
their real decision. Specifically, participants read:
Now take a moment to imagine that another person learned
about

the judgment you made. As often happens, misinformation got
out
and this other person thinks you chose: Yes, harm is appropriate
[No
harm is not appropriate]. Based on the information that you
would
[not] SMOTHER the baby, what would this person think of you?
From their perspective how well do you think they would say
each
trait describes you? THEY would think you are…
Participants indicated how they believed others would perceive
them on the same warmth (α = .89), competence (α = .87), and
morality items as Study 1. This resulted in a 2 (communication:
correct
vs. error) × 2 (decision: harm rejection vs. acceptance) × 2
(dimen-
sion: warmth vs. competence) quasi-experimental design, as
partici-
pants could not be randomly assigned to make a particular
judgment.
Consistent with Study 1 and past work (Rom et al., 2016),
warmth and
competence correlated moderately (r = .40), whereas morality
corre-
lated highly with warmth (r = .87) and less with competence (r
= .38).
Morality yielded results similar to warmth across condition
(replicating
previous work, Rom et al., 2016) but was not focus of the
current
manuscript, so we again relegated it to the supplement.
6.1.3. Results
We submitted warmth and competence ratings to a 2 (commu-

nication: correct vs. error) × 2 (decision: harm rejection vs.
accep-
tance) × 2 (dimension: warmth vs. competence) repeated
measures
ANOVA (see Fig. 2) with the first two factors between-subjects
and the
last factor within-subjects. There was a main effect of
communication:
participants gave higher personality ratings overall in the
correct
communication (M = 4.25, SD = 1.26) than communication error
condition (M = 3.98, SD = 1.35), F(1, 369) = 17.51, p < .001,
ηp
2 = .05. There was no main effect of decision, F(1, 369) = 0.46,
p = .499, ηp
2 = .001, but there was a main effect of personality di-
mension: participants gave lower warmth (M = 3.89, SD = 1.81)
than
competence ratings overall (M = 4.35, SD = 1.47), F(1, 369) =
7.30,
p = .007, ηp
2 = .02. In addition, there were significant 2-way interac-
tions between communication and personality dimension, F(1,
369)
= 19.26, p < .001, ηp
2 = .05, and between decision and personality
dimension, F(1, 369) = 4.43, p = .036, ηp
2 = .01, though not between
communication and personality dimension, F(1, 369) = 2.25,
p = .134, ηp

2 = .01. More importantly, we obtained the expected
three-way interaction, F(1, 369) = 49.02, p < .001, ηp
2 = .12.
Post-hoc contrasts largely replicated Study 1 in the correct com-
munication condition: Participants expected that others would
rate
them as warmer when they rejected harm, upholding deontology
(M = 5.06, SD = 1.49) than accepted causing harm, upholding
utili-
tarianism (M = 3.40, SD = 1.68), F(1, 182) = 19.70, p < .001,
ηp
2 = .10. Results for competence trended in the expected
direction,
but did not reach significance: participants expected that others
would
rate them as similarly competent when they rejected (M = 4.51,
SD = 1.19), rather than accepted, causing harm (M = 4.82,
SD = 1.19), F(1, 187) = 1.91, p = .168, ηp
2 = .01. Participants in the
error communications condition showed the opposite pattern.
Participants expected that others would rate them as less warm
when
they rejected (M = 2.94, SD = 1.93) rather than accepted
causing
harm (M = 4.42, SD = 1.68), F(1, 369) = 22.28, p < .001, ηp
2 = .11.
Again, ratings for competence trended nonsignificantly in the
expected
direction: participants expected others to rate them similarly on
com-

petence when they rejected (M = 3.66, SD = 1.40), versus
accepted
causing harm (M = 4.00, SD = 1.29), F(1, 369) = 2.25, p = .135,
ηp
2 = .01.
6.1.4. Discussion
Study 2 replicated the findings from Study 1 in the correct com-
munication condition: participants who rejected harm
(upholding
deontology) inferred that others would perceive them as
relatively
warmer but (nonsignificantly) less competent, whereas
participants
who accepted harm (upholding utilitarianism) inferred that
others may
perceive them as (nonsignificantly) more competent but less
warm.
Moreover, these meta-perception ratings flipped when
participants
imagined that a communication error occurred, and others
erroneously
believed they made the judgment opposite to the judgment they
actu-
ally made. Hence, meta-perceptions tracked the information
available
to others, rather than reflecting the judgments participants
actually
made. This finding rules out the possibility that the Study 1
meta-per-
ception findings were driven by individual differences in meta-
per-
ceptions among people who rejected versus accepted harm,
thereby

overcoming the limitation of employing quasi-experimental
designs.
However, thus far we have examined meta-perceptions using
only the
crying baby dilemma in American MTurk samples. To improve
gen-
eralizability, we examined whether these effects would replicate
using a
whole battery of dilemmas and an in-lab sample of German-
speaking
student participants.
7. Study 3
Study 3 examined whether the meta-perception findings in
Studies 1
and 2 would generalize to other dilemmas and samples. Thus,
we re-
cruited a laboratory sample of German-speaking university
students
and broadened the stimulus set by translating a standardized
battery of
10 dilemmas into German, and randomly presenting participants
with
one of the ten dilemmas from this battery (Conway &
Gawronski, 2013).
7.1. Method
7.1.1. Participants
We obtained 131 German university students (55 males, 75
females,
1 other, 2 no gender indication, Mage = 30.49, SD = 9.90) who
re-
ceived $0.25. Again, we aimed for ~50 participants per cell, and

Contents lists available at ScienceDirectComputers in Huma.docx

Contents lists available at ScienceDirectComputers in Huma.docx

Recommended

Recommended

More Related Content

Similar to Contents lists available at ScienceDirectComputers in Huma.docx

Similar to Contents lists available at ScienceDirectComputers in Huma.docx (20)

More from bobbywlane695641

More from bobbywlane695641 (20)

Recently uploaded

Recently uploaded (20)

Contents lists available at ScienceDirectComputers in Huma.docx