Content uploaded by Cristina Bicchieri
Author content
All content in this area was uploaded by Cristina Bicchieri on May 13, 2020
Content may be subject to copyright.
Content uploaded by Cristina Bicchieri
Author content
All content in this area was uploaded by Cristina Bicchieri
Content may be subject to copyright.
1 23
Ethics and Information
Technology
ISSN 1388-1957
Volume 13
Number 1
Ethics Inf Technol (2010)
13:5-15
DOI 10.1007/s10676-010-9258-
y
Studying the ethical implications of e-trust
in the lab
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
Science+Business Media B.V.. This e-offprint
is for personal use only and shall not be self-
archived in electronic repositories. If you
wish to self-archive your work, please use the
accepted author’s version for posting to your
own website or your institution’s repository.
You may further deposit the accepted author’s
version on a funder’s repository at a funder’s
request, provided it is not made publicly
available until 12 months after publication.
ORIGINAL PAPER
Studying the ethical implications of e-trust in the lab
Cristina Bicchieri •Azi Lev-On
Published online: 22 December 2010
Springer Science+Business Media B.V. 2010
Abstract The paper presents results of recent laboratory
experiments that study if and how computer-mediated
communication affects cooperation and trust. It is argued
that communication medium does not matter much for
trust-building and maintenance, whereas relevant pre-play
communication and group size can have a major influence.
The implications of the findings for the design of sites that
depend on trusting communities are discussed.
Keywords Trust Promise-keeping Social norms
e-trust Communication Trust games
Communication in the lab
In environments characterized by deep hostility and dis-
trust between populations, and especially where the barri-
ers to exchange are not only emotional but geographical
and physical as well, many hope that computer-mediated
communication (specifically the Internet) may generate
exposure to others’ culture, beliefs and opinions, expand
people’s horizons and develop mutual understanding.
The experimental laboratory is a uniquely apt environ-
ment to study trust, reciprocity and cooperation, as it
allows to manipulate and control for multiple variables,
desegregate their effects, and conduct multiple treatments
to rule out competing hypotheses. In this article we focus
on social dilemma and trust games, as such games are
frequently used to study the determinants of trust and
cooperation in strategic interactions.
Social dilemmas are choice situations involving inter-
dependent agents, where the choices of each influence the
welfare of all. These choice situations are ‘dilemmas’
because the strategic setting is such that short-term rational
decisions of narrowly self-interested agents lead to socially
sub-optimal outcomes, i.e. the Nash equilibrium is also
Pareto sub-optimal. In typical social dilemma experiments,
subjects are divided into groups of size greater than two.
All subjects receive an endowment and then decide to send
some, all, or none of this amount to a ‘group account’. The
amount the subjects do not send is theirs to keep. Then, the
amount accumulated in the ‘group account’ is multiplied
by the experimenters and is equally divided among all
group members. These games use the mixed-motive
structure of a social dilemma, where it is individually best
for subjects to keep their money in their personal account,
but all are better off if everyone makes a cooperative
decision and contribute their endowments to the group
account.
Close cousins of social dilemma games are trust games,
which are in essence sequential dilemma games. In a typ-
ical trust experiment, subjects are assigned to one of two
roles: first-movers and second-movers. Experiments con-
tain two decision periods. In the first decision period, each
first-mover receives an endowment and then decides to
The article is based on previous theoretic and experimental work by
Bicchieri (2006), Bicchieri and Lev-On (2007), Bicchieri, Lev-On
and Chavez (2010), Lev-On, Chavez and Bicchieri (2010), and Lev-
On (2009b).
C. Bicchieri (&)
University of Pennsylvania, Philadelphia, PA, USA
e-mail: cb36@sas.upenn.edu
A. Lev-On
Institute for Research in the Social Sciences,
Stanford University, Stanford, CA, USA
A. Lev-On
School of Communication, Ariel University Center, Ariel, Israel
123
Ethics Inf Technol (2011) 13:5–15
DOI 10.1007/s10676-010-9258-y
Author's personal copy
send some, all, or none of it to the second-mover. The
amount the first-mover does not send is hers to keep. In the
second decision period, the amount first-movers sent to
second-movers is multiplied by the experimenters. Then,
the second-mover can send some, all, or none of this
amount to the first-mover. The amount the second-mover
does not send is hers to keep.
In one-shot social dilemma and trust games, it might be
argued that pre-play communication, and in particular
promises exchanged by subjects, are ‘‘cheap talk’’ and
therefore should not be expected to be kept. But a robust
finding in the experimental literature is the positive effect
of (interactive and unrestricted) communication on coop-
eration, which elsewhere we denote as a ‘communication
effect’ (Bicchieri 2002; Bicchieri and Lev-On 2007; see
also Ostrom 1998). Ledyard (1995), in an extensive survey
of the experimental literature on public goods, singles out
communication and the marginal per capita return as the
two variables most conducive to cooperation. Sally (1995),
in a meta-analysis of 35 years of social dilemma experi-
ments, shows that the ability to communicate increases
cooperation over base rates by 40%.
Let us now highlight several key dimensions of the
communication effect:
1. As long as communication persists, cooperation rates
are high and stable (Frohlich and Oppenheimer 1998;
Ostrom and Walker 1991; Schmitt et al. 2000;
Kinukawa et al. 2000).
2. A standard finding in iterated social dilemma experi-
ments is that without communication cooperation
gradually declines (i.e. Isaac et al. 1985; Isaac and
Walker (1988); Kiesler et al. (1996); Ga
¨chter and Fehr
(1999)). But cooperation rates peak after communica-
tion, even when it takes place after a few rounds of
declining cooperation (Ostrom et al. 1992; Zheng et al.
2002; Isaac and Walker 1988). In fact, communication
not only improves cooperation in the round immediately
following it, but its effect carries over to a number of
subsequent iterations (Isaac and Walker 1988).
3. When there are two separate groups, the carryover
effect extends to the out-group. Orbell et al. (1988)
allowed agents to communicate and decide which
strategies to adopt for contributing to the production of
a public good. After the discussion, they informed the
subjects that their contributions would indeed be used
to provide a public good, but a public good that only
the members of another group would enjoy. Despite
the unexpected change of beneficiary, 59% of the
subjects gave to the out-group after discussion, signif-
icantly more than the 30% contribution rate in the
control, no-communication condition (but still less
than the 79% contribution rate obtained when
communication was allowed and the contributions
went to the original in-group beneficiaries).
A few studies demonstrate that the communication
effect exists in trust games as well. Charness and Duf-
wenberg (2006) allowed unrestricted written messages
from second-movers to first-movers before trust games.
They found significant differences between the communi-
cation and the no-communication conditions in terms of
both trusting and reciprocating behaviors.
1
Communication and pro-social norms
What can the ‘‘communication effect’’ be attributed to?
Dawes (1980) identified three elements of face-to face
(FtF) communication that make cooperation possible:
identification,discussion, and commitment. Experimental
results enable to rule out the first two as primary causes of
the communication effect, and suggest that the communi-
cation effect is caused neither by the ability to identify and
‘humanize’ other agents, nor by the content and dynamics
of generic discussion (Bicchieri 2002; Bicchieri and Lev-
On 2007). Communication is highly effective only when
participants can discuss the game, and collectively decide
how to act. In that context, participants typically make
promises (to cooperate, or reciprocate), and they keep them
when the time to act comes. What remains to be explained
is why a promise made in a one-shot, anonymous game is
taken so seriously that cooperation occurs even in the
absence of any sanctioning mechanism. Or why, in repe-
ated games, promises carry over not just to subsequent
rounds (when no new discussion occurred), but even to
agents who were not the original recipients of the pledge
and did not promise anything in return.
The communication effect has been explained by Bic-
chieri (2006) in terms of her theory of social norms, which
is the theoretical underpinning for our experimental
research program as well. By ‘social norms’ we refer to
informal behavioral rules that are not supported by formal
sanctions. Take a norm of promise keeping. For some
people, keeping promises is an important personal norm
that one would follow in any circumstance, irrespective of
what others do. For others, the decision to keep one’s
promise is conditional upon expecting most other people to
keep their promises as well, and upon the belief that one is
expected to fulfill one’s promises, too. In this second case,
we say that keeping promises is a social norm, and as such
its implementation is dependent upon the expectations that
individuals hold. There is much experimental evidence to
support the view that important pro-social norms such as
1
See also Ben-Ner and Putterman (2006).
6 C. Bicchieri, A. Lev-On
123
Author's personal copy
reciprocity, fairness or cooperation are social rather than
personal, as manipulating mutual expectations causes
major behavioral changes (Bicchieri and Xiao 2009; Bic-
chieri and Chavez 2010).
2
This view of social norms is crucial for explaining the
effects of communication on cooperation and reciprocity as
well. Cooperation/reciprocation do not occur just because
people focus on a relevant norm; it is also important that
the right kind of expectations are present, and certain types
of communication fare better than others in creating such
expectations.
According to Bicchieri (2006, see also Bicchieri and
Lev-On 2007), communication about the dilemma has a
twofold effect: it focuses agents on pro-social norms
(particularly the norm of promise-keeping), and it also
generates the kind of mutual expectations that support
norm-abiding behavior. Such expectations are twofold: on
the one hand individuals must have empirical expectations
about other people’s conformity with the relevant norm.
Since compliance with social norms is conditional,
doubting that a norm is in fact followed would diminish
one’s willingness to follow it. On the other hand it must
also be the case that individuals have normative expecta-
tions, i.e., they must believe that others think they ought to
obey the norm in question and may even be prepared to
sanction transgressions (Bicchieri 2006).
3
Especially in the
case of pro-social norms, there is a tension between self-
interest and what is good for the group or society. If one
expects others to cooperate, there may be the temptation to
defect, and the presence of normative expectations con-
siderably weakens such temptations.
Communication, when successful, generates a norma-
tive environment that is conducive to cooperation. There
are various ways in which social norms can become salient,
so that agents are led to focus on them. One way is to
observe other people’s normative or counter-normative
behavior (Schroeder et al. 1983; Pillutla and Chen 1999).
Another is to be exposed to written or verbal content that
‘calls to mind’ a specific norm (Cialdini et al. 1991). Yet
simply focusing people on a relevant norm might not be
enough to generate compliance, especially when there is
some ambiguity in the decision context, or individuals
receive conflicting messages (Bicchieri and Xiao 2009).
Interactive, direct communication among the subjects
involved in the decision situation, especially when the
content of such communication involves a discussion of the
decision context and mutual promises, is a highly effective
mechanisms for focusing people on social norms and
inducing compliance.
The communication effect has mostly been studied in
face-to-face (FtF) settings, but it is present in computer-
mediated environments as well, i.e. computer-mediated
communication (CMC) produces higher cooperation rates
than equivalent environments in which communication is
not allowed. Here are some relevant features of the com-
puter-mediated communication effect (Bicchieri and Lev-
On 2007):
1. The communication effect varies in degree according
to the richness of the communication channel. For
example, videoconferencing produces cooperation
rates very close to face-to-face communication,
whereas text-based communication produces much
less cooperation. Generally, the CMC effect approx-
imates the FtF communication effect the closer the
communication channel comes to reproducing the
features of face-to-face communication.
2. When using CMC, communication is more norma-
tively charged than FtF communication.
4
This could be
explained by the need to ‘compensate’ for the lack of
contextual cues in computer-mediated environments
(Frohlich and Oppenheimer 1998; Rocco 1998; Brosig
et al. 2003).
3. Compared to FtF communication, it takes more time to
establish cooperation, especially when using ‘poorer’
CMC channels.
4. Especially with asynchronous communication, it is
more difficult to establish ‘social contracts’ in CMC,
and even when such agreements are reached, they are
violated more frequently than agreements reached
using FtF communication.
Different communication contexts can thus hamper or
promote focusing on the relevant norms, and the formation
of expectations that are crucial in supporting norm-abiding
behavior. For example, people may be more inclined to
question the credibility of online promises, with detri-
mental consequences for cooperation. When promises are
involved, the success of face-to-face communication
depends on the availability of a variety of cues that allow
subjects to assess mutual intentions and form expectations
about each other, all of which helps in lending credibility to
mutual promises. Such indicators include visual cues (body
language, eye contact, facial expressions, and so on), ver-
bal cues (tone of voice, phrasing, fluency, manner of
expressing moral rhetoric, and so on), and social cues
(status, group membership, gender, and so on). Some of
these cues are usually correlated by agents with
2
By ‘pro-social norms’ we denote norms that further positive social
relationships.
3
Sanctions may be positive, as when one is praised for norm-
compliance, or negative, as when one is criticized, made feel guilty or
ashamed, or even ostracized by the relevant group (Bicchieri 2006
Ch.1).
4
For example, greater use of (empty) threats against potential free-
riders is common in CMC environments.
Studying the ethical implications of e-trust in the lab 7
123
Author's personal copy
trustworthiness, and their presence or absence can have
important motivational consequences via the formation (or
impairment) of mutual expectations of promise-keeping
behavior.
The study of computer-mediated communication and the
conditions under which mutual promising and collective
commitments are likely to take place, is the subject of
ongoing research. Some of our recent studies intended to
further explore the ‘communication effect’ in computer-
mediated settings, and refine the understanding of its
determinants. So far it has been shown that the medium of
communication matters; the question now shifts to what
happens when the medium in use for pre-play talk interacts
with additional variables.
Experiment 1: media richness and communication
relevance
In the first experiment surveyed below (Bicchieri et al.
2010), two dimensions of the communication effect in trust
games were manipulated: communication relevance (i.e., the
situational relevance of what is communicated), and the
richness of the communication medium (CMC vs. FtF).
Relevant communication may matter due to the possibility to
talk about the strategic situation that the participants are
facing, and to make non-binding pledges about future
actions.
5
To study the interaction between media richness and
relevance, 64 participants were recruited at the University
of Pennsylvania. Each experimental session consisted of a
sequence of three identical trust games. For each game, the
first-mover had 6 USD, any dollar amount of which could
be sent to the second-mover. The amount the second-mover
received from the first-mover was tripled. The second-
mover could then send any dollar amount back to the first-
mover.
6
Before playing the game, participants were
allowed a few minutes of discussion. Half of the
participants were engaged in face-to-face discussion, and
another half had computer-mediated chats. In roughly half of
the experimental sessions, participants were involved in
‘‘relevant’’ communication, i.e. they were allowed to discuss
any topic except those pertaining to their identities and their
decisions or earnings from the previous condition, if there
was one. In the remaining sessions, the instructions required
that participants discuss only the following questions
(adapted from Buchan et al. 2006), which were not relevant
to the game: What are the three most populated cities in the
world? What are the three most populated cities in the US?
How many people live in Philadelphia and the surrounding
suburbs? How many counties are there in Pennsylvania?
(‘‘irrelevant’’ communication).
7
In these last sessions, par-
ticipants were explicitly instructed not to talk with their
partner about the game they were about to play.
After making their decision, first-movers were asked
about their expectations of second-movers reciprocation.
We wanted to know whether expectations differed
depending on the communication medium and condition,
and if expectations predict the level of trust. Note that by
trust we mean the amount of dollars sent by first-movers
($0 through $6), by reciprocity we mean the amount
returned by second-movers, relative to the amount sent,
and by expected reciprocity we mean the amount the first-
mover expected to be returned by the second-mover, rel-
ative to the amount sent.
Table 1(Bicchieri et al. 2010) summarizes the respon-
ses across the five combinations of communication rele-
vance and medium. Both relevance and medium had large,
positive effects on trust, reciprocity, and expected reci-
procity relative to the control condition.
8
Relevant, face-to-
face communication had the largest effect on all three
variables, whereas relevant, computer-mediated commu-
nication had the second largest effect.
Figure 1shows the distribution of trust across the five
conditions. First-movers were most trusting in the relevant
Table 1 Mean trust, reciprocity, and first-mover expectations by communication relevance and medium (N =64)
Control (N =32) FtF-relevant (N =14) CMC-relevant (N =14) FtF-irrelevant (N =18) CMC-irrelevant (N =18)
Trust 2.63 (0.36) 5.57 (0.46) 5.14 (0.57) 4.17 (0.49) 3.28 (0.61)
Reciprocity 1.92 (.48) 7.57 (0.96) 5.14 (1.33) 3.33 (1.05) 1.94 (0.78)
Expected reciprocity 3.54 (0.53) 8.36 (0.69) 7.43 (0.96) 5.56 (0.91) 4.28 (0.93)
5
Support for this hypothesis is also found in Bouas and Komorita
(1996); Dawes et al. (1977); Ga
¨chter and Fehr (1999).
6
Participants were paired randomly with a different partner for each
game, and this was common knowledge. First-movers did not receive
feedback on the amount that the second-mover returned until the end
of the experimental sessions.
7
Bicchieri (2002) discusses several social dilemma experiments in
which the subject of conversation was relevant for the participants
(for example, a rise in college tuition), but the behavioral result was
not different from a no-communication condition. The data she
discussed show that only communication about the experiment have
an effect on cooperation.
8
In the control condition, no communication was allowed.
8 C. Bicchieri, A. Lev-On
123
Author's personal copy
communication conditions, in which the majority of par-
ticipants sent their entire endowment of 6 USD.
When we conducted a simultaneous regression of trust
on control, communication relevance, and medium, Ftests
revealed significant effects of control (i.e. communication
vs. no-communication) (F(1,93) =4.47, P=.037) and
relevance (F(1,93) =8.54, P=.004), but not of medium
(CMC vs. FtF) (F(1,93) =1.56, P=.22). Moreover, there
was no interaction between communication medium and
relevance (F(1,92) =0.01, P=.92). Controlling for the
other variables, first-movers had over five time greater odds
of sending each dollar when communication was relevant.
Relative to the other conditions, participants had three
times lower odds of sending each dollar in the control
condition (Bicchieri et al. 2010).
Next we wanted to determine whether expected reci-
procity predicted the first-mover’s level of trust.
9
The
estimated coefficients in Table 2show that trust increases
with expected reciprocity. When the first-mover expected
nothing to be returned, the predicted amount sent was only
.36 dollars. For each percent of the amount sent that the
first-mover expected to be returned, the first-mover sent an
additional .10 dollars. Thus, the median expected reci-
procity (proportion) of .45 resulted in a $5 increase in the
amount returned. The R
2
value of 0.79 indicates that a large
percentage of the variance in trust is explained by the
expected percentage reciprocity.
These results suggest that the behavior of first-movers
was strongly determined by their expectations of second-
movers’ reciprocation. The variable most conducive to
creating such expectations was—paraphrasing McLuhan
(McLuhan and Fiore 1967)—not the medium, but rather
the message. Though the richness of the medium of com-
munication (FtF vs. CMC) failed to produce significant
differences in first-mover investments, such investments
were significantly higher following unrestricted commu-
nication than restricted or no communication. Recall that
unrestricted communication could include strategic dis-
cussion of the game and promise-making; according to our
transcripts, all subjects who participated in the unrestricted
communication were involved in both.
The medium of communication, however, had an effect
on reciprocity. The pattern of second-mover return was
highly bi-modal, with participants returning either nothing
or exactly half of the maximum (i.e., returning 9 out of 18
USD). This pattern depended on the conditions, with
almost all second-movers returning 9 USD in the FtF-rel-
evant condition, and almost all returning nothing in the
control condition. Although this pattern was partly due to
the different levels of trust across conditions, it was also
clear that second-movers behaved qualitatively differently
across (medium of) communication conditions, even after
accounting for first-movers’ levels of trust. The probability
of returning each available dollar increased with the
amount trusted, but increased more rapidly for FtF than for
CMC. Our result is consistent with earlier findings that
individuals engage in positive reciprocation, especially
following verbal communication.
It should be noted, however, that actual reciprocity is
always lower than expected reciprocity, and this result is
constant across Trust games (Camerer 2003). Yet trusting
on the part of first-movers may be rational, insofar as
trusting acts as a signal, whose intended effect is to focus
the recipient on a reciprocity norm. If such a norm exists
and is shared, then it is rational to trust insofar as one
Table 2 Estimates for expected reciprocity (proportion) as a pre-
dictor of trust (N =64)
Variable Estimate SE T
Intercept 0.37 0.22 1.69
Expected percentage reciprocity 10.08 0.55 18.284
****
R
2
=0.78
****
P\.0001
0123 456
Control
Frequency
0.0 0.4 0.8
0123456
FtF-Relevant
0.0 0.4 0.8
0123456
CMC-Relevant
Amount Sent
0.0 0.4 0.8
0123456
FtF-Irrelevant
0.0 0.4 0.8
0123456
CMC-Irrelevant
0.0 0.4 0.8
Fig. 1 Distribution of trust by communication medium and relevance
9
Expected reciprocity was converted into the expected percentage
reciprocity (the amount expected to be returned divided by the
amount available 9100), and then regressed trust on this variable
using ordinary least squares. This conversion was necessary to control
for the dependency of the maximum amount that could be returned on
the amount sent.
Studying the ethical implications of e-trust in the lab 9
123
Author's personal copy
believes that in so doing one will trigger reciprocation even
when the material incentives to reciprocate are absent
(Bicchieri et al. 2010,2011). The fact that second-movers
usually return less than expected may be explained by a
theory of norm compliance as conditional upon having the
right sort of expectations (Bicchieri 2006). In the anony-
mous environment, there is no risk of being punished, and
thus the pull of the norm, though present, is less strong.
However, when relevant pre-play communication is
allowed, second-movers are focused on reciprocation, and
promises do matter, especially when first-movers show full
trust.
The data presented above only apply to dyadic com-
munication. Yet communication often involves groups,
thus it remained to be seen if the effects of communication
relevance and medium hold when individuals are involved
in group discussion, be it face-to-face or computer
mediated.
Experiment 2: group vs. dyadic pre-play talk
In a second experiment (Lev-On, et al. 2010), the richness
of the communication medium and the size of the com-
municating group were manipulated. We hypothesized that
dyadic communication would be more conducive to trust
and reciprocation than group communication, since in
dyadic communication the players directly communicate
with their counterparts in the actual game that follows, and
their actions have a clear consequence for the other
player—as agents’ choices directly punish or reward a
single identifiable person. Also, a player’s promise to
invest or reciprocate is directed to the person with whom
they later play the game, thus triggering an additional
motive—guilt aversion—for players not to break their
promises, in spite of their cheap-talk status. Yet group
communication may also be conducive to trust. A multi-
player pre-play conversation may encourage subjects to
focus on public reasons and channel the discussion into a
cooperative path, although participants are eventually
paired with only a single person from the group, without
knowing in advance who this person will be.
For this new experiment, 60 college students were
recruited. Again, each session consisted of three sequential
trust games. For each game, the first-mover had 6 USD,
any dollar amount of which he or she could send to the
second-mover. The amount second-movers received from
the first-movers was tripled by the experimenter. The
second-mover could then send any dollar amount back to
the first-mover. As in the previous experiment, participants
were paired randomly with a different partner for each
game, and this was common knowledge. First-movers did
not receive feedback on the amount that the second-mover
returned until the end of the experimental sessions.
Prior to the first game, participants were not allowed to
communicate, making it a no-communication/control con-
dition. Prior to the second game, participants communi-
cated in real time via computer-based text chat for five
minutes with the person they were paired with or, in the
group condition, for ten minutes with a group of eight
people, one of which was the (anonymous) person they
were paired with in the trust game.
10
Messages entered by
each participant appeared in a chat window visible to all
group members. Prior to the third game, participants
communicated face-to-face for two minutes with the per-
son with whom they were paired or, in the group condition,
they had seven minutes to communicate with an eight-
person group.
11
Participants then returned to their com-
puter stations and made their decisions in the game
privately.
Table 3(Lev-On et al. 2010) summarizes the responses
for the five combinations of group size and medium. Both
group size and medium had large, positive effects on trust,
reciprocity, and expected reciprocity relative to no com-
munication. Dyadic, face-to-face communication had the
largest effects on all three variables, whereas dyadic,
Table 3 Mean trust, reciprocity, and first-mover expectations by communication medium and group size
No-communication FtF-dyadic CMC-dyadic FtF-group CMC-group
Trust 3.03 (.46) 5.57 (.46) 5.14 (.57) 4.12 (.69) 3.94 (.68)
Reciprocity 1.83 (.60) 7.57 (.96) 5.14 (1.33) 3.62 (1.11) 2.12 (.95)
Expected reciprocity 3.50 (.67) 8.36 (.69) 7.43 (.96) 4.31 (1.17) 5.00 (1.14)
Parenthesized values are standard errors of the mean
10
The instructions specified that participants were allowed to discuss
any topic except those pertaining to (1) their identities or (2) their
decisions or earnings from the previous condition, if there was one.
Note that we had to allow more time to group discussion, in order to
let each participant have their say and have time to read others’
messages.
11
Again, we let group participants have more discussion time, to
give each of them the possibility of expressing their opinion. Note
that the times allotted to both dyadic and group FtF communication
are less than the times allotted to CMC participants. This is due to the
fact that exchanging computer messages takes more time than direct
verbal communication.
10 C. Bicchieri, A. Lev-On
123
Author's personal copy
computer mediated communication had the second largest
effect.
Figure 2shows the distribution of trust across the five
communication conditions. To test for effects of medium
and group size, we conducted a simultaneous regression of
trust on the dummy variables no-communication, dyadic,
and FtF. F-tests revealed significant effects of communi-
cation (v
2
(1) =4.88, P=.03) and group size (v
2
(1) =
26.56, P\.0001), but not of medium (v
2
(1) =1.39, P=
.24). Moreover, there was no interaction between medium
and group size (v
2
(1) =1.10, P=0.29). Thus, trust levels
depended on the presence of communication and whether
that communication was in dyads or groups, but again they
did not depend on whether communication was face-to-face
or computer-mediated.
Next we checked whether expected reciprocity predicted
the first-mover’s level of trust. The results again demon-
strate (in Table 4) that trust increases with the expected
percentage reciprocity. When the first-mover expected
nothing to be returned, the predicted amount sent was only
1.27 dollars. For each percent of the amount sent that the
first-mover expected to be returned, however, the first-
mover sent an additional .88 dollars. Thus, the median
expected percentage reciprocity of 50% resulted in a $4.4
increase in the amount returned. Again, the R
2
value of
0.61 indicates that a large percentage of the variance in
trust is explained by the expected percentage reciprocity.
Our results show, again, that the behavior of first-mov-
ers is strongly determined by their expectations of second-
movers’ reciprocation. This time, the variable most con-
ducive to creating such expectations was not the medium of
communication, but rather the number of communicators.
Investments in the dyadic communication conditions were
significantly higher than in the group communication
conditions, which were in turn significantly higher than in
the no-communication condition. In the dyadic conditions,
almost all first-movers sent their entire endowment, com-
pared to only 60% in the group conditions and 30% when
there was no communication.
As to the behavior of second-movers, their pattern of
returns was bi-modal, again, with many participants
returning nothing or exactly half of the maximum (i.e., 0 or
9 USD). This pattern depended on the communication
condition. For example, almost all participants in the FtF-
Dyadic condition returned 9 USD, but almost all partici-
pants in the no-communication and CMC-Group conditions
returned nothing. When the first-mover sent the entire
endowment, reciprocity was greatest in the dyadic condi-
tions, with 80% of second-movers returning at least half of
the amount received, compared to only 42% in the group
conditions and 44% when there was no communication.
Across group sizes, when first-movers sent less than their
entire endowment, second-movers tended to send back
little. Although this pattern was due in part to the different
levels of trust across conditions, second-movers behaved
qualitatively differently across conditions, even after
accounting for first-movers’ levels of trust. The probability
of returning each available dollar increased with the
amount trusted, but increased more rapidly for the dyadic
conditions, and most rapidly for the FtF-Dyadic condition.
As a general rule, higher levels of trust, reciprocation, and
expectations of reciprocity were recorded in the dyadic
conditions, compared to the group conditions. Since a
promise to trust/reciprocate was far more common in dyadic
communications, this result is not surprising. In the group
condition, however, when promises to trust/reciprocate were
collectively made, trust and reciprocation were far more
frequent than in the control, no-communication condition.
To summarize: communication richness (FtF vs. CMC)
failed to produce significant differences in first-mover
investments.
12
This time, the size of the communicating
012 34 56
No-communication
Frequency
0.0
0.4 0.8
0123 5 6
FtF-Dyadic
0.0 0.4 0.8
0123 456
CMC-Dyadic
Amount Sent
0.0 0.4 0.8
0123456
FtF-Group
0.0 0.4 0.8
0123 456
CMC-Group
0.0 0.4 0.8
4
Fig. 2 Distribution of trust by group size and medium
Table 4 Estimates for expected percentage reciprocity as a predictor
of trust (N =90)
Variable Estimate SE T
Intercept 1.27 0.29 4.41
****
Expected Percentage Reciprocity 8.75 0.73 11.94
****
R
2
=0.61
****
P\.0001
12
This finding does not conform to other experimental results (see
Bochet et al. 2006; Brosig et al. 2003; Frohlich and Oppenheimer
1998; Bos et al. 2001; Rocco 1998; Zheng et al. 2002) that found
significant differences in cooperation rates between ‘richer’ and
‘poorer’ communication conditions. In these studies, however, the
number of communicators in a group remained constant.
Studying the ethical implications of e-trust in the lab 11
123
Author's personal copy
group did make a difference: The amounts sent were sig-
nificantly higher in the dyadic communication conditions
than in the group communication and no-communication
conditions. Our results suggests that, in addition to the
influence of communication media found in earlier exper-
iments, there are additional variables (such as the content
of conversation and the number of discussants) that miti-
gate the perception of the credibility of promises and
generate expectations and behavioral consequences. In a
group context, unless all members promise to trust/recip-
rocate, it is more difficult to establish expectations con-
ducive to support such behaviors. Furthermore, promises
are much more frequent in groups engaging in face-to-face
communication than in groups that communicate via
computer, thus explaining the differences in trust/recipro-
cation we found among these two conditions.
Discussion and conclusions: the ethical implications
of e-trust lab findings
The experimental results that were surveyed show that
communication matters, but the ‘‘communication effect’’ is
embedded within the normative background primed by
experimenters. When only communication is manipulated
(no-communication vs. communication), cooperation and
trust tend to be significantly higher in environments where
communication is allowed. But when additional variables
are manipulated, things get more complicated. In dyadic
communication, the medium of communication has almost
no effect on trust, and the relevance of what is communi-
cated has a major effect. Reciprocity, on the contrary, is
affected by the richness of the communication medium. In
group communication, group size and relevance of com-
munication matter to trust, whereas the richness of the
communication media did not matter. For reciprocity,
again, the richness of the communication medium mat-
tered, along with group size.
Our data show that, in addition to the influence of
communication media found in earlier experiments, there
are further variables (such as the content of conversation
and the number of discussants) that mitigate the perception
of the credibility of promises and affect expectations (with
behavioral consequences). Indeed, the available data show
the advantages of dyadic over group communication and of
relevant over irrelevant communication, over and above
the impact of the richness of the communication medium.
Communication has an impact on pro-social behavior
because, among other things, it conveys information about
the other party, and this information can be used to assess
the credibility, trustworthiness, etc. of those we interact
with. Information, however, has a darker side. It has been
shown (Lev-On 2009a) that people can use information to
discriminate against other groups—for example, first-
movers may send more or less money depending on the
ethnicity of second-movers. Thus, when group identities
are involved, it seems that anonymity can actually produce
more trust. It remains possible that limited information
leads to stereotyping, focusing people, in the absence of
additional trust-supporting cues, not on pro-social norms,
but rather on norms of in-group favoritism that impair
possible interactions with members of opposing groups.
One should therefore be particularly cautious in assessing
the interaction of information, communication, and the
priming of social norms.
The ‘‘communication effect’’, particularly in computer-
mediated environments, is a complicated and nuanced
phenomenon. Richer communication can be conducive to
cooperation, but at other times it may jeopardize trust. And
quite often, the communication effect is mitigated by other
variables that focus subjects on pro- (or anti-) social norms.
The medium of communication does not matter in and of
itself, but due to its ability to focus subjects on pro-social
norms (promise-keeping, reciprocity, cooperation), and to
facilitate (or hamper) the formation of mutual expectations
that support norm compliance. As the experimental results
show, subjects can be focused on norms in a variety of
ways, and communication richness is only one among
them. Communication relevance is crucial, as it allows
participants to discuss their situation and make mutual
promises. For example, a rich communication medium
(FtF) that does not allow subjects to make promises to each
other may be detrimental to trust, as our experimental
results show.
But there is good news as well. Managers, organizers,
moderators and software designers have a significant
degree of control over online communicative and deliber-
ative environments, which includes managing membership
and content (i.e. agenda setting, facilitating and encour-
aging discussions, moderating discussions, preventing
‘flaming’ and removing inappropriate posts, archiving old
threads of discussion), framing and enforcing policies
regarding accepted behaviors and sanctions, as well as
technical and financial management (Preece 2000).
Familiarity of site managers with experimental results,
such as the ones we presented, can motivate institutional
design that would lead participants to focus on and adhere
to pro-social norms. Cooperation and trust may indeed be
facilitated, survive and be continuously adhered to in such
pro-social environments.
The results we presented here have interesting online
institutional design implications. For example, think of
virtual teams or distributed, ad-hoc workgroups where
individuals cooperate with other team members whose
identity they may not know in advance. Since dyadic, and
not whole-group communication, seems crucial for
12 C. Bicchieri, A. Lev-On
123
Author's personal copy
accomplishing a group’s goal, our experimental data show
that a ‘motivational’ conversation with all workgroup
members is not a proper substitute for direct communica-
tion, whenever possible, with the person whom one should
eventually trust and with whom one should cooperate. A
group’s goal may be fruitfully divided into small tasks,
each assigned to few members who would then commu-
nicate among themselves.
An additional implication involves the content of com-
munication. Many online exchange sites allow some form
of communication between future exchange partners. But
these forms of communication vary widely. It is likely that
the closer the sites come to emulating face-to-face com-
munication, the more conducive they are to triggering
mutually-beneficial exchanges. On the other hand, when
sites do not allow subjects to convey to each other much
information relevant to their future exchanges (for exam-
ple, when they provide only information about identities
but not about past behaviors), they may disable the for-
mation of empirical and normative expectations of trust
and trustworthiness, and thus jeopardize the success of
computer-mediated exchange.
The experimental results we presented further support the
arguments of scholars like Lessig (2006) and Flanagan et al.
(2008), who claim that technological designs embed values,
which in turn have cognitive and behavioral implications.
Our data show that their arguments apply to the formation of
e-trust as well. In our case, it is evident that the design of
computerized environments, within which subjects form
trust judgments, can focus subjects on certain aspects of the
issue at hand, dimming or neglecting other aspects.
According to the focus theory of norms, the empirical and
normative expectations that are formed in such environ-
ments have significant behavioral implications (Bicchieri
2006).
The important questions to be asked are thus how much
and how far we may go in manipulating the information
that is shared or presented to online participants, and what
design features would help build on communication effects
to increase trust when it is appropriate, and to shape
appropriate expectations about members’ participation.
The experiments we performed concentrated on a number
of features that are typically present in trust-supporting
environments: the number of discussants, the topics of
discussion, and the information available—which are all
relevant to the achievement of trust, especially among
members of heterogeneous groups. Web managers and
designers interested in promoting trust and cooperation
may not only find the experimental information useful but,
we want to argue, may frame their design choices in light
of such information,which in turn raises interesting ethical
concerns and dilemmas.Let us conclude by commenting
on three such concerns.
The first design dilemma involves the question of how
much information about the behavior of participants should
e-trust site managers provide. Experiments show that people
respond in a negative (i.e., uncooperative) way when mat-
ched with known free-riders. When information is available
about behavior of other subjects who participate in the game,
subjects with high normative expectations may be faced
with information about low investment levels of other
players. As a consequence, such players will change their
empirical expectations about what other players will do,
adjust their contribution accordingly by diminishing con-
tribution rates, and set off a snowball effect of reduced or
null contributions (see Kurzban et al. 2001; Wilson and Sell
1997; Isaac et al. 1985; Bicchieri and Xiao 2009).
Site designers, then, face a tough choice. Should they, in
the name of transparency and openness and in the hope of
luring additional contributors, make past contributions
visible and information transparent, and provide indicators
about members’ contributions and production levels during
the production of continuous public goods? Should they
take the risk-averse path and hide information, in order to
avoid betraying the expectations of contributors, if indeed
the going gets rough? Or should they select a middle
ground, for example by selectively allowing some pieces of
information to expire, or by presenting or weighing them
differently? A utilitarian view would suggest that, if the
probability of enhancing contributions is sufficiently high,
it makes sense to conceal potentially damaging informa-
tion. On the other hand, this policy violates participants’
reasonable expectation to know past contribution patterns.
A second design concern has to do with how to present
information. To understand the importance of this point, let
us look at reputation systems—that is, systems established
to overcome trust problems by gathering inputs from large
numbers of exchange partners (Lev-On 2009b). Resnick
et al. (2000) argue that the way feedback and ratings are
displayed in such systems can have tremendous impact
over the exchange decisions of participants, as they focus
the attention on certain aspects of the potential partners that
may have been otherwise overlooked. Again, these design
decisions can push subjects toward, or away from, trust and
cooperation. We may ask whether it is fair to manipulate
information to the effect of inducing trust and cooperation,
and whether the final result justifies such ‘hidden’ control.
A simplistic view of trust formation would suggest that,
the more visible to each other the participants are, the more
likely it is that they will come to trust each other. Several
online methods may be used to enhance the public visibility
of participants, such as queues, discussion threads, and
sophisticated visualization techniques. Disseminating per-
sonal information about participants, however, raises further
ethical concerns. Let us focus on a simple method to enhance
visibility: showing pictures of participants. Experiments
Studying the ethical implications of e-trust in the lab 13
123
Author's personal copy
show that physical appearance matters to trust; for example,
Eckel (2007a,b) found a ‘beauty premium’ in an online trust
game that involved participants’ pictures. Wilson and Eckel
(2006), in another game involving pictures, found both a
‘beauty premium’ and a weak positive effect of smiling on
monetary transfers. Such findings pose a dilemma to website
managers that care about facilitating trust among members:
Should they allow users to upload photos? Since individuals
may base their trust judgments at least in part on cues
derived from the pictures, and associate such cues with
trustworthiness, they may come to trust or refrain from
trusting based on appearances, which may lead to unfore-
seen consequences. It is interesting to note that a similar
problem may exist in e-voting systems that allow candidates
to include their pictures alongside their names, and enable
voters to get a glimpse of the candidates as they vote. Such
institutional designs may enable attractive political candi-
dates to ‘monetize’ their ‘beauty premium’—a factor that
may have been neutralized to some extent in earlier and
‘poorer’ voting technologies.
In both cases, a genuine attempt to decrease social dis-
tance, by means of pictures or other information, may lead
people to easily trust someone they should not reasonably
trust or, conversely, may impair trust formation when in
fact the other person is perfectly trustworthy. Knowing
people’s cognitive biases may thus lead to a form of ‘soft
paternalism’, where hiding some information, manipulating
it, or even preventing its diffusion may be justified by the
greater good (in term of social cooperation) such actions
are expected to produce.
Our aim here is not to support particular design or policy
choices. Rather, we want to highlight how familiarity with
experimental results such as those we have described can
raise key ethical questions for those who manage and
design online trust-supporting environments. As we have
seen, such environments introduce a variety of practical
and ethical concerns that may have far-reaching conse-
quences for trust and cooperation. IT managers and
designers have a significant control over the environments
in which trust judgments are made. As their choices have
ethical implications, awareness of experimental results on
the determinants of trust and cooperation can serve not
only as guide for usability and sociability (Preece 2000),
but as an ethical compass as well.
Acknowledgment Authors wish to thank three anonymous referees
for several useful comments and insights.
References
Ben-Ner, A. & Putterman, L. (2006). Trust, communication and
contracts: Experimental evidence.’ Working Papers 2006-23,
Brown University, Department of Economics.
Bicchieri, C. (2002). Covenants without swords: Group identity,
norms, and communication in social dilemmas. Rationality and
Society, 14(2), 192–228.
Bicchieri, C. (2006). The grammar of society: The nature and
dynamics of social norms. New York: Cambridge University
Press.
Bicchieri, C., & Chavez, A. (2010). Behaving as expected: Public
information and fairness norms. Journal of Behavioral Decision
Making, 23(2), 161–178.
Bicchieri, C., A. Lev-On & Chavez, A. (2010). ‘The medium or the
message? Communication richness and relevance in trust
games.’ Synthese, 176(1), 125–147.
Bicchieri, C., & Lev-On, A. (2007). Computer-Mediated communi-
cation and cooperation in social dilemmas: An experimental
analysis. Politics, Philosophy and Economics, 6, 139–168.
Bicchieri, C., & Xiao, E. (2009). Do the right thing: But only if others
do so. Journal of Behavioral Decision Making, 21, 1–18.
Bicchieri, C., Xiao, E., & Muldoon, R. (2011). Trustworthiness is a
social norm, but trusting is not. Politics, Philosophy and
Economics (forthcoming).
Bochet, O., Page, T., & Putterman, L. (2006). Communication and
punishment in voluntary contribution experiments. Journal of
Economic Behavior and Organization, 60, 11–26.
Bos, N., Gergle, D., Olson, J. S., & Olson, G. M. (2001). Being there
versus seeing there: Trust via video. Proceedings of CHI 2001:
291–2. New York: ACM Press.
Bouas, K. S., & Komorita, S. S. (1996). Group discussion and
cooperation in social dilemmas. Personality and Social Psy-
chology Bulletin, 22, 1144–1150.
Brosig, J., Ockenfels, A., & Weimann, J. (2003). The effect of
communication media on cooperation. German Economic
Review, 4, 217–241.
Buchan, N. R., Croson, R. T. A., & Johnson, E. J. (2006). Let’s get
personal: An international examination of the influence of
communication, culture, and social distance on other regarding
preferences. Journal of Economic Behavior and Organization,
60, 373–398.
Camerer, C. (2003). Behavioral game theory: Experiments on
strategic interaction. NJ, Princeton: Princeton University Press.
Charness, G., & Dufwenberg, M. (2006). Promises and partnership.
Econometrica, 74, 1579–1601.
Cialdini, R. B., Kallgren, C. A., & Reno, R. R. (1991). A focus theory
of normative conduct: A theoretical refinement and reevaluation
of the role of norms in human behavior. In M. P. Zanna (Ed.),
Advances in experimental social psychology, v. 24 (pp.
201–234). San Diego, CA: Academic Press.
Dawes, R. M. (1980). Social dilemmas. Annual Review of Psychol-
ogy, 31, 169–193.
Dawes, R. M., McTavish, J., & Shaklee, H. (1977). Behavior,
communication, and assumptions about other people’s behavior
in a commons dilemma situation. Journal of Personality and
Social Psychology, 35, 1–11.
Eckel, C. C. (2007a). People playing games: The human face of game
theory (presidential address). Southern Economic Journal, 73(4),
841–857.
Eckel, C. C. (2007b). People playing games: The human face of game
theory. Southern Economic Journal, 73(4), 841–857.
Flanagan, M., Howe, D., & Nissenbaum, H. (2008). Embodying
values in technology: Theory and practice. In J. van den Hoven
& J. Weckert (Eds.), Information technology and moral philos-
ophy. Cambridge University Press: Cambridge.
Frohlich, N., & Oppenheimer, J. (1998). Some consequences of e-
mail vs. face to face communication in experiment. Journal of
Economic Behavior and Organization, 35, 389–403.
Ga
¨chter, S., & Fehr, E. (1999). Collective action as a social exchange.
Journal of Economic Behavior and Organization, 39, 341–369.
14 C. Bicchieri, A. Lev-On
123
Author's personal copy
Isaac, R. M., & Walker, J. M. (1988). Communication and free-riding
behavior: The voluntary contribution mechanism. Economic
Inquiry, 26, 585–608.
Isaac, R. M., McCue, K., & Plott, C. R. (1985). Public goods
provision in an experimental environment. Journal of Public
Economics, 26, 51–74.
Kiesler, S., Sproull, L., & Waters, K. (1996). A prisoner’s dilemma
experiment on cooperation with people and human-like com-
puters. Journal of Personality and Social Psychology, 70, 47–65.
Kinukawa, S., Saijo, T., & Une, M. (2000). Partial communication in
a voluntary-contribution-mechanism experiment. Pacific Eco-
nomic Review, 5, 411–428.
Kurzban, R., McCabe, K., Smith, V. L., & Wilson, B. J. (2001).
Incremental commitment and reciprocity in a real time public
goods game. Personality and Social Psychology Bulletin, 27(12),
1662–1673.
Ledyard, J. O. (1995). Public goods: A survey of experimental
research. In J. H. Kagel & A. E. Roth (Eds.), Handbook of
experimental economics (pp. 111–194). Princeton: Princeton
University Press.
Lessig, L. (2006). Code, version 2.0. New York: Basic Books.
Lev-On, A. (2009a). The virtues of anonymity: Evidence from trust
games. Manuscript.
Lev-On, A. (2009b). Cooperation with, without trust online. In K.
S. Cook, C. Snijders, V. Buskens, & C. Cheshire (Eds.), ETrust:
Forming relationships in the online world (pp. 292–318). New
York: Russell Sage.
Lev-On, A., Chavez, A., & Bicchieri, C. (2010). Group and dyadic
communication in trust games. Rationality and Society, 22(1),
37–54.
McLuhan, M., & Fiore, Q. (1967). The medium is the message. New
York: Bantam Books.
Orbell, J. M., van de Kragt, A. J., & Dawes, R. M. (1988). Explaining
discussion-induced cooperation. Journal of Personality and
Social Psychology, 54, 811–819.
Ostrom, E. (1998). A behavioral approach to the rational choice
theory of collective action- presidential address of the American
political science association 1997. American Political Science
Review, 92, 1–22.
Ostrom, E., & Walker, J. M. (1991). Communication in a commons:
Cooperation without external enforcement. In T. R. Palfrey
(Ed.), Laboratory research in political economy (pp. 287–322).
Ann Arbor: University of Michigan Press.
Ostrom, E., Walker, J., & Gardner, R. (1992). Covenants with and
without a sword: Self-governance is possible. American Political
Science Review, 86, 404–417.
Pillutla, M. M., & Chen, X. (1999). Social norms and cooperation in
social dilemmas: The effects of context and feedback. Organi-
zational Behavior and Human Decision Processes, 78, 81–103.
Preece, J. (2000). Online communities: Designing usability and
supporting sociability. New York: John Wiley.
Resnick, P., Zeckhauser, R., Friedman, E., & Kuwabara, K. (2000).
Reputation systems. Communiations of the ACM, 43, 45–48.
Rocco, E. (1998). ‘Trust breaks down in electronic contexts but can
be repaired by some initial face-to-face contact’. Proceedings of
CHI 1998: 496–502. New York: ACM Press.
Sally, D. (1995). Conversation and cooperation in social dilemmas.
Rationality and Society, 7, 58–92.
Schmitt, P., Swope, K., & Walker, J. (2000). Collective action with
incomplete commitment: Experimental evidence. Southern Eco-
nomic Journal, 66, 829–854.
Schroeder, D. A., Jensen, T. D., Reed, A. J., Sullivan, D. K., &
Schwab, M. (1983). The actions of others as determinants of
behavior in social trap situations. Journal of Experimental Social
Psychology, 19, 522–539.
Wilson, R. K., & Eckel, C. C. (2006). Judging a book by its cover:
Beauty and expectations in a trust game. Political Research
Quarterly, 59(2), 189–202.
Wilson, R. K., & Sell, J. (1997). Liar, liar…cheap talk and reputation
in repeated public goods settings. Journal of Conflict Resolution,
41, 695–717.
Zheng, J., Veinott, E., Bos, N., Olson, J. S., & Olson, G. M. (2002).
Trust without touch: Jumpstarting trust with initial social
activities. Proceedings of CHI 2002: 141–6. New York: ACM
Press.
Studying the ethical implications of e-trust in the lab 15
123
Author's personal copy
