Content uploaded by Darren Gergle
Author content
All content in this area was uploaded by Darren Gergle
Content may be subject to copyright.
Being there versus seeing there: Trust via video
Nathan Bos, Darren Gergle, Judith S. Olson, Gary M. Olson
Collaboratory for Research on Electronic Work (CREW)
School of Information, University of Michigan
701 Tappan Suite E2420
Ann Arbor, MI USA 48019
+1 734-647-7730
{serp, dgergle, jsolson, gmo}@umich.edu
ABSTRACT
We studied the emergence of trust in a social dilemma
game in four different communication situations: face-to-
face, video, audio, and text chat. Three-person groups did
30 rounds of a social dilemma game and we measured
trust by the extent to which they cooperated vs. competed.
The face-to-face groups quickly achieved cooperative
behavior, while the text chat groups continued to compete
throughout. The video groups achieved the same levels of
trust as the face-to-face groups, although perhaps a bit
more slowly. The audio group was intermediate. These
results show that trust can emerge through mediated
communication.
Keywords
Trust, social dilemmas, communication, media
INTRODUCTION
There is increasing interest in the nature of trust over the
Internet. The organizational consultant Charles Handy [1]
has received widespread attention for his strong claim that
“trust takes touch.” Does it? A study by Rocco [2] showed
how powerful even brief direct contact can be. She found
that groups meeting over email could not develop enough
trust to reach the optimal outcome in a social dilemma,
whereas groups meeting face-to-face did so easily and
quickly. However, in a third condition, groups that
interacted via e-mail developed trust if they had a brief
face-to-face meeting prior to the experiment.
Our research sought to extend Rocco’s results. We asked
whether other forms of computer-mediated-communication
(CMC) might lead to the formation of trust. In particular,
we looked at groups doing a social dilemma task who
interacted over video, over phone conferencing, and via a
text chat. All these media are more interactive than e-mail,
but less “rich” than face-to-face interactions.
We expected that video might be sufficiently rich to allow
trust to emerge in a social dilemma game. Audio only was
less clear, since it clearly offers fewer cues than video.
Chat
seemed as if it would be close to e-mail, though it is a bit
more interactive than the latter. So our range of conditions
offers a way to assess the character of communication
needed for trust to emerge in this setting.
METHOD
Forty-five three-person groups played a game called
Daytrader, where they periodically communicated with
other players via one of four media: face-to-face,
videoconference, phone conference, or in an Internet
chatroom.
The Daytrader game is in the general class of games called
social dilemmas. Well-known social dilemmas include the
Prisoner’s Dilemma and the Problem of the Commons.
Social dilemmas are defined as situations where the best
interest of the group as a whole conflicts with the best
interest of each individual, so that if each looks out only
for oneself, all lose. In a multi-round social dilemma, the
maximum group benefit accrues only when the group
develops a level of trust, and each individual agrees to act
cooperatively.
In the Daytrader game, players are given 30 tokens each
round of the game and must decide whether to invest as an
individual or risk investing with the group. The individual
investment yields a guaranteed doubling of the payoff
every round. In the group investment all three players’
contributions are lumped together, multiplied by 3, and
then distributed back evenly. So an individual can receive
triple from the group when everyone contributes, but also
risks losing when other players hold back. There was also
a bonus of 90 tokens is given every five rounds to
whichever player made the most money in the previous
five rounds. This bonus has the effect of giving a windfall
profit to players who contribute less than other group
members. (When players invest identical amounts for
five rounds, i.e. cooperate, the bonus is split.) Groups play
30 rounds of this game in total, with discussions held
every five rounds via one of four media channels.
Subjects for this experiment were mostly students or others
affiliated with the University. All were unacquainted
before the study, and never interacted with each other
except via their particular experimental condition. Subjects
were paid according to how well they did in the game,
with each ‘token’ worth 1 cent, and each participant
guaranteed to make at least $15.
RESULTS
We used the total payoff earned by a group to measure the
degree of cooperation – the higher the total group payoff
the more the cooperation. By implication this measures
trust.
FTF Video Audio Text Chat
Communication Condition
5500
6000
6500
7000
7500
Mean Group Payoff
_
_
_
_
Figure 1. Average total group payoffs in four conditions
ANOVA showed a significant difference among the four
conditions (F(3,41) = 6.54, P < .002). The means and
deviations for the four conditions were: FTF n=9, mean=
7599, s.d.=211; Video n=14, mean=7531, s.d.=305; Audio
n=13, mean=7262, s.d.=520; Chat n=9, mean=6859,
s.d.=645. The four conditions produced results ordered as
we predicted, with face-to-face the most cooperative and
quick to build trust, followed by video, audio, and text chat
conditions. A post-hoc comparison using Bonferroni’s test
(regarded as a conservative test) found significant
differences in f-t-f vs. chat and video vs. chat.
Figure 2. Round by round group contribution averages
across four conditions
Figure 2 shows the trial-by-trail performance of the
groups. The top line (FTF) indicates that trust in the face-
to-face condition forms quickly, and continues throughout.
Video also converges at the maximum, but takes somewhat
longer, and the audio (phone) condition takes longer still,
with more variation throughout. The text chat average
does not converge on the maximum at any point, (although
some individual chat groups did cooperate fully by the
end).
Another noteworthy feature of figure 2 are the vertical
dropoffs occurring on a 5 round cycle, with a final dropoff
at the end. The dropoffs are the results of defections
within the game, typically where one player violates an
agreement and the other players rapidly retaliate. The tops
of the spikes are the rounds immediately after discussions,
when cooperation has been re-established. The dropoff at
the end, which is visible in all four conditions, is the result
of last-round defections at the end of the game. A t-test
comparing rounds 29 and 30 showed a large significant
difference (P<.000) between the group investments in
these rounds.
Post-questionnaires on group trust confirm the differences
between the conditions. A 10-item scale measuring the
trustworthiness of the group members (alpha = .96)
including items such as “the other players in the game
could be trusted” was significantly different between
conditions (F(3, 142) = 8.16, P<.000).
DISCUSSION
Videoconferencing may be as good as face-to-face for
building trust. In this experiment video was
indistinguishable from face-to-face, and both were
significantly better than text chat. Although we cannot yet
statistically separate the phone condition, it appears that
phone is somewhere in between text chat and video for
trust building.
It does appear that trust forms more slowly in mediated
conditions. Figure 2 shows that it took video and audio
some time to ‘catch up’ with the face-to-face groups.
Groups using mediated communications seem more likely
to form partial agreements (e.g. “lets all contribute 15,
then 20…”) rather than fully cooperate from the
beginning.
The defections on trial 30 suggest that the kind of trust
that was formed in all of these conditions may be
vulnerable to opportunistic behavior, even when face-to-
face. Late-round defections are a well-known phenomena
in social dilemmas. Our data may allow more insight into
when these are likely to occur, and how communication
affects it. Additional analysis will examine both slow trust
and defection events.
In sum, we have demonstrated that trust can emerge in
mediated conditions. These results are a major extension
of the work reported by Rocco [2], with application for
many forms of distributed collaboration. But our data
suggest that this kind of trust may be fragile. Additional
experimental studies can help clarify this. Field studies
will also be useful to examine whether these mediated
interactions can lead to enduring trust in real
organizations.
Acknowledgements
This work was supported by National Science Foundation
grant IIS 9977923 to the Olsons.
References
[1] Handy, C. (1995) Trust and the virtual organization.
Harvard Business Review. 73(3), 40-50.
Round Number
302826242220181614121086
Mean Group Payoff
280
260
240
220
200
180
160
140
120
FTF
Video
Audio
Text Chat
[2] Rocco, E. Trust breaks down in electronic contexts but
can be repaired by some initial face-to-face contact, Conference proceedings on human factors in computing
systems, 1998, pp496-502.
