424
allocation. the prediction from noncooperative game theory was that
subjects would invest according to the nash equilibrium – 8 tokens each
for a total of 64 tokens. subjects could earn considerably more if they re-
duced their allocation to a total of 36 tokens in the common-pool resource.
subjects in baseline experiments with multiple decision rounds substantially
overinvested – they invested even more tokens than predicted, so the joint
outcome was worse than the predicted nash equilibrium.
6
Building off prior public goods research (isaac and Walker 1988), we then
conducted a series of face-to-face communication experiments in which
the same payoff function was retained. after an initial ten rounds without
communication, subjects were told they could communicate with each other
in a group setting before returning to their terminals to make their own
private decisions. this provided an opportunity for “cheap talk.” the same
outcome was predicted in these experiments as in the baseline since a subject
could promise to cooperate but no external “third party” ensured that the
promise was fulfilled.
subjects used face-to-face communication to discuss strategies to gain the
best outcomes and then to agree – if possible – on what each subject should
invest. they learned about their aggregate investments after each round,
but not the decision of individual subjects. this gave them information as
to whether the total investments were greater than agreed upon. in many
rounds, subjects kept their promises to each other. in other rounds, some
defections did occur. subjects used information about the aggregate invest-
ment levels to scold their unknown fellow subjects if the total investment was
higher than they had agreed upon. the opportunity for repeated face-to-face
communication was extremely successful in increasing joint returns. findings
from communication experiments are consistent with a large number of
studies of the impact of face-to-face communication on the capacity of sub-
jects to solve a variety of social dilemma problems (see e. ostrom and Walker
1991; orbell, van de kragt, and dawes 1988; sally 1995; Balliet 2010).
in many field settings, resource users have devised a variety of formal or
informal ways of sanctioning one another if rules are broken, even though
this behavior is not consistent with the theory of norm-free, complete
rationality (elster 1989: 40–41). it was thus important to see if subjects in
a controlled experimental setting would actually use their own assets to
financially punish other subjects. after subjects played ten rounds of the
baseline common-pool resource game, they were told that in the subsequent
rounds they would have an opportunity to pay a fee in order to impose a
fine on another subject. We found much more sanctioning occurred in this
6 in simple, repeated public goods experiments, subjects initially tended to contribute at a higher level than
predicted by the nash equilibrium (isaac et al. 1984, 1985, 1994; isaac and Walker 1988, Marwell and ames
1979) and outcomes slowly approach the predicted nash equilibrium from a higher level. in common-pool
resource games, on the other hand, subjects initially achieved outcomes that were much worse than the
nash equilibrium that they then slowly approached from below (see also casari and Plott 2003).
425
design than the zero level predicted.
7
subjects did increase gross benefits
through their sanctioning but substantially reduced net returns due to the
overuse of costly sanctions.
8
sanctioning was primarily directed at those who
defected, but a few sanctions appeared to be directed at low contributors as
a form of revenge by those who had fined themselves. in a further design,
subjects were given a chance to communicate and decide whether or not
to adopt a sanctioning system of their own. subjects who decided to adopt
their own sanctioning system achieved the highest returns achieved in any of
the common-pool resource laboratory experiments – 90 percent of optimal
after the fines related to the small number of defections were subtracted
(e. ostrom, Walker, and Gardner 1992).
the predictions of noncooperative game theory are roughly supported
only when participants in a laboratory experiment do not know the
reputation of the others involved in a common-pool resource dilemma
and cannot communicate with them. on the other hand, when subjects
communicate face-to-face, they frequently agree on joint strategies and keep
to their agreements – substantially increasing their net returns. further,
communication to decide on and design a sanctioning system enables those
choosing this option to achieve close to optimal returns.
B. Studying Common-Pool Resources in Field Experiments
a series of field experiments have now been conducted by colleagues in
colombia to assess whether experienced villagers who are dependent on
resources make decisions about the “time spent in a forest” in a design that
is mathematically consistent with those reported on above. cardenas (2000)
conducted field experiments in rural schoolhouses with over 200 users of
local forests. he modified the design of the common-pool resource experi-
ments without, and with, face-to-face communication so that villagers were
asked to make decisions regarding “harvesting trees.” the outcomes of
these experiments were broadly consistent with the findings obtained with
university students.
in a different design, cardenas, stranlund, and Willis (2000) ran ten
rounds of baseline experiments with resource users from five villages who
were then given a chance to communicate face-to-face for the next set of
experiments. in five additional villages, participants were told after the
baseline rounds that a new regulation would go into force that mandated them
to spend no more than the optimal time in the forest each round. the prob-
ability of an inspection was 1/16 per round – a low but realistic probability
for monitoring rule conformance in rural areas in developing countries. if
the person was over the limit imposed, a penalty was subtracted from that
7 see henrich et al. (2006) in which field experiments were conducted in multiple countries testing whether
a much broader set of participants would also use punishments in public goods experiments. see
also henrich et al. (2004) for the reports of earlier field experiments of social dilemmas in fifteen small
communities.
8 similar findings exist for public goods experiments where punishers typically punish low contributors
(yamagishi 1986; fehr and Gächter 2002).