Beruflich Dokumente
Kultur Dokumente
http://rss.sagepub.com/
Published by:
http://www.sagepublications.com
Additional services and information for Rationality and Society can be found at:
Subscriptions: http://rss.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://rss.sagepub.com/content/16/3/319.refs.html
Werner Raub
ABSTRACT
Rationality and Society Copyright & 2004 Sage Publications. Vol. 16(3): 319–365.
www.sagepublications.com DOI: 10.1177/1043463104044682
not need to recover the costs of training and schooling. One way
of solving this problem is a contract between the employer and the
employee stipulating that the employee has to pay at least part of
the costs of training and schooling if he quits prematurely.
Another way of solving the problem is non-contractual. The
employee moves and acquires real estate close to his job. Thus,
if he quits prematurely and accepts an offer from another and
far away employer, he would have to move again and incur the
financial as well as social costs associated with moving.
We consider trust problems in the sense that the trustee has incen-
tives to abuse trust and the trustor has something to lose if trust
is abused (Coleman 1990: chap. 5). A well-known model of trust
problems is the standard trust game as introduced by Dasgupta
(1988: 59–61) and Kreps (1990: 100–1). The game is played by two
players. Player 1 is the trustor and player 2 is the trustee. The trustor
moves first – Coleman stresses the importance of this feature: there is
a time lag between the action of the trustor and the action of the
trustee – and chooses between placing trust and withholding trust.
We denote the placement of trust by C1 (with C indicating ‘coopera-
tion’) and withholding trust by D1 (with D indicating ‘defection’).
The game ends if trust is not placed, with payoffs Pi (i ¼ 1; 2) for
Figure 1. Extensive form of the Trust Game with Uncertainty and Incomplete
Information (TGUI) with payoffs for player 1 ðR1 > P1 > S1 Þ and player 2
ðT2 > R2 > P2 Þ; 0 < 1; maxðð S1 þ ð1 ÞR1 P1 Þ=
ð S1 þ ð1 ÞR1 S1 Þ; 0Þ < 1; dotted line indicates information set.
unexpectedly not returned later on, and costs that result from
reduced flexibility to respond to changes in exogenous conditions
(see also the concluding discussion). These costs arise if and only
if the trustee decides to post the hostage and do not depend on
how the game develops after hostage posting. Hence, these costs
are not only due if the trustee loses his hostage (in this case, one
could consider the transaction costs simply as an ingredient of the
value K2 of the hostage for the trustee) but also if trust is placed
and honored after hostage posting as well as if no trust is placed
after hostage posting. As an example, consider the car dealer’s trans-
action costs associated with providing a guarantee. Here, a core
element of his costs are the dealer’s (increased) costs of running a
garage so that repairs of cars under guarantee can be carried out.
Alternative assumptions on transaction costs are conceivable. For
example, Snijders (2000) employs the assumption that costs of
hostage posting arise if and only if the hostage is ‘accepted’ by the
trustor in the sense that she places trust (thus, no transaction costs
arise for the trustee if a hostage is posted but no trust is placed).
Our assumption on the transaction costs of hostage posting some-
what simplifies the analysis of the model, while using Snijders’s
assumption would not yield substantively different results.
A core assumption of our model is that the transaction costs of
hostage posting can differ between the two types of trustees. On
the one hand, allowing for differences in transaction costs seems
realistic when considering the examples. The costs of running a
garage that are related to the repair of cars under guarantee
depend on whether or not the dealer in used cars has lemons in
house. In the example of the employee who posts a hostage by
moving close to his job, an important ingredient of the transaction
costs involved in posting the hostage is due to undermining the
employee’s bargaining position vis-à-vis an alternative employer
offering a new position. These transaction costs emerge by definition
for an employee who is unreliable in the sense of our model, while a
reliable trustee does not have to incur these costs. While the assump-
tion that transaction costs of hostage posting can differ between dif-
ferent types of trustees seems to be empirically plausible, differences
in transaction costs are also interesting from our theoretical per-
spective of trust based on hostage posting. After all, one expects
from signaling theory (Spence 1974) that the signaling function of
hostages depends on differences in signaling costs for different
types of trustees. Transaction costs associated with hostage posting
Figure 2. Extensive form of the Hostage Trust Game with Uncertainty and
Incomplete Information (HTGUI) with payoffs and hostage value for player 1
ðR1 > P1 > S1 ; K1 Þ and player 2 ðT2 > R2 > P2 ; K2 Þ, and transaction costs
r and u for reliable and unreliable trustee (player 2);
0 < 1; maxðð S1 þ ð1 ÞR1 P1 Þ=ð S1 þ ð1 ÞR1 S1 Þ; 0Þ < 1; dotted
lines indicate information sets.
Lemma 4: Assume that best reply behavior implies that trust will not
be abused after hostage posting by the trustee and placement of trust
by the trustor. Generically, best reply behavior implies that the trustor
places trust after hostage posting if and only if
been posted, and the unreliable trustee abuses trust if no hostage has
been posted and trust is placed. This equilibrium is weakly Pareto-
dominated by the separating equilibrium.
5. Discussion
the hostage for the trustee is only required if the underlying trust
problem involves uncertainty ( > 0) so that the trustee may lose
his hostage even without opportunistic behavior.
It is helpful to relate these results to our examples of trust prob-
lems and of hostages as a mechanism to induce trust. We conjecture
that the guarantee as a hostage in the case of buying and selling used
cars is a relatively valuable hostage for the car dealer (K2 is high) so
that the hostage serves binding purposes. Likewise, the guarantee is
valuable for the buyer (K1 is high) and provides positive compen-
sation. This would be consistent with uncertainty involved in such
transactions and a relatively high likelihood of unfavorable con-
tingencies. We guess that the guarantee does not serve signaling
purposes. It seems that guarantees are typically provided for second-
hand cars that are not too old and therefore not likely to need
frequent and expensive repairs. Restricting guarantees to relatively
new cars would be consistent with an upper bound on the trans-
action costs of hostage posting as well as an upper bound on the
value of the hostage for the dealer. The deposit as a hostage in the
car rental example – typically a relatively minor amount – may
bind the client to abstain from reckless driving, assuming that the
incentive for reckless driving is mild (it would be interesting to see
if the deposit for sports cars is typically higher than the deposit
for family cars with a similar price tag). The hostage seems to pro-
vide rather low compensation for the trustor (it would be interesting
to see if the deposit varies between regions with regional differences
in, say, overall accident rates). The transaction costs associated with
posting the hostage are minor and presumably hardly differ between
different types of clients so that the hostage hardly serves signaling
purposes. Finally, consider hostages provided by an employee as
safeguards for investments of the employer in the training and
schooling of the employee. A contract stipulating that the employee
has to reimburse the employer if he quits prematurely presumably
serves binding purposes and also provides some compensation for
the employer. The hostage ‘moving and acquiring real estate close
to the job’ does not provide compensation for the trustor. This
hostage seems to serve not only binding but also signaling purposes.
After all, the transaction costs of posting the hostage differ severely
between employees who have or expect an outside offer and those
who do not have and do not expect such an offer.
Testable Predictions
Consider testable predictions for laboratory experiments. Obviously,
such experiments require explicit assumptions on how material
incentives provided by the experiment relate to utility functions of
subjects. In an experimental design, HTGUI could be implemented
by using a random device that determines the type of the trustee at
the beginning of the experiment in such a way that the trustor knows
the relevant probabilities but cannot observe the outcome produced
by the random mechanism. An approach for deriving predictions for
experiments is to vary some parameter of HTGUI that is interesting
from a substantive perspective so that different theorems on trust via
hostage posting apply. To illustrate, consider a first scenario with
r ¼ u and parameters of HTGUI so that the theorem on the pool-
ing equilibrium applies. Now compare the first scenario with a
second that differs only in that u is increased so that condition (9)
for the separating equilibrium is fulfilled. One would then predict
less hostage posting by unreliable trustees in the second scenario,
no difference with respect to placement of trust by the trustor
after hostage posting, and also no difference with respect to honor-
ing trust by unreliable trustees. Another promising approach for
experimental tests is to vary the probability of unfavorable con-
tingencies. For example compare the theorems on the pooling and
the separating equilibrium with Corollary 1 and consider that all
equilibria of HTGUI are such that both types of trustee do not
post a hostage and the trustor withholds trust if compensation is
too low (more precisely, compensation is too low if (7) is not ful-
filled). It follows that positive compensation should have a stronger
effect on hostage posting as well as on inducing trust by hostage
posting if the underlying trust problem involves a high probability
of unfavorable contingencies ( is large) than if one removes uncer-
tainty from the underlying trust problem ( ¼ 0). Research in this
direction seems to be particularly promising because previous experi-
mental evidence on hostages as a mechanism of cooperation in
dilemmas such as the trust game and the prisoner’s dilemma (Raub
and Keren 1993; Snijders 1996: chap. 6; Snijders and Buskens 2001)
seems to indicate that – empirically – positive compensation through
hostage posting has quite a strong effect on trust and cooperation.
It is easily seen that additional predictions for experimental tests
of the model can be generated by systematically varying one of the
parameters of the model so that HTGUI ‘moves’ between the sets
Research Agenda
Various questions on trust via hostage posting could and should be
tackled in future research. Some of these questions are inspired by
features of the examples that are not or not fully covered in our
model. First, consider the ‘expropriation hazard’ (Williamson
1985: 177) and the loss of flexibility (Becker 1991: 12-13) associated
with hostage posting. We have used the assumption that only the
trustee and not the trustor faces opportunities and incentives to
abuse trust. Our examples show, however, that this need not be
true. In the car rental example, a conflict between the rent-a-car
firm and the client may arise and the firm may keep the deposit of
the client strategically. Also, consider the employee’s hostage post-
ing by moving close to his job. This hostage is not subject to the
expropriation risk since the hostage as such is not valuable for the
employer but posting the hostage reduces the flexibility of the trustee
if contingencies change. Such changing contingencies can depend on
‘moves of Nature’ rather than on incentive-guided behavior of the
employer. An example is the risk that the firm goes bankrupt due
to market fluctuations rather than bad management of the employer
as owner/manager. However, contingencies may also change due to
strategic behavior of the employer after the employee’s hostage post-
ing. For example, when hiring the employee, the employer may offer
not only investments in future training and schooling but may also
promise future general policies that are attractive for the employee.
Such promises are often non-contractual and not legally binding or
enforceable. If the employee commits himself by moving, he loses
flexibility to react to the employer’s future deviations from the
promises. In the model developed here, we have included the expro-
priation risk and the expected costs of loss of flexibility in the trans-
action costs associated with hostage posting. This seems satisfactory
for risks that depend on moves of Nature rather than strategic
behavior of the partner. A more appropriate analysis that takes
the strategic behavior of the other player into account would require
to model hostage posting as a mechanism of cooperation in more
complex dilemma situations with possibilities for opportunistic
behavior from more than one player. Reciprocal hostage posting
by both players is an obvious mechanism for mitigating risks from
two-sided opportunistic behavior (see Williamson 1985: chap. 8 for
examples; Weesie and Raub 1996 for formal model building using
games with certainty and complete information; and Raub and
Keren 1993 for some experimental results).
While the expropriation risk and the loss of flexibility are
problems associated with hostage posting, one should not overlook
possible further benefits of hostage posting in addition to promoting
trust. One such benefit that merits closer analysis is that posting
the hostage may make the relation between trustor and trustee more
productive. For example, the employee who moves close to his
job may not only promote investments of the employer in the
employee’s training and schooling. Moving close to his job, more-
over, may imply an additional contribution to the employee’s
productivity and job satisfaction since he benefits from reduced
commuting time, better (opportunities for) contacts and interaction
with colleagues, etc. It should be easy to show that ‘productivity’ of a
hostage in this sense implies that the conditions for trust based on
hostage posting become less restrictive since posting the hostage
also increases the payoffs Ri from trust that is placed and honored.
A variant on this theme is that hostage posting by the trustee is
associated with costs for the trustor. For example, the used car
with a guarantee comes at a higher price for the buyer. This would
reduce R1 after a hostage has been posted. In this scenario, the con-
ditions for trust based on hostage posting become more restrictive.
We have motivated an analysis of hostage posting as a mechanism
of trust with the observation that dyadic embeddedness and network
embeddedness can be insufficient for stabilizing trust. Our model,
however, is one without these forms of embeddedness rather than
one in which dyadic and network embeddedness are too ‘weak’ to
stabilize trust without using hostages. For example, the dealer in
used cars may (want to) sell used cars infrequently but nevertheless
repeatedly to the same buyer or there may be some information
exchange about the dealer among buyers. Accounting for the com-
bined effects of learning and control through dyadic and network
embeddedness and of opportunities for hostage posting through
institutional embeddedness remains a challenging task for further
model building (see Weesie et al. 1998 and Raub and Snijders 2001
for some work in this direction).
A final feature of our examples that is not yet modeled in HTGUI
is hostage selection. The employee can post a hostage by moving to
his job or by signing a contract stipulating that he pays back training
and schooling costs of the employer after a quit. When to choose one
or the other of these hostages, when both? It seems reasonable to
assume that the transaction costs associated with posting different
hostages as well as the value of the hostage will be important factors
that affect the trustee’s decision but a more thorough analysis seems
useful (see Snijders 2000 for first steps in the analysis of the hostage
selection problem).
NOTE
I am indebted to Jeroen Weesie for comments on this paper and earlier joint work on
hostages as a mechanism of trust and cooperation. Comments by Marcel van Assen,
Davide Barrera, Vincent Buskens, Thomas Gautschi, Chris Snijders, and seminar
participants at Utrecht University are gratefully acknowledged. Comments by
anonymous reviewers were helpful when revising the paper. The research on which
the paper is based is part of the PIONIER-program ‘The Management of Matches’
of the Netherlands Organization for Scientific Research (NWO; grant PGS 50-370).
A first draft was written at the Department of Sociology, Berne University. The hos-
pitality of Andreas Diekmann and other colleagues in Berne was much appreciated.
REFERENCES
Akerlof, G. A. 1970. ‘The Market for ‘Lemons’: Quality Uncertainty and the Market
Mechanism.’ Quarterly Journal of Economics 89: 488–500.
Arrow, K. J. 1974. The Limits of Organization. New York, NY: Norton.
Axelrod, R. 1984. The Evolution of Cooperation. New York, NY: Basic Books.
Bacharach, M. and D. Gambetta. 2001. ‘Trust in Signs.’ In Trust in Society, ed.
K. S. Cook, pp. 148–84. New York, NY: Russell Sage.
Barber, B. 1983. The Logic and Limits of Trust. New Brunswick, NJ: Rutgers Univer-
sity Press.
Barrera, D. and V. Buskens. 2003. ‘Imitation and Learning under Uncertainty:
A Vignette Experiment.’ Utrecht. Mimeographed.
Becker, G. S. 1964. Human Capital. New York, NY: NBER, Columbia University
Press.
Becker, G. S. 1991. A Treatise on the Family, enlarged edn. Cambridge, MA: Harvard
University Press.
Burt, R.S. and M. Knez. 1995. ‘Kinds of Third-Party Effects on Trust.’ Rationality
and Society 7: 255–92.
Buskens, V. 2002. Social Networks and Trust. Boston, MA: Kluwer.
Buskens, V. and W. Raub. 2002. ‘Embedded Trust: Control and Learning.’ Advances
in Group Processes 19: 167–202.
Calvert, R. L. 1995. ‘Rational Actors, Equilibrium, and Social Institutions.’ In
Explaining Social Institutions, eds J. Knight and I. Sened, pp. 57–94. Ann Arbor,
MI: University of Michigan Press.
Camerer, C. and K. Weigelt. 1988. ‘Experimental Tests of a Sequential Equilibrium
Reputation Model.’ Econometrica 56: 1–36.
Coleman, J. S. 1990. Foundations of Social Theory. Cambridge, MA: Belknap Press of
Harvard University Press.
Dasgupta, P. 1988. ‘Trust as a Commodity.’ In Trust: Making and Breaking Coopera-
tive Relations, ed. D. Gambetta , pp. 49–72. Oxford: Blackwell.
Esser, H. 1993. Soziologie. Allgemeine Grundlagen. Frankfurt: Campus.
Frank, R. 1988. Passions Within Reason. New York, NY: Norton.
Granovetter, M. 1985. ‘Economic Action and Social Structure: The Problem of
Embeddedness.’ American Journal of Sociology 91: 481–510.
Hostage Game for the Trust Game with Uncertainty and Incomplete Information
(HTGUI)
H þ, H Trustee’s decision to post (H þ ) or not to post (H ) a
hostage
Ki Value of the hostage for player i ; K2 > 0
r ; u Transaction costs of hostage posting for reliable (r )
and unreliable (u ) trustee; r ; u > 0
a hostage in the equilibrium, the trustor would place trust after hos-
tage posting and would not place trust if no hostage has been posted.
Then, however, the unreliable trustee would be better off posting a
hostage rather than not posting a hostage, which would be incon-
sistent with the assumption that the unreliable trustee randomizes
with respect to his hostage posting decision. To summarize, the
equilibrium different from the pooling equilibrium described in the
theorem has to be such that the unreliable trustee randomizes with
respect to his hostage posting decision, the reliable trustee posts
no hostage at least with positive probability, and the trustor places
trust after a hostage has been posted.
Next, we observe that if an equilibrium different from the pooling
equilibrium exists, the trustor randomizes in that equilibrium after
no hostage has been posted: after no hostage posting, she places
trust with probability and withholds trust with probability 1
ð0 < < 1Þ. This can be seen as follows. If the trustor always
places trust after no hostage posting, the supposed equilibrium strat-
egy of the unreliable trustee could not be a best reply since he would
be better off by posting no hostage. If the trustor never places trust
after no hostage posting, both types of trustee would not behave
optimally with respect to their hostage posting decisions. Note
that the trustor randomizes in the equilibrium after no hostage has
been posted if she is indifferent in the equilibrium between placing
and withholding trust after no hostage posting. This requires that
condition (3) is fulfilled so that an equilibrium different from the
pooling equilibrium cannot exist if (3) is violated.
Denote the expected payoff for the unreliable trustee in the equi-
librium after no hostage posting by Uu ðH Þ and his expected payoff
after hostage posting by Uu ðH þ Þ. We must have
Uu ðH Þ ¼ T2 þ ð1 ÞP2 ¼ Uu ðH þ Þ
¼ ð1 ÞðR2 u Þ þ ðR2 K2 u Þ
and thus ¼ ðR2 K2 u P2 Þ=ðT2 P2 Þ It is easily seen that
0 < < 1. Also, denote the expected payoff for the reliable trustee
in the equilibrium after no hostage posting by Ur ðH Þ and his
expected payoff after hostage posting by Ur ðH þ Þ. We must have
Ur ðH Þ ¼ R2 þ ð1 ÞP2 Ur ðH þ Þ
¼ ð1 ÞðR2 r Þ þ ðR2 K2 r Þ
trustee. Since he uses a best reply strategy, it must be optimal for the
unreliable trustee not to post a hostage, given the equilibrium
strategies of the trustor and of the reliable trustee. By deviating
from his strategy and posting a hostage, the unreliable trustee
could obtain T2 K2 u if he abuses trust and (1 ÞðR2 u Þ þ
ðR2 K2 u ) if he honors trust after hostage posting and place-
ment of trust. The existence of the separating equilibrium implies
that P2 – the payoff of the unreliable trustee if he follows his equili-
brium strategy – is higher than both these alternative payoffs.
Assume now T2 K2 u > ð1 ÞðR2 u Þ þ ðR2 K2 u Þ.
This implies the first condition in (9) and the existence of the separ-
ating equilibrium then also implies the second condition in (9).
Assume T2 K2 u < ð1 ÞðR2 u Þ þ ðR2 K2 u Þ. This
implies the first condition in (10) and the existence of the separating
equilibrium then also implies the second condition in (10).
Finally, we have to show that (9) as well as (10) imply that the only
other equilibrium next to the separating equilibrium is the additional
equilibrium described in the theorem. Consider first properties of
an equilibrium if (9) holds. Using the Lemmas, the first condition
in (9) implies that the unreliable trustee abuses trust after no hostage
posting and placement of trust but honors trust after hostage post-
ing and placement of trust. However, the second condition in (9)
implies that the unreliable trustee prefers no hostage posting and
no placement of trust to hostage posting, placement of trust and
honoring trust. Thus, (9) implies that the unreliable trustee posts
no hostage. Assume now that the trustor places trust after hostage
posting. In this case, the best reply of the reliable trustee is to post
a hostage due to the third condition in (9). In that case, placing
trust after hostage posting would indeed be optimal for the trustor
due to the fourth condition in (9) and we get the separating equili-
brium. Assume, however, that the trustor withholds trust after
hostage posting. In that case, the best reply for the reliable trustee
would be not to post a hostage and we get the alternative equili-
brium described in the theorem. Note that hostage posting is not
an element of the equilibrium path of play induced by the alternative
equilibrium. This completes the proof of the theorem (note that it
has already been indicated in section 4 that the separating equili-
brium is a weak Pareto-improvement compared to the alternative
equilibrium).