Professional Documents
Culture Documents
Reviewed work(s): Source: The Journal of Conflict Resolution, Vol. 52, No. 5 (Oct., 2008), pp. 665-686 Published by: Sage Publications, Inc. Stable URL: http://www.jstor.org/stable/27638633 . Accessed: 04/12/2012 10:57
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Conflict Resolution.
http://www.jstor.org
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Journal of Conflict Resolution Volume 52 Number 5 October 2008 665-686 ? 2008 Sage Publications
Rational
Behavior
Massimiliano
Rational -
io-h ?
*
!*,?
Sender-Receiver
Landi School ofEconomics Singapore Management Domenico Colucci
in a Binary Choice
http://jcr.sagepub.com
^,^,^1
University, Singapore
Laden
The
weekend before the 2004 U.S. presidential elections, the Al Jazeera TV network aired a videotaped address fromOsama bin Laden (OBL, henceforth)
the American people. Apparently the message was not about the impend
targeting
talk of mine
the war
is for you and concerns the ideal way and its causes and results.
to prevent
The timing of the message and its very ending though, leave littledoubts about the fact that itwas indeedmeant to influence theAmerican voters. Quoting again from the message (Al Jazeera 2004):
Authors' Note: We thank Vincent Crawford and Domenico Menicucci for discussions and comments
on this article.
665
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
666
doesn't
House.
of Kerry,
and children for the sake of the squander his security, wealth, In conclusion, I tell you in truth, that your security is not in nor Bush, nor al-Qaida. No. Your security is in your own hands. to ascertain several reelect whether or not the message agreed Bush.1 For on had the
ways
However, helped
commentators George W.
actually
exam
in the balance, the campaign some improbable awaited hanging to tip the scale. Re-enter Osama bin Laden. By reminding us of 9/11 and invoked the only thing that could trump Iraq?and terrorism, bin Laden the President. His chilling reappearance reminded us of our peril, put Iraq in per
to the President's and played precisely success and strength?success and was never one to in Baghdad. Bin Laden strength that he so squandered remotely the American mind. He understand 9/11?and he pulled his spectacularly misjudged nemesis over the finish line. (p. 50) spective
A piece of (admittedly indirect) evidence supporting the above view can be found in Cohen et al. (2005): in late September 2004, the authors conducted an experi
ment,
and the recollection of the facts of 9/11 would tend to substantially displace votes fromKerry to Bush. This is corroborated by pre-electoral polls that found Bush to be clearly favored over Kerry in handling terrorism (Campbell 2005). These find sortie in the closing days of the campaign might have ings suggest thatOBL's affected the vote, helping President Bush's reelection. So we take it as a working hypothesis that the tape indeed influenced ballot
showing
that a shift
in the public's
attention
onto
themes
such
as
terrorism
derstand theAmerican voters?Was his implied position on U.S. election (opposing Bush and endorsing Kerry) a lie or a truthful revelation of his view? Did theAmeri
can voters OBL's by give message any credit is an example to the message? to influence of an attempt the result of an election a costless The number of cases in which message. using
behavior. The interpretationof ourmodel also benefits from assuming that the face value of OBL's words boiled down to opposing George W. Bush and hence endor model. sing JohnF. Kerry, but this is not strictlyrequired to build the we answers to seek the Did OBL really misun Ultimately, following questions:
an external
significant outside personalities or organizations endorse (more or less explicitly) one candidate is surely large. An example is the Economist's tough stand on Mr. Berlusconi (pictured as unfit to lead Italy) before both the 2001 and the 2006
Italian
personality
national television in May 2006, of one of the candidates for the incoming presi dential elections inMexico. Nevertheless, OBL's role and his message's timing, media make it a rather novel event in the political scenery content, and echo in the
general
elections.
Another
is
subcommander
Marcos's
endorsement,
on
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
667
and calls, in our opinion, for an explanation that should be anchored on empirically credible initial responses' behavior.2 We consider a game ? la Crawford and Sobel (1982) played between one sender
and a population of receivers. take The a sender has private information a about a state vari from able while the receivers binary action after observing public message
the sender.Majority rule determines the choice implemented, which, togetherwith the prevailing state, determines the payoffs for all players. In the benchmark case, Under this setup, it iswell known that sender and receivers have opposite interests. choose the action that, ex ante, maximizes
case, OBL's message could not have shifted in any equilibrium, the sender uses only uninformative messages of Bush. and the receivers
To account for this possibility, we introduce a structural model of interaction and behavioral that resembles rational Crawford's (2003) between fully players to the mislead Germans of the Allies' successful about where they attempt analysis would land on D-Day in 1944.While Crawford considers the transmission of infor
mation about intended
information. The uncertainty for both receivers and sender about the degree of sophistication of their opponent induces a game of incomplete informationwhose The interpretationof the events relies crucially on the relative likelihood ofmortal and sophisticated players. Nonetheless, in all the pure strategies equilibria of the game, the outcome most preferred by the (rational) sender is selected. These find ings extend the results of Crawford to our case. With respect to thementioned contest between OBL and U.S. voters, if a sufficientdeparture from full rationality
of the electorate message is assumed, succeeded then our model the race supports toward the idea that OBL's he preferred. pr? ?lectoral in tilting the outcome equilibria are studied as a function of the parameters governing this uncertainty.
play,
we
focus
on
the
transmission
of private
exogenous
mind can The strategic interactionbetween OBL and theU.S. voters we have in relative be explained as follows: Let the state of the world represent OBL's on as a influence Islamic Mid leader and his communities in the terrorist strength dle East and worldwide. This can be seen, from the point of view of theAmerican
electorate, voters
as a proxy
seriousness
of OBL's
imminent
threat
to
icy. In a "low"
on create
state, OBL
masses
is relatively weak
This makes the angered
Islamic
is poor:
conditions
that drive
U.S.
probably
as a better
one might thinkof resorting to thehistorical events as of 2004 Election Day as trary, a guide to shed lighton things,but this is hardly a way to avoid difficulties,because
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
668
itwould require some kind of counterfactual analysis for which scientific grounding is hard to find.
At no the time of the message, was on the game natural candidates' In this only positions respect, OBL were we and spelled that the U.S. out time for changes to focus available. believe and clearly, it is a sensible
assumption
between
voters.
We
choose tomodel
seems quite
unlike similar applications of strategic information transmission, such as Sobel (1985) and Benabou and Laroque (1992), the novelty of the scenario does not leave
room for
approach
learning
or
reputation
considerations.
The
same
lack
of
repetition Rational
rules
play the incomplete information game that is obtained by introducing the possibi lity of naive players and follow the standard equilibrium behavior. Naive players are built following the logic of level-k thinkingmodels that appear in the beha vioral game theory literature (see, e.g., Stahl andWilson 1994, 1995). In a nutshell, players form some simplified beliefs about their opponents' play and then best lar concepts that are typical of equilibrium reasoning. Different levels of sophisti cation imply different types of expectations. Thus, at the bottom of the scale, we
have case, level 0 players, whose behavior who is determined reveals the level 0 player is a sender an exogenous In our motive. by the true state of the world. One step respond to that. These players are naive in the sense that they do not use the circu
players
above is the class of level 1 receivers: They believe the population of sender to be formed by level 0 players only and therefore choose the strategy that is a best
response infinite to that. Moving sequence one step above in a recursive since we way, have we can generate game, only an a of behavioral types. Clearly, a finite
matters. it is only theirbehavior that Such a structure of boundedly rational decision rules has been shown to be a well-grounded starting point for modeling initial responses to novel situations. behavior in experi Examples beyond theworks previously cited include first-round mental guessing games, as in Nagel (1995) and Costa-Gomes and Crawford
(2006), and estimated
finite number of them behave differently.There is no loss of generality in restrict ing our analysis only to them, since from the point of view of the rational players,
Gomes, Crawford, and Broseta (2001). The related model of cognitive hierarchy developed by Camerer, Ho, and Chong (2004) also shows good fit to empirical data
from various
types
in a series
of two-person
normal-form
games
in Costa
with possibly boundedly Related research on transmission of private information rational players includes Kartik, Ottaviani, and Squintani (2006), and Chen (2006).
The
experimental
games.
pool of receivers (who may be partially naive) inwhich talk can be inherentlycostly.
The state of the world is a continuous random variable with an unbounded support.
first group
of authors
considers
a communication
game
between
a sender
and
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
669
Chen is similar but has a state that is uniform in [0, 1] and assumes nonstrategic types of players who are in fact optimizers given theirutility functions (so there is actually no bounded rationality in her model). Besides the finiteness of the state,message, we allow for and action spaces, the present article differs from both, in the sense that
a more Moreover, complex structure in Chen, the outcome rational of boundedly is not determined types for both sender rule. and receivers. by majority
that, of
determined
tively separable in the actions of different groups within the population of recei
vers.
by majority rule
payoff
Majority
per
change
the outcome
with
respect
to the one
The role of the sophisticated receivers mirrors the equilibrium play of the unin formed independents in Feddersen and Pesendorfer (1996), who mix between abstaining and voting for the candidate with less partisan support, to increase the probability that the informed independents be pivotal to the election. Yet, there is a substantial difference: The behavior of the sophisticated players in our model is When the prob purely driven by their strategic interactionwith their counterpart. a is sender the of sufficiently high, ability sophisticated sophisticated receivers
need which to vote requires so that the electoral them outcome does not depend of the mortal on the signal observed, to counterbalance the play receivers.
Our article is mainly motivated by the interest in the episode of OBL versus voters at the eve of 2004 elections and thereforeposits a clear clash in the interests
of those actors. (some Because events whether of which the analysis are mentioned payoff and might above) are be relevant where negatively voters may or to interpret other similar not be certain about correlated, we
the sender's
theirs
also extend our model to allow formore examples from real politics to fit in our framework. This part is akin to Sobel (1985) and entails that the receivers face an additional level of uncertainty in choosing what to do. The article is organized as follows: Section 2 discusses themodel in its general
form. relation Section between 3 extends sender the model and to account Section for uncertainty 4 concludes. about the payoffs cor receivers.
positively
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
670
Figure 1 the Actions Given (Um and Dm) of the Payoffs Majority of Receivers and the State of Nature (Two States), with b > 0. pi p^
pn
Urn
b-b
0, S^
pn
Urn Dm
0,
1,-1
receivers, who in turnchoose an action aeA ? {U,D}. The payoffs for all players depend on the state of nature and on the action of (themajority of) the receivers and are reported inFigure 1.
Notice that, in each state, the sender and the receivers have
A pure strategy for the sender is a function from the state of theworld space (S) to themessage space (M). A pure strategy for the receivers is a function from the message space (M) to the action space (A). So both types of players have four pure strategies. In what follows, (mi, m2) denotes the pure strategyby which message m\ is sent if the state is Si and message m2 if the state is S2. Similarly, (U, D) denotes the pure strategy that prescribes choosing action U if themessage is m\ and action D if the message ism2. Mixed strategies are probability distributions over the set of pure strategies. We
use
conflicting
interests.
three-dimensional simplex with respect to the set of pure strategies {(mi, mi), (mi, ^2), (^2, ^1), (^2, ^2)}. Similarly, for receivers, (?]? ?2, ?3, ?4) denotes an ele ment of the three-dimensional simplex with respect to the set of pure strategies {(?/, t/),(i/,D),(D,t/),(D,D)}.
For assigns (m2, mi), example, probabilities and the sender's 0, 0.5, mixed 0.1, and strategy 0.4 (0, 0.5, 0.1, 0.4) means that the sender (m}, m2), to the pure strategies (mi, mi),
the following
notation:
an element
of the
(m2, m2).
Finally let jidenote the posterior probability that the true state of theworld is S\ after observing message mj and r\ the probability that the state be Si given m2. These probabilities depend on p and on the strategy adopted by the sender, and are
computed, whenever possible, through Bayes' rule as follows:
+a3)
(1) (2)
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
671
In the remainder of this section,we firstdescribe the equilibria of the game with only rational players, then introduce bounded rationality along the lines of Crawford (2003) and look at the equilibria of the game between the sophisticated players.
We shall assume
veyed, the expected payoff fromchoosing either alternative in the election is the same.
We have in mind a scenario in which, ex ante, voters are indifferent between the can
that ph?\?p,
which
means
that, before
the message
is con
didates, whom are thereforeequally plausible to be elected.We use this condition as an approximation of the idea of a close race, as it was the case in the 2004 presidential
election. (i.e., where still pb?\?p Finally, be In this respect, for whom ours one is a model option of swing than voters. One could add partisan could voters voters citizens is better the other, irrespective a close election independent case, of the realiza be one
would of
assume
that
their
expected payoff as if theywere pivotal in the election. This restriction ismotivated by the need to reduce the set of equilibria, which is typically large in voting
games.5 brium. The The solution symmetry concept condition we work on all is that of the rational symmetric receivers sequential to adopt requires equili the same
(pure ormixed) strategy. Under full rationality, the following result (which is very much in the spirit of Crawford and Sobel 1982) holds: Proposition
receivers, true state
are such that the sender the strategies = choose 0C3) and the receivers (i.e., a2
(i.e., ?j + ?2 >
either
oc2 ^
0C3 or
0C3. A
best
receivers' posterior probabilities on the state equal the priors (see equations 1 and 2).
Therefore, receivers are indifferent between U and D. Suppose that on observing m\,
(rri2, m\).
respectively ra2, theymix in such a way to induce U (i.e., ?j + ?2 > 1/2), respec tivelyD (i.e., ?i -f?3 < 1/2):The sender's best response would be (m2,m\), which
violates a2 = 0C3. Likewise, we can dismiss the possibility that receivers mix
so
that ?i + ?2 < 1/2 and ?1 + ?3 > 1/2.Having ruled out receivers' strategies indu cing outcomes that change with themessage, we are leftwith three possibilities:
= and ?, + ?3> 1/2; and ?, + ?3< 1/2; ?i + ?2= l/2 ?,+ ?2> 1/2 ?, + ?2< 1/2
> < ?2 + ?4. Finally, observe that?, + ?3 1/2 if and only if ?2 + ?4
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
672
cases,
is indifferent a2 = 0C3.
between
sending
any message
in each
state and
there
Proposition
ests, receivers
at the voting stage is not affected by it.Thus, given the initial uncertainty about the
best way possible receivers action, vote. the message Their choice, sent does with not have any impact whatsoever remains in the or without any message, the same.
sent by OBL
frameworks.
toral outcome,
following: Players are assumed to be unsure about the degree of rationality of their opponents, who may be fully rational (and we call them "sophisticated") or follow a
behavioral norm of conduct (and we call them "mortal"). Mortal
beliefs about the distributionof opponents and maximize theirexpected payoff given
those beliefs. Iteration of best level responses 0 who sender expect generates always the sender the set of behavioral chooses (mi, m2). We to be a types we call con sider. We "truth assume The that a this type a therefore
players
have
naive
teller."
receivers
optimally choose (U, D) are called "believers." A senderwho thinks thepopulation of receivers ismade only of believers and thereforeoptimally chooses (m2,m\) is
called a "liar." Last, For the receivers who believe the sender is a liar and therefore
mally choose (D, U) are called "inverters." Notice thatother pairs of behavioral types
could emerge.
opti
would optimally choose (mi,m2), and receiverswho believe the sender does so would optimally choose (<7, ?>), but their strategies are indistinguishable from the firstpair of mortal types (truth tellers,believers). As a result, there is no loss of generality in
focusing our attention on the following pairs of types of sender-receivers:6
example,
sender
who
believes
that receivers
are
all
inverters
liar, or a sophisticated
type. Similarly,
the population's
considerations,
the structureof the game (payoffs, strategy,and uncertainty about types) is common knowledge (among sophisticated players). To understand how sophisticated players will behave in equilibrium, we need the following two intermediate results: Lemma 1. Letting w; be the probability that the sophisticated sender sends mes sage mi in state S?, then a sophisticated receiver's best response is given by (U,D) (D,U) ifxs(l+wi -w2)>l-2xt, ifxs{\+ wi - w2) < 1- 2xt.
given
that
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
673
Proof. A sophisticated receiver will choose the action maximizing his or her expected payoff. Upon observing message m\, the probability of being in state S] and S2 is proportional to, respectively, p(xt + w\xs) and (1 ?p)(x\ + W2XS). There
fore, since ph?\? p, U is optimal under m\ if Xt+ W\Xs >X\ +w2xs, that is, if a;v(1 + w\ ? w2) > ? 1 2xt.
Similarly, upon observing message m2, the probability of being in state S\ and S2 ? is proportional to, respectively, p(xt + (1 w\)xs) and (1 ? p)(x\ + (1 ? W2)xs). Therefore, U is optimal under m2 if
Xi+(\-W\ ? that is, if 1 )xs > xt + (1 w2)xs, ? w2)xs.
2xt >
(1 + w\
Observe that the only pure strategybest responses for the sophisticated receiver are (U, D) or (D, U). Therefore, let ? denote the probability of the sophisticated recei ver playing (?/,D) and 1? ? theprobability of playing (D, U). Lemma 2. The sophisticated sender's best response to (0, ?, 1? ?, 0) is given by (mi,m2)
(m2,mi)
< if ?jy,
if $ys>
2
-
~yb
-yh
Proof. The sophisticated sender knows that (U, D) will be played by a share ? + (1 ?)v?). + fys) of thepopulation, whereas (D, U) will be played by a share (y? (yb
Thus, in state S\, message m\ is best if
?>'v<^<^-
+ (l-?)Xv,
(3)
yi+ (l-P)ys<
Last, indifference between m\ and m2
is obtained
= l;
is best.
Therefore, the only possible configurations are (mi, m2) and (m2,mj), and the former is optimal whenever
in state S\m_?
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
674
We
are now
tial equilibria of the game between sophisticated players. The following proposition
covers all
in the position
to characterize
strategy
sequen
parameters'
configurations
except
case
1/2
max{x?,
receivers, EQ EQ EQ EQ
Furthermore, the outcome preferred by the sophisticated sender is achieved in all those equilibria.7 Proof. The proof uses Lemma 1 and Lemma 2 above. For example, for [(mi, which m2), (?/, D)] to be an equilibrium means (mi, m2) is best given ?=l, = < > that is and that is best 1,w2 = 0, y? (?/,D) requires ys given w\ 1/2, 1/2 y&, < cases are same > that is The other done in the x/ \?2xu way. implying 2x5 1/2. Proposition 2 shows that the (sophisticated) sender can take advantage of the
uncertainty, whether it is about the receivers or about himself or herself. Whenever
the population share of one mortal type of receiver is larger than 1/2 (equations 1 most preferred outcome by fool and 4), the sophisticated sender can just induce the are acting optimally, they the the receivers ing majority. Although sophisticated
cannot affect the outcome because they represent a minority of the voters.
On the other hand, when sophisticated receivers count (equations 2 and 3), and there is a probability of at least 1/2 that the sender is a mortal type, the sophisticated sendermimics the strategyof the least likelymortal and fools the sophisticated recei vers. Here, the majority of the receivers are acting optimally, since theprobability of a Yet, the outcome induced is still the facing sophisticated sender is smaller than 1/2. most preferredby the sophisticated sender.Last, observe that,from thepoint of view of a mortal sender, the outcome favors a liar only in Equations 2 and 4, while it favors the truthteller inEquations 1 and 3.
It remains to consider the case where
is inmixed strategy,as a quick glance at the best response functions suggests. The result is as follows: > mixed strategies sequen Proposition 3. Assume 1/2 maxj^, x?, y?,yb}. In the tial equilibria, sophisticated senders need to make sure that no transmission of
1/2
>
max{x,,
equilibrium
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
675
occurs an
(i.e., electoral
= oc2
^j^1 outcome
the not
receivers depend
choose
any
strategy (i.e.,
*(?i ^(?i
+ ?2) > V2 -yb> y s (?2 + ?4)> or ys (?i + ?2) < = + ?2)=V2-^ v,(?2 + ?4)).
1/2
on
the message
Proof Inspection of the best response functions depicted in Figures 2 and 3 shows that,when sophisticated players are pivotal, the equilibrium has to be in mixed strategies. Therefore, the probabilities about the state of nature have to be left unchanged by the message, for otherwise nonequilibrium behavior would
result. Note that
~
? Tft
+ _p[x, (ai +0C2K]_ + + + (1 -p)[xi + (<xi p[xt (cLi cl2)xs] +oc3)x,]' + p[x? fa + U4)XS] ' + a4)xj + (1 - p) [xt + (a2+ a4)xj p[xi+ (oc3
if and only if xt + U2XS = x\+ oc3x5.
As
a result,
= |i
r|
For the sophisticated receivers, itmust be that the electoral outcome does not depend on themessage observed (otherwise, the sender would have a profitable deviation), which means that the share of the population choosing U must exceed, fall short,or equal 1/2, irrespective of themessage. Thus, we need one of the fol
lowing cases:
> -,
z
l/2-^<^(?2 >
+ ?4).
Observe that in equilibrium, the sophisticated receivers need tomix in a way that (if the electoral outcome is tobe U), they mostly add to thebelievers when m 1 (so thatys when m2 (so that ys (?2 + ?4) + yb< 1/2). (?i + ?2) + yb > 1/2) and to the inverters In fact, this equilibrium resembles the one under full rationality (see proposition 1). First, when the shares (probabilities) of mortal players are equal, the two equilibria coincide. This is because the twomortal typesperfectlyoffset each other. Second, the equilibrium under full rationalityis the limitof theequilibrium with mortal and sophis
ticated
In general, the equilibria change continuously with respect to the composition of = 1) Suppose, for example, that the sender is (known to be) sophisticated (i.e., xs a and plays against population of mortal and sophisticated receivers. The equili
brium the players' population. Thus, our analysis can be carried forward to polar cases.
players,
when
to 1.
stitutes a majority, the sender plays the strategy that fools this type while
play
depends
on
the share
of mortal
receivers.
When
either mortal
type con
the
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
676
Best Responses
Sender
y*=i
y.=i
tf=1
sophisticated receivers will play like the other type (see equations 1 and 4 in propo
sition 2). If, on down the other to oc2= hand, the sophisticated receivers are in majority, to be we
are
with themixed strategyequilibria of proposition 3 inwhich the sender's strate left plifies the strategic interaction. The introduction of behavioral types causes the appearance of pure strategy are rational, inwhich the equilibria that do not exist when all players considered
outcome framework favors the (sophisticated) the happenings views sender. We can see how our analysis offers to a simpli a to evaluate of 2004 U.S. elections according gies boil 0C3. So assuming a sender who is known rational sim
fied model of the game that actually took place between OBL
voters. Depending on one's on the nature of OBL's
and theAmerican
various dif
ferent conclusions can be drawn. For example, ifwe think thatOBL was not only capable of mastering the game but also perfectly knowledgeable about American politics (in other words, ifwe believe OBL was very likely to be fully rational), thenwe can regard the final comment inKrauthammer (2004) thatOBL played against himself throughhis message as unrealistic. Indeed, even without attaching (very likely) sophisticated sender entail the deception of a majority ofmortal recei vers, either believers (in which case OBL was lying) or inverters (in which case
any specific interpretation to states and actions, the pure-strategy equilibria with a
rationality,
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
677
1 2
On the other hand, ifone believes that, likeKrauthammer (2004) suggests, OBL was naive about American politics, thenwe should look at the equilibria inwhich the sender is most likely a liar or a truth teller: Under those circumstances, the
sophisticated fooled by American the message, voters whatever would have avoided their mortal counterparts to be the true state of the world.
So far, our model depicts the case inwhich the interaction between sender and
receivers is clear cut, with As a
2004 message,
we point out that voters they framed are
perfect
negative
payoff
correlation.
Yet,
one
can
think
and-white
to be
(perhaps in a wider
Therefore,
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
678
Figure 4 the Actions Given (Um and Dm) of theMajority Payoffs of Receivers and the State ofNature (Four States) P1 F2 Un
1,-1
P1
1,1
P1 0,0 1,1
P1
0, S,
0,
1,-1
0,0
be the probability of therefore, that the state space is S = {S\, S2, S3, S 4} and let/?/ M = {m\,m2}, and a strategy for the sender is a state Si. The message space is still function from S to M.9 Receivers choose an action, aeA = {U,D}, and a strategy M toA. Notations for pure and mixed strategies for the receivers is a function from
follow pure the logic strategy of the previous the sender's section. message For example, (m\, m\, m2, m2) S\ and S2 denotes the where is mi in states and m2 in states
message is S3 and S4. In turn, (U, D) means that the receiver chooses action U if denote the posterior belief about mi, and action D if message ism2. Last, let^(r^) state Si when signal mi (m2) is observed. Remember that these probabilities depend on pi and on the strategy adopted by the sender and are computed by Bayes' rule
whenever Players' possible. payoffs as a function of the states and of the receivers' actions are
reported inFigure 4. Compared to the previous model, we have added complexity to the payoffs
structure, but we have restricted attention to the case where both alternatives
give
(5)
further that either one of the follow
(6) (7)
Inequality 6 means that the sender ismore likely to have positively correlated pay offs with the receivers (we call him a "friend"), whereas inequality 7 means the sender ismore likely to have negatively correlated payoffs with the receivers (we call him an "enemy"). Moreover, note that inequalities 5 and 6 imply thatP3> p\ whereas inequalities 5 and 7 imply thatP3 < p\ and P2< Pa and p2 > /?4,
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
679
This model
sender Given
is similar to Sobel
or an enemy, of M and
is a friend
as well
the nature
S, a strategy
corresponds two,
to messages
bria can be usefully distinguished in terms of the partitions they generate over the state space. To this end, we give the following definition: Definition 1. An equilibrium is said to be pooling if it induces P? and partially revealing if it induces Pl or P2. To begin with, consider the benchmark case of perfect rationality. Like in the base model, there are pooling sequential equilibria, inwhich the posterior beliefs
are
partitions
come thatdoes not depend on the signal received. Additionally, there cannot exist partially revealing equilibria that induce Pl. Sup
pose, egy for example, that there were When such an equilibrium observe where m\, the sender's pure strat is (mi, ra2, m2, mi). the receivers they know that the true state
equal
to the priors,
and
receivers
choose
a strategy
that induces
an electoral
out
is S\.Moreover, when m2 is sent, theyknow thatS2 is less likely to be the true state than S3 or S4, so theirbest response is (U, D). We now can see why the initial pro posed strategy for the sender cannot be an equilibrium strategy:He or she would
rather send m\ in state S4. In a similar vein, we can see
typePl cannot be part of a pure strategysequential equilibrium. Finally, a partially revealing equilibrium that induces Pl exists only when the states are partitioned into {Si, S3} and {S2, S4}, and the sender is a friend (i.e., con
dition
why
any other
partition
of the
best response is to choose (D, U) when the sender is a friend, and (U, D) when the
sender is an
6 holds).
In fact, In
suppose
the sender
chooses
(mi,
The
receivers'
enemy.
the former
case,
the
sender
does
to deviate
from
rather pick
partitions of
states inP2 cannot be supported as an equilibrium: Given the priors on these states, any strategy inducing P2 implies that it is optimal for the receivers to adopt either ([/,D) or (D, U). The only case inwhich the sender does not want to deviate is the
one shown above.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
680
This result parallels theorem 1 in Sobel (1995), which states that if the probabil ity of the sender being a friend is larger than 1/2, then there is an equilibrium in which themost preferred action for the sender is revealed with probability 1. Summing up, we have the following: Proposition
messages which
in which
the sig
are not
nal, there are partially revealing sequential equilibria, when condition 6 holds, in
pure strategies are as follows: 1 :[(ml,m2,m],m2),(D,U)}, 2 :[(m2,ml,m2,m{),(U,D)}.
EQ EQ
These results show thatwhen condition 6 holds, a selection issue among multiple
(pure about We about
meaning and signals are valued only for the information language has no intrinsic
the states now their of the world to the case they convey. in which as players in the have the same type of uncertainty and distinguish turn
strategies)
equilibria
arise.
This
is common
in communication
games,
since
between the two polar cases described by inequalities 6 and 7. The Sender
This case
opponent's
rationality
previous
section,
is an Enemy
entails P4>pi and which p\ > P3. the By states iterating Consider again where best a baseline type to a truth-telling sender we prefers obtain of a sender in the the
(mi, m2, mi, m2), choosing sense that the strategy separates states where
corresponds
type, U
the
from
he or she prefers
D.
responses,
the follow
truth-teller liar
believer inverter
The
notation
about
probabilities ys)
types
xs)
receivers'
To characterize the behavior of the sophisticated players, letw? denote the prob ability that the sophisticated sender sends message m\ in state S? while ? is the probability a sophisticated receiver chooses (U, D). Then, Lemma 3. The best response for a sophisticated receiver is given by (U,D) (D,U) > (Xt if (xt + W2Xs)p2 + W]Xs)pi+ (x? + H'3Xy)p3 + (x? + W4Xs)p4, if (xt + w2xs)p2< {xt + WjXs)pi+ (Xj + W3Xs)pi + (Xi + W4Xs)p4.
types
in the previous
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
681
(m2,mum2,mi)
Proof. A sophisticated receiver will choose the action maximizing expected payoff. The posterior probability of being in state S, induced by message mi is pro e {2,4}; similarly, the G {1,3} and to (x? + W{Xs)pi if / + WiXs)pi if / portional to (xt state S, having observed message m2 is proportional to probability of being in ? e {2,4}. Using these G {1,3} and to xt+ (1 ? Wi)xs)p? if / X[+ (1 Wi)xs)pi if / to is + w\xs)p\ + optimal if (xt probabilities, it is straightforward verify that (?/,D) > while + wixs)p3 + (xi+ w4x5)p4 (?>, U) is optimal if the oppo (xt (x] + W2XS) p2 site inequality holds. Therefore, in equilibrium, the receiver will only play either (?7,D) or (D, ?7).As for the sender, the fractionof those expected to play (U, D) will ? fall behind thatof those expected to play (?>, U) if (yb + ?y5) < (yi+ (1 ?)yj; that
is when
is (m2,mi, m2, mi) when theopposite inequality holds. The following proposition characterizes the pure strategy sequential Nash equi libria of the game: Proposition 5. Suppose conditions 5 and 7 hold. Then in the sequential equili
bria of the game between sophisticated sender and receivers, pure strategies are as
+y?, ?_y5
<
1/2:
In such
case,
m2)
is the best
strategy, whereas
it
follows:
1: [(m\,m2,m\,m2), 2: 3: 4: l(m],m2,m\,m2), l(m2,m\,m2,m\), if and only if and only if and only if and only if y? > if yh < if y? < if yh > and x? < and x? > and xt> and xt < -. -. -. .
EQ EQ EQ EQ
[(m2,m\,m2,m.\),(D,U)]
with the equilibria in the two states case of the previous Note the high similarity
section. There is a difference,
anytime the outcome preferredby the sender is implemented. In particular, in states S2 and S3, letting the sophisticated sender determine the outcome helps the receivers as well. This is due to the fact that in those states the interestsof the players are the
same.
though:
It is not
true anymore
lose
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
682
us
now
turn
to the
equilibria
the X], yb, y,-}. For max{if, for all i. This condition r\i ?pi
strategies, to be
which
arise
whenever we need
uninformative,
W\ =-h
which
tion 3.10 Observe that the problem for the receivers is the same as before: They mix so that the electoral outcome is not affected by the public need to find the right signal. Therefore, theirequilibrium strategy is the same as in proposition 3. The role of the sophisticated players in themixed strategies equilibrium is ana logous to the previous model: They need to put more probability on choosing the action that is consistent with themortal typewhose probability is smaller. Like in the base model, if the sender is known to be sophisticated, then the structure of the equilibria depends on the relative size of themortal receivers. In fact, if either of them is large enough to constitute a majority, the partially revealing pure strat egy equilibria of EQ 1 and EQ 4 in proposition 5 apply. Otherwise, if none is large enough so that the sophisticated players are pivotal, thenwe have the pool ing equilibria.
Therefore, we between can conclude sender and that the introduction receiver does not of uncertainty alter the main about result correlation of the base in payoffs
corresponds
closely
to the condition
model
now
The Sender
is a Friend
> P4 and 773 The sender is a friendwhen condition 6 holds and therefore/?2 >p\.
If we assume
again
naive
sender
adopts
m2),
iteration
of
In this case, believers play (D, U) instead of (U, D) as in the rest of the article. This is because themost likely states are S2 and S3 inwhich choosing, respectively, U mortal liars and invertersdo not fit in the scene: This and D is best. Also notice that
is because,
response to a believer type is now telling the truth. Having a simpler structureofmortal types, the conditions describing the sophis ticated players' best responses simplify accordingly: Lemma 4. Let conditions 5 and 6 hold. The sophisticated receiver's best
maintaining
the same
naive
type
as
in the rest of
the article,
the best
response is given by
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
683
(U,D) (D, U)
+ wAxspA, + w4x,p4.
(m2,mi,m2,mi)
if ?y5 >
-.
The proof is analogous to that of Lemma 3 and is therefore omitted. It is now straightforwardto characterize the equilibrium play in the pure strategy sequential equilibria. Proposition 6. Let conditions 5 and 6 hold. Then in the sequential equilibria of the
game between the sophisticated EQ EQ sender and receivers, pure strategies are given by
1 :[(m\,m2,mi,m2), 2 :{(m2,m\,m2,m\),
value.
The electoral outcome most favorable to the sophisticated sender is achieved in all
those
Remark thatEQ 2 is an equilibrium in which mortal receivers are fooled, but this does not produce any harm since there is a majority of sophisticated receivers. Conversely, with EQ 1, the vote results in unanimity. In both equilibria, however,
the outcome is the most
equilibria,
whatever
the state.
> max {.*:,, Similar to the enemy case, when 1/2 yb}, there are (mixed strategies)
pooling
preferred
by
the sender
in every
state.
dition theiraction on the signal. Unlike the enemy case, note that the set of equilibria does not expand when beha vioral types are introduced; rather, itbecomes reduced, provided that the amount of having less sophistication is good in the sense that iteliminates the pooling equilibria inwhich thepositive correlation in theplayers' payoffs is not exploited. Conclusion
sophisticated receivers is not too large. As a consequence, and perhaps surprisingly,
equilibria
in which
messages
are not
informative
and
receivers
do not con
We have analyzed a model with a sender interacting with a population of receivers. The sender has informationthat is relevant for the payoffs of both kinds of players. Receivers take a binary choice, and thefinal outcome is determined by majority rule.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
684
In addition, there is uncertainty, similar to Crawford (2003), about the sophistication of both sender and receivers.We find that the pure strategyequilibria of the game share theproperty that if the sender is sophisticated, the outcome will be favorable for
him when or her, whatever there are states the prevailing in which state. This result also the interests of the sender carries in the case through are not at odds with those of
the receivers.
lic on the eve of the 2004 U.S. presidential elections can help in tryingto rationalize a seemingly paradoxical event inwhich, according to some views, OBL "pulled his nemesis over the finish line" (Krauthammer 2004). Although we refrain from offer
on the matter, the reader can perhaps fill in the empty usefully the rationality of the various actors to see who won and regarding to our model. who lost according The voters the conflicting models interests between U.S. game we investigated ing personal views parameters slots our
Using
this framework
to analyze
OBL's
videotape
message
to the American
pub
and OBL in a simplifiedway alongside a hidden true motivation for themessage on OBL's part.A possible alternative (and equally interesting)description of what hap OBL's ultimate intended audience was not pened can be devised ifone believes that not (or only) theAmerican public but, for example, (radical) groups within the Isla mic communities worldwide.11 In that case, Farrell and Gibbons's (1989) model of cheap talkwith two audiences would be the natural startingpoint for the analysis, which can be the object of further research on the topic. Notes
1.The defeated candidate himself later ascribed a decisive importance to the videotape for the out come of the presidential race. Quoting fromNagourney (2005) "the attacks of Sept. 11were the "central deciding thing" in his contest with President Bush and that the release of an Osama bin Laden videotape theweekend 2. Another
historic defeat in the general elections 3 days after. Unlike OBL's message, it is not a case of a third party's endorsement. Yet, the events were so close to the elections and novel to call, again, for a model with initial responses. 3. This approach can be used for sports, where data can be (and in fact are) gathered (see, e.g., 2003; Walker andWooders 2001). Chiappori, Levitt, and Groseclose 2002; Palacios-Huerta 4. On the other hand, it is true thatworking under a more general assumption, such as pb > 1? p, generates a change in the structure of the equilibria. Yet, themain structure of the article remains unaf fected, since only the equilibria where the sophisticated receivers mix no longer exist. 5. If the population of voters were large but finite, looking at undominated strategies would do. With a continuum of voters the payoff irrelevance of voters' actions can be overcome by the notion of atom proofness as introduced byMeirowitz 6. The characterization of behavioral (2005). types depends on the definition of the level 0 players. In the and Colucci (2008), we show that themain results hold under a variety
before Election Day had effectively erased any hope he had of victory." episode that fits well this picture is the strategy adopted by Jos? Maria Aznar (and voters' response in the polls) following the 2004 Madrid train bombings that eventually brought him to a
working paper version of Landi of different but plausible definitions of level 0 types.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
Landi, Colucci
/Binary Choice
Sender-Receiver
Game
685
of the parameters
the Economist's in 2006 presidential elections inMexico, front pages examples of Marcos on Berlusconi, and Aznar's message right after theMadrid bombings seem to fit in this more general maintain the assumption of a binary message space because of the kind of application we have inmind, inwhich themessage clearly signals either one of two possible candidates in an election. 10. This condition is presented in terms of behavioral strategies Myerson (1991) rather than standard mixed strategies as in the previous section. Reasons of simplicity motivate this choice because mixed strategies for the sender are vectors in the fifteen-dimensional simplex. We are reluctant to use the expli cit wording
about behavioral strategies because of the possible confusion with the term behavioral types thatwe have used extensively in the present article. 11. This seems indeed to be the case for various more recent video or audio messages attributed to bin Laden.
Osama
References
Al Jazeera. 2004. Full speech. transcript of bin Ladin's (accessed October 2006). http://english.aljazeera.net/English/archive/
archive?Archiveld=7403 Benabou,
1992. Using privileged information to manipulate markets: Insiders, R., and G. Laroque. 107 (3): 921-58. gurus, credibility. Quarterly Journal of Economics Camerer, C, T. Ho, and J. Chong. 2004. A cognitive hierarchy model of games. Quarterly Journal of Economics 119 (3): 861-98. Campbell, J.E. 2005. Why Bush won the presidential election of 2004: Incumbency, ideology, terrorism, and turnout. Political Science Quarterly 120 (2): 219-41. Chen, Y. 2006. Perturbed communication games with honest senders and naive receivers. Unpublished paper, Arizona State University. Chiappori, P. A., S. D. Levitt, and T. Groseclose. 2002. Testing mixed strategy equilibrium when players are heterogeneous: The case of penalty kicks in soccer. American Economic Review 92 (4): 1138-51. Cohen, F., D. M. Ogilvie, S. Sheldon, J.Greenberg, and T. Pyszczynski. 2005. American roulette: The effect of reminders of death on support for George W. Bush in the 2004 Presidential election. 177-87. Analyses of Social Issues and Public Policy 5(1):
Costa-Gomes, M. A., and V. P. Crawford. 2006. Cognition and behavior in two-person guessing games: An experimental study. American Economic Review 96 (5): 1737-68. Costa-Gomes, M. A., V. P. Crawford, and B. Broseta. 2001. Cognition and behavior in normal-form games: an experimental study. Econometrica 69 (5): 1193-1235. Crawford, V. P. 2003. Lying for strategic advantage: Rational of intentions. American Economic Review 93 (1): 133-49. and boundedly rational misrepresentation
Crawford, V. P., and J. Sobel. 1982. Strategic information transmission. Econometrica 50 (6): 1431-51. 1989. Cheap talk with two audiences. American Economic Review 79 (5): Farrell, J., and R. Gibbons. 1214-23. Feddersen, T. J., andW. 404-24. Pesendorfer. 1996. The swing voter's curse. American Economic 2006. Credulity, Review 86 (3):
Krauthammer, C. 2004. How Bush almost let it slip away. The Time Magazine, Landi, M., and D. Colucci. 2008. Rational and boundedly rational behavior receiver game, 2d version (Singapore Management paper series 05-2008). Singapore: Singapore Management University.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions
686
Meirowitz, A. 2005. Informational party primaries and strategic ambiguity. Journal of Theoretical Politics 17 (1): 107-36. Myerson, R. 1991. Game theory: Analysis of Conflict. Cambridge, MA: Harvard University Press. Nagel, R. 85(5): 1995. Unravelling 1313-26. in guessing games: An experimental study. American Economic
Review
Nagourney,
A. 2005. Kerry says bin Laden tape gave Bush a lift. The New York Times, January 31. (accessed October 2006). http://www.nytimes.com/2005/01/31/politics/31kerry.html I. 2003. Professionals play minimax. Review of Economic Studies 70 (2): 395-415. Palacios-Huerta, Sobel, J. 1985. A theory of credibility. Review of Economic Studies 52 (4): 557-73. 1994. Experimental evidence on players' models of other players. Journal of Stahl, D., and P. Wilson. Economic Behavior 1995. On Economic and Organization 25 (3): 309-27. players' models of other players: Theory Behavior 10(1): 218-54. 2001. Minimax and experimental American evidence. Games and
-.
play atWimbledon.
Economic
Review
91 (5):
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 10:57:59 AM All use subject to JSTOR Terms and Conditions