Are Sanctions Effective?

A Game-Theoretic Analysis
Author(s): George Tsebelis
Source: The Journal of Conflict Resolution, Vol. 34, No. 1 (Mar., 1990), pp. 3-28
Published by: Sage Publications, Inc.
Stable URL: http://www.jstor.org/stable/174132
Accessed: 29/09/2010 10:44

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://links.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you
have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may
use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://links.jstor.org/action/showPublisher?publisherCode=sage.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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://links.jstor.org
Are Sanctions Effective?
A GAME-THEORETIC ANALYSIS

GEORGETSEBELIS
Universityof Californiaat Los Angeles

Although economic sanctions have been quite frequentin the twentieth century,a close
examinationof the low success rate(33 out of 83 cases) indicatesthatsendercountriesare not
able to select the appropriatecases. Moreover,analysts sometimes offer contradictoryadvice
for such selection. This articleprovidesa game-theoreticexplanationof these phenomena.Six
differentgame-theoreticscenarioslead to thesame equilibriumoutcome.This is a mixedstrategy
equilibrium.The success ratiois the outcomeof the selection of mixed strategiesby both sender
and receivercountries.Undera wide rangeof (specified) circumstances,the size of the sanction
has no impactupon the behaviorof the targetcountry.Finally,some empiricalimplicationsof
the game-theoreticanalysis are comparedto existing empirical generalizations,and further
implicationsfor empiricalresearchare discussed.

Dating as farback as 432 B.C., when Periclesenactedhis Megariandecree

prohibitingMegariansto tradeor travel on Athenianland, economic sanc-
tions have been an importantingredientof foreign policy making.In recent
history, sanctions have been applied for military purposes, to destabilize
foreign governments,protect human rights, and retaliate against terrorist
activities.They have been appliedcollectively by actorssuch as the League
of Nations and the OPEC countries, and unilaterallyby individualnation
statessuch as the UnitedStates,Soviet Union, UnitedKingdom,andCanada.
In the post-1960 period, sanctionswere used as frequentlyas two or three
times per year.
One would expect that such a long and dense experiencewould provide
accurateanswersto the questionsof whethersanctionswork, andif so, under
AUTHOR'S NOTE:Financialsupportfor this projectwas providedby UCLA'sAcademic
Senate. I would like to thankR. Bates, J. Frieden,J. Grieco,D. Lake,L. Martin,E. Rogers,and
S. Telhamifor manyuseful ideas, comments,andbibliographicalreferences.

JOURNALOF CONFLICTRESOLUTION,Vol. 34 No. 1, March1990 3-28

? 1990SagePublications,
Inc.

3
4 JOURNALOFCONFLICTRESOLUTION

what conditions. Answers to these questions would have decreased the

frequency of ineffectively applied sanctions:The sender countries would
have been able to anticipatethe results of their actions and avoid costly,
useless, andhumiliatingeconomic hostilities.
Nevertheless, the question of the effectiveness of sanctions remains a
question that elicits many contradictoryanswers.A quick look at the two
most recent and importantbooks on sanctionsillustratesthis contradiction.
Baldwin (1985: 3) tries to answerthe puzzle, "Why do statesmencontinue
to practice economic statecraftwhen 'everybody knows' that it does not
work?" Hufbauer and Schott (1985), in an impressive empirical study,
indicate that in 83 incidents involving economic sanctions after 1914, the
success ratewas a poor40%.Therefore,it is clearthateitherwe do not know
how to count success and failure (this is Baldwin's position) or that sender
countrieshave not learned to make a successful selection of cases for the
applicationof sanctions.This is the reasonHufbauerandSchottprovidenine
"commandments"for the architectsof sanctionpolicies.
The purposeof this articleis to investigatethe problemof authorityand
compliancein an anarchicenvironment.Whatarethe conditionsunderwhich
states will abstainfrom availableprofitablealternativesbecause of fear of
retaliationby other states? Do economic sanctionswork? Underwhat con-
ditions does their success rate increase?Why are sender countriesslow to
learnthe conditionsfor successful sanctions,as indicatedby the presentlow
success ratio?Finally, why, as we will see, do analysts give contradictory
prescriptions?
The article assumes states to be unitary and rational actors. This is a
common (althoughunrealistic)assumptionin the relevantliterature.Efforts
to studytheoreticallythe interconnectionbetweendomesticandinternational
politics are very recent(see Putnam,1988). The reason I do not follow this
more realisticpath is to keep the argumentas clear and simple as possible
andto show thatwe can answerthese questionswithoutnecessarilyreferring
to domestic politics. However, in several places I discuss the impact of
domestic politics or of otherplayers in the sanctionsgame. I first examine
six differentways of conceptualizingthe problem of sanctions as a game
between the sender and the targetcountry.These games make completely
different assumptionsabout the players: They operate under complete or
incomplete information,they have perfect rationalityor adaptivebehavior,
they move simultaneouslyor sequentially,and they have discreteor contin-
uous options.Regardlessof the differencesin the assumptions,however, all
six scenarios lead to the same equilibriumoutcome. Next, I examine the
propertiesof the common equilibriumof these games. Then, the conditions
underwhich sendercountrieswill applysanctionsregardlessof theirimpact,
 OFSANCTIONS S
TsebelisI GAME-THEORETICANALYSIS

or targetcountrieswill disregardsanctions independentlyof their severity,

are analyzed. A structuralkind of uncertaintyis introduced:One or both
countriesdo not know the "type"of theiropponent(whetherthe opponentis
"soft"or "tough").The thirdpartintroducesdomesticpolitics and problems
of internationalcooperation,and compares existing empiricalresults con-
cerningsanctionswith the predictionsof the game-theoreticmodel, explains
the low success rate of sanctions, and gives the reasonsfor policy prescrip-
tions that are poor and sometimescontradictory.

SIX SCENARIOS IN SEARCH OF AN EQUILIBRIUM VALUE

According to Galtung's(1967) definition, sanctionsare "[a]ctionsiniti-

ated by one or more internationalactors(the 'senders') againstone or more
others (the 'receivers') with either or both of two purposes:to punish the
receivers by depriving them of some value and/or to make the receivers
comply with certainnormsthe sendersdeem important."The opinionof the
academic communityin the late 1960s and early 1970s was that economic
sanctions are largely ineffective. Galtung (1967: 409) concluded that "the
probableeffectivenessof economic sanctionsis, generally,negative."Doxey
(1972: 547) claimed that "the deterrentand coercive force of sanctions is
weak on almost every count." Wallensteen (1968: 262) argued that the
"general picture is that economic sanctions have been unsuccessful as a
means of influence in the internationalsystem"; and for Adler-Karlsson
(1968: 9), "theoverallconclusionthatthe describedembargopolicy has been
a failure"is inescapable.In addition,Baldwin (1985) quotes severalofficial
reportsthat reachedthe same conclusion.
There is, however, a discrepancybetween the beliefs of the academic
communityand policies with respectto economic sanctions.Indeed,during
the same period,the numberof sanctionincidentsrose from5 in the 1965-70
period,to 13 between 1970 and 1975, to 22 between 1975 and 1980, andthen
droppedto 11 between 1979 and 1984 (see Hufbauerand Schott, 1985: 25).
This increase is not correlatedwith effectiveness. In fact, the success ratio
before 1973 was almost 45% while after 1973 it dropped to less than
one-third. The opinions of both academics and policy makers oscillate
between the belief that sanctions are ineffective on the one hand, and that
they can have successful policy outcomes on the other.This belief reflects
real differencesin the impact of sanctions (alas, with a differenceof phase
of 180 degrees so thatthese opinions are always in oppositionto the facts).
One reason for such diffused and contradictorybeliefs is the incredible
amountof noise that aggregate-leveldata contain:uncertaintywith respect
6 JOURNALOF CONFLICTRESOLUTION

to how to define whetheror not a sanctionwas successful, how to measure

the size of the penalty,how to measurethe impacton the economy of both
the senderandthe targetcountry,andhow to calculatethe dependenceof the
targetin an interdependenteconomic system where othercountriesmay fill
the gap in economic relationsandtrade.It seems thateach sanctionsincident
is uniqueand, therefore,any generalizationis impossible.
Compellingas this argumentmay seem, it did not preventpolicy analysts
from trying to control for different variables, classifying sanctions into
differentcategories,and providingpolicy prescriptionsfor successful sanc-
tions policies. We will do the same from a theoreticalpointof view. But first,
let us develop several stylized scenarios of sanctions. Such scenarios are
abstractionsand simplificationsof real situations,andeach one of them will
present unrealistic or undesirablefeatures. However, the diversity of the
assumptionsandthe unity of the conclusions of these models will persuade
the readerthat the results reportedhere are valid under a wide range of
conditions.
Let us assume that there are two extremeoptions for the receiver (from
now on, "target"country):eitherto violate a law, rule,norm,or standardthat
is of material and/or normative importanceto the sender country, or to
comply with it. These extremeoptions will be termed"violate"and "com-
ply."On the otherhand,the sendercountrycan choose betweenthe following
two extremes:eitherto "sanction"at maximumcapacityor to "notsanction."
Table 1 representsthe payoffs to each playerin each one of the fourpossible
extremeoutcomesof the game. Table 1 is the generalpayoff matrixthatwill
hold for all the possible variationsof the games.
It is reasonableto assume that if the senderwill not sanction, the target
would prefernot to comply with the standards.In algebraicterms,
ASSUMPTION1: b1 > dl

On the otherhand,it is reasonableto assumethatsanctionsalso inflict a cost

on the sender country.It would then be preferablefor the sender not to
sanctionif none of her interestswere hurt.In algebraicterms,
ASSUMPTION2: d2 > C2

On can also assume that sanctions at maximumcapacity have a deterrent

effect, and that the targetcountryprefersto avoid sanctions ratherthan to
violate the standardandbe sanctioned.I will relaxthis assumptionin the next
section, but for now,
ASSUMPTION3: cl > a,
 Tsebelis/ GAME-THEORETIC
ANALYSISOF SANCTIONS 7

TABLE 1
GeneralPayoff Matrixof SanctionsGame:Six Scenarios

Sanction No Sanction
Violate a1 a2 b1 b2
Comply c1 c2 d1 d2

NOTE: Assumptions are: b, > dj, d2 > c2, cl > a,, and a2 > b2.

Finally,the sendercountryprefersto reactandsanctionwhen its interestsare

violated ratherthanremaininactive.Again, in the next part,this assumption
will be put underscrutiny,but for the time being,

ASSUMPTION 4: a2 > b2

Having establishedthe orderof the differentpayoffs for each player,we can

concentrateon the differentscenariosfor economic sanctions.
Scenario 1: Completeinformation,rationality continuouschoices, and
simultaneous moves. Both the sender and the target country know each
other'spayoffs, andthey areunified,perfectlyrationalplayers.Eachone has
the option of choosing the level of his strategy.The target countryhas to
decide the level x of its violation (wherex = 1 meanscompleteviolation and
x = 0 means complete compliance). The sender countryhas to decide the
level of sanctions y (where y = 0 means no sanctions and y = 1 means
sanctionsat maximumcapacity).Whenx or y are0 or 1, the payoffs for each
country are given by Table 1. To write the payoffs of each player for each
pair of strategies,one additionaland importantassumptionhas to be made.
ASSUMPTION5: The payoffs of each playerare linearfunctionsof the
strategiesof both players.'
Assumption5, togetherwith the payoffsfromTable1, gives the following
utilities for each player as a functionof the strategieschosen by each:

ul= (dj-cj-bj+ al)xy + (cl- dj)y + (bl- dj)x + d, [11

u2= (d2- c2-b2+ a2)xy + (b2- d2)x + (c2- d2)y + d2 [2]2

Equations(1) and (2) enable us to calculatethe equilibriumstrategiesof

the sanctionsgame. Equilibriumstrategiesare defined as a pairof strategies
x* and y* that are optimal responses to each other.Indeed, if the players
1. The importanceof this assumptionwill be discussedat the end of this article.It is sufficient
here to say thatdifferentassumptionswould have producedthe same or similaroutcomes.
2. The readercan verify that for the extremestrategieswhere x and y are 0 or 1, functions
(1) and (2) have the values specified by Table 1. Because functions (1) and (2) are linearwith
respectto x and y, they arethe only linearfunctionswith this property.
8 JOURNALOF CONFLICTRESOLUTION

choose any other pair of strategies,one of them will have the incentive to
change his strategy.The other will also modify his strategy in response,
generatingan infinite cycle of responses.
The only equilibriumpair of strategiesfor the sanctionsgame described
by (1) and (2) is given by the equations:3
x* = (d2- c2)/ (d2- c2+a2-b2) [3]
Y*= (b1- dj) / (bl-dl+ cl-al) [4]
It can be shown that the equilibriumcalculatedby equations(3) and (4)
presentsall the desirablepropertiesof stabilityrequiredin game theory.In
particular,it can be shown to be regular(Harsanyi,1973), perfect (Selten,
1975), essential(WuWen-tsunandJiangJia-he,1962), andproper(Myerson,
1978).4
Scenario 2: Perfect information,rationality,discretechoices, and simul-
taneous moves. Considernow that all the assumptionsmade in scenario 1
still hold, except for one: Each countryonly has the two extreme options
available. The target country either can "violate" or "comply" with the
standard,and the sendercountryeithercan "sanction"or "notsanction."No
one pair of these strategies is a mutually best response, and the game
presentedin Table 1 has no pure strategyequilibria.It follows thatthe only
possible equilibriumof the discrete sanctionsgame is in mixed strategies.
The calculation replicates exactly the previous scenario: Each country,
instead of calculatingthe optimumlevel of its strategy,calculatesthe opti-
mum frequency of mixing its two pure strategies.The calculationslead to
exactly the same equilibriumas equations (3) and (4): The targetcountry
violates the standardswith frequency p* = x*, while the sender country
sanctionswith frequencyq* = y* (see Luce and Raiffa, 1957).
Baldwin (1985) arguesthatthe continuityof strategiesleads observersto
wrong inferencesbecause they have to interpretwhethersanctionswork or
not, while in fact each countrydoes not decide in a discretescenariobut in
a continuousone, and sets levels of sanctionsor of violations. Fromour two
stylized scenariosit becomes clearthatalthoughcontinuityof strategiesmay
create problems of counting, there is no essential conceptual difference
between the continuousand the discrete strategyscenarios.One can move
from the one to the other,translatinglevels into frequenciesand vice versa,
so in the remainingexamples I will drop the distinctionand use only the
easier case for expositionalpurposes.
3. For the derivation,see Ordeshook(1986: 131). Technicallythese equilibriumstrategies
are computedby setting aul/ax = 0 and au2/ay = 0.
4. For the proofsof these theorems,see van Damme (1984).
 Tsebelis / GAME-THEORETICANALYSIS
OF SANCTIONS 9

Scenario 3: Perfect information, rationality, continuous choices, and

sequentialmoves. Supposenow thatthe payoffs for the choice of the extreme
strategiesarethe ones in Table1, andthatall the remainingassumptionshold,
but thatthe two playersmove sequentiallyratherthansimultaneously.First,
the targetcountrydecides how much to violate the standard(decides x), and
then the sender country decides how much to punish (decides y). From
equations(1) to (4), it is clear thatif the targetcountrydecides to violate the
standardby more than x*, then it is in the interestof the sendercountryto
inflict maximumpunishment(respondwith y = 1), which would make the
target country regret its choice. On the other hand, if the target country
chooses anyx less thanx*, thebest responseof the sendercountryis to ignore
the violation (set y = 0), which will make the targetcountrywant to increase
its level of violation. Fromthis account,it follows thatno value less thanor
greaterthanx* can be an equilibriumstrategy,andtherefore,the equilibrium
of the game is given by (3) (the sendercan choose any response).
Scenario 4: Adaptivebehavior,alternatingmoves. Considernow thatthe
payoff matrixis as presentedin Table 1, but that no countryis a perfectly
rationalplayer.Both countriesdemonstrateadaptivebehavior and alternate
moves. Suppose thatin the beginning, the targetcountrychooses level x of
violation of the standard,and the sender countryadopts a level of sanction
y. The payoffs of the interactionaregiven by equations(1) and (2). Then the
targetcountrymodifies its behaviorto the value of the punishment.Subse-
quently,the sender countrymodifies the level of punishmentto agree with
the level of the offense. Then it is the turnof the targetto adjust,and so on.
These mutualadjustmentsaremadeproportionallyto the differencebetween
the currentpayoff and the maximumpossible payoff, given the strategyof
the opponent.
It is easy to calculatethat the optimumstrategyfor the targetcountryis
to violate the standardat the maximum(x = 1) wheneverthe sendercountry
appliessanctionsat a level less thany* [of equation(4)], and to comply with
the standardcompletely (x = 0) whenever the sender countryapplies sanc-
tions more than y*. Similarly,the optimumstrategyof the sendercountryis
to "sanction"at the maximum level (y = 1) whenever the target country
violates the standardat a level greaterthanx* [of equation(3)], and to "not
sanction"(y = 0) wheneverthe targetcountryviolates the standardat a level
less thanx*.
InAppendixA, I show thatthis processcan be formalizedby the following
differentialequations:
dx/dt= k(l-x) [(cl- dl+ b1-al)y - (b1- d1)]if y < y* [5]
10 JOURNALOF CONFLICTRESOLUTION

and dx/dt = -kx[(cl- dl+ bl- al)y - (bl- dl)] otherwise [5']

dy/dt = -ly[(d2- c2-b2+ a2)x + (c2-d2)] if x < x * [6]

and dy/dt = 1 (1- y) [(d2- c2- b2+ a2)x + (c2- d2)] otherwise [6']

where k and 1 arepositive constants.

In simple words, under the assumptionsA1-A4, if the sender country
sanctions,the targetcountrywill comply (A3); if the targetcountrycomplies
with the standards,the senderwill stop sanctioning(A2); if the senderstops
sanctioning,the targetwill violate the standards(Al); if the targetviolates
the standards,the senderwill sanction(A4); if the sendersanctions,the target
will stop violating the standards(A3); andso on. No matterwhich one of the
extreme combinations of strategies results from the choices of the two
players,one playerwill have the incentiveto modify her choice. I described
this cycling process between sanctions and no sanctions,violation of stan-
dards and compliance, in extreme terms. Equations (5) and (6) describe
smaller mutual adjustments.But the major question is, Where will this
process of mutualadjustmentsequilibrate?
In Appendix A, I prove that the unique equilibriumof this system of
differentialequationsis also given by (3) and (4). The interpretationof this
findingis the following:The heroicassumptionsof perfectrationalityarenot
requiredin orderto arrive at the same equilibriumin the sanctions game.
Even adaptive,myopicbehaviorleads to the same outcome.This conclusion
is very importantbecausethe argumentcan be madethatstatesarenot unified
actors, and consequentlycompeting coalitions inside each state will try to
adoptdifferentpolicies in orderto solve the sanctionsgame. These different
policies can then be modeled as a "tattonnement"process, where each
countrytries to find solutions that work underthe circumstances-that is,
given what the opponenthas done so far.
Scenario 5: Both-sided incomplete information, rationality, discrete
choices, and simultaneousmoves. To model incompleteinformation,I will
considerthatthereis randomnoise in the payoffs of each player,and when
they choose, they knowtheirown payoffsbutnot the payoffsof the opponent.
As I will arguein the next section, this noise in the payoffs may be the result
of either domestic politics or events relating to internationaleconomic
competition.Considerthe game of Table2, where 0 < e1 < 1, 0 < e2 < 1, and
x and y are independentand identically distributed,each with uniform
distributionover the intervalfrom0 to 1. Whenthe game is played, player 1
(the target)knows the value of x but not of y, andplayer2 (the sender)knows
 Tsebelis/ GAME-THEORETICANALYSIS
OF SANCTIONS 11

TABLE2
Payoff Matrixwith Two-SidedIncompleteInformation

Sanction No Sanction
Violate al + e1x a2 + e2Y b+ elx b2
Comply cl C2+ e2y dl d2

NOTE:Assumptionsare:b, > dl, d2 > c2, cl > al, a2 > b2; el and e2 small positive numbers;
x, y drawnfrom uniformdistributionsin the [0,1] interval.

the value of y butnot of x. If e1 ande2are0, thenthe game in Table2 becomes

exactly the same as the game with completeinformationin Table1. Thus, let
us assume thatel and e2 are very small positive numbers.Then x and y can
be interpretedas minor factors influencing the players' payoffs when the
strategies"violate"and "sanction"are chosen.
InAppendixB, I show thatwhene1ande2tendto 0, the uniqueequilibrium
of the game with incompleteinformationis againgiven by equations(3) and
(4). The interpretation of this finding is straightforward:Even under
incomplete information,as the perturbationsof the payoffs of each player
are reduced or as each player learns more about his opponent (so that e
tends to 0), the equilibriumstrategiesarethe same as the game underperfect
information.
Scenario 6: One-sided incomplete information, rationality, discrete
choices, and simultaneousmoves. This is a particularcase of the previous
game, where the payoffs of one of the opponentsareknownby bothplayers,
while the payoffs of the otherplayerare subjectto randomshocks.Using the
same assumptionsas the previous scenario concerning e1 or e2, and the
randomvariablex (if the payoffs of the targetcountryare unknownby the
sender)or y (if the payoffs of the sendercountryareunknownby the target),
the equilibriumof the game tends to (3) and (4) when e tends to 0 (see
Appendix B for the proof).
Six differentscenariosled to the same equilibrium.Some assumedsimul-
taneous, others sequentialmoves; some assumedperfect rationality,others
simple adaptivebehavior;some perfectinformation,and others incomplete
informationby one or both sides; and in some scenariosthe countrieshad
simple dichotomouschoices, in othersa continuumof strategieswas avail-
able. The convergenceof all these models to the same equilibriumshouldbe
interpretedas an indicationof the robustnessof this equilibriumto different
plausiblespecificationsof the sanctionsproblem.It is time now to examine
the propertiesof this equilibrium.
12 JOURNALOF CONFLICTRESOLUTION

THE ROBINSON CRUSOE FALLACY

The firstandprobablymost importantobservationfromequations(3) and

(4) is thatthe strategyof neitherthe targetnor the sendercountriesdepends
on their own payoffs; it dependson the payoffs of the opponent!Strangeas
this resultmight seem, it has been confirmedby the analysis of each one of
the six sanctionscenarios.Whetherthe strategiesare interpretedas discrete
or as continuous,andwhetherx or y representlevel or frequencyof sanctions
or of violations, the result remains the same. The strategyof each player
depends exclusively on the payoffs of the opponent.Let me single out this
result in the form of two theorems:
THEOREM1. Underassumptions1-4, modificationof the payoffs of the
targetcountryleaves the level or the frequencyof violation of the standards
(x*) unchanged.On the contrary,most of the time, it has an impact on the
severity or frequencyof sanctions(y*) imposedby the sendercountry.
THEOREM2. Underassumptions1-4, modificationof the payoffs of the
sendercountryleaves the level or the frequencyof sanctions(y*) unchanged.
On the contrary,most of the time, it has an impact on the level or the
frequencyof violations (x*) of the targetcountry.
The proof of both these theoremsis very easy. Inspectionof equation(3)
indicatesthatx* does not dependon the payoffs of the targetcountry,while
equation(4) indicates a monotonicrelationshipbetween the payoffs of the
target country and the equilibrium strategy of the sender country.5For
example, as long as assumptions 1-4 hold, an increase in the level of
maximumsanctions(a, in our model) will have the impactof reducingnot
the level or the frequencyof violationof the standard,butthe level or severity
of sanctionsby the sendercountry.
Although the mathematicsof this statementare straightforward,it still
flies againstourintuition.Therefore,whatis neededmorethanthe analytical
explanation is an intuitive explanation of theorems 1 and 2. Are they
mathematicalartifacts?If not, why are they so counterintuitive?Why does
the conventional wisdom expect that a change in the payoffs of a player
inducemodificationof his or herbehavior,while six differentgame-theoretic
models producethe opposite result?
Conventionalwisdom examines the problemof sanctionsfrom the per-
spective of one player only: the target country. It ignores the fact that
sanctions as well as violations are the outcome of the interactionof two
players- thatis, of a game betweenrationalplayersandnot a simpledecision
againstnature.The fallacyof applyingdecisiontheoryinsteadof gametheory
5. The monotonicrelationshipcan be shown by testing the sign of the first derivativeswith
respectto the payoffs.
 TsebelisI GAME-THEORETIC
ANALYSISOF SANCTIONS 13

when more than one rationalactor is involved I have called the Robinson
Crusoefallacy (see Tsebelis, 1989). This fallacy leads to wrong conclusions,
such as the expectationthatmodificationof the incentivesof one playerwill
modify his behavior,while in realityit modifiesthe behaviorof the opponent.
The reasonthatconventionalwisdom leads to such mistakenresultsis that:
(1) it considers simple decision problemswhere there is only one decision
maker,and (2) it confines itself to short-runanalysis.Indeed, as scenario4
indicates, it is very plausible that, in the short run, as sanctions increase
compliance will increase. However, once the sender country realizes this
changein the behaviorof the targetcountry,it will modify its own strategy-
thatis, reducethe severityof sanctions- andthe targetcountrywill respond
with further modifications, . . . and the new equilibrium will be the one
describedby equations(3) and(4), where modificationsof the payoffs of the
targetcountryhave no effect upon its behavior.
There are several corollaries of theorems 1 and 2. To decrease the
frequencyof violations (increasethe frequencyof compliance),a modifica-
tion of the payoffs of the sendercountryis required:an increaseof a2(make
"sanctions"easierfor the sendercountryto apply)or a decreasein b2(make
"not sanctions"more difficult once a standardis violated), or a decrease in
d2 (the value for the senderof harmoniousrelationsbetween itself and the
targetcountry)or an increaseof c2 (the value of wrongly appliedsanctions).
To decreasethe frequencyof sanctions,a modificationof the payoffs of the
targetcountryis required:an increaseof d1 (the value for the targetcountry
of harmoniousrelationsbetween itself andthe sendercountry)or a decrease
of a, (an increasein the cost of sanctionsto the targetcountry),or a decrease
of b1 (the value of unpunishedviolations) or an increaseof cl (the value of
wrongly appliedsanctions).
Case 1: Combinationof assumptions1-4. This is the most frequentand
most interestingcase of the sanctions game. It is possible, however, that
modificationsof the payoffs of one player modifies the initial assumptions
of the model. In particular,while it is always reasonableto assumethatif the
senderdoes not sanction,the targetprefersto violate the standardratherthan
to comply (assumption1), and thatif thereis no violation the senderprefers
not to sanctionratherthanto sanction(assumption2), assumptions3 and 4
are questionable.One can imagine, for example, that sanctions may not
provide sufficientincentive for the targetcountryto modify its behavior,or
that the sender countryprefers not to sanction even when its interests are
violated. To these points we now turn.
Case 2: Violationof assumption3, (cl < a,). In this case, the targetcountry
prefers to violate the standardno matterwhat the reaction of the sender
country,and the latterwill sanctionor not sanctionaccordingto whetheror
14 JOURNAL OF CONFLICT RESOLUTION

not a2 > b2. If a2 > b2, we will observe unsuccessfulsanctions.If, however,

a2< b2, as I suspecthas been the case quiteoften, no event will be observed.
At best, a diplomaticcomplaintwill be the only visible sign of discomfort
demonstratedby the sendercountry.
Case 3: Violation of assumption 4, (a2 < b2). In this case, the sender
countrywill never sanction.As a result,targetcountrieswill always violate
the sender's interests(assumption1). I believe that this is always the case
with small sendercountries;thatis why thereareno cases of sanctionswhere
the senderis a small country(see Hufbauerand Schott, 1985).
More realistic, however, is the case of a different kind of incomplete
informationthanthe one investigatedin scenarios5 and 6.6 Whathappensif
one or both of the playersdoes not know in which case she finds herself?For
example,what if the targetcountryis not surewhetherit is in the interestof
the sendercountryto sanction- thatis, it cannotdiscriminatebetween cases
1 and 3? Or what happenswhen the sendercountrydoes not know whether
sanctions would make the targetcountrymodify its behavior- that is, the
sendercountrycannotdiscriminatebetween cases 1 and 2? Or if combined
uncertaintymakesthe actorsbelieve thatthey couldbe in anyone of the three
cases?
Figure1 representsa case wherethe targetcountrydoes not knowwhether
the sender country is "soft" (that is, never sanctions) or "tough"(that is,
prefersto sanctionif its interestsareviolated).The probabilityof the sender
countrybeing soft is s, and this is common knowledge. The countrieshave
to decide their strategiessimultaneously,while the targetcountrydoes not
know the type of the sendercountry(althoughit knows the probabilitys). In
Appendix C, I examine this case of structuraluncertainty,and demonstrate
thatthe two essentialpropositionsof thispart- theorems1 and2- still hold.
The equilibriumstrategiesof bothcountriesarecalculated,andthe arguments
for why theorems1 and 2 hold for even more complicatedcases of uncer-
taintyaregiven. The mathematicsbecome morecomplicated;the functional
forms are differentfrom equations(1) and (2), but the essential conclusion
remains:Modificationof the size of the sanctiondoes not affectthe behavior
of the targetcountryat equilibrium;it affects the frequencyof sanctions.

FROM GAME THEORY TO THE REALITY OF SANCTIONS

Next, I will draw some inferences from the game-theoreticmodel of

sanctionsandcomparethemwith the findingsof the most massive empirical
6. 1 thankShibleyTelhamiforpointingout this case of "structural,"
as he called it, uncertainty
to me.
 TsebelisI GAME-THEORETICANALYSIS
OF SANCTIONS 15

1-s S

v/\q q NVq X XNV

al bi C1 di al b1 Ci di
t2 b2 C2 d2 s2 b2 C2 d2

Assumptions: cl> al b> di

t2>b2> s2 d2> C2

Figure 1

studyof sanctions,HufbauerandSchott'sEconomicSanctionsReconsidered
(1985). I will also drawadditionalinferencesthatmay be useful for further
empirical research.What happens if the sanction potential of the sender
countryincreases?Or what happensif the targetcountryis weak? Or how
should the argumentthatsanctionsarenoninstrumentalandpurelysymbolic
be treated?
16 JOURNALOF CONFLICTRESOLUTION

TABLE3
Possible Ordersof Payoffs, and
CorrespondingOutcomesin SanctionsGame

Name Conditions Outcome

Case 1 b >d d2 > c2 a2 > b2 cl > al Mixed strategies
Case 2 b >d d2> c2 al > cl Violation;
sanctionif a2 > b2
Case 3 b > d d2> c2 a2 > No sanctions

At this point, looking back at the conditions of cases 1, 2, and 3 of the

previous section will be very useful for the subsequentdevelopmentof the
argument.The purposeof Table3 is to recapitulatethe differentconditions
and outcomes of the sanctionsgame.
It is time to introducedomestic politics in each one of the countries,and
to assess its impact on the sanctions problem. I have examined elsewhere
situationsin which one actoris involved simultaneouslyin games in several
arenas.I have arguedthatsuch cases can be modeledas nestedgames, where
the situationin one arenainfluencesthe payoffs of the actorsin the principal
arena (see Tsebelis, forthcoming, a). According to this "nested games"
approach,if one considers the sanctions game as the principalarena,and
domestic politics in each country or other internationalfactors such as
economic competitionbetween sendercountriesas the secondaryarena,one
can take into account the impact of domestic politics or internationaleco-
nomic competitionon the sanctionsgame. Forexample,intransigenceon the
part of the opposition, or public opinion, may transformthe payoffs of a
governmentand make it willing to impose sanctionsalthoughit knows very
well that they will not be instrumentalwith respect to the target country.
Similarly,internationalcompetitionmay preventa countryfrom sanctioning
anotherif it knows thatits sanctionswill not be sufficientto inducethe target
countryto change its behavior.All these contextualfactors(with respectto
the sanctiongame) can be takeninto accountthroughcorrespondingchanges
in the payoffs of the countryinvolved in these nested games. Let us study
such factorsone at a time.
The interactionbetween domestic and internationalpolitics can be the
supportingstory for Figure 1. A moderatepartycompeteswith an extremist
party for power in the sender country,and the outcome of this struggle is
uncertain. The moderate party will follow a "soft" line in international
politics, while the extremistwill be "tough"(for operationaldefinitions of
these terms,see Figure1). The targetcountryhasto decidewhetherto violate
 TsebelisI GAME-THEORETIC
ANALYSISOF SANCTIONS 17

a standardwithout knowing which political line will prevail in the sender

country.Or,similarly,an internationalconferenceconcerningprobablesanc-
tions againsta targetcountryis held. This conferencecan succeed, in which
case the sendercountrieswill be "tough,"or the conferencecan fail, in which
case they will be "soft" and tolerate violations of a standard.The target
countryhas to decide without knowing the outcome of the conference, and
the conference has to decide without knowing the decision of the target
country.The models presented in this article can help us investigate the
strategiesof the countriesinvolved in such incidents.
First, let us concentrateon the targetcountry.As noted earlier,it has an
interest in violating the standardsif they are not going to be enforced by
sanctions(Al). Let us now assumethatthe sendercountryis small or weak
or thatit has little leverage,unableto inflict a sufficientlyseriouspenaltyon
the targetcountry.Or, alternatively,let us assume thatthe sendercountryis
strong,yet faces othercountriesthatarein economicor politicalcompetition
with it, so thatwhen the sender applies economic sanctionsothercountries
step in to alleviate the situationfor the targetcountry.Or,finally, consider
thatthereis strongoppositionagainstsanctionsin the sendercountry,so that
imposing sanctions is politically costly for the government.Under these
assumptions,the situationis describedas in case 2 in Table 2. The target
countrywill always violate the standardat the maximumlevel regardlessof
the reactionof the sendercountry.The sendercountrywill then eitherapply
unsuccessfulsanctionsor acquiesce.
But why would a countryapply sanctionsif it knows the payoffs of both
players, and therefore knows that sanctioningwill have no result? Such
behaviorcan be explainedif otherpotentialviolatorsaretakeninto account.
The sender countrymay want to use the sanctions in the specific case not
because of their effectiveness, but to send a signal to other small countries
that similarbehavioron theirpartwill not go unpunished.7
If, however,the strengthof the sendercountryincreasessufficiently,then
the situationwill be describedby case 1 in which the targetcountrywill apply
a mixed strategy:In a series of cases, the target countrywill violate the
standardsome of the time and comply the rest of the time. Similarly,the
sender countrywill apply a mixed strategy,sometimes choosing either to
sanctionor to not sanctionthe violator.Fromthis point on, a furtherincrease
of the sanctionpotential of the sender countrywill not have any deterrent
effect on the targetcountry.The frequencyof violation of the standardwill
remainunchanged.Whatwill change as the sanctionpotentialincreases is
the frequencywith which sanctionsareappliedby the sendercountry.In fact,
7. It requiresiteratedgameswith incompleteinformation.See KrepsandWilson (1982), and
for an applicationin internationalrelationsand the problemof hegemony,see Alt et al. (1988).
18 JOURNALOF CONFLICTRESOLUTION

as equation(4) indicates,this frequencywill be reducedat equilibrium.And

as AppendixC indicates,this logic remainsthe same, even in the case where
the senderor the targetcountrydo not know whetherthey arefacing a "soft"
or a "tough"opponent.
Let us now turnto the sendercountry.As demonstratedearlier,it will not
sanctionif no standardis violated (A2). Assume thatthereis a violationbut
with such small consequences that the cost of sanctions is higher than the
inconvenience from the violated standard.In this case, the sender country
will accept the violation of the standardwithout reaction.The situation is
described by case 3 in the previous section in which the observationwas
made thatthe sendercountryhas a dominantstrategyof no sanction.There
are two reasonswhy this situationmight occur:Eitherthe violation is small
or the costs of sanctionsarevery large.
However,if the inconvenienceof thesendercountryincreasessufficiently
(if it exceeds the costs of sanctioning),the situationwill be describedby case
1, in which bothcountrieswill applymixed strategies.In a seriesofincidents,
the targetcountrywill sometimesviolate and sometimesrefrainfrom viola-
tion,while the sendercountrywill sometimessanctionandsometimesrefrain
from sanctioning.The frequenciesof the respectivestrategiesaregiven again
by equations (3) and (4). From this point on, any furtherincrease in the
importanceof the standardwill have no impacton the behaviorof the sender
country. On the contrary,other things being equal, target countries will
reducethe frequencyof importantviolations,thinkingtwice beforeviolating
importantstandardsof importantsenders.
At this point, I have finished the theoretical investigation, and some
confrontationwith political situationsis in order.I will consider the most
extensive empiricalstudy of sanctions,which covers the whole universe of
sanctions incidents since the First WorldWar(Hufbauerand Schott, 1985;
from now on, H-S), and examine its findings in light of this game-theoretic
model. The H-S studysummarizesits empiricalfindingsin the formof "nine
commandments."
Commandment1 recommendsavoiding high policy goals. Indeed,high
policy goals were achievedonly 16%of the time. Such an outcome is due to
the fact thathigh policy goals occurwhen theviolated standardis considered
of vital importanceby the target country, in which case it is likely that
assumption A4 is violated and the target country prefers to violate the
standardsregardlessof sanctions(case 2 in the previoussection).
Commandment2 recommendsattackingthe weak. Empirically,there is
a correlationbetween economic weakness and susceptibilityto economic
sanctions.This commandmentis based on the idea that sanctioningstrong
opponentswill lead to case 2; the opponentmay be so strongthatsanctions
 Tsebelis/ GAME-THEORETIC
ANALYSISOF SANCTIONS 19

do not have a serious impact (violation of A4) and so she will continue to
violate the standard.
Commandment3 recommendsattackingallies. The reasoningis similar.
Enemieswill have structuredtheireconomy in such a way thatsanctionsare
not effective upon them; again the situationis describedby case 2.
Commandment4 asksfor no incrementalapplicationof sanctions.Strictly
speaking,thereis no time componentin the game-theoreticmodel presented.
Therefore,this rulecannotbe accountedfor directly.However,the reasoning
behind it is not time-related.Commandment4 tries to disallow time for the
economy of the targetcountryto adaptto the new situation.In this sense,
applying sanctions slowly is equivalent to applying ineffective sanctions:
sanctions with costs lower than the benefits of violation of the standard
(violating assumptionA4), leading again to case 2.
I arguethatcommandments5 and7 are contradictory,which accountsfor
theirdiscrepancies.Thatis why they mustbe examinedtogether.Command-
ment 5 recommendsmaximum economic sanctions, while 7 claims that
additional military policies are ineffective. It seems a country wants to
maximize economic sanctionsbecause it wants to maximize the impact of
sanctionson the targetcountry.If this logic is correct,thereis no reasonwhy
economic sanctions should not be accompaniedby diplomatic,military,or
other kinds of action. Additional pressure,no matterwhat its source and
nature,should increase the effectiveness of the project. Thus, 5 and 7 are
contradictory.However, if we consider the game-theoreticmodel, this con-
tradictionis explained.Increasingsanctionsis likely to transformthe game
from case 2 to case 1. However,once we are in case 1, any furtherincrease
in sanctionshas no impacton the equilibriumstrategyof the targetcountry,
as question(3) indicates.The only impactof highersanctionsis to reducethe
frequencyof sanctions.Thatis why militaryactionsdo not have any impact.
Because applicationof militaryactionindicatesthateconomicsanctionshave
already been applied at their maximum force, we are well inside case 1.
Consequently,additionalsanctions have no impact on the strategy of the
opponent(theorems1 and 2).
Commandment 6 recommendsthatsenderswith high costs of sanctionsnot
enter the game. High costs to the sendercountryplaces us in case 3 of the
sanctionsgame; therefore,sanctionsin this case are a mistake.If, however,
costs arebarelylower thanbenefits, the sanctionsgame is describedby case
1, and the frequencyof violationwill be high as indicatedby equation(3).
Commandment8 claims that effectiveness declines with the numberof
sendercountries.Becauseonly one sendercountryis assumedin my analysis,
this is again outside the frameworkof our model. However, it is a classic
collective-action problem.Each sender countryprefers to free ride on the
20 JOURNALOF CONFLICTRESOLUTION

sanctionsof the others.Therefore,the overall impactof sanctionsis reduced

up to thepoint of switchingfromcase 1 to case 2 whereviolationof standards
is the dominantstrategyfor the targetcountry.
Commandment9 links sanctions with broadereconomic and strategic
considerations;however,it does not provide any specific instructions.
These nine commandmentswere addressed to sender countries. The
commandmentstryto shiftthe situationfromcase 2 to case 1. However,even
if they were followed to the letterthey would not guaranteesuccess, because
in case 1 thereis always a percentageof unsuccessfulsanctionsspecified by
the probabilitiesof the equilibriumstrategiesof the two opponents.Let us
use commandment1 to demonstratethis point.
Commandment1 asks the sendercountryto lower policy goals in order
to shift fromcase 2 to case 1. However,loweringpolicy goals will not mean
automatic success. In case 3, mixed strategies are the only equilibrium
strategies.Therefore,we are likely to have cases of nonsanctionsas well as
cases of unsuccessfulsanctions.Moreover,as equations(3) and(4) indicate,
lowering the policy goals will have no effect on the frequencyof violation
by the targetcountrybutwill reducethe frequencyof sanctionsby the sender
country.
The reason why policy analysts and policy makers in industrialized
countriesare so slow to learn the conditionsof successful sanctions is that
they are committing the Robinson Crusoe fallacy. They assume they play
againstnaturewhile they actuallyplay againstrationalopponents-that is,
againstrationalactorsthemselvestryingto promotetheirown goals. Even if
all necessaryinformationwas collected, there still would be an irreducible
partgeneratedby the fact thatboth partiesfind themselves most of the time
in case 1, resultingin a situationin which mixed strategiesmustbe used. So,
in additionto the nine commandmentsof H-S, which are addressedto the
Westernpolicy maker,thereareseveralcommandmentsfor the targetcountry
thatlead to violationswithoutsanctionsor to nonviolations(if sanctionsare
imminentin case of violation).Such cases arenot includedin the H-S list for
the obvious reason that it is very difficult to collect nonevents.The model
presentedearlier can help us generate the conditions that would push the
sanctionsgame from case 1 to case 3, in which the outcome is no sanctions.
The exercise will not be solved here; however, policy analysts and policy
makersof targetcountriesfollow precisely this strategy.Thus, one of the
reasons that sanctions have such a low success rate is that ineffective
sanctionsare the goal of otherrationalactors(the targetcountries).
The argumentof selection bias in cases of deterrencehas been made
convincingly by Achen and Snidal (1989), and attemptsat more rigorous
empirical -studyof deterrenceare now underway (see Russett and Huth,
 ANALYSISOF SANCTIONS 21
TsebelisI GAME-THEORETIC

1989). Exactlythe same logic appliesto sanctions.Supposethatsome social

scientist persuadedby these argumentsdoes her best to collect cases of
nonsanctionsand add them to the H-S list in orderto eliminatethe selection
bias they present. An empirical study of this extended list would provide
much more complete, unbiased, and impartialinsights into the sanctions
phenomenon.
One naturaluse of these datafor the empiricallyminded social scientist
would be to provide the definitive answer on the effects of sanctions. The
standardway to answerthis questionwould be to comparethe frequencyof
compliancewith the standardswhen sanctionsare eitherpresentor absent.
If thereis a statisticallysignificantdifferencebetweenthese two frequencies,
the conclusion must be thatsanctionsare effective.
Now assume that we are in the equilibriumsituationdescribedby equa-
tions (3) and (4) of our model. Whatwould the empiricalfindingslook like?
The frequencyof compliancewith the standardswould be exactly the same
in either the presence or absence of sanctions. The data would look like
randomnoise! The conclusionwould be thatthereis no impact,yet we know
from the previous analysis that those frequencies exist in the first place
because sanctionsdo make a difference.Why is thereconflictbetween these
ideal8dataandtheoreticalanalysis?
Thereis no discrepancy:The empiricaldatalook like randomnoise because
thatis preciselywhatthey are.In fact,each one of the playerstookprecautions
to randomizehis strategy(or to follow any of the other five scenarios that
lead to the same outcome) in orderto avoid exploitationby his opponent.
This last theoretic expectation generatedby my model has important
consequences for both the empiricaltestabilityof the model and empirical
research.How does one test a model thatexpects the datato look like random
noise? And if one believes the underlyingtheory,how can one claim thatthis
theoryis the real reasonthatthe datalook like randomnoise? There aretwo
points to be made. First,the only way to discriminatebetween the expecta-
tions of my model and the high level of noise existing in empiricaldatais to
design situationswhereexternalnoise is reduced:laboratoryexperiments,as
opposed to historicalcases. Second, empiricaltests are not the only possible
tests of a theory.Othercriteriaareexcess content,plausibilityof assumptions,
andcongruencewith othertheories(see Lakatos,1970). The theoryI present
has excess content over othertheories because it explains why, contraryto
expectations,militarysanctions do not have any additionalimpact. More-
over, the assumptionsof my model, despite their simplicity, are plausible,
and the conclusions hold underreasonablecomplications.

8. In fact, so ideal as to be imaginary.

22 JOURNALOF CONFLICTRESOLUTION

However, throughout this article, there were two important simplifica-

tions. First, the payoffs of each actor were considered to be independent of
the payoffs of the other. Second, even when domestic politics or international
events were considered through the nested games approach, the unitary actor
assumption was never relaxed. The new actors influenced the payoffs of the
two principal actors; they were not considered as capable of strategic action
themselves. Such complications will increase greatly the realism of the
model.

APPENDIX A

Proof that scenario 4 leads to the same equilibrium:

For reasons of convenience, I repeatequations(1) and (2) of the main text, which
give the utilities for each countrywhen they adoptstrategiesx and y, respectively.

Ul = (dj- cl-bl+ al)xy + (cl- dj)y + (b1- dj)x + d, [1]

U2 = (d2- c2-b2+ a2)xy + (b2- d2)x + (C2- d2)y + d2 [2]

Both countriesdemonstrateadaptivebehaviorand alternatemoves. When the sender

countryappliessanctionsata level less thany* [of equation(4)], theoptimumstrategy
for the targetcountryis to violate the standardat the maximum(x = 1). In this case,
the optimumpayoff is given by (1) of the main text by substituting1 insteadof x.

maxu1 = (-b1+ a1)y + b1 [1A]

When the sendercountryappliessanctionsgreaterthany*, the optimumstrategy for

the target country is to comply with the standardcompletely (x = 0). In this case,
the optimumpayoff is given by (1) of the main text by substituting0 insteadof x.

maxul = (c1- dj)y + d, [2A]

Similarly,when the targetcountryviolates the standardmore than x*, the optimal

strategyforthesenderis to sanctionatmaximumcapacity(y = 1).Inthiscase,theoptimum
payoff for the senderis given by (2) of the main text by substituting1 insteadof y.

maxu2 = (a2- c2)x + c2 [3A]

Finally,when thetargetcountryviolates the standardless thanx*, theoptimalstrategy

for the sender countryis to ignore the violation (y = 0). In this case, the optimum
payoff is given by (2) of the main text by substituting0 insteadof y.

maxu2 = (b2- d2)x + d2 [4A]

 ANALYSISOF SANCTIONS 23
TsebelisI GAME-THEORETIC

The mutual adjustmentsare made proportionatelyto the difference between the

currentpayoff and the maximumpossible payoff, given the strategyof the opponent.
Consequently,from (1), (2), and (1A) - (4A) the differencebetween the levels of
violation of the standardby the targetcountryin time t and t+1 is:

dx/dt = k(1- x) [(dl- cl-bl+ al)y + (bl- dl)] if y < y* [5A]

and dx/dt = -kx[(dl- cl-bl+ al)y + (bl- dl)] otherwise [5A']

Similarly,the differencebetween thelevels of sanctionsby the sendercountryin time

t and t+1 is:

dy/dt = -ly[(d2- c2-b2+a2)x+ (c2- d2)] if x < x [6A]

and dy/dt = 1(1- y) [(d2- c2-b2+a2)x+ (c2- d2)] otherwise [6A']

where k and 1 are positive constants.

The system of differentialequations(5A) to (6A') is called LotkaVolteraequations
in the biology literature(see May, 1963, andHirschandSmale, 1974), and they have
not yet been solved in theirgeneralform. It is relativelyeasy, however, to calculate
the equilibriumof the system if we set the left-handside of the equationsequal to 0.
The unique equilibriumof the system is the one calculatedby equations(3) and
(4) of themaintext. To exclude otherpossible candidates,one can observetwo things:
(1) pairs where one of the variablesis 0 or 1 are not in equilibrium,because one of
the equations(5A) to (6A') is not 0, and so in the next roundthe variablethatwas 0
or 1 will have a differentvalue; and (2) pairswhere both variablesx and y are 0 or 1
are not in equilibrium,because again, one of the relevantequationsis not 0. So, the
system of (5A) to (6A') has a uniqueequilibrium,and this equilibriumis calculated
by equations(3) and (4).
QED

APPENDIX B

Proof thatscenario S leads to the same equilibrium:

Considerthe game of Table2, where0 < el < 1, 0 < e2 < 1, andx andy areindependent
and identicallydistributed,each with uniformdistributionover the intervalfrom0 to
1. When the game is played,player1 (thetarget)knowsthe value of x but not of y, and
player2 (thesender)knows thevalue of y butnot of x. Foreverypairof el ande2 there
is a uniqueequilibriumof the game, which is given by inequalities(1B) and (2B).
If

x > [el(cl-al) - (bl-dl+ cl-al) (a2-b2)] / [e12+ (bl-dl+ cl-al) (d2-c2+ a2-b2)I1B]
24 JOURNALOF CONFLICTRESOLUTION

play "violate,"otherwiseplay "comply."

If

y > [e2(a2-b2)- (a2-b2+d2-c2)(cl-al)] / [e22+(bj-dj+cj-aj) (d2-c2+a2-b2)] [2B]

play "sanction,"otherwiseplay "nonsanction."
PROOF.To calculate these inequalities,we reason as follows: Each player has to
maximize his own payoffs, given thathe observes the randomvariableaffectinghis
payoffs but not the payoffs of the opponent.Call p(x) the probabilitythat player 1
(the target)will play "violate"and q(y) the probabilitythatplayer2 (the sender)will
play "sanction."Player 1 has to choose "violate"when this is the actionmaximizing
his payoffs, no matterwhat the payoff of his opponent.So, for a value of x greater
thansome specified value x0, player 1 will choosz "violate."Similarly,player2 will
choose "sanction"for values of y greaterthansome specified value yo.
The expected utility of player 1 is given by

EU1 = (al+elx)pQ + (bl+elx)p(l-Q) + cl(l-p)Q + dj(1-p) (1-Q) [3B]

where

1 ~~~~~~~~[4B]
Q =f q(y)dy
0

EU1 becomes maximumwhen aEU,/ap is 0, or equivalentlywhen

bj-dj + elx - (bl-dl+cl-a1)Q =0 [SB]

So the strategyfor player1 is: Play "violate"when x is such thatthe left-handside of

(5B) is greaterthan0, and "comply"otherwise.
The expected utility of player2 is given by

EU2 = (a2+e2y)Pq+ (c2+e2y)q(1-P)+ b2(1 -q)P + d2(1-P)(1-q) [6B]

where

1 ~~~~~~~~[7B]
P =f p(x)dx
0
is 0, or equivalentlywhen
EU2 becomes maximumwhen 3EU2/Oq

c2- d2 + e2y + (a2- c2+d2 -b2)Q = 0 [8B]

So the strategyfor player2 is: Play "sanction"when y is such thatthe left-handside

of (8B) is greaterthan0, and "notsanction"otherwise.
However,because p(x) = 1 when x is greaterthanthe value xo calculatedfrom (5B)
and 0 otherwise,the integral
 Tsebelis/ GAME-THEORETICANALYSIS
OF SANCTIONS 25

P f p(x)dx = 1-xo
Similarly,because q(y) = 1 when y is greaterthan the value yo calculatedfrom (8B)
and 0 otherwise,the integral

1 ~~~~~~~~[10B]
Q = f q(y)dy = l-yo
0

Substitutionof P and Q from (9B) and (1OB)to (5B) and (8B) gives a linearsystem
of two equations and two unknowns (xo and yo). The solution of this system is
presentedby equations(1B) and(2B). It is easy to verify thatwhen e1 -O 0 and e2
0, the xOand yo tend to the following values:

xo- (a2-b2) I (d2-c2+a2-b2) [liB]

yo - (cl-al) I (bj-d1+cj-aj) [12B]

Equations(llB) and(12B) indicatethe limit frequencythateach playerchooses each

one of his pure strategies.Because the distributionof x and y is uniform,x will be
greaterthanxo exactly (1 - xo) of the time. Similarly,y will be greaterthanyo exactly
(1 - yo) of the time. Therefore,"violate"will be chosen with frequency(1 - xo) and
"sanction"with frequency(1 - yo). The readercan verify that these are the same
frequenciesas the ones calculatedin equations(3) and (4).
QED

Proof thatscenario 6 leads to the same equilibrium:

This is a specialcase of thepreviousproof,when eitherel or e2, butnotboth,is 0, while
theothertendsto 0. Letus assumethatel = 0, andrepeatthe stepsof the previousproof.
To facilitatecomparisons,the correspondingequationsarenumberedthe same way.

EU1 = a1pQ+ b1p(l- Q) + c1(1- p)Q + d1(1- p) (1- Q) [3B]

where

Q= q(y)dy [4B]

EU1 becomes maximumwhen OEU1/apis 0, or equivalentlywhen

bl - dl - (bl- dj+c1- a,) Q = 0 [5BJ

So the strategyfor player1 is: Play "violate"when the left-handside of (5B) is greater
than0, and "comply"otherwise.
26 JOURNALOF CONFLICTRESOLUTION

The expected utility of player2 is given by

EU2 = (a2+e2y)pq + (c2+e2y)q(1- p) + b2(1- q)p + d2(1 - p) (1- q) [6B]

EU2 becomes maximumwhen aEU2/aqis 0, or equivalentlywhen

C2- d2 + e2y + (a2- c2+d2-b2) Q = 0 [8B]

So the strategyfor player2 is: Play "sanction"when y is such thatthe left-handside

of (8B) is greaterthan0, and "notsanction"otherwise.
However, because q(y) = 1 when y is greaterthan the value yo calculatedfrom
(8B) and zero otherwise,the integral

Q = 1(y)dy = 1-yo [lOB]

Substitutionof Q from (1OB)to (5B) and(8B) gives a linearsystem of two equations
and two unknowns(p andYO).The limit values of these variableswhen e2 -.0 leads
to the equilibriumdescribedby equations(3) and (4).
QED

APPENDIX C

Calculationof the equilibriumstrategies in Figure 1:

The assumptionsare thatthe sendercountryis "soft"(with probabilitys) or "tough"
(with probability1 - s), and thatthis is common knowledge. So with probabilitys,
the two playersplay the right-handpartof the game tree, for which cl > a,, b, > dj,
and b2 > S2,d2 > c2; with probability(1 - s), the two playersplay the left-handside
of the game tree,for which cl > a,, b, > dj, and t2 > b2, d2 > c2.
The informationsets in Figure 1 indicate that the targetcountryhas to choose
between violate and not violate the standardwithoutknowing whetherits opponent
is soft or tough, while the sendercountryhas to choose whetherto sanctionor not,
without knowing the choice of the targetcountry(but knowing, of course, its own
type).
The targetcountry has two strategies:to violate or not violate. Assume that it
violates with probabilityp and does not violate with probability(l-p). The sender
countryhas fourstrategies:To sanctionunconditionally(whetherit is tough or soft),
to not sanctionunconditionally,to sanction if tough and not sanction if soft, and to
not sanctionif toughandto sanctionif soft. Of thesefourstrategies,two aredominated
and can be eliminated,so the sender country has two undominatedstrategies:to
sanctionwhen tough and to not sanctionwhen soft, and to not sanctionuncondition-
ally.Assume thatit follows thefirstwith probabilityq andthesecondwith probability
(1-q). We have to calculatethe equilibriumvalues p* and q*.
 Tsebelis/ GAME-THEORETIC
ANALYSISOF SANCTIONS 27

The expectedutilities from each strategyare the following:

EV = (1-s) [qa1+(1-q)bl]+ wb1 [1C]

ENV = (1-s) [qcl+(1-q)d1] + wd1 [2C]

ESNS = pt2 + (1-p)c2 [3C]

ENS = pb2 + (1-p)d2 [4C]

To calculatethe equilibriumvalue of q, we equate(1C) and(2C), andsolve for q. We

find
q* = (d1-bl) / [(1-s) (a1-cj+dj-bj)] [5C]

To calculatethe equilibriumvalue of p, we equate(3C) and(4C), andsolve for p. We

find

p= (d2-c2) / (a2-b2+d2-c2) [6C]

Simple observationof (5C) and (6C) indicatesthattheorems2 and 3 hold.

QED

More generally, in two-person games without pure strategy equilibria, the only
existing equilibriaarein mixed strategies,wherethepayoffs of eachplayeraffectonly
the behaviorof the opponent,so theorems2 and 3 hold regardlessof the natureof
uncertainty,as long as one of the subgames is the game of case 1 discussed in the
section on the RobinsonCrusoefallacy (see Tsebelis,forthcoming,b).

REFERENCES

ACHEN, C. andD. SNIDAL(1989) "Rationaldeterrencetheoryandcomparativecase studies."

WorldPolitics 41: 143-169.
ADLER-KARLSSON,G. (1968) WesternEconomic Warfare1947-1967: A Case Study in
ForeignEconomicPolicy. Stockholm:Almqvist & Wiksell.
ALT,J. E., R. L. CALVERT,and B. D. HUMES(1988) "Reputationandhegemonicstability:a
game-theoreticanalysis."Amer.Pol. Sci. Rev. 82: 445-466.
BALDWIN,D. A. (1985) Economic Statecraft.Princeton,NJ: PrincetonUniv. Press.
DOXEY,M. (1972) "Internationalsanctions:a frameworkfor analysiswith special referenceto
the UN and SouthernAfrica."Int. Organization26: 527-550.
GALTUNG,J. (1967) "Onthe effects of internationaleconomic sanctions,with examplesfrom
the case of Rhodesia."WorldPolitics 19: 378-416.
HARSANYI,J. C. (1973) "Gameswith randomlydistributedpayoffs:a new rationalefor mixed
strategyequilibriumpoints."Int. J. of Game Theory2: 1-23.
HIRSCH,M. W. and S. SMALE (1974) DifferentialEquationsDynamic Systems and Linear
Algebra.New York:AcademicPress.
28 JOURNALOF CONFLICTRESOLUTION

HUFBAUER,G. C. andJ. J. SCHOTT(1985) EconomicSanctionsReconsidered.Washington,

DC: InstituteFor InternationalEconomics.
KREPS,D. M. and R. WILSON(1982) "Sequentialequilibria"Econometrica50: 863-894.
LAKATOS,I. (1970) "Themethodologyof scientificresearchprograms," pp.91-196 in I. Lakatos
and A. Musgrave(eds.) Criticismand the Growthof Knowledge. Cambridge:Cambridge
Univ. Press.
LUCE, R. D. and H. RAIFFA(1957) Games and Decisions. New York:JohnWiley and Sons.
MAY,R. M. (1963) Stability and Complexityin Model Ecosystems. Princeton,NJ: Princeton
Univ. Press.
MYERSON, R. B. (1978) "Refinementsof the Nash equilibriumconcept." Int. J. of Game
Theory 7: 73-80.
ORDESHOOK,P. C. (1986) GameTheoryandPoliticalTheory.Cambridge:CambridgeUniv.
Press.
PUTNAM, R. D. (1988) "Internationalcooperationand the logic of 'two-level games."' Int.
Organization42: 427-460.
RUSSETT,B. andP. HUTH (1989) "Disputeinitiationand escalationin internationalpolitics,"
NSF grantproposal.
SELTEN, R. (1975) "Reexaminationof the perfectness concept for equilibriumpoints in
extensive games."Int. J. of Game Theory4: 25-55.
TSEBELIS, G. (1989) "The abuse of probabilityin political analysis: the Robinson Crusoe
fallacy."Amer.Pol. Sci. Rev. 83: 77-92.
TSEBELIS, G. (forthcoming, a) Nested Games: Rational Choice in ComparativePolitics.
Berkeley:Univ. of CaliforniaPress.
TSEBELIS,G. (forthcoming,b) "Moreon the RobinsonCrusoefallacy."Amer.Pol. Sci. Rev.
(forthcoming).
Van DAMME, E.E.C. (1984) Refinements of the Nash EquilibriumConcept. New York:
Springer-Verlag.
WALLENSTEEN,P. (1968) "Characteristicsof economic sanctions."J. of Peace Research3:
248-267.
WU WEN-TSUN and JIANG JIA-HE (1962) "Essentialequilibriumpoints of n-personnon-
cooperativegames."Sci. Sinica 11: 1307-1322.