Elección Social Revisión e Interpretación

Axiomatic Social Choice Theory: An Overview and Interpretation
Author(s): Charles R. Plott

Source: American Journal of Political Science, Vol. 20, No. 3 (Aug., 1976), pp. 511-596
Published by: Midwest Political Science Association
Stable URL: http://www.jstor.org/stable/2110686 .
Accessed: 10/06/2011 18:28
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.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://www.jstor.org/action/showPublisher?publisherCode=mpsa. .
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 [email protected].
Midwest Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to
American Journal of Political Science.
http://www.jstor.org
The Workshop
CHARLES R. PLOTT
CaliforniaInstituteof Technology
SocialChoiceTheory:
Axiomatic
andInterpretation*
An Overview
Normative concernsare centralto the studyof democratic theoryand thepolicy-
relevantareas of politicalscienceand economics.The samenormative concernsare
centralto thestudyof axiomaticsocialchoicetheory.Despitesimilaritiesinoriginand
in underlyingphilosophical the fieldshaveevolvedseparately
orientation, and within
different
methodologies. Thispaperattempts to surveythemostrecentmajorresults in
axiomaticsocialchoicetheorywithina unifying accessibleto nonspecial-
interpretation
ists.
Whyshould a politicalscientistbe interestedin somethinglike social

choice theory?At thisearlystageof thesubject'sdevelopment, it
studying
willbe of littlehelp to thosewhowantto understand broadquestionsabout
theevolutionof institutions. It mightbe of limitedhelp to thosewhostudy
narrowquestionsabout behavior,but only becauseof a close relationship
betweensocial choice theoryand theoriesof strategicbehavior.Political
scientistsinterestedin the normativeaspectsof systemsare the potential
customers, forit is out of concernfortheirproblem, at leasttheirproblemas
seenthrough theeyesof two economists, K. J. ArrowandDuncanBlack,in
the 1940s,thatthefieldseemsto havebeenborn.
The subjectbeganwithwhatseemedto be a minorproblemwithmajority
rule."It is just a mathematical said some(Dahl and Lindbloom,
curiosity,"
1953, p. 422). But intrigued and curiousabout thislittlehole,researchers,
not deterred by thepossiblyirrelevant,begandigging in thegroundnearbyas
describedbelow in the firstsection.Whattheynow appearto have been
uncovering is a giganticcavernintowhichfallalmostall of our ideas about
*Financial support supplied by the National Science Foundation is gratefully

acknowledged.An early draftof this paper was presentedat an MSSB conferenceon
Democratic Theory and Social Choice Theory,Key Biscayne, 1975. The commentsby
participantshave led to several changes.W. T. Jones and Paul Thomas providedmany
helpfuleditorialcomments,and I have enjoyed the benefitof many discussionson this
subjectwithcolleaguesJohnFerejohn,Mike Levine,and StevenMatthews.
AMERICAN JOURNAL OF POLITICAL SCIENCE, XX, 3, August1976 511

512 CharlesR. Plott
social actions.Almostanything we say and/oranyonehas eversaid about

whatsocietywantsor shouldgetis threatened withinternalinconsistency. It
is as thoughpeoplehavebeen talkingforyearsabouta thingthatcannot,in
principle,exist,and a majoreffortnow is neededto see whatobjectively
remainsfromthe conversations. The second sectionof the paper is an
attemptto surveythisproblem.
Aretheseexaggerated claims?If thisessayservesits purposewe willfind
out. The theoryhas not been translated intotheframeworks of politicaland
socialphilosophy withwhichpoliticalscientists are mostfamiliar, so its full
implications are yetto be assessed.Perhapsbyoutlining whatseemto be the
majorresultsand arguments withina singleintegrated themewhichmakes
themaccessibleto sucha broadprofessional audience,newmodesof thought
can be broughtto bearon theproblems andwe can successfully demonstrate
what social choice theoristshave been unable to demonstrate-that such
claimsarein factexaggerated. If theresults go theotherway,iftheclaimsare
notexaggerated, thenperhapsthepaperwillsupplythereaderwithsomenew
toolsandperspectives withwhichto attackold,butveryimportant problems.
The centralresultis broad,sweeping, andnegative. PaulSamuelsonratesit
as one of thesignificant intellectual
achievements ofthiscentury (Samuelson,
1967). It certainlyweighedheavilyin thedecisionto awardK. J.Arrowthe
Nobelprizein economics.Mostof the remaining workcan be interpreted as
attemptsto discreditthis centraldiscovery.An outlineof theseattempts
followinga development of the theoremformsthe secondsectionof the
paper.The firstsectionis devotedto providing severalexamplesof thetypes
of problemsthatthe majorresultclaimsare pervasive. Care mustbe taken,
however,not to confusetheexampleswiththemeaningof theresult.They
should be taken as symptomsof somethingmuch more seriousand as
exampleswhichbringa littlecloserto homewhatis otherwise a veryabstract
argument.
The thirdsectionof thepapersketcheswhatat firstamountedto a small
offshootfromtheeffortoutlinedin thesecondsectionbutnowseemsto be
becominga major,independent area of study.Such positiveresultsthatwe
do have are here. Veryearlyin the development of the literature,it was
discoveredthatit is possibleto describecertaininstitutions, processes,and
procedures in axiomaticterms, whereby a setof axiomscan be viewedas a set
of "parts" whichtakentogether"make" the process.For a puretheorist
constrainedonly by his imaginationand by logical consistency, several
interesting questionsimmediately occur.Whatare the axiomaticconstruc-
tions,the"parts"so to speak,of someof ourexisting processes? Whatis the
AxiomaticSocial Choice Theory 513
real differencein termsof their"parts"between,say,the Bordacountand

the Dodgson ballotingsystems?Whatdifference does it make in termsof
axiomaticallydescribedprocessbehaviorif people vote fortheirtop two
candidatesratherthanonly one? Even thistypeof analysismightprovide
toolsforthepractically Wehaveherea newvehicle
mindedpoliticalscientist.
throughwhichinstitutional comparisons can be made,and we haveno idea
abouthow faritmightcarryus. Ifone can axiomatically describeinstitutions
and processes,thenone can use thepowerof axiomaticmethodsto examine
processbehaviorand compareprocesses.In fact,one mightevenbe able to
take the desirable(axiomatic)"parts"of severaldifferentprocessesand use
the methodsto put themtogether speaking)brand
and forma (historically
newprocesswith"better"features thananyof thosefromwhichthe"parts"
weretaken.
Crazy?We reallydon't know.Perhapsthisessaywillhelp someoneprove
thatthe whole fieldis irrelevantand therebyallowexistingscholarsto rest
easilywiththeirwork.
PARADOXESOF VOTING
As an introduction to axiomaticsocialchoicetheory,let'sexaminea few

paradoxesaboutvotingprocesses."Paradoxes"mightbe thewrongword.In
factone of the real paradoxesis whywe regardthe following examplesas
paradoxes.Nevertheless, paradoxor not,everyone usuallyagreesthatthereis
definitelysomething unintuitiveandpeculiargoingon.
The standardexampleis the following majorityrulecycle.Supposethere
are threepeople 1,2,3tand threealternatives lx,y,z. The firstpersonfeels
x is best,y is second,andz is third.The secondpersonthinksy is bestandit
is followedin orderby z and thenx. The thirdpersonprefers z mostwithx
insecondplaceandy last.
In orderto saveprinting costs,let'sadopttheshorthand wayof represent-
ingsuchsituations usedby students of theseproblems.The notation"P" will
mean "preferred" and the notation"I" means"indifferent" and any sub-
scripton the letterssimplydenotesthe name of the personto whomthe
attitudeis ascribed.For example,Pi is the preferenceof individuali. So we
have:
firstperson xP1yP1z, (x is preferred

to y is preferred
to z)
secondperson YP2zP2x, (y is preferred
to z is preferred
to x)
thirdperson zP3xP3y. (z is preferred
to x is preferred
to y)
514 Charles R. Plott
Now whichalternativeshouldthissocietyget? Naturally, it shouldget

does it want?Naturally,
whatit wants.Whichalternative it wantswhatthe
Butthenthesocialpreference,
prefers.
majority PS, lookslikethis:
xPsyPszPsx
since2 and 3 preferz to x, I and 2 prefery to z, and I and 3 preferx to y.

Thisresultis calleda majority preference
rulecycleor majority cycle.
There are four standardresponsesto this. The firstis "ugh!" which
requiresno elaboration.The second responseamountsto throwing out the
implieddefinition of social preference."This cycledoesn'tmeananything;
thegroupwillchoosethebestthingwhentheymeet-justletthemvoteand
thingswill be OK as longas theycome to someagreement." The troubleis
thattheoutcomedependsonlyupon thevotingsequence.The outcomeis y,
x, or z, dependingonly upon whetherthe agenda sequence A, B, or C,
respectivelyis adopted.Thisis truein theoryand in fact(Plottand Levine,
1975). Now,whatkindof socialphilosophy shoulddependuponthat?
x z x
z y y
y x z
A B C
The thirdreactionamountsto a denialthatthephenomenon exists."Well,

you pulled that one out of a hat, but how oftenwill such a crazycase
occur?"Ifeverything is random,theansweris notwhattheaskerexpectedto
hear. In thiscase the answeris simply"almostalways."The probability of
thiseventapproachesone as thenumberof alternatives approachesinfinity
and it approachesit veryrapidly.In thissensethecycleis thecase and not
theexception.
But,one mightargue,everything is notrandom-people in
havesimilarities
preferences.How does thataffecttheprobabilities?A considerable amount
of workhas beenexertedin an effortto characterize thetypesof individual
preference configurations forwhichthemajorityruleprocessdoes not have
thiscyclicalproperty. Whathas beenlearnedwillbe outlinedbelow,but in
general,the resultshave not been encouraging.This line of questioning,if
anything, has supported theviewthatthephenomenon is pervasive.
The fourthreactionis a claimthatthereis no paradoxat all. "Whenthe
majorityrule cycle occurs,then societyis simplyindifferent-any of the

alternatives
in thecycleis as good as anyother."Now thisis a ratherclever
positionbut thereare stillproblems.Considerthe followingexample,this
timewiththreepeopleand fouroptions.
first
person yP1xPl wP1z,
secondperson xP2wP2zP2y,
thirdperson wP3zP3yP3x.
Now supposefirst,thatw is the statusquo-the alternative

whichinvolves
"doingnothing"-andsecond,thatthefollowingsequenceofvotesoccurs.
x
y
z
As thereadercan see ifhe tracesout theprocess,theultimatechoiceis z. But

everyoneprefers w, thestatusquo, to z, theoptionfinallychosen.So, here
we have a majorityrulecycle thatcontainstwo options,one of whichis
unanimously preferred to theother.Thatmeansthatproponents of thisidea
wouldhavesocietybe indifferent betweentwo alternatives eventhoughone
was unanimously preferred to the other.Processbehaviorlike thiswould
violateevena minimalconceptofefficiency. Aneconomistwouldask: "Why
shoulda societyacceptone alternative whenanotheralternative existswhich
wouldmakeeveryone betteroff?"Almostall socialphilosophers wouldagree
thatall conceptsof sovereignty requirethata peopleunanimousshouldnot
be refused.
"Well,"saystheskeptic,"that'seasy to cope with.Simplyfindsomeway
to firsteliminatetheParetodominated options,thosewhichareunanimously
beatenby someotheroption,andthenapplymajority ruleto whatremains."
Forlackof a bettertermcall this"modifiedmajority rule."
The problemwiththislineof argument is thatit does notsitwellwiththe
conceptof a social preference. To see this,supposepeoplehavetheprefer-
ences as listedin the exampleimmediately above and thatz and y are the
onlyoptionsopen.Bymajority rule,z willbe chosen,andwe wouldconclude
thatsocietyprefersz to y, or morestrongly, we wouldconcludethatz is
betterthany. Now supposetheoptionsare expandedto theset Xw,x,y,z[.
Usingthe processproposedabove we eliminatez fromconsideration, since
516 CharlesR. Plott
everyoneprefersw to z, and declarew, x, and y all "equallygood" or all

"equaHly preferred."Theyare all in themajorityrulecycle,andall arebetter
to theconclusionabove,thatz
thanz. But noticethisis in flatcontradiction
was sociallypreferred to y. Even thoughwe have not formally defineda
conceptof social preference, mostpeople would intuitively object to this
typeof inconsistency. We mustabandoneitherthisskeptic'sidea,and thus
majority ideasaboutsocialpreference.
rule,or ourintuitive
Let's trythe former.Afterall, thereare lots of processesotherthan
majorityruleandmanyof themaresuccessfully used.A primeexampleis the
Bordacount,or pointvotingsystem, describedas follows.Each personranks
theoptionsunderconsideration and assignseach a numberin accordwithits
levelin the ranking.That is, the leastpreferredgetsI point,thenextleast
gets2 points,etc. Pointsare all totaledand the optionwhichreceivesthe
greatestnumberof pointsis chosen.Considerthe four-option, seven-person
example,firstdiscoveredby Fishburn(1974b), given in Table 1. The point
totalsare shownin thebox. Clearlyy is best,and it is followedin rankingby
x, w, and z in order.But,supposez waseliminated fromearlyconsideration
(if not by somenaturalcause) sinceit getstheleastpoints,and besidesit is
TABLE I
Borda Count: A Four-Option,

Seven-PersonExample
Rankings
Individual
Mr.1 Mr.2 Mr.3 Mr.4 Mr.5 Mr.6 Mr.7
w x y w x y w
x y z x y z x
y z w y z w y
z w x z w x z
TotalPoints
w 18 15
x 19 14
y 200 13
z 13
Pareto-dominated by y. Withz gonethereare onlythreeoptions,so points

areassignedto w,x,y in magnitude fromI to 3 ratherthanI to 4 as before.
Withthisdone,the resulton the rightis discovered. Nowthesocialordering
w, x, and theny. Thisis exactlytheinverse of whatit wasin thefirstsetof
circumstances.
The socialpreference, as definedthrough the Bordacount,certainly goes
againstintuitiveideas about how a conceptlike social preference should
behave.So, we are back to a junctionwe havevisitedbefore.Eitherwe must
abandonour intuitiveideasaboutsocialpreference-and perhaps,as willbe
considered below,withthisgoesmuchof ourheritage, centuriesofworkand
reasoning in the fieldof social philosophy-orwe mustexamineadditional
processes.
We will take the latterroute first,since the formerroutewould be
presumptuous, at best,and perhapsgrosslywastefulof perfectly goodideas.
Havingtakenit,however,we willbe forcedto returnto thisjunctionagain
and beginto slowlyandpainfully seekwherethefirstrouteleads.Thatis the
current stateof thetheory.Somearestilledgingalongthesidesof thecavern
followingthe difficult pathsthatremainfromthe optimisticroute,while
othershave alreadyaccepteda positionat the bottomof thehole and are
trying to construct somestairswhichmightlead out.
THE GENERAL POSSIBILITY THEOREM
ProblemFormulation
At thispointeveryoneshouldagreethatsomething strangeand perhaps
unexpectedis goingon. Furthermore, the possibleramifications are very
broad. If the concepts,which help us speak about how we feel whole
societies,polities,and evenworldsshouldbehave,do not workat all forthe
simplecase of a societywitha handfulof peoplewithjust a fewalternatives,
thenperhapswe applythemat thegloballevelonlybecausewe do notreally
understand them.The simplecasesarea naturalprovinggroundforanytools
whichmightbe appliedto themorecomplicated.
Nowexactlywhatarethefactsandwhatis beingcalledintoquestion?The
factsare thatseveralcommonlyacceptedmeansof providing socialchoices
fromamonga fieldof contenders havesomeunusualproperties. The behavior
of these methodsdoes not sit well with our intuitivenotionsof social
preference, and we wishto knowwhattypeofsystem, ifany,wouldsitwell
withus.
SocialPreference
Perhapsthe firststep is to tryto providea precisestatement of our
intuition.At the base is someconceptof socialpreference.Anymethodof
socialchoiceshouldyieldthesociallymostpreferred alternative-the"best"
optionin the eyes of societyas a whole.The immediatetemptation is to
begin to definethe social preference
directly:whatdetermines the list of
priorities,
what specialconsiderationsneed to be included,etc. Such argu-
mentsand discussions are enlightening
and useful,but anyonewho has ever
engagedin such a discussionknowsthat forsome reasontheyneverend.
Perhapsby usinga verycleverinsightfirstusedbyArrow,we can avoidall of
that.Withoutagreeingon whethersocietyprefers x to y or not,we can all
agreeon thefollowing ofsocialpreference.'
principles
1. Preference Transitivity.
Ifsocietyprefers optionx to optiony,andit
prefers optiony to optionz, thenoptionx is preferred to optionz
(assumingof coursethatno relevantconsideration has changedin
theinterim).
II. Indifference Transitivity.
If societyis indifferent
betweenoptionx
and optiony, and it is indifferent betweenoptiony and optionz,
thenit is indifferentbetweenoptionx andoptionz.
III. Value/Feasibility Separation.The social attitude,preference or in-
difference, betweenx and y does not dependupon what other
optionsarefeasible.
IV. UniversalDomain. The social attitude,preference or indifference,
betweeneverypairof optionsis alwaysdefined,eventhoughthere
may be a greatdeal (all possible)of conflictamongthe opinions
(rankings of options)ofindividualmembers of thesociety.
Thinking about theexampleswe havealreadyreviewed willhelpus begin
to understand theseprinciples.
The firstexampleshowsthatmajority ruleas
a way of determining the social preference The
violatesthe firstprinciple.
"modifiedmajorityrule" and the Borda countexamplesviolatethe third
principle.The fourthprinciplesimplysays thatwe expectthe conceptof
social preference
to make sensein the particularcases (and othercases as
1 Formallyspeaking,we need more principlesthan the ones listed (which the reader
will probably implicitlyassume anyway): (i) for all x, xlsx; (ii) for all x and y, xIsy
impliesyIsx; (iii) xPsy and yIsz implies that xPsz; (iv) for any x and y we have one of
xPsy or yPsx or xIsy. We also need an importantPrinciple0 which states that a social
preferenceexists.
Axiomatic Social Choice Theory 519
well) whereindividual membersof thesocietyhavetherankings as givenin

theexamples.Morespecifically, it meansthatwe expectthesocialpreference
to be definedforeveryarrayof individualrankings. If you don't like the
implicitassumptionof individualtransitivity, we can expandthe domainto
includethecaseswhereindividual preferences canbe intransitive as well.
As it turnsout, these are very strongprinciples.We will ultimately
considerdoingaway withall of them.But,beforethathappens,let's show
whytheyare intuitively reasonableand how theyareimpliedby thewaywe
normally thinkaboutnormative socialphilosophy.
The firstthreeprinciplesare usuallyacceptedas reasonableat theindi-
viduallevelof cognition.Instancesof cyclicalindividualpreference or indif-
ferenceare itemsof greatcuriosity (Tversky,1969; Lichtenstein and Slovic,
1971), especiallyif theindividual exhibiting thecycleis willingto acceptit.
Violationsof theseprinciplesare consideredunusualand rareat best.The
thirdprinciplesimplydemandsthattheconceptof socialpreference between
alternativeshas an existenceindependentof the conceptof feasibility of
alternatives.It says we do not haveto be awareof whatwe areable to do
beforewe can even discusswhatwe wouldliketo do. The fourthprinciple
forcesus to take a stand on behalfof societyas a whole,even though
individualswithinsocietyhaveconflicting desires.
Some see thefirstthreeprinciples as following almostaxiomatically from
theword"best." Theywouldmaintainthatsocietyshouldhavewhatis best
and thatif x is betterthany and y is betterthanz, thenit followsthatx is
"best,"clearlybetterthanbothy andz. Herewe havethefirstprinciple with
the concept"better"substituted fortheword"preference" in theprinciples.
Conceptsof social welfaresimilarly implysuch principles. If the social
welfareis greaterwithoptionx thanwithoptiony, and if socialwelfareis
greaterwithoptiony thanwithoptionz, thenit followsfromtheconcept
"greater"thatsocial welfareis greaterat x thanat z. We thushavederived
PrincipleI, Preference Transitivity. PrincipleII, Indifference Transitivity,is
similarly trivial.PrincipleIII, Value/FeasibilitySeparation, follows since the
level of socialwelfaregenerated withoptionx is generally consideredto be
independent of whetheroptiony was or was notfeasible.So we can see that
conceptsofsocialwelfare necessitate acceptanceoftheseprinciples.
The listcan be multiplied. Mostconceptsregarding thenormative aspects
of socialchoicesutilizetheseprinciples plusmanymore.To rejectthesethree
wouldbe equivalentto rejecting anysocialphilosophy whichimpliesthem.
So, it would seem thatif we adopt at leastthese we can proceed.Butrecall
we have alreadybeen warnedthatthe analysisbelow will strongly suggest
thatthesethreeprinciples areexactlywhatwe cannotaccept.Eventhoughit
mayseemas thoughwe havesaid nothingbut theobvious,we havealready

descendedintothecavern.
Normative
ContentofSocialPreference
So farwe havesaidverylittleofa normative
nature.Wehavesaidthatweare
willingto channelnormative demandsthrough a conceptofsocialpreference,
but we have not made any normative demands.For example,shouldthe
social preferencehave anythingto do with individualpreferences? How
shouldthesocialpreference be defined?
Almostany libertarian philosopher would demandthatif a conceptlike
socialpreference
is to be used,thenit shouldhavesomesystematic connec-
tionwithindividualattitudes.For example,if everyonepreferred optionx
over optiony, then at least some would maintain,societyshouldnot be
indifferentbetweenthetwo optionsor perversely prefery to x. Thatis the
famousParetoPrinciple.It is actuallystronger
thantheprinciple we willuse.
V. Responsive.For each pairof alternatives,
x, y, thereis a patternof
attitudes,rankingR1 forpersonnumber1, ranking R2 forperson
number2, etc., such thaty is not preferredby societyto x, given
that the patternof attitudes(R1,, . .. ., Rn) prevails.
For some thiswould seem to be a veryminimalsovereignty condition.
Mostofus wouldreadilydemandmuchmore.It simplysaysthatno optionis
preferred by society to any other option, regardlessof the patternof
individualrankings.
People'spreferences prevailoverall otherconsiderations.
The socialpreferenceis "responsive"
to individualcitizens.If one acceptsthe
ParetoPrinciple,then one clearlyacceptsthis principle,since the Pareto
Principlestatesone of the circumstances,unanimity, whenthe patternof
preferencesnecessarily
outweighsotherconsiderations.
Process
Beforecontinuing let's checkour location.We are attempting
to under-
standthe paradoxesabove and see if we can finda process,a methodof
makinggroupdecisions,whichdoes not havesuchparadoxicalresults.Our
firstprinciples
attemptto make clearwhatconstitutes "paradoxicalbehav-
ior." The principles
are a typeof minimalexpectationaboutsystemperfor-
mance.The strangebehaviorof the examplesaboveis in violationof these
principles.
How can we findthoseprocesseswhichdo nothaveparadoxicalproper-
ties? There are literallymillionsof processes.In fact,by makingsmall

alterationsin procedures, millionsof variations
thereare literally ofanyone
process.How can we hope to even list themall, muchless examinethe
behavior ofeachundera variety of circumstances?
Thisalmostoverwhelming taskcan be solvedbya two-step approach.The
firststepis to considerprocessesonlyin termsofbehaviorpatterns. If many
different systemsof institutions alwaysbehaveidentically underidentical
circumstances, thenthereis no need to studythemall. We need onlystudy
one fromthe set, recognizing that all othersin the set are behaviorally
equivalent.So the idea is to solve firstthe simplerproblemof finding
behaviorpatterns consistentwithour principles. Then,havingidentified the
behaviorwe want,we canattemptto findinstitutions and ruleswhichinduce
thatbehavior.
Consistentwith this plan, the firststep is to identifywhat typesof
behaviorare commonto all systems.For example,if we definetheoutcome
oroutcomesin caseof a tieto be thechoiceofthesystem, thenthefollowing
two properties seemsufficientlyobvious.(1) The choice resulting fromany
systemmustbe one of thefeasibleoptions.(2) The choice resultingfromany
systemdependsin parton theattitudes, preferences, and wants of the people
in thesystem.Allsystems haveboth properties.
A thirdproperty, whichappearsto be commonto all processes,deserves
thestatusof a principle.People'sattitudes abouttheinfeasible alternatives do
not affectthe outcomeof theprocess,thesocialchoice,unlesstheyaffect
theirattitudeaboutthefeasible.The choice,at thispoint,needhavenothing
to do with a social preference.We will name the principleaccordingly
Independence of Infeasible
Alternatives.
Noticethatsome new conceptshave formally enteredwiththisand the
lastpostulate:(1) theset of feasibleoptions,and(2) socialchoice.Thereare
twodifferent conceptsof "feasibleset" floating about.The firstcomesfrom
economicsand is based upon physicaland technicalconsiderations. The
infeasible optionsare thosewhichone can imagine,but are impossible given
theresourcebase and technology. The secondis moreakinto parliamentary
processeswherebythe "feasible"alternatives are thosewhichare in some
senseunderconsideration. The firsttypeof set willbe calledthefeasibleset
andthesecondtypewillbe calledthecontender set.
The socialchoiceis theoptionthatresultsfromtheprocess.Naturally any
chosenoptionmustbe a feasibleoption.Wecan avoidmanyproblems laterif
we adopt a convention now regarding ties,ill-definedprocedures, and even
perhapsuncertainty of outcome.To allow forall thesepossibilities, we do
not demandthatthesocial choicenecessarily be a singleoption.Wecan say
the choice is a set of optionsas longas we understandthatthismeansone

optionfromthe chosenset willultimately be theoutcome,andwe neednot
evendevelopa termforthelatter.
Withthese ideas in mindwe can talk about a social choicefunction,
denotedas C(v,R1,. . . ,Rn). For any actual or imaginaryprocessit desig-
nates,for each feasibleset v and each arrayof individualrankings (R1,
. . . Rn), thesetof optionsthatareoutcomes(the"chosen"outcomes).With
thisnotationwe cannowformally statetheprinciple.
VI. Independenceof InfeasibleAlternatives. Suppose,where(R1, ...
Rn) is an array of rankings, one for each individual,and (R'1,
R'n) is a different arraysuch that for each individuali =
1,2, .. ,n,Ri and R'i areidenticaloverall optionsin thefeasibleset
v,then
C(v,Rl, . . . Rn) = C(v,R 1,*.*

The principlesays that if the feasibleset remainsthe same and the
outcomechanges,then the changein outcomemusthave been due to a
changein someone'sattitudeaboutsomeofthefeasibleoptions.Weusedthe
Borda countexample.If only w,x,y are feasible,thepointsone through
threeare assigned.If someone'spreference forthefourth option,z, changes,
it does not affectthenumbering of the threefeasiblealternatives,and thus
the social choice is not affected.Furthermore, if thesocialchoiceoverthe
feasibleset w,x,y changes,it could be due onlyto a changein assigned
numbers, andthusbe due to an alteredranking.
This axiom,whichArrowcalledIndependence of Irrelevant
Alternatives,
speaking,the mostcontroversial
is, historically of all the postulateslisted.
Muchof thiscontroversy is due to a mistakein Arrow'sownexplanation and
defenseoftheaxiom,whichdiffers fromtheone we areusinghere.2It would
be grosslymisleadingto indicatethat all of the controversy is over.For
example,we can easilythinkof procedureswhichviolatetheaxiom.How
thencan we sayinseemingly bold contradiction thatall processessatisfy it?3
2Arrow incorrectlyclaims that the Borda count violates the axiom. The mistake,
which was firstdiscoveredby Plott (1971), was based on a confusionof thisaxiom with
PrincipleIII above whichis, as we have alreadyseen,violatedby the Borda count.
'Suppose only those who preferLincoln to Johnson(neitheris feasible),get to vote
in the next election. If preferencesfor these infeasiblealternativeschange, then the
voting population and thus the outcome changes,even thoughno one's rankingof the
feasibleoptions changes.This processdoes not satisfyPrincipleVI.
Ratherthan followthis interesting argumenthere,we will postponethe

fora specialsectionlater.
discussion
RationalSocial Choice
Perhapsit is so trivialthatit neednot be mentioned thatanynormative

theoryof social choice demandsa correspondence betweenwhatsociety
prefers and whatis actuallychosen.Butiftherehasbeentruthin advertising
and we are now goingdown a dead-endroad,thenwe had betterbe clear
aboutwhatwe havedoneso we can findourwayback.
The principlestatesthatthe optionthatresultsfromtheoperationofthe
social processis the sociallymostpreferred of the feasibleoptions,and if
thereare severaloptionsfeasibleand equallygood at thetop of thesocial
ranking, thenanyone ofthemcan result.
VII. RationalChoice.If v is theset of feasibleoptions(R1, . . . ,Rn)are
theindividualrankings social
and if C(v,R1,. . . Rn) is the resulting
choice, then C(v,Rl, . . . Rn) = Jthe set of options in v whichare
for
best,or tied best, according to thesocial preference .
The Problem
Is therea processwhichdoes not behavein a paradoxicalway?Is therea

processthatchoosesin accordwithsomeconceptionof a socialpreference?
sort.In
distasteful
The answeris yes, but all examplesare of a particularly
orderto be clearwe needto add somesummary definitions.
Statement
Theorem
Let E be a universalset of options.These are all the optionsyou can

imagine,whetherfeasibleor not. Justput themtogether and callthema set.
Clearlyit is goingto be big,but thatis OK.
A socialpreference definitionis a functionwhichattachesto eachpattern
(n-tuple)of rankings of E, one ranking for each of then individuals,a single
ranking of the options of E for society. From the firstfour principles,
Preference Transitivity,Indifference Transitivity,Value/FeasibilitySepara-
tion, and Universal Domain, it makes senseto talk about a socialrankingover
all of theoptionsof E. PrincipleV, Responsive, indicatesthatitsparticular
formis goingto be functionally dependentupon individual rankings.Thus,
how we define the social preference depends, in part,upon the patternof
individualitypreferences,
and PrincipleVI willassurethattheseare all that
count.
A socialpreferencedefinition in casethereis an individual
is dictatorial io
suchthateitherSxPoy impliesxPsyforall x,yandregardless of therankings
of the otherindividualstor perverselyxPioy impliesyPsxforall x,yand
regardlessof therankings
of theotherin4ividuals. A dictatoris an individual
whosepreference dictatesthe social preference.By checking his preference
alone you can determinethe social preference. The implicationof the
dictator'srankingcan be eitherpositiveor negative.If the implication is
positive,thenwhenhe ranksany x overanyy, theoptionsare rankedthe
same way in the social preference.If the implicationis negative,then
wheneverhe ranksany x overany y, the social rankingbetweenthe two
optionsis theopposite.
We can now statea theorembased uponWilson's(1972a) versionof the
famousArrowGeneralPossibility Theorem.
Theorem.If E has at leastthreeoptionsand C(v,Ri,... Rn) is the choice

functionof some process(therebysatisfying
PrincipleVI, Independence of
InfeasibleAlternatives)and if C(v,R1, . . . ARn)
i) is definedforall finitesubsetsof E (any subsetof E canbe a feasible
set);
ii) is connectedthrough VII, RationalChoice,to a socialprefer-
Principle
encedefinition;
iii) the social preferencedefinitionsatisfiesPrinciplesI, II, III, (the
preferenceaxioms)IV, Universal Domain,andV, Responsive;
theneither
iv) thesocialpreference definition or
is dictatorial,
of individual
v) forall x and all y,xIsyregardless rankings.
Thatis a mouthful, but whatdoes it mean?It probablymeansthatifyou

are in agreement withPrinciplesI throughVII, thenyou necessarily have
whatsome would regardas a strangedefinition of socialpreference.Either
you havedefinedsocietyto be alwaysindifferent, oryourconceptionof the
socialpreference is definedin termsof theattitudeofa singleindividualwho
is eitheralwaysfavored, sinceif he ranksx overy so does society,or always
disfavored, sinceifhe ranksx overy societyranksy overx.
Thereis no questionabout thevalidityofthisresult.The theorem is true.
All of our principlesseemedreasonableenough,so whatis the trick?The
answerto thatquestionis simplynotin. Eachstepoftheargument mustnow
be retracedwitha greatdeal of care.Thisreexamination

processhasbeenthe
subjectofa greatdeal of attention.
Implications
Beforewe go to all thattrouble,however,let's look at the stakes.Some,

likemyself(Plott,1972)wouldclaimthattheconceptofsocialpreference itself
mustgo. Buchanan(1954a, b) was rightin his originalcriticism of Arrow,
thatthe conceptof socialpreference involvesan illegitimatetransfer of the
propertiesof an individual
to theproperties
ofa collectionofindividuals. For
me, the Arrowtheoremdemonstrates thatthe conceptof socialpreference
involvestheclassicfallacyof composition,
andit is shocking onlybecausethe
thoughts of socialphilosophersfromwhichwe havedevelopedourintuitions
aboutsuchmatters aresubjectto thesamefallacy.
In orderto see how extremethispositionis, we shouldinvestigate first
how broadlythe theoremcan be interpreted, and investigate secondthe
variousimplicitwaysin whichour principles of preference can appear.The
firstproblem, showsup in thefollowing
thatofinterpretation, criticism:
A theoryof committees andvotetabulationprocesseshas nothingto do witha whole
society,withits history,
heritage,
legalsystem,
rights,etc. It is to wholesocietiesand
thattheconceptofsocialpreference
socialsystems applies.
There is a two-headedanswerto this. First,one would expect that

conceptslikesocialpreference applicableto largecomplexsocietiesare also
applicablein the special case of a simpleone. If yourideas get you into
troublein the special case of a small uncomplicated societysuch as a
committeewithwell-defined options,theywill by definition get you into
troublein thegeneralcase.
Secondly,the chargeis based upon an unnecessarily narrowviewabout
whathas been demonstrated. To be sure,theexamplesinvolvedcommittees
and committeeprocesses,but nothingabout the formulation limitedthe
interpretations
to these.An optionor socialalternative couldbe a complete
descriptionof the amountof each type of commodity, the amountsof
varioustypesof workdone by each individual, theamountof each typeof
resourceused by each firm,the productionlevelof each firm,the typeof
government agenciesand theservices providedbyeach,etc.An optioncanbe
a verycomplicatedthing,but its degreeof complexity negatesnoneof the
arguments above unlessit bearson theinterpretationof one or moreof the
principles.
The setof feasibleoptionscouldbe a consumption possibilities

setandthe
processcould be a competitive process,a capitalisticprocess,a socialistic
process,or anyotherkindofprocess.4Thereis no need,forexample,forthe
processto be directedin thatsomejudge,administrator, or plannerusesthe
definedsocial rankingto determinethe best option and then directsits
implementation. The processcould be any type of game,votingprocess,
marketprocess,politicalprocess,etc.,as longas theinstitutions aredesigned
so that the resulting outcomes,equilibriums, winners,etc. are always,as
dictatedby PrincipleVII, RationalChoice,"best" accordingto the social
preference ranking.We concludethat thereis nothingabout the analysis
whichprecludestheapplication of ourresultsto wholesocieties.
"Even if the framework does apply to whole societies,what difference
does it make?No one reallyacceptsor uses theidea of a socialpreference
anyway,so no one caresif thereare deep logicalinconsistencies. The whole
enterprise is simplyan exercisein logic." It is thislineof criticism thatwe
will now attemptto refuteby demonstrating thatmanycommonlyused
conceptsareequivalent to theconceptof a socialranking.
One does not have to look farto findconceptslikesocial needs,group
wants,etc. Theseare simplyexpressions of priorities and arethusrankings of
options.In otherwords,they all implythe same technicalproperties as
preference. Ideas aboutwhatis good forsocietyareno different. If optionx
is betterforsocietythanoptiony and optiony is betterforsocietythan
optionz, thenoptionx is betterthanoptionz. The word"preference" does
not appear,but this is clearlya statementof PrincipleI with "better"
replacing"preference." All of thearguments applywithoutmodification. The
keyobservation hereis thatpossiblydifferent substantiveconcepts,e.g.,the
"preferred" as opposedto the"best,"havethesametechnicalproperties, and
thesetechnicalproperties causetheproblem.
Evenmoreformalized conceptssufferfromthesamedifficulty. Take for
exampletheconceptof economicwelfare.To different optionsone attaches
a number,perhapscomputedfromcertaintypesofeconomicconsiderations,
indicating thelevelof economicwelfare.Certainformsof cost-benefit analy-
sis are attemptsto operationalize such a formula.But indicatorsof social
welfareclearlyimplya rankingof social optionsaccordingto thenumbers
whichindicatethelevelsof welfare. The ranking satisfies
all ofourprinciples
of social preference,and thusthetheoremstandsas a criticism of anysuch
'The term comes from economics. It refersto all possible ways in which final
consumptiongoods can be distributedamong membersof society,given that the only
limitationsare resourcesand the productioncapacityof the society.
formula.The only admissibledefinitions of welfareare those whichare

dictatorial.
Rawls(1971) has diagrams whichgiveratherexplicitshapesto thesocial
ranking.Societyis betteroffwithone distribution of primary goods over
anotherif theworst-off personis betteroff.Sincethisranking all of
satisfies
our principlesof social preference it followsthat,if he is theorizingabout
processes,therebypickingup the remainder of ourprinciples, hissystemhas
thesameproblemsas do all others.In thisparticular case it seemsas though
theonlyadmissibledefinition of theworst-off person,a definition whichhe
does not supply,is thattheworst-off personalwaysremains thesameperson,
even in the extremecase in whichhe has acquiredall theprimary goodsin
society(Plott,1974).
So, we can see thatsocial choicetheoristshavea potentialaudiencelarger
thanthemselves. The majorresultsdo haveprofound andbroadimplications.
Let'sturnnowto a detaileddiscussion oftheprinciples.
The Analysis:An Overview
The problemwas, and is, thatwhenwe are preciseand clearabout our

intuitivenotionsabout social actions,we discoverdelicaciesand inconsis-
tencieswhich we had no idea existed.Is the problemthat our precise
statements failto captureour notions?Or, are our notionsactuallywrong?
Or, havewe simplyimposeda technicalfeatureon theanalysiswhich,once
modified,willeliminate all theproblems?
How one proceedson thesequestions,whatprinciples are firstexamined
and how theyare modified, is heavilydependent upontheidiosyncracies and
philosophicalpersuasions of theinvestigator. The analysishas differentimpli-
cationsfor different philosophies. Moreover,the formalstatements of the
principlescan be interpreted in a varietyof differentways.For example,the
Independenceof InfeasibleAlternatives can be interpreted as a normative
statementwhichshould, accordingto some, be embodiedin any social
preference definition;
it canbe interpreted as defininga typeofbinaryvoting
processonly(Pattanaik,1971); it canbe interpreted as a typeofinformation
conserving process(Brown, 1973a); it can be viewedas a principlewhich
prohibitsintensityconsiderations (Sen, 1970a); it can be confusedwith
PrincipleIII (Ray, 1973); it can be modifiedto implyPrinciplesI and II
(Hansson,1969a; 1973); it can be interpreted as above,etc.,etc.,etc. So, a
conclusionthat"PrincipleX is the realproblem"mustalwaysbe relative to
someparticular interpretation of thewholeset of principles and thepurpose
of theenterprise.Principles whichcan be disregarded on one interpretation,
and thus"solve"thedilemma, areabsolutelyessentialto someotherinterpre-

tation.So, by thetimewe takeall thepossiblephilosophies togetherwithall
thepossibleinterpretations
of theprinciples,we havea verylargenumberof
possibleimplications
of theanalysis.
As we proceedthen,to surveysome of the knownresults,we mustbe
carefulnot to judgethemfromone pointofviewalone.Furthermore, as the
readerprobesdeeperintothepapersreferenced here,he shouldnotexpectto
findotherauthorsmotivated by the interpretations
I havedeveloped;or by
some singleinterpretation;or indeed by any interpretation at all. Many
believethata deeperunderstanding of how the simplemathematical logic
itselfworksmaybe theverythingwe needto removetheparadoxes;so they
legitimatelyanalyzethemathematics involvedwithoutregardforinterpreta-
tions.
TheParetoPrinciple
The Pareto Principlecomes in two forms.One is a statementof a
normativeposition,whilethe otheris a statement
of a possiblebehavioral
law.The normativeversionis:
VIII. ParetoPrinciple.If x is availableandeveryone
ranksx abovey,then
y shouldnotbe chosen.
This axiomwas partof the originalArrowsystemof axioms.Withina
social preferenceframework where choice is necessarilygeneratedby a
preference, it would be stated"If everyoneprefersx to y, thensociety
prefersx to y."
Therehas been a greatdeal of unnecessary confusionaboutthisaxiom.
Contraryto whatmanythink,the axiom does not demandunanimity as a
conditionfor the determination of social preference. It does not say "If
societyprefers x to y, theneveryoneprefers x to y." It onlysaysthatwhen
thereis unanimity the social preferenceis determined, and it may also be
determined in manyothercases as well. To denythe axiom would be to
accept the positionthat thereare cases in whicheveryonein the society
prefersx to y butsocietydoesnot.Putthisway,theaxiomwouldseemto be
partof anylibertarian philosophyand could be listedalongwithPrinciple V
as partof thenormative structure
of a socialpreference definition.
The secondstatement of theaxiomis a statement alongbehaviorallines:
"If everyoneranksx overy andx is feasible, theny willnotbe chosen."This
is a hypothesisabout the possiblelong-runtendencyof social systems.
Postulateslike thisare frequently foundin coalitiontheories, forexample.
Any systemthat consistently violatedthisprinciple, the thesismaintains,

wouldsoon evolveintoa different system.Institutionswhichinduceprison-
er's dilemmasand theresulting forsocietiesas a wholearenot
inefficiencies
viableand will soon be transformed.5 The boundariesof thisthesisare still
underinvestigation, but if it is thoughtto be generallycorrect,thenit should
be listedalongwiththe Independence of InfeasibleAlternativesas a natural
property of processes.
Regardless of how thisaxiomis interpreted we knowone thingforsure.It
is not thecause of theproblemoutlinedabove,sinceit was not partof our
formulation even thoughit was partof Arrow's.Our principles implythat
theremustbe eithera "positive"dictatorwho alwaysgetswhathe wants
mostor a "negative"dictatorwhoalwaysgetswhathe wantsleast.Ifwe add
the ParetoPrinciplethen,due to a theoremprovedby Wilson(1972a), the
onlychangeis theremovalof thelatter"negative"dictator.
So regardlessof theinterpretation of theParetoPrinciple,it can holdall of
thetime,partof thetime,or never;itsacceptance,rejection, or modification
does nothingto helpus out of thedilemmawithwhichwestarted.
The Domainof SocialPreference
Somehowwe got froma lot of principles to a positionwherewe are

expectingour social preference and socialchoiceprocessto do a
definition
we expectour socialpreference
lot forus. In particular, definitionto handle
everypair of options,regardless of the patternof conflict(PrincipleIV,
UniversalDomain).6Secondly,we expectoursocialchoiceprocessto yielda
resultwhenany possiblesubsetof optionsis feasible,and regardless of the
patternof conflict(PrincipleVII, RationalChoice).Havewe beenunrealistic
and demandedmorethanis necessaryhere?It wouldbe easy to say yes if
somereasonableconsequences followedfromrelaxing theseprinciples.
Notice,now,thatto playthistune,we havea lot of strings to touch.We
can saythatthesocialranking neednotbe definedovereverypairof options,
or we can say thatit neednotbe definedforeverypatternof conflict, orwe
can use some combinationof the two.We can alterPrincipleIV, Universal
Domain,and a "technical"requirement buriedin a footnote,andsaythatfor
'For definitionsand examples of prisonerdilemma games see Luce and Raiffa

(1957).
6 listedin
readerwill recallthatwe had some additional"technical" requirements
footnote 1. The importantthrustof those assumptionswas completeness.For everypair
of optionswe had eitherpreferenceor indifference.
certainspecialoptions,the social rankingneed not be defined.7The social

preference wouldbe "incomplete."Alternatively, we can saythatforcertain
patternsof conflict,certainconfigurations of individualrankingsof the
options,thesocial ranking neednot be definedforanypairof options.That
is, societywouldnot be indifferent but neitherwouldit havea ranking. The
idea,then,is to limitthepatterns forwhichthedefinition
of conflict applies.
Of coursewe have to worryaboutPrincipleVII, sinceit demandsthatthe
socialchoicebe "preference based" regardless of thepatternofconflict, but
perhapswe can takecareofthat.
So basically,the idea is to eitherallow the preference definition to be
incomplete overpairs,orlimitthepattern ofindividual rankings thatwe wish
to consider,orboth.The former routebeginsto raiserather seriousproblems,
bothtechnicaland philosophical, withouroriginalformulations. It requiresa
reformulation of our originaldemands,the fullconsequencesof whichare
currentlyunknown.Whatdoes it mean to say, on some occasionwhen
societyby definition prefersx to y and y to z, thatthe ranking betweenx
and z is undefined?Froma technicalandmathematical pointofview,thereis
no problemwiththis,but the interpretation, relativeto our originalexam-
ples,hasyetto be thrashed out. See Fishburn(1973b, 1974d).
The secondroute,placinglimitations on the patternof conflict,is also
difficult,but heretherehas been someadvancedscouting.In somesensewe
would like to knowoverwhatdomain,whatpatterns of individual rankings
or whatpatternsof conflict,do we not get "impossibility results."Unfor-
tunately,that problemhas not even been well formulated, much less
answered, but we do haveanswersto some"closelyrelated"questionswhich
seemto indicatethatthiswholelineofquestioning is notthepathto follow.
weremodifiedto operateovercertainlimited
Theorem.If all our principles
of rankings
patterns thenthereis a dictatorover
acrossthesetofindividuals,
every"free"tripleofalternatives.
A "free"tripleis a setof threealternativesforwhichthesocialpreference

is definedforall patternsofrankings ofthethreealternativesacrossindividu-
als. The theorem, due to Murakami (1961), saysthatforeverythree-element
set overwhichwe fail to restrict thepatternsof conflictand on whichthe
socialpreference is defined,we pickup a dictator.Thisdictatorof thesocial
preference overa three-element set mayalso necessarilybe a dictatorover
ootnote6 applies herealso.

someotherthree-element set as well-it dependsupon thegeneralpattern of

restrictions.So, even if you are willingto live withdictatorsoverthree-
elementsets,you muststrategically choosethe patternsof conflictyou are
willingto consider;otherwise youwillendup withonlyone dictator, andthe
processof pickingtheseconflicts is goingto be tediousand perhapswithout
motivation or interpretation. Of coursewe will knowa lot moreonce the
puremathematical problemis solved.
Otherstudieshave takenthe moreindirectrouteof studying particular
processes.The processof simplemajorityrulehas receivedthemostatten-
tion,with the outstanding questionbeing:Overwhat patternsof conflict
does majorityrulecycle?Thereare two answers,depending uponhow one
interpretstheterm"patterns of conflict."
The Inada-Pattanaik-Sen approachhas been to tryto isolate a set of
rankingsof the alternatives such that if each individualin societyhas a
rankingwhichis a memberof thisset, thenthe majorityrule preference
wouldbe transitive.8 As a trivialapplicationoftheapproach,takea setwhich
has only two rankings. If everyindividualin societyhas one of thesetwo
rankings,thentherecanbe no majority rulecycle(thereaderis invitedto try
to findone). The mathematical problemis to getthesetas "big" as possible,
and the interpretive problemis to see if thereis any reasonto believethat
individualrankswould,or perhapsshould,be restricted to sucha set.
Now that the mathematical parthas been answered,the answerto the
interpretive part appearsto be "no." We have now, as a resultof recent
research,a setof axiomswiththeproperties:
i) If a set of rankingssatisfytheseaxiomsand if each individual has a
ranking fromtheset,thenthemajority is transitive.
preference
does not satisfytheaxioms,thenthereis a pattern
ii) If a set of rankings
of conflict,suchthattheranking ofeachindividualis in thesetandthe
majority preference
cycles.
The set of axiomsis veryrestrictiveand to datehasnotbeensuccessfully
applied at eitherthe behaviorallevel or normative level. For an excellent,
detailedanalysisof the extremely natureof thecompleteset of
restrictive
axiomsrelativeto usualassumptions inbehavioralmodelsregarding theform
of individualpreferences,thereadershouldconsultKramer(1974). Whenthe
numberof alternatives is large,usingthisaxiomset is almostlikeassuming
unanimity.
8For a good summarysee Fishburn(1973) or Sen (1971).

532 CharlesR. Plott
The second approachis to tryto place restrictions on the patternsof

conflictdirectly. In thecase ofmajority rule,whatpatterns of conflictscause
cycles?The news is bad. Essentially,all patternsfor whichthereis no
indifference cause problems.Studieswhichexaminedtheprobability of the
Paradox(DeMeyerand Plott,1970; Garmanand Kamien,1968; Niemiand
Weisberg,1968; Pomeranzand Weil, 1970; Gehrleinand Fishburn,1975)
havefoundthatwhenindifference is not allowedand all individual rankings
occurat random,the probability of a cyclebecomesone as thenumberof
alternativesgrowslarge.The approachis also veryrapid.
So thenewsshapesup as follows.Patternsof conflict whichdo notcause
problemsforma set of measurezero in theset of all possibleconflicts. In a
sense,theset of "troublefree"patterns ofconflict is "infinitely small"when
put againstthe backgroundof all possiblepatterns of conflict.If we are to
followthisroute,we needsomejustification forfocusing on thisverysmall
(relativeto thewholeset) patternof conflicts. Suppose,forexample,people
alwayshave"similar"attitudes. Whathappensthen?If one defines"similar"
to mean theproportion of peoplewhosepreferences can be put on a single
"scale" the answercan be foundin Niemi(1969). A measureof "social
homogeneity" is developedby Fishburn (.1973b)andanotherby Jamison and
Luce (1973). An index of "voterantagonism," based upon thenumberof
pairsof individuals whose preferences are in conflict, is developedby Kuga
and Nagatani(1974) and relatedto thefrequency of cycles.To makea long
storyshort,one can concludethatas similarity amongpreferences decreases,
then the likelihoodof a majoritycycle increases.Currently we have no
theoreticalreasonto suspectthatanyof theseindexestendto be "low,"but
thereis someevidence(Jamison,1975) whichsuggests thatthetheorymight
be overlypessimistic.
Whathappensif we don't requirethemajoritypreference relationto be
transitive?Afterall, it mightnot hurtifthereis a cycleas longas it is noton
top. Whynot simplydemandthatthereis a majority winnerovereverything
else? This means that we would be willingto accept a case wherew is
preferred by a majorityto x, y, and z, and therefore "wins" eventhough
thereis a cycleamongx, y,andz. It is a greatidea,butit fliesdirectly in the
faceof Principle I. Furthermore,ourwillingness to compromise likethisbuys
us verylittlein thecase ofmajority rule.Conditions on conflicts whichassure
theexistenceof sucha majority winnerwereworkedout forcertaincasesby
Plott(1967), and foran expandedclassof casesby Sloss(1973). In addition,
the probabilitywas examinedby DeMeyerand Plott (1970). The news
remainssad. Our compromise boughtus nothing. The theoretical probability
of such a majority winneras thenumberof alternatives growsto infinity is
conjecturedto be zero. Of course,untilsomeoneprovesit thereis alwaysa

rayofhope.
Well, so much for majorityrule. What about the existenceof some
preference definition?
Can't we at least get away fromthe impossibility
resultsby limitingthe patternof conflicts?Severalresultson thisbroader
problemcanbe foundin Fishburn (1973a). The outlookisnotgood,however.
Thispathhasbecomeexceedingly narrowandslippery. Perhapswe hadbetter
returnto wherewe startedand tryanotherroute.
TheNumberofPeople
The impossibility theoremstatedabove made no assumptions aboutthe
numberof people except that the numberis finite.Whathappensif this
assumption is removed?The questionis interesting primarily becauseit gives
us insightintothetechnicalstructure of theproblem.If impossibility prob-
lemsdo not occurwhenthereare infinitenumbersof people,perhapswe
need to rethink our formulations.Afterall, manymodelsof socialbehavior
proceedas if the populationwere infinitely large(Rikerand Ordeshook,
1973). Whenwe use proportions andpercentages to describepopulations and
assumethesepercentages are continuous,we are makingan implicit assump-
tionaboutthesize of thepopulation.9Thiswouldnotbe thefirsttimethat
the assumptionof a continuumhas providedjust the propersloppinessto
make a modelwork,and if it does workthenperhapswe shouldexplore
thoseinterpretations withinwhichsuchan assumption mightmakesense.
Fishburn (1970) posed theproblemand providedus withwhatseemedto
be a positiveresult.If all of our conditionsare satisfiedand if thereis an
infinitenumberof people,thenthereneednotbe a dictator.
The optimism generated by theFishburn resultwas soonerasedbyKirman
and Sondermann (1972). In theinfinitecase,ourconceptsnecessarily gener-
ate a peculiartypeof "invisibledictator"whichcan onlyexistin theinfinite
case. Therealwaysexistsa set of individuals whichdictates(if thisset is
unanimousforx overy, thensocietyprefersx overy). Furthermore, any
such set necessarilycontainsa propersubset whichalso has dictatorial
powers.So whileno singlepersonis dictator,the set of people who have
dictatorialpowersis "infinitely small."In orderto see this,imaginepeople
distributedcontinuously alonga line,witheach pointrepresenting a person.
9If thereare only ten people, it is impossibleforexactly95 percentof themto have

any particularcharacteristic.
534 CharlesR. Plott
Some dictatorial set thencouldbe takenas an interval. The resultsaysthat

thereis also a propersubsetof thissetwhichis also dictatorial.10
So, whatmomentarily appearedto be a trailturnsout to be a dead end.
Allowingthe populationto be infinitedoes not help at all. Let's go back
againand pickup anotherpath.
Alternatives
ofInfeasible
Independence
Herewe are at themostcontroversial principlein thewholeconstitution.
As we explorethis,thereadershouldbe informed oftworelatedfacts.First,
thereis a greatdeal of confusionabout theaxiom,due primarily to Arrow.
He chose a poor exampleto illustrate its meaning. In fact,because of this
confusionwe have used theword"infeasible," following Fishburn (1973a),
insteadof Arrow'sterm"irrelevant"in the axiom's name. Second, the
argument and interpretation we willgiveis "new" in thesensethatit is not
fullyintegrated into the Our positionwas firstadvancedin Plott
literature.
(1971) and is followedup in a fewplaces,but it is fairto saythatit hasnot
withstood thetestof a longperiodof professional scrutiny.
As it is stated,our axiomsaysnothingat all about thesocialpreference
definition. It operatescompletely independently ofanysuchconcept.It says
thatif the feasible set remains fixed and if individuals'preferencesoverthe
feasibleoptions remain fixed,then the social choice remains This
fixed. has
been a source of great confusion,sincemany peoplestate the axiom in terms
of the social preference definitionas opposed to the choice as we have
done." Since statingthe axiom in termsof a social preference definition
would forecloseour freedomof examiningthe axiom independently of a
socialpreference definition,we wishto avoid any such ties."2
l 'How could two sets be dictatorial?Notice that the preferenceof a dictator is

sufficientto determinea social preference,but it is not necessary.If he prefersx to y,
someone else's preferencecan prevail.The
then so does society,but if he is indifferent,
original theorem is actually that there is a sequence of dictators. The preference
determiningpowers of any one of themwould exist only in those cases wherethosein
frontof him are indifferent. We should also add, for the mathematicallyinclined,that
the "measure" of such a set may indeed be large.Currentlyno theoremshave exploited
the variousmeasuretheoreticpossibilities.
" The axiom would become: "The social preferencebetween x and y depends only
upon the patternof individualpreferencesforx and y, and not upon any preferencesfor
z." See Mayston(1974).
12 See Hansson (1973) for an unjustifiedcriticismof our position.The claim thereis
that Arrowdid not do what we are doing. The answeris that ifhe didn't do it the way
This principletellsus thatcertaintypesof modelbehaviorareinconsistent

withthe behaviorof any possiblerealsystem.It is based on a presumption
thatthereare(natural)behavioral principles (laws)whichgovernthebehavior
of all social systems.If such laws exist,they carryimplications for the
normative problemof finding a socialpreference definition.
Weare,afterall,
fundamentally interested in finding an acceptabledecisionmaking system, and
our searchfora social preference definition is a meansto thatend. If the
implementation of a socialpreference definition requiressystembehaviorin
violationof such laws, thenwe would rejectthe definitionon practical
grounds, ifnotmoralgroundsas well("ought"implies"can").
Whereare suchgeneralbehavioralprinciples to be found?The one we use
is a property of all of our current societalmodels.It is a property ofgeneral
economicmodels,spatialmodelsof politics-anyof our currentgamethe-
oretic,cooperativeor noncooperative models.Whether or notone can design
systems thatbothworkandviolatetheproposedlaw remains to be seen.
Remember,the principlesays that preferencechangesfor infeasible
optionswillnotchangethesocialchoice-theprocessoutcome(unlessprefer-
enceschangeforthefeasibleas well).Let's take,as an exampleof a proposed
processthatviolatestheprinciple, a processsimilarto the Bordacount,and
see whywe thinkit (as a process)willnot work.Supposewe are choosing
froma listof candidatesXw,x,y,z (, whowillreceivean offeras fullprofessor.
To this list is added a set of fourinfeasiblecandidates,JamesMadison,
ThomasJefferson, J. S. Mill, and Karl Marx,givinga total list of eight
candidates.Each memberof theappointment committee is to rankall eight
candidatesaccordingto hisevaluations and assigneachcandidatethevalueof
hisrank,withthelowestrankedreceiving thenumber1. If,forexample,your
ranking, R, fromfirstto lastwasMadison,Jefferson, feasiblecandidates w, x,
y, thenMill, feasiblecandidatez, and Marx,you would assignthemthe
numbers8, 7, 6, 5, 4, 3, 2, and 1, respectively. Now thefeasible candidate
withthe highesttotal scorewill receivetheappointment. Thisprocess,the
argumentclaims,violatesthe principle.In orderto see this,the argument
would go: simplynoticethatif you changedyourmindaboutMadisonand
Jefferson, two infeasible candidates, and rankedthemjustbelowx, theneach
of w and x would,in yournew ranking, R', receivetwo pointsmore,which
could conceivably be enoughto changethetotalvoteoutcome,makingone
of themthewinner.Noticenow thatrelativeto each other,preferences for
we are doing it (and thereis good reasonto believehe did), thenhe shouldhave done it
thisway, and a greatdeal of confusionwould have been avoided.
feasiblealternativeshaveremainedthesamebetweenthetwosituations, but
the social choice is (by assumptionthat this changein total pointswas
enoughto changetheoutcome)different. Thisis exactlythetypeofbehavior
our principledeclaresthatno processhas,and yetherewe see it. The critic
wouldsaytheprinciple is wrongandshouldbe discarded.
Have we demonstrated that the principleis wrongand shouldbe dis-
carded?It is not that easy. The justification of the principleagainstthis
attackrestson a deeper(gametheoretic) principle aboutthenaturalstrategic
behaviorof individualsinvolvedin such a process.The situationwe have
postulatedpresupposesa typeof "honest"reporting of preferenceswhich
could nothaveoccurredaccording to thistheory.Ifyou couldhavebeensure
of changingthe outcometo some morefavorableoutcomeby someslight
"misrepresentation" of yourpreferences, you wouldhavedone so. It could
neverhavebeento youradvantage to "honestly"reportyourfirstranking, R.
Since neitherMadisonnor Jefferson could possiblybe chosen,you have
nothingto lose by reporting themas beingrankedlowerthantheyreallyare.
The consequenceof thisbehavioris to breakthetiesbetweenyourprefer-
encesfortheinfeasible optionsandwhatyoureport.You wouldreportwhat
you see as strategically mostadvantageous. Since yourstrategicadvantages
have nothingto do withyourpreferences forinfeasibleoptions,whatyou
reporthas nothingto do withsuch preferences. Consequently,ifpreferences
fortheinfeasible change(withoutsomechangeforthefeasible)youwillnot
changewhatyou report,and thustheprocessoutcomewillnot change.But
that is exactlywhatPrincipleVI, Independenceof InfeasibleAlternatives,
declares.We concludethat the proposedprocesscould not behaveas re-
ported,andtheprinciple remainsintact.
Withtheargument above we can beginto geta feeling forwhatwe areup
against.We mightquibbleaboutwhether or notpeoplealwaysbehaveinsuch
a "dishonest"manner,but that would miss the point.In orderto use a
processwhichviolatesPrincipleVI we mustbe assured(givenour current
understanding) thatindividuals willneverbehavestrategically, and thatis an
entirely different
thing.
Well,we can see thatwe pickeda bad example,buthowcanwe generalize
PrincipleVI to all processes?Thatis doneby relying on thetheoryofgames.
As it turnsout, all processes,at least all reportedto date,can be modeled
withinthe game theoreticframework. Anythingyou have seen or can
describecan be modeledas a game,or so the gametheoretic practitioners
would claim. If you have seen it, thentheycan modelit. This is not to
say thattheirjob wouldbe easy or practical.It maytakethema lifetime or
two to do it,but thatis due to thecurrent limitationsofmathematics, etc.,
and does not bear upon the validityof principles whichmay governtheir
models.
Almost all game theoreticmodels satisfyour principle.The qualifier
"almost"is therefora purposeand signalsthe existenceof a pathwhich
mightbe fruitfully followed.Beforeseeingwhereit goes,let's examinethe
two majorcategoriesof the gametheoreticmodels-noncooperative games
and cooperativegames.The relationship betweenour principlesand nonco-
operativegameswas firstformally introduced byPlott(1971). In thenonco-
operativegame setup,each individualhas partialcontroloverthe option
resultingfromthe process.Votingis onlyone waysuchcontrolsare mani-
fested.In general,each individualhas his own set of feasibleactionsfrom
whichhe is freeto choose.The socialchoiceis theoptionwhichresultsfrom
any particular patternofindividual actions-eachfromhisownsetof feasible
actions.Suppose,forexample,a groupof two peoplein isolationfromall
otherconsiderations is to choose the overallroomtemperature in a room
withtwo heatingunits,each beingcontrolled by a different
individual.Each,
then,has a set of feasibleactionsdefinedby thetechnological of
capabilities
theheaterunderhis control.Each has partialcontroloverthesocialoptions
(overall room temperature). Game theoryseeks an understanding of the
choiceseach individual wouldmake,givenhis preferences amongthevarious
roomtemperatures, his recognitionof theinterdependence betweenthetwo,
etc. Thereare severalcompetingmodels,withinthisgeneralmodel,which
predictthe outcome.The most widelyacceptedis the Nash equilibrium
model,whichholds thatthe two willjockey aroundtheirindividualdials
untiltheyreachan overalltemperature-a socialchoice-suchthatneither, by
changing his dial alone,can improveupon thesituation accordingto hisown
tastes.More complicatedtheorieshave each individualmaintaining some
theoryabout how the otherindividualwill reactto any changehe might
make,etc. The generaland profoundmessageis,however, thatnoneof these
theoriesaboutwhattheoutcomewillbe relyuponhoweitherindividual feels
about room temperatures whichcannotbe achieved.The choice,then,is
independent of preferences forinfeasibleoptions.13
13This can be formalizedas follows: Let Ai be the set of actions reservedfor

n
individuali and let the CartesianproductE =rlH Ai be the universalset of social actions.
If v c E, let Ai(v) be the projection of v on Ai and let the admissiblesets of v be
v : iil Ai(v) = v . The choice functionis definedas C(v, R,, . . ., Rn) where Ri are
rankingsof E. The Nash equilibriumscan then be defined as N(v, R,..., =
.( a)i, * *, a v (aji, .
--v: ai-j1,ai, ai+ I .*, ) Ri (ai, . *,ai-j1, ai, aij+ , a)
for all aie Ai(v) and all i}. If C(v, R,. . ., Rn) C N(v, R,. . ., Rn) A k, then C(v,
Whatabout cooperativegames?The key idea thereis theidea of domi-

nance whichWilson(1971a, 1972b), Bloomfieldand Wilson(1972), and
Bloomfield(1976) have integrated into the social choice framework. An
optionx is saidto dominatean optiony in casethereis somegroupof people
each of whom prefersx to y and who collectivelyhave "the powerto
implement x overy." Dominancethenis the coincidenceof "power"and
preference."
In orderto getfromthe conceptof dominanceto theactualoutcomeof
theprocess,solutionconceptsare used. Solutionconceptsare modelswhich
use theidea of dominanceas a basic parameter in predicting whichoptions,
froma largeset of options,will be the resultantgroupchoice.Thereare
severalcompetingsolutionconcepts.For example,the core of a set of
feasibleoptionsis a set of options,each of whichis feasibleand each of
whichis undominated by any otherfeasibleoption.An applicationof the
resulting"core model"wouldbe a prediction thattheoptionresulting from
theprocesswouldbe somememberofthecore."
Thereare othersolutionconceptsas well,buttheyall sharewiththecore
the property thatthe predictedoutcomedependsonlyupon thedominance
configuration amongthefeasibleoptionsandnottheinfeasible options.So as
longas thedominancerelationdoes notdependuponindividuals' preference
for infeasibleoptions,then our PrincipleVI, Independenceof Infeasible
Alternatives,is satisfied.
We can now see whata processmustlook like if PrincipleVI and the
resultingconsequencesare to be avoided.The "rights andpowers"of individ-
uals and groupsmustbe sensitive to and dependupon theiropinionsabout
infeasibleoptions.All processeswhichdo not conformto the restrictions
imposedby Principle VI canbe foundamongthosein whichwhatyoucan do
dependsupon what you woulddo in somehypothetical situations.To date
no one has foundsucha process.The focusof theresearch effortis clear,but
the outlook is dim, as will be discoveredin the sectionof strategyproof
processes.
So herewe are.PrincipleVI has done a prettygoodjob ofsurviving. It is
RI . ., Rn) satisfiesPrincipleVI. Note there are obvious difficultieswith existence,

etc.,but those are not our concernhere.
14Let N be the set of individualsand let A(x,y) = (C C N: C has the power to
implementx over y}. We can say that x dominatesy in case thereis a C E A(x,y) such
thatxPiy forall i E C.
15 We can say thatC(v, R1, . . ., Rn) = {x E v:
yD(RI, . . ., Rn)x forno y E v}.
tarnishedand damageda little,but it is clearlystillin theair.Certainlythe

rubblein thispathcannotbe removedby simplehandwaving. Let'sgo back.
CardinalUtilities
This problemis so complicatedthatit would take a book-length paper

simplyto exploreall ofitsfacets.First,theconceptof cardinalutilityitselfis
unclear.Different scholarsuse thewordto meandifferent things.Secondly,
what it means to "compareutilities"across people has not been clearly
specified.Ifwe havenumbers representing people'sattitudes,it is easyto find
manynice functions (Hildreth,1953; Harsanyi,1955; DeMeyerand Plott,
1971; Sen, 1976), SO it is tempting to claimthatthe "obvious"wayaround
the problemis to allow interpersonal comparison of utilities.Judging from
the commentsone findsin the literature, the urgeto represent individual
attitudesby something morethanrankings is almostoverwhelming. However,
thenumbersmustbe interpreted, and closeexamination suggests thatthereis
nothing simpleor obviousabouthowone mightproceed.
Let's digressa littleon themeaningof cardinalutility. Whatdoesit mean
to say thatsomeonehas "cardinalutility"?The textbookansweris thathe
has a utilityfunctionuniqueup to linear(affine)transformations, but that
simplyinducesthesamequestionwith"uniqueup to linear(affine)transfor-
mation"substituted for"cardinalutility."Whatdoes it meanto haveone of
those?Unfortunately thereare severalanswers, noneofwhichseemto getto
theheartof theethicalquestionabouthowsocietyshouldchoose,whichwe
wouldlikeanswered. We consideronlytwohere.
The firstanswercomesfromthedecisionmaking-under-uncertainty litera-
ture.Certainpreference oflotteries
rankings arestructured
as ifthereisa utility
value attachedto each lotteryoutcomeand as if each lotteryis ranked
accordingto theutilityit is expectedto yield.For example,considerlottery
one, which yields one apple with probabilityPA and one orangewith
probability PO; and considerlotterytwo,whichyieldsan applewithproba-
bilityP'A and an orangewithprobability P'O. Somerankings overthesetwo
lotteries,alongwithothers,arestructured as iftheapplewouldyielda utility
(experience?)of magnitude UA and theorangewouldyieldone of UO, and
lotteryone would be rankedover lotterytwo in case PAUA + POUO >
P'AUA + P'0UO. Now thewords"utility,""experience," and "magnitude"
havenot been defined.Furthermore, theyhaveneverbeendefinedandneed
not be definedforthisconceptionof cardinalutility.Anynumbers willdo
for UA and UO as long as the resulting expectationcalculationsyielda
ranking of thelotteries identicalwiththepreference ranking underconsidera-
540 CharlesR. Plott
tion. Butwaita minute!Whyare theretwo conceptshere-expectation and

preference ranking? The answeris thattherearen't.Thereis onlya preference
ranking. We just searcharoundfornumberswhichallow us to expressthe
ranking as ifexpectation calculationsweregoingon.
Whatdo thesenumbershaveto do withour ethicalproblem?Well,they
are "cardinalutilities"in the sense they are "unique to linear(affine)
transformations" but theycarryno moreinformation at all thanthesimple
rankingof lotteriesfromwhich they were derived.Don't they express
"intensity of desire"?Onlyin thesamesensethatyourwillingness to pay a
maximumof $.25 for an apple and a maximumof $.05 for an orange
expressesan "intensity of desire."They onlyexpressthe factthatyou are
indifferent amongthe following threeoptions:(1) youhavethe$.25 and no
apple and no orange,(2) you do not havethe $.25 and haveno orangebut
you do have the apple,(3) you have$.20 and theorangebut you haveno
apple. Nothingmorethana rankingis involved.This typeof intensity was
neverprohibitedby our formulation of the problem(Campbell,1973).
Intensities like thiswere permissible all along,sincetheyonlyamountto a
ranking andwe stillobtainedan impossibility result.
Whathappened?Ouroriginalproblemformulation was muchbroaderthan
you mighthave expected,and in additionthis conceptionof "cardinal
utility"mayhavebeen muchnarrower thanyou wanted.By allowinglotter-
ies to be amongtheoptions,in thesamewaythatwe allowmoneyin handto
be a dimension of options,we haveno "external"measureof tradeoffs, and
thusare leftwithrankings alone.Thisis probablyoflittleconsequence, since
few would accept "desire to pay" as a measureof intensity on whichto
ultimately base moraljudgments. Our desireto pay is obviouslysensitive to
income.Likewise,tradeoffs betweenlotteriesare likelyto be judged inade-
quate becausetheyare"sensitive to attitudes towardsrisk."Judging intensity
in termsof the distanceone would runin orderto have one option over
anotherdependsupon"attitudetowardsrunning" judging and in terms of the
heightof a clifffromwhichone wouldjump in order to have one option over
anotherdependsupon"attitudetowardsheight,"etc.
So thereare manymeasuresof intensity. Theyhaveall beenwell recog-
nized for yearsbut forsome reasonnone have receiveduniversalor even
partialacclaim.Whynot? Perhapstheydo not sit well withour intuitive
notionsabout what an intensitymeasureshould give us. Perhapswe are
wantingmoreinformation thanis containedin a simpleranking, but what
it
could be?
The leadinginterpretation of whatwe are askingis thatwe wantto know
foreveryset of fouroptions,say 3w,x,y,z, whetheran individual's prefer-
ence foroptionw overoptionx is greateror less than his preference for
optiony overoptionz (Alt, 1971). The information neededthen,in addition

to thepreference ranking overoptions,is an additionalranking overpairsof
options.Bynowwe areaccustomedto roadblocks, so let'ssee whattheylook
likehere.
The firstand mostobviousproblemis one of establishing theexistenceof
the ranking overpairsof optionsas opposedto simplya ranking overoptions.
A preference rankingoveroptionscan be objectivelyconnectedto choice
behaviorthroughthe theoryof revealedpreference (Richter,1966, 1971).
Anothereasier,but perhapsmorecontroversial, way is to assumethatthe
word"preference" meanssomething, perhapseventhesamething,to every-
one. So, if you simplyask peoplefortheirpreference, theiranswercan be
takenas meaningful. However,the idea of a rankingoverpairsof options
does not seem to have thisoperationalsupport.Thereare folksout there,
e.g.,me,who wouldclaimthata ranking overpairsofoptionshasno general,
unique meaning.Whatdo you do with a personwho, whenasked if his
preference forw overx is greaterthanhis preference fory overz, responds
by sayingthathe does not understand thequestion?He maybe willingto
order"similarity" betweenpairsof options;he wouldsay,forexample,that
thesimilarity betweenw and x was greater thanthesimilarity betweeny and
z, butis unableto substitute theword"preference" in thiscontext.
The impulseis to probe,usingnewwordslike desire,want,strength, and
usingthe factthatalmosteveryonehas some experienceto whichhe will
simultaneously attach the words "intense"and "feeling,"untilyou have
founda questionthathe willanswerandwhichdoesindeedyieldan ordering
over pairs of options.Thermometer questions,"How warmdo you feel
towardx?" seemto be of thissort.Butoncetheordering is obtainedby one
meansor another,we face a new problemeven thoughthe intensity data
mightbe usedto accurately predictbehavior. Whatmoralsignificance can one
associatewithit? The veryprobingrequiredto getthedata has createdthe
possibilitythattheindividual has simplysubstituted forthequestionone of
themeasuresthatwe havealreadyrejected(amounthe wouldpay,heightof
cliff,etc.) as a basis for moraljudgments, even thoughsuch substitution
wouldmakeno difference at all forscientificand prediction purposes.How
do we guardagainstthat?I don'tthinktheansweris known.
Problemsdo not stophere.Evenifsomeconceptof cardinalutilitycanbe
establishedas a basis formoraljudgments, we facesomenewand different
problems.First,thereis a problemof constructing an acceptablesocial
preference definitionfromthe cardinaldata as opposedto theordinaldata
withwhichwe havebeenattempting to deal,and secondly,thereis a problem
ofdesigning a systemwhichwillbe sensitive to thesedata.
How do we definethe social preference to use acceptedmeasuresof
542 CharlesR. Plott
"utility"?We are facing,of course, the problemhere on interpersonal

comparisons. It is not a simplejob of addingup numbers, sinceourdataare
in theformof rankings of pairsofoptions.Wehaveno "unit"and no "zero"
on whichto fix a scale. As it turnsout,we havejust as delicatea problem
here as we did with our originalproblemwhenwe only had rankings of
optionsto workwith.A fundamental impossibility resultbased on cardinal
utilitiescan be foundin Schwartz(1970, 1972). The problemis thatvery
reasonableprinciples regardinghowone usesthesenewdatalead immediately
to contradictions.
Since theSchwartzresultmightnotbe theultimate impasse,let'scontinue
to look ahead. Supposewe have an acceptablesocial preference definition
based on such data. How are we goingto designa processwhichuses it?
Independenceof InfeasibleAlternatives, PrincipleVI, tellsus thatprocesses
in generalare not sensitiveto rankings of pairsof options.They are only
sensitiveto the rankings of options.For example,one of themajorconclu-
sions of nineteenth centuryeconomictheoristswas thatthe existenceof
cardinalutilitywas unnecessary foran explanationof the workingsof an
economicsystem.The systemwillworkthesameregardless of themagnitude
of cardinalutilities,as longas theordinalutilities remainthesame.Principle
VI is tellingus thateveniftherelevant parameters arein thesocialpreference
definition, thesystem willnotact on them.
So, we are stuck,evenhere,withour originalimpossibility result.Unless
we can come up withsome new kind of processwhichsystematically uses
these"intensity"data,we are blocked.The problemis now one thatwas
discussedearlier.If theprocessis sensitive to individualrankings of pairs,and
if it is not dictatorial,you will reportwhateverranking of pairsis to your
advantageto report.Your strategicbehaviorcauses the outcometo be
somethingotherthan what societyby definition prefers, therebycausing
PrincipleVII, RationalChoice,to be violated.Sinceno one elsecansee your
ranking of pairs,yourstrategic behaviorcannotbe prevented. Foran example
see DyerandMiles(1976).
Thereare evencomplicating factors.A naturalwayto proceedis to tryto
designa two-stage process.The first wouldsystematically enticeyou to reveal
yourtrueranking of pairs,whilethesecondwoulduse theinformation about
yourtruepreferences, thusobtained,as inputsto determine thesocialchoice.
But this directionhas been shownby Gibbard(1973) and Satterthwaite
(1975) to lead to dead ends.Theyhaveconcludedthatwithina wideclassof
specifications,thenecessary initialprocessdoesnotexist.
So herewe are once againfacinga return trip.By simplyclaiming thatwe
plan to use theconceptof cardinalutility, we solveno problems at all. Unless
one can see howto tiptoethrough theobjections, thisroutemustbe declared

a dead end. However,like theotherpaths,thisone doesnot reallyendin the
middleof spacewithabsolutely no forward alternativeleft,butwhatremains
a job fora dedicatedcreativescout.Let'sgo back.
is definitely
Axioms
The Transitivity
Well,we've examinedmost of the principlesnow, and have failedto
discharge or evensuccessfully modifyany of them.OnlyPrinciples I, II, III,
and VII remain.Principles I and II, Preferenceand Indifference Transitivity,
set forthsomeof thetechnicalproperties we mightexpecta socialpreference
to have. PrinciplesIII, Value/Feasibility Separation,and VII, Rational
Choice,on theotherhand,formtheverybasisforusinga conceptlikesocial
preferencein the firstplace. If those are removed,then we have really
changedthenatureofthewholeenterprise. Let's tryI and II first.
Evenin thecase of individual choicethereis evidencethatI and II do not
alwayshold. Recall,we obtainedtheseprinciples byan analogyto individual
preference, so we haveamplejustification forexploring whatmightresultif
theyareremovedormodified.
Whathappensifwe removeconditionII, the requirement thatthesocial
indifference is alwaystransitive? That meanswe are willingto live with
situations whenxI,y and yIlz but xP,z. Societyis indifferent betweenx and
y and it is, by definition,indifferent betweeny andz but,withnothing at all
changed,it prefers x to z. The socialindifference is no longerrequiredto be
transitive,but of coursethe socialpreference relationis stilltransitive. For
technicalreasons,attitudes withthisproperty arecalledquasi-transitive.
The answerwas suppliedin Sen (1970a, pp. 76-77), Mas-Colelland
Sonnenschein (1972, Theorem1).
Theorem.IfPrincipleII is removed,V is replacedby theParetoPrinciple, and

areretained,
ifall otherconditions thenthereexistsa setofindividuals,sayc,
suchthatforany pairof options, x,y , if xPiyforall members i of c, then
xPsy.'6 Furthermore, if xPiy forsome member(s)of c and yPixforsome
othermember(s), thenxIsy.
The theoremsays thatthe only social preference definitions whichare

consistentwith our set of principlesare oligarchies.That is, theremust
16 If the group is unanimouslyindifferent,then there could be a "second order"

groupwhich"takes over" in lexicographicfashion,the powersof the firstgroup.
necessarilyexista set of individuals suchthatwhentheyare united,unani-

mous in preference,thenthesocialpreference is decided,and if theyare in
thensocietyis indifferent.
conflict, If theyarenotin conflict,thenanysocial
"nonindifferences" must be due to unanimity amonga second coalition
(possiblydisjointfromthe first)whichis unanimousin opinionand Which
has "powers"whichbecomeactiveonlywhenthe firstcoalitionis indiffer-
ent.
Thougholigarchy is probablybetterthandictatorship, thistypeof social
preferencedefinition is not likelyto winwide acclaim.If one is opposedto
oligarchy,then some otherprinciplemustbe altered.Whathappensif, in
additionto eliminating PrincipleII, we alterPrinciple
I as follows?
I'. Acyclicity.The social preferenceis not cyclic.That is, for no
sequenceof options,x1,x2, . ,xm,do we havexl Px2 and x2Px3
and ... andxm_1PsXmandxmPsxli
The requirement that the social preference
be transitive,
i.e., xPsyand
yPsz impliesxPsz, has been dropped.Insteadwe requirethat the social
preference is not cyclic.We ruleoutbehaviorlikexPsyandyPszandzPsx,so
we knowthatxPsyand yPszimplythateitherxIsz or xPszis thecase.Ifwe
makethischange,we are no longerfreeto speakat all ofa socialpreference
ranking.No longerwillit necessarily makesenseto talkaboutthesecondand
thirdbest, etc,, which formthe essenceof the conceptof a ranking.17
Insteadwe can speak of a socialacceptability relation,whichwillbe desig-
natedas RS.18 By xRsywe willmeanthatx is just as acceptableas is y; by
xPsywe willmeanxRsybutnot yRsx;and by xIsywe will meanxRsyand
yR,x. In thisnewjargonour PrincipleI' has the implicationthatforany
finiteset of feasibleoptions,therenecessarily
existsomeoptionsthatarejust
as acceptableas any otherfeasibleoptionin the set (Sen, 1970a, Lemma
17 Suppose we have xlz, xPy, yPz. Now x seems clearlyacceptable,since it is just as
acceptable as both y and z, but we would probablyhesitateto say thatit is rankedfirst,

since it is not clear what would be ranked second. One mightargue that y should be
ranked second, but notice that there would be some ambiguity,since we would be
rankingit above z, which is just as acceptable as the thingrankedfirst.Thus we can see
thatin cases like thisthe concept of a rankingis not reallyappropriate.
" You will not find this term in the literature.Instead, the termsocial preference
relationis generallyused, even thoughthe termpreferencecan be confusingbecause it
does not mean the Ps we have been using.The Ps is called strictpreferencerelationin the
literature,and the relation Rs is called the preferencerelation.What we are callinga
social acceptabilityrelationSen (1970), forexample,would call a CollectiveChoice Rule
and if we added the requirementI' it would be called a Social Decision Function.
1* 1). If we take"optimum"to mean"nothingis better,"thenonlythebare

bonesof theconceptofan "optimum"remains.
Whathappensnow is not completely known.A completecharacterization
of thesocialpreference definitions withI' and ourotherprinciples
consistent
hasnotyetbeenobtained.
The following resultis due to Brown(1973a,b, 1974, 1975a,b),who has
been seekingto characterizethe definitionin termsof what he calls a
collegium.A collegiumis a set of individualssuchthatanymemberinside
theset has an absolutevetoand no memberoutsidethesethas a veto.For a
motionto carry,thecollegiummustbe unanimous and themotionmustalso
be ratified by someselectcoalitionotherthan(disjointfrom)thecollegium.
Theorem.If thesocialpreference definition
and processsatisfy
I', Acyclicity;
Separation;IV, Universal
III, Value/Feasibility Domain;VI, Independence of
VII, RationalChoice;andVIII, ParetoPrinciple,
InfeasibleAlternatives; then
thereexistsa specialset of individuals
CO and selectsetsofindividualsXC1,
C2, ... ,Ckt suchthatif xPiyfori E LCo U Cj forsomeCj fromamongthe
selectsets,thenxPsy.
This is an illuminating result,but problemsremain.If we use a process

based on a collegium, thentheinducedsocial preference maynot
definition
satisfyI', III, IV, VI, VII, andVIII. Furthermore,theconditions underwhich
themembersof thesetCOin thetheorem havethevetopowersof
necessarily
collegiummembersare stillunknown.Thatis, thesocial preference maybe
determined in manyinstanceseventhoughthe collegiumis not unanimous.
Nevertheless, thisresultand othershe providesshowthatcollegiums are in
somesenseclose to forming theonlypreference definitionscompatible with
ournewprinciples, butadditionalworkremains.
By replacingV witha different normativeprincipleto be usedin addition
to I', III, IV, VI, VII, and VIII, we cangetanotherviewofwhatis goingon.
The following principlesaysthateveryindividual hasa tie-breakingcapacity.
IX. StrongMonotonicity.If xIsy, givensome patternof individual
rankings(R1, . . . Rn), and one personchangeshis mind,he either
movesfromindifference movesfrompreference
to preference, to
or reverses
indifference, thensocietymovesfromindif-
preferences;
to preference
ference (in thedirectionofhischange).
The following in Mas-ColellandSonnenschein
resultwas presented (1972).
Theorem.If the processsatisfies Separa-

III, Value/Feasibility
I', Acyclicity;
tion;IV, Universal
Domain;VI, Independence of Infeasible
Alternatives;
VII,
546 CharlesR. Plott
RationalChoice; VIII, ParetoPrinciple;and IX, StrongMonotonicity;

then
thereis an individualio who has veto poweroverany pairof alternatives.
Thatis,ifxPioy,theneitherxIy orxP,y.
An interesting breakthrough was provided

by FerejohnandGrether (1974)
forthecase of a finitenumberof options.Supposethenumberof optionsis
M; thenumberof peoplewho preferx to y is N(x,y);and thenumberwho
prefery to x is N(y,x).A socialpreference
definition
we willcalltheFG rule
is
FG: xP,yifandonlyifN(x,y)> (M ) (N(x,y)+ N(y,x)).
This rulesatisfiesall of our principlesI' throughVI and in a senseis the

"best" rulewhichdoesso in thisfinitecase.'9 The troubleis thatifM is very
largerelative to thenumberof people,it is equivalentto unanimity.
That's it. Acyclicityis the weakestconditionwhichwe can use and still
havethesocialpreference resembleanything likean individual
preference.
It
is as relaxedas we can becomeand stilluse theconceptof an "optimum."
The road is not closed. Acyclicity may providethe answer,but informed
observers (Ferejohn,1976) say thatanyday nowwe will finda "collegium-
like"structure formingtheendof thepath.Let's return to theorigin.
Responsive
WhilePrincipleV andsimilaraxiomssuchas theParetoPrinciple look very
innocentindeed,and haveneverbeen questionedin thisliterature, I conjec-
turethatwhenjudgedfromthebroaderperspective of thehistoryof ideas,
theyare themostcontroversial of themall. Theysay thatthesocialprefer-
ence definitionshouldin somesensebe basedon individual preferences.The
definitioncan neveroperate in total disregardfor all preferences of all
individuals.
Beforegoingto whatmayseemto someto be theheartof theproblem,
let's dwellmomentarilyon whatcouldbe a misconception abouttheanalysis.
"Whyshould the social preference be sensitiveto individualpreferences?
Individualsare notablymisinformedand mayindeedbe undertheinfluence
19 Among those rules which are symmetric

in the arguments,it is the one forwhich
the P relationis the largest.
of somethinglike advertising and therefore unawareof their'true' prefer-

ences." Thereis a simpleanswerto this.We haveproblemsevenwhenthe
dataare thetruepreferences. If we had a theoryor methodto separateout
"true"preferences fromthose"undertheinfluence," and ifwe wouldwant
the social preference to be responsive to thetrueones,thenwe wouldstill
have all the problemswe have been outlining.Simplytake the individual
preferences we havebeen usingto be the"truepreferences" and startfrom
thebeginning.
Whatabout social preference definitionswhichare foundedon religious
codes and customs?Theseneednotbe responsive to individualpreferencesat
all. In fact,one cansee muchof democratic theoryas saying,"If youmustgo
through thisexerciseof findinga social preference
definition,thenPrinciple
V shouldbe partof it." Noticethatit is here,as opposedto transitivity,that
the referenceto democratictheoryis most useful.So those who reject
democratictheoryin generalare likelyto rejectthisprinciplealso. Such
individuals mayhaveproblems withinwhatever systemtheychoose,butthey
are not goingto be troubledby theanalysiswe are pursuing. In particular,
they should not have even found our initialexamplesparadoxical,and
probablystoppedreadinglongbeforenow. For the restof us, however,the
axiomremains, andwe mustsearchelsewhere.
RationalChoice
BrianBarryhas suggestedthattherationalchoiceaxiom,VII, maybe the
problem.20Why should we expect the "most preferred"to alwaysbe
chosen?Perhapswe shoulddefinethe social preference and thenlook for
processeswhichcome as close as possibleto yieldingthebestoutcome.Of
coursewe have to cometo gripswithwhatit meansto be "close," but the
idea seemssound.Whathappenswhenwe dropaxiomVII? It dependsupon
whatwe replaceit with,but in generaltheansweris thatwe currently
don't
know.The idea shouldbe pursued.
Value/Feasibility
Separation
Whathappensif axiomVII remains? Thereis onlyone axiomleft-axiom
III. Everything
elsemuststay,eitherbecauseofwhatnaturehas doneto us or
20 He raised this issue in conversation at a conference on "Social Choice and
DemocraticTheory,"Key Biscayne,1975.
becauseof "innocence."Let'sexaminewhatthisprinciple saysandsee where

it comesfrom.Likeit ornot,theonlychoice,givenwhatis currently known,
is likelyto be betweenthis principleand one of the dictator-like social
preference definitions.
PrincipleIII saysthatthesocial preference betweenanypairof
definition
options,say x, y, does not changeas the set withinwhichtheyare being
consideredchanges-allotherrelevantthingsbeingheldconstant.If we are
goingto use a conceptof preference at thesocietallevel,thenwe need this
principlesinceit formsa largepartof theheartof preference theory.Our
recognition withinourselvesof the independenceof our desiresfromour
opportunities, of our freedomto proceedwithvalue calculationsindepen-
dentlyof feasibility servesas a base formuchof thetheory's
considerations,
appeal.Doingawaywiththisis,in essence,doingawaywiththeconceptofa
socialpreference definitionaltogether.
Well,thattakesa littlethought. Whydidwe getintothesocialpreference
businessin the firstplace,andwhatarethealternatives? Wegotherebecause
of our intuitionand becauseof a lot of thingswe findin books.Of course
books can be wrongand our intuition canlead us astray.It wouldnotbe the
firsttimethatsomeoneillegitimately transferred theprincipleswhichgovern
his own behaviorto otherthings.Humanqualitiesare attributed to plants,
automobiles,and computers everyday. If we eliminatePrincipleIII, we are
admitting, as claimedby Plott(1972) and Schwartz(1972), thatwe have
madethesamemistakewithgroupsofindividuals.
Whatare the alternatives? Shouldn'tsocialchoicesfollowsomestandards?
Yes, but standards do not haveto look like preferences.Theycan look like
prohibitions againstcertainevents.That is, standardscan look like con-
straints,e.g.: "the outcomeshouldnot violateanyone'srights"; "Thou shalt
not kill"; "No one shouldstarveto death"; "No Paretodominatedoption
shouldbe chosen";etc. So, ar,alternative to theuse of a social preference
definition is to use constraints.
Ratherthanattempting to findwhatis most
preferred or mostacceptable,we can tryto findsomething whichis simply
acceptableas opposedto something whichis optimal.
SearchforChoicePrinciples
thenthereis no needto thinkin termsofa
If PrincipleIII is eliminated,
social preference definition on theprocessand
at all. We can focusdirectly
inquireabout the implicationsof behavioralconstraints. Are variouscon-
straints(which carryinteresting normativeimplications)consistent?What
kindof processmightcorrespond Can we invent
to certaintypesofbehavior?
new processeswhichhavethoseaspectswe likeabout existingprocessesbut

do nothavesomeofthemoredistasteful properties?
This changein focusmayinvolvea subtlechangein philosophical position.
Perhapswe are no longerattempting to definethegoodoptionswhichwould
serveas a standardagainstwhichprocessperformance isjudged.Wemayhave
insteadbeguna searchforacceptableprocessesand procedures whichareto
be judgedin termsof theirbehaviorin general, and notin termsofhowwell
theychoose the "proper"outcome.We mayhave shiftedour focusfrom
questionsabout "substance"to questionsabout "procedure."In thelitera-
tureno one takessideson thisissue,so we dropit fornow.
How do we proceedwithouta conceptofsocialpreference? We canplace
our normativedemands on the choice function,the process model
C(v,R1,... Rn), directly.For example,the Pareto Principlebecomes:If
everyone ranksx abovey andifx is feasible,theny ? C(v,R1,. . . Rn); i.e.,y
does not resultfromthe process.Othernormscanbe similarly stated.Notice
in particularthat the statementof this normdoes not requirea social
preference concept.We do not pickup a conceptofsocialpreference simply
becausewe wantto engagein normative The pointis subtleandis a
discourse.
frequent causeof controversy.
In spiteof our declarationthatwe havedone awaywiththeconceptofa
social preference,
it can creepintotheanalysisthrough thebackdoor."O K,
so we will no longerattemptto finda socialpreference definition. We can
forgetabout thatas long as we adopt the properprinciples to governthe
selectionof a social choice function.For example,don't you feel there
shouldbe some consistency betweenthe outcomewhichresultsfromsmall
feasiblesetsand theoutcomewhichresultsfromlargersets?"Ourintroduc-
toryexamplescan make this argumentseem mostappealing,but if your
answeris yes, thenyou are in greatdanger.Certainconsistency principles
governing choice functionsdirectlycarrythe implication thatsocietywill
chooseas if the choicesweregeneratedfromsomesocialpreference defini-
tion.Ifwe subscribe to one of them,we arebackwherewe started.
The classicexampleis the following, whichin theliterature is calledthe
WeakAxiomof RevealedPreference.
WARP.Suppose the social choice,when the feasibleset is v, is givenby

C(v,R,. Rn), and that it is C(v',Rl,... Rn) when the feasibleset is
v'-the rankings
of all individuals
remaining fixed.Ifv is a subsetofv', thenif
any option is in both v and C(v',Rl, . . . Rn), the set of options in both v and
C(v?,Rl, ....Rn) is exactly equal to C(v,R1, . . . Rn)
550 CharlesR. Plott
A picturemighthelp.(See Figure1.) Bybreaking theaxiomWARPintotwo

partswe can understand thediagrams.(i) If twooptionsaretiedfora "win"
in a contestwithina smallset,theneithertheytieagainoverthelargersetor
neitheris a "winner."(ii) If an optionwas a winnerfroma big set,thenit
shouldalso be a winnerinanysmallersubset.
principleis imposedand ifVI is retained(and ifv can
If thisconsistency
be anyfinitesubsetof E), thentheprocesschoosesas ifwe hadstartedwitha
social preference definitionwhichsatisfies1, 11,111,IV, V, and VII. The
resultsare all thesameas in thesocialpreference case,andwe endup witha
"dictator."
So, ifwe acceptall of WARP,thenwe areintrouble.Perhapsby dividing
it into"parts,"suchas in thediscussionabove,we couldkeepthe"desirable"
partand eliminatetherest.In doingso we wouldhavea betterunderstanding
of whatwe want in termsof processperformance. Thisidea has servedto
motivate muchcurrent research.
Of course,the troublewith the approachis that we have no general
guidelinesfor dividingup the axiom and no criterionfor "acceptable"
Axiom violated (part i) Axiom violated (part ii)
Axiom not violated
FIGURE 1
performance otherthanour own intuitionwhich,as we haveseen,is defi-

can come frominstitu-
nitelycapableof leadingus astray.Partialguidelines
tions.Otherscan come fromindividual behavioral as willbe seen
regularities,
in thenextsection,but currently whichgoverns
thereis no "metaprinciple"
thisportionof the area. Fromherethe subjectalso becomesverytechnical,
and since excellentsurveyshave been providedby Fishburn(1974c), Sen
(1976), and Richelson(1975a), we neednotcoverit.
and Loose Ends

Summary
A summary of whatwe have done would probablybe welcomeat this
point.Whatseemto be thestrongpointsand theweak pointsoftheformal
analysis?Whereare the variousinterpretations of the formalisms likelyto
hold,andwherearetheylikelyto breakdown?
The analysisstartedwitha fewexamplesof processbehaviorwhichserve
as convincing demonstrations for most thatthe worldis not at all as we
expected.Fromthesewe beginto developan overriding hypothesis about
whywe regardedthe behaviordemonstrated in theexamplesas peculiaror
"paradoxical."We foundthat theseexamplesviolatecertainprinciples of
preference whichare generallypresentin individualchoicebehavior.The
questionthen became "What processesdo choose in accord with these
principles?" The answerwhichhas been slowlyseepingin overthepastfew
yearsis now beginning to take a definiteshape.Thereappearto be none
whicharefreeof "dictator-like" features.
ThisnegativeanswerbeganwithArrow'sGeneralPossibility Theorem.At
first,since therewere so many abstractand complicated aspects of the
argument, thetheoremwas essentially ignored.Most scholars suspected that
buriedsomewherein the jargonwas a key assumption which, once made
explicit,wouldeithermakethewholethinggo awayor at leastprovethatthe
resultwas relevantto onlysome specialand isolatedsituations. Withsome
theycontinuedto base normative
justification, analysisuponconceptswhich
the theoremsuggestedwere inconsistent. Afterall, analysismustbegin
somewhere, and if all potentialproblemshad to be solvedbeforebeginning,
thebeginning wouldneveroccur.Within just thelast threeor fouryears,we
have movedourselvesto a muchbettervantagepointfromwhichwe can
surveywhatis goingon. Most of the obviouspotentialroutesaroundthe
resulthavebeen tried,and almosteveryaspectof thetechnicalconceptshas
beenexaminedin detail.
Table 2 providesa summary of severalof themajorideasabouthow to
dismantlethe result,and the typesof problemstheyencounter. The com-
552 CharlesR. Plott
TABLE 2
Attempts to Eliminate the General Possibility Theorem
Idea Answer
Eliminatethe Pareto It was not used, so thereis nothingto

Principle eliminate.Besides it only seems to pro-
vide some "direction" in the theorem.
Reduce the size of the It seems thoughverylittleconflictis

domain needed to generatenegativeresults.
More thanan amountadequate to cause
problemsis likelyto naturallyexist.
Remove Independenceof If the discussionis about processeswithinwhich

InfeasibleAlternatives individualsare placed in a positionwhere
theirdecisionsaffecttheirown total well-being,
thenthe axiom seems to be unavoidable.
Use CardinalUtilities a. In the sense of ratesof substitution,

amountsone would pay, and attitudes
towardlotteries,cardinalityis not precluded
if the optionsare viewedas social states.
b. Processesare not sensitiveto
cardinalityof the classicalvarietyso special
processeswhichare influencedby thesedata
need to be invented.Impossibility
resultsabout the existenceof such processes
have been discovered.
Remove Social Indiffer- It leads only to oligarchies.

ence Transitivity
Use Acyclicity It looks as thoughit will only lead to collegiums.
Remove Responsiveness This divorcesthe conceptsof social

preferenceand individualpreference.
Remove Value/Feasibility This is equivalentto doingaway withthe

Separation concept of social preference.
Remove Rational Social The consequencesof thisare unknown.

Choice
mentson thetablearegeneralizations aboutthestateoftheart.Theyarenot

claimsthatthe ideas listedshouldno longerbe pursued.Rather,thecom-
mentsunderlinethe importanceof doing even moreresearchalongthese
lines. That said, however,the generalizations point to a veryimportant
conclusionabouthow thisresearchshouldbe regarded. Untilrecently it has
not been unrealistic to presumethatthenegative results
wouldbe eliminated
through one of thesepathsof analysis.Therewas reallyno needforscholars
who used conceptslike "social preference," the "wantsand desiresof the
people," etc., to justifythe use of the conceptspriorto applicationor to
establishthattheconceptscouldbe consistently applied.The generalizations
on Table 2 indicatethatit maybe timefora shiftin theburdenof proof.
Now it is reasonableto supposethatthenegativeresultsshouldbe takenat
face value and thatour philosophical positionsmustbe alteredaccordingly.
If we are to taketheresultseriously, thenwhataretheconsequences? In
verygeneralterms,it seems that we cannotinquireabout whatkindsof
processes"givepeoplewhattheywant."The conceptof "process"seemsto
be inconsistent withtheconcept"whatthepeoplewant."The reasonappears
to be thatthelatterconcept,whichis actuallytheconceptof socialprefer-
ence in disguise,suffersfroma typeof fallacyof composition.It illegiti-
matelypresupposes thata property, preference,ofeachindividual in a group
is also a property ofthegroup.
For manyscholars,such a conclusionwill not be disconcerting at first,
sincetheyneverthoughtof a groupas havinga preference anyway.Econo-
mists,forexample,will usuallyaggressively avoidanysuchpresupposition.
But,thebasic problemis not theconceptof socialpreference itself.Rather,
the basic problemis the internalconsistency of "preference-like"behavior,
and thiscan be inducedby a greatnumberof concepts.Anyconceptwhich
inducesat the societallevel a consistency betweenchoicesover different
feasiblesets of the typerepresented by the axiomWARP,is likelyto have
problems.Theseincludeconceptswhichcarrytheconnotation of "optimum
outcome"or the"best outcome"or the"maximum"of anything, sinceeach
can be operationally equivalentto a preference. Now it would seemthat
anyoneusingtheseconceptsshouldat leastprovidethe restofus withsome
assurancethat his analysisdoes not founderon the mostobviousof the
inconsistencies.
Are thereimportant areaswherethe analysisdoes not apply?The key
problemis causedby theconsistency inducedby preference betweenchoices
over different feasiblesets. If in some instancesand forsome reasonthe
conceptof a feasibleset does not apply,thenitmaybe possibleto avoidthe
difficulties.What,forexample,is thefeasiblesetoflegislative enactments? Is
554 CharlesR. Plott
thisset differentfromthe universalset of all conceivablelegislative

enact-
ments?Does thisfeasiblesetchangefromtimeto time?Ifnot,thenwhenthe
optionsare piecesof legislationthe theoremmightnot apply.Whatabout
valuejudgments? Does thereexista feasiblesetwhichdiffers fromthesetof
all conceivablevaluejudgments?If valuejudgmentsare the optionsand if
individualsrankthem,thenwe mightbe able to talkabouta groupranking of
valuejudgments withoutgettingintotrouble.21
The "process"interpretation of Independenceof InfeasibleAlternatives
also carriesanotherimportant Theremaybe a good argument
qualification.
for the axiom if each optionis an entiresocial state.Whathappensif an
"option" is onlyone componentof a socialstate,as is thecasewithnormal
committeeprocesses?Then individuals can bringconsiderations externalto
theprocess-suchas bribes,side payments, theexternalnegotiations-to bear
in influencingtheprocess.The outcomeoftheprocesswouldseemto be very
sensitiveto committee members' attitudes
towardtheseexternaloptions,and
changesin preferences regardingthesethingsmay well resultin changesin
processoutcomes.As it standsnow, theapplicability of theaxiomin such
situationshasnotbeenestablished.
Perhapstheabovearguments willprovidethereaderwiththeflavorof the
negativeresultsand enablehim to explorethe possibleambiguities, when
examinedformthis point of view,of the philosophieswith whichhe is
familiar.Perhaps someone, when comparingphilosophicaloutlook, will
notice the key whichwill allow us to disregard the negativeresultscom-
pletely.We turnnow to the typesof analysisthathave proceededin full
recognition of theseproblems.
AXIOMATICMODELS OF PROCESSES,
PROCEDURES,AND INSTITUTIONS
Everyoneis dissatisfied
withnegativeresults.Evenwithinthemostsim-
such
plisticsettings cannotbe avoided.Yet, people do seemto get
results
along.Perhapstheydo not get along"optimally"but theydo seemto get
along.Maybeone shouldfocusthe analysison existing,acceptedprocesses-
the processesthatpeople use. Whatmakesthemtick?Whatare thediffer-
ences betweenprocesses?Whathappensif you make slightchangesin a
process?Perhapsifwe can answerthesequestions, thenwe willbe in a better
21
There is some controversyon thispoint.Compare Plott (1972) and Sen (1976).
positionto see whysomeprocessesare intuitively acceptableand othersare

not,andthusbe in a positionto improve upontheprocesseswe have.
Thatis exactlywhathas happened.If theliterature abovecan be citedas
partof a single-minded attemptto directly discredit
theimpossibilityresults,
theremaining literaturecan be seen as partof an attemptto finda different
approach.The huntinggrounds,possiblythe onlyones,are theexisting(in
use at some time) processesand the imagined(potential)processes.The
searchhas only recentlybegunand the problemsare veryhard.Nothing
earthshaking has been discoveredyet, but now and then glimmers occur
whichindicatewhattheefforts mightyield.Theseglimmers arethetopicsin
thissection.
Whilethereis not exactlyagreement on whatwe are lookingfor,there
appearsto be someagreement on somebackground conditionsforthesearch.
First,we are still talkingabout the use of axiomaticmethods,so we are
talkingabout a body of literature withinwhichthereis some agreement
about what constitutesa result.Secondly,we are still talkingabout the
essentiallynormativequestionof what processesare generallyacceptable,
whytheyare acceptable,and whatlies at thebase of suchjudgments. Taken
together,thesetwogroundrulesserveto narrowthefieldconsiderably.
The firstgroundrulemeansthattheefforts we revieware focusedupon
processmodelsand not necessarily the processesthemselves. Thisdoes not
mean that the processesremainunexamined.Giventhe currentstate of
knowledge, it reallymeansthe opposite,since onlya fewprocessescan be
modeledwiththedegreeof precisionrequiredby thesemethods.The practi-
tionerof theblackartmustgetin thereand putthemodeltogether beforehe
can evenbeginto theorizeaboutit.
The firstgroundrulealso meansthata modelmustbe explicatedas a set
ofaxioms.Sucha setof axiomscanbe viewedas a setof partswhichtogether
definethe whole thing.These "parts" can be interpreted as behavioral
regularitieswith normativeimplicationssuch as PrincipleVI above. Or,
sometimesan axiom can be interpreted as a proceduralfeatureof a larger
process-a description or a model of a particular institutionalfeatureof a
process.Examplesofbothwillbe coveredbelow.
If processescan be decomposedintoaxiomaticparts,we can thencompare
themby comparing the parts.Similarities
amongprocessescan be identified
in termsof the partstheyshare,the axiomstheyhave in common.Such
comparisons mayhelpus pindownthe particular aspectsof processeswhich
cause us to acceptor rejectthem.One can evenputpartsofvariousprocesses
together and see if theycombineintosomething whichwe like.The second
groundruleconcerning theacceptabilityof processesis simplya requirement
556 CharlesR. Plott
thatthe axioms,once isolated,shoulddo a goodjob alongtheselines.The

normative aspectsof a processmustbe easyto see,evaluate,and compareto
thoseof otherprocesses.
In thesubsectionsbelowwe willfirstreviewtheaxiomaticdecomposition
of existingprocesses.The secondsectionreviewssomeaxiomswhichmodel
or proceduralaspectsof processes.The thirdsectionandthe
specificfeatures
finalsectionare evenmorespeculative thanthefirsttwo.The thirdsection
coverssometheoretical processes,and the finalsectioncontainsa discussion
of some new possibleapplications, of a revealedinstitution's
the possibility
approachto thestudyof processeswhichhavenot receivedattention in the
literature.
A certainamountof technicalmaterialis includedin thissection.It is
unavoidableif the readerwantsto probedeeplyat all. For thosewho want
only to get a flavorof the analysis,I have triedto describewhat the
mathematics say,so you can be reasonably safeby simplyglancingoverthe
keepingan eyeopenforinterpretive
technicalmaterial, statements.
of ExistingProcesses
AxiomaticCharacterization
In a simplebut highlyoriginalpaper,K. 0. May(1952) openedthedoor
for a new line of inquiry.He provideda set of necessaryand sufficient
conditions,a set of axioms,fora "ballottabulating procedure"to be simple
majorityrule. Simplemajorityrulesatisfiesthe axioms;furthermore, any
methodof transforming a patternofballotsintoa "winner"whichsatisfies the
axiomsis functionally equivalentto (can be accurately modeledas) simple
majority rule.
Sincethen,threeadditionalprocesseshavebeencompletely characterized
in the sense that a set of necessaryand sufficient conditionshave been
isolated.Sets of necessary conditions(but not sufficient) havebeen formed
formanyprocesses. We willexamine brieflyonly those on whichtheworkis
complete-simplemajorityrule, Borda count, representative systems,the
competitivepricing mechanism-and one which is incomplete-competitive
electionprocesses.Onlythe firsttwoof thesewillbe examinedin anydetail.
That is enoughto showhow themethodof decomposition and comparison
works.
Beforegoingfurther, two different levelsof processmodelingshouldbe
distinguished: ballotor vote tabulatingmodels, and socialchoiceprocessesin
general.Ballot tabulatingprocedures, as shown in Figure2, take ballots,
of
votes,sequences votes,etc., in some operationalway and transform them
intotheoption(s)designated as the This
winner(s). is a very mechanical
thing
Social Choice
Processes
VotingProcesses
ballots
voteswinrs
-- wne() 1~
"0expres-
sions"
ballot tabulation
preferences procedure
attitudes social
information choice
decisionrules
< 0 game structure
OtherSocial Choice
Processes
FIGURE 2
whichmayinvolveno behavior,no strategy, no attitudes,no people,etc. In

contrastto ballottabulating othersocialchoiceprocessesinvolve
procedures,
people(at leastone) as well as all thecomplicating factorsthatpeoplebring
to
along. It is important notice that social choice processesmay have no
votingprocedures at all or any otherform of "group expression"
asidefrom
theimplemented processoutcome.
As one mightsuspect,sinceballottabulating proceduresaresimplerelative
to social choice processes,muchmoreis knownabout them.It shouldbe
noted,however,that a ballot tabulatingprocedurecan be a social choice
processiftheballotsarerankings, sincerely
reported byindividuals,andifthe
winneroptionis accurately implemented.Suchassumptions aregenerallynot
558 CharlesR. Plott
so thesearchforan acceptableballottabulating
justifiled, proceduremustalso
proceedalongwitha theoryofindividual behaviorand decisionanda theory
of implementation and controlof publicpolicy.In thislight,theidea of a
ballottabulating procedurecan be seenas partofa muchbroadereffort than
one mightinitiallybe led to believefromwhatis reportedhere.Formaliza-
tion at the voter behaviorend and at the policy controlend typically
necessitatesa mathematical theoryof information flow,organization
behav-
ior in general,and bureaucratic Thesevariablesgener-
behaviorin particular.
ate theirown independentproblems,and even at the mostnarrowballot
tabulating
levelof examination moreproblems arecertainlynotneeded.Life
is complicatedenoughas it is.
Rule
SimpleMajority
Thisis themodelA of the fleet-thefirstproduced.In orderto see what
makes it tick,let us firstnotice that this ballot tabulatingprocedureis
definedonlyforcases in whichtherearebuttwooptions.Call themx andy.
Each individualcan cast a vote in favorof x, cast a votein favorof y, or
abstain.
Call thenumber1 a vote forx, 0 an abstention and -1 a votefory. If Bi
represents the ballot of individual
i, it can takevalues1, 0, or -1. A ballot
tabulating procedure F, whichtakeseachpatternofballots,(B1,
is a function
B2,... ,Bn), one ballot for each of the n individuals, and assignsto it a
"summaryvote" or a "groupvote." If F (B1, . . ., Bn) = 1, thatis, if the
summary voteis 1,we sayx is thewinner.If it is 0 or-1 we say,respectively,
thatthereis a tieor thaty is thewinner.
A ballottabulating procedureis definedto be simplemajorityruleifit is
functionally equivalentto simplemajorityrule. That is, F is the simple
majority ruleballoting procedureincase
ifB1+B2+ . . . +Bn>O thenF(B1, . . . Bn)= I

ifB1+B2+.. .+Bn = 0 thenF(B1, . . . Bn)=
if B1+B2+. . . +Bn<O then F(B1, . . . Bn) = -
If the sum of all ballotsis positive,zero, or negative,the summary ballot

takesthevalue1,0, or-1.
How did we get into all of thismathematics so quickly?The trickis to
noticethatthisballot tabulatingprocedurecan be accurately modeledas a
functionwhichtakeseach arrayof ballotsintoa summary ballot.Thenwe
implicitly assume(if the interpretation is alongsocial choiceprocesslines)

thatthe resultof thesummary ballotis translated intosometypeof social
action.For example,if F(B1, . . . Bn) = 1, thenx is thewinnerand follows
thepathof implementation as all winnersdo. Once we geta function which
accurately modelstheprocess(and accuracyis veryimportant here),thenany
function whichbehavesin exactlythesamewaycanbe treatedas theprocess
itself.In thismethodology themodelbecomestheprocess.The realphysical
thingis droppedfromconsideration and returns at thosetimeswhenchecks
on progressand assumptionsare needed. So now any functionwiththe
mathematical properties listedaboveis thesimplemajorityruleballottabu-
latingprocedure.
Considerthefollowing fouraxioms:
Decisiveness.For each possiblearrayof ballots,(Bl, .. ,Bn) the sum-
maryballotF(B1, . . . ,Bn) takesa valueof either1,0, or-1. In otherwords,
the processis able to deal withany patternof ballots;it alwaysdefinesa
winneror a tie.
Symmetry.If the array(B'1, ... ,B' ) is a permutation of the array
(B1, ... 1B3), then F(B'1, . .. IBWO)= F(B1, . .. .Bn). If the ballots are
switchedamongvoters,it doesnotchangetheoutcome.
Duality. If Bi = -B'i for all i = 1, 2,..., n, then F(B1, .Bn) =
-F(B'I, ... , B'n). Iftheballot of each is
individual reversed, then the winner
is also reversed.
StrongMonotonicity. (PrincipleIX above).If Bi > B'i forall i and Bj > B'j
forsomej and if F(B'1, . .. IBWn) is 0 or 1, thenF(B1, . .. .Bn) = 1. Suppose
at leastone personchangeshisballotin thedirection of 1, he goesfrom-1 to
either0 or 1, or from0 to 1, and no one changesin thedirectionof -1,
thenif thesummary votewas 0 it is now 1, andifitwas I it is still1. It only
takesa changebyone personto breaka tie.
Theorem.A ballottabulating procedure is thesimplemajorityruleprocedure

(or its functionalequivalent)if and only if it exhibitsthe properties
of
Decisiveness,
Symmetry, DualityandStrongMonotonicity.
The theoremtellsus thatthefouraxiomstogether arethepartsthatmake

up majorityrule.Withthisresult,the normative theorybegins.A commit-
mentto thesimplemajorityruleballottabulating a
procedureis necessarily
commitment to thesefouraxioms,and a commitment to thesefouraxiomsis
a commitment
necessarily to simplemajorityrule.
The decomposition shouldhelpus identifythe"parts"thatare,or arenot,
acceptable.If by studying
individually theseaxiomswe canmakea judgment
about theiracceptability, thenwe are alreadycaughtup in thespiritof the

analysis.Suppose,forexample,thatx involves mydying;y does not;I prefer
x to y (I wantto die); you prefery to x; and thatthesocialchoiceis x. Now
supposeyourrankingand myranking are switched(I now no longerprefer
theoptionwhereI die but you do) whileall otherpeople'srankings remain
unchanged.Accordingto the symmetry axiom, the social choice should
remainthesame.Optionx, whereI die,should,according to theaxiom,still
be chosen.Now I don'tknowaboutyou,but forme, whereoptionssuchas
theseareinvolved, thesymmetry axiomtypeofbehavioris simplyunaccepta-
ble, and it is exactlyforthisreasonthatI personallywouldrejectthegeneral
use of majorityrule.The analysishelpsme identify exactlywhich"part"of
majority ruleI findunacceptable.The advertisementsaiditwould.
TheBordaCount
The Bordacountwas used in theintroduction as an exampleof a process
whichis subjectto certainpeculiarities.Whatareitsproperties? The question
was addressedindependently by Smith(1973) and by Young (1974a). We
willfollowYoung'sdevelopment heresincehe providesa completeanswer.
As withthe development of majorityruleabove,we will constrainthe
interpretation to ballot tabulatingprocedures.We willalso keep thediscus-
sion at theabstractsetting.Real settings willhavemanymorefeatures, but
theywillat leasthavethese,andtheyareall we need.
Let 9Z be a set of conceivabledecisionmakers. Presumably thereis an
infinite numberof them.Naturally we willbe facedwithonlya finitesubset
of themat anyone time.The thingto noticeis thatno conceptlikethiswas
neededforthedevelopment ofmajority rule.Currently,in orderto character-
ize the Borda countwe will need to comparehow theprocessworkswith
variousdifferent societies.
Let v = a1, a2, ...,ams be the set of feasibleoptions.This set will
remainfixedthroughout theanalysis.
We are thus startingout with threeconcepts:The set of conceivable
decisionmakers 9T;theset (yet to be specified)ofactualdecisionmakers;and
thesetof feasibleoptionsv.
If N is theset of actualdecisionmakerswithn members, we let R(1,N) be
theranking of theoptionsof v forthe firstpersonin N, R(2,N) theranking
of the second person,R(3,N) the rankingof the optionsin v forthethird
personin N, etc. So we needto identify boththeindividual and thegrouphe
is in.
Now suppose we have designedsome (perhapsstrange)social choice
processandhavemodeleditsbehaviorperfectly withthechoicefunction C(v,

R(1, N), R(2,N), . .
,R(,,N)). That is, the choice functionindicateswith
perfectaccuracywhattheoutcome,including ties,willbe once thedecision-
makersandtheirrankings areidentified.
If the processwe designedis the Borda count,the winningoptionsare
chosenas if by the followingformula.If thereare m feasibleoptions,an
optionreceivesa scoreof m + 1 - j fromeach individualforwhomit is the
jth mostpreferred. So if youprefer a3 firstanda7 secondandifthereare 10
alternativesin all, a3 receives10 pointsfromyou, a7 receives9, etc. The
optionwiththegreatest totalpointscoreis designated as the"winner."
Of coursethereare manyschemeswhichcould use thenumbersassigned
by each individual. Pointscould be multiplied, forexample,and/orraisedto
somepower.Pointsassignedby certainpeopleor assignedto certainoptions
could be given"extraweight,"etc. The schemewe designedcould involve
someverycomplicated computational gimmicks. The followingresulttellsus
something abouttherelationship of suchschemesto theBordacount.
Theorem.The processmodeledby the choice functionC(v, R(1,N),.

equivalentto it) ifandonlyif
R(n,N)) is the Bordacount(or is behaviorally
itsatisfies four(foreboding
thefollowing axioms.
buteasyto interpret)
Faithful.Suppose that N containsonly one individualand his most

optionis ak, then
preferred
C(v, R(1,N)) = ak
N1 and
Supposewe havetwo disjointsetsof decisionmakers,
Consistent.
N2,withn1 andn2 numbersofmembers then
respectively,
C(v, R(1,N,), . 'R(n1, N1 )) nC(v, R(1 ,NN), - nR(n2 N2))=

eithertheemptyset or C(v, R(l, N1), ' . . R(n 1,N 1), . . ,R(1,N 2)' ,R(n 2,N 2
and that
of thelabelsofalternatives
Neutral.SupposeHIis a permutation
(R(1,N), ...*R(n ,N)) and (R'(1,N), ... ,R(n ,N)) are two differentarraysof
rankings(for a fixed set of decisionmakersN) which have the propertythat,
foreach individuali and each pair of options aj and ak,
ajR(i,N)ak if and only if an(j) R'(i,N)an(k) for all i,j,k, then

ajCC(v, R(1,N), . ..R(n,N) if and only if alH,) C C(v,R'(1,N), ,R'(n,N)).
Cancellation.
For anyN and array(R( 1,N), . R(fl,N))suchthatforevery
pair of options(aj,ak) the numberof decisionmakers
who preferaj to ak
equalsthenumberwhoprefer ak to ai, then
C(v, R(1,N), ... 5R(n,N)) = v.
The interpretations of these axioms are as follows.Faithfulbehavior

meansthatthe processchoosesexactlyas an individual wouldin theevent
thatshe (he) is theonlydecisionmaker. Consistent behaviormeansthatthe
winner(s)whentwo groupsarethrown togetheris (are) exactlythewinner(s)
they had in commonwhen choosingseparately-when such winners-in-
commonexist.The processbehavesNeutrallyif the namesof the options
make no difference. That is, all optionsare treatedthe same way. The
Cancellationproperty says thatif thereis enoughconflict, in termsof the
numberwho preferone optionoveranother-asopposedto whoprefers one
optionoveranother-then everything ties.
Now let us see howtheBordacountdiffers frommajority rule-orat least
how it differsfromone of the"natural"waysof applyingmajorityruleto
setswithmorethantwo elements. Supposeourprocedure is one whichpicks
a majoritywinnerif one exists,and it picksa "top" cycleafterthePareto-
dominatedelementshavebeeneliminated ifa majoritywinnerdoes notexist.
Thisprocesswhichwe haveinvented fortheoccasionis Faithful, Neutral,and
satisfiesthe Cancellationproperty. The way thewinneris computedis not
thesameway theBordacountwinneris computed,butis thebehaviorofthe
processthesameas thebehaviorof the Bordacount?The answeris no. Our
processdoes nothavetheConsistency property andthuscannot,byvirtueof
thetheorem, be behaviorallyequivalentto theBordacount.
The following examplemakesthecase. Considertwo groupsof decision-
makersN1 and N2, withrespectively threeand fourmembers, wherev =
w,x,y. The preference rankingsaregivenas:
ofN1
Rankings ofN2
Rankings
R(i ,NI) R(2,NI) R(3,Nl) R(1,N2) R(2,N2) R(3,N2) R(4,N2)

w x y w w x x
x y w x x y w
y w x y y w y
Now,by definition C({w,x,y},R(1,Nl) ,R(2,N) , R(3,N1)) = {w,x,y}, where

C(., ) is the choice functionwhichmodelsour process,since thereis a
alternatives.
majorityrulecycleand thereare no Pareto-dominated And,by
definition
C( W,X,Y I R(1 ,N,s) XR(2,N 2 ,R( 3,N2 ), R(4,N2)) = jW,X
sincetheyaretiedbymajority ruleand bothbeaty.

Whatshouldbe theoutcomewhenbothgroupsarecombined?Ifyoufeel
thatoutcomesshouldbe consistent in thesenseof theconsistencyproperty
above, then you feel that w,x should be tied in the new situationof
combinedgroups.The pair w,x is what they chose in commonwhen
separate,so accordingto thisprinciplethatpair shouldbe tied whenthe
groupsare combined.But, in the processwe have definedthatdoes not
happen.For ourprocesswe get
C(Qwx,y , R(1,N ) ,R(2,N1) ,R(3,N1) ,R(1,N2),R(2,N2),R(3,N2) ,R(4,N2 W,

sincein thecombinedsocietyit is themajorityrulewinner.
In orderto see thatthe Borda countdoes satisfythe principlein this
particularsituation,notice the following.The BordawinnerswhenN1 is
tie.The Borda
aloneare all three w,x,y-thereis a three-way
choosing
winnerwhenN2 is choosingalone is x. The Bordawinnerwhenthegroups
are combinedis x, whichis also in accordwiththeconsistency property,the
onlythingtheychoosein common.
What is the point of all this? The processwe createdis difficult to
evaluate,but thismethodology at leastgivesus a place to start.We can see
that a choice betweenthe processwe createdand the Borda countturns
explicitly(whenthesearetheonlytwocontenders) uponhowone intuitively
regardsthe Consistency propertyIf it is accepted,thentheBordacountis
acceptedand the otherprocessis rejected.If the Consistency propertyis
rejected,then the processwe createdis acceptedand the Borda countis
rejected.The pointis thatthenatureof thisverycomplicated choicehasbeen
refinedto a choiceabouta well-defined behavioralproperty. The methodwill
not tell us whichprocessis, in some sense,better,but it does providea
cleardelineation
relatively aboutthenatureof thechoice.
Systems
Representative
Perhapsour discussionof axiomaticrepresentativesystemsshouldreally
be postponedto the sectionwherewe discusstheoretical and/orimaginary
processes,as opposed to thissectionwherewe are discussing
existingproc-
564 CharlesR. Plott
esses. However,the idea, firstintroducedby Murakami(1968), does catch

part of the spiritof real representative systems, especiallyif we keep the
discussionwithinthesimplecontextof a ballottabulating procedure. In hope
thatyou will be happyto see some progresson the simplestructure of a
representative system, we willdiscussithere.
As in thecaseof simplemajority ruleabove,letus restrict theanalysisto a
situationwherethereare onlytwooptions,x andy,andstudyexactlyhowa
representative structure would functionhere.Since thereare onlytwo op-
tions,ballot Bi of individuali takesvalues 1, 0, or -1, in case it is forx,
abstention, or fory,respectively.
Ratherthanattempta formaldefinition of a representative system,letus
taketwoexamples.
Example1. Thereare fiveindividuals 1,2,3,4,5. Decisionsare madeby
themajorityof a councilon whichMr. 1 has a vote,Mr.5 has a vote,and a
representative who alwaysvotesalongwiththemajorityof thosehe repre-
sents(Messrs.2, 3, and4) hasa vote.
Example2. Thereare fiveindividuals;1,2,3,4,5. Decisionsare madeby
themajorityof a councilon whichMr.1 hasa vote,a representative of Mr.1
and Mr.2 has a vote,and a representative of Messrs.3, 4, and 5 has a vote.
Representatives always vote in accord with the majorityof those they
represent.
Now we can model the behaviorof thesetwo systemsas followxs. Let
((Bl, ... ,Bk) ) represent a function whichtakesvalues1, 0, -1 as I Bi is
positive,zero,or negative.Noticethat( ( . . . ) ) is a ballotjustlikethe IB's,
it
takesthesamevalue,so it canfitintoa function justliketheB's. Example1
becomes( (Bl, ( (B2,B3,B4)), B5) ) and example2 becomes( ( ( (B13,B2)),
( (B3 ,B4,B5) ), B, ) ). The different representative systemsclearlybehave
differently. If (Bl ,B2,B3,B41,B5) = (1 ,1,-1,-1,-1) thenexample1 takesthe
value 1, while example2 takes the value 1. Notice that example2 is
equivalentto a collegium(discussedabove) wherebyindividual1 has the
powerto blockanything. He can cancelanyvotesof theothers.But,in order
for him to institutean option he must be accompaniedby one of the
coalitions in the family[ 2 I 3,4, I3,5 , 4,5 ] or the union of two or
moreof thesesets.That is, if 1 and 2 votetogether, thesummary votegoes
alongwiththem.If 1,3, and 4 get together, theydetermine the summary
vote,etc.
Example1 is interesting becauseit demonstrates thatrepresentative
sys-
temsand weighted votingsystems can be entirely
different evenat the
things,
verybasic ballot tabulationlevel of analysis.It is easy to show thatthis
exampleis not behaviorally equivalentto any weightedvotingsystem,even
thoughit is tempting to assertthatthevoteof,say,Mr. 1 is worthtwo or

threetimesthatof,say,Mr.2.
Now the questionthatMurakami askedwas: Whattypeof properties do
theserepresentative systems have?Whenis a ballottabulating systembehav-
iorallyequivalentto a representative system?He supplieda partialanswer,
but it remainedfor Fishburn(1973a), and subsequentlyFine and Fine
(1974), and Bartoszynski (1972) to supplycompleteanswers.As it turnsout,
it takesa lot of notationalmachinery to pose theanswerin a mannerwhich
beginsto makesense,and forthisthereadershouldgo to thesesources.
Whatis the difference betweensimplemajorityruleand representative
systems? A representative systemis behaviorallyequivalentto simplemajor-
ity rule if (in additionto satisfying of representative
all otherproperties
systems)it has the Symmetry property listedaboveas one ofthe"parts"of
simplemajority rule.Whataboutthedifference betweentheBordacountand
representative systems?If a representative systemsatisfiestheCancellation
property, listedas a partof the Bordacount,it is behaviorally equivalentto
the Borda count.22 Other comparisonscan be made, but for these the
originalsourcesneedto be consulted.
TheCompetitive Mechanism
Pricing
The finaltwo processesrequiremuchmoretechnicaldevelopment than

can appropriately be introducedhere.Bothshouldbe mentioned, however,
sincebothrepresent important stepsforward in our abilityto axiomatically
characterizecomplete social choice processes.In addition,each bringsa
different
slightly perspectiveto the whole enterprise.
The competitive pricing systemwasrecently axiomatizedby Sonnenschein
In
(1974). doing this,he had to introduce some conceptsexplicitly whichhad
previously beenhiddenin theabstractions of theanalysis.The majoridea is
thatof a privaterepresentation of a choicefunction. Thisis an abstractway
of describing (a) a structure of information and rights,(b) a conceptof
equilibrium, and (c) an individual'sbehaviorin responseto his rightsand
information,givenhispreferences. The threetogether definea processmodel,
an imaginarysystem,such that if the definitionsare correct,then the
equilibriaof the resulting imaginary systemare alwaysthechoicemadeby
thechoicefunction.
22 In this case it is also behaviorallyequivalent to simplymajorityrule,whichis the
Borda count when thereare only two alternatives.

A littleobscure?Let me tryagain.Supposeyouaregivena lot of dataon

systembehavior.In particular, you knowhow thesystemwillbehavein the
face of any (or almostany) arrayof individualrankings and feasiblesets.
Could you inventinstitutional structures of the type(a) above,and behav-
ioral theoriesof the types(b) and (c) above,so thattheprocessoutcomes
alwayscorrespondto those of your inventedsystem?That is, could you
inventa processwhichis behaviorally equivalentto theone forwhichyou
have data? While Sonnenschein was interestedin only special cases, the
answerin generalis "yes."
Now hereis the keyobservation. The competitive pricingmechanism is a
very special processin termsof (a), (b), and (c) above. In particular,
Sonnenschein showsthatthestructures (a), (b), and (c), whichconstitutethe
competitive pricingsystem,satisfya certainset of axioms.Furthermore, if
theprivaterepresentation of any choicefunction theseaxioms,then
satisfies
thereis a competitive pricingsystemwhichis behaviorally equivalentto the
processwhichgeneratedthe choice functionmodel.Thus,thestructure of
privaterepresentation of processesand theSonnenschein axiomsmayserveas
a framework withinwhichcomparisons can be madeat thetheoretical level,
betweenthe competitive priceprocessand otherprocesses.Clearlythisis an
important step forward, sincefreemarketprocessesare oftenconsideredto
be thesinequa nonofacceptableprocesses.
ElectionProcess
Competitive
Kramer(1975) has providedus withthesecondaxiomatictreatment ofa
completeprocess.The processis the competitive, majorityrule election
processin a spatialsetting. Theaxiomaticformulation is notcomplete, butit
is closeenoughto warrant citationin a reviewsuchas this.
He demonstrates that if candidatesin a two-party electionattemptto
maximizevotes,thentheywillconverge to themin-max set.A pointis in the
min-max set if themaximum-sized coalitionwhichwouldbenefitbya move
fromit is no largerthanthemaximum-sized coalitionwhichwouldbenefitby
a move fromany otherpoint.Unlikesome otherproposedsolutions,the
min-max set alwaysexists.The axiomaticstructure of a processwhichpicks
themin-max set is not known,so we cannotyetfullyevaluateand compare
the normative structure of thisprocessto thecompetitive pricingsystemor
to the otherprocesses.Fromour pointof view,theKramerresultdemon-
stratesthatcertainslightchangesin institutional structure result,theoreti-
cally,in a social choice functionwhichhas analyzable(in thesensewe are
usingtheterm)properties.
Procedural
Featuresof Processes
It is possibleto modelballottabulating procedures,and it is possibleto
modelsocial choiceprocesses.Is it also possibleto modelotherprocedural
featuresof processes?Can we identifythe contribution of individualor
isolatedprocedures to overallprocessbehavior?The answerto bothquestions
is in principleyes,althoughtheamountof workwhichhasbeendoneis not
evena dentin theamountof workthatneedsto be donebeforeanygeneral
successcanbe claimedforthemethods.
The analysiscan go two ways.We can identify a procedureand askabout
itsindependent Thesecretballotprocedure
contribution. is takenbelowas an
exampleof thisdirectionof analysis.We can also go theoppositedirection.
We can postulatesome type of modification of processbehaviorthatwe
want,and ask if thereis a procedurewhichwill inducethe behaviorwe
specify. Wereviewin thisconnection, theidea of pathindependent choice.
SecretBallot
Thereare manydifferent versionsof the secretballotprocedure, so we
mustbe a littlecarefulin identifyingwhatwe aretalking about.Wewillcalla
proceduresecretballotin case: (i) no one knowstheballotofanyparticipant
otherthanhimself,and everyoneknowsthatthepatternof information is
limitedin thatmanner;and (ii) once a ballot is cast,it does not bearthe
identityof the personwho cast it,nordoes it carryany identifying feature
suchas thetimeand placewhereitwas cast,whichwoulddistinguish it from
theballotcastbyanyotherperson.
We can now advancethe followinghypothesisabout the behaviorof
processeswhichhave thisfeature:If people's preferences are interchanged
whilethe feasibleset remainsunchanged, thenthesocialchoicewillremain
unchanged.The hypothesis can be put formallyas an axiomwhichwe have
seenbeforein a differentform.
Anonymity. If we haveany two arraysof rankings,
(R1,... 5Rn)and (R'1,
... .R'0), suchthateach is a permutationof theother(rankings
havebeen
switchedamong the individuals),then C(v,R1, . . .RO) = C(v,R'1, . . . 5R'n)
The symmetry axiom,whichwas an aspectof simplemajorityrule,is

impliedby thisaxiom.23
23 am indebted to JeffreyRichelsonforcallingto my attentiona mistakeabout the

of thisaxiom containedin a draftof thispaper.
interpretations
568 CharlesR. Plott
As a hypothesis about thebehaviorof any processwhichuses thesecret

ballotprocedure, thismodelis,as anygoodmodel,clearlyrejectable.In order
to rejectthe hypothesis one onlyneeds to finda processwhichdoes not
behaveas thismodel predicts.Take a hypothetical case of majorityrulein
whichthereare threepeople, 1,2, and 3. Mr. 1 has sometypeof personal
influence overMr.3, so thelatteralwaysvotesalongwithMr. 1, evenifhis
own preferences dictateotherwise.24 Now if Mr. 1 and Mr.2 switchprefer-
ences and Mr. 3's rankingremainsthe same, the axiom claimsthat the
outcomewill remainthesame.However,becauseof the personalinifluence,
Mr.3 changeshisvote to go alongwithMr.1. The outcomewillthuschange,
and theprocessbehaviorviolatestheaxiom.As a modeltheaxiomwouldbe
rejected.So, themodelis in principlerejectable.Ofcoursebeforethatreally
can happen,we need to supplya lot ofoperational and finda real
definitions
example.
The pointwe wishto establishis not thattheanonymity axiomis a well
workedout, accuratepropertyof processeswhichhave the secretballot
feature.It maybe changeda greatdeal,especiallyif phenomena suchas our
hypothetical exampleexist,beforewe have it exactlyright.The pointis
modestbut extremelyimportant-the potentialexistsforusingaxiomsas
behavioralrepresentations of processes.The dooris at
of proceduralfeatures
least open for a new way of studyingprocessesand variousaspectsof
procedure.
Let's takea disputeresolution processexampleto seehow themachinery
we have generatedso farcan be used. Suppose,basedon someconceptof
fairness,we acceptall of thefollowing:
1. Our social conflictsare to be resolvedby a committee whichis to
takeas datathewishesof theindividuals involvedin thedispute.
2. Committeemembersshouldnot be able to identify or distinguish
betweentheindividuals whosewishestheyareconsidering.
3. The committee shouldmakea judgmenton thebasisof thepattern
of conflictalone,and not thenatureof theoptions.Eachoptionis
givena label,and the keyis unknownto thecommittee members.
(So in makingtheirdecisionthecommittee knowsonlythepattern
of conflictoverthelabels.Thus,theprocedures prevent themfrom
consultingtheirownfeelings abouttheoptions).
24 I do not want to alienate preferencetheoristshere by taking a stand on what
constitutesa "preference"and whetheror not people's actions must be taken as an

operationaldefinitionof preferencein a "revealed" sense.
behaviorshouldsatisfythe
thatwe feelthecommittee's
Suppose further
standard:
4. Theconsistency of theBordacount.
property
Now we have a conclusion.The committeecan be replacedby a social

choicescoringfunction. Thereare manydifferent typesof scoringfunctions.
Basically,these are processeswherebyindividuals assignpointsto options
whichare thensummed.The winneris the optionwiththe highestpoint
total. As examples,take the following:everyoneassignsone pointto his
first-placeoptionandzeroto otheroptions;or,everyone assignstenpointsto
his firstoption,threeto hissecond,andzeroto all others;theBordacountis
an example;etc. Thereis not enoughinformation givenin theexampleto
determine whichsocial choice scoringfunction is theappropriate one. But,
we knowthatthereis an appropriate one and thatit is a memberof thisclass.
How can we supporta conclusionlike that?The naturalmodelsforthe
proceduralrestrictionstwo and threeare anonymity and neutralityrespec-
of
tively.The requirement consistency was imposed as a normative con-
straint.All we must do now is appeal to a theorem by Young (1974a, b), a
similartheoremby Smith (1973), and we concludethat the processis
necessarily equivalentto somesocialchoicescoringfunction.
behaviorally If
for
we noticed, example, thatthe cancellation was
property alwayssatisfied,
thentheprocesswouldbe behaviorally equivalentto theBordacount.So, if
the committeeprocessis expensiveto maintainor is not "error"free,then
replaceit withthebehaviorally (and normatively?) equivalent votingprocess.
PathIndependence
We haveseenthatan axiomcan be interpreted as a modelofa procedure

and as a normative Given
constraint. a procedure,e.g.,secretballot,we can
searchforan appropriatemodel(axiom).Likewise,givena normative demand
(axiom) about process performance, we can tryto findan institutional
procedurewhichinducestheappropriate systembehavior.In thissectionwe
willoutlinethe beginning(the end is stillunknown)of an attemptto apply
thelattermethodology.
The originof the problemis, likemanyof theproblemsexaminedin the
field,the cyclicalbehaviorof majorityrule.The claimwas made thatthe
outcome,whena cycleexists,is determined by thesequenceofvoting, andat
thetimewe discussedit,at least,thatdidnotseemto be verygoodthingto
a
have. In additiona theoremin Smith(1973) establishes thatthesequential
applicationof any social choice scoringfunctioncan producewhat some

would call bizarreresults.This naturallyraises the generalquestionof
whetheror notthereareanyprocessesat all whichhavetheproperty thatthe
outcomeis independent of the votingsequence.In thejargonto be intro-
duced below the questionis "Are thereany procedureswhichare path
independent?"
Are thereprocesseswhichgiveoutcomeswhichare independent of the
sequenceof voting?"Sure,"one mightanswer."Anyprocessthatdoes not
involvea sequence (e.g., the Borda count) mustby definitionhave that
property.""Don't be funny,dummy,"you may be temptedto remark.
"Find a nontrivial example."If you ventured theremark, hisreplymightbe
poignant,sinceyou did not tellhimexactlyhow you woulddecidewhatis
andwhatis not,andit is up to you to do so.
trivial
Whilereflecting on thatchallenge, let's explorewhyit mightbe necessary
to considersequencesof votesin thefirstplace. The requirement seemsto
stemfromthreedifferent sources.The firstis a simple normative require-
ment."I thinkthereshouldbe somerelationship betweentheoutcomeswhen
thereis a big feasible(contender)set and a smallfeasible(contender)set."
Well,thatneedsto be spelledout. Iftherequirement can be shownto be the
weak axiom of revealedpreference (WARP)discussedabove,thenwe know
immediately thatthereis a problem.That axiom takesus directlyback to
socialpreference definitions,and theassociatedimpossibility results.
The secondreasonforsequencesof votesseemsto be tiedto an informa-
tional processingproblem.Some of us seem to have limitations on the
amountof factualmaterialthatwe can handleand discuss."Giveus just a
fewoptionsat a time.We'll rejecta few and thenadd on a fewmorefor
consideration. Sooner or later we will have workedour way throughthe
wholefeasibleset by considering one set of contendersat a time."The
consequencesof thistypeof motivation havenot been carefully examined.
Theymayor maynotbe coveredbywhatwe discussbelow.
The final reason for decisionsto proceed in a type of "divideand
conquer" fashionseems to be due to a purelytechnicallimitationon
processesthemselves. Majorityrule,for example,is not even definedfor
anything other than two optionsat a time.It is by definition "binary,"and
can only be used to choose over large sets by resortingto some typeof
repeatedor sequential application.
If thereare only threefeasibleoptions,thereare stilla greatnumberof
possiblesequences.Some of thesearelistedon Figure3.25 Thereyou can see
25 See Farquharson(1969) fora game theoreticanalysisof some of these.

Example option
Question in the "blank" Tree diagram
Do we want or x x
do we not? y FYI
Do we want ? x x
Can we eliminate__? x ' LSIm
rv E
- Ofthetwo,which shall x
we elirninate?
Ex,y
LU
> t
E-li
El
prefer? 3xy 1__

x zD 3
FIGURE 3
SomePossibleWaysofApplying a
BinaryProcessto a Three-Element
ChoiceSituation
thatanswering thequestionon theleftis in effecta choicebetweenwhichof

the two sets at theendsof thebranchesis to receivefurther consideration.
Each of thesediagrams different
represents "paths"of choice.Ifwe arestuck
witha binaryrule,thenwe are forcedto applyit sequentially alongsome
"path"suchas one of these.
Onlythelastreasonhas receivedmuchattention, and it servesas motiva-
tionforan axiomaticdevelopment of theconceptof pathindependence. In
orderto developthe axioms,we need someidea about whatpeople,at any
stage,arebeingaskedto voteupon.Wepose thequestionas "Whichoption(s)
do we wantto eliminate?"as opposedto "Whichdo weprefer?"Let e(v;) be
the set whichis eliminatedby votefromthecontender set vi. We can then
definethe choice over the contenderset to be the set of thingswhich
survived.It is the set of optionswhichwerenot eliminated.In notational
form,whenthe"slash"means"setdifference" we have
C( Vi ) = vi\e( v; ) .
The notationsimplysaysthatthe"choice"fromthecontendersetis equal to

setwhichwerenoteliminated.
thesetof optionsin thecontender
is
PathIndependence.Supposev is any feasibleset.Wesaya choicefunction
in case foranytwosubsetsvi andvj ofv,suchthatv; Uvi=
pathindependent
v we have
C( v )=C C( vi )u Vji)
Thereis a lot of notationherebut theidea is, at base,rathersimple.We

have a machinewhichoperatesto eliminatesomeoptionsfromany set we
feedit, e.g., majorityruleoperatesto eliminatean option,unlessthereis a
tie,fromanypairof optionswithwhichit is presented. Whenall optionshave
been fedthrough themachinein one formor another, we implement one of
the optionsthatremains.If the machinehas theproperty of Path Indepen-
dence,thenfromanyfeasiblesetthesameoptionswillserveas candidatesfor
implementation, of how or in whatcombinations
regardless theoptionsare
fed to the machine.You can dividethe feasibleset into two contender
subsetsand "choose" overone subset,put thesurvivors and theothersubset
together,and choose over this new set. The ultimatesurvivors will be
independent of thewaytheproblemwasdividedintoinitialsubsets.
Of course,majority ruledoesnothavetheproperty ofPathIndependence.
To see thiswe simplytake a standardcyclicalcase withthreepeopleand
threealternatives. Suppose the preferences of Messrs.1, 2, and 3 are:

xP1yP1z; yP2zP2x; and zP3xP3y. Now if e(v) is the "machine"which
eliminates optionsfromthesetv,we havee((x,y}) = y,e( x,z}) = s, e((y,z )
= z, sincea majoritywouldvote to eliminatex, z, and y, respectively,
from
thesets,{x,O, {x,z}, (y,z). If C() is thesetofsurvivors,v\e(v), wherethe
slashis setdifference,
we have
C($C({x,y)),z ) = C({x,z4) = z
C(QC((y,z} ),x ) = C(y,z}) = x

C(QC({x,z}),y I) = C({z,yO) = y.
The survivor set is clearlynot independent

of thepathofchoice.A different
survivorset resultsforeachof thesedifferentwaysof dividing up thefeasible
setforpurposesofvoting.
So now we knowtwo things.Wehavesomeidea aboutwhyone mightbe
interestedin sequencesof votes,and we have at least one clearpictureof
whatit meansfortheoutcometo be independent of thesequenceof votes,
There are other ideas along this latterline (Plott, 1974; Parks, 1973;
Schwartz,1975b,c; Ferejohnand Grether, 1974). Rightnow,however,this
one seemsto be centralto theoveralltheme.
Now thatwe havesomeideas,whatcan we do withthem?Two linesof
researchemergefromhere.The firstis motivatedby some mistakenargu-
mentswhichoccurredearlyin the developmentof social choice theory.
Arrowclaimedthatthe transitivity typeaxiomsusedin thesocialpreference
definitionwerenecessaryto assurepathindependent choicebehavior.While
thisclaimis false,it is truethatpathindependent choicebehavioris "close"
to preference based behavior.26Manyof the resultsin theliteraturemake
26 Here is a simple example which exhibitspath independentbehaviorand is com-
pletely divorced from the "rational choice" structure.C(a) = a In a e [(x,y ,Ix,z},

{y,z}J and C({x,y,z}) = {x,y}. A proof that this functionis path independentcan be
foundin Plott (1973). An example or two would help you see why it mightbe true.
Q QC x,y }),xE ) =C x,y,z} )-x,y}
Now to see that choice functionis inconsistentwithpreferencetheory,simplynote that

the two-elementchoices indicate total indifference,but the choice over the three-
elementset is inconsistentwiththat.
theseconnectionsclearby showingwhatadditionalaxiomsmakethepath
independencerelatedaxioms equivalentto the weak axiom of revealed
preference (WARP)discussedabove.
The secondlineof investigation is notcompletely removedfromthefirst.
Whattypesof processesare pathindependent? The methodswe have for
findingthemare ratherbackward.We startwiththe basic processaxiom,
PrincipleVI, the Independenceof InfeasibleAlternatives. We thenadd an
axiom whichdemandsthatthe processyieldresultsforall (finite)feasible
sets and all patternsof rankings of optionsacrossthemembers of society.
Fromthispointaxiomssuchas Paretooptimality, strongmonotonicity, the
absenceof oligarchy, resultis obtained.
etc.,are added untilan impossibility
At the verypointwhenan impossibility resultis detected,theaxiomsare
tellingus thatwe havejust requiredtoo much.Usingthispointas a starting
point,we can thencontinueto exploresets of axiomswhichrequire"too
much,"and therebybeginto identify whatprocessesare not pathindepen-
dent,whichwe can thenuse to identify the classof processesthatare path
independent.
Beforewe get to some results,we need a slightmodification of our
Independence Alternative
of Infeasible axioms.Supposeindividual behavioris
"shortsighted" in thatwhenexpressing opinionsabout the feasibleoptions
undercurrentconsideration, the contenderset, no influenceis exertedby
preferences forfeasiblealternativeswhichare not in thecontender set. This
means,in part,that the individualdoes not engagein strategicbehavior.
Voting does not proceed in full anticipationof the outcomeof future
contests.27
In orderto modelthistypeof behavior,we willsimplychangethename
of PrincipleVI to Independenceof Noncontender
and the interpretation
Options.
of Noncontender
Axiom.Independence set.
Options.Letv be anycontender
If (R1, . . . Rn) and (R'1, . . . R'n) are two arraysof rankingssuch that for
27 Take the majorityrule process,with a cycle, outlined above in thissection.If the
sequence is x vs. y first,with the winnergoing against z, the ultimatewinneris z.

However, in anticipationof this problem Mr. 1 mightvote for y, his second choice,
ratherthan x, his firstchoice. If otherscontinuedto vote nonstrategically,his strategy
would change the resultfromz to y, which is, of course, an improvementforhim.The
axiom will be an attempt to rule out such behavior.Actually,individualsuninformed
about likely futurevotes do tend to vote this way. This is the basic reason why the
agenda exertssuch a powerfulinfluenceovercommitteechoices.
all i and forall contenderoptions,x, y in v, we have xRiy if and onlyif

xR'iy,then
C(v,Ri,... Rn) = C(v,R', * Rn)-
Literally,the axiom says thatif individuals'preferences fornoncontender

optionschangebut theirpreferences amongall contender optionsremainthe
same,thentheoutcome,thechoicefromthisset ofcontenders, remainsthe
same. In addition,if the feasibleset,of whichthecontender set is a proper
subset,expandsor shrinks, or if thesequenceof contestschanges,thechoice
at thisstageremainsthesameas longas thecontender setandattitudes about
theoptionsin thissetremainthesame.
Now hereis an interpretation of a resultdue to Blair,Bordes,Kellyand
Suzumura(1976).
Theorem.If the processsatisfies

PathIndependence,ParetoOptimality, and
Independenceof Noncontender Options,thenthereexistsan oligarchy.
Does thismeanthatour searchforpathindependent processesis a waste

of time?Not really,but it does mean thatthe particularnotionof path
independence we are usingnow is prettystrongand thatwe need to pursue
someof theweakernotions.In addition,we can see thatwe maybe wiseto
startthinkingaboutwhysomepathsmaybe moreacceptablethanothers.28
Againthe storystartsbecomingmoretechnicalfromhereon. The results
are of primary interestto thosewho wishto use the methods.Interested
readersshouldconsultA. K. Sen's (1976) reviewarticleforcomments and
references.
What have we learned fromthe exercise?It is possibleto use these
methodsin a searchforprocessesand procedures. The factthatwe did not
comeup witha moreappealingclassof processesdoes not reflecta faultof
the method;it reflectsa problemwith the demand that was made of
processes.The methodswereable to showus thatthedemandwas perhaps
moreunreasonable thanwe mighthaveimagined.
28 ProfessorMichael E. Levinehas arguedin conversationthatmanyof thediscussions

in legal circlescan be interpretedas attemptsto determinecriteriafordecidingwhichof
the many possible paths should be taken, because of an implicitrecognitionthat the
fairnessof the path is an importantparameterin determiningacceptabilityof the final
outcome.
Processes
andTheoretical
Social Indicators
The remaining literatureseemsto blend,to varying intotwovery
degrees,
broadand seemingly separateareas of inquiry.One branchmergesintothe
literatureon socialindicators The otherexpandsintothe
and socialstatistics.
generalliteratureon gametheory,strategic behavior,and themanysubstan-
tiveareaswheretheseabstract notionsapply.
Clearly,any attemptto limitthescope willbe arbitrary at best,but the
two guidelines withwhichwe begando a prettygoodjob: thedevelopment
mustbe axiomatic,and themajorimplications mustbe normative.To these
criteriaI have added threemore.First,theanalysismustinvolveindividual
attitudesdirectly.This eliminatesa greatnumberof studieson income
distribution,social position,etc., whichuse marketstatistics,wages,rents,
wealth,etc.,eventhoughsuchstatistics maybe used as proxiesforattitudes
or well-being(Atkinson,1975; Sen, 1973). Secondly,thematerial shouldnot
involvea largenumberof complicatedconceptswhichwe havenot already
introduced.The reviewis alreadyverylong.Some of thematerialhas been
coveredin otherreviews.Hurwicz(1973) has an excellentsurveyof theories
of decentralized processes,and Marschakand Radner(1972) haverecently
discussedthetheoryof teams.Unfortunately, someof thetheoreticalvoting
processessuchas generalscoringfunctions (Young,1975a),theKemenyrule
(Kemeny,1959; Levenglick, 1975; Merchant and Rao, 1975; Adelsmanand
Whinston,1975), and the processesbased on patternrecognition models
(Bowmanand Colontoni,1973, 1974; Blin and Whinston,1975; Young,
1975b) will not be covered.29Finally,the resultsshouldbe in somesense
"central"to the fieldof axiomaticsocial choice theory.Manystimulating
studiesexiston conceptsof liberalism.Perhapsifwe coversomecentralideas,
thereaderwillbe stimulated to pursuethesereferences.
Indicators
Nash-Hildreth-Harsanyi
"Indicator"modelsand thosewho deal withthemmustfacedirectly the
problem of interpersonal comparisons.
utility Such models usuallyhave two
parts:how numbersare to be combinedto forman index,andwhatnumbers
are to be combined.The functional formsusedto combinethebasicnumbers
are addition and multiplication.Since these differonly by a logarithmic
transformation, we can see thatany differences cannotreallybe
in indicators
29 I am indebtedto ProfessorAndrewWhinstonforcallingthesepapersto myattention.

in theformula forcombining thebasicnumbers. Instead,thedifferences must

be in the meaningof the basic numbersthemselves. If, forexample,the
numbersare "utilities,"thereis not likelyto be muchcontroversy about
whatto do withthem,but thereis likelyto be a lot of controversy about
whatthe"utilitynumbers"shouldbe.
The modelwe will outlinecannotbe foundexplicitlyin the literature.
Rather,I havetakenthelibertyof takingbitsand piecesfromthreeclassical
works-Nash(1950), Hildreth(1953), and Harsanyi(1955)-and putting
themtogether in a way consistent
withthespiritof moremoderntheoretical
developments (Sen, 1976). The functionalformof the indicatorwill be
multiplicative.The utilitynumberswillbe measuredin termsofprobabilities
as opposed to, say, just-noticeabledifferences (Goodmanand Markowitz,
1952) or someotherstandard.
Suppose the set of options v = {ao ,a,. . . ,am} is fixed and individuali
ranksao thelowest.Starting fromtherewe can calculate"utilitynumbers,"
ui(ak), whichare to be used as indicatorsof his "well-being." We say that
ui(ao) = 0 and thenfindanotheroption,saya1, whichhe ranksaboveao, and
giveit theutilitynumberui(a1) = 1. Now chooseany otheroption,ak,and
findthe probability P(k) such thatindividuali is just indifferent between
lottery1, whichgivesak withprobability P(k) and ao withprobability 1-
P(k), and lottery2, whichgivesa 50:50 chanceof getting eithera, or ao. A
utilitynumberforak is thencomputedto be ui(ak) = 1 -Whenthisis
2 P(k)
repeated for all k = 2,3, .. ,n we have a "utility function"foreach individ-
ual.
The bestoptionis theoptionin v forwhichtheproduct
w(ak) = ul(ak) U2u(ak) . ..- un(ak)
is thehighest.
Beforediscussingwhysomeonemightwantto do something likethis,let's
take an example. Let v = jao ,aI ,a2,a3,a4 and suppose thereare two
individuals)1,2t. Furthermore,
let's assumewe haveobtainedthedatagiven
inTable3.
Accordingto this method,option a3 is the "best," since it givesthe
highestvalueof "socialwelfare."
For the totallyuninitiatedthis must look ludicrous.Whereare the
valuejudgments? Whyshouldwe accepta3 as "best"?Whyshouldwe accept
thatmeasureof "social welfare"?Havingintroduced themeasurein sucha
naked form,let's go back and see whathas to be done in orderforthis
formula to makesense.
578 CharlesR. Plott
TABLE 3
LotteriesRankedEquivalentto
the Lottery[P(ao) = .5, P(a1 ) = .51
Mr.] Mr. 2
[P(a2) = 1/3 P(aO) = 2/31 [P(a4 ) = 4/5 P(aO) = 1/51

[P(a3) = 1/6 P(ao) = 3/41 [P(a2) = 1/5 P(ao) = 4/51
[P(a4 ) = 1/10 P(ao) = 9/10] [P(a3) = 1/20 P(aO) = 19/201
Derived UtilityFunctions Social Welfare
ai u1 (ai) u2(ai) ul (ai) -u2 (ai) = w(ai)

ao 0 0 0 *0 0
a, 1 1 1 1=1
a2 3/2 5/2 3/2 * 5/2 = 15/4
a3 3 10 3 10=30-max.
a4 5 5/8 5 5/8=25/8
1. Assumethatwe are willingto say thatall individuals are "equally

welloff"whenao prevails and thattheyare "equallywelloff"when
a, prevails.This is and mustbe a valuejudgment, withnothingto
supportit but us. If utilitynumbersare to be used, and if the
numbersare to reflecthow "welloff"in ouropiniontheindividual
is, thenu'(ao) = u2(ao) andu'(a1 ) = u2(a1) foranysetofnumbers
we chooseto assign.Thisequalityofnumbers followsfromourbasic
valuejudgments andnotviceversa.
In one documentedattemptto apply thisprocedure, theproblem
was to choose the flightplan of a spacecraft. Teams of research
each one of whichhad a different
scientists, experiment aboardthe
craft,were to choose the flightplan. Since different flightplans
would yield differential experimental benefits,the teamsall had
differentopinionsabout the alternative plans. In our jargon,each
flightplanis an option,and theset of all possibleflightplansis the
feasibleset. One distinguishedplan called "the AtlanticOcean Spe-
cial" (the mostdirectroutefromthelaunchpad intothe Atlantic
Ocean) was preferred theleastbyall teams,and so it playedtherole
of ao. Here,whereall teamshave "no data,"thedecisionmaker was
willingto say thatall teamswereequallywell off,eventhoughone
ofthemmighthaveworkedharder, spentmoremoney,etc.
The optionao servesto fix the "relativeorigins"of theutility

functions.30 The optiona, servesto fix the "scale" of the func-
tions,ifindeedtheconceptofutilityis to be used.
2. Assumethatwe are willingto say thatif foranytwooptionsai, aj,
the percentageincreasein utilityfor one individualdue to a
changefromai to aj isjustequal to thepercentage decreasein utility
fortheotherindividual, thenthetwooptionshave"equal welfare."
This is anotherjudgmenton our part.Thereis nothingarbitrary
about it, but it is certainly
subjectivein thatsomeotherjudgemay
feeldifferently. Thisassumption impliesthattheformof thewelfare
indicator willbe multiplicative.
3. Assumethatthe individuals'attitudestowardslotteries are consis-
tent with the expected utilityhypothesis.This meansthat the
individuals'preferences areas ifutilitynumbers
overlotteries forthe
alternatives exist,and the individuals'preferences forlotteriesare
determinedby the expectedvalue of utilitythe lotterywould
supply. In the economicsjargon,the ratio of utilitynumbers
u (aj)/u(ai)expressesthemaximumnumberof probability unitsthe
individual wouldshiftfroma lotterywhichgivesa; withcertainty to
ao, in orderto getaj withcertainty. The numbers express"strength
of preference" in thatsense.
4. Assumethat "strength of preference,"in the senseoutlinedin 3
above,is thebasisof utilitycomparisons. Thislastassumption is,of
course,stillanothersubjective judgment on ourpart.
If we accept thesebasic assumptions,
thenwe accept themultiplicative
indicatorstatedin the example,but whyshouldwe agreeto any of these
assumptions? Some may agreeand some maynot,and I see no framework
hereforresolving Whenthismethodwasused(DyerandMiles,
suchconflicts.
1976), participants pleasedwithit. Thereare other
were not particularly
measures,otherformulasforcombiningthe measures,etc. Perhapsone of
themwilldo. Allof them,however, willbearsomeverycloseanalogyto what
we haveoutlinedhere.
Whydidn'twe runinto an impossibility result?First,and mostimpor-
tantly,our methoddoes notsatisfytheUniversalDomainaxiom.Preferences
fortheindividuals arefixed,andtheanalysisdoesnotspecifyhowwe should
30 In the Nash frameworkit is called the "threatpoint." It may be thoughtof as a

"status quo." In morephilosophicaltermsit can be interpretedas theHobbesian"state of
nature."
proceedin case the preferences weresomething otherthanwhattheyare.In

particular,the choiceof ao and a, as pointsof comparison couldbe because
people'spreferences are whattheyare. Furthermore, ao is rankedlowestby
everyone.In orderto satisfya universaldomainrequirement, some means
mustbe specifiedfor choosingoptionsto play the role of ao and a, as
preferences change.
Secondly,theprocessmaynot satisfyValue/Feasibility Separation.What
happenswhenpreferences are fixedbut neitherao nora, is feasible?If the
originandscalechangewithfeasibility, thenourPrinciple III willprobablybe
violated.If on the otherhand,ao and a, are fixedregardless of thefeasible
set, thenIndependenceof InfeasibleAlternatives willbe violated.The mea-
surements would be in termsof attitudestowardslotterieswhichmaynot,
themselves, alwaysbe in the feasibleset, and therewould be problemsin
gettingtheprocessto be sufficientlyresponsiveto them.
If you tryto findan indicatorthatservesall of the functions whichwe
originallyaskedof a socialpreference definition,thenwe necessarily become
involvedin an inconsistency. Naturally,thosewho workon thisindicator
approachattemptto provideguidelines forthechoiceof thecriticalparam-
eterswhich,withinour simplesetting, tendto look a littlearbitrary.
Atleast
with this approachit is possibleto evaluateindicatorsin a simplesetting
wheretherealbasicethicalassumptions canbe clearlyseen.
SocialPreference
and Cooperative
Games
As we beginto explorethelogicof thesystemof axiomswe havelisted
throughout the review,manyquestionssurface.Whathappensif we relax
Independenceof InfeasibleAlternatives or modifyit in some naturalway
(Hansson,1973; G-ardenfors, 1973; Mayston,1974)? Does anything special
happenwhentheoptionsarelotteries (Fishburn,1972; Shepsle,1970, 1972)
or whenindividuals have randompreferences (Intriligator,
1973; Fishburn,
1975)? Can we eliminatesocialpreference definitions
altogetherandimpose
insteadvariousconsistency requirementson the choicefunctionitself(Fish-
burn,1974a,c;Parks,1971 ;Schwartz,1975b,c)or drop acyclicitywhile still
usingthe conceptof social preference (Blin, 1973)? Are thereadditional
conceptsof "cheatproof"processesand, if so, how do theyinteractwith
variousaxioms such as neutrality(Pattanaik,1973, 1974, 1975; Kelly,
1975)? Arethereprocesseswhichin somesensegiveeachindividual "partial
control"overprocessoutcomesin wayswhichare consistent withconcepts
of liberalism(Sen, 1970b,1970c;Blau,1975; Gibbard,1974)? Whathappens
if we giveindividuals the freedomto makebindingagreements about their
ballots,tradeballots (Coleman, 1973; Buchanan,1968; Rikerand Brams,

1973; Brams,1975; Schwartz,1975a),or evensellthem(Wilson,1969)?
The progressionof thequestionsin theparagraph abovedemonstrates how
the researchbeginsto blendinto purelybehavioralquestionsand intothe
properdomainof game theory.We have alreadyseen how the axiomatic
methodcan be usedto detectseveralinteractions betweennormative require-
mentsand systemstructure.For these results,only the bare bones of
behavioraltheorieswereused. Is it possibleto extendtheapplicationof our
axiomaticframework to the behavioralhypothesesfoundembeddedin co-
operativegametheory?
Wilson(1971a,1972b) discoveredthatifyouadopta certainpointofview
of cooperativegame theory,thenthe idea of a socialpreference definition
itselfbecomeslinkeddirectly withgametheory.The firststepis to undertake
thementalexerciseof equatingtheconceptof dominancewiththe concept
of socialpreference.So, ratherthansay xP,y,we wouldsayxD,y.The next
stepis to assumeour originalPrinciplesIII, IV, V, andVI (Value/Feasibility
Separation,Universal Domain,Responsive, Independence Alter-
of Infeasible
natives).PrinciplesI and II (Preference
Transitivity and IndifferenceTransi-
tivity)are unnecessary becausedominancerelationsgenerallydo not have
theseproperties.Thesetwoarereplacedbyprinciples whicharecharacteristic
of gametheoreticmodels.31PrincipleVII (RationalChoice)is unnecessary
becausewe willreplaceit withone of theexistingbehavioraltheories about
the relationshipbetweenthe dominancerelationand the outcome.These
behavioraltheorieswhichconnectthedominancerelationto a choicefroma
feasibleset are called solutionprinciples,32 or moreaccurately,theyare
solutionprinciplesforgameswithouttransferable utility(Aumann,1967).
31 Each restrictionhere can usually be identifiedas restrictingthe analysis to a

particular"type" of game. For example, in preferencetheorywe would certainlynot
allow eventslike xPsy and, simultaneously, yPsx,since thattendsto be alien to thevery
concept of preference.In game theory,however,to say that xDy impliesnot yDx is to
say thatthe class of gamesunderconsiderationis the class of "proper" games.
32 Two of the most widely referencedsolution principlesare the core and the Von
Neumann-Morgenstern solution. The followingtwo axioms make theseconceptsprecise

in our setting.The relation D(RI,.... Rn) is the dominance relation,which is func-
tionallyrelatedto the arrayof individualrankings.
Core Principle:For all v and all (RI, .. Rn)

C(v,D(RI ...,Rn) ) = {x E v: yD(RI ...,Rn)x for
no y e v}
Solution: C(v,D(RI, . . . Rn) ) is a subset of v such that
Von Neumann-Morgenstern
Since these solutionprinciplesare all competingforthe positionof a

universalbehaviorallaw,we are not free,theoretically,to selectfromamong
them.Presumably, naturehas alreadydone that.In fact,however, we do not
yet knowwhich,if any,of theseis universally true,or if thechoicefrom
amongthemis situationally determined. So withconfidence thatcan onlybe
supportedby ignorance,we select the principlewe want and pursueits
theoreticalimplications.If we are veryfortunate,someonewillgrindout an
implication which,by an appealto theworldof facts,willresolvesomeof the
competition byeliminating one of thecontenders.Thispossibilityis notat all
remote,sincewe have alreadydiscoveredthatsystembehaviorinducedby
solutionconceptsis closelyrelated(Wilson,1970; Plott,1974) to theprefer-
encebasedsystembehaviorwhichwe knowinducesimpossibility results.
The thirdstepis to determine whattypesof cooperative gamestructures
are compatiblewithany additionalnormative placedon system
restrictions
behavior.We saw, forexample,thatthe fullimplication of PrinciplesI, II,
and VII was a dictatorship. Wilson(1972b) showedthat veryrestricted
aspectsof theseprinciples inducedsystembehaviorwhichis onlyconsistent
withsimplegames.33Further analysis(Bloomfield andWilson,1972; Bloom-
field,1976) has led to theidentificationof constitutional
gamesandcharac-
teristic
functiongames.
The modeof analysislooksverypromising. We are beginning to see how
theessentiallynormative discourseswhichsocialchoicetheorywasoriginally
designedto analyzecanbe directly linkedto thebehaviorallyorientedtheory
of cooperativegames.
The Theoryof RevealedInstitutions
Those who come fromthe studyof marketsystems,or even physical

systems,to the studyof the choicebehaviorof groups,organizations,
legisla-
tive bodies,and othertypesof nonmarketdecisionprocessesare always
struckby the complexityof theproblems.Muchof thedifficulty seemsto
stem fromthe illusivenatureof rulesand institutions. In thissectionwe
explorean embryonictheoryabout how one mightidentify institutional
structures
by usingaxiomaticmodelsof processbehavior.
if x and y are in C(--), thenneitherxDy nor yDx, and ifz E v\c(..), thenthereis some x
E C(--) such thatxDz.
3 Simple games are systemsin which a coalition can be identifiedas "winning" in
the sense that,once formed,it has the power to implementany option.
The formalruleswhichguide social processes,those whichhave been

successfullycodified,are certainlydifficult to identify giventhemanyam-
in our language.
biguities Werethisnotthecase,thelegalprofession mightbe
considerably less prosperous. Evenmoredifficult to identify aretheinformal
rules and institutions such as informalauthority(delegatedby default),
information channels,obligations, long-term agreements, socialcodes,and so
forth.Sincethesearegenerally presumed to be evenmoreimportant thanthe
formalinstitutions, we faceparticularly difficultproblems bothin theareaof
policyanalysisand in thearea of basicresearch.Becauseinstitutions specify
the information and opportunities of processparticipants, predictionsde-
duced fromexistingbehavioralgeneralizations are verysensitive to assump-
tionsabout institutions. Withoutratherextensiveknowledgeaboutinstitu-
tions,we cannotbe confident of our analysisof theimpactofthosepolicies
whichwould alter partof the institutional structure.Furthermore, at the
basicscientific levelwe cannotdistinguish betweenbehavioral generalizations
whichare wrongbecause of a "misspecified parameter value,"the institu-
tions,andthosewhicharesimplywrong.
If we are givena lot of data about the behaviorof a system,then
axiomaticmethodscan helpus identify thetypesofinstitutions bothformal
and informal (or theirbehavioralequivalents)whichgovernthatbehavior.In
a sense,throughout thepaperwe havebeen constructing a theorywhichhas
thispotential.
The idea of a revealedinstitution has threeparts.First,we mustconsider
theformof theprimary data.Secondly,we mustconstruct themathematical
gametheoretic model(s)whichcouldhavegenerated thedata.Finally,we will
thetypesof institutional
investigate arrangements whicharecompatible with
thegamemodel.
The analysisbeginswithan accuratestatement of therelationshipbetween
theindependent variablesof theprocess,preferences and thefeasibleset,and
the dependentvariableprocessoutcome(s).We startwitha stylizedfactof
the old-fashioned "fittedcurve"variety.That is, we assumethatour data
have been givento us in a formwhichcan be summarized by a choice
functionC(v,Rl, . . .Rn).
. The functionsimplyindicates for everyvalue of
the independentvariables(v,R1, ... 5Rn) the option(s) whichwere chosen by
the process.Of coursewe alreadyhaveproblems, sincedata nevercomein
convenientforms,but let's continueand see whatwe mightbe able to do
underfavorablecircumstances. We wantto establishan "upperlimit"on our
theoretical
capabilities.
an exampleis introduced
In orderto facilitatethediscussion, in Figure4.
On theleft-hand side,individual
preferencesarelisted.Onlythree,casesI, II,
584 CharlesR. Plott
FIGURE 4
Revealed InstitutionsExample
Implied
IndividualPreferencesFeasible Set Choice Dominance Relation
Case (R1 ,R2,R3) v C(v,R1,R2,R3) D(R1 ,R2,R3)
x y z{ x,y,z} {x} - not yDx and not zDx

I y z x x,y} {x} . xDy and not yDx
z x y x,4z x,z} - not xDz and not zDx
y,z} {y}) yDz and not zDy
y x z +x,y,z} (y} - not xDy and not zDy

II x z y x,y} {y} yDx and not xDy
z y x x ,z} (xD xDzandnotzDx
Ay,z} y,z} C- not yDz and not zDy
z y x (x,y,z) {z} - not yDz and not xDz

III x z z x,y0 {x} xDy and not yDx
y x y x,z} zDx and not xDz
Y,Z z{ zDy and not yDz
and III, of the 2197 possibleconfigurations for threeoptionsand three

individualsare listed.34In the center,the feasiblesets and the resulting
choicesare listed.For eachconfiguration ofpreferences ofthethreeoptions,
thereare fourdifferent possiblefeasiblesets,excludingsingleton
sets.On the
right-handside of thefigureare thethreeimplieddominancerelations to be
discussedbelow.
The scope of theanalysiswillbe limitedby threemaintained hypotheses.
These are chosenbecausetheylead us back to conceptswe havediscussed
before.
1. The processcan be modeledas a cooperative
gamewithnontrans-
ferableutility.
3 Includingthe possibilityof indifference,

each individualcan rank threeoptions
thirteendifferentways. There are threeindividuals,so thereare 133 = 2197 different
configurations.
Thisassumption meansthatwe mustrecoverforeachvalueoftheparameters

an implieddominancerelation.Recall,we saythatan optionx dominatesan
optiony in case somecoalitionhas"thepowerto enforcex overy" and each
memberof the coalitionprefers x overy. Boththenotionof "power"and
thenotionof "enforce"willremainabstractthroughoutouranalysis.
The second and thirdassumptionsplace limitations on the formsof
cooperativegameswe willinvestigate.
2. The dominancerelationbetweenpairsof optionsdoes not depend
uponthefeasibilityof additionaloptions.
3. The outcomesare alwaysthe core of the cooperativegame.(The
coreis theset of feasibleoptionseach ofwhichis dominated
by no
otherfeasibleoption.)
Summarizing now,our firstobjectiveis to recovera function D(R1, . . ..
Rn) which assignsto each value of the variables(R1 Rn) a binary
relation,D, whichwe will call the implieddominancerelation.The first
hypothesis impliesthe existenceof sucha dominancefunction.The second
hypothesis. whichwe tentatively assumeis true,impliesthatthevariablev,
representing the feasibleset,is nonoperative as a variablein thisfunction, so
it can be droppedas a variablein the function statement. Thisassumption is
verysimilarto theValue/Feasibility SeparationPrincipleused in thesecond
section.Withtheseassumptions we can applyrepeatedly thethirdassumption
and recover theimplieddominancerelationfromthedata.
Let's see what typeof dominancerelationis impliedby thedata on the
figure.Considerfirstthefirstpattern of preferences, case I. Whenthefeasible
set is (x,y,z} thechoiceis x. By hypothesis 3, x is thecoreandis therefore
undominated. We conclude [not yDx and not zDx]. Continuenow to the
case wherethe feasibleset is (x,y}, and thechoiceis x. Since x is thecore,
we conclude [not yDx] whichis consistent withtheconclusionabove,and
we also conclude[xDy] forifwe had [notxDy], theny wouldbe undomi-
natedand feasibleand therefore, contraryto fact,in thecore and chosen.
Whenthe feasibleset is (x,z}. we conclude [not xDz and not zDx] since
both are chosen and therefore, by hypothesis, are not dominatedby any
otherfeasibleoption.Finally,whenthe feasibleset is {y,z). we conclude
[yDz and not zDy]. The argument is the same as thatappliedwhenthe
feasibleset is {x,y . Application of oursecondhypothesis meansthatwhena
dominancerelationis determined fora pair,it is also determined whenthat
pairis embeddedin a largerset. So, we are now freeto speakof theimplied
dominancerelationforthisconfiguration of preferences.
For case I, we have thusconcluded,the implieddominancerelationis
simplyxDy and yDz. For no otherpair(including{x,z}) does therelation

"dominance"hold. For thepatternincase II theimplieddominancerelation
is yDx and xDz. For case III it is zDx and xDy and zDy. All of these
conclusionsfollowfromtherepeatedapplicationoftheargument usedin the
preceding paragraph.
Whathavewe done? For each of our threecases of preferenceconfigura-
tionswe have,following proceduresimplicitin Wilson(1970) and suggested
in Plott(1974), attacheda dominancerelation.We haverecovered fromthe
data a function D(R1 , . . . Rn) which relates preferencesto the implied
dominancerelation.Of courseif thereweremorecases,we wouldprobably
have to resortto computertechniquesor mathematical simplifications,but
no immediate theoreticalproblemsareinvolved.
Once we have the functionD(R1,. . . Rn) whatdo we do? Dominance,
recall,occurswhen"power"and "preferences" coincide.Oncewe haveit,we
can deduce the powerrelationships implicitin theprocessbehavior(under
our threemaintainedhypotheses, of course).We let A(x,y,Rl,. . . Rn) be
the familyof coalitionswhichhavethe"powerto implement x overy" given
the patternof individualpreferencesR1, . . . Rn. If a coalition C is a member
of this family,thenwe have xDy in case everyonein C prefersx to y.
Considermajorityrule exampleand let C be some majoritycoalition.If
everyonein C prefers x to y, theyhave,sincetheyarea majority, "thepower
to implement x overy," so we wouldconcludexDy.
Our goal is now to findtheimplicitpowerrelationships as represented by
thefunction A. Againto limitthescopewe introduce a hypothesis.
A is independent
4. The function of(R1, . . . Rn).
This assumptionmeanswe can writeA as simplyA(x,y).Operationally the
assumption meansthattheinstitutions and
aresuchthatrights powers do not
depend upon preferences.35 Absolutemajorityrule,as opposed to simple
majorityrule,has thisproperty wherebythe "winningcoalition"does not
dependuponwhether or notsomeoneis indifferent.
We at last have a conclusion.Over the rangefor whichthe data are
reported,thisprocesswhichgeneratedthe data on figure4 is behaviorally
equivalentto a collegium.Mr.I hasthepowerto blockanyactionbutcannot
himselfinitiatean actionunlessaccompaniedby eitherMr. 2 or Mr. 3 or
both.Forthefunction A we have
3 Bloomfield and Wilson (1972) call the functionA(x,y) a constitution.We chose

not to use that terminology,since the termhad alreadybeen used to identifya closely
relatedmathematicalobject (Plott, 197 1).
A(X,y) - A(y,x) = A(X.z) = A(z x) = A(y,z) = A(z,y)
So therewe haveit. We beganwithsomedataaboutprocessbehavior. We

introduced (maintained)hypothesesabout the processwhichgenerated the
data. Fromthesewe deducedsome hypothesesabouttheinstitutions, per-
hapsinformal institutions,
implicit
in theprocessbehavior.
On paperthe idea has possibilities.Therehas evenbeen an attemptto
applya ratherremoteversionof theanalysis(McFadden,1975). Whatcan go
wrong?The readerprobablyalreadyhas a list. First,it is not obviousthat
stylizedfactscanbe generated in "realworld"datawithoutsometheory.In a
sensethat'swhateconometrics, theidentification
problem,and relatedmea-
surement theoriesare all about.Second,ourconclusions areverysensitiveto
our fourmaintainedhypotheses.If, forexample,we werenot givenchoice
data forall pairsof options,we could reachsubstantially different conclu-
sions, dependingupon whetherwe thoughtthe outcome was a Von
Neumann-Morgenstern solutionratherthanthe core. Operationalproblems
abound. How can one identify nonuniquechoices?In fact,forany really
complicatedprocess,how can you evenidentify whichsocialstateis chosen?
The analysissaysnothingabout randomelements, butsurelyanymeaningful
analysismusteven anticipatethe possibility of randompreferences. What
happenswhenfeasiblesetsarealwaysinfinite?
Extendingthelistsimplyunderlines the factthatwe arereallyin thearea
of pretheory now whereit is impossibleto tell whetherobjectionsto the
theoryreflectlimitationsof thetoolsor thelimitations
ofourimagination.
SOME NOTES ON THE OVERVIEW
In closing,a remarkor so aboutourperspectiveis in order.The readerhas

been warnedat severalpoints that our interpretation is not universally
acceptedor, in some cases,even recognized.The important turningpoints
shouldbe underlined.
First,our formulation of the impossibilitytheoremcannotbe found
explicitlyin theliterature.
Some,followingArrow,formulate theproblemas
a mappingfromindividualpreferences intoa socialpreference. Othersstart
witha choicefunctionand makeno explicitreference to a socialpreference.
We implicitly startedwitha social preferenceexistenceaxiom (an implicit
Principle0). From this and other axioms we could deduce the Arrow
formulation.
Sinceour formulation differs,our axiomsdifferalso. The Value/Feasibil-

itySeparationaxiomis new,forexample.Use of theterm"socialpreference"
definition is new.Our use of theconceptof "feasibility,"whilenot new,is
certainlyweddedto a particular subsetof theliterature.The sameholdsfor
all of our "process"interpretationsandevenourrepeateduse of theconcept
of "intuition"usedinmotivating theresearch.
The organization and motivation forthethirdsectionhas not appeared,
explicitly,elsewhere.The same holds true of the discussionof revealed
institutions.
Whenthe readerpicksup the challengewithwhichwe began,and ifhe
does indeeddemonstrate thatthewholefieldis irrelevant,he shouldcheck
withtheoriginalsourcesbeforemakinghis claim.Whilehe mayprovideme
withan incentive to "resteasily"withmywork,someofthescholarswhose
workis referenced heremighthesitateto join me. If, on theotherhand,the
readerhas foundthatthistheoryprovideshimwithnewtoolswithwhichto
attackold problems,he shoulddefinitely consultthe paperswe haverefer-
encedhere.In our effortto providea unifiedtheory,we haveonlytouched
uponthemanyexciting ideasand insightswhichexist.
Manuscript December17, 1975.

submitted
Finalmanuscript
received
March8, 1976.
REFERENCES
Adelsman,R., and Whinston, A. 1975. "The Equivalenceof ThreeSocial

DecisionFunctions,"PurdueUniversity,WP-75-155, November1975.
Alt,Franz.1971. "On theMeasurement et al.,
of Utility,"in J.S. Chipman,
eds., Preference,Utilityand Demand. San Francisco:HarcourtBrace
Jovanovich,Inc.
Arrow,K. J. 1963. Social Choiceand IndividualValues.New York: Wiley;
2ndedition1963.
__. 1959. "Rational Choice Functions and Orderings,"Economica, 1959,
vol. 26, pp. 121-127.
Notes on Rawls' Theoryof Jus-
1973. "Some Ordinalist-Utilitarian
1973,vol.70, pp. 245-263.
tice,"JournalofPhilosophy,
Atkinson,A. B. 1970. "On the Measurement of Inequality,"Journalof
EconomicTheory,vol.2, pp. 244-263.
1975.EconomicsofInequality,Oxford:Clarendon Press.
Aumann,RobertJ. 1967. "A Surveyof CooperativeGameswithoutSide
Payments," in M. Shubik,ed., Essaysin Mathematical Economics.Prince-

ton:Princeton UniversityPress.
Barbera,S. 1975. "The Manipulability of SocialChoiceMechanisms ThatDo
Not Leave Too Much to Chance,"mimeographed, Northwestern Univer-
sity,1975.
Bartoszynski. 1972. "Power Structurein DichotomousVoting,"Econo-
metricu, November1972,vol.40 pp. 1003-1019.
Batra,R. N., and Pattanaik,P. K. 1972a. "Transitive Multi-Stage Majority
Decisions with Quasi-transitive IndividualPreferences," Econometrica,
1972,vol.40, pp.1121-1135.
. 1972b. "On Some Suggestions forHavingNon-binary Social Choice
Functions,"Theoryand Decisions, 1972, vol. 3, pp. 1-11.
Black, R. D. 1958. The Theoryof Committees and Elections.New York:
Cambridge UniversityPress.
Blair,D. H.; Bordes,Georges;Kelley,JerryS.; andSuzumura, Kotaro.1976.
"Impossibility TheoremswithoutCollectiveRationality," Journal of Eco-
nomicTheory,forthcoming.
Blau, J. H. 1975. "LiberalValuesand Independence," Reviewof Economic
Studies,1975,vol.42, pp. 395-401.
Blin,J. M. 1973. Patternsand Configurations in EconomicScience. Cam-
bridge:D. ReidelPublishing Co., Dordrecht, Holland.
Blin,J. M., and Whinston, A. B. 1975. "Discriminant FunctionsandMajority
Voting,"Management Science,1975, vol.21, pp. 557-566.
Bloomfield, S. 1976. "A Social ChoiceInterpretation of thevonNeumann-
Morgenstern Game,"Econometrica, Jan. 1976, vol.44, pp. 105-114.
Bloomfield,S., and Wilson, R. D. 1972. "Postulates of Game Theory,"
JournalofMathematical Sociology,1972, vol. 2, pp. 221-234.
Bowman,V. J.,and Colantoni,C. S. 1973. "MajorityRuleunderTransitivity
Constraints," Management Science,1973,vol. 1, pp. 1029-1041.
. 1974. "FurtherCommentson MajorityRule underTransitivity Con-
Management
straints," Science,1974,vol.20.
Brams,StevenJ. 1975. GameTheoryandPolitics.NewYork:The FreePress.
Brown,D. J. 1973a. "AcyclicChoice,"CowlesFoundationdiscussion paper,
1973.
1973b. "Aggregation ofPreferences," forthcoming in Quarterly Journal
of Economics.
. 1974. "An Approximate Solutionto Arrow'sProblem,"Journalof
EconomicTheory,1974,vol.9, pp. 375-383.
. 1975a. "AcyclicAggregation overa FiniteSet of Alternatives," Cowles
FoundationDiscussionPaperNo. 391, Yale University, 1975.
590 CharlesR. Plott
. 1975b. "CollectiveRationality," CowlesFoundationDiscussionPaper

No. 393, Yale University, 1975.
Buchanan,J. M. 1954a. "IndividualChoice in Votingand in theMarket,"
JournalofPoliticalEconomy,August1954,vol.62, pp. 334-43.
_. 1954b. "Social Choice, Democracy,and Free Markets," Journal of
PoliticalEconomy,April1954,vol.62, pp. 114-123.
_. 1968. The Demand and Supply of Public Goods, Chicago: Rand
McNally.
Campbell,D. E. 1973. "Social Choiceand Intensity of Preference,"
Journal
ofPoliticalEconomy,1973,vol. 81, pp.211-218.
Coleman,JamesS. 1973. The Mathematics of CollectiveAction,Chicago:
Aldine.
Dahl, R. A., and Lindbloom,C. E. 1953. Politics,Economicsand Welfare,
HarperTorchbookedition.NewYork:Harperand Row.
DeMeyer,F., andPlott,C. R. 1970. "The Probability of a CyclicalMajority,"
Econometrica, 1970,vol. 38, pp. 345-354.
. 1971. "A Social WelfareFunctionUsing'RelativeIntensity' ofPrefer-
ence," QuarterlyJournalof Economics,February1971, vol. 85, pp.
179-186.
Dyer,J. S., and Miles,R. F., Jr.1976. "An ActualApplication ofCollective
Choice Theoryto the Selectionof Trajectories forthe MarinerJupiter/
Saturn1977Project,"Operations Research(fortheory)1976.
Farquharson, R. 1969. Theoryof Voting.NewHaven:Yale University Press.
Ferejohn,J. A. 1976. "Brown'sTheoryof CollectiveChoice."SocialScience
Working PaperNo. 117,California Institute
ofTechnology, 1976.
Ferejohn,J. A., and Grether, D. M. 1974a. "On a Classof RationalSocial
Decision Procedures,"Journalof Economic Theory,1974, vol. 8, pp.
471-482.
. 1974b. "Weak Path Independence," Social Science WorkingPaper No.
80, California
Institute
ofTechnology, 1974.
Fine,B., and Fine,K. 1974. "Social Choiceand IndividualRanking,"
Review
ofEconomicStudies,1974,vol.41, pp. 303-322,459-475.
Fine,K. 1972. "SomeNecessary and Sufficient
Conditions forRepresentative
Decisionon Two Alternatives, Econometrica, November1972,vol.40, pp.
1083-1090.
Fishburn, P. C. 1970. "Arrow'sImpossibilityTheorem:ConciseProofand
InfiniteVoters,"JournalofEconomicTheory,1970,vol.2, pp. 103-106.
. 1972. "Lotteriesand Social Choice,"Journal of Economic Theory,
1972,vol.5, pp. 189-207.
_ 1973a. The Theory of Social Choice, Princeton: PrincetonUniversity

Press.
. 1973b. "Voter Concordance, Simple Majorities, and Group Decision
Methods,"Behavioral
Science,1973,vol. 18,pp. 364-376.
_ 1974a. "On Collective Rationality and a Generalized Impossibility
Theorem,"Reviewof EconomicStudies,1974,vol.41, pp. 445-457.
__. 1974b. "Paradoxes of Voting," American Political Science Review,
1974,vol.68, pp. 537-546.
. 1974c. "Social ChoiceFunctions,"SIAM Review,January1974, vol.
16,pp. 63-90.
_ 1974d. "Impossibility Theorems without the Social Completeness
1974,vol.42, pp. 695-704.
Axiom,"Econometrica,
. 1975. "A ProbabilisticModel of Social Choice: Comment,"Review of
EconomicStudies,1975,vol.42, pp. 297-301.
__. 1976. "Dictators on Blocks: Generalizationsof Social Choice Impos-
sibility
Theorems," Journalof Combinatorial TheoryB, 1976,vol.20.
Fishburn, P. C., and Gehrlein,W.V. 1975. "Electionby Chancein a Worldof
UncertainVoters,"mimeographed, PennsylvaniaState University,
1975.
Girdenfors, P. 1973. "PositionalistVotingFunctions,"TheoryandDecision,
1973,vol.4, pp. 1-24.
Garman,M., and Kamien,M. 1968. "The Paradoxof Voting:Probability
Calculations," BehavioralScience,1968,vol. 13,pp. 306-323.
Gehrlein,WilliamV., and Fishburn,P. C. 1975. "The Probability of the
Paradox of Voting:A ComputableSolution,"mimeographed, Pennsyl-
vania State University, College of BusinessAdministration, University
Park.
Gibbard,A. 1973. "Manipulation of VotingSchemes:A GeneralResult,"
Econometrica, 1973,vol.41, pp. 587-601.
. 1974. "A Pareto-Consistent Libertarian
Claim,"Journalof Economic
Theory,1974,vol.7, pp. 388-410.
Goodman,Leo A., and Markowitz, Harry.1952. "Social WelfareFunctions
Based on Individual Rankings," American JournalofSociology,November
1952,vol.57, pp. 257-262.
Hansson,B. 1969a. "VotingandGroupDecisionFunctions," Synthese,1969,
vol.20, pp. 526-537.
. 1969b. "Group Preferences,"Econometrica, 1969, vol. 37, pp. 50-54.
. 1972. "The Existence of Group Preferences,"WorkingPaper No. 3,
TheMattiasFremlingSociety,Lund,Sweden,1972.
. 1973. "The IndependenceConditionin theTheoryofSocialChoice,"
TheoryandDecision,1973,vol.4, pp. 25-49.
592 CharlesR. Plott
Harsanyi,J. C. 1955. "CardinalWelfare,Individualistic Ethics,and Inter-

personalComparisonsof Utility,"Journalof PoliticalEconomy,1955,
vol.63, pp. 309-321.
Herzberger, H. G. 1973. "OrdinalPreference and RationalChoice,"Econo-
metrica,1973,vol.41, pp. 187-237.
Hildreth,C. 1953. "Alternative ConditionsforSocial Ordering," Economet-
rica,1953,vol.21, pp. 81-95.
Hurwicz,L. 1973. "The Designof Mechanisms for ResourceAllocation,"
American EconomicReview,May 1973,vol.63, pp. 1-30.
M. 1973. "A Probabilistic
Intriligator, Model of Social Choice,"Reviewof
EconomicStudies,1973,vol.40, pp. 553-560.
Jamison,Dean T. 1975. "The Probability MajorityRule: an
of Intransitive
EmpiricalStudy,"PublicChoice,Fall 1975,vol.23, pp. 87-94.
Jamison, D. T., and Luce, E. 1973. "Social Homogeneity andtheProbability
of IntransitiveMajorityRule,"JournalofEconomicTheory,1973,vol. 5,
pp. 79-87.
Kelly,J. S. 1974a. "NecessityConditionsin VotingTheory,"Journalof
. 1974b. "Voting Anomalies, the Number of Voters and the Numberof
1974,vol.42, pp. 239-251.
Econometrica,
Alternatives,"
__. and Social Choice Functions with a Large
1975. "Strategy-proofness
ImageSet," mimeographed, 1975.
Kemeny,J. 1959. "Mathematics withoutNumbers," Daedalus,1959,vol.88,
pp.577-591.
Kirman,A. P., and Sondermann, D. 1972. "Arrow'sTheorem,ManyAgents
and InvisibleDictators," Journal of EconomicTheory,1972,vol. 5, pp.
267-277.
Kramer,G. H. 1972. "SophisticatedVotingoverMultidimensional Choice
Space,"JournalofMathematical Sociology,1972,pp. 165-180.
. 1974. "On a Class of EquilibriumConditionsfor MajorityRule,"
1975. "A DynamicModelof PoliticalEquilibrium," CowlesFounda-
tionDiscussionPaperNo. 36, June1975.
Kramer,G. H., and Klevorick, A. K. 1974. "Existenceof a 'Local' Coopera-
tiveEquilibrium in a VotingGame,"Reviewof EconomicStudies,1974,
vol.41, pp. 539-547.
Kuga,K., and Nagatani,H. 1974. "Voter Antagonism and the Paradoxof
Voting,"Econometrica, 1974,vol.42, pp. 1045-1067.
Levenglick,A. 1975. "Fair and ReasonableElectionSystems,"Behavioral
Science,1975,vol.20, pp. 34-46.
Lichtenstein,S., and Slovic,P. 1971. "Reversalsof PreferenceBetweenBids
and ChoicesinGambling Decisions,"JournalofExperimental Psychology,

1971,vol.89, pp.46-55.
Luce,R. D., and Raiffa,H. 1957. GamesandDecisions.NewYork:Wiley.
Marschak, J.,and Radner,R. 1972.EconomicTheoryof Teams.NewHaven:
Yale University Press.
Mas-Colell,A., and Sonnenschein, H. 1972. "GeneralPossibility Theorems
and GroupDecisions,"Reviewof EconomicStudies,1972, vol. 39, pp.
185-192.
May,K. 0. 1952. "A Set of Independent, Necessaryand Sufficient Condi-
tions for Simple MajorityDecision,"Econometrica,1952, vol. 20, pp.
680-684.
Mayston,David J. 1974. The Idea of Social Choice.New York: St. Martin's
Press.
McFadden,D. 1975. "The RevealedPreference of a Government Bureau-
cracy: Theory," The Bell Journalof Economics, 1975, vol. 6, pp.
409-416.
Merchant, D. K., and Rao, M. R. 1975. "MajorityDecisionsandTransitivity:
Some Special Cases," Universityof Rochester,Graduate School of
Management, Working PaperSeries7506, October1975.
Murakami,Y. 1961. "A Note on the GeneralPossibility Theoremof the
Social Welfare Function," Econometrica, April1961,vol. 29, pp. 244-46.
. 1968.LogicandSocial Choice. London: Macmillan.
Nash,J. F. 1950. "The Bargaining Problem,"Econometrica,1950, vol. 18,
pp.155-162.
Niemi,R. G. 1969. "MajorityDecision-making withPartialUnidimension-
ality,"American PoliticalScienceReview,1969,vol.63, pp. 488-97.
Niemi,R. G., and Weisberg, H. F. 1968. "A Mathematical Solutionforthe
Probability of the Paradoxof Voting,"BehavioralScience,vol. 13, pp.
317-23.
Parks,R. R. 1971. "Rationalizations, Extensionsand Social ChoicePaths,"
Washington University,St. Louis,1971.
1973. "The Bergson-Samuelson WelfareFunction:An Impossibility,"
mimeographed, Washington St. Louis,1973.
University,
Pattanaik,P. K. 1971. Votingand CollectiveChoice.New York: Cambridge
University Press.
. 1973. "On the Stabilityof SincereVotingSituations,"Journalof
. 1974. "Stabilityof SincereVotingunderSome Classesof Non-binary
GroupDecisionProcedures," Journalof EconomicTheory,1974, vol. 8,
pp. 206-224.
__. 1975. "Strategic Voting without Collusion under Binary and Demo-
594 CharlesR. Plott
craticGroupDecisionRules,"ReviewofEconomicStudies,1975,vol.42,
pp. 93-103.
Plott, C. R. 1967. "A Notion of Equilibriumand Its Possibilityunder
MajorityRule,"AmericanEconomicReview,1967,vol. 57, pp. 788-806.
. 1971. "RecentResultsin theTheoryofVoting,"in M. Intriligator,
ed.,
Frontiers Economics.Amsterdam:
of Quantitative North-Holland.
. 1972. "Ethics,Social Choice Theoryand the Theoryof Economic
Policy,"JournalofMathematical Sociology,1972,vol.2, pp. 181-208.
__. 1973. "Path Independence, Rationality and Social Choice," Econo-
metrica,1973,vol.41, pp. 1075-1091.
. 1974. "Rawls Theoryof Justice:Some Impossibility Results,"paper
presented at thePublicChoiceSocietyMeetings, 1974.
Plott,C., and Levine,M. 1975. "On Usingthe Agendato InfluenceGroup
Decision:Theory,Experiments and Application,"Social ScienceWorking
PaperNo. 66, CaliforniaInstituteofTechnology, April1975.
Pomeranz,J. E., and Weil,R. L. 1970. "The CyclicalMajorityProblems,"
Communications of theACM,vol. 13, 1970,pp. 251-254.
Rawls,J. 1971.A TheoryofJustice.Oxford:Clarendon Press.
Ray, P. 1973. "Independenceof Irrelevant Alternatives," Econometrica,
1973,vol.32, pp. 987-991.
Richelson,Jeffrey. 1975a. "Conditionson Social ChoiceFunctions,"A. R.
Wagner & Co., BeverlyHills,November1975,mimeographed.
.1975b. "A Comparative Analysisof SocialChoiceFunctions," Behavioral
Science,1975,vol.20.
Richter,M. K. 1966. "RevealedPreference Theory,"Econometrica,1966,
pp. 635-645.
. 1971. "RationalChoice,"in J. S. Chipman,et. al., eds.,Preference,
UtilityandDemand,San Francisco:HarcourtBraceJovanovich.
Riker,W. H., andBrams,S. 1973. "The ParadoxofVoteTrading," American
PoliticalScienceReview,1973,vol. 67, pp. 1239-1247.
Riker,W. H., and Ordeshook, P. 1973. An Introduction to PositivePolitical
Theory,EnglewoodCliffs:Prentice-Hall.
Rosenthal,R. W. 1975. "VotingMajoritySizes,"Econometrica, 1975,vol.
43, pp.293-299.
Salles,M. 1975. "A GeneralPossibility Theoremon GroupDecisionRules
withPareto-Transitivity,"JournalofEconomicTheory,forthcoming.
Samuelson,P. A. 1967. "Arrow'sMathematical Politics,"in S. Hook, ed.,
Human Valuesand EconomicPolicy.New York: New York University
Press.
Saposnik,R. 1975. "Social ChoicewithContinuous Expression of Individual
Preferences," Econometrica, 1975,vol.43, pp. 683690.
Satterthwaite, M. A. 1975. "Strategy-Proofness and Arrow'sConditions,"

JournalofEconomicTheory,1975,vol. IO, pp. 187-217.
Schwartz,T. 1970. "On the Possibilityof RationalPolicy Evaluation,"
TheoryandDecision,1970,vol. 1, pp. 89-106.
. 1972. "Rationalityand theMythof theMaximum," Nous, 1972,vol.
6, pp.97-117.
. 1975a. "Vote Tradingand ParetoEfficiency," PublicChoice,Winter
1975,vol.24, pp. 101-110.
. 1975b. "Notes on the AbstractTheoryof Preference and Choice,"
mimeographed, Carnegie-Mellon University,
Pittsburgh, 1975.
. 1975c. "Choice Functions,'Rationality'ConditionsandVariations on
theWeakAxiomofRevealedPreference," mimeographed, Carnegie-Mellon
University,Pittsburgh,1975.
Sen, A. K. 1970a. CollectiveChoice and Social Welfare.San Francisco:
Holden-Day.
. 1970b. "InterpersonalAggregation andPartialComparability," Econo-
metrica,1970,vol.38, pp. 393-409.
. 1970c. "The Impossibility of a ParetianLiberal,"Joumalof Political
Economy,1970,vol.78, pp. 152-157.
1971. "Choice Functionsand RevealedPreference," Reviewof Eco-
nomicStudies,1971,vol.38, pp. 307-317.
.1973. On EconomicInequality.Oxford:Clarendon Press.
.1976. "Social ChoiceTheory:A Re-Examination," a paperdelivered
to
theEconometric Society,Toronto,1975, mimeographed, LondonSchool
of Economics,revised1976.
Sengupta,M. 1974. "On a Conceptof Representative Democracy,"Theory
andDecision,1974,vol. 5, pp. 249-262.
Shepsle,K. A. 1970. "A Note on Zeckhauser's 'MajorityRulewithLotteries
on Alternatives':theCase of theParadoxofVoting,"Quarterly Journalof
Economics,1970,vol. 84, pp. 705-709.
. 1972. "The Paradoxof Votingand Uncertainty," in R. G. Niemiand
H. F. Weisberg, eds.,Probability Models in CollectiveDecisionMaking.
Columbus:Merrill.
. 1974. "The Theoryof CollectiveChoice,"in C. Cotter,ed.,Political
ScienceAnnual,vol.5, Indianapolis:Bobbs-Merrill,pp. 1-87.
Sloss, Judith.1973. "Stable Outcomesin MajorityVotingGames,"Public
Choice,Summer1973,vol. 15,pp. 19-48.
Slutsky,S. 1975. "Abstentions and MajorityEquilibrium," Journalof Eco-
nomicTheory,1975,vol. 11,pp. 292-304.
Smith,J. H. 1973. "Aggregation of Preferencesand VariableElectorate,"
596 CharlesR. Plott
Smith,T. E. 1974. "On the Existenceof Most Preferred Alternatives,"

InternationalEconomicReview,1974,vol. 15,pp. 184-194.
Sonnenschein, Hugo. 1974. "An AxiomaticCharacterization of the Price
Mechanism," Econometrica, May 1974,vol.42, pp. 425-434.
Tversky,A. 1969. "Intransitivity of Preferences,"PsychologicalReview,
1969,vol.76, pp. 31-48.
Wilson,R. B. 1969. "An AxiomaticModel of Logrolling," AmericanEco-
nomicReview,1969,vol. 59, pp. 331-341.
. 1970. "The Finer Structureof RevealedPreferences," Journalof
. 1971a. "A Game-Theoretic Analysisof Social Choice,"in B. Lieber-
man,ed., Social Choice.New York: Gordonand Breach,SciencePublish-
ers.
. 1971b. "Stable CoalitionProposalsin MajorityRule Voting,"Journal
ofEconomicTheory,1971,vol. 3, pp. 254-271.
. 1972a. "Social ChoiceTheorywithouttheParetoPrinciple," Journal
ofEconomicTheory,1972,vol. 5, pp. 478-486.
. 1972b. "The Game-Theoretic Structureof Arrow'sGeneralPossibility
Theorem," JournalofEconomicTheory,1972,vol. 5, pp. 14-20.
1975. "On theTheoryof Aggregation," JournalofEconomicTheory,
1975,vol. 10, pp. 89-99.
Young, H. P. 1974a. "An Axiomatization of Borda's Rule," Journalof
. 1974b. "A Note on PreferenceAggregation,"Econometrica, 1974, vol.
42,pp. 1129-1131.
1975a. "Social Choice Scoring Functions," SIAM Journalof Applied
Mathematics,June1975,vol.28.
of New York,
. 1975b. "ExtendingCondorcet'sRule," CityUniversity
GraduateSchool,December1975.

Elección Social Revisión e Interpretación

Uploaded by

Copyright:

Available Formats

Elección Social Revisión e Interpretación

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Elección Social Revisión e Interpretación

Uploaded by

Copyright:

Available Formats

Axiomatic Social Choice Theory: An Overview and Interpretation

Author(s): Charles R. Plott

Whyshould a politicalscientistbe interestedin somethinglike social

*Financial support supplied by the National Science Foundation is gratefully

AMERICAN JOURNAL OF POLITICAL SCIENCE, XX, 3, August1976 511

social actions.Almostanything we say and/oranyonehas eversaid about

real differencein termsof their"parts"between,say,the Bordacountand

As an introduction to axiomaticsocialchoicetheory,let'sexaminea few

firstperson xP1yP1z, (x is preferred

Now whichalternativeshouldthissocietyget? Naturally, it shouldget

since2 and 3 preferz to x, I and 2 prefery to z, and I and 3 preferx to y.

The thirdreactionamountsto a denialthatthephenomenon exists."Well,

majorityrule cycle occurs,then societyis simplyindifferent-any of the

Now supposefirst,thatw is the statusquo-the alternative

As thereadercan see ifhe tracesout theprocess,theultimatechoiceis z. But

everyoneprefersw to z, and declarew, x, and y all "equallygood" or all

Borda Count: A Four-Option,

Mr.1 Mr.2 Mr.3 Mr.4 Mr.5 Mr.6 Mr.7

Pareto-dominated by y. Withz gonethereare onlythreeoptions,so points

THE GENERAL POSSIBILITY THEOREM

well) whereindividual membersof thesocietyhavetherankings as givenin

mayseemas thoughwe havesaid nothingbut theobvious,we havealready

ties? There are literallymillionsof processes.In fact,by makingsmall

the choice is a set of optionsas longas we understandthatthismeansone

C(v,Rl, . . . Rn) = C(v,R 1,*.*

Ratherthan followthis interesting argumenthere,we will postponethe

Perhapsit is so trivialthatit neednot be mentioned thatanynormative

Is therea processwhichdoes not behavein a paradoxicalway?Is therea

Let E be a universalset of options.These are all the optionsyou can

Theorem.If E has at leastthreeoptionsand C(v,Ri,... Rn) is the choice

Thatis a mouthful, but whatdoes it mean?It probablymeansthatifyou

be retracedwitha greatdeal of care.Thisreexamination

Beforewe go to all thattrouble,however,let's look at the stakes.Some,

There is a two-headedanswerto this. First,one would expect that

The setof feasibleoptionscouldbe a consumption possibilities

formula.The only admissibledefinitions of welfareare those whichare

The Analysis:An Overview

The problemwas, and is, thatwhenwe are preciseand clearabout our

and thus"solve"thedilemma, areabsolutelyessentialto someotherinterpre-

Any systemthat consistently violatedthisprinciple, the thesismaintains,

The Domainof SocialPreference

Somehowwe got froma lot of principles to a positionwherewe are

'For definitionsand examples of prisonerdilemma games see Luce and Raiffa

certainspecialoptions,the social rankingneed not be defined.7The social

A "free"tripleis a setof threealternativesforwhichthesocialpreference

ootnote6 applies herealso.

someotherthree-element set as well-it dependsupon thegeneralpattern of

8For a good summarysee Fishburn(1973) or Sen (1971).

The second approachis to tryto place restrictions on the patternsof

conjecturedto be zero. Of course,untilsomeoneprovesit thereis alwaysa

9If thereare only ten people, it is impossibleforexactly95 percentof themto have

Some dictatorial set thencouldbe takenas an interval. The resultsaysthat

l 'How could two sets be dictatorial?Notice that the preferenceof a dictator is

This principletellsus thatcertaintypesof modelbehaviorareinconsistent

13This can be formalizedas follows: Let Ai be the set of actions reservedfor

Whatabout cooperativegames?The key idea thereis theidea of domi-

RI . ., Rn) satisfiesPrincipleVI. Note there are obvious difficultieswith existence,

tarnishedand damageda little,but it is clearlystillin theair.Certainlythe

This problemis so complicatedthatit would take a book-length paper

tion. Butwaita minute!Whyare theretwo conceptshere-expectation and

C(v,Rl, . . . Rn) = C(v,R 1,.