The Irrationality of Intransitivity

Author(s): Gordon Tullock

Source: Oxford Economic Papers, New Series, Vol. 16, No. 3 (Nov., 1964), pp. 401-406
Published by: Oxford University Press
Stable URL: http://www.jstor.org/stable/2662577 .
Accessed: 11/03/2011 18:22

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=oup. .

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

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Oxford University Press is collaborating with JSTOR to digitize, preserve and extend access to Oxford
Economic Papers.

http://www.jstor.org
 THE IRRATIONALITY OF INTRANSITIVITY

By GORDON TULLOCK

A RECENT discussion of welfare economics 'summarizes the question of

transitivityas follows. There is not now conclusivereason to believe that
the assumptionof transitivityof preferencein individual choice is deeply
suspect. There are, however,groundsforbeing willingto make deep em-
pirical investigationsinto the validity of the assumption'.' At another
point the same author says: 'My conclusions indicate that we currently
possess no strongevidence to warrantdropping(the assumption of transi-
tivity).'2 Given the role that transitivityplays in much of the modern
literature,and in Rothenberg's book, his attitude towards it seems very
defensive.We do not have strongevidence against it; thereis no conclusive
reason to suspect deeplythe assumption. In sum, the evidence against
transitivityis not sufficientto condemn it, but Rothenberg is clearly
doubtful. In this he is more or less in accord with the bulk of current
opinion on the matter.
It is the purpose of this article to present a contraryview. It will be
argued that the assumption of transitivityis not particularlydoubtfulor
dubious. In view of the apparent suspicionof the assumptionby the bulk
of modernwelfare economiststhis mightseem to be a revolutionaryposi-
tion, but the practice of the same economistshas indicated relativelylittle
real doubt of the transitivenature of preferenceorderings. Article after
article has appeared in which this assumption is made, somethingwhich
would seem to indicate that the writershave at least some confidencein its
reliability. Further,the chains of reasoningbased upon this assumption
do not appear to have led to demonstrablyfalse conclusions about the
real world. This fact,in itself,is powerfulalthoughnot conclusiveevidence
forthe truthof the transitivityassumption.
The debate about the assumption of transitivityturns upon the inter-
pretation of certain real world experiences. If an individual is asked to
take his choice among a large number of pairs of alternatives,with each
alternativeappearing as one memberof quite a number of the pairs,3or a
large number of people are asked to judge between three alternatives
1 The Measurementof Social Welfare,Jerome Rothenberg (Prentice-Hall, Inc., Engle-
wood Cliffs,N.J.) 1961, p. 231.
2 Ibid., p. 19.

3 See ArnoldM. Rose, 'Astudy ofirrationaljudgements',The JournalofPolitical Economy,

Oct. 1957, pp. 394-402 for an example. In this case the choice among the alternatives
involved judgements of the severityof a crimeratherthan preference,but the transitivity
problem is the same.
402 THE IRRATIONALITY OF INTRANSITIVITY
grouped in a set of pairs,1it is highlylikelythat at least one case wherean
individual has chosen A over B, and B over C, then C over A will appear.
There are two possible explanations for this phenomenon. The first,
offeredby May,2 is that the preferenceschedules are actually intransitive.
The individual at one and the same time does preferA to B, B to C, and C
to A. The second, offeredby Rose,3 holds that the effectis an artifact
arising in the course of the experiment. This latter explanation is rather
complicated,since manythingscan go wrong. In the firstplace, the experi-
ment necessarilywill take time, and the subject may change his mind.
When he was asked to compare A and B he not only preferredA to B, he
also preferredA to C. The questions,however,forcedhim to think about
the generalsubject, and by the time the experimentergot around to asking
him to compare A with C he had changed his mind; he now prefersC to
A and to B. The apparent intransitivitymerelyreflectsthe change.
Anotherproblemwhichmay arise involves the factthat in many experi-
mentsofthis sortthe subject is requiredto choose betweenthe alternatives
even though he may be actually indifferent.This could easily lead to
apparent intransitivities.The subject may also be bored and simply fill
in the blanks on the experimentformhelter-skelterwithout really con-
sultinghis preferences.4They miayalso make a number of other 'errors'
in recordingtheirjudgements. The result of all of these factorsis that it
is quite possible foran experimentalsubject who has completelytransitive
preferencesto select A over B, B over C, and C over A.
Thus we have a well-establishedempiricalphenomenonand two possible
theoreticalexplanations forit. Rose in his article produced considerable
evidence that the second explanation was the true one, but the doubt on
the point expressed by Rothenberg would appear to justify another at-
tempt to demonstratethat intransitivityis not the explanation. For this
purpose we need only considerthree-elementintransitiveloops since any
larger order loop will necessarilycontain at least one three-elementloop
withinit. A demonstrationthat three-elementloops do not occur will also
serve to eliminatethe possibilityof largerintransitivecycles.
As a firststep I should like to presenta basic statementabout the real
world: A man confrontedwith a small (say 5) collectionof alternativeswill
either (a) preferone of them, (b) be indifferent among a subset of two or
more of the collectionbut preferany memberof that subset to the remain-
der of the collection,or (c) be indifferent among all the alternatives. It
seems unlikelythat anyone will seriouslyquestion this proposition,5but if
1 See Kenneth 0. May, 'Intransitivity, utility, and the aggregation of preferencepat-
terns', Econometrica,Jan. 1954, pp. 1-13. 2 Op. cit. pp. 7-13. 3 Op. cit. pp. 395-6.
4 Rose found that the number of apparent intransitivities in the judgements of each of
his subjects was inverselycorrelatedwith the time they took to complete the form(p. 399).
5 An ingenuous colleague has offered the criticismthat the individual in question might
 G. TULLOCK 403
someone does, it is easily tested empirically. Simple observation of a
cafeterialine would probably be the easiest 'experiment'. Anyone who
accepts this statementmust, if he is rational, agree that preferencesare
transitive.
The proof of transitivityis somewhat easier if we initiallyconfineour-
selves to a world in which indifferenceis impossible. In this restricted
world our basic statementshrinksto the simple formulathat a man will
preferone of a small group of alternatives. The proof of transitivityis a
simple example of reductioad absvurdur. If the individual is alleged to
preferA to B, B to C, and C to A, we can inquire which he would prefer
fromthe collectionof A, B, and C. Ex-hypothesi he must preferone, say
he prefersA to B or C. This, however,contradictsthe statementthat he
prefersC to A, and hence the alleged intransitivitymust be false.
There does not seem to have been any discussion of what a man with
intransitivepreferenceswould do if presented with a choice among all
elementsof the intransitiveloop simultaneously.He clearly would not be
able to choose,and he equally clearlycould not be called indifferentamong
the alternatives. The only pattern of behaviour consistentwith initransi-
tive preferenceswould be a closed loop in which the subject rapidly went
roundand roundhis preferential cycle. Kenneth 0. TMay'sarticle,'Intransi-
tivity,utility,and aggregationin preferencepatterns',' is surelythe locals
classicus forthe intransitivityhypothesis. In the course of the experiment
reported there he firstfound apparent intransitivitiesin the preference
orderingof 17 college students by paired comparisons. lie then presented
them with all three alternativessimultaneouslyand asked them to rank
them. They had no difficultyin doing so.2 This result is, of course,
completelyinconsistentwith the hypothesis that they had intransitive
preferenceschedules.
May, however, did not point out that the ability of his experimental
subjects to rank the alternatives proved that they had, at least at the
period withinthe experimentat which they made the ranking,transitive
preferenceschedules. He does not even discuss the point,but merelygoes
on to presenta mechanismto explain the intransitivitywhich he believes
his experimentsshowed. The simple explanation for May's results, of
course,is that the subjects changed their minds duringthe course of the
experimentand that the 'intransitivity'merelyreflectssuch changes. In
order to accept the intransitivityexplanation of the phenomenon it is
necessary to assume quite complex patterns. One would be that the
be totally ignorant of the alternatives and hence neitherhave preferencesnor be truly in-
different.This, of course,involves the question of exactly what we mean by 'indifference'.
Fortunately this semanticproblemcan be skipped. If the man in question has no preferences
with regard to the alternatives it is at least clear that he does not have intransitivepre-
ferences. 1 Econornetrica, Jan. 1954, pp. 1-13. 2 pp. 6-7.
404 THE IRRATIONALITY OF INTRANSITIVITY
subjects did not change theirminds duringthe period in which they were
asked the firstthreequestions,comparingA with B, B with C, and C with
A, but that they then all changed theirminds when they were asked the
fourthquestion which called forrankingall three alternatives.
There is an alternativeexplanation forMay's experimentalresultwhich
preserves intransitivityas conceivable. If we assume that his subjects'
preferenceswere not only transitive but also depended upon irrelevant
alternativesthenhis resultsare explicable. Theoreticallyit would be poss-
ible fora man to preferA to B, B to C, and C to A while at the same time
preferring A if presentedwith a collectionofA, B, and C, ifhis preferences
with respect to A and C were dependent upon whetheror not B was pre-
sent. If he prefersC to A when B is not presentbut prefersA to C when
B is present,then May's experimentalresults could be explained without
assuming that his subjects changed their minds duringthe course of the
experiment. Since the time of Occam, however,we have normallychosen
the simpler theory,and surelyin this case the hypothesisthat the sub-
jects had transitivepreferenceschedules but changed theirminds is the
simplerone.
So far, however, we have been discussing a simplifiedworld in which
indifference is absent. If we admit indifference the situationis more com-
plicated, although the same basic result will follow. Let us consider the
situationwhereit is alleged that a man prefersA to B, B to C, and C to A,
and that indifference is possible. If confrontedwith the collectionof A, B,
and C, then the individual might,in keeping with our basic statement,be
indifferent between A and B but preferboth to C. This would contradict
two links in the alleged intransitivitychain, the preferenceof A to B and
the preferenceof C to A. Hence, once again, the allegation of intransi-
tivity must be untrue if our basic statement is true. Finally, the indi-
vidual mightbe indifferent among the threealternatives. Spelled out, this
would involve not preferring A to B or C, not preferring B to A or C and
not preferringC to A or B. Clearly this contradictsall threelinks in the
alleged intransitivityloop.
If indifference is permitted,however,more complicatedtypes of intran-
sitivityare possible. An individual mightbe alleged to preferA to B and
B to C but to be indifferent between A and C. It would be tedious to go
throughthe various possibilities,but this would also be inconsistentwith
our basic statement. A more interestingcase would involve an individual
who was indifferent between A and B and also between B and C but who
preferredA to C. This is, of course,perfectlypossible.1 When confronted
I The intransitivityof the indifference
relation has been analysed by W. E. Armstrongin
a long series of articles beginningwith 'The determinatenessof the utility function', The
Economic Journal, Sept. 1939, pp. 453-67.
 G. TULLOCK 405

with the choice between A, B, and C, the individual will select A. Since
this would appear to contradicthis indifference between A and B, some
discussion seems in order.
Assume an individual wvho prefersblue to green and will always choose
the 'bluest' colour when given an opportunity.1He is, however, unable
to make infinitelyfinejudgements. The minimumchange of colour which
he is able to perceive is some small amount which we will denominate Y.
We present him with three colours,A, B, and C, in which A is the most
blue and C the most green,but with the differencebetween A and B less
than Y. The differencebetween B and C is also less than Y, but the dif-
ferencebetween A and C is greaterthan Y. If asked to choose which lie
preferswhen confrontedby these alternativesin pairs, our subject should
choose A to C, but be indifferent between A and B and B and C. If he is
presentedwith a choice between A, B, and C, he should have no difficulty
in selectingA because the presence of C makes it possible forhim to dis-
tinguishbetween A and B. He can see that A is bluer than C, but cannot
distinguishbetween B and eitherA or C. The realization that B must lie
betweenA and C easily follows. The same generalprocess could operate in
all cases whereindifference is foundto be intransitive. The addition of the
third alternative,by changing the informationavailable to the chooser,
increases the finenessof his discriminatorypowers. Hence his expressed
order is, in a sense, actually changed.
So farwe have been verylargelydiscussingthe logicofthe situation. Can
we design experimentswhich would serve to confirmor refuteour conclu-
sions ? The answerto thisquestionis yes,and henceour conclusionsare fully
' operational'. Speaking formyself,however,theresultsto be expectedfrom
the experimentsdiscussed below have been so obvious that they have re-
mained Gedankenexperimentsinstead of experimentsto be actually per-
formed.Perhaps one ofthereaderswillbe rathermoresceptical of my logic.
The firststep in any experimentwould be to give transitivitytests of
the sort developed by May and Rose to a number of subjects in order to
findapparent intransitivityloops. When such loops were found it would
be easy to distinguish true intransitivityfrom apparent intransitivity
arising fromexperimentaldifficultiesby simply confrontingthe subject
with all threeof the alternativeson the loop. If he is able to rank them,as
May's subjects were,it should be clear that the 'intransitivity'is only an
experimental artifact. This procedure would only be repeating May's
experiment,but his sample was rather small (only 17 subjects with in-
transitiveloops) and it mightbe argued that with a largersample a case
of genuine intransitivitymighthave turned up.
I Dr. James M. Buchanan suggestedthe followinganalysis. I am also indebted to him for
much other helpfulcomment.
406 THE IRRATIONALITY OF INTRANSITIVITY
May's experiment,however,did not cover anotherpossibility. His sub-
jects were required to choose between the alternatives, and could not
simplysay that theywere indifferent.It is possible that, with less restric-
tive conditionsa different result mighthave turned up. Perhaps some of
his subjects who had chosenA over B, B over C, and C over Al mighthave
expressed indifferencewhen confrontedwith all three alternatives as a
group. This result could be explained either in terms of experimental
problems,particularlythefactthat the subject may have changedhis mind,
or by a complicatedcombinationof true intransitivityand a most peculiar
formof interdependenceamong the alternatives. Experimental determi-
nation of whichof these was the correctexplanation would be easy. There
are a number of routines,but the simplestwould be to reduce the choice
once again. The subject could be told that since he was indifferent, the
experimenterwould make the choice. The experimentercould then say
that personally he disliked C and was indifferent between A and B and
hence would flipa coin to decide between them. The subject, if he has the
postulated intransitiveloop, should object and suggestthat A be marked
as the preferredalternative. If he does not, he clearly no longer prefers
A to B, and hence the apparent intransitivitymust be put down to a
change of mind rather than true intransitivity. This procedure can be
furtherelaborated to obtain even greaterassurance.
Note that the discussionhas involved no directassumptionofrationality
on the part of the personswho have preferenceschedules,only on the part
of the student consideringthem. If our basic statement about the real
world is accepted, then any intransitivepreferenceorderingmust involve
the simultaneous holding of directlycontrarypreferences. The subject
must preferA to B and B to A. Even a schizoid must split these two
preferencesbetween his two personalities. Our proof does not involve
the basic question of human rationality,but only bars one particulartype
of irrationality.

University
of Virginia
between A and C, if indifference
1 Or chosen A over B, B over C, but been indifferent
were permitted.