Rational Irrationality: A Framework for the Neoclassical-Behavioral Debate

Author(s): Bryan Caplan

Source: Eastern Economic Journal, Vol. 26, No. 2 (Spring, 2000), pp. 191-211
Published by: Palgrave Macmillan Journals
Stable URL: http://www.jstor.org/stable/40325987
Accessed: 21/11/2010 07:22

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=pal.

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.

Palgrave Macmillan Journals is collaborating with JSTOR to digitize, preserve and extend access to Eastern
Economic Journal.

http://www.jstor.org
 RATIONAL IRRATIONALITY:
A FRAMEWORK FOR THE NEOCLASSICAL-
BEHAVIORAL DEBATE
BryanCaplan
GeorgeMason University

INTRODUCTION

Economists havecharacterized beliefsas "rational," ifagentssatisfy Bayesian

probability axioms;or, more if
strongly,they also satisfy the rational expectations
assumption1 [Sheffrin,1996;Wittman, 1995].A diversebodyofexperimental evi-
denceshowsthatindividuals' beliefsdeviatefromthesestandardsofrationality
[Kahneman, Slovic,andTversky, 1982;Camerer, 1995;Rabin,1998].Criticsofthese
findings argue that the anomalies are suspect because financial incentives wereab-
-
sentorinadequate;peoplewouldbe morerational perhapsfullyrational - ifthe
monetary rewardswerelargeenough[Harrison, 1992;1990;1989;Wittman, 1995;
Friedman, 1998;Smith and Walker, 1993]. Defenders ofthe behavioral perspective
replythatanomaliesaregenerally robusttothiscriticism.
Thepurpose ofthecurrent articleis nottoresolvethiscontroversy, buttoprovide
a sharedframework fordebate.I present a modelof"rational irrationality" andshow
thatthemainpositions intheneoclassical-behavioral debateaboutbeliefs arespecial
casesofit.Inthismodel, is a goodlikeanyother,
irrationality andagentsoptimize by
trading Unlessotherwise
wealthforirrationality. stated, is interpreted
"irrationality"
as "deviations from rationalexpectations," ofwhichdeviation from Bayesianaxioms
is a subset.Thecentralassumption of the model of rational irrationalityis thatagents
perceive thepriceofirrationality without bias. On some level,they have rational
expectations abouttheconsequences ofirrationality, eventhough theytypically hold
a positive of
quantity irrationality in their consumption bundle.
Theupshotis thatitis notnecessary tosee theneoclassical andbehavioral ap-
proachesas twoirreconcilable paradigms. Theneoclassical-behavioral disputeover
beliefscaninsteadbe seenas a disagreement within "normal science"aboutparam-
etervalues;evenifresearchers cannotagreeabouttheirconclusions, atleasttheyare
askingthesamequestions. At the same time, the rational irrationality modeldoes
nottautologically definegenuineirrationality away:a keyfalsifiable implication of
themodelis that(compensated) demandforirrationality mustbedecreasing inprice.
Experimental findingsthatirrationality increases as monetary incentives risethus
differ inkindfrom otheranomalies and merit special attention and
[Hogarth Reder,
1987].

Bryan Caplan: DepartmentofEconomics,CenterforStudyofPublicChoice,GeorgeMasonUniver-

Drive,Fairfax,Virginia22030-4444.E-mail:bcaplan@gmu.edu
sity,4400 University

EasternEconomicJournal,Vol. 26, No. 2, Spring2000

191
192 EASTERN ECONOMIC JOURNAL

FIGURE 1
A TaxonomyofRationality

Thenextsection providesan elementarytaxonomy andirrational-

ofrationality
ityto clarify
thescopeofthecurrent Sectionthreepresents
paper'sinvestigations.
thesimplemodelofrationalirrationality.Sectionfouranalyzesfourspecialkindsof
indifference
wealth/irrationality curves- neoclassical, near-behav-
near-neoclassical,
-
ioral,andbehavioralandtheirimplications debate.
fortheneoclassical-behavioral
Thefifth sectionshowsthatseveralforms ofwealth-enhancing canalso
irrationality
beunderstood within therationalirrationality
framework. Sectionsixconcludesthe
paper.
A TAXONOMY OF RATIONALITY AND IRRATIONALITY

"Rationality,"in economic parlance,is equivocalin at leasttwoways.Notonly

has it beenascribedto bothpreferences andbeliefs,butin eachdomainthereis a
spectrum ofrationalitystandards, from leasttomostdemanding. In theinterestsof
clarity,this section an
provides elementary of
taxonomy rationality. By no means
intended to be exhaustive, its function is onlyto delimitthescopeoftherational
irrationalitymodel'sapplication.
Attheoutset, then,itis necessary todistinguishpreferences from beliefs[Mont-
gomery, 1996;Aumann, 1987]. An agent'spreferences indicate how he wouldbehave
what
(i.e., he wouldchoose to do) in all conceivablesituations.Anagent'sbeliefs indi-
catewhatprobability hewouldassigntoanyconceivable situation
actuallybeingthe
case. Figure1 illustrates thiscontrast byshowing preferences and beliefsas two
disjointsets.
Within thesetofpreferences, onecanthendrawthesub-set of"well-ordered and
stable"preferences. Thisis typically theweakesthurdleforpreferences toqualify as
"rational" [Becker,1962].Addingon morerestrictive - mostcommonly,
conditions
theexpected-utility axiomsofchoiceunderuncertainty - furthershrinks thesetof
rationalpreferences [Camerer, 1995].Ofcourse, bothmoreandlessdemanding ratio-
nalitycriteria canandhavebeenproposed [HarlessandCamerer, 1994].
 RATIONAL IRRATIONALITY 193

FIGURE 2
Indifference
Wealth/Irrationality Curves

IRRATIONALITY /"

Y* U- U-

WEALTH

Nextconsiderthesetofbeliefs, wherequitedifferent sensesof"rationality" ap-

theleastrestrictive
ply.Typically, definition requiresonlythatbeliefssatisfy the
familiarBayesianprobability axioms.Thesesetnolimitsonagents'priorprobabili-
in an
ties; principle, agent who is rationalin thisweaksensecouldbe grossly and
mistaken
systematically aboutwhattheworldis actuallylike.Imposing themore
rational
restrictive expectationsassumption rulesoutsuchcases,requiring thatagents
notonlysatisfytheBayesianaxiomsbutalsoholdunbiasedpriorprobabilities.
Theremainder ofthispaperdiscussestherationality ofbeliefsalone."Rational"
is usedinterchangeably with"satisfies therationalexpectations assumption." The
domainof"irrational" theshadedregionin Figure1,is theunionof(1) non-
beliefs,
and(2) Bayesianbeliefs
Bayesianbeliefs, thatfailtosatisfy therationed expectations
assumption.Thisis notmeanttosuggestthatthisis theonlyform of"irrationality
worth - ortoignore
considering theexperimental evidence inother
on"irrationality"
sensesoftheword.Rather,thecurrent paper focuses on violationsfrom rational
becausethisproblem
expectations is at onceimportant andtractable. Analyzing be-
haviorwithoutwell-ordered preferences butgivenwell-ordered
is quitedifficult; pref-
erences,analyzingirrationalbeliefsis - it will
be argued- a manageable task.2

RATIONAL IRRATIONALITY

Supposean agenthas well-defined preferences overbothpersonalwealthand

he
beliefs; caresabouthis wealth, butalsohas a "bliss wealth
belief9y*that(holding
constant) he wouldmostliketo believe[Akerlof and Dickens,1982;Akerlof,1989;
Caplan,1999a,1999b].Wealthshouldbeconceived inbroadtermstoincludenotjust
consumption value,butalsohumancapital,health,leisure,andsoon;in
andportfolio
one
fact, couldjustpartition all argumentsin theutilityfunctioninto"beliefs"
and
"everythingelse,"andusewealthas a synonym can
forthelatter.Thesepreferences
thenberepresented withindifference spaceas demon-
curvesinwealth/irrationality
stratedin Figure2. Wealthis onthex-axis,andtheabsolutevalueofthedeviation
from rationalexpectationsis onthey-axis;an agentwithrationalexpectations con-
194 EASTERNECONOMICJOURNAL

smeszeroy. The onlyunusualfeatureoftheseindifference curvesis thattheybend

backwardsat theblissbeliefy*; an agentis assumedtohave somespecificbeliefthat
he feelsattractedto,ratherthana contrariandesireto be as irrationalas possible.
As withutilitytheoryin general,one shouldnot read too muchpsychological
contentintothis choicemodel.A hungrypersonmightstate thathe buysfoodnot
becausehe "wants"itbutbecausehe "needs"it.A workercouldbalkat thesuggestion
thathe "revealshis preference forleisure"whenhe takesa breakaftera doubleshift.
An agentwhoholdssystematically biasedbeliefsthat"justseemobvious"tohimcan
be seen in the same light.
In psychologicalterms,cognitiveand motivational biases are different:themoti-
vationaldependon the emotions,butthe cognitivedo not [Nisbettand Ross, 1980].
But the rationalirrationality modeltreatsthemsymmetrically foranalyticalpur-
poses.3Still,the introspective experienceofthe effect ofincentivescoulddependon
the typeofthe bias. For a motivationalbias like over-confidence, an agent might
describehis responseto incentivesas: tryingto be reasonable,suppressinghis emo-
tions,or makingan effort to giveopposingargumentsa fairhearing.For cognitive
biases liketheavailabilitybias,4thesame agentmightsay thatincentivesmakehim
morelikelyto doubthis initialintuition,look at aggregateratherthan anecdotal
evidence,ask forexpertadvice,or spendextratimeresearchingthe question.5Mod-
elingcognitiveand motivationalbiases symmetrically does notimplythattheyare
introspectively equivalent.
In mostsituationsofpracticalinterest,systematically biasedbeliefshave a nega-
tiveimpacton personalwealth.Intuitively, an agentwhoconsistently respondsopti-
mallytothewaytheworldisn'talmostcertainly failstorespondoptimallytotheway
theworldis. As Nisbettand Rossputit,"Thecostsofwilly-nilly distortionsin percep-
tionare simplytoohighto make thema cure-allforthe disappointedor threatened
perceiver.In general,misperceptions makeus less able toremedythesituationsthat
threatenus or giveus pain than do accurateperceptions"[1980,234]. Suppose,for
example,thata doctorgenuinelythinksthat95 percentofthe peoplewhotestposi-
tivefora diseasehave it,whenin factmerely2 percentdo,as in thefamiliarbase rate
experiment[Casscells,Schoenberger, and Graboys,1978]. Earnestlyactingon this
biasedperceptionendangersthedoctor'scareerprospects, riskingmalpracticesuits,
dissatisfiedcustomers, and loss ofprofessional reputation.Similarly,an agentwhois
systematically over-confident, 80
mistaking percentprobability forperfectcertainty,
is likelyto makea widevarietyoflosingbets;a poorlycalibratedweatherforecaster
wouldbe likelyto lose his credibility and his audience[Lichtenstein, Fischhoff, and
Phillips, 1982].People who overestimate theirown and
abilities, accordingly bargain
formorethantheycan reasonablyexpectto get,mayreducetheiraverageearnings
[Babcockand Loewenstein,1997].
Agents'wealth/irrationality "budgetlines"therefore normallyhave the familiar
negativeslope as seen in Figure 3. A wealth/irrationalitybudgetline showsthecom-
binationsofwealthand irrationality that are feasible.For the sake ofconvenience
these expectedlosses are drawnas linear,i.e., proportional to the degreeofbias,
thoughthisneednotbe thecase. The budgetline'sintersection withthex-axisshows
an individual'swealthassuminghe strictly conforms to therationalexpectationsas-
sumption.The budgetline's intersection withthe y-axis,in contrast,indicatesthe
 RATIONALIRRATIONALITY 195

FIGURE 3
The Wealth/Irrationality
BudgetLine

IRRATIONALITY /"

' -
V
WEALTH

levelofrationalitynecessarytoactuallydrivean actor'swealthdowntozero.Ifwealth
is definedbroadly enough, thebudgetline'sintersection withthey-axiscouldbe in-
terpreted as thepointbeyond which more extreme wouldbe fatal.
irrationality
Rationallyirrationalagentschoosetheirutility-maximizingcombinationofwealth
andirrationality basedonan unbiased judgment aboutthetradeoffs. Thehigher the
private costof becomes,
irrationality the flatterthe budget linegets,andthe smaller
theoptimal purchase becomes
ofirrationality as Figure3 shows.Thecrucialassump-
tionis thatagentsonsomelevel have rational expectations abouttheslopeoftheir
wealth/irrationalitybudgetline;theyperceivetheimpactoftheirirrationality on
theirwealthwithout bias.Thisis whatdifferentiates from
rationalirrationality the
competing hypothesis ofunqualified
irrationality.Thepointhereis nottodefine un-
qualifiedirrationality away,butto lay thegroundwork forempiricalcomparison
byspelling outtherationalirrationalitymodel'sdetails.
Iftheprivate onwealthwerezero,thewealth/irrationality
impactofirrationality
budgetlinewouldbevertical, and an optimizing agentwouldalwayschoosehisbliss
beliefy*.Thispolarcaseplausibly arisesundermanycircumstances. Forexample, it
is pleasanttobelievethatyourjob is sociallybeneficial, butit unclearhowbiased
beliefsonthispointwouldbe costlyin termsofmaterialwealth[Klein,1994].The
sameappliesto manyreligious and politicalbeliefs:formostpeople,thereare no
practicalrepercussions ofdoubtingthetheory ofevolution thatone's
or believing
nationis the"bestintheworld"[Caplan,1999a].

THE NEOCLASSICAL-BEHAVIORAL CONTINUUM AND ITS PRACTICAL

SIGNIFICANCE

Curves
"Neoclassical"vs. "Behavioral"Indifference

In standard neoclassical
models,agentshavenopreferencesoverbeliefs.Beliefs
moredesirable
area toolforgetting commodities,notan endinthemselves."Neoclas-
sical"wealth/irrationality cantherefore
preferences indif-
as vertical
be represented
ference agentscareonlyabouttheirwealth,nottheiropinions.
curves:"neoclassical"
196 EASTERN ECONOMIC JOURNAL

FIGURE 4
"Behavioral"vs. "Neoclassical"Indifference
Curves
"Neoclassical"
Z Indifference
Curve
IRRATIONALITY

"Behavioral"
^ ~^r Indifference
Curve

Y* m"Near-Behavioral"
^v^^^ />^ Indifference
Curve

\ "Near-Neoclassical"
\^ Indifference
Curve

WEALTH

Theyworry abouttheiractualjobsafety, nothowsafetheybelieve their jobis [Akerlof

andDickens,1982].Similarly, they do not careabout self-image or how certainthey
are [Dickens, 1985;Lichtenstein, and
Fischhoff, Phillips,1982]. The implication is
thattheyhaverational expectationsifirrationalityhasanynegative impact onwealth,
however trivial.
In contrast, onbehavioral accounts, irrationalityandincentives are essentially
unrelated; people have an irreducible propensity to make and clingto systematic
errors.Cognitive anomaliesEireusuallyseenas inherently unresponsive to incen-
tives.Motivational anomaliestoo- though admittedly payoff-dependent intheory -
areoftenseenas payoff-independent in practice[Rabin,1998;Dickens,1986,1985;
Lichtenstein, and
Fischhoff, Phillips, 1982].AsPiattelli-Palmarini putsit,"Between
ourrationality andourcognitive pride,wewillchoosethelatter,andarewillingto
paythepriceforso doing"[1994,3].Thenaturalwaytodiagram thebehavioral posi-
tiononirrationality is withindifference curvesthatarehorizontal at theblissbelief
y*.Justas thepolar"neoclassical" agentis rational regardless ofhowweaktheincen-
tivesare,thepolar"behavioral" agentis irrational regardless ofhowstrong thein-
centivesareas canbe seenin Figure4.6
Criticsofbehavioral anomaliesoften maintain thatsubjectshaveinsufficient fi-
nancialmotivation to makethemthinkrationally[Harrison,1992; 1990; 1989;
Wittman, 1995;Smith,1991].Wittman forexampleremarks: "Mistakesare quite
likelyto occur when no one suffersfrom them (and are especially likelywhenthe
researcher is searching forwaysto tricktheunsuspecting subjects)...[G]iventhe
insignificantincentives inexperimental work,oneshouldbe surprised whenexperi-
mentalresultsdo confirm economic theory" [1995,41-2].Examining polarcasesof
neoclassicalandbehavioral indifferencecurvessuggests thatthiscriticism ofexperi-
mentalanomaliesis slightly offthemark.Withrationalrespondents, lackofincen-
tivesmerely explainsan increaseinthevarianceofbeliefs; itwouldnotaccountfor
theoccurrence ofthesystematic mistakes emphasized intheexperimental literature7
[Harrison, 1990,27-30;SmithandWalker, 1993].
 IRRATIONALITY
RATIONAL 197

Withinthe rationalirrationality framework, thereis a straightforward way to

reformulate thestandardneoclassicalcritique.Supposethatinsteadofperfectly ver-
tical wealth/irrationality indifference curves, actors have "near-neoclassical" indif-
ference curves:almostvertical,butcurvingoutwardsaroundy*> 0 as Figure4 shows.8
Undermostconditions, "near-neoclassical" actorscannotbe distinguished fromneo-
classicalactors:Neitheris irrationalifthe priceis non-trivial. The latentdifference
revealsitselfonlyas the priceofirrationality nears zero- -asit wouldin an experi-
mentwithoutmonetaryincentives.In thesechoiceenvironments, near-neoclassical
actorsconsumeappreciableamountsofirrationality, whereasperfectly neoclassical
actorsare as rationalas ever.Ifindifference curveswerereallyvertical,thenregard-
less oftheabsenceofincentives, criticsofbehavioraleconomicswouldlack an expla-
nationofsystematic experimental choiceanomalies.Butwith"near-neoclassical" pref-
erences, the main critique ofbehavioral economics is intelligible:people exhibitobvi-
ous irrationalbias in theabsenceofmaterialincentives, butat a smallpositiveprice
thisdisappears.
The presentpaperdoesnottrytoresolvethebehavioral-neoclassical controversy.
Whatthe rationalirrationality modelprovidesis a commonframework fordebate.
There are nottwo irreconcilable approaches, but two endpoints on a continuum of
possibilities.Ifthepolarneoclassicalpositionis empirically untenable,thepolar be-
havioralpositionis nottheonlyalternative.Economistswithneoclassicalpriors(i.e.,
thepriorjudgmentthatpeople'swealth/irrationality indifference curvesare vertical)
mightinsteadmakethemarginalmovetothe near-neoclassical view as experimental
anomaliesmultiply. Conversely, economists withbehavioralpriors(i.e.,thatpeople's
wealth/irrationality indifference curves are horizontal) neednotleap totheneoclassi-
if
cal view some evidence ofprice-sensitive irrationality accumulates.Theycouldin-
steadconcedeonlythatindifference curvesare "near-behavioral": theconsumption of
irrationality is pricesensitivebutpositiveevenat highprices.See Figure4.
Adoptingrationalirrationality as a framework can also illuminateHarrison's
[1989]claimthatit is more informative to check for deviationsfromrationalityin
payoffs ratherthanbeliefs.9Harrisonnotesthatin termsofbothstatisticaland eco-
nomicsignificance, experimental divergence fromfullrationalisty is oftenlarge,but
due to the flatnessof most payoff functions, divergence from income maximization is
usuallysmall.How can thisargumentbe related to the neoclassical-behavioral con-
tinuum?Supposethatan agentdeviatesfromrationality in neitherbeliefsnorpay-
offs.(See Table 1.) This is consistentwith bothneoclassicaland near-neoclassical
preferences: the agentmightbe rationalbecause he is congenitally rational,or be-
cause irrationality is costly.Similarly,ifan agent deviates from rationalityin both
beliefsand payoffs, thereis evidenceforbehavioralor near-behavioral preferences:
theagentmightbe irrationalbecausehe is congenitally irrational,orbecauseirratio-
nalityisn'tcostlyenough.Observeddeviationsfromrationality in beliefsalone,how-
ever,are consistentwitheverything exceptforpolar neoclassicalpreferences; only
neoclassicalpreferences are inconsistent withcostlessirrationality.10 Finally,note
thatthefourth boxin Table 1 is empty.Deviationsfromrationalbeliefsare whatdrag
payoffs belowtheirmaximumexpectedvalue. It is thusimpossibleto simultaneously
have rationalbeliefsand divergefromincomemaximization.
198 EASTERN ECONOMIC JOURNAL

TABLE 1
Beliefs,and Payoffs
Rational Irrationality,

Deviationfromrationality
in... is consistent
with... Beliefs

Payoff Yes No

Yes near-behavioral, -
behavioral

No near-
near-neoclassical, neoclassical,
behavioral,behavioral near-neoclassical

Practical Significance

Evenifbehavioral anomaliesdisappearin thefaceoffinancial find-

incentives,
ings ofbiased mistakes matter. As Smith and Walker observe, "[T]here areboth low-
stakeandhigh-stake economic decisions in life,andall are ofinterest" [1993,249].
Theformer candrawnear-neoclassical agentstodeviatefrom rationalexpectations;
thelattermayprovoke near-behavioralagentstosetasidetheirbiases[Caplan,1999a;
Camerer, and
1987;Frey Eichenberger, 1994].In addition, sometimes themarginal
costofirrationality is loweventhough thetotalstakesinvolved arehigh;as Russell
andThalernote,"themoreefficient themarket, thelessdiscipline themarketpro-
vides.In a fullyarbitraged market, all goods(assets)yieldthesamecharacteristics
perdollar(returns), thusindividuals can choosein anymannerwithout penalty"
[1985,1081].
Thepractical significance ofanydeviation from neoclassical preferencescannot
be confined tolab experiments becausethisis onlyoneformoflowstakedecision-
making[Kirchgssner and Pommerehne, 1993].Considersomeothers:theprivate
costofsystematic misestimates ofinflation is smalloversomeranges[Akerlof and
Yellen,1985].Contingent valuationsurveys overestimate thewillingness topayfor
environmental amenities [Harrison andKristrm, 1995].Moregenerally, surveysabout
issueswithexpressive valuemeritsuspicion duetothenegligible private ofbi-
cost
asedresponses [Brennan andLomasky, 1993;BoulierandGoldfarb, 1998].Aboveall
else,ifrationalexpectations failin experiments without adequateincentives, one
wouldexpectirrationality toplayanimportant roleindemocratic electionsduetothe
trivialprivateconsequences ofa vote[Caplan,1999b;Akerlof, 1989].
Wittman specifically triestocontrast hypothetical surveys andelections: "When
errorsinvolvelittlecost(forexample,answering a surveyquestionincorrectly in a
cognitive-psychology experiment), then littlecognitionwill be employed. When er-
rorsinvolvegreatcost(forexample, purchasing unreliable equipment forthemili-
tary),thengreater cognition willbeinvolved..." [1995,59].Considering theenormous
unlikelihood thata votewillchangean electoral outcome, itis ratherthesimilarity of
surveys and elections that stands out. Wittman's disanalogy conflates private and
socialcosts:inan election, errorsmayhavelargesocialcosts,eventhough theyrarely
 RATIONAL
IRRATIONALITY 199

will in laboratorysurveys.But the privateincentivestructureofsurveysand elec-

tionsis identical:in bothcases,theprivatecostoferrorsis effectively
zero [Brennan
and Lomasky,1993,38-41].
Similarly,even if behavioralanomaliesare onlyslightlysensitiveto material
incentives, it makesa practicaldifference.Near-behavioralagentswillmake a wide
of
range systematic mistakes, but iftheyencounterthesame situationrepeatedlythe
of
price irrationality may at last inducethem to correct
theirerrors.Thereare also
self-selection to consider:ifsomepeopleare closerto thebehavioralpolethan
effects
others,thenpeopleunwillingto abandontheirbiases evenat highpricesmayswitch
to activitieswheretheirbiases makeless difference. Thismakesthelong-runsupply
ofrationalitymoreelasticthan evidencefromexperiments withrandomlyselected
subjectsmightlead one to expect[Camerer,1987].

EmpiricalImplementation

Howexactly wouldempirical workers Twoapproaches

usethisanalysis? suggest
themselves.Thefirstis totestwhether (andwhen)rationalirrationalitycanbe re-
jectedin favorofa modelofunqualified The secondis to empirically
irrationality.
calibrate ofirrationality
theelasticity toincentives
withrespect undervariouscondi-
tions.

Can therational irrationality modelbe rejected?Therationalirrationality

modelis consistent witha muchbroaderrangeofobservations thanstandardneo-
classicalmodels, butitcannotbereconciled witha specialclassofanomalies (income
effectsaside).Rationalirrationalityleavesopenthepossibility thatbiasesfailtode-
forrationality
creaseas incentives intensify.Butitrulesoutthepossibility thatstronger
incentives forrationalityactuallymakesubjects lessrational.Thisis noteworthy be-
causeseveralextantsourcesreport findingsalongtheselines[Einhorn andHogarth,
1987;Camerer, 1995].Einhorn andHogarth rationalizetheseresultsas follows:"Per-
formance, however, dependsonbothcognition and motivation.Thus, ifincentivesize
canbe thought ofas analogoustothespeedwithwhichonetravelsin a givendirec-
tion,cognition determines thedirection. Therefore,ifincentivesarehighbutcogni-
one
tionisfaulty, gets to thewrong place faster."11
[1987,63] More straightforwardly,
onecouldsaythatindividuals simply perceivethewrong wealth/irrationalitybudget
constraint.12
Whattherational framework
irrationality highlightsis thatsuchfindings arenot
onemoreanomalyamongmany, but a to
challenge rationality on the deepestlevel.
As such,theydeservemorescrutiny: Harrison[1992,1990]findsthatthemarginal
incentives in manywell-known experiments havebeena fewpenniesorless,and
Thaleracknowledges thatthesefindings havenotbeenreplicated "atverylargestakes"
[1987,96] .
One recentand noteworthy studythatshedslighton thisquestionis Daniel
Friedman's [1998]studyofMonty Hall's3-doorsparadox.Friedman beginsbyexperi-
mentally the
demonstrating paradox'sstrength: subjectsswitch about 30 percentof
thetime,eventhough 100
theywouldswitch percent givenunbiased probabilityesti-
200 EASTERNECONOMICJOURNAL

mates.He thenestimateslinearprobability modelswith"switch"(=1 ifthe subject

switched, 0 otherwise)as thedependentvariable,and teststhesensitivity ofthebasic
resultsto modifications ofthe experiment. One variantis to movefromregularto
"intense"incentives, whichincreasesthe ex ante marginalcostofbiased estimation
five-fold- from$.10/turn to$.50/turn.13 Friedmanuses twodifferent variablesto cap-
turetheimpact:Intense,a dummyvariableequal to 1 ifa subjecthad "intense"incen-
tives,and Switchbonus, a continuous variableequal to(cumulative earningsforswitch-
ing - cumulative for
earnings notswitching).14
Friedmanreportsthatin a multipleregression, thecoefficient on Intenseis actu-
allynegative.Its discreteeffect is to reducethe"switch"percentageby9 percentage-
points,a declinesignificant at the 5 percentlevel. [Friedman,1998,940] However,
thecoefficient on thecontinuousmeasureSwitchbonus is positiveand significant. In
combination, thecoefficients on Intenseand Switchbonus indicatethatat least after
fiveturns,intenseincentivesyieldbetterperformance than regularincentives.As
Friedmanexplains,"SinceSwitchbonusalreadycapturesthemostrelevantpositive
aspectofintenseincentives,the significantly negativeIntensecoefficient shouldbe
thoughtofas measuringa residualimpact"[ibid.,944].
Fromthe standpointofthe rationalirrationality model,then,Friedman'sfind-
ings are not anomalous, even he
though presentsstrongevidencethatconsiderable
irrationality exists.The rationalirrationality modelmovesthe analyticalspotlight
fromthe"anomaly"in theusual sense oftheterm(violationoffullrationality) tothe
anomalousresponseofrationalityto strongermaterialincentives.Now supposein-
steadthatFriedmanhad foundthatthecoefficients onbothIntenseand Switchbonus
werenegative.This woulddefinitely countas empiricalevidenceagainstthe unlim-
ited applicabilityofthe rationalirrationality model,sinceit wouldimplythat con-
of
sumption irrationality riseswhen the of
price irrationality increases.
One potentially important caveatforempiricalresearcherstotakeintoaccountis
theshort-run versusthelong-runelasticityofrationality. SinceFriedman'sdiscrete
and continuousmeasuresofincentiveshave different signs,strictlyspeaking,his
resultspredictthatat firstsubjectsbecomeless rationalwhenthe stakes increase.
The positivecontinuouseffect ofincentivesovertakesthe negativediscreteeffect by
turnsix. In effect, then,the short-run responseofrationality to incentivesin the 3-
doorsgamemaybe negativeeventhoughthelong-runresponseis positive.
Divergencebetweenthe short-and long-runelasticityofrationalitywouldpre-
sumablybe farmorepronouncedifsubjectshad additionaltimeduringwhichto ac-
cess a libraryor the internet.For example,in an alternativeexperimentaldesign,
subjectsmightplayforan hour,receivean hourbreakof"freetime,"thenreturnto
playforone additionalhour.Wouldhigherstakes stillhave a temporarily perverse
effect?Or as incentivesrosewouldplayersbe increasingly inclinedto skiplunchand
searchout a correctexplanation[Freyand Eichenberger, 1994]?15

Calibratingtheelasticityofirrationality. A secondwayforempirical research-

erstousetherationalirrationalityframeworkis tocalibrate
theelasticity
ofirratio-
nalitywithrespect
toincentives. Assuming thatirrationality
declinesas thepriceof
it
rises,
irrationality is still
vitalto know"how much, under whatconditions?"Re-
 RATIONAL
IRRATIONALITY 201

turningtoFriedman,onecan roughlycalculatetheelasticityofirrationality in differ-

entconditions.
Extrapolatingfromthe probitversionofhis results,Friedmaninfersthatwith
intenseincentives, subjectswouldbe rational90 percentofthetimeafter57 turns.16
Followingthe same procedure[Friedman,1998,940] I calculatethatwithstandard
incentives,thecorresponding rateafter57 turnswouldbe only35 percent. Withnon-
intenseincentives,subjectswouldreachthe same level ofrationalityonlyafteran
estimated257 turns.In otherwords,after57 turnsofplay,increasingincentivesbya
factorof5 increasesthe percentageofrationalresponsesby a factorof2.6, and re-
ducesthenumberofturnsrequiredfor90 percentrationality bya factorof4.5. Focus-
ing on the moreimmediate effectof incentives,Friedman's resultsalso implythat
after10turns,theswitchrategivenintenseand non-intense incentiveswillbe 25 and
22 percentrespectively. The impliedelasticitystillhas the"right"sign.Yet 10 turns
out,magnifying theincentives fivetimesincreasesthepercentage ofrationalresponses
byonly14 percent.
Overall,then,Friedman'sresultssuggestthatthe short-run elasticityofirratio-
nality withrespect to incentivesis fairlylow,butthe long-runelasticityis quitehigh.
the
The reasonfordiscussing study,though, is notits specificconclusions,butrather
to illustratehowthe rationalirrationality modelmightredirectempiricalresearch.
Withintherationalirrationality framework, searchingfordeviationsfromfullratio-
nality remains useful. In fact,placingirrationality withina consistenttheoretical
framework wouldtendtodefusepurelymethodological objectionsagainstanomalous
experimental findings. The rationalirrationalityframework onlydemandsthatevi-
denceofanomaliesbe accompaniedby additionalworkon theirincentive-elasticity
underdifferent conditions.

INCENTIVES FOR IRRATIONALITY: POSITIVE ILLUSIONS, MORAL

CONSTRAINTS, AND SOCIAL PRESSURE

A simpleintuition liesbehindthenegative slopeofthewealth/irrationality bud-

getline:ifpeoplerespond optimallyto theway the world is not,then except by pure
chancetheyfailto respondoptimally to thewaytheworldis. Butthereare three
notableexceptions tothisprinciple.Thefirstcase is belief-dependent performance,
wherean agent'sbeliefsare actuallyan argument in his production function; the
second, socialpressure to deviatefrom rationality;the third, "self-serving bias" to
circumvent fixedmoralconstraints. In each of thesecases,agentsactually face an
upward-sloping budgetlineshownin Figure5. As in Figure3,
wealth/irrationality
thebudget lineis drawnas a straightlineforillustrativepurposes, butitcouldeasily
be non-linear.Thesteepertheline,themorewealthan agentforgoes ifherefuses to
deludehimself. In effect,
agents facea trade-off between a good(wealth), and a bad
inexcessofy*).
(irrationality Togetmoreofone,theymustenduremoreoftheother,
muchlikeonemight toleratemorepollution inordertoproduceadditional steel.
the
Notethat positively-sloped budget linereaches a tangency only at some point
moreirrational thany*.Evenifa person likesbeingrational (y*=0),thepositive slope
ofthebudgetlineprovides an incentiveforsystematicbiasinequilibrium. Asbefore,
202 EASTERN ECONOMIC JOURNAL

FIGURE 5
Social Pressure,and Moral Constraints
Performance,
Belief-Dependent

mRATTONALTIY j^ BudgetLineCreated
/ byBelief-Dependent
/7 Social
Performace,
y/ I orMoral
Pressure,
S Constraints
Y* -

WEALTH

theresponsiveness variesalongtheneoclassical-behavioral
ofbeliefsto incentives
continuum. Thecloseryourindifferencecurvesaretotheneoclassicalextreme, the
moreinclined youare tomanipulate yourviewstoyourmaterialadvantage.More
neoclassical
agentsreadily"psyche themselvesout"to enhancetheirperformance,
rationalize
to avoidbinding moralconstraints,and changetheiropinionsas social
fashionschange.Thedelusions pole,incontrast,
ofthoseclosertothebehavioral are
stable;theirviewsmaybebiased,buttheyarenotconveniently
relatively interpret-
ingrealitytoenhancetheirprospects.

Belief-Dependent
Performance

Supposethatsomeofan agent'sbeliefs arearguments inthatagent'sproduction

function.Theclearestexampleis theplaceboeffect, whereindividuals' healthreally
improves becausetheyfalsely believetheyarereceiving medicaltreatment.
effective
Taylorinterestingly notesthat"Placeboeffects are so powerful thatno newtreat-
mentordrugcanbe approved forgeneraluse in thepracticeofmedicine unlessits
effectivenesshasbeenevaluatedagainstthatofa placebo"[1989,118].Butthereare
manynon-medical examples too.Psychological
studieshavealsofound thatfrequently
individualswithrealistic- ratherthanover-optimistic- probabilityassessments are
morelikelytobe depressed. Withtheirmoodlesspositive, theyare often objectively
less successfulas a result[Camerer, 1995;Taylor,1989].Similarly, a studentwho
believesheis likelytodowellona testfeelscalmandconfident, andinconsequence
probably actuallydoesbetter.
Thereis alsoevidence thatthe"illusionofcontrol"is onnetwealth-enhancing.17
Stressimpairsperformance, andeventsbelievedtobe uncontrollable are normally
morestressful thanequallyunpleasant controllableones.Theproduction functionof
a personwiththe"illusionofcontrol" aboutan unpleasanteventis therefore less
impaired bystressthantheproduction functionofsomeone immune totheillusion.
AsTaylorexplains:
 RATIONAL
IRRATIONALITY 203

Whenan event- evena painfulorupsettingevent- is perceivedtobe

underpersonalcontrol, theeventdoesnotproduceas muchstressas
one perceivedto be uncontrollable.Whenpeople experienceuncon-
trollablestressfulevents,theyreact morenegatively.Their physi-
ological systemsresponddramatically,leading to an increase in
adrenalinsecretion, whichinturnhas accompanying sideeffectssuch
as the poundingofthe heart,nervousness,and sweatiness.Psycho-
logicaldistressis greaterwhena stressfuleventcannotbe controlled.
People who are understressthat theycannotcontrolalso perform
morepoorlyon othertasks. Concentration is limited,so it may be
to attendproperly
difficult to whattheyneed to do. [1989,75-6]

In termsofthe rationalirrationality model,agentswithany ofthese formsof

belief-dependent performance actuallyfacean upward-sloping budgetline- at least
oversomerangeas in Figure5. The flattertheslopeofthepriceline,thegreaterthe
impactofirrationalbeliefsonperformance. A personwithrationalexpectations about
theprobability ofsuccesscan rationallyexpecttobe less successfuland thereforeless
wealthy than the systematically overconfident.A personwith rationalexpectations
about the controllabilityofstressfuleventscan rationallyexpecta higherlevel of
stressand a lowerlevelofwealththana personsubjecttotheillusionofcontrol.As a
result,evenagentswithno intrinsictasteforirrationality (y*=0)on thesetopicswill
decidetobe irrationalin equilibrium.Perhaps thisevenexplainswhyoverconfidence,
theillusionofcontrol,and some otherbiases predominate empirically[Lichtenstein,
Fischhoff, and Phillips,1982; Langer,1982; Taylor,1989]: If over-ratingyourpros-
pectsactuallyimprovesyourprospects, evolutionary forcestend to weed out tastes
not
forrationality, irrationality [Cosmidesand Tooby,1996, 64-68; Waldman, 1994].

Social Pressure

Belief-dependent performance can mattereven fora solitarygame againstna-

ture,butthescopeforwealth-enhancing expandsin multi-player
irrationality games.
The powerofpositivethinkingaside, natureprovidesno incentiveto be irrational,
but otherhumansoftendo. Supposepeopleare morelikelyto hire,befriend, praise,
and givepoliticalpowerto those who share theirbeliefs,and more to
likely boycott,
shun,denounce,and persecutethosewho disagreewiththem[Kuran,1995; Klein,
1994;Becker,1974].In thisenvironment, less rationalitycouldeasilymakeyoumore
wealthy,giving rise to anotherinstance ofthe upward-sloping wealth/irrationality
budgetline.18The moreintensethe socialpressure(positive negative),theflatter
or
the slopeoftheline [Freyand Eichenberger, 1989].
Whatis remarkableis thatusingsocialpressure,a nucleusoftruebelieverswho
supportsomeirrationalview(thatis, with;y*>0)can generatepositively-sloped bud-
get lines foreveryone.Facing social pressure to share in popularillusions,people
withno intrinsicinclinationtowardsirrationality (thatis, with;y*=0)mayneverthe-
less opttobe irrational.A testsubjectmayhave no strongfeelingsabouttherelative
lengthoftwosticks.Whensurrounded byconfederates whoall claimthattheshorter
204 EASTERNECONOMICJOURNAL

stickis longer,though,test subjectsfrequentlysay- and perhaps convincethem-

-
selves thattheyagreewiththepopularposition.On a granderscale,a smallclique
ofcommitted Communists orbelieversinthecastesystemmaymakedissentso costly
thattheirdoctrinewinswidespreadacceptance[Kuran,1995]. The "truebelievers"
putpressureon theirclose associatesto sharetheirbeliefs,these associatesin turn
pressureothers,and irrationality
consequently "tricklesdown"fromactiviststo the
generalpublic.

Moral Constraintsand Self-Serving

Bias

Rabin[1995]distinguishes between"moralpreferences" and "moralconstraints."

Moralpreferences are in theutilityfunction;themoreintenselyan individualcares
about a moralgoal, the morehe wants at a givenprice.Moral constraints, on the
otherhand,shrinkone'spermissiblebudgetset. Peoplepursuemoralgoals because
theywantto,whereastheyobeymoralconstraints becausetheyfeeltheymust.Rabin
explainsthat,"Forgivenbeliefs,thereisn'ta bigbehavioraldistinction betweenthese
twomodelsofmorality," butmatterschange"whenpeoplecan manipulatetheirbe-
liefs"[1995,4]. Ifpeoplehave moralpreferences, mistakenbeliefsmisdirecttheirsin-
cereefforts to do good,makingthemsubjectively worseoff.But iftheyinsteadface
moralconstraints, mistakenpositivebeliefscan expandtheirbudgetset and make
themsubjectively betteroff.
For example,suppose an agentchoosesbetweentwo consumptivegoods,milk
and whiskey,shownin Figure6. The "amoral"budgetline showsthatcombinations
theagentcan financially afford;thetangencyshowsthemost-preferred bundle,mo-
ralityaside. Ifavoidingalcoholwerean internalizedmoralpreference, thiswouldbe
modeledbyshifting theagent'sindifference curve.The contrarypossibility,however,
is thatan agentavoidsalcoholbecause ofa perceivedexternalmoralconstraintthat
forbidsor discouragesits consumption.Unlikethe internalizedmoralpreference,
sucha moralconstraint wouldshifttheeffectivebudgetlinein alongthex-axis,leav-
ing the agent on a lower indifferencecurve.Assumingtheagentrealizesthetruealco-
hol contentofwhiskey, he has to choosewithinthe"moral"budgetset insteadofthe
"amoral"budgetset. The knowledgeassumptionis critical:The morethe agentun-
derestimates itsalcoholcontent, theless themoralconstraintmatters,andthehigher
his utility.
How can agentsmake moralconstraintsless binding?Rabin [1995] focuseson
beliefmanipulationthroughselectiveacquisitionofinformation, butbiased informa-
tionprocessingcan in principlebe just as effective. In fact,if agents choosetheir
normativeas well as positivebeliefs,theimpactofmoralconstraintsdisappearsen-
tirely;the"moralbudgetline"disappears,and theyselecta pointon a higherindiffer-
encecurvetangenttothe"amoralbudgetline."Ifbeliefsaboutmoralconstraints get
in thewayofself-interest, they couldjust abandon theirbeliefsin therelevant bind-
ing moralconstraints.Suppose,however,thatpeople'snormativebeliefsare fixed,
butpositivebeliefsare not.Thenan agentwhomaximizesutilitysubjectto personal
moralconstraints mightbe betteroffwithunreasonablepositivebeliefs,or"self-serv-
ing bias"19[Dahl and Ransom,1999; Babcockand Loewenstein,1997]. Facingwhat
 RATIONALIRRATIONALITY 205

FIGURE 6
MoralConstraints
MILK

Morally^^^V
Constrained^S.^^^. Morally
Optimum V^^ #^* Unconstrained
Optimum

"Moral" Budget Line -^^ ^s>^^ < "Amoral" Budget Line

WHISKEY

appearstobe a binding moralconstraint, he rationalizes hiswayaroundit,"cheat-

ing"onhismoralbeliefs bymanipulating hisnon-moral beliefs. Themoreherational-
izesthegreater hisutilityfrommaterialgoods.Forexample, thepersoninFigure6
might conveniently underestimate thealcoholcontent ofwhiskey, moving the"moral
budgetline"closertothe"amoralbudgetline."
NotethatFigures5 and6 illustrate twoquitedifferent tradeoffs.Figure6 shows
an ordinary tradeoffbetween twogoods,complicated onlyby the factthatperceived
moralconstraints shiftthebudget linein. Figure endogenizes budgetlinefor
5 the
Figure6; itshowsthatas positive beliefs
connected tomoralconstraints become more
biased,perceived moral constraints
relax, making a higher level ofwealth feasible.
Theonlythingholding backunlimited rationalization is a distasteforirrationality.
withchoiceoverbeliefs
In effect, thereis an additional margin (wealth/irrationality)
tooptimize along. Theindividualwith fixed normative beliefs and flexibledescriptive
thusfacesa positively-sloped
beliefs wealth/irrationality budget line.
Themoreonerousthemoralconstraint is,theflatter is thebudgetline'sslope;
thepayoff increaseswiththeabsolutedivergence
to rationalization betweenyour
morally-unconstrained utility-maximizing action and your morally-constrainedutil-
ity-maximizing The
action. Rabin
factors [1995] identifiesas mitigating self-serving
bias,suchas "salience "moral
injection," dogmatism," and"moral priming" makeequi-
libriumbeliefs morerationalbyflattening wealth/irrationality indifferencecurves.

EmpiricalImplementation

Howcouldempirical use thisanalysisofwealth-enhancing

researchers irratio-
nality?Therearetwodimensionsworth considering. The is
first to distin-
empirically
guishbetween beneficial and
irrationality low-cost The
irrationality. secondis toesti-
ofwealth-enhancing
matetheincentive-elasticity irrationality. discusses
Thissection
thedirection work
empirical along lines
these might take, using anoma-
overconfidence
lies[Lichtenstein, and
Fischhoff, Phillips, to
1982] illustrate.
206 EASTERN ECONOMIC JOURNAL

Beneficialirrationalityversuslow-costirrationality.Somediscussions of
wealth-enhancing irrationality[Taylor,1989] can beunclearwhether thosewith "posi-
tiveillusions" are(1) betteroffon balancebecausetheyaremuchhappierbutonly
slightlylessobjectively successful, offbecausetheyaremoreobjectively
or(2) better
successfulas wellas happier. Therational modelformally
irrationality distinguishes
thesetwocases:hypothesis (1) is just thestandardcase shownin Figure3, while
hypothesis (2) is thespecialcase shownin Figure5. Oneusefultaskforempirical
researchers wouldbe to determine whichofthesepossibilitiesis actuallythecase.
Forexample,doesoverestimating answergeneralinterest
one'sabilitytocorrectly
questions actually increase
the fraction ofcorrect byreducing
responses stress?
Onewaytogetat thisquestion, whichtomyknowledge has notbeentried,is to
combine standardtestsofgenerati knowledge withtwosortsofincentives: a reward
fora subject'stotalnumber ofcorrect responses, anda rewardforsubjectsto accu-
ratelyguessthenumber ofquestions oneanswered correctly.
Designating thenum-
berofcorrect answersas C,thesubject's believednumber ofcorrectanswers C, and
rewards/prices pxandp2,onesuchcompensation formulawouldbe: Payoff pxC-
=
-
p2(C C)2.Ifhypothesis (1) is correct,thenraising p2shouldreduceC , butleaveC
if
unchanged;hypothesis is then
(2) correct, raising p2shouldreducebothCandC.

Calibratingthe elasticityof irrationality.Forcases wherehypothesis (2)

the
applies, next task for empirical study is to calibrate the of
incentive-elasticity
The is
irrationality. procedureessentially the same as thatin the 3-doorsexperiment:
varybothincentives andotherconditions andsee howtheexpected degreeofratio-
nality
changes.Suppose, forexample, thatoneadministers a testof100general knowl-
edgequestions.One could estimate each subject's "blissbelief forC bysettingboth
pxandp2equal to0. The rest of indifference
subjects' curves could then bemappedout
byvarying/^ andp2.Forexample, =
onecouldset(pvp2) ($.20,$.00)foronegroupof
andset(pvp2)= ($.30,$.00)forthoseremaining.
subjects, IftheaveragevaluesofV
andC forthefirstgroupwere55 and58,andtheaveragevaluesofC andC forthe
secondgroupwere60 and 65,onecouldestimatethatsubjectswerewillingto in-
creasetheirbiasfrom 5 percent to8 percent inexchange foran extra$1.50worthof
income.
Themostunusualfacetofthisproblem is thattheindifference curvesbecome
backward-bending *.
afteryMoreover, while this show
paper'sfigures irrationality as
symmetric around^*, thisis notnecessarily thewaywealth/irrationality indifference
curvesactuallylook.Perhapsbeliefsare moremalleable(indifference curvesare
steeper)beyondy* than they arebetween 0 and y*,indicating agentsaremore
that
stubbornlycommitted to their wealth-impairing irrationality thantheyaretotheir
wealth-enhancingirrationality.Thus,calibrating theelasticity ofirrationality
requires
empiricalworkonbothcases.

CONCLUSION

Thecentralmessageofthispaperis thattheneoclassical-behavioral
debatecan
be analyzedas an elasticity "Pure
question. neoclassical" indif-
wealth/irrationality
 RATIONAL
IRRATIONALITY 207

ferencecurvesare vertical;such agentscare onlyabout wealth,not beliefs."Pure

behaviorial"indifference curvesare horizontalat y*; irrationality is constantand
unrelatedto materialincentives.Both endpointsand all ofthe intermediatecases
betweenthemfitcoherently intotherationalirrationalitymodel.Moreover, thisframe-
workis flexibleenoughto analyzediverseviolationsofrationalexpectations.Games
of"manversusnature"wheremoreirrationality has a negativeimpacton private
wealthare themostobvious.Butit can also be easilyappliedtogamesof"manversus
himself(forexamplebelief-dependent performance and fixedmoralconstraints) and
"manversussociety"(forexamplesocial pressure)wheremoreirrationality makes
youmateriallybetteroff.
Whilethecentralpurposeofthispaperis simplytoprovidea commonframework
fordebate,it also has two key substantivefindings.The firstis that experiments
whereincreasingincentivesforrationality makespeopleless rationalpose a unique
challenge.Such anomalies merit additional attentionbecause theyare as inconsis-
tentwithrationalirrationality as they are with rationality.Perhaps the long-run
ofrationalityis positiveeven ifthe short-run
incentive-elasticity responseis nega-
tive.Higherstakesin a rigidsettinghave beenfoundto makebiases worse[Thaler,
1987]. But what abouthigherstakes combinedwithsufficient timeto do morere-
search,ask experts,or experiment?
The secondconclusionis thateventhe staunchestcriticsofthe behavioralfind-
ingsmustconcedethatmanypeople'spreferences are notneoclassical,butonlynear-
neoclassical.This seeminglysmallconcessionhas strongimplications.Evolutionary
claimsaboutthepossiblescopeofirrationality mustdistinguish betweencases where
is privatelycostly,privatecostless,or evenprivatelybeneficial;onlyin
irrationality
thefirstcase are evolutionaryargumentsforperfectly neoclassicalpreferencesat all
compelling.Ifnear-neoclassical are
preferences widespread, then behavioral economics
providesa distortedpictureofmarketbehavior20, but ipsofactooffersa compelling
accountofmuchnon-market behavior.Mostnotably,real-worldagentsusuallylack
financialincentivesto be rationalin politics[Akerlof,1989; Caplan, 1999a, 1999b].
The implicationforexperimentaldesign:to simulate market conditions,you must
use privatematerialrewards,butto simulatepoliticalconditions, youmustnot.21

NOTES

For discussionand usefulsuggestionsI wouldlike to thankDon Boudreaux,TylerCowen,Pete

Boettke,JimSchneider, Geoffrey Brennan,Bill Dougan,Bill Dickens,MitchMitchell,Ed Lopez,J.C.
Bradbury,ToddZywicki, David Bernstein, RobinHanson,Dan Klein,AlexTabarrok,NickyTynan,
TimurKuran,RonHeiner,RogerCongleton, Fab Rojas,David Levy,KennethKoford, Daniel Fried-
man,two anonymousreferees,seminarparticipantsat GeorgeMason,participantsat the Public
ChoiceOutreachseminarand the PublicChoiceSocietymeetings,and membersofmyArmchair
Economists'listserv.Gisle Silva and Eric Cramptonprovidedexcellentresearchassistance.The
standarddisclaimerapplies.

1. theyhave characterized
In contrast, preferences and stable;or,
as "rational"iftheyare well-ordered
morestrongly, axiomsofchoiceunderuncertainty
iftheyalso satisfythe expected-utility [Becker,
1962; Camerer,1995]. The currentpaper focusessolelyon the rationalityofbeliefs.The second
sectionofthispaperpresentsa basic taxonomy ofirrationality.
208 EASTERN ECONOMIC JOURNAL

2. As an anonymous refereeputsit,"Indeed,ifonethinksofrationalexpectations as havinga 'correct'

understanding ofthe 'mechanisms' thatactuallydetermine outcomesin the real world,thensome-
onecouldhaverationalexpectations - thatis,understand and forecast'theworldverywell- yetstill
behave'irrationally/ meaningin a waythatdid notinvolvemaximizing his or herutility."
3. This paper takes the sensitivity ofcognitivebiases to incentivesas an empiricalquestion,as for
example,Tverskyand Kahnemanseem to: "Thisarticlehas been concernedwithcognitivebiases
thatstemfromthe relianceon judgmentalheuristics.These biases are notattributable to motiva-
tionaleffects such as wishfulthinking or the distortion ofjudgmentsbypayoffs and penalties.In-
deed, severalofthe severeerrorsofjudgmentreportedearlieroccurreddespitethe factthatthe
subjectswereencouragedto be accurateand wererewardedforthe correctanswers"[1982,18]. It
shouldbe notedhoweverthatotherresearchersappearto takeresponsiveness to incentivesas defi-
nitional;ifa formofirrationality decreasesas payoffs rise,itcouldn'thavebeencognitive in thefirst
place.As Piattelli-Palmarini putsit,"It is in factimportant to distinguish carefully betweencogni-
tiveillusionsand simpleerrorsofjudgmentorblundersdue to inattention, distraction,lackofinter-
est,poorpreparation, genuinestupidity, timidity, braggadocio, emotionalimbalance,and so on...It
shouldbe emphasizedthatcognitive scienceconsistsofprobingtheordinary mentalstructures ofthe
humanspecies,freefromsuchspuriouseffects as are due to motivation, emotions,or aggressivity"
[1994,141].
4. The availability bias,as TverskyandKahnemanexplain,arisesbecause"peopleassess thefrequency
of a class or the probability of an eventby the ease withwhichinstancesor occurrences maybe
broughtto mind"[1982,11] eventhoughfactorsotherthanfrequency and probability - suchas fa-
miliarity and salience- predictably affect theease ofrecallingexamples.
5. "Incentives do notoperatebymagic:theyworkbyfocusing attention and byprolonging deliberation.
Consequently, theyare morelikelytopreventerrorsthatarisefrominsufficient attentionand effort
thanerrorsthatarise frommisperception or faultyintuition" [Tverskyand Kahneman,1987,90].
But whycouldn'tincentives promptpeopleto be moreskepticalabouttheirintuitionorperception,
turnto aggregateevidence,double-check withan expert,or expendmoreresearchtime?
6. Considerthe specialcase ofbehavioralindifference curveshorizontalat y*=0.A personwithsuch
indifference curveshas rationalexpectations notbecauseit is in his self-interest, butbecausehe has
an intrinsic tastefortruth.He therefore staysrationalevenwhenirrationality has a positiveeffect
on wealth(see section4). "Truth-loving" indifference curvesare interesting at least fornormative
purposes,sincetheycapturetheideals ofscientific and philosophic objectivity.
7. Harrisonis aware ofthisdifficulty, but afterconsidering threepossibleexplanationshe concludes
thateconomists "havenousefulbusinessfussingaroundin an attempttomakesenseofunmotivated
behavior"[1992,1441].Smithand Walkeralso recognizetheneedto explain"whysubjectdecisions
are notjust randomresponsesin the absenceofsalientrewards"[1993,248] . Theirdecisioncost
modelexplainsbiasederrorsas theproductof"boundedrationality," plus truncation. It is notclear,
however,thattheirmodelcan explainprice-sensitive systematic errorswhentruncation problems
are notan issue.
8. Near-neoclassical preferences have no connection toRusselland Thaler'snotionofquasi rationality:
"[Dependingon howtheproblemis framed, it can be predicted whethertheagentwillchoosex ory.
We proposecallinganysuchregularyetnonrational behaviorquasi rational"[1985,1072].
9. Harrison[1989] actuallyuses slightlydifferent terminology: "messagespace" insteadof"beliefs,"
"expectedpayoff space"insteadofjust "payoffs."
10. Thispointis nearlyidenticaltowhatHarrisoncalls his "payoff dominance" critiqueofexperimental
findings.In termsofmyframework, howeverstrongtheexperimental case againstpolarneoclassi-
cal preferences, everywhere butthatpole is consistent withtheevidence.
11. The experimental findings Einhornand Hogarthdiscussedhereconcernutilitytheoryratherthan
rationalexpectations; elsewherein thesame volumeHogarthand Reder[1987,12] applythispoint
moregenerally.
12. I owe thisformulation to an anonymous referee.
13. Withregularincentives, a correct pickearns$.40,whilea wrongpickearns$.10;withintenseincen-
tives,a correctpickearns $1.00,whilea wrongpickearns -$.50. [Friedman,1998] For regular
incentives, E(payoff | switch)=$(2/3x.40+l/3 X.10)= $.30,E(payoff | don'tswitch)= $(2/3X.10+1/3 X.40)
= $.20, and thus the expectedmarginalcost of biased versus unbiased probabilityestimatesis
E(payoff | don'tswitch)=$.3O - $.20=$.10.Forintenseincentives,
| switch)-E(payoff E(payoff | switch)
= $.50,E(payoff | don't switch) = $.00, and thus E(payoff | switch)-E(payoff|don'tswitch)= $.50.
 RATIONAL IRRATIONALITY 209

14. Switchbonus shouldnotbe interpreted as a simplepracticeeffect, because the set ofindependent

variablesalso includesa trendvariable.
15. On a basicAltavistasearchof"3 doorspuzzle,"thefirstthreehitsexplainthecorrectanswer.Even
in a searchof"switchstay3,"one ofthefirsttenhitsoffers a correctexplanation.
16. Forthepartoftheexperiment withbothintenseand standardincentives, subjectsplayedeither12
or 15 turns.
17. The "illusionofcontrolis definedas an expectancy ofa personalsuccessprobability inappropriately
higherthantheobjectiveprobability wouldwarrant"[Langer1982,231]. Moregenerally, it consists
in treatingchanceeventsas controllable.
18. A secondpossibleresponseto socialpressureis toengagein "preference i.e.,pretendto
falsification,"
shareadvantageousviewswithoutactuallyadoptingthem[Kuran,1995].
19. Thisis onlyoneformofself-serving bias discussedin theliterature.Manyformsofself-serving bias
have been seen in normative, ratherthanpositive,beliefs,wherethe conceptofsystematic bias is
moredifficult to apply.Self-interest maybe empirically shownto influence perceivedfairness[Dahl
and Ransom,1999],but withouta measureof"truefairness"thereis no way to test forrational
expectations in the usual sense. In termsofwealth,moreover, self-servingbiases have oftenbeen
seen as coa/iter-productive: A personwhooverestimates his productivitywillprobablyearnless,not
more,as a result.Babcockand Loewensteinraise an important caveat whentheyask "whether it
benefitsa partyto be less biased,holdingconstantthebeliefsoftheotherparty"[1997,116].Ifthe
answerto theirquestionis yes fora givenformofself-serving bias, thenit can be analyzedwitha
standardnegatively-sloped budgetline.Iftheansweris no- less bias is harmfulgiventhebiases of
others - thenit shouldbe analyzedwitha positively-sloped budgetline.
20. Andnoteventhatin everycase,as Akerlof andYellen[1985]show,oras Russelland Thaler's[1985]
findings on consumers'detergent choicessuggest.
21. Rationality inpoliticsis in effecta publicgood,which,as numerousexperiments show,willtendtobe
under-supplied evenifit is notentirely absent[Ledyard,1995]

REFERENCES

Akerlof.G. The EconomicsofIllusion.Economicsand Politics,Spring1989,1-15.

Akerlof,G. and Dickens. W. The EconomicConsequencesofCognitiveDissonance.AmericanEconomic
Review,June1982,307-19.
Akerlof,G. and Yellen, J. Can SmallDeviationsfromRationality Make SignificantDifferencesto Eco-
nomicEquilibria?AmericanEconomicReview,September1985,708-20.
Aumann, R. CorrelatedEquilibriumas a ExpressionofBayesianRationality.Econometrica, January
1987,1-18.
Babcock, L. and Loewenstein. G. ExplainingBargainingImpasse: The Role ofSelf-Serving Biases.
JournalofEconomicPerspectives, Winter1997,109-126.
Becker, G. IrrationalBehaviorand EconomicTheory.JournalofPoliticalEconomy,February1962,1-
13.
JournalofPoliticalEconomy,June1974,1063-91.
. A TheoryofSocial Interactions.
Boulier, B. and Goldfarb,R. On the Use and NonuseofSurveysm Economics.JournalofEconomic
Methodology, 1998,1-21.
Brennan, G. and Lomasky,L. Democracy and Decision:ThePure TheoryofElectoralPreference. Cam-
bridge:CambridgeUniversity Press,1993.
Camerer, C. Do Biases in ProbabilityJudgments Matterin Markets?Experimental Evidence.American
EconomicReview,December1987,981-97.
. IndividualDecisionMaking,in The HandbookofExperimental Economics,editedby J.
Kagel and A. Roth.Princeton, NJ:PrincetonUniversity Press,1995.
Caplan, B. RationalIgnorancevs. RationalIrrationality.UnpublishedManuscript, GeorgeMasonUni-
versity, 1999a.
. The LogicofCollectiveBelief.UnpublishedManuscript,GeorgeMason University, 1999b.
Casscells, W.,Schoenberger,A., and T.
Graboys, Interpretation byPhysicians ofClinicalLaboratory
Results.New EnglandJournalofMedicine,November1978,999-1001.
Cosmides, L. and Tooby, J. AreHumansGoodIntuitiveStatisticiansAfterAll?Rethinking SomeCon-
clusions from the Literatureon Judgment Under Uncertainty.Cognition,January 1996, 1-73.
210 EASTERN ECONOMIC JOURNAL

Dahl, G. and Ransom, M. Does WhereYou Stand Depend on WhereYou Sit? AmericanEconomic
Review,September1999,703-27.
Dickens, W. OccupationalSafetyand Healthand Irrational'Behavior:A Preliminary Analysis,in Work-
ers*Compensation Benefits: Adequacy,Equity,and Efficiency , editedbyJ. Worrolland D. Appell.
Ithaca:ILR Press,1985,19-40.
. SafetyRegulationand Irrational'Behavior,in HandbookofBehavioralEconomics,vol.A,
editedbyB. Gilad and S. Kaish. Greenwich, CT: JAIPress,1986,325-48.
Einhorn, H. and Hogarth, R. DecisionMakingUnderAmbiguity, in Rational Choice:The Contrast
BetweenEconomicsand Psychology, editedby R. Hogarthand M. Reder.Chicago:University of
ChicagoPress,1987,41-66.
Frey, B. and Eichenberger, R. Anomaliesand Institutions. JournalofInstitutional and Theoretical
Economics145,September1989,423-437.
. EconomicIncentivesTransform PsychologicalAnomalies.JournalofEconomicBehavior
and Organization, March1994,215-34.
Friedman, D. MontyHall's ThreeDoors:Construction and Deconstruction ofa ChoiceAnomaly. Ameri-
can EconomicReview,September1998.933-946.
Harless, D. and Camerer, C. The PredictiveUtilityof Generalized Expected UtilityTheories.
Econometrica, November1994,1251-89.
Harrison, G. Theoryand MisbehaviorofFirst-Price Auctions.AmericanEconomicReview,September
1989,749-762.
. ThePayoffDominanceCritiqueofExperimental Economics.UnpublishedManuscript, URL
httpV/theweb.badm.sc.edu/glenn/dominance.pdf, 1990.
. Theoryand MisbehaviorofFirst-Price Auctions:Reply.AmericanEconomicReview,De-
cember1992,1426-1443.
Harrison, G. and Kristrm,B. On theInterpretation ofResponsesin Contingent ValuationSurveys,in
CurrentIssues in Environmental Economics,editedbyP. Johansson, B. Kristrm, and K. Maler.
Manchester:ManchesterUniversity Press,1995,35-57.
Hogarth, R. and Reder, M. Introduction: PerspectivesfromEconomicsand Psychology, in Rational
Choice:The ContrastBetweenEconomicsand Psychology, editedby R. Hogarthand M. Reder.
Chicago:University ofChicagoPress,1987,1-23.
Kahneman, D., Slovic, P. and Tversky,A. Judgment UnderUncertainty: Heuristicsand Biases. Cam-
bridge:CambridgeUniversity Press,1982.
Kirchgssner, G. and Pommerehne,W. Low-CostDecisionsas a Challengeto PublicChoice.Public
Choice,September1993,107-15.
Klein, D. If Government is So Villainous,How Come Government OfficialsDon't Seem Like Villains?
Economicsand Philosophy, April1994,91-106.
Kuran, T. PrivateTruths,PublicLies: The Social Consequences ofPreference Falsification.Cambridge:
HarvardUniversity Press,1995.
Langer, E. The IllusionofControl,in Judgment UnderUncertainty: Heuristicsand Biases, editedbyD.
Kahneman,P. Slovicand A Tversky.Cambridge:CambridgeUniversity Press,1982,230-8.
Ledyard, J. PublicGoods:A SurveyofExperimental Research,TheHandbookofExperimental Econom-
ics,in J. Kagel,and A. Roth.Princeton, NJ:PrincetonUniversity Press,1995.
Lichtenstein, S., Fischhoff,B. and Phillips, L. CalibrationofProbabilities: The State ofthe Artto
1980,in Judgment UnderUncertainty: Heuristicsand Biases, editedbyD. Kahneman,P. Slovic
and A. Tversky.Cambridge:CambridgeUniversity Press,1982,306-34.
Montgomery,J. Contemplations on theEconomicApproachto ReligiousBehavior.AmericanEconomic
ReviewPapersand Proceedings, May 1996,443-7.
Nisbett,R. and Ross, L. HumanInference: and Shortcomings
Strategies ofSocial Judgment. Englewood
Cuffs:Prentice-Hall, 1980.
Piattelli-Palmarini,M. InevitableIllusions:How MistakesofReasonRule OurMinds.New York:John
Wileyand Sons, 1994.
Rabin, M. Moral Preferences, Moral Constraints,and Self-Serving Biases. UnpublishedManuscript,
University ofCaliforniaBerkeley,1995.
. Psychologyand Economics.JournalofEconomicLiterature, March1998,11-46.
Russell, T. and Thaler, R. The RelevanceofQuasi Rationality in Competitive Markets.AmericanEco-
nomicReview,December1985,1071-82.
 RATIONAL IRRATIONALITY 211

Shefffrin,S. RationalExpectations. Cambridge:CambridgeUniversity Press,1996.

Smith,V. RationalChoice:TheContrastBetweenEconomics andPsychology.JournalofPoliticalEconomy,
Aueust1991,877-897.
Smith,V. and Walker,J. MonetaryRewardsand DecisionCostin Experimental Economics.Economic
Inquiry,April1993,245-61.
Taylor, S. PositiveIllusions:CreativeSelf-Deceptionand theHealthyMind. NY: Basic Books,1989.
Thaler, R. The Psychology and EconomicsConference Handbook:Commentson Simon,on Einhornand
Hogarth,and on Tverskyand Kahneman,in RationalChoice:The ContrastBetweenEconomics
and Psychology, editedbyR. Hogarth,and M. Reder.Chicago:University ofChicagoPress,1987,
95-100.
Tversky,A. and Kahneman, D. Judgment UnderUncertainty: Heuristicsand Biases, in Judgment
UnderUncertainty: Heuristicsand Biases, editedby D. Kahneman,P. Slovic,and A. Tversky.
Cambridge:CambridgeUniversity Press,1982,3-20.
. and . RationalChoiceand theFramingofDecisions,in RationalChoice:The
ContrastBetweenEconomicsand Psychology, editedbyR. Hogarth,and M. Reder.Chicago:Uni-
versityofChicagoPress,1987,1-23.
Waldman,M. Systematic Errorsand theTheoryofNaturalSelection.AmericanEconomicReview,June
1994,482-97.
Wittman,D. TheMythofDemocraticFailure:WhyPoliticalInstitutions Chicago:Univer-
are Efficient.
of
sity Chicago Press, 1995.