diffusion of ideas in academy

Thediffusionofideasintheacademy:Aquantitative

illustrationfromeconomics

ShaunP.HargreavesHeap∗,AshokParikh

SchoolofEconomicandSocialStudies,UniversityofEastAnglia,NorwichNR47TJ,UKReceived4October2004;receivedinrevisedform29April2005;accepted24August2005

Availableonline26October2005

Abstract

Ithaslongbeenrecognisedinpolicydiscussionthattheinfluenceofnewideasdependsnotjustontheirgenerationbutalsoonthewayinwhichtheydiffusethroughsociety.However,whilethespreadofnewideasinindustryhasbeenmuchstudied,therearenosimilarquantitativestudiesondiffusionwithintheacademy.Thispaperaddressesthatgapbyprovidingquantitativeevidenceontheadoptionoftwotechniquesforempiricallytestinghypothesesineconomics.Onthebasisofthisevidence,andcontrarytoacommonexpectation,thespreadofideasseemsneitherstraightforwardnorespeciallyrapidintheacademy.©2005ElsevierB.V.Allrightsreserved.

Keywords:Diffusion;Techniques;Academy

1.Introduction

Ithaslongbeenrecognisedinpolicydiscussionthattheinfluenceofnewideasdependsnotjustontheirgenerationbutalsoonthewayinwhichtheydiffusethroughsociety.Thereis,forexample,alargequanti-tativeliteratureonwaythatnewideasspreadthroughindustry(e.g.seeStoneman,1995,2001;Geroski,2000).Thesame,however,cannotbesaidofthelitera-tureonthedisseminationofknowledgeintheacademywherethereareseveralqualitativestudiesbutfewquan-titativeones(see,forexamplethesurveybyStephan,

Correspondingauthor.

E-mailaddress:s.hargreavesheap@uea.ac.uk(S.P.HargreavesHeap).

0048-7333/$–seefrontmatter©2005ElsevierB.V.Allrightsreserved.doi:10.1016/j.respol.2005.08.005

∗

1996).Thisissomewhatsurprisingasthereappeartobetwopotentiallydifferentviewsonthesubject.

Thethoughtthatacademymightfunctionlikea‘marketplaceforideas’hasalonghistoryandithasreceivedrecentsupportfromtheanalysisthatshowshowtheincentivesfornewideacreationthatarisefromthe‘prioritytodiscovery’normareremarkablysimilartothosecreatedbythepatentsysteminindus-try(seeDasguptaandDavid,1994).Inthiscontext,diffusionmightbeexpectedtofollowtherelativelystraightforwardlogisticpatternthatiscommonlyfoundinindustrystudies.Onemightalsosupposethatthehallmark‘openness’oftheacademywouldmakethespeedalongsuchapathratherrapid.Indeed,DasguptaandDavid(1994)explicitlyshowhowthe‘prioritytodiscovery’normcanencouragethetimelydisclo-

1620S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

sureofnewideasand,asthereisnoequivalenttothepaymentofaroyaltywhenanewideaisadopted,theconditionsintheacademyseemsetforrelativelyquickdiffusion.

However,asthesociologicalliteraturethathasbeenspawnedsinceKuhn’sStructureofScientificRevo-lutionsmakesclear,matterscanbemorecomplex.Thedecisiontoadoptanewideamustdepend,ifonemaintainsthelanguageofeconomics,ontherelationbetweenthebenefitsandcostsofusinganewideaand,whilenewideasmaybemorefreelyavailableintheacademythanindustry,thereseemstobenostrongrea-sonforsupposingthattheperceptionofbenefitswillbeanykeenerthere.Indeedthequalitativestudiesofdif-fusionintheacademyoftensuggest,ineffect,thattheabsenceofcompetitionallowstheperceptionofbenefittobeinfluencedbyextraneousfactors,likeconsidera-tionsofpersonalinterest,andsocomplicateanddelaytheprocessofdiffusion(e.g.seeBarnesandShapin,1979;Barnesetal.,1996).

Therearenoquantitativestudiesofdiffusionthathelpinjudgingthesecompetingclaims.Thepurposeofthispaperistoaddressthatgapbyconsideringwhetherthepatternofdiffusionintheacademyissimilartothatfoundinindustry,theparadigmcaseofamarketdrivenprocess.

Thisisimportantfromapolicyperspectivebecausetotheextentthatthemetaphorofthemarketdoesapply,thenpolicyneednotbeespeciallyconcernedwiththedetailsof‘local’institutions(likethepar-ticulararrangementsforfundinguniversities,hiring,promotingandfiringacademics,etc.)andtheassoci-ated‘local’sociologicalaspectsoftheacademy.Itisthe‘market’whichlargelydeterminesoutcomesinthesecircumstancesandtheremaybeavarietyof‘local’institutionsthatarecapableofoperating‘efficiently’insuchasetting.Policyinthesecircumstancesneedonlybedirectedtowardsmaintainingtheconditionsofcompetitioninthat‘market’.

Therehavebeensomequantitativestudiesofdiffu-sionintheacademybuttheyhavefocussedontheveryparticularissueofwhetheryoungoroldresearchersaremoreorlesslikelytoadoptanewidea(e.g.seeStewart,1986;Messeri,1988;Levinetal.,1995).Ifhasoftenbeenconjectured,theyoungaremorerecep-tivetonewideas,thenthiswouldhaveobviouspolicyimplications.Theevidencefromthesestudiesis,how-ever,moremixedandasaresulttherehasbeenlittle

pressureonthesegroundstoshiftthedemographicsoftheacademy.

Therelativelackofquantitativestudiesondiffusionintheacademyisunderstandablebecausetherearenosimpleoreasymeasuresfortheacademythatareanal-ogoustotheproportionoffirmsusinganewtechniqueinindustrystudies.Therearegoodreasonsforthisinthesensethatthereareproblemsassociatedwithsimplewaysofconstructingameasureofdiffusion.Wedis-cussthesedifficultiesinSection2andintroducetwo‘ideas’fromeconomicswheretheycanbeovercome.ThemaincontributionofthispapercomesinSection3wherewepresentanddiscusstheevidenceonthediffusionofthesetwoideasineconomics.Thesearethefirststudiesintheacademywhichplotdiffusioninawaythatiscommoninindustrystudiesandtheysug-gestthatinsomerespects,butnotall,diffusionintheacademyappearsneitherstraightforwardnorrapid.Inparticular,whilethereissomesupportforthemetaphorof‘marketplaceforideas’inthisdata,itismixed.Inaddition,journaleditorsplayappeartoakeyroleintheprocessofdiffusionandthisraisesanobviouspol-icyquestionconcerninghoweditorsareandshouldbechosen.

Ofcourse,itisdangeroustogeneraliseonthebasisofwhatis,ineffect,twoobservations.But,aswesug-gestintheconcludingsection,someevidenceisbetterthannoneand,attheleast,theevidencepresentedheresuggeststhatmoreworkneedstobedoneplottingdif-fusioninotherdisciplinesaswellasthediffusionofotherideasineconomics.

2.Dataondiffusion

Thecommonmeasureofdiffusioninindustrialstud-iesistheproportionoffirmsemployingaparticulartechnique.Dataontheproportionofoutputaccountedforbysometechniqueisrarelyavailablebutitisalsosometimesused.Theadoptionofanewideainindus-tryisthustakentobesynonymouswithitsactualuseinproductionanditseemsnaturaltofollowthesameprocedurewhenlookingattheacademy.Therearetwomeasuresthatmightbeusedforthispurpose:thepro-portionofresearchersemployingandtheproportionofknowledgeproducedusinganidea.Theyareanal-ogousto,respectively,thefirmandoutputmeasuresinindustry.Therearenodirectmeasures,though,of

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–16321621

eitherknowledgeproductionortheactivitiesofallresearchersintheacademy,letalonetheextentoftheirrelianceonaparticularidea.Nevertheless,itmaybetemptingtothinkthatindirectmeasuresofdiffusioncouldbeconstructedrelativelyeasily.

Forexampleoneapproachtothisprobleminthelit-eratureontheinfluenceofageonadoptionhasbeentotakeasampleofscientistsinaparticularfieldandrecordwhentheyhaveadoptedanidea(e.g.seeStewart,1986;Messeri,1988,onplatetectonicsandLevinetal.,1995,onevolutionarytheory),butitisnotwithoutdifficulty.First,thesampleofscientistsisnecessarilysmall,raisingdoubtsaboutitsrepresenta-tiveness.Second,itisnotalwayseasyinthesecasestojudgewhenindividualshaveactuallyadoptedapartic-ularideasincemanyscientistsarecautiousandhedgetheirviewswhendiscussinganewandcontroversialidea(likeplatetectonicsandDarwinianevolution).Analternativeapproach,thatmakesuseofthevar-iouscontemporarydatabasesonjournalarticlesandcitations,mightbetotakethearticlesinasuitablycho-sensampleofjournalsinaparticularsubjectareaasareasonableindexbothoftheknowledgeproducedandoftheactivitiesofallresearchersinthatsubjectarea(wesaymoreonthisbelow).Then,whenaparticu-larideacanbeassociatedwithanarticleorbook,itsspreadcanbeplottedthroughthework’scitationinthisgroupofjournalsovertime.Thismayseemlikeanobviousmethodbutoneimmediatedifficultyarisesbecause,oncetheideaiswellaccepted,itbecomespartoftheether,sotospeak,andtheoriginalpaperiscitedmuchlessfrequently.Ineconomics,forinstance,therationalexpectationshypothesisisincreasinglyusedwithoutcitingMuth(1961)justasthe‘lemon’smodel’isreferredtowithoutnecessarilycitingAkerlof(1970).Anotherdifficultywiththisapproachoccurswheneveranideacannotbeuniquelyassociatedwithasinglearticleorpieceofwork.ThusalthoughMuth(1961)isusuallytakenasthekeyoriginalarticlefortherationalexpectationshypothesis,thereissomecontroversyinthehistoryofthoughtonthis(seeKeuzenkamp,1991)andthereissignificantoverlapbetweenthecontentofthisideaandtheefficientmarketshypothesisinthefinanceliteraturewhereamorecommonreferenceisFama(1970).Finallythereistheproblemthatcitationneednotreflectacceptanceoftheideasinceapapermaybecitedasapointofdisagreementasmuchasoneofagreement.Thelastoftheseproblemsisfundamentalbecauseitaffectsanymeasurethatisbasedoncitationoranindex-icalcountofkeywords.Theonlywayarounditwouldseemtobetocheckeachcitationorkeywordtoseewhetherthereferencereflectsacceptanceorrejectionoftheidea;andthisthreatenstobecomeexcessivelyonerousthemomentthesetofjournalsissufficientlylargetobereasonablyrepresentativeandtherearemorethanafewyears.

Weavoidthesedifficultiesbyfocussingontwoideas.Theyareideaswithrespecttotechniquesofempiricaltestingineconomics.One‘idea’istousepre-existingdatasetsandapplyeconometrictechniquestotesthypotheses.Sincethesedatasetsarenottheproductofcontrolledexperimentsdesignedtotestaparticularhypothesis,thisapproachdependscriticallyoneconometrictechniques.Thishasbeenthepredom-inantmethodofempiricaltestingineconomicsduringthepost-warperiod.Theotherinvolvesrunninglabo-ratoryexperimentstogeneratefreshdatathatsupplyadirecttestofaparticularhypothesis.Althoughthisistheparadigmtechniqueforempiricaltestinginthenaturalsciences,ithasonlybeenusedwithanyfre-quencyoverthelast25yearsineconomics.Thechoiceofthesetwotechniqueshastheadvantagethattheirspreadcanbemonitoredrelativelyeasyinarepresen-tativesampleofjournals.Itonlyrequireseacharticletobescannedfortheuseofthetechniquesinceitwouldbedifficulttoconstrueuseasanythingotherthanacceptanceoftheideabehindthetechnique.Ourprecisemeasureofdiffusionistheproportionofpagesdevotedtoarticlesingeneraljournalswhichusethesetechniques.

Thechoiceofthesetechniqueshasanotheradvan-tage.Theadoptionofbothrequiresinvestmentinhumanandphysicalcapitalandso,inthissense,theyaresimilartonewtechniquesinindustry.Ofcourse,theseinvestmentsdifferbetweentechniquesandtheyvaryovertime,butthesameisthecaseinindustry(wesaymoreonthisbelow).Indeedthechangesoccurforthesimilarreasons(e.g.thecombinedinfluenceoflearningandfurthertechnologicalchanges).Thusinbroadterms,thecostsofadoptingthesetechniquesintheacademytakeasimilarformasthoseinindustryandso,ifcompetitionintheacademyfocusestheper-ceptionofbenefitsinasimilarway,thenonemightexpecttoseeasimilarpatternofdiffusionforthesetechniquesasthoseinindustry.

1622S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

Wechoosegeneraljournalsfortheobviousrea-sonthatspecialistjournalsmightexhibitabiaseithertowards/againstempiricaltestingortowardstheuseofoneorothertechniqueforempiricaltesting.Thiswouldbetheproblemwithlooking,say,attheJournalofEco-nomicTheoryortheJournalofRiskandUncertaintysincetheonehasvirtuallynoconcernwithempiricaltestingandthematerialoftheotherhaslentitselfpar-ticularlytotheexperimentaltechnique.

Anygeneraljournalwouldavoidthiskindofbiasandweselectedamixedsampleofsix‘top’andfour‘middle’rankingones.Themixtureisimportantifthesampleistoberepresentative,neverthelessitisweightedtowardsthe‘top’journalsbecausetheyaretypicallyregardedasthemostimportantsitesfortheproductionofknowledgeineconomics.Sincemostofthe‘top’journalsarelocatedintheUS,ourmiddlerankingonesaredrawnfromoutsidetheUStogivethesamplesomegeographicdiversity.Thefulllistofourjournalsis:••American•JournalEconometricaofEconomicPoliticalReview(AER).••Quarterly(Ecra).

Economy(JPE).•EconomicJournalofEconomics(QJE).•EconomicaJournal(EJ).•Oxford•ScottishEconomic(Eca).

•

AustralianJournalofPapersPolitical(OEP).

CanadianJournalEconomicofEconomicsPapersEconomy(AEP).(SJPE).(CJE).

Thefirstfivewouldfeatureinmostlistsof‘top’journals.Economicaisprobablyaborderlinecasenow,althoughitwouldnothavebeenduringthe1950swhentheapplicationoftheeconometrictechniquetoexistingdatasetswasdiffusing.OEPhasbeenincludedsothatthereareareasonablenumberofEnglishjournalsasthiswillallowacomparisonbetweendiffusionintheUSwiththatoftheUK;andtheSJPE,AEPandCJEwereincludedtoprovidefurthergeographicspread.Therewillalwaysberoomfordisputeoverwhetherthissampleisadequatelyrepresentative;wesetthisasidenowandfocusonsomefurtherpotentialdifficul-tieswithourproposedmeasure.Oneapparentprob-lemisthatbothtechniquesforempiricaltestinghavechangedovertime.So,apieceofeconometricsonapre-existingdatasetnowdoesnotembodythequitethesameideaasone,say,30yearsago.Thissame

difficultyarises,however,withindustrystudiesasthenewtechniquesofproductionalsoevolveovertime(e.g.seeKarshenasandStoneman,1995).Asaresultthesestudiesfocusonwhatis,ineffect,thespreadoftheunderlyingor‘generic’newidea,touseKarshenasandStoneman’s(1995)term.Bylookingattheuseofeconometricsonpre-existingdatasetsandexperimentswhichcreatenewdatasetsovertimewithoutdistin-guishingbetweenthekindofeconometricsortypeofexperiment,wearedoingthesame.Wearelookingatthespreadofthese‘generic’ideasforempiricallytestinghypothesesineconomics.

Anotherproblemconcernshowtointerpretthepro-portionsofpagesdevotedtoarticlesthatuseoneofthesetechniques.Thenaturalinterpretation,ifthejour-nalsareadequatelyrepresentative,isthatthisgivesameasureoftheproportionofknowledgeproducedintheacademythatusesthesetechniques.Butthisplainlyslidesovertheissueofhowtoaggregatepagesindiffer-entjournalsondifferenttopicsbytreatingeachequally.Ignorancemaywellcommendsuchastrategy,butitcannotbesourceofcomfort.

Thereisalsothequestionofwhetherthesepro-portionsarelikelytoreflecttheproportionsofactualacademicswhohaveacquiredthesetechniquesinamanneranalogoustotheindustrystudieswhichlookattheproportionoffirms.If(1)theproportionofaca-demicsacquiringthesetechniquesisreflectedintheproportionofsubmissionsinthisclasstothesejournalsand(2)theproportionofpublicationsofaparticulartypeinjournalsfollowstheproportionofsubmissionsofthistype,thenthemeasurewouldreflectthepropor-tionofacademicsusingthistechnique.

Whatmakesbothconditionsunlikelytoholdisthepresenceofaneditorialbias.Thisisplainwithrespecttothesecond,butitwillalsoaffectthefirstsinceacademicsarelikely,ceterisparibus,todirecttheirsub-missionstoajournalthatismorelikelytopublishtheirkindofresearch.Thuseditorialbiascouldaccountbothforwhytheproportionofpagesdevotedtoarticlesofacertaintypeinajournaldiffersfromtheaverageandwhythegrowthinthisproportioncanalsodifferacrossjournals.Since,journalsarewellknowntohavebiasesofthissort(e.g.theEconomicJournalunderKeynessharedhisscepticismregardingeconometrics),thisislikelytoproduceasignificantwedgebetweenourmea-sureofthespreadofanideainanyonejournalanditsactualspreadintheacademiccommunity.Thishigh-

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

Table1

Meansandstandarddeviationsofvariables(%ofpages)Nameofthevariable

EconometricmethodologyMean

PercentageoftotalpagestototalofalljournalsEconomicJournalEconomica

OxfordEconomicPapers

ScottishJournalofPoliticalEconomyAmericanEconomicReviewJournalofPoliticalEconomyEconometrica

CanadianJournalofEconomicsaAustralianEconomicPapersbQuarterlyJournalofEconomics

19.100011.170316.600017.400016.1919.275028.300023.525027.043425.0615.8250

Standarddeviation9.66308.443412.759712.050714.820712.391016.888.85495.958011.801710.1397

ExperimentaleconomicsmethodologyMean1.49621.78090.79480.15520.02.3109.1.72622.56810.10280.01.7586

StandardDeviation0.87571.67231.54330.71140.02.33961.76792.33960.360.02.2291

1623

Note:Foreachjournal,totalnumberofpagesoneachmethodologyisdividedbythetotalnumberofpagesineachjournalandisconvertedtopercentage.Foralljournalsthetimeperiodis1950–1990exceptingthetwo.a1968–1990.b1962–1990.

lightstheimportanceoffocussingonaggregatefiguresthatarederivedfromarepresentativesampleofgen-eraljournalsinthefurthersensethatbiasesofthiskindneedtowashoutintheaggregate(i.e.thereoughttobeanEconometricaforeveryEJ).

Thisdoesnotmeanthatthedataondiffusionwithinaparticularjournaliswithoutinterest.Quitethereverse,theindividualjournaldeviationsfromtheaver-agelevelsofdiffusionnotonlysupplyinformationonwhichwerethe‘leading’and‘lagging’journals,theyalsoprovideapointertotheexistenceofeditorialbias.Ofcourse,theremaybeotherreasonswhydiffusionwilldifferacrossjournals.Forexamplejournalscouldservedistinctacademiccommunitieswherethecon-ditionsinfluencingtheacquisitionofanewtechniquediffer.Butintheabsenceofanyevidenceregardingtheseothersourcesofdifference,theactualdifferencesindiffusionusingourmeasuresuggestapresenceofeditorialbias;andwereportonthislater.

Asweareconcernedwiththespreadintheuseoftechniquesforempiricaltesting,weignoredanyarti-clesthatwereexclusivelyconcernedwiththepurethe-oryofeachkindoftechnique.Thusdatawascollectedonthenumberofpagesdevotedtoarticlesthatactuallyemployeconometrictechniquesonpre-existingdatasetstotesthypothesesforeachyearfrom1950to1990andthesewerethenexpressedaspercentagesofthetotalnumberofpagesineachyear.Wealsocollecteddataontheproportionofarticlesusingeconometric

techniquesasthiswouldbeapossiblealternativemea-sureofdiffusion,but,asthetimeseriesonproportionofarticlesinajournalcloselyresemblesthetimeseriesontheproportionofpagestototalnumberofpagesinthatjournal,therewasnoobviousreasontopreferonemea-suretotheother.Weoptedfortheproportionofpages.Likewise,datawascollectedontheproportionofpagesdevotedtoarticlesusingtheexperimentaltechniqueinhypothesistestingforeachyearfrom1979to2003.Thestartyearswerechosenbecausetherewaslittleuseofeconometricsandexperimentstotesthypothe-sesbefore1950and1979,respectively.Theenddateforeconometricswasdictatedbythefactthatthetech-niqueappearedtohavefullydiffusedinthesejournalsby1990.Thiswasnotthecasefortheexperimentalmethodand2003wasourlastobservation.ThefullrunofyearsisnotavailablefortheCanadianJour-nalofEconomics(whichwasestablishedin1968),theAustralianEconomicPapers(whichstartedin1962)andtheScottishJournalofPoliticalEconomy,whichwasestablishedin1954.

Fig.1andTable1provideasummaryofthedataondiffusionlevels.

3.ThepatternofdiffusionintheacademyInthissection,wetestthehypothesisthatthediffu-sionintheacademyiswelldescribedbythemetaphor

1624S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

Fig.1.Percentageoftotalpagestooveralltotalpagesin10journals.

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–16321625

Fig.1.(Continued).

ofa‘marketplaceforideas’.Weapproachthisquestionintwoways.First,welookatdiffusionintheaggre-gateandconsiderwhetherthepatternofdiffusionofthesetechniquesinthesamplejournalsissimilartotheparadigmofamarketdrivenprocess,thediffusionofanewtechniqueinindustry.Secondweconsiderwhetherthepatternofdiffusionvariesacrosscountries.Thepointofthissecondtestisthat,evenifdiffusionintheaggregateresemblesamarketdrivenprocess,itmaybethecasethatthevarietyofnationalaca-demicinstitutionscreateratherdifferentmarketsinthissense,sothatthemetaphorofthemarketrequiresqualification.

Thelogisticandexponentialfunctionsarecom-monlyusedtodescribethepatternofdiffusioninindustrialstudies(seeGeroski,2000)andweusethisasourreferenceforwhattoexpectfromamarketdrivenprocess.Weestimatebothfunctionsonourtwoaca-demicdatasets.Ifneitherfunctionalformfitsourdatawell,thentherearegroundsforrejectingthehypothe-sisthatthepatternofdiffusioniswelldescribedbythemetaphorofthemarket.1

Itisparticularlyimportanttophrasethisintermsofrejectingthehypothesissincewhilemarketprocessesinindustrydothrowupthesepatterns,suchpatternscanalsoarisefromotherprocesses.

1

Inparticular,withy(t)equaltotheproportionofknowledgeusingthistechniqueintimet,thelogisticfunctionisgivenbyy(t)=

A

1+exp(−b−dt)

(1)

whereAequalsthefinalsaturationlevel.Sincey0=A/(1+exp(−b)),theparameter‘b’determineshowfarthepercentageofadoptionisbelowthesaturationlevelattimezeroandtheparameter‘d’controlsthespeedatwhichthesaturationlevelisreached.Thepointofinflexionoccursatt=(−b/t)whentheproportionofadoptersequalsA/2(i.e.halfthefinallevel).

Asthegraphsforeachtechniqueshow,thereisconsiderablefluctuationintheproportionsfromyeartothenext.Hence,weconsideredtwoversionsofthelogisticfunction,onewithandonewithoutARCH(AutoregressiveConditionedHeteroscedastic-ity).ARCHmodelsaccountforvolatilitythroughtheconditionalvariances,whicharerelatedtoeitherprevi-oussimilarvariancesorthevariablesthatcouldexplaintheconditionalvariances.Inneithermodelwasthereanyevidenceofunitrootinresiduals.Thenullhypothe-sisofunit-rootnon-stationaritywasdecisivelyrejectedbytheaugmentedDickey–Fullertest.Wefindonlyslightimprovementingoodnessoffitmeasuredbylike-

1626S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

Table2

LogisticandothermodelsusingtotalofeconometricmethodologyandexperimentaleconomicsmethodologyDependentvariableY1:5.7%growthp.a.Y2Y1Y2Y1

ModelusedExponentialExponentialLogisticLogisticLogisticwithfirstorderARCH

Coefficient,a50.3026*(5.73)8.2941*(0.5526)29.5601*(35.5084)4.1302(1.1107)28.7324*(44.7740)

Coefficient,b0.0244*(3.9976)0.0181(0.9779)−2.6929*(11.8191)−2.0623*(2.9348)−2.5951*(15.4768)

0.17*(10.4625)0.1286(1.5014)0.1725*(12.9979)Coefficient,d–

R2/unitroottest0.92300.62330.9543(−4.80)0.6392(−4.099)−5.36a(t-valueonlaggedresidual)

−88.63−15.7960−83.83ARCH:A0=1.76(1.43)A1=1.54(3.52)logL

Note:Theresultsoflogisticapproachonexperimentaleconomicsmethodologyarenotsignificantonthespeedofdiffusionimplyingthatthediffusionisnotsignificantlydifferentfromzero.Subsequentresultsdonotmakeanyfurtheruseofexperimentaleconomicsmethodology.Exponentialmodel:Yit=a(1−exp(−bt)).Logisticmodel:Yit=a/(1+exp(−b−dt)).

aFirstdifferencesofresidualsareregressedonthelastperiod’sresidualsandtheobtainedt-ratioistestedagainstthecriticalvalueobtainedfromresponsesurfaceestimatesforT=50.Theusedcriticalvalueis−3.5005.*Significantat5%level.

lihoodvaluewithARCH;andsointheremainderofthepaperwedidnotcontinuewithARCH.3.1.Discussion:theaggregatediffusionoftheeconometrictechniqueonhistoricaldatasetsThelogisticfunctionfitsthedatabetterthantheexponentialoneforthistechniqueofempiricaltesting(Table2);andthisisthecommonresultinindustrystudies.Thecoefficientshavetherightsignsandallaresignificant.Thuswecannotrejectonthisbasisthehypothesisthatthediffusionofthistechniqueforempiricaltestingintheacademyissimilartothatdrivenbymarketprocesses.Neverthelesswemayhaverea-sonforqualifyingthehypothesisifthesizeofanyoftheseestimatedcoefficientsseems‘strange’.Suchjudgementsarenoteasytomake.

Thereisnoexpectationthateitherthesaturationlevelorthespeedofdiffusion(thevaluesofthe‘b’and‘d’parameters)shouldbethesamefordifferenttechniques.Theevolutionoftheperceivedcostsandbenefitsofusinganewtechniqueareboundtovarywiththetechniqueinquestion(betheyindustrialoracademic).Nevertheless,whencomparedwithindus-trystudiesthathavefittedthesamedescriptivelogisticfunction,thespeedofdiffusionparameter‘d’fortheeconometrictechniqueseemsonthelowside.Table3givesMansfield’s(1968,19)estimatesoftheequiv-alentparameter‘d’forseveralnewtechnologiesintheUS.Hismeasureofdiffusion,asinmanyindustrial

Table3

DiffusionofnewtechniquesinindustryInnovation

IndustrialrobotsDiesellocomotives

CentralisedtrafficcontrolCarretarders

ContinuouswidestripmillBy-productcokeovenContinuousannealingShuttlecar

TracklessmobileloaderContinuousminingmachineTincontainer

High-speedbottlefillerPalletloadingmachineSource:Mansfield(19).

Estimateof‘d’0.280.200.190.110.340.170.170.320.320.492.400.360.55

studies,istheproportionoffirmsadoptingthenewtechniqueandhisestimatesaresimilartothosefoundinothercountrystudies(e.g.Davies,1979).2

Giventhecommonperceptionregardingthe‘open-ness’ofacademy,itisperhapsalittlesurprisingtofindthatthespeedofdiffusionforthistechniqueofempiricaltestingintheacademyappearstobeonthelowsiderelativetothissampleofindustrialtechniques.

Strictlyspeakingthecomparisonisnotquiterightasouresti-matesof‘d’refertodiffusioninwhatisaninternationalsetofjournalsandtheseareestimatesofindustrialdiffusioninnationalmarkets.However,thepointstillholdswhenwecompareestimatedthespeedofdiffusionindifferentnationalacademiesinSection3.3.

2

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–16321627

Ofcourse,thepre-existing/econometrictechniquemaynotberepresentativeofdiffusioningeneralintheacademy.EquallytheMansfieldsampleneednotbeindicativeofwhattypicallyhappensinindustry.Otherstudiesprovidesomecheckonthelatter.

Forinstance,Mansfield(1968)providesestimatesofintrafirmdiffusionforoneinnovation,thediesellocomotive.Suchevidenceisratherrareandinthiscasediffusionatthefirmlevelismeasuredbytheproportionofenginesthatarediesels.Theestimatesof‘d’byfirmrangeacrosshissampleof30railroadcompaniesfrom1.35to0.28,andsoisalwayslowerthantheinterfirmpaceofdiffusionofdieselsgiveninTable3(where‘d’fordieselengines=0.2).VonTunzelman(1978)alsolooksattheaggregatediffusion,ofsteamenginesinthe19thcentury,andestimatesthat‘d’is0.25which,again,issimilartotherangefoundinTable3forinterfirmdiffusion.

Inaddition,therearesomestudieswithafewdis-creteobservationsonoutputproportionsaccountedforbyanewtechnique.ForexampleittookFordandGen-eralMotors20yearstomovefrom0tohalftheleveloftheircurrentuseofindustrialrobots(seeMansfield,19).IntheUK,fromfirstintroduction,ittook3yearsforSpecialPressesinpapermakingtodiffuseto10%ofoutput,5yearsfortheBasicOxygenPro-cesstodiffuseto20%,4yearsforGibberalicacidinbrewingtoreach50%,6yearsforContinuousCast-ingtoaccountfor1%,6yearsforShuttlelessloomstomake1%and10yearsforAutomaticTransferLinesinvehicleproductiontodisseminateto30%(seeNasbethandRay,1974).Whileitisalittledifficulttomakeaclearcutcomparisonwiththesediscreteobservationsonthetimetakentoreachparticularsaturationlevels,theyseemtopointtothesameconclusionontherel-ativelyslowdiffusionofthepre-existing/econometrictechnique.

Thecomparisonisdifficultbecausewedonotknowwhatthefinalsaturationlevelisforthesetechnologiesanditisnotclearwhattheappropriatedateisforthefirstuseofeconometrics.Nevertheless,supposewemakeassumptionsthataremostfavourabletotheeconomet-ricsonpre-existingdata,soweassumethesaturationlevelis100%foreachoftheseindustrialtechnolo-giesand,totakeaccountoftheinterventionofWW2,wesaythattheeconometrictechniquewasfirstintro-duced10yearsbefore1950.Theestimatesfor‘b’and‘d’wouldthensuggestthathalfthefinalsaturationrate

foreconometricswasachieved15yearsafterthefirstobservationin1950.Hence,ittook25years,onthisbasis,foreconometricstodiffusefromitsfirstusetohalfitsfinalsaturationlevelandthisisslowcomparedwithfourofthesixtechnologiesreportedinthetext.3.2.Discussion:theaggregatediffusionoftheexperimentaltechnique

Thelogisticfunctionagaindescriptivelyfitsthedatabetterthantheexponentialoneforthistechnique,butthespeedparameter‘d’isnotsignificantlydifferentfromzero.Thusonthisbasis,werejectthehypothesisthatthepatternofdiffusionofthistechniqueissimilartothatfoundinindustrystudiesofamarketprocess.Thelevelofdiffusionjusthoversaround0.025%andwhilethereisnothingsurprisingperseaboutthislowfigure,thefactthatitshowsnoapparentstatisticalten-dencytoeitherincreaseordecreaseisquiteunlikewhatisnormallyfoundinmarkets3.3.3.Crosscountrydifferences?

Thesecondhypothesisthatwewishtotestiswhethertherearedifferencesinthepatternofdiffu-sionacrosscountries.WestartbynotingthatthedatainFig.1suggestthatpatternofdiffusionhasnotbeenthesameinalljournals.Wehavetestedforthisstatis-tically.Table4presentstheresultsoffittingalogisticfunctiontothespreadoftheeconometrictechniquewithjournalspecificeffects.

Threemodelswereestimated.Thefirstincolumn2constrainseachjournaltohaveidenticalfinalsatu-rationlevels(‘a’),identicaldiffusionspeeds(‘d’)andidenticallevelsofadoptionatthetimezero(‘b’)toprovideareferencepoint.Thesecondmodelreportedinthethirdcolumnrelaxessomeoftheserestrictionsandallowsboththebaseleveldiffusionparameter4

3

Itseemsanopen(andinterestingquestion)astowhethertheabsenceofanychangeinthelevelofdiffusionshouldbeinter-pretedasvery,veryrapiddiffusion(i.e.itreacheditssaturationlevelstraightaway)orasmovementatavery,veryslowpace(i.e.ithasnotmanagedtogetbeyonditsinitialtoeholdvalueinover20years).IofferonereasonforthefirstinterpretationinSection4.

4Theparameterbdeterminesbyhowmuchtheadoptionlevelisbelowthesaturationlevel.Attimezero,thisparameterdeterminesthelevelofdiffusioninthelogisticfunctionandhencewehavenameditasbaseleveldiffusionparameter.

1628

Table4

Paneldataanalysiswithlogistic

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

Modelwithcommona,bandd

ABDB1B2B3B4B5B6B7B8B10D1D2D3D4D5D6D7D8D10R2(SEE)logL(χ2)Mean(S.D.)

29.5933(21.85)−2.5756(7.96)0.1619(6.)

Modelwithcommonabutdifferentbandd39.8840(22.)

Modelwithalla,bandddifferent

−3.0926(5.99)−2.9602(5.44)−1.9043(5.61)−4.0252(5.65)−2.6138(5.94)−3.6080(4.44)−0.1880(0.76)1.7268(1.96)−2.0141(5.66)0.0911(4.91)0.1187(5.41)0.0758(5.49)0.1565(5.53)0.1195(5.90)0.21(4.36)0.0246(2.38)−0.0324(1.20)0.0718(5.25)

0.4946(9.2976)−12.59

18.87(13.04)

0.7083(7.2367)−1169.38(190.42)

18.87(13.04)

−3.04949(5.40)−3.2387(3.96)−3.4185(3.04)−4.5696(4.00)−2.9318(4.43)−3.4738(4.83)−1.5777(1.47)7.0739(0.55)−2.4263(3.01)0.9059(1.93)0.1433(2.88)0.2288(2.87)0.1942(3.37)0.1567(3.56)0.2448(4.63)0.4490(1.83)−0.1446(0.46)0.1629(2.61)0.7294(7.0530)−1156.17(26.42)

18.87(13.04)

A140.25(1.22)A235.40(5.87)A325.77(12.02)A434.99(8.25)A534.36(9.19)A1.75(20.)A725.22(19.31)A829.55(5.44)A1024.40(8.51)

ForCanadianJournalofEconomics,theBcoefficientisofoppositesigntowhatisexpected.ForCanadianJournalofEconomicsandAustralianEconomicPaperswhenindividuallogisticcurveswerefittedtheconvergencewasnotachieved.ThebestmodelintheabovetableistheunrestrictedonewhereeachjournalhasdifferentA’s,B’sandD’s.Journal1:EconomicJournal;journal2:Economica;journal3:OxfordEconomicPapers;journal4:ScottishJournalofPoliticalEconomy;journal5:AmericanEconomicReview;journal6:JournalofPoliticalEconomy;journal7:Econometrica;journal8:CanadianJournalofEconomics;journal9:AustralianEconomicPapers;Journal10:QuarterlyJournalofEconomics.

andthespeedofdiffusiontovarywhilekeepingthecommonsaturationlevel.Thethirdmodelincolumn4isthemostgeneralmodelandallowseachjournaltohaveaseparatesaturationlevel,diffusionspeedandbaseleveldiffusionparameter.Ofthethreemodels,therestrictedmodelsofcolumns2and3arerejectedbytheLikelihoodratioχ2-test(χ2=190.42for16degreesoffreedomandχ2=26.42for8degreesoffreedomandboththeseexceedthecriticallevel).Thuswepreferthemodelwithdifferentparametersforeachjournalwithimportantinferencethat,likefirmsinindustrystudies,therearesomeleadingjournalsandothersthat‘follow’.

EstimatesforAustralianEconomicPapershavebeenexcludedbecausetheydidnotconverge.Incolumns3and4,theCanadianJournalofEconomicshasanincorrectlysignedparametersbanddandwhenthelogisticwasattemptedonitsownforCanadianJour-nalofEconomics,theconvergencewasnotachievedwithnonlinearleastsquaresroutine.Inallothercases,thecoefficientsaresignificantandtakethecorrectsign,soforthemostpartdiffusioncanbedescribedbylogisticcurvesattheindividuallevelaswellasintheaggregatedata.ThefailuresoftheAEPandtheCJEinthisrespectarenotreallysurprising,eveniftheyrestrictourcrosscountrycomparisons,asthesejour-nalswereestablishedafterthetechniquehaddiffusedsignificantlywithintheacademiccommunityandso,ascanbeseenfromFig.1,theyimmediatelystartedatratherahighlevel.

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

1629

Table5

Proportioninjournal‘i’regressedontotalproportionJournalConstantCoefficientontotalproportionEJ

−1.9(−1.17)0.7*(9.14)Economica−4.5*(−2.06)1.11*(10.6)OEP1.7(0.57)0.81*(5.5)AER−2.9(−1.8)1.16*(15.1)JPE

−2.08(−0.9)1.59*(14.6)Ecconometrica15.4*(5.46)0.4*(3.05)QJE

0.51(0.23)

0.8*(7.67)

t-Statisticsinparenthesis.*Significantat5%level.

Wetriedtofitjournalspecificlogisticcurvestothediffusionoftheexperimentaltechniquewithineachjournalbut,liketheaggregatedata,withoutsuccess.5Nevertheless,itisapparentfromtheplotsofdiffusioninFig.1thatonlyfourjournalshaveplayedmuchroleinthediffusionofthistechnique(Econometrica,AER,EJandQJE);and,ineffect,wenowfindasimilardiver-sityacrossjournalswithrespecttotheparametervaluesfor‘a’,‘b’and‘d’inrelationtothetechniqueofecono-metrics.

Toexplorethesejournaldifferencesinrelationtotheeconometrictechniqueonhistoricaldatasets,weconsiderthehypothesisthatdiffusioninanyjournalisthesameastheaggregatediffusionofthetechniquebyrunningalinearregressionofindividualjournaldif-fusionlevelsonaggregatediffusionlevelsandtestingwhetherthecoefficientonaggregatediffusionissig-nificantlydifferentfromone.TheregressionresultsarereportedinTable5.

Forfourjournalswecannotrejectthehypothesisthatthecoefficientisequalto1:theyareEconom-ica,OEP,AERandQJE.Thehypothesisisrejectedintheremainingthreejournals.Asremarkedearlier,onepossiblecauseofdifferencesindiffusionacrossjournalsiseditorialbias.Intwoofthesejournals,thereispriorinformationpointingtotheexistenceofsys-tematiceditorialbiaswhichsupportstheinterpretationoftheresultshere.ThusEconometricaiswell-knownasthejournalthatplayedanearlyleadingrolewith

5

WefoundthatforEconomicJournaltheparameterbwassignif-icantbutnotdwhilewithrespecttoAmericanEconomicReview,QuarterlyJournalofEconomicsandEconometrica,neitherbnordweresignificant.ThesaturationlevelforEconometrica,EconomicJournalandAmericanEconomicReviewwas4.62%,3.94%and6.34%,respectively.

respecttoeconometrics(e.g.seeMorgan,1994)andtheequationinTable5pointspreciselytoanearlyincreasingandthendecreasingbiaswithrespecttoeconometrics.Likewise,itisfrequentlysuggestedthatKeynes’seditorshipoftheEconomicJournalleftabiasagainsteconometrics(e.g.seeStigleretal.,1995)andagaintheequationinTable5pointstopreciselythisintheearlypartoftheperiod.

Ofcourse,thevariedpatternsofdiffusioninthesethreejournalsmightstillbepartiallyexplainedbyadifferentevolutioninthebeliefsoftheacademiccom-munityformedaroundthesethreejournals.Thefactthattheseareall‘top’generaljournalsmakesthisseemunlikely.Thereis,however,onerespectinwhichthecommunitiesmighthavediffered,atleastintheearlypartoftheperiod,whichcouldhaveinfluencedthediffusionoftheeconometrictechnique.Itispossiblethatthejournalslocatedindifferentcountriesinitiallyhaddistinctacademiccommunitiesbothbecausetherewaslessinternationaltravelandbecausetheinstitu-tionsgoverninghiring,promoting,firingandthelikevariedandproduceddifferinglevelsofcompetitionintheUKandUS.SincetherearenostrongresultsfortheAEPandCJE,wecanonlysensiblecomparetheUSandUKinthisrespect.

Towardsthisend,wegroupedthejournalsaccord-ingtotheircountry,estimatingthelogisticcurvefordiffusioninbothcountriesandtestingthehypothesesthatbUK=bUSanddUK=dUS.TheresultsaregiveninTable6.

Models1–3testthehypothesisthatUSandUKjour-nalshavethesamevaluefor‘b’and‘d’andModel1isrejectedbyModel2whileModel3isnotrejectedbyModel2.Hence,wepreferthemodelwherewehavethesamesaturationparameterforUKandUSjournalsbuttheb’sandd’sdiffer.OnthebasisoftheModel3,wefindthatthediffusionspeedisnotablyfasterintheUS(althoughstilllowcomparedwithindustrystudies).TheinitialdegreeofdiffusionisalsomoreadvancedintheUSwith2.95%ascomparedwith1.75%fortheUKjournalsattimezero.Theinflexionpointwherehalfofthesaturationlevelisachievedis21.5yearsforUKjournalswhileitis12.3yearsforUSjournals.Althoughitispossibletointerpretthisasevidenceinfavouroftheexistenceoftwoacademiccommunities,theresultcouldequallyhavearisensolelyfromthecon-jecturededitorialbiasesintheEJ,EconometricaandtheJPEasthesewouldhaveproducedfasterdiffusion

1630

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

dddddd;ee):eetteett)6ctccctcctiiiirirrirtrttrtSstssstesseeerreerrUlerrtsetsses2drrlososodoeeoMLMMedLMo−m))))))M123456t((((((K;)031U)t2255381S...630315...dUB544859S000909111787−2dSdn724391...a766046...U−L422722t0008882g111777boKl−−−−−−−UK1dc-iUets−i1ntoabst**ms)-7)5−aimt4339841(B.9.od0909pCn.a01.1xS(0(eU+2c1eidt()/−ns*iat34asa96.-t=a2s05i.e-t01(tiBY:Kc3i)Udtlns*8i2e)at8a90d1.-t0oe1s0e-.1t0(m−(;))−ct-idtSSsintUUoa2t**ms2-0)9)dbmt4736od51174..Cn.−−a0601.(02(tKKUcU1bdit1ns)))*i)***at8d−a3137127-784736..t2s13.1.0679..5.−(pd-t050((010.1((0(SxUe+c2dit))b1(ns**i))at**484a−/)-433831.tS1s19.12.028...d-t03.(06(0107.(01(KUU21acb-ibtsi−−nt**(Koa32)*tpms5)055)-5392.92.xUmt33.1414e1...od272323Cn+a(a−−(−−(−−(1(=/t)icSY.)di)))t**ns12*i05.U)tat27307a..272:S-..t252.2s2522a5Ub-t−−(−−(−−(+2slKledacndi)))Udt***o−trnsi290576.1uata92.84.18amKo-...t252621(j1sb-t−−−(−(−−(=U;Kt)1i)UYtdd:ddn2−snal−a-aesadaiSn)c***oiBUoi32mt0s703)4)i42047mBermt0a58.4557..;eoot..)−028.8)bfCs8-9t21(320(52(t−sdb−pic)−−(hi(pt))1ss026bpxnit065...oa−xe.lt136(e+eis-222t(((p+1vta***x(elden815e1/l()337ra+/-605...aS%n20321U1oa355(=/it2tsaiaatuc)fi)0=Y+nfts)50taiiidt610a5..Y:Kc4fit.61s812:Ui6c-i(((11ntt***lgesd178elailin203bgade(So-709...odTaL1844a244Mo=tmi*YinthetwoUSjournalsthanintheoneBritishjournal.Furthermore,ifthereweredifferencesbetweenaca-demicnationalcommunities,thenitwouldbedifficulttoexplainwhytheremainingfourjournalsappearedtohavenosystematicformofeditorialbias.OnewouldhavetoconjecturethatthetwoBritishjournalsinthisgrouphadaneditorialbiaswhichoffsettheacademiccommunityeffect;andunlikethealternativeconjectureattheEJ,JPEandEconometrica,thereisnoobviouspieceofsupportingevidenceforthesebiasesintheOEPandEconomica.Thuswebelievethatthemostlikelyexplanationofthesejournaldifferencesisedito-rialbias.

4.Conclusion

Thispaperprovidesquantitativeevidenceonthediffusionoftwotechniquesforempiricallytestinghypothesesineconomics.Doesthisevidencesuggestthatthereisapolicyissuewithrespecttohowideasdiffuseintheacademy?Ifsowhatisit?

Broadlyspeaking,fewwillgetexcitedbypolicyquestionshereifitseemsthatideasdiffuseinarel-ativelystraightforwardandrapidmannerhere.ThequalitativeworkthathasbeendoneonthisquestiontendstoreinforcetheKuhnianobservationthattheadoptionofnewideas,particularlyintheearlyyears,isfarfromstraightforwardandrapid.Thequantitativeevidencesuppliedhereismoremixed.

Withrespecttotheeconometrictechnique,thepat-ternofdiffusionhasbeenrelativelystraightforward,mimickingthatfoundinthestudyofmarketprocessesinindustry.Itmayhavebeenalittleslowrelativetoindustrialspeeds,butthiscouldhavearisenfromtherelativeconfigurationofcostsandbenefitsassociatedwiththeadoptionofthisparticulartechnique.Thereisalsoaninterestingdifferenceinthespeedofdiffu-sionacrossjournals,muchliketheleadingandlaggingfirmsrevealedinindustrystudies.Thesedifferences,though,haveanationalaspect,withdiffusionoccur-ringmorerapidlyonaverageinUSjournalsthanUKones,andthismightsuggestthatthedifferencesinlocalacademicinstitutionshaveaffectedtheoperationoftheacademic‘market’inthesecountries.Themorelikelyexplanationisaneditorialbiasinthejournalsselectedfromthosecountries(andnotbroadercountryinstitu-tionaldifferencesoverthewayacademicknowledge

S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–16321631

productionisconstitutedinthesecountries).Thusonthebasisoftheevidenceofeconometrictechnique,weconcludethatthemetaphorofthe‘marketplaceforideas’seemssuitedtotheacademy;andthefunction-ingofthemarketseemstohavebeenaffectedbytheeditorialprocessesofkeyjournals.

Withrespecttotheexperimentaltechnique,mattersareratherdifferent.Thepatternofdiffusionisquiteunlikethatfoundinindustrystudies,suggestingthatthemetaphorofthe‘marketplaceofideas’doesnotapply(orifitdoes,themarketisworkinginanunusualway).Again,thereisasignificantdifferenceintheroleplayedbyparticularjournalsandthisreinforcestheearlierthoughtthattheprocessofdiffusionintheacademyhasbeenaffectedbytheeditorialpracticesofjournals.

Ofcourse,itisdangeroustogeneraliseonthebasisoftwoobservations.Butsomeevidenceisbetterthannoneand,withthatqualification,whatthisstudysug-gestsisthatdiffusionseemsneitheralwaysstraight-forwardnoralwaysrapidintheacademy.Thereisanobviousbroadpolicyquestionconcerninghowthepro-cessofdiffusionmightbeinfluencedthatarisesfromthisconclusionandthepaperpointsinoneparticulardirection.Howareandshouldtheeditorialfunctionofjournalsbeorganized?

Thisisnotanissue,withtheexceptionofHodgsonandRothman(1999),thathasreceivedmuchrecentattentionandsoitdeservesfurtherscrutiny.Theyfindthat‘top’journalsineconomicshavebecomeoligopolisticinthesensethattheyaredominatedbyeconomistswhohavebeentrainedandteachatjustafewinstitutionsandtheyworrythatthismightunder-minethedynamismofthedisciplinebyreducingthevarietyofnewideas.Thisisnotobviouslyconsistentwiththefindingherethatitisthese‘top’journalsthathavebeeninthevanguardofthediffusionofbothtech-niques,butthereisonewayinwhichtheevidenceontheexperimentaltechniquemightsupporttheiranal-ysis.Oligopoliesencourage‘collusion’andalthoughthisisusuallyregardedasdamaging,itneednotbe.Inparticular,‘collusion’,broadlyunderstood,mighthelpexplainwhythepatternofdiffusionfortheexperimen-taltechniquewasquiteunlikethatfoundinmarkets.Thusitmightbeargued,inawaythatinvitesamoresociologicalapproach,thatthisrelativelysmallgroupofeconomistswerequicktoappreciateandsharetheappreciationofthevirtuesoftheexperimentaltech-

nique.Asaresult,thetechniquediffusedquicklyandthisiswhyitsusehashovered,withoutshowinganytendencytoincrease,aroundthesamelevelofdiffusioninthesejournalsforover20years.Acknowledgments

TheauthorswouldliketothankBarbaraEberthandAmyShorefortheirresearchassistanceand,inthecaseofAshokParikh,theLeverhulmeTrustforfundinghisfellowship.References

Akerlof,G.,1970.Themarketfor‘lemons’:qualityuncertainty

andthemarketmechanism.QuarterlyJournalofEconomics84,488–500.

Barnes,B.,Bloor,D.,Henry,J.,1996.ScientificKnowledge:ASoci-ologicalAnalysis.AthlonePress,London.

Barnes,B.,Shapin,S.,1979.NaturalOrder.Sage,London.

Dasgupta,P.,David,P.,1994.Towardsaneweconomicsofscience.

ResearchPolicy23(5),487–521.

Davies,S.,1979.TheDiffusionofProcessInnovations.Cambridge

UniversityPress,Cambridge.

Fama,E.,1970.Efficientcapitalmarkets:areviewoftheoryand

empiricalwork.JournalofFinance25,383–417.

Geroski,P.A.,2000.Modelsoftechnologydiffusion.ResearchPol-icy29,603–625.

Hodgson,G.,Rothman,H.,1999.Theeditorsandauthorsofeco-nomicsjournals:acasestudyofinstitutionaloligopoly?Eco-nomicJournal109,165–186.

Karshenas,M.,Stoneman,P.,1995.Technologicaldiffusion.In:

Stoneman,P.(Ed.),HandbookofEconomicsofInnovationandTechnologicalChange.BasilBlackwell,Oxford.

Keuzenkamp,H.,1991.AprecursortoMuth:Tinbergen’s

1932modelofrationalexpectations.EconomicJournal101,1245–1253.

Levin,S.,Stephan,P.,Walker,M.,1995.Planck’sprinciplere-visited:

anote.SocialStudiesofScience25,275–283.

Mansfield,E.,1968.IndustrialResearchandTechnologicalInnova-tion.Norton,NewYork.

Mansfield,E.,19.IndustrialrobotsinJapanandtheUS.Research

Policy18,183–192.

Messeri,P.,1988.Agedifferencesinthereceptionofnewscien-tifictheories:thecasestudyofplatetectonics.SocialStudiesofScience18,91–112.

Morgan,M.,1994.TheHistoryofEconomicIdeas.CambridgeUni-versityPress,Cambridge.

Muth,J.,1961.Rationalexpectationsandthetheoryofpricemove-ments.Econometrica29,315–335.

Nasbeth,L.,Ray,G.,1974.TheDiffusionofNewIndustrialPro-cesses:AnInternationalStudy.CambridgeUniversityPress,Cambridge.

1632S.P.HargreavesHeap,A.Parikh/ResearchPolicy34(2005)1619–1632

Stoneman,P.(Ed.),1995.HandbookofEconomicsofInnovationand

TechnologicalChange.BasilBlackwell,Oxford.

Stoneman,P.,2001.TheEconomicsofTechnologicalDiffusion.

BasilBlackwell,Oxford.

VonTunzelman,G.,1978.SteamPowerandBritishIndustrialisation

to1860.ClarendonPress,Oxford.

Stephan,P.,1996.TheEconomicsofscience.JournalofEconomic

Literature34(3),1199–1262.

Stewart,J.,1986.Driftingcontinentsandcollidinginterests:aquan-titativeapplicationoftheinterestperspective.SocialStudiesofScience16,261–279.

Stigler,G.,Stigler,S.,Friedland,C.,1995.TheJournalsofEco-nomics.JournalofPoliticalEconomy103(2),331–359.