Was the Wealth of Nations Determined in 1000 B.C.?

Diego A. Comin William Easterly Erick Gong

Working Paper
09-052

Copyright 2008 by Diego A. Comin, William Easterly, and Erick Gong Working papers are in draft form. This working paper is distributed for purposes of comment and discussion only. It may not be reproduced without permission of the copyright holder. Copies of working papers are available from the author.

WastheWealthofNationsDeterminedin1000B.C.?*

DiegoComin

WilliamEasterly
September2008

ErickGong

Abstract
Weassembleadatasetontechnologyadoptionin1000B.C.,0A.D.,and1500 A.D. for the predecessors to todays nation states. We find that this very old history of technology adoption is surprisingly significant for todays national development outcomes. Our strong and robust results are for 1500 A.D. determining per capita income today We find technological persistence across longepochs:from1000BCto0AD,from0ADto1500AD,andfrom1500ADto the present. Although the data allow only some suggestive tests of rival hypothesestoexplainlongruntechnologicalpersistence,wefindtheevidence tobemostconsistentwithamodelofendogenoustechnologyadoptionwhere the cost of adopting new technologies declines sufficiently with the current level of adoption. The evidence is less consistent with a dominant role for populationaspredictedbythesemiendogenousgrowthmodelsorforcountry levelfactorslikeculture,genesorinstitutions Keywords:Technologyadoption,technologyhistory,economicdevelopment. JELcodes:O3,N7.

* WearethankfultoTobiasPfutzeforcommentsandresearchassistance,toTomokoWadaandJasonZhanShifor theirresearchassistance,forcomments,toPhilippeAghion,MichaelClemens,RafaelDiTella,AndrewGelman,Simon Gilgricht,GeneGrossman,PeteKlenow,MichaelKremer,RobertLucas,NathanNunn,PeterPeregrine,Louis Putterman,RomainWacziarg,DavidWeilandseminar/conferenceparticipantsatJohnsHopkins,Brown, Harvard/MIT,HBSandtotheNSF(Grant#SES0517910)andtheC.V.StarrCenterforAppliedEconomicsfortheir financialsupport. HarvardBusinessSchoolandNBER. NewYorkUniversityandNBER. UniversityofCaliforniaatBerkeley.

1.Motivation
Crosscountrydifferencesinpercapitaincomeinthepreindustrialworldweresignificantly smallerthantoday.Forexample,accordingtoMadison(2000),percapitaincomeintheUKin 1500ADwassixtyeightpercenthigherthaninMexico.In2000,itwasalmostthreetimeslarger. Thisobservationhasmotivatedtwoconclusions.First,thatpreindustrialcountryconditions, andinparticular,thetechnologyadoptionlevel,didnotvarymuchacrosscountries.1Second, thatpreindustrialcountryconditionsarealmostirrelevanttounderstandcurrentdevelopment dynamics.Aconsequenceoftheseconclusionsistheemphasisofeconomicdevelopment practitionersandstudentsonmoderndeterminantsofpercapitaincomelikequalityof institutionstosupportfreemarkets,economicpolicieschosenbygovernments,humancapital componentssuchaseducationandhealth,orpoliticalfactorslikeviolenceandinstability. Couldthisdiscussionbemissinganimportant,muchmorelongrundimensiontoeconomic development?Totheextentthathistoryisdiscussedatallineconomicdevelopment,itis usuallyeitherthedivergenceassociatedwiththeindustrialrevolution(e.g.Lucas,2000)orthe effectsofthecolonialregimes.2Isitpossiblethatprecolonial,preindustrialhistoryalsomatters significantlyfortodaysnationaleconomicdevelopment? Thispaperexploresthesequestionsbothempiricallyandtheoretically.Tothisend,weassemble anewdatasetonthehistoryoftechnologyover2500yearsofhistorypriortotheeraof colonizationandextensiveEuropeancontacts.Thefirstsignificantfindingisthattherewere importanttechnologicaldifferencesbetweenthepredecessorstotodaysmodernnationsas longagoas1000BC,andthatthesedifferencespersistedto0ADandto1500AD(whichwillbe thethreedatapointsinourdataset).Further,theseprecolonial,preindustrialdifferenceshave strikingpredictivepowerforthepatternofbothpercapitaincomesandtechnologyadoption acrossnationsthatweobservetoday.Ourstrongestresultsareforthedetailedtechnology datasetweassemblefor1500AD,whichisanexcellentpredictorofpercapitaincometoday. Wefindthattheresultsfor1500ADlargelycontinuetoholdwhenweincludecontinent dummiesandgeographiccontrols. Twoquestionsnaturallyfollowupourfindings.First,whatmechanismspropagatehistorical shocksthataffectthehistoryoftechnologyadoptionintothepresent?Second,whatdoour findingsteachusaboutexistinggrowthmodels?Toanswerthesequestion,wepresentasimple modelwherethestockoftechnologiesadoptedreducesthecostofadoptingnewtechnologies. Dependingonthestrengthofthiselasticity,historicaltechnologyadoptionwillorwillnothavea significanteffectontheadoptionoftechnologiesthathavecomealongsincetheindustrial revolutionandhenceoncurrentdevelopment.Hence,ourempiricalexplorationprovidesan estimateforthiselasticity.
1

Thisconclusion isnotnecessarilytrueasfollowsfromtheworkofKremer(1993),GalorandWeil(2000)andJones (2001,2005)andHansenandPrescott(2002).AllthesepapershaveMalthusianmodelswherefertilityincreaseswith technology.Inthiscontext,countriesmayhavesignificantdifferencesintechnologybutverysmalldifferencesinper capitaincome. 2 Anotable,honorable,andfamousexceptionisJaredDiamond(1997).

Oursimplemodelalsohelpsusthinkaboutalternativehypothesisthatmaygeneratethe observedimportanceofhistoricaltechnologyadoptiononcurrentdevelopmentand,more importantly,itguidesusinhowtoidentifytheminthedata. Onesuchexplanationisrelatedtoideasincreasingwithpopulationbecausetherearemore potentialthinkerstocreatethese(nonrival)ideas.Moreideasinturncreatesbettertechnology whichcansustainalargerpopulation,andthenthevirtuouscirclecontinues.Thisideadates backtoSimonKuznets,andJulianSimonandhasbeenformulatedmorerecentlybyKremer (1993),GalorandWeil(2000)andbythesemiendogenousmodelsofJones(1995,1999,2001, 2005),Kortum(1997),andSegerstrom(1998).Areducedformofthesepopulationandideas modelsintechnologyalonewouldalsoshowtechnologicalpersistence,intermediatedby population,andsoourfindingscouldprovideanotherpieceofevidenceinsupportofthese models. However,wethinkitisinterestingtoidentifywhetherthepersistenceoftechnologyalsoresults fromadirectpropagationmechanismofoldtonewtechnology,asopposedtobeingfully accountedforbythepopulationchannel.Sowecontrolforpopulationinourregressions.We findthatboththepersistenceoftechnologiesandtheeffectofhistoricaltechnologyadoption oncurrentdevelopmentpersistaftercontrollingforhistoricalpopulationandforbothhistorical andcurrentpopulation.Hence,ourfirstexplorationofthedatasuggestsadirecteffectofpast technologyonfuturetechnologythatisnotfullyexplainedbythepopulationchannel. Otherpotentialpropagationmechanismscurrentlydiscussedintheliteratureareinstitutions, cultureorthegeneticendowment.3Toidentifywhethertechnologyortheseotherfactorsare propagatinghistoricalshocksintothepresent,weexploitthecrosssectorvariationin technologyadoptionandexplorewhetherthepersistenceoftechnologyadoptionholdsafter includingcountryyeareffectsandcountrysectorfixedeffects.Wefindnotonlythatthe persistenceoftechnologyholdsalsowithinsectorsbutthatthemagnitudeofthiseffectis comparabletotheestimatedpersistencebeforeincludingcountryeffects.Further,the persistencewithinsectorsisnotdrivenbyanyspecificsector.Inourview,thesefindingsare suggestivethatthereisadirecttechnologyeffectevenifinstitutions,genesorculturealso mattersince,iftheselatterfactorsaccountedforallthepersistence,theywoulddosothrougha commoneffectontechnologyadoptionacrossallsectors.Thisishardtoreconcilewithour findingthatthemagnitudeoftheeffectofhistoricaltechnologyisvirtuallyunchangedafter includingthecountryeffects.Itisplausiblethattheknowledgefrompreviousadoption experiencehasanimportantsectorspecificitywhichwouldrationalizethewithinsector persistenceinadoption.
3 SpolaoreandWacziarg(2006)haveafascinatingexplorationoftheeffectofgeneticdistanceonlogincome distance.Theytakegeneticdistanceasadifferencebetweenallcharacteristicsverticallytransmittedfromparentsto children(notonlygenetic,butevenmoreimportantlycultural),andsuggestthatdifferencesinthesecharacteristics actasabarriertotechnology/developmentdiffusion.Theyfindthatcountriespopulatedbymoregeneticallydistant culturesalsohavemoredifferentpercapitaincomes.Thisfindingiscomplementarytooursbecausegeneticdistance isverypersistentandwasdeterminedinadistantpast.Itdiffers,however,inatleasttworespects.First,weexplore theeffectsoftechnologyadoptionhistoryoncurrentdevelopment.Ourlefthandsidevariablehasadirecteffecton development,whilegeneticdistancesurelydoesnothaveadirecteffectondevelopment.Itisaproxyforcostsof transferringtechnology.Second,byexploringarelationshipinlevelsweareabletopreservethetransitivityofour measures.Thisisnotthecasewhenlookingatdistances.AshrafandGalor(2008)haveanotherinterestinggenetic story,stressingthedifferencesingeneticdiversityacrossregionsoftheworldasadeterminantofcomparative economicdevelopment,throughthebeneficialimpactofdiversityontechnologicalcreativity.

 Insummary,whatourfindingsteachusisthattechnologyisremarkablypersistentoverthevery longrun,andthatverysignificantinputsthatshapethecurrentcrosscountrydistributionofper capitaincomeweredeterminedalongtimeagoandwerelikelytobepropagatedintothe currenttimesbythedynamicsoftechnologyadoption. OurpaperisclearlyrelatedtotheinfluentialbookbyJaredDiamond(1997)Guns,Germs,and Steel.OneimportantthesisinDiamondsbookisthatcurrentdevelopmentistiedtoancient technologies.Diamond,however,doesnotsystematicallytesttheeffectofancienttechnologies onmodernincomesorstudiesthepropagationmechanismsaswewilldohere.Perhapsforthat reason,theDiamondworkdidnotchangemuchthetendencyofdevelopmenteconomicsto focusonthemodernperiodoratmostthecolonialperiod. Inasimilarvein,economichistorianshavebeendebatingtheimportanceofpasttechnology adoptionfortheadoptionofsubsequenttechnologies,especiallywhattriggeredtheIndustrial RevolutioninEurope.Mokyr(1990,p.169)andRosenbergandBirdzell(1987)arguethat technologicalexperienceisalongwayfrombeingasufficientconditionfortheEuropean miracle,stressingtheearliertechnologicalleadofChina.Greene(2000),instead,arguesthat,in theWest,GrecoRomandynamismwaspartofalongcontinuumfromtheEuropeanIronAgeto medievaltechnologicalprogressandtheindustrialrevolution.Mokyr(1990,p.164)alsonotes howmanytechnologicaladvancespeteredoutthroughouthistorywithoutleadingtoa permanentstreamofinnovation. However,whileeconomichistoriansdisagreeonaninitialtechnologyadvantageasasufficient causeoftheindustrialrevolution,theirdescriptionoftechnologyhistoryreachesaconsensuson manymechanismsthatcausepasttechnologytohaveaneffectonfuturetechnology.(Thereis nocontradictionherepasttechnologycouldmatter,andyetnotbesufficienttoexplainwhy EuropeandnotChinaexperiencedtheindustrialrevolution,whichcouldalsodependonother factorssuchasinstitutionsandvalues).Specifically,theyprovidemanycasestudiesof technologicalinnovationtodocumentmechanismsthatsupporttheassumptionofourmodel thatahigherinitialtechnologylevellowersthecostofadoptingnewtechnologies(orinafew cases,raisesthebenefitsofnewtechnologies). INSERTTBLE1HERE InTable1,welistthesurprisinglylonglistofsuchmechanismsdetailedbyeconomichistorians (althoughsomeofmechanismsarerelatedtoeachother),alongwithmanyoftheillustrative exampleshistorianshavegivenfromtechnologyhistory.Theexamplesareskewedtowardsthe IndustrialRevolutionandaftersincethatiswherehistorianshavefocusedmostoftheirenergies andthehistoricalrecordismostcomplete(althoughmanyGreek/Romanandmedieval examplesdoappear).Itispossiblethatsomeofthemechanismsonlystartedtooperateatthe timeoftheIndustrialRevolution.However,wethinkitisplausiblethattheIndustrialRevolution wasonlyaspeedingupoftechnologydynamics(whichwewilldiscussbelow),andthatmanyof thesedynamicmechanismswillapplytoearliertechnologicaleras.Ifso,thenwehaveastronga prioricaseforthepersistenceoftechnologyoverlongperiods,whichwecantestagainstour longrundataset.

Ourempiricalmethodologyisverydifferentfromthecasehistoriesofthetechnologyhistory literature;wedontclaimittobesuperior,onlyavaluableadditionalmethodoftestingthe hypothesisoftechnologypersistence.Clearly,oneimportantbarrierthatmayhaveprevented otherresearchersfromimplementingthisstrategyisthelackofadataset.Oneofour contributionsistoassembleadatasettoexploremoredefinitelytheseissues. Therestofthepaperisorganizedasfollows.Section2presentsthedataset.Section3uncovers themainfindingsandshowstheirrobustness.Section4presentsasimplemodelthat rationalizesthefactsandsomeextensionsthatyieldsomeadditionalpredictionsthatallowus toidentifyseveralcompetinghypothesisaboutthenatureofthepropagationmechanism. Section5concludes.

2.Descriptionoftechnologydataset
Thehistoricaldatasetspresentedinthispapermeasurethecrosscountryleveloftechnology adoptionforover100countriesinthreeperiods:1000B.C.,0A.D.andtheprecolonialperiodin 1500A.D.Eachdatasetactsasasnapshotintime,capturingthelevelsoftechnology adoptionbycountrythroughouttheworld. Technologyadoptionismeasuredontheextensivemarginbydocumentingwhetheracountry usesaparticulartechnology,nothowintensivelyitisused.Forexample,inthedatasetfor1000 B.C.,weconsidertwotransportationtechnologies:packanimalsandvehicles.Acountryslevel oftechnologyadoptionintransportationisthendeterminedbywhethervehiclesand/ordraft animalswereusedinthecountryatthetime.Thetechnologiesthatweexaminechange betweentheancientperiod(1000B.C.and0A.D.)totheearlymodernperiod(1500A.D.)to reflecttheevolutionofthetechnologyfrontier. Ourfocusontheextensivemarginoftechnologyadoptionismotivatedbydataavailability constraintsandbyitshistoricalrelevance.Itismucheasiertodocumentwhetheratechnology isbeingusedinacountry(theextensivemargin)ratherthanmeasuringthedegreeofits adoption(theintensivemargin).4Inaddition,theextensivemarginhasbeenupto1875orsoa muchmoresignificantmargintoexplainthecrosscountryvariationintechnologyadoption (PulkkiandStoneman,2006). Thetechnologiesinourdatasetsarestateofthearttechnologies(atthetime)inproductive activities(i.e.activitiesthatenteredGDP)andforwhichithasbeenpossibletodocumentits presenceorabsenceforawiderangeofcountries.Ofcourse,itisaveryincompletelistsinceit doesnotcoverallthesignificantfrontiertechnologiesavailableatthetime.However,the numberoftechnologiescovered(12for1000BCand0ADand24for1500AD)issufficientto makepreciseinferencesaboutthetechnologicalsophisticationofeconomiesinthedistantpast. Morefundamentally,thecriteriaforincludingtechnologiesinouranalysisdonotgenerateany biasinourresultsbecausetheyinvolvetheclosenesstotheWorldtechnologyfrontierinthe distantpastwhich,ofcourse,isverydifferentfromthecurrenttechnologyfrontier.5
4 ItiswelldocumentedthattheChinesewereusingironfortoolsby0A.D;whatismoredifficulttoassessistheshare oftoolsconstructedfromironatthetime. 5 Anargumentinlinewiththeappropriatetechnologyliteratureisthatitisoptimalthatsometechnologiesarenot presentinsomeregions.Diamond(1997)forexample,arguesthathorseswereineffectiveinthetropics.Giventhe

 Arelatedissueisthatsomesectorsaremoredenselycoveredthanothers(i.e.for1500ADwe have8technologiesinmilitarybutonly2inmetalworking).Toavoidoverweightingsectors wherewehavebeenabletocollectdataonmoretechnologies,wecomputetheaverage adoptionrateineachsectorandthencomputetheoveralladoptionlevelbyaveragingthe sectoraladoptionlevels.Wehavealsoexperimentedwithalternativeaggregationapproaches obtainingverysimilarresults. Sinceourmainobjectiveistoanalyzetheeffectsthathistorictechnologyadoptionhasonthe currentstateofeconomicdevelopment,ourdatasetsarepartitionedusingmoderndaynation states.WeusethemapsfromtheCIAsTheWorldFactbook(2006)toputintoconcordancethe bordersofpresentdaynationswiththeculturesandcivilizationsin1000B.C.,0A.D.and1500 A.D.Forexample,thetechnologiesusedbytheAztecsandtheirpredecessorsduringpre colonialtimesarecodedastheonesusedbyMexicoin1500A.D.Incaseswhereacountryhad multiplecultureswithinitsbordersduringacertaintimeperiod,wetaketheculturewiththe highestleveloftechnologyadoptiontorepresentthatcountry.Thistechniqueisadirect consequenceofourgoalofmeasuringtheextensivemarginoftechnologyadoptionina country.Forexample,in1000B.C.thereweremultipleculturesresidingwithinCanadas moderndayborders.TheInitialShieldWoodlandwasthemosttechnologicallysophisticatedof theseculturesandwethereforeuseitsleveloftechnologyadoptiontorepresentCanadain 1000B.C. Theuseofthemostadvancedculturewithinaterritoryforacountrysleveloftechnologycould induceamechanicalcorrelationbetweentechnologyandcountrysize(asmeasuredeitherby populationorlandarea).Thelargerthesize,themoreculturesarebeingsampled,whichmakes themaximumofallcultureshigher.Forpopulation,thismechanicaleffectisreallythe KuznetsSimoneffectofpopulationontechnologymentionedintheintroduction,ifthemost advancedtechnologiesdoindeeddisseminatewithinthebordersofwhatistodaymeasuredas acountry.Wewilltestforthiseffectinourempirics.Forlandarea,thisalsocouldreflectareal economicphenomenonforthesamereasons,butitwouldinducereversecausalitybetween landareaandtechnology.Wewillexaminesomesimpletestsastowhetherthisaffectsour resultsintheempiricalsection. EachdatasetisconstructedfollowingthemethodologyusedbyGeorgeMurdockandother ethnologists(Murdock1967;Carneiro1970;TudenandMarshall1972;BarryandPaxson1971). Eachdatasetiscodedbyateamofresearcherssurveyingmultiplesourcesreducing,inthisway, thelikelihoodofmeasurementerror.Researcherstakedetailednotesincludingdirect quotationsandusing,whenappropriate,twoinferencetechniques:technologicalcontinuity (Basalla1988)andtemporalextrapolation(Murdock&Morrow1970:314).

technologiesinoursample,weviewthisargumentasveryannecdotal.Itisquiteclearthatvirtuallyallofthe technologiesinthethreehistoricaldatasetshavea(large)positiveeffectonlaborproductivityinvirtuallyall countries.Arelatedpointisthatourempiricalexplorationdoesnottrytoestablishthereasonsforthefailureto adoptatechnology.Rather,whetherthisfailurehadconsequencesinthe(very)longrun.Inasense,ifinhistorical timesitwasoptimalnottoadoptatechnology,itisevenmoresurprisingtofindnegativeeffectsinthe(very)long run.

Technologicalcontinuitystressesthatinnovationsarearesultofpreviousantecedents; innovationstypicallydonotspontaneouslyarisewithoutpreexistingtechnologies.6Weusethis techniquetoinferthatcountrieswithadvancedtechnologiesinaparticularsectoralsohad moreprimitiveones.Oneexamplethatillustratesthistechniquecomesfromthemilitary technologiesin1500A.D.Largewarshipswithover180gunsondeckwereconsideredthe pinnacleofmilitarytechnologyin1500A.D.(Black1996).Itisnotunreasonabletoassumethat acountrywithheavilyarmedwarshipsalsohadaccesstofieldartilleryandmuskets. Therefore,inPortugalandGermany,thepresenceoflargewarshipswasusedtoinfertheuseof bothmusketsandfieldartillery.Temporalextrapolationassumesthatatechnologymaintains somelevelofpersistenceovertime.Atechnologyadoptedfiftytoonehundredyearsearlieris assumedtostillbeinuse.7Inaddition,inmostofthecases,weareabletodocumentthatthe technologywaspresentin1550.Anexampleofthisisthecodingoftransportationtechnologyin 1500Turkey.WecodeTurkeyashavingthemagneticcompassinthe1500A.D.datasetbased onevidencethatitwasinuseintheOttomanEmpireby1450. Thedatasetsfor1000B.C.and0A.D.arederivedfromtheAtlasofCulturalEvolution8 (henceforwardabbreviatedasACE).(Peregrine2003)whilewecodedthedatasetfor1500 A.D.initsentirety.TheACEitselfisbasedontheEncyclopediaofPrehistory(Peregrine&Ember 2001)whosecompilationinvolvedmultipledatasourcesandover200researchers.The1500 A.D.datasetinvolvedseveralresearchersandover200sources. Itisimportanttonotethat,inamajorityofcases,thecodingoftechnologyadoptionisbasedon directevidenceofthepresenceorabsenceoftechnologiesinthecountries.Arelevant considerationcouldariseifwehadacivilizationcoveringvariousmoderndaycountriesandwe didnothaveanysourceofevidencethatthecodeforthecivilizationappliestoalltheindividual countries.Inthisevent,thestandarderrorsforourregressionswouldbemisleading.Toavoid thisproblem,for1500AD,wehavesearchedfordocumentationthatallowsustodeterminethe presenceorabsenceonthecountriesthatcorrespondtohistoricalempires.Forexample,for thecountriesthatcomposedtheOttomanempirein1500,wehavedocumentedthepresence orabsenceineachofthecountries. Inafewcases,wehavenotbeenabletodocumentdirectlytheabsenceofsomespecific technologyinagivencountryX.Inthesecases,however,wehavedirectevidenceofthe absenceofthetechnologyissomeneighboringcountryYthathasbeenproventodominate technologicallycountryXatthetime.Thisallowsustoinfertheabsenceofthetechnologyin countryX. AnexampleofthisisinourcodingofcommunicationsinSouthAmerica.Wehavelittle informationaboutthediffusionofthesetechnologiesin1500ADinBrazil,Uruguay,Paraguay, andColombia.SincetheIncaswerethemostadvancedcivilizationinSouthAmericaduringthat time,anytechnologiesabsentfromIncacivilizationwereassumedtobeabsentinthecountries
SeeBasalla(1988:3057)foranumberofcasestudiesdocumentingtechnologicalcontinuityortechnological evolution. 7 This_time_frame_rules_out_long_run_technological_regression_such_as_the_loss_of_some_Roman_achievements _inmedieval_EuropeortheChineseoceangoingvoyagestoEastAfrica. 8 Peregrine(2003)usesBP(BeforePresent)asthetimevariablewhencodinghisdatasets.WeconverttheBPtime periodstoeitherB.C.orA.D.Peregrines3000BPdatasetisusedforour1000B.C.datasetandPeregrines2000BP datasetisusedforour0A.D.dataset.
6

listedabove.TheIncasreliedonQuipus(lengthsofstringknottedatintervals)for communicationsandhadnowrittenrecords(Scarre1988:222;EncyclopediaBritannica2006). ThisimpliesthattheIncasdidnothavebooksormovabletypeprinting.Usingthisgeographic dominanceargument,weinferthatBrazil,Uruguay,Paraguay,andColombiadidnothavethese technologieseither. Thisgeographicdominanceargumentmight,however,haveconsequencesonthecomputation ofthestandarderrorsinourregressions.Despitethefewoccasionswherewehaveusedthe geographicdominanceargument,weavoidthisissuebyclusteringtheerrorsaroundtheareas oftechnologicalinfluenceusedtocodethedata.Further,wehavecheckedthatourresultsare robusttoclusteringtheerrorsbycontinents. Finally,therearetwopotentialconcernsininterpretingourdatathatwewanttoaddress directly.Thefirstisthatcountriesthatweremoreadvancedatthetimeweremorelikelyto leaverecords.Thesecondisthatcurrentlyrichcountriesmaybemorelikelytofindremainsthat documenttheexistenceoftechnologiesinthepast. Wedonotbelievethateitheroftheseconcernshasasignificantinfluenceonourdataset.This conclusionisbasedonthreereasons.First,rememberthatweusedirectevidenceofthe absenceofthetechnologiestocodethatthetechnologywasnotpresentinacountry.Thatis, lackofevidenceonthepresenceofthetechnologyisnotsufficienttocodeitsabsence.Second, moderndayarcheologistsdigwherevertheybelievetheycanfindremainsregardlessofthe originofthearcheologists.Indeed,mostofthemainarcheologicaldiscoveriesareindeveloping countriesandhavebeenfoundbyarcheologistsfromdevelopedeconomies.Finally,asweshow insection4.3,ourfindingsaboutthepersistenceoftechnologyadoptionholdevenwhenwe includecountry(fixedand/ortimevarying)effectsandexploitthecrosssectoralvariationin technologyadoption.Thattakescareofanycountrylevelbiasincludingbiasesinthereporting orcollectionofdata.

2.1TechnologyDatasetsfor1000B.C.and0A.D.
Thedatasetsfor1000B.C.and0A.D.measuretheleveloftechnologyadoptionforagriculture, transportation,communications,writing,andmilitaryon113and135countriesrespectively. TheACEdoesnotcontainanyvariablethatmeasuresdirectlythetechnologiesusedfor militarypurposes.Toassessacountrysleveloftechnologyadoptionforthemilitaryweusethe ACEdatasettodeterminewhichmetalswereavailableforeachculture.Metallurgyisintegral forthedevelopmentofmoreadvancedweapons(Macksey1993:216;Scarre1988;Collis 1997:29).Theprogressionfromstonetobronzeandfinallyironcorrespondedtoaprogression ofmorepowerfulweapons;stoneweaponswerereplacedbybronzeswordsanddaggers;iron weaponswereconsiderablystrongerthantheirbronzepredecessors(Hogg1968:1922).The relevantdatafromtheACEandhowitisusedcanbefoundinTable2. _________________InsertTable2here_________________________ Anexampleofhowacountryiscodedin1000B.C.and0A.Dwillbestillustrateour methodology.

KoreawasinhabitedbytheMumunpeoplesin1000B.C.TheMumunshadnotraditionof eitherwritingornonwrittenrecords.TheMumunshoweverdidrelyonagricultureasits primaryformofsubsistenceandusedpackanimalsfortransportation.InadditiontheMumuns producedmetalworkandusedbronzefortoolsbutnotiron(Rhee2001).Thecodingforthe MumunentryintheACEdataset(Peregrine2003)thereforeis: WritingandRecords=1 TechnologySpecialization=3 LandTransportation=2 Agriculture=3 Tools=2 Basedonthisdata,wecodeKoreain1000B.C.as: Communication:Mnemonicornonwrittenrecords=0;TrueWriting=0 Industry:Pottery=1;Metalwork=1 Transportation:Packordraftanimals=1;Vehicles=0 Agriculture:10%ormore,butsecondary=1;Primary=1 Military:Bronzeweapons=1;Ironweapons=0 Weaggregatethetechnologyadoptionmeasuresatthesectorlevelbyaddingalltheindividual technologymeasuresinthesectoranddividingthesumbythemaximumpossibleadoption levelinthesector.Inthisway,thesectoraladoptionindexisintheinterval[0,1].Theoverall adoptionlevelineachcountryandtimeperiodistheaverageoftheadoptionlevelacross sectors.Obviously,theoveralladoptionindexalsobelongstotheinterval[0,1]. TheadoptionlevelsinthefoursectorsjustreportedinKoreain1000B.C.arethefollowing: Communications=0 Industry=1 Transportation=0.5 Agriculture=1 Military=0.5

AndtheoverallleveloftechnologyadoptionforKoreain1000BCis0.6. 2.2TechnologyDatasetfor1500A.D.
Thetechnologydatasetfor1500A.D.encompasses113countriesandevaluatesthelevelof technologyadoptionacrossthesamefivesectors(agriculture,transportation,military,industry, andcommunications)asthepreviousdatasets.OurtechnologymeasuresoutsideEuropeare estimatedbeforeEuropeancolonization.Itisimportanttostress,therefore,thatourtechnology measuresin1500A.D.donotincorporatethetechnologytransferredbyEuropeanstotherest oftheworldafterEuropeanexplorationbeganaround1500. Obviously,therearealargernumberofsourcescoveringthetechnologyadoptionpatternsin 1500A.D.thanin1000B.C.or0A.D.Thisallowsustocollectadoptiondatafor24technologies

inthefoursectorsotherthanagriculturevs.the11technologiescoveredinthedatasetsfor 1000B.C.and0A.D.Asaresult,ourestimateoftheleveloftechnologyadoptionin1500A.D.is likelytobemoreprecisethanfortheearlierperiods.Notethat,asbefore,ourmeasuresattach equalweighttoeachofthefivesectors,soouroverallaverageisnotbiasedtowardssectorsin whichmoretechnologicalinformationisavailable.Soforexample,moreinformationis availableonmilitarytechnologies(8)thanindustrialtechnologies(2),butmilitaryandindustry haveequalweightsinouroverallindex.Forthesamereason,closelyrelatedtechnologieswithin asectorthatmightriskdoublecountingdonotincreasetheweightofthatsectorinour overallindex.Table3presentsthevarioustechnologiesmeasuredin1500A.D.9,10 ________________INSERTTABLE3HERE________________

2.3CurrentTechnology
Toexplorewhetherhistoricaltechnologydifferenceshavepersisteduntilcurrenttimes,we constructameasureofcurrenttechnologylevelbasedonComin,HobijnandRovito(2006).This measurecaptures(oneminus)theaveragegapintheintensityofadoptionoftenmajorcurrent technologieswithrespecttotheUS.Thesetechnologiesareelectricity(in1990),internet(in 1996),pcs(in2002),cellphones(in2002),telephones(in1970),cargoandpassengeraviation (in1990),trucks(in1990),cars(in1990)andtractors(in1970)allinpercapitaterms. Morespecifically,foreachtechnology,Comin,HobijnandRovito(2006)measurehowmany yearsagodidtheUnitedStateslasthavetheusageoftechnologyxthatcountryccurrently has.Wetaketheseestimates,normalizethembythenumberofyearssincetheinventionofthe technologytomakethemcomparableacrosstechnologies,taketheaverageacrosstechnologies andmultiplytheaveragelagbyminusoneandaddonetoobtainameasureoftheaveragegap intheintensityofadoptionwithrespecttotheUSwhoseadoptionlevelisone,byconstruction. Notethatthismeasureofcurrenttechnologyadoptiondiffersfromthehistoricalmeasuresin thatitincludestheintensivemargin.Thisisthecasebecauseinthelast100yearsorso,the extensivemarginoftechnologyhasdiffusedveryquicklyacrosscountries.Therefore,the intensivemarginoftechnologyadoptionistherelevantmargintoexplaincrosscountry differencesintechnology. Sincerichercountriestendtodemandmoreintensivelynewtechnologies,ourmeasureof currenttechnologyadoptionwillbecorrelatedwithpercapitaGDP.However,asshownin CominandHobijn(2008b),therearemanyotherdeterminantsoftechnologyadoptionabove andbeyondpercapitaincome.Hence,theusefulnessofhavingadirectmeasureofcurrent technologytoassessthepersistenceoftechnology.
9 Inouranalysiswehaveexperimentedwithsomealternativeaggregationschemessuchascollapsingthe technologiesofshipscapableofcrossingthevariousoceansintojustonetechnologywithoutanysignificantchange inourresults. 10 Oursectoralandoveralltechnologyadoptionmeasuresareveryrobusttoreasonablevariationsinthedefinitionof newtechnologies.Forexample,wehaveexploredtheeffectofaggregatingthetechnologytocrossthevarious oceansintooneuniquevariable.Thisresultsinaadoptionmeasurefortransportationwithacorrelationof0.988with respecttoours.Similarly,collapsingheavynavalgunsandlargeshipswith+180gunsintoauniquetechnologyresults inameasureofadoptioninmilitarywithacorrelationof0.996withourmeasure.

10

 Itisalsoworthwhilenotingthat,themajortechnologiesusedtosummarizethecurrentstateof technology,belongtofourofthesectorscoveredbythehistoricaldatasets(i.e.allbutmilitary). Weshalltakeadvantageofthisfeatureofthedatawhenestimatingthepersistenceof technologywithinsectorsinsection4.3.

3.Dataanalysis

3.1Crosscountrydispersionintechnology
WestartthedataanalysisbypresentinginTable4somedescriptivestatisticsfortheoverall technologyadoptionlevelin1000BC,0AD,1500ADandinthepresent. __________________INSERTTABLE4HERE_______________________ Theincreaseinthecrosscountryaverageoftheoveralltechnologyadoptionlevelbetween 1000B.C.and0indicatesthediffusionofthetechnologiesdescribedintheACE.Recallthatthe technologyadoptiondatasetfor1500A.D.containsdifferenttechnologiesthanthefirsttwo periods.Therefore,weshouldnotcomparetheabsolutelevelin1500ADwiththeprevious levels. Table4canhelpusexplorehowlargewasthecrosscountrydispersionintechnologyadoption. Thebinarynatureofourmeasuresofhistoricaltechnologyadoptionforindividualtechnologies providestwobenchmarkstointerpretthecrosscountrydispersionintechnologyadoption.11 First,themaximumrangefortheaverageadoptionlevelacrosscountriesistheinterval[0,1];0 foracountrythathasnotadoptedanyofthetechnologiesand1foracountrythathasadopted allthetechnologies.Second,themaximumcrosscountrydispersioninadoptionwouldoccur whenhalfofthecountrieshaveadoptedallthetechnologiesandtheotherhalfhasadopted none.Inthiscasethestandarddeviationoftheaverageadoptionlevelacrosscountrieswould be0.5. Similarly,acountrythathasnounitsofagivenmoderntechnologywillhaveameasureof currentadoption(forthattechnology)of0,whileacountrythathastheUSintensityofadoption willbemeasuredas1.Inafewcountriesandtechnologiesweobservethattheintensityof adoptionishigherthanintheUSandthecurrenttechnologyadoptionlevelishigherthanone. Figures1athrough1candTable5explorefurtherthecrosscountryvariationintheoverall technologyadoptionlevel.Table5exploresthevariationacrosscontinentsinoveralltechnology adoption.Figures1athrough1cpresentaworldmapwiththeoveralltechnologyadoptionlevel ineachcountryandhistoricalperiod.Weusefourcolorstoindicatetechnologyadoptionlevels between0and0.25,between0.25and0.5,between0.5and0.75andbetween0.75and1. Darkercolorsrepresentahigheroveralltechnologyadoptionlevel.Missingvaluesare representedinwhite. _________________INSERTTABLE5HERE_________________
11

Theexceptionstothisrulearethemeasuresoftechnologyadoptioninagriculture.

11

 Inallthreehistoricalperiods,EuropeandAsiapresentthehighestaveragelevelsofoverall technologyadoption,whileAmericaandOceaniapresentthelowest,withAfricainbetween.In currenttimes,theaverageadoptionlevelishighestinEuropeandOceania(drivenbyAustralia andNewZealand),followedbyAmerica,Asia,andAfrica. ____________INSERTFIGURES1a1cHERE______________________ Aglimpsetothefiguressufficestonotethatthereissubstantialvarianceinoveralltechnology adoptionbothacrossandwithincontinents.Tomakeobservationmoreprecise,wedecompose thecrosscountryvariationinoveralltechnologyadoptionbetweenthevariationwithin continentsandthevariationbetweencontinents.In1000BC,about65percentofthevariancein overalltechnologyadoptionisduetovariationwithincontinentsand35percentdueto variationbetweencontinents.Theseproportionsarereversedin0A.D.andin1500A.D.the shareoftotalvarianceduetothebetweencontinentcomponentrisesto78percent. Table6providesamoredetailedcomparisonoftheevolutionofoveralltechnologyadoptionin themostadvancedcountries.Thesecountriescorrespondtofourhistoricalcivilizations: WesternEurope,China,theIndiancivilizationandtheMiddleEasternpeoples.WesternEurope includesSpain,Portugal,Italy,France,UnitedKingdom,Germany,BelgiumandNetherlands.The IndiancivilizationincludesIndia,PakistanandBangladesh.Finally,theMiddleEasterncivilization includesSaudiArabia,UAE,Yemen,Oman,Iraq,Iran,Turkey,Syria,Lebanon,Jordan,Egypt, Libya,Tunisia,AlgeriaandMorocco. _________________INSERTTABLE6HERE__________________ In1000B.C.theMiddleEasternempiresandChinahaveanoveralltechnologyadoptionlevelof 0.95and0.9respectively,whileinIndiaandWesternEuropetheaverageadoptionlevelare0.67 and0.65respectively.In0A.D.IndiaandWesternEuropecatchupwithChinaandtheMiddle Easternempires.In1500A.D.WesternEuropehascompletedthetransitionandisthemost advancedofthefourgreatempireswithanaverageoveralladoptionlevelof0.94.China remainsaheadofmostcountrieswith0.88.TheIndianandtheMiddleEasternempireshave fallenbehindto0.7.Today,thegapbetweenWesternEuropeandtheotherthreehistorical empireshaswidenedconsiderably. WhydoourhistoricalrankingsdifferfromtheviewthatancientEuropeanswerebarbarians, whileChinaandtheMiddleEast/Islamiccivilizationswerewellintheleadformostofour sampleperiod?Basically,itisfortworeasons.First,ourmeasureismorecomprehensiveandit islesslikelytobeswayedbyhighlyvisibleinventionslikegunpowderinChina.Second,whatwe aremeasuringistheadoptionoftechnologiesratherthantheinvention(i.e.by1500, gunpowderwasalreadyadoptedinWesternEuropeandmostoftheArabworld).Most historiansagreethatEuropehadcaughtuptoandsurpassedtheIslamiccivilizationsometimein thelateMiddleAges.Appendix1detailswhatdifferencesintechnologyadoptionaffectedthe mosttherankings. TheusefulnessofTable6goesbeyonddescribingtherelativeevolutionoftechnologyinthe majorempires.Becauseourdatacollectionexercisegoesbeyondafewcasestudies,wecanput

12

inperspectivetherelativedynamicsoftechnologyintheempires.Inparticular,thelevelsof historicaltechnologyadoptionreportedfortheempiresinTable5until1500A.D.areallfairly high.Giventhecrosscountrydistributionoftechnologyreportedabove,thisimpliesthat whetherEuropeisaheadofChinaorviceversaissecondordercomparedtothetechnological advantageofthehistoricalempirescomparedtomostoftherestofthecountriesintheworld. Underneaththeoveralltechnologyadoptionmeasuresthereissignificantcrosssectoral dispersion.Toexplorethis,wesubtractthecountrysoveralltechnologyadoptionfromits adoptionlevelineachofthefivesectors.Then,wecomputethecrosscountrystandard deviationofthisvariabletomeasurethewithinsectorvariationintechnologyadoption.Table6 comparesthesemeasurestothecrosscountrydispersionintheoveralltechnologyadoptionfor eachofthethreeperiodswestudy. __________________INSERTTABLE7HERE_______________________ Themainfindingisthatthewithinsectorvariationintechnologyadoptionisapproximatelytwo thirdsofthecrosscountrydispersioninoveralltechnology.Insection4.3,weexplorefurther theimplicationsofthisstrikinglylargeheterogeneityinsectoraltechnologyadoption. Afinalcheckweconducttoshowthatourtechnologyadoptiondatasetsaresensibleisto correlatethemwiththecontemporaneousurbanizationrate.Severalauthors12haveusethe urbanizationrateasaproxyfordevelopmentlevelspeciallyforpremodernperiodssuchasthe onescoveredbyourtechnologyadoptiondataset. __________________INSERTTABLE8HERE_______________________ Theurbanizationratefor1000B.C.and0A.D.comefromPeregrinesACE,13whilethe urbanizationratefor1500A.D.comefromAcemoglu,JohnsonandRobinson(2002).Table8 reportstheestimatedcontemporaneouseffectofoveralltechnologyadoptiononthe urbanizationrate.Wefindthatthereisastrongandpositivecontemporaneousassociation betweentechnologyadoptionhistoryandthecontemporaneousurbanizationrates.This providesfurthersupportforthequalityofourmeasuresoftechnologyadoptioninprecolonial times.14

3.2Technologyhistoryandcurrentdevelopment
Weturnnexttostudyingwhethercenturiesold,precolonialtechnologyhistoryiscorrelated withdevelopmenttoday.Toanswerthisquestion,weestimatethefollowingregression
Acemoglu,JohnsonandRobinson(2002),forexample. Peregrine(2003)constructsameasureoftheurbanizationratethatcantakethreevalues.1ifthelargest settlementissmallerthan100persons.2ifitisbetween100and399persons.3ifitislargerthan400persons. 14 Theworkingpaperversionofthisarticleshowsthatthepositivecontemporaneousassociationbetweenoverall technologyadoptionandtheurbanizationrateisrobusttocontrollingfordistancetotheEquator.
13 12

13

yc = + Tc + uc

(1)

wherecurrentdevelopment,yc,ismeasuredeitherbythelogofPPPadjustedpercapitaincome in2002A.D.orbycurrenttechnologyadoption,Tcisthemeasureofhistoricaltechnology adoptionanducistheerrorterm. ___________INSERTTABLE9HERE_______________ ThefirstthreecolumnsofTable9reporttheestimatesofregression(1)whenycispercapita incomein2002andTcismeasuredsuccessivelybytheoveralladoptionlevelin1000B.C.,in0 andin1500A.D.(Tstatisticsareinparentheses.)Thetechnologyadoptionlevelin1000B.C.is positivelyandsignificantlyassociatedwiththelogofpercapitaGDPin2002.Technology adoptionin0A.D.isnotsignificantlycorrelatedtocurrentdevelopment.Theoveralltechnology adoptionlevelin1500A.D.ispositivelyandsignificantlyassociatedwithcurrentincomeper capita.Thismeasureoftechnologyin1500A.D.explains18percentofthevariationinlogper capitaGDPin2002. Inadditiontobeingstatisticallysignificant,theeffectisquantitativelylarge.Changingfromthe maximum(i.e.1)totheminimum(i.e.0)theoveralltechnologyadoptionlevelin1500A.D.is associatedwithareductioninthelevelofincomepercapitain2002byafactorof5. ___________INSERTTABLE10HERE_______________ ThefirstthreecolumnsofTable10showthatcurrenttechnologyisassociatedwithhistorical technologyadoptionverymuchinthesamewayascurrentpercapitaincome.Theassociation betweencurrenttechnologyandtechnologyin1000B.Cor0A.D.isinsignificant.However,there isastrongandsignificantassociationbetweentechnologyin1500A.D.andcurrenttechnology. Inparticular,changingfromthemaximum(i.e.1)totheminimum(i.e.0)theoveralltechnology adoptionlevelin1500A.D.isassociatedwithanincreaseinthetimelagintheintensityof adoptionwithrespecttotheUSof22percentoftheyearssincetheinventionofthetechnology. Figure2presentsthescatterplotbetweenoveralltechnologyadoptionlevelin1500A.D.and currentdevelopment.Thepositiverelationshipbetweenthesetwovariablesisquite transparent.Itisclearlynotdrivenbyoutliers.Inthebottomleftquadrantoftheplotwecan seemanyAfricancountriesthathadadoptedveryfewofthetechnologiesinour1500sample andthatarequitepoortoday.Europeancountriesareinthetoprightcorner. CountriesthatroughlycorrespondtoancientempiressuchasEgypt,Iran,China,India,and Pakistanweremiddleincomecountriesin2002andhadadoptedbetween70and90percentof thetechnologiesinour1500A.D.sample.Thesecountriesareslightlybelowtheregressionline inthebottomrightquadrantofFigure2.Thispaperdoesnotaddresssomewellknownpuzzles, suchasthefailureofChinatocapitalizeearlieronitstechnologicalprowess,orthestagnation followingtheearliertechnologicalprowessoftheIslamicempire.Theseareveryimportant puzzlesthatdeserve(andhavealreadyattracted)theirownliterature,butweareconcerned herewiththeglobalcrosscountryaveragerelationshipbetweenoldtechnologyandmodern income,andthesecounterexamplesarenotnumerousenoughtooverturntheaverageglobal relationship.

14

___________INSERTFIGURE2HERE______________ LatinAmericancountrieswerebehindthemediancountryintheoveralltechnologyadoption levelin1500buttodaytheyaremiddleincomecountries.Thisverylikelyhassomethingtodo withthelongperiodofEuropeansettlementinLatinAmerica,eventhoughtheEuropean settlersweregenerallyaminorityofthepopulation.Finally,inthetopleftcornerofFigure2we findtheNeoEuropes,theUS,Canada,AustraliaandNewZealand.Thesewereamongthe countrieswithmostprimitivetechnologyin1500A.D.butareamongtheWorldsrichest countriestoday.Thisisverylikelyduetothelargescalereplacementoftheoriginalinhabitants withEuropeansettlers.15 WewouldexpectthattheEuropeansettlersintheSpanishandPortuguesecoloniesandinthe NeoEuropesaffectedquitedramaticallytheprocessoftechnologytransfer(aswellasother factorswithwhichtechnologymaybeassociatedsuchashumancapitalaccumulationand institutionaldevelopment)inthesecountriesduringthecolonialperiod.Anotherplacewhere therewaslargescale(albeitstillminority)EuropeansettlementwassouthernAfrica.Ofcourse, therecouldbetechnologytransferinanycolonizednation,butthedurationandintensityofthe influenceofthesettlementprocessesinsouthernAfrica,LatinAmericaandtheNeoEuropes suggestaddingspecialcontrols.Europeansettlementwasmainlyafunctionofanexogenous eventthesusceptibilityofnativestoEuropeandiseases,whichcreatedlightlyinhabitedfertile landstobetakenoverbyEuropeansettlers(EasterlyandLevine2008).Further,thedifference inthedegreetowhichEuropeanscolonizerssubstitutedforthelocalpopulation(reflecting differentialsusceptibilitytobeingentirelywipedoutbyEuropeandiseases)justifiesthe distinctionbetweentheNeoEuropesandLatinAmerica/southernAfrica.16 Toformalizethisintuition,weusethefractionofEuropeansettlersintotalpopulationin1900 fromAcemoglu,JohnsonandRobinson(2001).17Thisfractionwasover90percentfortheNeo Europes,between15percentand65percentforSouthAfrica,LesothoandSwaziland,andmost countriesinLatinAmericaandtheCaribbean,andbelow15percentfortherestofnon Europeancountries. Basedonthis,wecreatetwodummies.ThefirstcapturespredominantEuropeansettlers,and takesavalueofonefortheUS,Canada,NewZealandandAustraliaandiszerofortherestof thecountries.TheseconddummyreflectslesserEuropeansettlerpredominancethaninthe neoEuropes,andtakesavalueofonefortheLatinAmericancoloniesofSpainandPortugal, SouthAfrica,LesothoandSwaziland,andiszerootherwise.18Thisyieldsthefollowingregression equation:

AverysimilarpictureemergeswhenthedependentvariableincurrenttechnologyadoptionasillustratedbyFigure A1. 16 IndigenousmortalityfromEuropeandiseaseswasabigfactorinOceaniaandtheAmericas,butnotsomuchin southernAfrica.Europeansettlementinthelatterprobablyreflectedtheattractionoflightlypopulatedlandsinthe temperatezonesthatEuropeanspreferred(EasterlyandLevine2008). 17 SimilarresultsareobtainedusingtheshareofpopulationfromEuropeandescentin1975fromAcemoglu,Johnson andRobinson(2001)orthefractionofEuropeansettlers100yearsafterfirstsettlementfromEasterlyandLevine (2006). 18 GiventhesmallvariationinhistoricaltechnologybothwithintheneoEuropesandwithinLatinAmerica,including theEuropeaninfluencedummiesisequivalenttoexcludingthesecountriesfromtheregression.

15

15

yc = + Tc + Majorc + Minorc + uc

(2)

Columns5through7inTable9andTable10reporttheestimatesofequation(2)withyc measuredbothbypercapitaincomein2002(Table8)andcurrenttechnology(Table9)andwith Tcmeasuredsuccessivelybytheoveralltechnologyadoptionlevelin1000B.C.,0and1500A.D. _______________INSERTFIGURES3,4AND5HERE_____________________ WefindthattheEuropeansettlementdummieshaveasignificantpositiverelationshipwith bothcurrentpercapitaincomeandcurrenttechnology.Further,whenincludingtheEuropean settlementdummies,thecorrelationbetweentheoveralltechnologyadoptionandcurrent developmentmeasuredbothbypercapitaincomeandtechnologyadoptionincreases.In particular,theeffectsofthetechnologyadoptionlevelsin1000B.C.and0oncurrentpercapita incomeandtechnologybecomestatisticallysignificant,andthecoefficientsapproximately double.Inotherwords,oncewecontrolforthemostobvioushistoricalexampleofreplacement oftheindigenoustechnologybytechnologiesbroughtbynewsettlers,technologyinancient timesbecomesanevenmoresignificantpredictorofdevelopmenttoday. Weacknowledgethattherecouldhavebeenotherpopulationmigrationsthattransferred technology,andoursinglingoutoftheinternationalEuropeanmigrationmaybeadhoc, althoughitseemstoustheprimacyofEuropeanmigrationoverthelast500yearsisnotreallyin doubt. Inanycase,ourresultsseemtoholdforotherpopulationmovementsaswell.Weknowfroma collaborativeexercisewithDavidWeilthatourfindingsholdalsowhenwecontrolmore comprehensivelyfortheinternationalmigrationflows.Specifically,weusePuttermanandWeil (2007)smatrixwhichgives,foreachcountry,thedistributionofitscurrentpopulationbyits origin.Wethenpremultiplythevectorofoveralltechnologyin1500ADbytheoriginmatrix andfindthattheoriginweightedmeasureoftechnologypredictscurrentpercapitaincome slightlybetterthantheregressorsincolumn7ofTable8.Wedonotreporttheseresultshereas PuttermanandWeil(2007)havenotyetmadetheirdatapublic.Also,wecontinuetofind significantcorrelationsinimportantspecifications(suchasthosealreadyreportedinTable7) evenwhentheEuropeaninfluencedummiesareexcluded. _______________INSERTFIGURES6,7AND8HERE_____________________ Afterincludingthesettlementdummies,anincreaseintheoveralladoptionlevelfrom0to1in 1000B.C.orin0A.D.isassociatedwithanincreaseinincomepercapitain2002byafactorof4. Asimilarincreaseintheoveralladoptionlevelin1500A.D.isassociatedwithanincreaseinper capitaincomein2002byafactorof19.Thisishalfofthecurrentdifferenceinincomepercapita betweenthetopandbottom5percentofthecountriesintheworld.19 AfterfilteringtheeffectofEuropeaninfluencethroughthetwodummies,theincrementsinthe adoptionofcurrenttechnologyassociatedwithhavingahigheradoptionofhistorical
TableA1showstherobustnessofthisfindingtousingasendogenousvariablepercapitaincomeinearlierpointsin time.Inparticular,thetableuses1913,1960and1990.
19

16

technologiesarealsodramatic.Forexample,movingfrom0to1intheadoptionlevelin1500 A.D.isassociatedwithareductionintheadoptionlagwiththeUSequivalentto45percentof theyearssincethetechnologyhasinvented.Thisismorethantwiceasmuchastheobserved crosscountrystandarddeviationincurrenttechnologyadoption. Similarly,20percentoftheincomedifferencebetweenEuropeandAfricaisexplainedby Africaslaginoveralltechnologyadoptionin1000B.C.,8percentisexplainedbythetechnology distancein0A.D.,and78percentisexplainedbyAfricaslaginoveralltechnologyadoptionin 1500A.D.ThisgivesaverydifferentperspectiveonAfricaspovertycomparedtotheusual emphasisonmoderngovernments.Italsoshiftsbackwardintimethehistoricalexplanationsfor Africaspoverty,comparedtotheusualemphasisofhistoriansontheslavetradeand colonialism.20 Figures3through8displaythescatterplotsofthetwomeasuresofcurrentdevelopment(i.e. percapitaincome3through5andtechnologyadoption6through8)andoverallhistorical technologyadoptionafterregressingthesevariablesontheEuropeaninfluencedummies.These figuresconfirmthesignificantassociationbetweencurrentdevelopmentandhistorical technologyafterconditioningontheEuropeaninfluencedummies.Clearly,thestrongest relationshipholdsbetweenoveralltechnologyadoptionin1500A.D.andcurrentdevelopment measures.

3.3Robustness
Weexploretherobustnessofthefindingsencounteredsofar.Westartbyexploringwhether weareidentifyingtheeffectofhistoricaltechnologyoncurrentdevelopmentthroughthecross continentvariationoralsothroughthewithincontinentvariation.Toanswerthisquestion,the firstthreecolumnsofTable11andTable12reporttheestimatesofregression(2)whenthe dependentvariableispercapitaincome(Table11)andcurrenttechnologyadoption(Table12) andweaddfourcontinentdummiestothecontrolset. Weextracttwomainconclusionsfromcolumns1through3.First,muchoftheeffectof technologyhistoryisdetectedfromthecrosscontinentvariation.Addingthecontinent dummieseliminatestheeffectofoveralltechnologyadoptionin1000B.C.oncurrent development(column1),andreduces,approximately,by60percenttheeffectoftechnology adoptionin0A.D.(column2)andin1500A.D.(column3)oncurrentdevelopment.Only1500 ADisstillsignificant.Theflipsideofthisisthatasignificantfractionoftheeffectsoftechnology adoptionhistoryin0A.D.and1500A.D.oncurrentdevelopmentisdrivenbythewithin continentvariation.Inparticular,thewithincontinentvariationinoveralltechnologyadoption in1500A.D.canstillaccountforcrosscountryvariationincurrentincomepercapitabyafactor of4.5andforaboutonestandarddeviationoftheobservedcrosscountrydispersionincurrent technologyadoption.21Insum,thepersistenceoftechnologyacrossthelast500years,orthe last2000years,isnotjustduetodifferencesbetweencontinents.
Therewassomeslavetradebefore1500A.D.acrosstheSaharaandalongtheIndianOcean.However,most accountsofthenegativeeffectsoftheslavetradestresstheAtlanticslavetrade,whichonlybecamenontrivialafter 1500A.D. 21 Oneinterestingquestioniswhethertheeffectofhistoricaltechnologyoncurrentdevelopmentworksthrougha positiveeffectontotalGDPorthroughanegativeeffectoncurrentpopulation.Intheworkingpaperweshowthat therelationshipworksentirelythroughtheeffectoftechnologyonGDP.Indeed,historicaltechnologyalsohasa
20

17

 _______________INSERTTABLE11HERE_________________ Gallup,SachsandMellinger(1999)havearguedthatthelatitudeisanimportantdeterminantof incomepercapita,withthetropicsatadisadvantage.HallandJones(1999),Acemoglu,Johnson, andRobinson2002,EasterlyandLevine2003andRodriketal.(2003)arguethattheeffectof tropicallocationisthroughinstitutions.Inprinciple,thereisnoreasonfortheseeffectstobe linear.Columns4through6inTable11and12reporttheestimatesofregression(2)after controllingforthedistancetotheEquatorbothwithalinearandaquadraticterm.22As emphasizedbythepreviousliterature,beingveryfarfromtheEquatortendstobeassociated withhigherlevelsofcurrenttechnologyandincomepercapita.Controllingforthelatitudeof countries,however,doesnoteliminatethestrongpositiveeffectofoveralltechnologyadoption in1500A.D.oncurrentdevelopment.Thiseffectremainsstatisticallysignificant,thoughthe associationoftechnologyadoptionhistoryon1000B.C.andin0A.D.oncurrentdevelopment becomeinsignificantaftercontrollingforthedistancetotheEquator.23Incolumns7through9 weadditionallycontrolforwhethercountriesarelandlockedreachingthesameconclusionsas incolumns4through6.Incolumns10through12,wefurtherexperimentwithatropical dummy(andthelandlockeddummy)showingthattheresultson1500ADbothforcurrent incomeandtechnologyarerobusttothisgeographiccontrol. Hence,ourrobustfindingisthat1500ADmattersforcurrentdevelopment,whiletheresultson 1000BCand0ADmatteringfortodaysdevelopmentarenotrobust.Westillhavetheresult thatthereispersistenceintechnologyinthemoreancientperiodsfrom1000BCto0AD,and from0ADto1500ADbutthevariationintechnologyunexplainedbypersistenceovereach longepochishighenoughtomakethedirectlinkfrom1000BCtothepresentmuchweaker (notsosurprisingly!)thanthelinkfrom1500ADtothepresent. _______________INSERTTABLE12HERE_________________ Aninterestingquestioniswhythegeographycontrolsreducethesignificanceofhistorical technologyadoptiononcurrentdevelopment.Ofcourse,therearemanypossibleexplanations forthisobservation.Oneconsistentwiththemodelpresentedinsection4isthatgeography affectsthediffusionoftechnologies.Technologiesdiffusefasterwithinthanacrosscontinents. Diamond(1997)advancedthehypothesisthatcertaintechnologiesdiffusealonglatitudes. Finally,landlockedterritoriesmaybelessexposedtonewertechnologiesandhenceadoptthem laterand/orlessintensively. _______________INSERTTABLE13HERE_________________ Aswehavenotedabove,theassociationbetweenlandareaandancienttechnologycouldbe reversecausality,sincealargerlandareacontainedalargersampleofculturesandtechnologies fromwhichwearecodingthebest.Weconclude,however,thatthisisnotaverysignificant

strongandrobustrelationshipwithcurrentpopulation,buttheeffectoncurrentGDPissolargethatmorethan compensatestheeffectonpopulation. 22 Theresultsareextremelyrobusttoallsortofpolynomialsinthelatitude. 23 SimilarresultsholdwhenincludingatropicaldummyinsteadofthedistancefromtheEquator.

18

concernbasedontworesultsreportedinTable13.First,percapitaGDPtodayandcurrent technologyareuncorrelatedwithlandarea.Second,thecorrelationbetweenhistorical technologyadoptionand,both,percapitaGDPtodayandcurrenttechnologyremainunaffected bythelandareacontrol.Sotheassociationbetweencontemporaneoustechnologyandland areadoesnotseemtoreflectanydominantsamplingeffect.

4.Amodelofendogenoustechnologyadoptionanddiscussion
Wenextpresentamodelthatprovidesasimplerationalizationforthefactsuncoveredsofar. Themodelisalsoaneffectivetooltohelpusthinkaboutalternativehypothesisandhowto identifytheminthedata.

4.1Baselinemodel
Generationsareindexedbyt.Countriesareindexedbyc.LetActdenotethenumberof technologiesadopteduptot.Therepresentativeagentinthecountrylivesforonegeneration anddecideshowmanynewtechnologiestoadoptduringthatperiod.The(gross)valuehe obtainsfromadopting Act newtechnologiestechnologyis

ct = ct Act

(3)

where ct isapotentiallytimevaryingparameter. Thecostofadoptingnewtechnologiesisgivenbythefollowingfunction.

C ct = bct

1 ) wherebctmaybetimevaryingand [ , implyingthatthecostisincreasingandconvexin thenumberofnewtechnologiesadoptedintheperiod.Notethat,inthisformulation,thecost ofadoptiondeclineswiththestockoftechnologiespreviouslyadoptedinthecountry.Thisisthe casebothbecauseexistingtechnologiesreducethephysicalcostofadoptingthenew technologiesandbecauseadopterslearnfrompreviousadoptionexperience.Thislearningmay beformal(e.g.training)and/orinformal(e.g.trialanderror)butasaresultofit,technological leadersarethefirsttoadoptnewtechnologies.Seeourdiscussionofallthemechanismsby whichthiseffectcouldoperateinTable1(althoughsomeofthemechanismsinTable1haveto dowiththebenefitofthenewtechnology,whichcouldbeeasilyincorporatedinourmodelas aneffectofinitialtechnologyonvalueratherthancostofadoption).24,25
Wehavealsodevelopedanextendedversionofthemodelthatallowsforcrosscountrydifferencesinthe technologyfrontier.Thisextensionpermitsustocapturecountryinteractionsandtheeffectofinternational migrationandcolonization.Thisgeneralizationofthemodelcouldalsoaccountforthegreatdivergenceduringthe XIXcentury.Allthetestableimplicationsofthesimplermodelgothroughinthismoregeneralversion.The Generalizationisavailableuponrequestfromtheauthors. 25 Ourstylizedmodelignorestheintensivemarginofadoption(i.e.howmanyunitsofthetechnologyareadoptedin thecountry).Thisisnotabadapproximationuptoapproximately1875,sincetheextensivemarginaccountsfora verysignificantshareofthecrosscountryvariationinadoption.Itiseasytosee,however,howasimilarargumentto theonemadeabovehelpsexplaincrosscountryvariationintheintensivemarginofadoptioninmorerecenttimes. (Thatis,thecostofadoptionfacedbyindividualadoptersdeclineswiththestockoftechnologiestheyadoptedinthe 19
24

(Act ) Act 1

(4)

 OptimaladoptionofnewtechnologiesyieldsthefollowinglawofmotionforAct.
1 Act ct Act 1 1 = Act 1 Act1 1 1

(5)

where

ct

ct bct

(6)

Itisquiteclearfrom(5)thatthelongrundynamicsoftechnologydependonthevalueof. Supposeforsimplicitythat ct isstationary.Thenifissmallerthan2,thenumberof technologiesadoptedgrowsatanacceleratingrateastechnologylevelrisesinthelongrun.This wouldalsoapplyacrosscountries,sothatmoreadvancedcountrieswillhavefastergrowthand therewillbedivergence.Ifisequalto2,Actgrowsendogenouslyinthelongrunatastationary rate;thereistechnologicalpersistencebutneitherconvergencenordivergence.Finally,ifis greaterthan2,Actgrowsatadecreasingrateinthelongrunastechnologyrises.Lessadvanced countrieswillgrowfasterthanmoreadvancedcountriesandtherewillbeconvergence.26 Oneinterestingfeatureofanacceleratinggrowthmodelforcertainparametervaluesisthatthe accelerationcanbeextremelygradualforlongperiods,butthenwhenAgetssufficientlyhigh, growthcansuddenlyaccelerateveryrapidly.Figure9showsasimulationofequation5inwhich growthisvirtuallynonexistentformillenniaandthenspikesuptomoderngrowthlevelsover thelasttwocenturies.27Anothermorebackwardareawithlowerinitialtechnology(assumedto behalfoftheinitialtechnologyofthemoreadvancedarea)wouldnothavethisspike.Soour simplistictoymodelcouldaccountforboththeaccelerationofgrowthwiththeindustrial revolutionofthelasttwocenturiesandthegreatdivergencebetweencountriesoverthesame period!Ofcourse,wedonttakethisentirelyseriously,butwethinkitisacuteillustrative demonstrationofthepotentialpoweroftechnologypropagationmodels.(Twobigreasonsnot totakethisexercisetooseriously:(1)asiswellknowninthenewgrowthliterature,increasing returnsquicklybecomesanembarrassmentasgrowthwillkeepacceleratingtoabsurdlevels.If themodelistoberescued,theremustbesomethingelsethatslamsthebrakesonfurther increasesingrowthonceitreachesmoderngrowthratelevels,suchasalimitonhowmanynew technologiescanbeabsorbedperunitoftimerelativetoexistingtechnologies.(2)Ouractual estimatesofdiscussedbelowshowittobeverycloseto2. Tomakefurtherprogress,itisusefultorewrite(5)usingthelogapproximationaroundthe deterministicsteadystateasfollows:

a ct =

1 1 ct + a ct 1 1 1

(7)

past.)Themodelalsoignoresothermicroaspectsoftheadoptiondecisionthatmaynotbeoffirstorderimportance tounderstandingtheadoptiondynamicssuchaswhoistheadopter,afirm,aperson,thegovernment,... 26 Notethat,ourmodelhastwowaysofgeneratingtheobservedaccelerationofthegrowthoftheworldfrontier: having >2oranincreasein ct .

27

The parameter values are ct =.14,=1.5,andinitialA=.0165.Weexperimentedwithdifferentparameter

valuesuntilwefoundanexamplelikethatshowninFigure9.

20

where ct log( ct ) and a ct log( Act ) . Iterating(7)backintimeweobtain

1 1 a ct = ct +1 (8) 1 + 1 a c 0 =1 Wefurtherassumethat ct = ct 1 + ct ,where E0 ct = 0, fort>0.Then,itfollowsthatthe

t

expectedleveloftechnologyadoptedattgiventheadoptionhistoryuptotime0is 1 t 1 + 1 a (9) E 0 (a ct ) = c0 1 c0 1 1 1 Toderivetherelationshipbetweenhistoricaltechnologyandcurrentpercapitaincome,wejust needtoassumethat,atleastinmoderntimes,percapitaincomeisdeterminedby contemporaneoustechnology.28Formally, (10) yct = act Hence,

t t

1 t 1 + 1 a (11) c0 E 0 ( y ct ) = c0 1 1 1 1 Thissimplemodeldeliversthreetestablepredictions.First,since't'isverylargeinoursample,
t

1 thecoefficientofac0in(11),canhelpuscalibrate.Inparticular,thecoefficient, 1 , 1 tends to 0if 1 <1.(Thatis,if>2.)Similar,thiscoefficientshouldbeverylarge,andin

particulargreaterthan1,if onlybebetween0and1if

1 >1.(Thatisif<2.)Finally,thecoefficientofac0in(11)can 1

1 1.(Thatisif2.) 1 1 1,thereispersistenceintechnologyadoption. 1

Second,equation(9)impliesthat,if

Oneimplicationofthisisthatthetechnologyadoptionhistoryhelpspredictcurrenttechnology adoption.Third,thereasonwhytechnologyadoptionhistorycanpredictcurrentdevelopmentis

28

ThismightnothavebeenthecaseinpreindustrialsocietieswherepopulationisdeterminedbyMalthusian dynamicsasinKremer(1993),GalorandWeil(2000),PrescottandHansen(2002)andJones(2001,2005),butitisa veryreasonableassumptionformodernsocietiessupportedbythefindingsinCominandHobijn(2008).

21

becauseitcanpredictcurrenttechnology.Therefore,aftercontrollingforcurrenttechnology historicaltechnologyadoptionshouldnothelpusforecastcurrentdevelopment.29 Thesepredictionscaneasilybetestedwithourdata.WehavealreadyobservedinTables9and 11thatoveralltechnologyadoptioninhistoricalperiodspredictscurrenttechnology.Toexplore furtherwhethertechnologyispersistent,weestimatethecrosscountrycorrelationof technologyadoptionacrossthehistoricaltimeperiods. _________________INSERTTABLE14HERE_______________ AsreportedinTable14,thecorrelationoftheoveralltechnologyadoptionlevelbetween1000 B.C.and0is0.62,between0A.D.and1500A.D.itis0.71andbetween1000B.C.and1500A.D. itis0.68.(Itisalsoreassuringthattheerrorrateonourtechnologicalmeasuresisnot disastrouslyhigh.) Wefindinstructivetomakeaquantitativeassessmentofthispersistence.Iftismeasuredin years,theautocorrelationsestimatedinthefirstcolumnofTable14implythatwhencalibrated annually,

1 0.9995. 1

Wealsoobservethishighpersistenceoftechnologyadoptionforthesectorlevelmeasures. Theaveragecorrelationcoefficientbetweenthetechnologyadoptioninasectorinoneperiod andinthesubsequentperiodisaround0.5.Technologyadoptionismostpersistentinmilitary, industryandtransportation.Thelowestcorrelationisincommunicationsbetween1000BCand 1500AD.Thislattercorrelationisstillstatisticallysignificantatthe5percentlevel.Alltheother correlationsreportedinTable14aresignificantatthe1/10000level. ________________INSERTTABLE15HERE_________________ AsshowninTables12and15,thispersistenceoftechnologyadoptionisnotdrivenbythe geographicalvariablessinceitholdsoncewecontrolforthecontinent,thelatitudeandwhether thecountryislandlocked. ________________INSERTTABLE16HERE_________________ Weconcludethisexplorationofthemodelpredictionsbyinvestigatingthethirdprediction, namelythattheeffectofhistoricaltechnologyadoptiononcurrentdevelopmentoperates throughcurrenttechnologyadoption.Table15showsthestrongpositiveassociationbetween currenttechnologyandpercapitaincomein2002.TheR2inthisregressionis80percent.The

29

Thereisafourthinterestingpredictionwhichwecannottestduetolackofdatawhichisthat,iftechnologyisso

persistentthat

1 1,factorsthataffectthehistoricalreturntoadoptingthetechnology(i.e.institutions)may 1

alsoaffectcurrentdevelopmentevenifinstitutionsarenotpersistentandevenaftercontrollingforhistorical technology.Thisisthecasebecausesuchfactorsaffectthechangeintechnologyattime0,henceaffectthelevelof technology(atleastshortly)after0andhenceaffectsubsequenttechnologyadoption.

22

importanceandsignificanceofthisvariableisrobusttocontrollingforthedistancetothe equatorandforthecontinentdummies. Incolumns4through8ofTable16weincludeourmeasureofoveralltechnologyadoptionin 1500ADasanadditionalcontrol.However,itsassociationwithcurrentdevelopmentbecomes insignificantoncewecontrolforcurrenttechnology.Thisisthecaseevenafterincludingthe Europeaninfluencedummies(columns7and8).Sincetechnology1500ADwasrobustand significantinourpreviousregressions,weinterpretthisfindingasevidencethattheeffectof historicaltechnologyoncurrentdevelopmentoperatesmostrobustlythroughtheeffectthat historicaltechnologyadoptionhasoncurrenttechnologyadoption.

4.2Population
Oneinterpretationofthefindingssofaristhattechnologyisaverypowerfulpropagation mechanism.Thatis,avarietyoffactors(e.g.institutions,geographyshocks,luck,...)affectthe returntoadoptingtechnologyinthedistantpastandthosehistoricaltechnologyadoption levelsarepropagatedintothepresentthroughtheeffectthattechnologyhasontheagent's incentivestoadoptsubsequenttechnologies. Analternativeview,alsoconsistentwiththemodelpresentedabove,isthatthepowerful propagationmechanismisnotthedynamicsoftechnologyadoptionbutthedynamicsofthe returntoadoptingtechnology( ct ).Thatis,ratherthanhaving

1 1,thepersistenceof 1

historicaladoptionresultsfrom1. Onevariablethathasbeenlinkedtoimprovedtechnologyispopulation.Kremer(1993) developsmodelswherethisisthecasebecausemorepopulationmeansmoreinventorsof ideas.Similarly,semiendogenousgrowthmodels(e.g.Jones(1995,1999,2001,2005),Kortum (1997),Segerstrom(1998))alsohavethefeaturethattheforcethatultimatelykeepstechnology improvingisthecontinuousexpansionofthescaleintheeconomythroughpopulationgrowth. Thereducedformofthesemodelsintechnologyalonewouldalsopredicttechnological persistence,andsothisfactaloneisnotenoughtoassesstheweighttobeattachedtodifferent models. Thesemodelsstresstheeffectofpopulationthroughthesupplyofideaswiththeincreased numberofthinkers.Wecangenerateasimilarpopulationeffectbyhavingpopulationaffectour returntoideas(asitobviouslydoeswithnonrivalideas).Oursimplemodelcanaccommodate theseinsightsbyassumingthat ct = lct ,where lct isthelogpopulationandevolvesaccording to lct = nc + lct 1 + lct , E0 lct = 0, fort>0.Then,actcanbeexpressedas:
t +1 1 1 a ct = nc (t + 1) + l c 0 + lcq 1 + 1 a c 0 q =1 =1 t t

(12)

Overthelonghorizonsweconsiderinthispaper, nc

lct lc 0 . Hence,(12implies) t

23

 1 1 t 1 t (t + 1) 1 1 l ac 0 + E 0 (a ct ) = (l ct l c 0 ) + c0 1 1 t 2 1 = Expression(13)impliesthatif

(13)

1 <1,historicaltechnologyshouldnotpredictcurrent 1

technologyoncewecontrolforhistoricalandcurrentpopulation.Further,failingtocontrolfor currentandpastpopulationinourregressionsmaycauseanupwardbiasinourestimateof

1 1 ifthereturntoadoptingtechnology, ct ,is higherinlargercountries.

Toidentifywhethertheobservedpersistenceoftechnologyisdrivenbytheomissionof populationinourregressionsorbecausetechnologyhasadirectpropagationmechanism,we controlforhistoricalandcurrentpopulationinourbaselineregression(2).Asweshowinthe appendix,thisidentificationstrategyisvalidregardlessoftheinfluencefromothercountries technologyintheadoptioncost.Table17and18reporttheestimationresults. ________________INSERTTABLE17HERE_________________ ThefirstthreecolumnsofTable17reporttheeffectoftechnologyandpopulationin1500AD oncurrentpercapitaincome(column1),currenttechnology(column2)andcurrentpopulation (column3).Themainfindingisthattheobservedeffectofhistoricaltechnologyoncurrent developmentandcurrenttechnologyisrobusttocontrollingforhistoricalpopulation.Indeed, theeffectofhistoricalpopulationoncurrentdevelopmentisnegativeandsignificant.Both historicaltechnologyandpopulationhavepositivesignificanteffectsoncurrentpopulation, althoughthesignificanceofhistoricalpopulationismuchhigher. Weconductsimilarexerciseswithtechnology(column4)andpopulation(column5)in1500AD asdependentvariablesandtechnologyandpopulationin0asindependentvariables.Our findingsareverysimilartothosereportedinthefirstthreecolumnsofTable17.Forexample, theeffectoftechnologyadoptionin0ontechnologyin1500persistsaftercontrollingfor populationin0. ________________INSERTTABLE18HERE_________________ Expression(13)impliesthattoidentifythepropagationmechanismthatgeneratesthe persistenceoftechnology,weshouldcontrolbothforhistoricalandforcurrentpopulation. Table18conductsthismorepropertestbyincludingcurrentpopulationascontrol.Regardless ofwhetherweusecurrentdevelopment(column1)ortechnology(column2)asdependent variables,westillfindthattechnologyin1500ADhasastrongandsignificanteffectoncurrent outcomes.Currentpopulationhasanegativeandsignificanteffectoncurrentdevelopmentand technology.Populationin1500hasaninsignificanteffect.Ashortcomingofthisexerciseisthe endogeneityofcurrentpopulation,whichwedonothaveagoodmeansofaddressingbecause ofalackofanyobviousinstrument.

24

Incolumns3and4ofTable18wecontrolfurtherforthecontinentdummies.Thisreducesthe magnitudeofthecoefficientsbuttheeffectoftechnologyadoptionin1500oncurrent developmentandtechnologystillremainssignificant. Weareinclinedtointerpretourfindingsassupportingadirecteffectofoldtechnologyonnew technology,evencontrollingforthepopulationchannel.Wedontclaimtohaverefutedthe populationandtechnologymodel,sinceourcoefficientsonpopulationareallovertheplaceand usuallyinsignificant;wethinkamoredetailedtestofthemanypredictionsofthepopulation andtechnologymodelswouldbeneededtopassjudgmentonthosemodels.Fornow,we concludeataminimumthatpopulationdynamicsarenottheONLYpropagationmechanism thatgeneratestheobservedeffectoftechnologyadoptionhistoryoncurrentdevelopment.

4.3Institutions,culture,genes
Theliteraturehasemphasizedatleastthreemorevariablesthatarelikelytoaffectthereturnto adoptingtechnologyandcouldbedrivingtheobservedpersistenceoftechnology.Theseare institutions,cultureandgenetics.Ifanyofthesevariablesaffectshistoricaltechnologyadoption andissufficientlypersistent,itsomissionfromourregressionwillbiastheestimateof andcouldrationalizethefactspresentedsofar.32 Furtherprogresscanbemadeintheestimationof

1 1

1 ifweassumethat,theeffectof 1

institutions,cultureandgeneticendowmentonthereturntotechnologyadoptionhasan importantsymmetriccomponentacrosssectors.Thatis,goodinstitutionspreserveproperty rightsandinduceagentstoadoptnewtechnologiesinallsectors(coveredbyourdataset).33 Underthissensibleassumption,wecanestimate

1 byexploitingthelargeobserved 1

withinsectorvariationintechnologyadoption(seeTable6).Thisintuitionisformalizedwiththe followingmultisectorextensionofourbaselinemodel. Supposethattheprofitsandcostsfromadopting Acst newtechnologiestechnologyinsectors are,respectively

cst = cst Acst

32

(14)

Commonmeasuresofinstitutionsdonotseemtobesufficientlypersistenttocauseanysignificantbias. Przeworski(2004)hascalculatedthatthecorrelationof'constraintsonexecutive'for59countriesfromPolityIVfrom firstyearofindependencetomostrecentyearavailableisonly.26.Similarly,wehaveestimatedthecrosscountry correlationinthedemocracyvariablefromPolityIVin2000withthesamevariablein1950tobe.33.Itseems remarkablethatthesewidelyusedmeasuresofinstitutionsarelesspersistentovera50yearperiodthantechnology adoptionovera2500yearperiod.However,thisisonlyonemeasureofinstitutions,anddeepermeasuresof institutionsmayshowmuchmorepersistence.Ifinstitutionsreallydohavelowpersistence,omittinghistorical institutionsfromourbaselineregressionshouldnotbiasourestimateof

33

1 . 1

Similarly,theforcesbywhichcultureandgenesaffecttechnologyadoption(e.g.higherwillingnesstoexperiment orhigherI.Q.)arerelevanttoallthesectorscoveredinourdatasets.

25

C cst = bcst

(Acst ) Acst 1

(15)

Notethatinthisformulation,thecostofadoptiondeclinesintheleveloftechnologyadoptedin thesector.Thentheoptimaladoptionoftechnologyinsectorsandcountrycimpliesthe followinglawofmotionfortechnology.

Acst cst Acst 1 = 1 Acst 1 Acst 1

where

(16)

cst

cst I S S = ct * cs * cst bcst

(17)

I I I S S Let cst log( cst ) beazeromean,iid,term, ct log( ct ) = ct 1 + Ict ,where E 0 Ict = 0 fort>0and

I S S cs log( cs ) .Notethat,inthisformulation, ct captures(potentiallytimevarying)factorsthataffectthe

S returntoadoptingtechnologyinallsectors, cs reflectsfixedfactorsthataffectthereturntoadopting
S technologyinsectorsofcountryc,and cst capturesotherfactorsthataffectthereturntoadopting technology.

and

a cst =
=1

I ct +1

+ +
S cs

S ct +1

1 1 1 + 1 a cs 0

(18)

t 1 t 1 t 1 t 1 I + 1 S + 1 a E (a ) = cs 0 0 cst c0 cs 1 1 1 1 1 1 1

(19)

Theempiricalcounterparttoequation(19)is

(20) Thisregressionincludesacountryeffect(ct1)thatcouldbetimevaryingandacountrysector fixedeffect(cs).Theidentificationof presentsthewellknownchallengeoftenencountered inconvergenceregressionsofalaggeddependentvariable.Tosolvethisproblemwefirst differenceequation(20)obtaining

acst = ct 1 + cs + * acst 1 + ucst

acst acst 1 = ct 1 ct 2 + * (acst 1 acst 2 ) + u cst

(21)

andinstrumenting acst 1 acst 2 with acst 2 .If acst 2 isuncorrelatedwith u cst ourestimateof

willbeunbiased.Wearegoingtoimplementregression(21)witht=2000AD,t1=1500AD

andt2=0AD.Wefinditsensiblethat u cst u cs 2000 u cs1500 isuncorrelatedtothetechnology adoptionlevelin0AD(i.e.1500yearsbefore). Inourlistofcurrenttechnologieswehavenomilitarytechnology.Hence,weestimate(21)using dataontheotherfoursectorsinourdataset.Since,fromourpreviousresults,theEuropean influencewillhaveaneffectonthegrowthoftechnologybetween1500and2000,wecontrol forthetwoEuropeaninfluencedummieswhenestimating in(21).

26

________________INSERTTABLE19HERE_________________ Table19reportstheestimatesof with(column1)andwithout(column2)thecountryfixed effectsinthegrowthrates.Themainfindingisthatweobserveverylargeandsignificant estimatesof inbothspecifications.Thereareseveralimportantimplicationsfromthese estimates.Recallthatbothoftheseregressionsincludecountryfixedeffectsinlevelsand countrysectorfixedeffectsinlevels.Thisissuggestivethatpersistentfactorsthathave remainedconstantoverthelast500yearsandwhichmayormaynothaveaffected asymmetricallytechnologyadoptioninthecountrysuchasgenesorgeographyarenotthe primarydriversoftheestimatesofthepersistenceoftechnology(i.e. ). Asecondinterestingobservationisthattheestimatesof arevirtuallyunaffectedbythe inclusionofthecountry(timevarying)effects.Thesetermscapturechangesatthecountrylevel between1500and2000whichaffectsymmetricallytheadoptionoftechnologyacrosssectors (withinthecountry),suchasanimprovementininstitutions(seeJones2001).Includingthese countryeffectsintheregressionsdoesnotaffectthesizeofourestimateof (indeedit increases).Ourresultssaythatifatechnologyinonesectorinonecountryimprovesmorethan averagecountrywidetechnologicalimprovementin1500,thattechnologicalimprovement showsupinbettertechnologyinthatsector(relativetothecountrywideaverage)in2000.This issuggestivethatomittedtimevaryingfactorsthataffecttheadoptionoftechnologyinmany sectorsintheeconomysuchasanimprovementininstitutionsarenotsolelyresponsiblefor theobservedhighpersistenceintechnologyadoption.34Notethatthisargumentstandsevenif theomittedvariablehasadifferentialeffectacrosssectorsaslongasithasasignificant commoncomponent.Wetakethisasverysupportiveofourhypothesisthatthereisadirect effectofoldtechnologyonnewtechnologyadoption,evenifinstitutionalandothercountry widefactorsalsomatter. Becauseofpotentialcrosstechnologydifferencesindiffusionnotcapturedbyourmodel,we exploretherobustnessofourestimatesof toincludingtechnologyspecificintercepts.Asis clearfromcolumns(3)and(4)ofTable19,thesignificanceof isunaffected.Interestingly,the similarityofthepointestimateswithandwithoutcountry(timevarying)effectsalsopersists afterincludingtechnologyspecificintercepts. Weconcludeourempiricalexplorationofthesourcesofpersistenceintechnologyadoptionby showingthatthesignificanceof isnotdrivenbyanysinglesector.Incolumns5through8of Table19wereporttheestimatesof aftereliminatingsuccessivelyoneofthefoursectors coveredinourdatasetrecallthatwehaveexcludedthemilitarytechnologiesfromthis analysis. Twoobservationsemergefromthisexercise.First,thesignificanceoftheestimateof isnot drivenbyanyspecificsector.Second,whencomparingtheestimatesof obtainedby exploitingthesectoralvariationintechnologyadoption(e.g.columns38ofTable19)withthe estimatesweobtainedwhenusingthecrosscountryvariation(e.g.columns3,6,9and12of Table12)itturnsoutthattheformerarenotsmallerthanthelater.
Wedonotknowwhethertheaspectsofculturewhicharerelevantforadoptionoftechnologyarefixedovera500 yearperiodortimevarying,butineithercase,theyarealsolikelytobetakencareoffbyoneofthetwoarguments posedabove.
34

27

 Asarguedabove,thisishardtoreconcilewiththeorieswheretheONLYmechanismthatinduces thepersistenceoftechnologyoperatesatacountrywidelevel.Instead,thisobservationcanbe accountedfornaturallybyoursimpletheorysincetheknowledgecreatedwhenadoptinga technologyinagivensectorislikelytoreducethecostsofadoptingsubsequenttechnologiesin thesectorbutnotsomuchinothersectors. Hence,weareinclinedtoconcludefromthefindingspresentedinthispaperthatthedynamics oftechnologyadoptionareadirectpropagationmechanismthattransmitssignificant technologyshocksofverydiversenatureintothedistantfuture.Theprimaryfeatureofour modelthatpredictsthisresultisthatthecostoftechnologyadoptionislowerthehigheristhe previousstockoftechnology.

5.Conclusions
Themainfindingofthispaperisasimpleone:centuriesoldtechnologicalhistoryisassociated withthewealthofnationstoday.Thisislargelyrobusttoincludingcontinentdummiesand geographiccontrols,soitisnotjustdrivenbyEuropevs.Africaortropicalvs.temperate zones.Therearetwosurprisingpartsofthefinding.Thefirstisjusthowoldthehistorycanbe andstillbecorrelatedwithmodernoutcomes.Ourmostrobustfindingisthattechnologyin 1500ADiscorrelatedwithdevelopmentoutcomestoday,itselfremarkablyoldwhenwe considerthatmosthistorydiscussionsofdevelopingcountriesstart(atmost)withEuropean contactandcolonization.Thesecondsurprisingaspectofourfindingishowlargeisthe magnitudeoftheassociationbetweenhistoricaltechnologyadoptionandcurrentdevelopment. Inourbaselinespecification,goingfromhavingnonetohavingadoptedallthetechnologies availablein1500ADisassociatedwithanincreaseincurrentpercapitaGDPbyafactorof17. Evenafterincludingabatteryofcontrols,thisfactorisover5. Inanefforttowardsunderstandingwhatdrivesthissurprisingcorrelation,wehavefound suggestiveresultsthattechnologyisverypersistent,thatthispersistenceisnotdrivenonlyby thepersistenceofpopulation,andthatitdoesnotdiminishwhenexploitingthesectoral variationintechnologyadoptionafterremovingthecountryaverageadoptionlevelinthe periodandcountrysectorfixedeffects(hencecontrollingforanyfactorsthatoperateata countrywidelevelsuchasinstitutions).Thisevidenceprovidessupporttothehypothesisthat thetechnologyadoptiondynamicsinwhichthecostofadoptingnewtechnologyfallswiththe stockofprevioustechnologyarethemechanismthatgeneratesthepropagationuncoveredin thedata.

28

 References Acemoglu,D.,S.JohnsonandJ.Robinson,(2002),ReversalofFortune:Geographyand InstitutionsintheMakingoftheModernWorldIncomeDistributionQuarterlyJournalof Economics,(November),v.117,iss.4,pp.123194 Adshead,S.A.M.(1988)ChinainWorldHistory.NewYork:St.Margin'sPress,Inc. Ashraf,QuamrulandOdedGalor(2008),HumanGeneticDiversityandComparativeEconomic Development,mimeo,BrownUniversity,June2008 Barry,H.,III,&L.M.Paxson(1971)"InfancyandEarlyChildhood:CrossCulturalCodes2," Ethnology.10,pp.466508. Basalla,G.(1988)TheEvolutionofTechnology.Cambridge,UnitedKingdom:Cambridge UniversityPress. Black,J.(1996)TheCambridgeIllustratedAtlasofWarfare:RenaissancetoRevolution1492 1792.Cambridge,UnitedKingdom:UniversityofCambridge. Carneiro,R.L.(1970)"ScaleAnalysis,EvolutionarySequences,andtheRatingofCultures"In:R. Naroll&R.Coheneds.AHandbookofMethodinCulturalAnthropology.GardenCity,New York:TheNaturalHistoryPress.pp:834871. Collis,J.(1984)TheEuropeanIronAge.NewYork:Routledge. Comin,D.andHobijn(2008a)"AnExplorationofTechnologyDiffusion"HBSw.p.#08093. Comin,D.andHobijn(2008b)"LobbiesandtechnologyDiffusion"ReviewofEconomicsand Statistics,forthcoming. DavidP.A.(1975)TechnicalChoice,InnovationandEconomicGrowth:EssaysonAmericanand BritishExperienceintheNineteenthCentury.CambridgeUniversityPress. Diamond,J.(1997)Guns,Germs,&Steel.NewYork:W.W.Norton&Company. Easterly,W.andR.Levine(2008),EuropeanSettlementandEconomicDevelopment,mimeo, Easterly,W.andR.Levine(2003)Tropics,germs,andcrops:theroleofendowmentsin economicdevelopmentJournalofMonetaryEconomics,50:1,January2003. EncyclopediaBritannica(2006h)"chariot,"EncyclopdiaBritannicaOnline. http://search.eb.com/eb/article9022532 EncyclopdiaBritannica(2006)"Inca"EncyclopdiaBritannicaOnline.19Aug.2006 http://search.eb.com/eb/article9042237

29

Erman,A.(1971)LifeinAncientEgypt.NewYork:DoverPublicationsInc. Gallup,J.,J.Sachs,andA.Mellinger(1999)"GeographyandEconomicDevelopment,", InternationalRegionalScienceReview,Vol.22,No.2,pp.179232,August1999. GalorO.,"FromStagnationtoGrowth:UnifiedGrowthTheory",inP.AghionandS.Durlauf, editors,HandbookofEconomicGrowth,2005,171293. Galor,O.,andD.N.Weil(2000),Population,TechnologyandGrowth:FromtheMalthusian regimetotheDemographicTransition,AmericanEconomicReview110:806828. Greene,K.(2000).TechnologicalInnovationandEconomicProgressintheAncientWorld:M.I. FinleyReConsidered.TheEconomicHistoryReview,NewSeries,Vol.53,No.1(Feb.2000), pp.2959. Hall,R.andC.Jones(1999)"WhyDoSomeCountriesProduceSoMuchMoreOutputper WorkerthanOthers?",QuarterlyJournalofEconomics,February1999,Vol.114,pp.83116. Hogg,O.F.G.(1968)ClubstoCannon,WarefareandWeaponsbeforetheintroductionof Gunpowder.London:GeraldDuckworth&Co.Ltd. Jones,C..GrowthandIdeas.inP.AghionandS.Durlauf,editors,HandbookofEconomic Growth,2005 Jones,C.."WasanIndustrialRevolutionInevitable?EconomicGrowthOvertheVeryLongRun" AdvancesinMacroeconomics,August2001,Vol.1,No.2,Article1. Jones,C.."Growth:WithorWithoutScaleEffects?"AmericanEconomicReviewPapersand Proceedings,May1999,Vol.89,pp.139144. Jones,C.."R&DBasedModelsofEconomicGrowth"JournalofPoliticalEconomy,August1995, Vol.103,pp.759784. Kerr,R.&N.Wood(2004)JosephNeedhamScienceandTechnologyinChina,Vol5Chemistry andChemicalTechnology,PartXII:CeramicTechnology.Cambridge,UnitedKingdom: CambridgeUniversityPress. Kipfer,B.A.ed.(2000)EncyclopedicDictionaryofArchaeology.NewYork:Lluwker Academic/PlenumPublishers. Klenow,P.andA.RodriguezClare(1997)"TheNeoclassicalRevivalinGrowthEconomics:HasIt GoneTooFar?"withAndrsRodrguezClare,NBERMacroeconomicsAnnual1997,B. BernankeandJ.Rotemberged.,Cambridge,MA:MITPress,73102 Kremer,M.PopulationGrowthandTechnologicalChange:OneMillionB.C.to1990,The QuarterlyJournalofEconomics,Vol.108,No.3(Aug.,1993),pp.681716.

30

Landes,D.(1969)UnboundPrometheus:TechnologicalChangeandIndustrialDevelopmentin WesternEuropefrom1750tothePresent,CambridgeUniversityPress Lucas,A.(1934)AncientEgyptianMaterials&Industries.London:EdwardArnold&Co. Macksey,K.(1993)ThePenguinEncyclopediaofWeaponsandMilitaryTechnology,Prehistoryto thePresentDay.NewYork:PenguinBooksUSA. McEvedy,C.andR.Jones(1978)AtlasofWorldPopulationHistoryPenguinBooks,London. Mokyr,J.(1990)TheLeverofRiches:TechnologicalCreativityandEconomicProgress.Oxford; NewYork;TorontoandMelbourne:OxfordUniversityPress. Mokyr,J.(2002).TheGiftsofAthena:HistoricalOriginsoftheKnowledgeEconomy.Princeton: PrincetonUniversityPress. Mokyr,J.(2005),TheIntellectualOriginsofModernEconomicGrowth,TheJournalof EconomicHistory,VOLUME65JUNE2005NUMBER2,285351. Mokyr,Joel(2005).LongtermEconomicGrowthandtheHistoryofTechnology,inthe HandbookofEconomicGrowth,editedbyPhilippeAghionandStevenDurlauf,Amsterdam: Elsevier,2005,pp.11131180. Murdock,G.P.(1967)EthnographicAtlas.UniversityofPittsburghPress. Murdock,G.P.&D.O.Morrow(1970)"SubsistenceEconomyandSupportivePractices"In:A. Spohered.Ethnology,9.No.3.pp:302330. Murdock,G.P.&C.Provost(1973)"MeasurementofCulturalComplexity"In:A.Tudened. Ethnology,12,No.4.pp:379392. Murdock,G.P.&S.F.Wilson(1972)"SettlementPatternsandCommunityOrganization," Ethnology.11,pp.254295. O'Brien,P.K.ed.(1999)AtlasofWorldHistory.NewYork:OxfordUniversityPress. Peregrine,P.N.(2003)"AtlasofCulturalEvolution"In:J.P.Grayed.WorldCultures,14,no.3.pp: 175. Peregrine,P.N.,&M.Embereds.(2001a)EncyclopediaofPrehistory.Vols18.NewYork:Kluwer Academic/PlenumPublishers. Peregrine,P.N.,&M.Embereds.(2001b)EncyclopediaofPrehistory.Vol9.NewYork:Kluwer Academic/PlenumPublishers. Pulkki,A.M.,andP.Stoneman(2006)InternationalDiffusionofNewTechnologies:Intensive andExtensiveMargins.MimeoUniversityofWarwick.

31

Rodrik,D.A.Subramanian,F.Trebbi(2004)InstitutionsRule:ThePrimacyofInstitutionsover GeographyandIntegrationinEconomicDevelopment.JournalofEconomicGrowth,June 2004,v.9,iss.2,pp.13165 Rosenberg,N.(2000)SchumpeterandtheEndogeneityofTechnology,Routledge:Londonand NewYork. Rosenberg,N.(1994)ExploringtheBlackBox:Technology,Economics,andHistory, CambridgeUniversityPress. Rosenberg,N.(1982)InsidetheBlackBox:TechnologyandEconomics.CambridgeUniversity Press. Rosenberg,N.andL.E.BirdzellJr.HowtheWestGrewRich:TheEconomicTransformationofthe IndustrialWorld,BasicBooks:NewYork,1986. Rouse,I.(1972)IntroductionToPrehistory,ASystematicApproach.NewYork:McGrawHillBook Company. Scarre,C.ed.(1988)HammondPastWorldsTheTimesAtlasofArchaeology.Maplewood,New Jersey:HammondInc. Spolaore,EnricoandRomainWacziarg,TheDiffusionofDevelopment,mimeo,Tuftsand Stanford,March2006 Tuden,A.&C.Marshall(1972)"PoliticalOrganization,"Ethnology.11,pp.436464. Wager,D.B.(1993)IronandSteelinAncientChina.NewYork:E.J.Brill.

32

Appendix1
InTable5wedocumenttheprogressionofadvancedcivilizations(WesternEurope,China,India, andtheArabEmpire)from1000BCto1500AD.InitiallyWesternEuropelaggedbehindother civilizationsbutassumedtechnologicalleadershipby1500AD.WesternEuropeinitiallystarted astheleasttechnologicalsophisticatedofthemajoradvancedcivilizationsin1000B.C.Itslowly positionwasduetoitssloweradoptionofcommunications,transportation,andmilitary technology.WesternEurope,withtheexceptionofItaly,hadnotyetadoptedwrittenrecords forcommunication,vehiclesfortransportation,andironweapons.Otheradvancedcivilizations atthetime,suchasChinaandtheArabEmpirealreadywereusingthesetechnologies. However,by0A.D.WesternEuropehadadoptedthesetechnologies,andby1500A.D.itwas themosttechnologicallyadvancedcivilization.WesternEuropesascendancyisprimarilydueto advancesintransportationandmilitarytechnology.Duringthe16thcentury,WesternEurope alonehadexpeditionsacrossallmajoroceans(ThefamousChineseexplorerChengHohad voyagedacrosstheIndianOceanasfarasEastAfricain14051431,butthesevoyagesturned outtobeanaberrationasthevoyageswerestoppedandshipsdestroyedbyasubsequent emperor).Thesejourneysrequiredadoptinganumberofshipbuildingandnavigational technologies.WesternEuropealsowasattheforefrontinnavalweapons.Thenaviesof WesternEuropedeployedlargewarships(inexcessof1500tondeadweight)witharmamentsof over180gunsbythe16thecentury.EvenadvancedcivilizationssuchasChinaandtheArab Empirehadnotyetadoptedtheseadvancedweapons.

33

 Figure1a:Overalltechnologyadoptionin1000B.C.

Figure1b:Overalltechnologyadoptionin0A.D.

Figure1c:Overalltechnologyadoptionin1500A.D.

34

 Figure2:Technologyin1500andcurrentdevelopment
11
IRL NOR CHE DNK AUT NLD BEL DEU FRA JPNHKG GBR FIN ITA SWE SGP NZL ESP 10 GRC MLT KOR PRT CZE SVK HUN SAU EST ARG POL LTU HRV ZAF CHL LVA MYS BWA MEX CRI RUS 9 URY BRA THA TON ROM TUN IRN COL BLZ GAB TUR NAM PAN DZABLR BIH FJI VEN PER SLV UKR PRY CHN GUY GTM PHL MAR ECU SYR EGY NIC IDN 8 IND BOL HND LSO PNG VNM AGO GHA GIN CMR PAK KHM SDN BGD LAO UZB SEN MNG MRT CIV MDA UGA NPL CAF BFA BENKEN 7 TCDNGA COG MLI ZMB NER ETH GNBMDG ZAR TZA SLE AUS USA CAN

.2 .4 .6 .8 1 Overall technology adoption level in 1500AD

Figure3:(Conditional)overalltechnologyadoptionin1000B.C.and(conditional)current development

35

log per capita income in 2002

JPN

GRC

1
HRV IRN TUR DZA BIH CHN 0 EGY MAR

BWA NAM

PAK SDN -1 MRT TCD

MOZ

BFA BEN KEN TJK MLI COG NGA ZMB ETH ZAR TZA

-2

-.5 0 .5 Residual overall technology adoption level in 1000BC

Residual log per capita income in 2002

NOR DNK CHE AUT NLD BEL FRA DEU FIN GBR SWE ESP PRT KOR CZE HUN EST POL LTU ARG MYS LVA ZAF RUS CHL MEX CRI BGR THA ROM GAB URY BRA BLZ KAZ USA COL PAN TKM BLR CAN UKR AUS VEN PER GUY SLV SWZ PRY GTM IDN ECU NIC IND PNG VNM AGO HND GHA GIN BOL CMR LSO KHM GMB BGD LAO UZB MNG SEN CIV UGA NPL MDA

2
ITA

36

Figure4:(Conditional)overalltechnologyadoptionin0A.D.and(conditional)current development
NOR IRL DNK CHE2 AUT NLD BEL FRA DEU ITA FIN GBR SWE ESP ISR GRC PRT KOR CZE OMN SAU1 POL CHL ZAF RUS MEX MKD IRN TUR KAZ DZA USA TKM LBN CHN PERJOR0 SLV EGY MAR SWZ SYR GTM

ARG URY GUY

BWA GAB BLZ NAM

HUN SVK EST LTU HRV LVA MYS CRI BGR THA ROM BRA BIHDOM BLR COL PAN UKR CAN VEN ALB AUS PHL

PRY

PNG

AGO GHA CMR GMB MNG CIV BFA BEN MOZ COG NGA ZMB GNB ZAR MWI SLE

NZL IDN ECU INDNIC VNM HND KHM LAO NPL

BOLPAK LSO SDN BGD UZB-1 SEN MRT UGA CAF KEN TCD TJK MLI YEM NER MDG ETH TZA -2

-.6 -.4 -.2 0 .2 Residual overall technology adoption level in 0AD

37

Residual log per capita income in 2002

JPN SGP

Figure5:(Conditional)overalltechnologyadoptionin1500A.D.and(conditional)current development
IRL NOR DNK CHE AUT 2 NLD DEU FRA JPN BEL GBR FIN ITA SWE SGP ESP GRC MLTPRT KOR CZE HUN SVK SAU EST 1 POL LTU HRV LVA ARG MYS BWA ZAF RUS CHL MEX CRI THA TON BLZ ROM TUN IRN GAB URY BRA TUR NAM DZABLR BIH USA FJI COL PAN UKR CAN CHN AUS VEN 0 GUY PHL SLV PER MAR PRY EGY NZL SYR GTM IDN ECU IND NIC PNG VNM AGO GHA GIN BOL HND KHM CMR LSO PAK SDN BGD LAO UZB -1 SEN MNG MRT CIV MDA UGA NPL CAF BFA BENKEN TCDNGA COG MLI ZMB NER MDG ETH GNB ZAR TZA SLE

-2

-.6 -.4 -.2 0 .2 .4 Residual overall technology adoption level in 1500AD

Residual log per capita income in 2002

38

Figure6:(Conditional)overalltechnologyadoptionin1000B.C.and(conditional)current technology
Residual overall current technology adoption
CHE NOR FIN SWE DNK NLD FRA DEU GBR AUT JPN BEL ITA ESP EST PRT HKG HRV GRC

.4

.2

BWA NAM MOZ

LVA MYS CZE HUN BGR POL URY ZAF USA CRI CHL ROM LTU RUS ARG VEN COL BRA THA GAB MEX CAN TKM AUS PAN ECU MNG BOL CUB PER UKR PNG ZWE ZMB GNQ GUY COG NIC HND GTM MDA BLR CMR KAZ CIV IDN SEN KEN SWZ PRY SLVIND GHA TJK LAO MMR SOM ZAR BEN BGD NGA TZA VNM GMB UZB UGA KHM BFA AGO NPL LSO GIN ETH MLI

TUR BIH IRN EGY MAR MRT DZA CHN PAK AFG TCD YUG SDN

-.2

-.5 0 .5 Residual overall technology adoption level in 1000BC

39

 Figure7:(Conditional)overalltechnologyadoptionin0A.D.and(conditional)current technology
Residual overall current technology adoption
CHE NOR FIN SWE DNK.4 NLD FRA DEU GBR IRL AUT JPN SGP EST HRV SVK LVA MYS HUN BGR URY ARG GAB PNG GUY BWA MNG ZMB COG NAM CMR PRYCIV GHA LBR MOZ ZAR BEN NGA GMB BFA SLE AGO GNB CRI CHL ROM LTU VEN COL BRA THA CAN NZL BIH AUS PAN PHL ECU CUB UKR NIC HND ALB DOM BLR IDN IND LAO MMR YUG VNM KHM NPL BEL ISR ITA HKG ESP PRT GRC ARE.2 CZE LBY LBN POL USA ZAF SAU RUS JOR OMN 0 MEX TUR TKM MKD IRN IRQ BOLEGY MRT PERMAR ZWE DZA GTM KAZ PAK CHN SEN KEN SWZ SLV SYR YEM TJK AFG SOM MDG CAF BGD TCD SDN TZA UZB UGA NER-.2 LSO ETH MLI

-.6 -.4 -.2 0 .2 Residual overall technology adoption level in 0AD

40

Figure8:(Conditional)overalltechnologyadoptionin1500A.D.and(conditional)current technology
FIN CHE NOR SWE .4 DNK NLD FRA DEU GBR IRL AUT JPN BEL SGP ITA HKG ESP EST HRV PRT GRC .2 SVK

MYS

URY ZAF USA CRI ROM SAU CHL RUS LTU ARG VEN COL BRA THA GAB CAN MEX NZL TUR AUS BIH IRN TUN PAN BWA ECU PHL IRQ MNG EGYUKR MAR MRT BOL CUB PNG PER ZWE ZMB GNQ GUY COG NIC HND DZA GTM NAM CMR CIV KEN SEN PRY BLR IDN PAK MDA CHN SLV IND SYR GHA LAO MMR AFG LBRZAR BENCAF MDG BGD TCDNGA YUG TZA VNM UGA SDN UZB KHM SLE AGO GNB NER ETH NPL GIN BFA LSO MLI

CZE LVA HUN LBY POL

-.2

-.5 0 .5 Residual overall technology adoption level in 1500AD

Residual overall current technology adoption

41

Figure 9: Simulated Growth Rate of Technology 0.03

0.025 0.02

g advanced initial technology

g backward initial technology

0.015 0.01 0.005

-1000

1058

1205

1352

1499

1646

1793

42

1940

176

323

470

617

764

-853

-706

-559

-412

-265

-118

911

29

FigureA1:Technologyin1500andcurrenttechnology
USA CAN NZL AUS FIN CHE NOR SWE DNK NLD FRA DEU GBR IRL AUT .8 BEL JPN ITA HKG ESP EST HRV PRT GRC SVK

1 Current Overall Technology Adoption

SGP

MYS CZE LVA HUN ZAF LBY POL CRI CHL ROM RUS SAU ARG VEN LTU COL BRA MEX THA TUR GAB PAN ECU BIH IRN BOL CUB HND TUN PER ZWE NIC GTM BWA PHL IRQ MNG EGYUKR MAR MRT PNG COG GNQ GUY SLV DZA PRY NAM ZMB MDA CHN CMR CIV KEN SEN IDN PAK IND BLR SYR GHA LAO MMR AFG LBRZAR BENCAF MDG BGD TCDNGA YUG VNM LSOTZA UGA SDN UZB KHM SLE AGO GNB NERBFA NPL GIN ETH URY MLI

.6

.4

.2

.2 .4 .6 .8 1 Overall technology adoption level in 1500AD

43

Table 1: Mechanisms for persistence of technology from economic history literature Mechanism for technology persistence Examples from economic history literature Cement masonry CT roads and aqueducts under Romans Complementarities between existing (Mokyr 1990, pp. 19-20), Roman water lifting CT power technology and new technology (CT is transmission ( gears, cams, chains) (Mokyr 1990, p. 21), complementary to) If a new technology is medieval waterwheels CT grain mills (Rosenberg and complementary to an old technology, then the Birdzell 1986, p. 154), medieval heavy plow CT fallow cost of adopting the new technology is lower. system CT draft animals feeding on fallow and fertilizing The more technologies in the initial technology set, the more new technologies there will be that field CT horse collar and nailed horseshoe (Mokyr 1990 pp. 32-35), compass CT astrolabe CT advances in astronomy are complementary. CT oceangoing ships in 15th century (Mokyr 1990, p. 47, Rosenberg and Birdzell pp. 72, 84), metallurgy CT Gutenberg press (Mokyr 1990, p. 48), new crops alfalfa and clover CT stall feeding of livestock CT animal fertilizer and abandonment of fallow system 1500-1750 (Mokyr 1990, p. 58) Chain mail CT breastplates and armor for soldiers (Rosenberg and Birdzell 1986, p. 58), Textile machinery CT Chemical innovation on detergents, bleaches mordants, and dyes (Landes 1969, p. 108) the lever , the wedge and the screw , the ratchet , the Recombination of old technologies to make new technology. Many new technologies are pulley , the gear , and the cant used to make Greek/Roman novel combinations of old technologies. The war machines (Mokyr 1990, p. 21), salt preservation of meat greater the number of old technologies, the and improved transportation in 15th century = long distance greater the number of possible combinations to trade in cattle from rural areas to cities (Rosenberg and make new technologies. Birdzell 1986, pp. 75-76), steam power + iron and steel metallurgy = factory machines (Rosenberg and Birdzell 1986, p. 146), steam engine + rails = railroad, refrigeration + steamships = long distance meat exports to Europe, electricity generation + conductors + meters + lamps = electric light (Rosenberg 1982, pp. 58-59), germ theory+ bactericidal molds + mass chemical production = penicillin (Mokyr 2002, p. 107) internal combustion engine + glider + propeller=airplane (Mokyr 2002, p. 114) Working machines to Galileos general theory of machine Feedback from technology to science. When principles (Mokyr 1990, p. 75), Steam engine to techniques work, this gives new evidence to thermodynamics (Mokyr 1990, p. 90; Rosenberg 1982, p. scientists to test theories why they work. 14), food canning to bacteriology and germ theory of disease Science will in turn be used to make further (Rosenberg and Birdzell 1986, p. 245), Wright brothers to innovations in technique. Science should be aerodynamics (Mokyr 2005, Mokyr 2002, pp. 96-97) defined very broadly beyond formal science to Rosenberg 1982, p. 157), discovery of transistor to field of include general understanding of laws of solid-state physics (Rosenberg 2000, p. 33), telegraph to nature or the way the world is, making mathematical physics (Mokyr 2002, p. 90) statements that are true or false. Technology is defined as the toolkit of techniques, which are not true or false, but simply work. Greek/Roman alphabetization, Arabic numerals, Gutenberg Feedback from technology to lower access printing, 19th century innovation in paper and printing, lower costs for knowledge. The more advanced are transport costs for people and books, postal services, certain access technologies, the easier it is to obtain general scientific knowledge, which then standard weights and measures, encyclopedias, ICT revolution (Mokyr 2002) lowers the cost of innovation and adoption.

Table 1 (continued): Mechanisms for persistence of technology from economic history literature Clockmaking and watchmaking beginning with Spillover of technology from one sector to medieval town clocks spillover to precision another. Technological ideas from one sector machining for factories in industrial revolution inspire new approaches to problems in other (Rosenberg and Birdzell 1986, pp. 148-150), sectors. Steam engine inventor Watts background was in mining industry, which required knowledge of metallurgy, chemistry, mechanics, and civil engineering (Mokyr 1990 p. 162); spillover from petrochemical industry for auto fuel to plastics and synthetic fibers (Rosenberg 2000 p. 92) Larger oceangoing ships and expanding world Economies of scale. Some technologies have fixed costs or internal economies of scale such trade in 15th century (Rosenberg and Birdzell that it is only worth adopting them at higher 1986, pp. 82-83), Mechanical reaper (David scale of operation of the economy. The greater 1975), Assembly line and interchangeable parts is initial technology, the greater the scale of (Mokyr 2002) operation of the economy. Gutenberg printing press, Ocean-going ships in Economies of scope of General Purpose 15th century, Electric engine for factories in Technologies. When a GPT is invented, its payoff is greater, the more technologies in virtually all sectors and multiple home which it can be used. Hence the greater is appliances (Rosenberg 1982, pp. 78-79), ICT initial technology, the greater the payoff to a (Mokyr 2002, pp. 112-113) GPT. Medieval optics made possible telescope and Feedback from technology to improved lab microscope (Rosenberg and Birdzell 1986, p. equipment. Improved lab equipment lowers 58), instruments measuring time, distance, the cost of scientific discovery, innovation, or adoption. The better is existing technology, the weight, pressure, temperature, Voltas battery, Petri dish (Mokyr 2005) advances in lens better the lab equipment. grinding to make better telescopes and microscopes (Mokyr 2002, pp. 97-100) Ocean-going ships since 15th century improved Learning by doing. Much of technological progress consists of learning how to make old sailing efficiency and ship design until advent techniques work better through small of steam power (Rosenberg and Birdzell p. adjustments, minor innovations, and adaptation 263), coal required to generate kilowatt-hour of to local circumstances. electricity fell drastically over the decades; semiconductors moved from a single transistor on a chip to more than a million such components (Rosenberg 1994, pp. 14-15), high pressure steam engine design and transmission (Mokyr 2002, p. 84); transition from Bessemer to Siemens Martin steelmaking process (Mokyr 2002, pp. 86-87)

Table 2: Coding Concordance Between ACE and the Technology Adoption Dataset ACE Dataset Technology Dataset for 1000 B.C. & 0 A.D. ( 0 = indicates absence of technology, 1 = presence of technology) Communication 0,1 0,1 Industry 0,1 0,1 Transportation 0,1 0,1 Agriculture 0 1 2 Military Bronze weapons: 0,1 Iron weapons: 0,1

Writing & Records 1 = None 2 = Mnemonic or nonwritten records 3 = True Writing Technological Specialization 1 = None 2 = Pottery 3 = Metalwork (alloys, forging, casting) Land Transport 1 = Human Only 2 = Pack or draft animals 3 = Vehicles Agriculture 1 = None 2 = 10% or more, but secondary 3 = Primary Military 1 = Stone Tools 2 = Bronze Tools 3 = Iron Tools

Table 3: Variables in the 1500 A.D. dataset Variable Description Military Standing Army Cavalry Firearms Muskets Field Artillery Warfare capable ships Heavy Naval Guns Ships (+180 guns), +1500 ton deadweight An organization of professional soldiers. The use of soldiers mounted on horseback. Gunpowder based weapons The successor to the harquebus (the common firearm of European armies) was larger and a muzzle-loading firearm. Large guns that required a team of soldiers to operate. It had a larger caliber and greater range than small arms weapons. Ships that were used in battle are considered "warfare" capable. Ships required significant advances in hull technology before they were capable of carrying heavy guns. Large warships that only state navies had the capability of building. Agriculture Hunting & Gathering Pastoralism Hand Cultivation Plough Cultivation The primary form of subsistence. The primary form of subsistence. The primary form of subsistence. The primary form of subsistence. Transportation Ships Capable of Crossing the Atlantic Ocean Ships Capable of Crossing the Pacific Ocean Ships Capable of Reaching the Indian Ocean Wheel Magnetic Compass Horse powered vehicles Any ship that had successfully crossed the Atlantic Ocean. Any ship that had successfully crossed the Pacific Ocean. Any ship that had reached the Indian Ocean from either Europe or the Far East. The use of the wheel for transportation purposes. The most common use was for carts. The use of the compass for navigation. The use of horses for transportation. Communications Movable Block Printing Woodblock or block printing Books Paper The use of movable block printing. The use of woodblock printing. The use of books. The use of paper. Industry Steel Iron The presence of steel in a civilization. The presence of iron in a civilization. 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0 1 2 3 0,1 0,1 0,1 0,1 0,1 0,1 0,1 0,1 Values

Table 4: Descriptive statistics of Overall Technology Adoption Period 1000BC 0 1500AD Current Number Obs. 113 135 130 109 Average 0.45 0.73 0.48 0.45 Std. Dev. 0.28 0.28 0.32 0.2 Min 0 0 0 0.15 Max 1 1 1 1

Note: Overall technology adoption measure in a country results from ave adoption across tecnologies in the sector and then averaging sectoral adoption measures across the sectors.

Table 5: Descriptive statistics of Overall Technology Adoption by Continent Period 1000BC Europe Africa Asia America Oceania 0AD Europe Africa Asia America Oceania 1500AD Europe Africa Asia America Oceania Current Europe Africa Asia America Oceania 34 42 33 22 3 0.63 0.31 0.41 0.47 0.73 0.19 0.08 0.15 0.17 0.32 0.27 0.13 0.23 0.34 0.36 0.87 0.54 0.76 1 0.92 32 39 25 24 9 0.86 0.32 0.66 0.14 0.12 0.07 0.2 0.19 0.07 0.04 0.69 0.1 0.07 0 0 1 0.78 0.88 0.26 0.13 33 40 34 25 3 0.88 0.77 0.88 0.33 0.17 0.15 0.2 0.15 0.17 0.11 0.7 0.6 0.6 0 0.1 1 1 1 0.6 0.3 30 34 23 24 2 0.66 0.36 0.58 0.24 0.2 0.16 0.31 0.25 0.12 0.14 0.5 0 0.1 0 0.1 1 1 1 0.4 0.3 Continent Number Obs. Average Std. Dev. Min Max

Table 7: Average Overall Technology Adoption in Advanced Civilizations

Civilization 1000BC 0.65 0.9 0.67 0.95 1 0.7 0.43 0.9 0.7 0.31 1 0.88 0.33 0.96 0.94 0.71 0 AD 1500 AD Current

W. Europe

China

Indian

Arab

Note: W. Europe includes Spain, Portugal, Italy, France, United Kingdom, Germany, Belgium and Netherlands. Indian Empire includes India, Pakistan and Bangladesh. Arab Empire includes Saudi Arabia, UAE, Yemen, Oman, Iraq, Iran, Syria, Lebanon, Jordan, Egypt, Libya, Tunisia, Algeria and Morocco

Table 7: Cross-country variation in overall technology vs. sectoral technology Std Std. of deviations from overall tech. Agriculture 0.35 0.25 0.2 0.13 0.12 0.11 0.14 0.2 0.2 0.12 0.26 0.24 0.18 0.26 0.32 0.18 0.16 0.22 0.23 Industry Military Transport. Comm. Overall 0.28 0.28 0.32 0.2

Period

1000BC

113

136

1500AD

130

Current

152

Note: Std. Overall is the cross-country standard deviation in overall level of technology adoption. Std of deviations from overall tech. is the cross-country standard deviation in technology adoption level in the sector after removing for each country the overall technology adoption level

Table 8: Urbanization rate and technology adoption history

Dependent Variable: Urbanization rate in 2.08 (10.48) 1.69 (6.99) 8.04 (2.57) 0.39 (1.25) 113 0.5 135 0.58 54 0.18 106 0.48 0.16 (0.84) 126 0.59 1.68 (7.39) 1.96 (15.21)

1000 B.C.

0 A.D.

1500 A.D.

1000 B.C.

0 A.D.

1500 A.D.

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

947 (2.87) -3.75 (0.56) 51 0.2

Distance from Equator

N R2

Note: t-statistics in parenthesis computed using clustered standard errors.

All regressions include a constant

Table 9: Technology History and Current Development

Log Income per capita in 2002 I 0.75 (1.87) 0.09 (0.19) 1.59 (3.22) 1.76 (8.33) 0.1 (0.51) 8.2 (28.22) 104 0.03 123 0 8.45 (19.62) 7.77 (22.1) 110 0.18 8.5 (45.64) 110 0.08 2.47 (8.41) 0.62 (2.17) 7.68 (22) 105 0.17 2.85 (8.18) 0.86 (1.87) 7.21 (9.63) 123 0.13 1.49 (1.84) 3.03 (6.6) 3.31 (11.43) 1.51 (4.84) 6.74 (27) 110 0.5 1.45 (3.19) II III IV V VI VII

Dependent Variable

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

Major European Involvement

Minor European Involvement

Constant

R2

Note: t-statistics in parenthesis computed using clustered standard errors. Major European Involvement is a dummy that is 1 for US, Canada, New Zealand and Australia. Minor European Involvement is a dummy that is 1 mostly for Latin American, Caribean Countries and southern Africa.

Table 10: Technology History and Current Technology

Dependent Variable I 0.085 (1.37) 0.014 (0.14) 0.22 (2.58) 0.5 (17.35) -0.02 (0.95) 109 0.02 130 0 117 0.12 134 0.19 0.58 (11.8) 0.04 (0.94) 109 0.23 0.66 (8.57) 0.08 (1.48) 130 0.25 0.24 (2.14) 0.45 (6.47) 0.73 (21.47) 0.19 (5.1) 134 0.5 0.175 (2.71) II III IV V VI VII

Current Technology

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

Major European Involvement

Minor European Involvement

R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include a constant Major European Involvement is a dummy that is 1 for US, Canada, New Zealand and Australia. Minor European Involvement is a dummy that is 1 mostly for Latin American, Caribean Countries and southern Africa.

Table 11: Primitive Technology and Current Development, Robustness

Log Income per capita in 2002 I 0.21 (0.36) 0.64 (1.6) 1.52 (1.94) 1.71 (9.28) -0.35 (1.75) 0.42 (1.79) 0.18 (1.05) 0.64 0.44 -0.18 2.24 (0.33) (0.15) (0.08) (1.38) 0.14 (0.93) -0.16 (0.53) 1.08 0.62 (0.5) (0.26) 0.36 (1.11) -0.54 (0.9) -0.69 (2.33) -1.12 (2.27) 1.55 (3.77) 0.39 (.54) 1.6 (2.37) 1.33 (2) 2.37 (4.06) 0.15 (0.34) 0.26 (0.67) 0.74 (1.29) 0.27 (0.93) -0.08 (0.31) 0.39 (0.95) II III IV V VI VII VIII IX X XI XII

Dependent Variable

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

Europe dummy

Africa dummy

Asia dummy

America dummy

Distance to equator

Distance to equator squared

5.01 5.72 4.33 2.82 4.58 3.35 (2.12) (1.64) (1.41) (1.34) (1.79) (1.05) -1.14 (4.49) -0.76 -0.74 -0.66 (6.53) (6.15) (3.1) 104 0.58 0.61 123 110 0.62 97 0.56 114 0.53 103 0.59 97 0.63 114 0.59 103 0.63 -0.7 (3.85) 104 0.41 -1.16 (5.28) -0.74 (4.4) 123 0.41 -0.46 (1.69) -0.52 (2.33) 110 0.54

Tropical Dummy

Land-Locked

R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include major and minor European involvement dummies and a co

Table 12: Primitive and Current Technology, Robustness

Current Technonology I V -0.02 (0.37) 0.08 (1.32) 0.17 (2.18) 0.31 (4.46) -0.03 (1.06) 0.04 (1.12) 0.03 (1.06) -0.59 (2.07) 1.92 (4.41) 1.8 (3.4) -0.54 (1.29) -0.88 (2.73) 2.06 (4.1) 0 0 (0.18) (0.32) -0.37 (1.35) 1.61 (3.75) -0.46 (1.39) 1.67 (3.83) -0.79 (2.41) 1.93 (3.82) -0.18 (4.42) -0.09 (5.12) 109 0.62 0.6 130 115 0.63 101 0.65 120 0.62 107 0.7 101 0.68 -0.09 (5.17) 120 0.66 -0.06 (1.83) 107 0.71 -0.11 (3.39) 109 0.41 -0.16 (5.71) -0.11 (3.81) 130 0.43 -0.07 (1.44) -0.055 (1.55) 115 0.52 0 (0.1) -0.03 (0.48) -0.1 -0.05 (1.89) (1.58) 0.22 0.2 (2.36) (2.24) 0.24 (2.82) 0.22 (2.49) 0.11 (1.91) 0.16 (2.12) 0.36 (4.13) 0 (0.14) VI VII VIII IX X XI XII -0.02 (0.32) 0.16 (1.76) 0.02 (0.52) II III IV

Dependent Variable

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

Europe dummy

Africa dummy

Asia dummy

America dummy

Distance to equator

Distance to equator squared

Tropical Dummy

Land-Locked

R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include major and minor European involvement dummies and a constant.

Table 13: Technology history, current technology and per capita GDP after controlling for arable land

Dependent Variable

Log per capita GDP 2002 I II 1.43 (2.83) 1.51 (1.79) 3.38 (7.37) 0.02 -0.008 -0.03 (0.43) (0.21) (0.88) NO 127 0 101 0.17 YES 120 0.14 108 0.55 -0.17 (4.13) III IV V

Current technology VI 0.19 (2.68) 0.25 (2.12) 0.53 (7.43) 0.01 -0.01 (0.48) (0.96) NO 113 0 106 0.23 -0.01 (1.74) YES 127 0.26 113 0.57 -0.03 (4.36) VII VIII

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

Log arable land area

Major and Minor european influence dummies N R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include a constant.

Table 14: Correlation of technology adoption measures over time

Overall 0.62 0 0.69 0 0.69 0 0.42 0 0.69 0 0.66 0 0.24 0.017 0.6 0 0.41 0 0.51 0 0.64 0 0.52 0 0.71 0 0.39 0 0.51 0 0.39 0 0.32 0 0.64 0

Agri.

Military

Industry

Comm.

Transport.

Correlation (1000BC, 0AD) p-value

Correlation (0AD, 1500 AD) p-value

Correlation (1000BC, 1500AD) p-value

Table 15: Correlation of technology adoption measures over time

Dependent Variable 0.66 (5.74) 0.66 (9.97) -0.69 (1.17) 1.3 (1.54) 0.17 (3.79) NO 103 0.51 105 0.7 94 0.66 103 0.75 -0.11 (2.39) 0 (0.1) 0.3 (0.4) 1.12 (1.68) 0.1 (0.12) 0.06 (1.61) 0.48 (0.91) -0.05 (0.1) -0.11 (0.25) 0.36 (3.92) 0.54 (1.67) -0.87 (1.86) -0.08 (1.82) YES 105 0.87 91 0.84 0.65 (2.08) -0.92 (1.92) -0.07 (1.55) 0.64 (5.24) 0.41 (6.3) 0.24 (2.44)

Overal technology adoption in:

0 A.D.

1500 A.D. 1500 A.D.

0 A.D.

1500 A.D.

1500 A.D.

Overall Technology Adoption in 1000 B.C.

Overall Technology Adoption in 0 A.D.

Distance from Equator

Distance from Equator squared

Land-Locked

Continent dummies

N R2

Note: t-statistics in parenthesis computed using clustered standard errors.

Table 16: Current technology and development

Dependent Variable: I 5.16 (22.42) 0.44 (1.63) 0.89 (3.05) -0.27 (2.11) NO NO 123 0.8 115 0.81 123 0.83 NO NO YES NO NO NO 106 0.83 0.72 (1.66) -0.33 (2.47) NO NO 99 0.85 YES NO 106 0.86 0.16 (0.51) 0.74 (1.17) 4.47 (16.27) 4.57 (16.31) 4.9 (17.6) 4.42 (13.38) 4.51 (11.55) II (log) per capita income in 2002 III IV V VI VII 4.55 (14.34) 0.57 (1.23) 0.53 (1.36) -0.27 (2.48) NO YES 99 0.88

VIII 4.83 (11.49) 0.78 (1.37)

Current Technology adoption

Overall Technology adoption level in 1500AD

Distance from Equator

Land-locked

Continent Dummies European Influence Dummies

YES YES 106 0.87

N R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include a constant

Table 17: Population technology and development

Dependent Variable: 3.35 (7.75) -0.16 (3.13) 0.69 (6.11) 0.04 (1.69) 105 0.55 110 0.53 115 0.72 98 0.49 -0.02 (1.67) 0.85 (19.95) 0.48 (1.7) 0.77 (14.15) 98 0.8 0.49 (6.15) 0.69 (1.37)

Y/L

Current Technology Population

1500 AD Technology Population

Overall Technology adoption level in 1500AD

(Log) Population in 1500 AD

Overall Technology adoption level in 0AD

(Log) Population in 0AD

N R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include a constant. The regressions with current dependent variables include European influence dummies.

Table 18: Population technology and development

Dependent Variable: 3.39 (8.12) -0.25 (2.19) 0.06 (0.49) NO 105 0.58 109 0.57 105 0.68 NO YES YES 109 0.69 0.02 (1.09) 0 0.01) 0.01 (0.61) -0.05 (2.94) -0.06 (0.52) -0.015 (0.76) 0.5 (6.62) 1.5 (1.98) 0.15 (1.73)

Y/L

Current Technology Y/L Technology

Overall Technology adoption level in 1500AD

(Log) Population in 2000 AD

(Log) Population in 1500 AD

Continent Dummies

N R2

Note: t-statistics in parenthesis computed using clustered standard errors. All regressions include a constant and European influence dummies

Table 19: Peristence of Technology within sectors

Dependent Variable: I 0.95 (7.66) YES YES NO 405 405 405 0.4 405 0.35 303 0.38 YES NO NO YES YES YES YES NO YES YES YES YES Agri. YES YES YES Comm. 300 0.45 0.85 (7.39) 0.31 (4.67) 0.41 (5.47) 0.17 (2.55) 0.27 (3.07) II III Technologycst IV V VI VII 0.46 (4.54) YES YES YES Transp. 305 0.18

VIII 0.34 (4.64) YES YES YES Indust. 307 0.53

Technologycst-1

Country-Sector Fixed effects Country time-varying effects Technology-specific trends Sectors excluded other than military

N R2

Note: Panel regression using technology adoption level in sector, s, in country, c, in year t. t includes years 2000 AD and 1500 AD. Most general regression is: Technologycst = cs+ ct+ s*t+ *Technologycst-1+ cst where cs is a country-sector fixed effect, ct is a country time-varying effect, s is a technology specific trend and cst is the error term.
ct- ct-1+ s+

Regressions are estimated taking time differences as follows: Technologycst-Technologycst-1 =

*(Technologycst-1-Technologycst-2)+

cst- cst-1

To avoid biases due to lagged dependent variable, (Technologycst-1 -Technologycst-2) is instrumented with Technologycst-2

t-statistics in parenthesis computed using clustered standard errors. All regressions include European influence dummies when t=2000.

Table A.1: Primitive technology and current development, robustness checks

1913 I -0.36 (0.62) 1.08 (1.62) 2.14 (4.91) 51 0.18 60 0.28 57 0.4 95 0.17 116 0.18 1.87 (4.85) 103 0.4 107 0.14 129 0.12 1.27 (2.6) 0.78 (2.32) 1.14 (2.28) 1.35 (1.81) 2.98 (6.47) 113 0.5 II III IV V VI IV V 1960 1990 VI

Dependent Variable: Log Income per capita in

Overall Technology adoption level in 1000BC

Overall Technology adoption level in 0

Overall Technology adoption level in 1500AD

N R2

Note: t-statistics in parenthesis computed using clustered standard errors. Major and minor European inf;uence dummies included in all regressions.

Table A2: Average adoption level by sector in the Empires Year Empire Sectors Agriculture 1000 B.C. W. Europe China Indian Arab 0 A.C. W. Europe China Indian Arab 1500 A.D. W. Europe China Indian Arab 1 1 1 0.85 1 1 1 0.94 1 1 0.5 0.527 0.6875 0.66 0.55 0.57 0.968 0.75 0.458 0.61 1 1 1 1 1 1 1 1 0.875 1 0.66 1 0.9375 1 0.83 1 1 1 1 1 1 1 1 1 1 1 0.83 1 0.125 1 0.33 1 0.5625 1 0.5 1 0.56 0.5 0.66 0.75 Industry Communications Transportation Military

Note: W. Europe includes Spain, Portugal, Italy, France, United Kingdom, Germany, Belgium and Netherlands. Indian Empire includes India, Pakistan and Bangladesh. Arab Empire includes Saudi Arabia, UAE, Yemen, Oman, Iraq, Iran, Syria, Lebanon, Jordan, Egypt, Libya, Tunisia, Algeria and Morocco