Reassessing patent propensity: evidence from a dataset of R&D awards, 1977-2004

(1)

Other uses, including reproduction and distribution, or selling or

licensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of the

article (e.g. in Word or Tex form) to their personal website or

institutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies are

encouraged to visit:

(2)

ContentslistsavailableatScienceDirect

Research

Policy

jou rn a l h o m e pa g e :w w w . e l s e v i e r . c o m / l o c a t e / r e s p o l

Reassessing

patent

propensity:

Evidence

from

a

dataset

of

R&D

awards,

1977–2004

Roberto

Fontana

a,∗

_,

_Alessandro

_Nuvolari

b

_,

_Hiroshi

_Shimizu

c

_,

_Andrea

_Vezzulli

d

a_Department_of_Economics_ad_Management,_University_of_Pavia,_Via_San_Felice_5,_27100,_Pavia_&_CRIOS_–_Bocconi_University,_Via_Sarfatti_25,₂₀₁₃₉_Milano, Italy

b_LEM_–_Sant’_Anna_School_of_Advanced_Studies,_Piazza_Martiri_della_Libertà_33,₅₆₁₂₇_Pisa,_Italy c_Institute_of_Innovation_Research_–_Hitotsubashi_University,_Tokyo,_Japan

d_UECE-ISEG,_Universitade_Técnica_de_Lisboa,_Rua_Miguel_Lupi,_20,_1249-078_Lisboa,_Portugal

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received28November2011 Accepted25May2012

Available online 16 September 2013 Keywords:

Innovation Patentpropensity R&Dawards

a

b

s

t

r

a

c

t

Itiswellknownthatnotallinnovationsarepatented,buttheexactvolumeofinnovativeactivities undertakenoutsidethecoverageofpatentprotectionand,relatedly,theactualpropensitytopatentan innovationindifferentcontextsremain,toamajordegree,amatterofspeculation.Thispaperpresents anexploratorystudycomparingsystematicallypatentedandunpatentedinnovationsovertheperiod 1977–2004acrossindustrialsectors.Themaindatasourceisthe‘R&D100Awards’competitionorganized bythejournalResearchandDevelopment.Since1963,themagazinehasbeenawardingthisprizetothe 100mosttechnologicallysigniﬁcantnewproductsavailableforsaleorlicensingintheyearprecedingthe judgments.WematchtheproductswinnersoftheR&D100awardscompetitionwithUSPTOpatentsand weexaminethevariationofpatentpropensityacrossdifferentcontexts(industries,geographicalareas andorganizations).Finallywecompareourﬁndingswithpreviousassessmentsofpatentpropensity basedonseveralsourcesofdata.

1. Introduction

WithintheEconomicsofTechnicalchangeandInnovation

Stud-ies(ETIS)literatureitis todaywidelyacknowledged thatmany

innovationsarenotpatented.Inprinciple,theremaybethreetypes

ofexplanationaccountingfortheinventor’sdecisionofnottaking

apatent(Basberg,1987).Theﬁrstexplanationisthatthe

innova-tionissimplynot patentable.Inthiscase,theinventorbelieves

thattheinnovationinquestiondoesnotrepresentsuitablepatent

matter(e.g.thepatentabilityof‘pure’softwareprogramwasstilla

matterofcontentioninmanyjurisdictionsnotsolongago).

Alter-natively,theinnovationisinprinciplepatentablebuttheinventor

mayanticipatethattheinventivestepembodiedinherinnovation

isnot‘high’enoughtobedeemedworthyofpatentprotectionby

patentexaminers.Inboththesetwoexamplesthedecisionofnot

patentingisdeterminedbythefactthatthisisnotactuallypossible

(orbelievedpossible).Thethirdpossibilityisthattheinventor,even

whenconceivingtakingapatentasafullyfeasiblecourseofaction,

∗ Correspondingauthor.Tel.:+390258363037;fax:+390258363399. E-mailaddresses:roberto.fontana@unibocconi.it(R.Fontana),

alessandro.nuvolari@sssup.it(A.Nuvolari),shimizu@iir.hit-u.ac.jp(H.Shimizu),

andreav@iseg.utl.pt(A.Vezzulli).

decidesnottopatenttheinnovationbecausesheactuallyprefers

todoso.Inthiscase,eventhoughtheinnovationispatentableand

worthpatenting,theinventorprefersindustrialsecrecyorother

alternativestrategiestoextractsomeeconomicreturnsfromher

innovation.Thisthirdcaseisthemostinterestingonefromthe

viewpointofinnovationscholars.

Theexistenceof‘appropriabilitystrategies’thatarealternative

topatentingwasinitiallydocumentedbyearlyeconomistsof

inno-vation(Kuznets,1962;Schmookler,1966;TaylorandSilberston,

1973).Lateron,thesurveystudiesbyMansﬁeld(1986)andLevin

etal.(1987)duringthe1980shighlightedthat,inmostindustries,

patentprotectionwasnotthetypicaltooladoptedbyﬁrmsforthe

extractionofeconomicreturnsfrominnovations,aﬁndingfurther

corroboratedbysubsequentresearchbothinUS(Cohenetal.,2000)

andEurope(ArundelandKabla,1998).Alltheseresearchresultsare

frequentlycitedandsurelyrepresentimportantpiecesofevidence

discussedintheinnovationliterature.However,asaptlynotedby

DeRasenfosse(2010),oncloserinspection,itisdifﬁculttoavoid

theimpressionthatthemajorimplicationoftheseﬁndings(i.e.

thata sizeableshareofinnovationsis neverpatented)hasgone

completelyneglected.Tobesure,manyempiricalinvestigations

acknowledgethelimitationsofpatentsasinnovationindicators.

However,oncetheselimitationsaregaugedagainsttheir

advan-tages(i.e.availability andrichnessofinformationtheyprovide)

(3)

theﬁnal choiceistorelyonpatentsifanythingbecauseofthe

sheerdifﬁcultyofconstructingsuitableindicatorsusingalternative

sources.1

Thisstateofaffairsisdeeplyunsatisfactoryasweincurtherisk

thateventhemostcarefullydesignedempiricalstudieswillprovide

uswitha partial,and sometimeseven distorted,representation

ofinnovativeactivities.Furthermore,sinceourunderstandingof

patentpropensityindifferentcontextsisstillrudimentary,inmany

cases,it isalsodifﬁculttoformulateasoundassessmentofthe

margin of error and of the biases involved in theadoption of

patentsasinnovationindicators.Consider,forinstance,the

con-ceptof‘propensitytopatent’usuallydeﬁnedintheliteratureas

theratiobetweenpatentsandR&Dexpenditures(Scherer,1983;

HallandZiedonis,2001).Thoughsurelylegitimate,weshouldnote

thatthisdeﬁnitionofpatentpropensityissimplydescribingthe

overallrelationshipbetweenpatentsandinnovativeeffortsandit

isonlyindirectlylinkedtotheactualdecisiontopatentornota

speciﬁcinnovation(seeDeRasenfosse,2010foramoreextensive

discussion).

Interestingly enough, economic historians and historians of

technologyhaveinsteadadoptedamore‘straightforward’

deﬁni-tionofpatentpropensity,namelytheshareofpatentedinnovations

in the total number of innovations occurring in a given time

period(Sullivan,1989;Moser,2005,2012).2_This

conceptualiza-tionofpatentpropensity,althoughintuitivelyappealing,isnotof

immediateempiricaloperationalizationbecauseitrequiressome

form of direct assessment of the total amount of innovations

occurringinagiventimeperiod.Still,historianshavedisplayed

considerableingenuitybothintheidentiﬁcationofsources

(alter-nativetopatents)thatcouldbeusedforformulatingquantitative

assessmentsofoverallinnovativeoutputindifferentcontextsand

periods,andinconnectingthesesourceswiththepatentevidence

forconstructing estimatesof patent propensity.In thisrespect,

therecentcontributionsofMoser(2005,2012)canberegarded

asamongoneofthemostsuccessfulexamplesofthisapproach.

Thispaperarguesthatthesehistoricalinvestigationssuggesta

frameworkofinquirythatcan,andshould,befruitfullyextended

withintheﬁeldofETIS.Inthepaperwepresentanapplicationof

thismethodusingadatabaseof‘important’industrialinnovations

occurredbetween1977and2004.Oursourceofdataisthe‘R&D

100Awards’competitionorganizedbythejournalResearchand

Development.Since1963,thisjournal(whichatthattimewascalled

IndustrialResearch)hasbeenawardingaprizeto100most

techno-logicallysigniﬁcantnewproductsavailableforsaleorlicensingin

theyearprecedingthejudgement.Thepotentialofthissourcewas

alreadyreckonedbyCarpenteretal.(1981)andScherer(1989).

ThissourcehasalsobeenmorerecentlyusedbyBlockandKeller

(2009)todocumenttheincreasingroleofpublicinstitutionsand

publicfundinginthegenerationofinnovationsintheUSeconomy

intheperiod1971–2006.

Though recentandtherefore not‘historical’ina strictsense,

thedatabase covers 30 years of innovations, several

manufac-turing industries, and different types of economic actors, both

corporationsandUniversitiesandPublicResearchOrganizations

(PROs).Thesedataseemparticularlyappropriateforstudyingthe

propensitytopatentforthefollowingreasons:(i)thedata

con-siderinnovationsthathavebeenrecognizedbyajuryofexperts

1_In_this_respect,_we_would_argue_that_a_large_bulk_of_the_most_recent_research oninnovationseemsimplicitlytoapplytopatentsWinstonChurchill’sfamousquip ondemocracy:“Democracyistheworstformofgovernment,exceptalltheothers”. Inthecaseofpatents,thejibewouldprobablysoundlike:“patentsaretheworst innovationindicator,exceptalltheothers”.

2_Moser_(2005,₂₀₁₂₎_uses_the_term_‘patenting_rate’_to_deﬁne_the_share_of_patented

inventionsinthetotalnumberofinventions.

assigniﬁcantandtheyshouldbecommerciallyfeasibleatthetime

oftheawards;(ii)mostoftheawardshavebeengrantedtolarge

corporationsaccountingforasizeableamountoftotalR&D

invest-ments;(iii)thedatacoverarelativelylongtimeperiodallowing

ustotakeintoaccountchangesinthedeterminantsofthe

propen-sitytopatent overtime.Usingthesedataweareabletoassess

systematicallytherelativeinﬂuenceof sector,organization,and

inventorspeciﬁccharacteristicsontheactualdecisionoftakingor

notapatent.

Ourstudyisbasedonasampleofabout3000innovationsthat

have receivedan award.Foreach innovationin ourdatasetwe

haveretrievedinformationconcerning:yearsoftheaward,

descrip-tionoftheinnovation,typeandnameofapplicantorganization(s),

applicationdomainoftheinnovation,country,andnameof

inven-tor(s).Theﬁrststepofouranalysisistomatchawardedinnovations

withpatentsusingthesearchengineoftheUSPTOwebsite.Then,

onthebasisoftheinventiondescriptioncontainedinthejournal,

weclassifyalltheawardedinnovationsinthirtydifferentsectorsof

activity.Inthisway,thedataallowathoroughcomparisonbetween

patentedandnotpatentedinnovationsacrossdifferentindustrial

sectors,countries,typesoforganizationandtypesofinnovation.

Ourresultshighlightthefollowingpatterns.First,alargeshareof

innovationsisnotprotectedbymeansofpatents.Second,weare

abletopointouttheexistenceofsystematicsigniﬁcantdifferences

inpatentingpropensityacrosssectors,geographicalareas,typesof

organizationandtypesofinnovation.

Thestructureofthepaperisasfollows.Section2reviewsthe

empiricalliteratureontheeffectivenessofpatentsas

appropriabil-itytools.Section3describesindetailourdatasource,ourmatching

procedureandthelimitations ofthedataset.Section4presents

ouranalysisofpatentpropensityacrossdifferentdimensions.

Sec-tion5comparesourﬁndingswiththoseofpreviousassessments

ofpatentpropensitycarriedoutusingdifferenttypesofdata.

Sec-tion6concludesanddrawssomemethodologicalimplicationswith

particularreferencetothepossibilityofextendingtheframework

ofinquiryadoptedbyeconomichistoriansandhistoriansof

tech-nologytocontemporarystudiesofpatentpropensity.

2. Patentsasindicatorsofinnovationandtheirlimitations

Scholarswithinthe ETIStradition haverelied intensivelyon

patents to investigate the sources, nature, and the effects of

innovativeactivities.Innovativeactivitiesareinherentlyelusive

phenomenawhichalmostbydeﬁnitionareboundtodefy

system-aticattemptsof(quantitative)measurement.Itisnotsurprising

thenthat theexistenceofpatentrecordshasbeenregardedfor

alongtime,mostlybyeconomists,butalsobyotherscholarsof

innovationwithdifferentdisciplinarybackgrounds,asanalmost

uniquesourceofinsightsintothenatureofinventiveactivities.

Themainmeritsofpatentrecordsasasourceformeasuring

inno-vationareeasytosummarize:(i)theyarebydeﬁnitionrelatedto

innovativeactivities;3 _(ii)_they_are_readily_available_(allowing_to

economizeeffortsofdatacollection);4 _(iii)_they_are_available_for

relativelylongperiodsoftime;(iv)theycontainasigniﬁcantdepth

ofinformation(inventors’namesandaddresses,ownershipofthe

3_In_the_words_of_Griliches_(1990,_p.₁₁₆₉₎_:_“[A]_patent_represents_a_minimal quan-tumofinventionthathaspassedboththescrutinyofthepatentofﬁceastoitsnovelty andthetestoftheinvestmentofeffortandresourcesbytheinventorandhis organi-zationintothedevelopmentofthisproductoridea,indicatingtherebythepresence ofanon-negligibleexpectationastoitsultimateutilityandmarketability”.

4_The_{‘accessibility’}_of_patent_as_a_source_has_greatly_increased_over_the_last₂₀ yearsorsothankstothecreationofonlinesearchenginesuchasESPACENET andtheeffortsofconstructionofdata-basescontaininginformationgatheredby patentrecordssuchastheNBER-USpatentdata-set(Halletal.,2001)andthe OECD-PATSTATdataset.

(4)

innovation,descriptionoftheinnovationanditsrelationwith

pre-viousones,asrepresentedbypatentcitations).Thesefactorshave

madepatentsthemostadoptedindicatorforscholarsinterested

especiallyinmeasuringtheoutputofinnovationactivities.

Althoughmuchprogresshasbeenachievedin thisway,itis

wellknownthatindicatorsofinnovationbasedonpatentssuffer

fromseverallimitations.Theselimitationscanbesummarized

cit-ingagainfromGriliches(1990,p.1169):“Notallinventionsare

patentable.Notallinventionsarepatentedandtheinventionsthat

arepatenteddiffergreatlyintheir‘quality’,inthemagnitudeof

inventiveoutputassociatedwiththem”.

Theﬁrstlimitationisclearlythemostobviousoneand

prob-ablytheeasiesttotackle.Somedomainsofinventiveactivitydo

notconstitutepatentablesubjectmatter.Thesolutionistoresort

toalternativeindicators forassessing inventiveoutputin these

areas.Thesecondlimitationisthatnotallpatentableinnovations

areactuallypatented.Thismeansthatincontextscharacterized

by low patent propensity, i.e. in environments in which ﬁrms

prefertoadoptalternativeappropriability strategies,theuseof

patentsasinnovationindicatormayresultinabiasedassessment

ofthevolumeofinnovativeactivities.5_The_existence_of_a_different

propensitytopatentacrossindustrieshasindeedbeenthemost

importantﬁndingofstudiesbasedonsurveysoftheattitudesof

R&Dpersonneltowardstheuseofpatents.Mansﬁeld(1986)

exam-ined how many patentable innovationswere actually patented

inarandomsampleoflargeUSﬁrmsindifferentindustries.His

resultshighlightedthatinsectorswherepatentsarenotregarded

asparticularlyeffective‘appropriability’mechanisms(i.e.primary

metals,electricalequipment,instruments,ofﬁceequipment,motor

vehicles,rubberandtextiles)around34%ofpotentiallypatentable

inventionswerenotpatented.Thispercentageisaround16%in

those sectors where patentsare consideredtobe more

impor-tant(i.e.pharmaceuticals,chemicals,petroleum,machinery,and

fabricatedmetalproducts).Resultsfromsubsequentsurveys

cor-roboratetheseearlyﬁndings.Insomeindustries,secrecyandlead

timesseemtobemoreimportantthanpatentingforappropriating

thereturnsfrominnovation(Levinetal.,1987).Moreover,patent

propensityvariesdependingonthetypeofinnovation,withﬁrms

morelikelytoapplyforapatentforproductinnovationthanfor

pro-cessinnovation(Cohenetal.,2000).6_Arundel_and_Kabla₍₁₉₉₈₎_also

ﬁndthatthe(salesweighted)propensitytopatentdiffersacross

innovationtypewithrelativelylowerratesforprocessinnovation

(24.8%)thanforproductinnovations(35.9%).

Among the available alternatives, secrecy seems to play an

importantrole inprotecting innovation.Lookingata sampleof

Europeaninnovativeﬁrms,Arundel(2001)ﬁnds thatsecrecy is

generallyratedasmorevaluablethanpatenting.Thisisparticularly

trueinthecaseofproductinnovation,thoughtheprobabilityof

beingconsideredmorevaluabledeclineswithﬁrmsize.Hussinger

(2007)carriesoutasimilaranalysisusingsalesﬁgurestoassess

theimportanceofalternativemeansofappropriationratherthan

5 _The_fact_that_firms_in_some_contexts_prefer_alternative_{appropriability}_strategies topatentingdoesnotimplythatpatentsarecompletelyirrelevantasameasure ofinnovation.Forexample,insemiconductors,appropriabilitystrategiesarebased moreonsecrecyandleadtimesthanonpatents,butfirmsareincreasinglyresorting topatentprotectioninordertousepatentsas“bargainingchips”innegotiation withotherfirms(HallandZiedonis,2001).Asaresult,eventhoughinthisindustry patentsaretakenforstrategicmotivesratherthanforreapingeconomicreturns fromaspecificinnovation,theystillprovideausefulmeasureofinnovativeactivities. 6 _The_survey_{questionnaire}_of_Levin_et_al.₍₁₉₈₇₎_did_not_contain_a_specific ques-tionaskingwhatpercentageofinnovationafirmtypicallypatented,butsimply containedaquestionaskingtoassesstherelativeeffectivenessofdifferent appropri-abilitystrategies.ThesurveyquestionnaireofCohenetal.(2000)insteadcontained aspecificquestionaskingrespondentswhatpercentageoftheirinnovationswas patented.

individualevaluations.Herﬁndings suggestthatsecrecyis

rela-tivelymoreimportantforinnovationsthatarenotcommercialized.

Thethirdlimitationisthatpatentsandinnovationsdiffergreatly

in theirtechnologicaland economic signiﬁcance.7 _In _particular,

severalstudieshaveshownthatthe‘size’distributionof

innova-tionsissharplyskewedwiththemajorityofinnovationsoflittleor

limitedtechnicalandeconomicsigniﬁcanceandarestricted

num-berof highlysigniﬁcantinnovations(Silverbergand Verspagen,

2007).Innovationscholarshaveattemptedtodealwiththis

prob-lembyweightingpatentsusingcitationsorotherinformationsuch

asclaims,andfamilysize(Trajtenberg,1990).Still,itis

acknowl-edgedthatthesemethodsrepresentonlyimperfectproxiesofthe

qualityoftheinnovationunderlyingthepatentinquestion.Infact,

themostsensibleuseoftheseproxiesofpatentvalueistousethem

as‘probabilisticmarkers’oftheunderlyingeconomicvalueofthe

patentsandemploythemfortheidentiﬁcationofgroupsof

poten-tiallyvaluablepatents(vanZeebroeck,2011).Clearlynottaking

properlyintoaccountthesevariationsintheunderlyingvalueof

patentsmayagainleadtobiasedassessmentsofinventiveoutput.

Economichistoriansseemtohavebeenmoresensitivetothe

limitationsofpatentsasinnovationindicatorsandhaveexplored

thepotentialitiesofalternativesources.Moser(2005,2012)has

constructedadatasetofinnovationsonthebasisofthecatalogues

ofnineteenthcenturyindustrialexhibitions(inhercaseshehas

usedtheCrystalPalaceexhibitionof1851andtheCentennial

exhi-bitioninPhiladelphiain1876).AccordingtoMoser(2002,pp.1–2),

thistypeofdatacovers“economicallyusefulinnovations”(i.e.the

commercial introduction of newproducts or processes), rather

than “inventions” (i.e.theadditionsto thestock of

technologi-calknowledge).Furthermore,exhibitiondatameasureinnovations

regardlessofwhethertheyarepatentedornot.Moser’sﬁndings

haveproducednovelinsightsonthesourcesofinnovationacross

countries and sectors duringthe secondhalf of thenineteenth

century.Herﬁndings showthatin1851,89% ofBritish

innova-tionsondisplayattheCrystalpalaceexhibitionwerenotpatented.

Evenamongprizewinninginnovations,84%werenotpatented.

Moreover,sheﬁndsthatpatentpropensityisaffectedbythe

char-acteristicsofthesectorswheretheinnovationoccurs,thelocation

oftheinvention(urbanvs.rural),andthequalityoftheinvention

(incrementalvs.radical).

Brunt et al. (2012)have insteadrelied uponinformationon

prizesandawards.Usingadatasetofawardsforinventions

pro-motedbytheRoyalAgriculturalSocietyofEnglandfrom1839to

1939,theyhavestudied,amongotheraspects,whetherandhow

prizesaffectinnovationandpatenting.Theirﬁndingspointtothe

presenceofapositiverelationshipbetweenprizesandpatenting.In

particular,thepropensitytopatentinthetechnologycategory

tar-getedbytheawardincreasedby42%forthoseinventionsawarded

a goldmedal.Moreover,patentsaremore likelytoberenewed

whentheyweretakenoutofawardedinventionsthussuggestinga

positiverelationshipbetweenawardsandthequalityofthepatent.

Theserecentcontributionsbyhistoriansclearlyillustratethe

potentialities of assessments of innovative output using data

sourcesthatarealternativetopatents.8 _In_particular,_we_would

liketosuggestthatinnovationscholarsshouldconsiderwith

par-ticularinterest prizeand exhibitiondata,becausesomeoftheir

intrinsiccharacteristicsarelikelytorenderthemlesspronetothe

7_In_the_words_of_Kuznets_(1962,_p.₃₇₎_:_“[T]he_main_difﬁculty_with_patent_statistics is,ofcourse,theenormousrangeinthemagnitudeoftheinventionscovered[...] patentedinventionsdodifferwidelyintheirpotentialeconomicmagnitude”.

8_{Interestingly,}_Schmookler,_one_of_the_pioneer_of_the_use_of_patent_statistic_in_the ﬁeldofETIS,wasalsooneoftheearlyscholarstoargueinfavourofcross-checking theassessmentsofinventiveoutputbasedonpatentwithdataon‘important inno-vations’(Schmookler,1966).

(5)

pitfallsthattypicallyaffectpatentdata.First,bothprizeand

exhi-bitiondatarefertovaluable,ortouseMoser’swords“economically

useful”,innovations.Inthecaseofprizesthisisalmostatautology

giventhattheyhavebeenrecognizedbyexpertsintheﬁeldas

supe-riortoalternativeavailablesolutionsandprobablyalsotoexisting

practices.Forthecaseofinnovationsdisplayedatindustrial

exhi-bitions,theireconomicandtechnologicalsigniﬁcancewilldepend

ontheexactcriteriathatanartefactmustsatisfyforbeingincluded

intheexhibition.Second,andmostimportantly,bothtypesofdata

typicallycompriseinnovationswithandwithoutpatents(Moser,

2012).Inotherwords,usingthistypeofdataallowsanassessment

oftheshareofinnovationsoccurringoutsideofthepatentsystem

aswellastheconstructionofamore‘direct’indicatorofpropensity

topatent.Thisisacrucialadvantagewithrespecttoothersources

ofdata.

3. The‘R&D100Award’database

This paper presents an extension of the method recently

employedbyhistorianstotheﬁeldofETIS.Usingasourceofdata

thatsofarhasreceivedlittleattentionweprovidenewestimates

ofpatentpropensityacrossindustriesandovertime.Oursourceof

dataisthe‘R&D100Award’competitionorganizedbythemagazine

ResearchandDevelopment(previouslycalledIndustrialResearch).

Themagazinewasfoundedin1959anditrepresentsprobablyone

ofthemostauthoritativeregularpublicationsforR&Dpractitioners.

Currentlyithasanestimatedmonthlyreadershipofover80,000.

Itisestimatedthatabout75%ofthereadersworksinhigh-tech

industries,whereastheremaining25%worksforgovernment

lab-oratories,universities,andsimilarorganizations.Over60%ofthe

readershavemanagerialorexecutivetypeofjobs.The‘R&D100

Award’competitionhasbeenrunningsince1963.Eachyearthe

magazineawardswithaprizethe100mosttechnologically

signiﬁ-cantproductsavailableforsaleorlicensingintheyearprecedingthe

judgement.Throughouttheyears,keybreakthroughsinventions

suchasPolacolorﬁlm(1963),theﬂashcube(1965),theautomated

tellermachine(1973),thehalogenlamp(1974),thefaxmachine

(1975),theliquid crystaldisplay(1980),theprinter(1986),the

KodakPhotoCD(1991),theNicodermantismokingpatch(1992),

Taxolanticancerdrug(1993),labonachip(1996),andHDTV(1998)

havereceivedtheprize.Inordertoapplyfortheprizeinventors,or

theiremployees,mustﬁllanapplicationformprovidingadetailed

descriptionoftheproductinquestion.Theprizeconsistsofaplaque

whichispresentedinaspecialceremony.Thereisnosumofmoney

involved.Theprizeisawardedbyajurycomposedofuniversity

professors,industrialresearchersandconsultantswithacertiﬁed

levelofcompetenceinthespeciﬁcareastheyarecalledtoassess.

Themembersofthejuryareselectedbytheeditorofthe

mag-azine.Themaincriteriaforassessmentaretwo:(i)technological

signiﬁcance(i.e.,whethertheproductcanbeconsideredamajor

breakthroughfromatechnicalpointofview);(ii)competitive

sig-niﬁcance(i.e.,howtheperformanceoftheproductcomparesto

rivalsolutionsavailableonthemarket).R&D100awardsare

acco-ladescomparabletotheOscarsforthemotionpictureindustryas

“theycarry considerableprestigewithinthecommunityofR&D

professionals”(BlockandKeller,2009,p.464).

Thetechnologicalsigniﬁcancerequirementistobeunderstood

infairlybroadterms:9

“[...]productsandprocessesthatcanchangepeople’slivesfor

thebetter,improvethestandardoflivingforlargenumbersof

9_All_the_following_quotes_concerning_the_rules_and_organization_of_the_R&D₁₀₀ competitionhavebeenretrievedfromthemagazinewebsite,www.rdmag.com, accessed23July,2010andon7April,2012.

people,savelives,promotegoodhealth,cleanupthe

environ-ment,etc.[...]AcureforcancerorAIDS.Anenginethatruns

onwater.Asafe,cheapmethodforcleaninguptoxicwaste.A

vehiclethat canﬂy800passengersfromNewYorktoTokyo

intwo hours.Adevicethat would cutautomotiveaccidents

oronethatwouldreduceworkplaceinjuries.Apollution-free

herbicidethatwouldincreasecropproductioninThirdWorld

countries”.

Accordinglyproductswithawidepotentialofapplicationare

preferredtothosecateringtoveryspeciﬁcsetsofuserneeds:

“Productsorprocessesthatsolveveryspecializedor

circum-scribedproblemscouldbejudgedless signiﬁcantthanthose

thatmeetlarger,morebroad-basedneeds.Forexample,anew

scientiﬁcinstrumentthatonlybeneﬁtsafewscientistsina

nar-rowﬁeldofinterestwouldhavedifﬁcultycompetingagainst

adevicewithmuchbroaderapplication.It woulddependon

howsigniﬁcantthetwoﬁeldsofinterestwereandhowmuch

thetechnicalimprovementscontributedtothesuccessofeach

device.”

Furthermore,forattainingtheprizethereshouldbeaproven

linkbetweentheeffectoftheinnovationandanimprovementin

technology:

“[...] these improvements must be attributedto signiﬁcant

breakthroughsintechnology.Ingeneral,thismeansyour

prod-uctshouldexhibitmultiplelevelsofimprovement-53times

faster,103greaterthroughput,503timesmoreaccurate-or,

preferably, orders of magnitude improvement over existing

technology.Again,we’relookingfor‘leapfrog’gainsin

perfor-mance,notexpected,incrementalimprovements.”

Additionallytheproductshouldalsorepresentamajor

improve-ment in comparison with alternative solutions already existing

on the market. For this reason, the applicant is requested

to provide a ‘competitive matrix’ illustrating how the

prod-uct compares with rival solutions already available on the

market:

“Thecompetitivematrixshouldshowhowyourproduct

com-parestoexistingproductsintermsofthecrucialfactorsinvolved

inthetechnology.Thisisyouropportunitytogivethejudges

aquickoverviewofhowyourproductbeatsthecompetition.

[...]Includeonlyfactorscrucialtothetechnology.Don’twaste

space(andthejudges’time)throwingineveryconceivable

fac-tor,justtopadyourentry.However,youmustlistallfactors

that are indeedcrucial to thetechnology, even ifyou don’t

‘win’thatparticularpoint.Forexample,ifyoufailtoinclude

‘hardness’inanentryinvolvinganewalloy,yourentrymaybe

lookeduponwithsuspicionbythejudges.Sometypicalfactors

youmightwanttoinclude:signal-to-noiseratio,weight,speed,

reliability, resolution, cost, accuracy, life expectancy, mean

time between failures, sensitivity, reproducibility, strength,

powerconsumption,productionyield,environmental

operat-ing,intensity,efﬁciency,size,outputrate,bandwidth,number

ofmaterialstested,stability”.

Theproductmustexistinmarketableform,i.e.it“musthave

beenﬁrstavailableforsaleorlicensingduringthecalendaryear

precedingthejudging”.Applicantsarenotrestrictedtoﬁrms,but

alsogovernmentallaboratories,universities,publicresearch

cen-tresareallowedtocompete.Incaseofproductsresultingfrom

researchcollaborations,theapplicationformrequirestoinclude

alltheorganizationsthathaveprovideda“signiﬁcantcontribution

to[the]creationoftheproduct”andtoprovideadescriptionoftheir

preciseroleintheproject.Hence,therulesofthecompetitionmake

(6)

properlyacknowledged.10_Finally_an_organization_may_submit_as

manyproductsasitwishesateachyearlycompetition.

ThereareanumberofcharacteristicsoftheR&D100awards

competitionthat,atleastprimafacie,appearparticularly

promis-ingforusingthisdatasourcetomeasureinnovativeoutput.First,

theR&D100awardscompetitionseemstorepresentagood

oppor-tunityforcompanies,governmentlaboratories,etc.toshowcase

theoutcomeoftheirinnovativeactivities.Thus,wecanexpectthat

theawardswillprovideuswithafairlyreliablesampleof

inno-vationsattainedbyR&Dperformers.Second,R&D100awardsare

grantedtoinnovationsthat,atleastinprinciple,shouldembody

asigniﬁcantimprovementovertheexistingstate-of-the-artthat

isclearlydocumented.Inotherwordsawardedinnovationsshould

represent,atleastinprinciple,atechnologicalbreakthrough.Third,

theselectionoftheawardsismadebywhatappearsacompetent,

authoritativejuryofexperts.Fourth,R&Dawardsmaybeassigned

both to patented and not-patented innovations. Finally, there

seemstobelimitedspacefor strategicbehaviourand attempts

toconditioningthejury,becausethenatureoftheprizeis

sim-plyhonoriﬁc.Alongsidetheseadvantages,somebiasesexistwhich

preventusfromconsideringtheseawardedinnovationsasafully

representativesampleofinnovativeactivities.Forexample,Scherer

wasstruckbythefactthatawardscoveredaverylimitednumberof

newweaponsystemsandarelativelyfewpharmaceuticalproducts,

bothsectorsnotoriouslycharacterizedbyhighR&Dinvestments.

Nevertheless,hestillregardedthesourceascapableofproviding

usefulinsightsonthenatureofinnovativeprocesses(Scherer,1999,

pp.67–68).

Retrievingtheinformationfromdifferentissuesofthe

maga-zine,wehaveconstructeda datasetofalltheR&D100awards

grantedintheperiod1977–2004.Ourdatasetcontains2802

inven-tions.Thetotalisnotequalto2800,becausetherequirementof

awarding100inventionswasapparentlyinterpretedwithsome

degreeofﬂexibility.Thus,theamountofawardsgivenineachyear

intheperiodweareconsideringrangesfrom97to109.Amajor

lim-itationofthedatasetisthatwedonothaveinformationonapplying

innovationsthatwerenotawardedtheprize.Forthisreason,we

cannotcontrolwhetherspeciﬁcfactors,besidesthespeciﬁc

tech-nicalandeconomicmeritsoftheinnovation,affectedtheselection

oftheawards.

3.1. Matchingawardswithpatentdata

In orderto assessthe propensitytopatent for theawarded

innovationsincludedinoursample,wehadtolookfora

possi-ble matchbetweeneach awarded innovationsand oneormore

USPTOpatents.11_We_do_not_expect_to_ﬁnd_an_exact_match_between

eachawardedinnovationandonepatent.Asnotedabove,awarded

innovationsrepresents‘products’availableforcommercialization

orlicense,soitispossiblethat,incertainareas,individual

compo-nentsofaspeciﬁcproductmaybeprotectedbydifferentpatents.

Weregardedtheawardedinnovationas‘patented’alsoincasesin

whichoneormorecomponentsoftheinnovationinquestionwere

actuallypatented.Tocarry outthematchingexercisewerelied

uponthefollowinginformationcontainedintheR&D100database:

(i)nameofawardwinningorganization(s),(ii)thenameof the

innovation,(iii)theyearofaward,(iv)thenameofdeveloper(s)

and(v)thedescriptionoftheinnovation.WehavesearchedUSPTO

10 _In_this_respect_it_is_also_important_to_note_that_the_rules_of_the_competition_state explicitlythat:“existingtechnologiespurchasedbythirdpartieswhothenconduct sales,[marketingandothercommercialization]efforts”areconsideredeligiblefor theawardonlyiftheoriginaldeveloperisincludedintheapplication.

11 _Given_the_nature_of_our_data_source,_the_most_obvious_choice_was_to_match_R&D 100innovationswithpatentstakeninUS.

patentsgrantedinatimeintervalrangingfrom3yearsbeforeto

3years aftertheaward.Thecriteriaforascertaininga ‘positive’

matchwerethenameoftheinventors, thenameofthe

organi-zationandtheconsistencybetweenthedescriptionofthe‘R&D

100’innovationandthetitleandabstractofthepatent(alsotaking

intoaccountthepossibilitythatoneormorecomponentsofthe

awardedinnovationcouldhavebeenpatentedasaseparateitem).

Inparticular,thepatentsearchprocedureentailedthefollowing

steps.First,thenameofdeveloperasInventorandthenameof

organizationasAssigneewereusedtosearchpatentsintheUSPTO

onlinedatabase.Ifanypatentswerefound,thepatenttitle and

abstractwerecheckedbylookingattheinformationprovidedby

theR&D100toseeifthepatentwascorrespondingtotheaward

winninginvention.Second,ifthenameofdeveloperwasnot

avail-able,abstractkeywordsearchwiththenameoforganizationwas

carriedout.Thekeywordswereselectedfromthetechnological

informationoftheinnovationscontainedintheR&D100list.Ifa

matchwasfoundatthisstage,afurthercheckwascarriedoutto

seewhetherthepatentwasrelatedwiththeawardwinning

inno-vationbycross-checkingtheinformationofthepatentandtheR&D

100innovation.

Weshouldnotethatthematchingprocessmaybesubjectedto

errors.Morespeciﬁcally,theremaybetwolimitationsinthepatent

searchingprocedureswehaveadopted.Theﬁrstisthetimespanof

thesearching.Thesearchconsidersasrelevanttotheinnovationa

patentobtainedinplus/minus3yearsfromtheyearoftheaward.

Itmeansthatthisprocedurecanoverlooktherelevantpatent(s)

thatweregrantedmorethan3yearsbeforeoraftertheyearofthe

award.Second,theproductnameanddescriptioncontainedinthe

R&D100databasemaynotalwaysprovideenoughinformationfor

theidentiﬁcationofoneormorepossibleunderlyingpatents.These

limitationsnotwithstanding,weareconﬁdentthatourmatching

procedureprovidedreliableresultsin mostofthecases.

Never-theless, inuncertaincases, the‘beneﬁtof thedoubt’wasgiven

toapositivematchinthesensethatweconsideredtheawarded

inventionascovered byapatent.12 _For _this_reason,_if_anything,

theadoptedmatchingproceduredoesnotcontainanyin-builtbias

leadingtoasystematicunderestimationofpatentpropensity.

3.2. The‘quality’ofR&D100awards

AshighlightedbyMoser(2012),oneof thechiefadvantages

of employing data onawardsand prizes asindicators of

inno-vative outputisthat, withrespecttopatents, this typeof data

shouldinprinciplecontainonlyrelatively‘important’innovations,

namelythoseinventionsdeemedworthyofreceivingtheprizeor

ofbeingputondisplay.Accordingly,theﬁrstexercisewecarried

outwasanattemptofcheckingwhetherourR&D100dataset

con-tainsinventionsthatareaboveacertainqualitythreshold.Thisis

donebyreplicatinganexerciseoriginallyperformedbyCarpenter

etal.(1981).Carpenteretal.(1981)usedthe1969and1970R&D

awardslistandmatchedtheseinventionswiththecorresponding

USpatents.Inthisway,theyobtainedasetof100patentswhose

technologicalsigniﬁcancehadbeen‘certiﬁed’bythegrantingofthe

award.Theythencomparedthecitationsreceivedbythisgroupof

patentswiththecitationsreceivedbyarandomsampleofpatents

distributedwithinthesametimecohort.Theirresultsshowedthat

thepatentscoveringtheR&D100awardsreceivedasigniﬁcantly

highernumberofcitationsthanthecontrolgroup.Thisobviously

suggeststhatR&D100innovationsareonaverageofbetterquality

thanthe‘average’patent.

12_In_this_respect_our_matching_procedure_was_also_robust_to_changes_in_the_‘time range’beforeandaftertheawardwasreceived.

(7)

Table1

Patentcitationsreceivedbyawardedinnovationsandbyarandomsampleofpatents(matchedbygrantedyearandtechnologyclass).

Number Mean Median Standarddeviation Min Max R&D100patents 535 12.88037 7 16.17822 0 137

Randomsample 5331 8.483024 4 14.11133 0 329

Note:Mann–WhitneytestrejectstheNullHypothesisofequalpopulations.

Table2

Totalawardedinnovationsandpatents.

Awardedinnovations Patentedinnovations Sharenotpatented(%)

Allthesample(1977–2004) 2802 255 90.9

Non-corporate 886 25 97.16

Corporateonly 1751 220 87.44

Ourresultsfortheperiod1977–2004conﬁrmtheearlyﬁndings ofCarpenteretal.(1981).ForeachR&D100innovationwithoneor

moreUSPTOpatentsweconstructeda‘matchedrandom’sampleof

tengrantedpatentsofthesamegrantedyearandofthesame

Inter-nationalPatentClassiﬁcation(IPC)class andthencomparedthe

numberofcitationsreceivedbypatentsinthisrandomsamplewith

thecitationsreceivedbythepatentscoveringaninventionwithin

theR&D100award.13_The_results_of_this_exercise_are_reported_in

Table1.14

The non-parametric Mann–Whitney test conﬁrms that the

mediannumberofcitationsofpatentsassociatedwithaR&D100

inventionissigniﬁcantlynotlowerthanthemedianoftherandom

matchedsample.Overall,thisexerciseconﬁrmsthatthe

innova-tionswhichreceivedaR&D100awardaremoresigniﬁcantfroma

technologicaloreconomicviewpointthanthe‘average’patentin

theirtechnologicalclass.

4. R&D100awardsandpatentpropensity

Thissectionpresentsourestimatesofpatentpropensitydeﬁned

as the share of patented innovations in the total number of

inventionsthat have receivedanR&D100award. Wecompute

ourestimatesacrossdifferentdimensionsofinnovativeactivities

(industry,geographicalarea,organizationandtypeofinvention).

4.1. Mostawardedinnovationsarenotpatented

Table 2reports thenumberof awardedinnovationsandthe

percentageofnotpatentedones.

As highlightedabove, theawardedinnovations containedin

theR&D100listrefertoproductsthatareavailableonthemarket

or for licence when the application is submitted. Hence, it is

possiblethat thedata willcontain a bias againstorganizations

suchasuniversitiesand PROsthat lack‘downstream’assets for

thecommercializationofaproduct.However,itisinterestingto

notethatresultsinTable2areconsistentwiththeresultsofBlock

and Keller (2009) showingthat a signiﬁcant share of awarded

innovations (more than 30%) are generated by non-corporate

typeoforganizations.Soitwouldseemthat,thepresenceofbias

notwithstanding, our data cover also a signiﬁcant segment of

the populationof non-corporate organization involved in R&D

13_Our_control_group_is_made_of_granted_patents_and_not_of_patent_{applications.}_At theUSPTOasizeablepercentageofpatentapplicationsarenotgranted.Inaddition tothis,manypatentapplicationsarediscontinued.Forthesereasonsitwouldhave beenmorerigoroustocomparethesampleofpatentedawardedinnovationswith patentapplications.However,thesedataarenotfullyavailableforUS,thustheneed torelyupongrantedpatents.

14_The_random_matched_sample_includes₅₃₃₁_patents_and_not₅₃₅₀_because_for somespeciﬁcyearsinsometechnologyclassesitwasnotpossibletocollectenough patentstocreatethematch.

activities.15 _Overall, _we _found _that ₂₆₉ _awarded _innovations

(slightlylessthan10%)werepatentedaccordingtoourmatching

criteria suggesting that the great majority of innovations were

notpatented.Thispercentageisslightlyhigher(12.56%)whenwe

consideronlyinnovationsthathavebeenmadebyﬁrms.

Thisestimatedpatentpropensityisinlinewiththeﬁndingsof

Moserwhoreportstotalpatentingratesbetween11%and14%for

theinventionsdisplayedattheCrystalPalaceexhibitionof1851

(Moser,2005,p.1221).Ofcourse,thisﬁndingshouldbeinterpreted

keepinginmindtheinherentlimitationsthedatasetdiscussedin

Section3.However,evenifweconsiderpossibleerrorsthatmay

haveledustounderestimatepatentingrates,theresultthatsucha

sizeableshareofmajorinnovationsisnotpatentedisremarkable.In

particular,ifweconsiderthat‘TheR&D100Award’isacompetition

aimedatacknowledgingtheoutputofformalizedR&Defforts,which

isnotoriouslyoneofthecontextswiththehighestpropensityto

patent and that, we are in principle dealing withbreakthrough

innovations,ourﬁndingsrevealthatpatentprotection,eveninthis

context,isactuallyamuchlessusedappropriabilitystrategythan

itis generallybelieved.Inthisrespect,ourﬁndings areactually

apowerfulcorroborationof theﬁndingsofMoser(2005).

Addi-tionally,theyarenotinconsistentwiththeresultsofboththeYale

(Levinetal.,1987)andtheCarnegieMellonsurvey(Cohenetal.,

2000)indicatingthatonlyinaveryrestrictednumberofcontexts

patentsareconsideredaseffectivetoolsforprotectinginnovation.

Theobvious policyimplicationisthat therecentdevelopments

towardsthestrengtheningofIPRregimesmayactuallyrepresent

astepgoinginthewrongdirection,asitwouldappear

consider-ingthepredominantshareofinnovativeactivitieswhichisactually

carriedoutwithoutresortingtopatentprotection.16

Fig.1displaystheevolutionovertimeofthepropensitytopatent

foralloursampleofinnovations.

Ourestimatedpropensitytopatentisneverhigherthan20%.

Moreover,contrarytowhathasbeensuggestedbyotherstudies

(KortumandLerner,1999),ourevidencedoesnotseemto

indi-catetheexistenceofsigniﬁcantstructuralbreaksinthetimeperiod

15_The_requirement_of_availability_for_sale_or_licensing_implies_that_the_awarded productmustbeavailableeitherforpurchaseorlicensingduringtheyear preced-ingtheawardandnotnecessarilyalreadylaunchedinthemarketonanextensive scale.Clearly,thisallowsalsoorganizationswithlimitedcapabilitiesinthe commer-cializationandinthe“downstream”developmentofnewproductstocompete,as alsoprototypesreadytobelicensedareeligiblefortheaward.Somenotable exam-plesofawardedproductsthatweredevelopedonlybyacademicinstitutionsare: the“CostThinFilmSolarCell”(UniversityofDelaware,1979);the“Low-Palmitic SoybeanOil”(IowaStateUniversity,1991);the“SC-54OralVaccine”(IowaState University,1996);the“Nanoruler”(M.I.T.,2004);the“Netsolve1.2”(Universityof Tennesse&U.C.S.D,1999)andthe“Chromium”(UniversityofVirginia&Stanford University,2004)software.

(8)

All industries and applicants 0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 2004 2001 1998 1995 1992 1989 1986 1983 1980 1977 % Inventions Patented

Fig.1. Patternofchangeinthepropensitytopatent.Allthesamples.

0% 5% 10% 15% 20% 25% 30% 2004 2001 1998 1995 1992 1989 1986 1983 1980 1977

All industries by applicant type

% Only corporate At least 1 academic At least 1 governamental

Fig.2.Patternofchangeinthepropensitytopatentbytypeofinventor.

considered.17_In _our_case,_it _is_suggested_that _the_propensity_to

patenthasbeenremarkablystableandpossiblycharacterizedby

aﬂuctuatingbehaviouraroundwhatseemstobeaconstantlevel

of10%.

Fig.2showstheevolutionofthepropensitytopatentbroken

downbytypeofinventors,distinguishingbetweencorporate(i.e.

ﬁrms)andnon-corporate(i.e.PROanduniversities)organizations.

Asexpected Fig.2 suggeststhat thepropensity topatent is

higherforﬁrmsthanforPROsanduniversities.Inseveralyears

thepropensitytopatentofPROsanduniversitiesisequaltozero

indicatingthatnoinnovationhasbeenpatented.Fromearly1990s

onwardsthepropensitytopatentofuniversitiesseemstoincrease.

Itwouldbetemptingtointerpretthisevidenceasaconsequence

of the strengthening of IPRs following the introduction of the

Bayh–DoleActinUS(MoweryandZiedonis,2002;Moweryetal.,

2001)thoughananalysisofthemechanismsunderlyingthistrend

fallsbeyondthescopeofthepresentpaper.18

17 _Hall₍₂₀₀₅₎_ﬁnds_several_structural_breaks_in_patent_application_series_at_USPTO

duringthe1967–1997timeperiod.Inourcase,boththePhillips–Perron(PP)andthe AugmentedDickey–Fuller(ADF)testsforunitrootrejectthenullhypothesisthat ourpatentpropensityseriescontainsaunitrootatthe1%signiﬁcancelevel,both afterincludingatimetrendandlaggedterms(uptothe4thorder)intheassociated regression.TheAndrews(1983)testforstructuralbreaksatunknownpoints(sup. F=8.6538,T=28)doesnotrejectthenullhypothesisof‘nostructuralchange’atthe 5%signiﬁcancelevel.

18 _A_preliminary_analysis_along_these_lines_actually_suggests_that_both_the_number ofpatentsandthenumberofawardedinventionsattributabletoUniversityand PROsinoursamplehaveincreased.

4.2. Patentingratesvaryacrosssectors

Awardedinnovationsareclassiﬁedbythemagazineinseveral

categoriesonthebasisoftheirtechnologicalcontent.The

classiﬁca-tionisnotconsistentovertimeandinsomecasestheinnovations

werenoteven assignedtoaspeciﬁccategory.Thus, inorderto

examinethedistributionofawardedinnovationsacrossdifferent

technologicalﬁelds,wehaveproceededasfollows.Firstwe

reclas-siﬁedeachawardedinnovationaccordingtoatechnology-oriented

classiﬁcationof30differentsectorsbasedontheco-occurrenceof

theIPCcodesproposedbytheObservatoiredesSciencesetdes

Tech-niques(OST).19_In_a_few_doubtful_cases,_we_have_relied_both_upon

theclassiﬁcationinproductcategoriesoftheR&D100awardsand

ontheinnovationdescription.Itisimportanttonotethatwehave

assignedeachawardedinnovationtoonlyoneofthe30OSTsectors.

Thesesectorshavebeenfurtheraggregatedinto5‘macro’

techno-logicalclasses(called‘OST5’henceforth)deﬁnedaccordingtothe

ISI-INIPI-OSTpatentclassiﬁcationbasedontheEPOIPC

technolog-icalclasses.

Table3displaysthesharesofawardedinnovationsthathave

beenpatented(patentingrates)classiﬁedbyboth5and30 OST

sectors.

The share of patented innovation variesconsiderably across

sectors. In terms of macro-sectors,the sector withthe highest

propensitytopatentischemical/pharmaceuticals,aresultwhichis

alsoinlinewiththeresultsoftheYaleandCarnegieMellonsurveys

ontheeffectivenessofpatentsforprotectinginnovationsinthese

ﬁelds.Finally,themacro-sectorwiththelowestpatentingrateis

instruments.20_In_this_case_we_should_remember_that_many

organi-zationsactiveinthissectorarenon-corporateinstitutionssuchas

universitiesandpublicresearchcentreswhichtraditionallydisplay

averylowpatentingattitude.

4.3. Patentingratesvaryacrossgeographicalareas

Table 4comparespatent propensities acrossdifferentworld

regions.

TheawardedinnovationswithatleastoneapplicantfromUS

showalowerpatentpropensitywithrespecttotheaveragelevel

ofthewholesample(−1.38%,statisticallysigniﬁcantat10%

signiﬁ-cancelevel),whereasapplicantsfromAsiatendtopatentmoretheir

innovationsthantheaveragelevel(+15.53%,statisticallysigniﬁcant

at1%signiﬁcancelevel).WeshouldnotethatthelargebulkofAsian

innovationsareawardedtoJapanesecompanies.Henceourresults

pointingtoasigniﬁcantlyhigherpropensitytopatentofAsian(and

especially Japanese)ﬁrmsareconsistentwithpreviousresearch

ontheaggressivepatternsofforeignpatentingofJapaneseﬁrms

inacomparativeinternationalperspective(Granstrand,1999,pp.

134–175).

It maybearguedthatthisﬁndingistheconsequenceof

dif-ferencesintheinstitutionalmixofUSandAsianprizewinners.In

ordertounderstandwhetherthisisthecasewithourdatawehave

comparedthepropensitiesacrosssubsetsofsimilarorganizations.

Table5conﬁrmsthattheseoveralldifferencesinpatentpropensity

ratesacrosscountriesarenotdrivenbydifferentinstitutionalmix

of US prize-winners with respect to other geographical areas.

InfactthehigherpropensitytopatentofAsianprize-winnersis

19_See_Hinze_et_al.₍₁₉₉₇₎_.

20_A_statistical_test_for_binary_variables_was_carried_out_to_check_whether_the dif-ferenceinproportionbetweeneachindustryandthetotalsampleisstatistically significant.Resultsrejectedthenullhypothesisofequalityintwocases:Chemical andPharmaceuticals(+4.6%withrespecttotheaveragepatentpropensityofthe wholesample,statisticallysignificantat5%significancelevel);Instruments(−2.8% withrespecttotheaveragepatentpropensityofthewholesample,statistically significantat1%significancelevel).

(9)

Table3

Patentingratesof‘R&D100’innovations.

OST5 OST30 Allapplicants Onlycorporate

No.ofinnovations Sharepatented No.ofinnovations Sharepatented Electrical

engineering

1 Electricalengineering&devices 274 0.1350 177 0.1751 2 Audiovisualtechnology 19 0.1053 12 0.1667 3 Telecommunications 32 0.1563 20 0.25 4 InformationTechnologies 255 0.0824 157 0.1210 5 Semiconductors 148 0.1149 90 0.1333 728 0.1126 728 0.1513 Instruments 6 Optics 198 0.1111 123 0.1545 7 Controltechnology 629 0.0493 384 0.0729 8 Medicaltechnology 125 0.1120 91 0.1209 27 Nuclearengineering 75 0.0400 41 0.0732 1027 0.0682 639 0.0954 Chemistry, Pharma 9 Organicchemistry 0 – 0 – 10 Polymers 47 0.1489 41 0.1463 11 Pharmaceutics 0 – 0 – 12 Biotechnology 87 0.0690 51 0.0588 14 Foodchemistry 0 – 0 –

15 Basicmaterialschemistry 42 0.2857 31 0.3870

176 0.1420 123 0.1703 Process engineering 13 Materialsmetallurgy 240 0.1458 167 0.2036 16 Chemicalengineering 220 0.1091 118 0.1525 17 Surfacetechnology 0 – 0 – 18 Materialsprocessing 8 0 0 – 20 Environmentaltechnology 154 0.0714 81 0.1235

24 Handling&printing 0 – 0 –

25 Foodprocessing 0 – 0 – 622 0.1125 366 0.1694 Mechanical engineering 19 Thermalprocesses 34 0.0882 23 0.1304 21 Machinetools 77 0.1169 47 0.1702 22 Engines 0 – 0 – 23 Mechanicalelements 43 0.0465 30 0.0333 26 Transport 27 0.0370 0 – 28 Spacetechnology 9 0.1111 0 – 29 Consumergoods 59 0.1017 44 0.1364 30 Civilengineering 0 – 0 – 249 0.0884 144 0.125 Total 2802 0.0960 1728 0.1336

conﬁrmedevenwhen consideringawardedinnovationswithat least one corporate applicant (columns 3–4) or only corporate applicants(columns5–6).

4.4. Multivariateregressionanalysis

Though interesting,thepreviousresultsonlyaccountforthe effectofasinglecharacteristicatatime(i.e.technologicalsector,

countryoforigin,applicanttype)onthepropensitytopatent.In ordertostudythejointeffectofallthesevariablesinanuniﬁed frameworkweperformaprobitmultivariateregressionanalysis usingtheprobabilitytopatent agiven innovationPr(PAT=1)as dependentvariableandasetofapplicantandinnovationspeciﬁc characteristicsasindependentandcontrolvariablesrespectively. Thelistof additionalvariables includes:dummiesfor the tech-nological sector (both atOST 5 and OST 30 aggregation level),

Table4

Patentingratesbyindustryacrosscountries.

Sector(OST5) FullSample USA Europe Asia Other

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) Total %Pat Total %Pat Diff.(4)–(2) Total %Pat Diff.(7)–(2) Total %Pat Diff.(10)–(2) Total %Pat Diff.(13)–(2) Elec.Eng. 728 11.2 597 8.21 −2.99%* ₁₆ ₀ _−11.2% ₈₆ _31.40 _+20.2%*** ₂₉ _3.45 _−7.75% Instruments 1027 6.8 876 5.58 −1.22% 56 8.92 +2.12% 50 22 +15.2%*** ₄₅ _4.44 _−2.36% Chemistry 176 14.2 158 13.92 −0.28% 8 0 −14.2% 5 20 +5.8% 5 0 −14.2% Proc.Eng. 622 11.2 559 11.09 −0.11% 17 11.76 +0.56% 25 12 +0.8% 21 0 −11.2% Mech.Eng. 249 8.8 207 7.73 −1.07%* ₉ _11.11 _+2.31% ₂₁ _23.81 _+15.01%** ₁₂ ₀ _−8.8% All 2802 9.6 2397 8.22 −1.38%* ₁₀₆ _7.55 _−2.05% ₁₈₇ _25.13 _+15.5%*** ₁₁₂ _2.68 _−6.92%**

Innovationswithmultipleapplicantsfromdifferentindustriesaredoublecountedinthetable. *_Difference_is_{statistically}_signiﬁcant_at_10%_{signiﬁcance}_level.

**_Difference_is_{statistically}_significant_at_5%_{significance}_level. ***_Difference_is_{statistically}_significant_at_1%_{significance}_level.

(10)

Table5

Patentingratesbygeographicalareaandorganizationtype.

Typeofapplicant Fullsample USA Europe Asia Other

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) Total %Pat Total %Pat Diff.(4)–(2) Total %Pat Diff.(7)–(2) Total %Pat Diff.(10)–(2) Total %Pat Diff.(13)–(2) Atleast1corporate 1916 12 1557 11.11 −0.89% 102 7.84 −4.16% 183 25.68 +13.68%*** ₇₄ _2.70 _−9.3%** Onlycorporate 1728 13.36 1389 11.74 −1.62% 94 7.45 −5.91%* ₁₈₂ _25.82 _+12.46%*** ₆₃ _3.28 _−10.08%**

ALL 2802 9.6 2397 8.22 −1.38%* ₁₀₆ _7.55 _−2.05% ₁₈₇ _25.13 _+15.5%*** ₁₁₂ _2.68 _−6.92%**

*_Difference_is_{statistically}_significant_at_10%_{significance}_level. **_Difference_is_{statistically}_significant_at_5%_{significance}_level. ***_Difference_is_{statistically}_significant_at_1%_{significance}_level.

Table6

Variabledescription.

Description Type

Dependentvariable

PAT =1iftheinnovationhasbeenpatented(seeSection3.1) Dummy Independentvariables

MAPPL =1formultipleapplicantorganizations,=0otherwise Dummy

NINV =numberofinventors Count

USA =1ifatleastoneapplicantisaU.S.organization,=0otherwise Dummy

EUROPE =1ifalltheapplicantsarefromcontinentalEurope,=0otherwise Dummy

ASIA =1ifalltheapplicantsarefromAsia,=0otherwise Dummy

PRO =1ifatleastoneapplicantisapublicresearchorganization,=0otherwise Dummy

Controls

dum19861995 =1theinnovationhasbeenawardedinthe1986–1995decade,=0otherwise Dummy

dum19962005 =1theinnovationhasbeenawardedinthe1996–2005decade,=0otherwise Dummy

ElectricalEng =1iftheinnovationbelongstotheElectricalEngineeringOST5macrosector,=0otherwise Dummy Instruments =1iftheinnovationbelongstotheInstrumentsOST5macrosector,=0otherwise Dummy ChemistryPharma =1iftheinnovationbelongstotheChemistry&PharmaOST5macrosector,=0otherwise Dummy ProcessEng =1iftheinnovationbelongstotheProcessEngineeringOST5macrosector,=0otherwise Dummy

dummies for the geographical macro areas (i.e. USA, EUROPE,

ASIA),andforthetimeperioddecades(1976–1985,1986–1995,

1996–2005)tocapturetheyearofaward.Anothersetofregressors

includes: a dummy (PRO) to account for innovations with at

least one publicresearch organization applicant (i.e. either an

academicoragovernmentalorganization),adummy(MAPPL)for

collaborativeinnovationstoindicatethepresence ofmorethan

oneapplicant,andacountvariable(NINV)thatreportsthenumber

of applicants.Allthe variables used in theregressions analysis

aresummarizedinTable6.Table7providesinsteadtheirmain

descriptivestatistics.

Resultsfromthemultivariateprobitregressionarereportedin

Table8.

Column(1)reportstheestimatedcoefﬁcientsforthemost

par-simoniousmodelwhichdoesnotincludesectorandtime-period

dummies.Column(2)includesbothtime-periodandOST5

macro-sectordummies.Columns(3–4)reporttheestimatedcoefﬁcients

andaveragemarginaleffects(AMEs)forthefullmodelwithboth

Table7

Descriptivestatistics.

Variable Obs Mean Std.Dev. Min Max PAT 2802 0.096 0.295 0 1 MAPPL 2802 0.256 0.437 0 1 NINV 2802 1.665 0.902 1 5 USA 2802 0.877 0.329 0 1 EUROPE 2802 0.038 0.191 0 1 ASIA 2802 0.067 0.250 0 1 PRO 2802 0.375 0.484 0 1 dum19861995 2802 0.356 0.479 0 1 dum19962005 2802 0.322 0.467 0 1 ElectricalEng 2802 0.260 0.439 0 1 Instruments 2802 0.366 0.482 0 1 ChemistryPharma 2802 0.063 0.243 0 1 ProcessEng 2802 0.222 0.416 0 1

time-periodandOST30macro-sectordummies(notreportedfor thesakeofclarity).

In terms of geographical and sectoral effect, ourregression analysisconfirmsthepreviousfindingsfromtheunivariate analy-sis.Ceterisparibus,awardedinnovationswithallapplicantsfrom Asia show a larger probability to be patented with respect to theexcludedcategory(whichincludesnon-US,non-Europeanand non-Asiancountries), followed byawarded innovationswithat least oneUS applicant. No significantdifferencein theaverage patentpropensityisdetectedbetweentheOST5macrosectors rep-resentedbythefourdummyvariablesincludedinthemultivariate regressionandtheexcludedone(MechanicalEngineering).

In terms of differences in the propensity across applicants and/orbetweencollaborativeandnon-collaborativeinnovations, weﬁndthatinnovationswithatleastonePROasapplicantshow alowerpatentpropensitywithrespecttotheexcludedcategory (only corporate applicants). Finally patent propensity increases with thenumber of inventors (NINV) whereas havingmultiple applicants(MAPPL)doesnotseemtoexertasigniﬁcanteffecton thepropensitytopatentanawardedinnovation.

Theformerfindingconfirmsourpreviouspointaboutthelower propensitytopatentfornon-corporateinventorsasaconsequence oftheirrelativelackofdownstreamcapabilitieswhencompared tofirms.Thelatterfindingseemstobeincontrastwithprevious studies,suchasBrouwerandKleinknecht(1999)andKleinknecht andvanderPanne(2011),whoinsteadfoundapositivecorrelation

betweenR&Dcollaborationsandpropensitytopatent.This

discrep-ancycanbeexplainedintermofdifferentresearchdesign.These

twostudiesemployedﬁrmleveldata(theCommunityInnovation

Survey)whereasouranalysisiscarriedoutattheinnovationlevel.

Moreoverthesestudiesadoptedacompletelydifferent

depend-entvariable(i.e.whethertheﬁrmappliedornotforapatentat

theEuropeanPatentOfﬁceorthenumberofﬁrm’sEPOpatents)

(11)

Table8

Estimationresultsforthepropensitytopatent.

Dependentvariable:Pr(PAT=1) (1) (2) (3) (4)

Coefficients Coefficients Coefficients Marginaleffects

USA 0.406** _0.387** _0.407** _0.045*** (0.165) (0.169) (0.169) (0.014) EUROPE 0.165 0.209 0.247 0.040 (0.216) (0.221) (0.222) (0.042) ASIA 0.971*** _0.965*** _1.006*** _0.240*** (0.196) (0.201) (0.201) (0.067) PRO −0.635*** _−0.617*** _−0.605*** _−0.077*** (0.083) (0.084) (0.085) (0.009) MAPPL −0.109 −0.087 −0.094 −0.013 (0.091) (0.091) (0.092) (0.012) NINV 0.153*** _0.144*** _0.138*** _0.019*** (0.037) (0.038) (0.038) (0.005) dum1986 1995 −0.157** _−0.128 _−0.017 (0.079) (0.082) (0.011) dum19962005 −0.144 −0.128 −0.017 (0.093) (0.099) (0.013) Electrical Eng 0.115 (0.129) Instruments −0.120 (0.128) ChemistryPharma 0.259 (0.169) ProcessEng 0.189 (0.137)

OST30SectorDummies Yes Yes

Constant −1.816*** _−1.755*** _−5.281***

(0.177) (0.206) (0.275)

Observations 2802 2802 2802 2802

Loglikelihood −811.436 −796.360 −792.076

ProbitMaximumLikelihoodestimateswithclusteredrobuststandarderrors(inparenthesis). **_{Statistically}_signiﬁcant_at_5%_level.

***_{Statistically}_signiﬁcant_at_1%_level.

pastcollaborationswithotherR&Dpartners).Resultsfromother

innovation-levelstudiessuchasMäkinen(2007)ﬁndinsteada

non-signiﬁcanteffectofcollaborationsonthepropensitytopatenta

giveninnovation.Moreover,ourresultcanbeunderstoodonthe

basisofthetwoeffectssuggestedbyPeetersandVanPottelsberghe

delaPotterie(2006)toexplainthepatentingofcollaborative

inno-vations.On the one hand there is a ‘need effect’ which refers

to“[...] a higherneed forpatent protectionresultingfromthe

mutualaccesstothepartners’knowledgebases”(PeetersandVan

Pottelsberghede la Potterie, 2006, p. 127). On the otherhand,

thereisa‘noveltyeffect’whichreferstoa“[...]potentiallymore

fundamentalandbreakthroughknowledgegeneratedbyR&D

col-laborationscomparedtoin-houseR&Dalone,whichwouldresultin

morepatents”(PeetersandVanPottelsberghedelaPotterie,2006,

p.127).Inourcasethe‘noveltyeffect’isprobablybetterproxiedby

thenumberofinventors(NINV),whichhasapositiveand

statisti-callysigniﬁcantcoefﬁcient,anddominatesthe‘needeffect’which

insteadisbetterproxiedbythedummyvariableMAPPL.

5. Reassessingpatentpropensity:areappraisalofthe empiricalevidence

Atthispoint,itisinstructivetocompareourﬁndingswithprior

estimatesofpatentpropensity.ThisisdoneinTable9.

Table9isbasedonanextensiverecognitionoftheexisting

liter-atureandsummarizestheﬁndingsofallthestudieswehavebeen

abletoidentify,thatprovidedirectorindirectempiricalestimates

oftheimportanceofpatentprotectionforproductinnovationsin

differentindustriesusingdifferenttypesofmethods.Thelastrow

ofthetablereportstheﬁndingsofthispaper.Ithastobestressed

thatthisexercisehasimportantlimitations,becausetheapproach

usedtoassesspatentpropensityisnotconsistentacrossstudies.

For instance,insomecasestheestimationof patentpropensity

wasnotthemaingoaloftheresearch(thisis,inparticular,the

caseforAcsandAudretsch,1990).Inothercases(Mansﬁeld,1986),

thesurveyaskedhowmanypatentableinnovationswereactually

patentedwhichisdifferentfromourfocusontheratiobetweenthe

numberofpatentedinnovationsandthetotalnumberof

innova-tions.Levinetal.(1987)askedrespondentstoassesstherelative

effectivenessofpatentsincomparisontootheralternative

appro-priabilitystrategies.Cohenetal.(2000)askedﬁrmstoreportthe

shareofinnovationsforwhichtheyhaveappliedforapatentwhich,

again,issomewhatdifferentfromthevariablewehaveconstructed

inthispaper.Theselimitationsnotwithstanding,webelievethatit

isusefultocompareourresultswiththemainﬁndingsconcerning

patentpropensityemergingfromtheliterature.

Itisimmediatelyinterestingtonotethat,consistentlywithwhat

wehavehighlightedintheintroduction,fourofthestudieslisted

inthetable(Thomson,2009;Moser,2012;MeisenzahlandMokyr,

2011;Nicholas,2011)arecontributionsofeconomichistoriansor

historiansoftechnology.

In terms of research strategies, we can draw a distinction

betweenstudiesusingasurveyapproach(Mansﬁeld,1986;Levin

etal.,1987;ArundelandKabla,1998;BrouwerandKleinknecht,

1999; Cohen et al., 2000) and other contributions that instead

estimatepatentpropensityusingindicatorsofoverallinnovative

output not based onpatents.Although providingvery detailed

snapshotsoninnovativeactivities,itiswellknownthatalsothe

datacollectedbymeansofinnovationsurveyssufferfromseveral

shortcomings.Themainoneisthatthistypeofdata,unavoidably,

reﬂectsthepersonaljudgementofactorsthatarerequiredto

moni-torandself-assesstheirowninnovativeactivitiesandperformance

(MairesseandMohnen,2010).

Turningourattention totheresearchstrategiesand sources

(12)

1790 R. Fontana et al. / Research Policy 42 (2013) 1780 – 1792 Table9

Estimatesofpatentpropensityindifferentstudies.

Averagepatent propensity(%) 1sthighest propensityto patent 2ndhighest propensityto patent 1stlowest propensityto patent 2ndlowest propensityto patent Indicatorofpatent propensity

Data&methods Period Geographical scope

Mansﬁeld(1986) 77 Petroleum(86%) Machinery(86%) Primarymetals (50%)

Motorvehicles (65%)

Propensitytopatent(sales weighted) Surveyofarandom sampleoflarge ﬁrms(over1 million$R&Din 1981) 1981–1983 USA

Levinetal.(1987) Drugs Organicchemicals Pulpandpaper Computers Perceivedeffectivenessof patentsasappropriation tool YaleSurvey(650 R&Dexecutives from130business lines) 1981 USA

AcsandAudretsch (1990)

Lumberand furniture

Foodandtobacco Petroleum Rubberandplastics Patents/Innovations SmallBusiness Administration InnovationDataset

1982 USA

ArundelandKabla (1998)

35.90 Pharmaceuticals (79.2%)

Chemicals(57.3%) Textiles(8.1%) Basicmetals (14.6%)

Propensitytopatent productinnovations(sales weighted) PACESurvey(604 largeﬁrms) 1993 Europe Brouwerand Kleinknecht(1999)

25.40 Rubberandplastic products(36.4%)

Pharmaceuticals, Chemicals, Petroleum(36.3%)

Basicmetals(9.9%) Class,clayand ceramics(11.8%)

Percentageofﬁrmsfor whichpatentare“very important”or“crucial”for productinnovations

CISSurvey 1992 Netherlands

Cohenetal.(2000) 49.12 Drugs(95.5%) Mineralproducts (79.2%)

Metals(2.97%) Steel(4.46%) Propensitytopatent productinnovations

CMSSurvey,1165 largeﬁrms

1991–1993 USA

Thomson(2009) 60.2 Electricity(87.5%) Agriculture: harvesting(85.7%)

Clocks(11.1%) Metalworking (40%) Shareofexhibitors displayingpatented innovations CatalogueofNew YorkGreat Exhibition 1853 USA Moser(2012) 11.10 Manufacturing machinery(29.8%)

Engines(24.6%) Miningand metallurgy(5%)

Chemicals(5.1%) Numberofpatented innovationsinthetotal numberofinnovations exhibited Catalogueof CrystalPalace GreatExhibition 1851 UK

MeisenzahlandMokyr (2011)

60.00 Textiles(81%) Ships(70%) Instruments, Construction(35%)

Mining(41%) Numberof“great inventors”with1patentor more

Analysisof biographical dictionaries

1660–1830 UK

Nicholas(2011) 44.6 Machineryand

machinetools (70.8%)

Auto(67.7%) Food(33.1%) Chemicals(34.9%) NumberofR&Dperforming ﬁrmswith1patentormore

Matchingpatents toﬁrmsinNRC surveys

1921–1938 USA

Thispaper 9.6 Chemistry,pharma (14.2%) Process engineering (11.2%) Instruments(6.8%) Mechanical engineering(8.8%) Numberofpatented innovationsinthetotal numberofinnovations awardedtheprize

R&D100 competition