a dynamic network measure of technological changerussellfunk.org/cdindex/static/funk_ms_2016.pdf ·...

MANAGEMENT SCIENCEVol. 63, No. 3, March 2017, pp. 791–817

http://pubsonline.informs.org/journal/mnsc/ ISSN 0025-1909 (print), ISSN 1526-5501 (online)

A Dynamic Network Measure of Technological ChangeRussell J. Funk,a Jason Owen-Smithb

a Strategic Management and Entrepreneurship, Carlson School of Management, University of Minnesota, Minneapolis, Minnesota 55455;b Department of Sociology, University of Michigan, Ann Arbor, Michigan 48103Contact: [email protected] (RJF); [email protected] (JO-S)

Received: December 10, 2012Revised: November 13, 2014;September 14, 2015Accepted: September 14, 2015Published Online in Articles in Advance:March 22, 2016

https://doi.org/10.1287/mnsc.2015.2366

Copyright: © 2016 INFORMS

Abstract. This article outlines a network approach to the study of technological change.We propose that new inventions reshape networks of interlinked technologies by shift-ing inventors’ attention to or away from the knowledge on which those inventions build.Using this approach, we develop novel indexes of the extent to which a new invention con-solidates or destabilizes existing technology streams. We apply these indexes in analysesof university research commercialization and find that, although federal research fundingpushes campuses to create inventions that are more destabilizing, deeper commercial tieslead them to produce technologies that consolidate the status quo. By quantifying theeffects that new technologies have on their predecessors, the indexes we propose allowpatent-based studies of innovation to capture conceptually important phenomena thatare not detectable with established measures. The measurement approach presented hereoffers empirical insights that support theoretical development in studies of innovation,entrepreneurship, technology strategy, science policy, and social network theory.

History: Accepted by Lee Fleming, entrepreneurship and innovation.Funding: Support for this research was provided by grants from the Rackham Graduate School and the

Interdisciplinary Committee on Organizational Studies at the University of Michigan, the NationalScience Foundation [Grants SES-0545634, SMA-1158711, and SMA-1262447], the Ewing MarionKauffman Foundation, and the Alfred P. Sloan Foundation.

Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2015.2366.

Keywords: networks • graphs • research and development • innovation • technological change • organizational studies • strategy • universities

Economic progress, in capitalist society, means tur-moil . . . .New products and new methods compete withthe old products and old methods not on equal termsbut at a decisive advantage that may mean death to thelatter. (Schumpeter 1942, p. 32)

1. IntroductionFoundational theories of technological change distin-guish between two types of new technologies. Thefirst “come to exist as entities that depart in somedeep sense from what went before” (Arthur 2007,p. 274; Mokyr 1990). Recombinant DNA made a radicaldeparture from existing drug discovery methods thathad relied on massive screening of potentially ther-apeutic compounds (Henderson and Cockburn 1996,Gambardella 1995). Inventions like this can transformthe fortunes of organizations and industries. A secondtype of new technology builds on and enhances its pre-decessors and therefore has the result of consolidating(not challenging) the status quo. In one classic study,Enos (1962) describes four waves of improvements inprocesses for manufacturing gasoline. Although newtechnologies of this type are sometimes dubbed “incre-mental,” they may make substantial improvementsover their predecessors and they account for the lion’sshare of economic and social welfare returns from

technological progress (Enos 1958, Rosenberg 1982,David 1990).

Despite the substantive and theoretical importanceof differentiating between these two types of technolo-gies, no quantitative measure exists to capture thisdistinction. Systematic explanations for why, when,and how particular inventions have differential effectson their environment remain elusive. Assessments oftechnological dynamics traditionally relied on detailedcase studies of particular technologies (Dosi 1982,Bresnahan and Greenstein 1996, Goldfarb 2005, Arthur2009). Quantitative studies exploded with the intro-duction of the National Bureau of Economic Research(NBER) U.S. Patent Citations Data File in the late1990s (Hall et al. 2002) and led to many productiveinsights. However, the patent literature’s focus on cita-tion counts and forward-citation-based measures ofimpact has become a limitation, because many dimen-sions of technological change cannot be seen by observ-ing the simple magnitude of an invention’s later use.Attending to the impact of a new invention withoutalso considering how it relates to already extant tech-nologies creates bias and ambiguity in analyses.

The inability to systematically measure fundamen-tal concepts has made theoretical development chal-lenging in many areas. Consider work on university

791

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.

http://pubsonline.informs.org/journal/mnsc/

mailto:[email protected]

mailto:[email protected]

https://doi.org/10.1287/mnsc.2015.2366

Funk and Owen-Smith: A Dynamic Network Measure of Technological Change

792 Management Science, 2017, vol. 63, no. 3, pp. 791–817, © 2016 INFORMS

research commercialization. Since the 1980 passageof the Bayh–Dole Act (Public Law 96-517, 94 Stat.3015), a law that allowed organizations to seek patentsand issue licenses on inventions made with federalfunding, scholarly debate has raged over the con-sequences of commercial activity for university sci-ence. Although proponents emphasize universities’ability to make transformative discoveries, detractorsworry that commercial engagement pushes campusesto focus on narrow pursuits that are attractive to indus-try. Empirical research has been unable to adjudi-cate between these positions. Part of the challenge isthat forward-citation-based measures of universities’patent impact are unable to determine whether corpo-rate engagement results in greater alignment betweenindustrial interests and academic priorities. Indeed,studies find that both industrial and more traditionalacademic funding sources yield higher-impact patents(e.g., Owen-Smith and Powell 2003). Such findings helpto establish that public and corporate partners bothselect high-quality projects, but they do little to adju-dicate the key substantive question for studies of aca-demic commercialization.

Computational advances are creating opportuni-ties for extracting deeper insights from large-scaledatabases like the U.S. Patent Citations Data File. In thearea of technological change, a quantitative measurethat leverages these advances to distinguish betweentechnologies that depart from or reinforce establishedtrajectories would allow empirical studies to bettermatch foundational theories and may facilitate newconceptual development. To create such a measure,we begin with a network conception of technolog-ical change. New discoveries are additions to pre-existing networks of complementary or substitutablecomponents. Inventions alter the structure of these net-works by creating relationships among technologiesand changing the way subsequent additions connect.Networks of technologies evolve as inventions consol-idate or destabilize the status quo by increasing ordecreasing the use of incumbent technologies.

We build on studies that distinguish between tech-nologies that render firm capabilities obsolete (andtherefore are competency destroying) and those thatimprove the value of capabilities (and therefore arecompetency enhancing) (Abernathy and Clark 1985,Tushman and Anderson 1986, Christensen 1997). How-ever, our approach differs from these works. First,studies of competency-enhancing and competency-destroying inventions focus on major changes, i.e., dis-continuities that fundamentally break with existingstandards (Tripsas 1997, Kaplan et al. 2003, Feldmanand Yoon 2012). By contrast, we suggest that a newtechnology’s influence on the status quo is a matter ofdegree, not categorical difference, and we make thisobservation a focal point of our indexes. Second, our

approach decouples the consolidating or destabilizingeffects of new technologies from organizations’ compe-tencies. Although the implications of new inventionsare felt forcefully in terms of how they influence incum-bents’ capabilities (Benner 2007; Sosa 2009, 2011), theireffects ripple more broadly through dynamic networksof prior and subsequent technologies.

Using our network model of technological change,we propose a measure, the CDt index, that quantifiesthe extent to which an invention consolidates or desta-bilizes the subsequent use of the components on whichit builds. Combining our CDt index with an impactweight yields an additional mCDt index that charac-terizes the magnitude of an invention’s consolidationor destabilization of the status quo. These indexes aresensitive to the enhancing and destructive effects ofimportant new technologies.

Applying these measures to the context of univer-sity research commercialization, we find clear evi-dence that, when universities have greater commercialengagement, they tend to create technologies that con-solidate the status quo but, when they receive morefederal funding for academic research, they tend toproduce more destabilizing inventions. Commercialengagement and federal support are both positivelyassociated with forward citations of university patents.This important finding points to the challenge of rely-ing on forward-citation-based impact measures to dis-tinguish among patents.

In what follows, we first review existing measures.Next, we present our CDt and mCDt indexes. Sub-sequent sections use patent data in four validationexercises. First, we demonstrate that our indexes dis-criminate among patents along dimensions that areknown to be associated with the destabilizing or con-solidating nature of technologies. Second, we presentcase studies that illustrate core features of our indexeswhile attesting to their ability to identify recognizedbreakthroughs. Third, we demonstrate the usefulnessof our approach with regressions that compare patents’impact with their scores on the CDt and mCDt indexes.Finally, we present several different aggregate ver-sions of our measures and evaluate their potentialfor organizational-level analyses in models that exam-ine how features of the academic research enterpriseexplain characteristics of university patents.

2. Measuring the E�ects of Technologies2.1. Existing MeasuresMany quantitative metrics treat technologies as vari-able in their impact, i.e., the extent to which they arelater used. Although the true impact of a technologyis difficult to measure, citations of papers and patentsare a common proxy (Griliches 1990, Trajtenberg1990, Harhoff et al. 1999, Hall and Trajtenberg 2004,Lanjouw and Schankerman 2004, Wuchty et al. 2007,

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.


Management Science, 2017, vol. 63, no. 3, pp. 791–817, © 2016 INFORMS 793

Schoenmakers and Duysters 2010). Impact measuresare attractive because they correspond to the intuitiveidea that new technologies that offer improvementsover the state of the art should become more widelyused than less valuable inventions. However, becausethey focus on the magnitude of future use, impact mea-sures of technologies miss the key substantive distinc-tion between new things that are valuable because theyreinforce the status quo and new things that are valu-able because they challenge the existing order. In short,impact measures are good for evaluating the extent towhich new inventions are used, but they are limitedbecause they offer no insight into how they are usedand what those uses do to shape future trajectories,a central objective of evolutionary theories of change(Arthur 2007, 2009; Dosi 1982; Nelson and Winter 1982;Verspagen 2007). This ambiguity makes it problem-atic to use impact measures to develop and test suchtheories.

We argue that, to understand a new technology’seffects, it is necessary to look not only at its impact,but also how the technology fits into existing trajec-tories. Measures that consider only the magnitude ofa technology’s later use miss the crucial point thatnew technologies emerge in environments that arecomprised of other technologies (Arthur 2007, 2009).Existing approaches are therefore unable to describethe substantial effects that inventions may have onthe subsequent use of their technological predecessorsor the evolution of broader technological trajectories.We describe a new measure designed to capture thesevery effects by appeal to network science.

2.2. Consolidating and Destabilizing TechnologiesOur measure has four core features. First, and mostimportant, it is structural in a network sense, able tocapture the extent to which future inventions that buildon a technology also rely on that technology’s pre-decessors. This second-order view of impact impliesthat an invention’s importance stems both from howit influences the use of other technologies and itsown direct use. Second, the measure is dynamic,able to account for variation in the extent to whichan invention alters the use of its predecessors overtime. Third, the measure is continuous, able to cap-ture degrees of consolidation and destabilization rang-ing from large-scale transformations to smaller-scale,incremental shifts. Finally, the measure is valenced,able to distinguish between consolidating and destabi-lizing technologies that may have similar impact butdifferent consequences for the status quo.

We developed our measure using utility patentsgranted by the U.S. Patent and Trademark Office(USPTO). Utility patents cover the majority of patentedinventions. They include any new or improved method,process, machine, manufactured item, or chemical

compound. Although other categories exist for designpatents, which are granted for ornamental designs offunctional items, and plant patents, which cover plantbreeds, most patents granted in the United States arein the utility category, over 90% in 1999 (Hall et al.2002). We focus on utility patents to help to avoid com-plexities that arise from differences in citation practicesacross categories.

Patents are attractive for our purposes because theUSPTO requires inventors to include citations of rele-vant technological predecessors (known as “prior art”)in their patent documents. Before being granted, allpatents go through an evaluation process in whichexaminers review these citations for comprehensive-ness. Applicants have incentives to make accuratecitations: listing irrelevant predecessors weakens apatent’s enforceability, but failing to acknowledgerelated work can invalidate a patent.1

2.3. Measure DevelopmentThe CDt index characterizes how future inventionsmake use of the technological predecessors cited by afocal patent. Our intuition is that citations of prede-cessors should decrease after a destabilizing inventionis introduced because the technology entails a breakwith past ways of thinking. By contrast, consolidatinginventions should be cited together with their prede-cessors and therefore increase citations of technologieson which they build. The networks in the bottom leftand right panels of Figure 1 illustrate this idea.

The bottom left panel of Figure 1 depicts a hypo-thetical patent that is maximally destabilizing. Thefocal invention (black diamond) cites four predeces-sor technologies (gray diamonds). These citations con-tribute to the impact of earlier inventions and indicatethe focal patent’s technological lineage. On the rightside of the panel, six subsequent patents (white cir-cles) cite the focal invention, but note that none ofthese future patents (or “forward citations”) also citethe focal invention’s technological predecessors. Thefocal patent is destabilizing because it directs the atten-tion of subsequent inventors and examiners away fromtechnologies that were relevant to its conceptualiza-tion. The bottom right panel of Figure 1 shows a focalpatent that has identical impact to the one in theleft panel (as measured by forward citations) but ismaximally consolidating. As illustrated by the whitesquares, each future patent also cites at least one of thefocal patent’s predecessors, and therefore this hypo-thetical patent appears to consolidate the use of theseearlier technologies.2

Drawing on the tools of network science, we devel-oped our measure by conceptualizing patents as nodesin tripartite networks (or “graphs”).3 A tripartitegraph, denoted G ⇤ (V1 ,V2 ,V3 ,E), is a network withthree generic types or categories of nodes, V1, V2,

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Figure 1. Illustrative Calculations of the CDt Index for Three Patents–2(0)(1) + (0) = 0

–2(1)(1) + (1) = –1

–2(1)(0) + (1) = 1

–2(1)(0) + (1) = 1

–2(1)(1) + (1) = –1

–2(1)(0) + (1) = 1

Compute the numeratorof the CDt index for eachforward citation.

Step 2Sum the values from Step 1to obtain the numerator.

16 = 0.17

Step 3Divide the value fromStep 2 by the total numberof forward citations, n, toobtain the CDt index.

Time (t)

Key

Cites focal patent (type f = 1)Cites predecessors (type b = 1)Cites focal patent and predecessors (type f = 1, b = 1)

–2(1)(0) + (1) = 1

–2(1)(0) + (1) = 1

–2(1)(0) + (1) = 1

–2(1)(0) + (1) = 1

–2(1)(0) + (1) = 1

–2(1)(0) + (1) = 1

(1) + (1) + (1) + (1) + (1) + (1) = 6

66 = 1Maximally destabilizing

–2(1)(1) + (1) = –1

–2(1)(1) + (1) = –1

–2(1)(1) + (1) = –1

–2(1)(1) + (1) = –1

–2(1)(1) + (1) = –1

–2(1)(1) + (1) = –1

(–1) + (–1) + (–1) + (–1) + (–1) + (–1) = –6

–66 = –1Maximally consolidating

Step 1

!–2fi bi + fi = (0) + (–1) + (1) + (1) + (–1) + (1) = 1

Note. For simplicity, we eliminate the weighting parameter wit by setting it to 1.

and V3, and ties (or “edges”) E that connect nodes ofdifferent types. The edges in our graphs are directedbecause citations point from one patent to another, andour graphs are acyclic because patents only cite inven-tions that are temporally prior and therefore there areno paths that begin and end with the same patent(or “node”).

Using this notation, let V1 consist of a focal patent, f ,that we seek to evaluate; V2 consist of predecessorinventions, b, cited by the focal patent; and V3 consistof a set of future patents, i, that eventually cite thefocal patent and/or its predecessors. Types f and bare fixed at focal patent’s issue date, but i can growas time passes and the patent and/or its predecessorsaccrue new citations. Our objective is to characterizehow a new patent, i, joins the network defined by theties (citations) between patents of type b and f . Newpatents may join in one of three ways: (1) i can cite thefocal patent’s predecessors (type b), (2) i can cite the

focal patent (type f ), or (3) i can cite the focal patentand its predecessors (both types f and b). For a focalpatent and vector i⇤ (i1 , i2 , . . . , in�1 , in) of future patentsthat cite the focal patent and/or its technological pre-decessors at time t, we define the CDt index as

CDt ⇤1nt

nXi⇤1

�2 fit bit + fit

wit, wit > 0, (1)

where

fit ⇤

(1 if i cites the focal patent (type f ),0 otherwise,

(2)

and

bit ⇤

8>><>>:

1 if i cites any focal patentpredecessors (type b),

0 otherwise,(3)

where nt is the number of forward citations in i, andwit indexes a matrix W of weights for patent i at time t.4

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



We emphasize that nt is the number of forward cita-tions of both the focal patent and/or its technologicalpredecessors and therefore this quantity differs froma traditional measure of patent impact. The multipli-cation by �2 in the numerator of Equation (1) ensuresthat the measure ranges from �1 to 1, with positive val-ues representing inventions that are more destabilizingand negative values highlighting inventions that aremore consolidating. Figure 1 gives examples of how tocalculate the CDt index for three different patents.

The CDt index captures the direction of an inven-tion’s effects on existing technologies, not the mag-nitude; therefore the measure does not discriminateamong inventions that influence a large stream of sub-sequent work and those that shape the attention of asmaller number of later inventors. By introducing aweighting parameter to Equation (1), the magnitudeof a focal invention’s future use can be incorporated.Formally,

mCDt ⇤mt

nt

nXi⇤1

�2 fit bit + fit

wit, wit > 0, (4)

where mt is a parameter that captures the magnitude ofuse of f at time t. In this formulation, mt differs from ntin that the former counts only citations of the focalpatent, whereas the latter includes citations of both thefocal patent and its predecessors. This measure differsfrom the CDt index by distinguishing among inven-tions according to their overall effect on a network ofinterlinked technologies. The CDt index captures thedirection of an invention’s effects, whereas the mCDtindex mixes both direction and magnitude.

In cases where neither the focal patent nor its techno-logical predecessors receive any citations from futurepatents during the measurement interval, the valueof nt is 0 and the indexes are undefined. When cal-culated five years after the focal patent’s issue date,undefined values of the CDt and mCDt indexes wererare, occurring in only 82,572 cases of 2.9 million, arate of 2.8%. Empirical analyses in later sections exam-ine the implications of these undefined values moreclosely.5

3. Assessments of Face Validity3.1. Descriptive Statistics and CorrelationsUsing data on the 2.9 million U.S. utility patentsgranted between 1977 and 2005, we examine the abil-ity of our indexes to discriminate among inventionsalong dimensions that are known from previous workto have associations with the destabilizing or con-solidating nature of new technologies. Our analysesexclude 34,889 patents that were outside the scope ofthe National Bureau of Economic Research’s (NBER’s)technology categorization system and 177 patents withsubstantial missing data. The basic covariates were

obtained from the Patent Network Dataverse (Li et al.2014) and supplemented with data taken from theUSPTO.

For simplicity, we set the weighting parameter wit inEquations (1) and (4) to 1 so that each future patent ithat eventually cites the focal patent and/or its pre-decessors contributes equally. Most of our analyses inthis and later sections evaluate impact and the CDtand mCDt indexes using forward citations made dur-ing the first five years after the focal patent’s issue,because annual citations of most patents reach theirpeak within this time frame (Jaffe and Trajtenberg2002). We denote these quantities as I5, CD5, and mCD5,respectively. In a small number of cases, we use all cita-tions made through the year 2010, regardless of thefocal patent’s issue date. For clarity, we label these val-ues I2010y , CD2010y , and mCD2010y to indicate the differ-ent approach to calculation.

Table 1 reports descriptive statistics and correlations.Values of the CD5 index roughly approximate a nor-mal distribution with a mode of 0, mean of 0.07, stan-dard deviation of 0.23, and a range from �1 to 1. Thedistribution leans slightly to the right, which suggeststhat modestly destabilizing inventions may be morecommon than slightly consolidating ones. This runscounter to prior research, which shows that destabiliz-ing technologies are rare. However, existing literaturedoes not account for the possibility that destabilizationis a matter of degree. Given that patents are somewhatcostly to obtain and applicants are required to demon-strate how their invention is nonobvious, useful, andnovel, this distributional skew may be expected. More-over, when interpreting the distribution, it is useful toobserve that the measure is only modestly correlatedwith impact (r ⇤ 0.03, p < 0.001). The distribution of themCD5 index is similar to the CD5 index, with a modeof 0, mean of 0.31, standard deviation of 1.75, and arange from �127.84 to 222.67, but the tails are longer.

Other estimates in Table 1 also add support that ourindexes function as intended. First, consider the cor-relations between assignee type and the CD5 index.The (small) negative correlation between the CD5index and firm assignee (r ⇤�0.00, p < 0.01) sug-gests that firms tend to produce more consolidatinginventions, whereas universities (r ⇤ 0.02, p < 0.001)and government laboratories (r ⇤ 0.02, p < 0.001)generate more destabilizing ones. These correla-tions are consistent with previous observations aboutthe differences between these types of organiza-tions. Companies tend to specialize in shorter-term,application-oriented development whereas public sec-tor organizations more often focus on longer-term,fundamental research (Dasgupta and David 1994,Rosenberg and Nelson 1994). Consistent with theseobservations, the CD5 and mCD5 indexes are also pos-itively correlated with acknowledgement of a “gov-ernment interest” (r ⇤ 0.02, p < 0.001, and r ⇤ 0.01,

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Table 1. Descriptive Statistics and Correlations

Variable Mean SD 1 2 3 4 5 6 7 8 9 10 11 12

1. CD5 index 0.07 0.23 1.002. mCD5 index 0.31 1.75 0.53 1.003. Impact (I5) 3.60 5.92 0.03 0.20 1.004. Government interest 0.02 0.14 0.02 0.01 �0.00 1.005. Nonpatent predecessors cited (log) 0.44 0.82 0.00 0.01 0.13 0.13 1.006. Predecessor patents cited 8.86 13.14 �0.17 �0.11 0.19 �0.01 0.26 1.007. Claims 14.21 12.32 �0.04 0.00 0.19 0.01 0.19 0.20 1.008. Distinctiveness 0.69 0.46 0.03 0.03 0.05 0.01 0.03 0.02 0.02 1.009. NBER—Chemical 0.18 0.38 0.01 �0.01 �0.07 0.02 0.07 �0.03 �0.02 0.06 1.00

10. NBER—Computers 0.14 0.35 �0.01 0.05 0.18 �0.01 0.05 0.03 0.09 �0.01 �0.19 1.0011. NBER—Drugs 0.09 0.29 0.02 �0.01 0.02 0.05 0.30 0.03 0.05 �0.04 �0.15 �0.13 1.0012. NBER—Electrical 0.19 0.39 0.03 0.04 0.04 0.03 �0.03 �0.03 0.00 0.01 �0.22 �0.20 �0.16 1.0013. NBER—Mechanical 0.20 0.40 �0.01 �0.03 �0.07 �0.03 �0.16 �0.02 �0.06 �0.03 �0.23 �0.21 �0.16 �0.2414. NBER—Others 0.19 0.40 �0.05 �0.04 �0.07 �0.04 �0.14 0.02 �0.05 0.00 �0.23 �0.20 �0.16 �0.2415. Government 0.02 0.14 0.02 0.00 �0.03 0.28 0.03 �0.03 �0.03 0.01 0.03 �0.01 0.01 0.0216. Firm 0.79 0.41 �0.00 0.02 0.08 �0.17 0.02 0.04 0.07 0.01 0.07 0.12 �0.07 0.0817. University 0.02 0.13 0.02 0.01 0.01 0.29 0.22 �0.01 0.04 0.01 0.02 �0.03 0.14 �0.0118. Median assignee experience (log) 4.43 3.35 0.03 0.05 0.09 0.06 0.09 �0.02 0.04 0.02 0.08 0.21 �0.08 0.1519. Median team distance (log) 1.56 2.19 0.00 0.01 0.06 0.04 0.14 0.05 0.09 0.02 0.08 0.02 0.08 �0.0120. Median team experience (log) 1.26 1.14 �0.02 �0.00 0.08 �0.02 0.10 0.10 0.10 �0.02 0.08 0.04 0.04 0.0521. Inventors 2.15 1.55 0.02 0.03 0.09 0.02 0.17 0.06 0.10 0.02 0.11 0.03 0.09 �0.0122. Examiner experience (log) 5.09 2.60 0.00 �0.01 �0.05 �0.02 �0.10 �0.03 �0.05 �0.03 0.02 �0.12 �0.04 �0.0023. Examiner workload 579.68 528.19 �0.03 0.00 0.07 �0.01 0.02 0.05 0.07 0.00 �0.08 0.12 �0.01 0.1024. Grant lag 2.08 1.97 �0.00 0.01 0.04 0.02 0.10 0.05 0.06 0.02 �0.02 0.10 0.04 �0.0125. Application year 1,991.84 8.14 �0.08 �0.02 0.13 �0.01 0.17 0.16 0.21 �0.04 �0.09 0.13 0.05 0.0626. Grant year 1,993.92 8.10 �0.09 �0.02 0.14 �0.01 0.20 0.17 0.23 �0.03 �0.10 0.15 0.06 0.05

Variable 13 14 15 16 17 18 19 20 21 22 23 24 25 26

13. NBER—Mechanical 1.0014. NBER—Others �0.25 1.0015. Government �0.02 �0.03 1.0016. Firm �0.02 �0.19 �0.26 1.0017. University �0.05 �0.05 �0.01 �0.25 1.0018. Median assignee experience (log) �0.08 �0.28 0.07 0.51 0.01 1.0019. Median team distance (log) �0.06 �0.10 0.03 0.16 0.06 0.17 1.0020. Median team experience (log) �0.05 �0.14 �0.05 0.22 �0.03 0.34 0.06 1.0021. Inventors �0.07 �0.12 0.03 0.19 0.04 0.25 0.49 0.07 1.0022. Examiner experience (log) 0.07 0.06 �0.00 �0.03 �0.03 �0.06 �0.04 �0.02 �0.05 1.0023. Examiner workload �0.02 �0.10 �0.01 0.04 �0.00 0.07 0.02 0.05 0.03 0.29 1.0024. Grant lag �0.04 �0.05 0.02 0.02 0.03 0.03 0.04 �0.00 0.04 �0.13 0.10 1.0025. Application year �0.06 �0.06 �0.04 0.07 0.04 0.15 0.11 0.20 0.15 �0.07 0.19 �0.14 1.0026. Grant year �0.07 �0.07 �0.04 0.08 0.05 0.16 0.12 0.20 0.16 �0.10 0.22 0.10 0.97 1.00

p < 0.001, respectively).6 Government interest patentsresult from federally funded research programs thathave been vetted by peer review and are ori-ented toward more fundamental discoveries. Finally,the small positive correlation between the num-ber of inventors and the CD5 (r ⇤ 0.02, p < 0.001)and mCD5 (r ⇤ 0.03, p < 0.001) indexes impliesthat larger teams may produce more destabilizingtechnologies, a potentially noteworthy addition toresearch on scientific collaboration (Wuchty et al. 2007,Singh and Fleming 2010).

Figure 2 shows the within-year variability of theCD5 index by NBER category. Consistent with cri-tiques about the exploding volume of patenting andperceived lowering of quality thresholds required forobtaining a successful grant (Jaffe and Lerner 2004),

the figure suggests that, across technology categories,the average patent of today is less destabilizing thanthose of the late 1970s and early 1980s. We also findsignificant (all p < 0.001) negative correlations acrosscategories between application and grant year and theCD5 index. However, Figure 2 also shows more outlierson the positive end of the CD5 index in later years.7

3.2. Case StudiesQuantitative descriptions of our indexes are consistentwith established findings about the correlates of var-ious types of inventions. We next examine how wellthey are able to identify and classify particular break-throughs. Table 2 reports the impact and CDt andmCDt indexes measured five years after issue (I5, CD5,and mCD5) and as of 2010 (I2010y , CD2010y , mCD2010y) for

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Figure 2. Distribution of the CD5 Index Among U.S. Utility Patents Granted Between 1977 and 2005

1.0

0.5

–0.5

CD

5 in

dex

–1.0

1980 1990

Drugs

Chemical Computers

Mechanical Others

Electrical

2000

0

1.0

0.5

–0.5

–1.0

1980 1990 2000

0

1.0

0.5

–0.5

CD

5 in

dex

–1.0

1980 1990 2000

0

1.0

0.5

–0.5

–1.0

1980 1990 2000

0

1.0

0.5

–0.5

CD

5 in

dex

–1.0

1980 1990 2000

0

1.0

0.5

–0.5

–1.0

1980 1990 2000

0

select inventions that have had a significant influenceon their fields.

Several patents on enhancements in oil and gas dril-ling are notable among inventions that have consoli-dated the use of their predecessors. These inventions donot establish new methods; they refine already widely

used technologies. For example, patent 4,573,530, “In-Situ Gasification of Tar Sands Utilizing a CombustibleGas,” assigned to Mobil Oil Corporation, describes animproved process for extracting carbon monoxide andhydrogen from tar sands. Process inventions like thishave become important in recent years because con-

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Table 2. Illustrative Patents

CD5 mCD5 Predecessors App. GrantPatent index index I5 I2010y cited year year Title Assignee

4,637,464 �0.20 �0.20 1 194 7 1984 1987 In Situ Retorting of Oil Shalewith Pulsed Water Purge

Amoco Corp.

4,573,530 0.17 0.17 1 192 6 1983 1986 In-Situ Gasification of TarSands Utilizing aCombustible Gas

Mobil Oil Corp.

4,658,215 0.29 0.57 2 200 4 1986 1987 Method for InducedPolarization Logging

Shell Oil Co.

4,928,765 0.00 0.00 0 195 10 1988 1990 Method and Apparatus forShale Gas Recovery

Ramex Synfuels,International, Inc.

6,958,436 �0.85 �127.84 150 150 5 2002 2005 Soybean Variety SE90346 Monsanto Co.5,015,744 �0.46 �17.12 37 173 4 1989 1991 Method for Preparation of

Taxol Using an OxazinoneFlorida State

University6,376,284 �0.30 �33.04 111 175 19 2000 2002 Method of Fabricating a

Memory DeviceMicron Technology,

Inc.6,063,738 �0.01 �0.39 51 178 12 1999 2000 Foamed Well Cement Slurries,

Additives and MethodsHalliburton Co.

4,724,318 0.09 3.57 42 145 2 1986 1988 Atomic Force Microscope andMethod for ImagingSurfaces with AtomicResolution

IBM Corp.

5,016,107 0.06 1.65 28 163 17 1989 1991 Electronic Still CameraUtilizing ImageCompression and DigitalStorage

Eastman Kodak Co.

6,285,999 0.16 5.22 33 193 7 1998 2001 Method for Node Ranking in aLinked Database

Stanford University

4,356,429 0.00 0.00 2 408 4 1980 1982 Organic ElectroluminescentCell

Eastman Kodak Co.

4,445,050 0.20 0.20 1 150 4 1981 1984 Device for Conversion of LightPower to Electric Power

None

5,010,405 0.60 5.40 9 159 2 1989 1991 Receiver-CompatibleEnhanced DefinitionTelevision System

MIT

4,237,224 0.81 46.77 58 282 1 1979 1980 Process for ProducingBiologically FunctionalMolecular Chimeras

Stanford University

4,399,216 0.70 11.13 16 339 2 1980 1983 Processes for Inserting DNAInto Eucaryotic Cells andfor ProducingProteinaceous Materials

ColumbiaUniversity

4,343,993 1.00 5.00 5 169 0 1980 1982 Scanning TunnelingMicroscope

IBM Corp.

4,683,202 1.00 42.00 42 2, 209 0 1985 1987 Process for AmplifyingNucleic Acid Sequences

Cetus Corp.

ventional gasification methods are difficult to use inremote regions like Northern Alberta, Canada, wherethe largest tar sands deposits are located.

The measure also identifies technologies that areknown to have destabilized the use of their pre-decessors. Patents 4,237,224, 4,399,216, and 4,683,202(the Cohen–Boyer patent on recombinant DNA, theAxel patent on eukaryotic cotransformation, and theMullis patent on polymerase chain reaction (PCR),respectively) are at the bottom of Table 2.8 These threeinventions from the late 1970s and early 1980s set thestage for the molecular biology revolution in pharma-ceutical R&D by making techniques for targeted invitro drug discovery possible (Powell and Owen-Smith

1998). That technological shift altered the internationalpharmaceutical industry in the 1980s and 1990s bychallenging traditional organic-chemistry-based drugdiscovery methods (Henderson and Cockburn 1996,Powell et al. 1996). These patents were also lucrativefor their assignees. The Cohen–Boyer patent generatedsome $255 million in licensing royalties for StanfordUniversity over its lifetime (Hughes 2001). A moreaggressive licensing policy at Columbia University ledthe Axel patent to yield some $790 million in revenues(Colaianni and Cook-Deegan 2009). Finally, some esti-mates place the total royalties for the Mullis patent at$2 billion (Fore et al. 2006).

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



The scanning tunneling microscope (STM, patent4,343,993) and the atomic force microscope (AFM,patent 4,724,318) together enabled the developmentof nanotechnology. Both instruments image and moveindividual atoms on material surfaces, which allowselectronics, medications, and other products to bedesigned and built atom by atom. Although bothinventions are destabilizing on our measure (i.e., theyhave positive scores), the AFM is less so than theSTM. This relative ranking accords with nanotechnol-ogy’s evolution. Although the STM was introduced fiveyears before the AFM, the AFM offered several majorimprovements, including three-dimensional render-ings and the ability to image living organisms (Youtieet al. 2008). The AFM was a radical improvement overprior microscopes, but it was less destabilizing becauseit built on (and cited) the more transformative STM.

Finally, consider patent 5,016,107, for an “ElectronicStill Camera Utilizing Image Compression and Digi-tal Storage.” This is an Eastman Kodak patent for anearly digital camera that employed novel techniquesfor image processing, compression, and recording ona removable storage medium (Lucas and Goh 2009).Relative to the perceived destabilization of digital pho-tography for silver halide imaging, this patent rankslow on the CD5 index with a score of 0.06. One expla-nation is that, unlike the technologies discussed above,a single patent did not cover the advent of digital pho-tography (Christensen 1997).9

The examples described above cover well-knownbreakthroughs from multiple industries and technol-ogy classes. They demonstrate that our approach dis-tinguishes between destabilizing and consolidatinginventions, as well as among degrees within each cat-egory. We now turn to a more detailed considerationof three inventions—glyphosate-resistant soybeans, amethod of ranking of online search results, and aeukaryotic cotransformation technique—that occupydifferent locations on the consolidating–destabilizingspectrum.

Consider Monsanto’s patent 6,958,436, titled “Soy-bean Variety SE90346.” This invention illustrates thecore premise of a consolidating invention. The patentdescribes a genetically engineered soybean that is resis-tant to glyphosate, an herbicide patented by Monsantoin the 1970s. Glyphosate is the active ingredient inMonsanto’s Roundup product line, the best-selling her-bicide worldwide. In addition to glyphosate tolerance,the seed integrates other desirable plant traits, includ-ing improved yield, immunity to various diseases, andresistance to shattering, by using genetic sequencesowned by Monsanto. Genetically engineered seedsincrease the value of the earlier patented chemical andbiological technologies by broadening potential appli-cation areas and excluding competitors (Graff et al.2003, Pollack 2009). In this particular case, Monsanto

has developed a plant variety engineered to resist dam-age from its market-leading herbicide.

The top panel in Figure 3 plots citations betweenthe Monsanto soybean patent, its predecessors, andsubsequent inventions. The line graph below the fig-ure tracks the patent’s annually updated CD5 index.In the network diagram, as in the measure, forwardcitations are divided into three types. Triangles arepatents that cite the focal patent’s predecessors but notthe focal patent. Circles represent patents that cite onlythe focal invention and not its predecessors. Squaresindicate patents that cite both the focal patent and itspredecessors.

The figure shows interesting features of this con-solidating invention. Note that, although the soybeanpatent has received 150 total citations (as of 2010), ithas never been cited independently of the technolo-gies on which it builds. The patent’s introduction alsovirtually eliminated independent citations of its prede-cessors (i.e., triangles).

The line graph that tracks the patent’s CD5 indexstarts at 0 but rapidly approaches �1. This patternimplies that the focal patent and its predecessors arecomplementary in a way that would not be expected ifthis invention had, for instance, opened a new methodof plant engineering.

Although not revealed in the figure, two additionalfeatures of the Monsanto soybean patent are sugges-tive. First, consider raw citation counts. In the five yearsbefore the focal patent was granted, the five techno-logical predecessors received a total of 61 citations, or2.44 citations per patent per year on average. In thefive years following the soybean patent grant, citationsof the predecessors increased by more than 600%, to anaverage of 15.28 per patent per year.10

The second suggestive feature is the antecedenttechnology’s ownership. By the time of application,Monsanto had acquired the firms that owned all butone of the predecessors cited by the focal patent.11

This observation fits with theories of technologicalinnovation that argue that incumbent firms strive toenhance the value of their knowledge bases (Sosa2011). It also suggests interesting possibilities for usingthis approach to examine corporate decisions aboutwhen to litigate competitor’s intellectual property. Wewould predict that the most likely patents to be chal-lenged in established industries are those that consol-idate the technological position of a significant com-petitor. A similar logic might be used in explanationsof merger and acquisition activities in technologicallyintensive fields, where complementary patent portfo-lios loom large.

Next, we examine patent 6,285,999, titled “Methodfor Node Ranking in a Linked Database.” This inven-tion, referred to as PageRank, covers the core algorithmused by Google to measure the importance of web

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Figure 3. Network Diagrams for the Monsanto (Top), PageRank (Middle), and Axel (Bottom) Patents

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

–1.0

0

I2010y: 150

CD2010y index: –0.85

mCD2010y index: –127.84

I5 : 150CD5 index: –0.85mCD5 index: –127.84

(3) 5,767,350 6,958,436(26)5,084,082

(15) 5,304,728

(163) 5,576,474

(164) 5,569,815

Patent 6,958,436, “Soybean VarietySE90346”

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

0

1.0

I2010y: 193

CD2010y index: 0.37

mCD2010y index: 71.41

I5: 33CD5 index: 0.16mCD5 index: 5.22

(62) 5,848,407

6,285,999

(87) 5,832,494

(14) 5,752,241

(97) 5,748,954

(30) 4,953,106 (21) 5,450,535

Patent 6,285,999, “Method For NodeRanking In a Linked Database”

1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

0

1.0CDt index

CDt index

CDt index

4,399,216(5) 3,800,0354,195,125(13)

Patent 4,399,216, “Processes For InsertingDNA Into Eucaryotic Cells and ForProducing Proteinaceous Materials”

I2010y: 339

CD2010y index: 0.95

mCD2010y index: 332.05

I5: 16CD5 index: 0.70mCD5 index: 11.13

6,014,678(31)

Notes. Black diamonds are focal patents, gray diamonds are predecessors, triangles cite predecessors but not the focal patent, squares cite thefocal patent and predecessors, and circles cite the focal patent but not predecessors. Node sizes are proportional to degree centrality, given inparentheses. The CDt index over time appears below each network.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



pages. Before PageRank, search engines used largelyineffective strategies to rank search results (Brin andPage 1998). PageRank, which is owned by Stanford Uni-versity but licensed exclusively to Google, proposeda new method that drew on insights from social net-work theory to rank web pages by the number of linksthey receive from other sites. PageRank’s destabiliz-ing effects are evident in Yahoo!’s June 2000 agreementto make Google its default search engine. Prior to theimplementation of Google’s search technology, Yahoo!was the market-leading search engine. After the intro-duction of this invention, Yahoo! could only maintainits position by adopting its competitor’s technology.

The middle panel of Figure 3 presents the networkof citations involving the PageRank patent, its prede-cessors, and subsequent patents. The figure is instruc-tive when compared to the Monsanto soybean patentabove it. Unlike the soybean patent, which quicklygarnered a substantial number of citations, PageRankappears to have been overlooked in years immediatelyfollowing its 2001 publication.12 Most citations between2002 and 2007 cite PageRank’s technological predeces-sors, but not the patent itself. This changes between2007 and 2010, when more subsequent inventions citePageRank without its predecessors. During this period,PageRank increasingly destabilizes its predecessors’use. This pattern of change is also visible in the CD5index trend line, which begins near 0 and rises to 0.37.

For our final case, we turn to the Axel patent oneukaryotic cotransformation, in the bottom panel ofFigure 3. This destabilizing discovery is one of thefoundational inventions of biotechnology. The patentcovers a method for inserting foreign genes intocells that then produce associated proteins. Alongwith the bacteriological method of recombinant DNAinvented by Cohen and Boyer, cotransformation is afundamental tool in biologically based drug devel-opment. In contrast to both Monsanto’s soybean andGoogle’s PageRank, citations of the predecessors citedby Columbia’s Axel patent effectively cease withintwo years of its publication. Of nearly 340 citations ofthe Axel patent, only one subsequent invention citesit together with its predecessors and only 14 citedthe predecessors on their own.13 Indeed, the trendline below this panel suggests that a swift change inthe CD5 index happened three years after issue andwas followed by a smooth rise to a maximum valueof 0.95.

The juxtaposition of the cotransformation and soy-bean patents is interesting when viewed in light ofresearch on technology paradigms and industry lifecycles. If the Axel invention and similar technologieslaid the foundations for molecular biology in the late1970s and early 1980s, then subsequent breakthroughsthat follow much later but operate within the sameparadigm should be more consolidating, as is the case

with the Monsanto patent. Tracing changes in the CDtindex of inventions that belong to particular lineagesover time might offer new means to evaluate indus-try evolution and the process by which knowledgeparadigms and technology standards coalesce, expand,and are swept away.

3.3. Regression AnalysesOur analyses so far have looked at the face validity anddiscriminatory power of the CDt and mCDt indexes rel-ative to patent impact. We found that (1) the CDt indexis essentially uncorrelated with impact, and (2) thetechnologies our approach identifies as consolidatingor destabilizing are sensible and echo published eval-uations of high-profile technologies.

In this section, we begin to explore the value of theCDt and mCDt indexes for substantive research. We dothis by considering the CDt and mCDt indexes along-side impact in sets of regressions that explore argu-ments from current literature about the determinantsof important new technologies. We begin by presentingmodels at the patent level and then consider adapta-tions of the CDt and mCDt indexes at the organizationlevel in analyses of university patenting.3.3.1. Patent-Level Analyses.Sample. Our data consisted of the 2.9 million U.S.utility patents that (1) were granted after 1976 butbefore 2006, (2) were available in full-text form fromthe USPTO (this criterion eliminated a small numberof patents, most of which had been withdrawn fromissue), and (3) were assigned to one of the six NBERtechnology categories. For each patent, we collectedcovariates at the patent, assignee, team, and examinerlevels.Patent Importance. We measured our three indicatorsof patent importance—the CD5 and mCD5 indexes andimpact (I5)—using citations received during the firstfive years after being granted. As noted previously, theCD5 and mCD5 indexes are undefined when neither afocal patent nor its predecessors receive any citations.To better understand the behavior of the indexes, wetherefore also considered models where we replacedthese undefined values with 0.Patent Covariates. One alternative interpretation ofour indexes is that, rather than capturing differencesin the effects of new technologies, they record varia-tion in opportunity. As a focal patent cites more pre-decessors, it also makes it easier for future patents tocite those predecessors, which in turn may cause thefocal patent to appear less destabilizing. To explore thispossibility, we include a variable, Predecessor patentscited, that counts the number of citations made by thefocal patent to U.S. utility patents.

In addition to prior inventions, U.S. patents mustalso disclose when they build on existing knowledge

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



that is not subject to patent protection (Fleming andSorenson 2004). Although these works (usually techni-cal manuals or scientific papers) are not incorporatedin our indexes, we include a count of Nonpatent pre-decessors cited (log) in our models, because these mayinfluence the nature and extent of a patent’s uses. Forexample, inventions that cite scientific papers may becloser to the frontiers of knowledge and therefore morelikely to open new technological domains.

Prior research also indicates that a patent’s scopeand distinctiveness are good indicators of its impor-tance (Hall and Trajtenberg 2004, Lanjouw andSchankerman 2004). Some patents describe improve-ments within a narrow area whereas others makebroader assertions. To capture these differences inscope, we followed earlier research and included in ourmodels a count of the number of Claims made by eachfocal patent. We also used prior work to guide our mea-sure of distinctiveness, which attempts to capture howdifferent an invention is from earlier ones by determin-ing whether it proposes novel combinations. Followingthis logic, we assign patents a Distinctiveness value of 1if they are classified with a previously uncombined setof USPTO subclasses and 0 otherwise (Fleming 2001,Funk 2014).

The context of discovery also influences a newtechnology’s importance. Although many contextualfactors matter, contemporary discussions often focuson the effects of sponsorship. We therefore includea dummy variable that captures whether a focalpatent declares a Government interest, meaning that theresearch leading to the patent was supported by fed-eral grants. Grant support may indicate one of severalthings that could bear on the CD5 or mCD5 indexes.First, because federal R&D funding is competitive,grant support may simply be an indicator for high-quality research. That in itself may bear on a patent’seventual impact but does not suggest a clear impli-cation for its destabilization. However, if the grant-funded research is higher quality and oriented to morefundamental objectives than privately funded R&D, wemight expect government interest patents to be moredestabilizing.

Finally, to evaluate the effect of changes in citationpatterns over time and across different kinds of inven-tions, we included indicator variables for the grant yearand NBER category of each focal patent (Mehta et al.2010): Computers, Drugs, Electrical, Mechanical, or Others(with Chemical omitted).

Assignee Covariates. Researchers have shown thatassignee characteristics may also influence the impor-tance of new technologies. Universities, governmentlaboratories, and private firms differ in terms of theirresearch missions and may generate different kindsof inventions (Dasgupta and David 1994, Rosenberg

and Nelson 1994, Stephan 2012). To examine poten-tial differences attributable to organizational form, weincluded indicators for whether each focal patent listedan assignee that was a Government entity (including for-eign and domestic), Firm, or University. Some patentshave multiple assignees of different forms, in whichcase we assigned them a value of 1 on each of the rel-evant indicators. The USPTO does not keep systematicrecords on assignees’ organizational forms and, there-fore, to obtain our indicators, we relied on featuresof assignee names. We began by manually coding asmall subsample, which we then used to train a naïveBayes classifier to label the more than 300,000 uniqueassignee names in the USPTO database.

Experience may also influence technological impor-tance. For instance, as an organization develops andpatentsmore inventions, itmayacquireaunique knowl-edge of particular components that when broughttogether form a destabilizing new technology. Bycontrast, there may also be an association betweenexperience and the production of more consolidatinginventions, as organizations seek to bolster or expanda particular technological trajectory. Recall the veryclear case of such a development strategy, apparent inMonsanto’s soybean patent. To capture these possibil-ities, we control for the stock of eventually successfulpatent applications that assignees made before apply-ing for the focal patent. When there was more than oneassignee we used the median stock of patents. We alsoadded a quadratic term in the models to account fordiminishing returns to experience.Team Covariates. Invention is a social activity(Wuchty et al. 2007) and the structure of the team thatcreates a new technology may influence that technol-ogy’s importance. Prior research indicates, for instance,that despite advances in information technology, geo-graphic proximity among inventors still matters (Funk2014). To capture these differences, we included ameasure of Median team distance (log). This variablerecords the median geographic distance among eachfocal patent’s inventors, adjusted for Earth’s curvature.It takes a value of 0 for patents with a single inven-tor. We also added Median team experience (log) anda quadratic term, which we measured as the mediannumber of ultimately granted patents applied for bythe inventors of the focal patent. To evaluate differencesin team size, we included a count of the number ofInventors listed on the focal patent in each model.USPTO Covariates. Our last set of covariates focusedon the USPTO examination process. As with assigneesand inventors, examiner experience may influence atechnology’s importance. Lemley and Sampat (2012)reported that examiners who have prosecuted morepatent applications tended to cite fewer predecessortechnologies and were also less likely to issue rejections

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



that would require applicants to revise their claims.We therefore added Examiner experience (log) to ourmodels, which we measured as the number of grantedpatents on which the examiner of the focal patent alsoserved as an examiner, before the focal patent’s appli-cation date. The time and energy an examiner hasto devote to review may also influence citations ofprior patents. To capture these differences, we added acovariate for Examiner workload. This variable countedthe number of open patent applications assigned to theexaminer at the same time as the focal patent by iden-tifying granted patents with application dates beforeand grants dates after a focal patent that were evalu-ated by the same examiner. This measure likely under-estimates the actual workload of examiners becausewe do not include applications that were eventuallydenied. We also controlled for Grant lag, which was thenumber of years between the grant year and applica-tion year of the focal patent (Mehta et al. 2010).14

Statistical Approach. We used ordinary least squares(OLS) regressions to model the CD5 and mCD5 indexesbecause both indexes are continuous and our data arecross-sectional. Models of impact (I5) used a negativebinomial specification because the variable takes ononly nonnegative integer values and likelihood ratiotests of overdispersion suggested that a Poisson processdid not generate the data. After excluding patents withundefined CD5 and mCD5 indexes we had 2,828,700observations. In models of impact (I5) and those thatset undefined values of CD5 and mCD5 to 0, our sam-ple size increased to 2,910,506. Correlations among thecovariates were low and the variance inflation factors(VIFs) and tolerances did not indicate multicollinearity.3.3.2. Patent-Level Results. Table 3 displays estimatesfrom four negative binomial models of patent impact(I5). Overall, the coefficients are consistent in sign andsignificance and are supportive of findings from ear-lier research. At the patent level, all four models showa positive association between Predecessor patents cited,Nonpatent predecessors cited (log), Claims, and Distinc-tiveness, which together suggest that higher-impactpatents tend to be broader in scope and novelty, whilealso building on existing streams of technology and sci-ence. Impact also varies across technological domains.Relative to Chemical (the omitted NBER category),patents in the Computers, Drugs, Electrical, Mechanical,and Others categories receive—according to estimatesin Model 4—between e0.120⇥1 ⇡ 1.13 and e0.710⇥1 ⇡ 2.03times more citations within five years of being granted.Also noteworthy, Models 2 and 3 suggest a negativeassociation between Government interest and impact,although that effect washes away in more fully speci-fied models.

There are significant associations between assignee-level characteristics and impact, such that patents

awarded to a Firm or University are more highlycited, whereas those with a Government assignee areless acknowledged by future inventors. Echoing someresearch on team science (Wuchty et al. 2007), patentswith more Inventors and greater Median team distance(log) receive more citations, as do those that havelonger Grant lag and those that are reviewed by exam-iners with higher Examiner workload. Finally, we findconsistent evidence that experience matters for patentimpact, but in different ways across levels. Increases inMedian assignee experience (log) and Median team experi-ence (log) are associated with higher impact. However,as Median examiner experience (log) increases, impactdecreases, a finding that is consistent with work byLemley and Sampat (2012).

OLS models of the CD5 and mCD5 indexes are shownin Tables 4 and 5, respectively. Although the coef-ficients in each table are internally consistent withrespect to their sign and significance, the estimatesand their interpretations differ from our impact mod-els. Beginning with the patent level, estimates in Mod-els 5–8 of Table 4 and Models 11–14 of Table 5 revealpositive relationships between reporting a Governmentinterest and the CD5 and mCD5 indexes, respectively.This finding makes sense in light of the federal gov-ernment’s tendency to fund basic research that seeksto open new areas.

Models in Tables 4 and 5 also suggest that the rela-tionship between technological importance and theuse of existing knowledge is more complex than isapparent from models of impact. Although Predecessorpatents cited and Nonpatent predecessors cited (log) bothhave a positive relationship with patent impact, theformer has a negative association with the CD5 andmCD5 indexes, whereas the latter has a positive associ-ation. These relationships are open to several interpre-tations. First, they may indicate that patents that buildon more patented inventions tend to consolidate thestatus quo, whereas those that build on more knowl-edge that is not subject to patent protection, includingscientific papers, tend to destabilize it. Second, the pos-itive association between Nonpatent predecessors cited(log) and our indexes may happen as flurries of newscientific discoveries create opportunities for techno-logical advances that are destabilizing (Azoulay et al.2007). Finally, the relationship may result from patentsbuilding on scientific and other knowledge that is oldbut has never been applied to a technological domain.

Coefficients on our variables for assignee type alsoraise the possibility that evaluating impact alone mayconceal important nuances. For instance, Model 4 inTable 3 reports a positive association between assign-ment to a Firm or University and a patent’s impact.Moreover, with respective increases in citations ofe0.092⇥1 ⇡ 1.10 and e0.036⇥1 ⇡ 1.04, the magnitudes of the

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Table 3. Negative Binomial Regression Models of Patent Impact (I5)

Model 1 Model 2 Model 3 Model 4

PatentGovernment interest 0.006 �0.081⇤⇤⇤ �0.090⇤⇤⇤ 0.009

(0.007) (0.006) (0.006) (0.007)Nonpatent predecessors cited (log) 0.049⇤⇤⇤ 0.045⇤⇤⇤ 0.052⇤⇤⇤ 0.042⇤⇤⇤

(0.001) (0.001) (0.001) (0.001)Predecessor patents cited 0.014⇤⇤⇤ 0.013⇤⇤⇤ 0.014⇤⇤⇤ 0.013⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Claims 0.014⇤⇤⇤ 0.014⇤⇤⇤ 0.015⇤⇤⇤ 0.014⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Distinctiveness 0.164⇤⇤⇤ 0.167⇤⇤⇤ 0.163⇤⇤⇤ 0.161⇤⇤⇤

(0.002) (0.002) (0.002) (0.002)NBER—Computersa 0.700⇤⇤⇤ 0.753⇤⇤⇤ 0.695⇤⇤⇤ 0.710⇤⇤⇤

(0.003) (0.003) (0.003) (0.003)NBER—Drugsa 0.277⇤⇤⇤ 0.248⇤⇤⇤ 0.218⇤⇤⇤ 0.269⇤⇤⇤

(0.004) (0.004) (0.004) (0.004)NBER—Electricala 0.389⇤⇤⇤ 0.425⇤⇤⇤ 0.386⇤⇤⇤ 0.400⇤⇤⇤

(0.003) (0.003) (0.003) (0.003)NBER—Mechanicala 0.099⇤⇤⇤ 0.116⇤⇤⇤ 0.077⇤⇤⇤ 0.120⇤⇤⇤

(0.003) (0.003) (0.003) (0.003)NBER—Othersa 0.108⇤⇤⇤ 0.109⇤⇤⇤ 0.045⇤⇤⇤ 0.140⇤⇤⇤

(0.003) (0.003) (0.003) (0.003)Grant year indicators Yes Yes Yes Yes

AssigneeGovernment �0.273⇤⇤⇤ �0.273⇤⇤⇤

(0.007) (0.007)Firm 0.113⇤⇤⇤ 0.092⇤⇤⇤

(0.003) (0.003)University 0.037⇤⇤⇤ 0.036⇤⇤⇤

(0.008) (0.008)Median assignee experience (log) 0.019⇤⇤⇤ 0.011⇤⇤⇤

(0.000) (0.000)Median assignee experience2 (log) �0.000 0.000⇤⇤

(0.000) (0.000)Team

Median team distance (log) 0.009⇤⇤⇤ 0.007⇤⇤⇤

(0.000) (0.000)Median team experience (log) 0.068⇤⇤⇤ 0.051⇤⇤⇤

(0.001) (0.001)Median team experience2 (log) �0.011⇤⇤⇤ �0.008⇤⇤⇤

(0.001) (0.001)Inventors 0.039⇤⇤⇤ 0.032⇤⇤⇤

(0.001) (0.001)Examiner

Examiner experience (log) �0.013⇤⇤⇤ �0.011⇤⇤⇤

(0.001) (0.001)Examiner experience2 (log) �0.000⇤⇤ �0.000

(0.000) (0.000)Examiner workload 0.000⇤⇤⇤ 0.000⇤⇤⇤

(0.000) (0.000)Grant lag 0.003⇤⇤ 0.005⇤⇤⇤

(0.001) (0.001)Constant �0.053⇤⇤⇤ �0.030⇤⇤⇤ �0.006 �0.118⇤⇤⇤

(0.007) (0.006) (0.006) (0.007)N 2,910,506 2,910,506 2,910,506 2,910,506Log-likelihood �6,722,030.578 �6,721,725.011 �6,730,242.211 �6,715,399.148

Note. Robust standard errors are in parentheses.aThe omitted NBER technology category is Chemical.⇤⇤p < 0.01; ⇤⇤⇤p < 0.001.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Table 4. OLS Regression Models of Patent CD5 Index

Model 9 Model 10Model 5 Model 6 Model 7 Model 8 Undefined⇤ 0 Undefined⇤ 0

PatentGovernment interest 0.005⇤⇤⇤ 0.012⇤⇤⇤ 0.012⇤⇤⇤ 0.005⇤⇤⇤ 0.005⇤⇤⇤ 0.005⇤⇤⇤

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)Nonpatent predecessors cited (log) 0.013⇤⇤⇤ 0.013⇤⇤⇤ 0.014⇤⇤⇤ 0.013⇤⇤⇤ 0.011⇤⇤⇤ 0.012⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Predecessor patents cited �0.003⇤⇤⇤ �0.003⇤⇤⇤ �0.003⇤⇤⇤ �0.003⇤⇤⇤ �0.003⇤⇤⇤ �0.003⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Claims �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Distinctiveness 0.013⇤⇤⇤ 0.012⇤⇤⇤ 0.013⇤⇤⇤ 0.012⇤⇤⇤ 0.013⇤⇤⇤ 0.012⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)NBER—Computersa 0.005⇤⇤⇤ 0.007⇤⇤⇤ 0.007⇤⇤⇤ 0.008⇤⇤⇤ 0.011⇤⇤⇤ 0.008⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)NBER—Drugsa 0.018⇤⇤⇤ 0.017⇤⇤⇤ 0.017⇤⇤⇤ 0.018⇤⇤⇤ 0.015⇤⇤⇤ 0.017⇤⇤⇤

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)NBER—Electricala 0.016⇤⇤⇤ 0.018⇤⇤⇤ 0.017⇤⇤⇤ 0.019⇤⇤⇤ 0.021⇤⇤⇤ 0.018⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)NBER—Mechanicala �0.004⇤⇤⇤ �0.004⇤⇤⇤ �0.006⇤⇤⇤ �0.003⇤⇤⇤ �0.001† �0.003⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)NBER—Othersa �0.015⇤⇤⇤ �0.016⇤⇤⇤ �0.019⇤⇤⇤ �0.014⇤⇤⇤ �0.012⇤⇤⇤ �0.014⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Grant year indicators Yes Yes Yes Yes Yes Yes

AssigneeGovernment �0.001 �0.004⇤⇤ �0.005⇤⇤⇤ �0.004⇤⇤

(0.001) (0.001) (0.001) (0.001)Firm �0.008⇤⇤⇤ �0.009⇤⇤⇤ �0.008⇤⇤⇤ �0.008⇤⇤⇤

(0.001) (0.001) (0.001) (0.001)University 0.005⇤⇤⇤ 0.004⇤⇤ 0.003⇤ 0.004⇤⇤

(0.001) (0.001) (0.001) (0.001)Median assignee experience (log) 0.002⇤⇤⇤ 0.002⇤⇤⇤ 0.002⇤⇤⇤ 0.002⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Median assignee experience2 (log) �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Team

Median team distance (log) �0.001⇤⇤⇤ �0.001⇤⇤⇤ �0.001⇤⇤⇤ �0.001⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Median team experience (log) �0.001⇤⇤⇤ �0.002⇤⇤⇤ �0.002⇤⇤⇤ �0.002⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Median team experience2 (log) 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Inventors 0.006⇤⇤⇤ 0.005⇤⇤⇤ 0.005⇤⇤⇤ 0.005⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Examiner

Examiner experience (log) 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Examiner experience2 (log) 0.000† 0.000⇤ 0.000⇤ 0.000⇤

(0.000) (0.000) (0.000) (0.000)Examiner workload �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤ �0.000⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Grant lag 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Imputed �0.105⇤⇤⇤

(0.000)Constant 0.130⇤⇤⇤ 0.110⇤⇤⇤ 0.122⇤⇤⇤ 0.120⇤⇤⇤ 0.109⇤⇤⇤ 0.120⇤⇤⇤

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)N 2,828,700 2,828,700 2,828,700 2,828,700 2,910,506 2,910,506R2 0.040 0.041 0.040 0.041 0.038 0.044

Note. Robust standard errors are in parentheses.aThe omitted NBER technology category is Chemical.†p < 0.1; ⇤p < 0.05; ⇤⇤p < 0.01; ⇤⇤⇤p < 0.001.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Table 5. OLS Regression Models of Patent mCD5 Index

Model 15 Model 16Model 11 Model 12 Model 13 Model 14 Undefined ⇤ 0 Undefined⇤ 0

PatentGovernment interest 0.031⇤⇤ 0.040⇤⇤⇤ 0.035⇤⇤⇤ 0.034⇤⇤⇤ 0.032⇤⇤⇤ 0.031⇤⇤

(0.010) (0.009) (0.009) (0.010) (0.009) (0.009)Nonpatent predecessors cited (log) 0.070⇤⇤⇤ 0.066⇤⇤⇤ 0.072⇤⇤⇤ 0.064⇤⇤⇤ 0.058⇤⇤⇤ 0.062⇤⇤⇤

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)Predecessor patents cited �0.015⇤⇤⇤ �0.015⇤⇤⇤ �0.015⇤⇤⇤ �0.015⇤⇤⇤ �0.015⇤⇤⇤ �0.015⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Claims 0.003⇤⇤⇤ 0.003⇤⇤⇤ 0.003⇤⇤⇤ 0.003⇤⇤⇤ 0.003⇤⇤⇤ 0.003⇤⇤⇤

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Distinctiveness 0.113⇤⇤⇤ 0.113⇤⇤⇤ 0.113⇤⇤⇤ 0.111⇤⇤⇤ 0.112⇤⇤⇤ 0.107⇤⇤⇤

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)NBER—Computersa 0.299⇤⇤⇤ 0.331⇤⇤⇤ 0.312⇤⇤⇤ 0.311⇤⇤⇤ 0.320⇤⇤⇤ 0.309⇤⇤⇤

(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)NBER—Drugsa 0.023⇤⇤⇤ 0.010⇤ 0.002 0.020⇤⇤⇤ 0.010⇤ 0.015⇤⇤⇤

(0.005) (0.005) (0.005) (0.005) (0.005) (0.005)NBER—Electricala 0.195⇤⇤⇤ 0.218⇤⇤⇤ 0.202⇤⇤⇤ 0.209⇤⇤⇤ 0.216⇤⇤⇤ 0.206⇤⇤⇤

(0.003) (0.003) (0.003) (0.004) (0.003) (0.003)NBER—Mechanicala 0.003 0.012⇤⇤⇤ �0.008⇤⇤ 0.016⇤⇤⇤ 0.023⇤⇤⇤ 0.016⇤⇤⇤

(0.003) (0.003) (0.003) (0.003) (0.003) (0.003)NBER—Othersa �0.043⇤⇤⇤ �0.041⇤⇤⇤ �0.069⇤⇤⇤ �0.030⇤⇤⇤ �0.021⇤⇤⇤ �0.029⇤⇤⇤

(0.003) (0.003) (0.003) (0.003) (0.003) (0.003)Grant year indicators Yes Yes Yes Yes Yes Yes

AssigneeGovernment �0.121⇤⇤⇤ �0.137⇤⇤⇤ �0.137⇤⇤⇤ �0.131⇤⇤⇤

(0.008) (0.008) (0.008) (0.008)Firm �0.038⇤⇤⇤ �0.051⇤⇤⇤ �0.046⇤⇤⇤ �0.049⇤⇤⇤

(0.003) (0.004) (0.003) (0.003)University 0.019 0.011 0.006 0.008

(0.012) (0.012) (0.011) (0.011)Median assignee experience (log) 0.016⇤⇤⇤ 0.013⇤⇤⇤ 0.013⇤⇤⇤ 0.013⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Median assignee experience2 (log) 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤ 0.001⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Team

Median team distance (log) �0.002⇤⇤ �0.002⇤⇤ �0.002⇤ �0.002⇤⇤

(0.001) (0.001) (0.001) (0.001)Median team experience (log) 0.007⇤⇤⇤ �0.001 0.000 �0.000

(0.001) (0.001) (0.001) (0.001)Median team experience2 (log) 0.001 0.001 0.000 0.001

(0.001) (0.001) (0.001) (0.001)Inventors 0.039⇤⇤⇤ 0.036⇤⇤⇤ 0.034⇤⇤⇤ 0.035⇤⇤⇤

(0.001) (0.001) (0.001) (0.001)Examiner

Examiner experience (log) �0.003⇤⇤⇤ �0.002⇤⇤ �0.002⇤⇤ �0.002⇤

(0.001) (0.001) (0.001) (0.001)Examiner experience2 (log) 0.000 0.000 0.000 0.000

(0.000) (0.000) (0.000) (0.000)Examiner workload �0.000 �0.000 �0.000 �0.000

(0.000) (0.000) (0.000) (0.000)Grant lag 0.005⇤⇤⇤ 0.005⇤⇤⇤ 0.005⇤⇤⇤ 0.005⇤⇤⇤

(0.001) (0.001) (0.001) (0.001)Imputed �0.373⇤⇤⇤

(0.002)Constant 0.288⇤⇤⇤ 0.178⇤⇤⇤ 0.242⇤⇤⇤ 0.220⇤⇤⇤ 0.188⇤⇤⇤ 0.226⇤⇤⇤

(0.006) (0.006) (0.006) (0.007) (0.007) (0.007)N 2,828,700 2,828,700 2,828,700 2,828,700 2,910,506 2,910,506R2 0.023 0.023 0.022 0.024 0.023 0.025

Note. Robust standard errors are in parentheses.aThe omitted NBER technology category is Chemical.⇤p < 0.05; ⇤⇤p < 0.01; ⇤⇤⇤p < 0.001.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



associations are nearly identical. This result is surpris-ing in light of prior research on the differences betweenapproaches to research and invention at firms and uni-versities (Dasgupta and David 1994, Rosenberg andNelson 1994). Estimates using our indexes are moreconsistent with these earlier arguments. The negativeand significant coefficients on the Firm indicator inModel 8 of Table 4 (for the CD5 index) and Model 14of Table 5 (for the mCD5 index) suggest that commer-cial entities produce patents that do more to consoli-date the status quo than other assignees. By contrast,the coefficients for the University assignee indicator arepositive, which suggests that academic organizationsgenerate inventions that destabilize existing technol-ogy streams, although only the coefficient on the CD5index is significant. In other words, both universitiesand firms tend to produce higher-impact inventionsthan other types of organizations, but that similaritymasks important differences in how the technologiesenter and change existing technological trajectories.

Results reported in Tables 4 and 5 also suggest thatthe CD5 and mCD5 indexes may offer insights intodistributed teams that are not apparent from impactalone. As noted above, Table 3 reports a positiveassociation between Median team distance (log) andpatent impact, but this finding is somewhat hard tointerpret in light of earlier research. Although somestudies find value in distributed collaboration (Owen-Smith and Powell 2004), others emphasize the chal-lenges of working across distances (Funk 2014). Thepositive and significant coefficients on Median team dis-tance (log) in Tables 4 (for the CD5 index) and 5 (forthe mCD5 index) imply that geographically distributedteams may have a greater tendency to create inventionsthat consolidate the status quo.

Finally, several coefficients are noteworthy becausethey help to further establish the face validity of ourindexes, although they do not contrast with the resultsof our impact models. Patents that bridge a previouslyuncombined set of USPTO subclasses, for instance,and therefore score a 1 on our measure of Distinctive-ness also challenge the status quo, as indicated by thepositive and significant coefficients in Tables 4 and 5.Moreover, consistent with the research on team struc-ture and invention, we also find, again in Tables 4 (forthe CD5 index) and 5 (for the mCD5 index), a positiveand significant association with Inventors: larger teamsproduce more destabilizing inventions. Last, models ofthe CD5 and mCD5 indexes both show positive coeffi-cients for Grant lag, which is consistent with the ideathat examiners require more time to evaluate patentsthat depart from the status quo and therefore theirprior experience.Robustness Checks. We also examined the robust-ness of our models to alternative specifications. First,to determine whether our focus on patents granted

between 1976 and 2006 influenced our findings, wereran our models on subsamples that began severalyears later and ended several years earlier. Our find-ings were similar to those presented here. Second, asnoted before, the CD5 and mCD5 indexes are unde-fined when neither a focal patent nor its predecessorsreceive any citations after the focal patent’s issue, andtherefore we had to exclude these patents from ourmodels. To gain some insight into whether these exclu-sions influenced our results, we estimated new modelswhere we set the CD5 and mCD5 indexes to 0. Theresults appear in Models 9 and 10 (for the CD5 index)and Models 15 and 16 (for the mCD5 index). Models10 and 16 are identical to Models 9 and 15, with theexception of an Imputed indicator variable that takes ona value of 1 for the newly added patents. Once again,the results are nearly identical to our preferred models.3.3.3. Organization-Level Analyses. The precedinganalyses examined the relationships between impactand the CDt and mCDt indexes and factors that areknown to have associations with the importance ofindividual patents. In this section, we move to a higherlevel of analysis to evaluate the ability of the CDtand mCDt indexes to offer insight into patenting byorganizations. Aggregate adaptations of the CDt andmCDt indexes may be useful in research consideringthe sources and consequences of technological changefor organizations and industries.

Our analyses focus on U.S. utility patents issued tothe 110 most research-intensive American universities.The data set covers a broad set of high- and low-impactinventions and therefore serves as a useful setting inwhich to examine the value of our approach for ana-lyzing portfolios of patents. Research on the organi-zational sources of important academic patents hasyielded mixed results. We believe that the ambigu-ity in this line of work may stem from its reliance onimpact measures of technological importance. Below,we demonstrate that, when analyzing portfolio-levelvariation, the CD5 and mCD5 indexes offer insightsinto the dynamics of academic patenting that are lessapparent in examinations of patent impact.

We consider four variations of the CDt and mCDtindexes that we have adapted for use with patentportfolios. We chose these four adaptations for theirdiversity and conceptual appeal, not because theyexhausted the list of possible approaches. The firstmeasure, CDmean

5 , is the average CD5 index amongpatents applied for by each university during year t+1.This is a simple approach to evaluating the consoli-dation or destabilization of multiple patents. Our sec-ond measure, mCDscale

5 , is a more complex adaptationof the mCD5 index that attempts to capture the overalldirection and magnitude of the effects of each univer-sity’s patent portfolio on existing technology streams.We compute the mCDscale

5 measure as the product of the

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



CDmean5 and the total impact (Itotal

5 ) of patents appliedfor during year t +1. The third and fourth adaptations,CDtotal�

5 and CDtotal+5 , are complementary measures that

attempt to separately capture the effects of universities’consolidating and destabilizing patents by countingcitations of patents with positive and negative valueson the CD5 index, respectively. Both the CDtotal�

5 andCDtotal+

5 indexes are computed based on patents appliedfor during year t + 1.

In what follows we briefly survey literature thatseeks to explain the scale and impact of academicpatenting. We then present our models of organiza-tion-level patent importance. Although the findingssuggest interesting new directions for research on aca-demic patenting and the economic value of universityR&D, we pay particular attention to the comparisonof effects across dependent variables in order to high-light multiple possibilities for analyses using aggre-gated versions of the CDt and mCDt indexes.

3.3.4. Research on Academic Patenting. In 1980,Congress passed the Bayh–Dole Act, a law that allowsorganizations that perform federally funded researchto file for patents and issue licenses on intellectualproperty they develop. The act accelerated a trendtoward research commercialization on campus. Overthe last 30 years, academic patenting has increased dra-matically. Some inventions have been highly lucrativefor the institutions that own them by generating newproducts, companies, and even industries.

Proponents of Bayh–Dole call it “prescient” (Cole1993) and herald the act’s importance in turning uni-versity science into an engine of economic devel-opment (Powell et al. 2007). In 2002, The Economist(2002, p. 3) called the act “possibly the most inspiredpiece of legislation enacted in America over the pasthalf century” and went on to attribute to it an impor-tant role in “reversing America’s precipitous slideinto industrial irrelevance.” By the same token, crit-ics of Bayh–Dole and commercialization bemoan the“selling” of the university (Greenberg 2007), linkthe act to the “corporate corruption” of academia(Washburn 2005), and argue that proprietary researchdiminishes universities’ scientific and public mission(Krimsky 2004).

Hyperbole aside, both the economic benefits andthe dangers of academic commercialization are oftenattributed to the type of inventions that universitiesproduce. For those with rosier views, it is the uni-versity’s capability to generate unexpected, market-creating inventions outside the channels of corporateR&D that make academic inventions valuable. In theterms we use here, the economic benefits of Bayh–Dole rely in part on academic scientists’ propensityto generate destabilizing inventions. Although criticsapproach Bayh–Dole from a range of philosophical

starting points, one thread running through most neg-ative appraisals is concern that attention to the com-mercial value of academic ideas leads to capture bycorporate interests, and with it to university scienceclosely wedded to industry priorities. Put differently,one significant concern about Bayh–Dole suggeststhat, as research commercialization becomes morewidespread, both published and patented science willtend to consolidate the status quo.

This tension has led many academic analysts toassess the costs and benefits of university researchcommercialization (Trajtenberg et al. 1997, Hendersonet al. 1998, Mowery et al. 2002). Two key themes in thisline of work emphasize the value of academic inven-tions and the relationship between increases in the useof proprietary science and the vitality of more funda-mental research and training.

Researchers pursuing the former question expressconcern over the relationship between the sourcesand types of R&D funding that support academicresearch and the impact of patented science on cam-pus. Although the results of some studies conflict,most find evidence that increases in the quantity ofacademic patenting do not decrease its impact at thecampus level (Mowery et al. 2002). These studies alsosuggest that connections with industry help academicinstitutions to learn to patent up to a point. Owen-Smith and Powell (2003) find a curvilinear (invertedU-shaped) relationship between ties to biotechnologyfirms and the impact of a university’s patent portfolio.They attribute diminishing returns to the likelihood ofcorporate capture of technology transfer priorities.

The relationship between academic (papers) andproprietary (patents) research outputs has also beenmuch examined. Owen-Smith (2003) demonstrates thatpatent flows are deeply intertwined with traditionallyacademic inputs and outputs such as federal grantsand publications. Sine et al. (2003) find that academicvisibility in the form of article citations increases thelikelihood that patents will be licensed by industry. Fora set of life science patents issued to 89 university cam-puses, Owen-Smith and Powell (2003) find a positiverelationship between the citation impact of life sciencepapers and the overall impact of a university’s patentportfolio.

Sample. To determine whether the CDt and mCDt in-dexes can offer any new insights into debates overacademic patenting, we collected data on a sampleof universities that mirror those studied in previousresearch. We focus on the 110 most research-intensiveU.S. universities (Owen-Smith and Powell 2003) andthe 55,322 utility patents awarded to these institutionsbetween 1976 and 2010. Our sample includes everyinstitution that has ever ranked among the top 100

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



nationally by federal obligations for science and engi-neering research. After defining our sample, we cre-ated a university ⇥ year panel with annually updatedvariables for each institution.Portfolio Importance. We considered six measures ofportfolio importance. The first two measures, vol-ume (V total) and impact (Itotal

5 ), capture the number ofpatents applied for by (and ultimately granted to) eachuniversity during year t + 1 and the total number ofcitations received by those patents, respectively. Vari-ants of these two measures have been widely used inprevious work on university patenting (Mowery andZiedonis 2002, Owen-Smith and Powell 2003). The lastof our six measures are the four portfolio variations onthe CDt and mCDt indexes discussed above: CDmean

5 ,mCDscale

5 , CDtotal�5 , and CDtotal+

5 . As in our earlier analy-ses, all measures are computed using citations receivedby patents in the first five years after being granted.

We frame the analyses we present here in terms ofgroups of right-hand side variables that operational-ize key concepts from the literature. In particular, weattend to measures related to individual university lev-els of experience with patenting, the sources of fund-ing and support for their research, and their scientificcapacity.Technology Transfer Experience Covariates. Priorwork reports mixed results about the effects of expe-rience with patenting. On the one hand, more experi-ence in the form of increasing patenting may help auniversity to better identify potentially commerciallyvaluable inventions and thus higher impact patents.On the other hand, experience may also be associatedwith declines in impact, because universities lookingto develop strong patent portfolios pursue more incre-mental inventions (Henderson et al. 1998). We measureeach university’s experience with technology transferas the volume of Patents applied for during year t. Wealso include a separate measure for Life sciences thatcaptures the proportion of recent patents applied forby each university that fall within the NBER’s Drugscategory.Industry Ties Covariates. Owen-Smith and Powell’s(2003) observation that greater corporate engagementleads to the capture of university technology transferefforts suggests that science supported by industrialpartners is likely to generate inventions that increasethe use of existing technologies. We thus propose thatindustry support of university R&D will lead universi-ties to pursue more consolidating discoveries. Industry-sponsored R&D (log) measures the dollar value, inmillions, of grants and contracts from corporationsto researchers on campus. This variable comes fromNational Science Foundation (NSF) surveys of campus-level R&D efforts available from the online WebCAS-PAR database.15 Our second measure, Industry R&D

ties, captures formal R&D relationships with firms intechnology-intensive industries. The information onR&D connections used to create this variable was con-tent coded from Securities and Exchange Commissionfilings made by a sample of 634 publicly traded firmsin high-technology sectors (see Buhr and Owen-Smith2010, Owen-Smith et al. 2015).Government Ties Covariates. Levels of support fromthe National Science Foundation (NSF), the NationalInstitutes of Health (NIH), and the Department ofDefense (DoD), the nation’s premier funders of basicscience, should yield research that is less connectedto existing industrial needs. Thus, more federal grantsand more patents derived from them should be associ-ated with destabilizing inventions. We track the level ofsupport (in millions of dollars) that flows to campusesfrom federal science agencies with three variables, NSFgrants (log), NIH grants (log), and DoD grants (log).Federal grant measures were extracted from the NSF’sWebCASPAR database.

Broadly speaking, if the process by which federal sci-ence agencies apportion R&D support is more likely tofund research focused on academic concerns divorcedfrom the needs of industry, then campuses that per-form more federally funded research and those whosepatents emerge from federally funded projects shouldproduce more destabilizing patents. Likewise, becausecorporate R&D funding is presumably linked to spon-sors’ proximate, market-driven goals, we expect cam-puses that pursue more industrially funded R&D toproduce patents that consolidate the use of existingtechnologies. Although any source of external supportseems likely to yield higher-impact patents, we antici-pate that industrial connections will be associated withconsolidating inventions, while public sector fundingwill lead to more destabilizing ones.Scientific Capacity Covariates. Although existingmeasures of publication impact suffer from many ofthe same limitations as the patent citation indexes wecritique above, we use them to provide some senseof the relationship between high-impact public andimportant proprietary science. We follow two lines ofreasoning. Noting first with Sine and colleagues (2003)that articles confer a “halo effect” on associated aca-demic patents while advertising their value to potentiallicensees, we expect both the volume and impact ofacademic science to be associated with higher-impactpatent portfolios. To the extent that highly-cited papersgenerate scientific attention because they report novel,basic-science discoveries that are far removed from thecurrent concerns of industry, we would expect higher-impact publications to be associated with more desta-bilizing inventions.

We use the count of Scientific articles published inthe preceding year to index the volume of academic

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



research on campus. In addition, we include a measureof the citation Impact factor of those articles as a proxyof the visibility and quality of a campus’s publishedresearch. Both measures are drawn from the Institutefor Scientific Information’s (ISI) University IndicatorsDatabase.Statistical Approach. Four of our measures of patentimportance—V total, Itotal


5 —arecounts that take on only nonnegative integer values.To capture time-invariant heterogeneity among uni-versities, account for overdispersion, and match theapproach taken in earlier research (Owen-Smith andPowell 2003), we model Itotal


5 usinga conditional fixed-effects negative binomial specifica-tion (Hausman et al. 1984). Our model of V total alsouses a negative binomial specification, but because itincludes a lagged dependent variable, we use randominstead of fixed effects.16 Finally, to model CDmean

5 andmCDscale

5 , we use a fixed-effects OLS specification. Allmodels include indicator variables for calendar year.3.3.5. Organization-Level Results. Table 6 presentsdescriptive statistics and correlations for all variablesused in these analyses. Table 7 reports results from afully specified model for each one of our dependentvariables. Nested specifications are available from theauthors.

Consider the first two columns of Table 7 (Models 17and 18), which present estimates for patent volume(V total) and impact (Itotal

5 ). The results of these modelsare broadly consistent with prior findings. Universi-ties that have greater prior experience with patent-ing (Patents) produce more and higher impact patents.A greater focus on Life sciences patenting, however, isassociated with diminished patent impact. This maybe the case because, unlike physical science and engi-neering technologies, the most sought-after life sciencepatents are those that offer exclusive protection forpotential drug candidates. Such patents are typically

Table 6. Descriptive Statistics and Correlations

Variable Mean SD 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1. Patents/volume (V total) 28.85 27.76 1.002. Impact (Itotal

5 ) 107.93 164.16 0.74 1.003. CDmean

5 0.10 0.12 �0.06 0.02 1.004. mCDscale

5 10.94 19.34 0.56 0.81 0.30 1.005. CDtotal+

5 59.39 99.31 0.69 0.95 0.07 0.83 1.006. CDtotal�

5 43.30 70.18 0.68 0.90 �0.04 0.66 0.74 1.007. Life sciences 0.48 0.21 �0.05 �0.17 0.15 �0.04 �0.20 �0.11 1.008. Industry sponsored 2.30 0.89 0.44 0.30 �0.10 0.20 0.27 0.29 0.02 1.00

R&D (log)9. Industry R&D ties 0.34 0.75 0.38 0.26 0.02 0.24 0.24 0.23 0.15 0.19 1.00

10. NSF grants (log) 0.65 1.27 0.42 0.32 �0.10 0.22 0.32 0.27 �0.41 0.29 0.16 1.0011. NIH grants (log) 0.67 1.32 0.44 0.20 0.04 0.23 0.18 0.20 0.49 0.38 0.35 0.11 1.0012. DoD grants (log) 0.63 1.42 0.43 0.34 �0.09 0.25 0.33 0.30 �0.28 0.44 0.22 0.63 0.22 1.0013. Scientific articles 1,919.69 1,179.77 0.62 0.38 �0.01 0.37 0.36 0.35 0.10 0.52 0.48 0.51 0.65 0.47 1.0014. Impact factor 21.87 10.12 0.15 0.33 0.30 0.47 0.33 0.27 0.30 �0.11 0.17 �0.27 0.25 �0.16 0.10 1.00

licensed exclusively and do not rely on suites of com-plementary intellectual property for their application.

Perhaps unsurprisingly, universities that producemore Scientific articles and higher-impact (Impact factor)research publications also develop more and higher-impact patents. Connections to corporate partners inthe form of Industry-sponsored R&D (log) and research-based university–industry alliances (Industry R&D ties)suggest more interesting associations. Increased indus-trial funding is modestly associated with more andhigher-impact patenting, but campuses with more cor-porate alliances patent less, and this type of corporateconnection has no significant relationship with patentimpact.

By the same token, increases in research fundingfrom key public sources (NSF grants (log), NIH grants(log), and DoD grants (log)) demonstrate largely pos-itive relationships with patent volume and impact.The cross agency funding differences in these mod-els warrant further scrutiny, but we expect that theyhave much to do with the relative proportion of eachfunding source in a university’s portfolio of researchsupport and the distinctive fields in which differentagencies concentrate their grant making.

Model 19 regresses these same variables on theaverage CD5 index (CDmean

5 ) for a given university.Just one variable, the impact factor of published sci-entific articles, offers any explanatory purchase onthis dependent variable, and that effect is marginal.Although the suggestion that higher-visibility scien-tific portfolios are associated with more destabiliz-ing patents is in accord with some of the patent-levelfindings we present in the prior session, the lack ofrobust relationships in Model 19 implies that an aver-age CD5 index might not be the best aggregate mea-sure for analyzing the kinds of diverse, decentralizedpatent portfolios that are commonplace on universitycampuses.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



Table 7. Regression Models of University Patenting

Model 17 Model 18 Model 19 Model 20 Model 21 Model 22V total Itotal

5 CDmean5 mCDscale

5 CDtotal+5 CDtotal�

5

Technology transfer experiencePatents 0.007⇤⇤⇤ 0.007⇤⇤⇤ �0.000 �0.219⇤⇤ 0.007⇤⇤⇤ 0.010⇤⇤⇤

(0.001) (0.001) (0.000) (0.065) (0.001) (0.002)Life sciences �0.177 �0.543⇤ 0.015 �8.831⇤ �0.490⇤ �0.777⇤⇤

(0.127) (0.214) (0.068) (4.364) (0.242) (0.274)Industry ties

Industry-sponsored R&D (log) 0.051† 0.118⇤ 0.019 �3.540⇤ 0.166⇤⇤ 0.090(0.029) (0.050) (0.012) (1.520) (0.053) (0.063)

Industry R&D ties �0.032⇤⇤ �0.005 0.001 �2.051⇤ 0.006 �0.027(0.012) (0.023) (0.005) (0.904) (0.028) (0.035)

Government tiesNSF grants (log)(centered) 0.061⇤ �0.024 0.006 3.163† 0.035 �0.057

(0.026) (0.043) (0.011) (1.710) (0.048) (0.054)NIH grants (log)(centered) 0.167⇤⇤⇤ 0.131⇤ 0.016 4.351† 0.212⇤⇤ 0.103

(0.039) (0.061) (0.025) (2.376) (0.067) (0.071)DoD grants (log)(centered) 0.019 0.093⇤⇤ 0.002 2.642⇤⇤ 0.047 0.100⇤

(0.016) (0.033) (0.007) (0.786) (0.034) (0.043)Scientific capacity

Scientific articles 0.000⇤⇤⇤ 0.000† �0.000 �0.021⇤⇤⇤ 0.000 0.000⇤

(0.000) (0.000) (0.000) (0.006) (0.000) (0.000)Impact factor 0.018⇤⇤⇤ 0.028⇤⇤⇤ 0.002† 0.902⇤⇤ 0.028⇤⇤⇤ 0.009

(0.003) (0.006) (0.001) (0.274) (0.006) (0.007)University fixed effects No Yes Yes Yes Yes YesYear fixed effects Yes Yes Yes Yes Yes YesConstant 1.714⇤⇤⇤ �19.452 0.029 39.160⇤ �18.873 �20.341

(0.142) (558.273) (0.070) (15.651) (427.032) (909.797)N 1,270 1,265 1,006 1,006 1,265 1,265R2 0.136 0.384Universities 109 108 108 108 108 108Log-likelihood �4,309.140 �4,239.365 870.336 �3,876.887 �3,707.129 �3,465.603

Note. Standard errors are in parentheses.†p < 0.1; ⇤p < 0.05; ⇤⇤p < 0.01; ⇤⇤⇤p < 0.001.

In contrast, Model 20 offers what we believe tobe the clearest and most compelling example of anaggregate version of our measures. mCDscale

5 weightsa standard forward citation measure of impact for agroup of patents by the CDmean

5 index. Comparing theseresults with those in Model 18 is thus illustrative ofthe benefits to be gained by incorporating the kindsof structural measures we propose into the analysisof portfolios of patents. Note first that, although highvolume patenting (Patents) and publishing (Scientificarticles) are both associated with increases in patentimpact (Model 18), the same measures are associatedwith higher-impact patents that consolidate the statusquo of their technology streams (Model 20). Univer-sities that produce higher-impact portfolios of articles(Impact factor), however, also develop higher-impactpatents that do more to destabilize the status quo.

Perhaps most tellingly, however, both measuresof engagement with corporate partners (Industry-sponsored R&D (log) and Industry R&D ties) are stronglyassociated with patenting that consolidates existing

technology trajectories. In contrast, increases in fund-ing from all three sources of public research support(NSF grants (log), NIH grants (log), and DoD grants(log)) are linked to more destabilizing patent port-folios. This analysis holds potential lessons for bothdefenders of and detractors from Bayh–Dole. For theformer, the implications are clear: the post–Bayh–Dolewave of university inventions depends on the founda-tion of academically important, publicly funded sci-ence. Thus, to preserve their economic contributions,the academic character of university science should beencouraged. For the latter, it appears that tighter indus-trial connections yield patents that reinforce existingtechnical arrangements, a finding that should deepenconcerns about the long-term effects of commercializa-tion on universities, their research, and ultimately theireconomic contributions.

Finally, turn your attention to Models 21 and 22,which examine variables that take a different approachto aggregation. Here we use the patent-level CD5 indexto differentiate between patents that destabilize or con-solidate the use of existing technologies. We then calcu-

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



late the forward citation impact for those patents. Bothmodels offer less clear signal than Model 20, but someinteresting features are apparent. Most notably, thecoefficients for variables that capture scientific capac-ity (Scientific articles and Impact factor) in these twomodels are consistent with the findings we reportedfor our mCDscale

5 index. A higher volume of scientificpapers is associated with higher-impact consolidatinginventions (Model 22, CDtotal�

5 ). In contrast, a higher-impact portfolio of scientific publications is associatedwith a more broadly used portfolio of patents thatdestabilize the use of the technologies on which theybuild (Model 21, CDtotal+

5 ).When considered holistically, these models suggest

that there is value in pursuing more aggregate formsof our measures, but more research is needed. To oureyes, the admittedly schematic findings presented inTable 7 imply that, for organizations like universitiesthat pursue research across many fields with multi-ple goals and partners, an integrated measure such asmCDscale

5 offers significant purchase. We suspect that thekinds of measures we created for Models 21 and 22 maybe more adapted to studying the effects of changes intechnology management and R&D practices for organi-zations, such as smaller firms, that pursue more explic-itly destabilizing or consolidating invention goals ina more homogeneous range of fields. Regardless, theadditional analytic flexibility that comes with ournetwork-based measurement approach may yield div-idends when applied to more specific examinations oftechnology strategy and its effects.

4. DiscussionWe began this article with a quote from Schumpeter.That insight served as a stepping stone for two gen-erations of research that views creative destructionas a motor for economic and social transformation.Although Schumpeter and others observe that whatmakes a technology important is the nature and extentof its use, quantitative evaluations of how, precisely, theuse of new technologies strengthens or challenges thestatus quo have remained elusive. The purpose of thisarticle has been to propose and evaluate measures—the CDt and mCDt indexes—that capture the effectsthat new inventions have on the use of their predeces-sors. Our approach differs from prior efforts by oper-ationalizing the distinction between technologies thatare valuable because they break from current stan-dards and those whose value derives from reinforcingthe trajectories from which they spring.

The CDt index has several attractive features thatmay make it useful for future research. By treating newtechnologies as additions to evolving networks, themeasure makes use of theoretically important struc-tural information on second-order forms of impact

that are overlooked by existing approaches. These net-work underpinnings also make the CDt index natu-rally suited to dynamic analyses over the course ofa particular technology’s existence. Our method alsooffers a continuous and valenced characterization ofan invention’s impact, which allows for clear, qualita-tive distinctions while also recognizing that the con-solidating or destabilizing character of technologies isoften a matter of degree. Moreover, because the valueof the CDt index can vary over time for a single patent,the measure allows for the possibility that some impor-tant proportion of the effect that a new invention hasis determined ex post in the context of its use ratherthan ex ante in the context of its discovery. Finally, byincorporating an impact weight into the CDt index, itis possible to obtain an indicator (the mCDt index) thatcharacterizes both technologies that are major depar-tures from the status quo and those that reinforce butoffer substantial improvements over existing technolo-gies. Together, these indexes capture the positive andnegative dimensions of technological change.

In addition to their methodological benefits, the CDtand mCDt indexes may also help to facilitate theoreti-cal development in some areas by clarifying conflictingempirical findings. For more than three decades, schol-ars have engaged in heated debates over the impli-cations of increasing commercial engagement for thenature and quality of university science. Althoughpatents are a potentially valuable source of data foradjudicating among competing views, empirical inves-tigations using forward-citation-based measures ofimpact have been inconclusive. Using the CDt andmCDt indexes, we found consistent evidence that whileincreases in federal support for academic researchappear to push universities to create technologies thatdestabilize the status quo, increases in commercial tiesare associated with university research that consoli-dates existing technology streams.

The indexes we propose are not without limita-tions and more remains to be done. Most broadly,our approach rests on the assumption that relevantnodes (patents) and ties (citations) can be meaningfullyidentified, and that the network under investigationevolves in some way. Many settings in which inven-tion occurs meet these criteria. However, in certaincases (for instance, where change is largely propelledby exogenous influences), the value of the approachwill be limited. Our indexes are also limited becausethey are undefined when a focal invention and its pre-decessors are never cited by subsequent technologies.In the context of U.S. patents, we found that thesekind of inventions are rare, occurring in 3% of caseswhen the CDt and mCDt indexes are measured fiveyears after the focal patent’s issue year. Nevertheless,inventions of this sort may still influence technologi-cal change (for instance, by leading inventors to believe

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



a particular field has been saturated) and thereforediverting attention from components that were in exis-tence before its introduction. A related limitation isthat, even when our indexes are defined, they maybe biased for inventions that offer improvements oversome existing technologies but use entirely new meth-ods and therefore do not build on those technologies,perhaps by using methods from a different domain.Clarifying these effects, both conceptually and withrespect to measurement, is an exciting opportunity forfuture research.

Our analyses did not explore the possibility thatsome inventions might simultaneously consolidate anddestabilize the use of predecessors in different technol-ogy streams. This phenomenon could occur if a par-ticular invention made its antecedents more valuablefor one community of users, but less so for others.Such patterns of alternating consolidation and desta-bilization may prove useful in identifying instanceswhere a technology fails to cite the predecessors itmost strongly destabilizes. When a new technologyis based on a totally different set of mechanisms,for instance, we might not see destabilization on ourmeasure for several patent generations. Our exampleof the Axel patent, which cited no examples of thechemistry-based drug discovery methods it eventu-ally supplanted, offers a case in point. The same effectmight also be present in cases where usage patternschange over the course of a technology’s life cycle.

Finally, our empirical analyses suffer from some lim-itations common to all research that relies on archivaldata like patents. For instance, many important inven-tions are never filed with the USPTO, although, ascritics suggest, many mediocre new technologies passthrough the examination process and are awardedpatent protection. Business and legal strategies maycreate additional noise or bias in the data through theintentional omission of legitimate citations or the inclu-sion of irrelevant references (Alcácer et al. 2009).

Although not without limitations, we believe thatthe CDt and mCDt indexes may help to expandresearch in multiple areas. Below, we address some im-plications of our approach where it is likely to be mostuseful.

4.1. Social NetworksThe indexes we propose may help to generate newinsights in research on social networks and invention(Burt 2004, Obstfeld 2005). A persistent challenge inthis area has been to distinguish between the effectsof more open, unconstrained social network configu-rations and more closed or cohesive structures. Somecurrent findings suggest that the former are more effec-tive for generating novel ideas whereas the latter aremore suited for mobilizing collaborators around theoften difficult work of development, but other research

points to the importance of more complex, hybridnetwork arrangements (Fleming et al. 2007, de Vaanet al. 2015). Rather than reflecting differences in theeffects of particular social network structures and theneed to consider more complex configurations, how-ever, the conflicting findings in this area may stemfrom noisy outcome indexes that mask fundamentaldifferences between network positions that facilitatedestabilizing ideas and those that support inventionsthat are more consolidating in nature. Closed groupsof inventors with many overlapping, cohesive rela-tionships, for instance, may collectively have deeperknowledge of a technological domain and thereforebe especially suited for developing ideas that consol-idate a trajectory. Alternatively, inventors who searchthrough networks of otherwise disconnected contactsmay leverage their vision advantage to pursue morefundamental insights that destabilize existing techno-logical arrangements.

4.2. Industries and FirmsThe CDt and mCDt indexes may also be useful forunderstanding industry evolution and its implicationsfor firms (Nelson and Winter 1982; Tushman andAnderson 1986; Benner 2007; Sosa 2009, 2011). Forinstance, recall our brief discussion of two related tech-nologies identified in Table 2: the scanning tunnelingmicroscope (STM) and the atomic force microscope(AFM). The STM was highly destabilizing because ofits major break with the technological standards inplace at the time of its introduction. By contrast, theAFM was only moderately destabilizing because itbuilt on and expanded the technological base createdby the STM. Yet the AFM has proven to be more impor-tant to the development of nanotechnology. Abstract-ing from this dynamic, one may hypothesize and useour indexes in tandem with qualitative and historicalcase data to examine the possibility that, in emergingsectors like nanoscale materials or tissue engineering,early inventions will be the most likely to challengethe incumbents’ capabilities, whereas later discoveriesmay enhance them as the technologies (and relatedorganizations, networks, and markets) mature.

The ability to distinguish between inventions thatconsolidate and those that destabilize the use of exist-ing technologies may also be valuable for examininga range of questions of relevance to technology strat-egy. For instance, consider debates over the effect ofmarket power on incentives to develop new technolo-gies. Much like research on social networks, this liter-ature has produced conflicting theoretical argumentsand empirical findings (Gilbert 2006, Ahuja et al. 2008).Although some contend that incumbents should pur-sue inventions in order to maintain their positions, oth-ers argue the opposite and claim that incentives aredepressed due to lack of competition. Ahuja and col-leagues (2008, p. 8) note that empirical efforts to resolve

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



this debate have been challenging, “since the maineffects are countervailing.” To the extent that incum-bents may differ with respect to the aims of their R&Defforts in relation to existing technology streams, themeasures we propose could help to disentangle someof these countervailing effects.

4.3. Science PolicyIn recent years, efforts to develop evidence-based sci-ence policy measures have triggered extensive workin what has come to be called the “science of sciencepolicy” (Lane 2010, Fealing et al. 2011, Powell et al.2012). Among other concerns in this burgeoning fieldis the question of how to support, identify, and eval-uate transformative research. We submit that variantsof the CDt and mCDt indexes might help to documentthe transformative effects of different collaborative orfunding models by, for instance, associating differinglevels of interdisciplinarity or scientific team structurewith the production of more or fewer findings thatenhance or replace existing knowledge bases. Our find-ings about the differing implication of higher-impactuniversity and corporate patents for established tech-nological arrangements and the tendency of federallyfunded grants to generate more destabilizing inven-tions offer cases in point. Funders and performers ofsuch research are under increasing political pressureto explain and justify their work in terms of its eco-nomic impact. However, findings like those we explorehere suggest the importance of more fundamental, aca-demic research as a means to open new technologicaltrajectories. As such, the CDt and mCDt indexes mayoffer a more effective means to evaluate the technologi-cal significance of scientific discoveries in a fashion thatjustifies novel funding approaches at the federal andstate levels. The findings we present in regard to thedisparate effects that industrial and public R&D find-ings may have for different types of academic patent-ing, and the suggestion that support from differentagencies is associated with different types of researchoutcomes, offer cases in point.

4.4. Geography and SpaceThe method we propose could also be of use in ongo-ing work on the causes and consequences of regionalindustrial agglomeration. Competition is stiffer withinestablished clusters than outside them (Stuart andSorenson 2003, Whittington et al. 2009). Work has doc-umented the different roles that physical proximity touniversities and other firms and geographically localversus distant ties play in the inventive performanceof biotechnology companies (Owen-Smith and Powell2004, Whittington et al. 2009). Both lines of work havesuggested that location in vibrant regions forces firmsto compete more vigorously to succeed, but researchin this area has not attempted to distinguish among

the types of inventions made by geographically clus-tered and geographically remote organizations. Adapt-ing insights on social isolation and creativity (Taylorand Greve 2006, Singh and Fleming 2010), one may pre-dict that firms located in “second-tier” regions (Mayer2011) or outside established clusters entirely are bet-ter positioned to generate inventions that destabilizeexisting streams of technology, whereas those in closerproximity to industry peers are well suited to work thatconsolidates and reinforces them. The ability to char-acterize regions along these dimensions may be valu-able for managers and urban planners, in addition toresearchers (Bettencourt et al. 2007).

At a more microlevel, our finding of a greater ten-dency for distributed teams of inventors to createinventions that consolidate the status quo suggests thatgeographic diversity may help collaborators to per-form better in some respects while also challengingtheir ability to create truly destabilizing new technolo-gies. Existing studies of team-based research suggestthat this could be the case, but pure impact mea-sures lack the ability to distinguish among such claims.If, as some research on scientific and technical teamssuggests, distance increases coordination challenges(Cummings and Kiesler 2005), then it may be thatdispersed teams are more effective in pursuing addi-tions to existing technological trajectories. In contrast,if serendipity is an important contributing factor fordestabilizing discoveries and face-to-face interactionsare a key mechanism for its recognition, then dispersedteams may be at a substantial disadvantage when itcomes to making more destabilizing inventions (Kaboet al. 2014).

4.5. Beyond Technological ChangeFinally, consider some possible uses of our dynamicnetwork approach far afield from technologicalchange. The CDt and mCDt indexes might be adaptedto studies of the evolving link structure of the WorldWide Web, where new pages will cite and be citedby others. Similarly the evolving structure of citationsamong judicial decisions that comprise an importantaspect of U.S. law might be examined in these terms.Rather than measuring the effects of new technologies,this family of indexes might offer new insight into morepolitical and cultural questions involving the dynamicsof political polarization in the blogosphere, the evo-lution of management fads and fashions, the visibil-ity of music artists and genres, the commercial andcritical impact of films, and the larger legal implica-tions of new court decisions. Evolving network data onrelationships of deference among entrants and incum-bents in different substantive domains are becomingmore available and are potentially valuable for researchquestions in multiple fields.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.



AcknowledgmentsThe authors thank Lee Fleming and the editors and review-ers at Management Science; Woody Powell; Jeannette Coly-vas; Brian Wu; John Chen; Ned Smith; Mark Suchman; JiyaoChen; Mary Benner; Andy Van de Ven; members of theQuantitative Methods Program; Corporate Strategy; and theEconomic Sociology workshops at the University of Michi-gan for helpful feedback; as well as audiences at the BrownUniversity Commerce, Organizations and Markets Seminar;the 2012 International Conference on Network Science inEvanston, Illinois; the 2012 Annual Meeting of the Academyof Management; the 2012 Cyberinfrastructure Days Confer-ence held at the University of Michigan; the ComputationInstitute at the University of Chicago; and the Strategic Man-agement and Entrepreneurship Seminar at the Universityof Minnesota. Bhaven Sampat generously provided data onUSPTO examiners. All errors are the authors’ own.

Endnotes1 Nevertheless, patent data have limitations. For example, inventorsmay not seek patent protection for their ideas. Moreover, the patentexamination process has been criticized by observers who argue thatthe USPTO grants many patents that fail to meet the legal require-ments of novelty and nonobviousness (Jaffe and Lerner 2004, Lemleyand Sampat 2008). Firms may have strategic reasons for omittingcitations, and examiners may fail to add these during their reviews(Alcácer et al. 2009).2 Although we do not include them in these diagrams, we also recog-nize and integrate into our measure the likelihood that later inven-tions will cite the focal patent’s predecessors without citing the focalpatent itself.3 Code for calculating the CDt and mCDt indexes along with valuesfor U.S. patents are available at http://www.cdindex.info.4 More nuanced approaches could, for example, determine the val-ues of W according to the issue date of i such that older citationshave greater or lesser influence over the measure, or by differentiallyweighting citations made by examiners and patent applicants.5 Our focus on types of nodes and binary relations among them dis-cards some structural data. We address these issues more in OnlineAppendix A, where we introduce variations on the CDt and mCDtindexes that allow for multiple focal patents.6 We identified government interest patents by extracting and codinginformation contained in the GOVT, GOVINT, and related fields of full-text patent documents, available from the USPTO.7 In unreported comparisons, we found few substantial differencesin the distribution of the CD5 index across broad (NBER) technologycategories. Computers patents (including software), however, appearto have less influence in terms of either consolidating or destabilizingthe status quo than other classes have.8 By convention, we refer to these patents using the last names oftheir authors. The inventors’ full names are, respectively, Stanley N.Cohen and Herbert W. Boyer, Richard Axel (with Michael H. Wiglerand Saul J. Silverstein), and Kary B. Mullis. Each of these patentscovers what was to become a fundamental method for research inmany fields of biology and the biotechnology industry. Cohen andBoyer established the fundamental technique of gene splicing. Axeland colleagues developed a method to insert genes into cells inorder to turn them into protein “factories.” Mullis invented an effi-cient method to produce large amounts of DNA from very smallsamples.9 The alternative version of the CDt index presented in OnlineAppendix A may prove useful for estimating the effects that patentfamilies have on their technological predecessors.

10 This calculation does not account for possible right truncation dueto delays between patent application and grant years and may alsobe inflated because of increases in the frequency of patenting andcitation over time. In Online Appendix B, we present exploratorymodels that attempt to adjust for these and other factors. The resultsare similar.11 Patent 5,084,082 is owned by DuPont.12 The PageRank patent also differs from Monsanto in that its back-ward citations come from a more diverse base of organizations,as would be anticipated for a more destabilizing invention. The sevenpredecessors cited by the patent are owned by a total of four differ-ent corporations and one university, none of which are Stanford orGoogle.13 One important question is whether or not the Axel patent’s highCD5 index of 0.95 could stem from the fact that it only cited twopieces of predecessor patents. There is a modest negative correlation(r ⇤�0.17, p < 0.001) between the number of citations of predecessorsand a patent’s CD5 index. However, some association between thenumber of predecessors cited and the CD5 index is anticipated asnovel inventions should have fewer predecessors available on whichto build.14 We thank Bhaven Sampat for sharing the data on USPTO examin-ers that made these analyses possible.15 http://ncsesdata.nsf.gov/webcaspar/.16 The results are similar if we drop the lagged term (Patents) andestimate the model using conditional fixed effects.

ReferencesAbernathy WJ, Clark KB (1985) Innovation: Mapping the winds of

creative destruction. Res. Policy 14(1):3–22.Ahuja G, Lampert CM, Tandon V (2008) Moving beyond Schum-

peter: Management research on the determinants of technologi-cal innovation. Acad. Management Ann. 2(1):1–98.

Alcácer J, Gittelman M, Sampat B (2009) Applicant and examinercitations in U.S. patents: An overview and analysis. Res. Policy38(2):415–427.

Arthur WB (2007) The structure of invention. Res. Policy 36(2):274–287.

Arthur WB (2009) The Nature of Technology: What It Is and How ItEvolves (Free Press, New York).

Azoulay P, Ding W, Stuart T (2007) The determinants of facultypatenting behavior: Demographics or opportunities? J. Econom.Behavior Organ. 63(4):599–623.

Benner M (2007) The incumbent discount: Stock market categoriesand response to radical technological change. Acad. Manage-ment J. 32(3):73–720.

Bettencourt LM, Lobo J, Helbing D, Kühnert C, West GB (2007)Growth, innovation, scaling, and the pace of life in cities. Proc.Natl. Acad. Sci. USA 104(17):7301–7306.

Bresnahan T, Greenstein S (1996) Technical progress and co-inventionin computing and the uses of computers. Brookings Papers onEconomic Activity: Microeconomics, Washington, DC, 1–83.

Brin S, Page L (1998) The anatomy of a large-scale hypertextual websearch engine. Comput. Networks ISDN Systems 30(1–7):107–117.

Buhr H, Owen-Smith J (2010) Networks as institutional support: Lawfirm and venture capital relations and regional diversity in high-technology IPOs. Res. Sociol. Work 21:95–126.

Burt RS (2004) Structural holes and good ideas. Amer. J. Sociol.110(2):349–399.

Christensen CM (1997) The Innovator’s Dilemma (Harvard BusinessSchool Press, Boston).

Colaianni A, Cook-Deegan R (2009) Columbia University’s Axelpatents: Technology transfer and implications for the Bayh-Doleact. Milbank Quart. 87(3):683–715.

Cole JR (1993) Balancing acts: Dilemmas of choice facing researchuniversities. Daedalus 122(4):1–36.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.

http://www.cdindex.info

http://ncsesdata.nsf.gov/webcaspar/



Cummings JN, Kiesler S (2005) Collaborative research across dis-ciplinary and organizational boundaries. Soc. Stud. Sci. 35(5):703–722.

Dasgupta P, David PA (1994) Toward a new economics of science.Res. Policy 23(5):487–521.

David PA (1990) The dynamo and the computer: An historical per-spective on the modern productivity paradox. Amer. Econom.Rev. 80(2):355–361.

de Vaan M, Stark D, Vedres B (2015) Game changer: The topology ofcreativity. Amer. J. Sociol. 120(4):1144–1194.

Dosi G (1982) Technological paradigms and technological trajecto-ries: A suggested interpretation of the determinants and direc-tions of technical change. Res. Policy 11(3):147–162.

Economist, The (2002) Innovation’s golden goose. (December 14):3.Enos JL (1958) A measure of the rate of technological progress in the

petroleum refining industry. J. Indust. Econom. 6(3):180–197.Enos JL (1962) Invention and innovation in the petroleum refin-

ing industry. Richard RN, ed. The Rate and Direction of InventiveActivity (Princeton University Press, Princeton NJ), 299–322.

Fealing KH, Lane JI, Marburger JH III, Shipp SS (2011) The Science ofScience Policy (Stanford Business Books, Stanford, CA).

Feldman MP, Yoon JW (2012) An empirical test for general purposetechnology: An examination of the Cohen—Boyer rDNA tech-nology. Indust. Corporate Change 21(2):249–275.

Fleming L (2001) Recombinant uncertainty in technological search.Management Sci. 47(1):117–132.

Fleming L, Sorenson O (2004) Science as a map in technologicalsearch. Strategic Management J. 25(8–9):909–928.

Fleming L, Mingo S, Chen D (2007) Collaborative brokerage, gen-erative creativity, and creative success. Admin. Sci. Quart. 52(3):443–475.

Fore J, Wiechers IR, Cook-Deegan R (2006) The effects of businesspractices, licensing, and intellectual property on developmentand dissemination of the polymerase chain reaction: Case study.J. Biomedical Discovery Collaboration 1:7.

Funk RJ (2014) Making the most of where you are: Geography, net-works, and innovation in organizations. Acad. Management J.57(1):193–222.

Gambardella A (1995) Science and Innovation: The U.S. Pharmaceu-tical Industry During the 1980s (Cambridge University Press,Cambridge, UK).

Gilbert R (2006) Looking for Mr. Schumpeter: Where are we in thecompetition-innovation debate? Jaffe AB, Lerner J, Stern S, eds.Innovation Policy and the Economy Vol. 6 (MIT Press, Cambridge,MA), 159–215.

Goldfarb B (2005) Diffusion of general-purpose technologies: Under-standing patterns in the electrification of U.S. manufacturing1880–1930. Indust. Corporate Change 14(5):745–773.

Graff GD, Rausser GC, Small AA (2003) Agricultural biotechnology’scomplementary intellectual assets. Rev. Econom. Statist. 85(2):349–363.

Greenberg DS (2007) Science for Sale (University of Chicago Press,Chicago).

Griliches Z (1990) Patent statistics as economic indicators: A survey.J. Econom. Lit. 28:1661–1707.

Hall BH, Trajtenberg M (2004) Uncovering GPTs with patent data.NBER Working Paper 10901, National Bureau of EconomicResearch, Cambridge, MA.

Hall BH, Jaffe AB, Trajtenberg M (2002) The NBER patent-citationsdata file: Lessons, insights, and methodological tools. Jaffe AB,Trajtenberg M, eds. Patents, Citations and Innovations (MIT Press,Cambridge, MA), 403–460.

Harhoff D, Narin F, Scherer FM, Vopel K (1999) Citation frequencyand the value of patented inventions. Rev. Econom. Statist.81:511–515.

Hausman J, Hall BH, Griliches Z (1984) Econometric models forcount data with an application to the patents-R&D relationship.Econometrica 52(4):909–938.

Henderson R, Cockburn I (1996) Scale, scope, and spillovers: Thedeterminants of research productivity in drug discovery. RANDJ. Econom. 27(1):32–59.

Henderson R, Jaffe AB, Trajtenberg M (1998) Universities as a sourceof commercial technology: A detailed analysis of universitypatenting, 1965–1988. Rev. Econom. Statist. 80(1):119–127.

Hughes SS (2001) Making dollars out of DNA: The first major patentin biotechnology and the commercialization of molecular biol-ogy, 1974–1980. Isis 92(3):541–575.

Jaffe AB, Lerner J (2004) Innovation and Its Discontents (Princeton Uni-versity Press, Princeton, NJ).

Jaffe AB, Trajtenberg M (2002) Patents, Citations, and Innova-tions: A Window on the Knowledge Economy (MIT Press, Cam-bridge, MA).

Kabo FW, Cotton-Nessler N, Hwang Y, Levenstein MC, Owen-SmithJ (2014) Proximity effects on the dynamics and outcomes of sci-entific collaborations. Res. Policy 43(9):1469–1485.

Kaplan S, Murray F, Henderson R (2003) Discontinuities and seniormanagement: Assessing the role of recognition in pharmaceu-tical firm response to biotechnology. Indust. Corporate Change12(2):203–233.

Krimsky S (2004) Science in the Private Interest (Rowman andLittlefield Publishers, Lanham, MD).

Lane J (2010) Let’s make science metrics more scientific. Nature464(7288):488–489.

Lanjouw JO, Schankerman M (2004) Patent quality and researchproductivity: Measuring innovation with multiple indicators.Econom. J. 114(495):441–465.

Lemley MA, Sampat B (2008) Is the patent office a rubber stamp?Emory Law J. 58:181–206.

Lemley MA, Sampat B (2012) Examiner characteristics and patentoffice outcomes. Rev. Econom. Statist. 94(3):817–827.

Li G-C, Lai R, D’Amour A, Doolin DM, Sun Y, Torvik VI, Yu AZ,Fleming L (2014) Disambiguation and co-authorship networksof the U.S. patent inventor database (1975–2010). Res. Policy43(6):941–955.

Lucas HC, Goh JM (2009) Disruptive technology: How Kodak missedthe digital photography revolution. J. Strategic Inform. Systems18(1):46–55.

Mayer H (2011) Entrepreneurship and Innovation in Second Tier Regions(Edward Elgar, Northampton, MA).

Mehta A, Rysman M, Simcoe T (2010) Identifying the age profile ofpatent citations: New estimates of knowledge diffusion. J. Appl.Econometrics 25(7):1179–1204.

Mokyr J (1990) Punctuated equilibria and technological progress.Amer. Econom. Rev. 80(2):350–354.

Mowery DC, Ziedonis AA (2002) Academic patent quality and quan-tity before and after the Bayh–Dole act in the United States. Res.Policy 31(3):399–418.

Mowery DC, Sampat BN, Ziedonis AA (2002) Learning to patent:Institutional experience, learning, and the characteristics of U.S.university patents after the Bayh–Dole Act, 1981-1992. Manage-ment Sci. 48(1):73–89.

NelsonRR,WinterSG(1982)AnEvolutionaryTheoryofEconomicChange(Belknap Press of Harvard University Press, Cambridge, MA).

Obstfeld D (2005) Social networks, the tertius iungens orienta-tion, and involvement in innovation. Admin. Sci. Quart. 50(1):100–130.

Owen-Smith J (2003) From separate systems to a hybrid order:Accumulative advantage across public and private science atResearch One universities. Res. Policy 32(6):1081–1104.

Owen-Smith J, Powell WW (2003) The expanding role of universitypatenting in the life sciences: Assessing the importance of expe-rience and connectivity. Res. Policy 32(9):1695–1711.

Owen-Smith J, Powell WW (2004) Knowledge networks as channelsand conduits: The effects of spillovers in the Boston biotechnol-ogy community. Organ. Sci. 15(1):5–21.

Owen-Smith J, Cotton-Nessler NC, Buhr H (2015) Network effectson organizational decision-making: Blended social mechanismsand IPO withdrawal. Soc. Networks 41:1–17.

Pollack A (2009) As patent ends, a seed’s use will survive. NewYork Times (December 17), http://www.nytimes.com/2009/12/18/business/18seed.html.

Powell WW, Owen-Smith J (1998) Universities and the market forintellectual property in the life sciences. J. Policy Anal. Manage-ment 17(2):253–277.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.

http://www.nytimes.com/2009/12/18/business/18seed.html

http://www.nytimes.com/2009/12/18/business/18seed.html



Powell WW, Koput KW, Smith-Doerr L (1996) Interorganizationalcollaboration and the locus of innovation: Networks of learningin biotechnology. Admin. Sci. Quart. 41(1):116–145.

Powell WW, Owen-Smith J, Colyvas J (2007) Innovation and emu-lation: Lessons from American universities in selling privaterights to public knowledge. Minerva 45(2):121–142.

Powell WW, Packalen K, Whittington K (2012) Organizational andinstitutional genesis: The emergence of high-tech clusters in thelife sciences. Padgett JF, Powell WW, eds. The Emergence of Orga-nizations and Markets (Princeton University Press, Princeton, NJ),434–465.

Rosenberg N (1982) Inside the Black Box: Technology and Economics(Cambridge University Press, Cambridge, UK).

Rosenberg N, Nelson RR (1994) American universities and technicaladvance in industry. Res. Policy 23(3):323–348.

Schumpeter J (1942) Capitalism, Socialism and Democracy (Perennial,New York).

Sine WD, Shane S, Gregorio DD (2003) The halo effect and tech-nology licensing: The influence of institutional prestige onthe licensing of university inventions. Management Sci. 49(4):478–496.

Singh J, Fleming L (2010) Lone inventors as sources of break-throughs: Myth or reality? Management Sci. 56(1):41–56.

Schoenmakers W, Duysters G (2010) The technological origins ofradical inventions. Res. Policy 39(8):1051–1059.

Sosa ML (2009) Application-specific R&D capabilities and the advan-tage of incumbents: Evidence from the anticancer drug market.Management Sci. 55(8):1409–1422.

Sosa ML (2011) From old competence destruction to new competenceaccess: Evidence from the comparison of two discontinuities inanticancer drug discovery. Organ. Sci. 22(6):1500–1516.

Stephan P (2012) How Economics Shapes Science (Harvard UniversityPress, Cambridge, MA).

Stuart T, Sorenson O (2003) The geography of opportunity: Spatialheterogeneity in founding rates and the performance of biotech-nology firms. Res. Policy 32(2):229–253.

Taylor A, Greve HR (2006) Superman or the fantastic four? Knowl-edge combination and experience in innovative teams. Acad.Management J. 49(4):723–740.

Trajtenberg M (1990). A penny for your quotes: Patent citations andthe value of innovations. RAND J. Econom. 21(1):172–187.

Trajtenberg M, Henderson R, Jaffe AB (1997) University versus cor-porate patents: A window on the basicness of invention. Econom.Innovation New Tech. 5(1):19–50.

Tripsas M (1997) Unraveling the process of creative destruction:Complementary assets and incumbent survival in the typesetterindustry. Strategic Management J. 18(S1):119–142.

Tushman ML, Anderson P (1986) Technological discontinuities andorganizational environments. Admin. Sci. Quart. 31(3):439–465.

Verspagen B (2007) Mapping technological trajectories as patent cita-tion networks: A study on the history of fuel cell research. Adv.Complex Systems 10(1):93–115.

Washburn J (2005) University, Inc. (Basic Books, New York).Whittington KB, Owen-Smith J, Powell WW (2009) Networks,

propinquity, and innovation in knowledge-intensive industries.Admin. Sci. Quart. 54(1):90–122.

Wuchty S, Jones BF, Uzzi B (2007) The increasing dominance of teamsin production of knowledge. Science 316(5827):1036–1039.

Youtie J, Iacopette M, Graham S (2008) Assessing the nature ofnanotechnology: Can we uncover an emerging general purposetechnology? J. Tech. Transfer 33(3):315–329.

Dow

nloa

ded

from

info

rms.o

rg b

y [1

41.2

13.1

68.1

90] o

n 20

Aug

ust 2

017,

at 0

5:36

. Fo

r per

sona

l use

onl

y, a

ll rig

hts r

eser

ved.

a dynamic network measure of technological changerussellfunk.org/cdindex/static/funk_ms_2016.pdf ·...

Documents