Singh_fleming_2010_lone Inventors as Soruce of Innovation

8/12/2019 Singh_fleming_2010_lone Inventors as Soruce of Innovation

1/16

MANAGEMENT SCIENCEVol. 56, No. 1, January 2010, pp. 4156issn 0025-1909 eissn 1526-5501 10 5601 0041

informs

doi 10.1287/mnsc.1090.1072 2010 INFORMS

Lone Inventors as Sources of Breakthroughs:

Myth or Reality?Jasjit Singh

INSEAD, Singapore 138676, Singapore, [email protected]

Lee FlemingHarvard Business School, Harvard University, Boston, Massachusetts 02163, [email protected]

Are lone inventors more or less likely to invent breakthroughs? Recent research has attempted to resolve thisquestion by considering the variance of creative outcome distributions. It has implicitly assumed a sym-metric thickening or thinning of both tails, i.e., that a greater probability of breakthroughs comes at the cost ofa greater probability of failures. In contrast, we propose that collaboration can have opposite effects at the twoextremes: it reduces the probability of very poor outcomesbecause of more rigorous selection processeswhilesimultaneously increasing the probability of extremely successful outcomesbecause of greater recombinant

opportunity in creative search. Analysis of over half a million patented inventions supports these arguments:Individuals working alone, especially those without affiliation to organizations, are less likely to achieve break-throughs and more likely to invent particularly poor outcomes. Quantile regressions demonstrate that the effectis more than an upward mean shift. We find partial mediation of the effect of collaboration on extreme outcomes

by the diversity of technical experience of team members and by the size of team members external collabo-ration networks. Supporting our meta-argument for the importance of examining each tail of the distributionseparately, experience diversity helps trim poor outcomes significantly more than it helps create breakthroughs,relative to the effect of external networks.

Key words : creativity; collaboration; invention; innovation; teams; quantile; diversity; networksHistory : Received December 25, 2006; accepted July 21, 2009, by Christoph Loch, technological innovation

product development, and entrepreneurship. Published online in Articles in Advance October 16, 2009.

1. IntroductionOur species is the only creative species, and it hasonly one creative instrument, the individual mind, andspirit of a man. Nothing was ever created by two men.There are no good collaborations, whether in music,in art, in poetry, in mathematics, in philosophy. Oncethe miracle of creation has taken place, the group can

build and extend it, but the group never invents any-thing. The precociousness lies in the lonely mind of aman. (Steinbeck 1952, pp. 130131)

Are lone inventors more likely to generate break-throughsor is that just a myth? Nobel-prizewinningauthor John Steinbeck offers an eloquent testimonial tothe creative abilities of the individual. He is not alone;writers, historians, and inventors have long champi-oned the lone inventor over the group in the realmof creativity and, in particular, in the invention of

breakthroughs (Schumpeter 1934; Mokyr 1990, p. 295;Hughes 2004, p. 53). Many creativity researchershave supported these arguments by elaborating theproblems of creative teams, including idea blocking,communication difficulties, and interpersonal tensions(Diehl and Stroebe 1987, Mullen et al. 1991, Dougherty1992, Runco 1999, Paulus and Brown 2003). Even

proponents of teamwork acknowledge that researchon the benefits of collaborative creativity remainssomewhat weak (Paulus and Nijstad 2003, p. 4).Such modesty notwithstanding, proponents of col-laboration have begun to successfully question themyth of the lone inventor. They have documentedmany disadvantages of individual effort (Sutton andHargadon 1996, Paulus and Nijstad 2003, Perry-Smithand Shalley 2003, McFadyen and Cannella 2004) andthe almost ubiquitous trend toward collaboration inscience research (Wuchty et al. 2007). Much disagree-ment remains (Paulus and Nijstad 2003), however, andin particular, whether lone inventors are more or less

likely to invent breakthroughs.The goal of this paper is to enlarge the solu-

tion space for these debates on the value of collab-oration, particularly with regard to the sources of

breakthroughs and the processes by which they areconceived and developed. We propose that researchon collaboration and creativity should place moreemphasis on theorizing about and examining theentire distribution of creative outcomes. This is sub-stantively important because such distributions tendto be extremely skewed, with most inventions being of

41


2/16

Singh and Fleming: Lone Inventors as Sources of Breakthroughs: Myth or Reality?42 Management Science 56(1), pp. 4156, 2010 INFORMS

little practical significance and a minute few being dis-proportionately impactful. We develop this argumentin the context of lone inventors, where a lone inven-tor is socially isolated and either does not work withcoinventors in a team, does not work for an organiza-tion, or both. Although most studies on collaborative

creativity focus on the issue of individuals versus col-laborative teams, and a few studies focus on garageinventors who do not work within an organization,very few studies consider the two contexts simultane-ously. We propose that many of the arguments linkingcollaborative teams with idea generation also general-ize to a comparison of ideas generated within versusoutside an organization.

Following most work in statistical theory and esti-mation, almost all research on creativity has consid-ered the influence of explanatory variables on theaverage or mean outcome. Motivated by an inter-est in particularly successful outcomes, or break-

throughs, recent work has empirically modeled thesecond moment or variance of creative outcome dis-tributions (Dahlin et al. 2004, Taylor and Greve 2006,Girotra et al. 2007, Fleming 2007). Greater variancecan increase the chances of a breakthrough becausethe associated increase in the mass in both tailsimplies a greater number of breakthrough outliers(Campbell 1960, Simonton 1999). These ideas paral-lel Marchs (1991) argument that structures, activi-ties, and mechanisms that increase mean performancethrough exploitation might be quite different fromthose that achieve breakthroughs through exploration

by increasing the variance in performance.

Unfortunately, the cumulative evidence on collab-oration and breakthroughs remains ambiguous andindeed often conflicting. Dahlin et al. (2004) demon-strate that independent or garage inventors (thosewho do not work for an organization) are over-represented in the tails of creative distributions. Con-sistent with this evidence, Fleming (2007) uses mean-variance decomposition models (King 1989) to show alower average and greater dispersion of creative out-comes by individuals who work alone. In contrast tothese results, Taylor and Greve (2006) demonstratethat collaboration in teams leads to higher variance indeviation from a normalized mean measure. Girotraet al. (2007) adopt an experimental approach and like-wise demonstrate higher variance in outcomes gener-ated by teams.

An examination of just the variance of outcomes,however, does not present the complete picture ofhow collaboration affects the distribution of creativeoutcomes. The hypothetical cumulative distributionfunctions shown in Figures 1(a) and 1(b) illustrate dif-ferent ways in which collaboration could affect theoutcome distribution, even assuming that collabora-tion is beneficial on average (many would dispute the

assumption; see Paulus and Nijstad 2003). The debateis often framed as whether the observed variance ofoutcomes is more in line with scenario (i) or (ii) inFigure 1(a): If individual inventors are associated withgreater variance of outcomes, scenario (i) is implic-itly assumed and collaboration is therefore judged as

being less desirable for achieving breakthroughs. Onthe other hand, if individual inventors are associatedwith lower variance, then scenario (ii) is implicitlyassumed and collaboration is considered more desir-able. Note, however, that both scenarios in Figure 1(a)assume symmetry in how collaboration affects theextremes: Neither allows the possibility that increasedlikelihood of breakthroughs could come without acorresponding increase in likelihood of particularly

bad outcomes. As scenarios (iii) and (iv) in Figure 1(b)illustrate, achieving greater variance is in reality nei-ther necessary nor sufficient for ensuring greater like-lihood of breakthroughs. Whereas lone inventors are

associated with greater variance in scenario (iii) andlower variance in scenario (iv), they are worse atachieving breakthroughs in both these scenarios. Inother words, the effects at the two tails need notinvolve a trade-off: collaboration can increase the like-lihood of breakthroughs while simultaneously reduc-ing the probability of particularly bad outcomes.1

Building on a stylized evolutionary model of cre-ativity, we explore why lone inventors might be lesslikely to invent breakthroughs and more likely toinvent failures. Supporting an argument that collab-oration improves the sorting and identification ofpromising new ideas, we find that working as a

part of a team or an organization, or both, trimsthe lower tail of the distribution of outcomes. Onthe other hand, and supporting an argument thatcollaboration enables more creative novelty, we findthat team and/or organization affiliation increases thelikelihood of creative outcomes toward extremely suc-cessful outliers. These beneficial effects on the dis-tribution of outcomes reflect more than an upwardmean shift: The effect is found to depend signifi-cantly on the quantile of the outcome distribution(Koenker and Bassett 1978). Consistent with an argu-ment that a greater diversity of knowledge enablesgreater recombinant opportunity and more rigorousassessment of that opportunity, we find two partial

1 These examples should not be taken literally. For ease of model-ing, the illustrative scenarios in Figure 1 were generated by assum-ing a normal distribution where collaboration increases the mean

but is allowed to increase or decrease the variance to differentextents. In most distributions, larger variance willin the limit havea greater probability of the very extreme outcomes at both ends(Fleming 2007). However, the probability mass where this happensmight be too trivial to be of economic significance. Furthermore,real-world outcomes need not obey a strictly normalor for thatmatter even symmetricdistribution.


3/16

Singh and Fleming: Lone Inventors as Sources of Breakthroughs: Myth or Reality?Management Science 56(1), pp. 4156, 2010 INFORMS 43

Figure 1(a) Illustrative Scenarios Where Greater Variance and

Greater Likelihood of Breakthroughs Occur Together

Lone inventor

Collaborative inventors

0

0.

2

0.

4

0.

6

0

.8

1.

0

p

0

0.

2

0.

4

0.

6

0.

8

1.

0

p

0 2 4 6 8 10

Value

0 2 4 6 8 10

Value

Scenario (i)

Scenario (ii)

Lone inventor


Notes. Scenario (i): Lone inventors achieve greater overall variance, greater

likelihood of breakthroughs, and greater likelihood of particularly poor

outcomes. Scenario (ii): Lone inventors achieve lower overall variance,

lower likelihood of breakthroughs, and lower likelihood of particularly poor

outcomes.

mediators of collaboration: the diversity of past tech-nological experiences of members in a collaborativeteam and the size of a teams external collabora-tion network. Supporting our meta-argument for theimportance of examining each tail of the distributionseparately, experience diversity helps trim poor out-comes significantly more than it helps create break-throughs, relative to the effect of external networks.

2. Lone Inventors, Creativity, and

Generation of BreakthroughsFollowing many researchers (Campbell 1960, Romer1993, Weitzman 1998, Simonton 1999), we view cre-ativity as an evolutionary search process across acombinatorial space. In the first phase of evolution-ary search, typically called the variation phase,people generate new ideas through combinatorialthought trials. The idea that novelty is a new com-

bination is at least as old as Adam Smith (1766).Given thorough historical search, novel technologiescan almost always be traced to combinations of prior

Figure 1(b) Illustrative Scenarios Where Greater Variance

and Greater Likelihood of Breakthroughs Do Not

Occur Together

0

0.

2

0.

4

0.

6

0.

8

1.

0

p

0

0.

2

0.

4

0.

6

0

.8

1.

0

p

0 2 4 6 8 10

Value

0 2 4 6 8 10

Value

Lone inventor


Scenario (iv)

Scenario (iii)

Lone inventorCollaborative inventors

Notes. Scenario (iii): Lone inventors achieve greater overall variance, lower

likelihood of breakthroughs, and greater likelihood of particularly poor

outcomes. Scenario (iv): Lone inventors achieve lower overall variance,

lower likelihood of breakthroughs, and greater likelihood of particularly poor

outcomes.

technologies (Basalla 1988). Science, music, language,art, design, manufacturing, and many other formsof creative endeavor have been described similarly(Gilfillan 1935, Romer 1993, Weitzman 1998).

In the second or selection phase, inventors eval-uate ideas to reject poor outcomes and identifythe most promising novelties. The processes withinthe generation and selection phases can be purelypsychologicalthat is, they can all occur within a sin-gle personor they can iterate between psychologicaland social-psychological processes. The purely psy-chological case would be an extreme example of alone inventor who has no interaction with collabora-tors or feedback of any kind. This archetypal exampleis probably rare in todays interconnected world; inour data, the example corresponds most closely to anindependent or garage inventor who works withouta team. At the other extreme of very social social-psychological processes, individuals work togetherclosely in both the generation and evaluation of ideas


4/16


(though, following Steinbeck, we believe that eachgenerative insight occurs within a single mind, afterwhich the insight may be shared, iterated on, andfurther recombined by collaborators). This archetypalexample is increasingly common in todays world(Wuchty et al. 2007), and in our data it corresponds

most closely to inventors who work in a team withinan organization.In the last or retention phase, members of a larger

creative community evaluate the selected ideas and goon to adopt a very few of them in their own creativesearches. This phase is mainly social. Indeed, exceptin the very rare cases when a purely objective mea-sure of the quality of an idea or invention can be used,it is completely social. Whereas objective measuresmay be possible in a univariate analysis of a particu-lar technology characteristic (such as transistor densityor miles per gallon), it is difficult to assess an intrin-sic and completely asocial value for most technologies

and even more difficult to make comparisons acrosstechnologies. Even expert assessment is still social,as the inventor(s) must necessarily communicate theidea to the experts. This argument is old; Hooker etal. (2003, p. 230), for example, argue, To be creative, avariation must somehow be endorsed by the field Creativity involves social judgment. (See also Simon-ton 1999, Csikszentmihalyi 1999.) Creative individu-als can incorporate their own prior work, but theirinfluence will be quite limited unless others pick upand build on their ideas. Following this evolutionarymodel, we define the ultimate success of a new idea as

its impact on future inventions.Independent of the ideas source (lone versus col-laborative), we propose that collaboration improvesthe effectiveness of the selection phase becausecollaborative selection will be more rigorous thanlone selection. We assume that individual inventorscreate and then immediately test their ideas and newcombinations within their own minds (Campbell1960). Most new ideas are quickly rejected; only afew are retained as the basis for continued search.The quality of even the retained ideas is still suspect,however, because individuals, whether experts(Simonton 1985) or nonexperts (Runco and Smith

1991), are notoriously bad evaluators of their ownideas. Teams have an inherent advantage in theidentification of the best ideas. A collaborative teamwill consider the invention from a greater variety ofviewpoints and potential applications; such broaderconsideration is more likely to uncover problems.Given the typically greater diversity of experience ona collaborative team, some member is more likely torecall having seen a problem with a similar inventionand argue to abandon or modify the approach. Inshort, collaborative creativity will subject individually

conceived ideas to a more rigorous selection processso that fewer poor ideas are pursued.

Anecdotal accounts by prolific and successful loneinventors support the argument; such inventors read-ily admit and joke about their inability to predictwhich of their inventions would prove to be break-

throughs (Schwartz 2004, p. 144). They often reporta division of labor between those who generate andthose who criticize: You wanted Charlie in the con-versation, because he would tell you when you werefull of it (Kenney 2006). The inventor of the alu-minum tennis racket, Styrofoam egg cartons, andplastic milk bottles reported that the problem withthe loner is that if you dont sift, you are liable tospend much time going down dead ends (Brefka2006). Referring to the inventor of a promising auto-mated language translator, a Carnegie Mellon profes-sor reports that Eli is a mad genius Both of those

words apply. Some of his ideas are totally bogus.And some of his ideas are brilliant. Eli himself cantalways tell the two apart (Ratliff 2006, p. 212). Thedual inventors of the Hewlett Packard thermal inkjetprinter were a prolific empirical tinkerer who gener-ated prototypes and a very methodical engineer whoexplained, documented, and criticized (Fleming 2002).

Arguments similar to benefits from affiliation withteams can also be made for affiliation with organiza-tions. We propose that a single independent inven-tor (the image here is of an antisocial individualworking in his or her garage) will be more isolatedthan a single inventor that works within an organi-

zation. The assumption is that an affiliated inventorwho does not collaborate will still enjoy more socialinteraction (among colleagues and technical experts)than an unaffiliated inventor. This assumption is con-sistent with perspectives that the ability to accumu-late and leverage knowledge provides a key reasonfor the existence of firms (Nelson and Winter 1982,Grant 1996). Accordingly, firms can be seen as socialcommunities that are a natural extension of teamswhen it comes to creation of new knowledge (Kogutand Zander 1992). Though there are surely excep-tions of highly connected yet independent inventors,

our argument depends on the typical independentinventor being more isolated than the typical affiliatedinventor.

Because isolated inventors will lack multiple and(to varying degrees) uncorrelated filters, they willuncover fewer potential problems and hence developmore dead ends. The individual inventor, lacking theadvantage of collaborative sorting, will develop morepoor ideas, with the result that a smaller proportionof her developed ideas will be used by others. Hence,we would expect collaboration to trim the lower tail


5/16


of the distribution of creative outcomes, giving ourfirst hypothesis.

Hypothesis 1. Lone inventors will invent a relativelygreater proportion of low impact inventions than collabo-rative inventors.

In addition to trimming the undesirable tail, col-laboration will also fatten the desirable end ofthe distribution. Repeating the oft-cited advantagesof diversity (Gilfillan 1935, Basalla 1988, Weitzman1998), we argue that collaboration should increasethe potential combinatorial opportunity for creatingnovelty. Each inventor brings a different set of pastexperiences and knowledge of potential technologiesto the search and thus increases the potential num-

ber of new combinations that can be generated. Newcombinations are more uncertain and more variablein their impact (Fleming 2001). In other words, theyshould increase the likelihood of both good and badoutcomes. Following the arguments for Hypothesis 1,however, collaboration should also provide a morerigorous selection process, such that the worse out-comes should be less likely to be developed. Col-laborative teams also generate more points in theupper tail because they can cycle through a greaternumber of iterations. Particularly if they work welland productively together, they can more efficientlygenerate and assess more potential options. Partly

because of their faster and more efficient selectionprocesses, such teams can spend more time on con-triving radically novel combinations that have greater

breakthrough potential. Immediate access to greater

diversity also makes teams more efficient in gener-ating radically new combinations in the first place.Thus, in addition to the assumption that teams investmore total effort in aggregate, we can expect them toiterate more quickly in the generation and selectionphases of creative search, thus generating more possi-

ble breakthroughs and avoiding more poor outcomes.For both the above reasonsgreater diversity

that enables greater recombinant opportunity and agreater volume of iterationscollaborative teams andaffiliated inventors should generate more potentialnovelty at the breakthrough end of the distribution.This leads to our second hypothesis.

Hypothesis 2. Lone inventors will invent a relativelylesser proportion of high impact inventions than collabora-tive inventors.

Note that these predictions do not depend on theinfluence of collaboration or affiliation on the aver-age outcome. Although we believe that collaborationshould also influence the mean positively (consis-tent with McFadyen and Cannella 2004), we arguethat our predictions involve more than a symmet-ric and upward shift of the mean (as could be real-ized by adding the same offset to every point in

a distributionwe will establish this empirically bydemonstrating effects of significantly different sizeson the two tails, in both logit models of extremeoutcomes and quantile regressions). And though ourdata do not afford an exogenous experiment in loneversus collaborative invention, we develop and test

additional observable implications of our theory withmediation analysis (Baron and Kenney 1986).The arguments for the first two hypotheses depend

heavily on the role of diversity. Following the logic ofHypothesis 1, we propose that diversity helps iden-tify a poor outcome before it is fully developed, thustrimming the lower tail. Diverse teams will containa greater variety of opinions and should be less vul-nerable to groupthink when assessing the value ofan idea (Janis 1972). Following the logic of Hypothe-sis 2, the diversity of team backgrounds should gen-erate greater recombinant opportunity and potentialapplications, thus fattening the upper tail. Typically,

as the size of a team increases, so should the aggre-gate diversity of experience within the team. Thesearguments imply that diversity mediates the value ofcollaboration, at both extremes. This leads to our thirdtestable hypothesis.

Hypothesis 3. The effect of collaboration on extremeinventive outcomes will be mediated by the diversity ofexperience of the team members.

The arguments for the value of collaboration shouldalso apply to indirect collaboratorspeople whowork with one or more of the team members onanother project but are not a part of the immediate

effort. These extended or supporting team mem-bers should provide benefits similar to those providedby the immediate team members. Such colleagues actas additional sources of recombinant diversity andshould therefore enhance the number of outcomes atthe top end of the distribution. They also act as addi-tional filters to trim the bottom end. For example,when an immediate team member describes a projectto nonteam colleagues, they can suggest overlookedpossibilities or problems.

Typically, as the size of a team increases, so shouldthe size of the external network that the team canaccess. Therefore, similar to the aggregated experience

diversity within the team, the size of the extendedcollaborative network should mediate the value ofcollaboration at both extremes. This implies the lastprediction.

Hypothesis 4. The effect of collaboration on extremeinventive outcomes will be mediated by the size of theextended social network of the team members.

3. DataWe examine the link between lone inventors and thedistribution of creative outcomes using U.S. Patent


6/16


Table 1(a) Raw Statistics Regarding Citation Impact of Patents from Individual Inventors vs. Teams

Standard Coefficient of 5th 10th 90th 95th 99th

Observations Mean deviation variation percentile percentile Median percentile percentile percenti le

Individual inventor 260438 949 1430 151 0 1 6 21 31 66

Team size= 2 136033 1103 1570 142 0 1 6 25 36 75

Team size= 3 67588 1252 1796 143 0 1 7 29 42 83

Team size= 4 29125 1373 2025 147 0 1 8 32 47 95Team size= 5 11906 1530 2218 145 0 1 8 36 53 105

Team size 6 10726 1777 2930 165 0 1 9 41 61 122

Overall 515816 1084 1631 151 0 1 6 24 36 76

Table 1(b) Raw Statistics Regarding Citation Impact of Unassigned vs. Assigned Patents

Standard Coefficient of 5th 10th 90th 95th 99th

Observations Mean deviation var iation percentile percentile Median percenti le percenti le percenti le

Unassigned 122553 822 1236 150 0 1 5 17 25 56

Assigned to firm 393263 1165 1728 148 0 1 7 26 39 80

Overall 515816 1084 1631 151 0 1 6 24 36 76

and Trademark Office (USPTO) patent data. Thesedata are attractive for several reasons. First, theyallow a systematic comparison of creative outcomeson both dimensions we are interested in: relativesuccess of individuals versus teams as well as ofindividuals working independently versus withinorganizations. Second, future citations received bypatents provide a systematic method of measuringimpact in a way that is comparable across out-comes. Third, the longitudinal nature of the dataallows us to examine a rich set of questions, includ-ing those requiring a historical account of past expe-

riences of inventors in terms of both the kind ofprojects they have worked on before and the peo-ple they have collaborated with. Finally, being ableto draw a large sample across a wide range of sec-tors increases the power of the statistical tests andmakes findings more general. We constructed thedata set from three sources: the USPTO itself, theNational University of Singapore, and the NationalBureau of Economic Research (Jaffe and Trajtenberg2002, Chap. 13). We applied inventor-matching algo-rithms similar to those previously employed by Singh(2005, 2008), Trajtenberg et al. (2006), and Fleminget al. (2007) to create a reliable patent-inventor map-

ping. Assignee names and parent-subsidiary match-ing were corrected based on procedures described inSingh (2007).

We follow the well-established tradition of usingthe extent to which a specific patent gets cited byfuture patents as a measure of its impact and ulti-mately its success. Firms and individuals differ intheir reliance on patents, often relying on alternativemeans of protecting their intellectual property (Levinet al. 1987). Nevertheless, conditional on a specificinnovation being patented, citations to that patent

help capture its overall economic, social, and tech-nological success. The number of citations a patentreceives has been shown to be correlated with sev-eral measures of value, including the consumer sur-plus generated (Trajtenberg 1990), expert evaluationof patent value (Albert et al. 1991), patent renewalrates (Harhoff et al. 1999), and contribution to anorganizations market value (Hall et al. 2005).2

We restrict our final sample to successful patentsfiled during the 10-year period 19861995, allowingsufficient historic as well as future time window forconstructing our measures, such as prior experience

of a patents inventors and future citation impact ofthe patent. In constructing these measures, we useinformation from all USPTO patents granted dur-ing 19752004. However, for comparability duringregression analysis, the actual sample analyzed (from1986 to 1995) was further restricted only to patentsarising from U.S.-based inventors.

A simple inspection of the data provides prelim-inary support for our arguments. As indicated inTable 1(a), patents resulting from teams appear to beassociated with more citations than those from indi-vidual inventors, a benefit that appears to increaseunambiguously with team size. Likewise, Table 1(b)

shows patents assigned to organizations also receivemore citations. The standard deviation of citation out-comes also increases with team and organization affil-iation, though there is no obvious relationship withthe coefficient of variation. The reported percentile

2 This is also consistent with the view of the USPTO: If a singledocument is cited in numerous patents, the technology revealedin that document is apparently involved in many developmentalefforts. Thus, the number of times a patent document is cited may

be a measure of its technological significance (Office of TechnologyAssessment and Forecast 1976, p. 167).


7/16


Figure 2(a) Cumulative Distribution of Observed Impact of Patents

from Individuals vs. Teams

0

0.

2

0.

4

0

.6

0.

8

1.

0

p

0 2 4 6 8 10

Normalized citation count

Individual inventor

Team size = 2Team size = 3Team size = 4

Team size = 5

Team size > 6

Figure 2(b) Cumulative Distribution of Observed Impact of

Unassigned vs. Assigned Patents

statistics suggest that team and organization affilia-tion is more likely to be associated with breakthrough(i.e., high-citation) outcomes while simultaneously

being less likely to be associated with low-value (i.e.,low-citation) outcomes. The raw data appear to bemost consistent with the theoretical scenario (iv) inFigure 1(b).

Figures 2(a) and 2(b), respectively, illustrate thecumulative distribution of observed outcomes fromindividuals versus teams (of different sizes) and fromunaffiliated inventors versus inventors whose patentsare assigned to organizations.3 Notice that the cumu-

lative distribution plots in both figures never inter-sect. In Figure 2(a), the outcome distribution forteams stochastically dominates the distribution for thelone inventor, with larger teams actually also dom-inating smaller teams. Likewise, in Figure 2(b), theoutcome distribution for patents assigned to organi-zations stochastically dominates the distribution for

3 To enable valid comparisons across different technologies, thesefigures are drawn using a citation impact measure that has beennormalized relative to the average citation impact in the same year-technology class cohort.

Figure 3 Relevant Coefficients and 95% Confidence Intervals from

Quantile Regression of Citations on Lone Invention

1

3

5

7911

Coefficientfrom

quan

tileregression

0 10 20 30 40 50 60 70 80 90 100

Quantile

Individual assigned

Team assigned Team unassigned

unassigned patents. Overall, these figures are oncemore consistent with a view that being a lone inven-tor decreases the probability of breakthroughs while

increasing the probability of particularly bad out-comes. Again, the evidence appears most consistentwith scenario (iv) from Figure 1(b). The next sectionintroduces regression models that enable us to exam-ine these issues in more detail and to formally test thehypotheses stated earlier.

4. Regression MethodologyWhereas previous research has primarily focused oneffect of collaboration on theaverageoutcome, we areinterested in the entire distribution of outcomes. Inparticular, we start by examining drivers of extreme

outcomesinventions that can be considered break-throughs versus those with little impact. We estimatelogistic regression models of the likelihood that apatents impact falls within one of these two extremes.At the upper tail, breakthrough inventions are definedusing an indicator variable cites_ p95 that is set to 1if and only if a patent is in the top 5% in termsof frequency of future citations received, comparedwith patents of the same application year and tech-nology class. Analogously, particularly poor outcomesare defined using an indicator variable citesEQ0 thatis set to 1 if and only if a patent receives no citations.4

Table 2 summarizes our key variables. Two indi-cator variables capture two different dimensions oflone inventor. The first variable, team, captureswhether a patent came from a single inventor (0) orfrom a team of two or more inventors (1). The secondvariable, assigned, captures whether the patent origi-nated outside the boundaries of any organization (0)

4 All findings reported in the paper are robust to using the top1% citation impact in defining breakthroughs and also to usingachievement of either 0 or 1 future citations in defining particularlypoor outcomes.


8/16


Table 2 Variable Definitions and Summary Statistics

Mean Std. dev. Min Max

Dependent variables

Cites_p95 Patent top 5% in citation impact 005 022 0 1

CitesEQ0 Patent receives no citations 007 025 0 1

Explanatory and control variables

Team Indicator that is 1 if and only if patent invented by more than one person 050 050 0 1Assigned Indicator that is 1 i f and only i f patent has an assignee firm 076 043 0 1

Claims Number of claims made by the patent 1466 1178 1 320

Patent_references Number of backward citations that the patent makes to other patents 1086 1208 0 745

Nonpatent_references Number of nonpatent references made by the patent 192 659 0 100

Average_experience Average number of previous patents for this teams inventors 557 1339 0 347

Joint_experience Number of past patents invented by the same team 190 671 0 274

Mediator variables

Experience_diversity Number of technology classes any team inventor has patented in before 616 952 0 234

Network_size Number of inventors at distance 2 in the teams collaborative network 1177 3012 0 991

or from within an organization (1).5 An interactionof these two variables tests the implicit hypothesis

that a doubly isolated inventoran individual work-ing outside any organizationis at the most severedisadvantage.

The regression analysis employs several controlvariables suggested by previous research: claims (thescope of the patent as measured by its number ofclaims), patent_references (number of references madeto previous patents), nonpatent_references (number ofreferences made to public sources outside of patents),average_experience(the average number of past patentsmembers of this team have been involved with), and

joint_experience (the number of past patents from thesame team). Because all these variables are highly

skewed, we use their logarithmic transformation inthe actual analysis.6 These variables control for thegreater resources of teams and the possibility of previ-ously developed collaborative advantage. Technologyfixed effects were used to account for systematic dif-ferences in citation rate across different technologies.Likewise, year fixed effects were used to account forany systematic differences over time, including thosearising from different observed windows of opportu-nity to be cited by future patents until 2004. Finally,to account for the possibility that error terms might becorrelated for observations involving the same inven-tor, we report robust standard errors that are clustered

on the identity of the first inventor.To generalize from examination of the two extremes

to the entire distribution of outcomes, we employ

5 All our findings are robust to using the actual number of inventorsbehind a patent rather than just an indicator variable (team) for thisnumber being 0 (for an individual inventor) or more than 1 (for ateam). We report the latter for ease of direct comparison with theother variable (assigned, which is also an indicator variable).6 In doing so, we first added one to any variables that can take avalue of 0. The results are robust to changing the size of the offsetor using the original variable.

a quantile regression approach (Koenker and Bassett1978; for its first application in the study of creativ-

ity, see Girotra et al. 2007). Unlike classical regression,which relates the mean of a dependent variable to theexplanatory variables, quantile regression estimateshow the relationship varies for different percentilesof the data. This allows an explanatory variable toexhibit different effects for different percentiles.

We test the mediator hypotheses with the proce-dure suggested by Baron and Kenney (1986). Fortesting Hypothesis 3, that benefits of collaborationoperate through the greater diversity of experiencethat teams bring to bear, we define experience_diversityas the number of distinct technology classes theinventor or inventors have patented in before. This

assumes that knowledge of more technical areas offersa greater diversity of ideas available for recombi-nation of criticism. For testing Hypothesis 4, that

benefits of collaboration operate through indirect col-laborative networks, we construct a measure of theexternal collaboration networks of the inventor(s):Network_size for the focal patent is defined as thenumber of unique inventors that are at a social dis-tance of not more than two from the focal inventor(s),i.e., are either recent collaborators or collaboratorscollaborators for one or more of the inventor(s).

5. ResultsA summary of definitions and key statistics for allvariables appears in Table 2, and a matrix of corre-lations among these variables is reported in Table 3.This section details regression analyses.

5.1. Lone Inventors and Extreme OutcomesThe analysis reported in columns (1)(4) in Table 4demonstrates that patents generated by inventorswith a team and/or organization affiliation are morelikely to end up as breakthroughs than those gen-erated by lone inventors. The magnitudes of these


9/16


Table 3 Correlation Matrix Among Variables

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

(1)Cites_p95 1000

(2)CitesEQ0 0062 1000

(3)Team 0058 0011 1000

(4)Assigned 0058 0012 0344 1000

(5) ln_claims 0089 0071 0100 0136 1000(6) ln_patent_references 0067 0082 0020 0002 0156 1000

(7) ln_nonpatent_references 0074 0011 0164 0168 0132 0089 1000

(8) ln_average_experience 0029 0036 0156 0272 0072 0016 0115 1000

(9) ln_joint_experience 0013 0038 0286 0010 0005 0016 0002 0655 1000

(10) ln_experience_diversity 0057 0012 0343 0337 0101 0018 0149 0873 0436 1000

(11) ln_network_size 0069 0014 0414 0380 0096 0006 0206 0628 0070 0669 1000

effects are substantial. For example, the estimates incolumn (3) imply that, keeping other variables attheir average value, patents from teams are 28% morelikely than patents from individuals to be in the 95thpercentile of citations. Similarly, assigned patents are

63% more likely than unassigned patents to be inthe 95th percentile of citations.7 Column (4) examinesinteraction effects between team and organizationaffiliation and finds the two to be complements:The citation impact is greatest for patents arisingfrom teams associated with organizations, with suchpatents being 2.11 times as likely to achieve a break-through compared with a lone inventor with neitherteam nor organization affiliation. There appears to belittle evidence that lone inventors are the sources of

breakthroughs.8

Next, we examine the other extreme of how beinga lone inventor affects the likelihood of inventing

particularly poor outcomes. As the results report incolumns (5)(8) of Table 4, patents from inventorswith team and/or organization affiliation are lesslikely to receive no citation at all. The magnitude ofthese effects is again substantial. For example, the esti-mates in column (7) imply that patents from teamsare 9% less likely than patents from individuals toreceive no citations. Similarly, assigned patents are14% less likely than unassigned patents to receiveno citations. Column (8) examines interaction effects

7 We have been somewhat conservative in using the number ofclaims, number of patent references, and number of nonpatent ref-

erences as control variables. A more aggressive interpretation couldbe that these three variables are also potential mediators throughwhich working in a team or an organization, or both, shapes thefinal outcome. Indeed, these variables are positively correlated withlikelihood of a patent being a breakthrough, and excluding themfurther increases the estimated effects ofteamand assigned, castinglone inventors in an even poorer light.8 This finding is robust to redefining breakthroughs based only oncitation counts calculated even after dropping citations an assignedpatent receives from future patents originating within the sameorganization. In other words, the results are not just a manifesta-tion of assigned patents generating significant within-organizationcitations.

between working in teams and working in an organi-zation and finds patents arising from teams associatedwithin organizations to be the least likely to fail (theyare 22% less likely to have no citations compared witha lone inventor with neither team nor organization

affiliation). Overall, lone inventors seem more likelyto end up in the left tail of the overall distribution ofoutcomes.9

To summarize, the evidence rejects a view thatbeing a lone inventor increases the probability ofachieving both extremely good and extremely badoutcomes. Instead, collaboration (both in terms ofteam affiliation and organization affiliation) is ben-eficial at both extremes: It increases the probabilityof breakthroughs while simultaneously decreasing theprobability of particularly poor outcomes. More gen-erally, the findings demonstrate why an analysis ofonly working alone versus collaborating achieves

greater variance and would be misplaced because itignores the very real possibility that breakthroughsneed not come at the expense of simultaneouslyincreasing the likelihood of poor outcomes.

5.2. Quantile Regression AnalysisFigure 3 plots the estimated coefficients and associ-ated 95% confidence intervals for the first three indi-cator variables from a quantile regression (Koenkerand Bassett 1978).10 This helps compare three cat-egories of patentsassigned patents from teams,assigned patents from individuals, and unassigned

9 This finding is also robust to dropping citations within the orga-nization as far as the team versus individual inventor distinctionis concerned. However, the finding that assigned patents are lesslikely to end up in the left tail than unassigned patents no longerholds if assignee self-citations are excluded, suggesting that theimpact of assigned patents with only a few citations is dispropor-tionately likely to be confined within the organization itself.10 To conserve space, the actual table of quantile regression esti-mates (including those for the control variables) has not beenincluded in the paper, but is available from the authors uponrequest. Also note that in Figure 3 a logarithmic scale has been usedfor the y-axis for easy visual comparison of different curves evenat the bottom tail of the distribution.


10/16


Table 4 Regression Analyses of Extreme Outcomes upon Lone Invention

(1) (2) (3) (4) (5) (6) (7) (8)

Dependent variable: Cites_p95 Cites_p95 Cites_p95 Cites_p95 CitesEQ0 CitesEQ0 CitesEQ0 CitesEQ0

Regression model: Logistic Logistic Logistic Logistic Logistic Logistic Logistic Logistic

Team 0347 0257 0125 00962

0019 0019 0017 0017Assigned 0573 0508 0182 0161

0025 0026 0016 0017

Team assigned 0780 0262

0030 0021

Team unassigned 0340 0176

0044 0032

Individual assigned 0535 0184

0031 0019

ln_claims 0473 0462 0457 0458 0309 0302 0302 0302

0012 0012 0012 0012 00088 00088 00088 00088

ln_patent_references 0243 0244 0239 0239 0315 0314 0314 0314

0012 0012 0012 0012 0010 00100 0010 0010

ln_nonpatent_references 0284 0283 0276 0276 00998 00984 00957 00959

0010 0010 0010 0010 0011 0011 0011 0011ln_average_experience 0146 0161 0111 0111 00111 00165 000463 000300

0012 0012 0012 0012 0011 0011 0011 0011

ln_joint_experience 0135 0212 0119 0121 0123 0149 0115 0118

0020 0019 0020 0020 0016 0013 0016 0016

Year fixed effects Included Included Included Included Included Included Included Included

Technology fixed effects Included Included Included Included Included Included Included Included

Observations 509,840 509,840 509,840 509,840 509,840 509,840 509,840 509,840

2 4,960 5,341 5,318 5,301 12,601 12,637 12,693 12,698

Degrees of freedom 50 50 51 52 50 50 51 52

Log likelihood 100,915 100,711 100,577 100,575 113,203 113,172 113,150 113,145

Note.Robust standard errors clustered by the first inventor are in parentheses.p


11/16


Table 5 Regressions of Experience Diversity and Network Size as Potential Mediators

(1) (2) (3) (4)

Dependent variable: Experience_diversity Experience_diversity Network_size Network_size

Regression model: Negat ive binomial Negative binomial Negat ive binomial Negative binomial

Team 0580 0723

00054 0012Assigned 0172 1331

00071 0024

Team assigned 0733 2146

00081 0035

Team unassigned 0488 1068

0013 0044

Individual assigned 0139 1468

00087 0035

ln_claims 00407 00405 00367 00369

00027 00027 00070 00071

ln_patent_references 000657 000631 0000752 0000112

00026 00026 00055 00055

ln_nonpatent_references 00297 00295 00882 00876

00024 00024 00053 00053ln_average_experience 1064 1065 1689 1688

00049 00049 00089 00090

ln_joint_experience 0114 0114 0983 0979

00057 00057 0012 0012

Year fixed effects Included Included Included Included

Technology fixed effects Included Included Included Included

Observations 509,840 509,840 509,840 509,840

2 104,975 105,715 83,382 82,695

Degrees of freedom 51 52 51 52

Log likelihood 1,113,549 1,113,415 1,194,324 1,193,965



12/16


Table 6 Mediation of Lone Invention by Experience Diversity and Network Size

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Dependent variable: Ci tes_p95 C ites_p95 C ites_p95 Ci tes_p95 Ci tes_p95 Ci tesEQ0 Ci tesEQ 0 Ci tesEQ 0 Ci tesEQ 0 Ci tesEQ 0

Regression model: Logistic Logistic Logistic Logistic Logistic Logistic Logistic Logistic Logistic Logistic

Team 0257 0137 0202 0102 0101 00962 00363 00848 00287 00249

0019 0020 0019 0021 0021 0017 0018 0017 0018 0018

Assigned 0508 0487 0447 0435 0429 0161 0150 0146 0138 01090026 0026 0026 0026 0027 0017 0017 0017 0017 0017

ln_experience_diversity 0331 0286 0279 0187 0182 0132

0019 0019 0020 0016 0016 0017

ln_network_size 0198 0177 0180 00470 00404 00604

00091 00091 00091 00081 00081 00085

ln_expdiversity ln_netsize 000739 00452

00068 00052

ln_claims 0457 0452 0462 0457 0456 0302 0298 0303 0299 0295

0012 0012 0012 0012 0012 00088 00088 00088 00088 00088

ln_patent_references 0239 0238 0238 0238 0238 0314 0313 0314 0314 0316

0012 0012 0012 0012 0012 0010 00100 00100 00099 00099

ln_nonpatent_references 0276 0270 0266 0261 0261 00957 00920 00920 00890 00881

0010 0010 0010 0010 0010 0011 0011 0011 0011 0011

ln_average_experience 0111 0197 0129 0369 0365 000463 0173 00611 0217 0187

0012 0023 0015 0024 0024 0011 0019 0014 0021 0021

ln_joint_experience 0119 00616 00488 00831 00786 0115 00871 00742 00530 00820

0020 0020 0020 0020 0021 0016 0015 0016 0015 0016

Year dummies Included Included Included Included Included Included Included Included Included Included

Technology dummies Included Included Included Included Included Included Included Included Included Included

Observations 509,840 509,840 509,840 509,840 509,840 509,840 509,840 509,840 509,840 509,840

2 5,318 6,178 5,612 6,465 6,559 12,693 12,652 12,769 12,736 12,795

Degrees of freedom 51 52 52 53 54 51 52 52 53 54

Log likelihood 100,577 100,251 100,133 99,896 99,895 113,150 113,040 113,123 113,021 112,970



13/16


Although our archival database enables this studyby furnishing enough data to model extremely rareevents, it also imposes a number of limitations. As istypical of most studies on collaboration that employarchival data, our results are based on a cross-sectional comparison of preexisting (and successful)

teams. Compared to laboratory teams, analysis of pre-existing and real-world teams provides greater exter-nal validity to our results, but at the cost of nothaving a random assignment of inventors to individ-ual versus collaborative effort in the form of teamand/or organization affiliation. Because we do nothave a model of how individuals get assigned tospecific teams or organizations, we cannot be sureif the same inventor(s) would have behaved accord-ing to our predictions in different situations. Selec-tion issues also remain, because publication data bydefinition only record successful submissions. Simi-lar unobserved influences might also complicate the

number of inventors that claim coauthorship of abreakthrough; when a particularly promising inven-tion is applied for, more peripheral inventors mayclaim to have contributed to the work and thus inflatethe explanatory variable.

One specific concern with this cross-sectionalapproach is that better or well-established inventorsmay have a greater propensity to engage in collabora-tion. This could lead to an upward bias in estimatedgains from collaboration. We tried to address thisissue through three robustness checks. First, to cap-ture the possibility that certain inventors might havea greater affinity for collaboration, we analyzed mod-

els that explicitly accounted for past projects for theinventor(s) of a patent being disproportionately col-laborative. This did not substantively alter our results.Second, we analyzed a subsample consisting onlyof inventors with no prior patents and found evenstronger advantage to collaboration. Third, we reranthe analysis employing panel data models with fixedeffects meant to capture time-invariant characteris-tics of the first inventor. The main findings remainedunchanged. Admittedly, even fixed effects models areinsufficient for ruling out unobserved individual char-acteristics that change over time. A natural exper-iment or an instrumental variable approach wouldhave been ideal to address many of these concerns,

but neither was found in our research context.Another concern, which applies not just to our

study but to most existing research on collabora-tion, is the difficulty of making conclusive statementsabout the net value of collaboration. In other words,would a particular set of inventors be more likely toinvent a breakthrough if they collaborated from thestart or if they worked alone and then pooled theirefforts at the end? (Or following Girotra et al. 2007,at what point in the process should inventors start

working together?) It is probable, for example, thatcollaboration imposes more administrative costs thanindividuals working alone (as a trivial but obviouscost, only one person can talk at a time duringmeetingsassumedly, such communication is instan-taneous and almost costless within an individual).

This question strikes as particularly important forfuture work, from both a theoretical and normativeperspective.

Against these qualifications, we should highlightthe unique strengths of our studya large scaleexamination of the entire distribution of creative out-comes, wherein we find strong and robust advan-tage to collaboration as well as significant mediationeffects consistent with theory. Even if a similarly pow-erful lab study could be performed (and require hun-dreds or even thousands of subjects, because of therarity of creative outliers), such a lab study would still

be unable to provide the long-term and social impact

of breakthroughs (as measured by citations) that occurand diffuse in the real world.

At a minimum, we hope this article prompts arethink of current approaches to researching cre-ativity. Theorists should increase their metaphoricaldegrees of freedom; rather than focusing only onthe mean or the symmetric variance, or both, of a dis-tribution, they should begin thinking about the dif-ferent effects a variable may have on the lower tail,the mean, and the upper tail. Diversity, for exam-ple, has been argued and shown to have frustrat-ingly wide and conflicting influences upon creativity.Proponents of diversity have argued and shown that

the collaboration of diverse individuals creates oppor-tunities for creative abrasion and technological bro-kering (Leonard-Barton and Swap 1999, Burt 2004);detractors of diversity have argued and shown thatdiversity leads to communication and interpersonalproblems (Dougherty 1992); those seeking resolutionhave argued and demonstrated that diversity helpscreativity until the point where the cognitive andcommunication challenges become too great, whichshould be observable in a nonmonotonic effect (Ahujaand Lampert 2001). These arguments are all reason-able and interesting. The mixed evidence adduced todate, however, should be reconsidered in light of adifferential effect of diversity on the tails, keeping inmind that previous results may have been biased byincorrect model specification, outlier data points, andsensitivity to heteroscedasticity. All of these proposedmechanisms for diversitys effect may operate, butthey may do so in different regions of the distributionand in surprising ways (such as an improvement inthe lower rather than upper tail).

Our results on external collaboration networks alsospeak to the controversy of how small world net-works influence creativity (small world networks are


14/16


characterized by individually tight clusters of net-works that are linked together sparsely; see Wattsand Strogatz 1998). Uzzi and Spiro (2005) demonstratepositive correlations between small world structureand creativity in Broadway musicals, and Fleminget al. (2007) demonstrate weak correlations of both

signs between small world structure and (1) sub-sequent patenting in regions and (2) citation andfuture use of a new creative combination. This con-flicting and weak evidence in prior empirical workon small worlds may be due to a failure to look atthe tails of the creative distribution. Consistent withour results on the importance of extended networksto the invention of breakthroughs, extended networksmay increase the skewness of breakthroughs due toenhanced contagion effects (Bikhchandani et al. 1992).Alternatively, if one thinks of the last evolutionarystage of retention as the aggregation of selectionchoices made by each member of the larger commu-

nity, then large and extended networks might increasethe proportion of failures as well. This would notoccur from differences in the intrinsic quality of ideas

but from the faster recognition and identification ofpoor versus promising ideas. Dense and extendedsocial networks would enable faster diffusion of tacitand inside information, and this might allow forfaster sorting of useless versus promising novelty.Note that this argument differs from an argument thatextended networks should increase average impact,

because of an increased number of social conduitsfor diffusion. How large social networks influencethe retention and distribution of invention outcomes

strikes us as an important topic for future research.Returning to the papers specific motivation, the

conflicting results for lone inventor variance (Dahlinet al. 2004, Fleming 2007 versus Taylor and Greve2006, Girotra et al. 2007) may have been driven byan increase in lone inventor failures rather than asymmetric increase in failures and breakthroughs.But the models of variance in these conflicting stud-ies implicitly assumed symmetry of the tails andtherefore masked any asymmetric influence of collab-oration (though Girotra et al. 2007 estimate a quantilemodel). Although focusing on the entire distribu-tion of outcomes will obviously increase the complex-ity of research, it should also provide more accuratemodels and more nuanced predictions of the phe-nomenon of creativity. Our results also highlight theneed for empirical and methodological caution ingeneral, wherein failure to account for different influ-ences of explanatory variables in different parts of thedistribution may provide only limited understandingof the phenomenon, particularly if the outcomes varygreatly.

Further progress using evolutionary analogies forcreativity will require a combination of laboratory and

archival methods. An evolutionary model of creativityis extremely difficult to study in its entirety becauseit unfolds over time and space (where space can bedefined quite broadly, including social, geographical,technical, or organizational) and over levels of anal-ysis (psychological; social-psychological; and social,

economic, or industrial). Some methods, such as lab-oratory experiments, are very well suited to under-standing the earlier stages of the process. With labexperiments, the subjects, sequence of psychologicaland social iteration, and other inputs to the processcan be carefully controlled and randomized. But labexperiments cannot observe the social stage of reten-tion by a larger community nor the huge numberof trials required to invent an outlier breakthrough.Other methods, such as econometric analyses of largesample archival data, are better suited to understand-ing breakthroughs and the acceptance and success ofideas but are less well suited to understanding the

underlying processes of generation because they onlyobserve an idea after it has been selected for pub-lication. Because lab methods and archival studiesappear complementary, the research streams should

benefit greatly from awareness and cross-fertilization.Furthermore, given the intuitive difficulty of concep-tualizing outliers and higher moments, researchersshould also incorporate formal models (e.g., Dahanand Mendelson 2001, Girotra et al. 2007, Kavadiasand Sommer 2007) and appropriate econometric tech-niques (e.g., Koenker and Bassett 1978) into theirresearch.

7. ConclusionIs the lone inventor more or less likely to invent a

breakthrough? Using U.S. patent authorship and cita-tion data, we establish empirically that the lone inven-tor is less likely to invent a breakthrough. This resultholds for inventors who work outside of organiza-tions and for inventors who work without coauthors(those who work alone and outside of an organizationare least likely of all to invent a breakthrough). Recentand conflicting research on the topic has focused onhow lone invention influences the variance of cre-ative outcomes. This research has assumed symmetric

influences of lone invention upon the second moment.We advance arguments for the importance of asym-metric influences and demonstrate that lone inventiondecreases the chances of a breakthrough and increasesthe chances of a relatively useless invention. Quantileregression supports these arguments and establishesthat the effect is more than a mean shift.

Considering the mechanisms that give rise to theseresults, we propose that collaboration trims the bot-tom of the distribution, because of a greater numberand variety of critical assessments; isolated inventors


15/16


are less effective than more social inventors at cullingthe bad ideas. Collaboration should also increase thenumber of successful outliers because the diversityof groups enables more novel combinations. Support-ing these arguments, greater diversity within a groupand extended collaboration networks beyond a group

appear to mediate the benefits of collaboration. Sur-prisingly, it appears that diversity has a relatively andsignificantly greater effect in trimming poor outcomesthan fostering breakthroughs. In contrast, extendedsocial networks also appear to be more beneficial tothe invention of breakthroughs than to the trimmingof poor outcomes.

The results do not support accounts of the heroiclone inventor, at least during the relatively recent timeperiod covered by our study; the story could be amyth, at least when considering the ultimate successof an invention. Although we agree with Steinbeckthat original insight occurs within an individual brain,

it also appears that creative processes can benefitgreatly from collaboration (though perhaps this is alsoconsistent with his elaboration that, the group can

build and extend it, Steinbeck 1952, pp. 130131).The stochastic dominance of collaborative efforts oversocially isolated efforts is particularly telling; at everypoint in the distribution of creative outcomes, collab-oration correlates with increased impact. Still, we can-not put the myth entirely to rest. It may be that thelone inventors of the 19th century were truly heroic,

but 20th century changes in technology and organiz-ing have increased the advantages of collaboration. Itis also possible that the opportunity costs of invent-

ing together may be greater than the costs of sim-ply working alone and then summing the individualefforts. In either case, the larger goal of this paper wasto motivate a reexamination of how we study creativ-ity. As illustrated by the significantly different influ-ence of diversity on the top and bottom of the creativedistribution, focusing solely on the meanor assum-ing symmetric influence of collaboration upon thetails of a distributioncan be methodologically insuf-ficient and substantively misleading. However futureresearchers think about and study these issues, wehope this work highlights the importance of studyingthe asymmetric moments of collaborative creativity.

AcknowledgmentsThe authors thank the Harvard Business School Divi-sion of Research and INSEAD for supporting this work;Henrich Greve, Eric Lin, Chris Liu, Matt Marx, KristenMerrill, Santiago Mingo, and Scott Stern for their comments;and participants in presentations at Stanford University,Harvard University, the University of Toronto, and theAcademy of Management and INFORMS annual confer-ences for their feedback. Departmental editor ChristophLoch, an associate editor, and two anonymous reviewersprovided exceptional help in developing the work.

References

Ahuja, G., C. Lampert. 2001. Entrepreneurship in the large cor-poration: A longitudinal study of how established firms cre-ate breakthrough inventions. Strategic Management J. 22(67)521543.

Albert, M. B., F. Narin, D. Avery, P. McAllister. 1991. Direct valida-tion of citation counts as indicators of industrially important

patents. Res. Policy 20 251259.Baron, R., D. Kenny. 1986. The moderator-mediator variable dis-

tinction in social psychological research: Conceptual, strate-gic, and statistical considerations.J. Personality Soc. Psych.51(6)11731182.

Basalla, G. 1988. The Evolution of Technology. Cambridge UniversityPress, Cambridge, MA.

Bikhchandani, S., D. Hirshleifer, I. Welch. 1992. A theory of fads,fashion, custom, and cultural change as informational cas-cades. J. Political Econom.100(5) 9921026.

Brefka, P. 2006. Personal interview, July 10, Boston.

Burt, R. 2004. Structural holes and good ideas. Amer. J. Sociol. 110(2)34999.

Campbell, D. 1960. Blind variation and selective retention in cre-ative thought as in other knowledge processes. Psych. Rev. 67

380400.Clogg, C., E. Petkova, A. Haritou. 1995. Statistical methods for com-

paring regression coefficients between models. Amer. J. Sociol.100 12611293.

Csikszentmihalyi, M. 1999. Creativity. Harper Collins, New York.

Dahan, E., H. Mendelson. 2001. An extreme-value model of concepttesting.Management Sci. 47(1) 102116.

Dahlin, K., M. Taylor, M. Fichman. 2004. Todays Edisons or week-end hobbyists: Technical merit and success of inventions byindependent inventors.Res. Policy 33(8) 11671183.

Diehl, M., W. Stroebe. 1987. Productivity loss in brainstorminggroups: Toward the solution of a riddle. J. Personality Soc.Psych.53 497509.

Dougherty, D. 1992. Interpretive barriers to successful productinnovation in large firms. Organ. Sci. 3(2) 179202.

Fleming, L. 2001. Recombinant uncertainty in technological search.Management Sci. 47(1) 117132.

Fleming, L. 2002. Finding the organizational sources of break-through invention: The story of Hewlett Packards ink jetinvention. Indust. Corporate Change 11(5) 10591084.

Fleming, L. 2007. Breakthroughs and the long tail of innovation.Sloan Management Rev. 49(1) 6974.

Fleming, L., C. King, A. Juda. 2007. Small worlds and innovation.Organ. Sci. 18(6) 938954.

Gilfillan, S. C. 1935. The Sociology of Invention. Follett Publishing,Chicago. (Reprinted 1970. MIT Press, Cambridge, MA.)

Girotra, K., C. Terwiesch, K. Ulrich. 2007. Idea generation andthe quality of the best idea. Working paper, INSEAD,Fontainebleau, France.

Grant, R. M. 1996. Toward a knowledge-base theory of the firm.

Strategic Management J.17 109122.

Hall, B., A. Jaffe, M. Trajtenberg. 2005. Market value and patentcitations. RAND J. Econom. 1 1638.

Harhoff, D., F. Narin, F. Scherer, K. Vopel. 1999. Citation frequencyand the value of patented inventions. Rev. Econom. Statist. 81(3)511515.

Hooker, C., J. Nakamura, M. Csikszentmihalyi. 2003. The groupas mentor: Social capital and the systems model of creativity.P. Paulus, B. Nijstad, eds. Group Creativity: Innovation ThroughCollaboration. Oxford University Press, New York, 225244.

Hughes, T. 2004. American Genesis: A Century of Invention andTechnological Enthusiasm 18701970. Chicago University Press,Chicago.


16/16


Jaffe, A., M. Trajtenberg. 2002. Patents, Citations and Innovations:A Window on the Knowledge Economy.MIT Press, Cambridge,MA.

Janis, I. 1972. Victims of Groupthink. Houghton Mifflin Company,Boston.

Kavadias, S., S. Sommer. 2007. The effects of problem structure andteam expertise on brainstorming effectiveness. Working paper,

Kenney, S. 2006. Personal interview, August 13, Cambridge, MA.King, G. 1989. Variance specification in event count models: From

restrictive assumptions to a generalized estimator. Amer. J.Political Sci.33(3) 76284.

Koenker, R., G. W. Bassett. 1978. Regression quantiles.Econometrica46 3350.

Kogut, B., U. Zander. 1992. Knowledge of the firm, combinativecapabilities, and the replication of technology. Organ. Sci. 3(3)383397.

Leonard-Barton, D., W. Swap. 1999. When Sparks Fly: Igniting Cre-ativity in Groups. Harvard Business Press, Boston.

Levin, R., A. Klevorick, R. Nelson, S. Winter. 1987. Appropriatingthe returns from industrial research and development: Com-ments and discussion. Brookings Papers on Economic Activity 3783831.

March, J. G. 1991. Exploration and exploitation in organizationallearning. Organ. Sci. 2(1) 7187.

McFadyen, M., A. Cannella. 2004. Social capital and knowledgecreation: Diminishing returns of the number and strength ofexchange relationships.Acad. Management J. 47(5) 735746.

Mokyr, J. 1990. The Lever of Riches: Technological Creativity and Eco-nomic Progress. Oxford University Press, New York.

Mullen, B., C. Johnson, E. Salas. 1991. Productivity loss in brain-storming groups: A meta-analytic integration. Basic Appl. Soc.Psych.12 323.

Nelson, R., S. Winter. 1982. An Evolutionary Theory of EconomicChange. Harvard University Press, Cambridge, MA.

Office of Technology Assessment and Forecast. 1976. U.S. Depart-ment of Commerce, Patent and Trademark Office, Sixth Report.U.S. Government Printing Office, Washington, DC.

Paulus, P., V. Brown. 2003. Enhancing ideational creativity ingroups: Lessons from research and brainstorming. P. Paulus,B. Nijstad, eds. Group Creativity: Innovation Through Collabora-tion. Oxford University Press, New York, 110136.

Paulus, P., B. Nijstad, eds. 2003. Group creativity: An introduction.Group Creativity: Innovation Through Collaboration. Oxford Uni-versity Press, New York, 312.

Perry-Smith, J. E., C. E. Shalley. 2003. The social side of creativity:A static and dynamic social network perspective. Acad. Man-agement Rev. 28 89106.

Ratliff, E. 2006. Me translate pretty one day. Wired 14(12) 210213.

Romer, P. 1993. Ideas and things: The concept of production is beingre-tooled.The Economist 328(7828) 7072.

Runco, M. 1999. Creativity need not be social. A. Montuori,R. Purser, eds.Social Creativity, Vol. 1. Hampton Press, Creskill,NJ, 237264.

Runco, M., W. Smith. 1991. Interpersonal and intrapersonal eval-uations of creative ideas. Personality Individual Differences 13295302.

Schumpeter, J. A. 1934.The Theory of Economic Development.HarvardUniversity Press, Cambridge, MA.

Simonton, D. 1985. Quality, quantity, and age: The careers of 10 dis-tinguished psychologists.Internat. J. Aging Human Development21 241254.

Simonton, D. K. 1999. Origins of Genius: Darwinian Perspectives onCreativity. Oxford University Press, New York.

Singh, J. 2005. Collaborative networks as determinants of knowl-edge diffusion patterns. Management Sci. 51(5) 756770.

Singh, J. 2007. Asymmetry of knowledge spillovers betweenMNCs and host country firms. J. Internat. Bus. Stud. 38(5)764786.

Singh, J. 2008. Distributed R&D, cross-regional knowledge inte-gration and quality of innovative output. Res. Policy 37(1)

7796.Smith, A. 1766. Lectures on Jurisprudence. (Reprinted 1982. Lectures

on Jurisprudence(Glasgow Edition of the Works and Correspondenceof Adam Smith, Vol. 5). Liberty Fund, Indianapolis.)

Steinbeck, J. 1952. East of Eden. Putnam Penguin, New York.

Sutton, R., A. Hargadon. 1996. Brainstorming groups in context:Effectiveness in a product design firm. Administration Sci.Quart.41(4) 685718.

Taylor, A., H. Greve. 2006. Superman or the Fantastic Four? Knowl-edge combination and experience in innovative teams. Acad.

Management J. 49(4) 723740.

Trajtenberg, M. 1990. A penny for your quotes: Patent citations andthe value of innovations. RAND J. Econom. 21 172187.

Trajtenberg, M., G. Shiff, R. Melamed. 2006. The names game:Harnessing inventors patent data for economic research.

NBER Working Paper 12479, National Bureau of Economic,Research, Cambridge, MA.

Uzzi, B., J. Spiro. 2005. Collaboration and creativity: The smallworld problem. Amer. J. Sociol. 111 447504.

Watts, D., S. Strogatz. 1998. Collective dynamics of small-world net-works. Nature 393 440442.

Weitzman, M. 1998. Recombinant growth. Quart. J. Econom. 113331360.

Wuchty, S., B. Jones, B. Uzzi. 2007. The increasing dominanceof teams in production of knowledge. Science 316(5827)10361039.

Singh_fleming_2010_lone Inventors as Soruce of Innovation

Documents

Transcript of Singh_fleming_2010_lone Inventors as Soruce of Innovation