Dahlander & Wallin (2006) Unlocking Communities As Complementary Assets

Research Policy 35 (2006) 1243–1259
A man on the inside: Unlocking communities as

complementary assets
Linus Dahlander a , Martin W. Wallin b,∗
a Innovation Studies Centre, Tanaka Business School, Imperial College London, South Kensington Campus, London SW7 2AZ, UK
b Department of Technology Management and Economics, Chalmers University of Technology, 412 96 Gothenburg, Sweden
Available online 20 October 2006
Abstract
Since Teece’s seminal paper explaining who were the gainers from technological innovation, increased globalization and the
information and communication technology revolution have brought new ways for firms to organize and appropriate from innovation.
A new more open model of innovation suggests that firms can benefit from sources of innovation that stem from outside the firm.
The central theme of this paper is how firms try to unlock communities as complementary assets. These communities exist outside
firm boundaries beyond ownership or hierarchical control. Because of practices developed by communities to protect their work,
firms need to assign individuals to work in these communities in order to gain access to developments and, to an extent, influence the
direction of the community. Using network analysis we show that some software firms sponsor individuals to act strategically within
a free and open source software (FOSS) community. Firm sponsored individuals interact with more individuals than interact with
them, and also they seek to interact with central individuals in the community. However, we can see differences in how individuals
interact, depending on whether their affiliation is with a dedicated FOSS firm or an incumbent in the software industry. Apparently,
some firm managers believe they need ‘a man on the inside’ to be able to gain access to communities.
© 2006 Elsevier B.V. All rights reserved.
Keywords: Complementary assets; Open innovation; Free and open source software; Social network theory
1. Introduction changed how managers need to think about these con-

cepts. We will make use of two of these concepts: the
The original aim of Teece’s (1986) paper was to appropriability regime and complementary assets.
explain why some firms fail to appropriate returns from The developments over the last two decades provide
innovation while others succeed. Teece uses three con- for simultaneous connect and disconnect between firms,
ceptual building blocks: the appropriability regime, com- bringing about new ways for firms to organize and appro-
plementary assets, and the dominant design paradigm. priate from innovation. In this paper we explain how the
We maintain that these building blocks are central to introduction of new ICTs can change the organization
the analysis of innovation in the 21st century. However, of innovation in the software industry through the use
over the years increased globalization and the informa- of free and open source software (FOSS). FOSS is an
tion and communication technology (ICT) revolution has extreme case of open innovation (Chesbrough, 2003),
developed by geographically dispersed communities that
consist of individuals who typically receive no financial
∗ Corresponding author. Tel.: +46 31 7721218; fax: +46 31 7721237.
compensation. Although FOSS can be used as inputs in
E-mail addresses: [email protected] (L. Dahlander), embedded devices to develop various kinds of products
[email protected] (M.W. Wallin). (Henkel, 2006), it can also be used to develop new soft-
0048-7333/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.respol.2006.09.011
1244 L. Dahlander, M.W. Wallin / Research Policy 35 (2006) 1243–1259
ware or services. The new organization of innovation in we do not analyze relationships between ﬁrms (which is
this part of the software industry calls for a question- the unit of analysis in the alliance literature), but rather
ing of the dichotomy between core and complementary we look at relationships between different individuals.
assets, which was fundamental in Teece’s seminal paper. However, we are interested in the role of ﬁrms in par-
In this paper we analyze how firms manage inno- ticipating in a specific kind of network structure—a user
vation across firm borders without ownership of key community. These communities, though, have developed
resources. We propose that some firms act strategically practices that prevent firms from claiming ownership, but
by claiming favourable network positions in a FOSS enable firms to sponsor peers or allow personnel to work
community. Drawing on social network theory (Burt, in the community. This is a key characteristic of FOSS
1992; Wasserman and Faust, 1994), we argue that these and an important determinant of how firms can interact
network positions enable firms to manage key resources and draw on resources developed within these commu-
without ownership in the legal sense. nities.
In the corporate governance literature it is often Our aim in this paper is to investigate sponsorship
argued that separation of ownership from control is an as an alternative way of organizing innovation in FOSS
efficient means to run corporations because it enables software development. We do this by looking at the indi-
mutual specialization of duties between risk-bearers and viduals affiliated to certain firms (i.e. being sponsored)
managers. In other words, ownership does not necessar- and examining how these individuals act in a user com-
ily indicate control. Similarly, we argue that control does munity, working on developing a particular software
not necessarily indicate ownership. There are other ways solution—GNOME. Our main data source consists of
in which firms can orchestrate resources in addition to 1659 individuals who sent 14,644 emails in a mailing
legal ownership, or hierarchy in the firm. Conventional list over a 6 year period. We gathered information from
wisdom, heavily influenced by the legal perspectives of the emails about who sent them and to whom. From this
the firm, is to internalize critical resources, i.e. to control unique data set we were able to reconstruct the social
by ownership. Today, innovation is seen as the prime networks by year, and to analyze the role that firm spon-
determinant of competitive advantage, which suggests sored individuals play in the community.
the need for control of critical aspects of the innova- The paper is structured as follows. Section 2 out-
tion process. However, recent developments along the lines the conceptual background to how we can think
open innovation paradigm (Chesbrough, 2003) suggest about communities as a complementary asset. Viewing
that firms need to reject the idea that control implies communities in this way helps to uncover how firms
ownership and open themselves up to the broad array participate in them. In Section 3 we develop a set of
of resources available to the firm. To do this, we argue, hypotheses related to how the communication patterns
managers must find new ways to conceptualize the ‘post- differ between firm sponsored individuals and hobby-
Chandlerian firm’ where innovation proceeds along less ists. Section 4 describes the research setting and the data
hierarchical lines (Langlois, 2003) since “the network of and Section 5 presents the results of firms’ participation
relationships between the firm and its external environ- in the community under study. Finally, we discuss our
ment can play an important role in shaping performance” findings and draw conclusions from the study.
(Laursen and Salter, 2006: 132). This raises important
questions about what is the core of the firm in a relational
2. Conceptual background
sense (Dyer and Singh, 1998; Lorenzoni and Lipparini,
1999; Dyer and Nobeoka, 2000; Gulati et al., 2000). Dyer
Teece’s (1986) seminal paper on profiting from tech-
and Singh (1998) contend that firms can generate rela-
nological innovation provided a framework for analyzing
tional rents by leveraging the complementary assets of
the rents accruing from innovation in terms of the appro-
alliance partners. We take their suggestion as a spring-
priability regime, complementary assets and dominant
board to investigate FOSS communities as a potential
design. Since Teece published this paper several new
complementary asset that firms can use in the innova-
trends have emerged, the outcome of which has become
tion processes.
what we now term open innovation. Chesbrough (2006:
The literature on alliances abounds with reasons for
2) defines open innovation as:
collaboration, e.g. reduced cost and risks of technolog-
ical development or market entry, scale economies in the use of purposive inflows and outflows of knowl-
production, reduced time to market, network effects, and edge to accelerate internal innovation, and expand the
opportunities for organizational learning (Gulati, 1998). markets for external use of innovation respectively.
The phenomenon under study here is different, though; Open innovation is a paradigm that assumes that firms
L. Dahlander, M.W. Wallin / Research Policy 35 (2006) 1243–1259 1245
can and should use external ideas as well as internal grated firms, then, should be the best positioned to benefit
ideas, and internal and external paths to market, as from existing complementary assets.
they look to advance their technology. The highly influential resource-based view in strate-
gic management contends that the firm consists of a
The key drivers of open innovation are: (1) global- bundle of resources, which are commonly defined as the
ization, i.e. the extension of geographical markets that stock of available factors that the firm owns and con-
enables an increased division of labour and specializa- trols (Amit and Schoemaker, 1993; Wernerfelt, 1984).
tion; (2) improved market institutions, e.g. intellectual Resources can be tangible – e.g. plants, equipment, nat-
property rights (IPR), venture capital (VC), and tech- ural resources, raw materials, finished goods – or intan-
nology standards that enable resources to more eas- gible, e.g. knowledge, patents, brand names. In this view,
ily transcend firm boundaries; (3) technological change the firm’s rationale is to use these resources, combining
which reduces the minimum efficient scale of produc- them into products and services, to maximize returns
tion that fuels increased specialization; (4) increased over time (Barney, 1986). This overly internal perspec-
labour market mobility, especially for specialist per- tive of the firm is somewhat mitigated by Dyer and
sonnel. In addition, technological change, such as the Singh’s (1998: 680) argument that critical resources may
ICT revolution, has affected the way firms communi- span firm boundaries and may be “embedded in inter-
cate, collaborate and coordinate. It is no surprise that firm resources and routines”. To be successful, the firm
the software industry itself has been greatly affected needs to generate relational rents. Interestingly, Dyer
by ICT developments because: (1) product duplication and Singh propose that one way for firms to generate
can be made electronically at marginal costs, i.e. there relational rents is to leverage the complementary assets
are almost no production costs; (2) transportation of of their alliance partners. That is, firm resources need
products can be achieved by electronic transmission at to be combined with complementary assets from out-
marginal costs, i.e. there are almost no logistical costs; side the firm to yield rents. Teece et al. (1997) propose a
(3) there is an installed base of generic hardware that similar argument about the role of dynamic capabilities
can run the new software, which results in fewer com- to build, integrate and reconfigure internal and external
patibility problems. As a consequence, new forms of competencies.
organizing innovative work are possible, with a greater In this paper we put forward the idea that a user
division of labour. Hand in hand with increased division community can be seen as a complementary asset. The
of labour comes the need for new mechanisms to inte- work developed in the community can be used in con-
grate tasks and responsibilities (Lawrence and Lorsch, junction with the firm’s internal expertise to develop
1967). The incentive to draw from external knowledge competitive products and services. Firms that engage
sources is particularly high when the industry is charac- in these communities, therefore, have a certain type of
terized by high levels of technological opportunity and business model (Chesbrough and Rosenbloom, 2002;
there are extensive investments in searches by other firms Chesbrough, 2003), which works as a cognitive script
(Klevorick et al., 1995), which is the case in the software and shapes the mindset of the firm towards looking for
industry. ideas in the community. Although this engagement in
Implicit in Teece’s framework is the divide between the community creates value for the firm, it is more dif-
development and production. There is a notion that ficult to appropriate because competitors may interfere.
inventions (ideas) are turned into innovations by their Because of the strong norms and mechanisms against
being funnelled through the value chain from product institutional protection in FOSS it is not possible to uti-
development, production and marketing. During this lize secondary markets for technology to gain access
process, complementary assets become key assets in and control. Business models are essentially systems
weak appropriability regimes. Traditionally, comple- for turning investments into revenue streams, but due
mentary assets are distinct firm level assets with well- to the mechanisms against strong institutional protec-
defined positions in the value chain. In the case of FOSS tion in FOSS communities direct profiting is probably
development the value chain is disintegrated. Individuals not a viable strategy. Firms need to find means of get-
can be contributors of code as well as early adopters of ting access to the developments in the community and
downstream products. In that sense, the community as a need to take indirect measures to convert the knowledge
complementary asset spans input to output. Teece argued created in a FOSS community into a complementary
that in industries with weak appropriability regimes own- asset. We argue that deploying resources, in the form
ership of complementary assets (in particular specialized of employees, to participate in FOSS communities is
complementary assets) determined profit. Fully inte- one way of gaining access to developments. It is impos-
sible to fend off competitors by traditional means, but firm may recognize their potential and make them a
because of the relational advantages to firms from a posi- job offer. Individuals are also driven by peer recogni-
tion in the network (Dyer and Singh, 1998), participation tion: other individuals in the community may value their
is justified. Building on a long tradition of scholarship in work.
how social networks influence innovativeness and per- Lerner and Tirole also address the question of why
formance (Baum et al., 2000; Burt, 2004; Powell et al., firms are involved in FOSS development. They offer
1996; Shan et al., 1994; Stuart, 2000), we develop an the economic rationale of free riding, arguing that firms
argument about how firm sponsored individuals partici- “expect to benefit from their expertise in some segment
pate in a user community. whose demand is boosted by the success of a comple-
von Hippel (1988) adopted a strategy to look at the mentary open source program” (Lerner and Tirole, 2002:
functional source of innovation, and discovered that 224). They also point out that firms may engage in FOSS
innovators are many times users that benefit from the for reasons not associated with revenue streams, such as
innovation by using it, rather than selling it in the mar- public relations and gaining legitimacy, or for reasons
ket place. User innovators are central to the development of learning-programmers who are employees of a firm
effort in being proactive and develop solutions to their participate to learn about their characteristics. In sum,
problems themselves, which firms can then, to vary- Lerner and Tirole see the situation as involving passive
ing degrees, access and use in their internal processes. firms that try to appropriate from spill-overs.
Such user innovations have been identified as frequent In another paper, Lee and Cole (2003) point to
by various researchers in different empirical settings the differences between firm-based and community-
(Freeman, 1968; Rothwell et al., 1974; Rosenberg, 1976; based models of knowledge creation, using the case
von Hippel, 1976, 1988). Sometimes users voluntarily of the Linux kernel. Several authors (Dahlander and
reveal their innovations and build on the work of oth- Magnusson, 2005; Lerner and Tirole, 2002; O’Mahony,
ers in a community in a democratic and cooperative 2003) have recognized that these models have become
manner (von Hippel, 2005). A user community is thus increasingly intertwined. In the community-based
defined as distributed groups of individuals focused on model, firms become gradually more active through
solving a general problem and/or developing a new solu- various means. Thus, when Lee and Cole argue for
tion supported by computer mediated communication. “a firm-based to a community-based model of knowl-
These communities many times safeguard their work edge creation”, we would rephrase this to “a firm and
from being appropriated by firms. By definition, to be community-based model of knowledge creation”. Sim-
able to utilize FOSS communities as a complementary ilarly, von Hippel and von Krogh (2003) suggest that
asset requires firms to themselves posses a substantial FOSS is characterized by successfully merging “private
amount of other resources that could be complemen- investment” and “collective action” into a “best of both
tary. In our own study, the ‘FOSS user community’ is worlds”. In fact, empirical research has provided evi-
referred to as the ‘FOSS community’. At some points dence of how firms allow employees to work on FOSS
we use the epithet ‘GNOME’ to point to the specific on company time, and that as many as 50% of develop-
data set developed for this study. ‘User community’ ers earn income from their engagement in FOSS (David
is used to refer to papers that use this term but that et al., 2003; Ghosh et al., 2002; Mitsubishi Research
do not necessarily deal with FOSS. A single ‘commu- Institute, 2004).
nity’ will be used in general discussion and in relation Relationships between firms have often been ana-
to claims not necessarily limited to user innovation or lyzed from a contractual point of view (Ahuja, 2000;
FOSS. Baum et al., 2000; Gulati, 1998; Powell et al., 1996).
But in the case of FOSS, firms cannot be represented as
3. Hypotheses the legal units of accrual if they want to participate in a
community. FOSS communities have strong interests in
A growing stream of research has analyzed the incen- ensuring that source codes remain free, and have well-
tives for participating in FOSS. Lerner and Tirole (2002) established practices to defend the shared achievements
address some of the economics behind the reasons that of developers. These practices include both social norms
individuals and firms engage in FOSS with seemingly no for source code remaining free, as well as legal mecha-
direct financial reward. Using the idea of signal incen- nisms such as reverse copyright schemes (copyleft) and
tives (Holmstrom, 1999), they propose that individu- non-profit organizations that make it impossible for firms
als are active in communities to increase future career to acquire source code and close it to future users. Firms
opportunities. Through their work in a community a that want to participate in communities therefore must
do so as peers (i.e. they must allow their employees to more diverse set of ideas and resources they can access.
work in these communities as peers). Access to commu- Central actors therefore have a more possibilities to com-
nities allows firms to get access to resources that cannot bine and exchange resources and exploit information
be bought in the market. to their advantage. Conversely, less central actors will
Much of the research on user communities focuses on find it more difficult to syndicate assets and orchestrate
hobbyists who develop and customize software in their responses in a competitive environment (Gnyawali and
spare-time with no direct monetary compensation, i.e. Madhavan, 2001).
hobbyists innovate outside their main fields of occupa- Based on the above reasoning we can summarize why
tion (Jeppesen and Frederiksen, 2006). Given the altruis- sponsored individuals are more active in FOSS. First,
tic story of FOSS communities (Kogut and Metiu, 2001; because they can be; they have a resource advantage
Lee and Cole, 2003) we would expect the vast majority over hobbyists; they can commit time and other resource
of users to work for the common cause and to receive to participate in FOSS problem-solving as part of their
no financial compensation for their involvement in the salaried work. Second, because they have to be. Indi-
community. However, although from this reasoning we viduals within a FOSS community may be suspicious
would expect the number of hobbyists to be greater than of a strong corporate agenda pushed by sponsored indi-
the number of individuals with affiliations to firms, we viduals. To break through the institutional protection
believe that the latter are more active. against firm intervention, sponsored individuals need to
Firm sponsored individuals are defined as individuals build legitimacy. Legitimacy in a professional network
whose work in a community is part of their employment is achieved through proof of skillfulness, and providing
with and affiliation to a commercial company. Individu- help to other individuals in the community.
als with a firm affiliation should have certain advantages
over non-sponsored individuals, i.e. hobbyists working Hypothesis 1. Being sponsored by a firm increases the
in their spare time. A firm sponsored individual can likelihood that an individual seek to form ties with other
devote all of his or her time to working in the commu- participants in the community.
nity. Hobbyists, on the other hand, assuming they need
to make a living elsewhere, cannot devote all of their There are two reasons why firm sponsored individu-
time to the community. This is not to say that hobby- als receive more help than hobbyists. First there is their
ists do not make critical contributions. On the contrary, resource advantage, which enables them to formulate
we expect both hobbyists and firm sponsored individu- interesting questions, which are easy to understand, and
als to make critical contributions to FOSS development answer. Second, providing help to a firm sponsored indi-
projects. Nevertheless, on average, we would expect that vidual is perhaps seen as signalling and may be thought
developers that are sponsored by a firm would devote to increase the chances of being offered employment.
more time, and to interact to a greater extent, in the FOSS Although conceptually distinct measures, due to reci-
community. procity, interacting with other individuals and being con-
Individuals sponsored by firms should have a deeper tacted are usually highly correlated. Because we expect
and more diverse knowledge base due to their resource sponsored individuals to interact frequently with others,
advantages. They have financial resources and can access they are also likely to have more individuals contacting
the results and developments from inhouse investments. them. On the other hand, because of suspicions about
Thus, they can combine the best of both worlds and act firms’ commercial interests, and the altruism of FOSS
as technological brokers (Hargadon and Sutton, 1997) community members, hobbyists could be less keen to
between the community and the firm, which is crucial if interact with firms. Too flagrant attempts by firms to out-
the community is to function as a complementary asset of source simple and non-challenging programming tasks
the firm. It can therefore be expected that firm sponsored will be recognized as such and ignored. Thus we propose
individuals will be able to participate more actively in the that:
functioning of the community.
Social network theory deals with how the structural Hypothesis 2a. Being sponsored by a firm increases
properties of the network influence the behaviour of the likelihood that participants in the community seek to
the nodes in the network. Network benefits can be both form ties with the individual.
information and control type (Burt, 1992). Benefits arise
because valuable information is not evenly distributed Hypothesis 2b. Being sponsored by a firm decreases
in the network. More central network individuals will be the likelihood that participants in the community seek to
the first to recognize new opportunities as they have a form ties with the individual.
Individuals with multiple relationships with other par- also to whom you connect. A tie to an individual who is
ticipants in the community can make their views known well-connected is more valuable than a tie to an individ-
to more people and benefit from exchanges with them. ual with no contacts (Bonacich, 1972, 1987; Friedkin,
Individuals that are ‘receivers’ can be considered to be 1991). Firm sponsored individuals may seek to inter-
prominent participants of the community because other act with well-connected individuals to gain knowledge
individuals seek to form ties with them (Wasserman and about critical developments and push their agenda.
Faust, 1994). The individuals who are most sought after
in the community can thus be said to be prestigious. Hypothesis 4. Being sponsored by a firm increases the
Firm sponsored individuals achieve credibility in the likelihood that an individual is tied to well-connected
community and legitimize their commercial endeavours participants in the community.
by interacting extensively with other individuals of the
community, both to give and to receive help. That is, they 4. Methods
have to invest in the community to gain access and are
thus less prestigious.1 4.1. Research setting
Hypothesis 3. Being sponsored by a firm decreases the We chose for our research setting a large FOSS com-
likelihood that an individual is prestigious in the community to analyze the role of individuals that have affili-
munity. ations with firms that actively participate. Hobbyists that
engage in FOSS communities are volunteers that are out-
Heterogeneity in network centrality measures equips side the hierarchical realm of firms. These communities
individuals with unique opportunities (Freeman, 1979), can encompass thousands of individuals who co-ordinate
which allow sponsoring firms to act strategically in the their tasks world-wide to produce high quality products
network. Sponsored individuals make frequent contri- primarily through computer mediated communication
butions in order to enhance their reputation within the and receive no direct financial compensation for shar-
community and are therefore better placed to watch ing their achievements with others (Lee and Cole, 2003;
developments. These individuals represent themselves von Hippel and von Krogh, 2003). These participants
in the community, but are able to bring expertise to are often regarded as being driven by ideological rather
the firm. Although the community has rules that do not than economic motives (Hertel et al., 2003). Although
allow firms to become formalized leaders in the project, these community participants do not have the organiza-
and licensing practices that prevent firms from clos- tional structure of a firm to work within, the community
ing the source code, sponsored individuals are still able has various mechanisms for governing decisions about
to achieve favourable positions. Network analysis pro- releases, co-ordinating work within the community and
vides more evidence of how FOSS communities give presenting the community to external parties (Fleming
space to firms. In our research setting, for example, Red- and Waguespack, 2005; Lee and Cole, 2003; O’Mahony,
Hat, Sun Microsystems and Ximian (former Helix Code) 2003; O’Mahony and Ferraro, 2004). The principles of
devote full-time paid employees to work on the project governance adopted vary across different FOSS com-
(German, 2003). Some of these people are responsible munities. For instance, the founder of Linux, Linus Tor-
for maintaining critical modules: e.g. RedHat for gtk+ valds, is responsible for the development of the Linux
and CORBA and Ximian for Evolution. kernel and is helped by a number of ‘lieutenants’. Tor-
The number of relationships firm sponsored individu- valds is a ‘wise man’ and he relies upon the trust from the
als have with other developers in a community is interest- individuals in the community. Other FOSS communities,
ing in that individuals who interact frequently with other such as GNOME, have a more formalized governance
individuals may occupy unique and advantaged posi- structure.
tions in the community. Individuals with many ties have All software is written in a programming language
access to many more parts of the network than individu- incorporating commands to enable it to work. When
als with fewer ties. But the number of ties in itself does writing software programs, programmers can add expla-
not necessarily translate into competitive advantage. It is nations and clarifications to make it easier for other
developers to interpret the code. In FOSS, this code is
1
open for anyone to screen, interpret, and evaluate. If
By prestigious, we refer to the social network term of the net
between in-degree and out-degree centrality. Of course, individuals
a user or developer experiences a programming error
that are helping other individuals in the community might be recog- (a so-called bug), they have the opportunity of either
nized by other individuals. fixing it themselves, or sending a detailed bug-report
to the software’s author. FOSS is coordinated through but operating under GPL. GNOME is therefore avail-
the Linux-kernel, which enables numerous projects to able in the public domain for anyone to use, develop
be developed with the objective of producing a solution and diffuse.3 de Icaza sent a draft announcement about
to a given problem through cooperative working. Each GNOME to several friends and mailing lists where he
project comprises a number of modules to which devel- expected to find individuals who would be interested
opers contribute code. FOSS communities vary in size in his idea. GNOME rapidly gained momentum and de
from one person to a few thousand who share and coor- Icaza and the other original participants focused more
dinate their work. Many communities document their and more on its development.
endeavours online, producing a chronology of events to In January 1998, the firm Red Hat initiated its
avoid any reinvention of the wheel. Advanced Development Labs, establishing the first
In our study we analyze the GNOME project funded team of developers to work on GNOME. Sev-
between 1999 and 2004. The GNOME environment is eral notable individuals in the GNOME community
a Windows-like desktop, and the GNOME development took up employment with Red Hat. In October 1999,
platform is a framework for building applications that two GNOME-based companies were incorporated in the
integrate with the rest of the desktop. In the effort to United States, namely Eazel and Helix Code.4 There
produce a user-friendly desktop and development tools, are other examples of corporations that have grown
GNOME has become a community where individuals out of individuals’ participation in GNOME, includ-
interact to solve problems and help each other. GNOME ing Henza and Gnumatic which were incorporated in
is an example of a FOSS community developed col- 2000.
lectively by several thousands of individuals who, to The code remains under a GPL and the community
different degrees, participate and help each other. continues to work collectively, although many individ-
GNOME was set up in 1996 by Miguel de Icaza to cre- uals receive financial compensation for their engage-
ate a collection of free libraries and applications designed ment in GNOME. Because in many ways GNOME is
to make Linux a user-friendly desktop.2 de Icaza and a one of the more hacker-oriented open source initiatives
few of his initial collaborators had had licensing con- which was founded as a reaction to less-open licens-
cerns about KDE, a similar FOSS desktop environment ing practices in another community, it is a good case
that relied on the Qt widget Toolkit developed by the for our study. GNOME, as a result, is an example of an
Norwegian firm Trolltech. Trolltech’s licence, however, extreme case of open innovation. Despite being hacker
did not comply with established FOSS practice and oriented, the number of individuals who have affili-
prevented users from modifying, distributing and redis- ations with firms and receive compensation for their
tributing copies of the code and therefore violated the involvement has increased since the community was
terms of the General Public Licence (GPL). de Icaza founded. As illustrated by the following quote, GNOME
contacted Trolltech to urge them to change their prac- is a research setting in which hobbyists and firm
tices: sponsored individuals jointly work on developing the
project:
Being a free software entusiast, I contacted Troll Tech,
the authors of Qt to propose an alternate licensing Since the project was started in 1998, GNOME has
scheme for Qt that would still allow them to build a grown tremendously. The technology has started to
company while empowering users but got no reply. mature, the amount of code contributed has exploded,
The Troll Tech FAQ at the time also contained sig- the number of developers (also known as GNOME
nificant errors regarding the GPL and ignored dual- hackers) who are contributing to GNOME has more
licensing schemes than doubled. With the emergence of a GNOME
http://primates.ximian.com/∼miguel/gnome- industry, and as corporate partners are embracing
history.html GNOME, the number of hackers who are paid to
When Trolltech refused to adopt a different practice
that would be compatible with FOSS norms, de Icaza ini- 3 O’Mahony (2002) states that licenses such as the GPL are a critical
tiated GNOME to provide similar functionality to KDE, cornerstone for FOSS for three reasons: (1) enabling communities to
collectively pool their work and avoid the risk of its being appropriated
by firms for proprietary use; (2) using the existing legal framework to
protest against proprietary software; (3) providing a normative ground
2 This description of the history of GNOME draws upon Miguel for users and developers so that improvements will be put back on the
de Icaza’s (2005) “The Story of the GNOME project” available at: web.
http://primates.ximian.com/∼miguel/gnome-history.html. 4 Helix Code later became Ximian which was acquired by Novel.
work full-time on GNOME has grown dramatically, interact to solve problems. They use several means such
in addition to the hundreds of volunteer hackers in the as Internet Relay Chat, private emails, and correspon-
community. dence in mailing lists. Participants in the community see
http://foundation.gnome.org/about/charter/ mailing lists as a vital means of communicating (Lee
and Cole, 2003). This was confirmed in interviews and
Because of its rapid growth, the community incor-
from secondary data available at the GNOME web site.
porated a formal structure and a foundation. As already
This influenced our methodology and confirms that ana-
described GNOME is not a collection of wise men who
lyzing activity on mailing lists identifies activity and
make decisions, but has a more formal organizational
relationships among the individuals subscribing to them.
structure. While the foundation is not directly involved
Mailing lists provide a simple communication technol-
in technical decisions, it does coordinate releases and
ogy that broadcasts to a list of subscribers via email.
decide which projects will be part of GNOME. Individ-
When a subscriber receives an email, he or she can
uals who have made significant contributions can apply
choose whether or not to reply to a request. This makes
for membership of the foundation, which needs to be
it possible to keep track of the developments that the
renewed every 2 years. Thus, total number of members
subscribers to the mailing lists discuss. We used longi-
changes every year. The condition for membership is
tudinal data relating to a key mailing list in the GNOME
providing contributions:
community.
Contributions may be code, documentation, trans- The development mailing list (http://mail.gnome.
lations, maintenance of project-wide resources, or org/archives/gnome-devel-list/) is our measure of the
other non-trivial activities which benefit the GNOME network, and contains discussions about future releases,
Project. design, architecture and other technical aspects and prob-
http://foundation.gnome.org/membership/ lem solving. GNOME has several other mailing lists
devoted to various topics. From its beginnings in 1999 to
Members are responsible for electing a Board of
December 2004 we identified in the mailing list 14,644
Directors, and can ask for referenda over any issues
mails on various topics. We cleaned out spam and anony-
under the jurisdiction of the foundation. Members are
mous messages. Authors with multiple identities were
eligible to vote in elections or become elected to the
reconciled through examination of names and emails,
Board of Directors. Individuals with leadership aspira-
and 1659 unique individuals were identified.
tion can announce their candidacy by sending a message
Some research has used mailing lists to follow activ-
to the foundation’s mailing list with their full name,
ity over time and to track major events (Lee and Cole,
e-mail, corporate affiliation if such is applicable, and
2003; von Krogh et al., 2003); we extended this by
reasons for wanting to serve as a director. Declaring
using the data to build social networks where a tie is
corporate affiliations is important. Although the 11 can-
represented by an email sent to a given person. Our
didates that receive the most votes will be elected to
data are unique in this regard, because the few analyses
the Board,5 with the rider that no more than four indi-
that exist on other data sources in FOSS communities
viduals from the same company can be included on the
treat ties symmetrically (O’Mahony and Ferraro, 2004;
board. In the case that there are more than 4 from the
Fleming and Waguespack, 2005). In addition to address-
same company, the four with the most votes will be
ing the neglected issue of how individuals construct
elected. The elected Board of Directors is the primary
affiliations with firms, our directional data allow us to
decision-making body of the GNOME foundation and is
enquire whether individuals receive more help than they
responsible for ratifying all the foundation’s decisions.
give. Relationship data can be directed and undirected,
Board decisions can be overturned by referendum. Ref-
the difference being that in undirected data individuals
erenda must be endorsed by 10% -of the members to be
differ only in the number of ties they have, whereas in
accepted.
directed data it is possible to distinguish for example
between the person that initiates a contact and the one
4.2. Data and variables
that receives an invitation, or the person that asks a ques-
tion and the one that answers it. In other words, more
Our focus on the GNOME project and the role of firms
information is conveyed in a directed than an undirected
in the community allows us to analyze how individuals
relation.
We define ties as the relationships among the indi-
5 After a referendum in September 2005, the members voted to viduals interacting. The network is constructed per year,
reduce the size of the board to seven. with a 3 year window of prior ties. In other words, a
tie disappears from the analysis if individuals cease to ized through the principal eigenvector of the adjacency
interact.6 Using these data on interactions among indi- matrix.
viduals combined with a range of information on the
attributes of the individuals as described below, we were 4.2.2. Independent variables
able to analyze our hypotheses. The independent variables are whether or not indi-
viduals have affiliations with firms that engage in the
4.2.1. Dependent variables GNOME community. Because we could not find any
Using the mailing list data we obtained, we derived available secondary data, we developed our own coding
various measures describing individuals’ interaction pat- scheme. We test for whether an individual has an affil-
terns for the 1659 individuals responsible for a total of iation with a firm, with the variable taking the value 1
14,644 mails, using UCINET 6 (Borgatti et al., 2002) and if the individual is coded as having an affiliation with
Pajek (de Nooy et al., 2005) for the years 1999–2004. We a firm that engages in the community a given year. Our
use four dependent variables to test our hypotheses: (1) data account for that the affiliations of individuals might
out-degree; (2) in-degree; (3) prestige; (4) eigenvector. change over the study period. Individuals may change
Below we explain our choice of measures. to a competitor firm, or take up employment with a firm
The number of relations with other individuals is mea- not engaged in the GNOME community.
sured by degree centrality, which calculates the total We used a combination of methods to retrieve affili-
number of ties with other individuals (Freeman, 1979). ations. Using the information in the emails themselves,
Central to the analysis is the difference between sending we were able to retrieve the domain names of the firms.
an email to one individual and receiving a reply from We coded whether an individual had sent at least one
someone else. Our network is therefore constructed to email from one of the firms that sponsored individuals
grasp directionality. An individual’s prominence, or the in the community in a given year. For the individuals
degree to which others seek to interact with a particu- coded as having affiliations with firms, we also exam-
lar individual, is measured through in-degree (INDEG) ined documentation available at the GNOME website to
centrality (Wasserman and Faust, 1994). In-degree cen- cross-check our results.
trality measures the number of ties that an individual has As already mentioned, many individuals involved in
with other individuals. the GNOME project are employed by firms specifically
The extent to which individuals help others and con- to work on the development of the project. Many individ-
tribute to the development of the community is measured uals have gone on to found firms based on the expertise
through out-degree (OUTDEG) centrality. This mea- gained in the GNOME project. We categorized affilia-
sures the total number of ties that an individual has tions depending on the type of firm: (1) SPONSDED
with other individuals (Wasserman and Faust, 1994). In- representing firms with a business model attuned to the
degree is thus the number of individuals that respond to a GNOME project (Helix Code, Red Hat, Eazel, Gnumatic
given individual’s emails, and out-degree is the number and Henzai) and (2) SPONSOTH representing incum-
of responses an individual makes to others. bents in the software industry that participate in the
The degree to which individuals interact with more GNOME community. By distinguishing between these
individuals than interact with them is measured as the two types of affiliations, it is possible to observe differ-
difference between in-degree and out-degree central- ences in firms’ strategies depending on whether they are
ity. Individuals that have more individuals that seek to explicitly focused on FOSS, or whether FOSS working
interact with them than they seek to interact with are is only a small part of their business. This allows us to
said to be prestigious (PRESTIGE) (de Nooy et al., capture the effect of these different focuses on the depen-
2005). dent variables. It also revealed that no individuals had
Following Bonacich (1972), we measure the extent sent emails from two different organizations in the same
to which an individual is connected to central individ- year, which supports our choice of method. By coding
uals by eigenvector (EIGENVEC) centrality. Eigenvec- the data annually, we can track mobility between differ-
tor centrality can be understood as a variant of degree, ent firms, and the move from hobbyist to firm employee,
where the importance of the nodes to which an indi- and vice versa.
vidual is connected is taken into account, conceptual-
4.2.3. Control variables
A wide range of the social network and innovation lit-
6 We ran the analyses experimenting with different decay times (1, erature proposes that individuals that are boundary span-
2, 3 and 4 years) and our findings are robust. ners (Allen, 1977; Tushman, 1977) and brokers (Burt,
1992) have more diverse mindsets, opinions and knowl- community’s website.7 The practice of having members
edge bases, which allow them to be more creative. Indi- was incorporated in 2000, so we only track membership
viduals that are able to integrate disparate parts of large for the 5 years between 2000 and 2004.
communities, such as GNOME, are arguably more able FOSS brings an open and distributed model of inno-
to interact with a wider range of individuals. Boundary vation that transcends geographical boundaries. Yet, as
spanners, for instance, are more likely to be aware that a shown by O’Mahony and Ferraro (2004), face-to-face
problem has been discussed in another part of the com- interaction is sometimes important in the governance
munity, and can use that information to solve the current of the community. We control for whether LOCATION
problem. Communities such as GNOME typically have influences communication patterns by coding whether
different mailing lists depending on the topic. Individu- an individual is geographically located in the US. This
als that are active in several mailing lists will have more was identified by the time zone from which the mail was
diverse knowledge bases, which will increase their likeli- sent (obviously by controlling for discrepancies from the
hood of being able to solve problems and of being attrac- same individual) and the domain of the emails. We also
tive to interact with. We therefore control for boundary checked various secondary data on the GNOME web site
spanning (BSPANNING) by measuring the number of to develop this measure and distinguish between individ-
mailing lists by year that an individual contributes to. uals from South America versus North America. In the
After excluding mailing lists containing only spam or fixed effects regression, this variable was dropped from
announcements of releases in the community, we were the analyses because of no variation within one panel.
left with 137 possible lists. An individual’s boundary We also decided to drop this from the random effects
spanning measure, therefore, can be an integer between 1 regressions as it had no effect on the results.
and 137. It is should be noted that the development mail- As shown by a number of writers (Hertel et al., 2003),
ing list (http://mail.gnome.org/archives/gnome-devel- FOSS communities are dominated by men. We control
list/) is one of the most active (as measured in the for GENDER by coding a dummy variable taking the
number of emails relative to the total number of emails value 1 if an individual is male. We discovered that 98%
contributed to mailing lists) and most important in the of the individuals in our analysis had male names and
community (subjective statements from participants and only 1% female names. The remaining 1% had gender
information about what the mailing lists deals with from neutral names and were impossible to code. We therefore
the community’s website). dropped this variable from the analyses.8
The experience of participating in the community
it can be argued will influence an individual’s inter-
4.3. Analysis of the data
action pattern. We therefore measure TENURE by the
numbers of years an individual has participated on any
The data are panel or cross-sectional time series
of the GNOME mailing lists. This is a more accu-
data, giving multiple measures per year. Our analyses
rate measure than using only the mailing list under
focus on 1659 individuals and 6020 individual years.
study (the development mailing list) because individu-
In some regressions we lose some observations due to
als might have been active in another mailing list before
the nature of the method. We used cross-sectional time
diversifying into development activities. More experi-
series ordinary least square (OLS) to predict EIGENVEC
enced individuals might have better knowledge about
and PRESTIGE, and cross-sectional time series negative
the latest developments in the project, which might
binomial regressions to predict INDEG and OUTDEG.
have allowed them to be more active and to send more
To control for unobserved heterogeneity, we used fixed
mails.
individual effects. This was confirmed by the Hausman
As pointed out in the description of the research
test of the null hypothesis that the coefficients estimated
setting, GNOME has a formalized hierarchy in which
by the efficient random effects estimator are the same
critical contributors can become members of the com-
as those estimated by the consistent fixed effects esti-
munity. Because members are considered to be critical
mator (Hausman et al., 1984). However, we decided to
contributors they may feel they have an obligation to
show the results using both random and fixed effects as
help resolve problems. Conversely, individuals who are
not members may feel that interacting with members will
increase their chances of becoming part of the hierarchy, 7 Only members of the community can vote in elections and be nom-
which positively influences the degree to which members inated in such elections.
have a high in-degree centrality. We therefore control for 8 The regressions including location and gender in the random effects
whether the individual is a MEMBER according to the regression analyses are available from the authors upon request.
Table 1
Descriptive statistics
Variable Observation Mean S.D. Min Max 1 2 3
1. INDEG 6020 3.94286 9.49825 0 192 1

2. OUTDEG 6020 3.94468 12.6355 0 318 0.9200 1
3. PRESTIGE 6020 −0.001827 5.38871 −126 36 −0.3947 −0.7231 1
4. EIGENVEC 6020 0.015191 0.027534 0 0.38 0.8909 0.8510 −0.4252
5. SPONSDED 6020 0.023754 0.152295 0 1 0.3245 0.3669 −0.2884
6. SPONSOTH 6020 0.011296 0.105688 0 1 −0.0048 0.0061 −0.0227
7. BSPANNER 6020 4.42591 5.36289 1 52 0.4514 0.4991 −0.3747
8. LOCATION 6020 0.455980 0.498100 0 1 0.0789 0.0932 −0.0793
9. MEMBER 6020 0.093356 0.290954 0 1 0.1901 0.2290 −0.2019
10. TENURE 6020 1.84784 1.39485 0 5 0.0834 0.0902 −0.0646
Variable 4 5 6 7 8 9 10
4. EIGENVEC 1
5. SPONSDED 0.3201 1
6. SPONSOTH −0.0064 −0.0167 1
7. BSPANNER 0.4871 0.3365 0.1020 1
8. LOCATION 0.1167 0.1112 −0.0347 0.1492 1
9. MEMBER 0.1866 0.1637 0.0305 0.3417 0.0662 1
10. TENURE 0.0092 −0.0503 0.3272 0.3272 0.0638 0.1848 1
the fixed effects regressions drop observations from the 5. Results

analyses.
We adopted various ways of analyzing the data Fig. 1 illustrates the network during 1999 of the
to achieve robust findings. Several analyses were run GNOME project mailing list that we analyzed. The nodes
including one variable at a time to test for the effects represent individuals that posted an email to the develop-
of the independent variables. To control for robustness, ment mailing list. Nodes that are boxes are individuals
we ran separate analyses for individuals who had written with a firm affiliation and ellipses are hobbyists. The
more than 5 and more than 10 mails, as it could be argued size of the node indicates the degree of activity. The
that the findings are due to ‘casual observers’ who only arcs connecting the nodes are mailing list interactions.
occasionally contribute a mail, but cannot be said to be
active in the community. We discovered no differences
in the results and, for reasons of space, we report here
only the analyses of the full sample.9 We also experi-
mented with different decay times (1, 2, 3 and 4 years)
and similar to other research (Fleming and Waguespack,
2005) opted for 3 years. We analysed a random sample
of 100 individuals to check whether our coding scheme
was robust.
The method in this paper avoids three problems inher-
ent in social network analysis. First, we avoid selection
bias as we are using a complete network and not a
restricted and non-random sample. Second, because of
the completeness of the data, there is no problem with
response bias. Third, we have combined existing data Fig. 1. Illustration of interactions in a GNOME mailing list in 1999.
sources available online, to circumvent any common Note: Illustration made with Pajek’s Fruchterman Reingold 2D algo-
method bias. rithm which minimises stress in the network (de Nooy et al., 2005).
The nodes illustrate individuals and the arcs the interactions among
the individuals. Boxes are individuals with affiliations with firms and
ellipses hobbyist. The sizes of the individuals represent the degree of
9 The results are available from the authors upon request. activity.
1254
Table 2
The effect of sponsoring
Model 1A INDEG Model 2A INDEG Model 1B INDEG Model 2B INDEG Model 3A OUTDEG Model 4A OUTDEG Model 3B OUTDEG Model 4B OUTDEG
BSPANNER 0.0365 (0.0023)*** 0.0353 (0.0023)*** 0.031 (0.0025)*** 0.0305 (0.0024)*** 0.0455 (0.0022)*** 0.0441 (0.0021)*** 0.0394 (0.0023)*** 0.0387 (0.0023)***
TENURE −0.0792 (0.0072)*** −0.0737 (0.0072)*** −0.076 (0.0074)*** −0.0718 (0.0073)*** −0.0781 (0.0076)*** −0.0711 (0.0076)*** −0.0742 (0.0077)*** −0.0689 (0.0077)***
MEMBER 0.2501 (0.0310)*** 0.2367 (0.0306)*** 0.2406 (0.0318)*** 0.2297 (0.0314)*** 0.2728 (0.0298)*** 0.2579 (0.0294)*** 0.2557 (0.0304)*** 0.244 (0.0301)***
0.3241 (0.0522)*** 0.29 (0.0541)*** 0.2916 (0.0483)*** 0.2488 (0.0498)***
L. Dahlander, M.W. Wallin / Research Policy 35 (2006) 1243–1259

SPONSDED
SPONSOTH −0.0881 (0.1452) −0.059 (0.1549) 0.0074 (0.1440) −0.0006 (0.1542)
CONSTANT 2.6959 (0.0793)*** 2.7291 (0.0815)*** 2.677 (0.0804)*** 2.7059 (0.0825)*** 2.4954 (0.0748)*** 2.5186 (0.0767)*** 2.53 (0.0769)*** 2.5511 (0.0788)***
Effects Random Random Fixed Fixed Random Random Fixed Fixed
Number of individual years 6020 6020 4328 4328 6020 6020 3933 3933
Number of individuals 1659 1659 1060 1060 1659 1659 951 951
ln r 3.664 3.807 3.283 3.398
ln s 0.502 0.502 0.422 0.423
Log likelihood −10729.4 −10711 −5534.62 −5521.05 −9750.76 −9733.57 −4942.82 −4930.98
Wald Chi2 349.382 419.075 237.171 288.263 585.692 664.46 389.634 445.014
Model 5A PRESTIGE Model 6A PRESTIGE Model 5B PRESTIGE Model 6B PRESTIGE Model 7A EIGENVEC Model 8A EIGENVEC Model 7B EIGENVEC Model 8B EIGENVEC
BSPANNER −0.3016 (0.0126)*** −0.2952 (0.0126)*** −0.297 (0.0153)*** −0.2963 (0.0153)*** 0.0014 (0.0001)*** 0.0014 (0.0001)*** 0.001 (0.0001)*** 0.001 (0.0001)***
TENURE 0.1685 (0.0268)*** 0.1475 (0.0272)*** 0.1589 (0.0278)*** 0.1469 (0.0280)*** −0.0012 (0.0001)*** −0.0011 (0.0001)*** −0.001 (0.0001)*** −0.001 (0.0001)***
MEMBER −0.954 (0.1336)*** −0.9051 (0.1342)*** −0.8592 (0.1360)*** −0.8301 (0.1362)*** 0.0009 (0.0006) 0.0006 (0.0006) 0.0005 (0.0006) 0.0003 (0.0006)
SPONSDED −1.7015 (0.2893)*** −1.0121 (0.3015)*** 0.0096 (0.0013)*** 0.0055 (0.0014)***
SPONSOTH −0.0656 (0.4239) −0.1844 (0.4495) −0.0047 (0.0020)** −0.0043 (0.0020)**
CONSTANT 1.2039 (0.1117)*** 1.2426 (0.1106)*** 1.0991 (0.0722)*** 1.1417 (0.0736)*** 0.0086 (0.0005)*** 0.0084 (0.0005)*** 0.0126 (0.0003)*** 0.0125 (0.0003)***
Effects Random Random Fixed Fixed Random Random Fixed Fixed
Number of individual years 6020 6020 6020 6020 6020 6020 6020 6020
Number of individuals 1659 1659 1659 1659 1659 1659 1659 1659
Sigma e 2.315 2.313 2.315 2.313 0.01 0.01 0.01 0.01
Sigma u 3.702 3.61 3.956 3.937 0.018 0.018 0.02 0.02
rho 0.719 0.709 0.745 0.743 0.755 0.748 0.792 0.79
R-square within 0.15 0.165 0.15 0.159 0.239 0.263 0.236 0.257
R-square between 0.094 0.095 0.094 0.096 0.047 0.052 0.048 0.053
R-square overall 0.121 0.135 0.12 0.129 0.207 0.23 0.203 0.223
Wald Chi2 689.93 730.795 568.44 641.318
R-squared 0.09 0.1 0.05 0.05
Note: Standard errors in parentheses. * Significant at 10%.
** Significant at 5%.
*** Significant at 1%.
Table 3 DED are compared to SPONSOTH and hobbyists in the

Comparing means of dependent variables in the three groups community.
Variable SPONSDED SPONSOTH HOBBYIST
1. INDEG 23.6993 3.514706 3.461525

6. Discussion
2. OUTDEG 33.66434 4.661765 3.204682
3. PRESTIGE −9.965035 −1.147059 0.2568428 The open innovation literature has provoked numer-
4. EIGENVEC 0.0716923 0.0135441 0.0138196 ous interesting questions about how firms can benefit
Note: Mean values reported, N = 6020. from innovations emerging from outside firm boundaries
(Chesbrough, 2003; Chesbrough et al., 2006). The cen-
tral theme in this paper is how firms try to unlock com-
munities which for them are a complementary asset that
Individuals at the centre of the network are seen to be
exists outside firm boundaries and outside their owner-
more connected than those near the periphery. The fig-
ship or hierarchical control. As explained in the concep-
ure shows that the distribution is heavily skewed. There
tual framework, firms participate in FOSS communities
is a handful of individuals that are responsible for the
because they feel that they can influence the direction
majority of contributions, and there are isolates on the
of development, gain legitimacy to use the software, and
fringe that never receive responses.10
benefit from the expertise of a large base of skilled users.
Table 1 presents the descriptive statistics. There is a
Because of licensing practices in the FOSS community
high correlation between the dependent variables. Mul-
that protect the software from being appropriated by
ticollinearity is not a problem because these are the
organizations, firms allow dedicated personnel to work in
dependent variables and are treated separately.
these communities. Despite the fact that competitors will
Table 2 presents the results from eight models predict-
also likely be involved, firms are keen for their employ-
ing INDEG, OUTDEG, PRESTIGE, and EIGENVEC.
ees to occupy critical positions in the project in order to
Each hypothesis is tested first by including only the
keep track of developments in the FOSS community.
control variables, and then including the independent
Prior research on open source communities has typi-
variable to illustrate the difference in the full model, from
cally depicted this collective participation as altruistic
the baseline. The models with A as an extension refer to
and collaborative (Lee and Cole, 2003; von Hippel and
results from the random effects, and those with B refer
von Krogh, 2003), but we show how firms participate in
to fixed effects.
seemingly open user communities. Thus we contribute to
Models 1A–1B, respectively, predict INDEG using
a more complete understanding about how these commu-
only the control variables with random and fixed effects.
nities operate and the role of individuals with affiliations
Models 2A–2B add the SPONSDED and SPONSOTH
to firms.
variables. They show that SPONSDED is positively
The empirical findings that can be drawn directly from
associated with INDEG and is significant. Models
the dataset derive from the differences in various mea-
3A–3B show the OUTDEG baseline model. Models
sures of firm sponsored individuals and hobbyists. We
4A–4B include the independent variables and again show
observed a heavily skewed distribution where a few indi-
that only SPONSDED is significant. Models 5A–5B
viduals, who could be seen as a group of experts, are
are the baseline models for PRESTIGE. Models 6A–6B
responsible for the majority of the emails. Although the
show that SPONSDED are less prestigious in network
largest proportion of the GNOME community is made
terms, suggesting that they interact with more individuals
up of hobbyists working in their spare time and receiv-
than interact with them. Models 7A–7B are the base-
ing no direct financial compensation, firm sponsored
line models for EIGENVEC. Models 8A–8B include the
individuals are very important. Firm sponsored individ-
independent variables and show that SPONSDED are
uals have certain resource advantages which can help
connected to more influential persons in the community
them gain favourable network positions; however how
and SPONSOTH to less important ones.
firms actively make use of such advantage in FOSS com-
To show the differences between SPONSOTH and
munities has not previously been investigated. Because
SPONSDED and the other individuals in the community,
of their resource advantage, firm sponsored individu-
we report the mean values of the dependent variables in
als can interact with many different individuals for the
Table 3. This clearly illustrates how extreme SPONS-
good of the project, which enhances their reputation
and legitimizes their commercial use of the project. Our
10 The figure shows sizes depending on the number of ties. Isolates results suggest that individuals sponsored by firms work-
are therefore not in the picture. ing exclusively on the GNOME project interact with a
greater number of individuals and that more individuals of the firm. Some firms have developed practices to
seek to interact with them. At the same time the number manage these resources by investing in community activ-
of individuals they interact with is greater than the num- ities. These practices, in turn, result in a blurring of the
ber of individuals who contact them. This can probably concepts of core and complementary assets. Accord-
be explained by the fact that these sponsored individuals ing to Teece (1986) a complementary asset is an asset
are keen to establish their legitimacy in the community which is needed in conjunction with the invention to
and also that other community participants may be sus- make sure it is possible to appropriate from the inven-
picious about their corporate agenda. tion, e.g. where markets do not or cannot exist. Under
In spite of the resources invested by companies to weak intellectual property protection, i.e. when there
become active members in FOSS communities, there is is a risk of knowledge spillovers to competitors, Teece
no clear evidence that firm sponsored individuals are able argued that the complementary assets of the firm are cru-
to orchestrate or stimulate debate within these commu- cial for appropriating the benefits from its innovations.
nities. To make a first analysis of how firms take part in He provided a few examples of complementary assets
these communities we make a distinction between FOSS such as specialist manufacturing, access to distribution
companies that have incorporated open source in their channels, service networks and complementary tech-
business models, and other firms in the software industry nologies. Firms can thus deploy complementary assets
that have devoted personnel to working in a community. and combine them with knowledge under weak appro-
This distinction is shown to be quite important. Individu- priability regimes to create successful and appropriable
als with affiliations with large incumbents in the software innovation. Teece suggested that firms supply the miss-
industry have no significant effect in the community. The ing pieces through their competencies. Firms start from
influence of SPONSDED (individuals sponsored by ded- what is there and identify what is missing. As the inven-
icated FOSS firms) however, was shown to be significant tion is channelled through complementary activities the
in the models, and the findings suggest that the employ- inventor can adjust its business model and profit from its
ees of firms who set out to draw from these communities invention even under a weak appropriability regime. The
are likely to be more active and important in the com- distinction between core versus complementary assets,
munity. Other software companies may not devote their however, assumes that the activities leveraged by these
best employees to working in the community or may only assets can in fact be separated in practice. From a the-
passively screen developments. Firm sponsored individ- oretical perspective, the distinction requires some kind
uals with an affiliation to dedicated firms are also likely to of atomistic perspective of knowledge. This is perhaps
be associated with well-connected individuals, while the particularly significant in the software industry, which
reverse is true for individuals affiliated to incumbents. is characterized by the difficulty of separating develop-
We suggest that firms use communities as comple- ment from production, and users from producers; thus
mentary assets. To gain access and legitimacy firms must a clear separation of complementary from core assets is
deploy employees to interact with other participants in challenging.
the community. The community as a complementary Should a clear separation between core and comple-
asset cannot be acquired through the market, but requires mentary assets emerge, the question of control through
participation, interaction and learning. Firm sponsored ownership remains. We speculate that more attention
individuals are to a greater extent than hobbyists con- needs to be paid to the managing of the key functions
nected to influential individuals in the community mea- in the firm that are necessary to appropriate returns from
sured in social network terms, which lends support to innovation, as opposed to questions of ownership of cer-
the thesis that firms to some extent can influence the tain resources, be they core or complementary. This is
direction of development in these communities. From of particular importance under situations beyond what
a broader perspective, our analysis raises a number of was covered by Teece (1986), i.e. situations where nei-
important questions concerning core and complemen- ther integration strategies nor contractual strategies are
tary assets, as well as ownership and control. First, it possible as is the case in FOSS. The challenge for man-
informs us about a general difficulty in the software agement is not in deciding which part of the integrated
industry to distinguish between core and complemen- value chain to own or not to own but to decide which
tary assets. Second, it informs us of a specific difficulty resources in a dissolving value chain it is necessary to
about ownership and control when private ownership of own in order to be able to manipulate internal resources
important assets is banned. to perform functions that will gain it access to the exter-
Obviously, as the alliance literature has taught us, crit- nal world. This facilitates access to crucial complemen-
ical resources are to be found outside the legal reach tary assets while at the same time providing gains in
the form of increased division of labour and increased these communities. Firms’ involvement in FOSS may
specialization. In the spirit of Teece (1986) we have ana- raise quality standards and create commitment incentives
lyzed complementary assets as a means to appropriate for developers not interested in the commercial aspects
from innovation. But there are also risks associated with of programming. As a result virtuous circles are formed.
innovation. In a recent paper, Hopkins and Nightingale However, it is a delicate balance. Through their active
(2006) argue that firms can use complementary assets participation, firm sponsored individuals gain central
to disappropriate the risks of technical change. Firms network positions in the FOSS community. And FOSS
can position themselves in networks and “disappropri- communities have safeguards against direct firm control.
ate some of the uncertain or undesirable consequences of The question of to what degree firms can make use of
innovation onto third parties” (Hopkins and Nightingale, this advantage to orchestrate the community remains to
2006: 361). In other words, firms can employ active be answered.
(appropriation) and passive (disappropriation) strategies. To summarize, we have demonstrated how firms work
Potentially, this can help explain the different strate- in a seemingly open and collective development pro-
gies employed by dedicated and incumbent firms in cess illustrating an extreme case of open innovation
the FOSS community. It might be the case that for distributed across time and space. But, more importantly,
SPONSOTH, being part of a community like GNOME we have also provided evidence of how firms deploy
allows the firms to disappropriate and transform risk, resources in the form of experts to get critical access to
while for SPONSDED it is a means of appropriating developments in the community. Apparently firm man-
benefits. agers believe they need ‘a man on the inside’ to get the
Our study has some limitations, and there are many job done.
opportunities for further research on this topic. We set
out to explore an area that in the past has been rather
superficially examined. Much effort went to coding indi- Acknowledgements
viduals’ affiliations with firms. A problem that emerges
from our study is that of endogeneity. The hypotheses are Comments from Paola Criscuolo, Simone Ferriani,
such that affiliations with firms influence the dependent Toke Reichstein, Ammon Salter and workshop partici-
variables. In practice, one should also assume that cen- pants at Cass Business School and Chalmers University
tral individuals are employed by firms because they have of Technology are greatly acknowledged. The advice
shown that they have the skills necessary to influence a and encouraging words from the Editor Henry Ches-
community. Much of this problem was overcome by ana- brough are appreciated. Three anonymous referees pro-
lyzing how positions in the network and affiliations with vided excellent feedback on earlier versions of the paper.
firms are associated, using panel data covering 6 years in Dahlander’s research was carried out in the Innova-
the GNOME project. Yet, this supports the main under- tion and Productivity Grand Challenge with financial
lying argument that even though hobbyists are greater in support from the Engineering and Physical Sciences
number than sponsored individuals, they are on average Research Council and the Economic and Social Research
less important to the community. Council through the AIM initiative. Wallin acknowl-
Lerner and Tirole’s (2002) argument is that individ- edges financial support from RIDE research centre at
uals participate in FOSS communities to enhance their Chalmers University of Technology and from the Euro-
career opportunities. In other words, important individu- pean Commission (FP6) Project KEINS - Knowledge-
als may become hired by firms. We believe that there are Based Entrepreneurship: Innovation, Networks and Sys-
other ways in which affiliation with firms and positions tems, contract number: CT2-CT-2004-506022. Errors
in networks come about. First, individuals that become and omissions remain our own.
critical contributors in the community can see the com-
mercial value of their endeavours and may establish their References
own firms. These individuals can both draw on their con-
siderable human capital in terms programming skills, Ahuja, G., 2000. Collaboration networks, structural holes and inno-
and also on their relational capital knowing whom to con- vation: a longitudinal study. Administrative Science Quarterly 45,
tact or whom to hire. Second, already established firms 425–455.
can dedicate individuals to work on projects in commu- Allen, T., 1977. Managing the Flow of Technology: Technology Trans-
fer and the Dissemination of Technological Information within the
nities to trace their developments. By making critical R and D Organisation. The MIT Press, Cambridge, MA.
contributions to the FOSS community firm sponsored Amit, R., Schoemaker, P.J.H., 1993. Strategic assets and organizational
individuals can catalyze the development processes in rent. Strategic Management Journal 14 (1), 33–46.
Barney, J., 1986. Strategic factor markets: expectations, luck, and busi- Ghosh, R.A., Glott, R., Krieger, B., Robles, G., 2002. Free/libre
ness strategy. Management Science 32, 1231–1241. and Open Source Software: Survey and Study. Part IV: Survey
Baum, J.A.C., Calabrese, T., Silverman, B.S., 2000. Don’t go it of Developers. International Institute of Infonomics/Merit, Maas-
alone: alliance network composition and startups’ performance in tricht, http://flossproject.org/report/.
Canadian biotechnology. Strategic Management Journal 21 (3), Gnyawali, D.R., Madhavan, R., 2001. Cooperative networks and
267–294. competitive dynamics: a structural embeddedness perspective.
Bonacich, P., 1972. Factoring and weighting approaches to status Academy of Management Review 26 (3), 431–445.
scores and clique identification. Journal of Mathematical Sociol- Gulati, R., 1998. Alliances and networks. Strategic Management Jour-
ogy 2, 113–120. nal 19, 293–317.
Bonacich, P., 1987. Power and centrality: a family of measures. Amer- Gulati, R., Nohira, N., Zaheer, A., 2000. Strategic networks. Strategic
ican Journal of Sociology 92, 1170–1182. Management Journal 21, 203–215.
Borgatti, S.P., Everett, M.G., Freeman, L.C., 2002. UCINET for Win- Hargadon, A., Sutton, R.I., 1997. Technology brokering and innovation
dows: Software for Social Network Analysis. Analytical Technolo- in a product development firm. Administrative Science Quarterly
gies, Harvard. 42 (4), 716–749.
Burt, R.S., 1992. Structural Holes: The Social Structure of Competi- Hausman, J., Hall, B.H., Griliches, Z., 1984. Econometric models for
tion. Harvard University Press, Cambridge, MA. count data with an application to the patents R&D relationship.
Burt, R.S., 2004. Structural holes and good ideas. American Journal Econometrica 52, 909–938.
of Sociology 110 (2), 349–399. Henkel, J., 2006. Selective revealing in open innovation processes: the
Chesbrough, H., 2003. Open Innovation: The New Imperative for Cre- case of embedded Linux. Research Policy 35 (7), 953–969.
ating and Profiting from Technology. Harvard Business School Hertel, G., Niedner, S., Herrmann, S., 2003. Motivation of soft-
Press, Boston, MA. ware developers in open source projects: an internet-based survey
Chesbrough, H., 2006. Open innovation: a new paradigm for under- of contributors to the Linux kernel. Research Policy 32, 1159–
standing industrial innovation. In: Chesbrough, H., Vanhaverbeke, 1177.
W., West, J. (Eds.),Open innovation: a new paradigm for under- Holmstrom, B., 1999. Managerial incentive problems: a dynamic per-
standing industrial innovation. Open Innovation: Researching a spective. Review of Economic Studies 66, 169–182.
New Paradigm. Oxford University Press, Oxford. Hopkins, M.M., Nightingale, P., 2006. Strategic risk management
Chesbrough, H., Vanhaverbeke, W., West, J., 2006. Open Innovation: using complementary assets: organizational capabilities and the
Researching a New Paradigm. Oxford University Press, Oxford. commercialization of human genetic testing in the UK. Research
Chesbrough, H., Rosenbloom, R.S., 2002. The role of the business Policy 35, 355–374.
model in capturing value from innovation: evidence from Xerox Jeppesen, L.B., Frederiksen, L., 2006. Why do users contribute to firm-
Corporation’s technology spin-off companies. Industrial and Cor- hosted user communities? The case of computer-controlled music
porate Change 11 (3), 529–555. instruments. Organization Science 17 (1), 45–63.
Dahlander, L., Magnusson, M.G., 2005. Relationships between open Klevorick, A.K., Levin, R.C., Nelson, R.R., Winter, S.G., 1995. On the
source software companies and communities: observations from sources and significance of interindustry differences in technolog-
Nordic firms. Research Policy 34 (4), 481–493. ical opportunities. Research Policy 24, 185–205.
David, P.A., Waterman, A.H., Arora, S., 2003. First full findings Kogut, B., Metiu, A., 2001. Open-source software development and
from FLOSS-US: The free/libre and open source developer distributed innovation. Oxford Review of Economic Policy 17 (2),
survey for 2003. SIEPR working paper. Available at: http://www- 248–264.
cepr.stanford.edu/programs/OpenSoftware David/David- Langlois, R.N., 2003. The vanishing hand: the changing dynamics
Waterman-Arora(2003) abs.html. of industrial capitalism. Industrial and Corporate Change 12 (2),
de Icaza, M., 2005. The Story of the GNOME project. Available at: 351–385.
http://primates.ximian.com/∼miguel/gnome-history.html. Laursen, K., Salter, A.J., 2006. Open for innovation: the role
de Nooy, W., Mrvar, A., Batagelj, V., 2005. Exploratory Social Network of openness in explaining innovation performance among UK
Analysis with Pajek. Cambridge University Press, New York. manufacturing firms. Strategic Management Journal 27, 131–
Dyer, H., Singh, H., 1998. The relational view: cooperative advan- 150.
tage and sources of interorganizational competitive advantage. Lawrence, P.R., Lorsch, J.W., 1967. Differentiation and integration in
Academy of Management Review 23, 660–679. complex organization. Administrative Science Quarterly 12 (1),
Dyer, J.H., Nobeoka, K., 2000. Creating and managing a high- 1–47.
performance knowledge-sharing network: the Toyota case. Strate- Lee, G.K., Cole, R.E., 2003. From a firm-based to a community-based
gic Management Journal 21, 345–367. model of knowledge creation: the case of the Linux kernel devel-
Fleming, L., Waguespack, B., 2005. Penguins, camels, and other birds opment. Organization Science 14 (6), 633–649.
of a feather: brokerage, boundary spanning, and leadership in open Lerner, J., Tirole, J., 2002. Some simple economics of open source.
innovation communities. Working Paper. Journal of Industrial Economics 50 (2), 197–234.
Freeman, C., 1968. Chemical process plant: innovation and the world Lorenzoni, G., Lipparini, A., 1999. The leveraging of interfirm rela-
market. National Institute Economic Review 45, 29–57. tionships as a distinctive organizational capability: a longitudinal
Freeman, L.C., 1979. Centrality in social networks: conceptual clari- study. Strategic Management Journal 20, 317–338.
fication. Social Networks 1, 215–240. Mitsubishi Research Institute, 2004. Free/libre/open source software
Friedkin, N.E., 1991. Theoretical foundations for centrality measures. Japanese developers online survey (FLOSS-JP). Available at:
American Journal of Sociology 96, 1478–1504. http://oss.mri.co.jp/floss-jp/report en.html.
German, D.M., 2003. The GNOME project: a case study of open O’Mahony, S., 2003. Guarding the commons: how community man-
source, global software development. Software Process Improve- aged software projects protect their work. Research Policy 32,
ment and Practice 8, 201–215. 1179–1198.
O’Mahony, S., Ferraro, F., 2004. Hacking alone? The effects of online Teece, D.J., Pisano, G., Shuen, A., 1997. Dynamic capabilities and
and offline participation on open source community leadership. strategic management. Strategic Management Journal 18 (7),
Working Paper. 509–533.
Powell, W.W., Koput, K.W., Smith-Doerr, L., 1996. Interorganizational Tushman, M.L., 1977. Special boundary roles in the innovation pro-
collaboration and the locus of innovation: Networks of learning cess. Administrative Science Quarterly 22, 587–605.
in biotechnology. Administrative Science Quarterly 41 (1), 116– von Hippel, E., 1976. The dominant role of users in the scientific instru-
145. ment innovation process. Research Policy 5, 212–239.
Rosenberg, N., 1976. Perspectives on Technology. Cambridge Univer- von Hippel, E., 1988. The Sources of Innovation. Oxford University
sity Press, New York, NY. Press, New York, NY.
Rothwell, R., Freeman, C., Horlsey, A., Jervis, V.T.P., Robertson, A.B., von Hippel, E., 2005. Democratizing Innovation. The MIT Press, Cam-
Townsend, J., 1974. SAPPHO updated—project SAPPHO phase bridge, MA.
II. Research Policy 3, 258–291. von Hippel, E., von Krogh, G., 2003. Open source software and the
Shan, W., Walker, G., Kogut, B., 1994. Interfirm cooperation and ‘private-collective’ innovation model: issues for organization sci-
startup innovation in the biotechnology industry. Strategic Man- ence. Organization Science 14 (2), 209–223.
agement Journal 15, 387–394. von Krogh, G., Spaeth, S., Lakhani, K.R., 2003. Community, joining,
Stuart, T.E., 2000. Interorganizational alliances and the performance of and specialization in open source software innovation: a case study.
firms: A study of growth and innovation rates in high-technology Research Policy 32, 1217–1241.
industry. Strategic Management Journal 21 (8), 791–811. Wasserman, S., Faust, K., 1994. Social Networks: Methods and Appli-
Teece, D.J., 1986. Profiting from technological innovation: implica- cations. Cambridge University Press, Cambridge.
tions for integration, collaboration, licensing and public policy. Wernerfelt, B., 1984. A resource-based view of the firm. Strategic
Research Policy 15 (6), 285–305. Management Journal 5 (2), 171–180.

Dahlander & Wallin (2006) Unlocking Communities As Complementary Assets

Uploaded by

Copyright:

Available Formats

Dahlander & Wallin (2006) Unlocking Communities As Complementary Assets

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dahlander & Wallin (2006) Unlocking Communities As Complementary Assets

Uploaded by

Copyright:

Available Formats

Research Policy 35 (2006) 1243–1259

A man on the inside: Unlocking communities as

Available online 20 October 2006

1. Introduction changed how managers need to think about these con-

1. INDEG 6020 3.94286 9.49825 0 192 1

the fixed effects regressions drop observations from the 5. Results

L. Dahlander, M.W. Wallin / Research Policy 35 (2006) 1243–1259

Table 3 DED are compared to SPONSOTH and hobbyists in the

1. INDEG 23.6993 3.514706 3.461525

You might also like