See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222836738

Motivation of Software Developers in Open Source Projects: An

Internet-Based Survey of Contributors to the Linux Kernel

Article  in  Research Policy · March 2003

DOI: 10.1016/S0048-7333(03)00047-7 · Source: CiteSeer

CITATIONS READS
1,078 2,958

3 authors, including:

Guido Hertel
University of Münster
148 PUBLICATIONS   5,908 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SPP 1921 - Intentional Forgetting in Organisationen View project

The Psychology of Migration and Refugee Integration View project

All content following this page was uploaded by Guido Hertel on 12 January 2018.

The user has requested enhancement of the downloaded file.

 Research Policy 32 (2003) 1159–1177

Motivation of software developers in Open Source projects: an

Internet-based survey of contributors to the Linux kernel
Guido Hertel∗ , Sven Niedner, Stefanie Herrmann
Institut fuer Psychologie, University of Kiel, Olshausenstr. 40, D-24 098 Kiel, Germany

Abstract
The motives of 141 contributors to a large Open Source Software (OSS) project (the Linux kernel) was explored with an
Internet-based questionnaire study. Measured factors were both derived from discussions within the Linux community as well
as from models from social sciences. Participants’ engagement was particularly determined by their identification as a Linux
developer, by pragmatic motives to improve own software, and by their tolerance of time investments. Moreover, some of the
software development was accomplished by teams. Activities in these teams were particularly determined by participants’
evaluation of the team goals as well as by their perceived indispensability and self-efficacy.
© 2003 Elsevier Science B.V. All rights reserved.
Keywords: Open Source Software; Linux; Virtual teams; Motivation; VIST model

1. Introduction to improve software development processes in other

OSS projects as well as in the corporate world.
The idea of Open Source Software (OSS) has gained In the following sections, we will first describe
more and more attention in the last years. OSS is usu- briefly the main principles of OSS development and
ally developed by a loosely-knit community of pro- discus suggestions about motivational processes that
grammers spread all over the world, who contribute to originate from the OSS community itself. Then, we
a software project via Internet without necessarily be- will present two theoretical models from social sci-
ing employed or paid by an institution. In most cases, ences that might provide a conceptual framework
such an institution does not even exist. The objective for the explanation of persons’ involvement in OSS
of the present study was to explore the motivation of projects. The first has its roots in research on voluntary
such persons who spend considerable time and effort action for social movements and related community
in OSS projects “for free”. The data were collected work (Klandermans, 1997; Simon et al., 1998) and
within the Linux kernel project, one of the most active specifies different classes of motives. If we understand
OSS projects in the world that has a relatively long the Linux community as a social movement with cer-
history and a high popularity (see Moon and Sproull, tain social and political interests, this model might
2002, for a recent documentation). In addition to sci- be suitable to explain the motivation of its members.
entific interests, a better understanding of motivational The second model specifies motivational processes of
processes within the Linux kernel project might help individuals when they work in small teams (Hertel,
2002; Karau and Williams, 2001; Karau et al., 2000).
∗ Tel.: +49-431-880-2980; fax: +49-431-880-1559. Since some of the Linux kernel development might
E-mail address: [email protected] (G. Hertel). be conducted in project teams, team-based motives

0048-7333/03/$ – see front matter © 2003 Elsevier Science B.V. All rights reserved.
doi:10.1016/S0048-7333(03)00047-7
1160 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

might also play an important role in order to un- assuring that once software was published under that
derstand developers’ activities in OSS projects. A license, the availability of the source code is guaran-
web-based questionnaire was developed based on teed for all future enhancements. The Linux kernel is
these two conceptual models and completed by mem- one example of software licensed under the GPL.
bers of the Linux kernel community. After describing The label “Open Source Software”, however, was
this questionnaire and its results, we will discus the coined to create a tamed variant of the Free Software
implications of the results for our understanding of Foundation’s ideological concept in order to enhance
the motivational processes in OSS projects, as well as the acceptance of OSS by software companies. Ac-
for the possible improvement of such projects. cording to Raymond (1999), it suffices that the source
code is publicly available and that it permits changes
by users. As a consequence, a number of compa-
2. Open Source Software nies such as Netscape (including the Mozilla project)
or Sun Microsystems (including the Solaris operating
In Open Source Software development, the source system and the StarOffice package) decided to pub-
code (i.e. the human-readable commands) of a com- lish the source code of parts of their software projects.
puter program is publicly available and usually shared Other examples of successful OSS products are the
via the Internet. Every Internet user with sufficient Apache webserver, the Free BSD operating system,
skills can join the project at any time, for instance by and the Mozilla web browser. However, none of them
downloading the source code and implementing ex- has yet reached the popularity of the Linux operating
tensions or corrections. Usually, software developers system.
contribute to OSS projects “for free”, either as a hobby
or during their regular working hours even when OSS
development is not part of their usual work. However, 3. The Linux kernel development process
some companies have recently started to sponsor OSS
development, and pay developers so that they can work Linux is a PC-based operating system that has been
on OSS projects full-time. Although this changes the developed as Open Source Software along the struc-
voluntary nature of OSS development, it does not af- ture of the UNIX operating system. The Linux system
fect the general OSS principle that the source code is consists of a large number of programs or modules
available to everybody. that are arranged around a kernel. The task of the ker-
The idea of OSS development is about as old as nel is to give the programs access to resources such
software development itself. In fact, when researchers as hard disk storage, random access memory, network
in the 1960s began to use computers for their work, band width, etc. The central role of the kernel makes
they often had to rely on open sharing of software it an essential part of the Linux operating system, cur-
code simply because commercial software solutions rently containing about two million lines of source
and support were not available at that time (Moon and code (Bovet and Cesati, 2001).
Sproull, 2002). Later, when commercial software de- The development of the Linux operating system was
velopment increased and often only the software ven- started in 1991 by the Finnish computer science stu-
dor had access to the source code of a program, OSS dent Linus Torvalds, who is still managing the project.
became an attractive alternative since it enabled the In the very beginning, Torvalds did not plan to start
users to adapt and improve the software according to a worldwide OSS project. In his first announcement
their personal needs. In the mid 1980s, R.M. Stallman, of Linux he called the project a “pet project ‘’, and
one of the first public advocates of the OSS idea, also his original motivation can be described as a mixture
introduced a political aspect to OSS, claiming that “in- of dissatisfaction with existing software solutions and
formation wants to be free”, and computer programs enjoyment to write software programs (Torvalds and
should not be owned by companies but be a public Diamond, 2001). However, over the years he and the
good (Stallman, 1994a,b). Following Stallman’s and Linux project have attracted several thousand devel-
others’ initiative, the Free Software Foundation cre- opers and other contributors from all over the world
ated the “general public license” (GPL), a legal paper (see Moon and Sproull, 2002, for more details). The
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1161

“credits” file of the Linux kernel alone contains al- on these subsystems particularly when we explore
ready about 400 contributors, and about 3500 people possible teamwork aspects of the Linux development
were subscribed to the Linux kernel mailing-list at the process. However, it should be noted that these sub-
time of this study (spring 2000). As a result of these systems are technical and not organizational entities,
activities, Linux products are today widely regarded and hence are not necessarily teams but can also con-
as being high in quality and reliability, and compete tain only one developer.
successfully with software products that are developed We derived the possible motives of Linux kernel
by large commercial companies. contributors from two different sources: first, we ana-
Although a formal organizational structure is lack- lyzed discussion papers that circulate within the OSS
ing, there are some general characteristics of the community itself, thereby building on the intuition and
Linux project that might be considered as structural experience of the persons involved. Second, as a more
conditions of successful OSS development (cf. Moon systematic and structural approach, we consulted theo-
and Sproull, 2002). Among them are (a) a general cul- retical models developed from social scientists who try
ture in which authority comes from competence, (b) to explain and predict more generally when and why
delegative and participative leadership principles com- individuals voluntarily engage in community projects.
bined with clear responsibilities, (c) a modular project Below, we describe these sources in more detail.
structure that decreases unnecessary complexity, (d)
a parallel release policy that simultaneously enables
rapid development and a stable working system, (e) a 4. Suggestions from the OSS community
motivating credit policy that not only acknowledges
the contributions of developers but also, for instance, While Stallmann’s publications (e.g. 1994a,b) on
documentation work, (f) clear rules and norms of OSS might be seen as rather ideological, first thoughts
the community that are communicated online, and about motivational processes in OSS projects can be
(g) simple but reliable communication tools that are found in publications by Raymond (1999). Accord-
available worldwide (e-mail, file transfer, Usenet dis- ing to Raymond, an OSS project is usually born by
cussion groups). Interestingly, most of these principles “scratching a developer’s personal itch” (Raymond,
are similarly considered as structural pre-conditions 1999, p. 32). He compares the development process
for successful distributed work in business organi- to a bazaar, where everyone can join and contribute,
zations (e.g. Duarte and Snyder, 1999; Lipnack and creating an inspiring, creative and democratic atmo-
Stamps, 1997). sphere. Raymond contrasted this bazaar style devel-
The activities of the Linux kernel programmers are opment with a rather hierarchical “cathedral style”
mainly organized by the Linux kernel mailing-list. which he observed in commercial software develop-
This mailing-list acts as a central place for discus- ment. In the bazaar model, the democratic discussion
sions about the technical and organizational aspects of changes in the software is assumed to assure that
of the kernel development. Since subscription to the only the best solutions are accepted for the source
mailing-list is open to everyone, joining and leaving code. On the other hand, it is also assumed that public
the project is easy. There is no formal criterion for scrutiny avoids flaws (bugs) in the program more ef-
being a Linux developer apart from the contributions fectively since every user is also a potential developer.
(patches) one submits. However, in all questions, Li- The quote “given enough eyeballs, all bugs (program-
nus Torvalds has still the final decision as the project ming errors) are shallow” (Raymond, 1999, p. 41) is
maintainer, and he is often referred to as a “benevo- famous among OSS developers and captures this idea.
lent dictator”. Besides the Linux kernel mailing-list, However, Raymond’s postulation of how OSS de-
there are a couple of mailing-lists related to more spe- velopment might work has also been criticized. For
cific technical aspects of the kernel development, such instance, Bezroukov (1999a,b) claimed that OSS de-
as the development of certain file system or hardware velopment is not as outstanding as Raymond stated,
drivers. These fields are called “subsystems” of the but resembles structures and processes of the scien-
kernel. We believe that most of the development takes tific community. According to Bezroukov, members
place in these subsystems, and therefore we will focus of both the scientific and the OSS community are not
1162 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

driven by monetary rewards but by competitive mo- 2002). Although these studies included developers
tives of status and reputation (Bezroukov, 1999a). In- from different OSS projects (Hars and Ou, 2002;
terestingly, Bezroukov also suggested that some of the Lakhani et al., 2002) or explored support systems in-
motivation to work in OSS projects might stem from stead of software development activities (Lakhani and
a perceived competition of OSS projects with com- von Hippel, 2000), the suggested motives of OSS de-
mercial software companies. Bezroukov further ar- velopers are mainly consistent with our speculations
gued that these competitive motives can also decrease presented above.
the quality of decisions, and he described burnout
syndromes and personal attacks (“flame wars”)
among developers as symptoms of such unproductive 5. Relevant models from social science
processes.
Linus Torvalds himself stated that he initially pub- In addition to a rather intuitive approach to under-
lished the source code of the first Linux kernel version stand the motives of OSS developers, we wanted to
because it was just “natural within the community” base the present investigation also on more systematic
(Torvalds, 1998), thereby referring to a (virtual) com- approaches as provided by theoretical models from so-
munity that already existed before the development of cial sciences. Two models from social psychology are
Linux was started, and that was based on Internet me- particularly relevant in this context.
diated discussions on Usenet. One of the main personal
motives for Torvalds was simply “fun to program” 5.1. Participation in social movements
(Torvalds and Diamond, 2001), and he assumed that
this also applies to a large number of his co-developers. A number of social researchers have explored moti-
However, Torvalds also admitted that the current suc- vational processes of persons who voluntarily engage
cess of the Linux project is connected with reputation in social movements such as the civil rights movement,
that might provide career advantages of software de- the labor movement, the peace movement, or in the
velopers, although this was not a goal at the beginning movements of specific social groups such as gay and
of the project. lesbian people, older people, etc. (e.g. Klandermans,
In summary, the discussion within the Linux kernel 1997; Omoto and Snyder, 1995; Simon et al., 1998).
community suggests two main motive classes of OSS Social movements can be defined as “effort(s) by a
developers: (a) intrinsic motivation (“fun to program”) large number of people to solve collectively a prob-
and personal challenges to improve existing software lem that they have in common” (Toch, 1965; see also
for own needs, and (b) social comparison motives such Simon et al., 1998). Although the OSS community
as competition with other developers (either within might not be a typical social movement, one can ar-
OSS projects or between OSS projects and commer- gue that some of the political and social goals of
cial software projects) and/or the interest to build a the OSS community—and the Linux community in
reputation that might be helpful for their occupational particular—can be understood as collective effort to
career. solve a common problem of those who participate (cf.
When the present study was conceptualized (Jan- Raymond, 1999). Examples of such goals are more au-
uary 2000), no empirical data were available that tonomy in modifying software according to personal
explored these speculations empirically beyond mere needs, or protecting the diversity of software solutions
demographic information about OSS developers (e.g. against too strong dominance of large economic enter-
Dempsey et al., 1999; Koch and Schneider, 2000). prises. Moreover, the voluntary nature of OSS contri-
Thus, one main objective of the study was to collect butions is another important feature similar to social
first empirical data that explore the prevalence and movements. Thus, even if one can question whether
impact of the mentioned motives within the Linux OSS communities such as the Linux community fulfill
kernel community. In the meantime, a few other em- all criteria of a social movement, there are at least a
pirical studies on motivation in OSS projects have number of similarities that let us assume that the under-
been presented (Hars and Ou, 2002; Lakhani and lying motives of those who contribute to OSS projects
von Hippel, 2000) or are in progress (Lakhani et al., are similar to participants in social movements.
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1163

One well established model that explains so- motives are assumed by Klandermans to determine a
cial movement participation was developed by person’s willingness to actively participate in a social
Klandermans (1997). According to his model, the movement.
motivation of participants in social movements de- Recent research has extended this conceptual
pends upon various (expected) costs and benefits. framework by the aspect of collective identification
Three basic types of expected costs and benefits are (Simon et al., 1998). Simon and his colleagues ar-
distinguished by Klandermans, comprising three dif- gued that persons not only weight costs and benefits
ferent classes of motives for the participation in and when they decide whether they want be involved in
contribution to a social movement. The first class is a social movement, but that they also feel and define
described as collective motives that are based on the themselves as members of a specific group-related
evaluation of the movement goals weighted by the to the social movement and behave according to the
perceived likelihood that these goals are reached. Such norms and standards of this group. Although one can
expectancy × value constructs have a long tradition argue that these identification processes are already
in social psychology (e.g. Ajzen and Fishbein, 1980; included within Klandermans’ model (e.g. as part
Atkinson, 1957; Vroom, 1964). According to this of his collective motives class), the empirical data
approach, the motivation to participate and contribute that Simon et al. documented seem to suggest that
to a social movement will be higher the higher a per- collective identification processes are another inde-
son values its goals and the more likely the person pendent factor for the explanation of social movement
perceives the attainment of these goals. Formally, this engagement. Interestingly, their data as well as the
relation is conceptualized as a multiplicative function data of other studies have shown that identification
of the subjective evaluation of the goals and of the with more specific subgroups are a better predictor
subjective expectation that these goals are reached. of willingness to contribute to the social movement
The second class of motives is called social motives than identification with the movement as a whole.
and contains expected reactions of significant others For example, identification as a feminist activist was
such as friends and family members. Motivation to a better predictor for the willingness to participate in
contribute to a movement should be higher the more women-related collective action than identification as
positive the expected reactions of significant others a woman (Kelly and Breinlinger, 1995). It seems that
are, weighted by the perceived importance of these sig- the closer a person identifies with active subgroups of
nificant others. This relation is formally also expressed a social movement, the more she/he is willing to con-
as a multiplicative function. It should be noted, how- tribute personally. Similar processes can be assumed
ever, that the term “social motives” can be misleading for projects of the OSS community.
because there are other motives that can be considered Thus, when we refer to the “Extended Klandermans
as “social” as well and that are not included in this Model” (EKM) we refer to four motivational compo-
concept, such as making new friends or socializing nents to explain voluntary action in social movements:
with others. Thus, for better discrimination, we will collective motives, norm-oriented motives, reward mo-
call motives related to reactions of significant others tives, and identification processes. These components
“norm-oriented motives” throughout this paper. are assumed to contribute additively to the motivation
Finally, the third class of motives includes reward of participants in a social movement or community
motives that result from expected costs and benefits project. When applied to OSS projects and the Linux
such as investments of time and money, making new kernel community in particular, we expect that the
friends, or risking one’s health (see also Simon et al., EKM components also explain and predict the motiva-
1998). Similar to the collective motives, these gains tion to contribute to the OSS project. Please note that
and losses are weighted by their expected likelihood most of the above mentioned motives derived from
and formally expressed as a multiplicative function. the discussion within the OSS community can be inte-
It is assumed that the motivation to contribute to a grated into the EKM components. For instance, intrin-
movement is higher the higher and the more likely sic motives (enjoying writing software) or expected
the expected gains are perceived. The opposite holds advantages of Linux-related activities for furthering
for expected losses. Together, these three classes of one’s career are good examples of reward motives.
1164 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

5.2. The VIST model of individuals’ motivation in of the present study was to explore whether teamwork
teams exists among Linux developers. If such teamwork
exists, models of individual’s motivation in (virtual)
The second model relevant for the understand- teams might contribute to the understanding of moti-
ing of developers’ motivation in OSS projects stems vational processes in OSS projects.
from research on motivational processes in small According to the VIST model (Hertel, 2002),
work teams. Building on expectancy × value mod- individuals’ motivation to work in a virtual team
els that explain motivational processes in individual is determined by four main components: valence,
work (e.g. Atkinson, 1957; Vroom, 1964), recent instrumentality, self-efficacy, and trust, which are ab-
research has extended these approaches to the more breviated by the label “VIST”. Valence is defined as
complex situation of teamwork in which (perceived) the subjective evaluation of the team goals similar to
values and expectancies are contingent to a num- the valence part of collective motives in the EKM
ber of additional factors (e.g. Karau and Williams, outline above. It is assumed that the motivation of a
2001; Shepperd, 1993). One current model (Hertel, team member is directly proportional to her/his sub-
2002; Hertel et al., 2002) has conceptualized these jective evaluation of the team goals. However, while
motivational processes particularly for de-located collective motives in the sense of the EKM refer to
or “virtual teams”, i.e. teams in which team mem- the more general goals of a social movement, the
bers work in different places and coordinate their valence component refers to more specific goals of a
work mainly via electronic media (cf. Lipnack and smaller team that might be a subgroup of the social
Stamps, 1997). movement.
OSS development and Linux kernel development In addition to this valence component, the other
in particular is usually realized via the Internet with- three VIST components contain expectancy concepts
out face-to-face contact of the contributors. Thus, the of different kinds. Instrumentality is defined as the
collaboration of the contributors can be considered perceived importance or indispensability of one’s own
as “virtual collaboration”. However, due to the high contributions for the group outcome. The higher the
number of participants and the ease of access, OSS perceived instrumentality of one’s own contributions,
projects are generally better understood as a commu- the higher should be the motivation to exert effort for
nity or collaborative network (Wellman, 1997) than as the team goals. On the other hand, if a person believes
a team. Communities usually include a large number that her/his contribution does not matter or cannot be
of people, and are open to anyone who wants to join identified, performance motivation should decrease
as long as she/he obeys some general behavior rules. considerably even when this person values the team
Collaborative networks are more restrictive in their goals highly. The third component self-efficacy is
access policy, relying on referral or reputation and adopted from work by Bandura (e.g. 1977) and in-
develop a more specific community code including cludes team members’ perceived capability of show-
sanctions for violating this code. However, the bound- ing the required activities for the team tasks. Thus,
aries of collaborative networks are still relatively self-efficacy describes the perceived contingency that
flexible, allowing a rather frequent change of collab- one’s own high effort leads to high performance.
orators. A “team”, in contrast, refers to a relatively When a team member believes that she/he is unable
small group of collaborating people (about 2–20) with to accomplish her/his part of the team task, her/his
clear and relatively stable team boundaries, functions, motivation should be low even if the team goals are
roles, and norms. Although OSS projects as the Linux highly valued and the contribution is perceived as
kernel project are more a community or collabora- necessary for the team’s success.
tive network, team-based approaches might still be Finally, trust as the fourth component is defined
relevant to understand developers’ motivation as long as the expectancy of team members that their efforts
as part of the development activities are realized in will be reciprocated and not exploited by other team
spontaneously formed project teams. Particularly in members (interpersonal trust), and that the electronic
the Linux kernel structure, such teams might exist for support system works reliably (trust in the system).
the subsystems mentioned earlier. Thus, one objective Trust is particularly relevant in virtual teams in which
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1165

misunderstandings and fear of exploitation can esca- with the movement in general. Accordingly, we mea-
late more quickly due to fewer face-to-face interac- sured identification related to three different levels of
tions (e.g. Jarvenpaa and Leidner, 1999). Together, group specificy, ranging from the general Linux user
each of the four components is assumed to contribute community over the Linux developer community as
positively to a member’s motivation in a virtual team. a moderately specific group to a certain Linux kernel
While related models of motivation in groups (e.g. subsystem as the smallest and most specific group
Karau and Williams, 2001) assume a multiplicative category. This granularity is reflected in the structure
function of at least some of these components, the of the Linux kernel as well as in the structure of
VIST model refrains from such specifications since the corresponding mailing-lists. Together, the compo-
the empirical base is still too shallow. Moreover, a re- nents of the EKM were expected to predict the level
cent meta-analysis of similar concepts for individual of engagement in the Linux development process as
work suggests that a multiplicative combination does indicated by time spent on Linux development and
not explain more variance than the single components by the willingness to engage in Linux development in
alone (Van Eerde and Thierry, 1996). Instead, the the future.
different components of the VIST model are assumed On the other hand, the VIST model was used to
to contribute rather additively to team members’ predict the motivation of those developers who ac-
motivation similar to predictors in a regression tually contribute software to the Linux kernel based
approach. on teamwork. Such teamwork was expected to occur
spontaneously in so-called “subsystems” of the Linux
kernel. Thus, the items related to the VIST model were
6. The present study only relevant for developers who contributed to such
subsystems based on teamwork. Accordingly, the va-
A web-based questionnaire survey was conducted lence, instrumentality, self-efficacy, and trust compo-
based on the two described motivational models in nents were measured with respect to the specific goals
order to explore the motives of OSS contributors and processes within these subsystem teams. Given
systematically. However, suggestions of possible mo- that such teamwork existed, we expected that the VIST
tives from discussions within the Linux community components would also predict part of the average
were integrated in the operationalization of the model time spent on Linux development (hours per week)
components. and the willingness to be involved in the subsystem
The Extended Klandermans Model was used to work in the future. Moreover, the VIST components
explore more general motives for participating in should predict more concrete performance criteria
the Linux kernel community, either by contributing such as estimated contributions of lines of computer
pieces of software or by following the discussion and source code and number of “patches” for the Linux
contributing comments and ideas to the mailing-lists. kernel submitted by a developer.1 Finally, indicators
Collective motives were measured with questions of participants’ satisfaction with the Linux kernel
about how much participants valued typical goals of development process such as recognition and com-
the Linux kernel community. Norm-oriented motives munication climate were measured for exploratory
were measured by exploring whether evaluations of reasons.
friends and family members play a role in participants’
involvement in the Linux development. Reward mo- 1 Computer source code is arranged in lines, where each line
tives were measured by asking participants to rate the corresponds to one instruction. Thus, the number of these lines
personal importance of possible gains and losses due of code can be used as an objective estimation of the quantity
to their involvement in the Linux kernel community. of contributions by one developer. Several lines of code, in turn,
Finally, the identification component was measured form a “patch” which is an atomic contribution to a software
related to three different group concepts in order to project, submitted as one file and addressing one single issue.
Consequently, the number of patches produced by one developer
explore whether identification with more specific sub- is an estimation of his or her number of issues addressed. For
groups might be more predictive for the participation reasons of anonymity, it was not possible for us to measure the
in social movements than unspecific identification contributions of the participants objectively.
1166 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

7. Method view of the results was published on the Internet along

with the raw data as a download file.
7.1. Recruitment and data collection procedure
7.2. The questionnaire
Data were collected between 15 February and 12
April 2000 via the Internet. About 4 weeks before the The variables measured in the questionnaire fall
survey started, the study was announced on relevant into four main categories. The first category included
mailing-lists of the Linux kernel community. More- demographic and context-related items. The second
over, a web-page was installed on which the goals and third categories contained motivational measures
and the planed procedure of the questionnaire study based on the EKM and the VIST model, respectively.
were explained including information about the per- Finally, the fourth category included measures of ex-
sons conducting the study (i.e. the authors of this pa- erted effort and concrete performance as criteria to
per). In order to increase the acceptability of the study be predicted by the motivational measures. Moreover,
as well as to increase the conceptual input, we made an additional questions addressed participants’ satisfac-
attempt to integrate the Open Source philosophy into tion with the Linux developing process for exploratory
the development of the questionnaire by inviting the reasons.
readers of the announcement to contribute and discus
ideas for the questionnaire on a related mailing-list.2 7.2.1. Demographic and context-related items
About 4 weeks after this first announcement, the Items of this category measured demographic data
final version of the questionnaire was installed on the such as the sex, age, nationality, and employment
Internet. The survey was then started with a second status of the participants. Moreover, it was measured
announcement on the same Linux kernel mailing-lists. whether participants received payment for their Linux
A lottery with small prizes (Linux-related products) developing work, whether they were able to accom-
was included as incentive to participate in the study. plish Linux-related work during their regular working
The e-mail addresses of participants who chose to hours, how much of their spare-time they devoted to
participate in this lottery were stored separately from Linux programming, and how long they had already
their questionnaire data in order to ensure anonymity. been engaged in Linux-related activities. Finally, it
As another incentive, it was announced that a brief was explored whether participants were active con-
summary of the anonymized results together with tributors to the Linux kernel or rather interested
the raw data of the study would be published on the readers of the discussion list.
Internet as soon as the study was finished. Similar as
the public discussion of the questionnaire items, this 7.2.2. Items measuring the components of the EKM
procedure integrated the Open Source philosophy in These questions were developed in a similar way
our research approach. to the study by Simon et al. (1998).3 The collective
The questionnaire items were completed either by motive component was measured by participants’ rat-
clicking answer options in drop-down boxes or by en- ings of how much they value the central goals of the
tering free answers. Since overlong questionnaires are Linux kernel community, which were derived from
often not completed on the Internet (e.g. Batinic and the mentioned sources within the community: improv-
Bosnjak, 2000), we restricted the length of the ques- ing the quality of the Linux kernel, improving the
tionnaire to a completion time of about 15 min. Af- quality of the subsystem the participant was working
ter completion of the questionnaire, the participants on, and the more general goal that “software should
clicked on a “send” button to transfer their answers to be free”. These items were answered on five-point
a server at the University of Kiel. At the end of the
study, the lottery was conducted, and a first short re- 3 Due to the mentioned space restrictions for Internet-based ques-

tionnaires we refrained from including expectancy weights for all

2 Unfortunately, the traffic on this mailing list was rather low. motives. The reported analyses were restricted to the collected im-
Only 15 persons subscribed to the list, and a real discussion never portance ratings. Such a procedure is appropriate since importance
started. ratings often already include expectancy considerations implicitly.
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1167

scales ranging from “very unimportant” (1) to “very on five-point scales from “disagree strongly” (1) to
important” (5). The norm-oriented motive component “agree strongly” (5). As with the EKM, the number of
was measured by one item which asked participants items was a compromise between issues of reliability
what relevant others (family, friends, colleagues) think and acceptability.
about their activities in the Linux kernel community.
The answer scale ranged from “very negatively” (1) 7.2.4. Criterion variables
to “very positively” (5). The reward motive compo- To test whether the measured indicators were in
nent was measured by eight items, which included fact valid predictors of participants’ motivation to
both motives derived from the mentioned discussion contribute to the Linux kernel development, sev-
within the Linux kernel community as well as mo- eral criterion measures were collected. First, as one
tives derived from the literature on social movements main indicator of participants’ engagement, partici-
(e.g. Klandermans, 1997). Participants rated the per- pants estimated how many hours they currently spent
ceived importance of the following gains and losses on on Linux development. Second, similar to Simon
five-point scales ranging from very unimportant (1) to et al. (1998), participants were asked whether they
very important (5): facilitating daily work due to bet- looked forward to being involved in Linux develop-
ter software, personal exchange with other software ment in the future, both in Linux more general and
developers, gaining reputation as an experienced pro- in their subsystem project (five-point answer scale
grammer inside the Linux community, having fun pro- from “strongly disagree” to “strongly agree”). Third,
gramming, improving one’s own programming skills, as more objective performance criteria, participants
career advantages, time loss, and lack of payment. Fi- estimated how many lines of code and how many
nally, the identification component was measured with “patches” (cf. footnote 1) they had contributed to
a three item scale adapted from Doosje et al. (1995; the Linux kernel. Finally, for exploratory reasons we
e.g. “I identify with the Linux user community”) that measured participants’ satisfaction with several pro-
was re-worded for each of the three different group cesses within the development process such as com-
categories identification with Linux users, identifica- munication, working atmosphere, recognition, and
tion as Linux developer, and identification with a cer- outcomes (five-point answer scales). Four items ad-
tain subsystem, respectively. dressed the Linux kernel development in general and
four items addressed processes within the subsystem,
7.2.3. Items measuring the VIST model components respectively.
These items were adapted from earlier work (Hertel,
2002; Hertel et al., 2002) and referred only to the 7.3. The sample
Linux kernel subsystems since these components were
considered as most relevant for the motivational pro- One hundred and forty-one participants (6 females,
cesses in (virtual) teams. Participants were instructed 135 males; mean age = 30 years, range: 16–54
to answer these questions only for the subsystem years) from 28 different countries took part in the
project that was most important to them. The valence study. About 48% of the participants lived in North-
component was measured by two items (e.g. “The ern America, 37% in Europe, and 7% in Australia.
success of this subsystem is very important to me”). Less than 3% lived in Asia, Southern America, and
The instrumentality component was measured by Africa, respectively. Most of the participants (67%)
three items (e.g. “I believe my personal contribution were full-time employees, 5% worked half-time, and
is crucial for the success of this subsystem project”). another 5% were unemployed. The remaining partic-
The self-efficacy component was measured by four ipants were students (23%). The participants were on
items (e.g. “I believe I have the necessary skills to average involved in Linux kernel projects for about
accomplish my tasks in this subsystem project”). Fi- 17 months (range: 1–98 months). About half of the
nally, the trust component was measured by three participants (69 persons; 49%) were actively engaged
items (e.g. “I believe that the other developers in in the Linux kernel developing process. These partic-
my project group invest high efforts in the develop- ipants categorized themselves either as “active Linux
ment of our subsystem”). These items were answered kernel developer” or as “maintainer of a module”, and
1168 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

were members of a kernel subsystem. We will call the VIST model are restricted to the participants of
this group the “developer group”. The other half of this subgroup.
the participants (72 persons; 51%) were readers of the
Linux kernel mailing-list. We will call this group the
“interested readers”. However, also the latter group 8. Results
spent some time on Linux kernel development (6.6 h
on average) according to participants’ self-ratings. The result section is divided into two parts. In
Participants of the developer group worked on av- the first part, we present the analyses related to the
erage 18.4 h per week on Linux development (range: components of the EKM as potential motivational
1–70 h). Twenty percent of the developers received a predictors of participants’ more general engage-
salary for their Linux programming work on a reg- ment in Linux-related activities. Apart from writing
ular base, and another 23% at least sometimes. The software, these activities include also reading and
remaining 57% indicated that they received no salary writing comments on the Linux kernel mailing-list.
at all for their work on the Linux kernel. Thus, not The EKM-related analyses were conducted across
all activities for the Linux kernel development were both the developer group and the interested reader
completely voluntary (see Hars and Ou, 2002, for sim- group. In the second part, we present the analyses
ilar results in other OSS projects). Not surprisingly, related to the VIST model components as potential
the more developers were paid for their Linux-related motivational predictors of the more specific activi-
work the more time they spent, r (n = 69) = 0.54, ties in subsystem teams. These analyses were only
p < 0.001. conducted with members of the developer group who
About 38% of the developer group could carry out worked in such teams. In both parts of the result
the Linux-related programming during their regular section, we first present tests whether our developed
working hours, although this did not imply that this scales were reliable and valid measures of the mo-
work was part of their official job. The remaining tivational constructs. Then, we document the means
62% carried out Linux-related programming outside and standard deviations of the measured constructs.
of their regular work. Paid developers were more likely Finally, regression analyses were conducted in order
to perform Linux-related programming within their to explore whether the motivational constructs signifi-
working hours, r (n = 69) = 0.69, p < 0.001. In- cantly predicted participants’ Linux-related activities.
vestigating how much of participants’ spare-time was
devoted to Linux-related activities showed that about 8.1. Preliminary analyses of the EKM items
40% of the developer group spent less than 10% of
their spare-time on Linux programming, 48% of the First, we tested the reliability and construct validity
developer group spent between 10 and 50% of their of our newly developed EKM scales based on factor
spare-time on Linux programming, and 12% of the analyses. Please note that such analyses also provide
developer group indicated that more than 50% of their information about the underlying motivational struc-
spare-time was devoted to Linux programming. No ture of the participating Linux community members.
meaningful correlation occurred between this measure A principal component analysis (see e.g. Bryant and
and the questions of payment, indicating that paid Yarnold, 1997, for an introduction) across all partic-
and non-paid developers spent more or less the same ipants revealed seven factors with eigenvalues > 1
amount of spare-time on Linux development. (varimax rotation). The first factor explained 22.9%
Importantly, 59% (39 participants) of the developer variance and was mostly determined by items address-
group indicated that they worked for their subsystem ing identification as a Linux developer as well as iden-
in a group, thus providing evidence that at least some tification with the subsystem (factor loadings between
spontaneous teamwork exists within the Linux kernel 0.38 and 0.85). This indicates that the participants did
community. These teams had an average 12 developers not discriminate much in their identification between
(range: 2–50) who were on average 28 months (range: the two group categories. Thus, we calculated for each
3–98 months) involved in the Linux kernel develop- participant one mean score for the specific identifica-
ment. The following team-related analyses based on tion as a developer and with the related subsystem
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1169

(alpha = 0.85). The second factor explained an addi- lated a mean score for each participant based on these
tional 11.2% variance and was mostly determined by five items. As noted by Simon et al. (1998), a high in-
the three items measuring the more general identifi- ternal consistency is not a necessary requirement for
cation as a Linux user (factor loadings between 0.82 the reward motive to be included in the analyses since
and 0.85). These items were combined to a scale score persons can be motivated by some outcomes but not
of identification as a Linux user (alpha = 0.77). To- by others, and this selectivity should vary across peo-
gether, these first two identification factors confirmed ple. Finally, the seventh factor which explained addi-
the assumed independence of identification processes tional 5.2% variance was mostly determined by the
related to more global and more specific group cate- (reversed) reward motive item measuring participants’
gories (cf. Simon et al., 1998). evaluation of time losses (factor loading = 0.87).
The third factor explained an additional 8.5% vari- Thus, a considerable tolerance of time investments on
ance and was determined by the two collective motive issues related to the Linux kernel seems to be another
items that addressed improving the quality of Linux independent characteristic of the members in the Linux
(general and subsystem level), and by two reward mo- kernel community.
tives that also addressed quality issues (facilitating Overall, the results suggest that the motives of per-
daily work with Linux software; career advantages due sons to participate in the Linux kernel community can
to experiences with Linux development; factor load- be categorized into seven major components that ex-
ings between 0.31 and 0.73). This factor seems to mea- plain a total of 65.6% variance. These factors largely
sure pragmatic motives related to the improvement of overlap with components specified by the EKM. Apart
the Linux kernel. A mean score of these items was from the distinction between more general and more
calculated for each participant as another predictor for specific group identification processes, norm-oriented
the following analyses (alpha = 0.71). motives could be distinguished as well. Collective and
The fourth factor explained additional 6.5% vari- reward motives were not as clearly distinguishable in
ance and was mainly determined by the item measur- our sample but were instead split into four more spe-
ing norm-oriented motives (factor loading = 0.78). cific factors measuring different aspects of possible
This result confirmed the assumed independence of outcomes: (a) a pragmatic component related to im-
norm-oriented motives from reward and collective mo- proving the Linux kernel for personal advantages, (b)
tives, as well as from identification aspects. The fifth a social/political component related to supporting free
factor explained additional 5.8% variance and was software and networking with other Linux developers,
mostly determined by the reward motive “having fun (c) a hedonistic component related to intrinsic motiva-
programming” (factor loading = 0.74). As suggested tion, and (d) a component mainly related to concerns
in the introduction section, hedonistic motives or in- of time losses.
trinsic motivation seem to be another independent fac- While these analyses demonstrated which motiva-
tor relevant for Linux-related activities. tional components could be reliably measured and dis-
The sixth factor explained additional 5.6% variance tinguished in our sample, in the following analyses we
and was mostly determined by the four remaining describe the relative importance of these motivational
reward motives (personal exchange with other soft- components for Linux-related activities.
ware developers, gaining reputation as an experienced
programmer inside the Linux community, improving 8.2. Means and correlations of EKM scores and
one’s own programming skills, lack of payment (re- Linux-related activities
versed); factor loadings between 0.39 and 0.66). More-
over, the third collective motive item (“software should As depicted in Table 1, the means of the mo-
be free”) showed its highest loading on this factor tivational components were all rather high. Thus,
(0.56). Thus, this sixth factor could be interpreted as participants identified highly both as a Linux user
social and political motives in the sense of supporting and as a Linux developer, indicated high pragmatic,
free software, and networking and socializing within norm-oriented, social/political, and hedonistic mo-
the Linux community. Although the scale consistency tives, and valued time losses due to Linux-related
of these items was low, alpha = 0.28, we still calcu- activities rather low.
1170 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

Table 1
Participation in Linux kernel development (n = 141): means and intercorrelations of seven motivational components and two effort criteria
Variable 1 2 3 4 5 6 7 8 9

1. General identification (Linux user) –

2. Specific identification (developer/subsystem) 0.08 –
3. Norm-oriented motives 0.09 0.29∗∗ –
4. Pragmatic motives 0.15 0.43∗∗ 0.33∗∗ –
5. Social/political motives 0.26∗∗ 0.18∗ 0.21∗ 0.31∗∗ –
6. Hedonistic motives 0.21∗ 0.28∗∗ 0.12 0.10 0.10 –
7. Time loss (rev. coding) 0.06 0.09 0.03 0.10 0.12 −0.07 –
8. Hours/week spent on Linux development −0.28∗∗ 0.32∗∗ 0.15† 0.16† 0.01 0.01 0.18∗ –
9. Willingness to be involved in the future −0.02 0.38∗∗ 0.12 0.40∗∗ 0.22∗ 0.19∗ 0.20∗ 0.22∗ –
Ma 4.0 3.6 3.7 4.2 4.1 4.6 3.4 12.5 4.2
S.D. 0.8 0.9 0.8 0.7 0.5 0.9 1.0 15.4 0.9
a Scale range of all scales except “hours/week spent on Linux development” varied between 1 and 5; see Section 7 for details. Higher

means indicate higher identification, stronger norm-oriented, pragmatic, social/political, and hedonistic motives, and lower importance of
time losses due to involvement in Linux development, and higher willingness to be involved in Linux development in the future.
† p < 0.10.
∗ p < 0.05.
∗∗ p < 0.01 (two-tailed).

However, comparing the means of participants in significant higher willingness to be further involved in
the developer and in the interested readers group Linux development (M = 4.5 versus 3.9), t(125) =
showed some interesting differences (see Table 2). 3.60, p < 0.001. More interestingly in terms of
First, participants in the developer group had higher motivational differences, participants of the devel-
means in the general engagement measures than par- oper group compared to participants of the interested
ticipants in the interested reader group. The partici- reader group showed marginally lower identification
pants of the developer group spent significantly more with the Linux community in general, M = 3.9 versus
hours per week on Linux development (M = 18.4 4.2, t(137) = 1.74, p < 0.09, but significantly higher
versus 6.6), t(136) = 4.82, p < 0.001, and showed identification with more specific categories such as

Table 2
Means of motivational components and effort criteria for the developer group and the interested readers group
Variable Developer group (n = 69) Interested reader group (n = 72) Difference

M S.D. M S.D.
1. General identification (Linux user) 3.9 0.9 4.2 0.7 p< 0.09
2. Specific identification (developer/subsystem) 4.0 0.7 3.3 1.0 p< 0.001
3. Norm-oriented motives 3.9 0.8 3.6 0.9 p< 0.06
4. Pragmatic motives 4.3 0.5 4.1 0.8 p< 0.07
5. Social/political motives 4.1 0.4 4.1 0.5 n.s.
6. Hedonistic motives 4.7 0.6 4.5 1.0 n.s.
7. Time loss (rev. coding) 3.6 1.0 3.3 1.1 p< 0.11
8. Hours/week spent on Linux development 18.4 18.0 6.6 9.2 p< 0.001
9. Willingness to be involved in the future 4.5 0.8 3.9 1.1 p< 0.001
Scale range of all scales except “hours/week spent on Linux development” varied between 1 and 5; see Section 7 for details. Higher
means indicate higher identification, stronger norm-oriented, pragmatic, social/political, and hedonistic motives, and lower importance of
time losses due to involvement in Linux development, and higher willingness to be involved in Linux development in the future. The
difference column indicates the results of t-tests (two-tailed) between the means of the two groups for each variable.
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1171

developers or a specific subsystem, M = 4.0 versus beta weight of the more general identification as a
3.3, t(137) = 4.49, p < 0.001. Moreover, participants Linux user showed a negative score (beta = −0.33,
of the developer group compared to participants of p < 0.001). As already apparent in the correlation
the interested reader group showed marginally higher analyses (Table 1), participants spent more hours on
means for the pragmatic motive, M = 4.3 versus 4.1, Linux-related activities when they identified with the
t(139) = 1.9, p < 0.07, and for the norm-oriented specific categories (Linux developer, subsystem) but
motive, M = 3.9 versus 3.6, t(134) = 1.97, p < 0.06. not as a Linux user in a more general sense. In addi-
Thus, those participants who accomplished most of tion, the less important participants rated time losses
the programming work (the developer group) identi- due to Linux development, the more hours they spent,
fied more strongly with more specific group categories beta = 0.15, p < 0.04 (directional test). All other
in the Linux community (as a developer or with a predictors showed no significant effects.
subsystem), and indicated higher concerns for reac- Performing this regression only for the developer
tions of significant others as well as higher pragmatic group showed very similar results. However, perform-
interests in improving the quality of the Linux kernel. ing this regression only for the interested reader group
Next, in order to explore which of the measured showed only a significant effect for the time loss com-
motives were predictive for the engagement of par- ponent, while the identification measures were not
ticipants, a number of correlational analyses were significant this time. Thus, while a low concern for
performed. As expected, most of the motivational time losses seems to foster Linux-related activities for
variables correlated significantly with the number of both groups of participants, the specific identification
hours participants invested in Linux development, as as a Linux developer or with a Linux subsystem is
well as with participants’ willingness to be further in- an important additional motive only for the developer
volved in Linux-related activities (cf. Table 1). Thus, group.
the measured EKM indicators covered the underlying Similar results were obtained when regression
motivation of participants to engage in more general analyses were performed with the same motivational
Linux-related activities quite well. predictors and willingness to be involved in Linux de-
As a more integrative step, we then performed a velopment in the future as criterion variable. This time
number of regression analyses to explore which of also the pragmatic motives component had a signifi-
the measured EKM-related indicators were the most cant effect (beta = 0.30, p < 0.01). Thus, the higher
relevant predictors for participants’ more general en- participants perceived personal rewards for their
gagement in the Linux kernel community. Compared Linux engagement, the more they were willing to be
to singular correlation analyses, a regression analysis involved in Linux-related activities in the future. Sep-
considers not only one predictor but a certain set of arate analyses of participants who received payment
predictor variables at the same time in the same analy- and participants who received no payment did not lead
sis. The results provide insight into the relative impor- to different results of the main regression analyses.
tance of each predictor variable in this specific context Together, these results suggest that while all com-
(e.g. Licht, 1997, for an introduction to the logic of ponents related to the EKM were significantly cor-
regression analyses). related with at least one of the criterion variables of
In the first regression, all seven motivational com- Linux-related engagement, the most important pre-
ponents were included as predictor variables, and the dictors for participants’ engagement were (a) a more
amount of hours participants spent on Linux-related specific identification as a Linux developer or with a
activities was entered as criterion variable. This anal- subsystem, (b) a considerable tolerance in respect to
ysis yielded a significant solution, F(7.123) = 5.31, time losses due to Linux development activities, and
p < 0.001 with R2 = 0.19 (adjusted for number of (c) a rather pragmatic interest in personal advantages
predictors), and revealed significant effects for both due to improving the Linux kernel quality.
general and specific identification components and for We now turn to the motivational processes of Linux
the time loss component. While specific identification developers who worked in a subsystem together with
as a Linux developer or with the subsystem showed other developers as a team. This form of collaboration
a positive beta weight (beta = 0.28, p < 0.01), the can be perceived as a “virtual team” which contains
1172 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

challenges particularly in respect to performance mo- To explore whether the VIST components were also
tivation due to low face-to-face contact (Hertel et al., predictive for the engagement of the participants, we
in press). The following analyses explore motivational correlated the VIST components with several criteria
processes in Linux subsystem teams based on the variables. Two of them measured participants’ moti-
VIST model. vation to work for the subsystem project (hours per
week spent on Linux development, willingness to fur-
8.3. Preliminary analyses of items related to the ther engage in Linux development) while the other
VIST model two criteria measured objective performance output
(number of patches, lines of code). As can be seen in
Prior to the main analyses, we again explored the Table 3, a number of significant correlations occurred
reliability and construct validity of our developed for the first three VIST components valence, instru-
scales.4 A principal component analysis with varimax mentality, and self-efficacy. Only the trust component
rotation revealed four factors with eigenvalues > 1 seemed to be rather unimportant in the Linux subsys-
that explained 68.9% variance. The empirical struc- tem teams. As a more integrative step, we again com-
ture was equivalent to the theoretically expected puted a number of regression analyses to explore the
structure of the VIST model, providing evidence for impact of the VIST components for each of the four
the assumed independence of its components. The performance criteria when all VIST components are
first factor explained 27.5% variance and was mostly taken into account simultaneously.
determined by the instrumentality items (factor load-
ings between 0.70 and 0.90). The second factor ex- 8.4.1. Hours spent on Linux development
plained additional 19.0% variance and was mostly Regressing the number of hours the developers
determined by the self-efficacy items (factor loadings spent on average per week for Linux development
between 0.64 and 0.72). The third factor explained an showed a significant effect for the instrumentality
additional 14.0% variance and was mostly determined component, beta = 0.34, p < 0.04. Thus, developers
by the trust items (factor loadings between 0.60 and spent more time on Linux development when they
0.87). Finally, the fourth factor explained additional felt that their contribution was highly important for
8.5% variance and was mostly determined by the va- the progress of the subsystem. This result is consis-
lence items (factor loadings between 0.64 and 0.89). tent with other research on motivational processes
Based on these results, we calculated mean scores for in virtual teams (Hertel et al., 2002, in press). The
the four scales for each participant, respectively (al- other VIST components did not improve the level of
pha of the trust scale = 0.66, all other alpha > 0.80). explained variance (R2 = 0.09), indicating that these
The following analyses of motivational processes components played only a minor role in the time
within the Linux subsystems were based on partic- invested by developers in this sample.
ipants who described themselves as developers and
who indicated that they worked for their subsystem 8.4.2. Willingness to increase participation in the
in a team with other developers (n = 39). subsystem in the future
Regressing this criterium on the VIST components
8.4. Means and correlations of VIST scores and showed a significant effect of the valence compo-
performance variables nent (beta = 0.40, p < 0.02) and the instrumental-
ity component, beta = 0.34, p < 0.04. Developers
were willing to spend more time in the subsystem the
Means and standard deviations of the VIST compo-
higher they valued its goals and the higher they per-
nents are documented in Table 3. These scores indi-
ceived their contribution as important for the project
cate a generally high motivation of developers in the
success. Including both components as predictors into
subsystem teams.
a regression of this criterium yielded the best solu-
4 Although the later analyses are restricted to developers who tion, R2 = 0.19. No further improvement was found
worked in subsystem teams, the validation of the VIST scales was by including the other VIST components into this
based on all participants of this study for power reasons. regression.
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1173

Table 3
Developers’ motivation in the Kernel subsystem teams (n = 39): intercorrelations between VIST components and effort and performance
criteria
Variable 1 2 3 4 5 6 7 8

1. Valence –
2. Instrumentality 0.19 –
3. Self-efficacy 0.07 0.46∗∗ –
4. Trust 0.21 0.03 −0.03 –
5. Hours/week spent on Linux development 0.19 0.34∗ 0.21 0.19 –
6. Increase participation in the subsystem (rev. coding) 0.40∗∗ 0.34∗ 0.09 0.06 −0.02 –
7. Number of patches 0.05 0.35∗ 0.36∗ 0.06 0.32∗ 0.09 –
8. Lines of code (log.-transformed) −0.13 0.08 0.32∗ −0.06 0.05 −0.33∗ 0.31∗ –
Ma 4.7 3.4 4.0 3.8 20.5 2.4 2.4 2.6
S.D. 0.5 1.0 0.6 0.8 19.4 1.0 1.6 0.9
a Scale range of all scales except “hours/week spent on Linux development”, “number of patches”, and “lines of code” varied between

1 and 5; see Section 7 for details. Higher means indicate higher valence, instrumentality, self-efficacy, trust, and willingness to increase
participation in the subsystem.
∗ p < 0.05.
∗∗ p < 0.01 (two-tailed).

8.4.3. Number of patches accepted components did not improve the amount of explained
The estimated number of patches a developer had variance, R2 = 0.10. Thus, only perceived own abil-
contributed to the Linux kernel and which had been ity was predictive for this performance indicator but
accepted is a performance output measure that is prob- none of the more group-related components of the
ably highly determined by the skills and experience of VIST model.
a person. However, since performance is always also Together, these results showed that motivational
a function of motivational factors, we still assume processes in virtual teams as specified by the VIST
that the VIST components might contribute to at least model can also be observed in the Linux kernel de-
some part of the variance of this criterion. Indeed, re- velopment process. The instrumentality and valence
gression analyses showed significant effects of the in- components were particular predictive for the mea-
strumentality component (beta = 0.35, p < 0.03) and sured motivational criteria such as time investment
the self-efficacy components (beta = 0.36, p < 0.03). and willingness to engage in the future. The output
Higher numbers of accepted patches were accompa- criteria were more strongly predicted by participants’
nied by higher self-efficacy and stronger feelings that perceived self-efficacy, presumably because objective
one’s own contributions were crucial for the subsys- performance is strongly determined by participants’
tem. Including both components as predictors into a programming skills and expertise. However, at least
regression of the number of patches yielded the for one of the two output criteria we could also ob-
best solution, R2 = 0.12, although this time the serve effects of perceived instrumentality of one’s
beta weights of both the instrumentality and the own contributions to the subsystem team. On the other
self-efficacy component were only marginally signif- hand, trust or fear of being exploited as the fourth
icant (beta = 0.24 and 0.25, respectively; p < 0.09, VIST component seemed to play only a minor role in
directional tests). the Linux subsystem teams, at least in this study.

8.4.4. Lines of code 8.5. Satisfaction and working climate

Regressing the (log.-transformed) estimations of
lines of code added to the Linux kernel as another For exploratory reasons, we also measured
performance output indicator showed only a signif- participants’ satisfaction with the Linux development
icant effect of perceived self-efficacy (beta = 0.32, process and the working climate in the subsystems.
p < 0.04, directional test), while the other VIST Specifically, we asked the participants to value their
1174 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

satisfaction with the communication processes, with were discussed with the involved persons in a public
the working atmosphere, with the recognition of mailing-list and the questionnaire data were pub-
contributions, and with the results of the subsystem. lished on the Internet. In a similar way, the develop-
These items were highly correlated and could be ment of the questionnaire items was both based on
combined as a scale (alpha = 0.76). The average rather intuitive considerations from involved mem-
satisfaction score of developers who worked in sub- bers of the Linux kernel community (e.g. Moon
system teams was high, M = 4.4 (S.D. = 0.6, scale and Sproull, 2002; Raymond, 1999; Torvalds and
range between 1 and 5), indicating that these devel- Diamond, 2001) as well as on systematic models from
opers were generally quite happy with the working social psychology. The obtained results showed that
processes. Similar questions addressing satisfaction these approaches complemented one another nicely.
with the Linux kernel development more generally Central motives that were discussed within the Linux
(alpha = 0.80) showed also high overall satisfaction, community could be integrated in the systematic so-
M = 4.2 (S.D. = 0.7; scale range between 1 and cial psychological models. However, these models
5). While no significant differences occurred between also suggested additional motives that could be con-
participants of the developer group and interested firmed in our empirical study. Together, the results
reader group in the general satisfaction score, partic- demonstrated that motivational processes within OSS
ipants of the developer group more often indicated projects, such as the Linux kernel development, do
that they had experienced “burnout” during their not differ completely from motivational processes in
Linux-related work. In the developer group, 17.4% other social communities and teams and can be ex-
indicated that they had experienced burnout at least plained within existing social psychological theories.
“once”, and 46.4% even “sometimes”. In the inter- With regard to the engagement for the Linux kernel
ested reader group, only 4.2% indicated that they had more generally (i.e. including reading and contribut-
experienced burnout “once”, and 18.1% “sometimes”. ing to the Linux kernel mailing-list), seven factors
have been identified as distinct motivational sources
for the participants. These factors were largely con-
9. Discussion sistent with the assumptions of the Extended Klander-
mans Model of voluntary action in social movements
One of the most compelling aspects of Open Source (Klandermans, 1997; Simon et al., 1998). Thus, en-
Software projects is that they are predominantly based gagement for the Linux kernel community seemed
on voluntary contributions from software developers to be driven by similar motives as voluntary action
without organizational support in a traditional sense within social movements such as the civil rights
(Moon and Sproull, 2002). One central question is, movement, the labor movement, or the peace move-
what motivates these persons to contribute to OSS ment. The main motivational factors were (a) a more
projects “for free”, and what rewards do they expect. general identification factor as Linux user, (b) a more
The objective of the present study was to explore these specific identification factor as a Linux developer or
motives empirically within one of the most prominent with a Linux subsystem, (c) pragmatic motives related
OSS projects: The Linux kernel project. Together with to the improvement of one’s own software and career
other recent studies (Hars and Ou, 2002; Lakhani and advantages, (d) norm-oriented motives related to reac-
von Hippel, 2000; Lakhani et al., 2002), the present tions of relevant others (family, friends, colleagues),
investigation is one of the first studies that provide (e) social and political motives related to support-
sound empirical data on motivational processes within ing independent software and networking within the
OSS projects. Linux community, (f) hedonistic motives such as pure
In doing so, we tried to integrate methodologi- enjoyment of programming, (g) and motivational ob-
cal and theoretical issues from the OSS community stacles related to time losses due to Linux-related
with research methods and models from social sci- activities (see Hars and Ou, 2002, for similar results
ences. For instance, while the statistical analyses of in a study including other OSS projects).
the questionnaire data were based on standard meth- All seven factors showed high mean scores and
ods from social sciences, the questionnaire items correlated positively with the measured criteria of
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1175

invested time for Linux-related activities and/or will- and performance in the Linux kernel subsystem
ingness to engage in Linux-related activities in the teams. This is a surprising result since fear that one’s
future. However, the more integrative regression anal- own efforts might be exploited by other users (or by
yses revealed that not all seven factors were equally commercial companies) certainly is an issue in OSS
predictive for participants’ engagement. With regard development (e.g. Moon and Sproull, 2002; Stallman,
to invested hours per week, the strongest predictors 1994a,b). For example, it has been vigorously dis-
were the specific identification as a Linux developer cussed on the Linux kernel mailing-list how to deal
or with a subsystem, and a considerable tolerance with contributions (so-called “modules”) that are not
of time losses due to Linux-related activities. This available under the General Public License. Perhaps,
is particularly true for those participants who could distrust and fear of exploitation play only a minor
be categorized as active developers in a more nar- role in Linux subsystems since the membership is
row sense. For the group of interested readers, only determined by contributions per se, implying that a
concern for time losses showed a significant effect developer who neither posts constructive messages to
when all seven possible predictors were integrated the mailing-list nor submits patches is not part of the
simultaneously in a regression analysis. This suggests subsystem.
that lack of time is one of the biggest obstacles for The exploratory data on satisfaction with the com-
participating in the Linux kernel project. munication processes and the working climate in the
Finally, as a third important predictor in the regres- subsystems and within the Linux community more
sion analyses, the pragmatic motive factor showed general showed quite positive results. Both partici-
significant effects in the regression of willingness to pants of the developer group and the interested reader
engage in Linux-related activities in the future. Thus, group seemed to enjoy their Linux-related activities
the interest to improve software for one’s own use (at least those who participated in our study). How-
and to increase one’s own career perspectives seemed ever, among the developers, there is also the danger
to be particularly important motives for planed ac- of negative side effects due to high voluntary engage-
tivities in this community. However, this motivation ment. More than 50% of participants in the developer
might diminish once a developer has actively joined group indicated that they had experienced feelings of
a project and learned that these expectations are not burnout at least once. Although this certainly is not a
always so easily fulfilled. diagnosis in a clinical sense, maintainer or organizer
In respect to more specific programming work of OSS projects should be aware of possible burnout
in subsystem teams, the four central factors derived effects and should take preventive steps against it.
from the VIST model (Hertel, 2002) could also be A number of other practical implications arise from
confirmed as underlying motives of participants. This the reported results that are not only relevant for the
result is in accordance with current research on vir- Linux kernel community and other OSS projects, but
tual teams in business organizations (Hertel et al., that might also be applied to software development
2002), suggesting that similar processes can be found within commercial enterprises if it follows a rather
in virtual teams of OSS projects with rather low level unstructured “bazaar”-like style. One important issue
of explicit organization. Similar to the EKM factors, is that identification processes are important for such
the VIST factors also showed generally high mean software development projects. In accordance with re-
scores. However, only the first three VIST factors search in other areas of voluntary engagement (Kelly
correlated significantly with at least one of the mo- and Breinlinger, 1995; Simon et al., 1998), this iden-
tivation/performance criterion measures. Participants’ tification should be related to smaller and more active
activities were particularly determined by (a) their units of a project rather than to the project in general.
subjective evaluation of the subsystem goals, (b) the Concerns about time losses, as another important
perceived importance of their own contributions for motivational factor, can be addressed by streamlining
the subsystem, and (c) the perceived personal ability the organizational procedures in OSS projects as well
to accomplish the tasks. as other software development processes so that inef-
On the other hand, perceived trust within the sub- fective steps and interactions are minimized. Helpful
system seemed to play only a minor role for motivation structural features that have been successfully realized
1176 G. Hertel et al. / Research Policy 32 (2003) 1159–1177

in the Linux project are a module structure that pre- (e.g. Wellman, 1997) might be included in such
vents unnecessary coordination requirements, and follow-up research. Second, the present study is based
simplicity of communication processes (Moon and on a cross-sectional correlational data set. Thus, con-
Sproull, 2002). In addition, attempts can be made to clusions about causal processes are not clear. For
increase the time-resources of participants whenever instance, high correlation between subsystem identifi-
possible. cation and time investments can either occur because
Pragmatic motives to improve one’s own software identification leads to increased time investments,
tools and to increase personal career chances were par- because time investments lead to increased identifica-
ticularly relevant for participants’ willingness to en- tion, or because both variables are affected by a third
gage in Linux activities in the future, but played only a variable such as pragmatic motives. These questions
minor role for the hours the participants actually have have to be addressed in complementing longitudinal
spent in the past. Thus, unrealistic expectations might and experimental studies.
be a danger for the motivation of participants in OSS In conclusion, the present study has contributed to
projects, and it might be fruitful to inform new par- our understanding of motivational processes in OSS
ticipants thoroughly about possible costs and gains of projects and has revealed various motivational forces
their activities in a project. Moreover, pragmatic mo- that contribute to a person’s willingness to engage
tives should be mainly relevant for software projects in such projects both at the community level as well
for which users have basic programming skills. Fur- as at the team level. The motivational forces are in
ther research might compare the demographic and mo- accordance with existing models of voluntary action
tivational structure of OSS projects depending on the and virtual teamwork. Moreover, this study also pro-
specific type of the software developed (operating sys- vides an example of how methodological principles of
tem, business application, etc.). the OSS community might be integrated in social re-
Finally, the results revealed that in OSS projects search. Together, we hope that this research helps to
some part of the development work is accomplished improve software development processes both in OSS
by (spontaneous) teams. The successful validation of projects as well as in other projects when the collab-
the VIST model within the Linux subsystems provides oration depends heavily on the voluntary engagement
a practical heuristic to address motivational challenges of persons in a rather low structured environment.
in such teams when members collaborate from dif-
ferent locations and different time zones. Similar to
virtual teams in business organizations (Hertel et al., Acknowledgements
2002), the perceived indispensability of one’s own
contributions for the team is an important motivational We thank SuSe Germany for sponsoring a number
factor together with a high evaluation of the team goals of lottery prizes for the participants in this study.
and a high sense of personal self-efficacy.
Before closing, we want to mention two important
References
limitations of the present study. First, although the
sample size was sufficient to conduct first quantitative Ajzen, I., Fishbein, M., 1980. Understanding Attitudes and
analyses, it certainly would be desirable to replicate Predicting Social Behavior. Prentice-Hall, Englewood Cliffs.
the presented results in follow-up studies with larger Atkinson, J.W., 1957. Motivational determinants of risk-taking
samples, either within the Linux community or within behavior. Psychological Review 64, 359–372.
Bandura, A., 1977. Self-efficacy: toward a unifying theory of
other OSS projects (see Lakhani et al., 2002, for an
behavioral change. Psychological Review 84, 191–215.
example of such a study that is currently in progress). Batinic, B., Bosnjak, M., 2000. Fragebogenuntersuchungen im
Apart from a higher reliability of the results and Internet [surveys on the Internet]. In: Batinic, B. (Ed.), Internet
higher representativity of the sample, it would also for Psychologists. Hogrefe, Göttingen, Germany, pp. 278–318.
be possible to prevent biases due to systematic differ- Bezroukov, N., 1999a. Open Source Software as a special type of
academic research (critique of vulgar Raymondianism), First
ences between responding and non-responding OSS Monday 4 (retrieved from the World Wide Web, 28 October
community members. In addition to self-report data, 2002: http://www.firstmonday.dk/issues/issue4 10/bezroukov/
objective performance criteria and network analyses index.html).
 G. Hertel et al. / Research Policy 32 (2003) 1159–1177 1177

Bezroukov, N., 1999b. A second look at the cathedral and Taetigkeitsfeld Informationsverarbeitung und Information-
the bazaar, First Monday 4 (retrieved from the World swirtschaft”.
Wide Web, 28 October 2002: http://www.firstmonday.dk/issues/ Lakhani, K.R., von Hippel, E., 2000. How Open Source Software
issue4 12/bezroukov/index.html). works: “Free” User-to-User Assistance, MIT Sloan School of
Bovet, P.D., Cesati, M., 2001. Understanding the Linux Kernel. Management Working Paper #4117 (May). Research Policy,
O’Reilly, Sebastopol. forthcoming.
Bryant, F.B., Yarnold, P.R., 1997. Principal-component analysis Lakhani, K.R., Wolf, B., Bates, J., DiBona, C., 2002. The Boston
and exploratory and confirmatory factor analysis. In: Grimm, Consulting Group Hacker Survey (retrieved from the World
L.G., Yarnold, P.R. (Eds.), Reading and Understanding Wide Web, 11 September 2002: http://www.bcg.com/open-
Multivariate Statistics, fourth ed. American Psychological source/BCGHackerSurveyOSCON24July02v073.pdf).
Association, Washington, pp. 99–136. Licht, M.H., 1997. Multiple regression and correlation. In:
Dempsey, B.J., Weiss, B., Jones, D., Greenberg, J., 1999. Grimm, L.G., Yarnold, P.R. (Eds.), Reading and understanding
A quantitative profile of a community of Open Source Multivariate Statistics, fourth ed. American Psychological
Linux developers, School of Information and Library Science, Association, Washington, pp. 19–64.
University of North Carolina, Chapel Hill, Technical Report Lipnack, J., Stamps, J., 1997. Virtual Teams. Wiley, New York.
TR-1999-05. Moon, J.Y., Sproull, L., 2002. Essence of distributed work: the
Doosje, B., Ellemers, N., Spears, R., 1995. Perceived intragroup case of the Linux kernel. In: Hinds, P., Kiesler, S. (Eds.),
variability as a function of group status and identification. Distributed Work. MIT Press, Cambridge, pp. 381–404. (Also
Journal of Experimental Social Psychology 31, 410–436. available on the World Wide Web: http://www.firstmonday.dk/
Duarte, D.L., Snyder, N.T., 1999. Mastering Virtual Teams. issues/issue5 11/moon/index.html, retrieved 28 October 2002.)
Jossey-Bass, San Francisco. Omoto, A.M., Snyder, M., 1995. Sustained helping without
Hars, A., Ou, S., 2002. Working for free? Motivations for obligation: motivation, longevity of service, and perceived
participating in Open-Source projects. International Journal of attitude change among AIDS volunteers. Journal of Personality
Electronic Commerce 6, 25–39. and Social Psychology 68, 671–686.
Hertel, G., 2002. Management virtueller Teams auf der Basis Raymond, E.S., 1999. The Cathedral and the Bazaar. Musings
sozialpsychologischer Modelle [management of virtual teams on Linux and Open Source by an Accidental Revolutionary.
based on social psychology models]. In: Witte, E.H. (Ed.), O’Reilly, Sebastapol, CA.
Sozialpsychologie Wirtschaftlicher Prozesse. Pabst Publishers, Shepperd, J.A., 1993. Productivity loss in performance groups: a
Lengerich, Germany, pp. 172–202. motivation analysis. Psychological Bulletin 113, 67–81.
Hertel, G., Konradt, U., Orlikowski, B., 2002. Managing distance Simon, B., Loewy, M., Stürmer, S., Weber, U., Freytag, P.,
by interdependence: goal setting, task interdependence, and Habig, C., Kampmeier, C., Spahlinger, P., 1998. Collective
team-based rewards in virtual teams. Manuscript under editorial identification and social movement participation. Journal of
review. Personality and Social Psychology 74, 646–658.
Hertel, G., Deter, C., Konradt, U., in press. Motivation gains Stallman, R.M., 1994a. The GNU manifesto (retrieved from the
in computer-supported teams. Journal of Applied Social World Wide Web, 28 October 2002: http://www.gnu.org/gnu/
Psychology. manifesto).
Jarvenpaa, S.L., Leidner, D.E., 1999. Communication and trust in Stallman, R.M., 1994b. Why software should not have owners
global virtual teams. Organization Science 10, 791–815. (retrieved from the World Wide Web, 28 October 2002: http://
Karau, S.J., Williams, K.D., 2001. Understanding individual www.gnu.org/philosophy/why-free.html).
motivation in groups: the collective effort model. In: Turner, Toch, H., 1965. The Social Psychology of Social Movements.
M.E. (Ed.), Groups at Work: Advances in Theory and Research. Bobbs-Merrill, Indianapolis.
Erlbaum, Mahwah, pp. 113–142. Torvalds, L., 1998. Interview with Linus Torvalds: what motivates
Karau, S.J., Markus, M.J., Williams, K.D., 2000. On the elusive free software developers? First Monday 3 (retrieved from the
search for motivation gains in groups: insights from the World Wide Web, 14 December 2001: http://www.firstmonday.
collective effort model. Zeitschrift für Sozialpsychologie 31, dk/issues/3 3/torvalds).
179–190. Torvalds, L., Diamond, D., 2001. Just for Fun: the Story of an
Kelly, C., Breinlinger, S., 1995. Identity and injustice: exploring Accidental Revolutionary. Harper Business, New York.
women’s participation in collective action. Journal of Van Eerde, W., Thierry, H., 1996. Vroom’s expectancy models
Community and Applied Social Psychology 5, 41–57. and work-related criteria: a meta-analysis. Journal of Applied
Klandermans, B., 1997. The Social Psychology of Protest. Basil Psychology 81, 575–586.
Blackwell, Oxford. Vroom, V.H., 1964. Work and Motivation. Wiley, New York.
Koch, S., Schneider, G., 2000. Results from software engineering Wellman, B., 1997. An electronic group is virtually a social
research into Open Source development projects using public network. In: Kiesler, S. (Ed.), Culture of the Internet. Erlbaum,
data. In: Hansen, H.R., Janko, W.H. (Eds.), Wirtschaftsuniver- Mahwah, pp. 179–208.
sitaet Wien. Working Paper #22 “Diskussionspapiere zum

View publication stats