Informs Management Science: This Content Downloaded From 193.226.62.221 On Sat, 24 Jun 2017 10:09:14 UTC

Understanding the Motivations, Participation, and Performance of Open Source Software
Developers: A Longitudinal Study of the Apache Projects

Author(s): Jeffrey A. Roberts, Il-Horn Hann and Sandra A. Slaughter
Source: Management Science, Vol. 52, No. 7, Open Source Software (Jul., 2006), pp. 984-999
Published by: INFORMS
Stable URL: http://www.jstor.org/stable/20110575
Accessed: 24-06-2017 10:09 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Management Science
This content downloaded from 193.226.62.221 on Sat, 24 Jun 2017 10:09:14 UTC
All use subject to http://about.jstor.org/terms
MANAGEMENT SCIENCE UHB.
Vol. 52, No. 7, July 2006, pp. 984-999 DOI i0.1287/mnsc.l060.0554
issn 0025-19091 eissn 1526-55011061520710984 @ 2006 INFORMS
Understanding the Motivations, Participation, and

Performance of Open Source Software Developers:
A Longitudinal Study of the Apache Projects
Jeffrey A. Roberts
John F. Donahue Graduate School of Business, Duquesne University, Rockwell Hall, Pittsburgh, Pennsylvania 15282,
jeff@roberts.net
Il-Horn Harm
Marshall School of Business, Bridge Hall 401T, University of Southern California, Los Angeles, California 90089,
hann@marshall.usc.edu
Sandra A. Slaughter
David A. Tepper School of Business, 354 Posner, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213,
sandras@andre w. emu. edu
Understanding
development. what motivates
Our study participation
contributes by revealing how theisdifferent
a central theme
motivations in developers
of OSS the research
are on open source software (OSS)
interrelated, how these motivations influence participation leading to performance, and how past performance
influences subsequent motivations. Drawing on theories of intrinsic and extrinsic motivation, we develop a
theoretical model relating the motivations, participation, and performance of OSS developers. We evaluate
our model using survey and archival data collected from a longitudinal field study of software developers in
the Apache projects. Our results reveal several important findings. First, we find that developers' motivations
are not independent but rather are related in complex ways. Being paid to contribute to Apache projects is
positively related to developers' status motivations but negatively related to their use-value motivations. Perhaps
surprisingly, we find no evidence of diminished intrinsic motivation in the presence of extrinsic motivations;
rather, status motivations enhance intrinsic motivations. Second, we find that different motivations have an
impact on participation in different ways. Developers' paid participation and status motivations lead to above
average contribution levels, but use-value motivations lead to below-average contribution levels, and intrinsic
motivations do not significantly impact average contribution levels. Third, we find that developers' contribution
levels positively impact their performance rankings. Finally, our results suggest that past-performance rankings
enhance developers' subsequent status motivations.
Key words : open source software; intrinsic motivation; extrinsic motivation; software development performance
History: Accepted by Eric von Hippel and Georg von Krogh, guest editors; received September 1, 2004. This
paper was with the authors 4 months for 2 revisions.
1. Introduction spectrum including scratching a "personal itch" (Ray

Open source software (OSS) communities cannot exist mond 1999, p. 4) with respect to software function
or prosper without the contributions of highly moti ality, enjoyment, and a desire to be part of a team
vated developers who are willing to donate their time (Ghosh 1998). Others liken the OSS community to a
and effort to the community. However, because these gift culture where the status of a participant depends
participants are often self-employed freelancers and on "what he gives away" (Raymond 1999). Alterna
volunteers, rather than traditional employees, it is not tively, Lerner and Tir?le (2002) suggest that OSS par
possible to solely rely on employment relationships ticipation may in part be explained by existing theories
or employment contracts to manage them. Thus, vital of labor economics. Lastly, as commercial companies
questions in OSS communities concern how to moti increase their involvement in OSS projects, there are
vate participants and how to direct, sustain, and influ more developers being paid to contribute, adding the
ence their behaviors (Markus et al. 2000). traditional incentive?pay?as a potential motivation
Several studies have revealed different motivations to participate and raising the issue of how paid partic
for contributing to open source projects (for a review ipation may affect other motivations to contribute.
see Rossi 2004). Often quoted motivations for partic Although there is no consensus in the OSS literature
ipating in OSS development projects cover a broad as to which motivation is most dominant, individual
984
Roberts et al.: Understanding the Motivations, Participation, and Performance
Management Science 52(7), pp. 984-999, ?2006 INFORMS 985
contributors could likely have multiple salient rea imperative to understand which types of motivations
sons for participating. This raises an important and are likely to generate more (or less) participation.
unanswered question that we address in this study: Our third question concerns the link between the
How are the motivations of contributors related, i.e., level of participation and performance ranking: How
are they independent, complementary, or contradic do levels of participation relate to changes in per
tory? Answering this question is significant because formance rankings? This question is relevant to OSS
an assumption in studies on OSS participation is that communities like Apache, which is the focus of our
motivations are complementary or "mutually rein study. The Apache projects were not originally orga
forcing" (e.g., Markus et al. 2000). However, if, for nized around a single person or primary contributor.
example, some motivations are negatively related to As such, the success of the Apache projects depends
others, increasing the level of those motivations may on shared leadership and the contributions of par
crowd out other motivations for participating. This ticipants. The projects are organized using a meri
issue is particularly relevant when considering how tocracy, or as Roy Fielding explains it: "The more
paid participation affects other motivations for open work you have done, the more you are allowed to
source contributions. In sum, understanding whether do" (Fielding 1999, p. 43). For the meritocracy to
an OSS participant's different motivations are in har be effective, promotions within the Apache commu
mony or at odds matters because it is the participant's nity should be based on contributions to the Apache
set of motivations, combined with knowledge, skills, projects. In answering our third question, we relate
and abilities, that produces the participant's behaviors a participant's promotion (or performance ranking)
to the level of his or her prior contributions to the
and performance (Mitchell and Daniels 2003).
In OSS development, the different motivations to Apache software code. This provides an important
validation of whether the meritocracy is functioning
participate have been generally classified as either as intended.
intrinsic or extrinsic (Rossi 2004). Intrinsic motiva
Finally, although motivation is an antecedent of
tion occurs when an activity satisfies basic human
behavior and performance, research in psychology
needs for competence, control, and autonomy. This
has also recognized the effect of performance feed
makes the activity interesting and likely to be per
formed for its own sake rather than as a means to back on motivation. For example, it has been shown
that feedback considered controlling tends to decrease
an end (Deci and Ryan 2000). In contrast, extrinsic intrinsic motivation. On the other hand, research
motivation stems from the environment external to
by Sansone (1986) suggests that feedback regarded
the task and is usually applied by someone other than
as competence enhancing can increase subsequent
the person being motivated (Johns 1996). Contribut intrinsic and extrinsic motivation. One of the tenets
ing to OSS projects for the sheer enjoyment of coding
of OSS projects is the frequent provision of feedback
is clearly an intrinsic motivation whereas being paid to contributors (Moon and Sproull 2002). As we have
to contribute is the quintessential extrinsic motiva noted, in some OSS projects like Apache, continued
tion. Other motivations, such as contributing to solve contribution is rewarded with a change in perfor
a problem of personal use benefit (use value) or con mance ranking. Therefore, the last question on our
tributing to enhance status or career opportunities are, agenda is: How does a change in performance rank
by definition, extrinsic, but following Deci and Ryan ing affect the subsequent motivations of OSS partic
(2000), contributors could internalize these motivations ipants? To the best of our knowledge, no studies of
so that they are self-regulated rather than externally OSS communities have considered how changes in
imposed. Following the literature in psychology (Deci performance rankings affect subsequent motivations
and Ryan 1987), we classify these motivations as inter for participating. The answer to this question has
nalized extrinsic motivations. Distinguishing the differ important implications for open source communities
ent types of motivations (pure extrinsic, internalized that wish to enhance or sustain the motivations of
extrinsic, and pure intrinsic) allows us to examine a their participants over the longer term.
second question: How do differences in OSS contrib In ?2 of the paper, we draw from the relevant
utors' motivations relate to differences in their par literature on psychology to develop our theoretical
ticipation? It is important to understand whether all model. Section 3 describes the empirical evaluation of
types of motivations affect OSS participation equally our model involving a longitudinal field study of the
or in the same way. Studies of OSS participants do not motivations, contributions, and performance of soft
often consider whether different motivations differen ware developers in the Apache Web server projects.
tially relate to participation. However, some motiva Section 4 presents the analysis and results. We dis
tions may strongly affect participation whereas others cuss our results in ?5 and conclude in ?6 by identi
may not be as salient. For leaders of OSS projects fying the contributions and limitations of our study
who are trying to attract developers to participate in and its implications for research and practice in OSS
projects or to sustain their level of participation, it is development.
986 Management Science 52(7), pp. 984-999, ?2006 INFORMS
Figure 1 Theoretical Model
Experience
2. Theoretical Framework exemplified by the level of their contributions to the

The theoretical framework for our study leverages source code. Over time, contributors' participation is
the general model of motivation and performance evaluated by the OSS community. This performance
in organizational and social psychology (Campbell evaluation may lead to a rise in a contributor's rank
and Pritchard 1976). In this framework, motivations within the community, which can, in turn, act as feed
vary across individuals and combine with individu back to influence the future motivations of contribu
als' knowledge, skills, and abilities to produce task tors. Figure 1 illustrates our theoretical model.
relevant behaviors. These behaviors contribute to
2.1. Intrinsic and Extrinsic Motivations in
individual performance. It is important to distinguish
motivation from behavior. Work motivation is the OSS Development
OSS development affords a particularly rich context
psychological force within an individual that deter in which to examine individual motivations. On one
mines the direction of the individual's behavior in an
hand, intrinsic motivations are likely to be impor
organization, the individual's level of effort, and the
tant as contributors have a high degree of auton
individual's level of persistence in the face of obsta omy and self-determination and are valued for their
cles (Kanfer 1990). Motivation has an important influ competence. On the other hand, the OSS commu
ence on performance because it focuses attention on nity provides extrinsic motivations such as reputation
particular task elements and produces effort as peo or status. Indeed, empirical studies of OSS contribu
ple work harder when they are motivated. Motiva tors find that the participants report a variety of rea
tion is a psychological state, whereas the outcome or sons for participating in projects (Hertel et al. 2003).
results of that state is behavior (Mitchell and Daniels Our model therefore includes an array of OSS partic
2003). Behavior also differs from performance because ipation motivations ranging from strictly intrinsic to
performance is an evaluation of the results of an strictly extrinsic to those that have both intrinsic and
individual's behavior usually by someone other than extrinsic characteristics.
the individual?it involves determining how well or Research has investigated the relationship between
poorly an individual has accomplished a task (Kanfer psychological needs and intrinsic motivation. Specif
1990). ically, intrinsic motivation has been linked to the
Applying the general model of motivation and per satisfaction of human needs for autonomy and com
formance to the OSS context, we stipulate that moti petence (Deci 1975). In the OSS context, human needs
vations vary across OSS contributors. Combined with for autonomy and competence are readily satisfied.
developers' knowledge, skills, and abilities, motiva As researchers have established, software develop
tions influence their participation in OSS projects as ment is an inherently motivating task because it is
complicated and creative but difficult to observe and Status seeking or career enhancement may both be
thus not easily monitored (Kirsch 1996, Weinberg internalized in an effort to increase one's standing in
1998). Compared to software development in an orga a reference group.
nizational setting, contributing to OSS projects allows Internalized extrinsic motivation based on identified
individuals even greater opportunities to express regulation is a more self-determined form of extrin
their creativity, enjoy their work, and experience a sic motivation. In this form of regulation, individu
sense of satisfaction and accomplishment (Lakhani als identify with an action and personally endorse
and Wolf 2005). It is these characteristics that draw it, leading to an identification that is accompanied
many competent, self-directed programmers into OSS by a higher degree of perceived autonomy. By defi
communities. nition, this type of motivation is extrinsic because it
At the other extreme, extrinsic motivation has been derives from the personal importance of the outcome
linked to the op?rant conditioning literature (Skinner rather than the performance of the task (Ryan and
1953) that advocates the use of incentives to reinforce Deci 2002). In OSS communities, an example of an
desired behavior. That is why extrinsic motivation is internalized extrinsic motivation is use value, or the
characterized by a strong focus on reward contingen desire to fix a bug or solve a problem of immediate
cies, which in Western-oriented economies is often relevance to the contributor. Some studies have iden
represented by pecuniary compensation. Although tified use value as a dominant motivation of OSS con
OSS communities do not have a profit motive per se tributors (e.g., Ghosh et al. 2002, Hertel et al. 2003, von
and hence do not offer monetary compensation to Hippel 2001). Although use value is extrinsic because
contributors, the pervasive commercial interest in of the personal benefit to the user (Markus et al. 2000,
many OSS products has generated a thriving OSS Rossi 2004), from a psychological point of view, use
industry. It is not unusual for third parties to employ value is internalized as a value of the OSS community
programmers for the specific purpose to contribute and transformed into a personally endorsed value.
to OSS projects. Examples include Time Warner's In this study, we examine the relationships between
engagement in Mozilla and IBM's involvement in different OSS motivations. The psychology litera
both the Linux and Apache OSS projects. ture on motivation has examined the relationships
Between intrinsic and extrinsic motivations there between intrinsic and extrinsic motivations (Lepper
exists a continuum of motivations that are consid and Henderlong 2000). We consider how these rela
ered a mix of intrinsic and extrinsic; they are clearly tionships may apply in the OSS context.
not intrinsic at the outset, but undergo an internal Experimental research in psychology (Deci 1971,
ization process that moves them away from strictly Lepper et al. 1973) has shown that under certain con
extrinsic motivations (Ryan and Connell 1989). Deci ditions, extrinsic motivations displace intrinsic moti
and Ryan (1987) refer to these motivations as internal vations. In cases where incentives are contingent upon
ized extrinsic motivations. According to Deci and Ryan performance, or individuals expect to be rewarded, or
(2000), internalization occurs when individuals assim incentives are tangible, external incentives undermine
ilate and reconstitute external incentives or contin characteristics of intrinsic motivation such as free
gencies, transforming these external incentives into choice behavior and self-reported interest. Multiple
their own motives and leading to self-regulation. Self meta-analysis studies (Rummel and Feinberg 1988,
regulation can take place in two ways: through intro Wiersma 1992, Deci et al. 1999) have found general
jection or through identification (Deci and Ryan 2000). support for this finding; Osterloh and Frey (2000)
Internalized extrinsic motivation based on introjected refer to this as the "crowding-out" effect. In general,
regulation is related to attainment of ego enhancement previous research has found that the undermining
and feelings of worth. In the OSS context, an example effect of external incentives is especially powerful for
of this motivation involves two related quests: for sta monetary compensations that are perceived to be con
tus and for career opportunities. Early on, Raymond trolling. The effects are larger for monetary rather
(1999) recognized status, or what he termed "ego than symbolic incentives and for expected rather than
boo," as an important driver of participation in OSS unexpected incentives. The crowding-out effect is also
communities. Similarly, OSS participation can be seen more observable for complicated rather than simple
as a move to enhance career prospects. As described tasks (Deci et al. 1999, Lepper and Henderlong 2000).
by Lerner and Tir?le (2002), OSS communities offer In the context of OSS projects, strictly extrinsic
an excellent setting in which a participant motivated motivations apply directly to those participants who
by career concerns can signal his or her abilities to the are remunerated for their activities. Drawing on the
labor market. Likewise, von Hippel and von Krogh logic of the crowding-out effect, we expect that mone
(2003) argue for the existence of private incentives tary incentives for involvement in OSS projects would
for the provision of a public good such as OSS. It weaken intrinsic motives for participating. When par
is important to note that motivations based on either ticipants are engaged in a traditional employment
relationship with a firm, the employer has the right Ryan (1982) finds that internalized extrinsic moti
to establish the policies, rules, and terms of employ vations based on introjection such as ego-enhancing
ment in order to structure the work environment and motivations reduce intrinsic motivation for the tar
assign tasks to employees. This suggests that partici get activity. Individuals may develop mastery for the
pants who are paid to contribute to OSS projects are purposes of gaining reputation, but find that moti
likely to have less autonomy in choosing which fea vation based on introjected regulation is quite con
tures they want to code and less freedom in how they trolling and less self-regulated. In the context of OSS
do their work. If, as Weinberg (1998) argues, program development, individuals who are highly motivated
ming itself provides the strongest intrinsic motivation by status could find themselves working on tasks they
for software development, "...if the programmer is may not necessarily enjoy but that are rather likely
given a chance to do it his way..." (p. 184), any to enhance their reputation in the community. For
restriction of participants' task autonomy should be example, writing open source software, and helping
negatively associated with their intrinsic motivations to test and debug it are critical ways to earn respect
to participate in OSS projects. in OSS communities (Markus et al. 2000). However,
a contributor who is motivated by reputation con
Hypothesis 1A (H1A). Contributors' intrinsic moti cerns may not particularly enjoy testing and debug
vations to participate in OSS projects are negatively related ging software, but may feel it necessary to do these
to being paid to participate. activities to gain status in the OSS community. This
suggests that motivations grounded in status-seeking
The literature in psychology suggests that other motivation are likely to be negatively associated with
extrinsic motivations, even those that are internal
the pure enjoyment of contributing.
ized (and thus not strictly extrinsic) could crowd out
intrinsic motivations. A meta-analysis by Deci et al. Hypothesis IC (H1C). Contributors' intrinsic moti
(1999) examined the results of 128 laboratory studies vations to participate in OSS projects are negatively related
of extrinsic and intrinsic motivation conducted over to their status motivations to participate.
the past 25 years. The results from this meta-analysis Much of the research in psychology has focused
show that most types of extrinsic motivations?espe on the effects of external incentives on intrinsic
cially those specifying contingencies related to the motivation. However, Ryan and Deci (2002) sug
task being performed?undermined intrinsic motiva gest that external incentives that promote feelings
tion. The authors found a positive effect of extrin of self-determination can promote self-determined
sic motivations on intrinsic motivation only when (i.e., internalized) extrinsic motivations. Performance
the feedback was both positive and verbal. With contingent incentives (such as pay) can influence how
respect to internalized extrinsic motivations that are individuals approach a task as well as their moti
based on the identification of values, Ryan and Deci vations during the performance period (Hennessey
(2002) observe that such identification is often com 2000) because such incentives can increase the impor
partmentalized and separated from one's other beliefs tance of doing well on a personal level. That is,
and values, and is characterized by a reduced self incentives that motivate individuals to strive for com
determination. Therefore, a contributor may identify petence can amplify or enhance their other extrin
with the OSS community through use-value motiva sic motivations. Following this logic, we expect that
tion only in a restricted, practical sense with lim being paid to contribute to OSS projects is comple
ited self-determination and reduced pure intrinsic mentary with other extrinsic motivations including
motivation. For example, a contributor who identifies use value and status. As a consequence, individu
strongly with an OSS community may choose to work als who are being paid to contribute to OSS projects
on tasks that are not inherently interesting because could also have a high use-value motivation because
completing the tasks provides value to the commu making contributions that improve the use value of
nity. In her review of the literature on OSS motiva the source code demonstrates competence for which
tions, Rossi (2004) states that use-value motivations they will be financially rewarded.
can provide a powerful explanation for why people Hypothesis 2A (H2A). Being paid to participate in
do tasks that may be uninteresting and mundane, and OSS projects is positively related to contributors' use-value
that are not "appreciated per se, for the intrinsic plea motivations to participate.
sure and enjoyment a programmer may derive from
them" (p. 5). Similarly, individuals who are being paid to con
tribute to OSS projects are likely to have higher sta
Hypothesis IB (H1B). Contributors' intrinsic moti tus motivations because an interest in attaining status
vations to participate in OSS projects are negatively related motivates individuals to demonstrate competence for
to their use-value motivations to participate. which they will be financially rewarded.
Hypothesis 2B (H2B). Being paid to participate in The literature on organizational psychology sug
OSS projects is positively related to contributors' status gests that the "ideal" intrinsic motivation is "in the
motivations to participate. work content itself" (Calder and Staw 1975, p. 539).
Tasks that are intrinsically motivating have a direct
2.2. Motivations and Participation in OSS and strong association between the activity and
Development the individual's purpose for performing the activ
According to the classic literature on op?rant condi ity. Therefore, engaging in the task directly satis
tioning (Skinner 1953), behavior that is rewarded with
fies the individual's goals. In an experimental study,
positive reinforcement is more likely to be repeated
Shah and Kruglanski (2000) find that the strength
in the future. In Western societies, monetary com
of the activity-goal association is positively related
pensation is the ultimate positive reinforcer to regu
to indices of intrinsic motivation, including the self
late economic activities in organizations. Hence, we
reported frequency of engagement in the activity and
expect that contributors who are paid to participate
the importance one places in doing the activity. Shah
in Apache projects would participate more intensely
and Kruglanski conclude that individuals who are
than those who are not paid. This is because their
intrinsically motivated to perform some activity will
wages act as a constant positive reinforcer of their
perform it very intensely. In the context of OSS devel
participation behavior.
opment, intrinsically motivated contributors should
Hypothesis 3 (H3). Being paid to participate in OSS have higher levels of participation because they like to
projects is positively related to contributors' level of code, and by coding they are directly satisfying their
participation. desires.
We also expect that individuals with higher levels Other psychologists have linked intrinsic motiva
of use-value motivations would exhibit higher levels tion to task participation via its effect on creativity,
of participation. In terms of use-value motivations, because an intrinsically motivated orientation to task
one of the most often cited drivers of OSS participa performance promotes characteristics that are essen
tion is the opportunity to create code that meets the tial for creativity (Amabile et al. 1986). For example,
specific needs of a developer (Raymond 1999). Fixing individuals with high levels of intrinsic motivation
a bug or solving a problem of immediate relevance focus more on the task, are more willing to take risks,
to the programmer provides a powerful motivation and will explore alternative strategies for performing
to create the software code in the first place (Lerner the task (Hennessey 2000, Osterloh and Frey 2000).
and Tir?le 2002). High levels of use-value motivation Higher creativity should lead to higher participation
therefore suggest a high level of participation. in OSS projects for several reasons. First, it should
Hypothesis 4A (H4A). Contributors' use-value moti focus the developer's attention on the task of cod
vations to participate in OSS projects are positively related ing. Second, it should help developers to persist in
to their level of participation. solving difficult or challenging problems by exploring
alternatives and "thinking outside the box." Indeed,
Status motivations should also be a strong driver
a survey study by Lakhani and Wolf (2005) provides
of participation. Raymond (1999) likened OSS com
empirical support for a link between creativity and
munities to gift cultures, where the sought-after sta
task participation. The researchers found that a per
tus is determined by the programmer's contribution.
sonal sense of creativity has the strongest associa
In addition, human-capital theory (Becker 1962) sug
tion with effort (hours worked) by contributors to
gests that individuals, endowed with differing apti
OSS projects. As such, we expect contributors with
tudes and abilities, will strive to acquire additional
knowledge and experience as long as the expected higher intrinsic motivations to participate more sub
incentives are greater than the expected costs. In a stantially in OSS projects, because intrinsic motivation
slight variation, signaling theory (Spence 1976) pre promotes the characteristics needed to perform soft
sumes that individuals showcase their education and ware development.
experience to signal imperfectly observable produc Hypothesis 5 (H5). Contributors' intrinsic motiva
tivity characteristics to current and future employers. tions to participate in OSS projects are positively related
Human-capital and signaling theories suggest that to their level of participation.
OSS contributors who are motivated by status con
cerns will participate as a way of improving and sig 2.3. Participation and Performance in
naling their programming abilities and competencies OSS Development
(Harm et al. 2006).
As we have noted, psychologists distinguish perfor
Hypothesis 4B (H4B). Contributors' status motiva mance from behavior. Performance is the outcome of
tions to participate in OSS projects are positively related an evaluation by others of an individual's behavior,
to their level of participation. and this behavior is often manifested by individuals'
task output (Mitchell and Daniels 2003). In the con As early as 1943, Hull proposed that motivation can
text of OSS development, several OSS communities arise from the reinforcer itself. He termed this incen
periodically evaluate the actual contributions of their tive motivation. Incentive motivation is dependent on
members and assign each member a certain perfor the strength of the incentive. As the size of an incen
mance ranking. These rankings are based on merit tive increases, so does the level of incentive motiva
and reflect the contributors' level of participation in tion and the likelihood of the individual behaving in
the OSS community. Advancement within the meri such a way as to bring about an even greater incentive
tocracy recognizes individuals' commitment and con in the future. This is because the receipt of extrinsic
tributions to the OSS projects (Fielding 1999). incentives imparts information about the likelihood of
receiving future extrinsic incentives for similar behav
Hypothesis 6 (H6). Contributors' level of participa iors (Lepper and Henderlong 2000). Expectations of
tion in OSS projects is positively related to their perfor future incentives can thus provide continued extrinsic
mance ranking. motivation for an individual to engage in previously
rewarded activities.
In considering the antecedents of participation and In the context of OSS communities, increases in per
performance, we control for individual contributors' formance ranking could increase contributors' inter
knowledge, skills, and abilities in terms of their level nalized extrinsic motivations in several ways. From
of education and experience. In software develop a psychological point of view, a developer who con
ment, both education and experience are very impor tributes code that improves the use value of the soft
tant antecedents of productive capacity (Ang et al. ware will experience a subsequent increase in rank,
2002). Thus, consistent with the literature on task per and this will reinforce the internalization process of
formance (Campbell and Pritchard 1976), we expect the use-value motivation through identification (Ryan
that developers' education and experience positively and Deci 2002). She could therefore expect that future
relate to their level of participation in OSS projects. use-value contributions will strengthen her identifica
tion with the community and lead to renewed inter
2.4. Past Performance and Subsequent nalization through future increases in rank. In this
Motivations in OSS Development way, a prior increase in rank could increase devel
Studies in psychology show that although incen oper's internalized extrinsic motivation to make use
value contributions in the future.
tives and feedback that are viewed as controlling can
decrease an individual's subsequent intrinsic moti Hypothesis 8A (H8A). An increase in contributors'
vation (Deci and Ryan 2000), the opposite is true performance ranking is associated with an increase in their
if these external evaluations are presented as indi subsequent use-value motivations to participate.
cators of personal competence (Pittman et al. 1980,
In addition, a promotion in rank is certainly asso
Ryan et al. 1983). Incentives that have an inform
ciated with greater status in the OSS community
ing aspect about task performance can increase indi
(Raymond 1999, Lerner and Tir?le 2002). Following
viduals' feelings of internal control and self-efficacy,
Ryan (1982) a promotion in rank in one time period
and can raise their level of enjoyment in the task could therefore reinforce internalization of the status
(Lepper and Henderlong 2000). This is because peo motivation through introjection. The increase in a con
ple tend to like to do things they think they are goodtributor's status motivation could then lead to future
at (Sansone 1986). Thus, all else equal, an increase source code contributions. Inasmuch as an increase in
in an individual's perceived competence at an activ status is seen as career enhancing (Lerner and Tir?le
ity should increase his or her level of intrinsic moti 2002, Harm et al. 2006), a rank increase will also lead
vation for the task. In the OSS context, we expect to a reinforcement of the status motivation through
that merit-based performance rankings would have introjection.
more of an informing aspect than a controlling aspect.
Hypothesis 8B (H8B). An increase in contributors'
An advance in rank communicates important infor
performance ranking is associated with an increase in their
mation to contributors about their ability and pro
subsequent status motivations to participate.
ductive capacity in software development as well as
the value of their contributions to the OSS commu
nity. Following Sansone (1986), an increase in ranking
3. Method
should therefore enhance contributors' sense of com We evaluate our hypotheses empirically, analyzing
petence, self-efficacy, and enjoyment in participating. archival data collected from OSS project records over
a period of four years, and from a targeted survey
Hypothesis 7 (H7). An increase in contributors' per of OSS participants. The following describes the set
formance ranking is associated with an increase in their ting of the data collection, each data source, and our
intrinsic motivations to participate. measures of key variables.
3.1. Research Setting of individuals within each project. To extract infor

We investigated three major OSS projects under the mation about individual contributions, we developed
control of the Apache Software Foundation (ASF). tools to mine submissions of the individual devel
The ASF includes a number of subprojects related opers. A submission to an OSS project is known as
to the development of a full-featured Web server a "patch"?an analogue to modification requests in
product offering. We studied the largest and most traditional software development environments. The
significant of these projects, including the Apache data encompassed contributions made and accepted
Web server project, which is a freely available source into any of our three target Apache projects. Data col
code implementation of an hypertext transfer proto lection was completed in January 2003 and included
col (http) server and is the project around which the all contributions from 1999 to 2002.2
Apache Group initially formed; the Jakarta project,
which currently consists of 19 Apache-related Java 3.2.2. Survey Data. A secure, Web-based survey
subprojects; and the XML project, which currently of Apache contributors was conducted to obtain
consists of 16 Apache-related XML subprojects. respondent motivations for participation in the
The Apache context is well suited for examining project. Dr. Roy Fielding, then chairman of the ASF,
the relationships between motivation, participation, introduced the survey to 1,301 uniquely identified
contributors via e-mail in November 2000. In all 233
and performance in OSS development. As a meri
e-mail invitations were undeliverable. Of the remain
tocracy, status, responsibility, and benefits are com
mensurate with contribution (Fielding 1999). There ing 1,068 contributors, 325 completed the instrument
are several observable levels of recognition or rank thereby yielding a response rate of 30%. Thirty-seven
within the ASF. In order of increasing status, these responses contained one or more missing pieces of
are developer, commuter, project management committee information and were thus dropped from further
member, and ASF member. In all cases, advancement analysis, yielding a usable sample of 288 responses.
within the hierarchy is in recognition of an individ An analysis of response bias using nonparametric
ual's commitment and contributions to an Apache tests of location and empirical distribution indicates
project. Although the number of attainable ranks is that our sample is representative of the overall popu
limited, the number of promotion opportunities at lation of Apache contributors.3
any rank is not constrained. For example, there is no
limit to the number of contributors who can achieve 3.3. Measures
the rank of developer or to the number of developers In the following paragraphs, we first define our mea
who can be promoted to the rank of committer. Simi sures of motivation followed by a description of
larly, a promotion to ASF member is not contingent onour measures of OSS participation and performance.
a number of predetermined positions (Fielding 1999). Our data form a panel covering years 1999 to 2002
As a result there is no rationing of promotion oppor (denoted as Periods 1 through 4) with measures of
tunities, and advancement reflects an objective mea individual participation (in Periods 2 and 3), perfor
sure of a positive peer review of one's performance.1 mance (in Periods 3 and 4), and past performance (in
3.2. Data Collection Periods 1 and 2). To this panel we add cross-sectional
survey data collected toward the end of Period 2,
3.2.1. Archival Data. All OSS work products are containing the indicators of respondent motivations.
placed in the public domain under various "free soft The panel provides the basis for the development of
ware" licensing arrangements. Apart from the source all subsequent measures and tests of hypotheses. To
and binary codes of the actual software programs, highlight the temporal nature of the measures and
Apache products include developer websites, change their relationships, we use the following notation. For
logs, documentation, and developer communications variables denoted as Xf2, the subscript t\ denotes the
in the form of e-mail archives. From these products, sum of variable X for periods a through z. Also note
we extracted two types of information: each contrib that, Xta, represents the cross-sectional value of vari
utor's progression along the Apache career path, and able X for period a.
each contributor's source code contributions to the
project.
3.3.1. Motivations. Following theoretical discus
sions of work motivation (Mitchell and Daniels 2003),
To assess a contributor's performance, we captured
the upward progression as a series of discrete transi we conceptualize OSS participation behavior as being
tions from one level to another in the ASF meritoc "driven" by different motivational underpinnings.
racy. This resulted in a time line for the promotion
2 A more detailed description of the archival data and the data
1 Further details on the research site can be found in the online extraction process are in the online appendix.
appendix on the Management Science website (http://mansci.pubs. 3 Detailed results from our assessment of response bias are reported
informs.org / ecompanion.html). in the online appendix.
These motivations can be thought of as existing on and operation suggest that it would be improbable
a continuum ranging from purely extrinsic to purely for participants to advance in the Apache meritocracy
intrinsic. Adopting this perspective, monetary com without substantive and sustained software code con
pensation is viewed as more strictly extrinsic than tributions.5 Thus, we measured participation based on
other types of extrinsic incentives (e.g., Calder and the number of source code contributions submitted
Staw 1975). Individuals are extrinsically motivated if and accepted by the project. A potential concern with
they are able to satisfy their needs indirectly, espe this measure is whether accepted source code con
cially through monetary compensation; indeed, the tributions equal submitted contributions, that is, does
"ideal" extrinsic incentive identified in the literature is consideration of only accepted contributions underes
strict "pay for performance" (Osterloh and Frey 2000, timate participation. To investigate how far this con
p. 539). Consistent with this idea, we measured extrin cern is salient for the Apache projects in our study,
sic motivation (EXTRINSIC^) as the average number we investigated all contributions submitted by a ran
of hours per week for which respondents were paid domly selected 10% of the contributors in our sam
for their Apache development efforts. ple. Because Apache does not distinguish or track the
We also conceptualized two internalized extrinsic number of "accepted" versus "rejected" submissions,
motivations: use value and status. To measure use-value we searched through e-mail archives to follow the his
motivations (USE VALUE^), we draw upon the con tory of each participant's interactions with the Apache
ceptualization of use value in the OSS literature (e.g., community. We found that the participation behavior
von Hippel 2001), using two-scale items that cap of these contributors was more similar to a revise- and
ture the extent to which solving bugs or problems, or resubmit process than to an accept-or-reject process.
adding needed features is important to developers in As described by other OSS researchers, the partici
motivating their participation. We assess status moti pants in our sample first engaged in getting-to-know
vations (STATUS^) using measures consistent with behavior by analyzing the source code or participat
the OSS literature regarding the motivating potential ing in discussions on message boards or both. Von
of status (Raymond 1999). Four-scale items capture Krogh et al. (2003, p. 16) have characterized this behav
the extent to which participants are motivated by sta ior as "a significant period of observation (lurking)."
tus considerations to make contributions. Often, the contributors' submissions were accepted
Finally, consistent with the literature on motiva without modification. Sometimes, contributors were
tion (Lepper and Henderlong 2000), we operational asked to make changes before their submissions were
ized intrinsic motivation in terms of the motivating accepted. This revise-and-resubmit procedure is simi
potential of the task itself. That is, intrinsic motivation lar to that described by Raymond (1999) and Markus
is the extent to which participants make code con et al. (2000). None of the code submissions for our
tributions because developing software is an activity random sample of Apache contributors was rejected
they enjoy and one that satisfies their needs for com outright. This suggests that our measure of partic
petence, control, or autonomy. We measured intrin ipation as code submitted and accepted provides a
sic motivation (INTRINSIC^) using four-scale items reasonable estimate of the level of participation for
designed to capture the extent to which OSS partic contributors.
ipants are motivated by aspects of the task itself to To capture the immediate and subsequent effects of
make contributions. motivations on behavior, our measure of participation
Following Anderson and Gerbing (1988) we as (PARTICIPATION^) is derived using contribution
sessed the psychometric properties of the motivation based metrics for Periods 2 and 3?the period cov
measurement scales used in this study for content, ering the survey and the following period. As a
convergent, and discriminant validity as well as for check of robustness, we derived several alternative
reliability. Overall, the tests provide strong support measures of contributions. The first measure is sim
for the reliability and validity of the motivation mea ply the cumulative number of patches submitted
sures used in this study4 and accepted into the software revision control sys
3.3.2. Participation. Although OSS volunteers pro tem for the particular year. The number of lines of
vide many different kinds of valuable services to their software code written or changed is a commonly
respective projects (Shah 2004), a principal partici used productivity metric in software development
pation behavior consists of authoring and maintain organizations (Boehm et al. 2000); thus, our second
measure is the cumulative number of lines of code
ing the software, i.e., writing lines of software code.
Insider accounts regarding OSS project organization submitted and accepted for the given year. Finally, to
account for potential productivity differences between
programming languages of the Apache subprojects
4 The online appendix lists the measurement scale items and their
sources for each measure of motivation. Results from the reliability
and validity analyses are also reported in the online appendix. 5 Personal exchange with Dr. Fielding.
under consideration, each contribution is converted Following the extant research (e.g., Ang et al. 2002),
to a common function point metric using industry we measured these constructs using demographic
standard language conversion factors (Boehm et al. survey items in which respondents reported their
2000). Our primary analysis uses the function point years of education (EDUCATION^), and their total
metric as the measure of contributors' participation. years of work experience (EXPERIENCE^).
Lastly, to control for project-level idiosyncrasies that
may influence participation, we operationalized each 4. Analysis and Results
measure of participation as the deviation of the mea Our theoretical model stipulates measurements in
sure from the applicable subproject mean. various time periods. Motivation is an antecedent
3.3.3. Performance. To operationalize individual of participation, and participation is an antecedent
performance we leveraged the fact that Apache oper of performance. Naturally, the data underlying these
constructs has to reflect this sequence. In our research
ates as a meritocracy (Fielding 1999). Promotion to a
higher rank within the Apache hierarchy is awarded design this requires the collection of data of partic
after one or more cycles of contribution followed ipation and performance of up to two years after
the measurement of the motivation constructs. Using
by a positive peer review and is, consequently, an
acknowledgement of an individual's substantive con
past performance as an antecedent of motivation
tributions to the project. This operationalization of
requires data collection of up to two years prior
to the measurement of the motivation constructs.
performance is consistent with the literature in psy
Including past performance in this model reduces
chology as Mitchell and Daniels (2003) explain, [per
the number of data points due to "late entry" into
formance is an] "outside standard that is... usually
the Apache career. Therefore, we follow a two-step
assessed by others" (p. 227). Measuring performance
estimation strategy. In the first step we estimate
as rank advancement in the Apache meritocracy sat
the motivation-participation-performance relation
isfies the important criteria that someone other than
ships with all respondents, thereby utilizing the maxi
the individual being evaluated is making the rat
mum number of available data points. To test the past
ing. In addition, meta-analysis studies in psychology
performance-motivation relationships of H7 and H8,
suggest that measuring performance over time using
we augment our primary model with our measure
measures of promotional progress is one of the most
of past performance, analyzing data only from those
reliable ways to measure performance, and that mea
respondents who had started their Apache careers as
sures of promotional progress also have higher valid of Period 1.
ity than other types of performance measures (Meyer
1987). Rank advancement in the Apache meritocracy 4.1. Model and Estimation
is clearly an indicator of promotional progress, and To test our hypotheses, we specified the follow
should therefore be a reliable and valid measure of ing structural equations in a simultaneous equation
performance. model (SEM):6
We operationalized our performance measure
PERFORMANCE^ = a0 + ax * PARTICIPATION^ + et
(PERFORMANCE^) as the number of the changes in
ASF rank experienced from Period 3 to Period 4?the PARTICIPATION^
period after our measures of motivation and partic
= ?0 + ft * INTRINSIC^ +?2* EXTRINSIC^
ipation. This temporal distinction between our mea
sures of motivation, participation, and performance is + ?33 * USE VALUER + ?i * STATUS^
consistent with the general model of motivation and + ?5 * EDUCATION^ +?6* EXPERIENCE,! + ?2
performance in psychology (e.g., Mitchell and Daniels
2003) in which the relationship between motivation, INTRINSIC,! = y0 + y1 * EXTRINSIC,2
behavior, and performance is properly considered as + y2 * USE VALUER + ^ * STATUSf22
a sequence and not as simultaneous events. As we did
+ j y4 * PERFORMANCE,!} + e3t
for our measure of participation, we control for pos
sible subproject differences in our performance mea USE VALUER = rj0 + Vl * EXTRINSIC,!
sure by operationalizing respondent performance as + {rj2 * PERFORMANCE,! ) + e4,
a deviation from the subproject mean performance.
STATUS,! = 4>0 + 4>x * EXTRINSIC,!
3.3.4. Contributor's Knowledge, Skills, and Abil
ities. The literature on performance has identified + j <f>2 * PERFORMANCE,! } + s5t.
individual characteristics such as knowledge and Our analysis approach follows the factor ana
skills as antecedents of participation. As described lytic (FA) SEM (i.e., FASEM) approach most com
earlier, these characteristics are difficult to measure
and are frequently assessed through the use of prox 6 Lagged PERFORMANCE (for Periods 1 and 2) is in equations for
ies, such as the level of education and experience. INTRINSIC, USE VALUE and STATUS only to test H7 and H8.
monly used to evaluate path analysis models with viewed a constrained version of the saturated model
latent variables (Anderson and Gerbing 1988). Due where 12 of the model paths are constrained to 0.
to nonnormality in our data (normalized Mardia's Thus constrained, the hypothesized model achieves
coefficient = 11.32), we estimated our model using a statistically indistinguishable level of performance
elliptically reweighted least squares (ERLS). ERLS from the fully saturated model. Consistent with the
has been shown to be superior to maximum like A^2, relevant model fit statistics are unaffected by the
lihood when estimating models where data exhibit additional constraints placed on the model.
even moderate departures from multivariate normal Figure 2 shows the estimated standardized path
ity (Sharma et al. 1989). coefficients and model fit statistics for both our
We evaluate model fit using a two-index compar primary and enhanced models.8 H1A, H1B, and H1C
ison strategy. Following Hu and Bentler (1997), we predicted a negative relationship between strictly
judged the adequacy of the hypothesized models by extrinsic motivations and intrinsic motivations and
first examining the x2 f?r significance and the ^2/df between the internalized extrinsic motivations (use
ratio for a value less than 3. Then, we compared value and status) and intrinsic motivations. These
the model standardized root mean squared resid hypotheses are not supported. The paths from USE
ual (SRMR) to a cutoff value of 0.08 and either a VALUEf2 to INTRINSIC^ and from EXTRINSIC^ to
CFI "close to" 0.95 or a root mean squared error of INTRINSIC^ are not significant in our model. The
approximation (RMSEA) where the 90% confidence path from ?TATUS^ to INTRINSIC^, while signifi
interval includes a value of 0.07 or less. cant, is positive, thus in the opposite direction than
predicted (y3 = 0.542, p < 0.01). H2A and H2B pre
4.2. Results dicted positive relationships between being paid to
We assessed model performance against several stan contribute to Apache and contributors' status and
dard model instantiations (Marsh 1994)7 The first is use-value motivations. These hypotheses are partially
a saturated model where all possible paths between supported. The path from EXTRINSIC^ to STATUS^
structural model variables, both manifest and latent, is positive and significant (fa = 0.157, p ? 0.04), pro
are estimated. The saturated model provides a basis viding support for H2B. However, the path from
for subsequent model comparisons as all other mod EXTRINSIC^ to USE VALUER, while significant, is in
els containing the same variables will be nested the opposite direction than predicted by H2A (y\x ?
within this model. The second is the null or indepen ?0.132, p = 0.05). H3 relates the strictly extrinsic moti
dence model where all covariances between structural vation of being a paid Apache developer to partic
variables are set to 0 and are thus unrelated. The null ipation. This hypothesis is supported, as the path
model is the base model for the computation of rel from EXTRINSICf2 to PARTICIPATION^ is positive
ative fit indices such as the CFI or the non-normed and significant (?\ = 0.152, p = 0.02). H4A and H4B
fit index (NNFI) as well as the theoretical model's x2 relate the use-value and status motivations to par
We computed x2 difference statistics (A^2) between ticipation. These hypotheses are partially supported,
the theoretical and alternative models under consid
as the path from STATUS,2 to PARTICIPATION^
eration as well as a comparison of model fit indices is significant and in the hypothesized direction of
(Bentler and Bonett 1980). H4B ()34 = 0.256, p < 0.01). However, the path from
Our hypothesized model fits the data very well. USE VALUE,2 to PARTICIPATION^, while signifi
The theoretical model's overall x2 statistic is signifi cant, is in the opposite direction than predicted by
cant. The x1 l?i ratio is well within acceptable range H4A (j33 = -0.208, p < 0.01). H5 relates participants'
with a ^2/df = 1.28. Model fit is judged acceptable intrinsic motivations to their level of participation.
using the SRMR (0.06) in combination with the CFI This hypothesis is not supported as the path from
(0.98). Taken together, these statistics indicate that our INTRINSIC^ to PARTICIPATION^ is not significant.
hypothesized model provides a highly acceptable fit Finally, H6 relates participation to changes in perfor
to the data. Utilizing the x2 difference test, we com mance ranking. The path from PARTICIPATION,^ to
pared the performance of the hypothesized model PERFORMANCE^ is positive and significant (ax =
with that of the alternative models. The hypothesized 0.178, p = 0.02) supporting H6.
model performance is clearly superior to that of the To test H7, H8A, and H8B, we enhanced our pri
independence model on measures of fit. Compared mary model by adding measures of participants' past
to the saturated model, the hypothesized model per performance and paths to relate them to the partic
forms quite favorably as well (A^2 = 1.22, Adf = 12, ipants' intrinsic motivations and internalized extrin
p < 0.001). In this case, the hypothesized model can be sic motivations (use value and status), respectively
7 The means, standard deviations, correlations, and covariances for 8 The results for cumulative number of patches and cumulative
the variables in our model and a summary of model fit evaluation number of lines of code are consistent with the reported results and
results are reported in the online appendix. are available on request from the authors.
Figure 2 Model Results: Standardized Path Coefficients
[0.055] i
Notes. Coefficients are shown for both primary and enhanced models. Coefficients for the enhanced model appear in brackets.
***p < 0.01 ,**/? = 0.05, *p = 0.10.
We then re-estimated the enhanced model including that contributors have multiple motivations to par
the original paths and variables and the newly added ticipate in OSS projects, and that some motivations
paths and variables. The enhanced model, like the are complementary, whereas others are not. Although
primary model, exhibits an excellent fit to the data. the previous OSS literature suggests that participants
The model's overall x2 statistic is significant with have multiple motivations for contributing, these
a ^2/df ratio of 1.2, well within acceptable range. motivations are thought to be "mutually reinforcing"
Model fit is judged acceptable using the SRMR (0.06) (e.g., Markus et al. 2000). Our findings suggest that
in combination with the CFI (0.98). Again, all A^2 this is not always the case. Specifically, and contrary
were computed between the hypothesized and alter to our expectations, we find no evidence of extrin
native models. In all cases, the Ax2 indicated in favor sic motivations crowding out strictly intrinsic motiva
of the hypothesized model; therefore, the hypothe tions.
sized model was retained as the best fitting model However, our results do suggest associations
and serves as the basis for the examination of our
among some motivations. Status motivations actually
remaining hypotheses. enhance intrinsic motivations. In addition, being paid
H7 positively relates past performance to intrinsic to contribute positively relates to participants' status
motivations. We find no support for H7, as the path motivations. A potential explanation for these com
from PERFORMANCE^ to INTRINSIC^ is not sig plementarities is offered by Sansone and Smith (2000),
nificant. H8A and H8B positively relate past perfor who suggest that extrinsic motivations can boost, reg
mance to the use-value and status motivations. These
ulate, and maintain interest in doing a task. For exam
hypotheses are partially supported. Although the ple, contributors' desires to further their careers may
path from past PERFORMANCE^ to USE VALUE,2 is enhance their inherent interest in making code contri
not significant, the path from past PERFORMANCE^ butions because making contributions can also help
to STATUS,2 is positive and significant, as we had pre
them to achieve higher status or to obtain better career
dicted (fa = 0.147, p = 0.05).
opportunities. On the other hand, we find that being
paid to contribute is negatively associated with partic
5. Discussion ipants' use-value motivations. In hindsight, this may
This study has examined the interrelationships not be too surprising. Extrinsically motivated devel
between the motivations, participation, and perfor opers may be more likely to view their contributions
mance of OSS developers. Our first finding reveals as part of their employment relationship and may
therefore have less personal use value for the Apache they may lose interest in making future contributions
software. An alternative explanation is that a contrib if there are no further salient problems or issues to
utor who receives an extrinsic incentive for his or her motivate them. This is consistent with the argument
use-value-motivated contribution will likely require put forth by Sansone and Smith (2000) who assert
such incentives in the future for submitting contri that, without motivations that help to sustain inter
butions that increase use value. In this sense, we do est, individuals could lose interest over time, even in
observe a crowding-out effect, not of an intrinsic moti activities they have previously found motivating. This
vation, but of an internalized extrinsic motivation. finding has important implications for those inter
Our second finding suggests that not all motiva ested in sustaining the participation of contributors
tions affect OSS participation equally or in the same to OSS projects: contributors who are primarily moti
way. Some studies identify participants' most "domi vated by use-value considerations may need subse
nant" motivations for engaging in OSS projects (e.g., quent incentives to sustain their level of participation.
Ghosh et al. 2002). However, these studies do not Our third finding suggests that the Apache meri
identify how differences in actual participation levels tocracy is operating effectively, and that promotions
are associated with different motivations for partici within the community are indeed based on actual
pating. As we predicted, being paid to contribute and contributions to the Apache projects. This is especially
status motivations are related to above-average partic important for a community like Apache that depends
ipation levels. However, contrary to our predictions, on shared leadership and the contributions of partic
we find no significant relationship between intrinsic ipants for its success.
motivations and participation levels. At first glance, Finally, we have found that past-performance rank
this finding is puzzling. Studies of OSS contributors ings enhance some motivations for participating but
have revealed that contributors do enjoy participat not others. We expected that an increase in rank
ing: developers have reported flow states and losing ing would boost contributors' subsequent intrinsic
track of time when working on OSS projects (Lakhani and extrinsic motivations. Our findings support this
and Wolf 2005). Why do these feelings of enjoy expectation for status motivations, as an increase in
ment in programming not translate into higher lev performance ranking is associated with a subsequent
els of code contribution? The literature on motivation
increase in contributors' status motivations to par
and performance offers some potential explanations. ticipate. This finding is consistent with the notion
Researchers have identified some disadvantages asso of incentive motivation where receiving a reward
ciated with intrinsic motivation (Osterloh and Frey increases motivation for an individual to perform
2000). Intrinsically motivated contributors tend to be previously rewarded activities. As Lerner and Tir?le
more autonomous and self-directed, which may lead (2002) have noted, a promotion in rank enhances
them to exhibit less desirable behaviors. In addition, a contributor's status in the OSS community and
Lepper and Henderlong (2000) suggest that intrinsic increases incentive motivation. Thus, an increase in
motivation may not be associated with better per status due to an advance in rank should amplify
formance if the aspects of the activity that make it the contributor's subsequent status motivations to
interesting come at the expense of attention toward participate. However, we find no significant associa
some outcome?in this case, being intrinsically moti tions between rank increases and intrinsic motivations
vated may not positively affect participation levels. or use-value motivations. Our results suggest that a
Because OSS contributors are self-directed, it is possi rank increase within the Apache community is neither
ble that their intrinsic motivation is not fully aligned considered controlling and thus diminishing intrin
with the mission of the OSS community. For exam sic motivation, nor is it viewed as informative on
ple, a potential contributor may be intrinsically moti competence and thus enhancing intrinsic motivation.
vated to work on perfecting one feature of the code This could be the result of the performance evaluation
that performs a relatively minor function, and may itself because rank increases are typically not accom
spend inordinate amounts of time perfecting the sin panied by detailed reports and feedback on perfor
gle feature rather than implementing a large number mance, but rather just announced. Therefore, it may
of contributions. Also contrary to our predictions, use not be clear to the contributor exactly what aspects of
value motivations are associated with below-average his or her performance are superior.
contribution levels. This result could be explained
by reconsidering the nature of use-value motivations.
Developers who contribute because they are moti 6. Conclusions
vated by use value want to solve a particular bug Our study makes several important contributions
that is causing them trouble or to add a particular fea to the OSS literature. First, our theoretical model
ture that they need to use. Once they have solved the and empirical evaluation increase the understanding
immediate problem or added the particular feature, of how motivations, participation, and performance
interrelate in OSS projects. Although some studies survey data on motivations with objective measures
have examined particular aspects of motivation or of participation and performance. This approach
participation, we are not aware of any study that helps to triangulate findings and mitigate common
has examined the system of interrelationships between method and source biases. In addition, we leverage
motivations, participation, and performance. Yet, one archival data to capture participation and perfor
must consider the motivational system in order to mance measures over a four-year period. This lon
understand how successfully an OSS community is gitudinal approach allows us to tease out potential
functioning in terms of attracting and sustaining par causality relationships between motivations, partici
ticipation. To our knowledge, this is also the first pation, and performance; and between past perfor
study that investigates how past performance affects mance and motivations. Our study focuses only on
subsequent motivations in OSS communities. Under projects within the Apache OSS community. Although
standing this relationship provides insight into the this potentially limits our findings in a strict sense
motivational effectiveness of feedback in the OSS to the Apache OSS community, we believe that our
setting. results could be applicable to other successful OSS
As a whole, our results have several implications communities such as Linux, Perl, and Mozilla that
for attracting and sustaining participation in OSS share an interesting relationship between the intrin
communities. First, our results suggest that OSS com sic and extrinsic motivations of contributors. Our
munities should largely welcome commercial efforts approach does offer some persuasive advantages.
by companies. While extrinsically motivated contrib Focusing on one OSS community enables us to link
utors have lower use-value motivations, they also developers to their actual code contributions, and to
exhibit greater status motivations, and being paid to link these code contributions to objective performance
contribute is associated with a higher level of con evaluations by others in the Apache community This
tributions to the source code. More importantly, in approach enables us to cleanly capture the relation
our setting, we could not detect any crowding out ships between past performance, motivation, partici
of (strictly) intrinsic motivations by extrinsic motiva pation, and subsequent performance. In addition, elic
tions. This is perhaps more important, as many suc iting broad participation in our study within one com
cessful OSS projects (e.g., Apache, Linux, SendMail, munity rather than limited participation across many
Mozilla) experience increased attention from leading communities helped us to obtain a representative
software producers who pay employees to contribute sample. That is why, our research design increases
to these projects. Second, developers with higher the internal validity, external validity, and statistical
status motivations appear to be the more substan conclusion validity of our results (Campbell and Cook
tive contributors. Therefore OSS communities may 1979).
want to nurture such motivations, perhaps by devot Our study opens up several important avenues for
ing distinct website space to recognize distinguished further research in the OSS arena. We were guided
developers or by promoting involvement in OSS by the literature on psychology and open source soft
communities as leverage in the labor market. Third, ware development to identify the most likely fac
the positive relationship between past-performance tors leading to OSS participation. However, it is
and status motivations indicates that a feedback sys possible that other factors are salient. For example,
tem provides a valuable service to the OSS com some researchers have suggested a motivation aris
munity by increasing these motivations. However, ing from a sense of obligation to the community
the current feedback system, which primarily is an (Lindenberg 2001) for contributing to OSS projects.
announcement of a rank increase, fails to enhance This motivation has certain similarities with the use
intrinsic motivations whereas feedback that indicates value motivation we have examined as use value
personal competence could increase intrinsic motiva is internalized based on identification with the OSS
tion (Pittman et al. 1980, Ryan et al. 1983). That is why community. Our results for use value suggest that
our results imply that OSS communities should cap this intrinsic motivation based on obligation to the
ture a competence component in the feedback system community may be associated with below-average
that stresses the competence of the contributor and contribution levels. However, the precise relation
that provides detailed information and feedback on ships between obligation motivations, participation
performance. This could, for example, include a list levels, and performance would need to be exam
ing of achievements and extraordinary contributions. ined in future research. Investigating the interrelation
Our study has several strengths and limitations. We ships among motivations, participation, and perfor
have drawn on the extensive literature on motivation mance in other OSS communities is another impor
in psychology and on the OSS literature to develop tant research extension. In addition, most of the lit
theoretically driven hypotheses. Our research design erature on the psychology of intrinsic and extrinsic
leverages multiple data sources, matching subjective motivation is validated via carefully designed and
controlled experiments. An experimental approach Deci, E. L., R. M. Ryan. 2000. The "what" and "why" of goal pur
suits: Human needs and the self-determination of behavior.
could be especially useful in revealing the processes
Psych. Inquiry 11(4) 227-268.
by which extrinsic motivations for contributing to Deci, E. L., R. Koestner, R. M. Ryan. 1999. A meta-analytic review
OSS projects are internalized and the mechanisms by of experiments examining the effects of extrinsic rewards on
which different internalization processes lead to dif intrinsic motivation. Psych. Bull. 125(6) 627-668.
ferences in participation. Experiments could also be Fielding, R. 1999. Shared leadership in the apache project. Comm.
ACM 42(4) 42-43.
helpful for designing feedback mechanisms to maxi Ghosh, R. A. 1998. Interview with Linus Torvalds: What motivates
mize the effect of past performance on future moti free software developers? First Monday 3(3).
vations. Further research on the motivational mech Ghosh, R. A., R. Glott, B. Krieger, G. Robles. 2002. Free/libre and
anisms underlying participation and performance is open source software: Survey and study. International Insti
tute of Infonomics, University of Maastricht, The Netherlands.
vital for effectively leveraging the advantages of Retrieved March 30, 2004, http://www.infonomics.nl/FLOSS/
"costless" OSS development. report/index.htm.
An online supplement to this paper is available on Harm, I., J. Roberts, S. A. Slaughter, R. Fielding. 2006. An empiri
the Management Science website (http://mansci.pubs. cal analysis of economic returns to open source participation.
Working Paper 2006-E5, Tepper School of Business, Carnegie
informs.org/ecompanion.html). Mellon University, Pittsburgh, PA, and Marshal School of Busi
ness, University of Southern California, Los Angeles, CA.
Acknowledgments Hennessey, B. A. 2000. Rewards and creativity. C. Jansone, J.
Funded in part by a grant from the Berkman Faculty Devel Harackiewicz, eds. Intrinsic and Extrinsic Motivation: The Search
opment Fund at Carnegie Mellon University. The authors for Optimal Motivation and Performance. Academic Press, San
Diego, CA, 55-78.
thank Roy Fielding for his support of the research and the Hertel, G., S. Niedner, S. Herrmann. 2003. Motivation of software
Apache developers for their participation in the study. They developers in open source projects: An Internet-based survey
also acknowledge the valuable comments of the associate of contributors to the linux kernel. Res. Policy 32(7) 1159-1177.
editor, two anonymous reviewers, and participants in semi Hu, L., P. M. Bentler. 1997. Selecting cutoff criteria for fit indexes
nars at Perm State University and Carnegie Mellon Univer for model evaluation: Conventional criteria versus new alter
sity on prior versions of the manuscript. natives. Structural Equation Model. 6 1-55.
Hull, C. L. 1943. Principles of Behaviour. Appleton-Centrury-Crofts,
New York.
Johns, G. 1996. Organizational Behavior: Understanding and Managing
References
Life at Work, 4th ed. HarperCollins, New York.
Amabile, T. M., B. Hennessey, B. Grossman. 1986. Social influence Kanfer, R. 1990. Motivation theory and industrial and organiza
on creativity: The effects of contracted-for reward. /. Personality tional psychology. M. Dunnette, L. Hough, eds. Handbook of
Soc. Psych. 50(1) 14-23. Industrial and Organizational Psychology, 2nd ed. Consulting
Anderson, J. C, D. W. Gerbing. 1988. Structural equation modeling Psychology Press, Palo Alto, CA, 75-170.
in practice: A review and recommended two-step approach. Kirsch, L. J. 1996. The management of complex tasks in organiza
Psych. Bull. 103(3) 411-423. tions: Controlling the systems development process. Organ. Sei.
7(1) 1-21.
Ang, S., S. A. Slaughter, K. Y. Ng. 2002. Human capital and institu
tional determinants of information technology compensation: Lakhani, K., B. Wolf. 2005. Why hackers do what they do: Under
Modeling multilevel and cross-level interactions. Management standing motivation and effort in free/open source software
Sei. 48(11) 1427-1445. projects. J. Feller, B. Fitzgerald, S. Hissam, K. Lakhani, eds.
Perspectives on Free and Open Source Software. MIT Press, Boston,
Becker, G. S. 1962. Investment in human capital: A theoretical anal MA.
ysis. /. Political Econom. 70(5) 9-^19.
Lepper, M. R., J. Henderlong. 2000. Turning "play" into "work" and
Bentler, P. M., D. G. Bonett. 1980. Significance tests and goodness "work" into "play": 25 years of research on intrinsic versus
of fit in the analysis of covariance structures. Psych. Bull. 88 extrinsic motivation. C. Jansone, J. Harackiewicz, eds. Intrinsic
588-606. and Extrinsic Motivation: The Search for Optimal Motivation and
Boehm, B., C. Abts, A. W. Brown, S. Chulani, B. K. Clark, Performance. Academic Press, San Diego, CA, 257-307.
E. Horowitz, R. Madachy, D. Reifer, B. Steece. 2000. Software Lepper, M. R., D. Greene, R. E. Nisbett. 1973. Undermining chil
Cost Estimation with COCOMO II. Prentice Hall, Upper Saddle dren's intrinsic interest with extrinsic reward: Test of overjus
River, NJ. tification hypothesis. /. Personality Soc. Psych. 28(1) 129-137.
Calder, B., B. Staw. 1975. The self-perception of intrinsic and extrin Lerner, J., J. Tir?le. 2002. Some simple economics of open source.
sic motivation. /. Personality Soc. Psych. 31(4) 599-605. /. Indust. Econom. 50(2) 197.
Campbell, D., T. D. Cook. 1979. Quasi-Experimentation: Design & Lindenberg, S. 2001. Intrinsic motivation in a new light. Kyklos
Analysis Issues for Field Settings. Houghton Mifflin, Boston, MA.
54(2/3) 317-342.
Markus, L., B. Manvile, C. Agres. 2000. What makes a virtual orga
Campbell, J. P., R. D. Pritchard. 1976. Motivation theory in indus
nization work? MIT Sloan Management Rev. 42(1) 13-26.
trial and organizational psychology. M. D. Dunnette, ed. Hand
book of Industrial and Organizational Psychology. Rand McNally, Marsh, H. W. 1994. Confirmatory factor analysis models of factorial
invariance. A multifaceted approach. Structural Equation Model.
Chicago, IL, 63-130.
1(1) 5-34.
Deci, E. L. 1971. Effects of externally mediated rewards on intrinsic
Meyer, H. 1987. Predicting supervisory ratings versus promo
motivation. /. Personality Soc. Psych. 18(1) 105-155.
tional progress in test validation studies. /. Applied Psych. 72(4)
Deci, E. L. 1975. Intrinsic Motivation. Plenum Press, New York. 696-697.
Deci, E. L., R. M. Ryan. 1987. The support of autonomy and the con Mitchell, T. R., D. Daniels. 2003. Motivation. Handbook of Psychol
trol of behavior journal of personality and social psychology. ogy. Industrial and Organizational Psychology, Vol. 12. Wiley,
/. Personality Soc. Psych. 53(6) 1024-1037. New York, 225-254.
Moon, J. Y, L. Sproull. 2002. Essence of distributed work: The case Sansone, C, J. L. Smith. 2000. Interest and self-regulation: The rela
of the linux kernel. P. Hinds, S. Kiesler, eds. Distributed Work. tion between having to and wanting to. C. Sansone, J. Harack
MIT Press, Cambridge, MA, 381-404. iewicz, eds. Intrinsic and Extrinsic Motivation: The Search for
Osterloh, M., B. Frey. 2000. Motivation, knowledge transfer, and Optimal Motivation and Performance. Academic Press, San Diego,
organizational forms. Organ. Sei. 11(5) 538-550. CA, 341-372.
Pittman, T. S., M. E. Davey, K. A. Alafat, K. V. Wetherill, N. A. Shah, S. 2004. Understanding the nature of participation and coor
Kramer. 1980. Informational versus controlling verbal rewards. dination in open and gated source software development com
Personality Soc. Psych. Bull. 6(2) 228-233. munities. K. M. Weaver, ed. Academy of Management Proceedings.
Academy of Management, New Orleans, LA, B1-B6.
Raymond, E. 1999. The Cathedral and the Bazaar: Musings on Linux
and Open Source by an Accidental Revolutionary, revised ed. Shah, J., A. Kruglanski. 2000. The structure and substance of intrin
O'Reilly & Associates, Boston, MA. sic motivation. C Sansone, J. Harackiewicz, eds. Intrinsic and
Extrinsic Motivation: The Search for Optimal Motivation and Per
Rossi, M. A. 2004. Decoding the free/open source puzzle. A sur
formance. Academic Press, San Diego, CA, 106-123.
vey of theoretical and empirical contributions. Working paper,
Dipartimento Di Economia Politica, Universit? degli Studi di Sharma, S., S. Durvasula, W. R. Dillon. 1989. Some results on the
Siena, Siena, Italy. behavior of alternate covariance structure estimation proce
dures in the presence of non-normal data. /. Marketing Res. 26
Rummel, A., R. Feinberg. 1988. Cognitive evaluation theory: A 214-221.
meta-analytic review of the literature. Soc. Behav. Personality
16(2) 147-164. Skinner, B. F. 1953. Science and Human Behavior. Macmillan,
New York.
Ryan, R. M. 1982. Control and information in the intrapersonal
sphere: An extension of cognitive evaluation theory. /. Person Spence, M. 1976. Competition in salaries, credentials, and signaling
ality Soc. Psych. 43(3) 450-461. prerequisites for jobs. Quart. }. Econom. 90(1) 51-74.
Ryan, R. M., J. P. Connell. 1989. Perceived locus of causality and von Hippel, E. 2001. Innovation by user communities: Learn
internalization: Examining reasons for acting in 2 domains. ing from open-source software. Sloan Management Rev. 42(4)
82-86.
/. Personality Soc. Psych. 57(5) 749-761.
Ryan, R. M., E. L. Deci. 2002. An overview of self-determination von Hippel, E., G. von Krogh. 2003. Open source software and the
theory: An organismic-dialectical perspective. E. L. Deci, R. M. "private-collective" innovation model: Issues for organization
Ryan, eds. Handbook of Self-Determination Research. The Univer science. Organ. Sei. 14(2) 209-223.
sity of Rochester Press, Rochester, NY, 3-33. von Krogh, G., S. Spaeth, K. Lakhani. 2003. Community, joining,
Ryan, R. M., V. Mims, R. Koestner. 1983. Relation of reward con and specialization in open source software innovation: A case
tingency and interpersonal context to intrinsic motivation: study. Res. Policy 32(7) 1217-1241.
A review and test using cognitive evaluation theory. /. Person Weinberg, G. 1998. The Psychology of Computer Programming. Silver
ality Soc. Psych. 45(4) 736-750. Anniversary Edition, 1st ed. Van Nostrand Reinhold, New York.
Sansone, C. 1986. A question of competence: The effects of compe Wiersma, U. J. 1992. The effects of extrinsic rewards in intrinsic
tence on task feedback on intrinsic interest. /. Personality Soc. motivation: A meta-analysis. /. Occupational Organ. Psych. 65
Psych. 51(5) 918-931. 101-114.

Informs Management Science: This Content Downloaded From 193.226.62.221 On Sat, 24 Jun 2017 10:09:14 UTC

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Informs Management Science: This Content Downloaded From 193.226.62.221 On Sat, 24 Jun 2017 10:09:14 UTC

Caricato da

Copyright:

Formati disponibili

Understanding the Motivations, Participation, and Performance of Open Source Software

Developers: A Longitudinal Study of the Apache Projects

Understanding the Motivations, Participation, and

1. Introduction spectrum including scratching a "personal itch" (Ray

Figure 1 Theoretical Model

2. Theoretical Framework exemplified by the level of their contributions to the

3.1. Research Setting of individuals within each project. To extract infor

Figure 2 Model Results: Standardized Path Coefficients

Potrebbero piacerti anche