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

net/publication/3412475

Open Source Versus Closed Source: Software Quality in Monopoly and

Competitive Markets

Article in IEEE Transactions on Systems Man and Cybernetics - Part A Systems and Humans · December 2005
DOI: 10.1109/TSMCA.2005.853493 · Source: IEEE Xplore

CITATIONS READS

34 414

4 authors, including:

Birendra K. Mishra Hsihui Chang

University of California, Riverside Drexel University
21 PUBLICATIONS 929 CITATIONS 72 PUBLICATIONS 2,291 CITATIONS

SEE PROFILE SEE PROFILE

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

Academic Research View project

All content following this page was uploaded by Hsihui Chang on 03 May 2016.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005 903

Open Source Versus Closed Source: Software Quality

in Monopoly and Competitive Markets
Srinivasan Raghunathan, Ashutosh Prasad, Birendra K. Mishra, and Hsihui Chang

Abstract—The open source model of software development has Championed by the philosophy that software is a public good
received substantial attention in the industry and popular media; and should be freely shared, used, and codeveloped by all, the
nevertheless, critics frequently contend that open source softwares open source movement challenges the traditional paradigm of
are inferior in quality compared to closed source softwares be-
cause of lack of incentives and project management, while propo- software as a proprietary good, to be guarded carefully via
nents argue the opposite. This paper examines this quality debate copyrights or patents and licensed or sold to users for profit.
by modeling and analyzing software quality, demand, profitabil- The traditional paradigm (referred to as “closed source”) is
ity, and welfare under open and closed source environments in based on the assumption that software development is a highly
monopoly and competitive markets. The results show no dominant specialized process, managed best by a localized team of highly
quality advantage of one method over another under all circum-
stances. Both open source and closed source qualities decrease in qualified developers, careful project management, and occa-
a competitive market. Conditions under which each method can sional enhancements in the form of new releases. In contrast,
generate higher quality software are examined. open source software is based on the principles of continuous
Index Terms—Competition, open source, software quality. improvement (implemented via frequent releases), collabora-
tion among developers and users irrespective of geographical
locations or employing firms, and adherence to open standards
I. I NTRODUCTION (implemented via open source licenses). As described by [3],
open source represents a “bazaar style” of software develop-
R ECENT years have seen an increasing interest in the
open source movement as a new paradigm for software
development. Open source refers to the use of shared source
ment, in contrast to the “cathedral style” emphasized by closed
source software development.
code, open standards, and collaboration among software de- One of the frequent criticisms of open source softwares is
velopers and users worldwide to build software, identify and that they are of lower quality compared to their closed source
correct errors, and make enhancements [1]. Unlike the tradi- counterparts. First, critics contend that the quality of open
tional (proprietary) paradigm of software development, users source software suffers from the “free rider” effect often as-
have free access to the source code, which they can modify sociated with public goods, where each user prefers others
to correct software bugs, port the software to new hardware to spend effort in improving the public good and then share
or software platforms, solve additional problems, create add-on in the benefits rather than investing in improvements them-
software programs, or simply use it for free [2]. Enhancements selves. Thus, improvement depends on the altruism of a few.
submitted by individual users are fed back to the original source Furthermore, attempts at commercial exploitation of freely
code for public use. This approach has led to the development donated software can result in an amelioration of altruis-
of the Linux operating system, Apache web server, Perl pro- tic tendencies [4]. However, proponents claim that software
gramming language, Sendmail electronic mail transfer agent, contributors are not necessarily altruistic; programmers con-
domain name system Berkeley internet name domain (DNS tribute to the source code for social recognition and pres-
BIND) system, and other Internet infrastructure software that tige in the open source community and to signal their talent
power some of today’s most powerful electronic commerce to prospective employers, which may subsequently translate
websites including Yahoo, Cisco Systems, and C-Net. Table I into job, consulting, or other career opportunities [5]–[7]. A
provides short descriptions of some commonly used open discussion of possible alternative motivations of open source
source software. programmers is provided by [8]. The focus here will be
on pecuniary, or economic, motives as opposed to altruistic
motives [7].
Manuscript received November 5, 2003; revised May 28, 2004 and Septem- Second, critics suggest that the lack of formal project man-
ber 20, 2004. This paper was recommended by Associate Editor H. R. Rao. agement in “bazaar style” software development undermines
S. Raghunathan and A. Prasad are with the School of Management, Univer-
sity of Texas at Dallas, Richardson, TX 75083 USA (e-mail: sraghu@utdallas. the product’s quality. However, open source proponents argue
edu; [email protected]). that the quality of open source software stems not from its
B. K. Mishra is with the Anderson Graduate School of Management, management but from its openness. Free access to the source
University of California at Riverside, Riverside, CA 92521 USA, on leave
from the School of Management, University of Texas at Dallas, Richardson, code ensures that the code is tested and retested by a worldwide
TX 75083 USA (e-mail: [email protected]). user base, leading to timely identification of bugs and opportu-
H. Chang is with the Anderson Graduate School of Management, Univer- nities for software enhancements, and is hence more reliable.
sity of California at Riverside, Riverside, CA 92521 USA (e-mail: hsihui.
[email protected]). Limited coordination in an open source environment may lead
Digital Object Identifier 10.1109/TSMCA.2005.853493 to some duplication of effort between programmers trying to

1083-4427/$20.00 © 2005 IEEE

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
904 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

TABLE I The proponents of both paradigms point to the inherent in-

EXAMPLES OF OPEN SOURCE SOFTWARE
centives built into respective approaches to build quality soft-
ware, with respect to software pricing and the approach to
software development, e.g., cathedral style versus bazaar style
[9]. A related issue of interest is whether the society as a
whole is better off under the closed source or the open source
paradigm. The free nature of open source software may benefit
consumers, but if the quality is low, then the value consumers
derive from the software will also be low. From a society
perspective, the profit realized by the closed source software
developed is also important.
The objectives of this research are threefold: 1) to analyze
the impact of inherent incentive structures of the closed and
open source approaches on software quality; 2) to determine
whether the open source or closed source approach always
results in a superior software quality and, if not, to identify
the conditions under which each approach is better; and 3) to
compare the societal welfare, defined as the sum of developer
profit and consumer surplus, in the open source and closed
source paradigms. The game theory was used to perform the
analysis. Software quality is analyzed in noncompetitive open
source and closed source markets, and then in a competitive
market where both open source and closed source softwares
compete for market share.
The findings indicate that despite the “free” nature of open
source software, built-in incentives for software developers
in the open source model can enhance software quality to
be comparable to or exceed the closed source quality. The
open source model becomes attractive when 1) a large pool of
programmers is willing to work on the open source software;
2) the programmers’ efforts are publicly recognized; and 3)
open source programmers perceive code contribution to be a fun
task or a hobby. If the open source and closed source softwares
have the same quality, then the social welfare is higher under
the open source than the closed source approach. If a closed
source software and an open source software compete in the
same market, then the open source quality is independent of
closed source software quality but is lower than the quality
when there is no competition. However, under competition,
the closed source quality decreases as the open source quality
increases. The social welfare is higher under competition than
in a monopoly.
For purposes of model tractability, several factors governing
the closed source and open source approaches are ignored in
this paper. Demand for software may depend on factors such
as functionality, standards, support, and legal issues. In the
proposed model, demand is a function of only quality and price.
While factors such as standards and licensing arrangements are
correct the same error, and hence, it is unclear whether having also important considerations in a firm’s decision to adopt the
multiple users spend resources on improving a common soft- software, these factors are covered to the extent they impact
ware necessarily improves its quality over having a closed firm utility for the software, in which case they fall under the rubric
with coordinated efforts and formal project management. While of “quality.”
process maturity concepts such as the capability maturity model The remainder of the paper proceeds as follows. Section II
(CMM) are not directly applicable to open source projects, reviews quality debate surrounding software development,
elements of these approaches such as processes for code sub- specifically focusing on open source software. Section III dis-
mission, peer review, and user involvement are strictly fol- cusses the modeling framework and assumptions. Section IV
lowed, particularly in larger projects, which also maintain good examines how quality improvement comes about in a monopoly
documentation. market with only one open source software. Section V

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 905

investigates quality improvement when closed source software provide clients with access to the source code (e.g., SAP R/3)
is in monopoly and compares the result to those in the preceding proclaim that any modification to the source code will result
section. Section VI models market competition between open in loss of customer support, thus limiting the product’s custom-
source and closed source softwares. Section VII concludes with izability to a predefined set of options. Furthermore, continuous
discussion and directions for future research. innovation and upgrades require soliciting suggestions and
feedback from software users. While open source programmers
actively solicit inputs from users worldwide via listservs, usenet
II. T HE Q UALITY D EBATE IN O PEN S OURCE S OFTWARE
groups, and bulletin boards, closed source software developers
Software quality has been extensively studied in software have limited interaction with a small group of users. In practice,
engineering. In this literature, the quality of a software is upgrades to open source software are released much more fre-
measured along dimensions such as correctness, understand- quently than that of closed source software, and consequently,
ability, completeness, conciseness, portability, consistency, open source software such as Linux seems to have devel-
maintainability, testability, usability, reliability, structuredness, oped faster than comparable closed source software such as
and efficiency [10], [11]. Despite advances in software process Windows NT. As an example, Linux supported 64-bit process-
innovations such as object-oriented analysis and design, func- ing since 1995 while NT became 64-bit ready in 2000, Linux
tion point analysis, and CMMs, the software industry has supports the latest networking standards such as Ipv6 but NT
continually been plagued by an inability to generate quality does not, and Linux provides ports to more software devel-
software products [12], [13]. It is noted in [13] that, “In the opment environments than does NT. However, open source
last 15 years alone, software defects have wrecked a satellite projects do seem to lag their closed source counterparts in
launch, delayed an airport opening for a year, destroyed a coordination of developers and project management, leading to
Mars mission, killed four Marines in a helicopter crash, in- some duplication of efforts by multiple developers, inefficient
duced a U.S. Navy ship to destroy a civilian airliner, and shut allocation of time and resources, and lack of attention to mun-
down ambulance systems in London leading to as many as dane software attributes such as ease of use, documentation, and
30 deaths.” support, all of which impact conformance to specifications.
The complex software systems that have become the central Indeed, industry comparisons of open source versus closed
nervous system of firms make the high quality of a software source softwares are inconclusive or slightly in favor of open
vital. They not only control a firm’s everyday operations but source. For instance, a comparative evaluation of Linux and
also enable or constrain the ability to make and implement Windows NT (respectively, open source and closed source
strategic business decisions [14]. Though software quality in- server operating systems) by [16] found that Linux is signifi-
vestments can reduce overall software life cycle costs by mini- cantly superior to NT in three out of nine dimensions (availabil-
mizing rework later on, many software manufacturers sacrifice ity, user satisfaction, and value for money), somewhat superior
quality in favor of other objectives such as shorter development in three other dimensions (operational features, support, and
cycles and meeting time constraints [15]. As one software scalability), comparable in two dimensions (interoperability
manager said, “I’d rather have it wrong than have it late. We and functionality), and inferior in one dimension (application
can always fix it later” [15]. The traditional conception of availability). Some system administrators prefer Linux over
software quality is centered on a product-centric conformance Windows NT because of Linux’s less hardware configuration
view of quality [14]. The focus of this perspective is to re- requirement, lightweight nature, better multiprocessing capa-
duce variability (by minimizing errors in design and coding) bilities, and networking support [17]; however, others prefer
and to conform to predefined specifications (e.g., in terms of Windows NT because of better security, ease of management,
error rates, speed, transaction load, etc.). Benchmarking soft- and superiority in Java performance.
ware based on predefined specifications assumes that it is The issue of cost comparison between an open source and
possible to specify ex ante the entire range of features expected a closed source has also been of interest to organizations, but
of a software. However, in reality, different users have differ- there are few academic studies available on precise and objec-
ent expectations of the same software and users’ expectations tive comparisons. It is still a new issue for academics. Among
of software evolve with time. For instance, some users may open source softwares, Linux is the one that has received the
view performance and reliability as key features of a soft- most attention. A few reports exist comparing it to Windows,
ware while others may consider ease of installation and main- but the results are inconclusive. For example, following is the
tenance as key features of the same software. The authors will abstract of an IDC report [18] sponsored by Microsoft on the
assume that howsoever quality is described, it may be measured total cost of ownership (TCO):
from the customers’ viewpoint as a utility providing attribute Linux is widely regarded as “free” because there is no or
such that a higher quality product is preferred by all users. little cost associated with software acquisition. However,
The authors can make a qualitative comparison of how after taking into account all costs, notably IT staffing,
quality might vary between open source and closed source soft- does Linux truly come at a lower cost than competing
wares. Open access to the source code allows greater opportu- platforms, such as Windows? IDC has completed a study
nities for customizing an open source software to the specific of five common workloads in enterprise computing that
needs of individual users than a closed source software. In fact, challenges the common industry perception that Linux is
such customization is encouraged and widely practiced by the “free.” Our in-depth study suggests that Microsoft Win-
open source community, while closed source softwares that dows 2000 offers lower total cost than a Linux solution

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
906 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

TABLE II
COST OF OWNERSHIP DATA (IDC 2002)

Windows 2000 and Linux Server Environment Five year total cost of ownership summary view for 100 supported users by workload and cost category
($) as reported by IDC 2002

in four of the five workloads common to most corporate TABLE III

COST OF OWNERSHIP DATA (ROBERT FRANCES GROUP)
IT environments. In these four workloads (network in-
frastructure, print serving, file serving, and security appli-
cations), the cost advantages of Windows are significant:
11–22% less over a 5-year period. The cost advantages
are driven primarily by Windows’ significantly lower costs
for IT staffing, generally the largest single component of
IT costs. For the fifth workload, Web serving, Linux had
a cost advantage of 6% compared with Windows 2000
over the 5-year period. IDC’s study confirms that low
Finally, as suggested by the open source examples listed
initial software acquisition costs are only one factor, not
in Table I, the open source model appears most viable for
the deciding one, in determining the 5-year total cost of
generic context-independent applications such as operating
ownership (TCO) for the two operating environments.
systems, network software, word processors, and spreadsheets,
The data reported by IDC are given in Table II.
where no specialized user inputs are required. The domain of
Contrary conclusions are reported by [19], as shown in
study for open source is likewise restricted to general purpose
Table III. This also takes into account software, hardware,
software used by a large number of end users who are price and
support, and maintenance costs, and found that Linux was the
quality takers (as opposed to a specific use software contract
least expensive platform to deploy and operate. Although some
requiring exact price and quality specifications). The model is
initial costs were higher at points, the ability to massively scale
thus not applicable for highly specialized business applications
the product horizontally without paying additional licensing
such as customized procurement systems or payroll systems,
fees can yield significant cost savings over the long term.
where programmers cannot code without knowing the users’
The lack of jury on the open source versus closed source
requirements, and the quality of the software product is
debate is reflected in the fact that closed source firms such as
determined by its meeting contractual specifications. Further,
IBM and Sun that profess value in the open source philosophy
open source programmers are less likely to contribute to highly
and routinely commit time, resources, and manpower to open
specialized applications because the small user base for such
source projects are not yet ready to abandon their internal
software limits the amount of rewards programmers can derive
closed source projects.
by contributing code.
When an individual programmer submits a proposed im-
The authors summarize this section with Table IV listing the
provement to an open source software, a coordinating commit-
key features and relevant differences between open source and
tee consisting of experienced programmers evaluates the merits
closed source softwares. These features provide the conceptual
of that proposal, examines possible side effects of the suggested
underpinnings for modeling open source and closed source soft-
improvement, compares it with alternative ideas submitted by
ware development within an economic framework. Also note
other programmers, and finally decides whether to incorporate
that open source software development resembles a tournament
the proposed improvements to the software, and if so, the
[20], [21] whereas closed source development resembles the
enhanced version is distributed as a new software release. As
standard production model of the firm. Using this intuition, the
long as developers return some of their improvements to this
authors proceed to modeling software quality developed under
committee, the software will continue to improve. Submission,
the two methods.
evaluation, and acceptance of improvements can be viewed as
a tournament, where one improvement is accepted from a set of
III. M ODELING F RAMEWORK AND A SSUMPTIONS
submissions, and the programmer submitting that improvement
gains visibility, respect, and future prospects within and outside The modeling framework is based on models of quality and
the open source community. price-based competition in a vertically differentiated market.

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 907

TABLE IV TABLE V
CONCEPTUAL UNDERPINNINGS OF THE MODELING FRAMEWORK LIST OF MAIN NOTATION

Quality and price-based competition have been analyzed in a

variety of settings such as service delivery [22]–[24], marketing
and supply chain channels [25], and economics. The primary
difference between this paper and prior work in this area is that
the authors model the impact of incentive structure on software
quality whereas prior research assumes that firms can set any
quality level they desire. This section presents a software devel-
opment model from both consumer and developer perspectives.
Table V provides a summary of the notation used in this paper.
On the consumer side, the authors assume that the consumer random variable with a deterministic component (function of
valuation for the software at quality Q takes the form programmer effort) and a random term. Thus
Qi (qi ) = qi + εi . (3)
V = tQθ . (1)
The cumulative distribution function (cdf) of Qi is denoted
The parameter θ ∈ (0, 1) captures the declining marginal FQi . Programmers’ costs C(qi ) are assumed to be increasing
value of quality. That is, a consumer’s value for a software is and convex, i.e., C  (qi ) > 0 and C  (qi ) > 0. The cost of effort
a concave function of its quality. The model assumes a scalar is typically considered to be convex by economists. Compared
measure of quality, which could be any one or an aggregate to spending the first hour on a project, spending an additional
of some or all of the software quality measures discussed hour after having spent say 8 h is more difficult due to phys-
in Section II. If the heterogeneity parameter t is uniformly ical and mental fatigue and the press of other duties. Also,
distributed between 0 and 1, the demand function is obtained initial improvements are easier to make because they are more
for the software at quality Q and price P , given a total of D obvious—like low-hanging fruit—compared to finding later
consumers improvements. Among convex functions, the quadratic form is


P selected for its tractability.
S(Q, P ) = D 1 − θ . (2) Programmers differ in their skill and experience. Presum-
Q
ably, all programmers can ultimately get a job done, but the
On the developer side, the software is developed by a set of more skilled and experienced ones will find it less costly in
programmers. Programmer i chooses a planned quality output terms of time and effort to do so. One can thus differentiate pro-
qi by applying the requisite effort. However, software im- grammer types by their cost of programming parameter. For
provements need not follow deterministically from the amount simplicity, the authors assume two types, high cost (H) or
of effort applied. Improvements may come randomly, in a low cost (L); however, this simplification is not critical to the
serendipitous manner, by “stroke of luck.” Additionally, some results here. The authors denote the number of programmers in
randomness is inherent in the subjective nature of the evaluation high- and low-cost types by NH = βN and NL = (1 − β)N ,
process. Therefore, the evaluated quality Qi is modeled as a respectively, where N is the total number of programmers

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
908 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

working on the software and β ∈ [0, 1] is the proportion of teers, assigns tasks and selects submissions for improvements,
high-cost programmers. and invests a lot of time in management. Larger projects, such
For the same planned output quality, a low-type programmer as Apache, have a formal core development team [26]. Fur-
incurs a lower cost compared to a high type. More specifically, thermore, there is some formalization from the fact that many
the programmer’s cost of quality is specified as Open Source projects are listed on http://SourceForge.net or
similar hosts, which provide tools such as concurrent versioning
2 2 system (CVS), mailing lists, and discussion forums. These may
cλH qH cλL qL
C(qH ) = or C(qL ) = (4) be considered as ad hoc coordination mechanisms [27].
2 2
The authors normalize the value of α to 1 in the case of
depending on the programmer’s type. In this specification, c is an open source software, while for a closed source software,
a cost parameter normalized to 1 for the closed source model α is likely greater than 1 because the closed source approach
and λH > λL > 0 indicates the relative costs for the high- is designed to achieve a coordinated and synergistic outcome.
and low-cost programmers. The parameter c will likely take a However, from a modeling perspective, the authors allow the
value of less than 1 in the open source model. This is because coordination cost parameter to take any value of fraction or
open source programming is viewed by some programmers as multiple. This implies that Open Source coordination can be
a hobby or it may be done for altruistic motives [7]. greater than or lower than in closed source.
The software’s aggregate quality is proportional to the sum The utility specification for programmers, for a reward
of the quality choices of the participating programmers. Partic- amount y, is given by
ipation can vary from a handful in small projects to hundreds
of programmers such as in the case of the Apache web server w( y) = y ν (5)
[26]. Not all of them gain economical rewards. In a study of
the Apache web server project, [6] concluded that higher status
where ν ∈ (0, 1) is the risk aversion parameter. An individual
in a merit-based ranking of contributors is associated with sig-
or entity is defined to be risk neutral, risk averse, or risk seeking
nificantly higher wages. Although, in the model here, only the
if the marginal utility from its payoff is a constant, decreasing,
top-ranking programmer obtains a reward, it can be modified so
or increasing function, respectively [28]. The power function
that there are rewards for several top-ranking
 contributors.
specification used here, and earlier for modeling the consumers’
The effective overall quality is α qi , i.e., the sum of
utility, is widely used in the economics and business literature
individual qualities multiplied by a coordination parameter. The
for capturing risk aversion, e.g., [29].
reason for summing qualities is that large programs are written
The model descriptions above apply to both open source and
in a modular manner. Modern programming languages such as
closed source software developments. The difference between
C++ and Java are well suited for collaborative environments.
the two approaches lies in the differences in parameter values
The reason is that they use object-oriented programming so
and in the manner in which the programmers are rewarded
that the program is modular, and modules can be independently
for their efforts. In the case of an open source, programmers
written, tested, and modified. In addition, there exist large
compete by participating in a tournament for recognition. In the
libraries of programs that can be incorporated into the current
case of a closed source, the firm shares a part of its profit with
project, greatly reducing programming time. Another feature
its programmers.
of these languages is their error handling capabilities. On en-
Next, the authors compare the qualities of open and closed
countering an error, the entire program does not hang, but it
source softwares in monopoly and competitive markets.
gives the signal that an error has been detected so that it can be
debugged. The user obtains utility from the parts of the pro-
gram that work but is denied utility from features that are
IV. Q UALITY U NDER O PEN S OURCE
buggy. The authors assume that the net utility, which is a mea-
sure of the overall quality of the program, is the sum of these The open source model is examined first in a noncompetitive
utilities, which is to say that the overall quality is the sum of the market. The discussion of the noncompetitive market is im-
individual qualities. portant. First, it provides a baseline for comparing competitive
The proportionality constant α accounts for the coordination results. Second, software markets tend to exhibit preferences
between programmers. Thus, when lack of coordination leads towards single dominant firms. Finally, in an early state of soft-
to overlap of programming efforts, the overall quality is not ware development, the software is often sheltered or isolated
improved to the same extent as when the effort is coordinated. from competition within a niche market of innovators and early
Closed source software development encourages and facilitates adopters. The authors assume that the open source environment
close collaboration between programmers. The reason why consists of a set of NOS ≥ 2 programmers that submit im-
open source projects might be less coordinated is that the provements to the open source code. A coordination committee
contributing programmers are located in different locations and costlessly evaluates submissions and selects a winner. The role
operate on a voluntary basis. They have different motivations of the coordination committee may be formal or informal. For
and capabilities. As a result, the open source project generally instance, the winner may be selected informally through mutual
works without explicit planning or schedules. However, coor- consensus among open source users, while a rigid hierarchical
dination might not be too poor either. In particular, most Open structure may govern which improvements, or variations of an
Source projects have a team leader who coordinates the volun- improvement, will be incorporated into the software, as in the

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 909

case of Linux. Let DOS be the number of open source users. This probability can be derived from Proposition 1 with a
Since the software is in a noncompetitive market and free, the specification of the random variable based on the Gumbel dis-
entire market may be considered adopters of the software, so tribution [30]. The above assumption has been used in the tour-
that DOS is equal to the size of the market D. nament literature [21], the literature on patent races [31], and
The winning programmer obtains a reward valued at R(D) the choice literature [30], [32]. Further, it is extensively used
with no reward for the rest of the field. In the event of a in competitive models as the multiplicative competitive interac-
tie, the winner is randomly chosen. The reward may include tion (MCI) model [33].
social recognition and prestige in the open source community, Although it is intuitive that high-cost programmers have a
job offers from firms, consulting opportunities, selection to the lower probability of winning, they do not voluntarily drop out.
board of directors of open source firms, and so forth [7]. Lerner The first-order condition is
and Tirole [7] propose that open source programmers have two  
incentives to participate: the ego gratification incentive and the 
NOS
R(D) υ
qj − qi
career concern incentive. The latter is a pure monetary concern j=1
and refers to future job offers, and shares in commercial open  2 − cλi qi = 0. (9)
source companies. The ego gratification incentive stems from 
NOS
qj
a desire for peer recognition. Both incentives increase with the j=1
number of users of the open source software; therefore, since
reward is a function of the total number of users of the open And note that if qi = 0 in this expression, indicating the
source software, R(D) can be interpreted as pertaining to both programmer dropped out, the left-hand side remains positive,
ego gratification and career advancement. The value of this indicating that the condition is not satisfied. Thus, the solution
reward increases with the size of the user base, i.e., R (D) > 0. is bounded away from zero for all programmer types, implying
Programmer i submits an entry with the expected payoff participation by both high- and low-cost types. Equations for
the high- and low-cost programmers can be written and solved
Ui = p(qi , q−i )R(D)υ − COS (qi ) (6) to yield Corollary 1.1.
Corollary 1.1: From Proposition 1 and the tournament win-
where p(·) is the probability of the programmer winning the ner assumption, the ratio of high- and low-cost programmers’
tournament and q−i is a vector of the endogenously determined planned qualities is qH /qL = λL /λH . The quality of the soft-
planned qualities of the remaining submissions. The probability ware is QOS = (NOS R(D)υ /c)((β/λH ) + (1 − β/λL )).
of any programmer winning the tournament is the probability Social welfare is DQθOS /2 (proof given in the Appendix).
that their entry is evaluated better than the remaining NOS − 1 This result leads to the following interesting insights.
entries, i.e., for programmer i 1) From Corollary 1.1, the quality improvement QOS is
increasing in ν, NOS , and D, and decreasing in β and c.
p(qi , q−i ) = prob [qi + εi ≥ max{qj + εj }j=i ] . (7) Lower risk aversion leads to higher quality levels because
the reward has greater expected utility for the program-
To evaluate this probability, the authors need to specify
mers. A greater number of programmers and a larger frac-
the distribution of max{qj + εj }j=i (details given in the
tion of highly productive programmers increase quality
Appendix). Having done so, the authors can substitute (7)
because of the higher cumulative quality. Improvements
into the programmer’s utility function and obtain the quality
due to D follow from the logic that if reward from effort
choice of the programmer and the expected software quality.
is increased, programmers are motivated to work harder.
Proposition 1 summarizes the results.
2) A particularly interesting result is that the quality of open
Proposition 1: The optimal planned quality for each pro-
source software is decreasing in c. Critics contend that
grammer qOS is the solution of the NOS equations
open source software development is dependent upon
∂p(qi , q−i ) ∂COS (qi ) people who consider it a hobby rather than a serious pur-
R(D)υ − = 0, i ∈ [1, NOS ]. suit, and hence, quality may suffer. In contrast, the model
∂qi ∂qi
∞ here suggests that such hobby considerations decrease
The winning software is 0 [1 − FYN ]
 0expected quality of the  the programmer’s cost of effort, and thereby, increase the
− −∞ FYN , where FYN ≡ N OS
j=1 FQJ and FQj is the cdf of
quality of open source software.
Qj (proofs given in the Appendix). 3) The role of NOS in quality improvement can have two
Proposition 1 provides a general characterization of soft- effects. Larger numbers of programmers decrease the
ware quality in the open source model. However, more specific likelihood of any individual obtaining the reward and,
insights require imposing more structure to this proposition. therefore, act as a disincentive to effort. At the same
For this purpose, the authors make the following assumption. time, more programmers will provide a larger number
Tournament Winner Assumption: The probability of pro- of improvements that cumulatively lead to higher quality
grammer i winning the tournament is given by software. Since the quality expression is increasing in the
number of programmers, the latter is the dominant effect.
qi
p(qi , q−i ) = . (8) 4) Some open source communities may require a critical

NOS
quality threshold in addition to a programmer winning
qj
j=1 the tournament in order to realize any reward. For a

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
910 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

quality threshold Q∗ , the expected reward is p(QOS ≥ would encourage employees to shirk. Therefore, programmer
Q∗ )p(qi , q−i )R(D)υ , not p(qi , q−i )R(D)υ . Although the compensation is generally based on the team’s total quality
analytical solution is intractable, it can be seen that a outcome to motivate programmers to work harder. For example,
quality threshold is detrimental to the open source soft- sharing of profits through stock option plan and bonus is a com-
ware quality since the reward is effectively reduced. If mon practice in many software development firms [37]. This
the quality threshold is sufficiently high, it is possible scheme is taken as the basis for the closed source model. The
that the open source model may suffer from the “public authors made the assumption that programmers who develop
good” problem because programmers unwilling or unable the closed source software get a share of the profits because of
to meet the quality threshold may simply stop making two reasons. First, sharing of profits through stock option plan
contributions. When the threshold increases, the effective and bonus is a common practice in many software development
reward decreases, causing the individual programmer ef- firms. Second, this assumption enables to tie software quality to
fort and the total quality to decrease. This, in turn, reduces reward obtained by programmers.
the probability that the quality will be higher than the The following sequence of actions is considered.
threshold, leading to a death spiral.
5) A question related to programmer heterogeneity is Stage 1) The firm announces its compensation plan and
whether open source quality can be improved if there hires MCS programmers.
is a more heterogeneous population of programmers. By Stage 2) Consistent with the open source model, each pro-
setting λH = 1 + ς and λL = 1 − ς and examining the grammer chooses a planned quality qi and the com-
sensitivity of quality to ς, it can be seen that bination of individual quality outputs determines
the total software quality.
∂QOS   Stage 3) The firm sells the software at a price P and realizes
sign = sign β− (1 − β) profits.
∂ς
 
−ς β+ (1 − β) . (10) This description of the closed source firm follows [37] with
the following generalizations. In [37], a fixed number of risk
neutral and identical team members share the profit. In the
Thus, if there are more high-cost than low-cost program-
present case, a firm decides how many programmers to hire and
mers (i.e., β > 1/2), the quality can improve with greater
how much profit to share with risk averse and heterogeneous
difference in programming costs between programmer types
programmers.
(i.e., higher ς). On the other hand, if the proportion of high-
With a heterogeneous pool of employees, it may be possible
cost programmers is small, greater cost differences reduce the
that the different employees are compensated differently using
software quality. In the latter case, it is better if programmer
the mechanism design. This outcome is called a separating
abilities are close to the average ability within the group.
equilibrium, i.e., an equilibrium where the firm offers different
amounts to different programmer types. The following proposi-
V. C LOSED S OURCE Q UALITY AND tion shows that this is not feasible.
Q UALITY C OMPARISONS Proposition 2: Under the assumptions that individual pro-
grammer qualities are unobservable and the compensation is
In closed source software development, a dedicated team of
based on profit sharing, a separating equilibrium does not exist
programmers is maintained to design and improve proprietary
(proof given in the Appendix).
software code. However, the effort input by individual program-
Therefore, the authors examine a pooling equilibrium where
mers is expensive to monitor, as discussed in the principal-agent
all programmers obtain an equal share. The game is solved by
literature in economics (e.g., [34]). Furthermore, it may be the
backward induction, solving the last stage first.
case that only team outcomes are observable but not individual
Stage 3 analysis: The demand for closed source software
outcomes. It is well documented in the literature on team
SCS (QCS , P ) is given by (2). The profit of the closed source
compensation that precisely knowing the effort expended by
firm is
members of the team is difficult for an outside manager [35]. On
the other hand, depending on inside members to regulate and
ΠCS (QCS ) = Max P SCS (QCS , P )
report upon each other creates even more complicated incentive P
mechanisms. This lack of observability, for example, creates the −SCS (QCS , P ) QθCS
well-known problem of free-riding, internal politics, trying to ⇒P = ∂SCS (QCS ,P )
= . (11)
2
take credit for other’s work, and the myriad of other problems ∂P
that managers have to deal with. Second, within a closed source
environment, programmers collaborate closely and the tasks The marginal cost of reproducing software is usually negligibly
cannot be made entirely independent. Itoh [36] has noted that a small and, therefore, excluded from (11). However, if there is
principal may want to reward an agent as a function of others’ a positive marginal cost of production, it can be scaled out
outputs, even if the outputs are independently distributed, when without affecting the intuition.
teamwork is desirable. Stage 2 analysis: The firm shares a proportion (1 − ρ)
As a result, compensation cannot be tied to individual per- of the profit with employees and retains the rest. Let CCS (qi )
formance metrics. Neither can a fixed salary be given since it denote the cost of the programmer’s effort in the closed source

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 911

environment. CCS (qi ) is given by (4). The utility function for TABLE VI
EFFECT OF ν ON QUALITY, PROFIT, AND WELFARE
programmer i is

υ
(1 − ρ)ΠCS (QCS )
Ui = − CCS (qi ) (12)
MCS

where programmer i receives a fraction 1/MCS of the profit

available to all programmers. The employee’s optimal quality
choice is obtained by solving the first-order condition (steps
given in the Appendix).
Stage 1 analysis: In the first stage, the firm maximizes its
share of the profit less variable expenses such as recruitment,
coordination, overhead, and equipment costs
TABLE VII
υ EFFECT OF θ ON QUALITY, PROFIT, AND WELFARE
Max [ρΠCS (QCS )] − C(MCS ). (13)
ρ,MCS

The authors assume that C(MCS ) = kMCS , where k is a

constant representing the cost for a one person change in MCS .
The expected quality QCS in the closed source model can now
be obtained.
Proposition 3: The ratio of the programmers’ quality levels
qH /qL = λL /λH . The price of the software is QθCS /2, the
quality is

 2 1
  2−θ
1

D θ α2 θ1+ θ υ υ+1 (1 − υ)1−υ β

λH + 1−β
λL
QCS =  
Proposition 4: In noncompetitive markets, the open source
8k 1−υ
software quality is higher, equal to, or lower than the closed
source software quality if
demand is DCS = D/2, and social welfare is given by  
3DQθCS /8 (proof given in the Appendix). >
This result leads to the following interesting insights. NOS R(D)υ =
 
<
1) The number of programmers is endogenously determined
 2  2−θ
2
and does not enter the quality expression. However, the 1 θ
D θ α2 θ1+ θ υ υ+1 (1 − υ)1−υ β 1 − β 2−θ
quality of the closed source software improves with c + .
8k 1−υ λH λL
higher demand (i.e., higher D), better coordination (i.e.,
higher α), more high-quality programmers (i.e., higher
Social welfare is higher, lower, or equal if
β), and lower costs (i.e., lower λH , λL , or k).
2) The effect of the risk aversion is difficult to derive ana-  
>
lytically. However, numerical simulations (see Table VI QθOS  =  0.75QθCS
for the parameters: D = 100, α = 1, θ = 0.5, β = 0.5, <
λH = 2, λL = 1, k = 1) show that higher programmer
risk aversion (i.e., lower ν) reduces quality. (The au- (proof given in the Appendix).
thors performed extensive numerical analysis for various Propositions 3 and 4 provide several qualitative insights that
parameter values. The results in all these simulations are likely to hold for other functional specifications.
were qualitatively similar to that presented in Tables VI 1) There is no universal quality dominance of open source
and VII.) over closed source and vice versa under all circum-
3) The effect of the consumer valuation parameter θ is stances.
also difficult to derive analytically. Numerical simulations 2) The open source model becomes more attractive in an en-
show that a higher value of θ reduces the quality but vironment where a large pool of programmers is willing
increases firm’s profit and welfare (see Table VII for the to work on open source software. The Internet revolution
parameters: D = 100, α = 1, ν = 0.9, β = 0.5, λH = 2, has been a major impetus for open source by allowing
λL = 1, k = 1). programmers worldwide to participate. Quality can be
Comparing the overall quality under open and closed source further enhanced a) if more programmers can be urged
environments, the authors get Proposition 4. to participate; b) if the programmers’ efforts are publicly

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
912 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

recognized; and c) if programmers perceive code contri- programmers in creating application programs. However, the
bution to be a fun task or a hobby. SDKs can be downloaded only after agreeing to a license whose
3) It is difficult to make prior assertions about programmer terms may be unsuitable for free software development.
ability in open source versus closed source environments. Licensing issues must be carefully considered so that the firm
Thus, the authors took β to be the same under both does not accidentally release its proprietary software as an open
circumstances in the analysis. One could argue the possi- source. This is because code under the general public license
bility that a closed source firm has a potential advantage (GPL) cannot be mixed with nonfree software. Alternative open
in that its hiring process may identify and exclude low- source licenses where mixing is permitted include the Berkeley
ability programmers so that only high-ability program- system distribution (BSD) license, the Mozilla public license,
mers are employed despite a heterogeneous labor pool. and the GNU Library GPL. Another approach is used by
On the other hand, it can be argued that open source pro- Sun, which has an Open Source office suite called http://www.
grammers are more skillful because they are self-selected. openoffice.org and a commercial version called StarOffice.
Also, since the source code is subject to extensive peer Beneficial contributions to http://www.openoffice.org can be
review, a level of personal accountability exists in open separately implemented into StarOffice.
source development that does not exist in the closed The firm must also remain vigilant about not making vol-
source model. If the open source programmers are of unteer contributors feel exploited, which it will if it profits at
higher quality compared to closed source, the results of the expense of their work [4]. This may cause them to lose
the model here will tilt in favor of the open source quality. altruism and stop contributing. Firms therefore offer rewards
4) Social welfare calculations reveal that for the same qual- to contributors including putting them on the payroll.
ity of software, open source generates greater social An example of a hybrid model used by Microsoft is described
welfare. by [38] and is called shared source.
If an open source software provides greater quality than a Microsoft is expanding its licensing model to include
closed source software, an interesting issue to consider is our Shared Source philosophy. Shared Source is a bal-
whether a closed source firm such as Microsoft can improve anced approach that allows us to share source code with
software quality by internally adopting an open source strategy customers and partners while maintaining the intellectual
of a tournament, where programmers work independently on property needed to support a strong software business. . .
the same problem and compete for a single reward. This means Over the years Microsoft has made millions of lines of
that the programmers do not divide their share of profit equally; source code freely available to developers through re-
rather, only one from the M programmers will get the entire sources such as SDKs, DDKs, and MSDN. . . The com-
profit shared by the firm. In Corollary 4.1, the authors provide mercial software model is just one model being utilized
a sufficient condition for when it is suboptimal for the firm in the software industry today. It is important to take
to use a tournament approach if the benefit of programmer into account the Open Source Software movement as an
coordination is sufficiently high, i.e., α is sufficiently high. example of an alternative model. . . the common traits are
Corollary 4.1: Let M ∗ be the optimal number of program- providing people with access to source code and allowing
mers hired by the closed source firm under the tournament others to modify and redistribute that code. As a result
approach. Compared to when all programmers receive equal of Microsoft’s statement of position today, many people
rewards, the firm’s profit is strictly lower when it uses the will attempt to say that Shared Source is Microsoft’s failed
tournament approach if α > M ∗υ /θυ (proof given in the attempt at being an Open Source Company. This could not
Appendix). be a more incorrect statement. Shared Source is not Open
Further, given that the number of programmers hired by a Source. The authors recognize that OSS has some benefits
closed source firm MCS is likely to be less than the number of such as the fostering of community, improved feedback,
open source programmers NOS , a closed source firm that is not and augmented debugging. The authors are always looking
already competitive cannot compete with open source firms by for ways to improve our products and make our customers
internally adopting an open source strategy. The cost of hiring more successful, and to that end the authors have incorpo-
enough programmers to make the tournament approach work rated these positive OSS elements in Shared Source.
would be formidable. So far, the authors have examined the improvement in quality
It may however be possible for a proprietary firm to obtain under open source and closed source when each environment
some benefits by collaborating with the Open Source commu- was modeled as a monopoly. The authors next examine quality
nity by implementing a hybrid development system where the implications in a competitive market where open source and
community is encouraged to develop add-on features or product closed source softwares compete for the same set of users.
extensions. For example, in the gaming software industry, it
is common for the major players such as Sony to tap into the
VI. C OMPETITION B ETWEEN O PEN AND C LOSED S OURCE
open source community to develop new features for a game
while the core of the software is architected and developed Let QCS and QOS be the qualities of the closed source and
in-house. A search of the Internet shows that a large number open source softwares, respectively, and P be the price of
of software firms such as Adobe, Microsoft, and Sun offer the closed source software. It is obvious that if QCS ≤ QOS ,
software development kits (SDK) for free. An SDK consists then all D consumers will adopt the open source software.
of an interface, development tools, and libraries that assist Competition arises only when QCS > QOS . (In the model here,

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 913

the demand for a software is a function of only its own and petition is that the closed source software takes away
competitor’s quality and price. Other factors such as industry some of the market for the open source software, reducing
standards and support may also influence a buyer’s decision to the potential reward of open source programmers. Thus,
adopt the open or closed source software. Such issues can be open source programmers are likely to put in less effort.
analyzed using location models such as linear or circular city 2) Open source quality is independent of the closed source
model. The authors separately analyzed the competition using quality. The reason for this counter-intuitive finding is
location models and found that the qualitative results of this that the closed source firm, which can use price to ma-
section remain the same.) In this case, it can be observed easily nipulate demand, finds it optimal to charge a price such
that only consumers with higher valuations, i.e., higher t, will that it gets the demand from all and only consumers on
buy the closed source software and the rest will adopt the open the upper half of the valuation spectrum (i.e., consumers
source software. The indifferent consumer is the one whose with t > 1/2). Because these consumers value quality
surplus is the same for both software. That is, the indifferent more than consumers on the lower half of the valuation
consumer’s t value is given by spectrum, they are more willing to pay a price for the
P higher quality closed source software. The open source
tQθCS − P = tQθOS ⇒ t =  . (14) firm caters to the consumers in the bottom half of the
QθCS − QθOS
valuation spectrum. Because the open source firm does
Thus, demands for the closed and open source softwares are not manipulate demand through price (i.e., it charges a
given by D(1 − P/(QθCS − QθOS )) and D P/(QθCS − QθOS ), zero price for its product), the open source firm gets its
respectively. demand from all and only consumers on the lower half
The authors assume the same sequence of actions described of the valuation spectrum. This demand structure occurs
in the previous section on closed source under monopoly with irrespective of the closed source quality.
the provision that under competition, in stage 2, open source 3) Closed source quality also decreases under competition
and closed source qualities are simultaneously determined. As from open source software. Conventional wisdom sug-
before, the closed source firm sets its price in stage 3. gests that competition will increase quality and decrease
For the closed source model, the authors obtain the quality as price, as in the case of automobiles (Banker et al. 1998)
follow. The optimal price and profit are and other industries. The authors find that for the software
 θ  industry, while competition decreases price, it does not
QCS − QθOS increase quality. The reason for this result is that the
P = (15) higher price charged due to higher quality is offset by the
 θ2 
D QCS − QθOS lower price (zero in the case of open source software) of
ΠCS = . (16) the competing product. Faced with a decline in revenue,
4
the closed source firm reduces its investment in quality.
Using an analysis similar to that of the previous sections, it
can be shown that
VII. C ONCLUSION
 θ 1−υ
CS QCS − QOS
Q2−θ θ
This paper examined software quality improvement under
(1 − ρ)D
υ
β 1−β two distinct methods of software development: the traditional
= MCS θυα2 + (17) closed source approach (e.g., MS Windows) and the emerging
4 λH λL
open source approach (e.g., Linux). Software quality is concep-
  D υ tualized as improvements to the software package not only in
NOS R 2 β 1−β terms of identifying and correcting errors but also enhancing
QOS = + (18)
c λH λL the functionality, scalability, maintainability, portability, and, in
general, enhancing the usability of the software. For both open
and welfare is
and closed source softwares, quality was defined in this study
3D(QOS + QCS ) in terms of acceptable enhancements to the code proposed by
. (19) the programmer community.
8
Using game theoretic analysis and modeling open source
Hence, the following result holds. and closed source software developments under monopoly and
Proposition 5: Under competition, 1) open source quality is competitive settings, the authors derived several useful and in-
independent of closed source quality; 2) open source quality teresting findings. They found that open source software quality
is lower than under monopoly; 3) closed source profit, quality, is not necessarily lower than closed source quality despite the
and price are equal or lower than under monopoly, and are absence of a proprietary interest. Programmers are motivated to
decreasing as open source quality increases; and 4) welfare improve open source quality for an opportunity for recognition
increases (proof given in the Appendix). and future potential benefits.
Many of the results in Proposition 5 have important implica- This study demonstrated that there is no universal dominance
tions for the software market. of open source or closed source approaches in terms of software
1) Open source quality is lower under competition than un- quality. Quality improvement under either scenario depends
der monopoly. The reason for the lower quality in com- on exogenous variables such as market structure (monopoly

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
914 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

versus competition), incentive structure (tournament versus entries to the tournament was not modeled. Restricting such
profit sharing), the number and quality of programmers, co- indirect collaboration between programmers may have led to
ordination between programmers, and endogenous variables underestimation of the quality of open source software in the
such as reward, programmers’ efforts, and number of users. It analysis. Second, the authors made certain simplifying assump-
is virtually impossible to stake a claim that one approach is tions such as quadratic cost functions. Since formal modeling
better than the other for all software and under all circumstances relies on abstraction to address some issues in a stylized setting,
without simultaneous consideration of all of the above factors. limitations are inherent to the modeling process, but analysis of
In other words, there is no “one size fits all” approach to soft- an abstract model can help understand, explain, or predict real
ware development, and hence the authors believe that industry issues in life. Future research may try to address some of the
claims touting one approach over the other are fundamentally more restrictive assumptions.
misguided. The adoption of Linux by large enterprises, e.g., Future research can extend the current analysis to a multi-
Diamler–Chrysler, attests that they found it to be lower TCO for period game. Issues relevant in a multiperiod game such as
the same quality level. Yet other car manufacturers remain with determining per-period quality improvements, prices and up-
Unix, a proprietary software. This is consistent with the predic- grade prices, and the effect of customers’ leap-frogging quality
tion that there is no clear-cut dominance of Open Source over advances may be modeled. It was assumed in this paper that
proprietary software or vice versa. Other than Linux, some data there is only one winner who receives the reward in the open
are available on the adoption of Apache, a web server soft- source tournament. It is possible that there will be multiple win-
ware that competes with Microsoft. A recent survey at http:// ners with different levels of rewards in this tournament. More
news.netcraft.com/archives/web_server_survey.html shows that complex interdependence of programmers’ efforts in determin-
about 70% of servers run on Apache and 20% on Microsoft. ing the total software quality, as opposed to the summation of
However, it is still possible to compare marginal improve- qualities used here, could also be examined. Other issues such
ments in quality between open and closed source developments as standards, licensing policies, and legal issues remain to be
due to each of the above factors, when all else are equal. investigated in the open source context. The model approach is
First, quality is likely to improve in both open source and necessarily an approximation of actual approaches from the in-
closed source with increasing market demand for the software. dustry. It would also be fruitful to continue to study approaches
Both closed source firms and open source communities will taken by industry on quality development in open source. This
have stronger incentive to improve softwares that are desired might include case comparisons, cross-sectional analysis, and
more by users, but for different reasons. Closed source firms longitudinal evolution of quality under the two methods.
see increased market demand as an opportunity for generating
more revenues through sales and licenses, while open source A PPENDIX
programmers have a higher opportunity to gain recognition
Proof of Proposition 1
among their community or signal their talent to potential em-
ployers (open source or otherwise). Second, the quality of open Consider the N th programmer and apply a statistical theorem
and closed source softwares also depends on the nature of [40, p. 183] as an intermediate step.
the market (monopolistic versus competitive) and competitors Theorem 1: If {Q1 , Q2 , . . . , QN −1 } are independent ran-
in the market. The quality of both open source and closed dom variables with cdfs {FQ1 , FQ2 , . . . , FQ,N −1 }, respec-
source softwares decreases in competitive markets. This result tively, and YN −1 ≡ max{Q1 , Q2 , . . . , QN −1 }, then the cdf of
 −1
suggests that competition from the open source movement may YN −1 is FYN −1 = N j=1 FQj .
hamper the innovation in the closed source software industry Proof of Theorem 1: Prob(YN −1 < Q) =  Prob(Q1 <
−1
and vice versa. Q) ∩ Prob(Q2 < Q)∩· · · ∩ Prob(QN −1 < Q) = N j=1 FQj .
Finally, some critics believe that open source software qual- Note that p(qi , q−i ) = prob(Qi − YN −1 ≥ 0). In other
ity suffers because programmers view code contributions as words p(·) is the cdf of the random variable (YN −1 − Qi ) and
a hobby rather than as a serious pursuit or being paid for can be determined using the convolution theorem [40, p. 186]
the same. Since many open source programmers come from
∞ 
N −1
a hacker culture, they perceive identifying bugs and fixes as a
fun or enjoyable activity, and hence such hobby considerations p(YN −1 − Qi ≤ 0) = fQi (x) FQj (x)dx.
−∞ j=1
are reasonable. The authors found that open source software
quality decreases with programmers’ cost of effort, and hobby N
If YN ≡ j=1 FQj is the distribution of the maximum
considerations increases quality by reducing programmers’ cost
quality, its expectation is obtained by the formula E(YN ) =
of effort. The enjoyment associated with open source code ∞ 0
improvement motivates programmers to create and share im- 0 (1 − FYN ) − −∞ FYN [40, p. 65].

provements, even in the absence of pay.
Like most other analytical models, the current model suffers Proof of Corollary 1.1
from a few limitations. First, it focused essentially on deter- Starting with (1) and the tournament winner assumption
mining the equilibrium quality of open source software and
qi
not on the process of development over time. In other words, a Ui = R(DOS )υ − COS (qi ). (A1)
dynamic model capturing features such as incremental quality 
NOS
qj
improvements via peer feedback and collaboration on future j=1

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 915

Solving for the quality chosen, the first-order condition is Proof of Proposition 3


NOS
The authors solve the game by backward induction, solving
R(DOS ) υ
qj − qi the last stage first. Stage 3 analysis gives

j=1
− cλi qi = 0

2 P

N ΠCS (QCS ) = Max D 1 − θ P. (A5)
qj P QCS
j=1

R(DOS )υ [(NH − 1)qH + NL qL ] The optimal price is QθCS /2. Inserting this into the profit
⇒ − cλH qH = 0 expression gains
[NH qH + NL qL ]2
R(DOS )υ [NH qH + (NL − 1)qL ] DQθCS
and − cλL qL = 0. ΠCS (QCS ) = . (A6)
[NH qH + NL qL ]2 4

Further, this implies Substituting this into the Stage 2 analysis gives the utility of
the programmer
R(DOS )υ
≈ qH υ
(NH qH + NL qL )cλH (1 − ρ)DQθCS qi2
− λi . (A7)
4MCS 2
R(DOS )υ
and ≈ qL (A2)
(NH qH + NL qL )cλL The first-order condition can be rewritten for the two pro-
grammer types separately as
where the symbol ≈ denotes “approximated by.” The approx-
υ
imation is reasonable if the input of a single programmer is αθυ (1 − ρ)D
small compared to the total quality. The second-order condition (QCS )θυ−1 = qL (A8)
λL 4MCS
is always met. Dividing the two equations, the authors get υ
λL /λH = qH /qL . Multiplying the first equation by NH and the αθυ (1 − ρ)D
(QCS )θυ−1 = qH . (A9)
second by NL and adding, the authors get the expected quality λH 4MCS
of the software
 An immediate conclusion from these equations is that the

 efforts of the two types are inversely proportional to their cost
υ
NH NL
QOS = R(DOS ) + parameter. Thus
cλH cλL

 qH λL
NOS R(D)υ β 1−β = . (A10)
= + . (A3) qL λH
c λH λL
This implies that the low-cost programmers choose a higher
Since there is no firm profit, social welfare is the surplus of quality level than the high-cost programmers. The authors
the consumers. This is given by now solve for software quality. Multiplying the first equation
by αMH and the second by αML and adding, the authors
1 solve as
DQθOS
D tQθOS dt = . (A4) υ
2 (1 − ρ)D β 1−β
0 MCS θυα2 + = (QCS )2−θυ
4MCS λH λL


 2−θυ
1
υ
(1 − ρ)D β 1−β
⇒ QCS = MCS
1−υ
θυα2 + .
4 λH λL
Proof of Proposition 2
(A11)
Proof by contradiction: suppose a separating contract
(x1 , QCS ) and (x2 , QCS ) is announced where x1 > x2 are the Now, in Stage 1, the firm maximizes its share of the profit
proportions of profit shared. These must satisfy the incentive
 υ  2−θυ
θ
compatibility constraints 1−υ (1−ρ)D β 1−β
ρD MCS θυα2 4 λH + λL
IC1 : Π1 (x1 , QCS | x1 ) > Π1 (x2 , QCS | x1 ) Max −kMCS
2
ρ,MCS 4
IC2 : Π2 (x2 , QCS | x2 ) > Π2 (x1 , QCS | x2 ). θυ
⇒ρ=1− and
2
Clearly, IC2 is violated since each employee prefers the  θ  2−θ
1
higher share. Intuitively, what this means is that since the firm
D α θ υ
2 2θ 3 θ(υ+1)
(1 − υ) 2−θυ β
λH + 1−β
λL 
can only observe the combined output of the employees, it MCS =   .
cannot discriminate between the employees since each one 64k 2−θυ
would pretend to be a low-cost programmer without any cost
incurred for lying.
(A12)

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
916 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

Substituting these back into the quality expression obtains Let M ∗ be the optimal value of MCS . If the closed source
firm does not use the tournament approach, its optimization
 2 1
  2−θ
1
model is given by
D θ α2 θ1+ θ υ υ+1 (1 − υ)1−υ β
λH + 1−β
λL
QCS =   .
8k 1−υ  υ  2−θυ
θ
1−υ (1−ρ)D β 1−β
(A13) ρD MCS θυα2 4 λH + λL
Max −kMCS .
Social welfare is the sum of the firm’s profits and consumer
ρ,MCS 4
(A20)
surplus. Since half the market is served by the firm, profit is
given by DP/2 = DQθCS /4. Surplus is Comparing (A19) and (A20), the authors can note that for
the closed source firm to be worse off using the tournament
1 approach, θυα2 > M ∗ is a sufficient condition.
υ


  DQθOS
D tQθOS − P dt = . (A14)
8
1
2
Proof of Proposition 5
Hence, welfare is equal to (DQθCS /4) + (DQθCS /8) = 1) The result is obvious from the QOS expression.
3DQθCS /8.
2) By comparing the expressions for QOS under monopoly
and competition, and noting that R is an increasing
function, the authors show this result.
Proof of Proposition 4 3) The authors prove this result by contradiction. They know
The proof follows from the expressions of QOS and QCS .
that when QOS = 0, QCS is same under competition and
monopoly. Let us assume a positive QOS . Let the optimal
M , ρ, and QCS for the competition case be MC , ρC , and
Proof of Corollary 4.1 QCCS , respectively. Let the optimal M , ρ, and QCS for the
monopoly case be MM , ρM , and QM CS , respectively.
If the closed source firm uses the tournament approach, then
the reward function for the programmers is given by (1 − ρ) Note that ∂QCS /∂QOS ≤ 0 under competition. Thus, the
DQθCS /4. In the tournament approach, α = 1 as in the open closed source can obtain strictly higher profit with M and ρ
source. From (1) and tournament winner assumption, utility is values equal to MC and ρC , respectively, in the monopoly

υ case because for these values of M and ρ, costs will be
qi (1 − ρ)DQθCS identical in the monopoly and competition cases but QCS , and
Ui = − COS (qi ). (A15)

NOS 4 consequently price and demand, will be higher in the monopoly
qj case. Hence, the optimal profit in the monopoly case cannot be
j=1
lower than that in the competition case.
Solving for planned quality, the first-order condition is Assume that QC M
CS > QCS . If the values of M and ρ in the
monopoly case are set to MC and ρC , respectively, then QCS

υ
(1 − ρ)D   will be higher than QM CS . Consequently, under monopoly, for
Qθυ−1 + (θυ − 1)Qθυ−2 qi = λi qi . (A16) these values of MC and ρC , the profit will be higher than the
4
optimal profit, which contradicts the optimality of MM and ρM .
Assuming that the quality of a single programmer is small The result that price is lower under competition follows from
compared to the total quality, the authors get the first-order the expression for price and the results that QCS is lower under
condition competition.
Welfare is given by the following expression, where the three

υ θυ−1
(1 − ρ)D Q terms, respectively, are the surplus of open source software
= qi . (A17) users, the profit for the firm, and the surplus of closed source
4 λi
software users:
Multiplying the equation for high-cost programmers by MH
1
and for the low cost by ML and adding, the authors get 2 1
DQθCS
υ D tQθOS dt + +D (tQθCS − P )dt
(1 − ρ)D β 1−β 4
MCS + = (QCS )2−θυ . (A18) 0 1
4 λH λL 2

DQθOS DQθCS DQθOS DQθCS

Now, in Stage 1, the firm maximizes its share of the profit = + + +
8 4 4 8
 υ  2−θυ
θ 3D(QθOS + QθCS )
ρD MCS (1−ρ)D β
+ 1−β = . (A21)
4 λH λL 8
Max − kMCS .
ρ,MCS 4
(A19)

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
RAGHUNATHAN et al.: OPEN SOURCE VERSUS CLOSED SOURCE: SOFTWARE QUALITY IN COMPETITIVE MARKETS 917

ACKNOWLEDGMENT ganizations: Competency rallying for software development,” Proc. Inst.

Elect. Eng., Softw., vol. 149, no. 1, pp. 3–17, 2002.
The authors thank A. Bhattacharjee, D. Harter, E. Haruvy, [28] A. Mas-Colell, M. D. Whinston, and J. R. Green, Microeconomic Theory.
and seminar participants at the University of Texas at Dallas New York: Oxford Univ. Press, 1995.
[29] A. K. Basu, R. Lal, V. Srinivasan, and R. Staelin, “Salesforce compen-
and the 23rd International Conference on Information Systems sation plans: An agency theoretic perspective,” Mark. Sci., vol. 4, no. 4,
(ICIS) in Barcelona 2002. pp. 267–291, 1985.
[30] D. McFadden, “Conditional logit analysis of qualitative choice behavior,”
in Frontiers in Econometrics, P. Zarembka, Ed. New York: Academic,
R EFERENCES 1974, pp. 105–142.
[1] T. O’Reilly, “Lessons from open-source software development,” Com- [31] G. Loury, “Market structure and innovation,” Q. J. Econ., vol. 93, no. 3,
mun. ACM, vol. 42, no. 4, pp. 32–37, 1999. pp. 395–410, 1979.
[2] G. Von Krogh, “Open source software development,” Sloan Manage. Rev., [32] R. D. Luce, Individual Choice Behavior. New York: Wiley, 1959.
vol. 44, no. 3, pp. 14–18, 2003. [33] L. G. Cooper and M. Nakanishi, Market Structure Analysis. Norwell,
[3] E. S. Raymond. (1999). The Cathedral and the Bazaar. [Online]. MA: Kluwer, 1998.
Available: http://www.tuxedo.org/~esr/writings/magic-cauldron/magic- [34] S. J. Grossman and O. D. Hart, “An analysis of the principal-agent prob-
cauldron.txt lem,” Econometrica, vol. 51, no. 1, pp. 7–46, 1983.
[4] E. Haruvy, A. Prasad, and S. P. Sethi, “Harvesting altruism in open source [35] B. Holmstrom, “Moral hazard in teams,” Bell J. Econ., vol. 13, no. 2,
software development,” J. Optim. Theory Appl., vol. 118, no. 2, pp. 381– pp. 324–340, 1982.
416, 2003. [36] H. Itoh, “Incentives to help in multi-agent situations,” Econometrica,
[5] G. P. Dwyer, Jr., “The economics of open source and free software,” vol. 59, no. 3, pp. 611–636, 1991.
Federal Reserve Bank of Atlanta, 1999, unpublished paper. [37] B. E. Hermalin, “Towards an economic theory of leadership: Leading by
[6] I.-L. Hann, J. Roberts, S. Slaughter, and R. Fielding, “Economic incen- example,” Amer. Econ. Rev., vol. 88, no. 5, pp. 1188–1205, 1998.
tives for participating in open source software projects,” in Proc. 23rd Int. [38] C. Mundie. (2001, May 3). The Commercial Software Model.
Conf. Information Systems, Barcelona, Spain, 2002, pp. 365–372. [Online]. Available: http://www.microsoft.com/presspass/exec/craig/05-
[7] J. Lerner and J. Tirole, “Some simple economics of open source,” J. Ind. 03sharedsource.asp
Econ., vol. 52, no. 2, pp. 197–234, 2002. [39] R. D. Banker, I. Khosla, and K. K. Sinha, “Quality and competition,”
[8] Y. Benkler, “Coase’s penguin, or Linux and the nature of the firm,” Yale Manage. Sci, vol. 44, no. 9, pp. 1179–1192, 1998.
Law J., vol. 112, no. 3, pp. 369–389, 2002. [40] A. Mood, F. A. Graybill, and D. C. Boes, Introduction to the Theory of
[9] E. Von Hippel and G. Von Krogh, “Open source software and the ‘private- Statistics, 3rd ed. Singapore: McGraw-Hill, 1974.
collective’ innovation model: Issues for organizational science,” Organ.
Sci., vol. 14, no. 2, pp. 209–313, 2003.
[10] B. W. Boehm, J. R. Brown, and M. Lipow, “Quantitative evaluation
of software quality,” in Proc. 2nd Int. Conf. Software Engineering, San
Francisco, CA, 1976, pp. 592–605.
[11] I. Gorton and A. Liu, “Software component quality assessment in practice:
Successes and practical impediments,” in Proc. 24th Int. Conf. Software
Engineering, Orlando, FL, 2002, pp. 555–558.
[12] D. E. Harter, M. S. Krishnan, and S. A. Slaughter, “Effects of process ma-
turity on quality, cycle time, and effort in software product development,” Srinivasan Raghunathan received the M.S. degree
Manage. Sci., vol. 46, no. 4, pp. 451–466, 2000. in business administration from the Indian Institute
[13] C. C. Mann. (2002). “Why software is so bad. . . and what’s being of Management (IIM) Calcutta, West Bengal, India,
done to fix it.” Tech. Rev. [Online]. Available: www.technologyreview. and the Ph.D. degree from the University of Pitts-
com/articles/mann0702.asp burgh, Pittsburgh, PA.
[14] C. K. Prahalad and M. S. Krishnan, “The new meaning of quality in the He is an Associate Professor in the School of
information age,” Harvard Bus. Rev., vol. 77, no. 5, pp. 109–118, 1999. Management, University of Texas at Dallas, Richard-
[15] S. Slaughter, D. Harter, and M. Krishnan, “Evaluating the cost of software son, TX. His research interests include economics
quality,” Commun. ACM, vol. 41, no. 8, pp. 67–73, 1998. of information systems, IT security, and information
[16] Bloor Research. (1999, Oct. 22). Linux Versus NT: The Verdict. [Online]. sharing in supply chains. His research is published
Available: http://www.itsecurity.com/news4/p222.htm in several journals including Management Science,
[17] C. Laird. (1998, Aug.). Linux Versus NT: Are You Getting the Information Systems Research, and various IEEE TRANSACTIONS.
Most From Your OS? [Online]. Available: http://www.sunworld.com/
sunworldonline/swol-08-1998/swol-08-linuxvnt.html
[18] IDC. (2002). Windows 2000 Versus Linux in Enterprise Computing.
[Online]. Available: http://www.microsoft.com/windows2000/docs/
TCO.pdf
[19] Robert Frances Group. (2002). Total Cost of Ownership for Linux
in the Enterprise. [Online]. Available:http://www.ibm.com/linux/RFG-
LinuxTCO-vFINAL-Jul2002.pdf
[20] J. R. Green and N. L. Stokey, “A comparison of tournaments and con-
tracts,” J. Polit. Econ., vol. 91, no. 3, pp. 349–365, Jun. 1983. Ashutosh Prasad received the M.S. degree in eco-
[21] S. Rosen, “Prizes and incentives in elimination tournaments,” Amer. Econ. nomics and the Ph.D. degree in marketing from
Rev., vol. 76, no. 4, pp. 701–715, 1986. the University of Texas at Austin, Austin, TX, and
[22] L. Li and Y. S. Lee, “Pricing and delivery-time performance in a compet- the M.S. degree in business administration from the
itive environment,” Manage. Sci., vol. 40, no. 5, pp. 633–646, 1994. Indian Institute of Management (IIM) Calcutta, West
[23] P. J. Lederer and L. Li, “Pricing, production, scheduling, and delivery- Bengal, India.
time competition,” Oper. Res., vol. 45, no. 3, pp. 407–420, 1997. He is an Assistant Professor in the School of Man-
[24] K. C. So, “Price and time competition for service delivery,” Manuf. Serv. agement, University of Texas at Dallas, Richardson,
Oper. Manag., vol. 2, no. 4, pp. 392–409, 2000. TX. His research is published or forthcoming in sev-
[25] A. Tsay and N. Agrawal, “Channel dynamics under price and service eral journals including the International Journal of
competition,” Manuf. Serv. Oper. Manag., vol. 2, no. 4, pp. 372–391, Research in Marketing, Journal of Business, Journal
2000. of Optimization Theory and Applications, Marketing Letters, and Marketing
[26] A. Mockus, R. T. Fielding, and J. Herbsleb, “A case study of open source Science. His research interests include new product introduction, network
software development: The Apache server,” in Proc. 22nd Int. Conf. effects, software marketing and piracy, advertising, and sales force issues.
Software Eng., Limerick, Ireland, 2000, pp. 263–272. Dr. Prasad received the School of Management’s Outstanding Undergraduate
[27] K. Crowston and B. Scozzi, “Open source software projects as virtual or- Teacher Award in 2001.

Authorized licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.
918 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 35, NO. 6, NOVEMBER 2005

Birendra K. Mishra received the M.S. degree in Hsihui Chang received the Ph.D. degree in account-
petroleum engineering and the Ph.D. degree in ac- ing from the University of Minnesota, Minneapolis,
counting from the University of Texas at Austin, MN.
Austin, TX. He is an Associate Professor of Accounting and
He is a Visiting Assistant Professor at the Ander- the Associate Dean at the Gary Anderson Graduate
son Graduate School of Management, University of School of Management, University of California,
California, Riverside, CA, and an Assistant Professor Riverside, CA. His current research interests include
at the School of Management, University of Texas at performance measurement, information systems, and
Dallas, Richardson, TX. He uses agency theory and data envelopment analysis. He has published more
game theory to model and analyze business prob- than 25 articles in various leading research journals
lems. His research has appeared or is forthcoming including the European Journal of Operational Re-
in Marketing Science, Management Science, Journal of Accounting Research, search, Journal of Accounting and Economics, Journal of Accounting and
Journal of Accounting and Public Policy, Journal of Management Accounting Public Policy, Strategic Management Journal, and The Accounting Review.
Research, and Journal of Business Research among others. Dr. Chang has received many awards for his research articles and one award
Dr. Mishra was awarded the Dean’s Excellence in Research Grant at the for his outstanding teaching. He has supervised an award-winning doctoral
Arizona State University. dissertation. He is an active member of various professional organizations and
has served as a consultant to several profit and nonprofit organizations.

Authorized
View publication stats licensed use limited to: Akademia Gorniczo-Hutnicza. Downloaded on April 17, 2009 at 04:25 from IEEE Xplore. Restrictions apply.