ArticlePDF Available

Open-Source vs. Proprietary Software

Authors:
  • IMT Atlantique, Brest, France

Abstract and Figures

The article studies technological competition between open-source and proprietary software using a model from interaction theory. We argue that the organizational structure of open-source software, allowed by openness of source codes and by the subsequent development of dedicated communities, is a key feature which, together with compatibility, can allow open-source software to overcome existing proprietary standards. This result depends on the distribution of adopters preferences, but holds when proprietary software producers try to react, even quickly. On the contrary, asymmetric hybridization strategies might successfully allow software producers to protect existing standards.
Content may be subject to copyright.
1
OPEN-SOURCE vs. PROPRIETARY SOFTWARE*
Jean-Michel Dalle
Ecole Normale Supérieure de Cachan
61, avenue du Président Wilson - F-94230 Cachan
Tel: (33) (0)49085995 - Fax: (33) (0)149085996
e-mail: < jmdalle@dir.ens-cachan.fr >
Nicolas Jullien
ENST Bretagne & ICI
Technopole de Brest Iroise - F-29285 Brest Cedex
Tel: (33) (0)298001245 - Fax: (33) (0)298001173
e-mail: < Nicolas.Jullien@enst-bretagne.fr >
* A preliminary version of this article was presented as an invited lecture to the
2001 ESSID Cargèse Summer School. We thank many participants for their
comments, and notably Paul A. David, Dominique Foray, Richard Nelson and
Edward Steinmuller.
2
Abstract:
The article studies technological competition between open-source and
proprietary software using a model from interaction theory. We argue that the
organizational structure of open-source software, allowed by openness of source
codes and by the subsequent development of dedicated communities, is a key
feature which, together with compatibility, can allow open-source software to
overcome existing proprietary standards. This result depends on the distribution
of adopters preferences, but holds when proprietary software producers try to
react, even quickly. On the contrary, asymmetric hybridization strategies might
successfully allow software producers to protect existing standards.
KEY-WORDS: open-source software, technological competition, interaction
theory, standards, hybridization.
3
1. The case with open-source software: from incentives to technological
competition
Open-source software has recently become a major interest both for the software
industry and for economic theory. What used to be a Linux
1
-hype has turned
into a growing and much-studied phenomenon. Many economic actors now
decide for an open-source strategy, i.e. adopt open-source software products and
even sometimes publish the sources of the programs they have written instead of
keeping them for themselves as used to be the case in the usual proprietary
model.
In this respect, the main question that has been studied for now in economics is
with incentives: why should so many individuals dedicate their work to open-
source projects from which they seem to get no reward, while these projects
might clearly benefit to many free riders? There are mainly two conflicting
explanations here, the first one relying upon some form of inter-individual
solidarity, possibly rooted in more or less rational behaviors (Lakhani & Von
Hippel, 2000; Harhoff, Henkel & Von Hippel, 2000), while the other is closer to
labor economics and, more specifically, to ‘career concerns’ (Lerner & Tirole,
2000). Although the apparition of an open-source-software economy i.e. of many
business firms dedicated to Linux and to other examples of open source software
is an important element in this debate (Dalle & Jullien, 2000; Jullien, 2001), as
it contributes to create a specific and dedicated labor market, might favor the
second explanation, the major problem is that ‘career concerns’ seem to apply
only to a very limited number of kernel developers rather than to much more
numerous obscure developers (Dalle & Jullien, 2001). Though evidence is not
clear about how open-source communities are actually organized, notably about
the actual proportion of kernel and obscure developers and about their
respective roles, and though both categories indeed seem to be essential, the key
to open-source software seems to be closer to the face that numerous obscure
developers correct infinitely many bugs and therefore critically improve software
efficiency (Raymond, 1998).
Anyway, an important consequence of these works is that open-source software
is undoubtedly here to stay. But then comes the next and for now neglected
question, about technological competition between proprietary and open-source
software. Indeed, if open-source software was always to lose against proprietary
software, then not only would it have attracted much less attention, but also it
would clearly be of considerably diminished interest. As a matter of fact, the
interest in open-source software precisely came from the fact that several
examples, and notably Linux, had gained significant market shares against
proprietary products. Conversely, if open-source software was proven to be able
to succeed against proprietary software standards, then it would not only
become an even stronger issue for the entire software industry but also an even
more interesting subject for economists. Most software technologies are indeed
subject to network effects and thus tend to give rise to de facto standards and to
monopolies (David, 1985, 1987). If open-source had the potential to durably
compete against existing proprietary software standards, then it might become
an extremely interesting tool for public policy to restore competition on quasi-
monopolistic markets.
Studying this “open-source vs. proprietary software” issue implies the use of a
technological competition model with network externalities, which have been
1
An operating system for computers now seriously challenging Microsoft Windows and
others on the server market.
4
used in economics since Arthur (1989). This paper is precisely dedicated to
adapting such a model to the competition between open-source and proprietary
software (section 2), and then to inferring several results with both passive
(section 3) and reactive (section 4) proprietary software producers.
2. A model of technological competition between open-source and
proprietary software
Following most recent literature on technological competition, at least since the
seminal work of Arthur (1989), we consider that the utility of a technology varies
with adoptions, and that adoptions are themselves subject to externalities and
network effects. As it has often been emphasized, these externalities can be local
and/or global
2
. Local externalities mostly account for compatibility issues and
for local word-of-mouth communication of information about technologies with
economic agents in the neighborhood. Global externalities account for the
improvement of the quality and performance (and price) of the technology during
its diffusion. We also account for idiosyncratic components in individual utilities,
as potential adopters are clearly heterogeneous: individual costs and benefits
associated with the use of a given technology clearly depend on several
idiosyncratic factors, and notably here on individual computer skills, a most
important characteristic for software technologies.
The utility of a given technology for agent “i” is therefore generally given by:
() () () ()
iGilikiu
++=
(1)
where
()
iu
is the utility of a given software technology for agent i,
()
iG
(resp.,
()
il
) accounts for global (resp., local) externalities, and
()
ik
accounts for
individual (idiosyncratic) adoption benefits or (individual switching costs.
We further consider that both local and global external economies are simply
functions of the local (resp., global) number of adopters of a similar technology.
Therefore:
() () ()()()
XGixlikiu ++= (2)
where
()
ix
is the number of adopters of the technology in agent i’s
neighborhood. Note that the last term of equation (2) does not depend any more
on each individual adopter but on the global number of adopters of the
technology, denoted by
X
.
We consider here an open-source technology, denoted by OS, and a proprietary
technology, denoted by P. Therefore:
() () ()()()
() () ()()()
PPPPPP
OSOSOSOSOSOS
XGixlikiu
XGixlikiu
++=
++=
(3)
Agent i will adopt open-source software iff:
2
See also Dalle (1995, 1997), Dalle & Foray (1998) for other considerations about
stochastic interaction models and technological competition.
5
() () ()( ) ( ) () () ()()()
PPPPPPOSOSOSOSOSOS
XGixlikiuXGixlikiu ++=>++= (4)
i.e. iff:
()() ()()
[]
() ()
[]
() ()
[]
ikikXGXGixlixl
OSPPPOSOSPPOSOS
>+
(5)
i.e. if local and global externalities “outweigh” indiosyncratic preferences.
We now turn to some specific characterization of the open-source vs. proprietary
software case.
Proposition 1:
() ()
[]
0> XGXGX
POS
(6)
Proposition 1 expresses the superiority of the open-source development model
over a proprietary one, i.e. the superiority of the open-source organization when
compared to a classical software development organization, as it indeed has early
been acknowledged by Microsoft (Valloppillil, 1998).
The 3 main reasons are the following:
(i) Developments in a proprietary organization are mostly ill-targeted
because developers are mainly not users, and therefore do not know which
functionalities to develop or improve first, or simply where the bugs are. On
the contrary, open-source communities benefit considerably from a “users as
innovators” organization (Von Hippel, 1988), and attract numerous
heterogeneous developers which, using their own idiosyncratic experience,
correct various bugs and suggest various new developments. As a
consequence, developments added to open-source software are considerably
more efficient for a given level of adoption than for proprietary software.
(ii) Proprietary software producers get incentives to release improved
versions only from time to time, so that users are in a way obliged to
regularly buy newer versions. Free bug corrections are pretty rare, and
usually limited to critical situations: proprietary software producers prefer to
wait for improvements to be sufficient to support the release of a new
version, i.e. an extra price. On the contrary, open-source software is very
regularly delivered to users through the release of successive versions which
add new functionalities and correct bugs and add minor improvements. As a
consequence, open-source software is also “continuously” more efficient than
proprietary software.
(iii) Finally, the performance of a proprietary technology depends of R&D
investments by its producer. These efforts tend to diminish for a monopolist
i.e. when network effects drive adoption toward a proprietary standard. More
generally, business-firms face a trade-off between investments and profits
which has an impact on which share of extra earnings associated with
increasing returns of adoption is dedicated to further R&D investments and
improvements of their technology. On the contrary, open-source communities
make no profits, while developers contribute for free. Furthermore, open-
source development tends to attract numerous (very) skilled workers which
prefer open organizational cultures. As a consequence, more developers will
generally contribute to a piece of open-source software than to a piece of
proprietary software for a given level of adoption.
To summarize, the efficiency of its organization allow open-source software to
benefit more from adoption externalities than proprietary software. This
6
organizational efficiency is directly implied by open-sourceness, which
guarantees that modifications and improvements will be redistributed and
simply that many developers will have access to the sources and will therefore be
able to locate bugs and to develop new features. To put it yet differently,
openness of sources is a key condition for open-source software to truly benefit
from the creativity of an open-source “community”, i.e. from numerous
independent software developers.
Proposition 2:
() ()
[]
0> xlxlx
POS
(7)
A dominant proprietary software producer gets no incentives to make its
technology compatible with its open-source competitor. On the contrary, any
open-source alternative to an existing proprietary software standard has to be
compatible with the existing standard, it has to be possible to use this open-
source technology in a network constituted of proprietary technologies. This is
typically the case for Linux and most examples of open-source technologies: as
for Linux, it is even now possible to emulate Windows on a Linux machine. As a
consequence, it is easier to adopt locally a compatible open-source technology
rather than a non compatible proprietary technology, for a similar level of local
adoption.
Formally:
() () ()
() ()
12
1
OS
P
lxlxly
lx lx
=+
=
(8)
where y denotes the number of proprietary software users in the neighborhood of
an agent with x open-source software neighbors. Function
2
l then stands for
positive local externalities associated with imperfect compatibility, while function
1
l is considered similar for open-source and proprietary software technologies
since we are dealing with similar technologies.
Another line of argumentation for proposition 2 has to do with local
informational externalities: open-source users tend to proselytize more than
proprietary software users, i.e. to disclose more information, both qualitative and
technical, about the open-source software they use. This is mainly due to
psycho-sociological reasons, but we wonder whether this reason would not apply
as soon as a we would consider any alternative, and notably open-source, to a
given dominant standard.
Note finally that neither proposition 1 nor proposition 2 depend on the price of
open-source software. As a matter of fact, price considerations would certainly
strengthen both, but we believe our results will be stronger if they do not rely on
price considerations: namely, the gratuity of open-source software is not
necessarily here to stay, if open-source software is here to stay. Furthermore,
price is only a limited part of adoption costs. Put differently, we would like to
prove that the main economic interest of open-source software is not gratuity.
If now we further consider that local and global externalities are simple linear
functions, i.e.:
()
()
() ()
()
()
() ()
;
;
OS OS OS OS P P P P
OS OS OS OS P P P P
lxi lxilxi lxi
GX GX GX GX
==
==
(9)
7
with
POSPOS
GandGll ,, as positive constants, then equation (5) becomes:
() ()
[][]
() ()
[]
ikikXGXGixlixl
OSPPPOSOSPPOSOS
>+
(10)
i.e.
() ()
[][]
() ()
[]
ikikXXGixixl
OSPPOSOSPOSOS
>+
βα
(11)
where of course:
00 >=>=
OSPOSP
GGandll
βα
(12)
As a consequence of propositions 1 and 2 we have:
POSPOS
GGandll >> (13)
And therefore:
11 <<
β
α
and (14)
Finally, let:
OSOSOSOSOS
GlGbandGla +=+=
(15)
Then (11) becomes:
( ) () ()
[][][]
() ()
[]
ikikXXbixixba
OSPPOSPOS
>+
βα
1 (16)
b represents here by definition the relative weight of global versus local
externalities (0<b<1) for a given software technology: the closer b will be to 1, the
stronger global externalities relative to local ones, and conversely when b is close
to 0. a stands for a measure of network effects, both local and global, for this
particular technology (a>0): the higher a, the stronger networks effects. a will
typically be very high for operating systems, and relatively high for many
software technologies
3
.
Then, without any restriction since hyperbolic tangent is a bijective function,
agent i adopts open-source software iff:
()() ()
[][][]
{}()
ikXXbixixbath
POSPOS
>+
βα
1 (17)
where
()
11 <
< ik are distributed according to idiosyncratic preferences of
agents in the population. Here, individual preferences are simply compared to
externalities and play the role of individual adoption thresholds. Note also that
the use of an hyperbolic tangent function does not only allows for this
normalization of individual adoption thresholds, but also sets boundaries to local
and global externalities, which is in line with the idea that the importance of a
new adoption diminishes with the number of adopters.
Finally, we will consider here 3 possible distributions of individual preferences
(adoption thresholds), namely:
3
a would be much lower, and b often close to 1, if this model was applied to other, non-
software technologies, while propositions 1 and 2 would not hold.
8
a. A uniform distribution, according to which all preferences are equi-probable.
Although this distribution has classically been assessed in many previous
works on technological competition with local and global interaction models,
it is clearly too restrictive and possibly misleading.
b. A normal, centered (gaussian) distribution: agents on average do not prefer
either technology. They are heterogeneous according to a given standard
deviation.
c. A bimodal distribution, which will allow us to account for the existence of two
subpopulations, as in Arthur (1989), while avoiding to use a double point-
mass distribution. This distribution will be assumed to be symmetric, i.e.
potential adopters have on average no preference for either technology. But
clearly half of them prefer open-source software, while the other half prefers
proprietary software. In both cases, theses preferences are distributed in a
normal way: such a bimodal distribution is indeed properly given by two
(non-centered) normal distributions, each with probability ½.
3. Results with passive proprietary software producer
a. Simulation protocol
For tractability reasons, we classically assume here that neighborhoods are
organized according to a 30x30 2-dimensional torus, i.e. each of the 900
potential adopters having 4 relevant local neighbors. We simulate technological
trajectories by choosing a random adopter at discrete times according to a given
distribution of idiosyncratic preferences
4
and to the existing set of local
neighborhoods: we then get his adoption behavior according to formula (17)
above, considering also the current global level of adoption. Initial conditions are
set to a uniform previous adoption of proprietary software, and we repeat this
algorithm up to 500 000 times. We measure the (possibly infinite) time needed
for open-source software to become a new standard in the place previously
occupied by proprietary software, here defined as a 70% market share. We repeat
this entire process for each α and β between 0 and 1 with step 0.1.
Most results exhibit a phase transition: α and β behave as state parameters
associated with a sharp discontinuity in diffusion regimes. Above critical values
of α and β, diffusion of open-source software is infinitely long – it seems as it will
never occur in economic time – whereas below them diffusion is almost sure and
fairly rapid. Figure 1 presents typical results of our simulations studies. The
large plateau corresponds to an infinitely long diffusion time (i.e. no open-source
standard will ever emerge).
4
For normal and bimodal distributions, we had to deal with negligible distribution tails
outside of the boundaries of our model (minus one and plus one): when it happened to be
the case, the algorithm simply went to the next step and skipped that particular adopter.
9
α
αα
α
β
ββ
β
Ti me
Figure 1: open-source software diffusion time depending on a and b (typical)
Instead of non-obvious 3-dimensional representations, we will present our
results here according to 2-dimensional phase diagrams. That is, we construct
(α,β) graphs delimiting ranges of values of α and β for which standardization
occurs for each software technology. Furthermore, according to our simulation
studies, most phase transition frontier between different regimes seem close to
linear combinations of α and β and are presented this way.
b. open-source is able to defeat proprietary software
Figure 2a represent the general outcome of technological competition between
open-source and proprietary software. For α and β are sufficiently low, open-
source software can eventually replace a proprietary software dominant
standard
5
. Put differently, the superior organization of open-source software
development projects and its compatibility with proprietary software have to be
sufficiently high to grant it with a decisive advantage over proprietary standards.
Clearly, several characteristics of open-source communities are here relevant in
this respect: communities should be large enough and creative enough for β to
be low enough. As for α, open-source software should be very well compatible
with existing proprietary software. One could even doubt whether this will be
enough, and refer then back to what we have named ‘proselytism’, the kind of
which has been created by Microsoft-phobia in favor of Linux and its fellows.
Thus a relevant question is here whether open-source software could defeat
proprietary software when it was not supported by some kind of psychological
feeling favoring the open-source alternative against the dominant proprietary
standard, as not all dominant proprietary standard might create as strong an
antagonistic response.
5
These results indeed complement previous results which showed that new technologies
had to be “better enough” than existing standards to have a chance to defeat them, or,
similarly, that non-linear diffusion and competition processes created endogenous
collective thresholds which had to be overcome for a new standard to emerge (Dalle,
1997).
10
β
ββ
β
α
αα
α
β
ββ
β
α
αα
α
O
S
P
Figure 2a & 2b: phase diagrams, uniform distribution of preferences, b=0.5, a=2.5
(plain line on both figures) and a=5 (dashed line on figure 1b)
β
ββ
β
α
αα
α
β
ββ
β
α
αα
α
O
S
P
O
S
P
Figure 3a & 3b: phase diagrams, uniform distribution of preferences, a=2.5, b=0.2
(3a) and b=0.8 (3b)
Figure 2b compares this result with results obtained for a technology with
stronger network effects (higher a): as expected,
α and β should be even lower in
this case since network effects first favor the existing standard, which is
therefore more difficult to challenge. Fig. 3a and 3b now present phase diagrams
when b varies i.e. when local externalities are more influent than global ones
(Figure 3a), and less (Figure 3b). Results here are as expected since either
α or β
play a more important role, except for some asymmetry while equation (17) is
symmetric. When local externalities have a strong influence (Figure 3a), open-
source software wins almost as soon as
β < 1, while when local externalities have
only a limited influence,
α should be somewhat lower for open-source software to
win. The importance of global superiority when local effects are strong is less
than the importance of local superiority when global effects are strong: in all
respect, an indirect proof of the importance of local network effects.
c. Results depend strongly on the distribution of adopter preferences
11
Figure 4 below presents similar results with different distributions of adopter
preferences. Clearly here, the uniform distribution is the most favorable
situation for open-source software, and also certainly the most unlikely in
reality. The least favorable situation is the normal distribution. An interpretation
of these results is of course that diffusion is easier when there are more
numerous ‘early’ adopters, i.e. adopters who prefer open-source even when
externalities strongly favor proprietary software. Finally, diffusion of open-source
software is easier with a bimodal distribution than with the normal one: here
again, when there exists a sub-population more favorable to the new open-
source software technology, it helps open-source software to be adopted more
easily.
Figure 4 also proves that differentiation could be a good strategy for open-source
software when it is subject to strong network effects. It is indeed easier for such
a new technology to become a standard when adopters mainly belong to two
subpopulations which prefer either technology – i.e. when the new technology is
able to differentiate from the existing proprietary standard – than when adopters
simply normally-distributed heterogeneous preferences. When you want to get
rid of a standard more easily (or with less difficulties), then you should first
differentiate and typically more specifically target a niche.
β
ββ
β
α
αα
α
Figure 4: phase diagram, a=2.5, b=0.5, uniform distribution of preferences (plain
line), centered normal distribution with 0.4 standard deviation (dotted line) and
bimodal distribution with twice 0.2 standard deviation (dashed line)
4. Results with reactive proprietary software producer
a. Simple reactions are likely to be ineffective
Results presented in section 3 above clearly hold only relative to the hypothesis
that proprietary software producers do not react against the diffusion of their
open-source competitor. This is most unlikely, and this is the result why we now
turn to study two of two alternative possible reaction strategies.
Let strategy A denote a passive “do-nothing” strategy, and let us first consider
the alternative strategy, denoted as B, according to which the proprietary
software producer reacts as soon as open-source software reaches x% market
share. It can for instance lower its price, although we have already mentioned
that price is not necessarily a major determinant of adoption i.e. that demand
can be relatively inelastic to price. Another possible reaction by this producer
12
could correspond to larger investments in R&D to improve the efficiency of its
product, be they supported by weaker profits or by debt. Finally, another
reaction, considered today by many proprietary software producers, would be to
rent their software instead of selling it, so that users would always benefit
immediately from bug corrections and new improvements. In all cases, these
evolutions represent in our model a significant increase in
β, since the utility of
the proprietary software technology for users is simply rendered higher for a
given level of adoption: therefore, the negative difference with the utility of the
open-source software solution is reduced. Typical results are given by Figure 5.
β
ββ
β
α
αα
α
Figure 5: phase diagram, a=2.5, b=0.5; uniform distribution with either strategy A
for the proprietary software producer or with strategy B with
β
β
+0.2 as soon as
open-source software reaches 5% market share (plain line for both, no difference);
centered normal distribution with 0.4 standard deviation with either no reaction
(dashed line) or strategy B (dotted line)
Although we consider a quick and important increase in
β, we get almost no
visible difference between strategy B and strategy A, although we would have
expected variations at least in line with the implemented variation of
β.
Technological competition is here extremely path-dependent: as soon as it has
started, it is almost impossible to make it deviate from its trajectory. An
interpretation is that initial conditions, i.e. the first adopters of open-source
software critically influence its diffusion, even with later a less important
advantage in terms of global externalities.
b. Hybridization can be more effective
Let us now consider technological hybridization as another possible strategy for
proprietary software producers. By technological hybridization, we mean that
either technology can adopt one or more feature of the other. This phenomenon
has recently been emphasized by economic historian D.A. Kirsch (1997) during
both the early and recent histories of the automobile. Early on, the success of
combustion engine vehicles was consistently related to the fact that they had
implemented an electric starter, taken from their electric vehicle competitor,
which allowed for a much easier start and therefore proved extremely important
for many users, and specially for women. Now, there is once again consistent
discussions about new “hybrid” vehicles which would possess both a combustion
engine and an electric engine (or a fuel cell). A new hybrid – variant – might
13
therefore be on its way, with the idea of combining the environmental interest of
electric vehicle and the autonomy and power of combustion engine.
Hybridization in economics works somehow as it does in genetics, where
fragments of chromosomes which account for characteristics of the parents get
hybridized and mixed together to become the genome of the child: here,
technological characteristics from the two technological ‘parents’ give birth to
technological ‘children’ with several characteristics from each parent: economic
hybrids however probably tend to remain closer to either one of their parents. As
we have suggested elsewhere (Dalle, Jullien & Simon, 2000), hybridization is a
key strategy for producers who face technological competition. It is not only that
they try to make their technology better by copying characteristics from other
technologies – of course they do –, but hybridization also plays a significantly
more important role as influences demand by modifying preferences within
different subpopulations of potential adopters who would prefer either
technology ab initio, and who might change their minds as the other technology
includes several features of their favorite. Modifying the underlying preferences
of potential adopters can prove an extremely valuable strategy, specially for a
dominant standard to resist technological invaders, as it influences the non-
linear dynamics of diffusion and competition processes.
As a matter of fact, we are already observing several examples of proprietary
software hybridizing itself with “open-source” features, typically retaining almost
all characteristics of proprietary licenses while only its source code becomes
accessible. This is why we now implement such a strategy for the producer of the
existing proprietary standard. We consider a bimodal distribution of potential
adopters: there exist two distinct subpopulations which prefer each available
technology. The proprietary software producers then implements features of
open-source software during competition. Formally, the distribution of adopter
preferences gets modified each time hybridization occurs: the left mode, i.e.
adopters who prefer open-source software, moves rightward, i.e. adopters prefer
open-source software “less” as proprietary software implements several features
of open-source software. The right mode is left unmodified: we consider that
adopters who used to prefer proprietary software still prefer proprietary software
even once it has implemented several characteristics of open-source software:
the new hybrid technology has simply been made even ‘better’, including new
characteristics without losing some therefore with no eviction effect, i.e. with no
adopter being ‘disappointed’ and switching back its preferences to open-source
software.
We simulate here two hybridization strategies, one quick and one slow. Namely,
strategy C corresponds to an hybridization where the left mode of the
distribution of adopters preferences moves to the right of 5% of the distance from
its peak and the origin each 1000 periods, while strategy D corresponds to the
same move but each 200 periods. Results presented in figure 6 show an
important influence of hybridization strategies on the outcome of technological
competition between proprietary and open-source software:
α and β have to be
significantly lower for open-source software to win when the proprietary software
producer adopts an hybridization strategy.
14
β
ββ
β
α
αα
α
Figure 6: phase diagram, a=2.5, b=0.5: bimodal distribution with either no reaction
(strategy A, plain line) or with slow hybridization strategy (strategy C, dashed line)
or with rapid hybridization (strategy D, dotted line) from the proprietary software
producer
A relevant question here would the sensitivity of users who prefer open-source
software to hybridization strategies: they should indeed be credible – recent
examples from Microsoft have proven that it might sometimes not be the case –
and should also render proprietary software close enough to open-source
software to convince users for which the openness of sources is a key condition
of their preferences. One could wonder whether it does not purely mean turning
proprietary software into open-source software, and if the ongoing invention of
less public license will retain enough characteristics of open-source.
5. Conclusion: strengths and weaknesses of open-source software
Open-source software can sometimes defeat proprietary software. However, this
is far from being always true. It relies on the efficiency of its organization.
Although the basics of this organization are determined by the openness of
sources, it relies of course on the community of developers and on its internal
organization. In a way, one could interpret the fact that many such communities
are not associated with ancillary business firms as a further development of their
‘internal’ organization. Our models points out that the organizational level is
here crucial, since it will crucially determine the outcome of the competition
between open-source and proprietary software. In what extent then are open-
source software communities able to organize themselves, to ‘self-organize’, and
is it sufficient to grant them with enough competitive power? This is still an open
question, which further studies should try to address: this is an important
question also if we were to consider public support to open-source projects, as is
already the case in Europe, both at national and at the European levels, as a
means to correct market failures – monopolies and closed de facto proprietary
standards – due to network effects, at least in software markets. Public
intervention cannot simply consider existing open-source communities and give
them a hint: economists have to suggest ways to help these communities
improve their organizational structure.
Apart from this, the competitiveness of open-source software also depends on its
compatibility with existing proprietary solutions, and on the distribution of
adopter preferences. This can be either a crucial strength or a crucial weakness
15
for open-source software. To take but one important example, think about the
very different destiny of Linux on the server market and on the global PC market:
it has won considerable market shares in the first, while it is still stagnant in the
second. A straightforward explanation for this, amongst other issues, is the lack
of an appropriate graphic user interface (GUI, or desktop) for Linux which is not
needed for servers since they are maintained by skilled users – geeks, once again
– who even prefer the older ‘command line’, while normal users need an efficient
GUI. In our framework, it means that the distribution of adopter preferences in
the server market allowed for the diffusion of Linux, whereas it did not in the
global market. A possible conjecture in this last respect might be that there were
not enough early adopters, that the left mode of a bimodal distribution was not
‘big’ enough.
A even more interesting point comes when we add that significant effort is now
being done to develop such a GUI for Linux, i.e. this time to hybridize open-
source with proprietary software. We have made simulations for this, which have
not been reported here as their main conclusion is that asymmetric
hybridization by an open-source newcomer has no discernible effect on the
outcome of its competition against a dominant proprietary standard. But this
absence of result is itself a result here, since it precisely means that developing a
GUI for Linux will probably not be sufficient to change its destiny on the PC
market. To put it differently, and although these issues are of course still
blurred, it might only be part of a potential solution, which should also include a
further improvement of Linux compatibility with Windows, and a further
increase in the efficiency of its development.
It is then an open question whether such an increase in organizational efficiency
is simply feasible, since it certainly depends not only on the number and the
personality of kernel developers, but also on the involvement of ancillary
business firms which would not only provide dedicated services, as is generally
the case with open-source software firms, but which would also act as quasi-
editors. But the question then is about how they would then earn money since
they could not sell free open-source software. Linux will perhaps never replace
windows in our offices: perhaps efforts in this direction are even misguided, all
the more so as they are diverting efforts from other projects which could win
their own open-source vs. proprietary software competition.
6. Bibliography
Arthur W.B. (1989), Competing technologies, increasing returns and lock-in by
historical events, Economic Journal 99: 116-131.
Dalle J.-M. (1995), Dynamiques d’adoption, coordination et diversité, Revue
Economique, 46: 1081-1098.
Dalle J.-M. (1997), Heterogeneity vs. externalities: a tale of possible technological
landscapes, Journal of Evolutionary Economics 7: 395-413.
Dalle J.-M., Foray D. (1998), The innovation vs. standardization dilemma : some
insights from stochastic interactions models), in Schweitzer F. & Silverberg G.
(eds), Evolution and Self-Organization in Economics, Duncker & Humblot:
Berlin, pp. 147-182.
Dalle J.-M., Jullien N. (2000), NT vs. Linux, or some explorations into the
economics of Free Software, in Ballot G., Weisbuch G. (eds), Application of
simulation to social sciences, Hermès: Paris, pp. 399-416.
16
Dalle J.-M. & Jullien N. (2001), ‘Libre’ software: turning fads into institutions,
miméo.
Dalle J.-M., Jullien N., Simon B. (2000), The economics of technological
hybridization from automobile to Libre software, or why Apple did win against
Microsoft, paper presented to the 5th WEHIA Conference, Marseille, June.
David P.A. (1985), Clio and the economics of QWERTY, American Economic
Review (Papers and Proceedings) 75: 332-337.
David P.A. (1987), Some new standards for the economics of standardization in
the information age, in Dasgupta, Stoneman Eds., Technology policy and
economic performance, Cambridge UP, Cambridge, Mass., pp. 206-239.
Harhoff D., Henkel J., Von Hippel E. (2000), profiting for voluntary information
spillovers: how users benefit by freely revealing their innovations, mimeo,
26p.
Jullien N. (2001), Impact du Logiciel Libre sur l’Industrie Informatique, PhD
thesis, 316 p.
Kirsch D.A. (1997), Technological hybrids and the automobile system: historical
considerations and future directions, mimeo, UCLA.
Lakhani K., Von Hippel E. (2000), How Open Source software works : ‘free’ user-
to-user assistance, MIT Sloan School of Management WP #4117.
Lerner J., Tirole J. (2000), The simple economics of Open Source, mimeo, 38p.
Raymond E.S. (1998), The Cathedral and the Bazaar, accessed on 19/04/01 at
http://www.tuxedo.org/~esr
Valloppillil V. (1998), Open Source Software, a (new?) development methodology,
Microsoft Internal Report.
Von Hippel E. (1988), The sources of innovations, MIT Press.
... 3 In the research arena, many have documented reasons explaining where contributors find their incentives to develop and improve software which is usually available for free. Among possible answers are career concerns and peer recognition (Lakhani and von Hippel, 2003;Weber, 2000;Dalle and Jullien, 2001;Lerner and Tirole, 2002). Meanwhile some stress the idea that open source is a form of gift economy based on altruistic motives (Raymond, 1999;Weber, 2000;Edwards, 2001;Kelty, 2001). ...
... Meanwhile some stress the idea that open source is a form of gift economy based on altruistic motives (Raymond, 1999;Weber, 2000;Edwards, 2001;Kelty, 2001). Other questions are interested in the analysis of open source software as a complex public good (Bessen, 2002); competition between open source software and proprietary software products (Dalle and Jullien, 2001;Casadesus-Masanell and Ghemawat, 2006); and the influence of public licenses on open source software development (Weber, 2000;Dalle and Jullien, 2001). ...
... Meanwhile some stress the idea that open source is a form of gift economy based on altruistic motives (Raymond, 1999;Weber, 2000;Edwards, 2001;Kelty, 2001). Other questions are interested in the analysis of open source software as a complex public good (Bessen, 2002); competition between open source software and proprietary software products (Dalle and Jullien, 2001;Casadesus-Masanell and Ghemawat, 2006); and the influence of public licenses on open source software development (Weber, 2000;Dalle and Jullien, 2001). ...
Article
ABSTRACT This study attempts to explore the price strategic competition between the proprietary
... Some scholars study the competition between proprietary software and its open source alternatives. Dalle and Jullien (2002) examine the technological competition between proprietary and open source software by an interaction theory model; Meng and Lee (2005) investigate the compatibility between open source software and its proprietary substitute; Lin (2008) analyzes the impact of users' expertise on the market in which proprietary software competes with open source software; Pradniwat (2008) and Gramstad (2014) study the competition between a commercial software and a free-of-charge open source substitute in the emergence of piracy; Xing (2010) considers the quantity competition between open source and proprietary software providers; Cheng, Liu and Tang (2011) examine how the network externalities affect the competition between proprietary and open source software; Xing (2013 Xing ( , 2014a) investigates the price competition between commercial open source and proprietary software; Zeroukhi and Pénard (2014) study whether public subsidies for open source software are socially desirable and how the extent of compatibility between open source software and proprietary software influences the optimal subsidy offered. However, they do not consider the impact of open source software on the quality of proprietary software. ...
Article
Full-text available
Purpose: A growing number of open source software emerges in many segments of the software market. In addition, software products usually exhibit network externalities. The purpose of this paper is to study the impact of open source software on the quality strategies of proprietary software vendors when the market presents positive network externalities. Design/methodology/approach: To analyze how open source software affects the quality of proprietary software, this paper constructs two vertical differentiation models: the basic model considers proprietary software monopolizing the market, and the extended model considers proprietary software competing with open source substitute. Findings: This paper mainly finds that the presence of open source software does not necessarily lead to the improvement of proprietary software quality. The network externalities and the compatibility between open source and proprietary software can change the impact of open source software on the quality of proprietary software and affect the quality strategies of proprietary software vendors under certain conditions. Originality/value: The main contribution of this paper is to examine the effect of open source software on the quality strategies for proprietary software vendors in software markets with network externalities. © 2015, Universitat Politecnica de Catalunya. All rights reserved.
... Recently, the academic literature has paid close attention to open source problem, in which the research on competition between open source and proprietary software is a hot area. From a technology, user skill and product innovation effort point of view, scholars have investigated the competition between open source and proprietary software (see [3], [4] and [5]). However they haven't analyzed how the emergence of open source software affects the R&D risk of proprietary software producer. ...
Article
Full-text available
By extending Cournot model, this paper investigates how the appearance of open source software (OSS) affects the R&D risk in a market initially monopolized by proprietary software (PS) producer. The software market exhibits network externality. It is found that (i) the R&D risk level of PS producer in the market monopolized by PS isn't higher than in this market emerging OSS; (ii) there exist some R&D cost functions making R&D risk level be lower when PS producer monopolizes the market than when it competes with an OSS producer.
Chapter
Open Source development projects are internet-based communities of computer programmers (von Hippel and von Krogh, 2003). Internet technology not only enables worldwide and almost cost-free distribution of software, but also enables a distributed production of software by users (von Hippel, 2001). The physical distance between the community members (programmers) prevents most face-to-face contact. This condition, together with the internet-based communication that is limited to written conversation and software code, provide the basis for our investigation. Using knowledge creation theory we discuss and analyse the Open Source software development process. How is new knowledge generated in Open Source projects? Who constitutes Open Source communities and how do people interact? Based on findings that indicate a direct sharing of tacit knowledge in Open Source communities (von Krogh, Spaeth and Lakhani, 2003), we propose how this may occur without co-location of the sharing parties. The knowledge creation process can be enabled by activities that take into account the emergent nature of Open Source projects. We address the question of what the role of a knowledge activist can be in Open Source and draw theoretical and practical implications.
Article
Full-text available
In the present times the most talked about issues in the knowledge driven economic system are Free and Open Source Software (FOSS) & Intellectual Property Right (IPR), both of which exist at poles apart. The question that prevails that is about the relevance of the either and that of the dominance of each one. The paper tries to probe into issues of general and specific relevance of FOSS and IPR as suppliers of certain set of utility and benefit to the user. It will also check the validity of the claim of FOSS and it licensing procedure comparing it with the user ship obligations of IPR protected products and services. The premise to the paper is that both FOSS and IPR have to exist and compliment each other ensuring a strong presence of resources in the public and private domain. The paper would certainly work on the area to validate the existence of conflict between the FOSS and IPR, or it is a mere false caution raised by groups pursuing either cause. Finally the paper would propose to demarcate the areas of dominance of FOSS and IPR and prove the utility at the socio-economic front.
Article
College students will not significantly contribute to open source because they do not know of its existence and because of the difficulty and lack of incentive to do so. With this contribution, open source would corner the software market. Through surveys of computer science students and our own personal attempt at contributing to the Open Source Movement, we defend the first part of our thesis. More specifically, we explore and detail students' responses to questions about their familiarity of the movement's existence and their willingness to contribute. Their answers showed that on the whole, many college students know little to nothing about the movement and even those that did, they would be unwilling to contribute. Furthermore, the question about what non-monetary incentive would be necessary for them in order to contribute demonstrated their willingness to assist the movement given that they received a reasonable compensation. The combination of these responses demonstrate that the open source movement needs to publicize more and offer some incentives for contribution in order to attract college students. The difficulty of contributing to open source will be shown by detailing our experience. By showing that in order to significantly contribute, we had to understand the operability of numerous lines of code and convince members of the movement of our credentials, we demonstrate the difficulty of substantially assisting Open Source. By exhibiting our code, we show that functional, easy-to-read, and well-documented code is not sufficient to obtain the credentials necessary for its acceptance and that prior experience is often the only way to get it. These two factors will show that it is difficult for college student newcomers to break into the development. The second part of our thesis is defended by articles, statistics, and the demand for code that's open and easy to modify. Many articles such as Is Microsoft Afraid of Open Source? suggest that the Open Source Movement poses a threat to the proprietary business model, but that except in cases of market server software, the threat is low at best [LS]. Statistics are then used to show how insignificantly college students are assisting the Movement when compared to the potential of this group for broad contribution. This potential is then emphasized by showing the significance of contribution to the few college students who do participate in the movement. For instance, "Dell, a company started by a college student, is now a multibillion dollar company," among other companies [O'R]. Using the combination of these two observations, we conclude that open source software can be improved in regards to both quantity and quality of functionality, operability, and ease of use. Through statistics of current demand for open source software, we extrapolate the damage to the proprietary market that these contributions will cause. [RWC] We then state that there is a high correlation between people who program in open source and actually use open source operating systems and software in order to emphasize the impact that college students switching to open source would have on the proprietary market as well. We then conclude that the creation of better software and the appreciation for open source that college students will obtain as a result of programming for it will cause the Open Source Movement to corner proprietary software.
Conference Paper
The social welfare differs from a software market when monopolized by open source software (OSS) and proprietary software (PS) were investigated by modifying Cournot model. Assuming the market exhibits network externality and software user expectations are fulfilled or myopic, the results show that the social welfare of the market monopolized by OSS may be lower than that of this market monopolized by PS. Therefore the social planner should not unconditionally encourage open source software when developing the software industry policy. Moreover, the probability of “the social welfare of the market monopolized by OSS is lower (or higher) than that of this market monopolized by PS” may be affected by user's software development skill, OSS's learning (maintenance or development) cost and network externality.
Article
By extending Cournot model, this paper explores the impact of open source software (OSS) to the price, output and profit of proprietary software (PS) and social welfare in a software market. In this paper, we assume that proprietary software producer aims at maximizing profit, the open source software is freely available and the market exhibits network externality. We find that the appearance of open source software in a software market monopolized by the proprietary software doesn't necessarily lower the price of proprietary software, decrease the output and profit of proprietary software producer and increase the level of social welfare, what happens in reality depends on the learning (maintenance or development) cost of open source software.
Article
By extending Hotelling game model, this paper explores compatible strategies between open source software (OSS) and proprietary software (PS). In this paper, we assume that PS producer aims at profit maximization, OSS producer pursues market share maximization and software user's expectations on network size may be fulfilled or myopic. For each type of user expectation, we study both cases of market fully covered and partly covered. The main findings are following: (i) in the case of myopic expectation, both OSS and PS producers choose maximum compatible degree in equilibrium. This conclusion is true in both fully covered and partly covered markets; (ii) in the case of fulfilled expectation, equilibrium compatible strategies depend on market coverage. When the market is partly covered, both OSS and PS producers choose maximum compatible degree in equilibrium. But when the market is fully covered, PS producer may choose maximum compatible degree or incompatibility as its equilibrium strategy (The choice depends on the learning (maintenance or development) cost of OSS, user's software development capacity and network externality); (iii) how the competitive strategies of OSS and PS producers are affected by the rival's compatible degree may be different in different user expectations (or market coverage).
Article
In recent years, the problem of open source is a very hot topic in software industries, however the studies about how open source software (OSS) affects competitive strategies of software industry are rare. Though expanding the classical Hotelling model, this paper investigates the product innovation decisions of proprietary software (PS) firm when open source software appears. In this paper, the model includes not only important factors (e.g. the network externality and software compatibility) affecting the traditional software industry competition, but also the OSS learning (maintenance or development) costs and user's software development skills. The following conclusions are found: (i) the compatibility degree between OSS and PS affects R&D decisions of PS; (ii) the R&D effort of PS increases with the learning (maintenance or development) costs of OSS; (iii) how network externality and user's software development skills affect the R&D effort of PS depends on the compatibility degree between OSS and PS.
Article
Full-text available
"Nowhere does history indulge in repetitions so often or so uniformly as in Wall Street," observed legendary speculator Jesse Livermore. History tells us that periods of major technological innovation are typically accompanied by speculative bubbles as economic agents overreact to genuine advancements in productivity. Excessive run-ups in asset prices can have important consequences for the economy as firms and investors respond to the price signals, resulting in capital misallocation. On the one hand, speculation can magnify the volatility of economic and financial variables, thus harming the welfare of those who are averse to uncertainty and fluctuations. But on the other hand, speculation can increase investment in risky ventures, thus yielding benefits to a society that suffers from an underinvestment problem.
Conference Paper
Open source software development represents a fundamentally new concept in the field of software engineering. Open source development and delivery occurs over the Internet. Developers are not confined to a geographic area. They work voluntarily on a project of their choice. As new requirements emerge, the software is enhanced bv the user/developers. In this paper we show a comparative study of open source and closed source software development approaches and present a software life cycle model for open source software development.
Article
The article presents an economic analysis of Libre software and of its sustainability as a new economic model for software. We underline the role of Libre software development communities and analyze incentives for both kernel and obscure developers. We emphasize the role of the so-called ‘public’ licenses to provide an appropriate institutional framework. We show that several features of Libre software also allow it to improve faster than proprietary software, and therefore to achieve strong market performance when competing against existing standards, even when proprietary software producers react. We illustrate our point using a simple local and global interaction model to study the technological competition between Linux and Windows on the server operating system (OS) market. We finally argue that if sufficient initial momentum could be created through public intervention Libre software could turn from a fad into an efficient economic institution to correct inefficiencies due to network externalities.
Article
There has been a recent surge of interest in open source software development, which involves developers at many different locations and organizations sharing code to develop and refine programs. To an economist, the behavior of individual programmers and commercial companies engaged in open source projects is initially startling. This paper makes a preliminary exploration of the economics of open source software. We highlight the extent to which labor economics, especially the literature on 'career concerns', and industrial organization theory can explain many of these projects' features. We conclude by listing interesting research questions related to open source software.
Article
I anatomize a successful open-source project, fetchmail, that was run as a deliberate test of some theories about software engineering suggested by the history of Linux. I discuss these theories in terms of two fundamentally different development styles, the "cathedral" model, representing most of the commercial world, versus the "bazaar" model of the Linux world. I show that these models derive from opposing assumptions about the nature of the software-debugging task. I then make a sustained argument from the Linux experience for the proposition that "Given enough eyeballs, all bugs are shallow," suggest productive analogies with other self-correcting systems of selfish agents, and conclude with some exploration of the implications of this insight for the future of software.
Article
Some very successful “open source” software products have been and are being developed, distributed, and supported in the field on a voluntary basis by and for users themselves — no supplier required2. The motives that induce users to contribute to an open source project “for free” and the mechanisms by which the various tasks can be effectively carried out are currently a subject of study for both practitioners and academics. To this point, explorations of the mechanics of and the incentives to participate in open source software projects have focused on the core tasks of developing and debugging and improving the open source software itself. Major motives used to explain why users would voluntarily work on these basic tasks include:1. a user’s direct need for the software and software improvements worked upon, 2. enjoyment of the work itself and 3. the enhanced reputation that may flow from making high-quality contributions to an open source project.