Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
COOPERATION AND DISRUPTIVE BEHAVIOUR –
LEARNING FROM A MULTI-PLAYER INTERNET
GAMING COMMUNITY
Michael Hahsler and Stefan Koch
Department of Information Business, Vienna University of Economics and Business Administration
Augasse 2-6, A-1090 Vienna, Austria
{michael.hahsler|stefan.koch}@wu-wien.ac.at
ABSTRACT
In this paper we report possibilities and experiences from employing Counter-Strike, a popular multi-player Internet
computer game and its resulting online community in research on cooperative behaviour. Advantages from using this
game include easy availability of rich data, the emphasis on team-playing, as well as numerous possibilities to change the
experiment settings. We use descriptive game theory and statistical methods to explore cooperation within the game as
well as the way the player community deals with disruptive behaviour. After a quick introduction to the basic rules of
Counter-Strike, we describe the setup of the Internet game server used. We then present empirical results from the game
server logs where cooperation within the game is analyzed from a game theoretic perspective. Finally we discuss the
applications of our results to other online communities, including cooperation and self-regulation in open source teams.
KEYWORDS
Game theory, cooperation, communication, internet community
1. INTRODUCTION
With the ever expanding possibilities of the Internet, understanding when and why individuals trust each
other and cooperate to achieve their goals together becomes more and more important. Concepts like
electronic commerce, groupware, virtual teams, digital universities and many more require as a prerequisite
established trust and cooperation in an electronic context. But unfortunately it seems that trust is harder to
achieve in the impersonal electronic context than it is with conventional face-to-face contact (Rocco, 1998).
In this paper we use a popular multi-player Internet computer game to investigate cooperative behaviour in a
pure electronic context. For this research we use a game theoretic approach.
In the area of normative game theory much work on strategic behaviour, where to cooperate is a possible
strategy, has already been done (Neumann and Morgenstern, 1944; Nash, 1950; Nash, 1950a; Nash 1951).
Normative game theory analyzes consequences of strategic behaviour by superrational players. However, in
real settings, like electronic consumer-to-consumer commerce, rationality of human players is limited.
Human players do not always exactly know what they want and it is also common that they do not posses
complete knowledge of the rules and the possible pay-offs of their various actions. This limits the direct
application of solution concepts of normative game theory, like the Nash equilibrium, dominant strategies,
incredible threats and so on. However, we use these concepts as benchmarks for descriptive game theory to
compare the normative solutions with the actual behaviour of real players.
Using computer games for academic purpose is not a new idea, for example Laird describes the research
of his group in building agents with artificial intelligence within the setting of the game Quake II (Laird,
2001), and a similar game, Doom, has been used as a user interface for system process management (Chao,
2001). For our research we use the wide-spread and freely available Counter-Strike modification for one of
the most highly acclaimed games in the last years, Half-Life. This environment offers excellent graphical
representation and therefore entertainment and high usage. In this game, two teams of several players have to
achieve conflicting goals, e.g. one team tries to rescue a number of hostages, while the other team tries to
prevent this. Counter-Strike therefore aims at enhancing the team-playing aspect while also offering text chat
and voice communication, and thus is especially suitable for analysing cooperative behaviour.
1/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
2. GAME AND EXPERIMENT SETUP
Using a multi-player computer game offers several advantages: incentive structure compatible for players,
virtual reality environment, almost perfect information for analyses using log files, possibilities to change the
setting, fun factor, communication maybe using several channels like text chat or voice, and presence of
emotion for the players. While the incentive compatibility ensures the players actions in accordance to the
goals, the virtual reality environment and fun factor guarantee a sufficient number of players. It is one of the
most consistent and robust findings in sociological literature that communication has a positive effect on
cooperation and trust (Kollock, 1998). As multi-player computer games today nearly always feature at least
one form of communication between the players, in the simplest variant a text chat facility, these effects can
also be included in research using these games. Using a multi-player computer game entails using human
subjects. As has been shown by Bazzan and Bordini (2001), the presence of emotions increases the rate of
cooperators. In their article, they have considered the emotions anger, joy, distress and pity, which each
individual agent might display. Given the incentive structure, human players might experience them all.
Counter-Strike is a game based on the Half-Life-Engine by Valve Software, distributed by Sierra Studios.
Since its release in 1998, Half-Life has won several awards, was named PC CD-ROM Game of the Year by
over 50 publications worldwide, and a community of programmers has used the game-engine to create new
modifications, maps and other tools. Counter-Strike is a modification that enhances the multi-player aspect of
Half-Life towards a more team-oriented approach in a setting of terrorism and police strike forces. After its
first beta release in June 1999, Counter-Strike has been one of the most widely played online-game
worldwide in the last years. Players have also formed so-called Clans, some based on themes like only
female players, which compete in tournaments and leagues both national and internationally. Membership
may be subject to prior tests of skill, and is expressed by adding the appropriate initials to the character name
when playing. In this game, there are two types of players/actors, forming two teams, Terrorists and Counter-
Terrorists, which constitute additional actors. There are several maps (or settings) which are played
consecutively. On each map a number of rounds are played. Depending on the map, each team has to achieve
a set goal in the round. The team’s goals are mutually exclusive, therefore ensuring strict conflict of interest
between the teams. Possible goals include for the Counter-Terrorists rescuing a number of hostage,
respectively preventing this for the Terrorists. When the goal is achieved, this team wins the round. Each
round has a time limit, and one team is defined to be winner when this limit is reached without prior goal
achievement by one team. The pre-defined winner is depending on the setting (in the hostage rescue example
above, the Terrorist team wins). Another way to win the game is by eliminating all members of the other
team. The game is played with mouse and keyboard. Mouse movement sets the sight and aiming of the
character, keyboard controls movement, e.g. forwards or backwards, and several other commands. Each
character can carry a limited number of weapons and equipment like bomb defuse kit or nightvision goggles.
Enough hits, depending on the weapon, eliminate a character. This player can then not re-enter the round, but
has to wait until it ends by one team’s success. In this time period, he can watch the proceedings of the other
players from several perspectives, and can chat with the other eliminated players. Each player has a certain
amount of starting money to be spent on weapons, ammunition and equipment in each map, and gains money
depending on his and his team’s success. A certain amount is awarded at the start of each round.
We installed the Half-Life server with Counter-Strike modification. Also employed is a modification
called AdminMod, allowing for additional types of votes and parameters in the game. In addition, the tool
PsychoStats is installed to provide statistics covering several game sessions on the World Wide Web based
on the log files. Both the game itself and the statistics on the WWW are accessible for everyone using the
Internet. Experiences and player opinions show that this server outperforms most others in the vicinity due to
processing power and high bandwidth. Due to the Internet’s limitations on connection speed, most players
come from countries in Central Europe. Game settings include a maximum number of 16 players and an
automatic punishment of one round suspension for team killings. An extension to AdminMod has been
programmed in order to record positional information for each player each second in the log files. This
information was necessary for the following analyses. We have also implemented a program in Java using
these positions to visualize the movements and positions of the players in a round. There are several
advantages of using especially Counter-Strike for the following analyses, foremost the incentive structure.
Counter-Strike allows for several incentives for the players. These include a high-score list within the game,
and when using a free statistics program and log analyzer (e.g. PsychoStats) a website with detailled ranking
according to overall performance with a skill system (see Figure 1), several other categories and an award
system. This adds an element of persistence to the game (Day, 2001), as these statistics span a time interval
which can be customized. In addition, the rules provide incentives for successful behaviour, as death within
2/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
one round only allows subsequent spectation of proceedings until the start of the next round. As this period of
forced inactivity is undesirable, realistic and successful playing will be encouraged.

Figure 1: Screenshot of WWW-statistics ranking players
Additional advantages of Counter-Strike are the virtual reality afforded by the Half-Life game-engine and
the Counter-Strike modification with very good visual and audio impressions, and a fun aspect, which can be
easily seen given the numbers for popularity and usage of this game. If so configured, the game records most
important events in a log file. In addition, APIs are provided for writing additional information into the logs.
We have used these APIs for including positional information in the log files, which are not recorded
automatically. Therefore the success and the behaviour of the players can easily be analysed. As a further
advantage, Counter-Strike offers several possibilities for changing the game and therefore the experiment
setting. These include the selection of available maps, the maximum number of players, and several other
variables of the game like automatic punishment of team killings, necessary number of votes for map
changes or player kicks, and the activation of automatic team balancing. Usage of a statistics program allows
for changing the ranking and award system, and the AdminMod extension allows for additional types of in-
game votes, and also for arbitrary limitations in the available equipment. Counter-Strike features several
communication channels, a text chat facility within the game, and since release 1.3 also a voice channel.
While the positive effects of communication on cooperation and trust are undisputed (Kollock, 1998), also
the modality of communication has been found to influence this effect, with voice communication
performing significantly better than text-to-speech, text chat or no communication (Jensen et al., 2000).
Therefore Counter-Strike will possibly capture this effect and together with higher ease-of-use during game-
play compared to typing text, cooperation is assumed to increase due to the new voice channel. The messages
passed using text chat are recorded in the log files, allowing for analysis of these exchanges between players.
As a last advantage, given the type of game and the incentive structure, emotions are experienced by the
players. This can be verified by both analysis of the text chat messages and listening in to the voice
communication. As has been shown, this presence of emotions increases the ratio of cooperators in an
iterated Prisoner’s Dilemma (Bazzan and Bordini, 2001).
3. COMMUNICATION WITHIN THE GAME
As has been said, Counter-Strike offers a text chat facility for communication. There are four possible
channels for communication, one for all players in the game who are still active in a given round, one for the
players already eliminated, and one for each team to allow for strategy coordination without the other team
listening in. All messages are recorded in the log files. These have been analysed to uncover what this facility
is used for. During a 14-day period, 6783 different words have been used, for a word count of 50494. All 319
words having at least 20 occurrences which total 36053 occurrences have been ordered in six broad
categories. One word could possibly be in several categories, and some words, e.g. ’I’ were not categorised at
all. Table 1 gives the results. As can easily be seen, the expression of emotions, both positive and negative, is
3/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
predominant. This proves the assumption that players experience emotions during play. As is not surprising,
the text chat is also used for communication concerning the game itself, for example to coordinate strategies.
These messages tend to be rather short, as they are used during the game, where time-consuming typing is
rather dangerous should an enemy be encountered. The text chat is also used to communicate about perceived
disruptive behaviour, for example to pledge a mistake or to acknowledge this. Discussions about cheating
behaviour do not seem very extensive. One explanation would be that it can not be unambiguously detected.
The high usage for social interactions like greetings when entering the game or trying to establish out-of-
game contacts, e.g. using instant messaging services like ICQ, seem to indicate a high level of both social
awareness and also persistence, as many players recognize each others from prior contacts on this server or
elsewhere. There are also some commands sent to the server using the text chat facility which are defined as
an own category.
Table 1: Categories of chatwords
CATEGORY EXAMPLES DIFF. WORDS TOTAL
Emotion lol,haha,shit 89 12289
Comm. concerning Game yes,teams,ok 33 5708
Social Interaction hi,cu,thx 14 2859
Comm. conc. Disr. Behaviour sorry,np,kick 14 1665
Game Commands mapvote,timeleft 7 1661
Comm. concerning Cheating kick,admin,noob 11 362
4. COOPERATION WITHIN TEAMS
For the analysis of cooperation within teams, a simplified model is necessary. One game is defined as the
play of one round, ending with the victory of one of the teams. A game consists of the two teams (Terrorists
and Counter-Terrorists), each with a variable number of players. Each of these players has two possible
actions (or moves) available, namely cooperate (C) or defect (D). All players choose their action for the game
simultaneously at the beginning. If only a single game is considered, players have empty information sets.
Each player’s strategy set or strategy space therefore consists of one strategy, playing cooperate (C) or defect
(D) each with a given probability. Playing cooperative (C) is defined using the spatial distance between the
player and his teammates during game-play. When the players stay together, they can help each other and are
much more effective. Players who for example stick behind waiting for the others to take most of the load
might have a chance for bigger payoff, but the team performance might decrease. Figure 2 shows a
screenshot of our visualizer tool, which shows as dots each player’s position in a given moment. In addition,
each player is surrounded by a circle depicting the maximum range for cooperative behaviour. One player is
defined as cooperating with another player if he is within a given radius centered on this player (the
cooperating players are highlighted in the figure). As the distance between the players can change during the
course of one round, a player is defined as playing the action cooperative (C) when is he is cooperating two
thirds of his time in a round. The time span when not enough players in a team remain (less than four) is not
counted. In addition, the first 10 seconds of a round are also discarded, as each team has one starting point,
therefore enforcing cooperative behaviour without conscious choice by the players.
4/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.

Figure 2: Screenshot of Visualizer with cooperating players
Each team also constitutes a separate player of different type in the game. The action set of each team
consists of two actions, namely cooperation within the team (C) or defection within the team (D). The
information set of each team consists of the action profile for all players in the respective team. The game
therefore consists of two stages, with players first choosing their actions, then the teams, having observed
their players’ choices. There is only one strategy in the teams’ strategy set, which is to play cooperation when
two thirds of the team’s players cooperate, or else play defection. Each player gets a payoff, which is the
expected utility as a function of the actions chosen by himself, his team and the opposing team. Therefore we
deal with a 2x2x2 payoff matrix, resulting in a cube. A player’s utility can be derived from the combination
of the number of kills he achieves, the presence or absence of his own death, the time actively having played
(until his death or the end of the game), whether his team was victorious and the degree of action he
experienced. This last component is measured by the number of hits to and by other players per minute
actively played. The preferences might differ between players according to their ranking of these goals.
From the data we have collected from the log files, we have empirically determined for a period of 15
days for all games the actions of each player, the actions of both teams and the resulting payoff for each
player. In the following we consider only one of the possible maps (de_dust), as the results depend on the
setting. During an interval of 15 days, about 6,000 rounds are played by about 1,500 different players. For
analysis, we first consider the aspect of the two teams playing against each other. This constitutes a collapsed
version of the 3-player 2x2x2 game. Table 2 gives the payoff matrix for this game. In this case, the only
payoff component shown is the number of kills achieved, which can be shown to be highly correlated with
the time actively having played and the team’s win. The number of kills shown is the mean for the players in
the team over all observed games. As can be seen, for the team T (Terrorist) playing defection (D) is the
dominant strategy. Given this finding, the team CT (Counter-Terrorist) should choose to play cooperation
(C). Therefore this combination constitutes a Nash equilibrium.
Table 2: Team T vs. Team CT
CT: C CT: D
T: C 0.720;0.662 0.634;0.708
T: D 0.738;0.643 0.682;0.610
Next, we consider the aspect of a Terrorist player playing with his team T (Table 3). This again
constitutes a collapsed version of the 3-player 2x2x2 game, with the enemy (CT) team collapsed using its
empirically determined action distribution. Analogous results for a CT player with his team CT are also
shown (Table 3). Again, the mean number of kills is shown as the only payoff component. As can easily be
seen, for both types of players, playing defect is the dominant strategy. Therefore, there is no conflict
between the Terrorist players and their team, as both would rationally choose to play D. For the other team,
there is a conflict, as it would be better for the team to play C in order to arrive at the Nash equilibrium, but
the individual player has incentive to play defect. We have investigated which actions are indeed taken by the
players and therefore teams. Figure 3 shows the game of a Counter-Terrorist player playing with his team in
extensive form. The edges of this tree are labelled with the action and it’s empirically determined probability.
Only the payoffs for the player are given at the leaf nodes, although this time the action component is also
given together with the number of kills. As can be seen, the players choose to play cooperate with more than
75%. Given the payoff presented above, this seems a counter-intuitive result. There are several possible
explanations for this behaviour: As can be seen from Figure 3, cooperation increases the payoff component
action to a large degree. Therefore incentive exists to adopt this behaviour. The high percentage for
cooperation would then hint at a high preference for action within the population. While we have at this
5/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
moment considered a single game, this game in fact is repeated. These repetitions will sometimes incorporate
the same players, which stay at the server or return later. As the communication shows, the players also do
develop social contacts, even using other channels. Therefore the situation allows for learning effects, for
example which players tend to follow cooperative strategies, and that cooperation might be favourable for the
team overall (Weibull, 1995; Fudenberg and Levine, 1998; Aumann et al., 1995).
Table 3: Player vs. Team
T: C T: D CT: C CT: D
P: C 0.700;0.701 0.668;0.726 P: C 0.647;0.658 0.638;0.687
P: D 0.707;0.701 0.793;0.726 P: D 0.718;0.658 0.741;0.688


Figure 3: Player vs. Counter-Terrorist team in extensive form
5. COOPERATION AND DISRUPTIVE BEHAVIOUR
Disruptive behaviour within the game, i.e. attacking the team mates or cheating using additional or
modified software, is possible. This can pose a problem for the other players (Day, 2001). We analysed how
the other players deal with this behaviour, and what causes it, e.g. a streak of failures or accusations of
cheating. The strategies available to other players are to resort to this behaviour themselves, i.e. attacking this
player, which might lead to more aggression and maybe degeneration of the game (’retaliation’ strategy), to
try to kick the player from the server using a vote, or to leave the server. The success of the ’vote’ strategy is
contingent on the experiment settings, i.e. the number of votes necessary for kicking as a percentage of
players in the game (currently 60% are used), and the behaviour of the other players during the vote. As
attacks on team mates are not visible to members of the opposing team, this might include some intercourse
using the chat facilities to obtain the necessary number of supporters. These messages are also logged and
can therefore be analysed. As has been shown in the respective section, this possibility is indeed used, for
example to pledge a mistake or for excuses. It can be assumed that acquaintances between players and
reputation will play an important role for the ability to convince others. It has indeed been found that initial
face-to-face contact can promote trust in electronic contexts, in which it is otherwise difficult to achieve
(Rocco, 1998). In addition, results have shown that structure in online interactions increases the chance for a
group to reach consensus in contrast to standard chat discussions, which are currently provided by Counter-
Strike (Farnham et al., 2000). This might impair the use of this ’vote’ strategy. In addition to these strategies,
the presence of an administrator, who has the means of kicking any player from the server at his disposal,
changes the situation. Players confronted with disruptive behaviour who know of the administrator’s
presence might do nothing hoping for him to act, or use the chat facility to encourage punishment by him.
Even if no administrator is online, e-mail messages might be sent to him complaining about other player’s
disruptive behaviour and wishing for retribution. Success of both on- and to an even greater extent offline
appeals might depend on acquaintance with the administrator and reputation of the players involved, i.e. the
accuser and the perpetrator. The empirical results retrieved from the log files for players employing both the
’vote’ and ’retaliation’ strategy are strong. The effects of different time spent playing by the players have
been eliminated. There is a significant correlation of 0.101 for the players between being attacked by a team
member and starting a vote to kick someone. Interestingly the correlation between being killed and starting a
vote is not significant. Empirical results seem to indicate that in this case a ’retaliation’ strategy is employed.
The correlation between being attacked and being killed by a team member at 0.587 demonstrates that not all
6/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
team attacks succeed in killing, either because this is not intended or other measures like self-defense prevent
this. Intention to kill might be missing because the attack has been a mistake, or because counter-measures by
other players or the server, which can be configured to enforce one round of inactivity after at teamkill, are
feared when actually following through with the attack. These might be avoided if the attacks are stopped
short of killing. Empirical results for the success of employing the ’vote’ strategy show that there are
significant if low correlations between having a kick vote started against a player, and this player having
attacked (0.115) or killed (0.111) a team member. There is also a relatively low correlation of 0.179 between
having a vote started and actually being kicked due to it, indicating that the votes rarely succeed. Of course,
there is a similar correlation between attacking and killing a team member of 0.615 to being attacked and
being killed. The same comments as given above apply again. Results for the ’retaliation’ strategy clearly
indicate that this strategy is more often employed. The correlations between being attacked or killed by and
attacking or killing a team member are considerably higher than for starting a kick vote, e.g. 0.612 for being
attacked and attacking compared to 0.101 for being attacked and starting a vote. It can also be seen that an
attack is more often answered by an attack than by a kill (0.612 compared to 0.316). In summary, the
empirical results clearly show that both strategies (’vote’ and ’retaliation’) are employed by the players, with
the ’retaliation’ strategy having a considerably higher acceptance. This may in part be explained given the
low success rate of the ’vote’ strategy. Of course, the experiment setting currently at 60% can be changed to
allow a vote to succeed with a smaller percentage of players, possibly altering this strategy’s acceptance. In
addition, retaliation can more easily be accomplished due to the user interface and has a quicker result.
6. APPLICATION TO A REAL WORLD SETTING
The results concerning cooperation and communication in the game are of interest for several fields in
which electronic communication, team-building and trust play an important role. One excellent example is
the development of open source software. In this model for software development, a large number of
voluntary participants cooperate to create and improve a software artefact (Raymond, 1999). The best known
example for this kind of project is the Linux operating system, others include the Apache Web server or the
Perl programming language. These programmers form a virtual organization without central management, as
all participation is strictly voluntary (Dafermos, 2001). There are several explanations for the motivation to
spend time on these projects without direct monetary compensation, including a reputation model, where the
main motivating force is recognition by the peers within the community, also termed ’gift culture’
(Raymond, 1999; Bergquist and Ljungberg, 2001), the sheer joy of hacking (Raymond, 1999; Torvalds and
Diamond, 2001), altruism or investment in the own human capital (Hars and Ou, 2001). Especially the
competition for status and reputation has been mentioned as a possible problem in this type of software
development (Bezroukov, 1999), as individual maximization does not necessarily lead to the successful
progress of the project overall. Therefore the situation faced is similar to the game setup in our experiments.
As the participants most often are geographically dispersed to an extent that face-to-face contact is not
possible, communication is limited to electronic channels, mainly mailing lists and e-mail. In addition, some
projects have also begun to implement voting mechanisms for decision-making (Fielding, 1999). The
resemblances therefore are striking between this setting and our experiment, allowing for transfer of results
from the game.
7. CONCLUSION
In this paper we have reported on the use of a popular multi-player Internet computer game in research on
cooperative behaviour. The game used is Counter-Strike, a free modification of one of the most highly
acclaimed games of the last years, Half-Life. Counter-Strike offers several advantages for our research,
including easy availability of rich data, the emphasis on team-playing, as well as numerous possibilities to
change the experiment settings. In addition, it offers several channels for communication and highly
engaging graphics in the game. This aspect can be seen from usage statistics on the Internet, where Counter-
Strike continues to be one of the most often played games, with users numbering from 40,000 to 60,000 at
any given moment. We have detailed how the communication facilities afforded by the game are used to
facilitate cooperation, to display emotions and to discuss matters like disruptive or cheating behaviour. We
next have presented an analysis of the cooperation in playing the game using descriptive game theory. The
results clearly show that at least in one team the individual maximization of payoff would result in a sub-
7/8

Preprint - To appear in Procs of the IADIS Intl. Conf. Web Based Communities 2004, Lisbon, Portugal, 24.26 March, 2004.
optimal team strategy and therefore performance. On the other hand, the results also show that cooperative
behaviour is nevertheless adopted to a high degree. We have presented several explanations for this. Also
analysed were the ways in which players deal with both disruptive and cheating behaviour. Both types were
defined, and possible counter-strategies for the other players were detailed. Results show that most often a
’retaliation’ strategy is employed against disruptive behaviour, as it is shown to be difficult to punish the
perpetrator using the vote mechanisms provided. We have then detailed one area, the development of open
source software, which shows remarkable resemblance to the experiment setting in the reputation games
which might arise in these virtual communities. Other possible areas in which results from this research could
be transferred would include digital universities, groupware or any other virtual community-building in
which a conflict of goals between the individual and the group might arise.
REFERENCES
Aumann, R.J. et al., 1995. Repeated Games with Incomplete Information. The MIT Press, Cambridge, Mass.
Bazzan A.L.C. and Bordini, R.H., 2001. A framework for the simulation of agents with emotions: Report on experiments
with the iterated prisoner’s dilemma. In Proceedings of the 5th International Conference on Autonomous Agents,
pages 292–299, Montreal, Quebec.
Bergquist. M. and Ljungberg, J., 2001. The power of gifts: Organising social relationships in Open Source communities.
Information Systems Journal, 11(4):305–320.
Bezroukov, N., 1999. A second look at the cathedral and bazaar. First Monday, 4(12).
Chao, D., 2001. Doom as an interface for process management. In Proceedings of the CHI 2001 Conference on Human
Factors in Computing Systems, pages 152–157, Seattle, Washington.
Dafermos, G.N., 2001. Management and virtual decentralised networks: The Linux project. First Monday, 6(11).
Day, G., 2001. Online games: Crafting persistent-state worlds. IEEE Computer, 34(10):111–112.
Farnham, S. et al., 2000. Structured online interaction: Improving the decision-making of small discussion groups. In
Proceedings of the ACM 2000 Conference on CSCW, pages 299–308, Philadelphia, PA.
Fielding, R.T., 1999. Shared leadership in the Apache project. Communications of the ACM, 42(4):42–43.
Fudenberg, D. and Levine, D.K., 1998. The Theory of Learning in Games. The MIT Press, Cambridge, Mass.
Hars, A. and Ou, S., 2001. Working for free? - Motivations for participating in Open Source projects. In Proceedings of
the 34th Hawaii International Conference on System Sciences, Hawaii.
Jensen, C. et al., 2000. The effect of communication modality on cooperation in online environments. In Proceedings of
the CHI 2000 Conference on Human Factors in Computing Systems, pages 470–477, The Hague, Amsterdam.
Kollock, P., 1998. Social dilemmas: The anatomy of cooperation. Annual Review of Sociology, 23:183–214.
Laird, J.E., 2001. Using a computer game to develop advanced AI. IEEE Software, 34(7):70–75.
Nash, J., 1950. The bargaining problem. Econometrica, 18:155–162.
Nash, J., 1950a. Equilibrium points in n-person games. In Proceedings of the National Academy of Sciences, USA,
volume 36, pages 48–49.
Nash, J., 1951. Non-cooperative games. Annals of Mathematics, 54:286–295.
Raymond, E.S., 1999. The Cathedral and the Bazaar. O’Reilly and Associates, Sebastopol, CA.
Rocco, E., 1998. Trust breaks down in electronic contexts but can be repaired by some initial face-to-face contact. In
Proceedings of the 1998 Conference on Human Factors in Computing Systems, pages 496–402, Los Angeles, CA.
Torvalds, L. and Diamond, D., 2001. Just for Fun: The Story of an Accidental Revolutionary. HarperCollins, NY.
Neumann, J.v. and Morgenstern, O., 1944. Theory of Games and Economic Behaviour. Princeton Univ. Press, N. J.
Weibull, J.W., 1995. Evolutionary Game Theory. The MIT Press, Cambridge, Mass.
8/8