PS YC HOLOGICA L SC IENCE
Research Report
Deciding to Defect
The Effects of Video-Game Violence on
Cooperative Behavior
Brad E. Sheese and William G. Graziano
Purdue University
ABSTRACT—This
experiment examined the effect of videogame violence on cooperative decision making. Participants (N 5 48) were randomly assigned to play either a
violent or a nonviolent version of the video game Doomt in
dyads. Following the video-game task, participants were
separated and given an opportunity to choose to cooperate
with their partner for mutual gain, withdraw from the
interaction, or exploit their partner for their own benefit.
Participants in the violent condition were significantly
more likely to choose to exploit their partners than participants in the nonviolent condition. These findings suggest that playing violent video games may undermine
prosocial motivation and promote exploitive behavior in
social interactions.
Recent research has found a positive association between
playing violent video games and aggressive feelings and behaviors (e.g., Anderson et al., in press; Bushman & Anderson,
2002; Sherry, 2001). Recent research has also provided initial
evidence of a negative association between playing violent
video games and prosocial behavior (Anderson & Bushman,
2001; Ballard & Lineberger, 1999; Wiegman & VanSchie,
1998). According to the General Aggression Model (GAM;
Anderson & Bushman, 2002), one way that exposure to violent
media may be linked to both pro- and antisocial behavior is
through the activation of aggression-related schemas that are
stored in memory. The activation of these schemas may alter the
way subsequent information is processed, such that neutral or
ambiguous cues may be interpreted as threatening or aggressive. Consequently, the activation of aggression-related schemas may alter appraisal and decision-making processes,
ultimately leading to more antisocial, and fewer prosocial,
behaviors.
Address correspondence to Brad E. Sheese, Department of Psychological Sciences, Purdue University, 703 Third St., West Lafayette,
IN 47907-2004; e-mail: brad@psych.purdue.edu. DOOM is a registered trademark of Id Software, Inc.
354
This study examined how video-game violence affects deliberate decisions to behave cooperatively or competitively and
expectations about how other individuals will behave. On the
basis of the GAM, we made two hypotheses. The first hypothesis
was that playing a violent video game would cause participants to
anticipate less cooperative and more competitive behavior from
interaction partners. The second hypothesis was that playing a
violent video game would cause participants to choose to compete, rather than cooperate, in subsequent interactions.
We used a modified Prisoner’s Dilemma (PD) game to examine
competitive and cooperative decision making. The PD has been
used extensively in previous research to examine conditions
that promote mutual cooperation (Axelrod & Hamilton, 1981).
For the current study, we used a modified version of the PD (PDALT; Insko, Schopler, Hoyle, Dardis, & Graetz, 1990) that allowed participants to choose among cooperating, defecting, and
withdrawing. In a traditional PD decision matrix, a decision to
defect may reflect either a decision to exploit a partner who is
trusted or a decision to defend oneself from an exploitive interaction partner. In PD-ALT, a ‘‘withdraw’’ choice that presents
an intermediate level of reward is available, making it possible
to distinguish between participants who trust their partners but
want to exploit them and participants who do not trust their
partners. The matrix is designed so that if participants do not
trust their partners, they should choose the option labeled ‘‘Y’’
(the withdrawal option). If participants do trust their partners
and want to cooperate, they should choose the option labeled
‘‘X’’ (the cooperation option). If participants trust their partners
but want to exploit them, they should choose the option labeled
‘‘Z’’ (the defection option).
The PD-ALT decision matrix allowed us to examine how
playing violent video games affected trust and decisions to
defect separately. Additionally, prior to making a decision for
themselves, participants were asked to anticipate their partner’s
decision. Examining the participants’ predictions and decisions
separately provided a direct method for testing the hypothesis
that violence should affect the participants’ perceptions of other
individuals’ intentions.
Copyright r 2005 American Psychological Society
Volume 16—Number 5
Brad E. Sheese and William G. Graziano
On the basis of the PD-ALT framework and our predictions
derived from the GAM, we anticipated the following pattern of
outcomes. First, we hypothesized that, relative to participants
who played a nonviolent video game, those who played a violent
video game would be significantly more likely to anticipate that
their partner would either withdraw or defect. Second, we hypothesized that participants would be significantly more likely
to choose to defect if they played a violent video game than if
they played a nonviolent game.
To test these hypotheses we developed a specially modified
version of the video game Doomt. In the nonviolent version of
the game, participants were asked to find the end of a threedimensional maze. In the violent version of the game, participants were asked to find the end of the same three-dimensional
maze, but weapons and simple computer-controlled opponents
were also introduced. The opponents, various kinds of fictional
monsters, were easy to kill with the available weapons and did
not present a significant challenge to the participants. They did,
however, present players with an opportunity to shoot and kill
opponents while in the process of finding their way through the
maze.
Teams were initially trained on how to play the video game
and were familiarized with the rules and objectives. They were
then given 25 min to complete as many mazes as possible.
After 25 min had elapsed, the participants were separated.
They were then presented with the PD-ALT decision matrix,
which showed three choices labeled ‘‘X,’’ ‘‘Y,’’ and ‘‘Z.’’ They
were told that the total number of points they had earned would
be multiplied by a fixed amount, determined by their decision
and their partner’s decision. If both partners decided to pick X
(cooperate), both partners’ scores would be multiplied by a
moderate amount (1.5). If both partners picked Z (defect), then
both partners would lose half their points (i.e., their scores
would be multiplied by 0.5). However, if one partner chose to
cooperate and the other chose to defect, the defecting partner’s
score would be multiplied by a large amount (2), whereas the
cooperating partner would lose half of his or her points. Finally,
if either partner picked Y (withdraw), both partners’ scores
would remain the same (i.e., their scores would be multiplied
by 1).
After successfully completing a brief quiz, to show they understood the instructions and the decision matrix, participants
were asked to anticipate what choice their partner would make.
Finally, they were asked to make the decision for themselves.
METHOD
RESULTS
Participants
Undergraduate students (N 5 48; 6 women) participated in
pairs. Participants were primarily (80%) non-Hispanic whites.
Procedure
Two participants, who were unfamiliar with one another, were
brought into a lab setting and randomly assigned to be either
‘‘Player A’’ or ‘‘Player B.’’ They were told that they would be
competing with all of the other study participants who had the
same designation, but that they were not competing with each
other. It was explained that when the study was completed, both
the best-performing Player A and the best-performing Player B
would be awarded $100.
Teams were then asked to play a multiplayer video game,
Doomt, in which they would complete a series of three-dimensional mazes. They were told that they would earn points for
every maze they completed within 25 min. Both players had to
reach the end of a given maze before the pair could advance to
the next maze.
Teams were randomly assigned to play either a violent or a
nonviolent version of Doomt. In the two conditions, the mazes
and the goal of the game were identical. However, in the violent
condition, both players were provided with weapons, and the
mazes included opponents that would attack the two players. In
the nonviolent condition, players had no weapons, and there
were no opponents within the mazes.
Volume 16—Number 5
The participants’ predictions of their partners’ choices were
used to calculate two outcome variables. Predicted trust, indicating whether participants anticipated that their partners
would trust them and not withdraw, was scored as follows.
Participants who anticipated that their partners would pick
either Choice X (cooperate) or Choice Z (defect) received a
score of 1 for predicted trust; participants who anticipated that
their partners would pick Choice Y (withdraw) received a score
of 0. Of the 48 participants, 42 anticipated that their partners
would trust them (21 in the violent condition, 21 in the nonviolent condition). Predicted defection, indicating whether the
participants anticipated that their partners would choose to
defect, was scored as follows. Participants who anticipated that
their partners would pick Choice Z (defect) received a score of
1; participants who anticipated that their partners would pick
Choice X (cooperate) or Choice Y (withdraw) received a score of
0. Of the 48 participants, 3 anticipated that their partners would
choose to defect (2 in the violent condition, 1 in the nonviolent
condition).
The participants’ own decisions using the PD-ALT matrix
were also used to calculate two outcome variables. The trust
score, reflecting whether the participant trusted his or her
partner and did not withdraw, was 1 for participants who picked
either Choice X (cooperate) or Choice Z (defect) and 0 for
participants who picked Choice Y (withdraw). Of the 48 participants, 43 chose to trust their partner and not withdraw (23 in
the violent condition, 20 in the nonviolent condition). The
355
Video-Game Violence
defection score, reflecting whether or not the participant chose
to defect, was 1 for participants who picked Choice Z (defect)
and 0 for participants who picked Choice X (cooperate) or
Choice Y (withdraw). Of the 48 participants, 8 chose to defect (7
in the violent condition, 1 in the nonviolent condition).
Nonindependence was assumed, because there was an insufficient sample size to test the independence of the participants on each of the dependent variables with a power equal to
or greater than .80 (Kenny, Kashy, & Bolger, 1998). Multilevel
modeling (Raudenbush & Bryk, 2002), conducted with HLM
(Version 5.05; Raudenbush, Bryk, Cheong, & Congdon, 2000),
was used to assess the effect of the manipulation of video-game
violence on each of the four dependent variables. Nonlinear
models were specified because the dependent variables were
binary. Models were specified such that individual participants
were nested within pairs. The condition (violent or nonviolent)
was dummy-coded and modeled as a fixed effect.
There was no evidence to indicate that participants assigned
to the violent condition (M 5 .86, SD 5 .34) were less likely
than participants assigned to the nonviolent condition (M 5 .86,
SD 5 .34) to anticipate that their partner would trust them, B 5
0.00, t(46) 5 0.00, p 5 1.00, d 5 0.00. There was also no
evidence to indicate that participants assigned to the violent
condition (M 5 .04, SD 5 .20) were more likely than participants assigned to the nonviolent condition (M 5 .08, SD 5 .06)
to anticipate that their partner would defect, B 5 0.74, t(46) 5
0.63, p 5 .53, d 5 0.19. Overall, there was no indication that
playing the violent version of the video game caused participants to anticipate less cooperation or prosocial behavior from
their partners.
A different pattern emerged in the participants’ own choices.
There was no evidence to indicate that participants assigned to
the violent condition (M 5 .96, SD 5 .20) were more likely than
participants assigned to the nonviolent condition (M 5 .83, SD
5 .08) to not trust their partner and choose to withdraw, B 5
1.53, t(46) 5 1.73, p 5 .18, d 5 0.51. However, participants
assigned to the violent condition (M 5 .29, SD 5 .46) were
significantly more likely than participants assigned to the
nonviolent condition (M 5 .04, SD 5 .20) to choose to defect,
B 5 2.24, t(46) 5 2.23, p 5 .03, d 5 0.66.
CONCLUSIONS
The current study is consistent with previous research indicating that playing violent video games may have negative effects on social behavior. The current study also extends
previous research by showing how playing violent video games
may affect subsequent decisions to cooperate or compete with
other individuals in high-stakes situations. More generally, the
current findings suggest that playing violent video games may
undermine prosocial and altruistic motivation and promote
competitive behavior in deliberate decision making.
356
Playing a violent video game appears to have contributed
directly to participants’ willingness to exploit their interaction
partners. However, contrary to predictions, there was no evidence that playing a violent video game altered participants’
perceptions of what their partners would do. Participants in the
violent condition were not any more likely than those in the
nonviolent condition to think that their partners would exploit
them or distrust them. Instead, participants in both conditions
generally expected that their partners would trust them and
cooperate with them. However, participants in the violent
condition were more likely than those in the nonviolent condition to exploit that trust and choose to defect. This pattern of
results suggests that playing a violent video game may have an
effect on cooperative behavior independent of alterations in the
perceptions of other people’s motives.
Anderson and Morrow (1995) suggested that the activation of
competitive knowledge structures can lead directly to aggression. Their findings were consistent with this hypothesis: In
comparison with priming cooperation, priming competition led
to more aggressive behavior in a subsequent video game, but did
not affect friendliness, liking, or hostility toward an interaction
partner. Our results suggest that the opposite may also be true,
that participating in violent game play may encourage people to
compete. In the current study, the addition of violent game play
into the cooperative maze task may have served to prime
competitive knowledge structures that were applied to the
subsequent PD-ALT interaction.
If this interpretation is correct, then the current findings are
remarkable considering that, in both conditions, participants
played a cooperative video game for 25 min before making their
decision. It could be plausibly argued that cooperatively engaging in violent behavior against a common enemy would actually increase feelings of cohesion and camaraderie and
promote subsequent cooperation (e.g., Deutsch, 1993; Sherif &
Sherif, 1953). However, the current results indicate that engaging in violent game play prompts the opposite pattern of
behavior. Violence, even when directed at third parties, appears
to be effective at undermining the emergence of cooperative
patterns of interaction.
The results presented here must be considered in light of
several limitations. First, the sample size of the current study
was insufficient for reliably detecting effects that are small or
even moderate in size. Consequently, caution is warranted in
interpreting the null effects reported. A second limitation is that
the findings may have limited generalizability because the study
used a convenience sample of undergraduate students who were
predominantly White males. Furthermore, given that the study
was conducted using dyads only, it is not clear if the effect
would generalize to single-player situations. Social interaction
during the violent game play may have been an important factor
in subsequent decisions to defect. A third limitation is the molar
nature of the manipulation. The addition of weapons and
monsters altered the basic maze task in a variety of ways.
Volume 16—Number 5
Brad E. Sheese and William G. Graziano
Consequently, additional research will be needed to determine
if it was the full manipulation or just a specific aspect of it that
was responsible for the observed effect.
Despite these limitations, the pattern of results is provocative
and worthy of additional study. Unlike in many of the published
studies in this area, participants in the current study were not
asked either to aggress against or to punish other people. Instead, they were given an opportunity to exploit their interaction
partners to earn a substantial sum of money for themselves. The
current study shows clear evidence that engaging in violent
game play makes people more likely to deliberately choose to
exploit their partners in this situation. Future research should
focus on examining further the psychological mechanisms, such
as aggressive cognitions, hostile affect, and increased physiological arousal (e.g., Anderson & Bushman, 2002), that may
account for this finding.
Acknowledgments—The authors thank Joshua Chartier and
Joey Mignogna for their invaluable assistance in conducting
this research. We also thank Jeremy Biesanz, Thomas Berndt,
and George Hollich for their thoughtful suggestions. This research was supported by Grant R01-MH50069 from the National Institute of Mental Health to William Graziano.
REFERENCES
Anderson, C.A., & Bushman, B.J. (2001). Effects of violent video
games on aggressive behavior, aggressive cognition, aggressive
affect, physiological arousal, and prosocial behavior: A metaanalytic review of the scientific literature. Psychological Science,
12, 353–359.
Anderson, C.A., & Bushman, B.J. (2002). Human aggression. Annual
Review of Psychology, 53, 27–51.
Anderson, C.A., Carnagey, N.L., Flanagan, M., Benjamin, A.J., Eubanks, J., & Valentine, J.C. (in press). Violent video games:
Volume 16—Number 5
Specific effects of violent content on aggressive thoughts and
behavior. Advances in Experimental Social Psychology.
Anderson, C.A., & Morrow, M. (1995). Competitive aggression without
interaction: Effects of competitive versus cooperative instructions
on aggressive behavior in video games. Personality and Social
Psychology Bulletin, 21, 1020–1030.
Axelrod, R., & Hamilton, W.D. (1981). The evolution of cooperation.
Science, 211, 1390–1396.
Ballard, M.E., & Lineberger, R. (1999). Video game violence and
confederate gender: Effects on reward and punishment given by
college males. Sex Roles, 41, 541–558.
Bushman, B.J., & Anderson, C.A. (2002). Violent video games and
hostile expectations: A test of the general aggression model.
Personality and Social Psychology Bulletin, 28, 1679–1686.
Deutsch, M. (1993). Educating for a peaceful world. American Psychologist, 48, 510–517.
Insko, C.A., Schopler, J., Hoyle, R., Dardis, G., & Graetz, K. (1990).
Individual-group discontinuity as a function of fear and greed.
Journal of Personality and Social Psychology, 58, 68–79.
Kenny, D.A., Kashy, D.A., & Bolger, N. (1998). Data analysis in social
psychology. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), The
handbook of social psychology (4th ed., pp. 233–265). New York:
McGraw-Hill.
Raudenbush, S.W., & Bryk, A.S. (2002). Hierarchical linear models:
Applications and data analysis methods (2nd ed.). Newbury Park,
CA: Sage.
Raudenbush, S.W., Bryk, A.S., Cheong, Y.F., & Congdon, R. (2000).
HLM 5: Hierarchical linear and nonlinear modeling. Lincolnwood,
IL: Scientific Software International.
Sherif, M., & Sherif, C.W. (1953). Groups in harmony and tension. New
York: Harper.
Sherry, J.L. (2001). The effects of violent video games on aggression: A
meta-analysis. Human Communication Research, 27, 409–431.
Wiegman, O., & VanSchie, E.G.M. (1998). Video game playing and its
relations with aggressive and prosocial behavior. British Journal
of Social Psychology, 37, 367–378.
(RECEIVED 6/14/04; REVISION ACCEPTED 9/23/04)
357