ArticlePDF Available

Abstract and Figures

Three experimental studies, one correlational study, and a meta-analysis tested key hypotheses concerning the short-term and long-term impact of exposure to violent video games. Experiment 1 found that violent video games in general increase the accessibility of aggressive thoughts. Experiments 2 and 3 found that playing violent video games increased aggression, even when arousal and affect were controlled. Experiments 2 and 3 also found that trait hostility and trait aggression were positively related to laboratory aggression. Furthermore, there was correlational evidence of a link between repeated exposure to violent video games and trait aggressiveness. Mediational analyses suggested that the trait effects and the violent video game effects on laboratory aggression were partially mediated by revenge motivation. The correlational study uncovered links between habitual exposure to violent video games, persistent aggressive cognitions, and self-reported aggressive behavior. A destructive testing regression approach found that the video game violence/aggression link remained significant even after stressing the link by partialling out sex, narcissism, emotional susceptibility, and Big Five personality factors. However, consistent with prior empirical and theoretical work emphasizing the importance of media violence in the creation of habitual aggressive patterns of thought, partialling out aggressive attitudes reduced the video game violent/aggression link to nonsignificance. The meta-analyses revealed significant effects of violent video games on aggressive behavior, affect, and cognition; on cardiovascular arousal; and on prosocial behavior. A best-practices meta-analytic approach revealed that contrary to media industry claims, better conducted studies tend to yield stronger effects of violent video games on aggression and aggression-related variables than do more poorly conducted studies.
Content may be subject to copyright.
In creating this electronic reprint, we have
attempted to keep the style, pagination,
and format as close to the published form
as possible. Nonetheless, some errors may
have occurred. If you discover an error,
please contact Craig Anderson using the
following email address: caa@iastate.edu.
This electronic reprint is provided as a
courtesy. Please do not post or
distribute this reprint in any fashion that
may violate the copyright of the original
publisher or the authors. Thank you for
your interest in this work.
VIOLENT VIDEO GAMES: SPECIFIC
EFFECTS OF VIOLENT CONTENT
ON AGGRESSIVE THOUGHTS
AND BEHAVIOR
Craig A. Anderson
Nicholas L. Carnagey
Mindy Flanagan
Arlin J. Benjamin, Jr.
Janie Eubanks
Jeffery C. Valentine
Three experimental studies, one correlational study, and a meta-analysis
tested key hypotheses concerning the short-term and long-term impact of
exposure to violent video games. Experiment 1 found that violent video games
in general increase the accessibility of aggressive thoughts. Experiments 2 and 3
found that playing violent video games increased aggression, even when arousal
and aVect were controlled. Experiments 2 and 3 also found that trait hostility
and trait aggression were positively related to laboratory aggression. Further-
more, there was correlational evidence of a link between repeated exposure to
violent video games and trait aggressiveness. Mediational analyses suggested
that the trait eVects and the violent video game eVects on laboratory aggression
were partially mediated by revenge motivation. The correlational study uncov-
ered links between habitual exposure to violent video games, persistent aggres-
sive cognitions, and self-reported aggressive behavior. A destructive testing
regression approach found that the video game violence/aggression link re-
mained significant even after stressing the link by partialling out sex, narcissism,
emotional susceptibility, and Big Five personality factors. However, consistent
with prior empirical and theoretical work emphasizing the importance of media
violence in the creation of habitual aggressive patterns of thought, partialling
out aggressive attitudes reduced the video game violent/aggression link to
nonsignificance. The meta-analyses revealed significant eVects of violent video
199
ADVANCES IN EXPERIMENTAL Copyright 2004, Elsevier Inc.
SOCIAL PSYCHOLOGY, VOL. 36 All rights reserved.
0065-2601/04 $35.00
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:01 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
games on aggressive behavior, aVect, and cognition; on cardiovascular arousal;
and on prosocial behavior. A best-practices meta-analytic approach revealed
that contrary to media industry claims, better conducted studies tend to yield
stronger eVects of violent video games on aggression and aggression-related
variables than do more poorly conducted studies.
I. Introduction
A. CRITICAL INCIDENTS
Violent video games are popular with adolescent and adult males and
females, are marketed to youth in ways that violate the video game industrys
own standards, and are easily obtained regardless of age (e.g., Buchman &
Funk, 1996; Federal Trade Commission, 2000; Walsh, 1999). School shoot-
ings by boys with a history of playing violent video games [e.g., West Paducah,
KY (December 1997); Jonesboro, AR (March 1998); Springeld, OR (May
1998), Littleton, CO (April 1999), Santee, CA (March 2001), Wellsboro, PA
(June 2003) and Red Lion, PA (April 2003)] heightened public debate about
the role played by this relatively new violent entertainment medium, including
a hearing by the U.S. Senate Committee on Commerce, Science, and Trans-
portation held on March 21, 2000. Other recent violent crimes linked to
violent video games include a violent crime spree in Oakland, California
(January 2003); ve homicides in Long Prairie and Minneapolis, Minnesota
(May 2003); beating deaths in Medina, Ohio (November 2002) and Wyoming,
Michigan (November 2002); and the Washington, D.C. ‘‘Beltway’’ sniper
shootings (Fall 2002). However, such public incidents and outcries do not
constitute scientic evidence of a true causal link. This chapter explores
current work by media violence researchers and presents ve new studies.
B. BRIEF HISTORY OF VIOLENT VIDEO GAMES
Video games rst emerged in the late 1970s, but in the 1990s violent games
came of age, with the rst-person shooter ‘‘Wolfenstein 3D’’ and the third-
person ghter ‘‘Mortal Kombat’’ leading the way. By the end of the 20th
century, even more graphically violent games were available to virtually
anyone who wanted to play them, regardless of age (Walsh, 1999). As early
as the mid-1990s, fourth grade girls reported playing video games more than
5 1/2 hours a week, and boys reported playing more than 9 hours a week
(Buchman & Funk, 1996). Furthermore, this same sample of fourth graders
reported that the majority of their favorite games were violent ones (58.9%
200 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:01 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QCby:CelestineJ.
for girls, 72.9% for boys). A survey of eighth and ninth grade students found
boys playing about 13 hours a week and girls about 5 hours a week (Gentile,
Lynch, Linder, & Walsh, 2004).
Data from the Cooperative Institutional Research Program (CIRP, 1998,
1999), which surveys entering college freshmen from more than 600 two- and
four-year colleges, reveal that older students also are playing a lot of
video games and that their time with such games is also increasing. In 1998
13.3% of the young men reported playing video games at least 6 hours per
week during their senior year in high school. By 1999 that gure had increased
to 14.8%. Increases are also occurring at the high end of the game playing
distribution. In 1998, 2% of the young men reported playing video games
more than 20 hours per week. By 1999, that gure had increased to 2.5%.
Another troubling aspect involves the lack of parental or societal oversight. A
recent survey of teens in grades 8 through 12 (Walsh, 2000) found that 90% of
their parents never check the ratings of video games before allowing a purchase,
and only 1% reported that their parents had ever kept them from getting a game
based on itsrating. Furthermore,ratings providedby the video game industrydo
not match those provided by other adults and game-playing youngsters. Spe-
cically, many games involving violenceby cartoon-like characters are classied
by the industry as being appropriate for general audiences, a classication with
which adults and youngsters disagree (Funk, Flores, Buchman, & Germann,
1999). Also,89% of the teens in Walshss urvey (2000) reported that their parents
never limit the amount of time they are allowed to play video games. Finally,
manyofthe most violent games have ‘‘demo’’ versionson theInternetthatcan be
downloaded for free by anyone. Of the boys in the sample who play video
games, 32% reported downloading them from the Internet (Walsh, 2000).
C. MEDIA VIOLENCE RESEARCH
Concern overvideo game violence would be misplaced if playing such games
had little impact on aggression. Decades of research have revealed that viewing
television and movie violence can cause short-term increases in aggression and
long-term changes in trait aggressiveness (e.g., Bushman & Anderson, 2001;
Bushman & Huesmann, 2001; Hearold, 1986; Huesmann & Miller, 1994; Paik
& Comstock, 1994; Wood, Wong, & Chachere, 1991). The research literature
on video games is smaller and less complete. Despite its relatively small size and
the methodological diYculties inherent in the rst studies of any ‘‘new’’ phe-
nomenon, a consensus is emerging that violent video games can cause increases
in aggressive behavior in children and in young adults (Anderson, 2000;
Anderson & Bushman, 2001; Dill & Dill, 1998; Sherry, 2001; Walsh, 2000).
We review this literature after considering theoretical issues.
VIOLENT VIDEO GAMES 201
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:01 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
II. The General Aggression Model
Good theoretical reasons support the belief that exposure to violent video
games will increase aggressive behavior (Anderson & Dill, 2000; Dill & Dill,
1998). The General Aggression Model (GAM) integrates existing theory and
data concerning the learning, development, instigation, and expression of
human aggression (Anderson & Bushman, 2002; Anderson & Carnagey,
2004; Anderson & Huesmann, 2003). It does so by noting that the enactment
of aggression is based largely on knowledge structures, such as scripts or
schemas, created by social learning processes.
In brief, GAM describes a multistage process by which two kinds of input
variables lead to aggressive (or nonaggressive) behavior. Figure 1 shows a
simplied version of the single-episode portion of GAM. Both personologi-
cal (e.g., trait hostility) and situational (e.g., recent violent video game play)
variables inuence behavior by aVecting the persons present internal state,
represented by cognitive, aVective, and arousal variables. Playing a violent
video game may inuence aggression by means of the cognitive route, for
example, if it primes aggressive thoughts or scripts, leading to hostile per-
ception, expectation, and attributional biases (e.g., Bushman & Anderson,
2002; Calvert & Tan, 1994; Crick & Dodge, 1994; Dill, Anderson, Anderson,
& Deuser, 1997; Kirsh, 1998). The three aspects of present internal state are
themselves interrelated, as indicated by the dashed lines connecting them.
For example, priming aggressive thoughts might subsequently increase feel-
ings of anger and a desire for revenge if the person is provoked. In such a
Fig. 1. The general aggression model: single episode cycle. Source: Anderson and Bushman,
2002.
202 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:01 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
case the cognitive eVect is regarded as the primary route of impact and the
aVective eVect as a secondary route of impact.
The present internal state inuences how a person perceives events, inter-
prets their meaning, and chooses behavioral responses. Whatever action the
person takes (e.g., aggressive or nonaggressive) inuences the present social
encounter. This sets the stage for the next round in the social interaction cycle.
Of particular relevance to the present article is the fact that nding signi-
cant links between exposure to violent video games and aggression does not
by itself reveal the primary route of the obtained eVect; it could have
occurred via cognition, aVect, or arousal, or some combination. However,
such ne distinctions are crucial to theoretical development and to the public
policy debate over whether parents should be given tools to help them
control their childrens access to violent games. In both the theory and the
public policy domains, the precise route(s) of primary impact are important
because of the developmental aspects of GAM.
Long-term eVects of media violence involve learning processes, such as
learning how to perceive, interpret, judge, and respond to events in the
physical and social environment. Various types of knowledge structures
(e.g., perception, interpretation, judgment, and action) develop over time
and are based on day-to-day observations of and interactions with other
people, real (e.g., family) and imagined (e.g., media). Each violent episode, as
outlined in Fig. 1, is essentially one more learning trial. Short-term eVects
become ingrained through the development of aggression-related knowledge
structures, which persistently color the persons expectations and perceptions
concerning social interactions, especially those with conictual content.
In a very real sense, a persons set of chronically accessible knowledge
structures denes that personspersonality. Figure 2 displays this develop-
mental aspect of GAM and identies ve types of variables that contribute
to the development of an aggressive personalityaggressive beliefs and
attitudes, aggressive perceptual schemata, aggressive expectation schemata,
aggressive behavior scripts, and aggression desensitization. Four of these
variables involve aggressive cognitions. For this reason, short-term eVects of
violent media on aggression via the cognitive route are particularly impor-
tant. Temporary mood states and arousal dissipate over time, but rehearsal
of aggressive cognitions can lead to long-term changes in multiple aspects of
aggressive personality. Furthermore, the literature on the development of
behavioral scripts suggests that even a few rehearsals can change a persons
expectations and intentions involving important social behaviors (Anderson,
1983; Anderson & Godfrey, 1987; Marsh, Hicks, & Bink, 1998). Figure 3
illustrates the dynamic aspects of the episodic and developmental portions
of GAM. Exposure to violent video games can serve as a proximate situa-
tional cause, increasing the likelihood that an aggressive behavior will occur
VIOLENT VIDEO GAMES 203
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:01 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
shortly after the exposure, but it also can serve as a distal environmental
modier, inuencing the development of aggression-related knowledge
structures and hence, aggressive personality.
III. GAM and Violent Video Games
A. BASIC ISSUES
Two levels of questions emerge from consideration of GAM and violent
video games. First, can violent video games cause increases in aggression?
Answering this question does not require a careful analysis of the multiple
processes by which input variables can inuence the expression of aggressive
behavior. It merely requires a body of research in which the eVects of violent
games are compared to nonviolent games or other appropriate control
Fig. 2. The general aggression model: developmental/personality processes. Source:
Anderson and Carnagey, 2004.
204 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
conditions. Much of the existing video game literature is of exactly this nature,
and it is at this level that a review of the literature reveals considerable support
for the hypothesis that playing violent video games can increase aggression
(Anderson & Bushman, 2001). Both experimental and correlational studies, on
average, yield signicant positive relations between exposure to violent video
games and aggressive behavior, with average eVect sizes in the rþ¼0.20
range (Anderson & Bushman, 2001). The experimental studies demonstrate
that a brief exposure to violent video games causes an immediate (and
presumably short-lived) increase in aggressive behavior. The correlational
studies link repeated exposure to violent video games with a variety of types
of real-world aggressive behavior, including violent criminal behavior. In
sum, despite its relatively small size and recent history, the research literature
has demonstrated that violent video games can increase aggression.
The second level of questions to emerge concerns specicity of violent
content eVects on aggression via the cognitive route. Nonviolent games can
Fig. 3. The general aggression model: overall view. Source: Anderson and Carnagey, 2004.
VIOLENT VIDEO GAMES 205
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
also increase aggressive feelings if, for example, they produce high levels
of frustration. They also can increase arousal, if they are suYciently demand-
ing and engaging. Thus, both violent and nonviolent games may inuence
aggression in the immediate situation if they increase aggressive feelings or
arousal. The real crux of the debate over eVects of violent video games lies in
their unique ability to directly increase aggressive cognitions, cognitions that
can have a much longer lasting impact if their repeated instantiation (by
repeated violent video game play, for instance) leads to persistent changes in
key knowledge structures, such as more positive attitudes toward aggression.
B. KEY QUESTIONS
Three key second-level questions remain unanswered by existing research on
violent video games. First, do violent video games generally increase aggres-
sive cognitions? Several studies have found signicant increases in aggressive
thoughts as a function of exposure to violent video games (see Anderson &
Bushman, 2001), but most have not explicitly controlled for other potential
diVerences between the target video games, such as diVerences in aVective or
arousal properties. In fact, only one published experimental study has suc-
cessfully controlled for both arousal and aVective features (Anderson & Dill,
2000, Study 2); the violent game yielded higher aggressive cognition scores.
Second, does violent video game content by itself cause short-term in-
creases in aggressive behavior tendencies? As GAM makes clear, to cleanly
test this specic violent content question in an experimental setting the
comparison violent and nonviolent video games should be equated on arousal
level and aVective factors such as enjoyment, frustration, and state hostility.
Such controls can be done by selecting violent and nonviolent games that do
not diVer on these factors, or by including measures of arousal and aVect and
using them as statistical controls. Only two published experimental studies
meet these criteria. Graybill, Strawniak, Hunter, and OLeary (1987) pre-
tested several video games and equated them on diYculty, excitement, and
enjoyment. Interestingly, this is one of the few experimental studies that failed
to nd a signicant eVect of video game violence on aggressive behavior. One
potential problem with this study is that the pretest ratings of the various
games were done by graduate students, whereas the participant population
was second through sixth graders. In other words, the ‘‘equating’’ process was
somewhat less than optimal. The other experimental study meeting these
control criteria was Study 2 of Anderson and Dill (2000), which found a
signicant increase in aggression attributable to diVerential violent content.
Third, is repeated exposure to violent video games associated with higher
aggression levels, and is this mediated by persistently elevated levels of
206 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
aggressive thoughts? To show such an eVect, signicant positive associations
between violent video game exposure, aggressive behavior, and persistent
aggressive thoughts would have to be found. Furthermore, the link between
video game violence exposure and aggression must be signicantly weakened
when the persistent aggressive thoughts measure is statistically controlled.
As of this writing, there are no published correlational studies of this type.
IV. Overview of the Present Studies
Experiment 1 used 10 video games to examine the eVects of violent content
on the accessibility of aggressive thoughts, physiological arousal, and aggres-
sive aVect. The results of Experiment 1 were used to select a pair of violent
and nonviolent video games matched on arousal and aVective dimensions
but diVering in violent content for use in Experiments 2 and 3. Experiment
2 tested whether these two games produce diVerent levels of short-term
aggressive behavior. Experiment 3 used two violent and two nonviolent
games and a diVerent aggression paradigm in an attempt to replicate the
specic violent video game content eVect on aggression. It also provided our
rst test of one factor that might increase or decrease the violent video game
eVect, specically the realism of the video game targets of aggression.
Experiments 2 and 3 also examined trait hostility and revenge motivation
eVects on aggression. Correlation Study 1 assessed violent video game expo-
sure, self-reported aggressive behavior, Big ve personality factors, and
attitudes toward violence. It tested a basic-personality-as-artifact hypothesis
as well as an aggressive-cognition-mediation hypothesis. The nal new study
was an updated meta-analysis of violent video game eVects on aggressive
behavior, thoughts, and aVect; physiological arousal; and prosocial behav-
ior. This analysis also compared average eVect sizes of the methodologically
best studies to those with signicant weaknesses.
V. Experiment 1
A. METHOD
1. Participants
Participants were 61 male and 69 female undergraduate students who
participated in partial fulllment of an introductory psychology research
requirement. Participants were asked to refrain from alcohol, caVeine,
VIOLENT VIDEO GAMES 207
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
tobacco products, and exercise for 12 hours prior to the study start time.
Participants were randomly assigned to play one of 10 video games.
2. Materials
a. Video Games. Ten video games were selected through a review of video
game sites on the World Wide Web, popular magazines, and retail outlets.
The ve violent games were: Dark Forces, Marathon 2, Speed Demon, Street
Fighter, and Wolfenstein 3-D. The ve nonviolent games were: 3-D Ultra
Pinball, Glider Pro, Indy Car II, Jewel Box, and Myst (see the Appendix for
a description of the 10 games). We attempted to nd games that did
not require extensive practice to achieve at least marginal prociency
and that were relatively involving for a college student population. Street
Fighter was played on the Super Nintendo Entertainment System, using a
19-inch Sony color television. All other games were played on a Macintosh
computer.
b. Video Game Experience. Participants estimated the average number of
hours per week spent playing video games in the past few months. Each
participant also rated his or her experience with the 10 video games used in
this study and an additional four games (SimCity, You Dont Know Jack,
Computer Cribbage, A10 Attack) using a 7-point unipolar scale anchored at
1 (Never have played), 4 (Have played some), and 7 (Have played often).
c. Video Game Ratings. After playing the assigned video game, partici-
pants completed a 6-item rating scale about the game, rating how diYcult
the game was to learn, how enjoyable the game was to play, how much action
the game had, how violent the game graphics were, how violent the game
content was, and how frustrating the game was (Anderson & Ford, 1986). For
all items, a response of 1 indicated ‘‘low’’ and a response of 7 indicated ‘‘high’’
on the adjective of interest. Preliminary analyses revealed that the two items
measuring the violence of video game content and graphics were highly
correlated, so they were averaged to form a composite measure of perceived
video game violence.
d. Word Completion Task. The word completion task (Anderson,
Carnagey, & Eubanks, 2003; Roediger, Weldon, Stadler, & Riegler, 1992)
involves examining a list of 98 words with one or more letters missing, and
lling in the missing letters. The missing letters are strategic, such that each
item can make more than one word. For instance, one item is ‘‘explo_e,’’
which may be completed as ‘‘explore’’ or ‘‘explode.’’ Participants were told
that their task was to ll in the blanks to make complete words. Participants
were given 3 minutes to complete as much of the task as they could. An
accessibility of aggressive thoughts score was calculated for each participant
by dividing the number of aggressive word completions by the total number
208 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
of completions. Forty-nine of the items can yield an aggressive word when
completed.
e. Cardiovascular Measures. Heart rate (HR), systolic blood pressure
(SBP), and diastolic blood pressure (DBP) were measured with an A & D
Medical automatic digital blood pressure monitor (model UA-751). The
blood pressure cuVwas attached to each participants nondominant upper
arm, approximately 1 inch above the elbow. All measurements were ob-
tained while the participants were seated. At each of three measurement
periods, HR, SBP, and DBP were collected twice, with approximately 1
minute elapsing between the end of the rst measurement and the beginning
of the second. The rst measurement period (baseline) was after signing the
consent form but before game play began. The second measurement was
during video game play, and the third was after the video game task.
3. Procedure
Participants were told that the study concerned the ways people learn
diVerent types of computer tasks. They were informed that we were interest-
ed in possible diVerences between simple and complex tasks. After complet-
ing consent procedures, participants entered a cubicle and completed the
Video Game Experience questionnaire. Participants then read directions for
the assigned game. The experimenter started the game for the participants.
Participants played for approximately 20 minutes. Physiological measures
were collected before the start of the game, approximately 10 minutes into
game play, and immediately after the game. Participants next performed the
word completion task and the Video Game Ratings measure. Participants
were debriefed and thanked for their participation.
B. RESULTS AND DISCUSSION
This study had two main goals. One was to compare violent and nonviolent
video games on key dimensions so that a pair of games, matched on all
dimensions except for violence, could be selected for further in-depth re-
search. The second was to test the hypothesis that playing violent video games
primes aggressive thoughts. For all analyses we tested a planned contrast that
compared the mean score of participants who had played one of the violent
games to the corresponding mean of those who had played one of the
nonviolent games.
Sex eVects were expected to occur on some measures (e.g., cardiovascular
measures, enjoyment of the games) and not on others. When preliminary
analyses revealed no sex eVects, sex was dropped from the statistical model.
VIOLENT VIDEO GAMES 209
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
1. Cardiovascular Measures
a. Blood Pressure. A 10 (Video game) 2 (Sex: female vs. male)
2 (Type: diastolic vs. systolic) 3 (Time of assessment: before video game
vs. during video game vs. after video game) ANOVA, with type and time as
repeated factors, revealed several interesting eVects. First, there were signi-
cant main eVects of sex [F(1, 86) ¼26.11, P<.001, d¼0.43], time [F(2, 172)
¼9.82, P<.001], and type [F(1, 86) ¼2312.64, P<.001, d¼4.02]. Males had
higher blood pressure than females (Ms¼93.46 and 86.27).
1
Blood pressure
increased from the baseline period (before playing the video game, M¼
89.92) to the video game play period (M¼91.73) and then decreased after
game play was complete (M¼87.94). Of course, SBP was greater than DBP
(Ms¼110.52 and 69.21, respectively). There was no main eVect of which
game was played [F(9, 86) ¼1.67, P>.10], nor was there any hint of a
violent vs. nonviolent game eVect [F(1, 86) ¼0.02, diVerence not signicant].
However, the sex, time, and type main eVects were all moderated by various
two-way interactions.
The type X sex interaction [F(1, 86) ¼18.33, P<.001, d¼0.32] resulted
from the fact that the sex eVect was larger for SBP (M
males
¼115.95 vs. M
females
¼105.8) than for DBP (M
males
¼70.97 vs. M
females
¼67.45). The type X time
interaction [F(2, 172) ¼11.88, P<.001] resulted from the fact that SBP
remained relatively constant across the three time periods (M
before
¼111.64
vs. M
during
¼111.09 vs. M
after
¼108.83), whereas DBP was highest during
video game play (M
before
¼68.21 vs. M
during
¼72.37 vs. M
after
¼67.06).
The omnibus time X game interaction [F(18, 172) ¼1.93, P<.02] is a bit
diYcult to comprehend. However, the more specic contrast testing the time
violent vs. nonviolent game interaction was also signicant [F(2, 172) ¼
5.37, P<.01] and accounted for much of the omnibus interaction.
2
On
average, participants who played one of the nonviolent games showed a
decline in blood pressure across the three time periods (M
before
¼91.41 vs.
M
during
¼90.41 vs. M
after
¼88.07). However, those who played one of the
violent games showed an increase in blood pressure during video game play,
followed by a decrease (M
before
¼88.44 vs. M
during
¼93.04 vs. M
after
¼
87.81). In other words, within the present sample of 10 games the violent
ones increased arousal, as measured by blood pressure, whereas the nonvio-
lent ones did not. This point is especially important to remember when
1
All reported means are the appropriate adjusted means. All P-levels are based on two-tailed
tests. Sample sizes diVer somewhat in diVerent analyses as the result of occasional missing
values. This occurred most frequently for the cardiovascular measures because of occasional
equipment malfunctions.
2
In fact, the residual omnibus time game interaction was nonsignicant [F(16, 172) ¼1.50,
P>.10].
210 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:02 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
selecting pairs of games for more in-depth research on the relation between
violent content and aggressive behavior. It is also important to note, however,
that the postgame assessment of blood pressure showed nearly identical
means for the violent and nonviolent games. Therefore, the potential prob-
lem of arousal being confounded with violent vs. nonviolent content may
not be as severe as many might assume.
b. Heart Rate. Preliminary analyses yielded no sex eVects. A 10 (Video
game) X 3 (Time of assessment) ANOVA with time as a repeated factor
yielded only one marginally signicant eVect, the time main eVect [F(2, 190)
¼2.82, P<.07]. HR was highest when assessed during game play (M¼
77.50), and was at about the same relatively low level before (M¼74.73) and
after (M¼74.28) game play. Indeed, a specic contrast comparing HR
during game play to the average HR before and after game play was
statistically signicant [F(1, 95) ¼4.43, P<.04, d¼0.30]. None of the
video game eVects (omnibus or violent vs. nonviolent contrast) approached
signicance (all Pvalues >.40).
2. Video Game Ratings
The mean ratings of the games on diYculty, enjoyment, action, frustra-
tion, and violence are shown in Table I. There was considerable variability
on all ve dimensions, which made selection of a pair of games diVering
primarily on violent content possible.
Ratings of game diYculty were aVected by game [F(9, 120) ¼7.86, P<.001].
The specic contrast comparing violent to nonviolent games was also signicant
and revealed the violent games to be more diYcult (M¼3.60) than the nonvio-
lent games (M¼2.80) [F(1, 120) ¼8.67, P<.01, d¼0.52]. However, there was
considerable variability within each game type and considerable overlap be-
tween the violent and nonviolent games. The violent game diYculty means
ranged from 2.09 (Wolfenstein 3D) to 4.38 (Street Fighter). The nonviolent
game diYculty means ranged from 1.54 (Ultra Pinball) to 4.83 (Myst).
Participantsenjoyment ratings were also aVected by game [F(9, 110) ¼2.23,
P<.05]. However, the comparison between violent and nonviolent games
showed that participants enjoyed the violent and nonviolent games equally
(F<1). In fact, the nonviolent games were enjoyed slightly more than the
violent games. The game sex interaction was also signicant [F(9, 110) ¼1.97,
P<.05]. However, the specic contrast testing the sex game violence
interaction was only marginally signicant [F(1, 110) ¼3.33, P<.08]. The
slight preference for the nonviolent games was marginally greater for females
(M
nonviolent
¼3.81, M
violent
¼3.53) than for males (M
nonviolent
¼4.23, M
violent
¼4.11).
Ratings of game action were signicantly aVected by game [F(9, 120) ¼
8.35, P<.001]. As shown in Table I, participants found the violent games to
VIOLENT VIDEO GAMES 211
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
contain more action (M¼4.09) than the nonviolent games (M¼2.68)
[F(1, 120) ¼31.67, P<.001, d¼0.99].
Participantsratings of frustration showed a signicant game eVect
[F(9, 120) ¼3.40, P<.001]. However, the contrast between violent and
nonviolent video games was nonsignicant, suggesting that participants
found the violent games (M¼4.40) and nonviolent games (M¼4.05)
equally frustrating [F(1, 120) ¼1.61, P>.20, d¼0.23].
Ratings of the violence of game content showed a signicant sex eVect
[F(1, 110) ¼4.82, P<.05, d¼0.40]. Females rated the games as more violent
than did males (Ms¼3.20 & 2.81, respectively). More importantly, the
violence ratings also yielded a signicant game eVect [F(9, 110) ¼45.81,
P<.001]. As expected, most of this omnibus game eVect (87%) was due
to the much higher violence ratings given for the violent games than for
the nonviolent games (Ms¼4.68 & 1.35, respectively,) [F(1, 110) ¼350.75,
P<.001, d¼3.38]. Furthermore, the sex by game type interaction was
not signicant [F(1, 110) ¼0.92, P>.3], indicating that the sex eVect did
not systematically vary as a function of whether the game was violent or
nonviolent.
TABLE I
Mean Rating of Video Game Difficulty,Enjoyment,Action,Frustration, and Violence
as a Function of Game, and Averages for Nonviolent and Violent Games
Nonviolent games DiYculty Enjoyment* Action Frustration Violence*
Myst 4.83 3.06 1.67 5.33 1.25
Jewel Box 1.77 4.38 2.62 3.69 1.00
Indy Car II 2.21 3.31 2.71 3.79 1.43
3-D Ultra Pinball 1.54 5.40 4.08 2.69 1.62
Glider Pro 3.62 3.94 2.31 4.75 1.41
Average 2.80 4.02 2.68 4.05 1.34
Violent games DiYculty Enjoyment Action Frustration Violence
Speed Demon 4.00 3.86 3.42 5.33 3.27
Dark Forces 3.27 4.10 3.18 4.27 4.35
Wolfenstein 3D 2.09 3.70 5.36 3.82 5.92
Marathon 2 4.25 3.69 3.67 4.25 4.86
Street Fighter 4.38 3.74 4.81 4.31 4.94
Average 3.60 3.82 4.09 4.40 4.67
Mean square error 2.37 2.35 2.01 2.38 .97
*Adjusted for sex eVects. Possible range of ratings was 1 to 7.
212 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
3. Selection of a Matched Pair
We conducted a number of additional analyses to select a violent/nonviolent
game pair most closely matched on irrelevant dimensions (i.e., diYculty,
enjoyment, action, and frustration) and diVering greatly on violence.
Although several pairings appeared to meet our criteria fairly well, we
ultimately chose Glider Pro and Marathon 2. We conducted a 2 (game)
2 (sex) 4 (rating dimension) ANOVA on ratings of these two games,
treating the four ‘‘irrelevant’’ rating dimensions (diYculty, enjoyment, ac-
tion, frustration) as a repeated measures factor. DiVerential eVects of game
would show up either as a game main eVect or a game rating dimension
interaction. The only eVect that approached statistical signicance was the
main eVect of rating dimension [F(3, 72) ¼3.93, P<.05]. The adjusted
means for diYculty, enjoyment, action, and frustration were 3.96, 3.81, 2.98,
and 4.51, respectively. This eVect, of course, is irrelevant to the issue of game
selection. All other eVects had Pvalues >.15. We also compared cardiovas-
cular eVects of these two games. DiVerential game eVects would appear
as an interaction involving the game and time of assessment variables.
None of the game time interactions (2-way or higher) were signicant.
Finally, we compared these two games on the violence ratings. The only
signicant eVect was a large eVect of game on the violence rating [F(1, 24) ¼
53.81, P<.001, d¼2.82]. Marathon 2 was rated as considerably more
violent than Glider Pro [Ms¼4.86 and 1.41, respectively]. The sex main
eVect approached signicance [F(1, 24) ¼4.24, P<.06, d¼0.79], with
females rating the games as more violent than males [Ms¼3.62 and 2.65,
respectively]. The sex game interaction did not approach signicance
[F(1, 24) ¼0.11, P>.70].
Thus, Glider Pro and Marathon 2 were well matched on the irrelevant
dimensions and diVered greatly on the desired dimension of violence. Of
note, these specic game comparisons were based on a relatively small
sample size. Therefore, in Experiment 2 on aggressive behavior (which used
these two games), we measured these same rating dimensions, as well as
blood pressure and HR, as additional statistical controls.
4. Accessibility of Aggressive Thoughts
The second major goal of this study was to test the hypothesis that playing
a violent video game can increase the relative accessibility of aggressive
thoughts. A series of analyses was conducted to compare the percentage of
aggressive word completions after violent versus nonviolent video game play.
Each analysis included a specic contrast comparing the average of the
nonviolent game conditions to the average of the violent game conditions.
VIOLENT VIDEO GAMES 213
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
Preliminary analyses indicated that sex of participant did not have any
signicant impact on the accessibility of aggressive thoughts measure.
a. All 10 Games. Participants produced a signicantly higher percentage
of aggressive words after violent games (M¼14.7) than after nonviolent
games (M¼12.5) [F(1, 120) ¼4.26, P<.05, d¼0.37]. Thus, as predicted by
GAM, playing violent games increased accessibility of aggressive thoughts,
relative to playing nonviolent games.
We conducted several similar analyses with various covariates added to
the model. When we controlled for the rating dimensions of diYculty,
enjoyment, action, and frustration in this way, the violent vs. nonviolent
contrast was still signicant [F(1, 116) ¼5.29, P<.05]. Similarly, the game
violence eVect was still signicant when we controlled for video game
experience (hours per week and average experience with 10 specic games)
[F(1, 110) ¼4.54, P<.05]. When physiological arousal changes were
statistically controlled (baseline to after video game play), the game violence
eVect was still signicant [F(1, 103) ¼5.74, P<.05]. However, when we
controlled for rated violence of the games, the game violence eVect became
nonsignicant [F(1, 119) ¼2.35, P>.12], as expected. In fact, controlling
for the rated violence of the games reduced the game violence eVect on
aggressive thoughts by 45%.
Finally, we performed an analysis that treated the specic games chosen to
represent the violent and nonviolent types as random eVects, rather than
xed (Winer, 1971). This entails using the between groups sums of squares,
within each of the violent and nonviolent game types, to estimate measure-
ment error, with 8 degrees of freedom. This tests the generalizability of the
video game violence eVect across specic games (Wells & Windschitl, 1999).
It also yielded a signicant game violence eVect [F(1, 8) ¼8.96, P<.05].
These results strongly support the hypothesis that violent content in video
games causes increases in the accessibility of aggressive thoughts, indepen-
dent of arousal and aVective inuences. This is the rst study to conclusively
demonstrate this speciceVect of violent video game content.
b. Glider Pro vs. Marathon 2. We further examined the violent video
game eVect on aggressive thoughts by using only the two matched games
Glider Pro and Marathon 2. Sex eVects did not approach signicance, so sex
was dropped. Despite the small sample size, Marathon 2 participants pro-
duced a signicantly higher percentage of aggressive words (M¼15.4) than
did Glider Pro (M¼10.8) [F(1, 26) ¼6.03, P<.05, d¼0.95]. A similar
statistical model that included changes in physiological arousal (baseline to
after video game play) as covariates yielded similar results. The eVect of
game was signicant [F(1, 21) ¼10.96, P<.01].
We then ran a series of analyses in which each of the video game rating
dimensions were entered as covariates. The analyses with diYculty, enjoyment,
214 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
action, and frustration each yielded a signicant game eVect on aggressive
thoughts [Fs(1, 25) >5.00, Pvalues <.05]. As expected, when rated violence
of the game was entered as a covariate the game eVect became nonsignicant
[F(1, 25) ¼2.42, P>.10], again demonstrating the specicity of violent
content eVects.
VI. Experiment 2
A. METHOD
1. Participants
Ninety-seven female and 93 male undergraduates participated, selected on
the basis of their responses to the Trait Hostility (TH) scale, administered
at the beginning of the semester as part of a battery of questionnaires. Half
of the males and half of the females were selected from the top and bottom
thirds of the TH distribution. All participated in return for partial course
credit for their introductory psychology class.
3
2. Design
The experiment can be conceived as a 2 (TH: high vs. low) 2 (Video
game: violent vs. nonviolent) 2 (Provocation pattern: increasing vs. am-
biguous) 2 (Sex: male vs. female) between subjects design. However, TH
was used as a continuous variable in all analyses, rather than a two-level
categorical variable. This regression approach is statistically more appropri-
ate and more powerful. For all analyses that included a covariate (e.g., TH),
we also tested all possible interaction terms involving the covariate. None of
these interactions were statistically reliable, so they were dropped from the
nal statistical models reported in this article. For dependent variables that
3
Six participants were dropped because they reported unusually high levels of video game
experience or number of hours per week spent playing video games. Our concern was that
participants who had extensive video game experience might respond somewhat diVerently than
the normal population. We used the standard procedures described by Tukey (1971) for
identifying outliers and ‘‘far outliers.’’ Six far outliers were identied, two on the basis of the
video game experience scale, four on the hours per week measure. For example, the far outliers
on the latter measure reported playing video games from 18 to 24 hours per week in recent
months. Five of the six far outliers were males. Supplemental analyses that included these
participants yielded the same patterns of aggression as those reported in the Results section,
albeit the eVect sizes were slightly smaller.
VIOLENT VIDEO GAMES 215
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
were not signicantly inuenced by sex of participant, the reported results
are based on analyses that dropped sex from the statistical model.
3. Materials
a. Trait Hostility. The Trait Hostility Scale is a 30-item self-report
inventory designed to measure chronic individual diVerences in aggressive-
ness or irritability, adapted from the irritability scale developed by Caprara,
Cinanni, DImperio, Passerini, Renzi, and Travaglia (1985) (see Anderson,
1997; Dill et al., 1997). Sample items include (1) Whoever insults me or my
family is looking for trouble, (2) Sometimes I shout, hit and kick and let oV
steam. The internal reliability of this scale is generally high; in the present
sample coeYcient alpha ¼.86.
b. Video Games. As described earlier, Marathon 2 (violent) and GliderPro
(nonviolent) were selected because they were similar on a variety of dimensions
in Experiment 1.
c. Competitive Reaction Time Task. A modied version of the Taylor
Competitive Reaction Time (CRT) task was used to assess aggressive behav-
ior. The CRT is a widely used and externally valid measure of aggressive
behavior (Anderson & Bushman, 1997; Anderson, Lindsay, & Bushman,
1999; Carlson, Marcus-Newhall, & Miller, 1989; Giancola & Chermack,
1998). Participants believe they are competing with another person to see
who can respond rst upon presentation of a tone. In the standard version of
the game, after each trial the ‘‘loser’’ receives an aversive punishment (e.g.,
loud noise), the intensity of which is supposedly set by the opponent. Prior to
each trial, each participant sets the punishment level that supposedly will be
delivered to the opponent, if the participant wins the trial. The possible
settings range from 0 (no noise) to 10 (100 db). These settings constitute
the measure of aggressive behavior.
In the present experiment we used a two-phase version of the task (Anderson,
Anderson, Dorr, DeNeve, & Flanagan, 2000; Bartholow & Anderson, 2002;
Lindsay & Anderson, 2000). In Phase 1 participants were told that their
opponent would set the intensity of the noise blast that the participant would
receive as ‘‘punishment’’ on ‘‘lose’’ trials, but that the opponent would not be
punished on trials that the participant won. It was further explained that in
Phase 2, the roles would be reversed so that the participant would set the
intensity of the noise blasts for the opponent, but the participant would not
receive punishments on ‘‘lose’’ trials. In actuality, a computer program
determined wins and losses as well as the noise intensities and durations
delivered to participants in Phase 1. After each trial the participant also saw
what noise level was ‘‘set’’ by the opponent, displayed on the computer
216 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
screen. All participants were given sample noise blasts of level ‘‘2’’ (60 db)
and ‘‘8’’ (90 db) before beginning Phase 1 of the task.
d. Provocation Pattern Manipulation. In Phase 1 of the CRT task, all
participants received the same randomly ordered series of 13 wins and 12
losses determined by the computer program. They also received blasts of
noise on the ‘‘lose’’ trials. The pattern of noise blasts was either an ambigu-
ous or an increasing provocation pattern (Anderson et al., 2000). Partici-
pants in both provocation conditions saw exactly the same punishment
settings8 in the low range, 9 in the middle range, and 8 in the high range,
and actually received the same punishments (4 in each range). In the ambig-
uous provocation condition the pattern of noise intensities was random,
whereas the increasing provocation pattern consisted of mostly low noise
intensities on the early trials, middle intensities on the middle trials, and
high intensities on later trials. On completion of Phase 1, the experimenter
reminded participants that in Phase 2 they would set noise intensities to be
delivered to their opponent, and that they would not receive noise blasts on
any trial in this phase.
4. Procedure
Participants were randomly assigned to one of the four experimental
conditions (video game provocation), with the constraint that equal
proportions of males and females and that equal proportions of high and
low TH participants were run in each condition. Same-sex pairs were run in
individual cubicles. On arrival, they were seated in cubicles and asked to
read and sign the consent form and to then read instructions for the video
game. Participants were told that the study was concerned with the ways
people learn diVerent types of computer tasks. Participants were informed
that we were interested in possible diVerences between simple and complex
tasks, and that we would be measuring HR and blood pressure throughout
the session. Participants played the assigned video game for 20 minutes.
Then the CRT task was explained. All participants were told that their
opponent would set the noise blasts in Phase 1 while they would set the
noise blasts for their opponent to hear in Phase 2. After the CRT task,
participants completed a questionnaire that included a manipulation check,
some motivation and aVect items, some demographic information, and
several suspicion check items.
Blood pressure and pulse were measured at ve diVerent points in time:
(1) before playing the video game, (2) about 10 minutes into the video game,
(3) after playing the video game, (4) after Phase 1 of the CRT task, and (5) aft-
er Phase 2 of the CRT task. Finally, the participants were debriefed. Care
VIOLENT VIDEO GAMES 217
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
was taken to ensure that each participant knew it was the computer and not
the other participant in the session that set the noise blasts in Phase 1 and
that the noise blasts they sent in Phase 2 were received only by the computer.
5. Measures
a. Aggressive Behavior. Aggressive behavior was operationalized as the
noise intensity (010) sent by the participants to their opponents in Phase 2 of
the CRT task. As is common with the CRT task, we examined four diVerent
intensity measures: intensity setting on the rst trial, and the average settings
on trials 29, 1017, and 1825. The early trials are the most important in
this two-phase version of the CRT, especially Trial 1, because it is the rst
opportunity the participant has to retaliate after being provoked. Note that
in the standard one-phase version Trial 1 occurs prior to any provocation,
provocations continue throughout the 25 trials, and therefore all trials are of
theoretical interest.
b. Cardiovascular Arousal. At each of the ve measurement periods,
blood pressure and pulse were assessed twice, with an interval of 1 minute
between the completion of the rst measurement and the beginning of the
second.
4
c. Questionnaires. After the last blood pressure and pulse measurements,
participants answered a number of questions about the experiment. One
item asked participants if they were ever ‘‘angry’’ during the reaction time
task. Responses were on a 5-point unipolar scale anchored at 1 (not at all),
2 (a little bit), 3 (somewhat), 4 (quite a lot), and 5 (a lot). Also included were
six items on which participants were to ‘‘indicate the extent to which this
motive describes your motive when deciding on where to set the noise
levels.’’ These items used the same 5-point scale described above. The six
items were (1) I wanted to impair my opponents performance in order to
win more; (2) I wanted to control my opponents level of responses; (3)
I wanted to make my opponent mad; (4) I wanted to hurt my opponent;
(5) I wanted to pay back my opponent for the noise levels he/she set; (6) I
wanted to blast him/her harder than he/she blasted me. The rst two items
represent instrumental reasons for aggressing. Responses to these two
4
Heart rate and blood pressure were assessed for two reasons. First, it was an important part
of the cover story. Second, we wanted to be able to statistically control for any eVects of the
video game manipulation on aggression that might be due to changes in arousal. As expected,
the video games did not produce diVerential changes in arousal. Including the cardiovascular
change measures as covariates in later analyses of aggressive behavior did not alter the pattern
of results in any appreciable way (but did reduce the error term and slightly increased the eVect
sizes of the video game and provocation manipulations). Thus, for the sake of simplicity these
arousal measures are not further discussed.
218 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
items were positively correlated (r¼.57, P<.001), and were combined
to form a scale labeled ‘‘Instrumental Aggressive Motivation.’’ The latter
four items represent a clearly revengeful type of aggressive motive, were
highly correlated, and were combined to form a scale labeled ‘‘Revenge
Motivation.’’ CoeYcient alpha for this scale was .74.
One question asked ‘‘Did the pattern of noise levels that you received
appear to be increasing, decreasing, or random?’’ This item was coded as
þ1 for increasing, 1 for decreasing, and 0 for random. We expected this
manipulation check to yield smaller scores in the ‘‘ambiguous’’ than in the
‘‘increasing’’ provocation condition, because the actual pattern was random
in the former and increasing in the latter. A 2 (video game) 2 (provocation
pattern) analysis of variance (ANOVA) conrmed this expectation. Those in
the increasing provocation conditions reported an increasing pattern of noises
(M¼.43), whereas those in the ambiguous condition reported that the
pattern seemed random [(M¼.05), F(1, 178) ¼43.91, P<.001, d¼0.95].
An open-ended question asked ‘‘What do you think the purpose of this
experiment was?’’ Responses to this item and to the nal oral debrieng were
used by the experimenter to rate each participant as either not suspicious (0),
slightly suspicious (1), or suspicious (2). Fourteen participants were classied
as suspicious. However, suspicion was unrelated to performance on the main
dependent variables, so all were kept.
Participants then completed a ‘‘Background Questionnaire’’ assessing de-
mographic information including height, weight, age, year in school, academic
major, and time since using alcohol, caVeine, and exercising. Participants also
estimated how many hours per week they had spent playing video games ‘‘in
recent months.’’ They then indicated ‘‘How much you have ever played each
of ...’’ 12 specic video games, using a 5-point unipolar scale anchored at 1
(Never have played) and 5 (Have played a lot). The 12 games listed were
Wolfenstein3D, You Dont Know Jack,Civilization II, A-10Attack, Computer
Cribbage, Street Fighter, Dark Forces, Myst, Indy Car II, Speed Demon, 3-D
Ultra Pinball, and Jewelbox. Ratings on these 12 items were combined into a
Video Game Experience composite, with a coeYcient alpha of .63. Finally, a
thorough oral debrieng was given, with notes recorded by the experimenter.
B. RESULTS
1. Aggressive Behavior
Four measures of aggressive behavior were based on the noise punishment
levels that participants set for their opponents during Phase 2 of the CRT
task. Recall that the key behavioral measure was noise intensity on Trial 1,
VIOLENT VIDEO GAMES 219
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
because it was the participantsrst opportunity to retaliate for noise blasts
received during Phase 1. The noise settings for the remaining 24 trials were
averaged into three blocks of eight trials each.
The key prediction was that exposure to the violent video game would
increase aggressive behavior in the two-phase CRT task. We believed that
this eVect was most likely to occur in the ambiguous provocation condition,
especially on the rst or early trials (a time video game provocation
pattern interaction). This is based on an outburst/social justice model of
aggressive behavior in the two-phase CRT rst described in the Anderson
et al. (2000) studies of temperature eVects. Experiment 5 of that work used a
very similar two-phase CRT task with both an ambigous and an increasing
provocation pattern and found that most of the interesting temperature
eVects occurred in the ambiguous pattern condition, especially on the early
trials.
We also expected that TH would be positively related to aggression. All of
these predictions were borne out.
We conducted a 2 (video game) 2 (provocation pattern) 4 (time: trial 1
vs. block 1 vs. block 2 vs. block 3) ANOVA, with the last factor as a repeated
measures factor, and with TH as a covariate. We then conducted a set of
planned contrasts examining the violent video game eVect separately for the
two provocation conditions, for each of the four aggression measures. Both
the provocation and the TH main eVects were signicant [Fs(1, 179) ¼5.81
and 4.69, respectively, Ps<.05]. Participants in the increasing provocation
condition set lower noise punishments than those in the ambiguous provo-
cation condition (Ms¼5.06 and 5.51, respectively, d¼0.18). Trait hostility
was positively related to noise punishment level (b¼0.23). This latter nding
provides additional evidence that the two-phase CRT task validly assesses
aggressive behavior.
There was also a signicant time eVect, a time provocation pattern
interaction, and a time video game provocation pattern interaction
[Fs(3, 537) ¼27.90, 19.81, and 4.06; Ps<.01]. Figure 4 illustrates both
the time main eVect and the time provocation pattern interaction. A
simple summary of these eVects is that participants tended to produce the
same pattern of noise punishments for their opponents in Phase 2 as they
had received from their opponents in Phase 1. On average, those in the
ambiguous condition gave moderate punishments across the four times,
whereas those in the increasing provocation condition delivered punish-
ments that started low and increased across time. Interestingly, this same
aggression pattern has occurred in the other two experiments that used the
two-phase CRT and both the ambiguous and the increasing provocation
pattern (Anderson et al., 2000, Experiment 5; Lindsay & Anderson, 2000,
Experiment 4).
220 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
To more thoroughly examine the video game eVects in the context of our
main predictions as well as in context of the three-way interaction, we
conducted 2 (video game) 2 (provocation pattern) ANOVAs with TH as
a covariate on each measure of aggressive behavior separately, along with
two planned contrasts testing separately the video game eVect in ambiguous
and increasing provocation pattern conditions. Table II presents the adjust-
ed means. Figure 5 plots the mean diVerences between the violent and
nonviolent video game conditions.
As can be seen, the pattern was essentially as predicted. On Trial 1, there
were signicant main eVects of provocation pattern and trait hostility, and
a marginally signicant video game provocation pattern interaction
Fig. 4. Noise punishment settings as a function of provocation pattern and time of aggres-
sion in Phase 2 of the competitive reaction time task, Experiment 2.
TABLE II
Aggressive Behavior (Average Noise Punishment Settings)as a Function of
Video Game,Provocation Pattern, and Time of Aggression
Video game Provocation pattern nTrial 1 Block 1 Block 2 Block 3
Violent Ambiguous 44 5.93 5.64 5.52 5.66
Nonviolent Ambiguous 46 4.86 5.55 5.48 5.47
Video game eVect 1.07 0.08 0.04 0.19
Violent Increasing 46 3.87 5.16 5.75 5.57
Nonviolent Increasing 48 3.98 4.96 5.38 5.81
Video game eVect 0.11 0.20 0.38 0.24
VIOLENT VIDEO GAMES 221
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:03 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
[Fs(1, 179) ¼22.35, 4.03, and 3.56, Ps<.001, .05, & .07, respectively]. The
ambiguous provocation pattern led to more aggression than the increasing
pattern (Ms¼5.39 and 3.92, d¼0.70). Trait hostility was positively related
to aggression (b¼0.35). The specic planned contrasts testing the video
game eVect separately for the ambiguous and increasing provocation pattern
conditions revealed that those who played the violent video game and who
had received the ambiguous provocation pattern delivered signicantly high-
er noise punishments to their opponents than did those in the corresponding
nonviolent game condition [F(1, 179) ¼5.72, P<.02, d¼0.5]. Interestingly,
those who received the increasing provocation pattern were unaVected by
the video games (F<1, d¼.05).
In Block 1 (Trials 29) only the provocation and TH main eVects
were even close to signicant. Those in ambiguous provocation conditions
delivered greater punishments to their opponents than those in the increas-
ing provocation conditions [Ms¼5.60 and 5.06, respectively; F(1, 179) ¼
7.02, P<.01, d¼0.39]. Trait hostility was again positively related to
aggression [b¼0.29, F(1, 179) ¼6.67, P<.02]. The video game eVect was
not signicant under either provocation condition (Fs<1). In Blocks
2 and 3, none of the eVects of video game, provocation, or TH were signicant
(Ps>.10).
2. Supplementary Analyses. Table III presents the correlations among
questionnaire variables. The most interesting ndings involved the mea-
sure of revenge motivation. Revenge motivation correlated positively with
feeling angry during the CRT task, instrumental aggressive motivation, and
Fig. 5. Violent video game eVect on noise punishment settings (violentnonviolent game
conditions) as a function of provocation pattern and time of aggression, controlling for trait
hostility, experiment 2.
222 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
experience with the 12 video games included in the questionnaire. To further
explore the latter relationship we split the video game experience scale into two
subscales and correlated each with revenge and instrumental aggressive moti-
vation. Experience with the six violent video games correlated signicantly
with revenge motivation (r¼.21, P<.01), and with instrumental aggressive
motivation (r¼.15, P<.05). Experience with the six nonviolent video games
did not correlate signicantly with either type of motivation (Ps>.5).
Revenge motivation also correlated positively with TH, which was measured
several weeks prior to the laboratory session. This suggests that the TH eVect
on aggressive behavior may have been mediated by revenge motivation. To test
this we ran the same regression ANOVAs on the four measures of aggression
but with revenge motivation as a covariate. The TH eVect disappeared. In
the repeated measures ANOVA the main eVect of revenge motivation was
highly signicant [F(1, 178) ¼52.08, P<.001], but the TH eVect did not
approach signicance (F<1). Similarly, on both of the individual aggression
measures that had previously yielded signicant eVects of TH (Trial 1, Block 1)
the revenge motivation eVect was quite strong [Fs(1, 178) ¼26.83 and 35.52,
bs¼0.89 & 0.65, respectively; Ps<.001], and the trait hostility eVects became
nonsignicant (Ps>.2). These results suggest that trait hostility inuenced
aggression through its eVect on revenge motivation.
Interestingly, revenge motivation was not aVected by the video game
manipulation (F<1). Furthermore, the violent video game eVect seen on
Trial 1 aggression in the ambiguous provocation condition was not dimin-
ished by the inclusion of revenge motivation in the statistical model
[F(1, 178) ¼6.96, p<.01]. The diVerence in adjusted means between the
violent and nonviolent video game conditions was essentially the same when
TABLE III
Correlations Among Questionnaire Variables
Variable
Revenge
motivation
Video game
experience
Hours
per week Angry
Instrumental
aggressive
motivation
Trait hostility .25* .02 .05 .06 .09
Revenge motivation .15
{
.08 .44* .26*
Video game experience .46* .05 .15
{
Hours per week .06 .12
Angry .07
Hours per week indicates hours per week spent playing video games.
*P<.001. NS range from 174 to 184.
y
P<.01.
{
P<.05.
VIOLENT VIDEO GAMES 223
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
revenge motivation was in the model (Ms¼5.89 vs. 4.80) as when it was not
in the model (Ms¼5.93 vs. 4.86). Thus, revenge motivation did not mediate
the video game eVect on aggression.
C. DISCUSSION
This experiment has four main ndings. First and foremost, playing a
violent video game for 20 minutes led to signicantly more aggression than
did playing a nonviolent game, in exactly the circumstances expectedon
the rst retaliation opportunity after an ambiguous pattern of provocations.
This video game eVect on aggression was specically due to violent content;
both the aVective and arousal routes were controlled by the selection of
games that do not diVer on these two dimensions. Second, TH was positively
related to aggressive behavior in the CRT task. Third, this eVect was
mediated by revenge motivation. Finally, the positive correlations between
past experience with violent video games and both of the aggressive motive
measures (revengeful and instrumental) suggest that repeated exposure to
violent video games might increase the likelihood that minor provocations
will elicit revengeful and instrumentally aggressive responses.
It is important to note that the violent video game eVect occurred only on the
rst trial in the ambiguous provocation condition. This is the most critical set
of circumstances for testing the violent content hypothesis in the two-phase
CRT. The lack of video game eVects in the increasing provocation condition
conrms earlier reports that in the two-phase CRT the ambiguous provocation
pattern results in more sensitive tests of other eVects than does the increasing
provocation pattern (Anderson et al., 2000; Lindsay & Anderson, 2000).
Therefore, the lack of a video game eVect in the increasing provocation pattern
condition should not be seen as a disconrmation of the violent content
hypothesis. Nonetheless, additional tests of the violent content hypothesis
are needed. Experiment 3 was conducted in part to provide such a test.
VII. Experiment 3
One major goal was conceptual replication. We made several changes to
maximize the gain from this replication test. First, we added two modied
games (one violent and one nonviolent) that have not been used in previous
studies, in addition to the two games used in Experiment 2 (Wells & Windschi tl,
1999). Second, we used the standard one-phase version of the CRT task in
which the participant and the ‘‘opponent’’ set punishment levels for each other
224 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
on the same set of 25 trials. In this paradigm, aggressive behavior emitted
during all 25 trials is relevant, unlike the later trials in the two-phase version.
Two additional changes allowed for a reasonable correlational test of the
hypothesis that exposure to media violence is positively associated with
aggressive behavior. We used the physical aggression subscale of the
Buss-Perry Aggression Questionnaire (Buss & Perry, 1992) as a measure of
‘‘trait aggressiveness.’’ We also created an overall media violence exposure
variable by combining a shortened version of the violent video games measure
used by Anderson and Dill (2000) with a measure of exposure to TV violence.
Finally, Experiment 3 provided a rst attempt to examine the eVects of
varying the realism or the ‘‘humanness’’ of the targets of aggression within a
game. It has often been assumed that cartoon-like characters in violent video
games (as well as television and movies) would elicit less aggression than
human characters, especially among participants older than 10 or 12 years
(i.e., participants who can clearly distinguish between the fantasy world of
cartoons and the real world). Half of the violent game participants played
the original version of Marathon 2, in which the targets are humanoid aliens
with green blood. The other half played the same game except that the
targets were given a human appearance and spouted red blood when shot.
A. METHOD
1. Participants
Two hundred fourteen undergraduate female and male students enrolled
in an introductory psychology class participated for course credit. Partici-
pants were randomly assigned to one of four conditions. Two of the condi-
tions involved playing a violent video game; the other two involved playing a
nonviolent video game. Ten participants were dropped for a variety of
reasons, including failure to complete the materials, high confusion about
how to play the games, declining to play the assigned video game, and giving
impossible answers to some questionnaire items. For instance, one partici-
pant reported watching 70 hours of television per week while taking a full
course load. Thus, the total nal sample size consisted of 134 females and 70
males. The ratio of female/male participants was approximately equal across
experimental conditions.
2. Materials & Design
a. Video Games. All video games were played on G3 Power Macintosh
computers. The CRT task was conducted with LCIII Macintosh computers.
Two of the games were essentially the same as those played in Experiment
VIOLENT VIDEO GAMES 225
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
2 (the green-blooded alien version of Marathon 2, labeled ‘‘Alien’’ hereafter
in this article, and Glider Pro). One minor change was made in the Alien
version; the human allies that appeared in the opening scene were removed.
The other two games were major modications of Marathon 2. The violent
modied version (labeled ‘‘Human’’) was identical to Alien except that
human gures with red blood replaced the green-blooded aliens. The nonvi-
olent version consisted of the same ‘‘world’’ as Marathon 2, but there were
no enemies at which to shoot, and the players main task was to explore the
world, nd replacement oxygen cylinders, and eventually nd the transport-
er in order to return to the ship. This game had a time element built into it,
such that the player could ‘‘die’’ if he or she failed to nd suYcient oxygen
supplies while searching for the transporter. This was labelled the ‘‘Explore’’
condition.
5
b. Aggression Paradigm. After playing the assigned video game, the
participants performed a standard version of the CRT task. Participants
were led to believe that they were competing against another college student,
and that they would be setting punishment levels for each other prior to each
trial. They were to click the mouse button as quickly as possible after hearing
a tone. Participants were told, ‘‘Once you both have hit the mouse button
and the winner is determined, the computer will display a sign that says
either YOU WON!or YOU LOST!If you lose the trial, you will receive
the noise your partner chose for you.’’ There were 25 such trials, the out-
comes of which were actually controlled by the computer. The ambiguous
win/loss and punishment patterns (described in Experiment 2) were used for
all participants. The main measure of aggression was the average noise
intensity level set by the participant.
c. Questionnaire. A questionnaire was administered at the end of the
CRT task. It contained the following measures that were used in Experiment
2: revenge motivation (coeYcient alpha ¼.85), instrumental aggressive
motivation (two items r¼.40), and angry feelings. Trait hostility was
replaced by the nine-item Physical Aggression subscale of the Buss-Perry
Aggression Questionnaire (coeYcient alpha ¼.84). The format and style of
the items is very similar to the TH scale used in Experiment 2, with scores
ranging from 1 to 5. Sample items are ‘‘There are people who pushed me so
far that we came to blows’’ and ‘‘I get into ghts a little more than the
5
We also modied the opening scenes of the Alien and Human versions to make it somewhat
easier, so that novice game players didnt‘‘die’’ right away. We thank Brian C. Anderson for
making these various changes to the Marathon 2 games. Note that to maximize the power of
the comparisons of the two violent game conditions (Alien vs. Human), we set up our
randomization schedule so that there would be about twice as many participants in the violent
conditions as in the nonviolent conditions. The nal sample sizes were: Alien ¼70; Human ¼62;
Glider Pro ¼36; Explore ¼36.
226 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
average person.’’ Although previous research has shown that this scale loads
on the same general factor as the TH scale used in Experiment 2 (Dill et al.,
1997), the advantage of the Buss-Perry physical aggression subscale is that
all nine items involve aggressive behavior, whereas the TH scale used in
Experiment 2 includes several diVerent types of items.
Video game experience was replaced with a composite measure of expo-
sure to video game and television violence. To assess exposure to violent
video games, we used a shortened version of the measure reported in
Anderson and Dill (2000). Participants listed up to three favorite video
games, indicated how often they have played each in recent years, and rated
the level of violence in the games. The video game violence exposure measure
is created by multiplying the violence rating for each game by the amount of
time spent playing that game, and averaging across the three games. Parti-
cipants also estimated how many hours per week they watch television, and
what proportion of time they watch violent television shows. The television
violence measure is simply the product of those two estimates. To create the
media violence exposure composite, we standardized the video game and
television violence scores and summed them.
d. Design. The overall design is thus a 2 (Sex: male vs. female)
2 (Content: violent vs. nonviolent) 2 (Game version: Old (Alien & Glider
Pro) vs. New (Human & Explore). The Physical Aggression scale was used as
a covariate.
6
3. Procedure
Participants were escorted into the laboratory on arrival. After being
seated by the video game equipment, participants read and signed a consent
form. The cover story explained that the study involved examining how
people learn simple and complex computer tasks. They were also told that
they would be playing two games, one a single-player game and the other a
competitive two-person game. Each participant was given an explanation of
how the controller worked and how to play the single-player game, followed
by instructions on how to play the CRT task. They were then told that they
had been randomly assigned to play the single-player game rst and were
instructed to play the designated game for 20 minutes. After playing for 20
6
This scale was not administered at the beginning of the session because we believed that
doing so would increase suspicion about the true purpose of the study. To check on the
possibility that the experimental manipulations might have systematically inuenced scores on
this ‘‘trait’’ measure, we conducted a 2 (Content: violent vs. nonviolent) 2 [Game version: Old
(Alien & Glider Pro) vs. New (Human & Explore)] ANOVA on the trait physical aggression
scores. None of the eVects approached signicance [all Fs<0.15, all Ps>.70]. Thus, using this
variable as a trait physical aggression covariate seemed appropriate.
VIOLENT VIDEO GAMES 227
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
minutes, the participant completed the CRT task. Participants were told that
they would not meet, see, or learn who their opponent was, but that their
opponent was the same sex. After completion of the CRT task, a question-
naire was administered to assess a number of aVective and motivational
variables, as well as suspicion. During the nal debrieng, the experimenter
probed for suspicion, explained all procedures, answered any questions, and
thanked the participant.
7
B. RESULTS
1. Aggressive Behavior
The main measure of aggressive behavior in this experiment was the aver-
age noise intensity level that participants set as punishments for their oppo-
nent, averaged across all 25 trials of the CRT task. Both the trait physical
aggression eVect and the main eVect of video game violent content were
signicant, replicating the main ndings of Experiment 2. Trait physical
aggression was positively associated with aggression [F(1, 195) ¼15.32,
P<.001, b¼.42]. More importantly, as shown in the left two columns of
Fig. 6, participants who had played a violent video game set higher punish-
ment levels than those who had played a nonviolent game [F(1, 195) ¼7.17,
P<.01, d¼.38, Ms¼5.41 and 4.83, respectively]. None of the other main
eVects or interactions were signicant [all Fs(1, 195) <3.20, all Ps>.07].
7
As in Experiment 2, the nal questionnaire included items asking participants about the true
purpose of the study. In addition, the Experimenter conducted a structured interview designed
to detect suspicion as well as to ease into the debrieng. The second author later examined the
participantswritten comments and the Experimentersnotes and rated each participant on a
four-point suspicion scale (03). Twenty-two participants indicated that they knew (or strongly
believed) that the study was about video games and aggression. These participants were
excluded from all analyses. During the year in which this experiment was conducted, a number
of stories about violent video games appaeared in national, regional, and student newspapers
and in a wide array of electronic media, which may account for the increased suspicion rate.
Unlike Experiment 2, in which only a few participants were so suspicious and in which suspicion
was unrelated to the key dependent variables, in the present study suspicion was highly related
to the key measure of aggression. Specically, the 22 highly suspicious participants produced
signicantly lower levels of aggression than did the remaining participants [F(1, 238) ¼26.47,
P<.0001, d¼.67]. This nding is similar to other research on suspicion in aggression studies, in
which suspicious participants tend to be relatively unaVected by experimental manipulations and
tend to display low levels of aggression, presumedly as a result of their deciding to not display
aggressive inclinations (Berkowitz & Donnerstein, 1982; Carlson, Marcus-Newhall, & Miller,
1990; Kruglanski, 1975). Fortunately, there was no systematic relation between condition of the
study and suspicion level [F(7, 238) ¼1.04, P>:40].
228 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
2. Supplementary Analyses
a. Correlations. Table IV presents the correlations among the key
questionnaire measures. The pattern of correlations was very similar to
that found in Experiment 2. Revenge motivation was positively correlated
with feeling angry, instrumental aggressive motivation, and trait physical
aggression. The latter suggests that the trait physical aggression eVect on
aggressive behavior reported earlier may have been mediated by revenge
motivation, similar to the TH eVect in Experiment 2. We return to this idea
shortly.
One of the most interesting correlational ndings concerned the relation
between media violence exposure and the trait physical aggression measure.
As shown in Table 4, the correlation was positive, statistically signicant,
and in the small to medium range. Trait physical aggression also correlated
positively with the number of hours spent with electronic entertainment in
general. To further test the hypothesis that violent content is most important
in this relation, rather than simply number of hours spent on electronic
entertainment, we conducted a simple regression analysis that included media
violence exposure and number of hours per week spent on video games and
television as predictors of trait physical aggression. The media violence mea-
sure remained a signicant predictor [F(1, 200) ¼18.79, P<.001, b¼.43].
When sex was also added to the model, the media violence eVect remained
signicant [F(1, 199) ¼6.68, P<.02, b¼.27].
Fig. 6. Covariate adjusted average noise intensity settings as a function of video
game violence, without (unmediated) and with revenge motivation as a mediating variable,
Experiment 3.
VIOLENT VIDEO GAMES 229
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
b. Mediation by Revenge. Revenge motivation was further analyzed with
a 2 (Sex: male vs. female) 2 (Content: violent vs. nonviolent) 2 (Game
version: Old (Alien and Glider Pro) vs. New (Human and Explore) ANCO-
VA, with trait physical aggression as a covariate. There was a signicant main
eVect of violent content on revenge motivation [F(1, 195) ¼8.24, P<.01,
d¼.41]. Participants in the violent game conditions reported higher levels of
revenge motives than did those in the nonviolent game conditions [Ms¼1.97
and 1.63, respectively]. There was also a signicant eVect of trait physical
aggression [b¼.43, F(1, 195) ¼31.83, P<.001]. This suggests that both the
trait physical aggression eVect and the video game eVect on aggression may
have been mediated by revenge motivation.
To test the mediation hypothesis, we added revenge motivation to the
ANCOVA model and examined the change in aggression score variance
accounted for by trait physical aggression and by the video game violence
manipulation. In both cases, the proportion of unique variance attributed to
the predictor dropped signicantly. Revenge motivation accounted for 77%
of the variance attributed to the trait physical aggression [F(1, 194) ¼13.27,
P<.001]. Nonetheless, the unique variance accounted for by trait physical
aggression still remained statistically signicant [F(1, 194) ¼3.92, P<.05].
Thus, it appeared that revenge motivation mediated most but not all of the
trait physical aggression eVect on aggression.
Revenge motivation accounted for 61% of the the violent video game
variance [F(1, 194) ¼4.89, P<.05]. However, when revenge motivation
was partialled out, the video game eVect was still marginally signicant
[F(1, 194) ¼3.16, P<.08]. Revenge motivation appeared to mediate much
TABLE IV
Correlations Among Key Questionnaire Variables
Variable
Revenge
motivation
Hours
per week Angry
Instrumental
aggressive
motivation
Media
violence
Trait physical aggression .39* .23* .13 .10 .36*
Revenge motivation .05 .48* .49* .07
Hours per week ——.04 .02 .55*
Angry —— .18
y
.00
Instrumental aggressive motivation ——— —.03
Hours per week indicates hours per week spent playing video games and watching television.
*P<.001. NS range from 203 to 204.
y
P<.01.
{
P<.05.
230 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
but not all of the video game eVect on aggression. The two right columns of
Fig. 6 show the video game violence eVect after revenge motivation is
partialled out of the adjusted means.
c. Alien vs. Human Targets. The two violent versions of Marathon
2 used in this experiment diVered only in whether the enemy targets were
green-blooded aliens or red-blooded humans. Supplemental analyses were
performed on these two violent game conditions. The rst set involved
ratings of the games provided by participants on the standard dimensions
used to select games in Experiment 1: diYculty, enjoyment, violence of
graphics and content (averaged), frustration, and action. The 2 (sex)
2 (game) ANCOVAs with trait physical aggression as a covariate yielded
no signicant game main or interaction eVects. Indeed, the only reliable
eVects to emerge from these analyses were main eVects of sex on the enjoy-
ment, violence, and frustration ratings [Fs(1, 127) ¼6.78, 8.94, and 4.06,
respectively; all Ps<.05]. Males enjoyed these two violent games more than
females (Ms¼3.88 and 3.09). They also rated the games as less violent
(Ms¼4.03 and 4.87) and less frustrating (Ms¼3.76 and 4.38) than females.
In sum, the two violent games were essentially equal on these game
dimensions, although men and women viewed them somewhat diVerently.
The most important question concerning the two violent game conditions
is whether the substitution of human characters with red blood increased
aggression in this short-term context. It did not [F(1, 127) ¼0.36, P>.50],
although the means were slightly in that direction (Ms¼5.49 and 5.33 for the
Human and Alien conditions, respectively).
C. DISCUSSION
Experiment 3 replicated the main ndings of Experiment 2 using two
similar violent games (the Alien and Human versions of Marathon 2) versus
two nonviolent games (Glider Pro and the exploration version of Mara-
thon 2) and did so with the more standard version of the CRT task. This
replication strengthens the hypothesis that violent content in a video game
can increase aggressive behavior after a minor provocation, even when the
arousal and aVective sequela are controlled. Furthermore, the positive asso-
ciation between trait physical aggression and CRT aggression further vali-
dates this noiseaggression paradigm. Finally, the trait physical aggression
eVect on aggressive behavior was partially mediated by revenge motivation,
similar to the TH eVect in Experiment 2.
Experiment 3 also found essentially equivalent eVects of the two violent
versions of Marathon 2the green-blooded alien or red-blooded human
VIOLENT VIDEO GAMES 231
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
enemies. This strongly contradicts conventional wisdom and some research on
television violence, both of which suggest that more realistic violence would
have a relatively greater impact. Of course, failure to nd a statistically signi-
cant eVect is not suYcient to claim that the eVect does not exist. Two
alternative explanations for the lack of a ‘‘realism’’ eVect are quite plausible.
First, it may be that the realism eVect occurs in repeated exposure long-term
contexts but not in a short-term context. This would be true if the realism eVect
operates as a kind of systematic desensitization eVect (Carnagey, Bushman, &
Anderson, under review). Second, the high realism condition in Experiment 3
(i.e., the Human-red blood version of Marathon 2) may not have been
suYciently realistic. Further work (not speculation) is needed on the realism
eVect.
Finally, Experiment 3 found additional correlational support for the link
between exposure to media violence and general level of trait physical
aggression. By themselves, such correlational data are not conclusive regard-
ing causality, but they lend support for the hypothesis that repeated exposure
to violent media leads to relatively high levels of trait aggressiveness.
VIII. Correlation Study 1
A. OVERVIEW
The primary purpose was to test the hypothesis that repeated exposure to
violent video games would be positively associated with aggressive behavior
and persistent aggressive cognitions (e.g., attitudes toward aggression), and
that these persistent aggressive cognitions would at least partially mediate
the violent video game exposureaggression link. A secondary purpose was
to examine associations between violent video game exposure and several
key personality indicators. Specically, we included Goldbergs Big Five
scales, the Narcissism scale, and Capraras Emotional Susceptibility scale.
If repeated exposure to violent video games does inuence basic personality
development, then associations should be found between such exposure and
personality indicators known to be related to habitual aggressive behavior
tendencies, such as Goldbergs agreeableness and conscientiousness factors,
narcissism, and emotional susceptibility. Of course, such associations could
reect the opposite causal direction. That is, basic personality factors favor-
ing aggressiveness might well predispose some individuals to enjoying and
playing violent video games. A cross-sectional correlation study of this type
cannot, by itself, conclusively rule out this alternative. However, this alter-
native further implies that any signicant correlations between violent video
232 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
game exposure and aggression should become nonsignicant when basic
personality factors are statistically controlled.
In sum, strong support for the hypothesis that repeated exposure to
violent video games increases aggressive behavior tendencies and does so
through changes in persistent aggressive cognitions would consist of three
parts. First, there must be signicant associations between violent video
game exposure and aggressive behavior tendencies. Second, these associa-
tions should persist even when basic personality factors are statistically
controlled. Third, these associations should be substantially reduced when
persistent aggressive cognitions are statistically controlled. Adequate testing
of these ideas requires a large sample size, so that lack of statistical power
does not lead to type I inferential errors.
B. METHOD
1. Participants
Three hundred seventeen male and 489 female college students at a
large midwestern university participated in a large mass testing question-
naire session for partial credit toward introductory psychology course
requirements.
2. Measures
a. Video Game Violence Exposure. Video game violence exposure (VGV)
was assessed using the same shortened version of the Anderson and Dill (2000)
measure described earlier in Experiment 3. Across the three video game items,
coeYcient alpha was .83, only slightly lower than the .86 alpha reported by
Anderson and Dill (2000) for their ve-game version of this measure.
b. Basic Personality. We measured seven basic personality factors, the
Big Five as well as two more specic factors that have been theoretically and
empirically related to aggression (narcissism and emotional susceptibility).
Goldbergs (1992) Big Five measure of basic personality structure consists of 20
items for each factor. CoeYcient alphas were: surgency (extraversion) ¼.87;
agreeableness ¼.91; conscientiousness ¼.89; emotional stability (neuroticism)
¼.80; intellect (openness) ¼.85. The 40-item Narcissism scale (Raskin & Terry,
1988) yielded an alpha of .84. The 27-item Emotional Susceptibility scale
(Caprara et al., 1985) yielded an alpha of .91.
c. Attitudes towards Violence. Two measures of attitudes toward vio-
lence were used. One was the recently revised 39-item Attitudes towards
Violence scale (ATVS; Anderson, Benjamin, Wood, & Bonacci, in press),
VIOLENT VIDEO GAMES 233
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:04 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
with an alpha of .92. The other was the 15-item Adolescent Attitudes
towards Violence scale (AATVS; Funk, Elliott, Urman, Flores, & Mock,
1999), with an alpha of .76.
d. Aggressive Behavior. Three measures of trait aggressive behavior were
used. The nine-item physical and the ve-item verbal aggression subscales of
the Aggression Questionnaire (Buss & Perry, 1992) yielded alphas of .85 and
.79, respectively. We also used the same standardized 10 physical aggression
items used by Anderson and Dill (2000) from the National Youth Survey
(Elliot, Huizinga, & Ageton, 1985), which includes behaviors that would be
considered criminal if known to police (e.g., assault, robbery). For ease of
exposition, the Buss-Perry physical aggression measure will be referred to as
mild physical aggression, whereas the National Youth Survey measure will be
referred to as severe physical aggression.
C. RESULTS AND DISCUSSION
1. Zero-order Correlations
As expected, VGV was positively related to each of the three aggression
measures and to both attitudes towards violence measures, as can be seen in
Table V. VGV was also negatively associated with the Big Five factors of
agreeableness and conscientiousness, as expected. VGV was also slightly
positively correlated with the Big Five factor of emotional stability. Finally,
VGV was positively correlated with narcissism and negatively correlated
with emotional susceptibility. There are a number of additional interesting
correlations in Table V, but because they are less central to the main
hypotheses we leave them for the reader to discover.
2. VGV–Aggression Link and Basic Personality
If the strong correlation between VGV and aggression is due solely to
spurious relationships with basic personality factors, such that aggressive
people also happen to like violent video games, then statistically controlling
for basic personality factors should eliminate the VGVaggression rela-
tionship. We used the destructive testing approach to assess this question
(Anderson & Anderson, 1996; Anderson & Dill, 2000). In this approach,
one rst determines whether a specic predicted relationship exists
(e.g., the VGVaggression correlation). If so, then theoretically meaningful
competitor variables (e.g., Big Five) are entered into the regression mode to
determine whether these competitors break the target relation. Of primary
interest is not whether the initial target link can be broken (i.e., made
234 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
nonsignicant), because even strong causal links between measured variables
can eventually be broken by adding more correlated competitors into
the model. Instead, the focus is the durability of the link, with consideration
given to the theoretical and empirical strength of the competitor variables.
Table VI presents the results of this destructive testing procedure, display-
ing the raw slopes linking VGV to each of the three aggression measures,
when only VGV is in the model (rst column of slopes) and when other
competitor variables are partialled out (columns 25). We rst added sex to
the model. The slopes decreased in magnitude but remained statistically signi-
cant.
8
In turn, we added the Big Five factors, then the narcissism measure,
and nally the emotional susceptibility measure. As shown in Table VI,
the VGVaggression link survived all of these competitor variables. These
TABLE V
Correlations Among Key Questionnaire Variables
Variable MPA VA SPA ATVS AATV B5Su B5Ag B5Co B5ES B5In NPI ES
VGV .312* .196* .166* .237* .321* .012 .161* .121* .076
y
.029 .159* .170*
MPA .438* .315* .478* .696* .012 .319* .202* .080
y
.043 .197* .070
y
VA ——.131* .209* .358* .180* .223* .078
y
.091
y
.055 .296* .068
SPA ———.250* .322* .003 .187* .130* .021 .100
y
.112
y
.039
ATVS ——— —.527* .097
y
.307* .137* .009 .220* .132* .007
AATV ——— — .068 .357* .235* .026 .108
y
.167* .045
B5Su ——— — .320* .198* .137* .380* .499* .098
y
B5Ag ——— — .607* .205* .489* .048 .066
B5Co ————.222* .415* .087
y
.146*
B5ES —————.055 .056 .566*
B5In ——————.262* .050
NPI ———————.212*
VGV, Video game violence exposure; MPA, mild physical aggression; VA,verbal
aggression; SPA, severe physical aggression; ATVS, Attitudes towards Violence scale; AATV,
Adolescent Attitudes towards Violence scale; B5Su, surgency; B5Ag, agreeableness; B5Co,
conscientiousness; B5ES, emotional stability; B5In, intellect; NPI, narcissism; ES, emotional
susceptibility.
*P<.001.
y
P<.05.
n¼806.
8
Because sex is highly correlated with exposure to violent video games, partialling out
sex eVects likely overcorrects for gender diVerences. Thus, the resulting estimate of the VGV
aggression relation is an extremely conservative one.
VIOLENT VIDEO GAMES 235
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
results support the view that the VGVaggression link is not a spurious
artifact of basic personality structure.
3. Persistent Aggressive Thoughts as a Mediator
Our next analysis used the same destructive testing approach, but here the
meaning is considerably diVerent because the ‘‘competitor’’ variables are
potential mediators of the long-term VGV eVect on aggressive tendencies.
Our application of GAM to the video game domain explicitly states that the
content of violent video games can create long-term changes in a host of
aggression-related knowledge structures, many of which can be indexed by
measures of general attitudes toward violence. Thus, we expected that add-
ing such attitude measures to the statistical model would result in substantial
reductions in the VGVaggression relation.
Once again, we began with a model having only VGV as the predictor, and
then added sex to the model (see columns 1 and 2 of Table VI). Next, we added
both attitudes toward violence measures to the model (instead of the person-
ality factors). As anticipated, the VGV aggression slope decreased substan-
tially for each of the three aggression measures. For mild physical aggression
the VGV slope dropped from a signicant .0098 (P<.005) to a nonsigni-
cant .0013 [F(1, 801) ¼0.28], an 87% decrease in the slope [(.0098 - .0013)/
.0098]. For verbal aggression the VGV slope dropped from a signicant
.0098 (P<.01) to a nonsignicant .0055 [F(1, 801) ¼2.44], a 44% decrease.
For severe physical aggression the VGV slope dropped from a signicant
.0058 (P<.02) to a nonsignicant .0031 [F(1, 801)1.96], a 47% decrease.
TABLE VI
Destructive Testing Results on the VGV/Aggression Relation
VGV slopes with various predictor variables in the model
Aggression measure VGV þSex þBig 5 þNPI þES
Mild physical aggression .0286* .0098
y
.0082
y
.0078
{
.0081
y
Verbal aggression .0182* .0098
y
.0075
{
.0068
{
.0071
{
Severe physical aggression .0097* .0058
{
.0056
{
.0053
{
.0055
{
df for Ftest of VGV eVect 1, 804 1, 803 1, 798 1, 797 1, 796
VGV, Video game violence exposure; Big Five,ve basic personality factors; NPI, narcissism
personality inventory; ES, emotional susceptibility.
*P<.001.
y
P<.01.
{
P<.05.
n¼806.
236 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
In sum, these results support the hypothesis that long-term eVects of
repeated exposure to violent video games on aggressive behavior tendencies
are at least partially mediated by changes in persistent aggressive cognitions.
Of course, as noted earlier the correlational nature of these results warrants
some interpretative caution and further integrative research.
IX. Updated Meta-Analysis
A number of new studies have become available since Anderson and
Bushmans (2001) meta-analysis of violent video game eVects. We added
all of the new studies we could locate to our database and made a
few procedural changes to address two specic questions. One question
concerned the possibility (frequently oVered by media representatives and
other critics) that the average eVect sizes reported in prior meta-analyses
might be inated by the inclusion of studies with potentially important
methodological weaknesses. In the present meta-analytic study, we identied
nine methodological weaknesses found in at least some violent video game
studies, and we categorized each sample as having none of them (the ‘‘Best
Practices’’ studies) or at least one (‘‘Not Best Practices’’ studies). We compared
the average eVect size results of this distinction. The second question concerned
potential diVerences between experimental studies (which allow stronger causal
statements on the basis of even a few studies) and correlational studies (which
typically involve more serious and realistic forms of aggression, and address the
long-term eVect issue).
A. METHODS
1. Study Sample
A complete list of included studies can be found at the following web page:
http://www.psychology.iastate.edu/faculty/caa/abstracts/20002004/04AESPref.pdf.
We included all studies that we could locate that had data testing a possible
link between exposure to violent video games and one of ve types of
outcome variables: aggressive behavior (dened as behavior intended to
harm another person), aggressive cognition, aggressive aVect, helping be-
havior, and physiological arousal. A given ‘‘study’’ might contain more than
one independent ‘‘sample’’ of research participants. For example, some
studies reported results separately for male and female participants. We used
one eVect size for each sample for each of the available ve dependent
VIOLENT VIDEO GAMES 237
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
variables. For example, if a sample had three diVerent (and valid) measures
of aggressive behavior and a composite of the three, we used the composite
measure. If a composite could not be obtained, we used the average of the
three separate eVect sizes.
2. Best Practices Coding
The following potential methodological problems were examined for each
sample:
1. Nonviolent video game condition contained violence, and there was no
suitable nonviolent control condition.
2. Violent video game condition contained little or no violence.
3. In a correlational study, the measure of video game exposure was not
specically tied to violent video games (e.g., the amount of time spent
on any kind of video game was measured instead of time spent on
violent video games).
4. Evidence that the violent and nonviolent conditions diVered signi-
cantly in ways that could contaminate the conditions, such as the
nonviolent condition being more diYcult, boring, or frustrating than
the violent condition.
5. A pre-post design was used, but only the average of the pre- and
postmanipulation measures was reported.
6. Each researc h session involved both a video game player and an observer,
but only the average of the player-observer measures was reported.
7. The aggressive behavior measure was not aggression against another
person (e.g., aggression against a non-human character, or against objects).
8. The outcome variable was physiological arousal, but arousal diVer-
ences between the violent and nonviolent video game conditions were
already controlled by pretesting, game selection, or both (i.e., equally
arousing violent and nonviolent games were intentionally chosen by the
researchers to control for potential arousal eVects on other outcome
measures such as aggressive behavior).
9. The outcome variable was aggressive aVect, but aVective diVerences
between the violent and nonviolent video game conditions were already
controlled by pretesting, game selection, or both (i.e., violent and
nonviolent games were intentionally chosen by the researchers to have
the same aVective impact, to control for potential aVective inuences
on other outcome measures such as aggressive behavior).
Some of these ‘‘weaknesses’’ are actually strengths for other aspects of the
same research. For example, if one wants to study whether violent video game
238 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
content (relative to a nonviolent video game) can increase aggressive behavior
even when there are no arousal diVerences between the games, pretesting and
selecting violent and nonviolent video games that produce equivalent levels of
arousal is an excellent methodological feature (as in our Experiments 13).
However, that same sample does not allow a good test of whether violent
video games on average increase arousal. Thus, for aggressive behavior this
sample would be coded as a ‘‘best practice’’ one, whereas it would be coded as
a‘‘not best practice’’ sample for physiological arousal.
For several samples it was possible to get eVect sizes for both a best
practices procedure and a not best practices procedure on the same outcome
variable. For example, several correlational studies reported both a best
practices measure of time spent on violent video games and a not best practices
measure of time spent on any type of video game.
B. RESULTS AND DISCUSSION
1. Best vs. Not Best Practices
Figure 7 presents the best vs. not best practices results. In each case the
methodologically best samples yielded average eVect sizes that were larger
than methodologically weaker samples. This was especially pronounced
for aggressive behavior and aggressive aVect, wherein the 95% condence
intervals for the best and not best samples did not overlap. These results
suggest that eVect size estimates that include methodologically weaker stud-
ies (e.g., Anderson & Bushman, 2001) underestimate the true eVect sizes of
exposure to violent video games.
2. Best Practices Samples: Experimental vs. Correlational
Figure 8 presents the average eVect sizes of the best practices samples
categorized by type of study. There are no consistent diVerences in eVect
sizes for the experimental versus correlational samples. Correlational studies
yielded a slightly larger average eVect on aggressive and helping behavior
than did experimental studies, whereas the opposite was true for aggres-
sive cognition and aVect. In all four of these cases the experimental and
correlational 95% condence intervals overlap. Furthermore, both the
experimental and correlational eVect sizes were signicantly diVerent from zero
for each outcome variable except physiological arousal, and that was because
there are no best practices correlational studies of this type. In sum, despite
the relatively small size of this research domain, there is considerable correla-
tional and experimental evidence linking exposure to violent video games with
increases in aggressive behavior and to several aggression-related variables.
VIOLENT VIDEO GAMES 239
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
X. General Discussion
These ve studies (three experiments, one correlational study, one meta-
analytic study) contribute to our understanding of human aggression from
both a personality processes and a situational eVects perspective. We believe
these results can best be understood within the GAM theoretical framework
described earlier.
A. SITUATIONAL EFFECTS
The main situational nding was that brief exposure to violent video
games increased aggressive behavior relative to nonviolent video games
matched on arousal and aVective dimensions. This occurred in both experi-
ments that measured aggression, regardless of whether the violent game had
Fig.7. EVects of violent video games on aggressive behavior, aggressive cognition, aggres-
sive aVect, helping behavior, and physiological arousal by best practices methodology. rþ,
average eVect size; K, number of independent samples; N, total number of participants. Vertical
capped bars are the upper and lower 95% condence intervals.
240 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
green-blooded aliens or red-blooded humans as the enemy targets,
and occurred in both versions of the CRT task. The lack of reliable sex X game
violence interactions suggests that the eVect was similar in men and women.
A second situational nding concerns the cognitive eVects of violent
video games. Experiment 1 demonstrated that violent games in general
produce increases in the relative accessibility of aggressive thoughts. The
present ndings make a very strong case for the hypothesis that violent
video games can (and do) cause increases in aggression because of the
violent content of such games, not just because of their arousal or aVective
properties. The present empirical results in combination with our theoretical
analysis also lend support to the concern that repeated exposure to violent
video games (or other violent media) might lead to development of an
increasingly aggressive personality, and that much of this developmental eVect
may be the direct result of the violent content. In short, repeatedly thinking
Fig.8. EVects of violent video games on aggressive behavior, aggressive cognition, aggres-
sive aVect, helping behavior, and physiological arousal by type of study for best practices
samples. r+, average eVect size; K, number of independent samples; N, total number of
participants. Vertical capped bars are the upper and lower 95% condence intervals.
VIOLENT VIDEO GAMES 241
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
about violent characters, choosing to be aggressive, enacting that aggressive
choice, and being rewarded for it can be conceived as a series of learning trials
inuencing a variety of types of aggressive knowledge structures. Violent
video games may well teach players to become more aggressive people.
One interesting diVerence between the results of Experiments 2 and 3
concerns revenge motivation. There was little evidence that the video
game eVect in Experiment 2 was mediated by an increase in desire for revenge,
but there was considerable evidence that much of the video game eVect was
mediated by revenge motivation in Experiment 3. The various procedural
diVerences between the two studies may account for these diVerences, particu-
larly the timing of various measurements in the two-phase versus the standard
one-phase CRT task. In any case, it may be useful in future research to explore
the possibility that the violent content of violent video games may increase
aggression by rst priming aggressive cognitions, which in turn increase desire
for revenge when mildly provoked.
B. PERSONALITY EFFECTS
Experiments 2 and 3 yielded several ndings of general interest from a
personality perspective. First, trait hostility (Experiment 2) and trait physi-
cal aggression (Experiment 3) were positively related to aggression in the
CRT task. This further validates these trait measures and the two versions of
the CRT task. Second, both of these eVects were largely mediated by revenge
motivation. Thus, it appears that one way in which highly hostile people are
predisposed to be aggressive against others is through increased revenge
motives that are aroused when mildly provoked. We believe that this is the
rst demonstration that such trait eVects on aggressive behavior operate
through increases in desire for revenge.
A third nding concerns the question of who is most susceptible to violent
video game eVects. There is some evidence from the television/movie violence
literature that highly aggressive people tend to be more strongly inuenced
by exposure to violent media than nonaggressive people (e.g., Bushman &
Huesmann, 2001). Such person situation interactions do not always occur
in the media violence literature; sometimes the opposite pattern occurs (e.g.,
Anderson, 1997). Also, even when this pattern occurs, it is not the case that
nonaggressive people are entirely unaVected. A common claim of skeptics
and media industry representatives is that only a very few disturbed in-
dividuals might be aVected at all. In the present studies neither trait hosti-
lity (Experiment 2) nor trait physical aggressiveness (Experiment 3)
interacted with the violent video game manipulation. That is, the violent
video game eVect on aggression was not reliably bigger (or smaller) for those
242 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:05 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
participants who scored high on these traits than for those who scored low.
Along these same lines, we did not obtain signicant sex video game
violence interactions on aggressive behavior. We suspect that with larger
sample sizes such person situation interactions will emerge in some
contexts. Nonetheless, the lack of such interactions in the present studies
suggests that violent video games inuence a sizeable proportion of people.
A fourth nding of interest, from Experiment 2, was that experience with
violent video games correlated positively (and signicantly) with both re-
venge motivation and instrumental aggressive motivation, whereas experi-
ence with nonviolent video games did not. Although Experiment 2 was not
explicitly designed to test these relations, it is interesting that this pattern ts
exactly what would be expected if repeated exposure to violent video games
does create more aggressive individuals. It also ts with prior research
designed to test such eVects (e.g., Anderson & Dill, Study 1, 2000).
Fifth, the correlational ndings of Experiment 3 that general media vio-
lence exposure is positively associated with trait physical aggression, even
when time spent on electronic entertainment and sex were statistically
controlled, support a long and increasingly strong line of research on media
violence eVects (e.g., Bushman & Anderson, 2001).
The correlational study provided support for the hypothesized link be-
tween repeated exposure to violent video games and increased aggressive
tendencies, and did so for three types of aggression: mild physical, verbal,
and severe physical aggression. This study also provided the rst correla-
tional support for the contention that such long-term eVects on aggressive
behavior are mediated by persistent aggressive thoughts, here indexed by
two diVerent attitudes toward violence measures. Furthermore, this study
provided the rst correlational evidence that the violent video game expo-
sure link to aggression persists even when a host of basic personality factors
are statistically controlled.
Finally, the meta-analysis further revealed consistent eVects of violent video
game exposure on aggressive behavior, cognition, and aVect, as well as on
arousal and prosocial behavior. The fact that these eVects were stronger in the
methodologically strongest studies, and occurred in both experimental and
correlation designs (with the exception of arousal), lends further support for
our application of GAM to both long- and short-term media violence eVects.
C. IMPLICATIONS
There are many questions requiring additional empirical work. One is the
need to identify specic features of violent video games that increase or decrease
their impact on aggressive thoughts, feelings, and behaviors. Experiment 3
VIOLENT VIDEO GAMES 243
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:06 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
provided one test of the hypothesis that the realism or ‘‘humanness’’ of the game
target might exacerbate the eVect. It yielded no support for that hypothesis.
Further tests using more extremely realistic and gory graphics are needed.
A second question concerns the long-term eVects of repeated exposure to
violent video games, especially on children and teens. Based on over 40 years
of research on television and movie violence, one reasonable expectation is
that repeatedly exposing youth to violent video games over a period of years
will have a sizeable negative impact on their development. Indeed, there is
reason to believe that the video game violence eVect will be larger than
television violence eVects because of the highly engaging and active nature
of video games compared with the relatively passive nature of watching TV.
Nonetheless, longitudinal research is badly needed to test this prediction and
to delineate protective and exacerbating factors.
A third set of questions concerns possible positive eVects of games de-
signed to promote prosocial behaviors. Do such games increase prosocial
and decrease antisocial behavior? Virtually no research exists on this topic.
However, video games are going to remain a major source of entertainment.
Therefore we believe it is important to oVer empirical evidence and quality
theory on which types of features promote a prosocial gaming experience, as
well as highlighting the potential antisocial eVects of games with violent
themes.
XI. Appendix
A. VIOLENT GAME DESCRIPTIONS
Dark Forces. This is a standard rst-person shooter. The player as-
sumes the role of a special ops guy in the Rebellion with the objectives
of stealing the Death Star plans and getting out alive. This game has a
fairly high level of violence, with weapons like a blaster rie and laser
pistol to kill enemy guards and stormtroopers.
Marathon 2. This is a standard rst-person shooter, in which the player
assumes the role of a space marine trapped in a base that has been taken
over by aliens. Your main goal is to retake it and not die. This game has a
high violence level, with the basic, underlying premise of the game being
to shoot anything that moves and kill or be killed. The targets are mainly
aliens, with some ‘‘compilers’’ or robots that the aliens use as slaves.
Speed Demon. This is a 3D combat driving game. The player drives a
heavily armed vehicle in a race with other similarly armed vehicles,
244 CRAIG A. ANDERSON et al.
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:06 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
shooting at and crashing into them as they do likewise. One gets points
for destroying other vehicles.
Street Fighter. This is a third-person ghting game, similar in many
ways to Mortal Kombat. The player chooses a character and then
engages in a series of ghts with other characters. Each character has
specic strengths and weaknesses.
Wolfenstein 3D. This is a rst-person shooter. The player assumes the
role of B. J. Blascowitz, an American soldier caught and taken prisoner
trying to inltrate a top-secret Nazi experimentation lab. The goal is to
shoot your way out of the prison and kill everything that moves. You
can pick up various weapons, including various guns. There is a very
high violence rate, with you shooting dogs and Nazi guards, with gory
bullet hits and ‘‘death poses.’’
B. NONVIOLENT GAME DESCRIPTIONS
3D Ultra Pinball. This is simply an electronic version of a pinball game,
complete with ippers, buzzers, bells, and various visual and auditory eVects.
Glider Pro. Players of this game control the forward and backward
motion of paper airplanes through a house. By ying over air ducts one
can gain lift. The glider can turn various items on and oV(such as light
switches, computers). One can earn points by ying over certain objects.
If the glider hits the oor or certain other items, it crumples and is
replaced by another glider.
Indy Car 2. The player assumes the role of driver in an Indy car race
with the goal of winning the race. If you bump other cars or drive too
fast on turns, you crash. This game can be played from the keyboard,
but is much easier with a steering wheel and pedals.
Jewel Box. This game is a colorful version of Tetris. Variously colored
shapes drop from the top of the screen. The player manipulates the
objects as they fall, trying to create lled rows. When a row is complete-
ly lled, it disappears and the player receives points.
Myst. This is a nonviolent exploration/mystery/adventure game with a
rst-person perspective. It begins on the Island of Myst. Players
begin by appearing on the island with no knowledge of how or
why they ended up there, or how to proceed. In an ancient library,
players discover two mysterious books, which lay out the basic mystery.
Players must then travel to several diVerent worlds (‘‘ages’’) to unravel
the mystery.
VIOLENT VIDEO GAMES 245
Comp. by:Saravanakumar Date:1/6/04 Time:18:27:06 Stage:1st Revises
File Path://FSC/Serials/SerialsProdEnv/SERIALSWOMAT/ASE/036/REVISES1/
45-3B2QCCORRECTIONS/3B2PS/ASE_036_C004.3D Proof by:Mary M.
QC by:Celestine J.
Acknowledgments
We thank Mary Ballard, Brad Bushman, and Jeanne Funk for their helpful comments on an
earlier version of this work.
References
Anderson, C. A. (1983). Imagination and expectation: The effect of imagining behavioral
scripts on personal intentions. Journal of Personality and Social Psychology, 45, 293305.
Anderson, C. A. (1997). Effects of violent movies and trait irritability on hostile feelings and
aggressive thoughts. Aggressive Behavior, 23, 161178.
Anderson, C. A. (2000). Violent video games increase aggression and violence. U.S. Senate
Committee on Commerce, Science, and Transportation hearing on ‘‘The Impact of Interactive
Violence on Children.’’ Tuesday, March 21, 2000. Available at: http://www.psychology.
iastate.edu/faculty/caa/abstracts/19951999/96AA.html.
Anderson, C. A., & Anderson, K. B. (1996). Violent crime rate studies in philosophical context:
A destructive testing approach to heat and southern culture of violence effects. Journal of
Personality and Social Psychology, 70, 740756.
Anderson, C. A., Anderson, K. B., Dorr, N., DeNeve, K. M., & Flanagan, M. (2000).
Temperature and aggression. In M. Zanna (Ed.), Advances in Experimental Social
Psychology (Vol. 32, pp. 63133). New York: Academic Press.
Anderson, C. A., Benjamin, A. J., Wood, P. K., Bonacci, A. M. (in press). Development and
testing of the attitudes toward violence scale: Evidence for a four-factor model. Aggressive
Behavior.
Anderson, C. A., & Bushman, B. J. (1997). External validity of ‘‘trivial’’ experiments: The case
of laboratory aggression. Review of General Psychology, 1, 1941.
Anderson, C. A., & Bushman, B. J. (2002). Human aggression. Annual Review of Psychology,
53, 2751.
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 meta-analytic review of the scientic literature. Psychological Science, 12, 353359.
Anderson, C. A., & Carnagey, N. L. (2004). Violent evil and the general aggression model. In A.
Miller (Ed.), The social psychology of good and evil (pp. 168192). New York: Guilford
Publications.
Anderson, C. A., Carnagey, N. L., & Eubanks, J. (2003). Exposure to violent media: The effects
of songs with violent lyrics on aggressive thoughts and feelings. Journal of Personality and
Social Psychology, 84, 960971.
Anderson, C. A., & Dill, K. E. (2000). Video games and aggressive thoughts, feelings, and
behavior in the laboratory and in life. Journal of Personality and Social Psychology, 78,
772790.
Anderson, C. A., & Ford, C. M. (1986). Affect of the game player: Short term effects of highly and
mildly aggressive video games. Personality and Social Psychology Bulletin, 12, 390402.
Anderson, C. A., & Godfrey, S. (1987). Thoughts about actions: The effects of specicity and
availability of imagined behavioral scripts on expectations about oneself and others. Social
Cognition, 5, 238258.
246 CRAIG A. ANDERSON et al.
VIOLENT VIDEO GAMES 247
Anderson, C. A., & Huesmann, L. R. (2003). Human aggression: A social cognitive view. In M.
Hogg & J. Cooper (Eds.), Handbook of social psychology (pp. 296-323). London: Sage
Publications.
Anderson, C. A., Lindsay, J. J., & Bushman, B. J. (1999). Research in the psychological
laboratory: Truth or triviality? Current Directions in Psychological Science, 8, 3-9.
Bartholow, B. D., & Anderson, C. A. (2002). Examining the effects of violent video games on
aggressive behavior: Potential sex differences. Journal of Experimental Social Psychology, 38,
283 290.
Berkowitz, L., & Donnerstein, E. (1982). External validity is more than skin deep: Some answers
to criticism of laboratory experiments. American Psychologist, 37, 245-257.
Buchman, D. D., & Funk, J. B. (1996). Video and computer games in the '90s: Children's time
commitment and game preference. Children Today, 24, 12-16.
Bushman, B. J., & Anderson, C. A. (2001). Media violence and the American public: Scientific
facts versus media misinformation. American Psychologist, 56, 477-489.
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.
Bushman, B. J., & Huesmann, L. R. (2001). Effects of televised violence on aggression. In D.
Singer & J. Singer (Eds.), Handbook of children and the media (pp. 223-254). Thousand Oaks,
CA: Sage Publications.
Buss, A. H., & Perry, M. (1992). The aggression questionnaire. Journal of Personality and Social
Psychology, 63, 452-459.
Calvert, S. L., & Tan, S. (1994). Impact of virtual reality on young adults' physiological arousal
and aggressive thoughts: Interaction versus observation. Journal of Applied Developmental
Psychology, 15, 125-139.
Caprara, G. V., Cinanni, V., D'Imperio, G., Passerini, S., Renzi, P., & Travaglia, G. (1985).
Indicators of impulsive aggression: Present status of research on irritability and emotional
susceptibility scales. Personality and Individual Differences, 6, 665-674.
Carlson, M., Marcus-Newhall, A., & Miller, N. (1989). Evidence for a general construct of
aggression. Personality and Social Psychology Bulletin, 15, 377-389.
Carlson, M., Marcus-Newhall, A., & Miller, N. (1990). Effects of situational aggression cues: A
quantitative review. Journal of Personality and Social Psychology, 58, 622-633.
Carnagey, N. L., Bushman, B. J., Anderson, C. A. (under review). Video game violence
desensitizes players to real world violence.
CIRP (1998, 1999). Cooperative Institutional Research Program Survey Results. Ames, IA: Office
of Institutional Research.
Crick, N. R., & Dodge, K. A. (1994). A review and reformulation of social information processing
mechanisms in children's social adjustment. Psychological Bulletin, 115, 74-101.
Dill, K. E., & Dill, J. C. (1998). Video game violence: A review of the empirical literature.
Aggression and Violent Behavior.- A Review Journal, 3, 407-428.
Dill, K. E., Anderson, C. A., Anderson, K. B., & Denser, W. E. (1997). Effects of aggressive
personality on social expectations and social perceptions. Journal of Research in Personality,
31, 272-292.
Elliot, D. S., Huizinga, D., & Ageton, S. S. (1985). Explaining delinquency and drug use. Beverly
Hills. CA: Sage.
Federal Trade Commission. (2000). Marketing violent entertainment to children: A review of self-
regulation and industry practices in the motion picture, music recording, & electronic game
industries. Report of the Federal Trade Commission. Available at: www.ftc.gov/reports/
violence/.
248 CRAIG A. ANDERSON et al.
Funk, J. B., Elliott, R., Urman, M. L., Flores, G. T., & Mock, R. M. (1999). The Attitudes
Towards Violence scale: A measure for adolescents. Journal of Interpersonal Violence, 14,
1123-1136.
Funk, J. B., Flores, G., Buchman, D. D., & Germann, J. N. (1999). Rating electronic games:
Violence is in the eye of the beholder. Youth & Society, 30, 283-312.
Gentile, D. A., Lynch, P. J., Linder, J. R., Walsh, D. A. (2004). The effects of violent video game
habits on adolescent hostility, aggressive behaviors, and school performance. Journal of
Adolescence, 27, 5-22.
Giancola, P. R., & Chermack, S. T. (1998). Construct validity of laboratory aggression
paradigms: A response to Tedeschi and Quigley (1996). Aggression and Violent Behavior, 3,
237-253.
Goldberg, L. R. (1992). The development of markers for the Big-Five factor structure.
Psychological Assessment, 4, 26-42.
Graybill, D., Strawniak, M., Hunter, T., & O'Leary, M. (1987). Effects of playing versus
observing violent versus nonviolent video games on children's aggression. Psychology: A
Quarterly Journal of Human Behavior, 24, 1-8.
Hearold, S. (1986). A synthesis of 1043 effects of television on social behavior. In G. Comstock
(Ed.), Public communication and behavior (Vol. 1, pp. 65-133). New York: Academic Press.
Huesmann, L. R., & Miller, L. S. (1994). Long-term effects of repeated exposure to media
violence in childhood. In L. R. Huesmann (Ed.), Aggressive behavior: Current perspectives
(pp. 153-186). New York: Plenum Press.
Kirsh, S. J. (1998). Seeing the world through Mortal Kombat-colored glasses: Violent video
games and the development of a short-term hostile attribution bias. Childhood, 5, 177-184.
Kruglanski, A. W. (1975). The human subject in the psychology experiment: Fact and artifact. In
L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 8, pp. 101-147). New
York: Academic Press.
Lindsay, J. J., & Anderson, C. A. (2000). From antecedent conditions to violent actions: A
general affective aggression model. Personality and Social Psychology Bulletin, 26,
533-547.
Marsh, R. L., Hicks, J. L., & Bink, M. L. (1998). Activation of completed, uncompleted, and
partially completed intentions. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 24, 350-361.
Paik, H., & Comstock, G. (1994). The effects of television violence on antisocial behavior: A
meta-analysis. Communication Research, 21, 516-546.
Raskin, R., & Terry, H. (1988). A principal-components analysis of the narcissistic personality
inventory and further evidence of its construct validation. Journal of Personality and Social
Psychology, 54, 890-902.
Roediger, H. L., Weldon, M. S., Stadler, M. L., & Riegler, G. L. (1992). Direct comparison of
two implicit memory tests: Word fragment and word stem completion. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 18, 1251-1269.
Sherry, J. L. (2001). The effects of violent video games on aggression: A meta-analysis. Human
Communication Research, 27, 409-431.
Tukey, J. W. (1977). Exploratory data analysis. Menlo Park, CA: Addison-Wesley.
Walsh, D. (1999). 1999 Video and Computer Game Report Card. Minneapolis: National Institute
on Media and the Family. http://www.mediaandthefamily.org/1999vgrc2.html.
Walsh, D. (2000). Interactive violence and children: Testimony submitted to the Committee on
Commerce, Science, and Transportation, March 21, 2000. Downloaded from http://
www.mediaandthefamily.org/senateviolence-full.html on March 24, 2000.
VIOLENT VIDEO GAMES 249
Wells, G. L., & Windschitl, P. D. (1999). Stimulus sampling in social psychological
experimentation. Personality and Social Psychology Bulletin, 25, 1115-1125.
Winer, B. 1. (1971). Statistical principles in experimental design. New York: McGraw-Hill.
Wood, W., Wong, F. Y., & Chachere, J. G. (1991). Effects of media violence on viewers'
aggression in unconstrained social interaction. Psychological Bulletin, 109, 371-383.
... Although this interaction between trait aggressiveness and aggressive behavior has been replicated several times (see Anderson & Dill, 2000;Arriaga et al., 2006;Bartholow et al., 2005;Gentile et al., 2004;Giumetti & Markey, 2007;Saleem et al., 2012), there are also some cases in which the interaction was not significant (Anderson et al., 2004;Carnagey et al., 2007;Gentile et al., 2009;Konijn et al., 2007). For instance, Anderson et al. (2004) measured trait aggressiveness and then randomly assigned both male and female participants to play either a violent or a nonviolent video game. ...
... Although this interaction between trait aggressiveness and aggressive behavior has been replicated several times (see Anderson & Dill, 2000;Arriaga et al., 2006;Bartholow et al., 2005;Gentile et al., 2004;Giumetti & Markey, 2007;Saleem et al., 2012), there are also some cases in which the interaction was not significant (Anderson et al., 2004;Carnagey et al., 2007;Gentile et al., 2009;Konijn et al., 2007). For instance, Anderson et al. (2004) measured trait aggressiveness and then randomly assigned both male and female participants to play either a violent or a nonviolent video game. Next, all participants played a competitive reaction time game in which they could punish their opponent by delivering a noxious blast of white noise (the aggression outcome). ...
... Moreover, we tested two video games in the present research, but future studies should be undertaken to generalize these findings to other video games (see Wells & Windschitl, 1999). In addition, future research might benefit from identifying more accurately what is the minimum duration of playing needed for each effect to emerge (Anderson et al., 2004) and could benefit from including This document is copyrighted by the American Psychological Association or one of its allied publishers. ...
Full-text available
Article
Objective: The present research examined whether trait aggressiveness was more associated with aggressive behavior in relevant situations (playing a high-violence video game as compared to a lowviolence video game) and when participants had an agent (perpetrator) rather than a victim role. Method: Two studies were conducted with female undergraduate students. In Study 1, female participants first reported their level of trait aggressiveness. After completing the scale, participants were randomly assigned to a behavioral manipulation of the agent or victim role that required participants to either give or receive noise blasts. Afterward, they were randomly assigned to play a high or low-violence video game. Finally, we assessed a measure of aggression as the dependent variable. Study 2 used a similar design and procedure and was intended to generalize the results to behavioral intentions of aggression using a priming task of the role manipulation through an imagination task. Results: Across two studies, playing a highversus low-violence video game moderated the relationship between trait aggressiveness and aggressive behavior only when female participants were cast in a role relevant to the trait (agent) but not when the role was less relevant to trait aggressiveness (victim). Conclusions: Trait aggressiveness was most predictive of aggressive behavior when female participants adopted an agent (perpetrator) role and also played a high-violence video game. This research supports the idea that women acted on their traits more when the situation and the role are more relevant to the trait, because the trait seems like a more valid basis of behavior.
... Social interaction relies on the involvement of others while the attentions of engagement depict the combined action with other members of society (Vivek et al., 2014). With regard to violent videogames, whenever a player perceives a certain difficulty which hinders their enjoyment and engagement, it increases their tendency to be aggressive (Anderson et al., 2004). While the violence in the game can be in the form of animosity or a composite of physical and verbal aggression , it is a significant causal factor for physically aggressive behavior in its consumers (Anderson et al., 2008). ...
Full-text available
Article
Debate on violent games and their effect on aggressive behavior remains inconclusive. This study aims to study the predicting role of cognitive, affective, and behavioral engagement states in violent videogames on aggressive behavior, which remains nebulous to date. We visited gaming zones and administered the study survey to collect data from violent videogame users. We collected 208 valid responses that were further analyzed. The present study used SmartPLS (3.3.3) software to perform partial least squares structural equation modeling (PLS-SEM) analysis in two stages. In the first stage, the measurement model assessment reported that cognitive, affective, behavioral, and aggressive behavior proved to be reliable reflective-formative composite constructs. Whereas, the second phase illustrated that cognitive engagement in violent videogames fails to impact aggressive behavior. The other two engagement states (affective and behavioral) in violent games showed a positive impact on aggressive behavior. Our study contributes to aggressive behavior literature by understanding how violent videogame engagement states impact aggressive behavior, which is crucial to recognize aggression so that steps can be taken toward addressing it. This study also contributes methodologically by utilizing the hierarchical component model (HCM) approach to estimate, specify, and validate the hierarchical structure of higher-order constructs (i.e., consumer violent videogame engagement dimensions (cognitive, affective, and behavioral) and aggressive behavior) as reflective-formative composite models.
... Some researchers believe that playing video games as complex technological and social phenomena can incur negative because it keeps the individual from playing creative games, teaches wrong beliefs and behaviors, and takes away the opportunity to develop better motor skills (Anderson et al., 2004;Hamlen, 2009). However, some research has highlighted the positive effects of these games. ...
Full-text available
Article
Background In recent years, interest in digital games has grown significantly and at the same time a range of games in 2D and 3D has been created. While 2D games cannot give depth to objects, 3D games creates spatial depth. The growth of the gaming industry has raised concerns among some psychologists. There is still much controversy about positive or negative effects of these games on cognitive performance. Executive functions are a set of cognitive abilities that play a decisive role in one’s purposeful behavior. The present study aims to investigate the effect of 2D and 3D video games on some of these functions—flexibility, inhibition, and continuous attention—in male students aged 18–26 years. Method Forty-five male students at Persian Gulf University participated in the study. They were matched based on intelligence score and experience of playing video games. The participants were divided into three groups of 15 (two experimental groups and one control group). Participants in the two experimental groups played 2D and 3D video games over 20 sessions. They completed the Wisconsin Card Sorting Test, continuous performance test, Stroop test, and Raven intelligence test. Results Data were analyzed via covariance and the results showed that playing 2D and 3D video games positively and significantly affects inhibition and playing 3D games has a significant positive effect on flexibility. Conclusion Based on our findings, we discuss the different effects of 2D and 3D video games, in line with the study’s theoretical framework, and give suggestions for future studies.
... The results from this study indicate statistical significance of board size on accounting information variables and stock market return. It is seen that increasing the size of the board improves stock market return which supports Anderson et al. (2004) but contrary to Jensen (1993). The findings from this study also support Fama and Jensen (1983) who argued that the role of the board involves monitoring managerial behaviour, which is likely to be more effective with a large board size. ...
Article
Cyber‐aggression is a serious social problem worldwide. Its risks have been frequently explored, and violence exposure in daily life has been regarded as an important risk factor of cyber‐aggression. However, the longitudinal association between violence exposure in daily life and cyber‐aggression has not yet been examined, and the mechanisms underlying the link between violence exposure and cyber‐aggression remain largely unclear. Based on the General Aggression Model and Script Theory, we explored the circular relation between violence exposure in daily life, hostile automatic thoughts, and cyber‐aggression. The current study adopted a longitudinal design to address these issues among 941 college students. The results indicated violence exposure in daily life predicted hostile automatic thoughts and cyber‐aggression 6 months later; hostile automatic thoughts predicted violence exposure and cyber‐aggression 6 months later; and cyber‐aggression predicted hostile automatic thoughts and violence exposure 6 months later. Moreover, each of them plays a mediating role in the association between the other two variables. These results support and expand the General Aggression Model and Script Theory that violence exposure, aggressive cognition, and aggression facilitate each other. This also provides theoretical guidance on reducing cyber‐aggression in daily life.
Full-text available
Article
The goal of this study is to offer an overview of the trends that characterize the current research on prosocial media effects. To this end, we conducted a systematic literature review. The period reviewed was from 2017 to 2021. It found four general trends: The first one is related to research on prosocial children´s media. The second deals with the examination of the effects of mixing prosocial and violent content and practices in the same media text. The third reflects the importance of exploring what happens during media use in relation to user behaviors above all in multiplayer video games social dynamics. Finally, the fourth trend found is related to the exploration of the possibilities opened up by immersive media.
Chapter
Exposure to media violence can increase the likelihood of aggressive and violent behavior, and also lead to other harmful effects on youth as well –including children's play with real guns, risk glorification, sexual violence, stereotyping of gender and minorities, and decline in prosocial behavior. The psychological processes underlying harmful media violence effects are explained, using the General Aggression Model. Appropriate interventions by parents and school systems, and others are explored, to promote more informed parent, caregiver, policymaker, and consumer decisions.
Full-text available
Chapter
Bu bölümde; dijital oyunların kişilik, duygu ve davranışlar üzerindeki etkileri ele alınmış, dijital oyun oynama alışkanlıklarının oluşmasında ve devam etmesinde rol oynayan, güdüleyen unsurlar, özellikle de psikolojik ihtiyaçlar çerçevesinden teorik olarak değerlendirilmiştir. Dijital oyun oynama alışkanlıklarının kişilik üzerindeki olası etkileri ve tartışmalı bir konu olan dijital oyunlar ve saldırganlık ilişkisi ele alınarak dijital oyunlar ve kişilik arasındaki ilişkiler karşılıklı olarak incelenmiştir. Dijital oyun oynamanın olumsuz ve olası olumlu sonuçları ele alınmış, bu konudaki tartışmalar gözden geçirilmiştir. Sonuç kısmında ise dijital oyunların olumsuz sonuçlarını azaltmak için bağımlılık düzeyindeki dijital oyun oynama alışkanlıklarına dair yapılması gerekenlerden söz edilmiştir ve bu konularla ilgili olarak çocuklara, gençlere, yetişkinlere, anne babalara ve öğretmenlere yönelik önerilere yer verilmiştir.
Article
The continuous growth of production and the development of new video games have influenced many aspects of human lives, particularly the young generation. While video games may tap on the entertainment side, primarily aiming to entertain players, they may profoundly impact multi-literacy development. This qualitative study utilizes narrative inquiry, which involves collecting the data from the experiences of a teenager―a native video gamer―who has spent a significant amount of time since childhood playing video games. The data were collected from diverse sources across time, namely individual interviews, phone calls, verbal protocols, diaries, and WhatsApp texting. In this study, the teenager reveals how the pleasure of video games resulted in developing multi-literacy and learning the English language subconsciously. An analysis of the narrative provides insights into how playing video games results in multi-literacy development and impacts the native video gamer’s cognition, affection, language learning, and cultural knowledge.
Full-text available
Article
The external validity of artificial "trivial" laboratory settings is examined. Past views emphasizing generalizability of relations among conceptual variables are reviewed and affirmed. One major implication of typical challenges to the external validity of laboratory research is tested with aggression research: If laboratory research is low in external validity, then laboratory studies should fail to detect relations among variables that are correlated with aggression in "real-world" studies. Meta-analysis was used to examine 5 situational variables (provocation, violent media, alcohol, anonymity, hot temperature) and 3 individual difference variables (sex, Type A personality, trait aggressiveness) in real-world and laboratory aggression studies. Results strongly supported the external validity of trivial laboratory studies. Advice is given on how scholars might handle occasional descrepancies between laboratory and real-world findings.
Full-text available
Article
Past research has shown that simply imagining oneself deciding to perform or refusing to perform a target behavior produces corresponding changes in expectations about oneself, whereas imagining someone else has no impact on expectations about oneself (Anderson, 1983b). The present experiment further examined this specificity effect and the proposed underlying mechanism. It was hypothesized that imagining oneself in a behavioral script would influence expectations about oneself, but not expectations about another person, and that imagining another person in a behavioral script would influence expectations about that person's behavior, but not expectations about one's own behavior. Furthermore, it was hypothesized that previously imagined scripts would be easier to imagine (i.e., more available), and that ease of imagination would be correlated with behavioral expectations. All hypotheses were supported. The role of imagined scripts in a variety of decision domains is discussed.
Article
A new questionnaire on aggression was constructed. Replicated factor analyses yielded 4 scales: Physical Aggression, Verbal Aggression, Anger, and Hostility. Correlational analysis revealed that anger is the bridge between both physical and verbal aggression and hostility. The scales showed internal consistency and stability over time. Men scored slightly higher on Verbal Aggression and Hostility and much higher on Physical Aggression. There was no sex difference for Anger. The various scales correlated differently with various personality traits. Scale scores correlated with peer nominations of the various kinds of aggression. These findings suggest the need to assess not only overall aggression but also its individual components.
Chapter
About a quarter of a century ago, a young American radical, Stokely Carmichael, commented that violence was as American as apple pie! At least in terms of prevalence, nothing much seems to have changed since that time. The frequency of violence directed by one human being at another was appallingly high then and is appallingly high now. The United States is not the most violent society in the world. That distinction belongs to some of the less developed countries ravaged by wars, terrorism, drug battles, and general lawlessness. Nor is violence as endemic now as it has been during many of the last 20 centuries. Among the highly developed Western societies, however, the United States has scored at the top for the past several decades on most objective measures of interpersonal violence. For example, homicides in the United States rose from an overall rate of about 5 per 100,000 to 10 per 100,000 between World War II and the 1980s and have remained at about that level. Of course, the rate in some inner-city ghettos may be 10 times this rate (100/100,000) and the rate for certain age cohorts may be 3 times this rate (e.g., 30/100,000 for males 18 to 24). In comparison, no other highly developed Western society has a rate much above 3 per 100,000 and most are below 1 per 100,000. Rates at these levels are cause enough for concern and also reflect increases since World War II, but the sustained rates in the United States are a national tragedy. In some urban areas of the United States the most common cause of death for young males is now homicide.
Chapter
Social psychological research of the past years has been greatly preoccupied with the question of the validity of results obtained in the typical laboratory experiment. In the late 1950s and the early 1960s, several authors warned that pervasive sources of contamination may be operative in psychology experiments, and that the data collected therein may not be what they seem. This chapter attempts a critical review of the literature on subject-artifacts. The notion of artifact has been explicated. The chapter discusses the possible reasons for the quick concessions to the claims for subject artifacts. However, the gist of the discussion revolves about the validity of the critiques. The chapter outlines several distinct strategies for the study of subject-artifacts and examines the unique difficulties of each approach. The empirical evidence bearing on the part played by the various artifacts has been considered in an attempt to estimate the extent of the biases involved.
Article
The intention-superiority effect is the finding that response latencies are faster for items related to an uncompleted intention as compared with materials that have no associated intentionality. T. Goschke and J. Kuhl(1993) used recognition latency for simple action scripts to document this effect. We used a lexical-decision task to replicate that shorter latencies were associated with uncompleted intentions as compared with neutral materials (Experiments 1 and 3). Experiments 2-4, however, demonstrated that latencies were longer for completed scripts as compared with neutral materials. In Experiment 3, shorter latencies were also obtained for partially completed scripts. The results are discussed in terms of the activation and inhibition that may guide behavior, as well as how these results may inform theories of prospective memory.