ARTICLE EXPLICATION ASSIGNMENT

Full Length Article

Impact of violent video game realism on the self-concept of aggressiveness assessed with explicit and implicit measures

Joerg Zumbach a,⇑, Caroline Seitz a, Matthias Bluemke b a University of Salzburg, Austria b University of Heidelberg, Germany

a r t i c l e i n f o

Article history: Received 4 February 2015 Revised 7 July 2015 Accepted 11 July 2015 Available online 17 July 2015

Keywords: Violent computer games Aggression Immersive virtual environments

a b s t r a c t

We compared the standard 2D representation of a recent violent computer game to its 3D representation realized by shutter-goggles in a lab experiment. Assuming that the higher degree of realism of media vio- lence would impact stronger on players in a pretest–posttest design, we analyzed the influence of violent video game exposure on implicit and explicit measures of aggressiveness. According to an explicit ques- tionnaire on aggressiveness, participants reported having becoming more peaceful, whereas an Implicit Association Test on aggressiveness (Agg-IAT) indicated that the association between self and aggressive behavior became stronger after violence exposure, confirming the unique utility of Agg-IATs in media research. The 3D visualization mode, however, did not further strengthen this association, and a media- tion model of increases in aggressiveness by participants’ flow experiences was not supported. When inspecting flow experiences, an interaction effect between gender and visualization mode was evident: Male participants were more likely to have flow experiences in the high-realism (3D) format, whereas female participants were more likely to experience flow in the standard (2D) mode. We discuss the find- ings in the context of automatic information processing in aggression, and we contend possible changes in automatic behavioral precursors due to media influence.

! 2015 Elsevier Ltd. All rights reserved.

1. Introduction

This contribution investigates the psychological consequences of a potent factor in violent video and computer games (VVG): a player’s immersion into realistic game scenarios. The present experiment has been designed to test whether the impact of VVG on automatic aspects of players’ aggressiveness concept is a func- tion of the game’s realism, potentially mediated by a player’s higher flow experiences in more realistic scenarios. The introduc- tion will first briefly review short- and long-term effects of violent media exposure and theoretical explanations. Then potential mediators between game realism in VVG and subsequent changes of aggressiveness will be discussed. Drawing on current dual-process models of social cognition and behavior, it is con- tended that the observation of possible effects should encompass more objective measures of aggressive dispositions, such as implicit measures, as they capture crucial automatic processes that matter in the generation of aggressive behavior and may be shifted by VVG.

1.1. Violence exposure in video games

Computer game features help players get immersed into the games. Carefully calibrated challenges and rewards, well-balanced with elements of chance, engage a player in various tasks. Not denying the positive effects on players’ enjoyment, social ties, mental relaxation, visual acuity and other cognitive benefits (e.g., Adachi & Willoughby, 2013; Granic, Lobel, & Engels, 2014; Green & Bavelier, 2006, 2012), what has concerned many research- ers is that violence exposure can stimulate aggressive cognition, affect, and behavior, while reducing empathy and prosocial behav- ior (e.g., Anderson & Bushman, 2002; Bartholow, 2005; Bushman & Anderson, 2001). In line with social learning theory (Bandura, 1977), gratification after observing aggressive models is held responsible for acquiring aggressive behavioral scripts (Crick & Dodge, 1994). Contemporary approaches underscore learning even in the absence of immediate reinforcement (e.g., mimicry of aggressive scripts; Bushman & Anderson, 2001; Bushman & Huesmann, 2006; Huesmann & Kirwil, 2007).

One prominent model for comprehensively explaining detri- mental VVG effects in the short and long run is the General Aggression Model (GAM; Anderson & Bushman, 2002; Anderson & Dill, 2000; Bushman & Anderson, 2002; for a critical note see

http://dx.doi.org/10.1016/j.chb.2015.07.018 0747-5632/! 2015 Elsevier Ltd. All rights reserved.

⇑ Corresponding author at: School of Education, University of Salzburg, Hellbrunner Str. 34, 5020 Salzburg, Austria.

E-mail address: [email protected] (J. Zumbach).

Computers in Human Behavior 53 (2015) 278–288

Contents lists available at ScienceDirect

Computers in Human Behavior

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p h u m b e h

Ferguson & Dyck, 2012). In the short run, situational factors—mod- erated by personality traits (Bushman, 1995)—can trigger aggres- sive scripts (short-time effects refer here to minutes up to hours; i.e., on the level of a state). The supposed mechanisms are unspeci- fic arousal, angry emotions, and aggressive cognitions. A multitude of primary lab experiments as well as meta-analyses have demon- strated a significant causal effect of computer game violence on aggression (Anderson, 2004; Anderson and Bushman, 2001; Anderson et al., 2003). Long-term effects have been described for emotional desensitization to violence and chronic activation of aggressive behavioral scripts (Carnagey & Anderson, 2004; Cline, Croft, & Courier, 1973; long-time effects refer to time periods of months and years; i.e., they are rather on a trait level). Paralleling research on TV violence (Huesmann, Moise-Titus, Podolski, & Eron, 2003), long-term studies have shown that con- suetudinary use of VVG throughout childhood significantly pre- dicts physical aggression during adolescence (Anderson et al., 2008).

Despite these stable effects, the cause-effect debate is still going on, partly because meta-analytical conclusions diverge (Anderson et al., 2010; Ferguson & Kilburn, 2009). To the extent that pre- formed attitudes and gamer’s explicit self-views contrast with autonomous effects, which may even be hidden from introspec- tion, a reconciliation seems difficult to achieve, especially if players feel as if they need to defend themselves against accusations of negative effects from common leisure activities (Kneer, Glock, Beskes, & Bente, 2012; Przybylski, Deci, Rigby, & Ryan, 2014). One way to partially reconcile the contradictory evidence then, apart from critically discussing any sophisticated meta-analytical strategies, is to highlight the limits of self-report measures of aggressive dispositions (Bender, Rothmund, & Gollwitzer, 2013) and to use objective implicit measures as a complement. The latter are based on objective indicators; being less blatant, they are more difficult to control than subjective measures. It has been firmly established that implicit measures of the aggressiveness self-concept sensitively reflect the influence of different computer games, and that implicit measures, under controlled conditions, reflect the impact of violent, but not nonviolent game conditions (Bluemke, Friedrich, & Zumbach, 2010; Bluemke & Zumbach, 2012; Uhlmann & Swanson, 2004). What is still unclear, though, is whether implicit measures reflect the influence of game realism as sometimes found with subjective measures, and whether the altered psychological state actually serves as a mediator in this process.

1.2. Involvement, identification, and realism in computer gaming

We use as a starting point the recent computer games that often attempt to maximize the realism of computer games. Here, virtual reality is intended to help players immerse themselves totally in a game, and, thus to increase realism (Hoorn, Konijn, & van der Veer, 2003; Konijn & Hoorn, 2005). Modern Immersive Virtual Environment (IVE) technologies, which are typically head-mounted displays with 3D visualization and motion-based input devices, are used more frequently nowadays, partly with the approved goal to establish aggressive skills, for instance, in military training units (Persky & Blascovich, 2008; Tamborini & Skalski, 2006). This drives our overall research question: Do VVG effects become stronger, the more real the games feel, and can this be demonstrated independently of subjective self-reports?

Our first working hypotheses are here: (1) Directing one’s game character through a virtual world connects the player’s self with the virtually executed behaviors to some extent in the short run, and if exposed to violent actions repeatedly, a player’s self-concept may be altered such that aggressive schemata are

advanced as the dominant ones (Crick & Dodge, 1994). (2) This automatic association between self and aggressive schemata will be picked up by implicit measures (though it can be censored from an explicit view of the self).

As a preliminary answer, research has shown that the mere involvement of a computer game player with a game character (avatar) is sufficient for stronger media effects than when watching similar content passively (Carnagey & Anderson, 2004). Identifying with the virtual protagonist can blur the distinction between the virtual and the real world and exacerbate subsequent aggressive behavior (Konijn, Nije Bievank, & Bushman, 2007). Support for the relevance of a player’s involvement in realistic VVG comes from a lab experiment by Fischer, Kastenmüller, and Greitemeyer (2011). When participants created their own avatars so that these were optically reminiscent of the players themselves, the detri- mental effect of violence exposure on aggression was markedly intensified in comparison to pre-selected avatars, and this effect was mediated by players’ involvement (cf. Hoffner & Buchanan, 2005).

In another study, Tamborini et al. (2004) compared VVG effects of a modern IVE with those of a classic desktop version of the game. Participants in the IVE condition reported that they experi- enced being present in the game (so-called ‘‘telepresence’’) more strongly and felt more hostile affect afterward (cf. Eastin & Griffiths, 2006). Persky and Blascovich (2008) showed that, in com- parison to less immersive desktop platforms, IVE platforms pro- moted particularly the experience of presence and aggressive thoughts, not necessarily of aggressive behavior; they also found higher levels of aggressive cognition among active players than among passive game observers. Krcmar, Farrar, and McGloin (2011) showed that IVE participants reported higher levels of aggression to the extent that they strongly identified with their characters, suggesting that experienced realism mediated between platform and individual effects (cf. Ivory & Kalyanaraman, 2007). Rajae-Joordens (2008) observed long-term effects of using 2D and 3D devices for watching movies and playing a VVG. The 3D device significantly intensified not only aggressive, but all sorts of emotional experiences. It appears that blurring the distinction between virtual world and reality by increasing telepresence aggravates any emotional experiences, and possibly also unwar- ranted VVG effects.

The different processes—identification, involvement, and expe- rienced realism—feed into a common signal: the experience of immersive flow, that is, the total absorption in a task, which is a specific state of mind and powerful explanatory concept. According to Csikszentmihalyi (1991), flow is an increased sense of psychological immersion, a holistic sensation that people feel when acting with total involvement. The primary antecedents of flow in computer-based tasks are congruency between skills and challenges and narrowly focused attention, complemented by sec- ondary antecedents such as interactivity and telepresence (Hoffman & Novak, 1996; Novak, Hoffman, & Yung, 2000). Telepresence refers to the perception of being present in virtual reality, once one is fully immersed. Equally, immersion has been defined as the state of flow when navigating a virtual environment (Koh & Kim, 2003). Hence, flow experiences and immersion are mutually related in the sense that flow is an affective outcome that follows from the cognitive phenomenon of interpreting spatial, acoustic, tactile, visceral, and social cues as if one were enveloped by a virtual world (Grinberg, Careaga, Mehl, & O’Connor, 2014). In a cyclic fashion, the more flow a gamer experiences, the stronger the sense of immersion into virtual reality. Though immersion and flow can be disentangled, they are conceptually and empirically related (Faiola, Newlon, Pfaff, & Smyslova, 2013); many authors simply equate the two (see also Huang, Chiu, Sung, & Farn, 2011). Also the scale we used (see Methods) refers to telepresence

J. Zumbach et al. / Computers in Human Behavior 53 (2015) 278–288 279

and immersive aspects of flow; hence for the present purpose we will use the terms ‘‘flow’’ and ‘‘immersion’’ interchangeably.1

The occurrence of flow itself is fostered by the realism of the environment and the presence of the player. We propose that experiencing a game as realistic increases identification with a game character and involvement with the game. Sufficient playing skills provided, experienced sensory realism in a 3D environment will nourish psychological realism. Along this psychological chain, the overlap between virtual and real world might become so strong that a player’s real self becomes highly blended with game charac- ter executing aggressive scripts in the virtual world, thereby informing a player’s affect, cognition, and behavior (Klimmt, Hefner, & Vorderer, 2009).2 Any disruptions of flow (e.g., technical annoyances, distancing oneself mentally from the avatar) are likely to weaken the connection between the player’s self and his game character. In sum, we come to our second set of working hypotheses that (3) playing VVG in 3D mode shapes a player’s aggressive dispo- sitions and self-concept of aggressiveness more than in 2D mode, and (4) that immersive flow is a crucial mediator.

Though the aforementioned studies inspire adopting a causal chain model, the previously mentioned IVE studies either did not manipulate realism at all, counteracted the realism manipulation by having gamers play in an unrealistic invulnerable ‘‘god-like’’ mode, counteracted the notion of realism by playing an avatar from a third-person rather than first-person perspective, or did not vary exclusively 2D and 3D realism (invoking confounders). For instance, sometimes participants played at typical 2D desktops but were compared to players wearing an unusual—and restrict- ing—3D helmet plus gloves that both were meant to insulate the users from environmental distractions. Alternatively, one group played an old-fashioned game while the other group played a mod- ern game simply related by label (Doom 1 and 3), while at the same time significant differences existed between the quality of games, movements, player interactions, graphic resolutions, and gory details. According to the Media Comparison Paradigm (Bluemke & Zumbach, 2012; Bluemke et al., 2010), media compar- isons are only valid within ‘‘zones of comparability’’. In these zones either the semantics are identical (i.e., what specific media content participants encounter) while the syntactics vary (i.e., how partic- ipants interact with the media content), or the semantics vary while the syntactics are identical.

The goal of our research was to compare realism conditions while both content and interactivity remain identical (cf. Przybylski et al., 2014, for well-controlled interactive and struc- tural aspects of media effects). Furthermore, a renewed inspection with the help of implicit measures seemed warranted as research on the mediating model has yielded inconsistent results. The medi- ation account mostly held for explicit self-report measures, yet when using rather objective indicators, the mediational model occasionally broke down (Jeong, Biocca, & Bohil, 2012; Persky & Blascovich, 2008). This role of different measures of aggression and their role within research on media influences on aggression is another important and central issue here.

1.3. Measuring automatic and controlled processes in aggression

Explicit measures require participants to introspect, deliberate, and report accurately. Implicit measures utilize the performance in

simple categorization tasks. Participants’ associations are inferred on the basis of their objective response latencies, which are taken to reflect automatic attitudes, stereotypes, self-evaluations, or trait-like aspects of the self-concept (Greenwald & Banaji, 1995). More specifically, these constructs are captured in an associative, non-propositional form a few hundred milliseconds after stimulus presentation (Gawronski, 2009). Implicit measures have benefits over explicit measures when predicting behavior (a) in clearly automatic behavioral domains, (b) in situations of low cognitive or motivational control, or (c) when participants are unable or reluctant to admit the attribute in question to themselves or others (Greenwald, Poehlman, Uhlmann, & Banaji, 2009). Embraced across nearly all psychological domains (e.g., Haines & Sumner, 2013; Roefs et al., 2011), they have not had much impact on the field of aggression yet (cf. Bluemke & Teige-Mocigemba, 2015).

The Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998) is one of the most successful implicit measures (for limitations see De Houwer, Teige-Mocigemba, Spruyt, & Moors, 2009). By selecting aggressive and non-aggressive behavior stimuli IATs can target at the spontaneous association between the self and aggressive behavior, and the resulting IAT scores can reflect automatic components of aggressive dispositions that are not fully represented by questionnaires (Banse & Fischer, 2002; Banse, Messer, & Fischer, 2015). With regard to the impact of VVG on implicit measures, Uhlmann and Swanson (2004) were the first to show that such an aggressiveness-IAT (Agg-IAT) corre- lated in a trait-like manner with long-term exposure to VVG. Furthermore, in their lab experiment Agg-IAT scores adequately reflected randomly allocated violent or non-violent computer game play (see Lemmer, Gollwitzer, & Banse, 2015, on the question of trait vs. situation-specific variance in the Agg-IAT. Bluemke et al. (2010) showed that the Agg-IAT was sensitive to both peaceful and violent game content even when controlling arousal confounds. In another study an IAT-variant differentiated between groups of computer game users (Bluemke & Zumbach, 2012). Never did explicit measures fully converge with implicit measures.

These findings indicate a dissociation rather than an overlap between explicit and implicit measures, which is in line with mounting evidence on the utility of implicit measures to explain parts of the variance in aggressive behavior via an automatic (as opposed to a controlled) pathway that can trigger behavioral scripts but is not captured by explicit measures (Richetin & Richardson, 2008). For instance, in studies on incremental validity, the IAT predicted aggression when trait-variables indi- cated, or situational manipulations created, low self-monitoring, self-control, or executive functioning (e.g., Bluemke & Friese, 2012; Denson, Capper, Oaten, Friese, & Schofield, 2011; Richetin, Richardson, & Mason, 2010; Schmidt, Zimmermann, Banse, & Imhoff, 2015).

Recent dual-process models of social cognition can explain both the dissociation between explicit and implicit measures, as well as the incremental validity of implicit measures in behavioral predic- tion. They conjecture that information processing can be reflective, effortful, slow, and flexible as opposed to automatic, effortless, fast, and inflexible. For instance, according to the reflective-impulsive model (RIM; Strack & Deutsch, 2003, 2004), human cognition pro- motes the activation and execution of behavioral scripts either via a rather reflective or a rather impulsive pathway, with a complex interplay between the two systems of information processing. Compatible with this notion, the GAM integrates an automatic, effortless, and fast route (called ‘‘impulsive action’’) next to a reflective, but more resourceful and slower way to the activation of aggressive scripts (called ‘‘thoughtful action’’). The mechanisms leading from perceptual input (cues) to aggressive behavior may involve associative priming of stereotypes, hostile interpretations, and behavioral scripts (Berkowitz, 1993; Crick & Dodge, 1994) as

1 The label ‘‘immersion’’ has also been used for one of three factors underlying player motivations, summarizing empirically correlated aspects of discovery, role-playing, customization, and escapism (Yee, 2006). However, other authors clearly conceptualize immersion as a subscale of flow experiences, with immersion being the strongest contributor for learning success in e-learning games (Fu, Su, & Yu, 2009).

2 Note that positively valenced scripts can be acquired too (Bluemke et al., 2010; Sestir & Bartholow, 2010). We will return to this in the Discussion.

280 J. Zumbach et al. / Computers in Human Behavior 53 (2015) 278–288

well as reduced self-control due to angry emotions and unspecific physiological arousal or excitation transfer (Batinic & Appel, 2008; Bushman & Huesmann, 2006; Zillmann, 1978). Whereas the GAM does not specify how impulsive precursors are to be assessed, the RIM designates implicit measures for tapping—imperfectly though—into associative structures residing in the impulsive system.

Taking up the notion of dual-process models and applying it to the domain of aggression, one can argue that violent media are likely to affect not only reflective precursors of behavior but also automatic information processing, that is, accessible scripts and spontaneous associations (e.g., Bushman, 1998). These automatic processes are known to predispose some individuals to spontaneously aggress with higher likelihood than others—even if individuals are unaware of these determinants, or unwilling to report on them (Berkowitz, 2008; Richetin & Richardson, 2008; Todorov & Bargh, 2002). From the perspective of dual-process mod- els, by using rather blatant and deliberate measures to demonstrate violence–exposure or realism effects, one misses out spontaneous cognitive processes that occur involuntarily. Relying exclusively on transparent questionnaires thus entails the risk of underesti- mating aggressive tendencies and not uncovering VVG effects (Bender et al., 2013; Mills & Kroner, 2006). Yet with the Agg-IAT there is a measure for exploring nuances of altered automatic asso- ciations after VVG exposure in 2D or 3D mode. Having executed aggressive behavioral scripts repeatedly during game play in virtual reality, a player’s game character is associated with aggression. And the stronger a player has immersed himself into a world that feels more real than virtual, the more likely it is that the association between the player’s self and aggressive behavior becomes strengthened too, as reflected in an individual’s Agg-IAT score.

1.4. Study goal and hypotheses

We tested the short-term effects of the current generation of computer games on spontaneous self-associations with aggressive behavior. We focused exclusively on playing a VVG in a traditional 2D vs. an enhanced realistic 3D visualization mode. Previous research has documented that changes in Agg-IAT scores are speci- fic to the degree of violence in game contents, with nonviolent games not changing Agg-IAT scores even after repeated measure- ment (Bluemke et al., 2010; Uhlmann & Swanson, 2004). Such a repeated measurement design was warranted, because irrelevant cognitive factors confound the performance in implicit measure- ment procedures and inflate between-participant variance in response latencies, thereby reducing the sensitivity of the implicit measure to experimental treatments (Fiedler, Messner, & Bluemke, 2006). When not controlling preexisting individual differences by means of a baseline measure prior to game play, power to obtain 3D gaming effects may be too low (Banse et al., 2015; Bluemke et al., 2010).

Hypothesis 1. Change scores of an explicit measure of aggressive- ness, which relies on introspection and subjective self-report, should not reveal negative effects on a player’s self-concept after violence exposure, because of short-term effects are difficult to detect with trait questionnaires, but also due to lack of introspection and/or underreporting as a result of social desirable responding or protection of the status of the gamer in-group (Bender et al., 2013).

Hypothesis 2. We hypothesized that violence exposure during a modern computer game increases aggressive dispositions on an implicit measure. The Agg-IAT change scores would indicate stron- ger spontaneous associations of self with aggressive behavior from

pretest to posttest (VVG main effect; Bluemke et al., 2010; Klimmt, Hefner, Vorderer, Roth, & Blake, 2010).

Hypothesis 3. The magnitude of the VVG main effect should depend on game realism. Playing the identical violent shooter in an immersive 3D environment should intensify the game experi- ence and lead to even higher Agg-IAT change scores (moderation effect; Persky & Blascovich, 2008).

Hypothesis 4. Finally, we predicted that realism effects as evident in Agg-IAT change scores would be mediated by players’ flow expe- riences during game play. Stronger involvement and absorption in the 3D condition should amplify the flow experiences and promote the link between self and aggressive behavior (mediation account; e.g., Fischer et al., 2011; Krcmar et al., 2011; Ravaja et al., 2004). In this mediation model flow correlates both with 2D vs. 3D visualiza- tion mode and Agg-IAT changes, thereby accounting at least partly for the direct relationship between realism and changes in aggres- siveness (Baron & Kenny, 1986).

To our knowledge, this mediation model has never been applied to the automatic associations in a well-controlled experi- mental design that kept genders balanced and game content iden- tical. All 3D studies so far relied on indicators of reflective processes (explicitly endorsed statements), though implicit mea- sures are well suited to tap into subtle aspects of automatic infor- mation processing (Gawronski, 2009) and changes in the self-concept (Klimmt et al., 2010). As the impact of realistic, highly involving first-person shooters on Agg-IAT scores is to date unknown, we also compared the effect size on the Agg-IAT to pre- vious outcomes based on static game scenarios with relatively low realism, as one might suspect stronger VVG effects after highly realistic game play.

Furthermore, in previous studies involving the Agg-IAT men had rather been underrepresented in the lab (Bluemke et al., 2010), but overrepresented in an online study (Bluemke & Zumbach, 2012), warranting a further check with a gender-balanced sample. We did not expect a severe influence of gender on the predicted data pattern, yet it has been noted that gendered self-identification with an avatar can foster aggression (Anderson & Murphy, 2003; Krcmar et al., 2011). Thus, if gender effects moderated the link between violence exposure and aggres- siveness such that men generally showed higher Agg-IAT changes than women, we would expect this pattern to be even more pro- nounced in the highly realistic game environment.

2. Method

2.1. Participants and design

Sixty-nine participants from the University of [removed for review] (52% males; Mage = 24.46 yrs., SD = 5.86) participated in an experiment on computer visualization modes in exchange for course credit. Another four participants are not included in this sample due to incomplete data (technical failure, unmotivated responding in the IAT with more than 50% errors in at least one critical block, or complete drop-out). Due to listwise deletion for the occurrence of missing data, the reliability analyses based on Cronbach’s Alpha may be based on fewer than 69 participants. As regards the experimental procedure, participants were randomly assigned to the 2D or 3D condition. The pretest–posttest design afforded implicit and explicit measures of aggressiveness, taken before and after the computer game. In addition, we assessed potentially mediating variables at posttest.

Participants reported owning a personal computer (100%), a Sony playstation (45%), a Microsoft Xbox (14%), or a Nintendo

J. Zumbach et al. / Computers in Human Behavior 53 (2015) 278–288 281

Wii (13%). On average they played 0.60 h a day (SD = 1.33) and 5.91 h a week (SD = 9.19). As regards gender differences, 22% of men vs. 6% of women reported owning an Xbox, and the likelihood to play computer games was double for men (36% vs. 18%), in line with Gentile, Lynch, Ruh Lindner, and Walsh (2004). Men played twice as many hrs/week than women, Ms = 7.64 (SD = 9.83) vs. 4.01 (SD = 8.17), in line with Lemmens and Bushman (2006). Thirty-three participants stated that they do not play computer or video games at all (6 of them male).

2.2. Material and experimental procedure

After providing informed consent, participants reported socio-demographic characteristics. First, they worked on the impli- cit measure (Agg-IAT), then an explicit aggression questionnaire. This order was fixed to prevent priming of specific aggressive con- tent and subsequent effects of aggression suppression (e.g., Glock & Kneer, 2009). Next, the 10-min treatment provided one experi- mental group with the realistic and highly graphic first person shooter ‘‘Call of Duty: Modern Warfare 2’’ (Act 1, Level 2; Infinity Ward/Activision Blizzard, 2011) on an IBM-type personal com- puter (PC) attached to a 120 Hz LCD flat screen (21’’ screen width). Another group used special shutter-goggles (nVIDIA GeForce 3D Vision) that allowed players to experience the same game at the same PC and LCD-screen in a realistic and flicker-free 3D manner. Ten minutes of a mildly difficult level were presented to partici- pants. The player switched into the role of a soldier within a team (other players were CPU-controlled) and the screen provided a first person perspective of the game environment and player’s extrem- ities and weapons. By moving through the virtual environment the task was to shoot enemy soldiers while protecting oneself against enemy fire. Usually a level was completed after having destroyed all enemy forces and conquered specific targets. Apart from asking for the subjective flow experiences during the game, implicit and explicit measures of aggressiveness were taken a second time. Including a debriefing about the study background, the experiment took between 45 and 55 min. We used SPSS-20 for the statistical analyses.

2.2.1. Implicit measure of aggressiveness The Agg-IAT assessed the speed while sorting stimulus words in

a double-barreled sorting-task (Uhlmann & Swanson, 2004). To keep the repeated-measurement procedures succinct, we used a standard five-block structure with a fixed order of the crucial blocks to reduce extraneous variance that might attenuate correla- tions and mediating paths. The following blocks required partici- pants to correctly categorize stimuli with the help of two response keys: (1) attribute words (i.e., behavioral stimuli such as HIT or TALK) as either aggressive or peaceful (20 practice trials), (2) target words (such as ME or YOU) as related to self or other (20 practice trials), (3) attribute and target words when self + peaceful (as well as other + aggressive) shared one response-key (40 critical trials), (4) target stimuli with inverted response-key assignments (20 practice trials), and (5) target and attribute words when self + aggressive (other + peaceful) shared one response-key (40 crit- ical trials). Participants had to correct their responses if they made a categorization error. The stimuli were behavioral words taken from studies that had successfully used German Agg-IATs (see Banse et al., 2015; Bluemke & Friese, 2012; Bluemke & Zumbach, 2012; Bluemke et al., 2010).

The first two trials in each critical block were discarded to com- pensate for orientating reactions and slow responses in the absence of practice trials in the critical blocks. Individual IAT effects were computed as so-called D-scores (Greenwald, Nosek, & Banaji, 2003), that is, as latency difference between the mean latencies in the critical blocks 3 and 5, standardized by an

individual’s pooled standard deviation in the critical blocks. In line with recommendations for IAT scoring procedures for cognitively taxing treatments, no error penalties were applied to erroneous trials (see Bluemke & Zumbach, 2012; Schoel, Bluemke, Mueller, & Stahlberg, 2011). Higher scores indicate higher aggressiveness, and for most participants negative IAT scores (indicating a peaceful self-concept) resulted, yet given a fixed block order and the absence of an IAT calibration procedure of zero scores, the absolute magnitude of the IAT effects should not be overinterpreted. D-scores were reliable at both pretest and posttest (Spearman– Brown corrected reliability of IAT effects based on odd and even trials), r1 = r2 = .75 (90%-confidence intervals CI90 = .62–.83).

2.2.2. Explicit measure of aggressiveness A German version of the Aggression Questionnaire (BPAQ; Buss

& Perry, 1992) with 29-items and a 5-point rating scale format (ranging from ‘‘I fully agree’’ to ‘‘I do not agree at all’’) assessed facets of self-perceived aggressiveness. The items form subscales pertaining to physically aggressive behavior (e.g., ‘‘I have threat- ened people that I know well’’), verbally aggressive behavior (e.g., ‘‘My friends say that I am belligerent’’), anger (e.g., ‘‘I erupt due to nothing’’), and hostility (e.g., ‘‘If people are kind to me I wonder what they want’’). Two items did not enter the analyses to satisfy the intended four-dimensional structure (Von Collani & Werner, 2005). Analyses were based on the BPAQ means across all items, which were reliable, Cronbach’s a1 = .84 (CI90 = .79–.89, n1 = 62), and a2 = .88 (CI90 = .84–.91, n2 = 60) at pretest and postt- est, respectively.

2.2.3. Involvement, immersion, and flow experiences Participants’ flow experiences during the game were assessed

with a modified version of the Flow Short Scale adapted to game play in the present study (Engeser & Rheinberg, 2008; Rheinberg, Vollmeyer, & Engeser, 2003). The scale is based on the flow-concept as suggested by Csikszentmihalyi (1991) and has been externally validated in predicting learning performance within a statistics program or success in an Assessment Center subtask (cf. Engeser & Rheinberg, 2008; Rheinberg & Vollmeyer, 2003). Thirteen items with 7-point rating scales were averaged so that they formed the three relevant subscales absorption by activity (4 items, e.g., ‘‘I was totally absorbed in what I was doing during the game’’; a = .78, CI90 = .70–.85, n = 68), fluency of perfor- mance indicating the right skill level (6 items, e.g., ‘‘The right thoughts/movements occurred of their own accord during the game’’; a = .86, CI90 = .82–.90), and outcome importance encompass- ing potentially distracting thoughts (3 items, e.g., ‘‘I was worried about failing’’; a = .85, CI90 = .79–.90). Note that absorption resem- bles the concept of presence or ‘‘being present’’, that is, the factor accounting for realism effects on explicit reports of aggressive feel- ings and thoughts (Persky & Blascovich, 2008). The first two sub- scales were substantially correlated, r = .62 (p < .001), whereas outcome importance correlated with the other subscales only weakly, rs = .27–.33 (ps < .05). Hence, we aggregated absorption and fluency items to index the overall flow experience (a = .88, CI90 = .84–.91, n = 67).

3. Results

We inspected pretest–posttest changes due to game play. The main analysis of explicit and implicit measures of aggressiveness used a 2 (time: pretest vs. posttest) ! 2 (realism: 2D vs. 3D) ! 2 (gender) ANOVA with repeated measurement on the first factor. Flow experiences were analyzed in the same manner, yet—in the absence of a baseline score—without the repeated measurement factor time.

282 J. Zumbach et al. / Computers in Human Behavior 53 (2015) 278–288