Effect of music tempo in First-Person Shooter on arousal and aggression ∗ AdnanM.Niazi DepartmentofElectricalEngineering,Mathematics&ComputerScience,UniversityofTwente,7500AE,Enschede,TheNetherlands ARTICLE INFO ABSTRACT ArticleHistory: Thepresentstudy usedabetweensubjects, post-test only designto investigate theimpactof Submittedon21April2011 background music(lowtempo musicvs.hightempo music)on arousal andaggression ofa playerplayinga firstperson shooter videogame:Counter-Strike.Arousalwasmeasuredusing Skin Conductance Level(SCL)andaggression using an Implicit Attitude Test(IAT).Results Keywords: indicatenosignificantdifferencesin thearousal andaggression between thetwo conditions Tempo implyingthatmusictempoplaysnoroleinmanipulatingthearousalandaggressionofaplayer Arousal ingamingcontext.Thestudydid,however,findasignificant correlation betweenarousal and Aggression aggressioninhightempo condition, suggesting ageneral trend thatmore aroused subjects ImplicitAttitudes showed more aggressive cognitions. No significant correlation was found in low tempo GameMusic condition. 1. Introduction A recent longitudinal work using late elementary and ju- nior high school aged students in the United States and With the development of technology and popularization of Japanshowedthathabitualviolentvideogameplayinearly electronicproducts,videogameshavebecomeanimportant schoolyearswaspredictiveofphysicalaggressioninthelater part of life of adolescents and adults alike. Worldwide, the years(Anderson,Sakamoto,Gentile,Ihori,Shibuya&Naito, video games sales is expected to top $48.9 billion in 2011 2008). Anderson et al. (2008) argues that violent video and $68 billion in 2012, making it one of fastest-growing game exposure, leading to an increase in physical aggres- componentofthemediasectorworldwide. sionacrosstwoverydifferentcultures,isastrongtestament With video games, infiltrating our lives more than even be- tothepowerofviolentvideogamesinincreasingandsteer- fore, it’s no wonder their moral, social, and psychological ing aggression, particularly in young children. In summary, implications are constantly under the microscope with ma- experimental (e.g., Farrar & Krcmar, 2006), cross-sectional jority of studies focusing mainly on the relation between (Anderson et al., 2004), and longitudinal studies (Ander- video games and aggression. Experimental evidence now son et al., 2008) have now demonstrated that violent video suggests that exposure to violent video games is in fact re- gamesarerelatedtoincreasesinaggressionformen(Farrar lated to an increase in aggression (see Anderson & Bush- & Krcmar, 2006), women (Anderson & Murphy, 2003) and man(2001),Sherry(2007),andAnderson(2004)formeta- children(Andersonetal.,2008)alike. analyses). More specifically, violent video games can ma- However, meta-analyses (Sherry, 2007) have raised ques- nipulateaggressivecognitions(Anderson&Dill(2000);An- tions about how game features and participant characteris- derson, Carnagey, Flanagan, Benjamin & Jr. (2004); Kirsh tics may play a role in moderating or mediating aggressive (1998);Tamborini,Eastin,Skalski,Lachlan,Fediuk&Brady outcomes. Sherry (2007), suggests that variations in game (2004)),andaggressiveaffect,leadingtofeelingsofhostility stimulus, ageoftheplayer, andamountoftimethegameis (Anderson&Dill(2000);Farrar&Krcmar(2006);Tamborini played,amongothervariables,causedifferentoutcomes. For et al. (2004)), and resulting in aggressive delinquency. Ex- example, Carnagey & Anderson (2005) found that violent perimentalworkalsofoundexposuretoviolentgamestobe videogamesthatrewardviolentbehaviorsleadtoincreases linkedtothephysiologicalarousal(Barlett,Harris&Baldas- in hostile emotions, aggressive thinking, and aggressive be- saro,2007). havior. Barlett et al. (2007) reports that playing a violent videogameusingalightgunasopposedtoatraditionalcon- ∗StudentNumber:s1029789 trollerleadstomoreaggression. Eastin(2006)reportsthata Emailaddress:[email protected](AdnanM.Niazi) 1 gendermatchbetweenselfandgamecharactercanincrease to variations in light and temperature, variations in the au- aggressive thoughts in female game players. Farrar & Krc- ditory stimuli may also elicit analogous adaptive responses. mar(2006)havefoundthatvideogameplaywiththeblood For example, heart rate, blood pressure, breathing rate and feature activated can increase aggression. Thus, features of amplitude, and other physiological manifestations of affect- violent games can serve to heighten or dampen effects on lessarousalinvoluntarilyincreaseuponexposuretofastand subjects. stimulatingmusic(Lundin,1985). Inaddition,fromaninformation-theoreticperspective,faster 1.1. VideoGamesandMusic tempi produce higher rates of information and levels of in- formational density (Crozier, 1981), which should increase “Sometimescertaingameshavenoorreallybadmu- the "interestingness" of the music and thereby contribute to siclikeCounterstrike,soIputonsomeothermusic” listeners’ pleasure (Berlyne, 1974). Tempo may also con- tributeindirectlytopleasureviaarousalinthatfastermusic Subject-110 (Huiberts,2010) may seem more exciting. Of course, as tempo increases to One game feature that has garnered much popular interest a very high speed, one can expect an eventual decrement by the video game players, but not much psychological re- in pleasure (Holbrook & Anand, 1990). The present study, search, is the game music. The continuing growth of the however,doesnotexploreanextremerangeoftempi(Presto videogameindustryhassparkedanincreasedinterestinthe music) that would produce such non-monotonic effects. If musicwrittenforgamesoundtrack. Today’scomposersoften theresultsoftheexperimentprovideapositiveanswer,that workwithlargeorchestra’sproducingscoresthatoftenrival music tempo does indeed play a role in moderating aggres- film music in scale, complexity, and artistic ambition (Ben- sion,itisenvisagedthatfutureexperiments,usingasimilar nun, 2010). With game revenues outstripping those of the methodology, will be designed to investigate more specific film industry, developers are keen more than ever before to questionsabouttherelationshipbetweenthemusicandag- rivalthecinematicexperienceofHollywoodmovies. There- gressioninallvideogamesgenresingeneral. sultisthat,gamemusicisbecomingmoreandmoreadaptive andinteractive. Insteadofloopingthesoundoverandover again, modern game music is constantly adapting itself to 1.2. TheGeneralAggressionModel(GAM) thegameplayoftheplayer. Thisdynamicandadaptivemu- sicprovidesanimmenselyimmersiveexperience,thelikesof The ‘General Aggression Model’ as proposed by Anderson whichhaveneverbeenseenbefore. Researchattributesthis & Dill (2000), tries to assimilate various factors which ac- immersion to the “tight linkage between visual, kinaesthetic, count for aggressive behavior, in a unified model by merg- andauditorymodalities(Laurel,1993),theuseofrealisticau- ing together aspects of various theories that try to explain dio samples (Jørgensen, 2006) and spatial sounds (Murphy human aggression from different perspectives. It acknowl- & Pitt, 2001). Although this study, due to lack of technical edgesthatexposuretoviolentvideogamescanincreaseag- expertise, does not implement an adaptive dynamic game gressionbothinshorttermandlongterm. Italsotakesinto music scheme, it does however, implement a linear non- account that personal and situational factors can influence dynamic game music scheme. The aim of this study is to theaggressiveinternalstateofahumanviaanarousal,affec- find the effect of music tempo on the physiological arousal tiveandcognitiveroute(Raessens&Goldstein,2005). Play- and the aggression of a player while playing a first person ing violent video games can increase aggression by priming shooterlikeCounter-Strike. aggressive thoughts, increasing hostile feelings, or arousal. These three factors – cognition, affect, and arousal – are Tempo, the speed or pace of music, is perhaps the most ba- highly interrelated and activating one tends to activate the siccomponentofmusic’stemporaldimension(Duerr,1981; othertwo. Kellaris & Kent, 1993) and has long been held to be an im- portant determinant of listeners’ reactions (Hevner, 1937; Theenactmentofaggressioninvideogamesislargelybased Rigg,1940). AsGundlach(1932)observed,“Itappearsquite oncognitivefactorslikeknowledgestructures(e.g. learning certainthatmanypiecesofmusiccanelicitinmanyauditorsa ofaggressivescriptsorschemas). Feelingsandarousallevel, fairly uniform characterization solely through factors resident which are not per se specific effects of violent video games, withinthemusicalstructure"(p. 642). Forexample,fastmu- operate only as moderator variables. Cognition works on sictendstobeperceivedasmoreexcitingandslowmusicas twoinformationprocessingsystems,whichinfluencebehav- morerelaxing; musicpitchedinmajorkeystendstobeper- ior in a qualitatively different way. In the implicit system, ceived as happier than music pitched in minor (or atonal) informationisprocessedbyunconscious,automatic,andin- modalities. Theory and recent findings in musical aesthet- tuitive processes. Using this system unconsciously leads to icsallowustogobeyondBruner(1990)generalproposition activation in an associative network. The resulting behav- that "the components of music are capable of having main as iorisuncontrolledandimpulsive. Intheexplicitsystem,in- well as interactive effects on [feeling states] ..." (p. 99). For formation is processed by conscious, controlled, and reflec- example,onecananticipateapositivecontributionoftempo tiveprocesseswhichleadstothoughtfulandcontrollablebe- to arousal. Just as the human body adapts physiologically havior(Strack&Deutsch,2004).Whicheversystemisfinally 2 used depends on whether higher-order cognitive processes H1: The arousal will be significantly higher in the group are engaged or not (Berkowitz, 1993), the extent of which that played the first-person shooter with fast-paced music dependsontheparticularsituation. comparedtothegroupthatplayedthesamegamebutwith slowpacedmusic. 1.3. Theroleofmusicasmoderatorvariable H2: Thegroupthatlistenstofast-pacedmusicwhileplaying willshowhigheraggressiononimplicitaggressivemeasures Based on the GAM, Anderson & Bushman, 2001 conducted comparedtothegroupthatplayedthesamegamebutwith an experiment to examine the effect of violent versus non- slowpacedmusic. violent games on aggressive cognitions. The arousing ele- mentsinthegameswereheldconstant,sothatitcanbecon- H3: Arousal and aggression will be correlated. Highly cluded that aggressive knowledge structures were primed arousedparticipantswillshowmoreaggressivecognitions. solely by the violent content of a game. The study unveiled that violent content in video games led to a rise in aggres- sivecognitions. However,theexperimentleavesthequestion 2. Method openastowhatrole‘arousal’playsincreatingtheseaggres- sivecognitions. Thecurrentstudytriestofindananswerto thisquestionbyusingtempotomanipulatearousalandfind- ingitseffectonaggression. Previousstudieshaveshownthat 2.1. Design tempo is one of the features which correlates most strongly with physiological arousal (Gomez & Danuser, 2007). Car- A Posttest-Only between group design was adopted for the pentier & Potter (2007) reported in their experiment that experiment. One group played Counter-Strike with low fast-paced music elicits greater skin conductance level than tempo (Adagio) music and sound while in the other group, slow-paced music. We hypothesize that high arousal, in- subjects played Counter-Strike with high tempo (Allegro) duced by the high tempo music, will intensify the effect of music and sound. This study adopts the definition of mu- the violent game on aggressive cognitions. The mechanism sicandsoundasprovidedbyGrimshawetal.,2008. Sound behind this expected moderator effect of music can be ex- referstoalldiegeticsoundsoriginatingfromthegameenvi- plained by the ‘Excitation transfer theory’ by Zillerman in ronment such as gunshot sounds, footstep sounds, ambient which the arousal caused by one stimulus (the music) am- sounds etc. By music, we refer to the non-diegetic sound plifies the arousal response of another stimulus (aggressive (musicalsoundtrack)thatdoesnotoriginatefromthegame action). The valence of the second stimuli will then be am- charactersbutispresenttoprovideamoreimmersiveexpe- plified regardless of whether the valence of the two stimuli riencetotheplayer. is congruent or not. In order for this to work, two condi- tions have to be met. Firstly, the second stimulus should occur before the arousal from the first stimulus completely has decayed (Tannenbaum & Zillmann, 1975). Secondly, a 2.2. Participants misattributionofexcitationshouldtakeplacesothattheex- periencedarousalcanbefullyattributedasbelongingtothe There were 37 (22 male) participants who took part in the second stimulus (Cantor, Bryant & Zillmann, 1974). Both experiment. Around 76% of the respondents were Dutch criteria are met in our study. The music starts parallel with whiletheremaining24%wereallGerman. Theaverageage thegame,soanoverlapofthestimuliisguaranteed. Because of the sample was 22.65(SD = 5.58) years. Psychology and ofthe‘immersive’natureoffirst-personshooters,therespon- CommunicationStudiesstudentsmadeupfor41%and24% dentswon’tpaymuchattentiontothebackgroundmusicand ofthesamplerespectively,andfortheirparticipationtheyall attribute it’s arousal to the action in the game. Grimshaw, received 1 course credit. The remaining 35% of the partici- Lindley&Nacke(2008)describesthisas‘thefeelingofbeing pantswereeitherUniversityofTwenteemployeesorfriends encapsulatedinsidethegameworldandnotfeelinginfrontof andfamilyfromoutsidetheUT. amonitoranymore’. Accordingtotheirexperiment, ‘immer- sion’ was highest in the condition where both, non-diegetic 60% of the respondents played video games in one form or background music and diegetic game sounds were present. the other. The reported weekly game usage was 4.6 (SD = Theirconditionresemblesourtwoconditions. Soweexpect 7.22)hours. ThetwomostpreferredgamegenreswereFirst- ahighdegreeof‘immersion’inourexperiment. personShooters(40.5%)andRacingGames(32.4%). 60%of thosewhoplayedgames,reportedtoplayingtheirgameson Based on prior research, the present study assumes that a aPCand22%reportedplayingvideogamesonanxBox360. brief exposure to violent video games (situational input) temporally creates a more violent self-concept through the Allrespondentslistenedto2.95(SD=2.29)hoursofmusic cognitive route. We are interested whether arousal, caused daily. The most popular music genres among the respon- by background music, can strengthen this short-term effect dentswereRock(57%),Dance&Electronic(38%),andPop throughexcitationtransfer. Charts(32%) 3 2.3. Materials usedSupra-auralheadphonesduringgameplay. ThePCand headphone volume controls were turned to full for all the 2.3.1. VideogameandMusic participants. The sound levels were lowered internally in the game settings. Under these conditions the participants AportableversionofCounter-Strike1.6byValveCorporation in both conditions were subjected to music levels of 78 ± was used for the experiment. Because of the blood and in- 2dB, where the ± 2dB variation accounts for the difference tenseviolencecontainedinit,Counter-Strikehasearnedan between Laptop soundcards and the headphones vendors. ESRB (Entertainment Software Rating Board) rating of +M Duringgaming,participantsusedaGlockpistolandanMP5 (suited only for 17+ aged audience). It was chosen for this submachinegun,thesoundlevelsofwhichweresetto87dB studybecausesurveysfrom2002–2005revealedthatitwas ±2dBand85dB±2dB,respectively. consistentlythetopmostgameplayedonline. Standard distributions of the game only have diegetic 2.4. Procedure sounds. Therefore, two bespoke versions of the game were designedtoincorporatethehighandlowtemponon-diegetic Astheparticipantsenteredtheroom,oneoftheresearchers music in the game. The high tempo music was 140 BPM welcomedthemandhadsomesmalltalk. Atthesametime, ‘TRANCEFORMED 009’ by Matt Helden(excerpt extracted anotherresearcherstrappedontheQ-sensortothelefthand from 42:47 to 53:40) and the slow tempo music was the wristoftheparticipantandtheEDA(Electro-DermalActivity) 60 BPM ‘A Rose in Haiti’ by Mykle Anthony Alexander. The recording was started and a ten minute timer was started. Beats per minute measure for both soundtracks was deter- Theparticipantswerethengivenaninformedconsentform. minedbymanualtappinginNativeInstrument’sTraktorPro. Aftersigningtheconsenttheparticipantswereaskedforany Both tracks belong to the genre of Electronic Music. The prior experience with playing Counter-Strike. If the partic- high tempo track belongs to the subgenre trance, which is ipant reported no prior experience with the game, then the characterized by tempo values between 120 and 145 BPM. researcherexplainedthebasicgamecontrolsusingaprinted Theslowtempotrackbelongstothesubgenre,calledDown- manual. After10minuteselapsed,thegamewasstartedfor step/Downbeat. Unliketrance,thissubgenreischaracterized theparticipantandtheresearcherplacedamarkerintheQ bylaid-backmelodies. Bothsubgenreshaveclearbeats. The sensordatatomarktheendofthebaselineperiodandbegin- musictrackswerechosenaccordingtothefollowingcriteria: ningofthegamingperiod. Forthenext10minutes,thepar- absence of lyrics, not having a remarkably positive or neg- ticipantplayedthegamewhiletheresearcherwaitedoutside ative valence, clearly identifiable BPM, electronically made theroom. Afterallowingthesubjecttoplaythevideogame and not a recent hit in the charts. In this way we ensured for10minutes,theresearcherenteredtheroom,stoppedthe thatbothsoundtracks,moreorless,hadthesameattributes, game, placed a second marker, unhooked the sensor from except for their tempo. The exclusion of lyrics is important the wrist of the participants, started the IAT for the partic- because they can transfer meaning which could influence ipants and got out of the room. After completing the IAT mood. Bothsoundtracksweremonophonic,hadabitdepth test,ademographicsquestionnaireautomaticallypoppedup of16(CDQuality),andweresampledat11025Hz. for the subjects to respond to. After filling in the question- naire,theparticipantsweredebriefedabouttheexperiment Participants in both conditions played as a terrorist against iftheywantedto, andthankedfortheirparticipationinthe 3 counter-terrorists. The counter-terrorist bots were armed experiment. onlywithapistol,whereastheparticipant,playingasterror- ist had an MP5 gun and a Glock pistol. The music track 2.5. Measures started playing automatically at the beginning of the first round. The music then played uninterruptedly for the next Thedependentvariablesinthisexperimentarearousal,and 11minutesofgameplay. Theselectedmusictracksweredig- aggressionmeasuredviaskinconductanceandimplicitatti- itally looped to fit the 11 minutes durations. The difficulty tude test, respectively. These two dependent variables were level was set to Easy and the map was set to de_dust2on2. alsomeasuredusingaLikert-scalequestionnaire. The Counter-Strike settings panel was completely locked so the participants could not change any settings of the game. 2.5.1. SkinConductanceLevel Thisensuredthateveryallparticipantsplayedessentiallythe samegame. Skin conductance is one of the most robust non-invasive methods of measuring physiological arousal. Skin conduc- tance refers to how well the skin conducts electricity when 2.3.2. VideoGameandPCVolume anexternaldirectcurrentofconstantvoltageisapplied. The The loudness of music has been found to contribute posi- detected changes in electro dermal activity (EDA) indicate tivelytobothretrospectivedurationestimatesandperceived processes related to sympathetic arousal. The electric prop- pace of music stimulus (Kellaris, Mantel & Altsech, 2008). erties change within seconds and are closely related to psy- Therfore, a strict control of the audio volume was prac- chologicalprocesses(Figner&Murphy,2009). Weareinter- ticed to avoid introducing a confound. The participants estedintheoverallarousalandexcitementwhilethesubjects 4 play the game. Therefore, this research uses the Skin Con- thermore, because a moving average uses overlapping win- ductanceLevel(SCL)whichembodiesthelongtermarousal dows,itiscomputationallyveryslow. level of the player. However, how a person feels, cannot Owing to the issues, associated with the moving averaging be solely derived from the arousal level and thus not as- filter, we used a detrending technique for separating SCRs sessed in this study. Emotions are always determined by arousal level (high/low) and a corresponding valence (pos- from SCL. This technique never overestimates the SCL, is itive/negative). The emotional state of, for example, being computationallymuchfasterthanthetraditionalmovingav- eragefilter,andyieldsmuchbetterqualitySCRsandSCL. happycomparedtothatofbeingupsetjustdifferontheva- lence dimension. Both have a high arousal level (Barrett, The technique is based on the fact that SCRs die out very 1998). quickly and the GSR activity returns to the SCL trend. So if we could detect the minimas of SCR in small consecutive The skin conductance of the subject was measured using time windows, then we could interpolate these minimas to Affectiva Q-Sensor. The Q-Sensor is a wearable, wireless get an accurate representation of the SCL activity. Because biosensorGSR(GalvanicSkinResponse)deviceforgalvanic wearedetectingGSRminimaseverytime,ineachtimewin- skinresponseteststhatmeasuresemotionalarousal. Besides dow, therefore by design, this technique could never over- measuring the skin conductance, it also measures the tem- estimateSCL.Thisapproachofbaselinetrendestimationby perature and 3-axis acceleration. However, for our experi- fitting a curve locally to the intensity minima works a lot mentweonlyusedtheskinconductancedata. Thiswireless fasterthanthemovingaveragefilter,andyieldsmuchbetter sensor was housed in a wrist band that was attached to the SCLestimation. Thewholeprocesscanbedividedintothree distant end of ulna on the left hand of the participant. The steps: samplerateforthesensorwassetto32Hz. Step1: DividetheentireGSRactivityintoanumberofnon- Affectiva, the vendor of Q-sensor, recommends that the overlapping windows. For our case, the size of the overlap- recordings should be used after 15 minutes of hooking the pingwindowwassetto480samples,whichat32Hzsample sensortothesubject,butourpretestsrevealedthatthedata translatesto15secondsofdata. Therationalebehindusing becomes stable and reliable after 8 minutes. Therefore, 10 awindowsizeof480samplesisthattheSCRstake10to20 minutesofbaselinedatawascollected,thefirst8minutesof seconds to die down to the SCL level. Therefore, selecting which was discarded and only the last two minutes of data awindowlengthinthisrangeyieldsbetterresultsthanany were used for determining the baseline mean. The gaming otherwindowsize. interval was defined as all the data collected 10 minute af- Step2: FindaglobalminimumpointintheGSRinsideeach ter placing the first marker M1. Only the first 9 minutes of ofthewindowsproducedinthepreviousstep. this10minutedatawasused. Thelastoneminutewasdis- Step 3: Linearly, quadratically, or cubically interpolate each cardedtoavoidtheedgeeffectsofthefilterusedtofilterout oftheseconsecutivepointstoobtainasmoothwaveformthat theSCRs(SkinconductanceResponses). wouldquiteaccuratelyrepresenttheskinconductancelevel without any overestimation. For our case, we used linear 2.5.2. IsolatingSCLfromSCRs interpolation. Skin conductance level (SCL) is used in this study to inves- The filter was designed and implemented in Matlab(cid:13)R tigating the effect of game music on the general long term R2010b. The comparison of results between this algorithm arousaloftheplayer. Ahigherskinconductancelevelmeans and the traditional moving average filter are explained in higher arousal and vice versa. However, the skin conduc- Figure 1. The blue trace in this figure represents the raw tance level is masked by short term physiological responses unprocessed GSR activity of a subject. It contains both SCL thatshowupintheGSRrecordingsasSkinConductanceRe- and SCR’s. The red trace shows the result of using a 5120 sponses (SCRs). Skin conductance responses show up vari- point moving average filter on the raw GSR. The filter re- ous peaks in the GSR. Usually in psychological studies, the movesSCR’sbutresultsinoverestimatingtheSCLinregions norm is to average the GSR activity over a certain period, of high SCR concentration. The green trace shows the SCL usingaslidingwindowfilter(alsocalledmovingaveragefil- generated by our non-overlapping window based spline fit ter), to smooth out the SCRs in order to get a measure for algorithm. This filter follows local minimas in consecutive theSCL.Mathematically,amovingaveragefilterisgivenas: timewindowsandhence, neveroverestimatestheSCL.The window size is set to 480 samples (or 15 seconds at 32 Hz x¯(n)= 1 (cid:88)N x(n−k) samplingrate). TheblacktraceshowstheSCR’sobtainedus- N ingourinon-overlappingwindowbasedsplinefitalgorithm. k=1 ItwasobtainedbyjustsubtractingtheSCL,thatweobtained where N in the above equation is Window Size. Although using this filter, from GSR because SCR = GSR - SCL. The movingaveragefiltersmoothsouttheSCRs,butitcanmas- SCR’s are very clean with no baseline SCL in it. The pink sively overestimate the SCL in the regions where there is a trace was obtained by subtracting the SCL obtained using highSCRconcentration(Figner&Murphy,2009). Itcanalso themovingaveragefilter,fromtheGSR.Asyoucansee,the underestimate the SCL in certain regions of the GSR. Fur- 5 failureofthemovingaveragefiltertoaccuratelydetectSCL, results in a failure to produce pure SCRs. The SCR’s gener- ated by this filter contains reminiscences of the SCL in the regionswhereithadoverorunderestimatedtheSCL.[Note: Adcoffsetof-1wasaddedtothiswaveformwithoutwhich the two SCR waveforms (pink and black one)overlap and thedifferenceisdifficulttoobservethen]MarkerM1wasset whenthegamewasstartedanditmarkstheendofthebase- linedurationandstartofthegameplay. MarkerM2wasset after10minutesofgameplay. Onlythelasttwominutesof data before marker M1, was used for baselining because in theselasttwominutesthesensordataisreliableandstable. ThelastminuteofthedatabeforemarkerM2wasdiscarded toremovecorrupteddataproducedbytheedgeeffectofthe filter. So only nine minutes of SCL activity after marker M1 Figure 1: Comparison Result of ‘Moving Average’ and ‘Local MinimaSplineFit’algorithmforseparatingSCLandSCR’s wasused. getthemeanvalueofskinconductanceduringgaming. 2.5.3. NormalizingSCL 1 1(cid:88)7280 Skin Conductance level, like any other psychophysiological µ =MeanGamingSkinConductance= x (i) output variables, displays marked individual differences in g 17280 i=1 Ng the maximum and often in the minimum levels of which Finally the baseline mean skin conductance was subtracted subject is capable. Some subjects, known as Labiles, show from gaming mean skin conductance to get a baseline cor- a very large variation in their skin conductance, whereas rectedvalueofskinconductanceduringgaming. somesubjectsknownasStabiles,showaverysmallvariation their skin conductance. This was demonstrated in one of µ=µ −µ g b thestudiesdonebyLykken(1968)inwhichoneofthesub- This process of first filtering the data, normalizing it, and jects, who was relaxing for 30 minutes had minimum skin finding the mean skin conductance level, was repeated for conductance which was 2 times larger than another subject theGSRdataofeachofthe37respondents. who was blowing up balloons to bursting. Surely, the first subjectwasn’tmorearousedthenthesecondone. Suchvari- ations in range, are generally unrelated to the underlying variable of interest, and must be corrected so as to remove 2.6. ImplicitAssociationTest their influence. Lykken, Rose & Luther(1966) introduced a methodofnormalizingtheskinconductancetoremovethis To measure the aggressive self-concept of the subjects af- inter-subject variability in the range. The method works by ter playing the video game, an IAT (Implicit Association Test finding the global minimum and maximum in the skin con- or Implicit Attitude Test) was conducted. A self-concept is ductanceofasubjectandtransformingtheoriginaldatainto the total sum of beliefs that people have about themselves anormalizedversion: and which guide the processing of self-relevant information x(i)−x (Brehm, Kassin & Fein, 2005). The IAT or other implicit x (i)= min N x −x measures are useful for assessing aggressiveness for three max min primaryreasons. First,explicitmeasuressuchasself-reports Where x(i)and xN(i)representtheoriginalandnormalized tap only into processes involved in the explicit information samplesrespectively. xmin and xmax aretheglobalminimum processing mode. Second, the IAT has been demonstrated andglobalmaximuminthedataofasubjectrespectively. to have incremental validity over and above explicit mea- sures. StudieshaveshownthattheIAThasreliabilityaround After the skin conductance was normalized, it was split 0.80 for internal consistency, and 0.60 for test-retest stabil- into baseline data and gaming data with respect to the first ity as well as good construct and predictive validity (Das- marker. All 3840 samples, corresponding to two minutes of gupta & Greenwald, 2001; Greenwald & Farnham, 2000). datacollected beforethefirst markerM1, wereaveraged to Finally, aggressive behavior is typically socially undesirable getthemeanvalueforbaselineskinconductance. and reports of such are therefore subject to social desirabil- 1 3(cid:88)840 ity biases, so the predictive validity of explicit measures for µ =MeanBaselineSkinConductance= x (i) b 3840 Nb aggressive behavior is questionable. Implicit measures may i=1 help to improve the prediction of aggressive behavior. This All 17280 sample points, corresponding to nine minutes of isduetothehard-to-controlspontaneousassociationswhich data collected after the first marker M1, were averaged to suchtestsreveal(Bluemke,Friedrich&Zumbach,2010). 6 TheImplicitAssociationTest(IAT)isacomputerbasedsort- Concept-Attribute Pairing. In the fifth stage of the task, re- ing task that indirectly measures the strength of the asso- spondentssortitemsfromboththeattributeandtargetcon- ciation between two contrasted target categories (e.g., Me cept categories again, except that the response key assign- versus Others) and two bipolar attribute categories (e.g., mentsnowrequireOthersorPeacefulitemstobecategorized Peaceful versus Aggressive) via a computerized classification with one key and Me or Aggressive items to be categorized task(Richetin & Richardson, 2008). The IAT obliges the re- withtheotherkey, theoppositeassociationfromtheearlier spondents to identify stimulus items and categorize them block. Respondents sort stimulus items with this response into one of these four superordinate categories. Associa- assignmentforthenext20trials. Thisisthepracticeblock. tion strengths are measured by comparing the speed of cat- Block7–(IncompatibleTest2)isexactlysimilartotheblock egorizing members of the superordinate categories in two 6, but in this block 40 trails are done instead of 20. This different sorting conditions (Nosek, Greenwald & Banaji, block isalso referred toas the Criticalblock). The latencies 2005;Uhlmann&Swanson,2004). of the responses generated in this critical block are used in The IAT’s procedure has 7 blocks, with block 3 & 4 and 6 & thecalculatingthefinalDscore. 7providingthemostcriticaldata(Noseketal.,2005). TheIATeffectiscalculatedusinglatencydatafromSteps3& Block 1–(Target Compatible Practice) involves Learning The 4and6&7. Intheaboveexample,sortingthestimulusitems Concept Dimension. The respondents sort items from two faster when Me or Aggressive (and Others or Peaceful) share different concepts into their superordinate categories (e.g., aresponsekeythanthereversepairingsindicatesastronger ‘I’ for Me and ‘they’ for Others). Categorizations are made association strength between Me and Aggressive (and Oth- using two different keys on a computer keyboard that are ers or Peaceful), which means that the respondent is more mappedtothesuperordinatecategories(e.g.,the‘E’keyfor aggressivethanpeaceful. Meandthe‘I’keyforOthers)andstimulusitemsappearse- The IAT was designed in InquisitTM 3.0.4.0 by Millisecond quentiallyinthemiddleofthecomputerscreen(seeTable1). Software. BecausetheIATusesquitealotofcomplexadjec- Respondentsperform20trialswiththesesortingrules. tives in each category, it is imperative that implicit attitude Block 2–(Attribute Practice) involves Learning The Attribute test be carried out in the respondents’ native language. Be- Dimension. Therespondentsperformthesametaskwiththe cause almost all the participants were either Dutch or Ger- same two keys but now sort items representing two poles man, therefore the IAT was designed in three different lan- of an attribute dimension (e.g., Violent for Aggressive and guages: English, German, and Dutch (refer to Tabel A.2 to Friendly for Peaceful). Respondents perform 20 trials with seeallthestimulswordsusedinallthethreelanguages). Be- thesesortingrules. fore starting each IAT, instructions appeared on screen that asked user to select their native language. Once the choice Block 3–(Compatible Test 1) involves Learning Concept- was made, the IAT started in that language. An ITI (Inter AttributePairing. Inthisstep,thetwosortingtasksofblock1 Trial Interval) of 250 ms was used. Furthermore, the com- and2arecombinedsuchthat, onalternatingtrials, respon- patible blocks (2, 3 & 4) and incompatible blocks (5, 6 & dents are identifying a stimulus as Me or Peaceful and then 7) were administered in counterbalanced order across the a word as Others or Aggressive. In this case, one key (‘E’) is sampletocontrolforblockordereffects. thecorrectresponsefortwocategories (MeorPeaceful)and the other key (‘I’) is the correct response for the other two categories (Others or Aggressive). Respondents perform 20 2.6.1. CalculatingIATDScore trialswiththesesortingrulesinthisblock(oftenreferredto Greenwald, Nosek & Banaji (2003) described in detail the asthePracticeblock). scoring algorithm for calculating the IAT effect. It involves Block4–(CompatibleTest2)isexactlysimilartotheblock3, calculating the difference in average response latency be- butinthisblock,40trailsaredoneinsteadof20. Thisblock tween the two sorting conditions and dividing by the stan- is often referred to as the Critical block). The latencies of dard deviation of all latencies for both sorting tasks. Thus, theresponsesgeneratedinthiscriticalblockareusedinthe the IAT score (called D) is a calculation of effect size for an calculatingthefinalIATDscore. individual’sresponsesinthetask. TheIATDscoreisrelated toCohen’sd. Block5–(TargetIncompatiblePractice)involvesLearningRe- versedConceptDimension. Inthisstageofthetask,onlystim- Before the D score for a subject is calculated, all the trails ulusitemsforthetargetconcepts(MeandOthers)aresorted havingalatencygreaterthan10,000msarediscarded(Lane, for20trials,butthistimethekeyassignmentisreversed. In Banaji, Nosek & Greenwald, 2007). After cleaning the data thepresentexample,itemsinOtherscategorywouldnowre- from these outliers, the following calculations are done on quirean‘E’keyresponseanditemsintheMecategorywould theremainingtrailstocalculatetheIATDscore: requirean‘I’keyresponse. Step 1: Find the mean response latency µ , of all the trails 3 Block 6–(Incompatible Test 1) involves Learning Reversed inblock3. Similarly, themeanresponselatencyi.e. µ4, µ6, µ forblock4,6&7respectively,arecalculated. 7 7 Table1: DifferentBlocksofanImplicitAssociationTest(IAT) Block CategoryLabels StimulusItems CategoryLabels Block1: Me Others • ◦ LearningTheConceptDimension I ◦ • (20Trials) Them Block2: Aggressive Peaceful ◦ • LearningTheAttributeDimension Friendly • ◦ (20Trials) Violent Block3: PracticeBlock Me Others LearningConcept-AttributePairing or or (20Trials) Aggressive Peaceful • ◦ & I • ◦ Block4: CriticalBlock Violent ◦ • DoingConcept-AttributePairing Them ◦ • (40Trials) Friendly Block5: Others Me • ◦ LearningReversedConceptDimension Them ◦ • (20Trials) I Block6: PracticeBlock Me Others LearningReversedConcept-AttributePairing or or (20Trials) Peaceful Aggressive • ◦ & I ◦ • Block7: CriticalBlock Violent ◦ • DoingReversedConcept-AttributePairing Them • ◦ (40Trials) Friendly SchematicdescriptionandillustrationofanImplicitAssociationTest. TheIATconsistsofseriesof7discriminationtasks. Thereare20trialsin eachblockexceptforblock4and7,whichcontain40trialseach. Theblackdotsindicatesthecorrectresponseandthewhitedotsrepresentsa wrongresponse. Block3&6arereferredtoas‘PracticeBlocks’andBlock4&7arereferredtoas‘CriticalBlocks’. Theresponselatencydata collected during these performing the Practice and Critical Blocks, is used to calculate the IAT D score. The compatible blocks (2,3 & 4) and incompatibleblocks(5,6&7)wereadministeredincounterbalancedorderacrossthesampletocontrolforblockordereffects. Step2: Calculatethepooledstandarddeviationofthelaten- Step 6: Calculate the IAT D score by averaging the D score ciesσpooled(3,6) inthepracticeblocks3and6. forpracticeandcriticalblocks. Step3: Calculatethepooledstandarddeviationofthelaten- D= Dpractice+Dcritical ciesσpooled(4,7) inthecriticalblocks4and7. 2 Step 4: Divide the difference of mean latencies of practice TheresultingIATscore(calledD)rangesfrom-2to+2. Pos- blocks3and6bytheirpooledstandarddeviationtogetthe itiveIATscoresindicateapeacefulself-conceptandnegative Dscoreforpracticeblocks. scores an aggressive self-concept (Greenwald & Farnham, 2000). µ −µ D = 6 3 practice σ pooled(3,6) 2.7. Self-ReportMeasures Step 5: Divide the difference of mean latencies of critical blocks4and7bytheirpooledstandarddeviationtogetthe Besides obtaining general demographics data, the follow- Dscoreforcriticalblocks. ing explicit self-report measures were also obtained by ask- µ −µ ing the subjects to give a self-evaluation on a 7-point-Likert D = 7 4 critical σ Scale. pooled(4,7) 1. Howaggressiveareyoufeeling? 8 2. How interesting the music was while you were playing indicatorforarousal,wefoundnodifferences. However,the thegame? hypothesis is true when looking at the self-reported arousal 3. Howarousingwasthemusicwhileyouwereplayingthe of the subjects. Hypothesis 2, i.e. ‘The group which listens videogame? to high tempo music while playing will show higher aggres- siononimplicitaggressivemeasures’ couldnotbeconfirmed. Theresponsevalue1fortheabovethreequestionaretrans- Theindirectassumptionbehindthehypothesiswasthatthe lated as ‘very arousedŠ, ‘very interestingŠ, and ‘very arous- high BPM condition would score lower on the IAT-test, as ingŠrespectively. Whereas,aresponsevalueof7means‘very a consequence of the high arousal. Hypothesis 2 thus has calm’, ‘very boring’, and ‘very relaxing’ respectively for the to be rejected. This result is not entirely surprising because threequestionmentionedabove. thetwoconditionsdidnotdiffersignificantlyintheirarousal (i.e. skinconductancelevel)aswell. 3. Results 3.3. CorrelationbetweenArousalandAggression According to the excitation transfer theory, the arousal pro- 3.1. Distributionofthedata ducedwhilstlisteningtomusicshouldgettransferredtothe in-game actions and that would consequently have ampli- Kolmogorov Smirnov test was carried on all relevant data to testforthenormalityofdata. TheIATDscore(Z =0.46; p fied the violent acts in the game. This should have been = 0.98), skin conductance data (Z = 1.03; p = 0.22), and reflected in the IAT-score. We neither found a difference in bodilyarousalnorontheIATscoresbetweenthegroups. The the two Likert scale questions: ‘How aggressive the subjects feltaftergaming?’ (Z=1.29;p=0.07),and‘Howinteresting arousalwemeasuredisthusunrelatedtothemusic. the music was while you were playing the game?’ (Z = 1.05; Nevertheless, it would be interesting to see if the arousal p = 0.22), all turned out to be normally distributed. Only levelwhilegamingandtheaggressiveaffectiscorrelatedin the third Likert scale question ‘How arousing the music was anywaytotheIATscore. Wehandlethefollowingallocation whenplayingthevideogame?’ (Z=1.418;p=0.036)hada ofcorrelations: non-normaldistribution. r≤0.30 LowCorrelation 0.30<r<0.60 ModerateCorrelation None of the variables above contained outliers, where an r≥0.60 HighCorrelation outlier was defined as a point in a data set which is located morethan1.5XInterquartileRangeabovethethirdquartile Foraggressiveeffect, thecorrelationwiththeIAT(r=0.32) orbelowthefirstquartile(Moore&McCabe,2005). wasmoderatepositivelysignificant(α<0.05). Arousalwas measured directly via skin conductance and indirectly via self-indication. We are most interested in the first one, be- 3.2. Differencesbetweenthetwoconditions cause itis the directmeasure of arousal. APearson correla- A two-sided independent sample t-test with an alpha of 0.05 tionbetweenSCLandIAT-scorewasconducted,onceforall wasusedtocomparethegroupwhichlistenedtofast-paced subjects (n = 37) and once for each condition apart. There music (n = 18) with the group which listened to low-paced is a moderate negatively significant correlation (r = -0.31) music (n = 19). The t-test reveals that the two conditions between SCL and IAT score (α < 0.05) in general. When do not differ significantly from each other in their skin con- splittedintotwogroups,itrevealsthatthereisastrongneg- ductancelevel(T =-1.100;p=0.279),theirIATscore(T = ative significant correlation (r = -0.62) between IAT scores -0.107;p=0.916),andtheirself-reportedaggressivefeeling andSCLinthehigh-tempogroup,whereasthereisnosignif- measure (T = -1.475; p = 0.15). However, the group that icantcorrelationinthelowtempogroup(α 0.01). Probably listenedtohightempomusicfoundthemusicmoreinterest- thisisanartifact. Inthefast-pacedconditionwerebychance ingascomparedtothegroupthatlistenedtothelowtempo more subjects with more extreme skin conductance. That music(T =-3.19;p=0.003). couldexplainwhythecorrelationjustbecamesignificantin that condition (see Figure 2). Contrary to the skin conduc- Tocomparethetwogroupsontheirself-reportedarousalon tance,self-indicatedarousalshowsnosignificantcorrelation a7-pointLikertscale(where1standsforbeing‘veryaroused’ withIAT-score. Hypothesis3i.e. ‘Arousalandaggressionwill and 7 for ‘very relaxed’), a Mann-Whitney U test with an al- showcorrelation. Highlyarousedparticipantswillshowmore pha of 0.05 was used. The test indicated that self-reported aggressivecognitions’isconfirmedpartly. arousalwasgreaterforthegroupthatlistenedtohightempo music (Mdn = 2.5) compared to the group that listened to thelowtempomusic(Mdn=5)and(U =68.5,p=0.007). 4. Discussion Hypotheses 1, i.e. ‘The arousal will be significant higher in the group which played video game with high tempo music’ Thepresentstudyattemptedtoinvestigatethepotentialrole cannotbeconfirmedfully. Forskinconductance,whichisan of arousal as a moderator for aggressive cognitions. It was 9 Figure2: CorrelationbetweenSCLandIAT-score 4.1. Limitationsofthestudy The study failed to corroborate existing studies about the effect of music tempo on arousal. There could be several possibleexplanationstoit. Itmaybebecausethelowtempo musicwasnotreallylowtempoatall. Tempoisverydifficult thingtomeasure. Wetriedtomeasureitusingfivedifferent professional DJ softwares and each one of them gave dif- ferent BPM readout. Ultimately, it was decided to resort to manualcountingbytappingthebeatsinNativeInstruments’ Traktor Pro software. However, this technique of manual tapping the beats, requires one to have a good background inmusictocorrectlyrecognizeandregisterabeatasshown byoneofthestudiesdonebyMcAuley&Semple(1999). So it could be that the lack of formal musical background, led to wrong estimates of BPM using the manual tapping tech- nique, resulting in wrong estimates of the tempo. Future studiesshouldusesoundtracksthathavetheirtemporatings verifiedbyamusicalexpert. A further limitation of this study is that the subgenre of the lowandhightempomusicarenotexactlythesame. Astudy by Kellaris & Kent (1993) shows that genre, has moderat- ing effect on the tempo and arousal. It could be a reason thatsubgenreofthelowtempomusicwassooverpowering Note: Thecirclesandsquaresrepresentparticipantdatainlowand andarousingcomparedtothehightempomusicthatitcom- hightempoconditionsrespectively. Theredandblacklinesshowthe linearfittodatainhighandlowtempoconditionsrespectively. pletely masked the effect of tempo and in effect, the study started measuring the effect of subgenre rather than the ef- fect of tempo, as originally intended. Future studies should keepthegenre, subgenre, tonality, andtextureofthemusic the same while manipulating the tempo. Keeping these pa- rameters the same, while manipulating the tempo is a very hypothesized that high tempo music will induce high levels difficult task. It can be done in certain software packages, of arousal which will consequently result in higher aggres- butitcompletelydestroysthemusicalaestheticsandfidelity sivetendenciesandviceversa. Theresultsrevealedthatmu- of the soundtrack in the process. Human ear can easily de- sic tempo plays no role whatsoever in modulating arousal tectthatthetempooftheoriginalsoundtrackhasbeendoc- in the gaming context. This is sharp in contrast to various tored. However, a study by Richards, Fassbendera, Bilginb musicstudieswherethetempoofmusicwasshowntomod- & Thompson (2008), advocates the use of a software Able- ulate the arousal and pleasure. The only possible explana- tonLivebyDAW(DigitalAudioWorkstation). Inthatstudy, tiontothiscontrastisthatthosestudiesusedmusicinstan- the tempo of a music track was varied by as much as 21- daloneformanddidn’tcouplethemusicwithavideogame. 26% and the doctored soundtracks were presented to three The complex gameplay coupled with various other factors musicalexpertswhowereunabletodetectanyirritatingdis- (discussed below), might have contributed to the complete tortionsandmanipulations. Futurestudiesshouldtherefore maskingoftheeffectofthetempoonarousal. Tofindifthis explore the use of this software. An alternative could be to was the case, a further experiment should be conducted on composethelowandhightempomusictracksusingprofes- our two chosen soundtracks. But this time, only the music sional musicians who can change tempo while keeping the shouldplayandnotthevideogame. Ifourlowtemposound- genre,texture,andtonalitymoreorlessthesame. trackresultsinlowarousalandhightempomusicresultsin higharousaleffect,thenitwouldconfirmourhypothesisthat One shortcoming of our game music design is that it’s Non- video games do have an influence in masking the arousing Dynamic Linear Sound i.e. it loops over and over again and effectsofmusic. isunaffectedbyplayers’inputandgameplay. Modernvideo gamesuseAdaptiveNon-diegeticSoundsi.e. musictracksthat The study did, however, find a significant correlation be- occurinreactiontothegameplay. Theadaptivenon-diegetic tweenarousalandaggression,showingageneraltrendthat music provides a more immersive experience to the player. more aroused subjects showed more aggressive cognitions Futurestudiesshouldincorporatethiskindofadaptivemusic andviceversa. ThisresultisinlinewiththeExcitationtrans- tomimicmoderngames. fertheory. Arousalseemingly,wastransferredtotheviolece ingameactionsandreflectedintheIATscores. The IAT used for the experiment has many flaws. To start 10
Description: