Chat with us, powered by LiveChat Psychology - Essayabode - Page 438

Please read the attached pages. - respond to questions - Summary responding question Experimentalevidenceofmassive-scaleemotionalcontagionthroughsocialnetworksAdamD.I.Kramera,1,JamieE.Guilloryb,2,andJeffreyT.Hancockb,caCoreDataScienceTeam,Facebook,Inc.,MenloPark,CA94025;andDepartmentsofbCommunicationandcInformationScience,CornellUniversity,Ithaca,NY14853EditedbySusanT.Fiske,PrincetonUniversity,Princeton,NJ,andapprovedMarch25,2014(receivedforreviewOctober23,2013)Emotionalstatescanbetransferredtoothersviaemotionalcontagion,leadingpeopletoexperiencethesameemotionswithouttheirawareness.Emotionalcontagioniswellestablishedinlaboratoryexperiments,withpeopletransferringpositiveandnegativeemotionstoothers.Datafromalargereal-worldsocialnetwork,collectedovera20-yperiodsuggeststhatlonger-lastingmoods(e.g.,depression,happiness)canbetransferredthroughnetworks[FowlerJH,ChristakisNA(2008)BMJ337:a2338],al-thoughtheresultsarecontroversial.InanexperimentwithpeoplewhouseFacebook,wetestwhetheremotionalcontagionoccursoutsideofin-personinteractionbetweenindividualsbyreducingtheamountofemotionalcontentintheNewsFeed.Whenpositiveexpressionswerereduced,peopleproducedfewerpositivepostsandmorenegativeposts;whennegativeexpressionswerere-duced,theoppositepatternoccurred.TheseresultsindicatethatemotionsexpressedbyothersonFacebookinfluenceourownemotions,constitutingexperimentalevidenceformassive-scalecontagionviasocialnetworks.Thisworkalsosuggeststhat,incontrasttoprevailingassumptions,in-personinteractionandnon-verbalcuesarenotstrictlynecessaryforemotionalcontagion,andthattheobservationofothersÕpositiveexperiencesconstitutesapositiveexperienceforpeople.computer-mediatedcommunication|socialmedia|bigdataEmotionalstatescanbetransferredtoothersviaemotionalcontagion,leadingthemtoexperiencethesameemotionsasthosearoundthem.Emotionalcontagioniswellestablishedinlaboratoryexperiments(1),inwhichpeopletransferpositiveandnegativemoodsandemotionstoothers.Similarly,datafromalarge,real-worldsocialnetworkcollectedovera20-yperiodsuggeststhatlonger-lastingmoods(e.g.,depression,happiness)canbetransferredthroughnetworksaswell(2,3).TheinterpretationofthisnetworkeffectascontagionofmoodhascomeunderscrutinyduetothestudyÕscorrelationalnature,includingconcernsovermisspecificationofcontextualvariablesorfailuretoaccountforsharedexperiences(4,5),raisingim-portantquestionsregardingcontagionprocessesinnetworks.Anexperimentalapproachcanaddressthisscrutinydirectly;how-ever,methodsusedincontrolledexperimentshavebeencriti-cizedforexaminingemotionsaftersocialinteractions.Interactingwithahappypersonispleasant(andanunhappyperson,un-pleasant).Assuch,contagionmayresultfromexperiencinganinteractionratherthanexposuretoapartnerÕsemotion.Priorstudieshavealsofailedtoaddresswhethernonverbalcuesarenecessaryforcontagiontooccur,orifverbalcuesalonesuffice.Evidencethatpositiveandnegativemoodsarecorrelatedinnetworks(2,3)suggeststhatthisispossible,butthecausalquestionofwhethercontagionprocessesoccurforemotionsinmassivesocialnetworksremainselusiveintheabsenceofex-perimentalevidence.Further,othershavesuggestedthatinonlinesocialnetworks,exposuretothehappinessofothersmayactuallybedepressingtous,producinganÒalonetogetherÓsocialcomparisoneffect(6).Threestudieshavelaidthegroundworkfortestingthesepro-cessesviaFacebook,thelargestonlinesocialnetwork.Thisresearchdemonstratedthat(i)emotionalcontagionoccursviatext-basedcomputer-mediatedcommunication(7);(ii)contagionofpsy-chologicalandphysiologicalqualitieshasbeensuggestedbasedoncorrelationaldataforsocialnetworksgenerally(7,8);and(iii)peopleÕsemotionalexpressionsonFacebookpredictfriendsÕemotionalexpressions,evendayslater(7)(althoughsomesharedexperiencesmayinfactlastseveraldays).Todate,however,thereisnoexperimentalevidencethatemotionsormoodsarecontagiousintheabsenceofdirectinteractionbetweenexperiencerandtarget.OnFacebook,peoplefrequentlyexpressemotions,whicharelaterseenbytheirfriendsviaFacebookÕsÒNewsFeedÓproduct(8).BecausepeopleÕsfriendsfrequentlyproducemuchmorecontentthanonepersoncanview,theNewsFeedfiltersposts,stories,andactivitiesundertakenbyfriends.NewsFeedistheprimarymannerbywhichpeopleseecontentthatfriendsshare.WhichcontentisshownoromittedintheNewsFeedisde-terminedviaarankingalgorithmthatFacebookcontinuallydevelopsandtestsintheinterestofshowingviewersthecontenttheywillfindmostrelevantandengaging.Onesuchtestisreportedinthisstudy:Atestofwhetherpostswithemotionalcontentaremoreengaging.Theexperimentmanipulatedtheextenttowhichpeople(N=689,003)wereexposedtoemotionalexpressionsintheirNewsFeed.Thistestedwhetherexposuretoemotionsledpeopletochangetheirownpostingbehaviors,inparticularwhetherex-posuretoemotionalcontentledpeopletopostcontentthatwasconsistentwiththeexposureÑtherebytestingwhetherexposuretoverbalaffectiveexpressionsleadstosimilarverbalexpressions,aformofemotionalcontagion.PeoplewhoviewedFacebookinEnglishwerequalifiedforselectionintotheexperiment.Twoparallelexperimentswereconductedforpositiveandnegativeemotion:OneinwhichexposuretofriendsÕpositiveemotionalcontentintheirNewsFeedwasreduced,andoneinwhichex-posuretonegativeemotionalcontentintheirNewsFeedwasreduced.Intheseconditions,whenapersonloadedtheirNewsFeed,poststhatcontainedemotionalcontentoftherelevantemotionalvalence,eachemotionalposthadbetweena10%andSignificanceWeshow,viaamassive(N=689,003)experimentonFacebook,thatemotionalstatescanbetransferredtoothersviaemotionalcontagion,leadingpeopletoexperiencethesameemotionswithouttheirawareness.Weprovideexperimentalevidencethatemotionalcontagionoccurswithoutdirectinteractionbe-tweenpeople(exposuretoafriendexpressinganemotionissufficient),andinthecompleteabsenceofnonverbalcues.Authorcontributions:A.D.I.K.,J.E.G.,andJ.T.H.designedresearch;A.D.I.K.performedresearch;A.D.I.K.analyzeddata;andA.D.I.K.,J.E.G.,andJ.T.H.wrotethepaper.Theauthorsdeclarenoconflictofinterest.ThisarticleisaPNASDirectSubmission.FreelyavailableonlinethroughthePNASopenaccessoption.1Towhomcorrespondenceshouldbeaddressed.Email:[email protected]:CenterforTobaccoControlResearchandEducation,UniversityofCalifornia,SanFrancisco,CA94143.8788Ð8790|PNAS|June17,2014|vol.111|no.24www.pnas.org/cgi/doi/10.1073/pnas.1320040111 90%chance(basedontheirUserID)ofbeingomittedfromtheirNewsFeedforthatspecificviewing.ItisimportanttonotethatthiscontentwasalwaysavailablebyviewingafriendÕscon-tentdirectlybygoingtothatfriendÕsÒwallÓorÒtimeline,ÓratherthanviatheNewsFeed.Further,theomittedcontentmayhaveappearedonpriororsubsequentviewsoftheNewsFeed.Fi-nally,theexperimentdidnotaffectanydirectmessagessentfromoneusertoanother.Postsweredeterminedtobepositiveornegativeiftheycon-tainedatleastonepositiveornegativeword,asdefinedbyLinguisticInquiryandWordCountsoftware(LIWC2007)(9)wordcountingsystem,whichcorrelateswithself-reportedandphysiologicalmeasuresofwell-being,andhasbeenusedinpriorresearchonemotionalexpression(7,8,10).LIWCwasadaptedtorunontheHadoopMap/Reducesystem(11)andintheNewsFeedfilteringsystem,suchthatnotextwasseenbytheresearchers.Assuch,itwasconsistentwithFacebookÕsDataUsePolicy,towhichallusersagreepriortocreatinganaccountonFacebook,constitutinginformedconsentforthisresearch.Bothexperimentshadacontrolcondition,inwhichasimilarpro-portionofpostsintheirNewsFeedwereomittedentirelyatrandom(i.e.,withoutrespecttoemotionalcontent).Separatecontrolconditionswerenecessaryas22.4%ofpostscontainednegativewords,whereas46.8%ofpostscontainedpositivewords.Soforapersonforwhom10%ofpostscontainingposi-tivecontentwereomitted,anappropriatecontrolwouldwith-hold10%of46.8%(i.e.,4.68%)ofpostsatrandom,comparedwithomittingonly2.24%oftheNewsFeedinthenegativity-reducedcontrol.Theexperimentstookplacefor1wk(January11Ð18,2012).ParticipantswererandomlyselectedbasedontheirUserID,resultinginatotalof??155,000participantsperconditionwhopostedatleastonestatusupdateduringtheexperimentalperiod.Foreachexperiment,twodependentvariableswereexaminedpertainingtoemotionalityexpressedinpeopleÕsownstatusupdates:thepercentageofallwordsproducedbyagivenpersonthatwaseitherpositiveornegativeduringtheexperimentalperiod(asinref.7).Intotal,over3millionpostswereanalyzed,containingover122millionwords,4millionofwhichwerepositive(3.6%)and1.8millionnegative(1.6%).IfaffectivestatesarecontagiousviaverbalexpressionsonFacebook(ouroperationalizationofemotionalcontagion),peo-pleinthepositivity-reducedconditionshouldbelesspositivecomparedwiththeircontrol,andpeopleinthenegativity-reducedconditionshouldbelessnegative.Asasecondarymea-sure,wetestedforcross-emotionalcontagioninwhichtheoppositeemotionshouldbeinverselyaffected:Peopleinthepositivity-reducedconditionshouldexpressincreasednegativity,whereaspeopleinthenegativity-reducedconditionshouldex-pressincreasedpositivity.Emotionalexpressionwasmodeled,onaper-personbasis,asthepercentageofwordsproducedbythatpersonduringtheexperimentalperiodthatwereeitherpositiveornegative.Positivityandnegativitywereevaluatedseparatelygivenevidencethattheyarenotsimplyoppositeendsofthesamespectrum(8,10).Indeed,negativeandpositivewordusescarcelycorrelated[r=??0.04,t(620,587)=??38.01,P<0.001].Weexaminedthesedatabycomparingeachemotionconditiontoitscontrol.Afterestablishingthatourexperimentalgroupsdidnotdifferinemotionalexpressionduringtheweekbeforetheexperiment(allt0.13),weexaminedoverallpostingrateviaaPoissonregression,usingthepercentofpostsomittedasaregressionweight.Omittingemotionalcontentreducedtheamountofwordsthepersonsubsequentlyproduced,bothwhenpositivitywasreduced(z=??4.78,P<0.001)andwhennegativitywasreduced(z=??7.219,P<0.001).Thiseffectoccurredbothwhennegativewordswereomitted(99.7%asmanywordswereproduced)andwhenpositivewordswereomitted(96.7%).Aninteractionwasalsoobserved,showingthattheeffectwasstrongerwhenpositivewordswereomitted(z=??77.9,P<0.001).Assuch,directexaminationofthefrequencyofpositiveandnegativewordswouldbeinappropriate:Itwouldbeconfoundedwiththechangeinoverallwordsproduced.Totestourhypothesisregardingemotionalcontagion,weconductedweightedlinearregressions,predictingthepercentageofwordsthatwerepositiveornegativefromadummycodeforcondition(experimentalver-suscontrol),weightedbythelikelihoodofthatpersonhavinganemotionalpostomittedfromtheirNewsFeedonagivenviewing,suchthatpeoplewhohadmorecontentomittedweregivenhigherweightintheregression.WhenpositivepostswerereducedintheNewsFeed,thepercentageofpositivewordsinpeopleÕsstatusupdatesdecreasedbyB=??0.1%comparedwithcontrol[t(310,044)=??5.63,P<0.001,CohenÕsd=0.02],whereasthepercentageofwordsthatwerenegativeincreasedbyB=0.04%(t=2.71,P=0.007,d=0.001).Conversely,whennegativepostswerereduced,thepercentofwordsthatwerenegativedecreasedbyB=??0.07%[t(310,541)=??5.51,P<0.001,d=0.02]andthepercentageofwordsthatwerepositive,conversely,increasedbyB=0.06%(t=2.19,P<0.003,d=0.008).Theresultsshowemotionalcontagion.AsFig.1illustrates,forpeoplewhohadpositivecontentreducedintheirNewsFeed,alargerpercentageofwordsinpeopleÕsstatusupdateswerenegativeandasmallerpercentagewerepositive.Whennegativitywasreduced,theoppositepatternoccurred.Theseresultssug-gestthattheemotionsexpressedbyfriends,viaonlinesocialnetworks,influenceourownmoods,constituting,toourknowl-edge,thefirstexperimentalevidenceformassive-scaleemotionalcontagionviasocialnetworks(3,7,8),andprovidingsupportforpreviouslycontestedclaimsthatemotionsspreadviacontagionthroughanetwork.Theseresultshighlightseveralfeaturesofemotionalconta-gion.First,becauseNewsFeedcontentisnotÒdirectedÓtowardanyone,contagioncouldnotbejusttheresultofsomespecificinteractionwithahappyorsadpartner.Althoughpriorresearchexaminedwhetheranemotioncanbecontractedviaadirectinteraction(1,7),weshowthatsimplyfailingtoÒoverhearÓafriendÕsemotionalexpressionviaFacebookisenoughtobuffer??1.505.05.15.25.35.4??1.80??1.70??1.60Positive Words (per cent)Negative Words...