Psychology and morality of political extremists - EPJ Data ...

We explain our definition of political extremism, data collection procedure, plausible confounding variables and how to adjust for them, ... Skiptomaincontent Advertisement SearchallSpringerOpenarticles Search Psychologyandmoralityofpoliticalextremists:evidencefromTwitterlanguageanalysisofalt-rightandAntifa DownloadPDF DownloadPDF Regulararticle OpenAccess Published:14May2019 Psychologyandmoralityofpoliticalextremists:evidencefromTwitterlanguageanalysisofalt-rightandAntifa MeysamAlizadeh  ORCID:orcid.org/0000-0001-6696-64711,IngmarWeber2,ClaudioCioffi-Revilla3,SantoFortunato4&MichaelMacy5  EPJDataScience volume 8,Article number: 17(2019) Citethisarticle 19kAccesses 19Citations 47Altmetric Metricsdetails AbstractTherecentriseofthepoliticalextremisminWesterncountrieshasspurredrenewedinterestinthepsychologicalandmoralappealofpoliticalextremism.Empiricalsupportforthepsychologicalexplanationusingsurveyshasbeenlimitedbylackofaccesstoextremistgroups,whilefieldstudieshavemissedpsychologicalmeasuresandfailedtocompareextremistswithcontrastgroups.WerevisitthedebateoverthepsychologicalandmoralappealofextremismintheU.S.contextbyanalyzingTwitterdataof10,000politicalextremistsandcomparingtheirtext-basedpsychologicalconstructswiththoseof5000liberaland5000conservativeusers.Theresultsrevealthatextremistsshowalowerpositiveemotionandahighernegativeemotionthanpartisanusers,buttheirdifferencesincertaintyisnotsignificant.Inaddition,whileleft-wingextremistsexpressmorelanguageindicativeofanxietythanliberals,right-wingextremistsexpressloweranxietythanconservatives.Moreover,ourresultsmostlylendsupporttoMoralFoundationsTheoryforpartisanusersandextendittothepoliticalextremists.Withtheexceptionofingrouployalty,wefoundevidencessupportingtheMoralFoundationsTheoryamongleft-andright-wingextremists.However,wefoundnoevidenceforelevatedmoralfoundationsamongpoliticalextremists. IntroductionSincelate2016,severalhateandviolentrallieshavebeenheldinU.S.,U.K.,Poland,Germany,Canada,andRussia,JewishandAfrican-AmericaninstitutionsandMosquesacrosstheU.S.havebeenthreatenedwitharmedprotestsorattackedbyloneactors,andimmigrantshavebeentargetedinsuspectedhatecrimesintheUnitedStatesandothercountries.Manyoftheseincidentswerelinkedtofar-rightandalt-rightsupporters,whichincludebutnotlimitedtoNeo-FascistMovementinItaly(CastelliGattinaraetal.[18]),NationalActioninBritain(Macklin[63]),right-wingmilitantsinRussia(Enstad[32]),HiveterrorismandrefugeecrisisinGermany(Koehler[56]),DeathSquadandtheHungarianArrowsNationalLiberationArmyinHungary(Mareš[65]),andRacialist,Anti-Federalist,andChristianFundamentalistideologiesintheU.S.(SweenyandPerliger[90],Windischetal.[101]).Ontheotherhand,during2016and2017,far-leftmovementsintheU.S.suchasAntifawereactivelyengaginginviolentactionsattackingalt-rightdemonstrators,protestagainstalt-rightspeakersinuniversities,andthreattodisruptfar-rightaffiliatedparades(Beinart[13]).Ahigh-profileexampleofsuchincidentswasthecaseofCharlottseville,VirginiainAugust2017,atwhichmembersofAntifaclashedwithalt-rightsupportersusingsomeviolentmethods(LaFree[62]).WhiletheantifascistmovementsseemedtobedisappearedwiththeendofWWII,theyareonriseintheUnitedStatesandEurope,inpartduetothegrowthofneo-Nazism(LaFree[62],Arlow[10]).Thisrecentriseofhateandviolencedemandsbetterunderstandingofthecausesofpoliticalextremismanditsactorsandhascaughttheattentionofresearchersacrossvariousdomains(e.g.Bonikowski[15],Jaskoetal.[48],Kurzmanetal.[61],Kruglanskietal.[59],Webberetal.[99]),includingpoliticalpsychology.Doestheappealofextremistgroupsreflectpsychologicalormoraldifferences?Inotherwords,dopoliticalextremistspossessdistinguishedpsychologicalormoralprofilescomparedtonon-extremists?Thisquestiontapsintocompetingpsychologicalexplanationsforinvolvementinextremistpoliticalgroups.Inpsychology,morethan60yearsofresearchhasfocusedonpredictorsofleft-versusright-wingideology(e.g.Adornoetal.[1],Altemeyer[8],Sidanius[83],Rokeach[79]).AstheauthorsofTheAuthoritarianPersonalityputit:“Ideologieshavefordifferentindividuals,differentdegreesofappeal,amatterthatdependsupontheindividual’sneedsandthedegreetowhichtheseneedsarebeingsatisfiedorfrustrated”(Adornoetal.[1]).Inotherwords,politicalpsychologistsbelievethatourbeliefsservepsychologicalfunctionsandtherearenaturalcorrespondencesbetweenourpsychologicalstatesandpoliticalorientation(Jost[49]).Althoughtheliteratureonpsychologicalandmoralcovariatesofpoliticalattitudesandorientationisrich,theextenttowhichitsfindingscanbegeneralizedtobothextremesofthepoliticalspectrum(leftandright)isquestionable(GreenbergandJonas[39],Jostetal.[53]).Indeed,individualpsychologicalandmoralcorrelatesofnon-violentleft-andright-wingpoliticalextremistshavenotbeenwell-researched(Prince[76]).Oneobviousreasonforthisshortageofresearchissimplythefactthatextremistswouldnotvolunteertoparticipateinexperimentalorfieldstudies.Researchonthesusceptibilitytopoliticalextremismhasreliedonsurveysandethnographicstudies.Surveyshavebeenadministeredtorandomsamplesaswellasconveniencesamples(e.g.collegestudents)toidentifythosewhoseresponsesfallattheextremesoftheliberal-conservativespectrum(e.g.Jostetal.[53],Kruglanskietal.[58],vanProoijenetal.[96])andwhobelievethattherealizationoffundamentalsocialandeconomicchangesrequiresmilitantactionoutsidetheelectoralprocess.Animmediateproblemisthatelicitedresponsestoaninterviewerarenotequivalenttovoluntaryexpressionsofsupportfor,agreementwith,andendorsementofextremistgroupsandactivities,includingthosethatarenon-violent.Otherstudieshaveusedfieldobservationofextremistgroups(Canetti-Nisimetal.[17],AtranandGinges[11]),buttheseworkslackpsychologicalmeasures,failtocomparepoliticalextremistswithpropercontrolgroups,andsufferfromselectionbias.Computationalstudiesalsosufferfromlackofappropriatedataforvalidation(e.g.Alizadehetal.[4,6],FlacheandMacy[35],AlizadehandCioffi-Revilla[2,3],RavandiandMili[77]).Widespreaduseofsocialmediabyextremistorganizationsandtheirfollowersprovidesresearcherswithunprecedentedopportunitiestostudytheprofilesofthosewhoaresusceptibletoextremistappeals(e.g.Magdyetal.[64],RoweandSaif[81],Davidsonetal.[24],Ferrara[33]).Twittermessageshavebeenshowntorevealunderlyingpsychologicalattributesthatarereflectedinwordusages,amethodwhichgoesbackto1950s(e.g.theHarvardGeneralInquirer)andrecentlymasteredbyJamesPennebaker(Pennebakeretal.[70]).Inthispaper,weintroduceanoveldatasettostudythepsychologicalandmoralprofilesofpoliticalextremistsintheUnitedStates.WeanalyzeTwittermessageswrittenbyover500,000Americanfollowersofnon-violentU.S.extremistindividuals/organizationsandidentifiedthetruesupportersofextremeideologies,thensampled10,000Twitterusersfromtheidentifiedextremistspool.Wecomparethetext-basedindicatorsofpsychologicalandmoralvariableswithresultsforfollowersofthefivemostliberalandthefivemostconservativeU.S.Senatorsaccordingtotheir2018DW-NOMINATEscores(PooleandRosenthal[74]).Wecontrolforothercontributingfactorsincludingthefollowerscount,friendscount,actioncount,actionfrequency,andtenurebyusinginversepropensityscoreweighting.InSect. 2,wereviewvariouspsychologicaltheoriesassociatedwiththeroleofcertainty,anxiety,happiness,andmoralfoundationsonpoliticalorientation,andwhetherornottheyaregeneralizabletothepoliticalextremismcontext.Wediscussopposingtheoriesanddevelophypothesestobetestedwithournewdataset.Weexplainourdefinitionofpoliticalextremism,datacollectionprocedure,plausibleconfoundingvariablesandhowtoadjustforthem,andmethodstoinferpsychologicalandmoralvariablesfromusers’tweetsinSect. 3.WepresentourresultsinSect. 4andconcludethepaperinSect. 5.TheoreticalandempiricalbackgroundPsychologicalconstructsandpoliticalextremismThegeneralformofhypothesisthatweexplorehereiswhetherornotthereexistsomepsychologicalandmoralvariablesthatcovarywithpoliticalorientationandextremity.Specificvariablesthathavebeenhypothesizedorempiricallyshowntocorrelatewithpoliticalideologyincludeambiguityintolerance(Frenkel-Brunswik[36],FibertandRessler[34]),uncertaintyavoidance(SorrentinoandRoney[88]),fearandaggression(Adornoetal.[1],Altemeyer[9]),negativeaffect(Tomkins[94]),needforcognitiveclosure(KruglanskiandWebster[60],Kruglanksi[57]),needfororderandstructure(Altemeyer[9],SmithandGordon[87]),integrativecomplexity(Sidanius[83,84],Tetlock[93]),anxiety(Greenbergetal.[41]),group-baseddominance(Prattoetal.[75]),systemjustificationtendencies(JostandBanaji[50]),self-esteem(Jostetal.[51])andmoralfoundations(HaidtandJoseph[45],HaidtandGraham[44]).However,onlyfewoftheseresearchexplicitlyinvestigatedtheextenttowhichtheirfindingsaredrivenbyorgeneralizabletopoliticalextremists.Ontheotherhand,notallofthesevariablescanbemeasuredbasedonsocialmediadata.Therefore,inthispaper,weonlyfocusonthosevariablesthathavebeenhypothesizedtocorrelatewithpoliticalextremismandaremeasurablethroughsocialmediadata.Thisincludescertainty,anxiety,positiveandnegativeemotions,andthefivemoralconstructsdiscussedintheMoralFoundationsTheory(HaidtandJoseph[45]).Itshouldbenotedthatwedonotyetknowtherelationshipbetweenthetext-basedindicatorsofpsychologicalandmoralvariablesandtraditionalwaysofassessingtheminpsychology,nordoweproposeourmethodasareplacementforthem.Rather,wewanttohighlightthepotentialofonlinedataasacomplementtoexistingmethodsofstudyingpsychologicalandmoralprofilesofpoliticalextremists.CertaintyGenerally,twoopposingviewshavebeenproposedforhowcertaintycovarieswithpoliticalextremism.Themainstreamandwell-studiedviewisthatofnegativerelationshipbetweenpoliticalextremismandcertainty.Theyarguethatnaturallyextremisminvolvesasortofdeviancyfromwhatmostpeopletendtobelieveinorconsiderreasonable.Theextenttowhichpeopledeviatefromthenormconstructsaspectrumwhereinsomebeliefsorbehaviorsmayconsidermoreextremethantheothers.Therelationshipbetweenpoliticalextremismandcertaintyrisesfromthecostanddifficultyassociatedwithmaintainingthedeviantpositioninasociety.Infact,beinganonconformistrequireseffortand“peoplecansavethemselvesconsiderableeffortandenergybygoingalongwiththecrowd”(Alquistetal.[7,p. 81]).Whetherthenormisestablishedthroughcentraltendencyofpeopleoranauthorityfigure,deviatingfromitischallengingandrequiressubstantialenergeticresources(Kruglanskietal.[58],Kruglanskietal.[59]).Becauseofthatdifficulty,extremiststendtoprotecttheirnonconformityfrommajoritypressuresbyholdingthemwithconsiderablecertainty(Webberetal.[99]).Inasimilardiscussion,GreenbergandJonas[39]proposedthat,whetherleftorright,needstoavoiduncertaintywouldbehigherforextremiststhanthoseatthecenter.H1a:: PoliticalExtremistsscorehigherontext-basedindicatorsofcertaintycomparedtotheirnon-extremistcounterparts. Ontheotherhand,Sidanius[84]observesthatholdingextremeattitudesrequiressomedegreeofsophisticationandcomplexity.Accordingtothisview,extremistscanmanagethepressureassociatedwiththeirdevianceandmayhavehigheruncertaintytolerance.Empiricalevidencebasedonquestionnairestudyofonehundredandeightundergraduatestudentsrevealsthatuncertaintytolerance(ratherthanavoidance)isassociatedwithideologicalextremity,especiallyleft-wingextremism(Jostetal.[53]).Therefore,assumingthathigheruncertaintytolerancemeansorisassociatedwithlesscertainty,wecanhypothesizethat:H1b:: Politicalextremistsscorelowerontext-basedindicatorsofcertaintycomparedtotheirnon-extremistcounterparts. H1c:: Left-wingextremistsscorelowerontext-basedindicatorsofcertaintycomparedtoright-wingextremists. AnxietySimilartocertainty,researchontherelationshipbetweenpoliticalextremismandanxietyormentaldistressisinconclusive.Theextantliteratureidentifiestwoperspectives.Buildingontheempiricalfindingsthatconservativesscorehigherthanliberalsonthe“perceptionofdangerousworld”scale(Altemeyer[9],Duckitt[30]),firstgroupoftheoriespostulatethatright-wingextremistssufferfromgreateranxietythanothers.Jostetal.[51]foundthatfearofthreatandlossanddeathanxietyaresignificantpredictorsofpoliticalconservatism.WhileCrowsonetal.[23]andGreenbergandJonas[40]arguethatanxietyisassociatedonlywithextremeattitudes,Jostetal.[53]showedthatitisassociatedwithpoliticalconservatisminparticularandnotideologicalextremityingeneral.Morerecently,RoccattoandRusso[78]foundapositivecorrelationbetweenanxietyandright-wingextremismundersocietalthreattosafetycondition.H2a:: Right-wingextremistsscorehigherontext-basedindicatorsofanxietythanothers. Asecondgroupofresearchers,drawingonTerrorManagementTheory(Greenbergetal.[42]),SocialIdentityTheory(Tajfel[91]),thefrustration-aggressionhypothesis(Dollardetal.[29])andSystemJustificationTheory(JostandHunyady[52]),arguethatideologicalextremityisabufferingfactorforpsychologicalhealth.Accordingtothisview,holdingextremeviewpointsyieldslowerlevelsofanxiety.Intwoseparatestudieswhichaskedsubjecttofilloutaquestionnaire,VanHielandDeClercq[95]foundlessadverseeffectsoffacilitatorsofmentaldistressamongthoseparticipantswhoscorehigheronauthoritarianismscale.Theirresultsfurthershowedthat“authoritarianismactsasabufferagainstmentaldistress”forpeoplewithD-typepersonality.Morerecently,drawingonlongitudinalsurveydatafrom18countries,Vargas-Salfateetal.[97]concludedthatconservatismisnegativelyandsignificantlyrelatedtoanxietyanddepression.Furthermore,itisarguedthat,unliketheirmoderatecounterparts,supportersofthefarrightshowauthoritarianinclination(Bobbio[14,pp. 72–79]).Thus,wecanhypothesizethat:H2b:: Right-wingextremistsscorelowerontext-basedindicatorsofanxietythanothers. HappinessThereisalongstandingdebateabouttherelationshipbetweenconservatismandhappiness.ApublicopinionsurveyconductedbyPewResearchCenterin2006showedthatwhereas47%ofconservativerepublicansconsideredthemselvesas“veryhappy”,only28%ofliberaldemocratsfeltso(Tayloretal.[92]).Sincethen,manystudieshavesupportedorrejectedtheseresults.Drawingonsystem-justificationtheoryanddatafromtencountries,NapierandJost[66]foundthatright-wingideologyispositivelycorrelatedwithhappinessandlifesatisfaction,andthatthereexistsahappinessgapbetweenleft-andright-wingsupporters.Usingalongitudinalsurveyofonlinesubjectsfrom18countries,Vargas-Salfateetal.[97]showedthathavingsystemjustificationattitudeisassociatedwithhigherlifesatisfaction.Earlierworkhasidentifiedconservatismasaformofsystem-justifyingbelief(JostandHunyday[52]).However,right-wingideologyhasbeenreportedtobenegativelycorrelatedwithindicatorsofpsychologicalwell-beingsuchaslifesatisfaction,positiveaffect,orabsenceofnegativeaffect.Forexample,PetersonandDuncan[72]foundthatwomenwhoscorehighonauthoritarianscalehavelesspositiveaffectcomparedtothosewhoscorelowonthesamescale.Inanotherstudy,Duriezetal.[31]showedthatauthoritarianismisassociatedwithdevelopingdepressivesymptoms.Moreover,therearestudieswhofoundnosignificantrelationshipbetweenright-wingattitudesandpsychologicalwellbeing(e.g.Butler[16],Onraetetal.[68]).Althoughtherearemanyempiricalstudiesontherelationshipbetweenpositiveemotionandpoliticalorientation,oursearchtofindarelatedstudyonpoliticalextremitywasfruitless.Tofillthisgapandbuildingontheobservedrobustoccurrenceofalineareffect(Jostetal.[51]),weassumethattherelationshipbetweennegative/positiveemotionandpoliticalorientationismonotonouslyanduniformlylinear.Hence,ifthesefindingsaretrue,onecouldhypothesizethat:H3a:: Conservativesscorehigherontext-basedindicatorsofpositiveemotionandloweronnegativeemotioncomparedtoliberals. H3b:: Extremistsscorehigherontext-basedindicatorsofpositiveemotionandloweronnegativeemotioncomparedtonon-extremists. H3c:: Right-wingextremistsscorehigherontext-basedindicatorsofpositiveemotionandloweronnegativeemotioncomparedtoleft-wingextremists. Ontheotherhand,accordingtothesecondsetofstudies,wehypothesizethat:H3d:: Liberalsscorehigherontext-basedindicatorsofpositiveemotionandloweronnegativeemotioncomparedtoconservatives. H3e:: Extremistsscorelowerontext-basedindicatorsofpositiveemotionandhigheronnegativeemotioncomparedtoconservatives. H3f:: Right-wingextremistsscorelowerontext-basedindicatorsofpositiveemotionandhigheronnegativeemotioncomparedtoleft-wingextremists. MoralfoundationsandpoliticalextremismMoralpsychologistshavearguedthattheleftdiffersfromtherightintheemotionalresonancewithdifferent“moralfoundations”(HaidtandJoseph[45],HaidtandGraham[44],Haidt[43]).TheMoralFoundationsTheory(MFT)proposesthatliberalsendorsevaluesoffairness/reciprocity(e.g.thenumberoneprincipleinmakinglawsshouldbeensuringfairtreatment)andavoidanceofharm(e.g.compassionforthosewhoaresuffering)moreenthusiasticallythanconservatives,whereasconservativesendorsevaluesofingrouployalty(e.g.loyaltytoingroupismoreimportantthanindividualconcerns),obediencetoauthority(e.g.lawmakersshouldrespecttraditions),andenforcementsofpurity,divinity,andsanctitystandards(e.g.thegovernmentshouldhelppeoplelivevirtuously)moreenthusiasticallythanliberals.Grahametal.[38]developedmethodstomeasureeachofthesefivemoralfoundationsandempiricallysupportedtheabovehypothesesaboutmoraldifferencesbetweenliberalsandconservatives.Theobservedpatternhasbeenreplicatedmanytimes(e.g.Daviesetal.[25],Dayetal.[27],NilssonandErlandsson[67],Yilmazetal.[102]).However,theextenttowhichMFTcanbegeneralizedtopoliticalextremistsisyettobetestedwithempiricaldata.Therefore,ourfirstsetofhypothesesabouttherelationshipbetweenpoliticalextremismandMFTcanbestatedasfollowings:H4a:: Left-wingextremistsscorehigherontext-basedindicatorsoffairness/reciprocitythanright-wingextremists. H4b:: Left-wingextremistsscorehigherontext-basedindicatorsofharmavoidance/carethanright-wingextremists. H4c:: Right-wingextremistsscorehigherontext-basedindicatorsofingrouployaltythanleft-wingextremists. H4d:: Right-wingextremistsscorehigherontext-basedindicatorsofobediencetoauthoritythanleft-wingextremists. H4e:: Right-wingextremistsscorehigherontext-basedindicatorsofpuritythanleft-wingextremists. Furthermore,inordertolinkMFTtopoliticalextremism,GrahamandHaidt[37]arguethatsincesuccessfulpoliticalmovementsmusthaveastorythatexplainsthecurrentproblemsandtheirsolutions(Westen[100]),MFT“providesthemostcomprehensibleaccountofthehooksinthemoralmindtowhichagoodideologicalnarrativecanattach”.Whilepreviousresearchtendedtodefineextremismintermsofonlyharmavoidance/caremoralfoundation,GrahamandHaidt[37]suggestthatitcanbedefinedbasedonothermoralfoundationsaswell.Theyfurtherarguethatthe“elevation”or“sacralization”ofamoralfoundationisamajorcauseofextremism.Forexample,incaseofwhitesupremacists,thenarrativeprovidedinTheTurnerDiaries(Pierce[73])emphasizesoningroupandpurityfoundations.Thewhiteraceanditspurebloodaresacralized,andself-sacrificeandloyaltyforitsprotectionandsurvivalarepaintedasmoralideals.OrincaseoftheWeatherUnderground,whichwasaleft-wingextremistgroupin1970s,theirprimaryareaofsacralizationwasblackvictimsinwhiteAmerica,whichledthemtopossesssacralizedharmandfairnessmoralfoundations.Hence,oursecondsetofhypothesesregardingtheassociationbetweenMFTandpoliticalextremismcanbestatedasfollowings:H4f:: Left-wingextremistsscorehigherthanliberalsontext-basedindicatorsoffairness. H4g:: Left-wingextremistsscorehigherthanliberalsontext-basedindicatorsofharmavoidance. H4h:: Right-wingextremistsscorehigherthanconservativesontext-basedindicatorsofingrouployalty. H4i:: Right-wingextremistsscorehigherthanconservativesontext-basedindicatorsofobediencetoauthority. H4j:: Right-wingextremistsscorehigherthanconservativesontext-basedindicatorsofpurity. DataandmethodsIdentifyingpoliticalextremistgroupsinU.S.andtheirfollowersDeterminingwhichpoliticalideologyisextremeandwhichisnotisaverychallengingandcontext-dependenttaskandusuallyinvolvescontroversies.Anysortofdefinitionleavesconsiderableroomforinterpretation.Inthispaper,welimitourscopetowhitesupremacyandneo-Naziideologiesasright-wingextremist(RWE)andAntifaasleft-wingextremist(LWE)ideologyinthecurrentU.S.politicalarena.Weidentified25whitesupremacyandneo-NazigroupsthathaveactiveTwitteraccountsbyconsultingtheSouthernPovertyLawCenterwebsite(SPLC2018)andlisttheirnamesandTwitterhandlesinTable A1intheAppendixsection.ForAntifagroups,wereliedonmanualsearchonTwittertoidentifypopularofficialandlocalchaptersofthemovementandcameupwithalist16verifiedAntifaaccounts.Theverificationwasperformedthroughcross-checkingourlistwiththoselistedatblocktogether.org,acrowdsourcingwebapplicationintendedtosharealistoffakeAntifaaccounts.ToobtainmorevalidatedAntifaaccountssothatwehaveequalnumberofleft-andright-wingextremistseedaccounts,wecollected4527friends(i.e.thosewhoarebeingfollowed)and5,639,256friendsoffriendsofourinitial16Antifaaccountsandbuilttheirfriendshipnetwork.Weperformk-coredecompositiononthefriendshipnetworktoobtainthemaincoreofit.Thek-coreofagraphisformallydefinedasthemaximalsubgraphwithnodesofatleastdegreek.Themaincoreisthenon-emptygraphwithmaximumvalueofkandcanbeusedtoidentifythemostinfluentialnodesofagivennetwork(e.g.Kitsaketal.[55]).Previousresearchalsousedk-coredecompositiontocharacterizetheefficiencyofthespreadofinformation(Conoveretal.[21])ordisinformation(Shaoetal.[82]).Fromtheobtainedmaincore,wetake9userswhohavethehighestdegreesandmanuallychecktheirTwitterpagestomakesuretheyareassociatedtoAntifa.Themanualprocessincludeslookingforeithermentioningof“Antifa”inthename,Twitterhandle,orbiodescriptionofthegroup,orhighvolumesharingofcontentpostedbyotherknownAntifapages(morethanhalfoftherecent50tweets),plusthenumberoffollowers(onlyincludethosewithmorethan5000followers).Finally,weaddtheseobtainedinfluentialuserstoourlistof16LWEextremistaccountstoformourfinallistof25AntifaaccountsandreporttheirnamesandTwitterhandlesinTable A2intheAppendixsection.WecallthesetwosetsofLWEandRWEextremistindividual/organizationlistsas“seedaccounts”.Wegetthefollowersoftheseseedaccounts,andfromthoseuserswhopassedthepreprocessingstep(seeSect. 3.3),weconsiderafollowerasasupporterorsympathizerofanextremepoliticalideologyifs/hefollowsatleastthreeofthecorrespondingaccountsfromourseedaccounts.ControlgroupsTotestourhypotheses,wecomparethepsychologicalprofileofpoliticalextremistswithqualifiedfollowersofthetopfivemostliberalandconservativeU.S.Senatorsaccordingtotheir2018DW-NOMINATEscore(PooleandRosenthal[74]).Table 1demonstratesthelistofSenatornamesineachpoliticalcategory.WeperformallpreprocessingstepsmentionedinSect. 3.3onfollowersoftheseSenators.Inaddition,weexcludemutualfollowersbetweeneachpairofpoliticalideologies.Finally,weonlyconsideruserswhofollowatleastthreeSenatorsfromeachcategory.Table 1ListofthefivemostliberalandconservativeU.S.senatorsaccordingtotheir2018DW-NOMINATEscoresFullsizetablePreprocessingoffollowersWecrawledallfollowersoftheseedaccounts(Table 2).Incaseofliberalsorleft-wing(LW)andconservativesorright-wing(RW)followers,aftercollectingthetotalnumberofuniquefollowers(Table 2),weuniformlysample10,000followersatrandomfromeachoftheseedaccounts(i.e.50,000LWand50,000RWintotal)andperformtherestofthepreprocessingandanalysisonthesesamples.WeonlykeepthoseuserIDsthattheirlanguageisEnglish,arefromtheU.S.,andarenot“verified”.Weimposethe“notverified”constrainttoexcludepotentialjournalists,newsanchors,andcelebrities.Wefurtherexcludeuserswhomentioned“journalist”or“\(\mbox{RT}\neq\mbox{Endorsement}\)”intheirbio.Wealsoexcludemutualfollowersbetweenpoliticalgroups(i.e.thosewhofollowatleastoneseedaccountfromatleasttwopoliticalgroups).Afterthesepreprocessingsteps,weusebotometer(Davisetal.[26],Varoletal.[98])toidentifybots.Usingmorethanathousandfeaturesincludingfriends,tweetcontent,tweetsentiment,networkproperties,andtemporalpatterns,botometerprovidestwoscoresbetweenzeroandoneforEnglishspeakingusersanduniversalusers,wherezeroindicatesthehighestclassifierconfidenceforahuman,andoneforabot.Anyscorein-betweenmeansthattheclassifierisnotcertainabouttheaccountandwehavetomakeadecision.Weusethe0.7thresholdfortheEnglishspeakerscoreandexcludealluserswithscoresbeyondthatfromourdata.Next,tomakesurethatthefollowersoftheseedaccountsarereallyaffiliatedwiththepoliticalgroups,followingBarberá[12],weonlyconsiderthosefollowerswhofollowatleastthreeoftheseedaccountsfromeachpoliticalgroup.Table 2DatasummaryFullsizetableFinally,tocontrolforanypotentialdifferencesbetweenfollowersofthoseseedaccountsthatbelongtogroupsandthosethatbelongtoindividuals,weremovedallfollowerswhowereonlyfollowingindividualRWEaccounts.IncaseofLWEseedaccounts,thereisonlyoneindividualaccount.Thisresultsin9400LW,12,034RW,7665LWE,and10,983RWEaffiliatedTwitterusers(Table 2).Wecollectedupto3200tweetsoftheseusersandanalyzetheirtimestamps.Welimitouranalysistothosetweetswhichhavepostedinapriorthreemonthsofthedateofourdatacollection,whichwasonMarch15,2018.WeexcludethoseuserswhohavenotpostedinthistimeperiodorthosewhoseoldestavailabletweetinourdatasetisafterDecember15,2017andcalltheremainingusersas“qualifiedusers”inTable 2.Wemakethelatterrestrictiontomakesurethattheusers’tweetsarerepresentativeoftheirtemporalchangesoverthecourseofthreemonths.Finally,weuniformlytake5000usersatrandomfromthequalifiedusersandestimatetheirtext-basedpsychologicalandmoralvariables(seeSects. 3.5and3.6onhowtoestimatethepsychologicalandmoralindicatorsfromtweets).ThesummaryoftheusersateachstepofthedatacollectionprocessislistedinTable 2.PreprocessingoftweetsBeforeweproccedtoestimatingthepsychologicalandmoralprofileofextremistsandnon-extremists,weneedtoperformsomepreprocessingonourtextdata.First,weconvertalltweettextstolowercaseandremoveallURLs,usermentions,andpunctuationsfromthetext.Wefurtherremoveretweetsfromourcorpusofdatasinceretweetsarenottheoriginalpostsoftheauthorsandshouldnotbeconsideredasemotionalexpressionsoftheusers.Tocontrolfortemporalvariations,weonlyconsiderthosetweetswhichhavebeenpostedwithinapriorthreemonths.InferringandvalidatingpsychologicalindicatorsArichbodyofresearchhasshowntherelationshipbetweenlinguisticusageandemotion(Pennebakeretal.[71]).Weusewell-validatedLinguisticInquiryandWordCount(LIWC)lexicontomeasurethesetofpsychologicalvariablesmentionedinH1–H3.LIWCusesfrequencypercentagestogaugeindividuals’preferencesregardingspecific“function”wordsaswellas“content”wordsthatarechosentoconveysemanticinformation.WeconstructedpsychologicallanguageprofilesusingtheLIWC2015lexicon(Pennebakeretal.[69]).Tomeasurecertainty(H1),weconsiderthecertaintywordslistfromthecognitiveprocessescategory.Examplesofthecertaintywordsandword-phrasesinclude“always”,“never”,and“certain”.Toquantifytheanxiety(H2)andhappiness(H3),weusetheanxiety,positiveemotion,andnegativeemotionwordslistsfromtheaffectiveprocessescategory.Examplesoftheaffectiveprocesseswordsandword-phrasesinclude“love”,“nice”,and“sweet”forpositiveemotion,“hurt”,“ugly”,and“nasty”fornegativeemotion,and“worried”and“fearful”foranxiety.Weusetheseaffectiveandcognitiveprocesseswordliststocountthenumberofusagesacrossallthetweetsinauser’sTwitter“timeline”expressedasaproportionoftheuser’stotalwordcount.ForareviewofusingTwitterdatainhealthandwell-beingresearchseeSinnenbergetal.[85].AlthoughLIWChavebeenvalidatedandusedinmanycontexts,tothebestofourknowledge,ithasneverbeenusedontextoriginatedfrompoliticalextremists.Therefore,thequestionremainswhetherornotitcanproperlyestimatethetext-basedpsychologicalprofileofpoliticalextremists.Therefore,beforeusingLIWC,weneedtovalidateitsperformanceonextremists’tweets.Toaccomplishthevalidationtask,weuniformlysample100tweetsatrandomfromthecorpusofextremists-generatedtweets.Then,thefirstauthorevaluatedtheextenttowhicheachofthesampledtweetscommunicatedeachofthefourpsychologicalconstructs(i.e.anxiety,certainty,negativeandpositiveemotion)usinga7-pointLikert-typescale.Next,werunLIWContweetsandcomputetheratioofhitsassociatedwitheachofthefourpsychologicalmeasures.Finally,wecomputethePearsoncorrelationcoefficientbetweenthehand-codedandLIWC-generatedscoresandreporttheresultsinTable 3.Ascanbeseen,therearesignificantandstrongpositivecorrelationsbetweenthehand-codedandLIWC-generatedscoresacrossallfourpsychologicalconstructs,whichindicatesthattheLIWCdictionarywordsandtermsforanxiety,certainty,positiveemotion,andnegativeemotionaresufficientlyrobusttodetectthecorrespondingpsychologicalconstructsintweetspublishedbyAmericanleft-andright-wingpoliticalextremistsidentifiedinthisstudy.Table 3CorrelationstatisticsbetweenLIWC-generatedandhand-codedpsychologicalscoresfor100politicalextremist-writtentweetsFullsizetableInferringmoralfoundationsGrahametal.[38]developedMoralFoundationsDictionary(MFD),whichcontainswordlistsassociatedwitheachofthefivemoralfoundationsintroducedintheMFT.Examplesofthewordsinclude“safe”,“peace”,and“endanger”forharmavoidanceandcare,“fair”,“equal”and“disproportion”forfairnessandreciprocity,“together”,“nation”,and“traitor”forin-groupandloyalty,“obey”,“law”,“tradition”,and“illegal”forauthorityandrespect,and“piety”,“innocent”,and“trashy”forpurityandsanctity.Grahametal.[38]appliedtheMFDonsermonsintextformandtheresultswereconsistentwithMFT.UsingMFDtoanalyze12yearsofnewscontentrelatedtostemcellresearch,CliffordandJerit[20]foundconsistentresultswithMFTwithrespecttoharmavoidanceandpuritymoralfoundations.Theyfurthershowedthatwordlistsrelatedtotheotherthreefoundationsrarelyappearedintheirdatasetthroughcontentanalysisofasmallnumberofrandomlyselectedarticles.SimilartoLIWCdictionary,wecouldnotfindanypreviousstudythathasvalidatedtheapplicationofMFDinthepoliticalextremismcontext.Therefore,wetakethesameprocedureasdescribedaboveintheSect. 3.5andreportthevaliditystatisticsinTable 4.Theresultsshowsignificantandstrongcorrelationsbetweenhand-codedandMFD-generatedscoresacrossallfivemoralfoundations.Table 4CorrelationstatisticsbetweenMFD-generatedandhand-codedmoralscoresfor100politicalextremist-writtentweetsFullsizetableConfoundingcovariatesTherearemanyvariablesthatmightcontributetothetext-basedpsychologicalindicatorsofTwitterusers.Hence,withoutcontrollingforcommoncauses,ourresultswouldbeconfounded.InthecaseoflanguageanalysisofTwitterusersthroughword-countapproach,ananalystshouldselectvariablesthatmightaffectthedistributionofthewordsamongindividuals.Table 5listsasetofcovariateswhichwemeasuredforuseascovariatesofpsychologicallanguageofdifferentgroups.Forexample,onecouldhypothesizethatuserswhopublishedmoretweetsaremorelikelytogetmatchedwithLIWCandMFDdictionaries,andthus,gethigherscores.Tobetterdemonstratethecovariateimbalanceacrossthegroups,thedistributionofthecovariateslistedinTable 5(exceptfortopic)areplottedinFig. A1intheAppendixsection.Table 5ListofcovariatesFullsizetableOneimportantlatentconfoundingvariablethatcouldimposebiasonourresultsistopicsofthetweets.Thatis,sincedifferentpoliticalgroupsmightdisproportionallytalkmore/lessaboutcertaintopicscomparedtotheothers,somewordsaremore/lesslikelytobeusedbymembersofthoseparticularpoliticalgroups.Ifthisisthecase,andthosefrequentlyusedwordsareassociatedwithsomeoftheLIWCorMFDcategories,thatwouldcastdoubtsonourresults,becausethepotentialobservedpsychologicaldifferencesbetweenpoliticalgroupsarethendriveninpartbythosehighlytopic-relatedwords,notthepoliticalideologyorextremityoftheusersinthosepoliticalgroups.Forexample,itwouldbehardtotalkaboutguncontrolissueswithoutusingcertaintermsthatmightbefoundinLIWCandMFDdictionaries,includingtermssuchas“control”,“own”,and“power”.Therefore,weshouldcontrolforthesetopicsbeforecomparingtext-basedpsychological/moralprofilesofdifferentpoliticalusers.Whenwecontrolfortopics,infactweareconditioningouttheaveragelevelofpsychological/moralconstructsinthosetopics.LatentDirichletAllocation(LDA)isapopularmethodfortopicmodelingontextdata.However,standardLDAwouldnotworkwellfortweetsbecausetheyareshort,andasingletweetusuallytalksaboutonlyonetopic.Therefore,unlikeLDAwhichyieldsadistributionofvarioustopicsforadocument,weuseaTwitter-LDAmethod(Zhaoetal.[103]),whichassignseachtweettoonlyonetopic.Sinceweareinterestedincontrollingforgeneraltopics(e.g.elections,guncontrol,hatespeech,etc.),noteventsandstories,wesetthenumberoftopicsat20anditerationsat1000andusedTwitter-LDA’sdefaultsettings(Zhaoetal.[103])toestimatethetopicsoftweets.TheworddistributionofthetopicsalongwiththeirsuggestednamesarelistedinTable A3intheAppendixsection.Figure 1showsthedistributionoftopicsacrossthefourdifferentpoliticalgroups.Whilethefrequencydifferencesofsometopicsaresmallacrossthegroups(e.g.entertainment,photography,andsocialmediaactivity),therearetopicsthattheirfrequencydifferencesbetweenthefourgroupsofusersarelarge(e.g.election,sport,racial,religious,communityevents,andblacklivesmatter(BLM)andenvironment).Theseresultsfurtheremphasizetheimportanceofadjustingfortopicsinouranalysis.However,notalltopicsareeligibleorrequiredtobeadjustedfor.Weshouldcontrolforatopicifit:1.: Issemanticallymeaningful(i.e.itisnotnoisyoutcomeoftheLDA); 2.: DoesnotoverlaptoomuchwiththewordcategoriesofinterestinLIWCandMFD. Figure 1DistributionofTwitter-LDA-generatedtopicsacrossvariouspoliticalgroups.Thenumberoftopicsissetto20.Whilethefrequencyofoccurrenceofsometopicsaresimilaramongthegroups(e.g.entertainmentandphotography),therearetopicsthattheirfrequencydifferencesbetweengroupsarehigh(e.g.sportandracial).Wecontrolforatopicifitissemanticallymeaningful(i.e.itisnotanoisyoutcomeoftheLDA)anddoesnotoverlaptoomuchwiththewordcategoriesofinterestinLIWCandMFD.Asaresult,weremoveNoise1,2,&3,Feelings,andPleasuretopicsandcontrolledfortheremaining15topicsFullsizeimageAsaresult,weshouldnotcontrolfor“Noise1”,“Noise2”,and“Noise3”topicssimplybecausetheyarenotrepresentativeofsemanticallymeaningfultopics.Furthermore,weshouldnotadjustfor“Pleasure”and“Feelings”becausetheyoverlaptoomuchwiththepositiveemotioncategoryofLIWC.CovariatesadjustmentSincetheusersarenotrandomlyassignedtoeachofthefourgroups,ourobservationalstudyoftheirsocialmediaactivitieswouldsufferfromselectionbias.Therefore,weshouldidentifyconfoundingvariablesandcontrolforthemsothatwecancharacterizemeandifferencesthataremorelikelytobeaboutthelinkbetweenpoliticalorientation,politicalextremity,andtext-basedpsychologicalindicatorsofpsychologicalandmoralconstructs.Inaddition,wehaveamulti-valuedtreatmentsexperimentwithfourlevels,eachrepresentingadifferentgroupofpoliticalusers.Thus,reducingcovariateimbalancebetweenthemisnotatrivialtask,sincemostoftheexistingapproachesandtoolsaredesignedforbinarytreatments.AccordingtoRosenbaumandRubin[80],ifwehaverelevantinformationonasetofcovariatesX,andthepotentialoutcomesareindependentofthetreatment,thenwecanestimateanunbiasedestimatorusingonlythepropensityscoreandtheobservedoutcome.Thepropensityscoreistheconditionalprobabilityofbeingtreatedatsomepointincovariatesspace,\(P(T=1|X)\),whereTisthetreatmentstatus,with\(T=1\)meaningtreatedand\(T=0\)meaningnontreated.Here,thefourgrouplabels(i.e.LW,RW,LWE,andRWE)areusedasthetreatmentindicator.However,therearetwomaindifficultiesinusingpropensityscores:(1) evenaslightmisspecificationintheestimationofpropensityscorescanresultingettingabiasedestimate(e.g.SmithandTodd[86],KangandSchafer[54]);and(2) balancingcovariatesbetweenmorethantwogroupsofsubjectsisnottrivial.Totackletheseissues,ImaiandRatkovic[46]introducedCovariateBalancingPropensityScore(CBPS)methodology,whichestimatesthepropensityscoresforeachobservationwhileoptimizingthecovariatebalance.Italsogeneralizeswelltomulti-valuedtreatments.Oncethepropensityscoresarecomputed,theycanbeusedforweighting,matching,regression,stratification,oracombinationofthem(ImaiandRatkovic[46]).SeeImbens[47]andStuart[89]forextensivereviewofpropensityscoremethods.Inthispaper,weusetheinverseoftheestimatedpropensityscoresasweightstocreateabalancedsampleoftreatedandcontrolobservations.ThemethodisknownasInverseProbabilityWeighting.Animportantadvantageofusingweightingoverotherpossibleapproachesisthatwedonotloseanyofoursubjects.Let\(T_{i,j}\)beanindicatorvariabledenotingwhetheruserihasreceivedjthtreatment(i.e.whetheritbelongstoLWE,LW,RW,orRWE),and\(e_{i,j}\)denotesthepropensityscoreassociatedwiththeuserireceivingtreatment j.Then,formulti-valuedtreatments,theweightscanbeobtainedfromEq. (1): $$w_{ij}=\sum_{j=0}^{J-1}\frac{T_{i,j}}{e_{i,j}}.$$ (1) Figure 2comparesthecovariateimbalancemeasuredasdifferenceinmeansbetweenourfourtreatmentgroups(LWiscodedasgroup1,LWEas2,RWas3,andRWEas4)beforeandafterweighting.Eachpointontheplotisacovariateandeachboxplotrepresentsthemedian,minandmax,upperandlowerquartilesofthecovariatesforeachcontrast.Comparingthecovariateimbalancebefore(theupperpanel)andafter(thelowerpanel)weightinginFig. 1showsthatapplyingtheweightsobtainedfromtheCBPSmethodissignificantlyreducedthecovariatesimbalance,measuredasabsolutedifferenceofstandardizedmeans,acrossallfourtreatmentgroups.Figure 2Comparingthecovariateimbalancebeforeandafterweighting.Eachpointontheplotisacovariateandeachboxplotrepresentsthemedian,minandmax(excludingtheoutliers),upperandlowerquartilesofthecovariatesforeachcontrast.Comparingthecovariateimbalancebefore(theupperpanel)andafter(thelowerpanel)weightingshowsthattheweightingissignificantlyreducedtheimbalanceacrossthetreatmentgroupsFullsizeimageResultsWeuseAnalysisofVariance(ANOVA)totestforcorrelationofpoliticalorientationandpoliticalextremitywithcertainty,anxiety,positiveandnegativeemotions,andthefivemoralfoundationsmentionedintheMoralFoundationTheory(MFT)acrossconservatives,liberals,left-wingextremists,andright-wingextremists.ThenweuseTukey’sHSDtestforperformingposthocmultiplepairwise-comparisonbetweenmeansofthegroups.Finally,tomakesurethattheresultsarenotdrivenbysamplesize,foreachofthepsychologicalandmoralvariables,weuniformlysample25percentoftheusersfromeachgroupatrandomandperformtheTukey’sHSDtestandreporttheresultsintheAppendixsection.CertaintyInthissection,weuseANOVAandTukey’sHSDtesttoassesswhetherpoliticalextremistshavehigher/lowertext-basedindicatorsofcertaintythannon-extremists(H1aandH1b),andleft-wingextremistshavelowertext-basedindicatorsofcertaintythanright-wingextremists(H1c).Aone-wayANOVAisconductedtotestforcorrelationofpoliticalorientationandpoliticalextremitywithtext-basedindicatorsofcertaintyacrossvariouspoliticalgroupsandtheresultsarereportedinTable 6.Theresultsshowthatthereisasignificantcorrelationbetweenpoliticalextremityandtext-basedindicatorsofcertaintyatthe0.05level(\(F(1,1)=5.44\),\(p=0.02\)).However,theeffectsizeisnotsignificantwhenweuseaBonferroniadjustedalphalevelof0.0055(0.05/9).Therefore,wecannotrejectthenullhypothesisagainstH1a.Table 6ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithtext-basedindicatorsofcertaintyFullsizetableTukey’sHSDtestresultsinTable 7showthatnoneofthemeandifferencesacrossthefourgroupsaresignificant.Amongthem,wecanseethatalthoughthemeanscorefortext-basedindicatorsofcertaintyofright-wingextremistsishigherthanleft-wingextremists(\(\mathit{Mean}\\mathit{Difference}=0.0001\),\(\mathit{CI}=[-0.0007,0.0009]\),\(p=0.98\)),themeandifferenceisnotsignificantatthe0.0055Bonferroniadjustedalphalevel.Therefore,wecannotrejectthenullhypothesisagainstH1c.Theresultsfromthesmallersamplesizewhenweonlyconsider25percentofthedatashowprettymuchthesimilarpattern(Table A4intheAppendix).Theonlyexceptionisthatthemeandifferencebetweenright-wingextremistsandliberalsbecomessignificant(\(p=0.001\)).Takentogether,theresultsofthissectionsuggestthat,therearenosignificantlydifferencesinhowoftenliberals,conservatives,left-wingextremists(LWE),andright-wingextremists(RWE)uselanguageindicativeofcertaintyintheirtweets.Table 7Multi-groupmeancomparisonsofcertaintyusingTukey’sHSDtestFullsizetableAnxietyTheANOVAresultstotestourhypothesesonwhetherright-wingextremistsshowhigherlanguageindicativeofanxiety(H2a),orlowerthanothers(H2b)arereportedinTable 8.Theresultsshowthatcontrollingforpoliticalextremity,politicalorientationisareliablepredictorofthetext-basedindicatorsofanxiety(\(F(1,1)=38.41\),\(p<0.001\)),evenifweusetheBonferroniadjustedalphalevelof0.0055.However,therelationshipbetweenpoliticalextremityandlanguage-basedanxietyscoresacrossthefourpoliticalgroupsisnonsignificant(\(F(1,1)=0.81\),\(p=0.367\)).Table 8ANOVAresultstotestforcorrelationofeffectofpoliticalorientationandextremitywithtext-basedindicatorsofanxietyFullsizetableTheresultsofthepost-hocmulti-grouppairwisecomparisonsofthemeansusingtheTukey’sHSDtestarereportedinTable 9.Right-wingextremists(RWE)scorelowerontext-basedindicatorsofanxietythanleft-wingextremists(\(\mathit{Mean}\\mathit{Difference}=-0.001\),\(\mathit{CI}=[-0.0014,-0.0006]\),\(p<0.001\)),liberals(\(\mathit{Mean}\\mathit{Difference}=-0.0005\),\(\mathit{CI}=[-0.0008,-0.0001]\),\(p<0.001\)),andconservatives(\(\mathit{Mean}\\mathit{Difference}=-0.0003\),\(\mathit{CI}=[-0.0007,0]\),\(p=0.07\)).Exceptforthecomparisonwithconservatives,allofthemeandifferencesaresignificantattheBonferroniadjustedalphalevelsof0.0055.Theresultsfrom25%samplesizeareingeneralagreementwiththisfinding(Table A5intheAppendix).Theonlyexceptionisthatalthoughthemeandifferencebetweenright-wingextremistsandliberalsissignificantat0.05level(\(p=0.03\)),itisnonsignificantwhenweusetheBonferroniadjustedalphalevelof0.0055.Therefore,weshouldrejectH2a.However,wecannotrejectthenullhypothesisagainstH2b.Nevertheless,theresultssuggestthatright-wingextremistsscoresignificantlylowerontext-basedindicatorsofanxietythanliberalsandleft-wingextremists.Table 9Multi-groupmeancomparisonsofanxietyusingTukey’sHSDtestFullsizetableHappinessInthissection,wetestforhypothesesregardingmeandifferencesoftext-basedindicatorsofpositiveandnegativeemotionsinamulti-groupstudyofliberals,conservatives,left-wingextremists,andright-wingextremists.TheANOVAresults(Table 10)showthatbothpoliticalorientationandpoliticalextremityaresignificantpredictorsofthetext-basedindicatorsofpositive(PoliticalOrientation:\(F(1,1)=11.24\),\(p=0.001\);PoliticalExtremity:\(F(1,1)=73.11\),\(p<0.001\))andnegativeemotions(PoliticalOrientation:\(F(1,1)=12.08\),\(p=0.001\);PoliticalExtremity:\(F(1,1)=73.17\),\(p<0.001\)),evenifweusetheBonferroniadjustedalphalevelof0.0055.Table 10ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithpositiveandnegativeemotionsFullsizetableNowthatthepredictorsarefoundtobesignificant,weturntotheresultsofpost-hoccomparisonsusingtheTukey’sHSDtestinTable 11.Wecanseethat,whileconservativesuselowernumberofwordsindicativeofpositiveemotion(\(\mathit{Mean}\\mathit{Difference}=-0.0004\),\(\mathit{CI}=[-0.0023,0.0014]\),\(p=0.93\)),andnegativeemotion(\(\mathit{Mean}\\mathit{Difference}=-0.0003\),\(\mathit{CI}=[-0.0015,0.0008]\),\(p=0.90\))thanliberals,noneofthemeandifferencesissignificantatthe0.0055Bonferroniadjustedlevel.Theresultsfromasmallersamplesizewhereweonlyconsider25percentofthedatashowsimilarpattern(Table A6intheAppendix).Therefore,weshouldrejectbothhypothesesH3a(conservativesshowhigherpositiveemotionandlowernegativeemotionthanliberals)andH3d(liberalsshowhigherpositiveemotionandlowernegativeemotionthanconservatives).Table 11Multi-groupmeancomparisonsofpositiveandnegativeemotionsusingTukey’sHSDtestFullsizetablePost-hoccomparisonsusingtheTukey’sHSDtestindicatethat,whencomparingleft-wingextremists(LWE)andliberals(LW),LWEscorelowerontext-basedindicatorsofpositiveemotion(\(\mathit{Mean}\\mathit{Difference}=-0.0065\),\(\mathit{CI}=[-0.0084,-0.0047]\),\(p<0.001\))andhigheronnegativeemotionthanliberals(\(\mathit{Mean}\\mathit{Difference}=0.0038\),\(\mathit{CI}=[0.0026,0.005]\),\(p<0.001\)),andbothmeandifferencesaresignificantatthe0.0055Bonferroniadjustedalphalevel.Comparingright-wingextremists(RWE)andconservatives(RW)revealsthesamepattern,withtheexceptionthatthemeandifferenceinpositiveemotionisnotsignificantattheBonferroniadjustedalphalevelof0.0055(\(\mathit{Mean}\\mathit{Difference}=-0.0023\),\(\mathit{CI}=[-0.0042,-0.0004]\),\(p=0.01\)).Therefore,theresultssupportH3eandrejectH3bforleft-wingextremists.Tukey’sHSDresultsfromthe25%samplesizeanalysisshowsimilarpattern(Table A6intheAppendix).Therefore,weshouldalsorejectH3bforRWE,butwecannotrejectnullhypothesisagainstH3eforRWE.TheresultsofTable 11furthershowthat,whencomparingright-wingextremistsandleft-wingextremists(i.e.H3cvs.H3f),right-wingextremistsscorehigherontext-basedindicatorsofpositiveemotion(\(\mathit{Mean}\\mathit{Difference}=0.0038\),\(\mathit{CI}=[0.0019,0.0058]\),\(p<0.001\))andloweronnegativeemotion(\(\mathit{Mean}\\mathit{Difference}=-0.002\),\(\mathit{CI}=[-0.0032,-0.0008]\),\(p<0.001\)).BothmeandifferencesaresignificantattheBonferroniadjustedalphalevelof0.0055.ComparingLWEandRWEinTable A6intheAppendix,whereonlyconsidered25%ofthedata,demonstratethesamefindings.Therefore,theresultssupportH3candrejectH3f.Takentogether,theresultssuggestthatanyobserveddifferencesintext-basedindicatorsofpositiveandnegativeemotionsbetweenliberalsandconservativesarenotstatisticallysignificant.However,ingeneral,extremistsuselesslanguageindicativeofpositiveemotionandhigherofnegativeemotionscomparedtonon-extremists.TheonlyexceptionisbetweenRWEandRWinpositiveemotion,whichifuseaconservativesignificancethreshold,thedifferenceisnotsignificant.Finally,RWEhavehighertext-basedpositiveemotionandlowertext-basednegativescorescomparedtoLWE.MoralfoundationstheoryInthefollowingfivesub-sections,wepresentresultsontestinghypothesesabouttherelationshipbetweenpoliticalorientation/extremismandtheMoralFoundationsTheory(MFT).Thegeneralformofhypothesesweareinvestigatingherearewhetherornot: (a) MFT’spredictionsofconservativesspeakingmoreenthusiasticallythanliberalsaboutobediencetoauthority,ingrouployalty,andpurity,andliberalsspeakingmoreenthusiasticallyaboutfairnessandharmavoidancecanbegeneralizedtothepoliticalextremists;and (b) MFT’spredictionofelevatedmoralfoundationsamongextremistscanbeempiricallysupported. FairnessWeuseANOVAtotestforthecorrelationofpoliticalorientationandpoliticalextremitywithtext-basedmeasuresoffairnessandreporttheresultsinTable 12.Theresultsshowthatpoliticalorientationisasignificantpredictorofthelanguageusageofwordsindicativeofappreciationoffairness(\(F(1,1)=16.41\),\(p<0.001\)),evenifweusetheBonferroniadjustedalphalevelof0.0055.Incaseofpoliticalextremity,althoughitisasignificantpredictoroftheoutcomevariableatthe0.05level(\(F(1,1)=5.21\),\(p=0.022\)),itbecomesanonsignificantoneattheBonferroniadjustedsignificancelevelof0.0055.Table 12ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithtext-basedindicatorsoffairnessFullsizetablePosthoccomparisonsofmeandifferencesusingtheTukey’sHSDtest(Table 13)showthatalthoughconservativesscorelowerontext-basedindicatorsoffairness(whichisinagreementwithMFT),themeandifferenceisnotsignificantattheBonferroniadjustedalphalevelof0.0055.Inaddition,theresultsdemonstratethatright-wingextremists(RWE)usesignificantlylowernumberofwordsindicativeoffairnessthanleft-wingextremists(LWE)andthemeandifferenceissignificantattheBonferroniadjusted0.0055level(\(\mathit{Mean}\\mathit{Difference}=-0.0003\),\(\mathit{CI}=[-0.0005,-0.00021]\),\(p<0.001\)),whichsupportsH4a.Nonetheless,theresultsshowthatalthoughLWEscorehigherthanLWontext-basedindicatorsoffairness,themeandifferenceisnonsignificant(\(\mathit{Mean}\\mathit{Difference}=0.0001\),\(\mathit{CI}=[-0.0001,0.0002]\),\(p=0.87\)).Therefore,wecannotrejectthenullhypothesisagainstH4f.ResultsfromperformingTukey’sHSDtestonthesmallersamplesizedatawhereweonlyconsider25%oftheusersfromeachgrouprevealthesamepattern(Table A7intheAppendix).Table 13Multi-groupmeancomparisonsoffairnessusingTukey’sHSDtestFullsizetableIngeneral,theresultsofthissectionareinagreementwithMFT’spredictionsofliberalsspeakingmoreaboutfairnessthanconservatives,andleft-wingextremistsspeakingmoreaboutfairnessthanliberals(i.e.elevatedmoralfoundationsamongextremists).However,ourresultsdonotsupportthesignificanceofthesedifferences.Nevertheless,theresultssupportourhypothesisofLWEscorehigherthanRWEontext-basedindicatorsoffairness.Harmavoidance/careTheANOVAresultsfortestingthecorrelationofpoliticalorientationandpoliticalextremitywiththetext-basedmeasuresofharmavoidance(care)aredemonstratedinTable 14.Thereisaseeminglysignificantrelationshipbetweenpoliticalorientationandlanguageusageindicativeofharmavoidanceacrossliberals,conservatives,left-wingextremists,andright-wingextremists(\(F(1,1)=4.5\),\(p=0.034\)).However,themeandifferenceisnotsignificantwhenweusetheBonferroniadjustedalphalevelof0.0055.Ontheotherhand,politicalextremitydoesnotseemtosignificantlycorrelatewithtext-basedindicatorsofharmavoidanceacrossthefourpoliticalgroups(\(F(1,1)=0.91\),\(p=0.34\)).Table 14ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithtext-basedindicatorsofharmavoidanceFullsizetableWeuseTukey’sHSDtestforaposthocmeancomparisonsbetweenthegroups(Table 15)andresultsshowallpairwisemeandifferencesarenonsignificant(Bonferroniadjustedalpha =0.0055).Samepatternistruewhenweonlyconsider25percentofthedata(Table A8intheAppendix).Moreparticularly,wecanseethat,contrarytoMFT’spredictions,conservativesscorelowerthanliberalsontext-basedindicatorsofharmavoidance,though,themeandifferenceisnotsignificant(\(\mathit{Mean}\\mathit{Difference}=-0.0003\),\(\mathit{CI}=[-0.0008,0.0002]\),\(p=0.41\)).Inagreementwithourhypothesisofleft-wingextremists(LWE)expressingmorelanguageindicativeofharmavoidancethanright-wingextremists(i.e.H4b),theresultsofTable 15showthatRWEonaveragescore0.0004unitlowerthanLWEonharmavoidancelanguage-basedconstruct,however,themeandifferenceisnotsignificant(\(\mathit{Mean}\\mathit{Difference}=-0.0004\),\(\mathit{CI}=[-0.0009,0.0001]\),\(p=0.2\)).Inaddition,contrarytoMFT’sprediction,left-wingextremistsscorelowerthanliberalonlanguageindicatorsofharmavoidanceorcare(\(\mathit{Mean}\\mathit{Difference}=-0.0001\),\(\mathit{CI}=[-0.0006,0.0004]\),\(p=0.94\)),though,themeandifferenceisnonsignificant.Therefore,weshouldrejectH4g.Table 15Multi-groupmeancomparisonsofharmavoidanceusingTukey’sHSDtestFullsizetableTakentogether,theresultsofthissectiondiscouragethelinearextensionofMFT’spredictionsaboutharmavoidancetothepoliticalextremismcontext.Morespecifically,theresultssuggestthat,atleastwhenitcomestolanguageusageonTwitter,regardlessofpoliticalorientationandextremity,individualsarenotreallydifferentonhowmuchtheytalkaboutharmavoidanceandcare.IngrouployaltyTheANOVAresultsoftestingthecorrelationofpoliticalorientationandpoliticalextremitywithtext-basedmeasuresofingrouployaltyacrossdifferentpoliticalgroups(Table 16)showasignificantrelationshipwithpoliticalextremity(\(F(1,1)=6.065\),\(p=0.014\)).However,itbecomesnonsignificantwhenweusetheBonferroniadjustedalphalevelof0.0055.Acrossliberals,conservatives,left-andright-wingextremists,politicalorientationdoesnothaveasignificantcorrelationwithtext-basedexpressionofingrouployalty(\(F(1,1)=3.65\),\(p=0.056\)).Table 16ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithtext-basedindicatorsofingrouployaltyFullsizetableTofurthertestourhypothesesofwhetherright-wingextremistsexpressmorelanguageindicativeofingrouployaltythanleft-wingextremists(i.e.H4c),andright-wingextremistsexpressmorelanguageindicativeofingrouployaltythanconservatives(i.e.H4h),weuseposthoccomparisonsofmeansusingTukey’sHSDtestandreporttheresultsinTable 17.First,inagreementwithMFT’sprediction,theresultsshowthatconservativesscorehigherontext-basedindicatorsofingrouployaltythanliberalsandthemeandifferenceissignificantattheBonferroniadjustedalphalevelof0.0055(\(\mathit{Mean}\\mathit{Difference}=0.0005\),\(\mathit{CI}=[0.0001,0.001]\),\(p=0.0052\)).Second,andcontrarytoourhypothesisH4c,itisactuallyleft-wingextremiststhatscorehigherthanright-wingextremistsontext-basedindicatorsofingrouployalty(\(\mathit{Mean}\\mathit{Difference}=-0.0001\),\(\mathit{CI}=[-0.0006,0.0003]\),\(p=0.89\)),though,themeandifferenceisnotsignificant.Hence,weshouldrejectH4c.Third,andcontrarytoMFT’spredictionofelevatedingrouployaltyamongright-wingextremists,onaveragetheyscorelowerthanconservativeontext-basedmeasuresofingrouployalty,andthemeandifferenceissignificantattheBonferroniadjustedalphalevelof0.0055(\(\mathit{Mean}\\mathit{Difference}=-0.0006\),\(\mathit{CI}=[-0.001,-0.0002]\),\(p=0.0022\)).Therefore,werejectH4h.Analysisofthesmallersamplesize(25%ofthedata)revealsthesamepattern(Table A9intheAppendix).Theonlyexceptionisthatthemeandifferencebetweenconservativesandliberalsisnonsignificantatthe0.0055Bonferroniadjustedlevel(\(p=0.025\)).Table 17Multi-groupmeancomparisonsofingrouployaltyusingTukey’sHSDtestFullsizetableGenerally,whilesupportingtheMFT’sclaimofconservativesspeakingmoreaboutingrouployaltythanliberals,theresultsofthissectioncastseriousdoubtsontheextenttowhichthisobservationcanbegeneralizedtopoliticalextremists.Infact,theresultssuggestthat,onaverage,right-wingextremistshavetheleastexpressionofingrouployaltyintheirtweetscomparedtousersintheotherthreepoliticalgroups.ObediencetoauthorityTheANOVAresultstotesthypothesesontheextenttowhichMFTcanbegeneralizedtothepoliticalextremists(i.e.H4d),andwhetherornotMFT’spredictionofelevatedobediencetoauthorityamongtheright-wingextremistscanbeempiricallysupported(i.e.H4i)arereportedinTable 18.Theresultshowthatpoliticalorientationisasignificantpredictoroftext-basedindicatorsofobediencetoauthority(\(F(1,1)=56.05\),\(p<0.001\)),evenifweusetheBonferroniadjustedsignificancelevelof0.0055.However,politicalextremity’srelationshipwithlanguageusageindicativeofobediencetoauthorityisnonsignificant(\(F(1,1)=56.05\),\(p=0.12\)).Table 18ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithtext-basedindicatorsofobediencetoauthorityFullsizetablePosthoccomparisonsofthemeansofgroupsusingtheTukey’sHSDtest(Table 19)indicatesthatalthoughconservatives(RW)scorehigherontext-basedmeasuresofobediencetoauthority,themeandifferenceisnotsignificant(\(\mathit{Mean}\\mathit{Difference}=0.0003\),\(\mathit{CI}=[-0.0001,0.0007]\),\(p=0.18\)).Theresultsfurthershowthatright-wingextremists(RWE)scorehigherthanleft-wingextremists(LWE)ontext-basedmeasuresofobediencetoauthority(\(\mathit{Mean}\\mathit{Difference}=0.0017\),\(\mathit{CI}=[0.0013,0.0021]\),\(p<0.001\)).Inaddition,wecanseethatRWEusesignificantlyhigherlanguageindicativeofobediencetoauthoritythanconservatives(\(\mathit{Mean}\\mathit{Difference}=0.0009\),\(\mathit{CI}=[0.0005,0.0013]\),\(p<0.001\)).Thelattertwomeandifferencesaresignificantatthe0.0055Bonferroniadjustedalphalevel.Therefore,resultssupportbothH4dandH4i.PerformingtheTukey’sHSDtestonthesmallersamplesize(25%ofthedata)showssimilarresults(Table A10intheAppendix).Table 19Multi-groupmeancomparisonsofobediencetoauthorityusingTukey’sHSDtestFullsizetableGenerally,theresultsofthissectionsupportMFT’spredictionaboutcross-groupvariationsinobediencetoauthorityvalues.Theresultssuggestthat,intermsofthefrequencyofusingtext-basedsignalsofobediencetoauthority,conservativesscorehigherthanliberals,right-wingextremistsscorehigherthanleft-wingextremists,andright-wingextremistsscorehigherthanconservatives.Withtheexceptionofmeandifferencesbetweenconservativesandliberals,allotherfindingsaresignificantattheBonferroniadjustedalphalevelof0.0055.PurityTable 20presentstheANOVAresultsofthetestforcorrelationofpoliticalorientationandpoliticalextremitywithtext-basedmeasuresofpurityacrossliberals,conservatives,left-wingextremist(LWE),andright-wingextremists(RWE).Wecanseethatpoliticalorientationisasignificantpredictorofthetext-basedmeasuresofpurity(\(F(1,1)=8.54\),\(p=0.003\)),evenattheBonferroniadjustedalphalevelof0.0055.However,thepoliticalextremity’srelationshipwiththedegreetowhichindividualsuselanguageindicativeofpurityisnotsignificant(\(F(1,1)=0.34\),\(p=0.56\)).Table 20ANOVAresultstotestforcorrelationofpoliticalorientationandextremitywithtext-basedindicatorsofpurityFullsizetableTheresultsoftheTukey’sHSDposthoccomparisonsofmeanstotestthehypothesesofright-wingextremistsscorehigherthanleft-wingextremistsonlanguage-basedindicatorsofpurity(H4e)andright-wingextremistscorehigherthanconservatives(H4j)aredemonstratedinTable 21.First,inagreementwithMFT,wecanseethatconservativesexpressmorelanguageindicativeofpuritythanliberals(\(\mathit{Mean}\\mathit{Difference}=0.0003\),\(\mathit{CI}=[0.0001,0.0006]\),\(p=0.086\)),however,themeandifferenceisnonsignificantatthe0.0055Bonferroniadjustedlevel.Second,althoughright-wingextremistsscorehigherthanleft-wingextremistsontext-basedmeasuresofpurity,themeandifferenceisnotsignificant(\(\mathit{Mean}\\mathit{Difference}=0.0002\),\(\mathit{CI}=[-0.0001,0.0005]\),\(p=0.25\)),whichinturn,rejectsH4e.Finally,RWEonaveragescorehigherthanconservativesontext-basedindicatorsofpurity,butthemeandifferenceisnotsignificantatthe0.0055Bonferroniadjustedlevel(\(\mathit{Mean}\\mathit{Difference}=0.0001\),\(\mathit{CI}=[-0.0003,0.0003]\),\(p=0.997\)).Therefore,wecannotrejectthenullhypothesisagainstH4j.Analysisofthe25percentsamplesizerevealprettymuchthesamepattern(Table A11intheAppendix):conservativesscorehigherthanliberalsandright-wingextremistsscorehigherthanleft-wingextremists,butbothmeandifferencesarenonsignificant.However,inTable A11,right-wingextremistsscorelowerthanconservatives,butthemeandifferenceisnotsignificantatthe0.0055Bonferroniadjustedlevel(\(p=0.16\)).Table 21Multi-groupmeancomparisonsofpurityusingTukey’sHSDtestFullsizetableEffectsizesResultsthusfarindicatethattherearelanguagedifferencesacrossliberals,conservatives,left-wingextremists,andright-wingextremistsintermsoftheproportionofwordsthatappearintheLIWCandMFDdictionaries.Butanimportantquestionremainsthatwhatdotheselanguagedifferencesmeanpsychologicallyandmorally?Onewaytoanswerthisquestionistomeasurethecorrelationofthesetext-basedindicatorswithactualquestionnaire-basedpsychologicalandmoralprofilesofindividuals.Thechallengehereisthat,asmentionedintheintroductionpartofthispaper,politicalextremistsdonotsimplyvolunteerforpsychologicalstudiesandthusitishardtofindreliableresultsbasedonarelativelylargesample.Anotherpossiblewaytofindananswerforthevalidityofthemeandifferenceswouldbetocomparethemwithtext-basedpsychologicalandmoralindicatorsofindividualswhohavesomesortofpsychologicaloremotionaldisorder.However,thereareatleasttwoproblemsassociatedwiththisapproach.First,althoughpreviousapplicationsoftheLIWCdictionariessupportthesignificanceofthisapproachinmentalhealthresearch(e.g.ChungandPennebaker[19],DeChoudhuryetal.[28],Coppersmithetal.[22]),thecomparabilityofthegroupsintermsofothervariablesofinterestorthesufficient/necessarydiagnosticcriteriaforanyofthedisordersarequestionable.Second,webelievethesekindsofcomparisonsareinappropriatebecausetheymightinadvertentlyimplyimpreciseclaimsaboutthepopulationunderstudy.DiscussionInthispaper,wetestedhypothesesaboutthepsychologicalandmoralprofilesofpoliticalextremistsandtheirdifferenceswithpartisanusersusingTwitterdataofU.S.left-andright-wingextremistgroups.WesummarizedallhypothesesandcorrespondingresultsinTable 22.Theresultsoffernewinsightstothemoralprofileofpoliticalextremistswhilesettlingsomepreviouslydebatedtheoriesofthepsychologicalappealofpoliticalextremityandorientation.Languageanalysisoftweetswrittenby10,000extremistsand10,000non-extremistpartisanusersshowedthatregardlessofpoliticalorientation(i.e.leftorright),extremistsshowalowertext-basedindicatorsofpositiveemotionandhighertext-basedindicatorsofnegativeemotioncomparedtonon-extremists(i.e.liberalsandconservatives).Inaddition,wefoundthatright-wingextremistsusemorewordsindicativeofpositiveemotionandlowerwordsindicativeofnegativeemotioncomparedtoleft-wingextremists.Table 22SummaryofhypothesesandresultsFullsizetableIncaseofanxiety,whileleft-wingextremistsscorehigherthanallothergroupsonlanguage-basedmeasuresofanxiety,right-wingextremistsscorelowerthanconservatives.Infact,right-wingextremistsexpresstheleastlanguageindicativeofanxietyamongallfourstudiedpoliticalgroups.Incaseofcertainty,surprisingly,wefoundnoevidenceofsignificantdifferencesbetweentheaveragetext-basedindicatorsofcertaintyacrossliberals,conservatives,left-,andright-wingextremists.Overall,thepatterncastseriousdoubtsonmainstreamviewofseeingextremistsasdogmaticandrigidindividualswithhighattitudecertaintyandanxiety.Instead,itlendssupporttotheoriessuchastheSystemJustificationTheory(JostandHunyady[52])whicharguethatideologyingeneralandpoliticalextremityinparticularoperateasabufferingfactorformentaldistress.Forexample,theobservationsthatright-wingersexpressmorepositiveemotionandlessnegativeemotionwordsthanleft-wingersareanalogoustopreviousfindingsthatshowedconservativesarehappierthanliberals(e.g.NapierandJost[66]),whichcanbeexplainedbytheideathatconservativesystem-justifyingideologiesserveasapalliativefunction(JostandBanaji[50]).Inaddition,theresultsresonatewithSidanius’s[84]viewofcharacterizingpoliticalextremistsassophisticatedindividualswhocanmanagethesocietalpressureofbeingdeviantandhavehigheruncertaintytolerance.However,anotherexplanationcouldsimplybethefactthathavingaRepublicanpresidentmakesright-wingusersmoresatisfiedwiththepoliciesandthereforetheyaremorelikelytousewordswithpositiveemotionconnotations.OnecanconductalongitudinalstudyandcomparethepsychologicalindicatorsoftheusersanalyzedinthispaperwiththeircorrespondingpsychologicalindicatorsinthetimeofObama’spresidency.WithrespecttotheMoralFoundationsTheory(MFT),althoughourresultsareingeneralagreementwiththeMFT,theydemandtocarefullyinvestigatetheroleofpoliticalextremiststounderstandhowmuchoftheobserveddifferencesbetweenliberalsandconservativesaredrivenbytheindividualsatboth(extreme)endsofthepoliticalspectrum.OurresultssupporttheMFT’spredictionofconstativesscoringhigherthanliberalsonobediencetoauthority,ingrouployalty,andpurityandloweronfairness,however,onlytheingrouployaltymeandifferencewassignificant.Theresults,however,donotsupporttheMFT’spredictionofliberalsscoringhigherthanconservativesontext-basedindicatorsofharmavoidanceandcare.Infact,ourresultsshowquitetheoppositedirection,though,themeandifferencewasnotsignificant.Thiscanbeattributed,atleastinpart,tothevalidityoftheMoralFoundationsDictionary(MFD)itself.InourefforttovalidatetheMFDontweetswrittenbyAmericanpoliticalextremists,wefoundtheleastcorrelationandthecorrespondingp-valuebetweenthehand-codedtweetsandMFD-generatedscoresfortheharmavoidancecategory.Nevertheless,ourresultsshowthat,withtheexceptionofingrouployalty,MFTcanbegeneralizedtopoliticalextremismcontext.Wefoundright-wingextremistsspeakmorelanguageindicativeofobediencetoauthorityandpurityandlesslanguageindicativeoffairnessandharmavoidancethanleft-wingextremists(themeandifferencesinfairnessandobediencetoauthorityweresignificant).Surprisingly,incaseofingrouployalty,theresultspaintleft-wingextremistsastheoneswhousemorewordsandphrasesindicativeofingrouployaltythanright-wingextremists(thoughthemeandifferencewasnonsignificant).However,ourresultswereinconclusivewithrespecttoMFT’sargumentofelevatedmoralfoundationamongpoliticalextremists.Ifthiswasthecase,right-wingextremistsshouldhavescoredhigherontext-basedmeasuresofobediencetoauthority,in-grouployalty,andpuritycomparedtoconservatives,andleft-wingextremistsshouldhavescoredhigherontext-basedmeasuresoffairnessandharmavoidancethanliberals.Amongthefivemoralfoundations,ourresultswereconsistentwiththeabovepredictionincasesofobediencetoauthority,fairness,andpurity.Buttheonlysignificanteffectsizewasforobediencetoauthorityandtheothertwomeandifferenceswerenonsignificant.Ontheotherhand,theresultsshowtheoppositedirectionforharmavoidanceandingrouployalty.Infact,itwasconservativeswhospeakmoreaboutingrouployaltythanright-wingextremists,andthatliberalswhospeakmoreaboutharmavoidancethanleft-wingextremists.Ourresultsprovidenoevidenceofacausalrelationshipbetweenanyoftheanalyzedpsychologicalormoralvariablesandpoliticalextremism.Thosewhoaremoreemotional/moralmaybemoresusceptibletoextremistappealsorthosewithextremistviewsmayexpressthemselveswithmoreemotion/morality,orboth.Forinstance,becominganextremistcouldleadonetoexpressnegativeemotion,inpartbecausealmostnoonesharesyourviews,ratherthan(orinadditionto)beingasourceofitsappeal.Furthermore,theremaybepsychologicaldifferencesbetweenextremistswhoexpresstheirpoliticalopinionsandemotionsonTwitterandthosewhodonot,orthosewhodonotevenusesocialmedia.Nordotheseresultsruleoutthealternativetheorythatmaterialdeprivationandpoliticaloppressionencourageextremistviewsandemotionalagitation.Moreover,althoughtweetscanreflectvoluntaryexpressionsofsupportforcertaingroupsorideas,usersarenotfreefromdesirabilityorself-presentationsbiases.Forinstance,theymightchoosenottopubliclyexpresstheirrealopinionsandemotionsbecausetheyafraidtogetsuspendedbyTwitterorisolatedbytheirfollowers.Wealsocannotruleoutanygeographicalbiasimposedonourresults(seeAlizadehetal.[5]forareviewoftheeffectofspatialproximityonpeerinteractionandopiniondynamics)sincewedonothaveaccesstothelocationoftheusers.Wealsohavenoevidenceaboutextremistorviolentbehavior.Ouranalysisisfocusedonthosewhofollowandsupportnon-violentextremistsandisthereforelimitedtothepsychologicalprofilesassociatedwithattractiontoextremistviews.Wedonothavedataforthefollowersofviolentextremistgroupsorbehavioralmeasuresoftheiractivitiesoffline.Theresultsareneverthelessusefulinprovidingevidenceaboutthepsychologicalandmoralprofilesofthosewhoareattractedtoextremistviews. AbbreviationsLIWC: LinguisticInquiryandWordCount LW: Left-Wingers(Liberals) LWE: Left-WingExtremists MFD: MoralFoundationsDictionary MFT: MoralFoundationsTheory RW: Right-Wingers(Conservatives) RWE: Right-WingExtremists References AdornoTW,Frenkel-BrunswikE,LevinsonDJ,SanfordRN(1950)Theauthoritarianpersonality.Harper,NewYork GoogleScholar  AlizadehM,Cioffi-RevillaC,(2014)Distributionsofopinionandextremistradicalization:Insightsfromagent-basedmodeling.In:The6thInternationalConferenceonSocialInformatics,pp 348–358 Chapter  GoogleScholar  AlizadehM,Cioffi-RevillaC(2015)Activationregimesinopiniondynamics:comparingasynchronousupdatingschemes.JArtifSocSocSimul18(3):8 Article  GoogleScholar  AlizadehM,Cioffi-RevillaC,CrooksAT(2015)Theeffectofin-groupfavoritismonthecollectivebehaviorofindividuals’opinions.AdvComplexSyst18(01n02):1550002 MathSciNet  Article  GoogleScholar  AlizadehM,Cioffi-RevillaC,CrooksAT(2017)Generatingandanalyzingspatialsocialnetworks.ComputMathOrganTheory23(3):362–390 Article  GoogleScholar  AlizadehM,ComanA,LewisM,Cioffi-RevillaC(2014)Intergroupconflictescalationleadstomoreextremism.JArtifSocSocSimul17(4):4 Article  GoogleScholar  AlquistJL,AinsworthSE,BaumeisterRF(2013)Determinedtoconform:disbeliefinfreewillincreasesconformity.J ExpSocPsychol49:80–86.https://doi.org/10.1016/j.jesp.2012.08.015 Article  GoogleScholar  AltemeyerB(1996)Theauthoritarianspecter.HarvardUniversityPress,Cambridge GoogleScholar  AltemeyerB(1998)Theother“authoritarianpersonality”.In:ZannaMP(ed)Advancesinexperimentalsocialpsychology,vol 30.AcademicPress,SanDiego,pp 47–92 GoogleScholar  ArlowJ(2019)Antifawithoutfascism:thereasonsbehindtheanti-fascistmovementinIreland.IrPolitStud,1–23 AtranS,GingesJ(2012)Religiousandsacredimperativesinhumanconflict.Science336(6083):855–857 Article  GoogleScholar  BarberáP(2015)Birdsofthesamefeathertweettogether:BayesianidealpointestimationusingTwitterdata.PolitAnal23(1):76–91 Article  GoogleScholar  BeinartP(2017)Theriseoftheviolentleft:Antifa’sactivistssaythey’rebattlingburgeoningauthoritarianismontheAmericanright.Aretheyfuelingitinstead?TheAtlantic’sPoliticsandPolicy.September.https://www.theatlantic.com/magazine/archive/2017/09/the-rise-of-the-violent-left/534192/ BobbioN(1996)Leftandright:thesignificanceofapoliticaldistinction.UniversityofChicagoPress,Chicago GoogleScholar  BonikowskiB(2016)Nationalisminsettledtimes.AnnuRevSociol42:427–449 Article  GoogleScholar  ButlerJC(2000)Personalityandemotionalcorrelatesofright-wingauthoritarianism.SocBehavPersIntJ28(1):1–14 Article  GoogleScholar  Canetti-NisimD,HalperinE,SharvitK,HobfollSE(2009)Anewstress-basedmodelofpoliticalextremismpersonalexposuretoterrorism,psychologicaldistress,andexclusionistpoliticalattitudes.JConflResolut53(3):363–389 Article  GoogleScholar  CastelliGattinaraP,O’ConnorF,LindekildeL(2018)Nocountryforactingalone?Theneo-fascistmovementandlone-actorterroristattacksinItaly.PerspectTerror12(6):136–148 GoogleScholar  ChungC,PennebakerJW(2007)Thepsychologicalfunctionsoffunctionwords.SocCommun1:343–359 GoogleScholar  CliffordS,JeritJ(2013)Howwordsdotheworkofpolitics:Moralfoundationstheoryandthedebateoverstemcellresearch.J Polit75(3):659–671 GoogleScholar  ConoverMD,GonçalvesB,FlamminiA,MenczerF(2012)PartisanasymmetriesinonlinepoliticalactivityEPJDataSci1(1):6 Article  GoogleScholar  CoppersmithG,HarmanC,DredzeM(2014)MeasuringposttraumaticstressdisorderinTwitter.In:EighthinternationalAAAIconferenceonweblogsandsocialmedia GoogleScholar  CrowsonHM,ThomaSJ,HestevoldN(2005)Ispoliticalconservatismsynonymouswithauthoritarianism?.J SocPsychol145(5):571–592 GoogleScholar  DavidsonT,WarmsleyD,MacyMW,WeberI(2017)Automatedhatespeechdetectionandtheproblemofoffensivelanguage.arXivpreprint.arXiv:1703.04009 DaviesCL,SibleyCG,LiuJH(2014)Confirmatoryfactoranalysisofthemoralfoundationsquestionnaire.SocPsychol45:431–436 Article  GoogleScholar  DavisCA,VarolO,FerraraE,FlamminiA,MenczerF(2016)Botornot:asystemtoevaluatesocialbots.In:Proceedingsofthe25thinternationalconferencecompaniononWorldWideWeb.InternationalWorldWideWebConferencesSteeringCommittee,pp 273–274 Chapter  GoogleScholar  DayMV,FiskeST,DowningEL,TrailTE(2014)Shiftingliberalandconservativeattitudesusingmoralfoundationstheory.PersSocPsycholBull40(12):1559–1573 Article  GoogleScholar  DeChoudhuryM,GamonM,CountsS,HorvitzE(2013)Predictingdepressionviasocialmedia.In:SeventhinternationalAAAIconferenceonweblogsandsocialmedia GoogleScholar  DollardJ,DoobLW,MillerNE,MowrerOH,SearsRR(1939)Frustrationandaggression.YaleUniversityPress,NewHaven Book  GoogleScholar  DuckittJ(2001)Adual-processcognitive-motivationaltheoryofideologyandprejudice.In:ZannaMP(ed)Advancesinexperimentalsocialpsychology,vol 33.AcademicPress,SanDiego,pp 41–113 GoogleScholar  DuriezB,KlimstraTA,LuyckxK,BeyersW,SoenensB(2012)Right-wingauthoritarianism:protectivefactoragainstorriskfactorfordepression?EurJPers26(5):536–549 Article  GoogleScholar  EnstadJD(2018)Right-wingterrorismandviolenceinPutin’sRussia.PerspectTerror12(6):89–103 GoogleScholar  FerraraE(2017)Computationalsocialsciencetogaugeonlineextremism.arXivpreprint.arXiv:1701.08170 FibertZ,ResslerWH(1998)IntoleranceofambiguityandpoliticalorientationamongIsraeliuniversitystudents.J SocPsychol138(1):33–40 GoogleScholar  FlacheA,MacyMW(2011)Smallworldsandculturalpolarization.JMathSociol35(1–3):146–176 MathSciNet  Article  GoogleScholar  Frenkel-BrunswikE(1949)Intoleranceofambiguityasanemotionalandperceptualpersonalityvariable.JPers18(1):108–143 Article  GoogleScholar  GrahamJ,HaidtJ(2012)Sacredvaluesandeviladversaries:amoralfoundationsapproach.In:Thesocialpsychologyofmorality:exploringthecausesofgoodandevil,pp 11–31 Chapter  GoogleScholar  GrahamJ,HaidtJ,NosekBA(2009)Liberalsandconservativesrelyondifferentsetsofmoralfoundations.JPersSocPsychol96:1029 Article  GoogleScholar  GreenbergJ,JonasE(2003)Psychologicalmotivesandpoliticalorientation-theleft,theright,andtherigid:commentonJostetal.(2003) GreenbergJ,JonasE(2003)Psychologicalandpoliticalorientation—theleft,theright,andtherigid:commentonJostetal.(2003).PsycholBull129:376–382. Article  GoogleScholar  GreenbergJ,PyszczynskiT,SolomonS,RosenblattA,VeederM,KirklandS,LyonD(1990)EvidenceforterrormanagementtheoryII:theeffectsofmortalitysalienceonreactionstothosewhothreatenorbolstertheculturalworldview.JPersSocPsychol58(2):308 Article  GoogleScholar  GreenbergJ,SolomonS,PyszczynskiT(1997)Terrormanagementtheoryofself-esteemandculturalworldviews:empiricalassessmentsandconceptualrefinements.AdvExpSocPsychol29:61–139 GoogleScholar  HaidtJ(2012)Therighteousmind:whygoodpeoplearedividedbypoliticsandreligion.Vintage HaidtJ,GrahamJ(2007)Whenmoralityopposesjustice:conservativeshavemoralintuitionsthatliberalsmaynotrecognize.SocJusticeRes20(1):98–116 Article  GoogleScholar  HaidtJ,JosephC(2004)Intuitiveethics:howinnatelypreparedintuitionsgenerateculturallyvariablevirtues.Daedalus133(4):55–66 Article  GoogleScholar  ImaiK,RatkovicM(2014)Covariatebalancingpropensityscore.JRStatSoc,SerB,StatMethodol76(1):243–263 MathSciNet  Article  GoogleScholar  ImbensGW(2004)Nonparametricestimationofaveragetreatmenteffectsunderexogeneity:areview.RevEconStat86(1):4–29 MathSciNet  Article  GoogleScholar  JaskoK,LaFreeG,KruglanskiAW(2016)Questforsignificanceandviolentextremism:thecaseofdomesticradicalization.PolitPsychol38(5):815–831 Article  GoogleScholar  JostJT(2017)Ideologicalasymmetriesandtheessenceofpoliticalpsychology.PolitPsychol38(2):167–208 Article  GoogleScholar  JostJT,BanajiMR(1994)Theroleofstereotypinginsystem-justificationandtheproductionoffalseconsciousness.BrJSocPsychol33(1):1–27 Article  GoogleScholar  JostJT,GlaserJ,KruglanskiAW,SullowayFJ(2003)Politicalconservatismasmotivatedsocialcognition.PsycholBull129:339–375 Article  GoogleScholar  JostJT,HunyadyO(2002)Thepsychologyofsystemjustificationandthepalliativefunctionofideology.EurRevSocPsychol13:111–153 Article  GoogleScholar  JostJT,NapierJL,ThorisdottirH,GoslingSD,PalfaiTP,OstafinB(2007)Areneedstomanageuncertaintyandthreatassociatedwithpoliticalconservatismorideologicalextremity?PersSocPsycholBull33(7):989–1007 Article  GoogleScholar  KangJD,SchaferJL(2007)Demystifyingdoublerobustness:acomparisonofalternativestrategiesforestimatingapopulationmeanfromincompletedata.StatSci22(4):523–539 MathSciNet  MATH  Article  GoogleScholar  KitsakM,GallosLK,HavlinS,ChenX,LiljerosF,MuchnikL,StanleyHEMakseHA(2010)Identificationofinfluentialspreadersincomplexnetworks.NatPhys6(11):888–893.https://doi.org/10.1038/nphys1746 Article  GoogleScholar  KoehlerD(2018)RecenttrendsinGermanright-wingviolenceandterrorism:whatarethecontextualfactorsbehind‘hiveterrorism’?PerspectTerror12(6):72–88 GoogleScholar  KruglanskiAW(2013)Thepsychologyofclosedmindedness.PsychologyPress GoogleScholar  KruglanskiAW,GelfandM,GunaratnaR(2012)Terrorismasmeanstoanend:howpoliticalviolencebestowssignificance.In:ShaverPR,MikulincerM(eds)Meaning,mortality,andchoice:thesocialpsychologyofexistentialconcerns.AmericanPsychologicalAssociation,Washington,pp 203–212 Chapter  GoogleScholar  KruglanskiAW,JaskoK,ChernikovaM,DugasM,WebberD(2017)Tothefringeandback:violentextremismandthepsychologyofdeviance.AmPsychol72:217–230 Article  GoogleScholar  KruglanskiAW,WebsterDM(1996)Motivatedclosingofthemind:“seizing”and“freezing”.In:Themotivatedmind.Routledge,London,pp 68–111 GoogleScholar  KurzmanC,KamalA,YazdihaH(2017)Ideologyandthreatassessment:lawenforcementevaluationofmuslimandright-wingextremism.Socius3:2378023117704771 GoogleScholar  LaFreeG(2018)IsAntifaaterroristgroup?Society55(3):248–252 Article  GoogleScholar  MacklinG(2018)‘Onlybulletswillstopus!’—thebanningofnationalactioninBritain.PerspectTerror12(6):104–122 GoogleScholar  MagdyW,DarwishK,WeberI(2016)#FailedRevolutions:usingTwittertostudytheantecedentsofISISsupport.FirstMonday21:2 Article  GoogleScholar  MarešM(2018)Right-wingterrorismandviolenceinHungaryatthebeginningofthe21stcentury.PerspectTerror12(6):123–135 GoogleScholar  NapierJL,JostJT(2008)Whyareconservativeshappierthanliberals?PsycholSci19(6):565–572 Article  GoogleScholar  NilssonA,ErlandssonA(2015)Themoralfoundationstaxonomy:structuralvalidityandrelationtopoliticalideologyinSweden.PersIndividDiffer76:28–32 Article  GoogleScholar  OnraetE,VanHielA,DhontK(2013)Therelationshipbetweenright-wingideologicalattitudesandpsychologicalwell-being.PersSocPsycholBull39(4):509–522 Article  GoogleScholar  PennebakerJW,BoydRL,JordanK,BlackburnK(2015)ThedevelopmentandpsychometricpropertiesofLIWC2015 PennebakerJW,FrancisME,BoothRJ(2001)Linguisticinquiryandwordcount:LIWC2001.LawrenceErlbaumAssociates71,Mahway. GoogleScholar  PennebakerJW,MehlMR,NiederhofferKG(2003)Psychologicalaspectsofnaturallanguageuse:Ourwords,ourselves.AnnuRevPsychol54(1):547–577 Article  GoogleScholar  PetersonBE,DuncanLE(2007)Midlifewomen’sgenerativityandauthoritarianism:marriage,motherhood,and10yearsofaging.PsycholAging22(3):411 Article  GoogleScholar  PierceWL(1978)TheTurnerdiaries.NationalVanguardBooks,Hillsboro GoogleScholar  PooleKT,RosenthalH(1985)Aspatialmodelforlegislativerollcallanalysis.AmJPolitSci29:357–384 Article  GoogleScholar  PrattoF,SidaniusJ,StallworthLM,MalleBF(1994)Socialdominanceorientation:apersonalityvariablepredictingsocialandpoliticalattitudes.JPersSocPsychol67(4):741 Article  GoogleScholar  PrinceJ(2016)Psychologyofextremism.Extremism,counter-terrorismandpolicing,51 RavandiB,MiliF(2019)Coherenceandpolarizationincomplexnetworks.JComputSocSci,1–18 RoccatoM,RussoS(2017)Right-wingauthoritarianism,societalthreattosafety,andpsychologicaldistress.EurJSocPsychol47:600–610 Article  GoogleScholar  RokeachM(1960)Theopenandclosedmind.BasicBooks,NewYork GoogleScholar  RosenbaumPR,RubinDB(1983)Thecentralroleofthepropensityscoreinobservationalstudiesforcausaleffects.Biometrika70(1):41–55 MathSciNet  MATH  Article  GoogleScholar  RoweM,SaifH(2016)Miningpro-ISISradicalisationsignalsfromsocialmediausers.In:ProceedingsofthetenthinternationalAAAIconferenceonwebandsocialmedia(ICWSM2016),pp 329–338 GoogleScholar  ShaoC,HuiPM,WangL,JiangX,FlamminiA,MenczerF,CiampagliaGL(2018)Anatomyofanonlinemisinformationnetwork.arXivpreprint.arXiv:1801.06122 SidaniusJ(1985)Cognitivefunctioningandsociopoliticalideologyrevisited.PolitPsychol6:637–661 Article  GoogleScholar  SidaniusJ(1988)Politicalsophisticationandpoliticaldeviance:astructuralequationexaminationofcontexttheory.JPersSocPsychol55(1):37 Article  GoogleScholar  SinnenbergL,ButtenheimAM,PadrezK,ManchenoC,UngarL,MerchantRM(2017)Twitterasatoolforhealthresearch:asystematicreview.AmJPublHealth107(1):e1–e8 Article  GoogleScholar  SmithJA,ToddPE(2005)DoesmatchingovercomeLaLonde’scritiqueofnonexperimentalestimators?JEconom125(1–2):305–353 MathSciNet  MATH  Article  GoogleScholar  SmithMR,GordonRA(1998)Personalneedforstructureandattitudestowardhomosexuality.JSocPsychol138(1):83–87 Article  GoogleScholar  SorrentinoRM,RoneyCJ(1986)Uncertaintyorientation,achievement-relatedmotivation,andtaskdiagnosticityasdeterminantsoftaskperformance.SocialCogn4(4):420–436 Article  GoogleScholar  StuartEA(2010)Matchingmethodsforcausalinference:areviewandalookforward.StatSci25(1):1 MathSciNet  MATH  Article  GoogleScholar  SweeneyMM,PerligerA(2018)Explainingthespontaneousnatureoffar-rightviolenceintheUnitedStates.PerspectTerror12(6):52–71 GoogleScholar  TajfelH(1982)Socialidentityandintergrouprelations.CambridgeUniversityPress,Cambridge GoogleScholar  TaylorP,FunkC,CraighillP(2006)Arewehappyyet?RetrievedApril3,2018,fromthePewResearchCenterwebsite:http://www.pewsocialtrends.org/2006/02/13/are-we-happy-yet/ TetlockPE(1984)CognitivestyleandpoliticalbeliefsystemsintheBritishHouseofCommons.JPersSocPsychol46:365–375 Article  GoogleScholar  TomkinsSS(1963)Leftandright:abasicdimensionofideologyandpersonality.In:WhiteRW(ed)Thestudyoflives.Atherton,Chicago,pp 388–411 GoogleScholar  VanHielA,DeClercqB(2009)Authoritarianismisgoodforyou:right-wingauthoritarianismasabufferingfactorformentaldistress.EurJPers23(1):33–50 Article  GoogleScholar  vanProoijenJW,KrouwelAP,BoitenM,EendebakL(2015)Fearamongtheextremeshowpoliticalideologypredictsnegativeemotionsandoutgroupderogation.PersSocPsycholBull4:0146167215569706 GoogleScholar  Vargas-SalfateS,PaezD,KhanSS,LiuJH,deZúñigaHG(2018)Systemjustificationenhanceswell-being:alongitudinalanalysisofthepalliativefunctionofsystemjustificationin18countries.BrJSocPsychol57:567–590 Article  GoogleScholar  VarolO,FerraraE,DavisCA,MenczerF,FlamminiA(2017)Onlinehuman-botinteractions:detection,estimation,andcharacterization.arXivpreprint.arXiv:1703.03107 WebberD,BabushM,Schori-EyalN,Vazeou-NieuwenhuisA,HettiarachchiM,BélangerJJ,MoyanoM,TrujilloHM,GunaratnaR,KruglanskiAW,GelfandMJ(2018)Theroadtoextremism:fieldandexperimentalevidencethatsignificanceloss-inducedneedforclosurefostersradicalization.JPersSocPsychol114:270–285 Article  GoogleScholar  WestenD(2008)Thepoliticalbrain:theroleofemotionindecidingthefateofthenation.PublicAffairs GoogleScholar  WindischS,SimiP,BleeK,DeMicheleM(2018)Understandingthemicro-situationaldynamicsofwhitesupremacistviolenceintheUnitedStates.PerspectTerror12(6):23–37 GoogleScholar  YilmazO,HarmaM,BahçekapiliHG,CesurS(2016)ValidationofthemoralfoundationsquestionnaireinTurkeyanditsrelationtoculturalschemasofindividualismandcollectivism.PersIndividDiffer99:149–154 Article  GoogleScholar  ZhaoWX,JiangJ,WengJ,HeJ,LimEP,YanH,LiX(2011)ComparingTwitterandtraditionalmediausingtopicmodels.In:Europeanconferenceoninformationretrieval.Springer,Berlin,pp 338–349 GoogleScholar  Downloadreferences Acknowledgements Notapplicable. Availabilityofdataandmaterials ThedatasetsgeneratedandanalyzedduringthecurrentstudyarenotpubliclyavailableduetotheTwitter’sDeveloperAgreement.However,theuserIDsandtweetIDsareavailablefromthecorrespondingauthoronrequest. FundingThisstudywassupportedinpartbytheQatarComputingResearchInstitute.M. AlizadehwaspartiallyandC. Cioffi-RevillaisfundedbyONR-MinervaGrantNo. N00014130054.AuthorinformationAuthorsandAffiliationsEmpiricalStudiesofConflictProject,WoodrowWilsonSchoolofPublicandInternationalAffairs,PrincetonUniversity,Princeton,USAMeysamAlizadehQatarComputingResearchInstitute,Doha,QatarIngmarWeberComputationalSocialScienceProgram,DepartmentofComputationalandDataSciences,GeorgeMasonUniversity,Fairfax,USAClaudioCioffi-RevillaCenterforComplexNetworksandSystemsResearch,SchoolofInformatics,Computing,andEngineering,IndianaUniversity,Bloomington,USASantoFortunatoSocialDynamicsLaboratory,DepartmentofSociology,CornellUniversity,Ithaca,USAMichaelMacyAuthorsMeysamAlizadehViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarIngmarWeberViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarClaudioCioffi-RevillaViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarSantoFortunatoViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarMichaelMacyViewauthorpublicationsYoucanalsosearchforthisauthorin PubMed GoogleScholarContributionsAllauthorscontributedtothedesignofthestudy.MAandIWcollectedthedata.MAanalyzedthedata.Allauthorscontributedtothemanuscript.Allauthorsreadandapprovedthefinalmanuscript.CorrespondingauthorCorrespondenceto MeysamAlizadeh.Ethicsdeclarations Competinginterests Theauthorsdeclarenocompetinginterests. AdditionalinformationPublisher’sNoteSpringerNatureremainsneutralwithregardtojurisdictionalclaimsinpublishedmapsandinstitutionalaffiliations.AppendixAppendix Table A1Listofwhitenationalistandneo-nazigroupsintheU.S.whohaveactiveTwitteraccountonDecember2017Fullsizetable Table A2ListofAntifagroupsinU.S.whohaveactiveTwitteraccountonDecember2017Fullsizetable Table A3Listofthewordsineachtweet-LDAestimatedtopicsalongwiththeirsuggestedpotentialnamesFullsizetable Table A4Multi-groupmeancomparisonsofcertaintyin25%samplesizedatausingTukey’sHSDtestFullsizetable Table A5Multi-groupmeancomparisonsofanxietyin25%samplesizedatausingTukey’sHSDtestFullsizetable Table A6Multi-groupmeancomparisonsofpositiveandnegativeemotionsin25%samplesizedatausingTukey’sHSDtestFullsizetable Table A7Multi-groupmeancomparisonsoffairnessin25%samplesizedatausingTukey’sHSDtestFullsizetable Table A8Multi-groupmeancomparisonsofharmavoidancein25%samplesizedatausingTukey’sHSDtestFullsizetable Table A9Multi-groupmeancomparisonsofingrouployaltyin25%samplesizedatausingTukey’sHSDtestFullsizetable Table A10Multi-groupmeancomparisonsofobediencetoauthorityin25%samplesizedatausingTukey’sHSDtestFullsizetable Table A11Multi-groupmeancomparisonsofpurityin25%samplesizedatausingTukey’sHSDtestFullsizetable Figure A1DistributionofthecovariateslistedinTable 5(exceptfortopics)acrossdifferentpoliticalgroupsFullsizeimage Rightsandpermissions OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttribution4.0InternationalLicense(http://creativecommons.org/licenses/by/4.0/),whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedyougiveappropriatecredittotheoriginalauthor(s)andthesource,providealinktotheCreativeCommonslicense,andindicateifchangesweremade. ReprintsandPermissionsAboutthisarticleCitethisarticleAlizadeh,M.,Weber,I.,Cioffi-Revilla,C.etal.Psychologyandmoralityofpoliticalextremists:evidencefromTwitterlanguageanalysisofalt-rightandAntifa. EPJDataSci.8,17(2019).https://doi.org/10.1140/epjds/s13688-019-0193-9DownloadcitationReceived:25December2017Accepted:25April2019Published:14May2019DOI:https://doi.org/10.1140/epjds/s13688-019-0193-9SharethisarticleAnyoneyousharethefollowinglinkwithwillbeabletoreadthiscontent:GetshareablelinkSorry,ashareablelinkisnotcurrentlyavailableforthisarticle.Copytoclipboard ProvidedbytheSpringerNatureSharedItcontent-sharinginitiative KeywordsPoliticalextremismPsychologicalprofileLanguageanalysisLIWCMoralfoundationstheory DownloadPDF Advertisement