carolyn m. parkinson1* 2 3 - biorxiv · carolyn m. parkinson1*, adam m. kleinbaum2, and thalia...

SpontaneousNeuralEncodingofSocialNetworkPosition

CarolynM.Parkinson1*,AdamM.Kleinbaum2,andThaliaWheatley3

1DepartmentofPsychology,UniversityofCalifornia,LosAngeles,1285FranzHall,Box

951563,LosAngeles,California,USA90095

2TuckSchoolofBusiness,DartmouthCollege,100TuckHall,Hanover,NewHampshire,USA

03755

3DepartmentofPsychologicalandBrainSciences,DartmouthCollege,6207MooreHall,

Hanover,NewHampshire,USA03755

*Correspondingauthor:

CarolynM.Parkinson,Ph.D. Telephone:(310)206-8177

DepartmentofPsychology Fax:(310)206-5895

UniversityofCalifornia,LosAngeles E-mail:[email protected]

1285FranzHall,Box951563

LosAngeles,CA90095

Keywords:socialnetworkanalysis,socialperception,functionalMRI,multi-voxelpattern

analysis

.CC-BY-NC-ND 4.0 International licenseacertified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under

The copyright holder for this preprint (which was notthis version posted January 9, 2017. ; https://doi.org/10.1101/098988doi: bioRxiv preprint

https://doi.org/10.1101/098988

http://creativecommons.org/licenses/by-nc-nd/4.0/

1

SpontaneousNeuralEncodingofSocialNetworkPosition

Humansformcomplexsocialnetworksthatincludenumerousnon‐reproductivebondswithnon‐kin.Navigatingthesenetworkspresentsaconsiderablecognitivechallengethoughttohavecomprisedadrivingforceinhumanbrainevolution.Yet,littleisknownabouthowandtowhatextentthehumanbrainencodesthestructureofthesocialnetworksinwhichitisembedded. By combining social network analysis and multi‐voxel pattern analysis offunctional magnetic resonance imaging (fMRI) data, we show that social networkinformation about direct relationships, bonds between third parties, and aspects of thebroadernetworktopologyisaccuratelyperceivedandautomaticallyactivateduponseeingafamiliarother.



https://doi.org/10.1101/098988


2

Unlikemanyotherspeciesthatenactsocialbehaviorinlooseaggregations(e.g.,

swarms,herds),humansformgroupscomprisedofmanylong-term,intense,non-

reproductivebondswithnon-kin1.Thecognitivedemandsofnavigatingsuchgroupsare

thoughttohavecomprisedadrivingforceinhumanbrainevolution2.Yet,littleisknown

abouthowandtowhatextentthehumanbrainencodesthestructureofthesocial

networksinwhichitisembedded.Here,wecharacterizedthesocialnetworkofan

academiccohort(N=277),asubsetofwhom(N=21)completedafunctionalmagnetic

resonanceimaging(fMRI)studyinvolvingviewingvideosofindividualswhovariedin

termsof“degreesofseparation”fromthemselves(socialdistance),theextenttowhich

theyarewell-connectedtowell-connectedothers(eigenvectorcentrality–EC),andthe

extenttowhichtheyconnectotherwiseunconnectedindividuals(brokerage).

Understandingtheseaspectsofothers’socialnetworkpositionsrequirestrackingnotonly

directrelationships,butalsobondsbetweenthirdpartiesandthebroadernetwork

topology.Pairingnetworkdatawithmulti-voxelpatternanalysis,weshowthatsocial

networkpositioninformationisbothaccuratelyperceivedandspontaneouslyactivated

uponencounteringfamiliarindividuals.Thesefindingselucidatehowthehumanbrain

encodesthestructureofitssocialworld,andunderscoretheimportanceofintegratingan

understandingofsocialnetworksintothestudyofsocialperception.

Relationshipsareintrinsictohumanbehavior.Everydayinteractionsareshapednot

onlybyourownrelationships,butalsobyknowledgeofbondsbetweenthirdpartiesand

thebroadersocialnetworksinwhichweareembedded.Well-connectedindividualscan

effectivelythreatenorbolsterone’sreputation3,thosewhobridgeotherwisedisparate

groupscanefficientlyseekandspreadinformation4,andknowledgeofmutualties



https://doi.org/10.1101/098988


3

influencesinformation-sharingandtrust5.Humansocialintelligencerests,inpart,ona

calculusthatinheresinanunderstandingofsocialnetworkstructure.

Isknowledgeaboutothers’socialnetworkpositionsactivatedonlywhenexplicit

goalsrequireit,orspontaneously,wheneverweencounterfamiliarindividuals?Itmaybe

efficienttoprocesssuchinformationonlywhenourgoalsrequireit(e.g.,determininghow

toobtaininformation;forecastingtherepercussionsofasocialmisstep).Alternatively,it

maybebeneficialtoactivatesuchknowledgespontaneouslywhenencounteringothers,

giventheimportanceofsocialnetworkpositiontomanyaspectsofbehaviorandto

impressionsofstatusandcompetence3,6.Humansspontaneouslyregisteragreatdealof

informationwhenperceivingotherpeople(e.g.,intentions,traits,emotions7,8),presumably

tofacilitateappropriate,beneficialsocialinteractions.Thus,thebrainmayrunseveral

social“daemons”–efficient,backgroundprocessesthatspontaneouslyregisterinformation

usefulforpredictingthesocialrepercussionsofpotentialactions,and,morebroadly,to

informcognitionandbehavior.

Totestwhetherthebrainspontaneouslyencodesthesocialnetworkpositionsof

familiarothers,wescanned(fMRI)membersofarealworldsocialnetwork(seeFig.1;

Methods)astheyviewedbriefvideosof12classmates(Fig.2).Theonlytaskwasto

indicatewhenthesamevideowaspresentedtwiceinarow(seeMethods)inorderto

ensureattentionwithoutanyinstructionstoretrievesocialrelationshipknowledge,or

personknowledgemoregenerally.Therefore,weconsideranysocialnetworkposition

informationencodedwhileparticipantsperformthistasktoberetrievedspontaneously

(i.e.,withoutinstruction).



https://doi.org/10.1101/098988


4

Eachclassmateineachparticipant’sstimulussetwascharacterizedaccordingto

threemetricsderivedfromthesocialnetworkdata:geodesicsocialdistancefromthe

participant;EC;andconstraint,aninversemeasureofbrokerage.Geodesicsocialdistance

referstotheminimumnumberofintermediarysocialtiesrequiredtoconnecttwo

individuals.ECisaprestige-basedcentralitymetricthatconsidersnotonlyhowmany

connectionsagivenindividualhas,butalsothecentralitiescharacterizingeachcontact9.

HighECimpliesthatanindividualiswell-connectedtowell-connectedothers;lowEC

impliesthatanindividualhasfewfriendswhotendtobeunpopular.Prestige-based

centralitymetricsareparticularlyusefulforcharacterizingsocialstatus,giventhatbeing

namedasafriendbyapopularindividualshouldincreaseone’ssociometricstatus(i.e.,the

extenttowhichsomeoneislikedbyothers)morethanbeingnamedbysomeoneless

popular9.Individualswhoconnectotherswhowouldnototherwisebeconnectedoccupy

networkpositionslowinconstraint,andhavethecapacitytoserveas“brokers”of

resources(e.g.,information)inthenetwork.Becauseofthestructureoftheirlocalsocial

ties,brokerscancoordinatebehaviorandtranslateinformationacrossstructuralholesin

networks4.

Toprobeforthespontaneousencodingofsocialnetworkpositioninformation,we

usedrepresentationalsimilarityanalysis(RSA),whichdistillsfMRIresponsepatternsinto

representationaldissimilaritymatrices(RDMs)thatindicatethedegreetowhichparticular

brainregionsdistinguishbetweensetsofstimuliormentalstates10.BecauseRDMsare

abstractedawayfromthespatiallayoutofneuroimagingdata(i.e.,theyareindexedby

experimentalcondition;Fig.3),RSAaffordstheevaluationofthedegreetowhichsimilarity

structurescontainedinparticularbrainregionsreflectthoseofdataacquiredusingother



https://doi.org/10.1101/098988


5

modalitiesofmeasurementorcomputationalmodels10(here,thesocialnetworkdata).

Specifically,inthecurrentstudy,weusedagenerallinearmodel(GLM)decomposition

searchlightapproach11.NeuralRDMswereiterativelyextractedwithin9-mmradius

spherescenteredateachpointineachparticipant’sbrain.Withineachparticipant,each

localneuralRDMwasmodeledasaweightedcombinationofRDMsbasedonpropertiesof

thesocialnetworkpositionsoftheindividualsinthatparticipant’sstimulusset(Fig.3).

Usingthistechnique,participants’brainsweremappedintermsofthedegreetowhichthe

representationalcontentoflocalneuralresponsestofamiliarotherscouldbeexplainedby

thoseindividuals’positionsintheirsocialnetwork,andintermsofwhereinformation

aboutspecificsocialnetworkpositioncharacteristicwascarriedreliablyacross

participants(Fig.4).

Wehypothesizedthatgeodesicsocialdistancewouldbespontaneouslyencoded,

giventheimportanceofthisinformationfordeterminingself-relevance.One’simmediate

socialtiesareobviouslymostself-relevant.Giventheimportanceofreputation

managementforhumanbehavior12,individualstwo“degreesaway”mayberelatively

importanttoidentifyandmonitor:Negativeinteractionswithsuchindividualscould

damagerelationshipswithone’sdirectconnections.Similarly,sharingmutualfriendsmay

enhancetrust,giventhepotentialreputationcostsofbadbehavior5.Associaldistance

betweenpeopleincreases,theirrelevancetoeachotherdecreases.Wepredictedthatsocial

distance-relatedinformationwouldbecarriedinthelateralsuperiortemporalcortex

(STC)andinferiorparietallobule(IPL),aswellasthemedialprefrontalcortex(MPFC),

givenpastresearchimplicatingtheseregionsinencodingsocialdistance13andself-

relevancemoregenerally14.



https://doi.org/10.1101/098988


6

Socialdistancewasreliablysignaledinalargeclustercenteredinthelateral

posteriorSTCandextendinginferiorlythroughoutposteriorlateraltemporalcortex(LTC),

andsuperiorlytotheanterioraspectoftheIPL(seeFig.4;TableS1).Pastresearch

demonstratedthatmulti-voxelresponsepatternsinthisregionencodeegocentricspatial

andabstract(e.g.,social)distanceswhenexplicitlyjudging14ormentallynavigating15such

distances14;thecurrentfindingssuggestthatthisregionalsoencodesegocentricdistances

spontaneously(i.e.,intheabsenceofanyexplicitdistancetask).Thus,whenencounteringa

familiarindividual,knowledgeofagent-to-agentrelationshipsappeartobespontaneously

retrieved,suchthatrepresentationsofotherpeopleinthisregionareorganizedintermsof

whethersomeoneisafriend,afriend-of-a-friend,orfartherremovedfromoneselfinsocial

ties.Ithasbeensuggestedthatsomeregionswithinposteriorparietalcortex,suchasthe

anteriorIPL,whichhavewell-establishedrolesinrepresentingandnavigatingphysical

space,analogouslyrepresentmoreabstractrelationships(e.g.,socialtiesbetween

agents)16,17.Thecurrentresultssuggestthatwhenencounteringfamiliarindividuals,

humansmayspontaneouslyretrieveknowledgeofwheretheyarelocated,relativeto

oneself,inamentalmapof“socialspace”.

AlthoughtheLTCandIPLregionsthatcarriedinformationaboutsocialdistance

herehavepreviouslyimplicatedinencodingsocialdistance13,14,someregionspreviously

implicatedinsignalingsocialdistancewerenotimplicatedinthecurrentstudy.For

instance,previousresearchhasimplicatedMPFCindistinguishingfriendsfromstrangers13,

andarecentstudyimplicatedthehippocampusandposteriorcingulatecortex(PCC)in

trackingsocialdistancesbetweenparticipantsandcharactersinaninteractivegame18.

Differencesbetweenthecurrentresultsandthoseobservedinpreviousinvestigations



https://doi.org/10.1101/098988


7

likelyreflectdifferencesindataanalyticapproachesandinhowsocialdistancehasbeen

operationalized.Inthecurrentstudy,participantsonlysawpersonallyfamiliarindividuals,

andsocialdistancewasoperationalizedintermsofgeodesicdistanceintheirreal-world

socialnetwork.Inpreviousneuroimagingstudies,theterm“socialdistance”hasbeen

operationalizedinwidelyvaryingways,suchasthepresenceofsocialties13,thestrengthof

socialties14,anddistancefromoneselfinatwo-dimensional(affiliationxstatus)

representationalspace18.Giventhatthesevariableslikelyhavedifferentialconsequences

forsocialcognitionandbehavior,itisnotsurprisingthattheyareencodedbyatleast

partiallydistinctneuralsubstrates.

Whereassocialdistanceisinherentlyrelativetotheperceiver,otheraspectsof

familiarothers’socialnetworkpositions,suchasthedegreetowhichone“bridges”

differentareasofthenetworkandthenumberoffriendssomeonehas,areincreasingly

thoughttobelargelystable,possiblyheritable,dispositionaltendenciesthatshapesocial

behavior19,20.Therefore,wehypothesizedthatECandconstraintwouldbeencodedin

brainregionsinvolvedinencodingothers’traitsandbehavioraltendenciesmoregenerally,

suchastheMPFC,whichiswidelyimplicatedininferringandencodingperson

knowledge21andinintegratingknowledgeofpersonalitytraitsinordertosignalindividual

identity22.

InformationaboutECwasreliablycarriedinbrainregionsthatencodeindividual

identitywhenimaginingothers’actions22(i.e.,MPFC)andviewingfaces23,24(e.g.,temporal

pole;fusiformgyrus;seeFig.4andTableS2),suggestingthatsociometricstatusmay

compriseadimensionofmeaningfororganizingmentalrepresentationsofothers.ECwas

alsoencodedinmedialparietalcortex(precuneus,PCC),aregionpreviouslybeenshownto



https://doi.org/10.1101/098988


8

encodeextraversion22,whichismodestlycorrelatedwithEC25,suggestingthatthisregion

mayencodedispositionaltendenciescommontobothextraversionandEC.Inaddition,

recentworkhasalsoshownthatthemedialparietalcortex,aswellasotherregions

involvedininferringothers’mentalstates,intentions,andtraits(e.g.,MPFC;

temporoparietaljunction),respondspreferentiallytowell-likedindividuals,whichis

thoughttoreflectperceiversbeingpreferentiallymotivatedtounderstandtheinternal

statesofpopularothers26.Thecurrentfindingsareconsistentwiththenotionthatbrain

regionsthatrepresentothers’internalstatesandbehavioraltendencies(e.g.,PCC,MPFC)

tracksociometricstatus,andsuggestthatlikeotherfacetsofsocialstatus(e.g.,

dominance27,prestige28),ECmaymodulateattentiontotheinternalstatesofothers.Future

behavioralstudiesshoulddirectlytesttheimpactofEConsocialattention.

InformationaboutECwasalsoreliablycarriedinunexpectedregions,suchas

extrastriatevisualcortex(EVC).Thisresultisunlikelytobeduetolow-levelvisual

characteristicsofstimuli,aseachparticipanthadauniquestimulusset,andbecausevideos

correspondingtoeachindividualineachstimulussetwerehorizontallymirroredonhalfof

trials(seeMethods).However,thisfindingmaynonethelessreflecttheeffectsofsocial

statusintermsofsocialtiesonvisualattention.Peopletendtopreferentiallyorienttoward

high-statusindividualsandtothelocioftheirattention,presumablytoobtainbehaviorally

relevantinformationaboutoursurroundings28–30.GiventhatECisreliablysignaledinEVC

responsepatterns,futureresearchshouldtestifvisualattentionisalsopreferentially

allocatedtocentralactorsinone’ssocialnetwork.

EC-basedRDMswerealsosignificantlyrelatedtoneuralRDMsinbrainareas

previouslyimplicatedinevaluatingsocialstatusintermsofdominance,prestige,and



https://doi.org/10.1101/098988


9

morality,suchastheventralMPFCandventrolateralprefrontalcortex(VLPFC)31–33.The

involvementoftheventralMPFCinsocialstatusencodinghasbeensuggestedtoreflecta

moregeneralroleinassessingthevalueofstimuli32,whereastheVLPFChasbeen

suggestedtoencodesocialstatusinordertoappropriatelymodulatebehavioral

responding,whichisthoughttobeaprimaryfunctionofstatuscues31.Wesuggestthat

theseregionslikelyencodeECforsimilarreasons,ashighECindividualshavehigh

behavioralrelevanceandvalueassocialpartners.Forexample,individualsconnectedto

well-connectedothersmaybeprotectedfrommistreatmentbecausetheyaremorelikely

tobedefendedbyothers,whothemselvesaremorelikelytobedefended.Lessriskis

associatedwithwrongingalowECindividual,giventhatlowECindividualshavelittle

influenceonthespreadofinformationandotherresources3.

Thecurrentresultssuggestthatwhenencounteringafamiliarindividual,the

degreetowhichthatindividualiswell-connectedtowell-connectedothersshapes

processesrelatedtovaluation,behavioralmodulation,attention,andencodingothers’

internalstates,dispositionalcharacteristics,andidentities.Manyofthesefindingsechothe

knowneffectsofotherdimensionsofsocialstatus(e.g.,statusconferredbydominance).

Althoughagreatdealofpastpsychologicalandneuroimagingresearchonsocialstatushas

focusedonphysicaldominance,wenotethatovertphysicalviolenceisrelativelyrarein

contemporaryhumangroups34andthatsocialsupportandreputationmanagementare

centraltoeverydayhumanlife12.Socialpowerinsuchgroupsmayberelativelyless

contingentonindividualstrengthandphysicalaggression,andmoredependentongroup

dynamicsandaffiliativerelationshipmaintenance.Thus,sociometricstatusislikely



https://doi.org/10.1101/098988


10

especiallyrelevanttomodernhumans,andmeritsfurtherattentioninsocialperception

andneuroscienceresearch.

InadditiontosocialdistanceandEC,diverseaspectsofsocialcognitionand

behavior(e.g.,decidinghowtoeffectivelyseekorspreadinformation;trustdecisions)

wouldbenefitfromencodingnetworkconstraint.Lowconstraintindividualscanbroker

theflowofinformationbetweengroups,andthus,exertadisproportionateinfluenceonthe

flowofideasandresources4.Additionally,individualsinrelatively“closed”localnetworks,

characterizedbyhighconstraint,suffergreaterreputationcostsforbadbehavior;

correspondingly,constraintcanfostertrustandcooperation4.Giventhedearthofprevious

researchinvestigatingtheperceptionofconstraint,wemadenospecificpredictionsabout

whichbrainregionswouldbeinvolvedinencodingthisfacetofsocialnetworkposition.

LargeclustersspanningbothrightandleftlateralSTCcarriedinformationabout

constraint(TableS3),asdidasmallerclusterinthesupplementarymotorarea(SMA).

AlthoughthelateralSTCandSMAareimplicatedinbiologicalmotionprocessing35and

actionunderstanding36,respectively,thisfindingwasnotattributabletotheamountof

movementinvideos(seeSI).Aperceiver’sknowledgeofthenetworkconstraintofan

individual,orofassociateddispositions,mayimpacthowthatperceiverattendstothat

individual’smovements.Forexample,becausebrokersmaybeperceivedasexceptionally

charismaticorinteresting(e.g.,becausetheyoftenserveassourcesofnovelinformationor

opportunities4),theymaycommanddifferentialamountsofattentiontotheirexpressions

andgestures.Brokersmayalsodifferintheamountofsocialmeaningcarriedintheirfacial

andbodilymovements(e.g.,fidgetingaimlesslyvs.usingmovementtoexpressoneself

coherently).ThelatterexplanationwouldbeconsistentwithevidencethattheSTS



https://doi.org/10.1101/098988


11

respondstothesocialmeaning,ratherthanamount,ofmovementindynamicdisplays37.

Futurestudiescouldarbitratebetweenthesehypothesesbytestingifstrangersareableto

differentiatebetweenindividualshighandlowinconstraintbasedontheirobserved

movements.Ifso,thiswouldsuggestthatnetworkconstraintisencodedinlateralSTC

becausethisaspectofsocialnetworkpositionisapparentinhowindividualscarry

themselves.Ifnot,thiswouldbeconsistentwiththeinterpretationthatperceivers’

knowledgeofanindividual’snetworkconstraint,orofqualitiesrelatedtothisaspectof

socialnetworkposition,influenceshowperceiversattendtothatindividual’sexpressions,

gestures,andbodilymovements.

Afterscanning,participantswereaskedabouttheirperceptionsofeachSNA-

derivedmetricofinterestforeachindividualintheirstimulusset(seeSI).Thisallowedus

totesttheaccuracyofparticipants’perceptionsofothers’socialnetworkpositions,andto

evaluatehowwellparticipants’perceptionsmatchedthedatausedtoconstructtheir

stimulussets.Post-scanratingsindicatedthatparticipants’explicitperceptionsofthe

socialnetworkpositionsoftheindividualsintheirstimulussetscloselymatchedreality.

Veridicalconstrainthadasignificanteffectonperceivedconstraint(ß=19.44SE=2.01,p<

.0001),andveridicalEChadasignificanteffectonperceivedEC(ß=14.95,SE=0.93,p<

.0001).Further,subjectiveratingsofsocialcloseness(ß=-31.00,SE=1.62,p<.0001),

proportionofsocialtimespenttogether(ß=-22.74,SE=1.84,p<.0001),andfrequencyof

discussions(ß=-33.77,SE=1.89,p<.0001)variedasafunctionofgeodesicnetwork

distance(seeMethodsandFig.5).

Althoughparticipantshadconsciouslyaccessibleknowledgeofthesocialnetwork

positioncharacteristicsstudiedhere(Fig.5),thetaskusedinthefMRIstudy(aone-back



https://doi.org/10.1101/098988


12

memorytask)didnotrequireparticipantstoretrievesocialrelationshipknowledge,and

fromparticipants’perspectives,thesocialnetworkquestionnaireandfMRIstudywere

ostensiblyunrelated.Nevertheless,upto40%ofthevarianceinsimilaritystructuresof

localfMRIresponsestopersonallyfamiliarotherscouldbeexplainedmerelyby

characteristicsofthoseindividuals’positionsintheperceiver’ssocialnetwork(Fig.4b).

Thesefindingsareconsistentwithbehavioralevidencethathumansspontaneously

activateknowledgeaboutotherpeopleuponencounteringtheminordertoinform

cognitionandbehavior7,8,andsuggestthathumansspontaneouslyactivatecomplex

knowledgeaboutotherpeople’spositionsintheirsocialnetworkswhenviewingthem.

Thesefindingsareconsistentwithpsychologists’mountingappreciationforthe

importanceofbothdirectandindirectrelationshipknowledgetoeverydaycognitionand

behavior.Everydayinteractionsareinfluencednotonlybyinformationthatwouldbe

availabletoanyobserver,butalsobypatternsofpersonalandthird-partyrelationships.By

adoptinganinterdisciplinaryapproachcombiningtheoryandmethodsfromneuroscience,

psychology,andSNA,wecanbegintouncoveradeeperunderstandingofhowthehuman

brainnegotiatestheintricaciesofeverydaysociallife.

Methods

Part1:Socialnetworkcharacterization

Participants.ParticipantsinPart1ofthestudywere275first-yearMastersof

BusinessAdministration(MBA)studentsataprivateuniversityintheUnitedStateswho

participatedaspartoftheircourseworkonleadership(91females;184males).Thetotal

classsizewas277students;twostudentsfailedtocompletethequestionnaire(i.e.,



https://doi.org/10.1101/098988


13

responserate=99.3%).Allprocedureswerecompletedinaccordancewiththestandardsof

theDartmouthCommitteefortheProtectionofHumanSubjects.

Socialnetworkcharacterization.Tocharacterizethesocialnetworkofallfirst-year

students,anonlinesocialnetworksurveywasadministered.Participantsfollowedane-

mailedlinktothestudywebsitewheretheyrespondedtoasurveydesignedtoassesstheir

positioninthesocialnetworkoffirst-yearstudentsintheiracademicprogram(seeSI).The

surveyquestionwasadaptedfromBurt38andhasbeenpreviouslyusedinthemodified

formusedhere25,39.Itread,“Considerthepeoplewithwhomyouliketospendyourfree

time.Sinceyouarrivedat[institutionname],whoaretheclassmatesyouhavebeenwith

mostoftenforinformalsocialactivities,suchasgoingouttolunch,dinner,drinks,films,

visitingoneanother’shomes,andsoon?”

Aroster-basednamegeneratorwasusedtoavoidinadequateorbiasedrecall.

Classmates’nameswerelistedinfourcolumns,withonecolumncorrespondingtoeach

sectionofstudentsintheMBAprogram.Nameswerelistedalphabeticallywithinsection.

Participantsindicatedthepresenceofasocialtiewithanindividualbyplacingacheckmark

nexttohisorhername.Participantscouldindicateanynumberofsocialties,andhadno

timelimitforresponding.

SocialnetworkanalysiswasperformedusingtheRpackageigraph40,41.Threesocial

network-derivedmetricswereextractedforeachnode:constraint,ECandgeodesic

distancefromeachclassmate,asdescribedingreaterdetailbelow.

Constraint.Theconstraintofactoriisgivenbythefollowingequation,wherePij

correspondstotheproportionofi'sdirectsocialtiesaccountedforbyhis/hertietoactorj.



https://doi.org/10.1101/098988


14

Theinnersummationapproximatestheindirectconstraintimposedonibyotheractors,q,

whoaresociallyconnectedtobothiandj(i.e.mutualfriendsofiandj):

𝑪𝒐𝒏𝒔𝒕𝒓𝒂𝒊𝒏𝒕𝒊 = (𝑷𝒊𝒋 + 𝑷𝒊𝒒𝒏

𝒒~𝟏𝑷𝒒𝒋)𝟐

𝒏

𝒋~𝟏

Anunweighted,undirectedgraphwasusedtoestimateconstraint;i.e.,thepresence

ofanysocialtie,irrespectiveofitsdirection,wasusedtocomputetheconstraintofeach

node.Constraintisaninversemeasureofnetworkbrokerage.

EC.Agraphconsistingofnodesconnectedbyedgescanbecharacterizedbyan

adjacencymatrixA,populatedbyelementssuchthataij=1ifnodesiandjaredirectly

connected,andaij=0ifthesenodesarenotconnected.TheECofeachnodeisgivenbythe

eigenvectorofAinwhichallelementsarepositive.Therequirementthatallelementsof

theeigenvectormustbepositiveyieldsauniqueeigenvectorsolution(i.e.,that

correspondingtothegreatesteigenvalue).Here,whencomputingEC,thedirectionalityof

thegraphwaspreserved;intheeventofasymmetricrelationships,onlyincoming,rather

thanoutgoing,tieswereusedtocomputeEC.

Socialdistance.Geodesicsocialdistancereferstothesmallestnumberof

intermediarysocialtiesrequiredtoconnecttwoindividualsinanetwork.Individualswho

aparticipantnamedasfriendshaveadistanceofonefromhim/her.Individualswhoma

participant’sfriendsnamedasfriends(butwhowerenotnamedasfriendsbythe

participant)haveadistanceoftwofromtheparticipant.Individualswhowerenamedas

friendsbyclassmatesatadistanceoftwofromtheparticipant(butnotbytheparticipant

orhis/herfriends)haveadistanceofthree,andsoon.



https://doi.org/10.1101/098988


15

Part2:Neuroimagingstudy

Participants.AsubsetofindividualswhohadcompletedPart1participatedina

subsequentneuroimagingexperiment.Participantswereinformedduringclassaboutthe

opportunitytoparticipateinanfMRIstudythatwasostensiblyunrelatedtotheonline

questionnaireinPart1,andthattheywouldreceive$20/hourascompensationandimages

oftheirbrains.Allparticipantswereright-handed,fluentinEnglish,andhadnormalor

corrected-to-normalvision.Participantsprovidedinformedconsentinaccordancewiththe

policiesoftheDartmouthCollegeCommitteefortheProtectionofHumanSubjects.

Twenty-fourparticipants(12females)completedthefMRIstudy.Thesamplesizewas

chosenbasedonpreviousfMRIstudiesusingsimilarparadigmsandRSAmethods11,42.One

participantwasexcludedduetoimageartifact,andtwowereexcludedbecausetheyscored

lessthan65%correctontheone-backmemorytaskusedinthescanner(thisthreshold

wasbasedonwhathasbeenusedpreviouslyinsimilarstudies43).Consequently,we

analyzeddatafrom21participants(10females,aged25-33,M=27.95,SD=2.16).Asa

within-subjectsdesigninvolvingnogroupallocationwasused,blindinginvestigatorsto

between-subjectsconditionsandrandomassignmentofparticipantstoconditionswere

notapplicable.

Imageacquisition.ParticipantswerescannedattheDartmouthBrainImaging

Centerusinga3TPhilipsAchievaInterascannerwitha32-channelheadcoil.Anecho-

planarsequence(35msTE;2000msTR;3.0mmx3.0x3.0mmresolution;80x80matrix

size;240x240mmFOV;35interleavedtransversesliceswithnogap;3.0mmslice

thickness)wasusedtoacquirefunctionalimages.Functionalrunsconsistedof180

dynamicscans,foratotalacquisitiontimeof360sperrun.Ahigh-resolutionT1-weighted



https://doi.org/10.1101/098988


16

anatomicalscanwasacquiredforeachparticipant(8.2sTR;3.7msTE;240x187FOV;

0.938mmx0.938mmx1.0mmresolution)attheendofthescanningsession.Foam

paddingwasplacedaroundsubjects’headstominimizemotion.

Stimuli.Eachparticipant’scustomizedstimulussetconsistedofshortvideosoffour

individualsateachofthreegeodesicdistances(i.e.,one,two,andthree)fromthe

participantinthesocialnetworkoffirst-yearMBAstudents.ThetwohighestandlowestEC

individualsateachsocialdistancewereincludedinthestimulusset(Fig.2).

Thevideosusedasstimuliconsistedofindividualsintroducingthemselves(e.g.,“Hi

mynameis[firstname],andyoucancallme[firstname/nickname]”).Avideoofthiskind

wasmadeinvolvingeachstudentatthebeginningoftheacademicyearasaresourcefor

otherstudentsandfaculty.Videosweretruncatedto2s,beginningwhenthesubjectbegan

tosaytheword,“Hi,”andwerepresentedwithoutsound.PriortoenteringthefMRI

scanner,participantswereshowneachvideowithsoundtofamiliarizethemselveswiththe

stimuli.

fMRIparadigm.ThefMRIstudyconsistedof10runsandfollowedarapidevent-

relateddesignwithaninter-trialintervalconsistingof4soffixation(Fig.2c).Fournull

events,eachconsistingofanadditional2soffixation,wererandomlyinsertedintoeach

run.Ineachrun,fourrepetitionsof14eventcategories(12identities;1nullevent;1catch

trial)werepseudo-randomizedsuchthattherewerenoconsecutiverepeatsofthesame

category.Horizontalmirroringwasrandomlyappliedtohalfthepresentationsofeach

stimuluswithineachruntoreducesimilaritieswithinidentitiesduetolocallow-level

visualfeatures.Catchtrialsinvolvedseeingthesamestimulusatthesamemirroringlevel

astheimmediatelypreviousstimulus(ortwotrialsbackifacatchtrialfollowedanull



https://doi.org/10.1101/098988


17

event).Participantswereinstructedtopressabuttonwhenanidenticalvideowas

presentedtwiceinarow(i.e.,forcatchtrials).

Post-scanquestionnaire.Afterscanning,participantswereaskedabouttheir

subjectiveperceptionsofeachsocialnetworkmetricofinterestforeachindividualintheir

stimulusset,aswellasquestionsassessingtiestrength(seeSI).Becausetheconstraint

questionaskedaboutbrokerage(i.e.,whichindividualswerelowinnetworkconstraint),

responsestothisitemweremultipliedby-1.Toalleviateskewinthenetworkdata,

eigenvectorcentralitiesandnetworkconstraintvalueswerelog-transformedpriorto

analysis.

Thecorrespondencebetweenparticipants’post-scanratingsandthesocial

networkpositioncharacteristicsoftheindividualsintheirstimulussetswasassessed

usinglinearmixedmodelsusingtheR40packagelme444.Foreachofthefivequestions(see

SI),amodelwasconstructedwithparticipants’ratingsasthedependentmeasureandthe

relevantsocialnetworkpositioncharacteristicasafixedeffect,aswellasrandom

interceptsandslopesforeachparticipant.Totestthesignificanceoftherelationship

betweenparticipants’ratingsandsocialnetworkdata,p-valueswerecomputedusing

Satterthwaite’sapproximationfordegreesoffreedom45asimplementedinlmerTest46.

fMRIdatapreprocessing.ForfMRIdataanalysis,datawerepreprocessedand

averagevoxel-wisehemodynamicresponsestoeachidentitywereestimatedusingAFNI47.

Pre-processingstepsincludedapplyingAFNI’s3dDespikefunctiontoremovetransient,

extremevaluesinthesignalnotattributabletobiologicalphenomena,slice-timing

correctiontocorrectforinterleavedsliceacquisitionorder,alignmentofthelastvolumeof

thefinalruntothehigh-resolutionanatomicalscan,registrationofallfunctionalvolumes



https://doi.org/10.1101/098988


18

totheanatomical-alignedfinalfunctionalvolumeusinga6-parameter3-Dmotion

correctionalgorithm,spatialsmoothingusinga4-mmfull-widthathalf-maximumGaussian

kernel,andscalingeachvoxeltimeseriestohaveameanamplitudeof100.Priorto

regression,consecutivevolumeswheretheEuclideannormofthederivativesofthemotion

parametersexceeded0.3mmwereexcludedfromfurtheranalysis,aswerevolumesin

whichmorethan10%ofbrainvoxelswereidentifiedasoutliersbytheAFNIprogram

3dToutcount.

ParameterestimateswereextractedforeachvoxelusingaGLMthatconsistedof

gamma-variateconvolvedregressorsforeachof13predictors(oneforeachofthe12

identitiesintheparticipant’sstimulusset;oneforcatchtrials),aswellas12regressorsfor

eachofthesixdemeanedmotionparametersextractedduringvolumeregistrationand

theirderivatives,andthreeregressorsforlinear,quadratic,andcubicsignaldriftswithin

eachrun.Thisprocedureremovedvariancecausedbyregressorsofnointerest,and

resultedinanestimateoftheresponseofeachvoxeltoeachtrialtype.

GLMdecompositionsearchlight.UsingPyMVPA48andSciPy49,aGLMdecomposition

searchlight11wasperformedwithineachparticipant’sdata.Asphere(radius=3voxels)

wasmovedthroughouteachparticipant’sbrain.Ateachpointinthebrain,thelocal

distributedpatternsofneuralresponsestoeachpersoninthestimulussetwereextracted

withinaspherecenteredonthatpoint,andthepairwisecorrelationdistancesbetween

themwerecalculatedtoconstructalocalneuralRDM(Fig.3a-c),whichwasdecomposed

intoaweightedcombinationofpredictorRDMsusingordinaryleastsquares(OLS)

regression(Fig.3d).TherewerethreepredictorRDMs,onecorrespondingtoeachsocial

networkpositionmetricofinterest.PredictorRDMswereconstructedbytakingthe



https://doi.org/10.1101/098988


19

Euclideandistance(i.e.,theabsolutevalueofthenumericaldifference)betweenthe

relevantsocialnetworkpositionmetricsforeachpossiblepairofidentitieswithineach

participant’sstimulusset.EachpredictorRDMforeachparticipantwasthenz-scored.Next,

foreachRDM(e.g.,theEC-basedRDMforagivenparticipant),thevarianceaccountedfor

bytheremainingtwopredictorRDMs(e.g.,thesocialdistanceandconstraint-basedRDMs

forthatparticipant)wasremovedusingOLSregression.Thus,theresultantpredictor

RDMsweremadeorthogonaltooneanotherpriortoperformingtheGLMdecomposition

searchlight.

Ateachsearchlightcenter(i.e.,ateachvoxel),theGLMdecompositionprocedure

yieldedaβvaluecorrespondingtoeachsocialnetworkderivedmetricofinterest,aswell

asanR2valuecorrespondingtohowmuchthevarianceinthesimilaritystructureoflocal

neuralresponsepatternscouldbeexplainedbythesocialnetworkpositionsofthe

individualscomprisingagivenparticipant’sstimulusset.

Groupanalysis.Eachsubject’smapsofregressioncoefficientsandR2valueswere

transformedtostandard(Talairach50)spaceusingAFNI:Anatomicalscanswerelinearly

alignedtotheTalairach50templateusingthe@auto_tlrcalgorithminAFNI,andthesame

transformwasusedtoaligneachparticipant’ssearchlightresultstostandardspaceprior

togroupanalysis.Toidentifyareasthatreliablycontainedinformationabouteachspecific

aspectofsocialnetworkpositionacrossparticipants,theregressioncoefficientsforeach

socialnetworkposition-derivedRDMweretestedagainst0acrossparticipantsusingone-

tailedone-samplet-tests.Morespecifically,FSL’srandomise51,52programwasusedto

performpermutationtestsandtogenerateanulldistributionofclustermassesformultiple

comparisonscorrection(cluster-formingthreshold:p<.01,two-tailed;5,000



https://doi.org/10.1101/098988


20

permutations;10-mmvariancesmoothing).Allreportedresultshavebeenthresholdedata

family-wiseerrorrateof5%.

Dataandcodeavailability:Thedatathatsupportthefindingsofthisstudyareavailable

fromthecorrespondingauthoruponrequest.Thecodeusedfortheanalysesalsois

availableuponrequest.



https://doi.org/10.1101/098988


21

Acknowledgements

ThisworkwassupportedbyagraduatefellowshipfromtheNeukomInstitutefor

ComputationalScienceandaDartmouthGraduateAlumniResearchAward.Theauthors

wishtothankWillHaslettforassistancewiththeopticalflowanalysis.

AuthorContributions

CP,AMKandTWconceivedofanddesignedthestudy.CPandAMKcollectedthedata.CP

analyzedthedata.CP,AMK,andTWwrotethepaper.



https://doi.org/10.1101/098988


22

References

1. Shultz,S.&Dunbar,R.I.M.Bondednessandsociality.Behaviour147,775–803(2010).

2. Dunbar,R.I.M.&Shultz,S.Evolutioninthesocialbrain.Science(80-.).317,1344–7(2007).

3. Ellwardt,L.,Labianca,G.&Wittek,R.Whoaretheobjectsofpositiveandnegativegossipatwork?Soc.Networks34,193–205(2012).

4. Burt,R.S.,Kilduff,M.&Tasselli,S.Socialnetworkanalysis:Foundationsandfrontiersonadvantage.Annu.Rev.Psychol.64,527–47(2013).

5. Burt,R.S.&Knez,M.Kindsofthird-partyeffectsontrust.Ration.Soc.7,255–292(1995).

6. Brent,L.J.N.Friendsoffriends:Areindirectconnectionsinsocialnetworksimportanttoanimalbehaviour?Anim.Behav.103,211–222(2015).

7. Uleman,J.S.,Newman,L.S.&Moskowitz,G.B.inAdvancesinExperimentalSocialPsychology(ed.Zanna,P.)28,211–279(AcademicPress,1996).

8. Todorov,A.,Gobbini,M.I.,Evans,K.K.&Haxby,J.V.Spontaneousretrievalofaffectivepersonknowledgeinfaceperception.Neuropsychologia45,163–73(2007).

9. Bonacich,P.Powerandcentrality:Afamilyofmeasures.Am.J.Sociol.92,1170(1987).

10. Kriegeskorte,N.,Mur,M.&Bandettini,P.Representationalsimilarityanalysis:Connectingthebranchesofsystemsneuroscience.Front.Syst.Neurosci.2,4(2008).

11. Chikazoe,J.,Lee,D.H.,Kriegeskorte,N.&Anderson,A.K.Populationcodingofaffectacrossstimuli,modalitiesandindividuals.Nat.Neurosci.17,1114–1122(2014).

12. Tennie,C.,Frith,U.&Frith,C.D.Reputationmanagementintheageoftheworld-wideweb.TrendsCogn.Sci.14,482–8(2010).

13. Krienen,F.M.,Tu,P.-C.&Buckner,R.L.Clanmentality:Evidencethatthemedialprefrontalcortexrespondstocloseothers.J.Neurosci.30,(2010).

14. Parkinson,C.,Liu,S.&Wheatley,T.Acommoncorticalmetricforspatial,temporal,andsocialdistance.J.Neurosci.34,1979–87(2014).

15. Gauthier,B.&vanWassenhove,V.Timeisnotspace:Corecomputationsanddomain-specificnetworksformentaltravels.J.Neurosci.36,(2016).

16. Yamazaki,Y.,Hashimoto,T.&Iriki,A.Theposteriorparietalcortexandnon-spatialcognition.F1000Biol.Rep.1,74(2009).

17. Parkinson,C.&Wheatley,T.Oldcortex,newcontexts:Re-purposingspatialperceptionforsocialcognition.Front.Hum.Neurosci.7,645(2013).

18. Tavares,R.M.etal.Amapforsocialnavigationinthehumanbrain.Neuron87,231–243(2015).

19. Fowler,J.H.,Dawes,C.T.&Christakis,N.A.Modelofgeneticvariationinhumansocialnetworks.Proc.Natl.Acad.Sci.U.S.A.106,1720–4(2009).

20. Burt,R.S.Network-relatedpersonalityandtheagencyquestion:Multiroleevidencefromavirtualworld.Am.J.Sociol.118,543–591(2012).

21. Wagner,D.D.,Haxby,J.V.&Heatherton,T.F.Therepresentationofselfandpersonknowledgeinthemedialprefrontalcortex.WileyInterdiscip.Rev.Cogn.Sci.3,451–470(2012).

22. Hassabis,D.etal.Imagineallthepeople:Howthebraincreatesandusespersonality



https://doi.org/10.1101/098988


23

modelstopredictbehavior.Cereb.Cortex24,1979–1987(2014).23. Kriegeskorte,N.,Formisano,E.,Sorger,B.&Goebel,R.Individualfaceselicitdistinct

responsepatternsinhumananteriortemporalcortex.Proc.Natl.Acad.Sci.U.S.A.104,20600–20605(2007).

24. Nestor,A.,Plaut,D.C.&Behrmann,M.Unravelingthedistributedneuralcodeoffacialidentitythroughspatiotemporalpatternanalysis.Proc.Natl.Acad.Sci.U.S.A.108,9998–10003(2011).

25. Feiler,D.C.&Kleinbaum,A.M.Popularity,similarity,andthenetworkextraversionbias.Psychol.Sci.26,593–603(2015).

26. Zerubavel,N.,Bearman,P.S.,Weber,J.&Ochsner,K.N.Neuralmechanismstrackingpopularityinreal-worldsocialnetworks.Proc.Natl.Acad.Sci.112,201511477(2015).

27. Jones,B.C.etal.Facialcuesofdominancemodulatetheshort-termgaze-cuingeffectinhumanobservers.Proc.R.Soc.bBiol.Sci.277,617–24(2010).

28. Dalmaso,M.,Pavan,G.,Castelli,L.&Galfano,G.Socialstatusgatessocialattentioninhumans.Biol.Lett.8,450–2(2012).

29. Klein,J.T.,Shepherd,S.V&Platt,M.L.Socialattentionandthebrain.Curr.Biol.19,R958-62(2009).

30. Shepherd,S.V,Deaner,R.O.&Platt,M.L.Socialstatusgatessocialattentioninmonkeys.Curr.Biol.16,R119-20(2006).

31. Marsh,A.A.,Blair,K.S.,Jones,M.M.,Soliman,N.&Blair,R.J.R.Dominanceandsubmission:Theventrolateralprefrontalcortexandresponsestostatuscues.J.Cogn.Neurosci.21,713–724(2009).

32. Cloutier,J.&Gyurovski,I.Ventralmedialprefrontalcortexandpersonevaluation:Formingimpressionsofothersvaryinginfinancialandmoralstatus.Neuroimage100,535–543(2014).

33. Karafin,M.S.,Tranel,D.&Adolphs,R.Dominanceattributionsfollowingdamagetotheventromedialprefrontalcortex.J.Cogn.Neurosci.16,1796–1804(2004).

34. Pinker,S.Declineofviolence:Tamingthedevilwithinus.Nature478,309–311(2011).

35. Grossman,E.D.,Battelli,L.&Pascual-Leone,A.RepetitiveTMSoverposteriorSTSdisruptsperceptionofbiologicalmotion.VisionRes.45,2847–2853(2005).

36. Mukamel,R.,Ekstrom,A.D.,Kaplan,J.,Iacoboni,M.&Fried,I.Single-neuronresponsesinhumansduringexecutionandobservationofactions.Curr.Biol.20,750–6(2010).

37. Wheatley,T.,Milleville,S.C.&Martin,A.Understandinganimateagents:Distinctrolesforthesocialnetworkandmirrorsystem.Psychol.Sci.18,469–474(2007).

38. Burt,R.S.StructuralHoles:TheSocialStructureofCompetition.(HarvardUniversityPress,1992).

39. Kleinbaum,A.M.,Jordan,A.H.&Audia,P.G.Analtercentricperspectiveontheoriginsofbrokerageinsocialnetworks:Howperceivedempathymoderatestheself-monitoringeffect.Organ.Sci.26,1226–1242(2015).

40. RCoreDevelopmentTeam.R:ALanguageandEnvironmentforStatisticalComputing.RFoundationforStatisticalComputing.(2012).

41. Csardi,G.&Nepusz,T.Theigraphsoftwarepackageforcomplexnetworkresearch.InterJournalComplexSyst.1695(2006).



https://doi.org/10.1101/098988


24

42. Connolly,A.C.etal.Therepresentationofbiologicalclassesinthehumanbrain.J.Neurosci.32,2608–2618(2012).

43. Said,C.P.,Moore,C.D.,Engell,A.D.&Haxby,J.V.Distributedrepresentationsofdynamicfacialexpressionsinthesuperiortemporalsulcus.J.Vis.10,1–12(2010).

44. Bates,D.,Mächler,M.,Bolker,B.&Walker,S.Fittinglinearmixed-effectsmodelsusinglme4.J.Stat.Softw.67,1–48(2015).

45. Satterthhwaite,F.E.Anapproximatedistributionofestimatesofvariancecomponents.Biometrics2,110–4(1946).

46. Kuznetsova,A.,Brockhoff,P.B.&Christensen,R.H.B.lmerTest:Testsinlinearmixedeffectsmodels.Rpackageversion2.0-33(2016).

47. Cox,R.W.AFNI:Softwareforanalysisandvisualizationoffunctionalmagneticresonanceneuroimages.Comput.Biomed.Res.29,162–73(1996).

48. Hanke,M.etal.PyMVPA:Aunifyingapproachtotheanalysisofneuroscientificdata.Front.Neuroinform.3,3(2009).

49. Oliphant,T.E.SciPy:OpensourcescientifictoolsforPython.Comput.Sci.Eng.9,10–20(2007).

50. Talairach,J.&Tournoux,P.Co-PlanarStereotaxisAtlasoftheHumanBrain.(ThiemeMedicalPublishers,1988).

51. Winkler,A.M.,Ridgway,G.R.,Webster,M.A.,Smith,S.M.&Nichols,T.E.Permutationinferenceforthegenerallinearmodel.Neuroimage92,381–397(2014).

52. Jenkinson,M.,Beckmann,C.F.,Behrens,T.E.J.,Woolrich,M.W.&Smith,S.M.FSL.NeuroImage62,782–790(2012).



https://doi.org/10.1101/098988


25

Figure1.Socialnetworkcharacterization.Thesocialnetworkofafirst-yearcohortofMBA

studentswasreconstructedbasedonresponsestoonlinequestionnairesadministeredto

allmembersof the class (N=275;99.3%response rate).Nodes indicate students; lines

indicate reported social ties between them. For ease of visualization, only mutually

reportedsocialtiesareillustrated.AsubsetofthesestudentsparticipatedinanfMRIstudy.

Orangenodes indicate fMRI studyparticipants; graynodesdenoteothermembersof the

graduateprogram.Nodesizeisproportionaltoeigenvectorcentrality.



https://doi.org/10.1101/098988


26

Figure2.Stimulussetconstructionandparadigmforneuroimagingstudy.(A)Thegeodesic

distancebetweeneachfMRIstudyparticipantandeveryotherstudentinthenetworkwas

characterized. An alternative visualization of the network is shown in which nodes are

organized intohorizontal layersaccording todistance fromaparticularparticipant.Each

participant’sstimulussetwascomprisedof12ofhisorherclassmates:thetwolowestand

twohighesteigenvectorcentralityindividualsatdistancesofone,two,andthreefromthe

participant in the network (e.g., the classmates signified by the two smallest and two

largest nodeswithin each layer in (A)). (B) During the fMRI study, participants viewed

brief(2s)videosofthe12individualsintheirstimulussetsseparatedby4-6soffixation.

Inordertomaintainattention,aone-backtaskwasused(i.e.,participantswereinstructed

touseabuttonpress to indicatewhenan identicalvideowaspresented twice inarow).

Frames from this participant’s video clip are reproduced with permission from the

individual.



https://doi.org/10.1101/098988


27

Figure 3. GLM decomposition searchlight. (A) A spherical searchlight was moved

throughouteachparticipant’sbrain.(B)Ateachpointinthebrain,distributedpatternsof

neuralresponsestoeachindividualintheparticipant’sstimuluswereextractedwithina9-

mmradiusspherecenteredonthatpoint.(C)Ateachsearchlightcenter,aneuralRDMwas

generatedbasedonpairwisecorrelationdistancesbetweenlocalneuralresponsepatterns

toeachclassmateintheparticipant’sstimulusset.(D)EachlocalneuralRDMwasmodeled

asaweightedcombinationofRDMsconstructedbasedonthepairwiseEuclideandistances

(i.e.,theabsolutevalueofnumericaldifferences)betweenindividualsineachparticipant’s

stimulussetintermsofsocialdistance,eigenvectorcentrality,andnetworkconstraint.



https://doi.org/10.1101/098988


28

Figure4.Neuralencodingofsocialnetworkposition.(A)Distinctbrainregionsencodedifferentpropertiesofpeers’socialnetworkpositions(socialdistance=purple;eigenvectorcentrality=orange;constraint=green).Betavaluesindicatetheextenttowhichtheinformationcontainedinlocalmulti-voxelresponsepatternstoparticipants’classmatescouldbepredictedbasedonpropertiesofthoseindividuals’socialnetworkpositions;p<.05,FWE-corrected.(B)TheR2valuecorrespondingtotheGLMdecompositionperformedateachsearchlightcenterindicatestheextenttowhichtheinformationcontainedinlocalmulti-voxelresponsepatternscanbeexplainedbythesocialnetworkpositionsoftheclassmatesbeingviewed.Voxel-wiseR2valuesaveragedacrosssubjectsaredepicted;redcontoursindicateclustersofvoxelsthatreliablysignaledoneormoreofthetestedaspectsofsocialnetworkpositionacrossparticipants.ResultsareprojectedontoacorticalsurfacemodeloftheTalairach50N27brainusingPySurfer(https://github.com/nipy/PySurfer).



https://doi.org/10.1101/098988


29

Figure5.Associationsbetweenperceivedandactualsocialnetworkcharacteristics.Blackdashedlinesdepicttherelationshipsbetweenperceivedandactualsocialnetworkcharacteristicsacrossallparticipants(fitusinganordinaryleastsquareslinearmodel).Solidpurple,orangeandgreenlinesdepicttheserelationshipsforeachsubjectforsocialdistance,eigenvectorcentrality,andconstraint,respectively.(A)Neuroimagingstudyparticipants’subjectiveratingsofsocialcloseness,proportionofsocialtimespenttogether,andfrequencyofdiscussionswiththeindividualsintheirstimulussetsvariedaccordingtogeodesicnetworkdistancefromtheminthenetwork(allp’s<.0001,seemaintext).(B)Participants’estimatesoftheeigenvectorcentralityoftheindividualsintheirstimulussetswerecloselyrelatedtothoseindividuals’actualeigenvectorcentralities(p<.0001,seemaintext).(C)Participants’estimatesofthenetworkconstraintofindividualsintheirstimulussetswerealsoassociatedwiththeactualconstraintofthoseindividuals’positionsinthesocialnetwork(p<.0001,seemaintext).Asdescribedinthemaintext,self-reportdatawasobtainedafterscanning;networkconstraintandeigenvectorcentralitywerelog-transformedpriortoplottingandanalysistoalleviateskew.Perceivednetworkconstraintratingsweremultipliedby-1priortoplottingbecausetherelevantquestionaskedparticipantstorateperceivedbrokerage(whichisinverselyrelatedtonetworkconstraint).Analysesofbehavioralratingswereconductedusinglinearmixedmodelsthatincludedby-subjectrandomslopesandintercepts.



https://doi.org/10.1101/098988


S1

SupplementaryInformation

I.Opticalflowanalysis

Toquantifytheamountofmovementwithineachvideoclipusedasastimulusin

theneuroimagingexperiment,theaverageopticalflow(i.e.,thepatternofapparentmotion

betweenconsecutivevideoframes)wascomputedforeachvideothatwasshowninthe

fMRIstudy.Giventhatthevideosusedasstimuliwererecordedbyastablecameraagainst

aplain,staticbackground,opticalflowestimatesforthesevideoscaptureoftheamount

thateachindividualmovedhisorherfacialfeaturesandheadinthevideoclip.Farneback’s

algorithmformotionestimation1asimplementedinOpenCV2wasusedtoestimatethe

averagemagnitudeofopticalflowineachvideo.Thismethodextractsapixel-wisemotion

vectorforeachpairofsequentialframesinwhicheachpixelischaracterizedbya

magnitudeandadirection.Toestimatethemagnitudeofmotionwithineachframepair,

themagnitudevalues(withoutrespecttodirection)weresummedacrosspixels.To

computethemeanmagnitudeofopticalflowforagivenvideo,themotionmagnitude

estimateswereaveragedacrossframeswithinthatvideo.

Inordertotestwhetherornotindividualdifferencesinnetworkconstraintare

relatedtomovementinthevideosusedasstimuli,thecorrelationbetweennetwork

constraintandaveragemotionmagnitudewasassessedamongthe88individualswhose

videoswereusedasstimuliinthefMRIstudy.Giventhatdistributionsofbothvariables

werehighlyskewed,datawerelog-transformedpriortoanalysis.Theresultsofthis

proceduresuggestthatinthestimuliusedinthecurrentstudy,networkconstraintand

amountofmovementwerenotsignificantlycorrelated,r=-0.12,p=0.28(seeFig.S1).



https://doi.org/10.1101/098988


S2

II.Post-scanquestionnaire

Participantsperformedthepost-scanquestionnaireona13”MacBooklaptopusing

Psychopy3.Participantsfirstviewedaninstructionscreenthatread,“Nowyouwillseethe

samepeoplewhoyousawinthescanner.Youwillbeaskedquestionsabouteachperson.

Thesequestionsrelateonlytothisperson’sinteractionswithinthe[institutionname]MBA

cohort.Weunderstandthatpeoplehavemanysocialcirclesthattheyparticipatein(perhaps

includingfamily,friendsoutsideof[theinstitution],othercontacts,etc.).Forthesequestions,

pleasejustconsiderinteractionswithintheMBAcohort.Youwillbepresentedwitha

continuousratingscaleforeachquestion.Youcanchooseanypointalongthecontinuumto

respond.Pressanykeytocontinue.”Duringthesurvey,videosofthe12individualsfromthe

participant’sstimulussetwerepresentedinarandomorder.Participantsrespondedtoall

questionsaboutagivenindividualsequentially,andthesamevideothathadplayedinthe

scannerrepeatedonaloop(withoutsound)abovethequestiontextandresponsescale

(seeFig.S2).

Participantswerepresentedwithquestionsconcerninglaydefinitionsof

eigenvectorcentrality(“Insocialnetworkanalysis,scientistsassessaconstructthat

measureshowmanyfriendsapersonhas,andhowmanyfriendsaperson’sfriendshave.How

wouldyouratethispersononthisconstruct?”Responsesrangedfrom“Low(fewfriendswho

havefewfriends)”to“High(manyfriendswhohavemanyfriends)”)andconstraint(“Social

networkanalystsalsoassessaconstructcalled‘brokerage’thatmeasureshowmuchaperson

connectsgroupsofpeoplewhowouldn’totherwisebeconnected.Usingthisdefinition,how

highisthisindividualin‘brokerage’?”Responsesrangedfrom“Low(thispersonnever

connectsdistinctgroupsofpeople”to“High(Thispersonoftenconnectsdistinctgroupsof



https://doi.org/10.1101/098988


S3

people)”).Responsestotheitemassessingbrokeragewerereversescoredinorderto

estimateperceivednetworkconstraint.

Participantswerealsopresentedwiththenamegeneratorthathadoriginallybeen

usedtoconstructthenetwork(“Considerthepeoplewithwhomyouliketospendyourfree

time.Duringthelastmonth,isthisoneoftheclassmateswhoyouhavebeenwithmostoften

forinformalfriendshipactivities,suchasgoingouttolunch,dinner,drinks,films,visitingone

another’shomes,andsoon?”Responsesrangedonacontinuumfrom“Noneofmysocial

activitiesinthepastmonthhaveincludedthisperson”to“Allofmysocialactivitiesinthepast

monthhaveincludedthisperson”),aswellasquestionsdesignedtoassesstiestrength

(“Howcloseareyouwiththisperson?”Responsesrangedfrom“Distant”to“Lessthanclose”

to“Close”to“EspeciallyClose”)andfrequencyofinteractions(“Onaverage,howoftendoyou

talktothisperson(anysocialorbusinessdiscussion)?”Responsesrangedfrom“Lessoften”

to“Monthly”to“Weekly”to“Daily”).



https://doi.org/10.1101/098988


S4

References1. Farnebäck,G.inImageAnalysis:LectureNotesinComputerScience(eds.Bigun,J.&

Gustavsson,T.)2749,363–370(SpringerBerlinHeidelberg,2003).2. Bradski,G.TheOpenCVLibrary.DrDobbsJ.Softw.Tools25,120–125(2000).3. Peirce,J.W.PsychoPy-PsychophysicssoftwareinPython.J.Neurosci.Methods162,

8–13(2007).



https://doi.org/10.1101/098988


S5

Figure S1.Relationshipbetweennetwork constraint andmovementduring videos.

Theamountofmovementofthe88individualswhosevideoswereusedasstimuliwasnot

significantly related to the constraint characterizing those individuals’ positions in the

socialnetworkoffirst-yearMBAstudents,r=-0.12,p=0.28.



https://doi.org/10.1101/098988


S6

Figure S2. Post-scan questionnaire. Following scanning, participants responded to

questions about their subjective perception of each aspect of social network position of

interest for each individual in their stimulus set. A screenshot of the question

correspondingtonetworkconstraint(reverse-scored)isshown.



https://doi.org/10.1101/098988


S7

Table S1: Brain regions where local neural information content is associated with the social distance of the individuals being viewed. Hemi Size

(mm3) COG x

COG y

COG z

Location

R 4,397 49.6 -46.6 7.6 IPL (SMG), STG, STS, MTG Hemi = hemisphere; COG = center of gravity; L = left; R = right; IPL = inferior parietal lobule; SMG = supramarginal gyrus; STG= superior temporal gyrus; STS = superior temporal sulcus; MTG = middle temporal gyrus. All reported results are significant at a statistical threshold of p < .05, FWE-corrected. All coordinates are in Talairach space.

Table S2: Brain regions where local neural information content is associated with the eigenvector centrality of the individuals being viewed. Hemi Size

(mm3) COG x

COG y

COG z

Location

L 24,483 -42.4 -18.5 33.6 IPL, IFG, Ins., pre-central gyrus R 8,768 21.7 26.8 -4.9 MPFC, IFG, aIns., ant. PHG, TP L 7,716 -32.7 12.9 -6.1 aIns., IFG L, R 7,552 -9.4 -41.8 40.4 PCC, precuneus R 6,802 20.4 -48.0 -2.8 PHG, LG, FG R 6,065 48.0 -29.4 38.4 IPL, precuneus, post-central gyrus L, R 5,233 -0.8 -44.4 65.4 Precuneus, post-central gyrus L, R 4,961 -0.6 -83.3 28.5 EVC

Hemi = hemisphere; COG = center of gravity; L = left; R = right; a = anterior; IPL = inferior parietal lobule; IFG = inferior frontal gyrus; Ins. = insula; MPFC = medial prefrontal cortex; PHG = parahippocampal gyrus; TP = temporal pole; PCC = posterior cingulate cortex; LG = lingual gyrus; FG = fusiform gyrus; EVC = extrastriate visual cortex. All reported results are significant at a statistical threshold of p < .05, FWE-corrected. All coordinates are in Talairach space.

Table S3: Brain regions where local neural information content is associated with the constraint of the individuals being viewed. Hemi Size

(mm3) COG x

COG y

COG z

Location

R 11,872 51.5 -16.0 -3.0 STS, STG, MTG, ITS, pIns. L 7,739 -51.6 -38.5 7.1 STS, STG, MTG, pIns. R 4,363 11.5 -5.7 58.4 SMA, dorsal premotor cortex Hemi = hemisphere; COG=center of gravity; L = left; R = right; STS = superior temporal sulcus; STG = superior temporal gyrus; MTG = middle temporal gyrus; ITS = inferior temporal sulcus; pIns. = posterior insula; SMA = supplementary motor area. All reported results are significant at a statistical threshold of p < .05, FWE-corrected. All coordinates are in Talairach space.



https://doi.org/10.1101/098988


carolyn m. parkinson1* 2 3 - biorxiv · carolyn m. parkinson1*, adam m. kleinbaum2, and thalia...

Documents