Transcript
  • 1

    Psychophysiologicalmodeling-Currentstateandfuturedirections

    DominikR.Bach1.2,3,GiuseppeCastegnetti1,2,ChristophW.Korn1,2,4,SamuelGerster1,2,Filip

    Melinscak1,2,TobiasMoser1,2

    1ClinicalPsychiatryResearch,DepartmentofPsychiatry,Psychotherapy,andPsychosomatics,PsychiatricHospital,UniversityofZurich,Zurich,Switzerland

    2NeuroscienceCenterZurich,UniversityofZurich,Zurich,Switzerland3WellcomeTrustCentreforNeuroimagingandMaxPlanck/UCLCentrefor

    ComputationalPsychiatryandAgeingResearch,UniversityCollegeLondon,UnitedKingdom

    4InstituteforSystemsNeuroscience,UniversityMedicalCenterHamburg-Eppendorf,Hamburg,Germany

    Correspondence

    DominikR.Bach,DepartmentofPsychiatry,Psychotherapy,andPsychosomatics,PsychiatricHospital,UniversityofZurich,Lenggstrasse31,CH-8032Zurich;

    Switzerland.Email:[email protected]

    Abstract

    Psychologistsoftenuseperipheralphysiologicalmeasurestoinferapsychologicalvariable.Itisdesirabletomakethisinverseinferenceinthemostpreciseway,ideally

    standardizedacrossresearchlaboratories.Inrecentyears,psychophysiological

    modelinghasemergedasamethodthatrestsonstatisticaltechniquestoinvert

    mathematicallyformulatedforwardmodels(psychophysiologicalmodels,PsPMs).

    ThesePsPMsarebasedonpsychophysiologicalknowledgeandoptimizedwithrespect

    totheprecisionoftheinference.Buildingonestablishedexperimentalmanipulations,

    knowntocreatedifferentvaluesofapsychologicalvariable,theycanbebenchmarkedin

    termsoftheirsensitivity(e.g.,effectsize)torecoverthesevalueswehavetermedthis

    predictivevalidity.Inthisreview,weintroducetheproblemofinverseinferenceand

    psychophysiologicalmodellingasasolution.Wepresentbackgroundandapplicationfor

    allperipheralmeasuresforwhichPsPMshavebeendeveloped:skinconductance,heart

    period,respiratorymeasures,pupilsize,andstartleeyeblink.ManyofthesePsPMsare

    taskinvariant,implementedinopen-sourcesoftware,andcanbeusedofftheshelffora

  • 2

    widerangeofexperiments.Psychophysiologicalmodelingthusappearsasapotentially

    powerfulmethodtoinferpsychologicalvariables.

    Keywords

    analysis/statisticalmethods,autonomicnervoussystem,computationalmodeling,

    electrodermalactivity,heartrate,pupillometry

    1Introduction

    Peripheralphysiologicalmeasurementsareoftenusedtoinferpsychologicalvariables.

    Forexample,totestapsychologicalinterventionthatreducesfearmemory,aresearcher

    maybeinterestedinquantifyingthestrengthoffearmemoryfromskinconductance

    responses(SCR).Thispsychologicalperspectiveistheoppositeofabasic

    psychophysiologicalperspective,inwhichresearchersaimtodescribehowpsychologicalvariablesimpactontheperipheralmeasure.Toenableinferenceona

    psychologicalvariable,thepsychophysiologicalmappingfromthisvariabletothe

    measuredsignal(the"forwardmodel"instatisticalterminology)needstobeknown

    withsomecertainty,anditneedstobeexploitedinthebestpossibleway(themodel

    needstobe"inverted"),toarriveatthemostpreciseestimateofthepsychological

    variable.Thisisthepsychophysiologicalinverseproblem(Figure1a).

    Psychophysiologicalmodelingisastatisticalframeworktosolvethisproblemina

    principledmanner(Bach&Friston,2013).Itcanprovideexperiment-invariant,off-the-

    shelfapplicationsthatimproveoncurrentmethodsforinverseinferenceandthereby

    suggestmeaningfulmethodologicalstandardstoenhancereproducibility.

  • 3

    Indeed,psychophysiologicalmodelingapproacheshavebeenappliedtoanalyzeawide

    varietyofexperiments:toinferattentionalvariablesfrompupilresponses(e.g.,deGee

    etal.,2017;deGee,Knapen,&Donner,2014),toinferfearlearningfromSCR(e.g.,Bach,

    Weiskopf,&Dolan,2011;Bulganin,Bach,&Wittmann,2014;Tzovara,Korn,&Bach,

    2018)andstartleeyeblink(Bach,Tzovara,&Vunder,2017),andtoquantifyautonomic

    arousalduringperception(e.g.,Bach,Seifritz,&Dolan,2015;Hayesetal.,2013;Koban,

    Kusko,&Wager,2018;Koban&Wager,2016;Sulzeretal.,2013),decision-making(e.g.,

    Alvarez,etal.,2015;Bach,2015a;deBerkeretal.,2016;Nicolle,Fleming,Bach,Driver,&

    Dolan,2011;Talmi,Dayan,Kiebel,Frith,&Dolan,2009),andrest(Fanetal.,2012).

    Thisreviewisstructuredinthefollowingway.First,wediscusshowtocompare

    methodsforinverseinferenceonpsychologicalvariablesandintroducetheconceptof

    predictivevalidity.Wethenpresentpsychophysiologicalmodelingasanovelapproach,

    includingaspecificimplementationcreatedbytheauthorstogetherwithrelated

    methods.Inthemajorpartofthereview,wegiveatutorial-styleoverviewofthevarious

    forwardmodelsandinversionmethodsdevelopedoverthepastdecade,fordifferent

    physiologicalmeasures.Thefieldismovingrapidly.Whileninemethodologicalarticles

    Figure1.A:Thepsychophysiologicalinverseproblem.Top:psychophysiologicalperspective(forwardinference,e.g.,DoesaversivememoryinfluenceSCR?).Bottom:psychologicalperspective(inverseinference,e.g.,Doesmyprocedureestablishaversivememory,asindexedbySCR?)B:Benchmarkinganinverseinferencemethodbyassessingpredictivevalidity:Whatisthesensitivityforinferringtheeffectofaknownexperimentalmanipulation.

  • 4

    onthetopicwerepublishedbetween1993-2013,11suchpaperscameoutinthe5

    yearssincethelastreviewonthetopic(Bach&Friston,2013).Thislastreview

    containedahistoricalperspectiveonthedevelopmentofmodelsforSCRinthe1990s

    and2000s(Alexanderetal.,2005;Bach,Flandin,Friston,&Dolan,2009;Barry,

    Feldmann,Gordon,Cocker,&Rennie,1993;Limetal.,1997)andontheemergent

    critiqueofoperationalism(Green,1992);here,weapproachtheproblemina

    systematic,nonhistoricalmanner.

    2Predictivevalidity

    Allanalysismethodsforpsychophysiologicalsignalsarebasedonsomeknowledge

    abouttheforwardmappingfrompsychologytophysiology.Aplethoraof

    psychophysiologicalliteraturehasaddressedsuchforwardmappings.However,even

    withaperfectforwardmodeltherearedifferentwaysofmakinginverseinference.For

    example,onecandefinedifferentpossibletimewindowstodetectanSCRpeakafteran

    experimentalevent.Extendingthepeakwindowmayincreasethesensitivityofthe

    methodtodetectatrueevent-relatedresponse,butdecreaseitsspecificitybecause

    experiment-unspecificpeaksmaybemistakenforevent-relatedones.Crucially,the

    optimalbalanceisdifficulttointuitasis,forexample,evidentfromthecoexistenceof

    differentpeakdetectionwindowsinanalysisrecommendations(Boucseinetal.,2012),

    andsometimesevenwithinthesamelaboratories.Hence,itwouldbedesirableto

    quantitativelyevaluateaninverseinferencemethod.

    Toassessthequalityofinverseinference,onewouldideallycomparetheinferredwith

    theactualvalueofthepsychologicalvariable(i.e.,with"groundtruth").Ofcourse,

    groundtruthisneverknownforpsychologicalvariables.1Tosolvethisconundrum,we

    havepragmaticallyproposedtouseanexperimentalmanipulationthatcanbeassumed

    toinfluencethepsychologicalvariableinacertainwayandisknowntoimpactonthe

    peripheralmeasure.Onecanthenevaluatemethodsbytheirsensitivitytodetectthe

    impactofthisexperimentalmanipulation(Figure1b).Wehaveintroducedtheterm

    predictivevalidityforthismeasure(Bach,Daunizeau,Friston,&Dolan,2010),sinceit

    1Thisistrueformanyareasofscienceandtechnology.

  • 5

    evaluateshowwellthepsychologicalvariablecanbepredicted.2Predictivevalidity

    analysishasoftenbeenperformedonacategoricalexperimentalmanipulation(e.g.,

    anticipatingthreatorsafetyinfearconditioning).Althoughitmaythusappearonthe

    surfacethatpredictivevalidityboilsdowntoclassificationperformance,oneusually

    aspirestoinferpsychologicalvariablesonacontinuousscale,sothatthemethod

    extendstosituationsinwhichthepsychologicalvariablevariesparametricallyacross

    morethantwolevels.Crucially,sincemostpsychophysiologicalmeasuresarerelatively

    unspecific,validationexperimentsrequirethattheexperimentalconditionsdifferon

    onlyonedimension,thepsychologicalvariableofinterest.Thisistrueforanyinverse

    inferencemethod.Onceamethodwithhighpredictivevalidityisidentified,onecan

    applythismethodtoother(methodologicallysimilar)experimentstoinferthesame

    psychologicalvariable.

    Thus,inavalidationexperiment,agoodinferencemethodprovidesanestimatorofthe

    (known)psychologicalvariablethathassmallervariancethananyothermethod.Fora

    categoricalvalidationexperimentwithtwolevels,thissimplymeans-becausethescale

    ofthepsychologicalvariableisarbitrary-thatthestandardizeddifferenceinthe

    estimatedpsychologicalvariablebetweenthesetwolevelsshouldbelarge.Thiscanbe

    evaluatedbyregardingtheeffectsize,orteststatistic,ortheresidualsumsofsquaresin

    apredictivemodel,orthemodelevidenceofthatpredictivemodel.Allofthese

    quantitiesaremonotonicallyrelated.Usingmodelevidenceadditionallyallowsusto

    makestatementswhethertwomethodsaredecisivelydifferent(Bach&Friston,2013;

    seeAppendixEquation1).Atthesametime,thedifferenceintheestimated

    psychologicalvariablebetweentworandompartitionsofthesameexperimental

    conditionshouldbezeroonaverage.

    Predictivevaliditycanbeharnessedtovalidateanyinverseinferencemethod,includingoperationalanalysis.Forthepsychophysiologicalmodelingapproach,someadditional

    considerationsarewarranted.Here,psychologicalvariablesareestimatedbyoptimizing

    thegoodness-of-fitoftheforwardmodel.Yet,forcomparisonofdifferentmethods,the

    goodness-of-fitoftheforwardmodelisnotasuitablecriterion.Thegoaloftheforward

    2Becausethepsychologicalvariableisknownapriori,onecouldalsocallit“retrodictivevalidity”.

  • 6

    modelistopredictthesignal,andthegoalofinferenceistofindthemostprecise

    estimateofthepsychologicalvariable.Thesetwogoalscanalignwith,beorthogonalto,

    orevenopposeeachother.Intuitively,onecouldassumethat,iftheforwardmodelis

    knownwithcertaintyandformulatedinmathematicalterms,oneshouldeasilybeable

    toinvertthemapping.However,thereareseveralstatisticalreasonswhythisintuition

    isincorrectinthegeneralcase(althoughitmaybecorrectunderspecific

    circumstances).First,theforwardmodelmayuseparametersthatoneisnotinterested

    ininferring.Forexample,thebestknownforwardmodelforSCRassumesthatthe

    strengthoramplitudeofpsychologicalinputintothesystemisdifferentoneachtrial

    (Gerster,Namer,Elam,&Bach,2017).However,manyresearchersarenotinterestedin

    psychologicalvariablesonatrial-by-trialbasisbutonlyintheaveragepsychological

    variablewithinanexperimentalcondition.Thestandardgenerallinearmodel(GLM)

    inversionapproachforSCR(Bachetal.2009;Bach,Friston,&Dolan,2013)willnormallyyieldthesameconditionwiseestimates,regardlessofwhetherestimationwas

    doneonatrial-by-trialbasisfollowedbyaveraging,oronacondition-by-conditionbasis.

    Inthiscase,thesimplermodelyieldsthesameinferenceonthepsychologicalstate,

    althoughempiricallyitcannotpredictSCRdatasowell,becauseitwouldassumethe

    same(average)SCRamplitudeforeachtrial(Bachetal.,2013).Hence,precisionofthe

    forwardmodelandoftheinferenceareunrelated.Anexamplewheretheyareopposed

    isgivenbyindividualresponsefunctionsforSCR.Allevidencesuggeststhatthemapping

    fromsudomotornerveactivitytoskinconductancedependsonsubject-specific

    anatomicalproperties,andisvariablebetweenpersons(Bach,Flandin,Friston,&Dolan,

    2010;Gersteretal.,,2017).Hence,aforwardmodeltakingthisheterogeneityinto

    accountwillhaveabettergoodness-of-fitthanamodelassumingacanonicalresponse

    functionacrosssubjects,aswehavealsoshownempirically(Bachetal.,,2013).Atthe

    sametime,itcanbedifficulttoestimatetheshapeofanindividual'strueresponse

    functionfromalimitednumberoftrialswithshortintertrialintervals,andensuing"overfitting"canmakeinferenceonthepsychologicalvariableworse,reducing

    predictivevalidity(Bach,Friston,&Dolan,2013).

    Tosummarize,predictivevalidityallowsastatementonthequalityofinverseinference,

    regardlessofthemethodunderstudy.Itcanbeusedtobenchmarkpsychophysiological

    models,operationalmethods,orevenmachine-learningmethodsthattrytofind

    statisticalregularitieswithoutanyknowledgeofthepsychologicalorbiophysical

  • 7

    relationships(Grecoetal.,2017;Greco,Lanata,Valenza,DiFrancesco,&Scilingo,2016;

    Greco,Valenza,&Scilingo,2016).Asastatisticalframework,ithasapotentialto

    improveinverseinference,tostandardizemethodsacrosslaboratories(byselectingthe

    bestone),andtoprovideanobjectivemeansforqualitycontrolwithinandbetween

    laboratories.Wewillreturntotheselatterpointsinthediscussion.

    3PsychophysiologicalModeling

    Thegoalofinverseinferenceistofindthebestpossibleestimatorofapsychologicalvariable,fromameasureddatatimeseries.Thisincludesso-calledoperationalmethods,

    which"operationalize"(i.e.,equateanoisyversionof)thepsychologicalvariablewitha

    singlephysiologicaldatafeature,forexample,apeak-to-troughmeasure.Because

    operationalmethodsuseoneoraverysmallnumberofdatafeatures,ratherthanthe

    entiretimeseries,theymaysufferfrominformationloss.Psychophysiologicalmodeling

    isawayofusinganentiredatatimeseriesforinference(Figure2).Inanutshell,a

    psychophysiologicalmodel(PsPM)isaformal,quantitativemodelthatmapsa

    psychologicalvariableontoanobserveddatatimeseries.PsPMsarespecifiedin

    mathematicalform.TheearliestPsPMsweredevelopedforSCRandexplicitlyconstitutedasequenceoftwomodels(Figure3a):aneuralmodelthatspecifiesthe

    mappingofthepsychologicalvariableontosudomotornerveactivity(SNA),anda

    peripheral(effectororgan)modelthatspecifieshowSNAmapsontomeasuredSCR

    (Alexanderetal.,2005;Bachetal.,2009;Limetal.,1997).ForSCR,thissplitisuseful

    Figure2.Operationalanalysis(top)assumesthatselecteddatafeaturesare"equivalent"toapsychologicalvariable,wheretheselectionofdatafeaturesisoftenbasedoninformalmodels.Psychophysiologicalmodelling(bottom)estimatesthemostlikelypsychologicalvariable,giventheentiredatatimeseriesandastandard(experiment-invariant)responsemodel.

  • 8

    becausetheperipheralmodelcanbeevaluatedonitsownbyintraneuralstimulation

    andrecordingsfromwellaccessibleperipheralnerves(Gersteretal.,2017).Forsome

    othermeasures,theperipheralmodelcanbeapproximatedbyspecificstimuli,for

    example,onecanuseluminancechangestoelucidatepupilmechanics.However,for

    mostPsPMsthathavebeencreatedtodate,neuralandperipheralprocessesaremore

    difficulttoseparateexperimentallyastheefferentnervesarelessaccessible(atleastin

    humans),andincurrentmodelstheyareeithercollapsed,orthedistinctionisonlyused

    formathematicalconvenience.

    3.1Hybridapproaches

    TheempiricaldistinctionbetweenneuralandperipheralmodelsforSCRhasearlyon

    motivatedahybridapproach(Alexanderetal.,2005),engendered,forexample,inthe

    Figure3.A:Basicformalismofmostpsychophysiologicalmodels.B:Relatedhybridapproaches(e.g.Ledalab,cvxEDA)useastandardresponsemodeltoinfera(noisy)timeseriesofneuralinputs,andselectdatafeaturesofthattimeseriesas"equivalent"tothepsychologicalvariable.C:Lineartimeinvariant(LTI)systemslieatthecoreofallexistingpsychophysiologicalmodellingandhybridapproaches.Aneuralinputisconvolvedwithacanonical(experiment-invariant)responsefunction,toyieldapredictionforthemeasuredsignal.

  • 9

    softwaresLedalab(Benedek&Kaernbach,2010a,2010b)orcvxEDA(Greco,Valenza,

    Lanata,Scilingo,&Citi,2015).Inthisapproach,adeterministicperipheralmodelis

    invertedtocomputeanoisySNAtimeseriesfromthetimeseriesofmeasuredSCRdata.

    Tomakeinferenceonpsychologicalvariables,somedatafeatures(peak-to-trough

    measures)oftheSNAtimeseriesareextracted,inlinewithmoretraditionaloperational

    analysis(Figure3b).Thissecondmappingisheuristicallymotivated,notquantitatively

    specifiedorevaluated.TwodifferentstudygroupshavecomparedLedalabwithpeak-

    to-troughmethodsontheonehandandafullPsPM-basedapproachontheother,in

    paradigmswithrelativelyshortintertrialintervals.Onaverage,thehybridLedalab

    approachwasfoundtoyieldsimilarpredictivevalidityasdirectlyusingpeak-to-trough

    measuresoftheSCRdata,anditspredictivevaliditywassurpassedbyinversionoffull

    PsPMs(Bach,2014;Green,Kragel,Fecteau,&LaBar,2014).Systematicevaluationof

    cvxEDAhasnotyetbeenconducted.Notably,theseapproachesdifferfromthePsPMapproachalsointheirperipheralforwardmodelsandstatisticalmethodsformodel

    inversion.Theycouldpossiblybeextendedtodirectlyestimatepsychologicalvariables,

    butretainingtheirspecificforwardmodelsandinversionmethods.

    3.2Lineartimeinvariantsystems

    AllPsPMsandhybridmodelsthathavebeenproposeduptotodaycontainattheirheart

    alineartimeinvariant(LTI)system(Figure3c).ALTIsystemisasystemtheoutputof

    whichdoesnotexplicitlydependontime(timeinvariance),andtheoutputtothesumof

    twoinputsisjustthesumoftheoutputsoftheindividualinputs(linearity).Thefirst

    principleimpliesthatdifferentresponseshapesareexplainedbydifferentinputs.

    Accordingtothelinearityprinciple,themagnitudeofaresponsedoesnotdependonthe

    baseline.LTIsystemsareunambiguouslydescribedbythemathematicaloperationof

    convolution(overlapintegral)ofaninputtimeserieswitharesponsefunction(RF),

    whichcorrespondsinsignalprocessingtermstoalinearfilter(seeAppendixEquation2).LTIsystemsconstituteamathematicalsimplificationofrealbiophysicalsystems,

    whichcontainmanyparametersthatcannotbeusefullyconstrainedfrommeasured

    data.WewillreportforeachmeasurehowcanitbedescribedbyaLTIsystem.

    3.3Modelinversion

    ManyPsPMsassumethatexperimentaleventsrapidlyengageapsychologicalprocess,

    whichthenfeedsintothephysiologicalsystemwithconstantlatencyandshape.Under

  • 10

    theseassumptions,theamplitudeofthepsychologicalinputcanbeestimatedina

    generallinear(convolution)model(Bachetal.,2009),similartostandardapproaches

    foranalysisoffMRIs(Friston,Jezzard,&Turner,1994).Inanutshell,theRFisconvolved

    withatimeseriesofimpulses(deltafunctions)centeredonexperimentaleventsfor

    eachcondition,andtheensuingtimeseriesformcolumnsinthedesignmatrixofa

    multipleregressionmodel.Theestimatedcoefficientofanindividualeventcolumn

    constitutestheamplitudeestimateforthatcondition(seeAppendixEquation3-5).

    Looselyspeaking,theRFisregressedontotheobservedresponse,andtheregression

    coefficientistheestimateoftheinputamplitude.Iflatencyand/orshapeofthe

    psychologicalinputcannotbeassumedtobeconstant,thentheyneedtobeestimatedas

    well,andtheinversionmodelbecomesnonlinear,forexample,inSCRmodelsforfear

    conditioning(Bach,Daunizeauetal.,2010)orstartleeyeblinkresponse(SEBR)models

    (Khemka,Tzovara,Gerster,Quednow,&Bach,2017).

    4Psychophysiologicalmodelsfordifferentmeasurements

    Table1givesanoverviewofthedifferentpsychophysiologicalmodelsproposeduntil

    today.

  • 11

    Table1:Psychophysiologicalmodelsdevelopeduntiltoday.PSR:pupilsizeresponses.SCR:skinconductanceresponses.HPR:Heartperiodresponses.RPR:respirationperiodresponses. 1RAR:respirationamplituderesponses.RFRR:respiratoryflowrateresponses.SEBR:startleeyeblinkresponses. 2Measure

    ment

    Psychological

    variable

    Neuralmodel Peripheralmodel Model

    inversion

    Software

    implementation

    Publishedin

    SCR Genericphasicarousal Instantaneousimpulse LTIsystemwithparameters

    fromempiricaldata

    GLM PsPM Bach,2014;Bach,Flandin,Friston,&

    Dolan,2009;Bach,Flandin,Friston,&

    Dolan,2010;Bach,Friston,&Dolan,2013

    Genericphasicarousal

    (validatedforfear

    conditioning)

    ConstrainedGaussian

    impulse

    LTIsystemwithparameters

    fromempiricaldata

    Variational

    Bayes

    PsPM Bach,Daunizeau,Friston,&Dolan,2010;

    Staib,Castegnetti,&Bach,2015

    Generictonicarousal

    (validatedforanxiety

    andcognitiveload)

    Gaussianimpulses

    withconstantshape

    andunconstrained

    onset

    LTIsystemwithparameters

    fromempiricaldata

    Variational

    Bayes

    PsPM Bach,Daunizeau,Kuelzow,Friston,&

    Dolan,2011;Bach,Friston,&Dolan,2010

    Generic Notspecified LTIsystemwithparameters

    fromtheoreticalconsiderations

    Deterministic

    inversefilter

    Ledalab Benedek&Kaernbach,2010a;Benedek&

    Kaernbach,2010b

    Generic Discreteimpulseswith

    unconstrainedonset

    LTIsystemwithparameters

    fromtheoreticalconsiderations

    Convex

    optimisation

    cvxEDA Greco,Valenza,Lanata,Scilingo,&Citi,

    2015

    PSR Luminanceadaptation Instantaneousimpulse Combinationof2LTIswith

    parametersfromempiricaldata

    GLM PsPM Korn&Bach,2016

    Attention Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    OLSestimation

    infrequency

    domain

    Pupil Hoeks&Levelt,1993

    Fearconditioning Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    GLM PsPM Korn,Staib,Tzovara,Castegnetti,&Bach,

    2017

    HPR Notyetspecified Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    GLM PsPM Paulus,Castegnetti,&Bach,2016

    Fearconditioning Instantaneousimpulse LTIwithparametersfrom GLM PsPM Castegnetti,etal.,2016

  • 12

    empiricaldata

    RPR Notyetspecified Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    GLM PsPM Bach,Gerster,Tzovara,&Castegnetti,2016

    RFRR Notyetspecified Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    GLM PsPM Bach,Gerster,Tzovara,&Castegnetti,2016

    RAR Notyetspecified Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    GLM PsPM Bach,Gerster,Tzovara,&Castegnetti,2016

    Fearconditioning Instantaneousimpulse LTIwithparametersfrom

    empiricaldata

    GLM PsPM Castegnetti,Tzovara,Staib,Gerster,&

    Bach,2017

    SEBR Genericstartlereflex Instantaneousimpulse

    withvariablelatency

    LTIwithparametersfrom

    empiricaldata

    Template

    matching/GLM

    PsPM Khemka,Tzovara,Gerster,Quednow,&

    Bach,2017

  • 13

    4.1Skinconductance 3

    4.1.1Forwardmodel 4

    Skinconductanceisoftenusedtoinferphasicortonicsympatheticarousalgeneratedby 5

    awiderangeofpsychologicalstimuliandtasks(Boucsein,2012).Openingofsweat 6

    glands,elicitedviathesympatheticnervoussystemwithnegligibleparasympathetic 7

    transmission,causesphasicincreasesofskinconductancethataretermedSCR(see 8

    Boucsein,2012,forthephysiologyofSCR).SlowCfiberscarryingimpulsestothesweat 9

    glandsaretermedsudomotor(SN),andtheiractivitycanbemeasuredbyintraneural 10

    recordings.Fromtheirendterminal,theneurotransmitteracetylcholinediffuses 11

    throughtheskintoreachsweatglands,aprocessonthetimescaleofuptoasecond.In 12

    thehistoryofpsychophysiologicalmodeling,itwasrecognizedearlyonthat 13

    nonoverlappingSCRcanbewelldescribedbyasimpleresponsefunction,and 14

    overlappingSCRcanbeseenasbeinggeneratedbyaLTIsystem(Alexanderetal.,2005; 15

    Bachetal,2009).AllpublishedSCRmodelshavethereforeassumedthatthemapping 16

    fromSNAtoSCR(butnotnecessarilyfrompsychologicalvariabletoSNA)iswell 17

    describedbyaLTIsystem. 18

    19

    Twotypesofresponsefunctionshavebeenproposed:onebasedonabiophysicalmodel 20

    ofthesweatgland,withparameterssetfromtheoreticalconsiderations(Alexanderet 21

    al.,2005;Benedek&Kaernbach,2010a,2010b;Grecoetal.,2015),andapurely 22

    phenomenologicalfunctionwithparametersfittedtoadatabaseof1,278SCRsfrom64 23

    individualsinsixdifferentexperimentalconditions(Bach,Daunizeauetal.,2010;Bach 24

    etal.,2009;Bach,Flandinetal.,,2010).Whiletheformerapproachappearstheoretically 25

    morerigorous,manyofthebiophysicalparameterswerenotknownfromphysiological 26

    researchandhadtobeguessed.Theensuingforwardmodelhasnotbeensystematically 27

    evaluated,anditisunclearhowwellthisresponsefunctionfitsactualSCR.Incontrast, 28

    thelattermodelisdefinedbyitsfittoempiricaldata.Thisphenomenologicalresponse 29

    functionismathematicallydescribedbyaGaussian-smoothedexponential(Bach, 30

    Flandinetal.,2010)orathird-order(linear,constant-coefficient)ordinarydifferential 31

    equation;seeAppendix-Equation6,7). 32

    33

    TherearegoodempiricalargumentstomotivatetheuseofLTIsystemstomodelthe 34

    SNA/SCRrelationship,providedthatSCRdataarehigh-passfiltered.Threekindsoftests 35

    havebeenexploitedtoevaluatetheforwardmodel.Indirecttestsmadetheauxiliary 36

  • 14

    assumptionthatSNburstsfollowexternalstimulationwithconstantshapeandlatency 37

    (thisassumptionisnotpartoftheLTIsystem).Thesetestsshowed,thatforshortevents 38

    (<1sduration)thatareseparatedbyatleast30s,morethan60%ofthevariancein 39

    (high-passfiltered)SCRcanbeexplainedunderaLTImodel(Bach,Flandinetal., 40

    2010),supportingtheplausibilityofthetimeinvarianceapproximation.Intheabsence 41

    ofstimulation,baselinevarianceexceededtheresidualvarianceunderstimulation, 42

    implyingthattheresidualvarianceisduetonoiseratherthanLTIviolations(Bach, 43

    Flandinetal.,2010).SCRtopairsofstimuliseparatedbydifferentintervals(2-9s)do 44

    notdependontheinterval,inlinewiththelinearityprinciple(Bach,Flandinetal., 45

    2010).Amoredirecttestofthetimeinvarianceprincipleisfurnishedbyintraneural 46

    recordings,whichshowthat60%-75%ofSCRvarianceisexplainedbyaLTImodelthat 47

    takesSNactivityasinput,althoughthisisstillsufferingfrominterferingnon-SN(e.g., 48

    vasomotornerve)activity(Gersteretal.,2017).Athirdapproachreliesonintraneural 49

    stimulationwhileblockinginterferingnervetrafficbyregionalanesthesia.Here,SNis 50

    stimulatedatdifferentrepetitionfrequencies,thussimultaneouslyaddressingthe 51

    linearityandtimeinvarianceprinciple.Inthiscase,93%-99%ofSCRvarianceis 52

    explainedunderaLTImodelwhenstimulationfrequencyisbelow0.6Hz.Abovethis 53

    stimulationfrequency,theLTImodelcannotbeusefullyappliedduetostrong 54

    nonlinearities(Gersteretal.,2017);however,thislimitationshouldbelargelyirrelevant 55

    formostpsychologicalexperimentswithslowerstimulationrates.Tosummarize,it 56

    appearsthatundersuitableconditions,aLTImodelisnotjustanapproximationbut 57

    ratheranaccuratedescriptionofbiophysicalrealityintheSNA/SCRsystem.Notably, 58

    tonicskinconductancecomponents,whicharefilteredoutinallmodelingapproaches, 59

    donotappeartobemodeledbyafiniteLTIsystem. 60

    61

    TheLTImodeldoesnotmaketheassumptionthatSCRshapeisconstantbetween 62

    individuals.Onthecontrary,allevidencesuggeststhatthisisnotthecase.Nevertheless, 63

    inferencebasedonacanonicalresponsefunctionalreadyprovidesbetterinferencethan 64

    operationalmethods.Wehaveshownthatthisinferencecanbefurtherimprovedby 65

    allowingsomevariabilitybetweenindividuals,butonlyifthepossibleindividual 66

    responsefunctionsareverystronglyconstrainedtoavoidoverfitting(Bachetal.,2013; 67

    Staib,Castegnetti,&Bach,2015). 68

    69

  • 15

    ForthemappingfrompsychologicalvariabletoSNA,threedifferentforwardmodels 70

    havebeendevelopedandevaluated.First,shortstimulielicitrapidSNAwithconstant 71

    Q6latency(Bachetal.,2009).Itappearsthatthemappingfrompsychologicalvariable 72

    toSCRislargelyinvarianttothetypeofexperimentstimulus(aversivewhitenoise 73

    bursts,aversiveelectricstimulation,aversivepictures,auditoryoddballs,andavisual 74

    detectiontask;Bach,Flandinetal.,,2010),whichmotivatestheuseofthismodelfor 75

    phasicarousalindependentoftheelicitingstimulusorexperiment.Forlongerstimulior 76

    anticipationofstimuli,amodelwithvariableSNAlatencyandshapecanbeused,and 77

    thesetwoparametersareestimatedfromtheexperimentaldata(Bach,Daunizeauetal., 78

    2010).Thismodelwasmotivatedspecificallytoanalyzefearconditionexperimentsin 79

    whichparticipantsareexposedtoconditionedstimuli(CS),oneofwhich(CS+)predicts 80

    anaversiveevent(US).Typically,thereisatimedelaybetweenCS+andUS,andso 81

    participantswillanticipatetheUSduringCS+and(toadiminishingextent)duringCS- 82

    presentationandexpressSCRatsome(unknownandpossiblyvariable)timepoint 83

    duringthisinterval.Notably(anddifferentfrommodelsdiscussedlater),responsesto 84

    CS+andCS-arethoughttobegovernedbythesamepsychologicalprocess,buttobe 85

    quantitativelydissimilar.Hence,inthisapplicationofthemodel,aconditioned 86

    sudomotornerveresponsetoeachCSisestimated,andthedifferenceintheiramplitude 87

    betweenCS+andCS-constitutestheinferenceonfearmemory.Finally,toaccountfor 88

    spontaneousSCRfluctuations,amodelisproposedinwhichconstant-shapeSNbursts 89

    occurwithvariableonsetandamplitude,whichareestimatedfromthedata(Bach, 90

    Daunizeauetal.,,2011). 91

    92

    4.1.2Inferenceonpsychologicalvariables 93

    Theconstant-latencyforwardmodelisinvertedinaGLMapproach(Bachetal.,,2009) 94

    andhasbeenevaluatedonindependentdatasets(Bach,2014;Bachetal.,2013)in 95

    comparisontodifferentpeak-scoringmeasures(Boucseinetal.,2012)andtomeasures 96

    fromthehybridLedalabapproach(Benedek&Kaernbach,2010a,,2010b).Predictive 97

    validitywasassessedforthecomparisonofnegativearousingversusneutralpicture 98

    presentation,positivearousingversusneutralpictures,picturepresentationversusno 99

    stimulus,andfearfulversusangryfacepresentation.Predictivevaliditywasdecisively 100

    higherfortheGLM-basedapproachonthemajorityofcomparisons,anditwasnever 101

    decisivelysurpassedbyanothermethodintheremainingcomparisons(Bach,2014).In 102

    anevaluationstudyfromanindependentlaboratory,aGLMapproachhadhigher 103

  • 16

    predictivevaliditythanpeak-scoringorLedalabfordistinguishingCS+andCS-in 104

    aversivelearning(Greenetal.,,2014).Fordistinguishingfivedifferentphasesofafear 105

    generalizationexperiment,peakscoringappearedtohavehigherpredictivevalidity 106

    (Greenetal.,2014).However,notethatthepredictivevalidityevaluationassumesthat 107

    distinguishabledifferentpsychologicalstatesarecreatedbytheexperiment,whichis 108

    lesswellestablishedforthelattermanipulation.Thisstudydidnotallowforassessing 109

    whetheranymethodwasdecisivelybetterorworsethananother.3 110

    111

    Theflexible-latencyforwardmodelisinvertedinavariationalBayesapproach(Bach, 112

    Daunizeauetal.,2010)andwasevaluatedonindependentfearconditioningdatasets 113

    (Staibetal,2015)incomparisontodifferentpeak-scoringmeasures(Boucseinetal., 114

    2012)andtomeasuresfromthehybridLedalabapproach(Benedek&Kaernbach, 115

    2010a,2010b).PredictivevaliditywasassessedforthecomparisonofCS+versusCS-, 116

    andwasfoundtobedecisivelyhigherforthemodel-basedapproachthanforpeak- 117

    scoringorLedalabmeasures(Staibetal.,2015). 118

    119

    Finally,theflexible-onsetmodelisinvertedwiththesamevariationalBayesapproach 120

    (Bach,Daunizeauetal.,2011)andcanbeusedtoinfertonicarousal,whichisoften 121

    operationalizedasthenumberofspontaneousskinconductancefluctuations(Boucsein 122

    etal.,2012).Thismethodwasevaluatedwithrespecttodistinguishingpublicspeaking 123

    anxietyfromrestornonpublicspeakinganxiety,andmentalloadfromrest.Predictive 124

    validityofthemodel-basedapproachandofanautomatedpeak-countmeasurewas 125

    comparable(Bach&Staib,2015).Furthermore,undertheperipheralLTImodel,the 126

    areaunderthecurveofatimeseriesequalsthenumberofspontaneousfluctuations 127

    timestheiramplitude.Whilethisrelationhasbeenempiricallyvalidated,thesamestudy 128

    alsoshowedthatthenumberofspontaneousfluctuationsaloneallowsbetterinference 129

    ontonicarousalthantheareaunderthecurve(Bach,Friston,&Dolan,2010). 130

    131

    4.1.3Futuredirections 132

    3Bayesfactorsarereportedinthisstudyaswell,buttheyarenotcomparablebetweenthemodelsevaluated.Comparisonofmodelevidenceisonlypossibleifthedependentvariableinthemodelisthesame(Burnham&Anderson,2004)butGreenetal.(2014)useestimatesofpsychologicalstateasdependentvariable,whichareobviouslydifferentbetweenmethods.

  • 17

    Theconstant-latencyGLMapproachappearsfairlymature.Ithasbeenoptimizedwith 133

    respecttomodelcomplexityanddatapreprocessing(Bachetal.,,2013)andevaluated 134

    bytwodifferentlaboratories(Bach,2014;Greenetal.,2014).Currentresearchfocuses 135

    onincrementalimprovements,suchasdatapreprocessingandmodelingof 136

    nonlinearitiesunderspecificconditionssuchasrapideventsuccession,orwhenthe 137

    sweatductsqualitativelychangetheirresponsebehavior(Tronstad,Kalvoy,Grimnes,& 138

    Martinsen,2013). 139

    140

    Theflexible-latencymodelisalsoinamaturestageandhasbeenoptimized(Staibetal., 141

    2015),butaformalevaluationbydifferentresearchlaboratoriesislacking.Intermsof 142

    theforwardmodel,aquestionremainsastounderwhatconditionsadditional 143

    complexityaffordedbyestimatingresponselatencyimprovesinferenceonthe 144

    psychologicalvariable(i.e.,Whatisthelevelofvariabilityinresponselatencythat 145

    shouldmotivatepreferringthismethodtotheconstant-latencyGLMapproach?).This 146

    questionisawaitingempiricalinvestigation.Aweaknessoftheflexible-latencymodelis 147

    itscomplexity,suchthatitisimpossibletoestimateallparametersatthesametimeon 148

    standardPCsandclustercores.Parametersarethereforeestimatedinatrial-by-trial 149

    fashionsuchthatestimationerrorsthatoccurinonetrialwillpropagateintothenext 150

    one.Whilesometechnicaltricksreduceadetrimentalimpactofthissequential 151

    estimation,itmaybepossibletoimproveonparameterestimationwithaglobal 152

    optimizationmethodthatevaluatestheentireparameterspaceatthesametime. 153

    154

    Finally,theflexible-onsetmodelistheleastmatureandhasbeenevaluatedononlya 155

    smallnumberofquestionsanddatasets;thereisroomforoptimizationofthismethod. 156

    157

    BeyondSCR,othermeasuresalsoallowinferenceonSNAand,thus,onthesame 158

    psychologicalvariables.Itremainstobedeterminedwhethermeasuressuchasskin 159

    potentialandskinsusceptance(Tronstadetal.,2013)orthememristorpropertiesofthe 160

    skin(Pabst,Tronstad,&Martinsen,2017)yieldadditionalinformationorcanhelp 161

    reducenoiseintheinverseinference. 162

    163

    4.2Pupilsize 164

    4.2.1Forwardmodel 165

  • 18

    Awiderangeofpsychologicalvariablesimpactsonpupilsize(e.g.,deGeeetal.,2014; 166

    Joshi,Li,Kalwani,&Gold,2016).Pupilsizeiscontrolledbytwoantagonistmuscles:the 167

    radialM.dilatatorpupillae,whichreceivessympatheticinnervationviapreganglionic 168

    neuronsfromthespinalcordandpostganglionicneuronsfromthesuperiorcervical 169

    ganglion,andthecircularM.sphincterpupillae,whichreceiveparasympatheticinput 170

    frompreganglionicneuronsin 171

    theEdinger–Westphalnucleuswithinthemidbrainandpostganglionicneuronsinthe 172

    ciliaryganglion(McDougal&Gamlin,2008).TheEdinger–Westphalnucleusreceives 173

    bothluminance-mediatedinputsfromtheretinaandappearstorelayinputsthatarenot 174

    relatedtoluminance(e.g.,fromthelocuscoeruleus;Joshietal.,2016;Liu,Rodenkirch, 175

    Moskowitz,Schriver,&Wang,2017).Thisinnervationmotivatesusingluminance 176

    changestoprobethebiophysicsofphasicpupilresponses.Thefactthatthepupilis 177

    controlledseparatelybybothbranchesoftheautonomicnervoussystem,andby 178

    antagonisticmuscleswithdifferentmechanicalproperties,alreadysuggeststhatpupil 179

    responsesmaybestbemodeledbytwoparallelLTIsystems.Asaparsimonious 180

    description,amodelisproposedthatdoesnotsplitthesystemintocontributionsofthe 181

    twomusclesbutratherseparatesaslowerdilation/constrictionresponsefromafaster 182

    componentthatonlyoccursforconstrictions(seeAppendixEquation9,10and 183

    parametersforPSR_dilandPSR_con;Korn&Bach,2016).Interestingly,theformer 184

    componentdirectly(exponentially)relatestoluminance,whilethecontributionofthe 185

    lattercomponentappearsindependentfromtheamountofluminancechange(Korn& 186

    Bach,2016).Becauseofthedifferenttimeconstantsofthetwosystems,themodel 187

    makesaninterestingcounterintuitiveprediction:lightflashesshouldleadtopupil 188

    constriction,butbriefdarknessperiodsshouldalsoleadtoconstriction,whenthefaster 189

    responseinducedbythereturntolightprecedestheslowerresponseinducedbythe 190

    darknessperiod.Indeed,thispredictionwasconfirmedbyexperimentalobservationin 191

    ourownlaboratory(Korn&Bach,2016)andbyothers(Barbur,Harlow,&Sahraie, 192

    1992).Thisforwardmodelexplainedaround60%ofsignalvariance,speakingtothe 193

    validityofLTIapproximation(Korn&Bach,2016).Furthermore,itwasusedtoinferthe 194

    neuralinputintothepupilsystemfordifferentpsychologicaltasks(visualdetection, 195

    auditoryoddballdetection,listeningtoemotionalwords).Theinferredinputlatency 196

    meaningfullyrelatedtoknownunderlyingpsychologicalprocesses(Korn&Bach,2016). 197

    Thisalsosuggeststhatthefasttimecourseofpupilresponsesimpliesanecessityto 198

    buildneuralforwardmodelsthataretosomeextentspecifictothepsychological 199

  • 19

    processstudied-different,forexample,fromtheunspecificSCRmodels-becausethe 200

    timeconstantsofthepsychologicalprocessesdiffer.Onesuchmodelwasdevelopedfor 201

    fearconditioning(Korn,Staib,Tzovara,Castegnetti,&Bach,2017).Here,wefoundthat 202

    CS-responsesaredifferentbetweenexperiments,dependingontheperceptualmodality 203

    andphysicalpropertiesoftheCS.However,theaddedimpactoftheCS+wasrather 204

    constantacrossexperimentsandcouldbemodeledwithaconstant-latencyCS-evoked 205

    neuralinputthatpeaksbetweenCSandUS(Kornetal.,2017).Thisneuralinput 206

    appearedtobetime-lockedtotheCS,fordifferentintervalsbetweenCSandUS. 207

    DifferentfromSCRmodels,thismeansthatthereisnocommonCSresponsethatdiffers 208

    onlyinamplitudebetweenconditions.Instead,themodelseekstoestimatetowhat 209

    extentaspecificCS+componentisexpressedoneach(CS+andCS-)trial.SincethisCS+ 210

    componentmaynotbeorthogonaltotheexperiment-specificCSresponse,thismeans 211

    thatallestimates(CS+andCS-)areonlyinterpretableuptoaconstant:itispossibleto 212

    interpretdifferencesbetweentrialsets,ortemporalchangesintheestimatedCS 213

    response,butnotthemagnitudeofresponseestimatesforindividualtrialsor 214

    conditions.Formathematicalconvenience,neuralandperipheralmodelarecollapsed 215

    intooneresponsefunction,whichmakesGLMinversionpossible(Kornetal.,2017;see 216

    AppendixEquation9andparametersforPSR_FC).Anotherpsychologicalmodelwith 217

    similarstructurebutadifferentresponsefunctionwasproposedtocapturetheimpact 218

    ofattentiononpupilsize(Hoeks&Levelt,1993). 219

    220

    4.2.2Inferenceonpsychologicalvariables 221

    PupilPsPMshavebeenusedfortwopurposes.Thefirstisdirectinferenceona 222

    psychologicalvariable.InferenceonCS+memoryinfearconditioningisimplementedin 223

    aGLMinversionapproach,andyieldspredictivevaliditysuperiortopeakscoringor 224

    area-under-the-curvemeasures(Kornetal.,2017).Giventheshortlatencyandsignal- 225

    to-noiselevelssimilarorbetterthanSCR,inferencecanbeperformedonasingle-trial 226

    level.Similarly,inferenceonattentionhasbeenusedinnumerousstudies(e.g.,deGeeet 227

    al.,2014,2017;Knapenetal.,2016),althoughtoourknowledgethismethodhasnot 228

    beensystematicallyinvestigatedandvalidatedbeyondtheinitialdevelopmentdataset 229

    (Hoeks&Ellenbroek,1993;Hoeks&Levelt,1993).Anotherapplication,distinctfromall 230

    othermodelspresentedinthisarticle,istoinferthetimecourseofapsychological 231

    process.Thisispossiblebecauseluminance-relatedresponsesaremediatedbynear- 232

    instantaneousneuralactivityandtherebyallowcharacterizingpupilbiomechanics. 233

  • 20

    Comparingameasuredpupiltimecoursewiththetimecourseofaluminance-related 234

    responsethereforeaffordsestimatingtheneuralinputintothepupillarysystem,and 235

    thusmakesinferenceonthedynamicsoftheunderlyingpsychologicalprocess(Korn& 236

    Bach,2016). 237

    238

    4.2.3Futuredirections. 239

    Pupil-sizemodelingisarelativelynewapproachandrequiresfurtherstudy. 240

    Importantly,anindependentvalidationofpsychologicalinferenceisyetlacking.PsPMs 241

    existforluminance-conditioned,fear-conditioned,andattention-relatedresponses,and 242

    appeartocruciallydependonthetimingofthepsychologicalprocessunderstudy. 243

    Potentially,pupil-sizemodelingthusoffersamoreprecisewindowintothetemporal 244

    dynamicsofcognitiveprocessesthanmanyotherpsychophysiologicalvariables.Finally, 245

    alotofresearchiscurrentlybeingdoneonpupilsizeanditsrelationtocognitive 246

    processes,inhumansandotherspecies(e.g.,Eldar,Cohen,&Niv,2013;Joshietal., 247

    2016).Itislikelythatnewmodelsandmethodswillemanatefromthisbasicresearch. 248

    249

    4.3Heartperiod 250

    4.3.1Forwardmodel 251

    Heartrateorheartperiodareoftenusedtoinferemotionalarousal,forexample,while 252

    watchingpicturesorduringfearconditioning(Berntson,Quigley,&Lozano,2007; 253

    Bradley,Codispoti,Cuthbert,&Lang,2001)andcanbemeasuredwith 254

    electrocardiogramorpulseoxymeters.Cardiacrhythmisgeneratedlocallyintheheart, 255

    butmodulatedunderslowersympatheticandfasterparasympatheticinfluence 256

    (Akselrodetal.,1981).Sympatheticstimulationfrequencyappearstolinearlyscalewith 257

    heartperiodchanges,notheartrate(Berntson,Cacioppo,&Quigley,1995).Therefore 258

    currentPsPMsforphasiccardiacresponsesmodelheartperiod,whichismappedonto 259

    thefollowingRspikeandlinearlyinterpolated.Itappearsthatvariousshortstimuli 260

    inducephasicheartperiodresponses(HPR),andsixresponsecomponentscouldbe 261

    identifiedinasystematicinvestigationseeAppendixEquation8andparametersfor 262

    HPR_E1-E6).However,neitherthisstudynorpreviousresearchbasedonoperational 263

    methodsallowadefiniteconclusionastowhichpsychologicalvariablesinfluenceeach 264

    ofthecomponentsandrelatedly,whetherthesecomponentsareindependently 265

    controlled.Incontrast,awell-replicatedphenomenonisfear-conditionedbradycardia 266

  • 21

    (Castegnettietal.,2016).Thisbradycardiaresponseappearstobeaddedtoastimulus- 267

    specificHPRthatoccursforbothCS+andCS-infearconditioning,similartowhatis 268

    observedforpupilsizeandwiththesamelimitationsforinterpretationofresponse 269

    estimates.However,differentfrompupilresponses,itappearstobetime-lockedtothe 270

    USwhentheCS-USintervalisvaried(Castegnetti,Tzovara,Staib,Gerster,&Bach,2017; 271

    Castegnettietal.,2016).Furthermore,relatingthebradycardiaresponsetothefirst 272

    responsecomponentelicitedbyshortstimulirevealedaputativeneuralinputthat 273

    peaksattheUS(Castegnettietal.,2016).Pragmatically,thePsPMforfear-conditioned 274

    HPRcollapsesaneuralandperipheralsystemintooneresponsefunction,allowingGLM 275

    inversion(seeAppendixEquation9andparametersforHPR_FC). 276

    277

    4.3.2Inferenceonpsychologicalvariables 278

    TherangeofpsychologicalvariablesthatcanbeinferredfromphasicHPRtoshort 279

    stimuliappearsunclearatpresent.Fear-conditionedbradycardiaisanotable,well- 280

    studiedexception.UsingtheGLMapproach,fearmemorycouldbeinferredwithhigher 281

    predictivevaliditythanwithpeak-scoringmethods(Castegnettietal.,2016).Unlikefor 282

    SCRandPSR,attemptstoperformsingle-trialanalysesinourownlaboratoryhavenot 283

    succeeded,probablybecauseheart-periodtimeseriesaredominatedbyrespiratory 284

    arrhythmia,andmanytrialsarerequiredtoreducetheimpactofthisnoisecomponent. 285

    286

    4.3.3Futuredirections 287

    ElucidatingtheforwardmodelforHPRappearsanimportanttaskbothinthecontextof 288

    PsPMsandoperationalapproaches.Thefidelityofinferenceonfearmemoryhasbeen 289

    demonstratedinseveraldatasetsbutrequiresindependentconfirmationfromdifferent 290

    laboratories.Castegnettietal.(2017)havesuggestedthatfear-conditionedbradycardia 291

    couldpotentiallybe-atleastpartly-inducedbyincreasedthoraxpressureinducedvia 292

    respirationamplituderesponses;thismaybeanotherinterestingavenueofresearch. 293

    294

    4.4Respirationmeasures 295

    4.4.1Forwardmodel 296

    Mostrespiratorypsychophysiologyresearchhasfocusedonhowpsychologicalstates 297

    onatimescaleof10-20suptominutesinfluencerespirationparameters(Boiten, 298

    Frijda,&Wientjes,1994;Grassmann,Vlemincx,vonLeupoldt,&VandenBergh,2015; 299

    Ritzetal.,2010;Vlemincx,VanDiest,&VandenBergh,2015;Wuyts,Vlemincx,Bogaerts, 300

  • 22

    VanDiest,&VandenBergh,2011),butrelativelylittleisknownaboutphasic 301

    respiratoryresponses.PsPMshavebeendevelopedforrespiratoryperiod,respiratory 302

    amplitude,andrespiratoryflowrateresponsestobriefexternalevents,allmeasured 303

    withasimplesingle-chestbeltsystemasstandardlyemployedinfMRIlaboratories(see 304

    AppendixEquation8andparametersforRPR,RAR_E,andRFRR).Externaleventscause 305

    responsesinthesethreemeasuresthatarecapturedwithLTIsystems,butasyetitis 306

    notclearwhichpsychologicalstatescouldbeinferredfromtheseresponses(Bach, 307

    Gerster,Tzovara,&Castegnetti,2016).Incontrast,wehaveshowninseveral 308

    experimentsthataCS+infearconditioningelicitsabiphasicrespiratoryamplitude 309

    responsethatcanbemodeledinaLTIsystem,thusallowingGLMinversion(Castegnetti 310

    etal.,2017)(seeAppendixEquation9andparametersforRAR_FC).Theapproachand 311

    itsinterpretationaresimilartothatforpupilsizeandheartperiod. 312

    313

    4.4.2Inferenceonpsychologicalvariables 314

    Itappearsthatfearmemorycanbeinferredfromrespirationamplituderesponses,and 315

    predictivevalidityofthisinferenceishigherforaGLMinversionthanpeakscoring 316

    (Castegnettietal.,2017);however,itislowerthanformanyotherpsychophysiological 317

    measures.AdistinctadvantageoftherespiratoryPsPMcouldbethatitonlyrequires 318

    single-chestbeltdata,whichisstandardlyavailableinmanyMRIscanners. 319

    320

    4.4.3Futuredirections 321

    Moreresearchisrequiredtoelucidatetherangeofpsychologicalvariablesthatcanbe 322

    inferredfromrespiratorymeasures.Modelingmoresophisticatedrespiratorymeasures 323

    couldbeaninterestingpossibility. 324

    325

    4.5Startleeyeblinkelectromyogram 326

    4.5.1Forwardmodel 327

    Differentfromthepreviouslydiscussedmeasures,whichareunderthedirectinfluence 328

    ofapsychologicalvariable,theimpactofpsychologicalvariablesonstartleeyeblinkis 329

    onlymodulatoryandrequireselicitationofastartleresponsetorevealit.Thisstartle 330

    eyeblinkresponse(SEBR)itselfhasratherstereotypicaldynamics,whileitsamplitudeis 331

    modulatedbydifferentpsychologicalvariables(Yeomans,Li,Scott,&Frankland,2002). 332

    Thismodulationhasbeensuggestedtobalancetheprotectiveutilityofthestartle 333

    responsewithitsmetabolicandopportunitycost(Bach,2015b).Importantly,SEBR 334

  • 23

    dissociatesCS+andCS-infearconditioning,aphenomenontermedfear-potentiated 335

    startle(Brown,Kalish,&Faber,1951).APsPMwasdevelopedtomodeltheimmediate, 336

    briefSEBRtostartleprobesintheabsenceofanypsychologicalmanipulation(Khemka 337

    etal.,2017).Thismodelexplainedabout60%ofsignalvarianceunderLTIassumptions. 338

    Forinference,theneuralinputwasallowedaflexiblelatency,tobettercaptureslight 339

    latencyvariationbetweenindividualsandtrials(Khemkaetal.,2017).Ascommoninthe 340

    literature,themodelassumesthattheshapeofSEBRisindependentofthepsychological 341

    orcognitivestate,whichonlyimpactsonitsamplitude. 342

    343

    4.5.2Inferenceonpsychologicalvariables 344

    ThisPsPMwasemployedtoinferfearmemory,bothduringacquisitionandmemory 345

    recallunderextinction(Khemkaetal.,2017).Predictivevalidityoftheinferencewas 346

    comparedtofourpeak-scoringmethodswithdifferentpreprocessingsteps.Foreachof 347

    threeexperiments,adifferentpeak-scoringmethodperformedbest,butacrossall 348

    experiments,thePsPMapproachyieldedhighestpredictivevalidity(Khemkaetal., 349

    2017). 350

    351

    4.5.3Futuredirections 352

    TheimpactofdifferentpreprocessingmethodsonSEBRanalysisappearsnotwell 353

    understood.ThisleadstoaheterogeneouspicturewhencomparingPsPMwithdifferent 354

    peak-scoringmethods,andshouldbeafocusoffutureresearch.Startle-independent 355

    eyeblinksareatypicalsourceofnoiseinSEBRresearch,andmodelingtheseeyeblinks 356

    couldbeanimportanttopicforfurtherinvestigation. 357

    358

    4.6Combiningpsychophysiologicalmodels 359

    Inpsychophysiologicalresearch,differentmeasurementmethodsareoftenused 360

    simultaneously(inthespiritofconvergentoperationalization),butnotcommonly 361

    combinedforstatisticalinferenceonapsychologicalvariable.InaPsPMapproach,this 362

    maybeapossibilityandcouldimproveinferenceundercircumstanceswhereseveral 363

    measuresareindicativeofthesamepsychologicalvariable.Inordertoenablesuch 364

    combination,itwouldbedesirabletoclarifyqualitativelythattwomeasuresare 365

    impactedbythesamepsychologicalvariable,andtoinvestigatequantitativelythe 366

    dimensionalstructureofthesemeasuresacrossdifferentindividuals. 367

    368

  • 24

    3695Discussion 370

    Psychologicalinvestigationreliesonsolvingtheinverseproblem:makinginferenceon 371

    essentiallyunobservablepsychologicalvariables.Psychophysiologybenefitsfrommany 372

    decadesofresearchontheforwardmappingfrompsychologicalvariablesonto 373

    physiologicalmeasures.Thishasallowedthebuildingofpreciseandexplicitforward 374

    models,whichcanbespecifiedinmathematicalformandinvertedtoyieldinferenceon 375

    thepsychologicalvariable:psychophysiologicalmodeling.Buildingonsimple 376

    experimentalmanipulationsthatyieldaknownpsychologicalstate,methodscanbe 377

    evaluatedintermsoftheirpredictivevalidity(i.e.,thefidelitywithwhichtheyrecover 378

    theknownstate).Thisallowscomparisonofoperationalaswellasmodel-based 379

    methodsandhasrevealedthatinmanycasesPsPMsallowmorepreciseinferencethan 380

    traditionaloperationalmethods,forexamplepeakscoring. 381

    382

    Asalimitation,thisconclusionisbasedonalimitednumberofstudiesanddatasets, 383

    mostofwhich-withoneexception(Greenetal.,2014)-comefromourlaboratoryand 384

    arethusbasedonthesamerecordingequipmentandrathercomparableexperimental 385

    procedures.ItispossiblethatthePsPMsandinversionmethodsdevelopedinthis 386

    contextoverfittheseexperimentalcircumstancesanddonotgeneralizewelltodata 387

    acquiredindifferentcontexts,forexample,usingdifferentexperimentaltimingsor 388

    involvingothertypesofartifacts.Notably,thesamelimitationappliestothevarietyof 389

    operationalmethodsthathaveoftenbeendevelopedinandforspecificlaboratories 390

    suchthatamultitudeofoperationalanalysismethodscoexistsintheliterature.Wehave 391

    providedexamplesforapplicationofthePsPMsindifferentlaboratories,whichprovides 392

    circumstantialevidencethatoverfittingisnotamajorissuefortheapproachpresented 393

    here.However,toentirelyruleoutsuchpossibility,itwouldbedesirabletocomparethe 394

    PsPMswithdifferentoperationalanalysesmethodsinmanyexperimentalsituations. 395

    396

    Researchonpsychophysiologicalmodelingunderlinesthenecessityforprecise 397

    specificationofforwardmodel,andthusforthedetailedandmeticulousworkofbasic 398

    psychophysiology.TheapplicationofPsPMsremainsrestrictedtosituationsinwhich 399

    thismappingiswellknown.Indeed,PsPMsexistonlyforasmallnumberof 400

    experimentalscenarios.However,theseincludestandardexperimentssuchasfear 401

    conditioningandpictureviewing,andmaywellcomprisealargeproportionofapplied 402

  • 25

    psychophysiologyresearch.Forthese,PsPMsofferimprovedinferencecomparedto 403

    currentlyusedmethods.Forexploratoryresearch,operationalmethodswiththeir 404

    flexibilitymaybemoreappropriate.However,asanadvantageofmodel-basedmethods 405

    theirprecisespecificationimplementationreducesresearcherdegreesoffreedom. 406

    Analysisflexibilityisaproblemrecognizedacrosstheentirefieldofpsychology 407

    (Simmons,Nelson,&Simonsohn,2011)andpossiblymoreprevalentwithflexible 408

    operationalmethods.Whilestandardizationofmethodsisanongoingeffort(Boucsein 409

    etal.,2012;Lonsdorfetal.,2017),PsPMoffersrationalcriteriabeyondcommunity 410

    consensusforchoosingthestandards,namely,thequalityoftheinference.Notably,all 411

    methodsdiscussedinthisreviewareavailableinopen-sourcetoolboxes,mostofthem 412

    intheMatlab-basedPsPMtoolbox(pspm.sourceforge.net). 413

    414

    Asaframeworkforevaluatingmethods,wehaveproposedtobenchmarktheir 415

    predictivevalidity,thatis,theirabilitytorecoverknownpsychologicalstates(Bach& 416

    Friston,2013).Wenotethatthisframeworkhaspotentiallymanymoreapplications 417

    thanevaluatingPsPMsorcomparingthemtooperationalmethods.Forexample,it 418

    allowspoweranalyses.Ifthefidelityofamethodisknowninastandardexperiment,it 419

    isoftenpossibletoderivebest-caseadditionalassumptionsthatdefineminimum 420

    samplesizesrequiredtoachieveadesiredpowerlevel.Wehaveprovidedan 421

    experimentalexampleforthisinthecontextofaninterventiontoreducesynaptic 422

    plasticityduringfearconditioning,asmeasuredwithSEBR(Bachetal.,2017).Asan 423

    importantinsight,poweranalysisbasedonpredictivevalidityresearchhasrevealed 424

    thattherequiredsamplesizes-especiallywhenusingtraditionaloperationalmethods- 425

    canbemuchhigherthanwhatisthestandardinthefield.Consequently,studiesnot 426

    basingtheirsamplesizeonthisorotherformalanalysesmaybeunderpowered. 427

    Anotherpotentialapplicationisqualitycontrol.Labscancomparetheirmeasurement 428

    methodsbetweeneachother,assuremeasurementfidelityovertime,orbenchmark 429

    researchtrainees,usingpredictivevalidityofstandardmeasures.Finally,apotentially 430

    powerfuluseofthisframeworkistheoptimizationofexperimentaldesign.Iftheeffect 431

    ofastandardexperimentonapsychologicalvariableisknownapriori,thenonecan 432

    chooseanexperimentaldesignthatbestallowsdetectingthiseffect.Asanexample,this 433

    approachcanhelptofindtheoptimalbalanceofretentiontrialstomeasurefear 434

    memoryrecall,forwhichwehaveprovidedanempiricalexample(Khemkaetal.,2017). 435

    436

  • 26

    Tosummarize,thefieldofpsychophysiologicalmodelingismovingrapidlyandhasin 437

    partsalreadymatured.Withthesedevelopments,wehopetohaveprovidedtask- 438

    unspecifictoolsthatfreeresearchers'resourcesfromhavingtodevelopdataanalysis 439

    proceduresforeverystudy,andinsteadfocusonthepsychologicalorcognitive 440

    questionstheywanttoanswer. 441

    442

    Acknowledgments 443

    Theauthorsthanktheircolleagueswhoovertheyearshelpedestablishingthesoftware 444

    PsPMandthemethodsdescribedinthisreview:JeanDaunizeau,RaymondJ.Dolan, 445

    GuillaumeFlandin,KarlJ.Friston,GabrielGräni,SaurabhKhemka,PhilippC.Paulus, 446

    LinusRüttimann,MatthiasStaib,AthinaTzovara. 447

    448

    References 449

    Akselrod,S.,Gordon,D.,Ubel,F.A.,Shannon,D.C.,Berger,A.C.,&Cohen,R.J.(1981). 450Powerspectrumanalysisofheartratefluctuation:aquantitativeprobeofbeat- 451to-beatcardiovascularcontrol.Science,213,220-222. 452

    Alexander,D.M.,Trengove,C.,Johnston,P.,Cooper,T.,August,J.P.,&Gordon,E.(2005). 453Separatingindividualskinconductanceresponsesinashortinterstimulus- 454intervalparadigm.JournalofNeuroscienceMethods,146,116-123. 455

    Alvarez,R.P.,Kirlic,N.,Misaki,M.,Bodurka,J.,Rhudy,J.L.,Paulus,M.P.,&Drevets,W.C. 456(2015).Increasedanteriorinsulaactivityinanxiousindividualsislinkedto 457diminishedperceivedcontrol.Translationalpsychiatry,5,e591. 458

    Bach,D.R.(2014).Ahead-to-headcomparisonofSCRalyzeandLedalab,twomodel- 459basedmethodsforskinconductanceanalysis.BiolPsychol,103C,63-68. 460

    Bach,D.R.(2015a).Anxiety-LikeBehaviouralInhibitionIsNormativeunder 461EnvironmentalThreat-RewardCorrelations.PLoSComputBiol,11,e1004646. 462

    Bach,D.R.(2015b).AcostminimisationandBayesianinferencemodelpredictsstartle 463reflexmodulationacrossspecies.JTheorBiol,370,53-60. 464

    Bach,D.R.,Daunizeau,J.,Friston,K.J.,&Dolan,R.J.(2010).Dynamiccausalmodellingof 465anticipatoryskinconductanceresponses.BiolPsychol,85,163-170. 466

    Bach,D.R.,Daunizeau,J.,Kuelzow,N.,Friston,K.J.,&Dolan,R.J.(2011).Dynamiccausal 467modelingofspontaneousfluctuationsinskinconductance.Psychophysiology,48, 468252-257. 469

    Bach,D.R.,Flandin,G.,Friston,K.J.,&Dolan,R.J.(2009).Time-seriesanalysisforrapid 470event-relatedskinconductanceresponses.JNeurosciMethods,184,224-234. 471

    Bach,D.R.,Flandin,G.,Friston,K.J.,&Dolan,R.J.(2010).Modellingevent-relatedskin 472conductanceresponses.InternationalJournalofPsychophysiology,75,349-356. 473

    Bach,D.R.,&Friston,K.J.(2013).Model-basedanalysisofskinconductanceresponses: 474Towardscausalmodelsinpsychophysiology.Psychophysiology,50,15-22. 475

    Bach,D.R.,Friston,K.J.,&Dolan,R.J.(2010).Analyticmeasuresforquantificationof 476arousalfromspontaneousskinconductancefluctuations.InternationalJournalof 477Psychophysiology,76,52-55. 478

    Bach,D.R.,Friston,K.J.,&Dolan,R.J.(2013).Animprovedalgorithmformodel-based 479analysisofevokedskinconductanceresponses.BiolPsychol,94,490-497. 480

  • 27

    Bach,D.R.,Gerster,S.,Tzovara,A.,&Castegnetti,G.(2016).Alinearmodelforevent- 481relatedrespirationresponses.JNeurosciMethods,270,147-155. 482

    Bach,D.R.,Seifritz,E.,&Dolan,R.J.(2015).TemporallyUnpredictableSoundsExerta 483Context-DependentInfluenceonEvaluationofUnrelatedImages.PLoSOne,10, 484e0131065. 485

    Bach,D.R.,&Staib,M.(2015).Amatchingpursuitalgorithmforinferringtonic 486sympatheticarousalfromspontaneousskinconductancefluctuations. 487Psychophysiology,52,1106-1112. 488

    Bach,D.R.,Tzovara,A.,&Vunder,J.(2017).Blockinghumanfearmemorywiththe 489matrixmetalloproteinaseinhibitordoxycycline.MolPsychiatry. 490

    Bach,D.R.,Weiskopf,N.,&Dolan,R.J.(2011).Astablesparsefearmemorytracein 491humanamygdala.JournalofNeuroscience,31,9383-9389. 492

    Barbur,J.L.,Harlow,A.J.,&Sahraie,A.(1992).Pupillaryresponsestostimulusstructure, 493colourandmovement.OphthalmicPhysiolOpt,12,137-141. 494

    Barry,R.J.,Feldmann,S.,Gordon,E.,Cocker,K.I.,&Rennie,C.(1993).Elicitationand 495habituationoftheelectrodermalorientingresponseinashortinterstimulus 496intervalparadigm.InternationalJournalofPsychophysiology,15,247-253. 497

    Benedek,M.,&Kaernbach,C.(2010a).Acontinuousmeasureofphasicelectrodermal 498activity.JournalofNeuroscienceMethods,190,80-91. 499

    Benedek,M.,&Kaernbach,C.(2010b).Decompositionofskinconductancedataby 500meansofnonnegativedeconvolution.Psychophysiology,47,647-658. 501

    Berntson,G.G.,Cacioppo,J.T.,&Quigley,K.S.(1995).Themetricsofcardiac 502chronotropism:biometricperspectives.Psychophysiology,32,162-171. 503

    Berntson,G.G.,Quigley,K.S.,&Lozano,D.(2007).CardiovascularPsychophysiology.In 504J.T.T.Cacioppo,L.G.;Berntson,G.G.(Ed.),HandbookofPsychophysiology.NewYork 505City:CambridgeUniversityPress. 506

    Boiten,F.A.,Frijda,N.H.,&Wientjes,C.J.(1994).Emotionsandrespiratorypatterns: 507reviewandcriticalanalysis.IntJPsychophysiol,17,103-128. 508

    Boucsein,W.(2012).Electrodermalactivity.NewYorkSpringer. 509Boucsein,W.,Fowles,D.C.,Grimnes,S.,Ben-Shakhar,G.,roth,W.T.,Dawson,M.E.,& 510

    Filion,D.L.(2012).Publicationrecommendationsforelectrodermal 511measurements.Psychophysiology,49,1017-1034. 512

    Bradley,M.M.,Codispoti,M.,Cuthbert,B.N.,&Lang,P.J.(2001).EmotionandMotivation 513I:DefensiveandAppetitiveReactionsinPictureProcessing.Emotion,1,276-298. 514

    Brown,J.S.,Kalish,H.I.,&Faber,I.E.(1951).Conditionedfearasrevealedbymagnitude 515ofstartleresponsetoanauditorystimulus.JournalofExperimentalPsychology, 51641,317-328. 517

    Bulganin,L.,Bach,D.R.,&Wittmann,B.C.(2014).Priorfearconditioningandreward 518learninginteractinfearandrewardnetworks.FrontiersinBehavioral 519Neuroscience,8,67. 520

    Castegnetti,G.,Tzovara,A.,Staib,M.,Gerster,S.,&Bach,D.R.(2017).Assessingfear 521learningviaconditionedrespiratoryamplituderesponses.Psychophysiology,54, 522215-223. 523

    Castegnetti,G.,Tzovara,A.,Staib,M.,Paulus,P.C.,Hofer,N.,&Bach,D.R.(2016). 524Modelingfear-conditionedbradycardiainhumans.Psychophysiology,53,930- 525939. 526

    deBerker,A.O.,Rutledge,R.B.,Mathys,C.,Marshall,L.,Cross,G.F.,Dolan,R.J.,& 527Bestmann,S.(2016).Computationsofuncertaintymediateacutestress 528responsesinhumans.NatCommun,7,10996. 529

  • 28

    deGee,J.W.,Colizoli,O.,Kloosterman,N.A.,Knapen,T.,Nieuwenhuis,S.,&Donner,T.H. 530(2017).Dynamicmodulationofdecisionbiasesbybrainstemarousalsystems. 531Elife,6. 532

    deGee,J.W.,Knapen,T.,&Donner,T.H.(2014).Decision-relatedpupildilationreflects 533upcomingchoiceandindividualbias.ProceedingsoftheNationalAcademyof 534SciencesoftheUSA,111,E618-625. 535

    Eldar,E.,Cohen,J.D.,&Niv,Y.(2013).Theeffectsofneuralgainonattentionand 536learning.NatureNeuroscience,16,1146-1153. 537

    Fan,J.,Xu,P.,VanDam,N.T.,Eilam-Stock,T.,Gu,X.,Luo,Y.-j.,&Hof,P.R.(2012). 538Spontaneousbrainactivityrelatestoautonomicarousal.TheJournalof 539Neuroscience,32,11176-11186. 540

    Friston,K.J.,Jezzard,P.,&Turner,R.(1994).AnalysisoffunctionalMRItime-series.Hum 541BrainMapp,1153-171. 542

    Gerster,S.,Namer,B.,Elam,M.,&Bach,D.R.(2017).Testingalineartimeinvariant 543modelforskinconductanceresponsesbyintraneuralrecordingandstimulation. 544Psychophysiology. 545

    Grassmann,M.,Vlemincx,E.,vonLeupoldt,A.,&VandenBergh,O.(2015).Theroleof 546respiratorymeasurestoassessmentalloadinpilotselection.Ergonomics,1-9. 547

    Greco,A.,Guidi,A.,Felici,F.,Leo,A.,Ricciardi,E.,Bianchi,M.,Bicchi,A.,Citi,L.,Valenza, 548G.,&Scilingo,E.P.(2017).Musclefatigueassessmentthroughelectrodermal 549activityanalysisduringisometriccontraction.ConfProcIEEEEngMedBiolSoc, 5502017,398-401. 551

    Greco,A.,Lanata,A.,Valenza,G.,DiFrancesco,F.,&Scilingo,E.P.(2016).Gender-specific 552automaticvalencerecognitionofaffectiveolfactorystimulationthroughthe 553analysisoftheelectrodermalactivity.ConfProcIEEEEngMedBiolSoc,2016, 554399-402. 555

    Greco,A.,Valenza,G.,Lanata,A.,Scilingo,E.,&Citi,L.(2015).cvxEDA:aConvex 556OptimizationApproachtoElectrodermalActivityProcessing.IEEETransBiomed 557Eng. 558

    Greco,A.,Valenza,G.,&Scilingo,E.P.(2016).Investigatingmechanicalpropertiesofa 559fabric-basedaffectivehapticdisplaythroughelectrodermalactivityanalysis.Conf 560ProcIEEEEngMedBiolSoc,2016,407-410. 561

    Green,C.D.(1992).OfImmortalMythologicalBeasts:OperationisminPsychology. 562Theory&Psychology,2,291-320. 563

    Green,S.R.,Kragel,P.A.,Fecteau,M.E.,&LaBar,K.S.(2014).Developmentandvalidation 564ofanunsupervisedscoringsystem(Autonomate)forskinconductanceresponse 565analysis.IntJPsychophysiol,91,186-193. 566

    Hayes,D.J.,Duncan,N.W.,Wiebking,C.,Pietruska,K.,Qin,P.,Lang,S.,Gagnon,J.,Bing, 567P.G.,Verhaeghe,J.,Kostikov,A.P.,Schirrmacher,R.,Reader,A.J.,Doyon,J., 568Rainville,P.,&Northoff,G.(2013).GABAAreceptorspredictaversion-related 569brainresponses:anfMRI-PETinvestigationinhealthyhumans. 570Neuropsychopharmacology,38,1438-1450. 571

    Hoeks,B.,&Ellenbroek,B.A.(1993).Aneuralbasisforaquantitativepupillarymodel. 572JournalofPsychophysiology,7,315-315. 573

    Hoeks,B.,&Levelt,W.J.M.(1993).PupillaryDilationasaMeasureofAttention-a 574QuantitativeSystem-Analysis.BehaviorResearchMethodsInstruments& 575Computers,25,16-26. 576

    Joshi,S.,Li,Y.,Kalwani,R.M.,&Gold,J.I.(2016).RelationshipsbetweenPupilDiameter 577andNeuronalActivityintheLocusCoeruleus,Colliculi,andCingulateCortex. 578Neuron,89,221-234. 579

  • 29

    Khemka,S.,Tzovara,A.,Gerster,S.,Quednow,B.B.,&Bach,D.R.(2017).Modelingstartle 580eyeblinkelectromyogramtoassessfearlearning.Psychophysiology,54,204-214. 581

    Knapen,T.,deGee,J.W.,Brascamp,J.,Nuiten,S.,Hoppenbrouwers,S.,&Theeuwes,J. 582(2016).CognitiveandOcularFactorsJointlyDeterminePupilResponsesunder 583Equiluminance.PLoSOne,11,e0155574. 584

    Koban,L.,Kusko,D.,&Wager,T.D.(2018).Generalizationoflearnedpainmodulation 585dependsonexplicitlearning.ActaPsychol(Amst),184,75-84. 586

    Koban,L.,&Wager,T.D.(2016).Beyondconformity:Socialinfluencesonpainreports 587andphysiology.Emotion,16,24. 588

    Korn,C.W.,&Bach,D.R.(2016).Asolidframeforthewindowoncognition:Modeling 589event-relatedpupilresponses.JVis,16,28. 590

    Korn,C.W.,Staib,M.,Tzovara,A.,Castegnetti,G.,&Bach,D.R.(2017).Apupilsize 591responsemodeltoassessfearlearning.Psychophysiology,54,330-343. 592

    Lang,P.J.,Bradley,M.M.,&Cuthbert,B.N.(2005).Internationalaffectivepicturesystem 593(IAPS):Affectiveratingsofpicturesandinstructionmanual.TechnicalReportA-6. 594Gainesville,FL:UniversityofFlorida. 595

    Lim,C.L.,Rennie,C.,Barry,R.J.,Bahramali,H.,Lazzaro,I.,Manor,B.,&Gordon,E.(1997). 596Decomposingskinconductanceintotonicandphasiccomponents.International 597JournalofPsychophysiology,25,97-109. 598

    Liu,Y.,Rodenkirch,C.,Moskowitz,N.,Schriver,B.,&Wang,Q.(2017).Dynamic 599LateralizationofPupilDilationEvokedbyLocusCoeruleusActivationResults 600fromSympathetic,NotParasympathetic,Contributions.CellRep,20,3099-3112. 601

    Lonsdorf,T.B.,Menz,M.M.,Andreatta,M.,Fullana,M.A.,Golkar,A.,Haaker,J.,Heitland,I., 602Hermann,A.,Kuhn,M.,Kruse,O.,MeirDrexler,S.,Meulders,A.,Nees,F.,Pittig,A., 603Richter,J.,Romer,S.,Shiban,Y.,Schmitz,A.,Straube,B.,Vervliet,B.,Wendt,J., 604Baas,J.M.P.,&Merz,C.J.(2017).Don'tfear'fearconditioning':Methodological 605considerationsforthedesignandanalysisofstudiesonhumanfearacquisition, 606extinction,andreturnoffear.NeurosciBiobehavRev,77,247-285. 607

    McDougal,D.H.,&Gamlin,P.D.R.(2008).Pupillarycontrolpathways. 608Nicolle,A.,Fleming,S.M.,Bach,D.R.,Driver,J.,&Dolan,R.J.(2011).Aregret-induced 609

    statusquobias.JournalofNeuroscience,31,3320-3327. 610Pabst,O.,Tronstad,C.,&Martinsen,O.G.(2017).Instrumentation,electrodechoiceand 611

    challengesinhumanskinmemristormeasurement.ConfProcIEEEEngMedBiol 612Soc,2017,1844-1848. 613

    Paulus,P.C.,Castegnetti,G.,&Bach,D.R.(2016).Modelingevent-relatedheartperiod 614responses.Psychophysiology,53,837-846. 615

    Ritz,T.,Kullowatz,A.,Goldman,M.D.,Smith,H.J.,Kanniess,F.,Dahme,B.,&Magnussen, 616H.(2010).Airwayresponsetoemotionalstimuliinasthma:theroleofthe 617cholinergicpathway.JApplPhysiol(1985),108,1542-1549. 618

    Simmons,J.P.,Nelson,L.D.,&Simonsohn,U.(2011).False-PositivePsychology: 619UndisclosedFlexibilityinDataCollectionandAnalysisAllowsPresenting 620AnythingasSignificant.PsychologicalScience,22,1359-1366. 621

    Staib,M.,Castegnetti,G.,&Bach,D.R.(2015).Optimisingamodel-basedapproachto 622inferringfearlearningfromskinconductanceresponses.JNeurosciMethods,255, 623131-138. 624

    Sulzer,J.,Sitaram,R.,Blefari,M.L.,Kollias,S.,Birbaumer,N.,Stephan,K.E.,Luft,A.,& 625Gassert,R.(2013).Neurofeedback-mediatedself-regulationofthedopaminergic 626midbrain.Neuroimage,83,817-825. 627

  • 30

    Talmi,D.,Dayan,P.,Kiebel,S.J.,Frith,C.D.,&Dolan,R.J.(2009).Howhumansintegrate 628theprospectsofpainandrewardduringchoice.JournalofNeuroscience,29, 62914617-14626. 630

    Tronstad,C.,Kalvoy,H.,Grimnes,S.,&Martinsen,O.G.(2013).Waveformdifference 631betweenskinconductanceandskinpotentialresponsesinrelationtoelectrical 632andevaporativepropertiesofskin.Psychophysiology,50,1070-1078. 633

    Tzovara,A.,Korn,C.W.,&Bach,D.R.(2018).HumanPavlovian 634 fearconditioningconformstoprobabilisticlearning.PLOS 635 ComputationalBiology,14,e1006243. 636Vlemincx,E.,VanDiest,I.,&VandenBergh,O.(2015).Emotion,sighing,andrespiratory 637

    variability.Psychophysiology,52,657-666. 638Wuyts,R.,Vlemincx,E.,Bogaerts,K.,VanDiest,I.,&VandenBergh,O.(2011).Sighrate 639

    andrespiratoryvariabilityduringnormalbreathingandtheroleofnegative 640affectivity.IntJPsychophysiol,82,175-179. 641

    Yeomans,J.S.,Li,L.,Scott,B.W.,&Frankland,P.W.(2002).Tactile,acousticandvestibular 642systemssumtoelicitthestartlereflex.NeurosciBiobehavRev,26,1-11. 643

    644

    645Appendix 646

    647

    Modelevidence 648

    Toquantifymodelevidence,PsPMusesthefollowingidentitytocomputeAkaike 649

    InformationCriterion(AIC): 650

    wherenisthenumberofdatapointsinthepredictivemodel,Listhemaximumofthe 651

    likelihoodfunction,andkthenumberofparametersinthepredictivemodel.kis 652

    constantforallmethodsthatareevaluatedinamethodscomparison.Thepredictive 653

    modelusestheaprioridefinedpsychologicalvariableasdependentvariable(this 654

    ensuresitisthesameacrossallmethods),andthedesignmatrixcontainstheestimated 655

    psychologicalvariableand(possiblysubject-specific)interceptterms. 656

    657

    LTIsystems 658

    Lineartimeinvariantsystemsaredefinedbythefollowingconvolutionoperation: 659

    whereu(t)istheinputintothesystemattimet,histheimpulseresponsefunction(RF), 660

    andτisadummyvariableoverwhichintegrationisperformed.Notethatsinceweare 661

    dealingwithacausalsystem(i.e,.intime),wehaveexplicitlysetthelowerlimitofthe 662

    !"# = −2 ln ! + 2! = ! ln !""! + 2! (1)

    ! ! = !×ℎ = ! ! − ! ℎ ! d!!

    ! (2)

  • 31

    integraltozero,suchthataninputthatoccursaftertcanhavenoimpactuponthe 663

    outputattimet. 664

    665

    GLM 666

    AGLMcanbewrittenas 667

    whereYisthevectorofobservations,βisavectorofinputamplitudeparameters,andϵ 668

    istheerror.Xisthedesignmatrixinwhicheachcolumnisobtainedbyconvolving 669

    impulsefunctionsatknowntimepointswitheachcomponentoftheRF.Eachcolumn 670

    takestheform: 671

    where!!(!)istheneuralinputwithunitamplitudeforconditioni,andjistheindexof 672theRFcomponent.Finally,Xalsocontainsacolumnfortheintercept.Themaximum- 673

    likelihoodamplitudeestimates!canbecomputedusingtheMoore-Penrose 674pseudoinverse!!,forexampleimplementedintheMatlabfunctionpinv.m: 675

    676

    Skinconductanceforwardmodel 677

    (a)PhenomenologicalRFdescribedasaGaussian-smoothedexponential: 678

    withestimatedparameters:!! = 3.0745 !forpeaklatency;! = 0.7013fordefinitionof 679therisetime;!! = 0.3176 and!! = 0.0708todefinethetwodecaycomponents. 680(b)PhenomenologicalRFdescribedasthird-order(linear,constant-coefficient)ordinary 681

    differentialequation: 682

    with!! = 1.342052,!! = 1.411425,!! = 0.122505,!! = 1.533879 683 684

    ! = !" + ! (3)

    ! = !!(!)×ℎ!(!) (4)

    ! = !!! (5)

    ℎ ! ∝ ! ! − !!

    !!! ! + !! ! !",

    ! ≥ 0,

    ! ! = 12!" !! !!!!

    !!!! ,

    !! ! = !!!!

    (6)

    ! = !!! + !!! + !!! − ! ! − !! = 0 (7)

  • 32

    ModelswithGaussianresponsefunctions 685

    Gaussianfunction: 686

    Parametersseetable(HPR_E1-6:constant-latencyheartperiodresponses;RPR: 687

    respirationperiodresponses;RAR_E:constant-latencyrespirationamplituderesponses; 688

    RFRR:respiratoryflowrateresponses). 689

    Responsefunction(RF) ParametersofGaussianFunction

    !(mean) !(standarddeviation)HPR_E1 1.0 1.9

    HPR_E2 5.2 1.9

    HPR_E3 7.2 1.5

    HPR_E4 7.2 4.0

    HPR_E5 12.6 2.0

    HPR_E6 18.85 1.8

    RPR 4.20 1.65

    RAR_E 8.07 3.74

    RFRR 6.00 3.23

    690

    Modelswithgammaresponsefunctions 691

    Gammafunction: 692

    Parametersseetable(HPR_RC:fear-conditionedheartperiodresponses;PSR_dil: 693

    luminance-evokedpupildilation;PSR_con:luminance-evokedpupilconstriction;PSR_FC: 694

    fear-conditionedpupilsizeresponses;RAR_FC:fear-conditionedrespirationamplitude 695

    responses;SEBR:startleeyeblinkresponses).Notethattheamplitudeparameterisleftfree 696

    forallmodelsotherthantheluminancemodelsinwhichithasaphysicalinterpretation. 697

    698

    Responsefunction(RF) ParametersofgammaFunction

    !(shape) !(scale) !!(onset) !(amplitude)

    ! = 1! 2! !! !!!

    !!!! (8)

    ! = ! − !!!!!!!

    !!!!!

    !!Γ ! (9)

  • 33

    HPR_FC 48.5 0.182 -3.86 -

    PSR_dil 2.40 2.40 0.2 0.77

    PSR_con 3.24 0.18 0.2 0.43

    PSR_FC 5.94 0.75 0.002 -

    RAR_FC(early) 2.570×10! 3.124×10!! −8.024×10! -RAR_FC(late) 3.413 1.107 7.583 -

    SEBR 3.5114 0.0108 -

    699

    Modelforsteady-statepupilsize 700

    701where!isthezscoredsteady-statepupildiameterand!!is the respective illuminance 702level in (in [!"] ). The parameter values are: 703! = 49.79;! = −0.50 !!" ;! = −1.05. 704 705 706

    ! !! = ! + !!!!! (10)


Recommended