Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
ContentslistsavailableatSciVerseScienceDirect
Biological
Psychology
j o u r n a l ho me p ag e :w w w . e l s e v i e r . c o m / l o c a te / b i o p s y c h o
Intentional
retrieval
suppression
can
conceal
guilty
knowledge
in
ERP
memory
detection
tests
夽
Zara
M.
Bergström
a,∗,
Michael
C.
Anderson
b,c,
Marie
Buda
c,
Jon
S.
Simons
c,
Alan
Richardson-Klavehn
daSchoolofPsychology,UniversityofKent,CanterburyCT27NP,UK
bMRCCognitionandBrainSciencesUnit,15ChaucerRoad,CambridgeCB27EF,UK
cBehaviouralandClinicalNeuroscienceInstituteandDepartmentofPsychology,UniversityofCambridge,DowningStreet,CambridgeCB23EB,UK dDepartmentofNeurology,FacultyofMedicine,OttovonGuerickeUniversityofMagdeburg,Zenit1Building,LeipzigerStrasse44,39120Magdeburg,
Germany
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received16March2012 Accepted23April2013 Available online xxx Keywords: Episodicretrieval Event-RelatedPotentials Memorysuppression Guiltyknowledge Cognitivecontrola
b
s
t
r
a
c
t
Brain-activitymarkersofguiltyknowledgehavebeenpromotedasaccurateandreliablemeasuresfor establishingcriminalculpability.Testsbasedonthesemarkersinterpretthepresenceorabsenceof memory-relatedneuralactivityasdiagnosticofwhetherornotincriminatinginformationisstoredin asuspect’sbrain.Thisconclusioncriticallyreliesontheuntestedassumptionthatremindersofacrime uncontrollablyelicitmemory-relatedbrainactivity.However,recentresearchindicatesthat,insome circumstances,humanscancontrolwhethertheyrememberapreviousexperiencebyintentionally suppressingretrieval.Weexaminedwhetherpeoplecoulduseretrievalsuppressiontoconcealneural evidenceofincriminatingmemoriesasindexedbyEvent-RelatedPotentials(ERPs).Whenpeoplewere motivatedtosuppresscrimeretrieval,theirmemory-relatedERPeffectsweresignificantlydecreased, allowingguiltyindividualstoevadedetection.Ourfindingsindicatethatbrainmeasuresofguilty knowl-edgemaybeundercriminals’intentionalcontrolandplacelimitsontheiruseinlegalsettings.
© 2013 The Authors. Published by Elsevier B.V. All rights reserved.
1. Introduction
Recent suggestions that technological advances now allow
us todecode criminal guilt from brain activitydata have
gen-erated intensive interdisciplinary debate within the scientific
community(Garland&Glimcher,2006;Greely&Illes,2007;Sip,
Roepstorff,McGregor,&Frith,2008;Wolpe,Foster,&Langleben,
2005).Severalemerging companiesare advertising commercial
implementations of brain activity guilt detection (e.g. No Lie
MRI,http://www.noliemri.com/;BrainFingerprintingLaboratories,
http://www.brainwavescience.com/), and attempts to introduce
evidencefromsuchtestsincriminaltrialsarefrequentlyreported
ininternationalmedia(Giridharadas,2008;McCall,2004;Miller,
2010). However, despite widespread interest and discussion,
empiricaldataconcerningthevalidityofthesebrainactivity-based
methodsissparse.Oneprominentconcernisthatmostresearch
todatehasbeenconductedoncompliantparticipantswithlittle
夽 Thisis anopen-accessarticledistributedunderthetermsoftheCreative CommonsAttributionLicense,whichpermitsunrestricteduse,distributionand reproductioninanymedium,providedtheoriginalauthorandsourcearecredited.
∗ Correspondingauthor.Tel.:+4401227827507;fax:+4401227827030. E-mailaddress:[email protected](Z.M.Bergström).
motivation to hide theirguilt, whereas real criminals mayuse
countermeasurestrategiestoavoiddetection.Inviewofthe
impor-tantsocietal,legalandethicalimplicationsofbrainactivitycrime
detection,itisvitaltovalidatethesemethodsbeforetheyarewidely
adopted,and, inparticular,toevaluatehowwelltheyworkfor
uncooperativesuspectsmotivatedtoconcealincriminating
knowl-edge.
Memorydetectionaimstoestablishculpabilitybydetermining
fromphysiologicalorbehaviouralcorrelatesofmemoryretrieval
whetherasuspecthasknowledgeofacrimethatonlyaguilty
per-sonwouldpossess(Meegan,2008).Scalp-recordedEvent-Related
Potentials(ERPs)areoftenusedinthesetypesoftestas
inexpen-siveandnon-invasivemeasuresofreal-timeneuralactivity(e.g.
Allen, Iacono,& Danielson,1992; Rosenfeld, Angell, Johnson, & Qian,1991;VanHooff,Brunia,&Allen,1996).ERPshavegained
popularityasanalternativetotraditionalautonomicmeasuresin
memorydetectionstudies(Lykken,1959;seeBen-Shakhar&Elaad,
2003,for review),partlybecausetherapidand process-specific
brainresponsesreflectedinERPsarebelievedtobemore
resis-tanttocountermeasuresthanotherphysiologicalandbehavioural
measures(e.g.Lykken,1998;seediscussioninBen-Shakhar,2011).
Recently however,researchershavechallengedthisassumption,
showing that ERP memory detection tests may be more
vul-nerable than previously thought (e.g. Mertens & Allen, 2008;
0301-0511/$–seefrontmatter © 2013 The Authors. Published by Elsevier B.V. All rights reserved.
Rosenfeld,2006;Rosenfeld,Soskins,Bosh,&Ryan,2004,although seeRosenfeldetal.,2008).Inthispaper,wedescribea
counter-measurethathasnotbeenexploredbeforeintheliterature,which
questionsoneof thefundamentalassumptionsofbrain-activity
memorydetection.
InanERPversionofatypicalguiltyknowledgetest(GKT,Lykken,
1959),ERPs arerecorded whileparticipantsengageina
crime-irrelevanttargetdetectiontaskthatincludesremindersofsome
incriminatinginformation(e.g.Farwell&Donchin,1991;Farwell
&Smith,2001).Participantsareaskedtodiscriminatebetweena
setoftargetitems(forexamplewordspresentedonacomputer
screen)andanothersetofirrelevantcontrolitems(otherwords)
bypressing one button for targetsand anotherfor irrelevants.
Thisprocedureproducesanenlargedpositivemid-parietal
com-ponenttermedthe“P300”(seePolich,2007,forreview)intheERP
waveformaround300–900msspecificallyaftertarget
presenta-tions,thoughttoindexprocessesrelatedtoparticipants’conscious
recognitionoftargetsasmeaningfulstimuli.BecauseP300
ampli-tudesaretypicallyinverselyrelatedtothesubjectiveprobabilityof
astimulus(Donchin,1981),asmallproportionoftargetsare
typ-icallyintermixedwithalargerproportionofirrelevantstomake
targetssubjectivelyrare, thusenhancingP300differences.
Cru-cially,toassess thepresenceorabsence ofguiltyknowledge, a
smallproportionofcrimereminders(“probes”)arealsopresented
aspartoftheirrelevantset.Totrulyinnocentsuspects,suchcrime
probesareindistinguishablefromirrelevantitemsandthuselicit
nospecialbrainresponse.Toguiltysuspects,theprobesstandout
basedontheircrime-relatedmemorystatus,andthisrecognition
elicitsanenhancedparietalP300similartotargets.Thus,guilty
suspectsshowenlargedparietalP300stobothprobesandtargets
sincebothtypeselicitrecognition,whereasinnocentsuspectsonly
showenlargedparietalP300stotargetsandnottoprobes.
Memory detection tests make the crucial assumption that
reminders of incriminating information uncontrollably elicit
recognition-relatedbrainactivity.Thisassumption gains
plausi-bilityfromthefactthat theGKTmemoryprobesdirectlyname
details from the crime, thus constituting exceedingly potent
retrievalcuesforapersonallysignificantevent,makingitappear
extremelyunlikely thatifa related memoryis present,thecue
will not automatically evoke recognition and its neural
mark-ers.Theinevitabilityofsuchretrievalisquestioned,however,by
recentevidencethatthebrainactivitycorrelatesofgeneral
mem-oryretrievalmaybeundermorevoluntarycontrolthanhasbeen
previouslyassumed(e.g.Dzulkifli&Wilding,2005;Herron&Rugg,
2003;Rissman,Greely,&Wagner,2010).Sofar,thestrongest
evi-dencethatmemoryretrievalcanbeintentionallypreventedstems
fromthe Think/No-Think (TNT) memorysuppression paradigm
(Anderson&Green, 2001).In thistask, participantsare trained
onpairsofstimuli(typicallyweaklyrelatedwords),andarelater
presented with the first item of each pair as a reminder, and
areaskedto eitherthink of theassociate item (theThink
con-dition),or tocompletely preventtheassociate fromcomingto
mindbysuppressingretrieval(theNo-Thinkcondition).Thinkand
No-Thinkremindersarerepeatedlypresented,typicallyrandomly
intermixedinequalproportions.fMRIevidencefromthisparadigm
suggeststhat peoplecanengageresponseoverridemechanisms
mediated by the lateral prefrontal cortex to suppress retrieval
bymodulating memory-related activityin thehippocampus in
responsetoreminders(Andersonetal.,2004;Benoit&Anderson,
2012;Butler&James,2010;Depue,Curran,&Banich,2007;Levy& Anderson,2012;Paz-Alonso,Ghetti,Anderson,&Bunge,2013).
Mostcriticallyformemoryguiltdetectionresearchwhichis
pre-dominantlyERP-based,askingparticipantstosuppressunwanted
memoriesinaTNTtaskcausesmemory-specificreductionsofan
ERPeffectwithsimilarpolarity,topographyandtimingtotheP300
component(Bergström,DeFockert,&Richardson-Klavehn,2009a,
2009b; Bergström,Velmans, De Fockert, &Richardson-Klavehn, 2007;Hanslmayr,Leipold,Pastötter,&Bäuml,2009;Mecklinger, Parra,&Waldhauser,2009).However,thesefindings havebeen
interpretedasvoluntarysuppressionoftheERPmarkerofepisodic
recollection,which isspecificallycorrelatedwiththeamountof
episodic detail that is consciously recollected in response toa
reminder(e.g.Vilberg,Moosavi,&Rugg,2006;seeRugg&Curran,
2007,forreview).Becauseoftheirsimilarcharacteristics,parietal
effectsrelatedtorecollectionand“classic”P300effectsrelatedto
stimulusevaluationaredifficulttoteaseapart.Thisisparticularly
thecaseintaskswhereepisodicremindersaresubjectivelyrare,
suchastheGKT,sincerecollection-relatedandclassicP300effects
maybothcontributetoparietalERPamplitudes.Theepisodic
rec-ollectioneffectishowevermorelikelytobeleft-lateralizedthan
P300effects.Furthermore,althoughtheP300ishighlysensitiveto
subjectiveprobability,theparietalrecollectioneffectmaynotbe
(Herron,Quale,&Rugg,2003;seealsoSmith&Guster,1993).These
ERPeffectsarethusthoughttoindexseparablecognitiveprocesses,
althoughthepreciserelationshipbetweenthemisstillunclear.
Theimplicationsoftheaforementionedresearchforcriminal
guiltdetectiontestsneverthelessremaintobeestablished,since
there are many features of such tests that have not yet been
exploredinthecontextofretrievalsuppression.Noresearchhas
examinedwhetherretrieval-relatedbrainactivitycan be
inten-tionallysuppressedforobjectsoreventsdirectlynamed bythe
reminderitself,inparticularwhencomparingthesepotentcuesto
novelcontrolcuesthatareunlikelytoelicitmemory-relatedbrain
activity.Moreover,it remains unclear whethermemory-related
brainactivitycanbesuppressedwhenremindersaresubjectively
rare,asintypicalGKTresearch.Ifmnemoniccontrolispossible
under conditions of exceedingly strong and subjectively
unex-pectedretrievalcueshowever,itraisesthepossibilitythatguilty
suspectsmotivatedtoconcealtheirknowledgemaybeableto
sup-pressbrainactivityelicitedbyincriminatingprobesduringmemory
testingandhenceeludedetection.
Furthermore, prior TNT research hasshown that repeatedly
suppressingretrievalinresponsetoaremindercandramatically
reduce theubiquitousbeneficial effects ofreminders on
reten-tion,andevenimpairrecallperformanceoftheavoidedmemories
comparedtoitemsinabaselineconditionthathavebeenneither
recallednorsuppressedsinceinitiallearning(e.g.Andersonetal.,
2004;Anderson&Green, 2001;Bergströmetal.,2009b; Depue etal.,2007;seeAnderson&Huddleston,2011,forareviewand
meta-analysis).Thisfindingimpliesthatretrievalsuppression
dur-ingaguiltdetectiontestmaybesuccessfultotheextentofactually
impairinglaterretentionoftheincriminatingmemories.
Todeterminewhetherpeoplecancontrolbrainresponsesto
remindersthat mightbeexpectedtoelicitincriminating
recog-nition,wecreatedanoveldesignthatcombinedelementsfrom
bothGKTandTNTparadigms,andaskedparticipantstovoluntarily
suppressmemoriesofasimulatedcrime.ERPswererecorded
dur-ingthreephasesofamemorydetectiontestthatdirectlyprobed
centraldetailsfromaprevioushomeburglarysimulation.Inone
controlphase,participantsweretrulyinnocentofthetestedcrime.
Inasecond“guiltycooperative”phase,participantswereaskedto
remembertheircrime.Finally,in athird “guiltyuncooperative”
phase,participantswereaskedtosuppresscrimerecalltoevade
detection.Followingtheguiltyknowledgetest,wecompared
mem-oryforrepeatedlysuppressedandrepeatedlyrememberedcrime
detailstomemoryforotherdetailsthatwereencounteredduring
theinitialburglarysimulationbutdidnotappearinthe
interven-ingdetectionphase(i.e.abaselinecondition),toassesswhether
retrievalsuppression of crime memorieshad lasting effectson
memoryaccessibility.
The relative probabilities of item categories were varied
experimentpresentedeachcategorywithequalprobability,which
allowed an assessment of whether retrieval-related activity to
exceedingly strong reminders of a crime can be voluntarily
suppressedintheabsenceofprobabilityeffects.Thesecond
exper-imentreducedtheprobabilityofprobesandtargetstoinvestigate
whether the neural response to crime reminders can be
sup-pressedevenwhenprobesaresubjectivelyunexpected.Ifpeople
cansuppresscrimerecallwhenmotivatedtodoso,parietalP3001
amplitudesforprobes shouldbereduced intheguilty
uncoop-erativephasecompared totheguiltycooperativephase,leading
tosignificantlypoorerguilt detectionrates.Iftheyareperfectly
successfulatsuppressingretrieval,P300amplitudesduring
sup-pressionof guilty knowledge should be indistinguishable from
thoseobservedduringtheinnocentcontrolphase.
2. Materialsandmethods 2.1. Participants
Experimentoneincludeddatafrom24right-handednativeGermanspeakers(15 females)withameanageof24(range18–35).Experimenttwoincludeddatafrom 24right-handednativeEnglishspeakers(14females)withameanageof21(range 18–35).Allparticipantshadnoknownhistoryofneurologicalorpsychiatricdisease andhadnormalorcorrected-to-normalvision,andgavewritteninformedconsent. ExperimentswereconductedinaccordancewiththeguidelinesofthelocalEthics CommitteesoftheUniversityofMagdeburg(Experimentone)andtheUniversityof Cambridge(Experimenttwo).
2.2. Design,materialsandprocedure
ExperimentonewasconductedinGermanattheOttovonGuericke Univer-sityofMagdeburg,Germany,whereasExperimenttwowasconductedinEnglishat theUniversityofCambridge,UK.Visualandverbalmaterialswerekeptassimilar aspossibleacrossexperiments.Bothexperimentsconsistedofthreemainphases: (1)aninitialburglarysimulation;(2)amemorydetectionphase,whereERPswere recorded;and(3)afinaltestphase,whereparticipants’memoryfortheburglary taskwasassessed.
Uponarrival,participantscompletedthecrimesimulationtaskonacomputer, whichwasdesignedtoleadtorich,elaborativememoryencodingof24photographs ofdistinctcommonobjects(e.g.agoldwatch,drawnatrandomforeachpersonfrom alargersetofobjectsthatwereassemanticallyunrelatedaspossible)without inten-tionallearningattempts.Participantswereaskedtovividlyimaginebeingaburglar whowasbreakingintohouseswiththeaimofstealingvaluables.Duringeachtrial,a pictureofaroominteriorwaspresentedfirst,andparticipantswereaskedtoimagine thattheyhadbrokenintotheroomandjudgewhethertheythoughtthatparticular roomwaslikelytocontainsomethingvaluable.Second,fournumbersappearedon thebackground,markingparticularlocationsintheroom(e.g.adrawer). Partic-ipantswereaskedtosearchthroughthelocationsbypressingthecorresponding numbersonthekeyboardinordertofindahiddenobject.Whentheypressedthe correctoptionanobjectappeared,superimposedontheroompicture.Next, partic-ipantscompletedthreeratingtasksontheobject-roompicture.First,theydecided whethertheobjectwasthekindofobjecttheywouldexpecttofindintheparticular room.Second,theyrateditsvalue,andthird,decidedwhethertheywantedtosteal it.Therewasnouppertimelimittorespondandthecomputeronlymovedonto thenextquestionafteraresponsehadbeenrecorded.Eachofthejudgementscould onlybegivenaftera4sdelayofviewingthepicture,meaningthateachtrialwasa minimumof16slong.
Foreachparticipant,wordsnamingonerandomlydrawnsubsetofobjectsfrom theburglarytask(e.g.“watch”)werelaterpresentedasprobesduringtheguilty cooperativeblock,andwordsnaminganothersubsetofobjects(alsorandomly drawn)werepresentedasprobesduringtheguiltyuncooperativeblock.Therest ofthestudyobjectswerenotpresentedduringtheEEGphase,butwereusedasa behaviouralbaselineforthefinalmemorytests.Otherobjectwordsforwhichthe pictureswerenotpresentedduringstudywererandomlyassignedtothetarget, irrelevantandinnocentprobeERPconditions,usingadifferentsetforeachphase. Apriorcontrolexperiment(seenextsection)withanindependentgroupthat per-formedoursimulatedburglarytaskconfirmedthattheyrecognisedthesecrime probesasreferringtothepreviouslyencounteredobjectswithceilingaccuracy, andoverwhelminglyratedtheseprobesaselicitingautomatic,involuntaryrecallof
1 WerefertothelateparietalERPpositivityintheseexperimentsasaP300effect,
inlinewithtypicalGKTresearch.However,asdiscussed,thefunctionalsignificance ofparietalERPpositivitiesintheP300time-windowisunclear,sincetheseparietal positivitiesmightberelatedbothtostimulusevaluationprocessesandtoepisodic recollection.
theburglaryobjects.Thus,presentingthesecrimeprobesinourmemorydetection testwouldbeexpectedtostronglyandreflexivelyelicitretrievaloftheassociated objectsstoredinmemory.
Followingcrimesimulation,participantswerefittedwithanEEGcapand com-pletedthememorydetectionphaseonacomputer.Theyweretoldthatreminder wordsthatreferredtoobjectsfromtheburglarywouldbepresentedonthescreen, inordertotrytomakethemrememberdetailsfromthecrimesothattheirbrain activitycouldbeusedascriminalevidence.However,thesereminderwordswould beintermixedwithotherwordsthatdidnotrefertotheburglarywhiletheywere performingatargetdetectiontask,whichwouldbeconductedinthreeblocks.Before eachblockofthememorydetectiontest,participantspracticedablock-specificlist oftargetwordsandwereinstructedtodetectandpressonebuttonwheneverthey recognisedatargetword,andpressanotherbuttonforanyotherwords(buttons andresponsehandcounterbalancedacrossparticipants).
Followingtargetlearning,morespecificinstructionsfortheparticular mem-orydetectionblockweregiven(ordercounterbalancedacrossparticipants).Inthe guiltycooperativeblock,participantswereinstructedthatiftheyrecognisedan objectwordascorrespondingtoapictureseenduringthecrimesimulation,they shouldtrytorememberasmanydetailsaspossibleaboutthatobjectfromthe bur-glary(analogoustoablockedThinkconditionintheTNTparadigm).Intheinnocent controlblock,participantsweretoldthattheywouldbetestedforacrimetheyhad notcommittedsoshouldsimplyfocusonthetargetdetectiontask(withthe reason-ingthataninnocentsuspectinarealcriminalinvestigationwouldofcourseknow themselvestobeinnocent).2Intheguiltyuncooperativeblock,participantswere
toldtotrytheirbesttocompletelystopanymemoriesofthecrimefromcoming tomindatall,withoutself-distraction(analogoustoablockedNo-Thinkcondition, withinstructionssimilartodirectsuppressiongroupinBergströmetal.,2009b;
Benoit&Anderson,2012).Itwasstronglyemphasisedinallphasesthattheyshould
alwaysreadandpayfullattentiontoeachwordtheentiretimeitwasonthescreen, andonlypressthe“recognition”buttonfortargetsandthe“non-recognition”button foranyotherwords.Thus,therewerenodifferencesinovertresponserequirements acrossphases.
Duringeachmemorydetectiontrial,wordswerepresentedfor2sinwhiteon ablackbackground,precededbya1swhitefixationcross,andfollowedbya0.5s greenfixationcross.Buttonpressesandblinksweretimedtothegreenfixationin ordertoavoidcontaminatingrelevantEEGwithmotor-relatedartefacts.The propor-tionofitemsintheprobe,irrelevantandtargetcategorieswasmanipulatedacross experiments.InExperimentone,eachblockcontainedanequalproportionofeight itemsineachcategory,inlinewithpreviousmemorysuppressionERPresearch thathaspresenteditemstobesuppressedatequalprobabilitytootheritems(e.g.
Bergströmetal.,2007,2009a,2009b;Hanslmayretal.,2009;Mecklingeretal.,2009).
Thislistof24wordswaspresentedtwelvetimes,randomlyintermixedeachtime. InExperimenttwo,eachblockcontainedfourprobes,sixteenirrelevantsandfour targets,i.e.withprobabilitiesofapproximately0.17,0.67and0.17respectively, whicharetypicalproportionsinguiltyknowledgetests(Allenetal.,1992;Farwell
&Donchin,1991;Mertens&Allen,2008;Rosenfeldetal.,2004),alsopresented
twelvetimeseachrandomlyintermixed.
Followingtheguiltdetectionphase,participantscompletedtwofinalteststhat aimedtodetectwhethertheearlierretrievalsuppressionmanipulationaffected laterretentionoftheavoidedmemories.Memorywascomparedforrepeatedly suppressedobjectpictures(i.e.objectpicturesforwhichcorrespondingwordswere presentedasprobesintheuncooperativephase)torepeatedlyrememberedobject pictures(i.e.objectpicturesforwhichcorrespondingwordswerepresentedas probesinthecooperativephase)andtomemoryforbaselineobjectsthatwere simplypresentedduringtheinitialburglarysimulationbutdidnotappearinthe guiltyknowledgetestphase.Becauserecognitionmemoryforobjectsthemselves waslikelytobeatceiling(asestablishedinourvalidationstudyforprobewords), thetestsfocusedonmoredifficultdetailsoftheencodingepisode.Allparticipants completedtwotestsinthesameorder:locationrecallandobjectrecall.Forthese recalltests,participantswerepresentedwithallbackgroundroompicturestogether withthefourmarkedlocationswheretheyhadsearchedforahiddenobject.First, theywereaskedtotrytorememberthelocationinwhichtheyhadfoundtheobject duringtheburglarysimulationtask.Nexttheywereaskedtorecalltheobjectthat theyfoundinthatlocation.Therewasnolimittoresponsetimesineithertest,and picturesstayedonthescreenuntilananswerwasgiven.Finally,participants com-pletedaquestionnairewheretheyratedtheirperceiveddifficultyandsuccessat suppressingcrimedetailsintheuncooperativeblockandtheirperceiveddifficulty andsuccessatrememberingcrimedetailsinthecooperativeblock.
2Onemightworrythattellingparticipantsthattheywereinnocentmayhave
somehowledtodifferentERPeffectsintheinnocentphasecomparedtotheguilty phases.Importantlyhowever,topreviewtheresults,thiswasnotthecase.There werenomeasurabledifferencesbetweentargetandirrelevantERPwaveforms acrossthethreephases,demonstratingthatERPsintheinnocentphasewerenot affectedbyinstructions.
2.3. Validationrecognitiontest
Inordertovalidatethatusingobjectwordsasprobesbroughtmemoriesof thecrimesimulationtomind,anindependentgroupofparticipants(N=10) com-pletedthecrimesimulationfollowedbyanold/newrecognitiontest,wherewords referringtoobjectsfromtheburglarywereintermixedwithnewobjectwords, andparticipantsjudgedwhetherawordcorrespondedtoanobjectfromthe bur-glary.Ifparticipantsrespondedthattheyrecognisedanobjectword,theynextrated whethertheburglarymemorycametomindautomaticallyorthroughintentional effortwhenseeingtheword.
2.4. EEGrecordingandanalysis
InExperimentone,EEGwasrecordedreferencedtotheleftmastoidusinga BrainVisionBrainAmpamplifierfrom60Ag/AgClscalpelectrodesembeddedinan Easycap.InExperimenttwo,EEGwasrecordedreferencedtoCzusinga Electri-calGeodesicNetamps200systemwitha128-channelHydroCelGeodesicSensor Net.Inbothexperiments,signalswereacquiredat250Hz(bandwidth0.1–70Hz) andanalysedusingEEGLAB7(Delorme&Makeig,2004).ThecontinuousEEGdata werere-referencedtoanaveragemastoidreference,filtereddigitallywitha band-passof0.3–30Hz(two-wayleast-squaresfiniteimpulseresponsefilter),andwere thenseparatedintoepochstime-lockedtothewordstimulusonset.Concatenated epochswerecorrectedfromartefactsusingIndependentComponentAnalysis(see
Bergströmetal.,2009b,fordetails).Anytrialsthatstillcontainedvisibleartefacts
followingartefactcorrectionwereremoved.Next,ERPswereformedfortheprobe, irrelevantandtargetitemswithineachguiltycooperative,guiltyuncooperativeand innocentblock(i.e.nineERPconditionsintotal)includingonlytrialsforwhich participantsgaveacorrectresponsewithintheallocatedtime.Onlyaverysmall proportionoftrials(8%inExperimentoneand6%inExperimenttwo)weredeleted intotalaftertheseexclusioncriteria.
Basedonourstrongapriorihypotheses,wefollowedalargebodyof previ-ousP300researchbyfocusingthemainanalysesonthemid-parietalelectrodesite (Pzin10/20nomenclature),whereP300effectsareconsistentlylargestandwhere selectiveanalysesaretypicallyfocusedinGKTresearch.ParietalERPamplitudes werefirstanalysedatthegrouplevelusingparametricstatistics.Because mem-orydetectiontestsaremeanttofunctionasdiagnostictoolsfordeterminingthe guiltofanindividualperson,wealsoestimatedthereliabilityofP300differences withineachparticipant’ssingletrialdatausingnonparametricbootstrap resam-pling(Efron&Tibshirani,1986;Rosenfeldetal.,2004).Priortobootstrapanalysis, individualEEGtrialswerefurtherlowpassfilteredwithan8Hzcut-offinorderto increasesignal-to-noiseratio.Subsequently,foreachparticipant,asetof individ-ualtrialsofthesamesizeastheoriginalprobesetweredrawnatrandomwith replacement,andaveragedtocreateabootstrappedprobeERP.Thesame proce-durewasrepeatedusingirrelevanttrials,creatingabootstrappedirrelevantERP.To createabootstrapped“base-to-peak”measure(e.g.Rosenfeldetal.,2004),the max-imumamplitudeoftheirrelevantP300wasmeasuredbyfindingthe100ms-long timewindowwiththemostpositiveamplitudewithintherangeof400–900ms(this valuebeingthedifferencebetweenthebaselineandtheP300peak,sincetheepochs werebaseline-correctedpriortobootstrapping),andthisvaluewassubtractedfrom thecomparableprobeP300maximumamplitude,creatingaP300differencevalue. Using200iterationsofthisprocedure,adistributionofP300differencevalueswas created.Inordertostatewith95%confidencethatprobeP300sweremorepositive thanirrelevantP300s,thevalueat1.65standarddeviationsbelowthemeanofthis distributionofdifferencesshouldbegreaterthanzero.
Wecomparedbootstrapresultsforthebase-to-peakP300methodabovewith bootstraptestsoftwootherstatistics,meanbaseline-correctedmid-parietal ampli-tudeacrossthewholeP300time-window(400–900ms)andP300peak-to-peak difference(thedifferencebetweenthemaximumP300amplitudevaluedescribed aboveandthe100ms-longtime-windowwiththemostnegativevaluefollowingthe P300peakuptoamaximumlatencyof1600mspost-stimulus;Soskins,Rosenfeld,&
Niendam,2001).Classificationratesacrossthesestatisticswerecomparedatstrict
(95%confidence)andliberal(90%confidence)thresholds.Usingthesethresholds andmeasures,individualsforwhichprobeP300swerereliablymorepositivethan irrelevantP300swereidentifiedandassignedasguilty.
Inordertofacilitatecomparingourclassificationresultswithpreviousresearch, wealsocalculatedReceiver-OperatingCurves(ROC)andassociatedAreaUnder Curves(AUCs)fromthebootstrapdistributionsbycomparinghit-ratesinthetwo Guiltyconditions(i.e.theproportionofparticipantscorrectlyidentifiedasguiltyin theseconditions)withfalsealarmratesintheInnocentcondition(i.e.proportionof participantsincorrectlyidentifiedasguiltywheninnocent)acrossdifferent thresh-olds.TheAUCprovidesausefulsummarystatisticofclassificationperformance irrespectiveofaparticularthreshold,allowingdetectionratestobecomparedacross studiesthatusedifferentclassificationcriteria(seeBen-Shakhar&Elaad,2003).The AUCrangesfrom0to1,andwhereasanAUCvalueof1meansperfect classifica-tionacrossallthresholds(peoplewerealwaysdetectedwhenguiltybutneverfalsely identifiedassuchwheninnocent,irrespectiveoftheclassificationthreshold)anAUC valueof0.5meansclassificationisatchance(peoplewerenomorelikelytobe clas-sifiedasguiltyintheGuiltyphasethantheInnocentphase).AUCvaluessmallerthan 0.5meanthatpeopleweremorelikelytobewronglyclassifiedasguiltywhen inno-centthancorrectlydetectedwhenactuallyguilty.Apreviousmeta-analysisfounda
meanAUCvalueof0.87across42GKTstudiesthatusedskinconductancemeasures andamock-crimeproceduresimilartothecurrentparadigm(Ben-Shakhar&Elaad, 2003)whereasanewer,stillunpublishedmeta-analysisfoundameanAUCof0.93 across32GKTstudiesthatusedP300-basedmeasuresandavarietyofprotocols (Ben-Shakhar,personalcommunication).
Inthecurrentstudy,AUCswerecalculatedwiththenon-parametrictrapezoid method,andtheCooperativevs.InnocentAUCwasstatisticallycomparedagainst theUncooperativevs.InnocentAUCbyestimatingthestandarderrorsoftheAUCs usingjack-kniferesamplingandcalculatingaz-scoreforthedifference,as rec-ommendedbyHanleyandHajian-Tilaki(1997).Thez-scoreformulaincludeda correctionforcorrelationsbetweenAUCsinducedbythewithin-subjectsdesign, asestimatedbythecorrelationbetweenthejack-knifepseudo-values(seeHanley
&Hajian-Tilaki,1997,fordetails).
Finally,wealsoconductedawhole-headanalysisinordertoinvestigatethe pos-sibilitythatalthoughguiltysuspectsmaybesuccessfulatsuppressinglatepositive parietalERPs,theirguiltmayneverthelessberevealedbyotherERPeffects,such aseffectsrelatedtothecognitivecontrolprocessesthatarerecruitedtosuppress retrieval.Previousresearchhasindicatedthatthesecontrolprocessesaremanifest asearlyfronto-centralorcentro-parietalERPnegativities(Bergströmetal.,2009a,
2009b;Mecklingeretal.,2009)orfrontalslow-driftnegativities(Hanslmayretal.,
2009).Becauseoftheexploratorynatureofthisquestion,weusedadata-driven approachwithmultivariatenon-rotatedspatiotemporalPartialLeastSquares(PLS,
McIntosh&Lobaugh,2004).PLSisapowerfultechniquethatallowsexamination
ofdistributedpatternsofspatialandtemporaldependenciesintheERPdatawith minimalassumptionsregardingthetiminganddistributionofpotentialeffects. PLSanalyzesthe“cross-block”covariancebetweenamatrixofdependent meas-ures(thespatiotemporalERPdistribution)andasetofexogenousmeasures,inthis caseorthogonalcontrastvectorsrepresentingdifferencesbetweenexperimental conditions(thenumberofcontrastsequaltothedegreesoffreedom),thereby con-strainingthesolutiontocovarianceattributabletotheexperimentalmanipulation. InnonrotatedPLS(Bergströmetal.,2009a,2009b;McIntosh&Lobaugh,2004)the sumsofsquaresofthecross-blockcovariancebetweeneachcontrastmatrixand thespatiotemporaldatamatrixaredirectlytestedforsignificanceusingrandom permutationtest.
Inthecurrentanalysis,PLSwasconductedseparatelyforeachofthethree detec-tionphases,eachphasesplitintoearly(0–400ms),middle(400–800ms)andlate (800–1200ms)time-windowsinordertoincreasetemporalresolution.Two con-trasts,onetestingthedifferencebetweenprobesandirrelevantsandanothertesting thedifferencebetweenprobesandtargetswereassessedforsignificanceineach time-windowwith1000permutations.
3. Results
3.1. Validationrecognitiontestandpost-testquestionnaire
Therecognitiontestconfirmedthatparticipantscould
discrim-inatebetweencrime remindersand newwordswithveryhigh
accuracy(98%correct,SEM0.7%)andveryhighconfidence(average
rating 2.9 (SEM 0.02) on a 1–3scale where 3 is highly
confi-dentand 1isnotconfident),and thatcrimememoriescameto
mindveryautomatically(averagerating3.9(SEM0.06)ona1–4
scalewhere4isautomaticrecalland1highlyeffortfulrecall).The
validation testthus confirmedthatcrimeobjectswere strongly
encoded, and that presenting thesecrime probes in our
mem-orydetectiontestwouldbeexpectedtostronglyandreflexively
elicitretrievaloftheassociatedobjectsstoredinmemory.
Con-sistentwiththisexpectation,participantsretrospectivelyjudged
uncooperativerecallsuppressiontobemoreeffortful(t(47)=14.9,
P<0.0001;Cohen’sd=2.89,calculatedasthedifferencebetween
meansdividedbythepooledstandarddeviationtoensure
unbi-asedeffectsizeestimates;Dunlap,Cortina,Vaslow,&Burke,1996)
andlesssuccessful(t(47)=11.1,P<0.0001,d=2.28)than
cooper-ativecrimerecall,which didnotinteractwithexperiment.This
findingindicatesthat recallin responsetodirect cuestocrime
relatedobjectswashighlyautomaticandhadtobeintentionally
suppressed,andthateffortfulsuppressionwasrecruitedinboth
experiments.
3.2. FocalERPresultsatthemid-parietalsite
3.2.1. Grouplevelanalyses
Inbothexperiments,groupaverageERPsatmid-parietal
Fig.1.Groupaveragemid-parietalERPsandscalpmapscontrastingdifferentitemtypeswithinblocksinExperimentone(leftcolumn)andExperimenttwo(rightcolumn). Topographicmapsshowthemeandifferencebetweenprobesandirrelevantsbetween450and800ms.
probes depending on instructions to recall or suppress crime
memories(Fig.1),with largerprobe P300s duringcrime recall
thancrimesuppression. Thescalp mapsshowthat theaverage
differencebetweenprobesandirrelevantsbetween450and800ms
in thecooperative phase displayedthe canonical enhancement
oftenfoundinguiltyknowledgestudies,relativetotheinnocent
phase. Incontrast,when subjectswereuncooperative and
sup-pressedretrieval,thisenhancementwaswilfullyavoided.Statistics
on mean ERP amplitudesbetween 450 and 800ms (when the
P300effectwasmaximal)atthemid-parietalelectrodeconfirmed
thatatthegrouplevel,voluntarymodulationsofmemory-related
P300effectsweresuccessful,asrevealedbysignificantinteractions
betweenthecriticalitemcategories(probe/irrelevant)andguilty
phase(guilty cooperative/guilty uncooperative)in both
experi-ments(Experimentone:F(1,23)=5.57, P<0.05, Partial2=0.20;
Experimenttwo:F(1,23)=10.19,P<0.01,Partial2=0.31).These
interactionsbetweenitemtypeandguiltyphasewerenot
signifi-cantlymodulatedbytheorderinwhichparticipantscompletedthe
differentphases,sinceaddingphaseorderasabetweensubjects
variabledidnotchangethesignificanceofthetwo-way
interac-tionsineitherexperiment,andthethree-wayinteractionsbetween
phaseorder×itemtype×guiltyphasewerenotsignificant(both
Fs<1,P>0.7).
Plannedt-testsrevealedthatinExperimentonewhen
partici-pantswereinthecooperativephase,probeselicitedmorepositive
parietalP300sthanirrelevants,(t(23)=3.1,P<0.01,d=0.48),
repli-catingpriorresearchwithmemorydetectiontests,althoughtarget
P300s were still slightly larger than probe P300s (t(23)=2.8,
P<0.05,d=0.31).Bystarkcontrast,whenparticipantswereasked
tobeuncooperativeandsuppresstheirknowledgeinExperiment
one,theirERPsboreremarkableresemblancetothoseobservedin
theinnocentphase:probeP300swerenotdifferentfromirrelevant
P300sineithertheuncooperativephase(t<1,n.s.,d=0.03)inwhich
crimeknowledgewaspresentinmemory,orintheinnocentphase
(t=1,n.s.,d=0.14)inwhichacrimememorywasabsent.
Signifi-cantlylargerP300stotargetsthantoprobeswereobservedinboth
innocent(t(23)=5.3,P<0.001,d=0.77)anduncooperativephases
Fig.2.Groupaveragemid-parietalERPscontrastingthesameitemtypesacrossblocksinExperimentone(leftcolumn)andExperimenttwo(rightcolumn).
In Experimenttwo,where targetsandprobecategorieswere
oflowerprobabilitythanirrelevantitems,parietalP300sto
tar-gets were indeed almost twicethe magnitude of target P300s
inExperimentone,consistentwithpreviousevidencethatP300
sizeis ofteninversely related tostimulusprobability(Donchin,
1981). P300 amplitudes to probes were also enhanced in the
low probability experiment, in particular in the cooperative
phasewhereprobe P300sweresignificantlymore positivethan
irrelevant P300s (t(23)=6.0, P<0.001, d=1.03), although target
P300s were still larger than probe P300s (t(23)=3.1, P<0.01,
d=0.48).Intheuncooperativephase,probeP300swereenhanced
comparedtoirrelevantP300s(t(23)=2.9,P<0.01,d=0.48),
indi-catingthatsuppressionofmemory-relatedactivitywassomewhat
lesspronouncedwithlowprobabilitythanwithequal
probabil-itycrime reminders.Similarly to Experimentone, targetswere
associated with significantly larger P300s than probes in both
innocent(t(23)=7.6,P<0.001,d=1.81)anduncooperativephases
(t(23)=6.7,P<0.001,d=0.99)ofExperimenttwo.
It is necessary toshowthat thereduced neural responseto
crimeprobesinExperimentsoneandtwowasnotduetosome
phase-genericprocesssuchaspayinglessattentiontoallitems
duringtheuncooperativephase,butratheragenuineandtargeted
efforttocontrolmemory.ComparingERPsacrossthecooperative,
uncooperative,andinnocentphases(Fig.2)showedthatthe
vol-untarymodulationsofP300swerememory-specificbecausethey
wererestrictedtocrimeprobes(dashedboxes),withno
modu-lationsoftargetorirrelevantP300s.T-testsrevealednoreliable
differencesinP300amplitudefortargetsorforirrelevantsineither
experiment(allts≤1,allds<0.17).Incontrast,P300responsesto
probeitemsweresignificantlyreducedintheuncooperative
com-paredtothecooperativephaseinbothExperimentone(t(23)=2.7,
P<0.05,d=0.50)andExperimenttwo(t(23)=3.0,P<0.01,d=0.52).
InExperimentone,P300responsestoprobesduringthe
uncooper-ativephasewerereducedtothepointofbeingindistinguishable
from probes during the innocent phase (t<1, n.s., d=0.07); in
Experimenttwo,uncooperativeprobeP300sweremorepositive
thaninnocentprobeP300amplitudes(t(23)=2.3,P<0.05,d=0.57),
confirmingthatsuppressionwasnotcompleteinthelow
probabil-ityexperiment,thoughclearlysuccessfultoa significantextent.
These findings converge to indicate that modulations of P300
amplitudewereentirelyselectivetocrimeprobes,reflecting
vol-untarymodulationsofmemory-specificbrainactivity.
3.2.2. Individualguiltclassification
The overall pattern of results in the individual guilt
confidencethresholds(Fig.3).Theseresultsshowthatbootstrap
testingofbase-to-peakP300andmeanP300amplitudesproduced
moreaccurateclassificationthantheP300peak-to-peakmethod,
inparticularforExperimentone.Usingastrictthresholdof95%CI
Fig.3.Percentageofparticipantsclassifiedasguiltyusingdifferentstatisticsforthe bootstraptestatdifferentguiltclassificationthresholds.
resultedinacceptablefalsepositiveratesforallstatistics,whereasa
liberalthresholdof90%CIresultedinreducedspecificity(in
partic-ularforthebase-to-peakandpeak-to-peakP300statistics),without
muchincreaseinsensitivity.Withaliberal90%confidencelevel,
anunacceptablyhighpercentageofparticipantswereerroneously
classified as guilty in the innocentphase(i.e. false alarms) for
thebase-to-peakandpeak-to-peakP300measures(approx.30%).
Allmeasuresshowedmorereasonablefalsealarmrates(12.5%or
lower)withastrict95%confidence threshold,exceptthe
peak-to-peakmeasurewhich wasstillratherhighinExperimenttwo
(17%). The peak-to-peakmeasure wasalsoparticularly
insensi-tiveatdiscriminating betweenprobesand irrelevantsacrossall
phasesinExperimentone,likelybecausethepost-P300negative
peak(Soskinsetal.,2001)wasnotverypronouncedwithequal
probabilityprobes.3
InExperimentone,forallmeasures,thepercentageof
individ-ualparticipantsclassifiedasguiltywashighlysimilaracrossthe
innocentphase,inwhichtherewasnocrimeknowledge,andthe
uncooperativephase,inwhichtheparticipanthadguilty
knowl-edge.Thissimilaritybetweeninnocentanduncooperativephases
contrastedwiththecooperativephase,in whichdetectionrates
werenotablyhigher(withtheexceptionofthepeak-to-peak
mea-sure,asdescribedabove).InExperimenttwowithinfrequentguilty
knowledgeprobes,correctdetectionratesduringtheguiltyphases
oftheexperimentwereoverallhigherthaninExperimentone.Even
under thesemore favourable conditions, however, suppressing
crimememoriessubstantiallyreduceddetectionratescompared
towhenparticipantsrecalledtheircrime.
Oncloserinspection,aroundhalfofparticipantsinExperiment
onedidnotshowreliableP300enhancementstotargetscompared
toirrelevantitemswithanyparticularmeasure.Thelowdetection
rateinthisexperimentwasthuslikelyaresultofreduced
signal-to-noise(SNR)ratiobecauseofoverallsmallerP300swhenprobes
werepresentedmorefrequently.Inordertoassessthepatternof
detectionforsubjectswithgoodSNR,wealsotestedclassification
foreachmeasureandthresholdwhenincludingonlyparticipants
thatshowedareliabletarget-irrelevantdifferenceinthat
analy-sis(sincethisdifferencecanbeconsideredabenchmarkagainst
whichtheprobe-irrelevantdifferencecanbecompared,e.g.Farwell
&Donchin,1991,seealsoIacono,2007),aspresentedinTable1.
InExperimentone,theseanalysesselectivelyenhanceddetection
ratesduringthecooperativephasesbyonaverage24%whereas
averagedetectionratesfor theuncooperative phaseswereonly
increasedby4%.InExperimenttwo,themajorityofparticipantshad
reliabletarget-irrelevantdifferencessotheconditionalisedanalysis
onlyincreasedaveragedetectionratesby4%inbothphases.Thus,
whenincludingonlyhighSNRparticipants,cooperativeguilt
clas-sificationwasmoresimilaracrossequiprobableandrareprobes.
Importantlyforourconclusions,theconditionalisedanalysis
con-firmedthatthelowuncooperativedetectionrateinExperimentone
wasnotduetolowSNR.Rather,lowSNRappearedtoobscurethe
differencebetweencooperativeanduncooperativephases,since
thisdifferencewasenhancedwhenparticipantsthatfailedtoshow
reliabletarget-irrelevantdifferenceswereexcluded.
3Thepeak-to-peakP300measurehasbeenproposedassuperiorfordetecting
guiltyknowledgecomparedtootherbootstrapmeasures(Soskinsetal.,2001),and ourfindingsappearinconsistentwiththisargument.However,thepeak-to-peak measurereliesonaccuratelyestimatingthepost-P300negativepeakthatwasnot verypronouncedinourdata,particularlyinExperimentone.Sinceourexperiments (particularlyExperimentone)hadsomenotabledesigndifferencesfromprevious GKTprotocols,thesedifferencesmayhavealteredthecognitiveprocessesinvolved duringthetaskandtherebyaffectedtheERPmorphology.Thus,wedonotthink ourresultsspeakclearlytotheissueofwhichbootstrapmeasureisgenerallymost effectiveforguiltyknowledgedetectionacrossdifferentparadigms.
Table1
ConditionalisedguiltclassificationandAreaUnderCurveresultsacrossexperimentsandbootstrapmeasures.
Phase CI Experimentone Experimenttwo
Base-to-peak Mean Peak-to-peak Base-to-peak Mean Peak-to-peak
% N % N % N % N % N % N
Conditionalisedguiltclassification
Cooperative 95% 50 10 67 9 29 7 76 21 53 17 70 20 90% 58 12 63 11 33 9 73 22 76 21 82 22 Uncooperative 95% 14 7 10 10 11 9 39 23 28 18 55 22 90% 14 14 17 12 25 12 48 23 35 20 61 23 Innocent 95% 0 13 0 11 0 14 14 21 6 17 17 24 90% 13 15 25 12 20 15 32 22 5 19 33 24
Areasundercurves(standarderrorsinparentheses)
Cooperativevs.innocent 0.60(0.08) 0.66(0.07) 0.47(0.08) 0.84(0.07) 0.87(0.06) 0.83(0.06)
Uncooperativevs.innocent 0.47(0.10) 0.52(0.14) 0.45(0.08) 0.69(0.09) 0.70(0.09) 0.76(0.08) AUCdifferencepairedZ-test
Z-score 1.61 1.62 0.17 2.13 2.34 1.10
P-value 0.054 0.053 0.431 0.017 0.010 0.077
Note:CI,confidenceinterval;%,percentageofparticipantsthathadreliablylargerP300sforprobesthanirrelevantsoutofthosethatalsohadreliablelargerP300sfortargets thanirrelevantsintherelevantanalysis;N,numberofparticipantsincluded.
Table2
Significancevaluesofthecontrastsinthewhole-headPLSanalysisasestimatedby1000permutations.
Phase Comparison Experimentone Experimenttwo
0–400ms 400–800ms 800–1200ms 0–400ms 400–800ms 800–1200ms
Cooperative Probesvs.irrelevants 0.344 0.006 0.029 0.400 0.001 0.001
Probesvs.targets 0.120 0.001 0.183 0.302 0.005 0.020
Uncooperative Probesvs.irrelevants 0.619 0.345 0.390 0.216 0.158 0.450
Probesvs.targets 0.023 0.000 0.236 0.042 0.000 0.009
Innocent Probesvs.irrelevants 0.758 0.748 0.950 0.787 1.000 0.907
Probesvs.targets 0.126 0.000 0.229 0.002 0.000 0.001
The AUC summary measures of classification performance (Table 1) confirmed the general picture that, even though the
group-levelstatistics werehighlysignificant,individual
classifi-cationwasoverall poorinExperiment one.In Experimenttwo,
classificationperformance calculated onthebasis of the
Coop-erative“hits”versusInnocent“falsealarms”ROCwassimilarto
previousskin-conductance GKTresearch (Ben-Shakhar &Elaad,
2003) but still lower than previous P300-based GKT research
(Ben-Shakhar,personal communication).Importantly,
classifica-tionperformance waslower inboth experiments and usingall
measureswhencalculatedbasedon“hits”intheUncooperative
condition.ComparingtheAUCmeasureswithajack-knifebased
pairedZ-test(Hanley&Hajian-Tilaki,1997)revealedsignificantly
ormarginallysignificantlyhigherCooperativethanUncooperative
AUCvaluesforallbootstrapmeasuresinbothexperiments,withthe
exceptionofthepeak-to-peakmeasureinExperimentonesince
thismeasurefailedtodetectguiltyknowledgeinbothphasesof
Experimentone(Table1).TheAUCanalysisthusconfirmedthatthe
suppression-inducedreductioninindividualdetectionrateswas
notdependentonarbitraryclassificationcut-offpoints.
Insum,theindividualbootstrapresultsconfirmedthata
size-able proportion of participants successfully suppressed crime
recallandevadeddetection.Althoughexactsuccessratesvaried
somewhatdepending onthe specificclassification method and
threshold,theglobalpatternwasconsistent,with22%more
partic-ipantsclassifiedasguiltywhencooperatingthanwhensuppressing
theircrimememories,averagedacrossexperiments,measuresand
thresholds.
3.3. ERPwhole-headPLSanalysis
The result of the PLS analysis testing pairwise contrasts in
each of thedetection phases and time windowsare presented
inTable2.Therewerehighlysignificantwhole-headdifferences
betweentargetsandprobesinalldetectionphasesinboth
exper-iments, with the differencebeing maximally significant in the
400–800ms time-window. The whole-headdifference between
probesandirrelevantswasalsohighlysignificantduringthe
coop-erativephases,between400and800msinExperimentoneand
between400and1200msinExperimenttwo.However,therewere
nosignificantdifferencesbetweenprobesandirrelevantsinthe
innocentoruncooperativephasesineithertime-windowandin
eitherexperiment.Thesewhole-headPLSresultsthuscapturedthe
sameP300patternasthefocalparietalanalysis(withthe
excep-tionthattheP300differencebetweenprobesandirrelevantsin
Experimenttwodidnotreachsignificanceinthewhole-head
anal-ysis),withnoadditionalERPindicationsofguiltacrosstheglobal
spatiotemporal datawhen participantswereuncooperativeand
suppressingcrimerecall.
3.4. Finalrecallresults
Meanaccuracy onthefinal testsispresented inTable3.On
the four-choicelocation accuracy task, performance was above
chance in all conditions, but there were no significant main
Table3
Proportioncorrectresponsesacrossthelocationandobjectfinalrecalltests.
Test Condition Experimentone Experimenttwo
M SEM M SEM Location Recall 0.65 0.05 0.63 0.05 Suppress 0.55 0.04 0.67 0.04 Baseline 0.60 0.04 0.62 0.04 Object Recall 0.67 0.05 0.46 0.05 Suppress 0.52 0.05 0.41 0.05 Baseline 0.46 0.04 0.40 0.04
effectsor interactions. When participantswere asked to recall whichobjectsappearedineachroom,therewasamaineffectof condition(F(2,92)=11.4,P<0.001,Partial2=0.20),butno
inter-actionwithexperiment(F(2,92)=1.2,P>0.3,Partial2=0.03).The
conditionmaineffect wascaused bysignificantly higherobject recallaccuracy(againstaBonferronicorrected˛=0.0167forthree family-wiseposthoctests)for repeatedlyrecalled objectsthan for repeatedlysuppressed items(t(47)=2.5, P<0.0167, d=0.37) andforbaselineitems(t(47)=5.0,P<0.001,d=0.68),and trend-levelhigher objectaccuracyfor suppressedthanbaselineitems (t(47)=2.0,P=0.046,d=0.27)collapsedacrossexperiments.
4. Discussion
Thecurrentresearchisthefirstdemonstrationthat memory-related brainactivitynormallyelicited by remindersof incrim-inatingknowledgecanbeintentionallysuppressed.Inlinewith typicalguilty knowledge research,we used exceedingly strong retrievalcues as crimeprobes – object wordsas reminders of elaboratelyencodedobjectpictures–which wouldhavegreatly magnified recall prepotency compared to previously published retrievalsuppression studies(Andersonet al., 2004; Bergström etal.,2007,2009a,2009b;Depueetal.,2007;Hanslmayretal., 2009;Mecklingeretal.,2009).Moreover,ERPstothesepowerful
cueswerecomparedwithERPstopreviouslyunseencontrolitems,
unlikelytoelicitmemory-relatedbrainactivity.Despitethese
chal-lenges,suppressionofparietalERPpositivitywashighlysuccessful,
reducingdetectionratesofguiltyindividuals.
4.1. Theoreticalimplications
Importantly,voluntarymodulationsofparietalERPamplitudes
werespecifictocrimeprobes,withnosignificantmodulationsof
ERPstotargetsorcontrolitemsacrossphasesineitherexperiment.
ThisresultisnoteworthybecauseitconfirmsthattheprobeERP
effectswereunlikelyduetosomephase-genericprocess,suchas
participantspayinglessattentiontoallitemsintheuncooperative
phase.Rather,theERPdifferencesbetweenprobesacross
cooper-ativeanduncooperativephasescanbeconfidentlyinterpretedas
successfulvoluntarymodulationsofmemory-specificbrain
activ-ity.In fact, when we presentedfrequent reminders of a crime,
guiltyparticipants’effortstosuppressretrievalrenderedparietal
amplitudesindistinguishablefromthoseexhibitedwhentheywere
innocent.ThissimilarityofERPprofilesacrosstheuncooperative
andinnocentphasessuggeststhatwhenprobesand irrelevants
areequallyprobable,suspectscanvoluntarilysuppressretrieval
andmodulateneuralresponsestocrimeprobes,disguisingstored
knowledgeso thattheyappear innocent. Thisresult is thusan
importantextensiononpreviousfindingsthatthelateparietalERP
positivitythatistypicallyinterpretedasacorrelateofepisodic
rec-ollectionislargelyundervoluntarycontrol(Bergströmetal.,2007,
2009a,2009b;Dzulkifli&Wilding,2005;Hanslmayretal.,2009; Herron&Rugg,2003;Mecklingeretal.,2009).Thisabilityto
con-trolretrievalmayoriginate,inpart,fromthepreviouslyestablished
capacitytointentionallymodulatemnemonicactivityinthemedial
temporallobesandotherbrainareasinvolvedinmemory
repre-sentationandretrieval(Andersonetal.,2004;Benoit&Anderson,
2012;Butler&James,2010;Depueetal.,2007;Levy&Anderson, 2012).
Inasecondexperiment,wereducedtheprobabilityofcrime
reminderstomakethemsubjectivelyunexpected,inorderto
inves-tigate whether probability-sensitive parietal ERP effects would
be amenable to suppression. This issue is important because
the majority of GKT applications use low-probability designs.
A significant proportion of participants successfully concealed
crime memories even under these more demanding
circum-stances,althoughsuppressionwassomewhatlesspronouncedthan
withequal probabilitycrimereminders. Thisfindingsuggests a
boundaryconditionforvoluntarycontrolofmemory-relatedbrain
activity,indicatingthatsubjective expectednessisanimportant
factorforsuppressionsuccess.
The precise functional significance of this result is
how-ever unclear, because parietal ERP positivities in the P300
time-window arethought toindex additivecontributions from
multiple independent sources (Johnson, 1986). In memory
tasks, probability-sensitive cognitive processes related to
stim-ulus evaluation (e.g. Sutton, Braren, Zubin, & John, 1965) and
probability-insensitivecognitiveprocessesthattrackepisodic
rec-ollectionsuccess(e.g.Vilbergetal.,2006)maybothproduceERP
positivitiesthatoverlapatparietalscalpregions.Itistherefore
pos-siblethatvaryingtheprobeprobabilityacrossexperimentsmay
havechangedtherelativecontributionsoftheseunderlying
com-ponentprocessestoscalpamplitudes.InExperimentone,parietal
ERPpositivitiesmayhaveprimarilybeenrelatedtoepisodic
rec-ollection(ER).InExperimenttwo,theenhancedP300amplitudes
may have beenthe result of additiveepisodic recollection and
probability-related‘oddball’(O)processes(i.e.ER+O).One
specu-lativeinterpretationofourdataisthatthesedistinctprocessesare
differentiallysusceptible tovoluntarycontrolmechanisms,with
theepisodicrecollection-relatedERcomponentbeingmore
sus-ceptiblethantheprobability-sensitiveOcomponent.Thisaccount
could explain why the relative difference between cooperative
(ER+)anduncooperative(ER−)probeswassimilaracrossthetwo
experimentseventhoughERPamplitudestoprobeswereoverall
enhancedinbothconditionsinExperimenttwo(O+).Alternatively,
infrequentremindersmaysimplymakeitmoredifficulttoengage
thecognitivecontrolmechanismsrequiredtosuppressretrieval
duetoalackofpreparation(cf.Hanslmayretal.,2009)orpractice.
Inadata-drivenwhole-headanalysis,wetestedforalternative
ERPeffectsthatmighthavebeenusedtodiagnoseguiltwhenthe
P300wassuppressed,suchaseffectsrelatedtothecognitivecontrol
processesrecruitedtosuppressretrieval.Surprisingly,eventhough
thebehaviouraldataandself-reportsindicatedthatretrieval
sup-pressionrequiredagreatdealofeffort,thisanalysisfailedtoreveal
reliableevidenceofcontrol-relatedERPeffects,4incontrastto
pre-viousliterature(Bergströmetal.,2009a,2009b;Hanslmayretal.,
2009;Mecklingeretal.,2009).Oneexplanationforthis
discrep-ancymaybethatpreviousresearchhasrequiredparticipantsto
switchonatrial-by-trialbasisbetweensuppressingand
retriev-inginformationfromthesameeventcontext,whereasthecurrent
taskrequired participantstosuppresscrimeretrievalacrossan
entirephaseoftheexperiment.Suchblockedretrievalsuppression
mayinvolvemoresustainedpatternsofbrainactivationthanthose
observedinpriorresearch.Sustainedbrainactivitycanbedifficult
tomeasurewithERPs,sinceERPsareprimarilysensitiveto
tran-sientbrainactivitypatternsthataretime-lockedtotheonsetof
externallydefinedevents.
Thefinalrecalltestconductedaftertheguiltdetectionphase
confirmed that theretrieval suppression manipulation affected
later retention of the tested memories. Compared torecall for
objectsinthecooperativephase,objectsrepeatedlycuedinthe
uncooperativephaseshowedreliablypoorermemory,indicating
4TherewasatrendformorenegativeERPsacrossfrontalsitesforprobesinthe
sametime-windowastheparietalP300,particularlyinthecooperativephase(see scalpmapsinFig.1).Thesefrontalnegativitieswerethusnotspecificallyrelatedto suppression,butratherseemedinverselyrelatedtoparietalP300positivities,and thusmayreflectthenegativepoleofadipolarfieldstemmingfromthesame gener-atorastheP300.Sincethefrontalnegativitiesdidnotprovideadditionaldiagnostic informationregardingguiltorinnocenceinourdata(althoughseeFarwell&Smith, 2001)wedonotreportfurtherontheseeffects.
thatparticipantswereabletolimitcrimerecalldespiterepeated
reminders. However, suppressed items were still somewhat
strengthenedcomparedtobaselineobjectsthatwerenever
pre-sentedin theintermediate GKTtask. Thisresult contrasts with
previousfindingsthatrepeatedlysuppressingretrievalinresponse
toareminder cansignificantlyimpairrecallperformanceofthe
avoidedmemoriescomparedtobaseline(e.g.Anderson&Green,
2001;Andersonetal.,2004;Bergströmetal.,2009b;Depueetal.,
2007).The trendlevel enhancement for suppresseditems
rela-tivetobaselineitemsindependentlyconfirmsthesubjectivelyvery
highdifficultyparticipantsreportedfortherecallsuppressiontask,
confirmingthatthemnemonic controlevidentin reduced
pari-etal amplitudewas theproduct of effortful control.Such great
intrusivenesswasexpected,giventhatthecuesdirectlyreferred
totheobjectsthemselvesand thuslikelysharedmany
overlap-pingfeatureswiththeto-be-suppressedmemories(seeAnderson
&Spellman,1995).Therefore,retrievalsuppressionislikelytobe
moredifficultwiththeoverlappingcuesusedinGKTcompared
tothenon-overlappingcue-associatepairsusedinothermemory
suppressionresearch(e.g.Anderson&Green,2001).Nevertheless,
itispossiblethatunderadifferentprotocol,criminalsmaybeable
tosuppressmemoriesoftheircrime tothepoint of significant
below-baselineforgetting.
4.2. Practicalimplications
Afewpreviousstudieshaveshownthatcountermeasurescan
degradeERP-basedmemoryguiltdetection.Somehaveinvolved
trainingparticipantsoncovertresponsestoirrelevantcontrolitems
(e.g.Rosenfeld et al., 2004,2008; Meixner & Rosenfeld, 2010),
withtheaimofincreasingP300stoirrelevantsandthusmaking
probesandirrelevantsmoresimilar.Othershavetrained
partici-pantstomakeadditionalcovertresponsestotargetitemsinorder
toenhance theirrelative salience, thereby reducingthe
atten-tionalresourcesavailableforprocessingofprobesandirrelevants
(Mertens&Allen,2008).Importantly,noprevious studieshave
assessedwhethermemory-relatedbrainactivityinresponsetocrime
reminderscanbevoluntarilyandspecificallycontrolled.Ourstudy
isthusthefirsttochallengethecriticalassumptionthat
memory-relatedbrainactivityisautomaticallyelicitedwhensuspectsare
presentedwithcrimereminders.Thecurrentresultsshowthatthis
assumptionofmemorydetectiontestsisnotalwaysjustified.
Thegeneralizabilityofourfindingsissomewhatcomplicated
bythemultitudeofdifferentguiltyknowledgeprotocolsand
clas-sificationtechniquesthathavebeendevelopedbydifferentgroups
(e.g.Allenetal.,1992;Farwell&Donchin,1991;Rosenfeldetal.,
1991,2004).Althoughwekeptourdesignassimilaraspossibleto
thetypeofprotocolthathasarguablybeenthemostprevalentin
theGKTliterature,wedidintroducesomenoveldesignelements
inordertobeabletomanipulateretrievalsuppressionand
investi-gatesubsequenteffectsonmemory(Anderson&Green,2001).For
example,weemployedaveryelaborateencodingphasetoensure
thatcrimememorieswouldbehighlyintrusive,andour
manip-ulationuseddifferentinstructionsfrompracticalapplicationsof
GKTresearch.Themajority of appliedGKTprocedures uselow
probabilityprobes,soalthoughourfirstexperimentwas
theoret-icallyimportantforpriormemorysuppressionwork,thesecond
experimentwithlowprobabilityprobeswasmostrelevantto
prac-ticalapplications.Therefore,whilstourdesignemphasisedinternal
validity,aspectsofthedesignhadreducedecologicalvalidity.
Fur-thermore,newlydevelopedprotocolshavebeendemonstratedas
moreaccurateand resistantto othertypesof countermeasures
(Rosenfeldetal.,2008;Rosenfeld&Labkovsky,2010).Whetheror
notretrievalsuppressioncanbesuccessfullyappliedinthese
alter-nativeprotocolsisanempiricalquestion.Furtherresearchisalso
requiredtodeterminewhethersuppressionattemptscanreduce
guiltdetectionbasedonbehaviouralorANSmeasuresofmemory
(reviewedinBen-Shakhar&Elaad,2003).
Itisalsocrucialthatsuppressioncountermeasuresareassessed
outside thelaboratory.First,memoriesofa realcrimemay
dif-ferinintrusivenessfromthoseofacrimesimulation,whichcould
affectsuppressionsuccess.Second,realcriminalswilllikelydiffer
intheirmotivationtocontrolretrievalfromtypicalresearch
vol-unteers.Sinceouraimwastoexperimentallydemonstratepeople’s
capacitytocontrolmemory-relatedbrainactivity,wemanipulated
whetherourvolunteersshouldretrieveorsuppresscrime
memo-ries.Ofcourse,inarealcrimesetting,asuspectismorelikelyto
attempttosuppressthanintentionallyretrievecrimememories.
Incontrast,volunteersparticipatinginlab-basedstudiesonguilty
knowledge testing are likely morecooperative than real
crimi-nalsuspects.Typicalresearchvolunteerswithoutcountermeasure
instructionswillhavelittlemotivationtosuppressretrieval,and
mayeven intentionallyretrievecrimememoriesinresponse to
demandcharacteristics(e.g.Orne,1962).Suchcooperationfrom
volunteerswouldleadtoanover-estimationofthetest’sabilityto
detectmemoriesinarealcrimesetting.
Nevertheless,despitemethodologicaldifferences,thecurrent
researchhasimportantpracticalimplicationsbecauseitchallenges
theassumptionthatmemory-relatedbrainactivityisoutsideof
vol-untarycontrol,andhencemoreresistanttocountermeasuresthan
otherphysiologicalandbehaviouralmeasures(e.g.Lykken,1998;
seeBen-Shakhar,2011).Our findingsshowedanunprecedented
degreeofsuccessatintentionalcontrolovermemory-relatedbrain
activity,withlittlespecial training,andthat thiscontrolcanbe
marshalledasaneffectivecountermeasureduringmemory
detec-tiontests.Theseresultsthusdemonstratethatinprinciple,retrieval
suppressionposesachallengetoguiltdetectionteststhatrelyon
brain-activitymarkersofmemory.
The ability of uncooperative suspects motivated to disguise
theirguilttosuppressmemory-relatedneuralactivityestablished
hereraisesconcernsregardingthevalidityofERP-basedmemory
detectiontestsasameanstoestablishcriminalguiltorinnocence.
Togetherwithpreviousevidence(Mertens&Allen,2008;Rosenfeld
etal.,2004),thecurrentfindingspointtoaspecificproblemwith
enhancedfalsenegativesinmemorydetectiontests,meaningitis
particularlyriskytoconcludethatasuspectisinnocentbasedona
negativeresult.Theseconcernsareimportantbecauseithasbeen
arguedthatERPP300-basedmethodscanreliablydemonstratethat
asuspectlacksknowledgeofacrime,andthatthesetechniques
havebeeninstrumentalinwinningthereleaseofcriminalsuspects
inreallegalcases(seeMcCall,2004).Althoughnoteveryparticipant
managedtoavoiddetectionunderallcircumstances,ourresults
indicatethat,evenundersituationsmostfavourableforguilt
detec-tion,atleastanadditional20%ofguiltycriminalsuspectscould
bemisclassifiedandpotentiallysetfreeasaresultofintentional
retrievalsuppression,whichmayhavedirereallifeconsequences.
Althoughscepticismhasprevailedwhenthistypeofevidencehas
beenconsideredinUSlegalcases,boththemethodsandtheirlegal
standingarestillevolving,asrecentlynotedinScienceMagazine
(Miller,2010).
5. Conclusions
Across two experiments, we have demonstrated that the
absenceofareliablyenhancedbrainresponsetocrimereminders
isnotunequivocalevidencethatrelevantmemoriesareabsentin
thatperson’sbrain.Instead,thepresenceorabsenceof
memory-related brain activity appears to primarily track whether that
personishavingasubjectiveexperienceofremembering(cf.Allen
&Mertens,2009;Rissmanetal.,2010).Anabsenceof
notrememberinganassociatedcrimeatthatspecifictime,butdoes
notdeterminethattheyhavenosuchcrimememoriesstoredin
theirbrain.Aninnocentverdictinaguiltyknowledgetestcould
arisebecauseasuspectistrulyinnocent,becausetheyhave
for-gottentheparticulardetailsofthecrimethatarebeingtested,or
becausetheyarehighlymotivatedtodisguisetheirknowledgeand
areintentionallysuppressingcrimememories.
Acknowledgements
WethankChristinaMeierfordatacollectionhelp,andProf.
Ger-shonBen-Shakharand ananonymousreviewerforconstructive
feedbackduringthereviewprocess.Thisworkwassupportedby
DFGGrantsRI-1847/1-1,SFB779TPA7,andSFB779TPA10N
(A.R.-K.),theAlexandervonHumboldtFoundation(A.R.-K.),theFederal
State of Sachsen-Anhalt (A.R.-K.), MRC Grant MC-A060-5PR00
(M.C.A.),BBSRCGrantBB/G014795/1(J.S.S.),andbyajointaward
fromtheMRCandtheWellcomeTrust(J.S.S.).
References
Allen,J.J.,Iacono,W.G.,&Danielson,K.D.(1992).Theidentificationofconcealed
memoriesusingtheevent-relatedpotentialandimplicitbehavioralmeasures:
Amethodologyforpredictioninthefaceofindividualdifferences.
Psychophysi-ology,29,504–522.
Allen,J.J.B.,&Mertens,R.(2009). Limitationstothedetectionofdeception:True
andfalserecollectionsarepoorlydistinguishedusinganevent-relatedpotential
procedure.SocialNeuroscience,4,473–490.
Anderson,M.C.,Ochsner,K.N.,Kuhl,B.,Cooper,J.,Robertson,E.,Gabrieli,S.W.,etal. (2004). Neuralsystemsunderlyingthesuppressionofunwantedmemories.
Science,303,232–235.
Anderson,M.C.,&Green,C.(2001).Suppressingunwantedmemoriesbyexecutive
control.Nature,410,366–369.
Anderson,M.C.,&Huddleston,E.(2011).Towardsacognitiveandneurobiological
modelofmotivatedforgetting.InR.F.Belli(Ed.),Nebraskasymposiumon
moti-vation(Vol.58)Trueandfalserecoveredmemories:Towardareconciliationofthe
debate(pp.53–120).NewYork:Springer.
Anderson,M.C.,&Spellman,B.A.(1995).Onthestatusofinhibitorymechanismsin
cognition:Memoryretrievalasamodelcase.PsychologicalReview,102,68–100.
Ben-Shakhar.(2011).Countermeasures.InB.Verschuere,G.Ben-Shakhar,&E.
Mei-jer(Eds.),Memorydetection:TheoryandapplicationoftheConcealedInformation
Test(pp.200–214).Cambridge:CambridgeUniversityPress.
Ben-Shakhar,G.,&Elaad,E.(2003).Thevalidityofpsychophysiologicaldetectionof
informationwiththeGuiltyKnowledgeTest:Ameta-analyticreview.Journalof
AppliedPsychology,88,131–151.
Benoit,R.,&Anderson,M.C.(2012).Opposingmechanismssupportthevoluntary
forgettingofunwantedmemories.Neuron,76,450–460.
Bergström,Z.M.,DeFockert,J.,&Richardson-Klavehn,A.(2009a). Event-related
potentialevidencethatautomaticrecollectioncanbevoluntarilyavoided.
Jour-nalofCognitiveNeuroscience,21,1280–1301.
Bergström,Z. M.,De Fockert, J.,&Richardson-Klavehn, A.(2009b). ERP and
behaviouralevidencefordirectsuppressionofunwantedmemories.
Neuroim-age,21,1280–1301.
Bergström,Z.M.,Velmans,M.,DeFockert,J.,&Richardson-Klavehn,A.(2007).
ERPevidenceforsuccessfulvoluntaryavoidanceofconsciousrecollection.Brain
Research,1151,19–133.
Butler,A.J.,&James,K.H.(2010).Theneuralcorrelatesofattemptingtosuppress
negativeversusneutralmemories.Cognitive,Affective,&BehavioralNeuroscience,
10,182–194.
Delorme,A.,&Makeig,S.(2004).EEGLAB:Anopensourcetoolboxforanalysisof
single-trialEEGdynamicsincludingindependentcomponentanalysis.Journal
ofNeuroscienceMethods,134,9–21.
Depue,B.E.,Curran,T.,&Banich,M.T.(2007).Prefrontalregionsorchestrate
sup-pressionofemotionalmemoriesviaatwo-phaseprocess.Science,317,215–219.
Donchin,E.(1981).Surprise!...Surprise?Psychophysiology,18,493–513.
Dunlap,W.P.,Cortina,J.M.,Vaslow,J.B.,&Burke,M.J.(1996). Meta-analysisof
experimentswithmatchedgroupsorrepeatedmeasuresdesigns.Psychological
Methods,1,170–177.
Dzulkifli,M.A.,&Wilding,E.L.(2005). Electrophysiologicalindicesofstrategic
episodicretrievalprocessing.Neuropsychologia,43,1152–1162.
Efron,B.,&Tibshirani,R.(1986).Bootstrapmethodsforstandarderrors,confidence
intervals,andothermeasuresofstatisticalaccuracy.StatisticalScience,1,54–75.
Farwell,L.A.,&Donchin,E.(1991).Thetruthwillout:Interrogativepolygraphy(lie
detection)withevent-relatedbrainpotentials.Psychophysiology,28,531–547.
Farwell,L.A.,&Smith,S.S.(2001).UsingbrainMERMERtestingtodetectknowledge
despiteeffortstoconceal.JournalofForensicScience,46,135–143.
Garland,B.,&Glimcher,P.W.(2006). Cognitiveneuroscienceandthelaw.Current
OpinioninNeurobiology,16,130–134.
Greely,H.T.,&Illes,J.(2007).Neuroscience-basedliedetection:Theurgentneed
forregulation.AmericanJournalofLawandMedicine,33,377–431.
Giridharadas,A.(2008,September).India’snoveluseofbrainscansincourtsisdebated. TheNewYorkTimes.http://www.nytimes.com/
Hanley,J.A.,&Hajian-Tilaki,K.O.(1997). Samplingvariabilityofnonparametric
estimatesoftheareasunderreceiveroperatingcharacteristiccurves:Anupdate.
AcademicRadiology,4,49–58.
Hanslmayr, S., Leipold, P., Pastötter,B., & Bäuml,K.-H. (2009). Anticipatory
signatures of voluntary memory suppression. Journal of Neuroscience, 29,
2742–2747.
Herron,J.E.,Quayle,A.H.,&Rugg,M.D.(2003).Probabilityeffectsonevent-related
potentialcorrelatesofrecognitionmemory.CognitiveBrainResearch,16,66–73.
Herron,J.E.,&Rugg,M.D.(2003).Retrievalorientationandthecontrolof
recollec-tion.JournalofCognitiveNeuroscience,15,843–854.
Iacono,W.G.(2007).Detectionofdeception.InJ.T.Cacioppo,L.G.Tassinary,&G.
G.Berntson(Eds.),Handbookofpsychophysiology(3rded.,Vol.58,pp.688–703).
Cambridge,UK:CambridgeUniversityPress.
Johnson,R.(1986). AtriarchicmodelofP300amplitude.Psychophysiology,23, 367–384.
Levy,B.J.,&Anderson,M.C.(2012).Purgingofmemoriesfromconsciousawareness
trackedinthehumanbrain.JournalofNeuroscience,32,16785–16794.
Lykken,D.T.(1959).TheGSRinthedetectionofguilt.JournalofAppliedPsychology,
43,385–388.
Lykken,D.T.(1998). Atremorintheblood:Usesandabusesoftheliedetector.New
York:PlenumTrade.
McCall, B. (2004, February). Brain fingerprints under scrutiny. BBC News.
http://www.bbc.co.uk
McIntosh,A.R.,&Lobaugh,N.J.(2004).Partialleastsquaresanalysisofneuroimaging
data:Applicationsandadvances.Neuroimage,23(Suppl.1),S250–S263.
Mecklinger,A.,Parra,M.,&Waldhauser,G.T.(2009). ERPcorrelatesofintentional
forgetting.BrainResearch,1255,132–147.
Meegan,D.V.(2008). Neuroimagingtechniquesformemorydetection:Scientific,
ethical,andlegalissues.AmericanJournalofBioethics,8,9–20.
Meixner,J.B.,&Rosenfeld,J.P.(2010).CountermeasuremechanismsinP300-based
concealedinformationtest.Psychophysiology,47,57–65.
Mertens,R.,&Allen,J.J.B.(2008).Theroleofpsychophysiologyinforensic
assess-ments:Deceptiondetection,ERPs,andvirtualrealitymockcrimescenarios.
Psychophysiology,45,286–298.
Miller,G.(2010,May).fMRIliedetectionhearingends,decisionstilltocome. Scien-ceInsider.http://news.sciencemag.org/scienceinsider
Orne,M.T.(1962).Onthesocialpsychologyofthepsychologicalexperiment:With
particularreferencetodemandcharacteristicsandtheirimplications.American
Psychologist,17,776–783.
Paz-Alonso,P.M.,Bunge,S.A.,Anderson,M.C.,&Ghetti,S.(2013). Strengthof
couplingwithinamnemoniccontrolnetworkdifferentiatesthosewhocanand
cannotsuppressmemoryretrieval.JournalofNeuroscience,33,5017–5026.
Polich,J.(2007). UpdatingP300:AnintegrativetheoryofP3aandP3b.Clinical
Neurophysiology,118,2128–2148.
Rissman,J.,Greely,H.T.,&Wagner,A.D.(2010). Detectingindividual
mem-ories throughtheneuraldecodingofmemorystatesandpastexperience.
ProceedingsoftheNationalAcademiesofSciencesofUnitedStatesofAmerica,107,
9849–9854.
Rosenfeld,J.P.(2006).‘Brainfingerprinting’:Acriticalanalysis.TheScientificReview
ofMentalHealthPractice,4,20–37.
Rosenfeld,J.P.,Angell,A.,Johnson,M.,&Qian,J.-H.(1991).AnERP-based,
control-question lie detector analog:Algorithms for discriminatingeffects within
individuals’averagewaveforms.Psychophysiology,28,319–335.
Rosenfeld,J.P.,&Labkovsky,E.(2010). NewP300-basedprotocoltodetect
con-cealedinformation:Resistancetomentalcountermeasuresagainstonlyhalfthe
irrelevantstimuliandapossibleERPindicatorofcountermeasures.
Psychophys-iology,47,1002–1010.
Rosenfeld,J.P.,Labkovsky,E.,Winograd,M.,Lui,M.A.,Vandenboom,C.,&Chedid,E. (2008). TheComplexTrialProtocol(CTP):Anew,countermeasure-resistant,
accurate P300-based methodfor detection ofconcealed information.
Psy-chophysiology,45,906–919.
Rosenfeld, J. P., Soskins, M., Bosh, G., & Ryan, A. (2004). Simple, effective
countermeasurestoP300-basedtestsofdetectionofconcealedinformation.
Psychophysiology,41,205–219.
Rugg,M.D.,&Curran,T.(2007).Event-relatedpotentialsandrecognitionmemory.
TrendsinCognitiveScience,11,251–257.
Sip,K.E.,Roepstorff,A.,McGregor,W.,&Frith,C.(2008).Detectingdeception:The
scopeandlimits.TrendsinCognitiveSciences,12,48–53.
Soskins,M.,Rosenfeld,J.P.,&Niendam,T.(2001).Thecaseforpeak-to-peak
mea-surementofP300recordedat0.3Hzhighpassfiltersettingsindetectionof
deception.InternationalJournalofPsychophysiology,40,173–180.
Smith, M. E., & Guster, K. (1993). Decomposition of recognition memory
event-relatedpotentialsyieldstarget,repetition,andretrievaleffects.
Electroen-cephalographyandClinicalNeurophysiology,86,335–343.
Sutton,S.,Braren,M.,Zubin,J.,&John,E.R.(1965).Evoked-potentialcorrelatesof
stimulusuncertainty.Science,150,1187–1188.
VanHooff,J.C.,Brunia,C.H.,&Allen,J.J.(1996).Event-relatedpotentialsasindirect
measuresofrecognitionmemory.InternationalJournalofPsychophysiology,21,
15–31.
Vilberg,K.L.,Moosavi,R.F.,&Rugg,M.D.(2006). Therelationshipbetween
elec-trophysiologicalcorrelatesofrecollectionandamountofinformationretrieved.
BrainResearch,1122,161–170.
Wolpe,P.R.,Foster,K.R.,&Langleben,D.D.(2005). Emergingneurotechnologies