published:27November2012doi:10.3389/fpsyg.2012.00532
Thecurrentandfuturestatusoftheconcealedinformationtestforfielduse
IzumiMatsuda1*,HiroshiNittono2andJohnJ.B.Allen3
123
NationalResearchInstituteofPoliceScience,Chiba,Japan
GraduateSchoolofIntegratedArtsandSciences,HiroshimaUniversity,Higashi-Hiroshima,JapanDepartmentofPsychology,UniversityofArizona,Tucson,AZ,USA
Editedby:
WolfgangAmbach,InstituteforFrontierAreasofPsychologyandMentalHealth,Germany
Reviewedby:
FrankM.Marchak,VeridicalResearchandDesignCorporation,USA
GershonBen-Shakhar,TheHebrewUniversityofJerusalem,Israel
DonaldKrapohl,NationalCenterforCredibilityAssessment,USA*Correspondence:
IzumiMatsuda,NationalResearchInstituteofPoliceScience,6-3-1Kashiwanoha,Kashiwa,Chiba227-0882,Japan.
e-mail:izumi@nrips.go.jp
TheConcealedInformationTest(CIT)isapsychophysiologicaltechniqueforexaminingwhetherapersonhasknowledgeofcrime-relevantinformation.Manylaboratorystud-ieshaveshownthattheCIThasgoodscientificvalidity.However,theCIThasseldombeenusedforactualcriminalinvestigations.OnesuccessfulexceptionisitsusebytheJapan-esepolice.InJapan,theCIThasbeenwidelyusedforcriminalinvestigations,althoughitsprobativeforceincourtisnotstrong.Inthispaper,wefirstreviewthecurrentuseofthefieldCITinJapan.Then,wediscusstwopossibleapproachestoincreaseitsprobativeforce:sophisticatedstatisticaljudgmentmethodsandcombiningnewpsychophysiologicalmeasureswithclassicautonomicmeasures.Onthebasisoftheseconsiderations,weproposeseveralsuggestionsforfuturepracticeandresearchinvolvingthefieldCIT.
Keywords:concealedinformationtest,fieldapplication,probativeforce,statisticaljudgment,combinationofmeasures
OVERVIEW
TheConcealedInformationTest(CIT)assessesanexaminee’scrime-relevantmemoryonthebasisofdifferencesinphysiologi-calresponsesbetweencrime-relevantandcrime-irrelevantitems(Lykken,1959).Althoughmanystudieshavesupportedthevalid-ityoftheCIT,ithasnotbeenwidelyusedinfieldsituations.Thereappeartworeasonsforitsunpopularity.First,someexam-inersappeartopreferanalternativemethodtermedtheControlQuestionTest(CQT),eventhoughthevalidityoftheCQThasbeenseriouslyquestioned(Ben-Shakhar,2002).Second,theCITisbelievedtobedifficulttoapplyinnon-laboratoryfieldsettings.InJapan,however,theautonomic-basedCITisroutinelyappliedsuccessfullyincriminalinvestigations.Evenso,CITresultshavenotbeenwidelyinfluentialincourtsettings.
Inthispaper,wereviewthecurrentstatusoftheCITinthefieldandlaboratorystudies,withthegoalofoutliningstepsthatcancontributetoanincreasedprobativevalueoftheCITincourt.First,wereviewhowJapaneseexaminershavetriedtoovercomethedifficultiesoftheCITforfieldapplication.Second,wereviewstatisticalmethodsthatcanbeusedtosupportjudgmentsinfieldCITapplications,andinvestigatenewmeasuresthatcanbeaddedtothecurrentCITimplementations.
Throughoutthispaper,wewillemphasizeviewpointsrelevanttofieldapplications.Inthefield,anexamineeisoftennotwillingtotakethetestanddoesnotcomplywithinstructions.Therefore,inJapan,aclassicautonomic-basedCIThasbeenused,whichsimplyconsistsofonecrime-relevantitemandseveralcrime-irrelevantitemsanddoesnotrequireanovertbehavioralresponse.ThispaperwillfocusonhowthisexistingfieldCITcanbeexpanded,butitwillnotreviewotheralternativeapproaches.Forexample,othermemorydetectionorliedetectionteststhatarestillinthe
laboratorystage,suchastheautobiographicimplicitassociationtest(Sartorietal.,2008),showpromisebutareoutsideofthescopeofthispaper.
CURRENTSTATUSOFFIELDCIT
WHATISTHECIT?
TheCIT,alsoknownastheguiltyknowledgetest(GKT;Lykken,1959),isusedincriminalinvestigationstoexaminewhetherapersonrecognizescrime-relevantinformationthatinnocentpeo-plewouldnotknow.IntheCIT,anexaminerpresentsseveralitemstoanexaminee,oneofwhichisacrime-relevantitem.Theitemsareselectedsuchthatinnocentexamineeswouldnotbeabletodistinguishthecrime-relevant(critical)itemfromthecrime-irrelevant(non-critical)items.Eachitemispresentedonceinablockandthisblockisrepeatedseveraltimesindifferentpresen-tationorders.DuringtheCIT,theexaminerrecordsphysiologicalresponsestotheitems.Inthecasethattheresponsesdonotdifferbetweenthecriticalandnon-criticalitems,theexaminerwouldinferthattheexamineedoesnotrecognizethecriticalitem.Ontheotherhand,inthecasethattheresponsesdifferbetweenthecriticalandnon-criticalitems,theexaminerwouldinferthattheexamineerecognizesthecriticalitem.Thus,theCITcanprovideimportantforensicinformationforthepoliceandthejusticesys-tem,identifyingindividualswithkeyinformationaboutthecrime.Suchindividualsmaybeguiltyofcommittingthecrime,orhaveotherusefulinformationaboutthecrimeiftheywerenottheperpetrator.
TheCITisconsideredtohaveasolidscientificfoundation,asmanylaboratorystudieshavedemonstrateditseffectiveness(forareview,seeBen-ShakharandElaad,2003).Althoughpublishedfielddataarerelativelyscarce(Elaad,1990;Elaadetal.,1992;Hira
www.frontiersin.orgNovember2012|Volume3|Article532|1
Matsudaetal.CurrentandfutureoffieldCIT
andFurumitsu,2002;Osugi,2010),theresponsepatternofthevariousphysiologicalmeasuresinfieldCITsaresimilartothoseobservedinlaboratoryCITs(i.e.,skinconductanceincrease,heartratedecrease,respirationsuppression,andfingerpulsevolumedecreaseforcriticalitemsascomparedtonon-criticalitems;Elaad,1990;Elaadetal.,1992;Osugi,2010;Verschuereetal.,2011).
POTENTIALPROBLEMSINTHEFIELDAPPLICATIONOFTHECIT
Todate,theCIThasnotbeenwidelyusedinfieldsettings.Thismayreflect,inpart,thebeliefthattheCITisdifficulttoapplyinfieldsettingsforavarietyofreasons(Krapohl,2011).First,theCITcanproducefalsepositivecases.Criticalitemsthatonlyaguiltypersonknowsaresometimesdifficulttofind.Someinnocentexamineesmayknowthedetailsofthecrimethroughanynum-berofmeans,includingmediareportsandrumors(i.e.,informedinnocentexaminees;forareview,seeBradleyetal.,2011).Otherinnocentexamineesmay,viarepeatedinterrogationsorrepetitionsofcrimedetails,cometohavefalserecollectionsforcrime-relevantitems(AllenandMertens,2008).IftheseinnocentexamineestaketheCIT,theywouldshowdifferentresponsesforcriticalandnon-criticalitems,resultinginfalsepositiveoutcomes.Second,theCITisvulnerabletofalsenegativeoutcomes.Ifcriticalitemsareselectedthatarenotmemorabletotheperpetratorofthecrime,itisunlikelytoberecognized,thusproducingafalsenegativeoutcome.Evenifexamineesdohavecrime-relevantmemoriesandrecognizethecrime-relevantitem,physiologicaldifferencessometimesmightnotbeobserved.Forexample,althoughskinconductanceistypicallymeasuredintheCIT,onestudyreportedthatapproximatelyoneoutoffourpeoplewereelectrodermalnon-responderstoorientingstimuli(VenablesandMitchell,1996).Third,somestudieshaveshownthattheCITisvulnerabletophysicalcountermeasures(e.g.,pressingthetoesagainstthefloorwhennon-criticalitemsarepresented)aswellasmentalcounter-measures(e.g.,countingnumberseachtimeanon-criticalitemappears;forareview,seeBen-Shakhar,2011).Inthenextsection,wewillintroducehowJapaneseCITexaminershaveattemptedtoovercomethesethreeproblems.
CURRENTFIELDUSEOFTHECITINJAPAN
anexamineradvisescriminalinvestigatorstoconducttheCITatanearlystageoftheinvestigationinordertomakeitlesslikelythatcrime-relevantitemsbecomeknowntoawideraudienceovertime.WhenanexaminerisrequestedtoconducttheCIT,he/shefirstconsultswithinvestigators.Anexamineralsochecksmediareportsrelatedtothecrimeandtotherecordofinvestigation.Furthermore,beforeconductingeachCIT,anexaminerpresentsalltheitemsintheCITtoanexaminee,andaskstheexamineeifthereareitemsthathe/sherecognizesorfeelsdifferentfromtheothers.Iftheexamineepointsoutthecrime-relevantitem,theexaminerwouldnotadministertheCITquestionaboutthatitem.
Preventionoffalsenegativecases
JapaneseCITexaminersstrivetoselectcriticalitemsthataguiltypersonshouldremember.Theytrytoavoidusingperipheralfea-turesofthecrime,andinsteadusecentralfeaturesascriticalitems(Carmeletal.,2003;Gameretal.,2010;NahariandBen-Shakhar,2011).Inaddition,beforeeachCIT,anexaminerexplainsthemeaningofeachquestionitemtoanexaminee,inorderthattheexamineewillunderstandwhattheexaminersareasking.
However,evenwhenanexamineemightrecognizeacriticalitem,he/shesometimesmaynotshowadifferentphysiologicalresponsebetweenthecriticalandnon-criticalitems.Oneofthestrategiestoavoidthistypeoffalsenegativecaseisthesimultane-ousmeasurementofmultiplevalidatedresponses.InJapan,anewpolygraphsystemhasbeenusedsince2003,whichsimultaneouslyrecordsskinconductance,heartrate,pulsevolume,andrespira-tion.Thesemeasuresarethoughttoreflectthedifferentaspectsofaphysiologicalresponse.Laboratorystudiesshowthatcombiningthesemultiplemeasurescouldreducefalsenegativerateswhilemaintaininglowfalsepositiverates(e.g.,Gameretal.,2008a).
Counter-countermeasures
Inspiteofthethreeproblemsoutlinedabove,theCIThasbeenofficiallyandsystematicallyusedinJapanforthelast50years.About100trainedexaminersperformabout5,000CITsperyear(Osugi,2011).Allexaminers(whoarenotinvestigators)belongtoaforensicsciencelaboratoryofaprefecturalpoliceheadquarter.TheCQT(Reid,1947)isnolongerused.TheresultsoftheCIThavebeenacceptedasevidenceincourtsincethe1960s.AlthoughJapan’ssuccessfulapplicationoftheCITinthefieldhasattractedattentionfromforeignresearchersandexaminers,notmuchhasbeenwrittenabouthowthepotentialproblemsforfielduseoftheCIThavebeenaddressedinJapan.Therefore,potentialsolutionsarereviewedbrieflybelow,andmoredetailsareavailablefromOsugi(2011).
Preventionoffalsepositivecases
Toguardagainstphysicalcountermeasures,anexaminermon-itorsanexaminee’sbehaviorandhis/herphysiologicalresponsescarefullyduringtheCIT.Whentheexaminerthinksthattheexam-ineeisintentionallyapplyingcountermeasures(e.g.,frequentbodymovements,sighs,orsniffing),heorshewouldinstructtheexami-neetorefrainfromsuchactivities(Osugi,2011).AlthoughspecificsensorstodetectphysicalcountermeasureshavenotbeenappliedinJapanyet,itmaybeusefultointroduce,forexample,pressure-basedsensorsincorporatedinthetestchairandfloorpads,whichhavebeenusedinsomeothercountries.
Previousstudieshavesuggestedthatmentalcountermeasuresaffectskinconductance,butdonotaffectrespiration(Ben-ShakharandDolev,1996;Hontsetal.,1996).InJapan,anexaminermeasuresmultipleautonomicindicesincludingrespiration,whichcanservetolessenthechancethatcountermeasureswillchangetheoutcomeoftheCIT.Tomeasureanexaminee’sphysiologicalresponsefromvariousresponsechannelscanthuscontributetoreducingtheeffectofunobservablementalcountermeasures.
Otherattempts
JapaneseCITexaminersmakeeveryefforttopreventfalsepositivecasesthrougheverystepintheprocess,frompre-examprepara-tiontotheactualadministrationoftheCIT.Onaroutinebasis,
ExaminersinJapanalsouseotherprocedurestogetmoreaccu-rateand/orinformativeresults.First,examinersalwaysconductapretestbeforeaskingaboutcrime-relevantinformation.Inthepretest,anexaminerasksanexamineetomemorizeanumber
FrontiersinPsychology|CognitiveScienceNovember2012|Volume3|Article532|2
Matsudaetal.CurrentandfutureoffieldCIT
onacardinprivateandthenpresentsseveralnumbersincludingthememorizednumber.Thepretestnotonlyhelpstheexami-neetounderstandtheCITparadigm,butalsohelpstheexam-inertoknowthephysiologicalresponsepatternoftheexamineewhenheorsherecognizesanitem.Consideringtheresponsepattern,theexaminerconductsthesubsequentCITs.Forexam-ple,iftheexamineeshowedhighreactivityinskinconductanceresponseinthepretest,theexaminerjudgestheresponsesofsubsequentCITspayingmoreattentiontotheskinconductanceresponse.
Second,anexaminersometimesusesasearchingCIT.ThesearchingCITisdifferentfromthetypicalCITinthatanexaminerdoesnotknowwhichitemiscrime-relevantinadvance.Forexam-ple,ifaweaponhasbeenmissing,anexaminercanaskanexamineeabouttheplacewherehe/sheabandonedaweapon,suchas“WasaweaponabandonedinareaA,areaB,...,orareaE?”Indeed,thejudgmentismoredifficultforasearchingCITthanforausualCITwithknownsolutions,becausetheexaminerhastojudgenotonlywhethertheexamineehasrecognitionbutalsowhichitemtheexamineerecognizes.Additionally,inthecasethatthequestionitemsdonotcoverallpossibilities,thefindingofnophysiologicaldifferencesbetweenitemscannotsupportanexaminer’sconclu-sion“theexamineedoesnotrecognizethecrime-relevantitem;”instead,thisfindingcanonlysupporttheconclusionthat“theexamineedoesnotrecognizeanyitemsinthisquestionset.”Butifanexaminerdevelopsanappropriatequestionset,thesearch-ingCITcansuggestpotentialnewcrime-relevantinformationofwhicheveninvestigatorshavenoknowledge.Intheaboveexam-ple,iftheresponsesdifferbetweenareaAandotherareas,theinvestigatorswillfocusinvestigationonareaAandconsequentlymayfindthemissingweapon.
Third,inJapan,anexamineronlydecidesonwhetheranexam-ineerecognizeseachcrime-relevantitemandneverintegratestheresultsofmultipleCITquestionstojudgewhethertheexamineeisguiltyorinnocent.Itistheinvestigators’task,ratherthantheexam-iner’stask,tointegratetheresultsacrosstheCITquestionsandevaluatetheexaminee’slikelihoodofguilt.Someauthors,how-ever,havearguedthatexaminersshouldintegrateresultsacrossmultipleCITquestionsinordertoobtainmorestatisticallyreli-ableandrobustresults(Ben-ShakharandElaad,2002).However,Japaneseexaminershavemaintainedtheapproachofonlyadopt-ingajudgmentforeachCITquestion.Oneofthejustificationsforconductingthetestinthismanneristhatitallowstheexaminertoclarifywhichitemstheexamineerecognizesandwhichitemstheexamineedoesnot.Forexample,inthecaseofatheftthatwasconductedbyagroupofperpetrators,informationindicat-ingwhethertheexamineeknowseachcrime-relevantitemmaybecomeacluetorevealwhatrolehe/sheplayedinthecrime(e.g.,amajorculpritorjustalookout).Thus,treatingresultsfromeachCITquestionseparatelycanfacilitateinvestigationsofcasesinvolvingmultiplesuspects,andprovidedetailstoguideandfacilitatetheinvestigators’continuinginquiriesforanytypeofcase.Additionally,asdescribedabove,Japaneseexaminerssome-timesusesearchingCITs;insuchcaseswhereanexaminerdoesnotknowwithcertaintywhichalterativeisthecriticalitemforagivenCITquestion,itisdifficulttointegrateCITresultsacrossquestions.
ValidityofthefieldCITinJapan
OnearticlehasreportedonfieldCITdatasetsusingthecurrentpolygraphsysteminJapan.Kobayashietal.(2009)analyzedthedataof113CITquestionsobtainedfrom38examinees(33menand5women,meanage=36.4,SD=12.5).Subsequentinves-tigationsconfirmedthatalloftheseexamineesrecognizedthecriticalitemsoftheseCITquestions.ForeachCITquestion,theresponseswerecomparedbetweencriticalandnon-criticalitemswithattest.Ifthepvaluedidnotexceed0.10,theexamineewasjudgedasrecognizingthecriticalitem.Thecorrectdetectionrateswere52.5%fortheskinconductanceresponse,49.5%forheartrate(averagein16–20saftertheitemonset),38.1%forrespi-rationlinelength(averagein0–15s),and26.2%fornormalizedpulsevolume(averagein6–10s).Itshouldbenotedthatthesevaluesarecorrectdetectionrates(i.e.,sensitivities)forindivid-ualCITquestionsusingasinglemeasure.AlthoughKobayashietal.didnotreportthedata,combiningthevariousphysiolog-icalmeasuresshouldincreasetheoveralldetectionrate.IntheactualfieldCIT,examinersarriveataconclusionbycombiningalloftheavailablemeasures.Inaddition,toaddressthespeci-ficityoftheCIT(i.e.,howwelleachmeasurecorrectlyindicatesnon-recognitionofcriticalitemswhenexamineesdonothaverecognition),alargerdatasetincludingbothguiltyandinnocentsubjectswouldberequired.
IMPROVINGTHEPROBATIVEFORCEOFTHECITINCOURTAlthoughtheCIThasbeenwidelyusedforcriminalinvestigationsanditsresultshavebeensometimesacceptedasevidenceincourtinJapan,theCITresultsarenotconsideredsufficientlystrongthattheytypicallydirectlyaffecttheoutcomesincourt.ToimprovetheprobativeforceoftheCIT,webelievethefollowingtwoapproachesaremostpromising.
Thefirstapproachistousestatisticalmethodstointerprettheresults.InfielduseoftheCITinJapan,CITresultsaremainlyderivedthroughtheexaminers’visualinspections(Osugi,2011).Ifthejudgmentisunderpinnedbystatisticalmethods,theCITresultswouldbecomemoreconvincingforjudges.Moreover,suchanapproachiswell-justifiedintheliterature:statisticalactuar-ialjudgmenthasgreaterreliabilityandvaliditythanjudgmentsbasedonvisualimpressions(Dawes,1979).Inlaboratorystud-ies,Lykken’sscoringandz-scoreaveraginghavebeencommonlyusedfordecision-making(Meijeretal.,2011).Lykken,1959scor-ingisbasedontherankofthecriticalitemamongallitemsindescendingorderoftheresponsevalues.Z-scoreaveragingusestheaveragestandardizedresponsevalueacrossblocksandmeasures(Ben-Shakhar,1985).Althoughthesetwomethodsaresimpleandclear,theydohavedrawbacks.Wewillreviewthesetwometh-odscriticallyandcomparethemwithotherproposedmethodsbelow.
ThesecondapproachistoaddnewmeasurestocurrentfieldCITtoincreaseitsaccuracy.InthecurrentfieldCIT,heartrate,skinconductance,respiration,andpulsevolumearerecorded.Newmeasurescanbeintroducedeitherbyimprovingquantifi-cationmethodsofcurrentlyrecordedresponsesorbyrecordingnewresponsechannels,suchasreactiontime,facialresponses,activationsusingfunctionalmagneticresonanceimaging(fMRI),andfeaturesoftheelectroencephalogram(EEG)andevent-related
www.frontiersin.orgNovember2012|Volume3|Article532|3
Matsudaetal.CurrentandfutureoffieldCIT
potential(ERP).Wewillreviewthesenewmeasuresandevaluatethesefromtheviewpointoffieldapplication.
STATISTICALEVALUATIONMETHODS
2011).Forexample,evenwhentheresponsetothecriticalitemmightbethreetimesaslargeasthenextlargestresponse,thescorewouldbethesameaswhenitisonlyslightlylarger.
Z-scoreaveraging.Z-scoreaveragingiswidelyusedinlaboratorystudiestocapturequantitativedifferencesbetweenitems(Ben-Shakhar,1985;Figure1).Inthismethod,aresponsetoeachitemisfirststandardizedusingthemeanandSDofeachmeasurewithinablock.Theaimofthestandardizationis(1)tocanceloutthedifferencesinphysiologicallevelsamongblocksand(2)totreatmultiplemeasuresthathavedifferentunitsinthesamedimen-sion.Ifameasuretypicallydecreasestoacriticalitem(e.g.,heartrate,respiration,orpulsevolume),itsz-scoreismultipliedby−1.Thescoresforthecriticalitemarethenaveragedacrossallblocksandallmeasures.Wethenjudgewhethertheaveragedz-scoreissignificantlyhighenoughtoexceedtypicalcutpointsusingthestandardnormaldistribution.ThismethodneedsnoparameterestimationaprioriandthusiseasytoapplytofieldCIT.
However,thismethodhastwodisadvantages.First,thismethodassumesthatforeverysubject,allmeasuresrespondinthenor-mativeexpecteddirection.Itthusdoesnotconsiderindividualdifferencesinresponsepatterns.Thephysiologicalmeasuresthatresponddistinctivelybetweencriticalandnon-criticalitemsaresometimesdifferentbetweenexaminees(Matsudaetal.,2006).Forexample,Osugi(2011)reportedresultsfromfielddatainwhichaguiltyexamineeshowedconstantdistinctiveresponsesonlyinrespiration.Insuchacase,withanincreasingnumberofmeasures,theaveragez-scorewillbecomesmallerandthusmightleadtoafalsenegative.Second,thismethoddoesnotconsiderthedifferencesingeneralaccuraciesamongmeasures.Forexample,inlaboratorystudies,accuracyisusuallyhigherforskinconductancethanforothermeasures(i.e.,heartrate,res-piration,andpulsevolume;e.g.,Ben-ShakharandElaad,2003;Gameretal.,2008b).However,withz-scoreaveraging,allmea-suresareweightedequally.Itmightbepreferableifeachmeasurewereweightedaccordingtoitsaccuracy.
Proposedstatisticalmethods
Here,wereviewstatisticalmethodsthathavebeenusedinpre-viousstudies.First,wereviewstandardstatisticalmethodssuchasLykken’sscoringandz-scoreaveraging.Wethenreviewfiveotherproposedmethods:logisticregressiondiscrimination,latentclassdiscrimination,Bayesianclassification,multivariatenormaldistributiondiscrimination,anddynamicmixturedistributiondiscrimination.Finally,weoutlinerecommendationsfortheiruse.
Standardstatisticalmethods
Lykken’sscoringmethod.ThisisatraditionaldiscriminationmethodproposedbyLykken(1959;Figure1).Thismethodassignsascoreof2ifthecriticalitemelicitedthelargestresponse,ascoreof1ifthecriticalitemelicitedthesecondlargestresponse,andascoreof0otherwiseineachblock.Iftheaverageofthescoresacrossblocksexceedsathreshold,itisjudgedthattheexamineerecognizesthecriticalitem.
Lykken’sscoringmethodhasseveraladvantages.First,thismethodisverypractical.Itcanbeusedwithoutquantificationandparameterestimations.Second,becauseresponsesarerankedwithineachblock,correctionisnotrequiredevenifphysiologicallevelschangebetweenblocksasaresultofhabituation.
However,Lykken’sscoringmethodhasitsdrawback:thismethoddoesnottakeintoaccountquantitativedifferencesbetweenresponsestocriticalandnon-criticalitems(Meijeretal.,
Toovercomethedisadvantagesofz-scoreaveraging,otherstatis-ticalmethodshavebeenproposed:logisticregressiondiscrimi-nation,latentclassdiscrimination,Bayesianclassification,multi-variatenormaldistributiondiscrimination,anddynamicmixturedistributiondiscrimination.WewillexplainthesemethodsbelowandinFigure2,andevaluatethesemethodsfromtheviewpointoffieldapplication.Inparticular,wewillfocusonwhetheranewmethodovercomesthelimitationsofz-scoreaveraging.Logisticregressiondiscrimination.Thismethodconsidersthedifferencesinaccuracyamongmeasuresbyallocatingaweighttothez-scoreofeachmeasure(Gameretal.,2006,2008b;Figure2A).TheweightsareacquiredfromtheCITdatasetsofpreviousexami-nees,wheregroundtruthhasalreadybeenestablished.Eachweightreflectstheeffectivenessofthemeasureforestimatingrecognition.Iftheseweightsareall1,theresultwillbethesameastheoneofz-scoreaveraging.
Thismethodispracticalandwidelyusedinvariousresearchdomains.Ifthesamplesizeislarge,theweightparameterswillbeestimatedquitestably.
FIGURE1|Illustrationsofthestandardstatisticalmethods:Lykken’sscoringandz-scoreaveraging.Z_HR,az-scoreforheartrate;Z_SCR,az-scoreforskinconductanceresponse;Z_PV,az-scoreforpulsevolume;p,probability.Lykken’sscoringassignsascoreof2ifthecriticalitemelicitedthelargestresponse,ascoreof1ifthecriticalitemelicitedthesecondlargestresponse,andascoreof0otherwiseineachblock.Inz-scoreaveraging,z-scoresaresimplyaveragedacrossblocksandmeasures.Z-scoresmaybemultipliedby−1ifasmallerresponseischaracteristicofrecognition.
FrontiersinPsychology|CognitiveScienceNovember2012|Volume3|Article532|4
Matsudaetal.CurrentandfutureoffieldCIT
FIGURE2|Illustrationsofthefiveproposedstatisticaldiscriminationmethods.Z_HR,az-scoreofheartrate;Z_SCR,az-scoreforskin
conductanceresponse;Z_PV,az-scoreforpulsevolume;p,probability.(A)Thelogisticregressionmethodissimilartoz-scoreaveraging,buteachz-scoreisweightedaccordingtotheaccuracyofthemeasureestimatedfrompreviousdatasets.(B)Thelatentclassdiscriminationmethodisatwo-layermodelofthelogisticregressionmethod.Thereisanappropriateregressionformulaforeachclass,andtheresultoftheregressionformulaissummedacrossclasseswithaweightofthelikelihoodofanexamineebelongingtoaclassaccordingtohis/herpretestresult.(C)TheBayesianclassificationmethodcalculatestheprobabilityofrecognitionby
multiplyingpriorprobabilitiesandtheprobabilitiesthatastandardizedresponsevalueofeachmeasureexceeds/doesnotexceedathresholdintherecognitioncondition.Hereisthecasethataparticipant’sheartratechangeandskinconductanceresponseexceededthethreshold,whilehis/herpulsevolumedidnotexceedthethreshold.(D)Inthemultivariatenormaldistributionmethod,aguiltymodel(two-distributionmodel)andaninnocentmodel(one-distributionmodel)areappliedtotheobtainedresponsesinaCIT(eachsmallcirclerepresentsaresponsetoacritical(yellow)oranon-critical(white)item).Thebetterfittedmodelwillbeselected.(E)Thedynamicmixturedistributionmethodusestimeseriesandisanextendedversionofthemultivariatenormaldistributionmethod.Inthismethod,aguiltymodel(representingtimeserieswithamixtureofthreedistributions)andaninnocentmodel(representingtimeserieswithamixtureoftwodistributions)areappliedtotheobtainedtimeseriesinaCIT.Themodelthatfitsthetimeseriesbestisselected.
www.frontiersin.orgNovember2012|Volume3|Article532|5
Matsudaetal.CurrentandfutureoffieldCIT
Ontheotherhand,thismethoddoesnotsufficientlyconsiderindividualdifferencesinresponsepatterns.Thisisbecausetheparametersarecalculatedtobefittedtothenormativeresponsepattern.Similartoz-scoreaveraging,ifaguiltyexamineeshowsdistinctiveresponsesonlyinasmallnumberofmeasures,thismethodmightproduceafalsenegative.Additionally,thelogisticregressionmethodmayunderperformthez-scoreaveragingifthesamplesizeisnotlargeenoughtoreliablyestimatetheparameters(c.f.,Dawes,1979).
Latentclassdiscrimination.Thismethodisanextendedversionofthelogisticregressionmodelthatconsidersindividualdiffer-encesinresponsepatterns.Asmentionedbefore,inthefieldCIT,anexaminerconductsapretestusingcardstocapturetheresponsepatternofanexaminee.However,theresultsofthepretestarenotconsideredinmoststatisticalmethods.Therefore,Matsudaetal.(2006)proposedthelatentclassdiscriminationmethod(Figure2B).Inthismethod,previouslyobtainedexamineesaregroupedintoseveralclasses,foreachofwhichadiscriminantfor-mula(e.g.,logisticregressionformula)iscalculatedandfittotheresponsepatternoftheexamineesbelongingtothatclass.Itisthenestimatedifagivenexamineerecognizesacriticalitemusingthefollowingprocess.First,theprobabilitythattheexamineewouldrecognizethecriticalitemiscomputedbyapplyingthediscrimi-nantformulaofeachclasstohis/herstandardizedresponsevalues.Second,theprobabilitythattheexamineebelongstoaclassiscomputedbyusinghis/herpretestdata.Finally,therecognitionprobabilityiscalculatedbysummarizingeachclass’srecognitionprobabilityacrossallclasseswithaweightoftheprobabilityfortheclassthattheexamineebelongsto.Inthismanner,eachexamineecanbedistinguishedthroughhis/herresponsepattern.
Thismethodconsidersseveralresponsepatternsaslatentclasses.Inaddition,theaccuraciesofthemeasureshavebeenreflectedasparametersofadiscriminantformulaineachclass.Moreover,theseparameterscanbeestimatedstablywithalargedatasetofpreviousexaminees.
However,factoringinthepretestdatacanalsobecomeadraw-backinpracticalapplications.InJapan,about5–6CITsaretypi-callyconductedafterthepretest.Ittakesabout2or3htofinishalltheCITs(Osugi,2011).Therefore,aresponsepatternmaychangefromthepretesttothelastCITforanexaminee.Inaddition,thismethodisbasedonamorecomplex,hierarchicalmodel,andconsequentlyneedstoestimatemoreparametersthanthelogisticregressionmethod.Thisimpliesthatthelatentclassdiscrimina-tionmethodrequiresalargerdatasetthanthelogisticregressionmethodforparameterestimation.
Bayesianclassification.Thismethodcombinesmultiplemea-suresbyusingcomputationsbasedonBayes’theorem(Allenetal.,1992;Figure2C).Thisapproachcalculatestheprobabilitythatanexamineerecognizesanitemusing(1)thesensitivity/specificityofeachmeasure(i.e.,theprobabilitythataresponsevalueexceeds(ordoesnotexceed)athresholdintheconditionthatanexamineerecognizes(ordoesnotrecognize)theitem)and(2)apriorprob-ability(i.e.,theprobabilitythattheexamineeshowsthedistinctiveresponsebychancetoeachitem,whichisdeterminedbythenum-berofitemsinthetest).Thismethodalsousesawithin-subjects
standardization,sothatlargeindividualdifferencesinresponsemagnitudeareeliminated,andthepatternofresponsesacrosscrit-icalandnon-criticalitemsisretained.First,foreachstandardizedmeasure,thesensitivity,specificity,andthresholdarecalculatedfromapreviouslyobtaineddataset.Thestandardizedresponsevalueofagivenexamineeisthencomparedtothethreshold.Iftheresponsevalueexceeds(ordoesnotexceed)thethreshold,thesen-sitivity(or1−sensitivity)isenteredintoBayes’formulatocalculaterecognitionprobability.Similarly,thespecificityor1−specificitycanbeenteredintoBayes’formulatocalculatetheprobabilityofafailuretorecognizecrime-relevantitems.
Asthismethodtreatsresponsesasbinarydata–thatis,whetheraresponseexceedsthethresholdornot–quantitativedifferencesbetweenitemsarenotfullycapturedwiththismethod.Ontheotherhand,thankstodealingwithbinaryvalues,thismethodisnotexcessivelyaffectedbyoutliers.Controllingtheinfluenceoffactorsthatwillproduceoutliersisdifficultinthefieldsituationascomparedwiththelaboratorysituation.Forthisreason,forfieldCITapplications,theBayesianclassificationmaybepreferredtotheotherstatisticalmethods.
Multivariatenormaldistributiondiscrimination.Incontrasttologisticregression,latentclassdiscrimination,andBayesianclassification,whichrequirepreviouslyobtaineddatatoestimatetheirparameters,themultivariatenormaldistributionmethodrequiresonlytheCITresultsofthecurrentexaminee(Adachi,1995;Figure2D).Iftheexamineerecognizesacriticalitem,thedistributionoftheresponsesshoulddifferbetweencriticalandnon-criticalitems(i.e.,guiltymodel).Incontrast,iftheexami-needoesnotrecognizethecriticalitem,thedistributionshouldnotdifferbetweencriticalandnon-criticalitems(i.e.,innocentmodel).BoththeguiltymodelandtheinnocentmodelareappliedtothegivenresponsesintheCIT.Iftheguiltymodelbetterfitstheresponsesthantheinnocentmodel,theexamineeisjudgedasrecognizingthecriticalitem.
Thismethodonlyrequiresthatresponsestocriticalandnon-criticalitemsdiffer,anddoesnotrequireapreviousdataset.Inaddition,thismethodhasnoassumptionsoftypicalresponsepat-terns.Therefore,itcandealwithvariousresponsepatterns,eveniftheresponsepatternisverydifferentfromthetypicalnormativepattern.
However,withthismethod,wecanestimatemodelparame-ters(i.e.,meanandSDofdistributions)onlyfromthegivendata.Thesamplesizeisthusthenumberofrepetitions;forexample,ifeachitemisrepeatedfivetimes,thesamplesizeisfive,whichistoosmalltobeusedtoestimatestableparameters.Inaddition,althoughtheaccuracyofeachmeasurecanbecalculatedbasedonpreviousdatasets,thismethoddoesnotusepreviousdatasets.Therefore,thedifferencesinaccuracybetweenmeasurescannotbetakenintoaccount.
Dynamicmixturedistributiondiscrimination.Inordertoesti-matestablemodelparametersbyusingonlythegivendata,theextendedversionofthemultivariatenormaldistributionmethod–thedynamicmixturedistributionmethod–wasdeveloped(Mat-sudaetal.,2009a;Figure2E).Similartothemultivariatenormaldistributionmethod,thismethodpreparesaguiltymodelandan
FrontiersinPsychology|CognitiveScienceNovember2012|Volume3|Article532|6
Matsudaetal.CurrentandfutureoffieldCIT
innocentmodel,butappliesthesemodelstotimeseriesdata.Theguiltymodelrepresentstheresponsetimeseriesusingthreedis-tributions:anon-responsedistributioncorrespondingtothebaselevel,acriticalresponsedistributioncorrespondingtoresponsestothecriticalitem,andanon-criticalresponsedistributioncor-respondingtoresponsestothenon-criticalitems.Incontrast,theinnocentmodelrepresentstheresponsetimeseriesusingtwodis-tributions:anon-responsedistributionandapooledcritical/non-criticalresponsedistributioncorrespondingtoresponsestobothcriticalandnon-criticalitems.TheguiltyandinnocentmodelsareappliedtothetimeseriesoftheCITdata.Ifthetimeseriesismorecompatiblewiththeguiltymodelthanwiththeinnocentmodel,theexamineeisjudgedasrecognizingthecriticalitem.
Similartothemultivariatenormaldistribution,thismethodrequiresnopreviousdatasetandnoassumptionoftypicalresponsepatterns.Therefore,thismethodisveryflexibleandcaneasilyaccommodateindividualdifferencesinresponsepatterns,evenifanindividual’sresponsepatternisverydifferentfromthetypicalnormativeresponsepattern.Additionally,becausetimeseriesdataareused,stablemodelparametersmaybeestimatedwiththetypicalnumberorrepetitionsintheCIT.
However,sincethismethoddoesnotdependonpreviousdatasets,theaccuracyofeachmeasurecannotbetakenintoaccount.Furthermore,thismethodrequirescomplexcalculationsforparameterestimations(i.e.,Gibbssampler).Givencurrenttechnology,ittakesatleastabout10mintofinishthecalcula-tionoftheparameters.Ifthecalculationalgorithmisimproved,thismethodmightbeideallysuitedtofieldCITuse.
Summaryofstatisticalmethods
Table1summarizestheadvantagesanddisadvantagesofthevar-iousstatisticalmethods.Asthetableshows,aperfectstatisticalmethoddoesnotexist.Morestudiesarerequiredtocontinuetoimproveexistingmethods.
However,themostpromisingmethodatpresentwouldappeartobethelatentclassdiscriminationmethodorthedynamicmix-turediscriminationmethod.Table1showsthemethodologicaladvantagesofthelatentclassanddynamicmixturedistributionmethodsascomparedtotheothermethods,recognizingthattheirparametercalculationsarecomplex.Furthermore,superiorityofthesetwomethodsintermsofdiscriminationperformancewas
demonstratedempirically(Matsudaetal.,2009a).Inthisstudy,19guiltyparticipantswerediscriminatedfrom15innocentpartic-ipantsbyusingthelogisticregression,latentclass,multivariatenormaldistribution,anddynamicmixturedistributionmeth-ods.Thediscriminationperformancewashigherforthelatentclassandforthedynamicmixturedistributionmethodsthanforthelogisticregressionandthemultivariatenormaldistributionmethods.Ofcourse,thisresultshouldbeverifiedbyusinglargernumberoffieldCITdatasets.Inaddition,theirdiscriminationperformanceshouldbealsocomparedwiththatoftheBayesianclassificationmethod,whichisexpectedtoberobustinthefaceofoutliers.
Methodsrequiringpreviouslyobtaineddatasetsmayhavelim-itedutilityforfiledCITapplications.Suchmethods(i.e.,thelogisticregression,latentclass,andBayesiandiscriminationmeth-ods)requiretheparameterstobeestimatedfromthefieldCITdataforwhichvalidgroundtruthdataareavailableforeachexaminee.However,theexactconfirmationofthisknowledgeisverydiffi-culttoobtaininthefieldsituation,sinceitisdifficulttoknowwithabsolutecertaintywhoisguiltyandwhoisinnocentinafieldcase.Itmaytakearatherlongtimetocollectasufficientnumberofappropriatefielddatasetsforparameterestimation.Iftheparametersareestimatedfromaninsufficientnumberoffieldsamples,thesemethodsmayunderperformthesimplez-scoreaveraging(Dawes,1979).Incontrast,methodsthatrequireonlythecurrentdataset(i.e.,themultivariatenormaldistribu-tionanddynamicmixturedistributionmethod)haveastrongadvantageforfieldusesincetheydonotrequireapreviouslyobtaineddataset.Butthisalsoindicatesthatthelattermethodsmaybemoreinfluencedbymissingvaluesandmeasurementarti-factsthantheformermethods.Evenwhenadoptingthelattermethods,evaluatingtheirgeneralizabilitywillrequireusingafielddataset.
ADDITIONALMEASURES
InordertoimprovetheprobativeforceoftheCITincourt,itwouldbealsopromisingtouseadditionalmeasuresthatcanpotentiallyincreasetheaccuracyoftheCIT.ThecurrentfieldCIT,thatisbasedonmeasuresofautonomicresponses(i.e.,skinconductance,heartrate,respiration,pulsevolume),hasbeenworkingwellsofarinJapan.Therefore,itwouldbemorepromisingtoaddnewmeasures
Table1|Comparisonofstatisticalmethodsintermsoffeaturesthatareimportantforfieldapplication.Statisticalmethod
Flexibilityforindividualdifferences
Considerationofaccuracydifferencesamongmeasures
Standard:Z-scoreaveraging(A)Logisticregression(B)Latentclassdiscrimination(C)Bayesianclassification(D)Multivariatenormal
distribution(E)Dynamicmixture
distribution
LowLow
High(assumesubgroupshavingdifferentresponsepatterns)Medium
High(noassumptionofatypicalresponsepattern)
High(noassumptionofatypicalresponsepattern)
No
No
Stable
High
YesNo
YesNo
StableUnstable
MediumMedium
NoYesYes
NeedofpreviousdatasetforparameterestimationNoYesYes
StabilityofparameterestimationNoparametersStableStable
LowMediumHighComplexityofmodel
www.frontiersin.orgNovember2012|Volume3|Article532|7
Matsudaetal.CurrentandfutureoffieldCIT
totheautonomic-basedCITinsteadofalteringthecurrentfieldCITcompletelytousealternativemeasures.Inthissection,wewillreviewadditionalCITmeasuresthatcanbeobtainedbyusingtwoapproaches.Thefirstapproachistorefinethequantificationoftheclassicautonomicresponses.Thesecondapproachistoimplementnewphysiologicalmeasurestoaugmenttheautonomicresponsesusedcurrently.
Quantificationofnew/refiningaspectsofautonomicresponses
amplitudeofthecyclebytheaveragevoltageduringthecycle.Thenormalizedpulsevolumeisadvocatedasamorevalidmeasurefortheassessmentofvasculartonethantheusualpulsevolume(Sawadaetal.,2001).ThevalidityofthenormalizedpulsevolumehasalsobeenconfirmedinaCITstudy(Matsudaetal.,2009a).
Addingnewmeasures
TheImprovementofcurrentquantificationmethodsisasimplewaytoincreaseaccuracyofthecurrenttest.Here,wewillreviewsomeexamplesofhowquantificationmightberefined.
Respiration.RespirationhasbeenoperationalizedasrespirationlinelengthinalmostallCITstudies(forareview,seeGamer,2011a).Therespirationlinelengthisdefinedasthesumofthemovingdistancesoftherespirationcurveinaspecifiedtimeinterval.Therespirationlinelengthdecreaseswhenrespirationissuppressed(i.e.,shorterrespiratorytimeandsmallerampli-tude),andthusisagoodmeasurefortheCIT.However,thelinelengthisbiasedbyhowthepartsoftherespiratorycyclesareincludedinthetimeinterval.Toaccountforthisbias,Elaadetal.(1992)shiftedthestartingpointofthetimeintervalslightly,calculatedthelinelengthforeachshift,andthenaveragedthelinelengthsforallshifts.However,eventhismethodcannotremovethebiascompletely(Figure2inMatsudaandOgawa,2011).
Tofullyresolvethisbiasproblem,anewquantificationmethod–aweightedaveragerespirationlinelength–hasbeenrecentlyproposed(MatsudaandOgawa,2011).Thismethodcal-culatestherespirationlinelengthpercycle,weightsitwiththeproportionthatthecycleoccupiesinthetimeinterval,andthenaveragestheweightedlinelengthsacrossallcyclesinvolvedinthetimeinterval.Thediscriminationperformancewassignificantlybetterfortheweightedaveragerespirationlinelengththanforthetraditionalrespirationlinelength.
Moreover,thereisanundeniablepossibilitythatchangesinrespiratoryrateandamplitudeareelicitedindependentlyintheCIT.Toextractmorepreciseinformationfromrespiration,respi-ratoryrate,andamplitudecouldbemeasuredseparately.Inordertoquantifythese,theuseoftheweightedaveragemethodwouldbepreferable(e.g.,Matsudaetal.,2009a).
Pulsevolume.Recently,pulsevolumehasbeenquantifiedasfin-gerpulsewaveformlengthinawaysimilartothatofrespirationlinelength(ElaadandBen-Shakhar,2006;Vandenboschetal.,2009).Thefingerpulsewaveformlengthcanreflectbothpulserateandamplitudeinformation.Asmentionedabove,thelinelengthisaffectedbywhichproportionofacycleisincluded.However,theeffectofthisbiasismuchsmallerforpulsevolumethanforrespiration,becausethecycletimeofapulseismuchshorter.Ontheotherhand,sinceheartrateiscomputedwithanelectrocar-diograminJapan,themeasurementoffingerpulsevolumelengthisredundant.
InJapan,normalizedpulsevolumehasbeenappliedtothefieldCITtoevaluatevasculartonemoreaccurately.Thenormal-izedpulsevolumeiscomputedperpulsecyclebydividingthe
Newphysiologicalorbehavioralmeasurescanberecordedinadditiontoautonomicresponsesinthefield,particularlyiftherecordingiseasyandstable.Here,wewillreviewreactiontime,facialfeatures,fMRIactivations,andEEG/ERPfeatures.Reactiontime.Onepossiblemeasurethathasbeenconsideredisreactiontimeafteritemonset(forareview,seeVerschuereandDeHouwer,2011).Somestudiesreportedhighaccuracyofindividualclassificationusingreactiontime.Forexample,Allenetal.(1992)reportedasensitivityof0.950andthespecificityof1.000.
However,inthecurrentsituationinthefield,theremaybeproblemswithusingreactiontime.First,reactiontimecanbecontrolledintentionally.Itmightthereforebeeasiertousecoun-termeasuresthataffectreactiontimethanthosethataffectauto-nomicresponses.Infact,somestudiesusetheresponsetimeasameasureofcountermeasures(Rosenfeldetal.,2008;WinogradandRosenfeld,2011).Second,itisuncertainwhetherexamineeswouldfollowtheinstructions,suchas“respondasquicklyandaccuratelyaspossible.”Unliketheautonomic-basedCIT,areaction-timetaskrequiresexamineestorespondactively.Evenwhenexami-neesareinnocent,however,theymaynottakethetestwillinglyandthusmaynotcooperate.Inaddition,attributesoffieldexam-ineesaremorediversethanthoseofparticipantsinlaboratorystudies.Forexample,elderexamineeshaveslowerandmorevari-ablereaction-times,whichmightrenderthismeasurelessusefulinsomepopulations.
Despitetheselimitations,researchmightprofitfromfurtherexaminationofreactiontimeintheCIT.Itisaneasilyobtainedmeasure,andindividualdifferencesinresponsetimesmightnotbeofconcernifquantifiedusingwithin-subjectmetrics(z-scores).Moreover,itmightbepossibletoidentifyreaction-timeresponsepatternsthatwouldsuggestwhenreactiontimecan,andwhenitcannot,provideusefulinformation.
Facialfeatures.FacialexpressionshavepotentialasameasureincurrentfieldCITexaminations.Becauseafaceisusuallynotcov-ered,itiseasytorecordtheinformationwithoutattachingspecialelectrodes(i.e.,witharemote-sensingtechnique).
Itiswell-knownthatliedetectioncanmakeuseoffacialmuscleactivity(Ekman,2001).However,asfarasweknow,nostudyhasreportedtheuseoffacialmusclechangesintheCIT,butautomatedFacialactioncodingsystem(FACS;Little-wortetal.,2011)mightmakethisaneasypossibilitytoexplorefurther.Ontheotherhand,facialskinsurfacetemperaturehasbeenmeasuredintheCIT(Pollinaetal.,2006).Inthisstudy,thetemperatureincreasedforcriticalitemscomparedtonon-criticalitemsinaregionbelowtheeyes.Itsindividualclas-sificationresultwasasensitivityof0.917andaspecificityof0.917.
InformationrelatedtotheeyeshasalsobeenappliedtotheCIT.Startleeyeblinksreducedmoreforcriticalitemsthanfor
FrontiersinPsychology|CognitiveScienceNovember2012|Volume3|Article532|8
Matsudaetal.CurrentandfutureoffieldCIT
non-criticalitems(Verschuereetal.,2007).Temporaldistributionsofblinksdifferedbetweencriticalandnon-criticalitems(Fukuda,2001).Pupilsizesincreasedmoreforcriticalitemsthanfornon-criticalitems(BradleyandJanisse,1981;LubowandFein,1996).LubowandFein(1996)reportedasensitivityof0.50–0.70andaspecificityof1.00usingpupilsizes.
ThusavarietyoffacialmeasuresshowsomepromiseforuseintheCIT,butnonehavebeenextensivelyresearched.Therefore,futureresearchshoulddetermineifuseofthesefacialmeasurescanincreasethevalidityofthecurrentautonomic-basedCIT.fMRI.RecentresearchhasutilizedfMRIinCIT-likeexperiments(forareview,seeGamer,2011b).Noseetal.(2009)reportedtheaccuracyoffMRIintheCIT:thesensitivitywas0.84andthespecificitywas0.84.However,theuseoffMRIinthefieldwouldbedifficultatthepresenttime.First,theequipmentforfMRIisexpensiveandnotportable.Second,examineesmustbeextremelycooperativeastheyarenotabletomoveduringthefMRIscanningandwouldhavetotoleratethenoiseduringthetest.Third,someexamineescouldnotbetestediftheyhavemetalintheirbodiesthatwouldmakefMRIunsafe.Althoughtechnicalimprovementofrecordingsandanalysesareexpectedinfutureresearch,fMRImeasuresmayinherentlycarrynomoreornolessweightthanothermeasuresusedintheCIT.
EEG/ERPs.ManylaboratorystudieshavemeasuredEEGdur-ingtheCITandreportedsignificantdifferencesinERPcom-ponentsbetweencriticalandnon-criticalitems,especiallyP3amplitudes(Rosenfeldetal.,1988;FarwellandDonchin,1991;Allenetal.,1992;Rosenfeld,2011).Arecentmeta-analysisshowedthattheP3measureismoreeffectivethanthetraditionalauto-nomicmeasuresindetectingparticipants’concealedknowledge:Cohen’sdwas2.55fortheP3amplitudeand1.72forskincon-ductanceresponse(Ben-ShakharandMeijer,2012).ThisresultissimilartothatofAllenandIacono(1997),inwhichtheycomparedtheareaunderROCcurvefromtheirERPdatatopublishedskinconductancedata.TheincreaseoftheP3ampli-tudeisthoughttoreflectthesignificanceofthecriticalitemfortheexaminees(Rosenfeld,2011),whichisoftenembeddedwithinanoddballparadigm.Inaddition,recentstudieswithratherlonginter-stimulusintervals(>7s)reportedtheincreaseoftheN2(Matsudaetal.,2009b,2012;GamerandBerti,2010)andthelatepositivepotential(Matsudaetal.,2009b,2012)forthecriticalitem.
DuetotheprogressofrecordingandanalysistechniquesithasbecomeeasiertomeasureEEGinfieldsituations.Infact,anEEGcanberecordedwithapolygraphsystemcur-rentlyusedinfieldCITinJapan,althoughthestimuluspre-sentation/controlsystemforithasnotbeenequippedyet.ArecentstudymeasuredERPsunderthestandardprotocoloftheautonomic-basedfieldCIT(Matsudaetal.,2011).Thisstudyshowedthatlatepositivepotentialsignificantlydifferedbetweencriticalandnon-criticalitems,evenwheneachitemwaspre-sentedonlyfivetimes.Importantly,includingthelatepositivepotentialimprovedthediscriminationperformanceofthestan-dardautonomic-basedCIT.Furthermore,Rosenfeld(2011)have
proposedanewprotocoloftheERP-basedCITinordertomakethetestresistanttocountermeasures(“complextrial-basedCIT”),andhavereportedhighaccuracies.Collectivelythesestudiesindi-catethatfeaturesoftheERPwouldbepromisingadditionstothefieldCIT.
Moreover,althoughmoststudiesquantifiedEEGinthetimedomain,somerecentstudiesfocusedoninformationinthefre-quencydomain(Abootalebietal.,2006,2009;Zhaoetal.,2011).Thesestudiesshowthatdifferencesinwaveletfeaturescanreflectthedifferencesbetweencriticalandnon-criticalitems.Further-more,thefrontalasymmetryofleftandrightEEGalphapowermayhavepromiseasanewmeasure.FrontalEEGasymmetryisanindexofthebasicemotionaldimensionofapproachversuswith-drawal(CoanandAllen,2004).IntheCIT,relativerightfrontalalphaactivitywassignificantlylowerforcriticalitemsthanfornon-criticalitems(Matsudaetal.,submitted).Thisresultsuggeststhatthecriticalitemwouldelicitwithdrawal-orientedmotivationandemotion,whichmaybeanadditionalindicatorofrecognitionofthecriticalitem.
SUMMARY
Inthepresentpaper,wereviewedhowtheCIThasbeenusedforfieldcriminalinvestigationsinJapan,andsuggestthatwithappro-priatetrainingandinstitutionalsupport,theCITcanfrequentlybeusedinfieldapplications.Wealsoreviewedvariousstatisticalmethodsandpotentialnewmeasures,whichmaycontributetoimprovedvalidityandincreasedprobativevalueoftheCIT.Wesuggestedthatmorestudiesofthesevariousstatisticalmethodsarerequiredbeforeapplyingthestatisticalmethodsinthefield.WealsohighlightedthepromiseofaddingnewquantificationofexistingmeasuresandaddingnewmeasuressuchasEEG/ERPindicestothecurrentfieldCIT.ItshouldbeanimmediategoaloftheJapaneseCITexaminersandresearcherstoimprovethepro-bativevalueofthefieldCITbyintroducingstatisticaljudgmentmethodsandthenaddingnewmeasurestothecurrentCIT.
Despiteimprovementsinmeasuresandstatisticalassessment,itisimportanttorememberthattheCITisnotatesttojudgewhetheranexamineeisguiltyorinnocent.TheCITcanshowonlywithrelativelyhighprobabilitywhethertheexamineerecog-nizesthecrime-relevantitem.Theexamineemayhaveobtainedcrime-relevantinformationbyanynumberofmeans,onlyoneofwhichisbybeingtheperpetratorofthecrime,whileothersincludeaccidentalexposureviamediaorinterrogations,orexposureviaarelationshipwiththeperpetratorofthecrime;agoodexaminerofcoursepayscloseattentiontoremovethesepossibilities.However,theCITresultcanbeusedasonescientificindicatorofwhetheranindividualmayhavebeeninvolvedinthecrimeunderinvestiga-tion.GiventhefundamentallysoundparadigmoftheCIT,andthepromiseofimprovementsusingmoresophisticatedstatisticsandadditionalmeasures,wehopethattheuseoftheCITwillincrease,withJapan’simplementationservingasausefulmodel.
ACKNOWLEDGMENTS
ThisstudywassupportedinpartbyKAKENHI24730650.WethankTokihiroOgawa,MichikoTsuneoka,andthereviewersfortheirhelpfulcomments.
www.frontiersin.orgNovember2012|Volume3|Article532|9
Matsudaetal.CurrentandfutureoffieldCIT
REFERENCES
Abootalebi,V.,Moradi,M.H.,andKhalilzadeh,M.A.(2006).Acom-parisonofmethodsforERPassess-mentinaP300-basedGKT.Int.J.Psychophysiol.62,309–320.
Abootalebi,V.,Moradi,M.H.,andKhalilzadeh,M.A.(2009).AnewapproachforEEGfeatureextractioninP300-basedliedetection.Com-put.MethodsProgramsBiomed.94,48–57.
Adachi,K.(1995).Statisticalclassi-ficationproceduresforpolygraphtestsofguiltyknowledge.Behav-iormetrika22,49–66.
Allen,J.J.B.,andIacono,W.G.(1997).Acomparisonofmethodsfortheanalysisofevent-relatedpotentialsindeceptiondetection.Psychophys-iology34,234–240.
Allen,J.J.B.,Iacono,W.G.,andDaniel-son,K.D.(1992).Theidentificationofconcealedmemoriesusingtheevent-relatedpotentialandimplicitbehavioralmeasures:amethodologyforpredictioninthefaceofindivid-ualdifferences.Psychophysiology29,504–522.
Allen,J.J.B.,andMertens,R.(2008).Limitationstothedetectionofdeception:trueandfalserecollec-tionsarepoorlydistinguishedusinganevent-relatedpotentialproce-dure.Soc.Neurosci.4,473–490.Ben-Shakhar,G.(1985).Standardiza-tionwithinindividuals:asimplemethodtoneutralizeindividualdif-ferencesinskinconductance.Psy-chophysiology22,292–299.
Ben-Shakhar,G.(2002).“Acriticalreviewofthecontrolquestionstest(CQT),”inHandbookofPolygraphTesting,ed.M.Kleiner(SanDiego:AcademicPress),103–126.
Ben-Shakhar,G.(2011).“Countermea-sures,”inMemoryDetection:The-oryandApplicationoftheConcealedInformationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cambridge:CambridgeUniversityPress),200–214.
Ben-Shakhar,G.,andDolev,K.(1996).Psychophysiologicaldetectionthroughtheguiltyknowledgetechnique:effectsofmentalcoun-termeasures.J.Appl.Psychol.81,273–281.
Ben-Shakhar,G.,andElaad,E.(2002).Effectsofquestions’repetitionandvariationontheefficiencyoftheguiltyknowledgetest:areex-amination.J.Appl.Psychol.87,972–977.
Ben-Shakhar,G.,andElaad,E.(2003).Thevalidityofpsychophysio-logicaldetectionofinformationwiththeguiltyknowledgetest:
ameta-analyticreview.J.Appl.Psychol.88,131–151.
Ben-Shakhar,G.,andMeijer,E.(2012).Skinconductance,respiration,heartrate,andP300intheconcealedinformationtest:ametaanalysis.Int.J.Psychophysiol.85,324.
Bradley,M.T.,Barefoot,C.A.,andArsenault,A.M.(2011).“Leakageofinformationtoinnocentsuspects,”inMemoryDetection:TheoryandApplicationoftheConcealedInfor-mationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cam-bridge:CambridgeUniversityPress),187–199.
Bradley,M.T.,andJanisse,M.P.(1981).Accuracydemonstrations,threat,andthedetectionofdecep-tion:cardiovascular,electrodermal,andpupillarymeasures.Psychophys-iology18,307–315.
Carmel,D.,Dayan,E.,Naveh,A.,Raveh,O.,andBen-Shakhar,G.(2003).Estimatingthevalidityoftheguiltyknowledgetestfromsimulatedexperiments:theexter-nalvalidityofmockcrimestud-ies.J.Exp.Psychol.Appl.9,261–269.
Coan,J.A.,andAllen,J.J.B.(2004).FrontalEEGasymmetryasamoder-atorandmediatorofemotion.Biol.Psychol.67,7–49.
Dawes,R.M.(1979).Therobustbeautyofimproperlinearmodelsindecisionmaking.Am.Psychol.34,571–582.
Ekman,P.(2001).TellingLie.CluestoDeceitintheMarketplace,Politics,andMarriage.NewYork:Norton.Elaad,E.(1990).Detectionofguiltyknowledgeinreal-lifecriminalinvestigations.J.Appl.Psychol.75,521–529.
Elaad,E.,andBen-Shakhar,G.(2006).Fingerpulsewaveformlengthinthedetectionofconcealedinformation.Int.J.Psychophysiol.61,226–234.Elaad,E.,Ginton,A.,andJungman,N.(1992).Detectionmeasuresinreal-lifecriminalguiltyknowledgetests.J.Appl.Psychol.77,757–767.
Farwell,L.A.,andDonchin,E.(1991).Thetruthwillout:Interrogativepolygraphy(“liedetection”)withevent-relatedbrainpotentials.Psy-chophysiology28,531–547.
Fukuda,K.(2001).Eyeblinks:newindicesforthedetectionofdecep-tion.Int.J.Psychophysiol.40,239–245.
Gamer,M.(2011a).“Detectingcon-cealedinformationusingautonomicmeasures,”inMemoryDetection:TheoryandApplicationoftheCon-cealedInformationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cambridge:CambridgeUniversityPress),27–45.
Gamer,M.(2011b).“Detectingofdeceptionandconcealedinfor-mationusingneuroimagingtech-niques,”inMemoryDetection:The-oryandApplicationoftheConcealedInformationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cambridge:CambridgeUniversityPress),90–113.
Gamer,M.,andBerti,S.(2010).Taskrelevanceandrecognitionofcon-cealedinformationhavedifferentinfluencesonelectrodermalactivityandevent-relatedbrainpotentials.Psychophysiology47,355–364.
Gamer,M.,Gödert,H.W.,Keth,A.,Rill,H.-G.,andVossel,G.(2008a).Electrodermalandphasicheartrateresponsesintheguiltyactionstest:comparingguiltyexamineestoinformedanduninformedinno-cents.Int.J.Psychophysiol.69,61–68.Gamer,M.,Verschuere,B.,Crombez,G.,andVossel,G.(2008b).Com-biningphysiologicalmeasuresinthedetectionofconcealedinformation.Physiol.Behav.95,333–340.
Gamer,M.,Kosiol,D.,andVossel,G.(2010).Strengthofmemoryencod-ingaffectsphysiologicalresponsesintheguiltyactionstest.Biol.Psychol.83,101–107.
Gamer,M.,Rill,H.G.,Vossel,G.,andGödert,H.W.(2006).Psychophys-iologicalandvocalmeasuresinthedetectionofguiltyknowledge.Int.J.Psychophysiol.60,76–87.
Hira,S.,andFurumitsu,I.(2002).PolygraphicexaminationsinJapan:applicationsoftheguiltyknowledgetestinforensicinvestigations.Int.J.PoliceSci.Manag.4,16–27.
Honts,C.R.,Devitt,M.K.,Winbush,M.,andKircher,J.C.(1996).Men-talandphysicalcountermeasuresreducetheaccuracyoftheconcealedknowledgetest.Psychophysiology33,84–92.
Kobayashi,T.,Yoshimoto,K.,andFuji-hara,S.(2009).Jitsumupolygraphkensanogenjyo[Thecontemporarysituationoffieldpolygraphtests].Jpn.J.Physiol.Psychol.Psychophysiol.27,5–15.
Krapohl,D.J.(2011).“Limitationsoftheconcealedinformationtestincriminalcases,”inMemoryDetec-tion:TheoryandApplicationoftheConcealedInformationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cambridge:CambridgeUniversityPress),151–170.
Littlewort,G.,Whitehill,J.,Wu,T.,Fasel,I.,Frank,M.,Movellan,J.,etal.(2011).“Thecomputerexpres-sionrecognitiontoolbox(CERT),”inProceedingsofIEEEInternationalConferenceonAutomaticFaceandGestureRecognition,298–305.
Lubow,R.E.,andFein,O.(1996).Pupil-larysizeinresponsetoavisualguiltyknowledgetest:newtechniqueforthedetectionofdeception.J.Exp.Psychol.Appl.2,164–177.
Lykken,D.T.(1959).TheGSRinthedetectionofguilt.J.Appl.Psychol.43,385–388.
Matsuda,I.,Hirota,A.,Ogawa,T.,Taka-sawa,N.,andShigemasu,K.(2006).Anewdiscriminationmethodfortheconcealedinformationtestusingpretestdataandwithin-individualcomparisons.Biol.Psychol.73,157–164.
Matsuda,I.,Hirota,A.,Ogawa,T.,Takasawa,N.,andShigemasu,K.(2009a).Within-individualdiscrim-inationontheconcealedinfor-mationtestusingdynamicmix-turemodeling.Psychophysiology46,439–449.
Matsuda,I.,Nittono,H.,Hirota,A.,Ogawa,T.,andTakasawa,N.(2009b).Event-relatedbrainpoten-tialsduringthestandardautonomic-basedconcealedinformationtest.Int.J.Psychophysiol.74,58–68.
Matsuda,I.,Nittono,H.,andOgawa,T.(2011).Event-relatedpotentialsincreasethediscriminationperfor-manceoftheautonomic-basedcon-cealedinformationtest.Psychophys-iology48,1701–1710.
Matsuda,I.,Nittono,H.,andOgawa,T.(2012).Decomposingcognitiveprocessesunderlyingtheconcealedinformationtestbyevent-relatedpotentials.Int.J.Biomed.SoftCom-put.Hum.Sci.18,5–11.
Matsuda,I.,andOgawa,T.(2011).Improvedmethodforcalculatingtherespiratorylinelengthinthecon-cealedinformationtest.Int.J.Psy-chophysiol.81,65–71.
Meijer,E.H.,Verschuere,B.,andBen-Shakhar,G.(2011).“Practicalguide-linesfordevelopingaCIT,”inMem-oryDetection:TheoryandAppli-cationoftheConcealedInforma-tionTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cam-bridge:CambridgeUniversityPress),293–302.
Nahari,G.,andBen-Shakhar,G.(2011).Psychophysiologicalandbehavioralmeasuresfordetectingconcealedinformation:theroleofmemoryforcrimedetails.Psychophysiology48,733–744.
Nose,I.,Murai,J.,andTaira,M.(2009).Disclosingconcealedinfor-mationonthebasisofcorti-calactivations.Neuroimage44,1380–1386.
FrontiersinPsychology|CognitiveScienceNovember2012|Volume3|Article532|10
Matsudaetal.CurrentandfutureoffieldCIT
Osugi,A.(2010).Gapandconnec-tionbetweenlaboratoryresearchandfieldapplicationoftheCITinJapan.Int.J.Psychophysiol.77,238.Osugi,A.(2011).“Dailyapplicationoftheconcealedinformationtest:Japan,”inMemoryDetection:The-oryandApplicationoftheConcealedInformationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cambridge:CambridgeUniversityPress),253–275.
Pollina,D.,Dollins,A.,Senter,S.,Brown,T.,Pavlidis,I.,Levine,J.,etal.(2006).Facialskinsurfacetemper-aturechangesduringa“concealedinformation”test.Ann.Biomed.Eng.34,1182–1189.
Reid,J.E.(1947).Arevisedquestion-ingtechniqueinlie-detectiontests.J.Crim.LawCriminol.37,542–547.Rosenfeld,J.P.(2011).“P300indetect-ingconcealedinformation,”inMem-oryDetection:TheoryandAppli-cationoftheConcealedInforma-tionTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cam-bridge:CambridgeUniversityPress),63–89.
Rosenfeld,J.P.,Cantwell,B.,Nas-man,V.T.,Wojdac,V.,Ivanov,S.,andMazzeri,L.(1988).Amodified,event-relatedpotential-basedguiltyknowledgetest.Int.J.Neurosci.42,157–161.
Rosenfeld,J.P.,Labkovsky,E.,Wino-grad,M.,Lui,M.A.,Vandenboom,C.,andChedid,E.(2008).Thecomplextrialprotocol(CTP):anew,countermeasure-resistant,accurate,P300-basedmethodfordetectionofconcealedinfor-mation.Psychophysiology45,906–919.
Sartori,G.,Agosta,S.,Zogmaister,C.,Ferrara,S.D.,andCastiello,U.(2008).Howtoaccuratelydetectautobiographicalevents.Psychol.Sci.19,772–780.
Sawada,Y.,Tanaka,G.,andYamakoshi,K.(2001).Normalizedpulsevolume(NPV)derivedphoto-plethysmographicallyasamorevalidmeasureofthefingervasculartone.Int.J.Psychophysiol.41,1–10.Vandenbosch,K.,Verschuere,B.,Crombez,G.,andDeClercq,A.(2009).Thevalidityoffingerpulselinelengthforthedetectionofconcealedinformation.Int.J.Psychophysiol.71,118–123.
Venables,P.H.,andMitchell,D.A.(1996).Theeffectsofage,sexandtimeoftestingonskinconductanceactivity.Biol.Psychol.43,87–101.
Verschuere,B.,Crombez,G.,Koster,E.H.W.,VanBockstaele,B.,andDeClercq,A.(2007).Startlingsecrets:startleeyeblinkmodulationbycon-cealedcrimeinformation.Biol.Psy-chol.76,52–60.
Verschuere,B.,andDeHouwer,J.(2011).“Detectingconcealedinfor-mationinlessthanasecond:responselatency-basedmeasures,”inMemoryDetection:TheoryandApplicationoftheConcealedInfor-mationTest,edsB.Verschuere,G.Ben-Shakhar,andE.Meijer(Cam-bridge:CambridgeUniversityPress),46–62.
Verschuere,B.,Meijer,E.,andDeClercq,A.(2011).Concealedinformationunderstress:atestoftheorient-ingtheoryinreal-lifepoliceinterro-gations.LegalCriminol.Psychol.16,348–356.
Winograd,M.R.,andRosenfeld,J.P.(2011).Mockcrimeapplica-tionofthecomplextrialprotocol(CTP)P300-basedconcealedinfor-mationtest.Psychophysiology48,155–161.
Zhao,M.,Zheng,C.,andZhao,C.(2011).Anewapproach
forconcealedinformationidentificationbasedonERPassessment.J.Med.Syst.36,2401–2409.
ConflictofInterestStatement:Theauthorsdeclarethattheresearchwasconductedintheabsenceofanycom-mercialorfinancialrelationshipsthatcouldbeconstruedasapotentialcon-flictofinterest.
Received:31July2012;accepted:10November2012;publishedonline:27November2012.
Citation:MatsudaI,NittonoHandAllenJJB(2012)Thecurrentandfuturesta-tusoftheconcealedinformationtestforfielduse.Front.Psychology3:532.doi:10.3389/fpsyg.2012.00532
ThisarticlewassubmittedtoFrontiersinCognitiveScience,aspecialtyofFrontiersinPsychology.
Copyright©2012Matsuda,NittonoandAllen.Thisisanopen-accessarticledis-tributedunderthetermsoftheCreativeCommonsAttributionLicense,whichpermitsuse,distributionandreproduc-tioninotherforums,providedtheoriginalauthorsandsourcearecreditedandsub-jecttoanycopyrightnoticesconcerninganythird-partygraphicsetc.
www.frontiersin.orgNovember2012|Volume3|Article532|11
因篇幅问题不能全部显示，请点此查看更多更全内容