1 Eyewitness Identification Laura Mickes Royal Holloway, University

EyewitnessIdentification
LauraMickes
RoyalHolloway,UniversityofLondon,UnitedKingdom
DepartmentofPsychology
[email protected]
ScottD.Gronlund
UniversityofOklahoma,UnitedStates
DepartmentofPsychology
[email protected]
1
Keywords:eyewitnessidentification,eyewitnessmemory,discriminability,reliability,
receiveroperatingcharacteristicanalysis,confidence-accuracyrelationship,confidence
accuracycharacteristicanalysis,diagnosticityratio,lineup,showup,systemvariables,
estimatorvariables,sequential,simultaneous
Abstract
Wepresentresultsthatchallengethenotionthateyewitnessmemoryisnotoriously
inaccurate.Thecoreofthischallengearisesfromviewingresearchfromthejoint
perspectivesofdiscriminability(theabilitytodistinguishinnocentfromguiltysuspects)
andreliability(thelikelihoodthepersonidentifiedisactuallytheperpetrator).Wealso
introduceaneworganizationalframeworkthatclassifiesvariablesaccordingtothe
differentstagesofmemoryastheymapontothedifferentstagesofthecrime/criminal
justiceprocess.Thediscriminability/reliabilityperspectiveandtheneworganizational
frameworkhavemajorimplicationsforhoweyewitnessidentificationevidenceshouldbe
utilizedbythecriminaljusticesystem.
Acknowledgements
ThisworkwaspartiallysupportedbytheEconomicandSocialResearchCouncil
[ES/L012642/1]toLauraMickes.
WethankKayolanGanevforhisvaluableassistance.
2
1
Introduction
Eyewitnesseshaveabadrapforbeinguntrustworthy.Althougheyewitnesseshave
difficultywithstandingsuggestionsandfeedback,theirmemoriesalsohavebeenfaulted
duetoaninabilitytouseconfidencelevelstoreflectthelikelyaccuracyoftheir
identification.Afterall,faultyeyewitnessidentificationshaveplayedarolein70%ofthe
wrongfulimprisonmentoftheinnocent(InnocenceProject,2015).Aglanceatthe
eyewitnessliterature,themediacoverage,andwhatiswrittenintextbooks,makesit
seemasifthememoryofaneyewitnessisfunctionallyuseless.Take,forexample,the
statementwritteninarecentnewsarticlecoveringacrimeinwhichthereweremultiple
eyewitnesses:“Whileeyewitnessaccountsareoftenfickle–andevenshockingly
inaccurate–theimmediacyoftheresponseinthiscasehelpedpoliceactquickly.The
videowascertainlykey,giventheunreliabilityofhumanmemory”(TheChristianScience
Monitor,2016).Isthenotoriousreputationmerited?Caneyewitnessmemoryneverbe
trusted?Thischapterpresentsresultsthatchallengethatnotion.
1.1 Systemvs.EstimatorVariables
Researchershavedissectedthevariablesthataffecteyewitnessaccuracyand
studiedeachconstituentpart.Thosepartsoftenhavethenbeengroupedintowhether
theycanbeusedtomakedecisionstoimprovethesystem(e.g.,decisionsmaderegarding
howtoconductalineup,orwhatinstructionstoreadtoaneyewitness)orwhetherthey
involvevariablesthataffecttheaccuracyofanidentification(e.g.,across-race
identification,lengthoftheretentioninterval).Thesevariablestypicallyhavebeen
categorizedassystemandestimatorvariables(Wells,1978),respectively.Asystem
variableisavariablethatisunderthecontrolofthelegalsystemandanestimator
3
variableisnot.Thatdistinction,however,onlyappliestopolicymakerswhoneedtomake
decisionsabouthoweyewitnessevidenceshouldbegatheredbythepolice,becausehow
alineupisconductedcanbespecified,butwhethertheeyewitnessandtheperpetrator
arethesameracecannot.However,thereisnosuchsystem/estimatordistinctionwhenit
comestothecourtoflaw.Thatis,forjudgesandjurorswhoneedtodeterminethelikely
culpabilityofadefendantbasedontheidentificationofaneyewitness,thereisnocontrol
overanyvariablewhatsoever.Take,forexample,lineuppresentation.Whethertopresent
aneyewitnessasimultaneous-orsequentially-presentedprocedureisunderthecontrol
ofthepolice,andthusisasystemestimatorattheinvestigativestage.Butwhenthatcase
movestothecourtoflaw,whetherasimultaneousorsequentiallineupwaspresentedis
avariablethatisnotunderthecontrolofthejudgeandjurors.
1.2 Anewconceptualizationofvariablesthataffectmemory
Figure1depictsanewwaytoconceptualizethevariablesthataffecteyewitness
memory,andhow,overthecourseofacrimeanditssubsequentinvestigation,thenature
ofthevariableschanges.Variablesarecategorizedaccordingtoitsoccurrenceinastage
inmemoryandthestageinthecrimeandsubsequentcriminalproceedings.Theeventsof
thecrimeareencoded,stored(maintainedinmemory)duringthetimebetweenthe
crimeandthereportingofthecrime,andfinallythememoryisretrievedwhenreporting
thecrimeandattemptingtoidentifytheperpetratorfromalineup.
Toprovidecontext,considerthefollowingexampleinwhicharobberyis
witnessed.Iftheperpetratorisofadifferentracethanthewitness,thenthatmayaffect
lateridentificationbecauseofthecrossracebias(Meissner&Brigham,2001,a
phenomenonthoughttooccurduringEncoding).Thetimebetweenwitnessingthe
robberyandreportingthecrime(Storage)wouldalsoaffectmemorybecauselonger
4
retentionintervalswilldecreasememoryperformanceduetotheeffectsofforgettingand
interferenceandcreateopportunitiesforinfluencessuchasmisinformation(e.g.,Loftus,
2005)andsuggestibility(e.g.,Zaragoza&Lane,1994;Lindsay&Johnson,1989).Whenthe
witnessattemptstoidentifytheperpetratorfromalineup(Retrieval),thetypeoflineup,
suchasasimultaneouslineup,maymakeadifferenceinmemoryperformancecompared
toifmemorywastestedon,say,asequentiallineup.
1.3 TwoTypesofAccuracy
TheorganizationalstructureinFigure1providesaroadmapforthetypesof
analysesthatareimportanttothestakeholdersthatdealwitheyewitnessevidence.Both
setsofstakeholdersareinterestedinaccuracy,butindifferenttypesofaccuracy,and
therefore,resultsfromdifferenttypesofanalysesarerelevant.Onetypeofaccuracy,the
typethatismostoftenconsideredinlaboratorylist-learningmemoryexperiments,is
discriminability.Inlist-learningexperiments,alistofitems,calledtargets,arepresented
duringthestudyphaseandthen,duringthetestphase,targetsanditemsthatwerenot
presentedduringthestudyphase,calledlures,arerandomlyintermixedandpresentedto
aparticipant.Theparticipantindicateswhethereachitemwasorwasnotonthelist.
Likewise,inaforensically-relevantexperiment,duringthestudyphase,aparticipantwill
typicallyviewavideoofamockcrimewheretheperpetratoristhetarget.Memoryfor
theperpetratoristhentestedusingalineup.
<Figure1nearhere>
Discriminabilityistheabilitytodistinguishthetarget,ortheguiltysuspect(the
individualwhocommittedthecrime),fromalure,ortheinnocentsuspect(anindividual
whothepoliceincorrectlysuspectofcommittingthecrime).Discriminabilityistypically
measuredbyd'(aparametricmeasureofdiscriminability,orsensitivity),percentcorrect,
5
orareaundertheReceiverOperatingCharacteristic(ROC)curve(anon-parametric
measureofdiscriminability)(seeMacmillan&Creelman,2005).Approachestomeasuring
lineupdiscriminabilityarediscussedindetailinsection2.2.
Adifferenttypeofaccuracy–onethatshouldbeconsideredwhendetermining
eyewitnessreliability(i.e.,thelikelihoodthattheidentifiedsuspectisguilty)–appliesto
subsetsofeyewitnesseswiththesamediscriminability.Toevaluateaccuracyofthistype,
positivepredictivevalue(forexample,theproportioncorrectoralikelihoodratioateach
levelofconfidence)isconsidered.Thisistypicallymeasuredwithcalibrationanalysis
(Juslin,Olsson,&Winman,1996),oracloserelative,confidence-accuracycharacteristic
(CAC)analysis(Mickes,2015).Calibrationanalysis,especiallyCACanalysis,providesthe
informationthatismostrelevanttojudgesandjurors,whoarechargedwithdetermining
culpability,becauseitindicatestheaccuracyforagivenlevelofconfidencereportedby
theeyewitness.Approachestomeasuringcalibrationarediscussedindetailinsection2.3.
Decisionsaboutvariablesthatarepartoftheinvestigationprocess,thatmight
increasediscriminability,shouldbeinformedbyresultsfromROCanalysis(asdepictedin
Figure1).Thisisthetypeofaccuracythatpolicymakersmustconsiderwhenmaking
decisionsregardingwhetheroneidentificationprocedureissuperiortoanother.
However,courtroomdecisionsabouteyewitnessreliabilityshouldbeinformedbyresults
fromcalibration-typeanalyses.Whenitcomestoadvisinglegalprofessionals,ortestifying
incourtsasexpertwitnesses,thisdistinctionmustbekeptinmind.
Webeginourchapterbyexpandingupontheassessmentofdiscriminabilityand
reliability.Asmentioned,discriminabilitycanbeassessedusingROCanalysis,andwewill
introducethistechniqueanddiscussseveralargumentsthathavebeenmaderegardingits
use.Thiswillbefollowedbyadiscussionofwaystomeasurepositivepredictivevalueto
6
assessreliability.Next,wewillturntoareviewofseveralvariablesknowntoaffect
eyewitnessmemory,andwillconsidertheimpactofthesevariablesondiscriminability
andreliability.Thesevariableswillbeorganizedaccordingtothememoryphaseupon
whichtheyhavethebiggestimpact(encoding,storage,orretrieval).Wewillclosewith
theimplicationsofthechangingviewsofeyewitnessidentificationrevealedbythe
assessmentofdiscriminabilityusingROCanalysesandtheassessmentofthereliability
usingcalibrationandCACanalyses.
2
TestingandMeasuringEyewitnessMemory
Eyewitnessidentificationprovidesevidenceduringthecourseofcriminalcases.One
waypoliceinvestigatorsgleaninformationfromaneyewitnessistoadministeratypeof
identificationprocedure,whichisessentiallyarecognitionmemorytestofthe
eyewitness’smemoryfortheperpetrator.Identificationproceduresincludelineups
(photosorvideos),livelineups,showups,fieldviews,andmugbooks(NationalResearch
Council,2014).Thevariablesthataffecteyewitnessaccuracyareoftenmeasuredinthe
labusinglineupsorshowups,andthatiswhatwefocusoninthischapter.
2.1 Lineups
Astandardpolicelineupcontainsonesuspect(whoiseitherguiltyorinnocent)
andmultiplefillers(peoplewhoareknowntobeinnocent).Figure2showsaschematicof
a6-personsimultaneousphotolineup.Theimagesofeachlineupmember(photosor
videos)areeithersimultaneouslyorsequentiallypresented,dependingonthepolicy
adoptedbyaparticularcountryorbyaparticularjurisdictionwithinacountry.
<Figure2nearhere>
7
Therearethreepossiblechoicesthateyewitnessescanmakewhenfacedwitha
lineup.Theycanidentify1)thesuspect,2)oneofthefillers,or3)noone.Allpossible
outcomesareinTable1.Ifthetargetispresentinthelineup,andtheeyewitness
identifiesthem,thenthatisacorrectidentification(alsoknownasahit).Ifthetargetis
notpresentinthelineup,andtheeyewitnessidentifiestheinnocentsuspect,thenthatis
afalseidentification(alsoknownasafalsealarm).Thesedecisionsareinredboldedfont
becausethesearethedecisionsthatmatter,andthusthefocusistypicallyplacedonthe
identificationoftheguiltyandinnocentsuspects.Thatisbecausefillersareknowntobe
innocent,andalthoughpickingoneisanerror,thereisnoriskofthefillerbeing
wrongfullyinvestigatedorconvicted.Iftheeyewitnessdoesnotidentifyalineupmember,
andifthetargetisinthelineup,thenthatisamiss;andifthetargetisnotpresentinthe
lineup,thenthatisacorrectrejection.Inrealpoliceinvestigations,itisunknownwhether
thesuspectisinnocentorguilty.Inlaboratorystudies,bycontrast,itisknownwhether
thesuspectisinnocentorguilty.Thus,inalaboratorystudy,thecorrectIDrateisthe
proportionofguiltysuspectsidentifiedfromthetarget-presentlineups.ThefalseIDrate
istheproportionofinnocentsuspectsidentifiedfromthetarget-absentlineups.Correct
IDratesandfalseIDratesarethenusedtoassessthediscriminabilityofalineup
procedure.
<Table1nearhere>
TherearedifferentapproachestoobtainingthefalseIDrate.First,anindividual
couldbedesignatedastheinnocentsuspect,andthuscomputingthefalseIDrateis
straightforward(i.e.,thenumberoftimestheinnocentindividualwasidentifieddivided
bythenumberoftarget-absentlineups).Often,thereisnoactualinnocentsuspectina
labstudy,andthereareseveraldifferentapproachestocomputingthefalseIDrate.One
8
approachistorandomlydesignateonefillerinatarget-absentlineuptoserveasthe
innocentsuspectafterdatacollection(suchthatanyidentificationofthatindividual
wouldcountasaninnocentsuspectID).Asecondapproachistodesignatethefillerwho
ismostoftenmisidentifiedastheinnocentsuspect(andanyidentificationofthat
individualwouldcountasaninnocentsuspectID).Athirdapproachistoestimatethe
numberofinnocentsuspectidentificationsfromthenumberoffillerIDsfromtargetabsentlineups.ThisestimateisobtainedbydividingthenumberoffillerIDsbythe
numberoflineupmembers(e.g.,sixforthelineupinFigure2).Thatestimatedvalueis
thendividedbythenumberoftarget-absentlineupstoestimatethefalseIDrate.
Dependingonthefairnessofthelineup,thoseapproachesshouldyieldsimilar
conclusions.
2.2 Discriminability
2.2.1 Pastmeasureofdiscriminability:Thediagnosticityratio
Formanyyears,claimsaboutdiscriminabilitywerealmostalwaysmadeonthebasis
ofthediagnosticityratio.Thediagnosticityratioisalikelihoodratioinwhichthecorrect
IDrateisdividedbythefalseIDrate.Forexample,basedonthehigherdiagnosticityratio
thatwasfrequentlyobservedforthesequentiallineupprocedure,thatprocedurewas
declareddiagnosticallysuperiortothesimultaneousprocedure(Steblay,Dysart,Fulero,&
Lindsay,2001;Steblay,Dysart,&Wells,2011).Thesameconclusionwasreachedwhen
thesimultaneousprocedurewascomparedtotheshowupprocedure,alsousingthe
diagnosticityratio(Steblay,Dysart,Fulero,&Lindsay,2003).However,thediagnosticity
ratioisnotthebestwaytomeasurediscriminabilitybecauseitconflatesresponsebias,
whichisthelikelihoodofpickingsomeonefromalineup,withdiscriminability(Wixted&
Mickes,2012;Gronlund,Wixted,&Mickes,2014).Instead,ReceiverOperating
9
Characteristic(ROC)analysisisamoresuitablemeasureofdiscriminability(National
ResearchCouncil,2014).
2.2.2 Non-parametricmeasureofdiscriminability:Receiveroperating
characteristic(ROC)analysis
ROCanalysisiswidelyusedinotherfields,particularlyinappliedfields,suchas
diagnosticmedicine(Lusted,1971a,1971b;Swets,1988;Swets,Dawes,&Monahan,
2000).Inmedicine,ROCanalysishasbeenthepreferredmethodologysincethe1970sfor
identifyingthemostaccurateprocedureforidentifyingadiseaseinapatient.Thegoalin
diagnosticmedicineisalwaystoidentifytheprocedurethatyieldsthebest
discriminability(inthiscase,theabilityofamedicalproceduretodistinguishbetween
thosewhohavethediseasefromthosewhodonot).Tomaximizediscriminability,the
bestprocedureistheonethatreducesbotherrorsthatcanoccur:1)incorrectly
identifyingadiseaseinapatientwithoutthedisease,and2)failingtoidentifythedisease
inapatientwhohasthedisease.Theformerisanalogoustoincorrectlyidentifyingan
innocentsuspectinalineuporshowup(afalseID),andthelatterisanalogoustonot
identifyingtheguiltysuspectinalineuporshowup(amiss).Becauseidentifyingadisease
inapatientisconceptuallythesameasidentifyingtheguiltysuspectinalineup,ROC
analysiswasadaptedforuseinlineupdata(Wixted&Mickes,2012;Mickes,Flowe&
Wixted,2012).
AnROCplotisaplotofthecorrectIDrateandfalseIDrateforeverylevelof
responsebias(confidence)andisshownusinghypotheticaldatainFigure3.The
correspondingdatausedtoplottheROCcurvesareshowninthetablebelowinredfont.
TheboldedvaluesrepresenttheoverallcorrectIDratesandfalseIDratesforboth
procedures(andthevaluesusedtocomputediagnosticityratios).InFigure3,theleftmost
10
pointisthecorrectIDrateandfalseIDratereflectingthoseidentificationsmadewiththe
highestlevelofconfidence(100%confident,themostconservativeresponsebias).The
nextpointtotherightisthecorrectandfalseIDratepairforthoseidentificationsmade
withmediumconfidenceandabove.TherightmostpointisthecorrectrateandfalseID
rateforthoseidentificationsmadewithlow,mediumorhighlevelsofconfidence.In
otherwords,therightmostpointistheoverallcorrectIDrateandfalseIDrate.The
dashedlineisthelineofchanceperformance,andifthepointsfellonthatlineitwould
meanthattheeyewitnesshasnoabilitytodistinguishinnocentfromguiltysuspects.The
furtherthecurvefallsabovethelineofchanceperformance,thebetterthe
discriminability.Thatis,thebettereyewitnessescandistinguishinnocentfromguilty
suspects.InFigure3,itisclearthatProcedureAgivesrisetogreaterdiscriminabilitythan
ProcedureB,becausetheROCisfurtherfromthelineofchanceperformance.Tomeasure
thestatisticaldifferencesbetweenthecurves,thepartialareaunderthecurve(pAUC)is
computedandcompared(asdescribedindetailinGronlund,Wixted,&Mickes,2014).
<Figure3nearhere>
ToconductROCanalysis,onemustcollectdatafromhundredstothousandsof
participantsperconditionbecausethereisonlyonetrialperparticipant(justlikethe
experienceofarealeyewitness).Despitelineupdataconsistingofonedecisionper
eyewitness,thecombineddataset,whichcontainsindividualswithdifferentlevelsof
discriminabilityanddifferentcriterionsettings,canneverthelessbeusedtoassess
discriminability.Noadditionalassumptionsneedtobemadeabouttheformofthe
underlyingmemorydistributionsorcharacteristicsofthedecisionprocess.
11
2.2.2.1 ArgumentsregardingROCAnalysis
TheUSNationalAcademyofSciencescommitteecommissionedtoassessthe
currentstateofeyewitnessidentificationresearchmadethefollowingstatementabout
ROCanalysis:
“ROCanalysisisapositiveandpromisingstep,withnumerousadvantages.For
example,theareaundertheROCcurveisasingle-numberindexof
discriminability.Moreover,thisindexreflectsaparameter-freeapproachto
binaryclassificationperformance;theoutcomeisentirelydata-dependent
andthusidenticalacrossallusersdrawingfromthesamedataset(Green&
Swets,2009).Mostimportantlyforitsapplicationtotheproblemof
evaluatingeyewitnessperformance,theROCapproachpossessesadistinct
advantagebecausethedimensionsofanalysis–discriminabilityandresponse
bias–mapdirectlyontothemechanisticparametersofcausalmodelsof
humanrecognitionmemory(seeChapter4).”NationalResearchCouncil,
2014;p.59)
Despitethecommittee’ssupportofROCanalysisforlineupdata,itsuseisnotwithoutits
critics(Lampinen,2016;Wells,Smalarz,&Smith,2015).Wellsetal.claimedthat:“Yes,the
NationalResearchCouncil(NRC)reportgotitwrongbyinterpretingROCanalyseson
lineupsasmeasuresofunderlyingdiscriminability,”(p.325,Wells,Smith,&Smalarz,
2015).Eachoftheirargumentsisdescribedbelow(anddiscussedindetailinWixted&
Mickes,2015a;Wixted&Mickes,2015b).
2.2.2.1.1
Anti-ROCargument#1:EyewitnessidentificationanddiagnosticmedicineROCs
aredifferent
ThefirstargumentisthateyewitnessidentificationROCsanddiagnosticmedicine
ROCsaredifferent.Thefactthatlineupshavefillers,andmedicaltestsdonot,maybe
consideredasproblematic.Asmentionedpreviously,ROCanalysisisthepreferred
methodofanalysisinmedicinewhenthequestioniswhichdiagnostictestisbetterat
discriminatingdiseasefromnon-diseasestates(e.g.,Zweig&Campbell,1993).Lineupsare
12
conceptuallyanalogouswhenthequestioniswhichlineupprocedure,forexample,is
betteratdiscriminatingguiltyfrominnocentsuspects(Mickesetal.2012).Although
argumentsagainstthisclaimhavebeenadvanced(Wells,Smalarz,&Smith,2015),itis
clearthecaseinfavorofthesimilaritybetweenROCsfromdiagnosticmedicineandROCs
fromeyewitnessidentificationiscompelling.Consider,forexample,detection-plusquadrant-localizationtasksinwhichradiologistsarepresentedwithanx-raysectionedin
fourquadrantsandtheirtaskistoidentifythelocationofatumor,ifpresent,fromoneof
fourquadrants.Thisisisomorphictoaneyewitnessbeingaskedtoidentifythelocationof
aperpetrator,ifpresent,inoneofsixpositionsofalineup.Inboththex-rayandthe
lineup,theparticipantispresentedwithatarget-present(tumororguiltysuspectpresent)
ortarget-absent(notumorornoguiltysuspectpresent).
ThetoppanelofFigure4(fromWixted&Mickes,2015a,Figure2)showsROC
curvesconstructedusingtwoapproacheswiththedatareportedbyStarr,Metz,Lusted,
andGoodenough(1975).Oneapproachconsidersallidentificationsmadefromatargetpresentx-ray(thisincludescorrectIDsofthetumorand“filler”IDs,whichare
identificationsmadetoanyquadrant),withthecorrespondingratesplottedonthey-axis.
Allidentifications(toanyquadrant)madeintarget-absentx-rays,andthecorresponding
rates,areplottedonthex-axis.ThisapproachyieldstheROCcurvelabeled“ROC”.The
secondapproachgivesrisetolocationROC(LROC),andincludesonlytumorIDsfromthe
quadrantinwhichthetumorislocatedascorrectIDs.Thisapproachisanalogoustolineup
ROCs(whichonlycountguiltysuspectIDsfromtarget-presentlineupswhencomputing
thecorrectIDrate).
<Figure4nearhere>
13
Asacomparison,theROCandLROCplotfromtheStarretal.(1975)study,andthe
ROCandLROCfromaneyewitnessstudy(Palmer,Brewer,Weber,&Nagesh,2013),are
placedsidebyside.InthePalmeretal.study,theROCcurveincludesallidentifications
fromtarget-presentandtarget-absentlineups.TheLROCcurveonlyincludescorrectIDs
madefromtarget-presentlineups.Thefalsealarmrateiscomputedfromalltarget-absent
fillerIDs(bottomhorizontalaxis)andthetophorizontalaxisshowstheestimatedfalseID
rate(bydividingthex-axisvaluesbylineupsize).Thus,thefactthattheLROCcurveand
theROCcurvearethesamewhetherfillerIDsareincludedisevidencethatincludingfiller
IDsdoesnotchangeconclusions.Thefactthatthetwoplotsareundeniablysimilar
underscoresthepointthatdiagnosticmedicineROCsandeyewitnessidentificationROCs
areinherentlyalike.
2.2.2.1.2
Anti-ROCargument#2:ROCanalysisignoresfilleridentifications
ThesecondargumentisthatROCanalysisneedstoincludefilleridentificationsto
meaningfullymeasurediscriminability.Inthepast,identificationsoffillers(alsoknownas
“foils”)werenotconsideredofimportanceintheappliedsense.Wells&Turtle(1986)
madethispoint:
“Note,however,thattheidentificationofafoil,althoughatrueidentification
error,isa"known"error.Thatis,inanactuallineuporphoto-spread
situation,theidentificationofafoilwillbedetectedasanerror.The
identificationofafoildoesnotresultinchargesbeingbroughtagainstthe
identifiedperson.Inotherwords,theidentificationofafoilisnotafalse
identificationintheforensicsense.Throughouttheremainderofthisarticle,
weusethetermfalseidentificationonlytorefertotheidentificationofan
innocentsuspect;wecallinaccurateidentificationsoffoilsfoilidentifications.”
(emphasisadded;p.321)
SincetheimportationofROCanalysisforlineupdata,however,therehasbeenmuch
discussionaboutfilleridentifications.Wells,Yang,andSmalarz(2015)claimedthatROC
analysisisproblematicbecauseidentificationsmadetofillersarenottreatedaserrors.A
14
newtermwasthenintroduced,fillersyphoning,thatdescribestheideathatinnocent
suspectsarelesslikelytobechosen(andthereforeprotected)becausefillersarechosen
instead(Wells,Smalarz,&Smith,2015).TheyarguethatsomeofthedifferencesacrossID
procedures,forexample,occurduetodifferentialfillersyphoning,andnottodifferential
discriminability.TheyfurtherarguethatROCanalysismasksfillersyphoningbecauseit
ignoresfilleridentifications.Therearethreemajorchallengestothatargument.First,filler
identificationswerenotincludedinthediagnosticityratioeither,makingROCanalysisno
differentinthatregard.Second,asindicatedabovebyWellsandTurtle(1986),onlythe
identificationsofguiltyandinnocentsuspectsmatter,notfilleridentifications.Third,filler
identificationscanbeincludedinROCanalysis,andtheconclusionsdonotchange.
Thereisanimportantdifferencebetweendiagnosticx-rayROCsandlineupROCs,
anditspeakstotheirrelevanceoffillerIDsinlineupROCs.Indiagnosticmedicine,“filler”
IDs(i.e.,decidingthatatumorispresentinaquadrantthatdoesnotincludeatumor)is
anunknownerror.Thatis,adoctorwillneedtoperformanadditionaltesttodetermine
whetherthequadrantheorshechoseshouldbeclassifiedasahit(tumorpresent),afalse
alarm(notumorpresentinthatquadrant),orevenafalsealarmandamiss(notumor
presentinthechosenquadrantbutatumorwaspresentinanadjacentquadrant).
However,thepoliceneednotconductanyadditionaltestwhenafillerischosen;they
knowthisindividualtobeinnocent.Astrictanalogofthex-rayandlineupsituations
wouldoccuronlywhenpriorknowledgeexistsofwhereatumorislocated(aswouldbe
thecasewhenaseniordoctor,whoknowstheanswer,istrainingajuniordoctorwho
doesnot).OnlyinthissituationwouldtheseniordoctorknowtoignoretheIDfroma
“filler”quadrant,justlikethepolicealreadyknowregardingtheIDofafiller.
15
2.2.2.1.3
Anti-ROCargument#3:Thediagnosticityratioiswhatthelegalsystemwantsto
know
Thethirdargumentisthatthediagnosticityratioiswhatthelegalsystemreally
wantstoknow.However,resultsfromboththediagnosticityratioandROCanalysisare
usefulforthelegalsystem,andwhichanalysisshouldbeconsidereddependsonthe
questionofinterestbecausethetwoanalysesaddressdistinctquestions.Whentryingto
answerthequestion:whatistheprobabilitythataparticularsuspectwhohasbeen
identifiedfromalineupisguilty,thediagnosticityratio,notROCanalysis,isuseful
(Mickes,2016;Mickes,Flowe,&Wixted,2012).Butwhentryingtoanswerthequestion:
whichprocedureisdiagnosticallysuperior,ROCanalysis,notthediagnosticityratio,is
required.ThisdistinctionisnewtoeyewitnessIDresearch,butthedistinctioniswell
understoodinmedicine.Policymakersandleadersofhealthorganizations(e.g.,American
DiabeticSociety)endorsethetestthatyieldsthebestdiscriminabilitybasedonresults
fromROCanalysis.Medicaldoctorsandtheirpatients,ontheotherhand,needtoknow
foragivendiagnosis,howlikelyisitthattheyhavethedisease(e.g.,diabetes).Inthis
case,resultsfromalikelihoodratioarerelevant.Failingtounderstandthisdistinction,and
consequentlyusingthediagnosticityratiotomakeconclusionsaboutdiscriminability,has
resultedinmisleadingconclusions.
2.2.2.1.4
Anti-ROCargument#4:ROCanalysismeasures“psychological”discriminability,
notobjectivediscriminability
ThefourthargumentisthatROCanalysismeasures“psychological”
discriminability,not“objectivediscriminability”,butinfactROCanalysiscanbeused
eitherforappliedpurposesortheoreticalpurposes.Forappliedpurposes,ROCanalysis
assessesobjectivediscriminabilityandmerelygraphicallydepictsthedata.Inthiscase,no
16
theoreticalconsiderationsareneeded.Indeed,justlikeindiagnosticmedicine,thetypical
useofROCanalysisineyewitnessidentificationisforappliedpurposes.Considerdata
fromWetmoreetal.(2015)inwhichtheycomparedperformanceonashowup(onlya
suspect,nofillers,areshown)toperformancefromafairlineup.
Toprovideaconcreteexample,imagine200lineupsand200showups.Figure5
depictsidentifiedguiltysuspects(inblack)andidentifiedinnocentsuspects(inred)for
bothashowup(onthetopleft)andalineup(onthetopright).Consideringonlythefair
lineupsandshowups(usingthe“innocentweak”suspect),Wetmoreetal.found,forthe
showup,theoverallcorrectIDratewas61%andtheoverallfalseIDratewas42%,and,
forthelineup,thecorrectIDratewas67%andthefalseIDratewas10%(asshowninthe
toppanels).Notheoreticalconsiderationsareneededtoappreciatethatthelineupyields
higherobjectivediscriminabilityinthatthereareoverallmoreguiltysuspectsand
simultaneouslyfewerinnocentsuspectsidentifiedwiththelineupcomparedtothe
showup.ThefigurebelowshowstheWetmoreetal.(2015)dataplottedinROCspace.
TherightmostpointsoutlinedinredrepresenttheoverallcorrectandfalseIDratesfor
theshowupandlineup.WhetherconsideringonlythemostliberalpointontheROC
(pointsoutlinedinred),ortheentireROCcurves,thestoryisthesame:thelineupis
superior.
<Figure5nearhere>
ROCanalysiscanalsobeusedfortheoreticalpurposestomeasure“psychological
discriminability”,whichisdiscriminabilityinthemindofaparticipant.Thiscanbe
mimickedusingasimplesignaldetectionmodel(toppanelofFigure6)thatmapsontothe
lineupsituation(forasimilarinstantiation,seeClark,2003).Themodelassumesthata
17
witnessfirstdeterminesthelineupmemberwiththegreatestmemorystrength,andthen
identifiesthatlineupmemberifhismemorystrengthexceedsaresponsecriterion,c
(otherwise,thelineupisrejected).ThismodelassumesGaussiandistributionsofmemory
strengthforfillers,innocentsuspects,andguiltysuspectswithmeansofμFiller,
μInnocent,andμGuilty,respectively,andstandarddeviationsof1(assumingequal
variance).A6-membertarget-presentlineupisconceptualizedas5randomdrawsfrom
theFillerdistributionand1randomdrawfromtheGuiltydistribution.The
conceptualizationofatarget-absentlineupdependsonwhetherthelineupisfair(i.e.,the
suspectdoesnotstandoutamongstthefillers)orunfair.Ifthelineupisfair,the
distributionforfillersandtheinnocentsuspectarethesame,anda6-membertargetabsentlineupisconceptualizedas6randomdrawsfromtheFillerdistribution.However,
ifthelineupisunfair,a6-membertarget-absentlineupisconceptualizedas5random
drawsfromtheFillerdistributionand1randomdrawfromtheInnocentsuspect
distribution.Inallcases,themodelassumesthattheeyewitnessalwaysselectsthelineup
memberwiththegreatestassociatedmemorystrengthprovidedthatthisstrength
exceedsaresponsecriterion,c.Theabilityofeyewitnessestodiscriminateinnocentfrom
guiltysuspectsisofinterest,whichisrepresentedbythedistancebetweenthemeansof
theμInnocentandμGuiltydistributions(i.e.,thegreaterthedistance,thegreaterthe
discriminability).
Figure6showstwosignaldetectionmodels.Themodelonthetopleftrepresents
afairlineupwheretheinnocentsuspectandthefillersresembleeachother(i.e.,μFiller=
μInnocent)andthemodelonthebottomleftrepresentsanunfairlineupwherethe
innocentsuspectmatchestheguiltysuspectmuchmorethanthefillersdo(i.e.,μFiller<
μInnocent).ThecorrespondingfairandunfairlineupROCcurvesontherightshowthe
18
factthattheareaunderthecurveisgreaterforthefairlineupthantheunfairlineup.
Thus,ROCanalysescanalsobeusedtomakesenseofpsychologicaldiscriminability(i.e.,
therearethreedistributionsinanunfairlineup,inafairlineup,themodelcanbereduced
toatwodistributionmodel).Althoughthereareobviousparallelsbetweenthetwotypes
ofdiscriminability,objectivediscriminabilityinformsusabouttheaccuracyofaprocedure
(andrequiresnotheoreticalassumptions)whereaspsychologicaldiscriminabilityinforms
usaboutthecapabilitiesofanindividual,aconclusionthatrestsonseveraltheoretical
assumptions.
<Figure6nearhere>
2.2.3 Parametricmeasureofdiscriminability:d'
BecauseROCanalysiswasonlyrecentlyintroducedtotheeyewitnessliterature,
manyofthevariablesthathavebeeninvestigatedinthepast,andwhichwereview
below,haveyettobeexaminedusingROCanalysis.Thatmeansthatwhathaspreviously
beendeclaredaperformancedifferencecouldinfacthaveaffectedresponsebiasandnot
discriminability.However,d'(aparametricmeasureofdiscriminability,i.e.,thedifference
betweenμInnocentandμGuiltyinthetoppanelofFigure6)generallyprovidesacloser
approximationtothetruththanthediagnosticityratiointhatd’islittleaffectedby
responsebiases(Mickes,Moreland,Clark&Wixted,2014).Figure3showsthesuspectID
rates,fillerIDrates,andnoIDsforalllevelsofconfidencefortarget-presentandtargetabsentlineupsforthetwoproceduresintheROCplot.Allthreesummarystatistics(the
diagnosticityratio,pAUC,andd')arepresentedaswell.Ineachcase,ProcedureA
outperformsProcedureB.
19
Theproblemwithusingthediagnosticityratiotomeasurediscriminabilitycanbe
easilyillustrated.Foreachlevelofconfidence,thediagnosticityratiocanbecomputed,so
forProcedureA,thediagnosticityratioforidentificationsmadewithlow,medium,and
highlevelsofconfidenceare6.8(.536/.079),10.0(.357/.036),and22.0(.157/.007),
respectively.ForProcedureB,thediagnosticityratioforidentificationsmadewithlow,
medium,andhighlevelsofconfidenceare2.5(.257/.103),3.1(.143/.046),and4.0
(.057/.014),respectively.Thus,thediagnosticityratioincreasesmonotonicallyas
respondingbecomesmoreconservative.Itisanerrortoassumethatthehigher
diagnosticityratioismoreaccurate(i.e.,resultsinbetterdiscriminability)becauseitis
anotherwayofsayingthatconservativerespondingismoreaccurate.Inthisexample,
regardlessofresponsebias–liberaltoconservative–discriminabilityisthesameacross
theROCcurve.Thed'values,ontheotherhand,donotmarkedlychangeasresponding
becomesmoreconservative.ForProcedureA,d'=1.4foridentificationsmadewithlow
confidenceandd'=1.5foridentificationsmadewithmediumandhighlevelsof
confidence.ForProcedureB,d'=0.6foridentificationsmadewithlow,medium,andhigh
levelsofconfidence.Thestabilityofd'acrosschangingresponsebiasesiswhyd'isamore
appropriatemeasureofdiscriminabilitythanthediagnosticityratio.Therefore,toassess
thecurrentstateofknowledge(sections3,4and5),wecomputed'valuestoassess
discriminabilityinthecaseswhereROCanalysishasyettobeconducted. 2.3 Reliability
Thereliabilityofanidentificationisassessedbyconsideringpositivepredictive
value.Positivepredictivevalueistheprobabilitythatanidentifiedindividualisactually
theperpetrator.Themosteffectivewaytomeasurepositivepredictivevalueistotake
confidenceatthetimeofidentificationintoaccount.Highlyconfidentwitnessestestifying
20
inacourtoflawarepersuasivetojurors(Penrod&Cutler,1995),however,this
confidenceisnotnecessarilydiagnosticofaccuracy.Therearemanypowerfulforcesthat
canadverselyinfluenceconfidence,suchasprovidingfeedbackimmediatelyafteran
identification(e.g.,Bradfield,Wells,&Olson,2002;Wells&Bradfield,1998),providing
misinformation(e.g.,Loftus,2005),andmakingsuggestivestatements(e.g.,Zaragoza&
Lane,1994;Lindsay&Johnson,1989;Loftus&Palmer,1974).Thus,confidenceexpressed
attrialisanalogoustoconsideringevidencefromacontaminatedcrimescene.However,
confidenceexpressedataninitialidentificationattempt(assumingthatthelineupandits
administrationarefair)isdiagnosticofaccuracy(Wixted,Mickes,Clark,Gronlund,&
Roediger,2015).However,socialmediaplatforms(e.g.,Facebook,Instagram)have
presentednewwaysforeyewitnessestotrytofindtheperpetrator,whichmay
contaminateasubsequentidentificationprocedureadministeredbythepolice(whichis
assumedtobetheinitialidentificationattempt).Consequently,thepoliceshouldinstruct
witnessesnottosearchforfacesonsocialmedia,justastheyshouldinstructwitnesses
nottospeaktooneanother(TechnicalWorkingGroupforEyewitnessEvidence,1999;see
Paterson,Kemp,&Ng,2011),andinquireifawitnessdidsearchforfacespriorto
administratingthelineuptest.Toreiterate,itisinitialconfidenceadministeredbythe
policethatwerefertothroughoutthischapter.Moreover,wefocusontheaccuracy
associatedwithhighconfidenceIDsbecausethesurprisinglyhighlevelsofaccuracy
associatedwiththesehighconfidenceIDs,andthecapabilityofeyewitnessesto
compensateforvariablesthatadverselyaffectdiscriminability,iscentraltoourclaimthat
eyewitnessevidencecanbefarmoretrustworthythanistypicallythought.
21
2.3.1 Pastmeasureoftheconfidence-accuracyrelationship:Thepointbiserialcorrelationcoefficient
Theconfidencethataneyewitnessexpressesatinitialidentification,andthe
associatedaccuracy,werelongthoughttobeweaklyrelated.Earlyresearchyieldeda
point-biserialcorrelationcoefficientofonly.07,whichpromptedthefollowingstatement,
“...theeyewitnessconfidence–accuracyrelationisweakundergoodlaboratory
conditionsandfunctionallyuselessinforensicallyrepresentativesettings”(p.165,Wells&
Murray,1984).Adecadelater,thatsentimentwasechoed,“Amajorsourceofjuror
unreliabilityistheirrelianceonwitnessconfidence,whichisaweakindicatorof
eyewitnessaccuracyevenwhenmeasuredatthetimeanIDismadeandunderrelatively
‘pristine’laboratoryconditions”(p.830,Penrod&Cutler,1995).Sporer,Penrod,Readand
Cutler(1995)conductedameta-analysisthatshowedarelativelystrongpointbi-serial
correlationcoefficientof.41whenonlychooserswereincluded.Nonetheless,the
overarchingmessagewasstilltobecautiousaboutconfidence.
Thesemisconceptionsarosebecauseamisleadingstatisticwasusedtoassessthe
relationshipbetweenconfidenceandaccuracy.Juslinetal.(1996)showedthatthepointbiserialcorrelationcoefficientcanmaskthestrengthoftheconfidence-accuracy
relationship,andarguedthatcalibrationanalysisshouldbeconductedinstead.Sincethen,
Brewerandcolleagueshaveshowntimeandtimeagainthattherelationshipistypically
strong,whenusingthemoreappropriatecalibrationanalysis(e.g.,Brewer&Wells,2006;
Palmer,Brewer,Weber,&Nagesh,2013;Sauer,Brewer,Zweck,&Weber,2010),and
whenthefocusisononlychoosers(i.e.,thosewhomadeanidentification)ataninitial
identificationattempt.Below,wediscussthreewaystomeasurepositivepredictivevalue
usingconfidence.
22
2.3.2 Calibrationanalysis
Calibrationanalysismeasurestherelationshipbetweenthesubjectiveprobability
thatanidentificationiscorrectandtheobjectiveprobabilitythattheidentificationis
correct.Therearedifferentapproachestocomputingcalibration,andoneapproach
involvesincludingthefillerpicks.Usingthatapproach,tocomputethedependent
variable,thecorrectsuspectIDsaredividedbythesumofthecorrectIDsandincorrect
IDs(includingthefillerIDs)foreverylevelofconfidence.Thatisthenplottedasaccuracy
asafunctionofconfidence.ThetoppanelofFigure7showsthecalibrationplotof
hypotheticaldata.
2.3.3 Confidence-accuracycalibration(CAC)analysis
Acloserelativeofcalibrationanalysisthatonlyconsiderssuspectidentificationsis
confidence-accuracycalibration(CAC)analysis(Mickes,2015).TocomputetheCAC
dependentvariable,thecorrectsuspectIDsaredividedbythesumofthecorrectsuspect
IDsandincorrectsuspectIDs(fillersareexcluded).OneadvantageCACanalysishasover
calibrationanalysisisthatthescaledoesnotneedtobefrom0-100%,andinstead
confidencecanbeassessedusinganytypeofordinalscale.Moreimportantly,CAC
analysisspecificallyaddressesthequestionthatjudgesandjurorshave(unlike
calibration):Whatisthelikelihoodthatthissuspect(becauseonlysuspectIDs,notfiller
IDs,areadvancedtotrial)isguilty?ThemiddlepanelofFigure7showstheCACplotofthe
samehypotheticaldataasinthetoppanel.NotethatchanceaccuracyinaCACplotis50%
ifanequalnumberoftarget-presentandtarget-absentlineupsareused.
<Figure7nearhere>
Anotherwaytomeasurepositivepredictivevalueistocomputethediagnosticity
ratioforeachlevelofconfidence.Asalreadystated,thediagnosticityratioisnota
23
measureofdiscriminability,butitcanbeusedtomeasurereliability.Thebottompanelof
Figure7showsthediagnosticityratiosbyconfidenceforthehypotheticaldatausedinthe
panelsabove.InallthreepanelsofFigure7,confidenceisassociatedwithhigherlevelsof
accuracyforProcedureAthanProcedureB.Becausejudgesandjurorsarethe
stakeholdersofthisinformation,itmightbeeasiertopresentCACanalysistothem
becauseitismoreunderstandablethanadiagnosticityratio.Forexample,sayingthathigh
confidenceidentificationsareassociatedwith99%accuracyismoreunderstandablethan
sayingthediagnosticityratiowas103.
Wenowhavelaidouthowtomeasurediscriminability(ROCanalysis,ord’ifROC
analyseshavenotyetbeenconducted)andhowtoassessreliabilitybymeasuringthe
relationshipbetweenconfidenceandaccuracy(CACanalyses).Itistimetoturnour
attentiontopriorresearchinvolvingthemanyvariablesthataffecteyewitnessmemory.
Weorganizethisdiscussionbyconsideringwhethertheprimaryimpactofthesevariables
takesplaceduringtheencoding,storage,orretrievalphasesofmemory.
3
VariablesthatAffectEncoding
3.1 WeaponFocus
Weaponfocusreferstoareductionindiscriminabilityfortheperpetratorandthe
detailssurroundingacrimethatinvolvedaweapon(Loftus,1979;Loftus,Loftus,&Messo,
1987).Theweaponfocuseffectisapopularphenomenontoinvestigate.Asearchofthe
term“weaponfocus”yields1,530resultsinGoogleScholar(March15,2016).Theeffect
likelyoccursduringencodingbecauseattentionisdirectedtotheweaponratherthanon
theotheraspectsofthecrimeortheperpetrator.Twoprominenttheories,bothproposed
intheoriginalempiricalweaponfocusstudies,attempttoaccountfortheeffectand
continuetodominatetheliterature.Theseaccountspositthatweaponfocusisdueeither
24
tohigharousalortothepresenceofanunusualobject(Loftusetal.).Loftusetal.,
creditingEasterbrook(1959),proposedthatthepresenceofaweaponcauseshigharousal
thatmightconsequentlynarrowaneyewitness’attentionalfocus.Theyalsoproposedthat
loweraccuracymightbeduetoatendencytofixateonanunusualobject.Thetwo
possibilitiesarecommonlyreferredtoasthearousal/threathypothesisandtheunusual
itemhypothesis,respectively.Sincethen,investigationshavetestedthepredictionsof
bothhypotheses,withsomesupportforeachaccount.
Someofthestudiestestrecallofevents,and/ortestrecognitionofthe
perpetratorfromalineup.Inmanyofthelatterstudies,memorywastestedonlyon
target-present(e.g.,Shaw&Slotnick,2001;Pickel,1998)ortarget-absentlineups(e.g.,
Hulse&Memon,2006;Maass&Kohnken,1989),whichisproblematicformeasuring
discriminability(e.g.,Mickes&Wixted,2015;seeRotello,Heit,&Dube,2015).The
conclusionofameta-analysisofweaponfocuseffectstudieswasthattheeffectwas
dependentonthelevelofthreat(thelargerthethreat,thebiggertheeffect)and
retentionintervals(thelongertheretentioninterval,thesmallertheeffect).
ThefirstweaponfocusexperimentinwhichROCanalysiswasconductedwas
recentlypublished(Carlson&Carlson,2014).Inthisexperiment,participantswatcheda
videoofacrimeinwhichtheperpetrator,whohadadistinctivefeatureornot,hada
weaponornot,andthenweretestedonasequentialorsimultaneouslineup.
Discriminabilitywaslowerintheweaponpresentconditionthanintheweaponabsent
conditionwhenthedistinctivefeaturewasnotpresent.Thus,usingROCanalysis,the
weaponfocuseffectwasfound.
Thenextquestionis:whatistheeffectofweaponfocusonthereliabilityofanID
(therelationshipbetweenconfidenceandaccuracyasassessedbyCACanalysis)?
25
Regardlessofwhetheraweaponcauseslowerdiscriminability,whatmattersinthe
appliedsenseare,caneyewitnessestakeafactor(likeweaponpresence)intoaccount
whenreportingtheirlevelofconfidence(i.e.,theCACresults).Thesearetheresultsthat
caninformjudgesandjurorswhendecidingaboutdefendants’culpability.Ifaccuracyfor
identificationsmadewithhighconfidenceisthesameregardlessofthepresenceor
absenceofaweaponatencoding,thenthatisamorerelevantconsiderationthanif
memoryisoverallworseifaweaponispresentduringthecrime.WelimitedtheCAC
analysistotheweaponandnoweaponconditionsfromCarlsonandCarlson(2014)(and
excludedtheconditioninwhichanartificialfeaturewasaddedtothetarget).OurCAC
analysisrevealedthathigh-confidenceaccuracywasveryhighwhetheraweaponwas
presentornot(e.g.,averageaccuracy,collapsedacrosscondition,was97%).
Inamorerecentinvestigationofthereliabilityoftheweaponfocuseffect,
participantswereassignedtooneofthreeconditions:weaponpresent,weaponpresent
butconcealed,andweaponabsent(Carlson,Dias,Weatherford,&Carlson,2016).There
wasalargediscriminabilitydifferencebetweentheweaponpresentandweaponabsent
andconcealedconditions,butthestorywasdifferentforreliability(andreplicatedour
CACresultsoftheirearlierstudy).CACanalysisrevealednostatisticaldifferencesamong
conditionsfortheidentificationsmadewithmediumandhighlevelsofconfidence.Ifthis
resultcontinuestoreplicate,thesearetheresultsthatjudgesandjurorsneedtoknow:
Notthatthepresenceofaweaponmayyieldlowerdiscriminability,becausethereisno
controloverwhetheraweaponwaswieldedornotduringacrime,butthefactthatatthe
highendoftheconfidencescale,thelikelihoodthatthesuspectisguilty(thereliability)is
similar,andhigh,irrespectiveofthepresenceofaweapon.
26
3.2 ExposureDurationandDividedAttention
Thesamedistinctionbetweendiscriminabilityandreliabilityshouldbemadewhen
itcomestoothervariablesthataffecteyewitnessidentification,suchasexposure
duration,dividedattention,etc.Noonewoulddisputethatlongerexposuretimesto
stimuli,includingaperpetrator,shouldgenerallyresultingreaterdiscriminabilitythan
shorterexposuretimes.Norwouldanyonedisputethatgreaterattentionpaidduring
encodingshouldgenerallyleadtogreaterdiscriminability.Palmeretal.(2013)conducted
experimentsinwhichtheymanipulatedtwovariablesthataffectencoding:exposure
duration(5secondsvs.90seconds;Experiment1),attention(fullvs.divided;Experiment
2),andonevariablethataffectsretrieval:retentioninterval(immediatelytestedvs.tested
afteradelay;Experiment1).InExperiment1,oneexperimenterapproachedpotential
participantsinpublicplacesandaskediftheywouldtakepartinanexperiment.Ifthey
agreed,asecondexperimenterwouldstepintoviewforeither5secondsor90seconds;
theparticipantsweretestedimmediatelyor6-8dayslater.Collapsingacrosstheretention
intervalconditions,ROCanalysisshowedthatthe90-secondencodingconditionyielded
greaterdiscriminabilitythanthe5-secondcondition.Whatabouttheconfidence-accuracy
relationship?Palmeretal.providedthedatatoconductCACanalysissothatwecould
answerthequestion:Didthoseinthe5-secondconditionappreciatethefactthatthey
onlysawthetargetbrieflyandadjusttheirconfidenceaccordingly?Yes.Identifications
madewithhighconfidencehadthesameaccuracyregardlessofexposureduration(the
CACresultsoftheirexperimentarereportedinMickes,2015).
InExperiment2,participantswatchedavideooftwotargetindividuals.
Participantsinthedividedattentionconditionhadtorespondtolowandhightoneswhile
watchingthevideo.Thoseinthecontrolconditionweretoldtoignorethetoneswhile
27
watchingthevideo.Asexpected,discriminabilitywaslowerinthedividedattention
conditioncomparedtothefullattentioncondition.Importantly,similartotheexposure
durationfindings,participantswereabletocalibratetheirconfidencetoreflectthefact
thattheyweremorelikelytomakeanerrorinthedividedattentioncondition.
Identificationsmadewithhighconfidencewerehighlyaccurateforbothconditions(over
96%).Onceagain,participantscanadjusttheirconfidencetoreflectthelikelihoodthey
aremakinganerrorwiththesevariablesthataffectmemoryatencoding.
3.3 Cross-RaceBias
Thecross-racebiasisaphenomenoninwhichpeoplearemoreaccurateat
recognizingfacesofindividualsofthesameracethanindividualsofanotherrace(Chance
&Goldstein,1981;Malpass&Kravitz,1969;Meissner&Brigham,2001).Thelower
discriminability(i.e.,abilitytodiscriminateoldfromnewfaces)forotherracefacesis
generallyindicatedbyanincreasedfalsealarmrateandoccasionallybyareducedhitrate
(Meissneretal.,2001).Participantsalsogenerallyexhibitamoreliberalresponsebiasfor
cross-racefaces,whichmeansthatparticipantsaremorelikelytoendorsecross-race
facesaspreviouslyseen,irrespectiveofwhethertheyactuallywere.Thecross-racebias
hasbeenfoundacrossavarietyofconditionsandacrossavarietyofraces(e.g.,Ng&
Lindsay,1994;Platz&Hosch,1988;Wright,Boyd,&Tredoux,2001).
Thecross-racebiasisthoughttoariseduringencoding.Bornstein,Laub,Meissner,
andSusa(2013)wereunabletoreducethedeficitbydeliveringcautionaryinstructionsat
retrieval.Golby,Gabrieli,Chiao,andEberhardt(2001)usedfunctionalMRIandfound
greaterfusiformfaceareaactivation,thefirststageofface-specificprocessing,forsame
racefaces.However,JohnsonandFrederickson(2005)foundthattheinductionofpositive
28
emotions,eitherpriortoorafterencodingthesameraceandcrossracefaces,eliminated
thecross-racedeficit(byraisingd'forthecross-racefaces).
Attemptsatexplainingthecross-racerecognitiondeficithavetakentwogeneral
approaches.Oneapproachattributesthedeficittodifferentialperceptualexperience
(e.g.,differencesintheamountorqualityofcontact,Brigham&Malpass,1985;Malpass
&Kravitz,1969),whichhasresultedinawide-rangeofsuggestionsforhowdifferential
perceptualexperiencecouldbemanifested(e.g.,Freeman,Pauker,&Sanchez,2016).For
example,DeGutis,Mercado,Wilmer,andRosenblatt(2013)foundthatown-racefaces
wereprocessedmoreholisticallythancross-racefaces,andlinkedthisincreaseddegreeof
holisticprocessingtotheownracerecognitionadvantage.Facespaceexplanations(e.g.,
Valentine,1991)documentedadenser(i.e.,moreconfusable)psychological
representationofcrossracefaces,whichwouldariseifparticipantsfocusonthewrong
featuresfordiscriminatingcrossracefaces(Papesh&Goldinger,2010).Goldinger,He,
andPapesh(2009)documentedthatparticipantsmadefewerandlongereyefixationsto
cross-racefaces,andalsoreportedthatsomeparticipantsexertedlesscognitiveeffort(as
indexedbypupildiameter)tocrossracefaces.
Thesecondapproachattributesthecross-racedeficittodifferentialsocial
categorization.Bernstein,Young,andHugenberg(2007)demonstratedthepowerofsocial
categorizationtodifferentiallyinfluencememoryforin-group(same-race)versusoutgroup(cross-race)faces.WhiteparticipantsviewedaseriesofphotosofWhitefaceson
redandgreenbackgrounds.Thoseparticipantsweretoldthatthephotosonared
backgroundwenttotheiruniversity(in-group)andthoseonagreenbackgroundwentto
theirrivaluniversity(out-group).Participantsshowedgreaterd'forthein-groupphotos
29
(d'=1.23vs.0.94).Acontrolgroupexhibitednomemorydifferenceforthephotos(d'=
1.08and1.14,forredandgreenbackgrounds).Consequently,itappearsthatthecrossraceeffectcanbeconsideredacross-categoryoranout-groupeffect.
Themostcomprehensivecurrenttheory,theCategorization-IndividuationModel
(Hugenberg,Young,Bernstein,&Sacco,2010),tiesthesetwoapproachestogetherby
proposingthattheabilitytorememberfacesthatbelongtoadifferentcategory/raceisa
functionofthreefactors:socialcategorization,motivatedindividuation,andperceptual
experience.Thecoreofthemodelinvolvesopposingprocessesthatoperateatencoding:
Individuationactstodistinguishcategorymembersfromoneanother(somethingthat
peopledomorenaturallyforown-groupmembers),whereascategorizationclusters
categorymembersalongshareddimensions(e.g.,race,Levin,1996).Accordingtothis
theory,theout-groupbiasarisesfromthetendencytoattendtoidentity-diagnostic
(individualistic)featuresofin-groupindividualsbutcategory-diagnosticfeatures(e.g.,skin
tone)ofout-groupindividuals.Thistheoryalsoprovidesexplanationsregardingfactors
thatcreate,diminish,oreliminate,anout-groupmemorydeficit.
Onefactorthatresultsinout-groupindividualsbecomingmoresalient/memorable
iswhenthoseindividualsarethreatening.Ackermanetal.(2006)foundthatthecrossracebiaswaseliminatediftheotherracefacesdisplayedangryemotionsbecausethreat
inducesindividuation,ratherthancategorization.Likewise,ShriverandHugenberg(2010)
diminishedthemagnitudeofthecross-racedeficitifotherracefaceswereattributed
greaterpowerorstatus(e.g.,aprestigiousoccupationaltitlelikedoctor)becausethreat
andpowerenhancethelikelihoodofindividuation,reducerelianceoncategorization,and
therebyenhancedrecognitionmemory.
30
Hourihan,Benjamin,andLui(2012)foundthatparticipantsalsoareworseat
makingmetamemoryjudgmentstocross-racefacesregardingwhethertheywill
subsequentlyremembertheface.Thishaspotentialimplicationsfortherelianceon
eyewitnessconfidencewhenacross-raceIDisinvolved.Ifeyewitnessesarenotawareof
thisdifficulty,theymaynotbeabletoadjusttheirconfidencejudgmentaccordingly.But
CACanalysesconductedonDodsonandDobolyi(2015)showedthatconfidencewasvery
informativeofaccuracybutracewasminimallyimportant(theaccuracyforhigh
confidencesameraceandcrossraceIDsbothexceed95%).Again,theseparticipantswere
abletoadjustthelikelihoodthattheymadeacorrectIDdespitethefactthattheywere
morelikelytomakeanerrorwhenmakingacrossraceID.Asimilarfindingwasreported
byNguyen,Pezdek,andWixted(inpress)(forexperimentsinwhichperformancewas
abovechancelevels),whore-analyzeddatafromfourcross-racefacerecognition
experiments.
3.4 StressandArousal
Arousalsignifiesgeneralphysiologicalandpsychologicalactivation;stressarises
fromanimbalancebetweenthephysicalandpsychologicaldemandsandtheabilityto
respond(Hoscheidt,LaBar,Ryan,Jacobs,&Nadel,2014).Stressandarousalhavecomplex
effectsonmemory.Forexample,stressappearstoimpairmemoryforneutralinformation
(Payne,Nadel,Allen,Thomas,&Jacobs,2002),butenhancememoryforemotionally
arousingmaterials(Buchanan&Lovallo2001).Theresearchconductedonlineupsfocuses
ondiscriminability;moreresearchneedstobecompletedassessingpositivepredictive
value.Wecannevermakescrimelessstressful,butweneedtodetermineifeyewitnesses
areabletocompensateforthatstresswhentheyassesstheirconfidence.
31
Ameta-analysisbyDeffenbacher,Bornstein,Penrod,andMcGorty(2004)
concludedthathighlevelsofstressadverselyimpactseyewitnessIDsofatarget.Across
27tests,correctIDaccuracywas.42versus.54forhighstressandlowstress,respectively.
Interestingly,theyfoundthatincreasedstressadverselyaffectedtarget-presentaccuracy,
nottarget-absentaccuracy.ValentineandMesout(2008)testedindividualsvisitingthe
LondonDungeonwhowereexposedtoanindividualduringthetourthattheylaterwere
askedtoIDfromatarget-presentlineup.ValentineandMesoutdidnotmanipulatestress
butratherclassifiedparticipantsbasedonamediansplitbasedontheiranxietyscorein
reactiontotheDungeonperformance.Amongthelowanxiousindividuals,75%chosethe
target,butonly18%ofthehighanxiousindividualschosethetarget.Unfortunately,
withoutatarget-absentlineupforcomparison,andgivenlargedifferencesinresponse
biasesacrossthetwogroups(highanxiousmoreconservative),conclusionsaretentative.
Morganetal.(2004)conductedthemostcompellingexperimentexamining
memoryforaperpetrator.Itisthemostcompellingbecauseitinvolvesfargreaterlevels
ofstressthancanethicallybeusedinthelaboratory.Participantsweresoldiers
undergoingprisoner-of-wartraining,whichinvolvedsleepandfooddeprivation.Aftera
weekofclassroomtraining,participantswereconfinedinamockprisonerofwarcamp,
andwhileinisolationweresubjectedtolow-stressandhigh-stressinterrogationsfrom
differentinterrogators.Participantswerethreatenedwithphysicalviolenceinthehighstressinterrogation.IDattemptsweremadeapproximately24hoursaftercompletionof
thetraining.Despiteinteractingwiththeirinterrogatorsformorethan30minutes,
participantshadgreatdifficultysubsequentlyidentifyinganinterrogator(somevieweda
livelineup,othersaphotolineup).Wetookthesimpleaverageoftheliveandphoto
lineupdata,andfoundthatd'wasgreater(andthecriterionmoreliberal)when
32
identifyingthelow-stressinterrogator(d'=2.33,c=0.66)thanthehigh-stress
interrogator(d'=1.36,c=1.20).Morganetal.(2013)usedasimilarsampleofparticipants
undergoingprisoner-of-wartraining.Allparticipantsexperiencedhighstress
interrogationsbutvariedinwhether,andhow,theywereexposedtomisinformation.
Somegroupssubsequentlyvieweda9-persontarget-absentlineup,fromwhich53%of
thecontrolparticipantsmadeafalsepositiveselection.Thoseindividualsinthe
photographicmisinformationcondition(whowereexposedtoafillerphotoandaskeda
seriesofquestionsabouttheinterrogationwhileviewingthisphoto)hadafalseIDrateof
91%.Notsurprisingly,themajorityoftheseselectionsinvolvedselectionofthefiller.
Inarecentinvestigationoftheeffectofstressonlineupperformance,participants
watchedavideoofamockcrimewhileeitherbeingexposedtoastressor(coldpressor)or
not,andlatermemoryforthetargetinthevideowastestedfromalineup(Sauerlandet
al.2016).Cortisollevelsvalidatedtheefficacyofthestressmanipulation,yetneitherROC
analysisnorCACanalysisyieldedadifferencebetweenthegroups.Butinterestingly,
identificationsmadewithhighconfidencewerehighinaccuracy(i.e.,over95%proportion
correct)regardlessoflevelofstress.However,theauthorsacknowledgedthatthesample
sizewassmall,somoreresearchstillneedstobeconductedinthisarea.
IntheHoustonPoliceDepartmentfieldstudy,over300realeyewitnessesto
robberieswhowerefacedwithalineupdecisionmadeconfidencejudgmentsintheir
identifications(Wells,2014).Manyofthesewitnesseswerevictimsandthuslikelyhad
experiencedsomedegreeofstressduringencodingofthecrime.Despitethat,the
identificationsmadewithhighconfidencewereassociatedwithhighaccuracyand
identificationsmadewithlowaccuracywereassociatedwithlowaccuracy(Wixted,
33
Mickes,Dunn,Clark&Wells,2016).Theseresultssuggestthattheseeyewitnesseswere
highlyreliable.
Insum,stressappearstoadverselyimpactdiscriminabilityforthetargetofan
event,bethatamockperpetrator,anactorintheLondonDungeon,oraninterrogator.If
thepolicecouldcontrolthelevelofstressexperiencedbyeyewitnessesatthetimeofa
crime,thentheyshouldensurelowstresstoensurehigherdiscriminabilityduringalater
identification.Butofcoursetheycannot,sothekeyquestionforthelegalsystem
concernstheeffectofstressonreliability.InitialCACresultsandimplicationsofthe
HoustonPoliceDepartmentfieldstudysuggestthatstressdoesnotimpairreliability(i.e.,
stressdoesnotcauseeyewitnessestomistakenlyidentifyinnocentsuspectswithhigh
confidence).Ifconfirmedbyadditionalresearch,thiswouldbeafactthatjudgesand
jurorsshouldbemadeawareof.
4 VariablesthatAffectStorage
4.1 VerbalOvershadowing
Theverbalovershadowingeffectpositsthatmemoryisadverselyaffectedafter
providingaverbaldescriptionofapreviouslypresentedstimulus(e.g.,aface).Schooler
andEngstler-Schooler(1990)coinedthetermbasedonresultsfromseveralexperiments
inwhichparticipantsviewedavideoofamockcrime.Theyfoundthatthosewhoverbally
describedtheperpetratorwerelessabletocorrectlyidentifytheperpetratorfromalater
lineuptestthanthoseinthecontrolcondition.InExperiment1oftheoriginalpaper,
participantsweretestedontarget-presentlineupprocedures,andthecorrectIDratewas
.64inthecontrolconditioncomparedtoonly.37intheverbaldescriptioncondition.This
counterintuitivefindingsparkedmuchinterestandfollow-upresearch:theoriginalpaper
34
hasbeencited763times(GoogleScholarsearchretrievedApril8,2016).Notethatno
conclusioncanbereachedregardingdiscriminabilitywithoutatarget-absentcondition.
Thefollow-upresearchdoesnottellaconsistentstory.Thelackofconsistencywas
theimpetusforameta-analysisconductedbyMeissnerandBrigham(2001).Basedon
resultsfrom29investigationsoftheverbalovershadowingeffecttheauthorsconcluded
thattheeffectwasreal,butsmall.Otherinvestigationswereconductedonlistlearning
experimentsinwhichtheverbalizationconditionsfaredbetterthanthecontrol
conditions.Thechangeinparadigms(i.e.,listlearningvs.forensicallyrelevant
experiments)mightaccountfortheinconsistenciesreportedsincetheoriginalverbal
overshadowingeffectpaper(whichusedaforensicallyrelevantone-trialparadigm).
Therearethreemainhypothesestoexplaintheverbalovershadowingeffect.The
contentaccount(e.g.,Meissneretal.,2001)holdsthattheverbaldescriptioninterferes
withthememoryofthetarget,causingareductionindiscriminability(i.e.,theabilityto
distinguishbetweentheinnocentandguiltysuspect).Thecriterion-shiftaccount(Clare&
Lewandowsky,2004)holdsthatverbalovershadowingreflectsachangeinresponsebias
(i.e.,thelikelihoodtochoosesomeonefromalineup)ratherthanachangein
discriminability.Theprocessingaccountholdsthattheswitchfromvisualtoverbal
processing(Schooler,2002)affectsbothdiscriminabilityandresponsebias(Chin&
Schooler,2008).
Becauseofthelackofconsistentfindings,andbecausetheimportanceof
conductingdirectreplicationshasrecentlybeenhighlightedinthefieldofpsychology
(Pashler&Wagenmakers,2012),Experiment1andExperiment4intheoriginalSchooler
andEngstler-Schooler(1990)paperwerechosenasoneofthefirstpre-registered
replicationreportprojects(OpenScienceCollaboration,2015).Independentlaboratories
35
respondedtothecalltoattempttoreplicatethefinding,andthemeta-analysisfromthat
workcaninformusaboutthetruesizeoftheverbalovershadowingeffect(Alognaetal.,
2014).Theresultsrevealedasmall,butsignificant,verbalovershadowingdeficit,which
waslargerifthedescriptiontaskwasseparatedintimefromtheoriginalevent.
Intheoriginalpaper,participantswereonlytestedontarget-presentlineups.But
togetacompletepictureofhowverbalizationaffectsparticipants’abilitytodiscriminate
innocentfromguiltysuspects,oneneedstoalsomeasurefalseIDrates(whichare
collectedfromtarget-absentlineups).ItiswellestablishedthatwhilethecorrectIDrate
couldbelowerinoneconditionthisdoesnotmeanthatthatconditionisworse,perse,
becausethefalseIDratecouldbelowerinthatconditionaswell.Despitereplicatingthe
originalresult(Alognaetal.,2014),becausetherewasnowaytomeasurefalseIDs,the
effectverbalizationhasonidentificationperformance(discriminability)remainsunclear
(Mickes,2016;Mickes&Wixted,2015;Rotelloetal.,2015).
Recently,twoexperimentsinvestigatingtheeffectsofverbalovershadowingwere
completedthatallowedROCandCACanalyses(Seale-Carlisle&Mickes,2015).Although
theoriginalandreplicationstudiesused8-personlineups,wereduceditto6-person
lineups.Inbothexperiments,thereducedcorrectIDratesreplicated.Intheexperimentin
whichthedescriptionoccurredimmediatelyafterthestudyphase,theROCcurveswere
notsignificantlydifferent(i.e.,discriminabilitywasnotdifferent).CACanalysisrevealed
thatidentificationsmadewithhighconfidencewereequallyreliableaswell.Inthe
experimentinwhichthedescriptionoccurred20minutesafterthestudyphase,and
immediatelybeforethetestphase,theROCcurvesrevealedadifference.Thatis,
discriminabilitywaslowerintheverbalizationcondition.However,CACanalysisrevealed
thathighconfidenceidentificationsstillwereequallyreliable,andbothhighlyaccurate.
36
TheCACresultsarereassuringgiventhatthecriminaljusticesystemreliesonverbal
reportsofcrimes.Whythesedifferencesexistduetothetimingofthedescriptionsshould
bethesubjectoffutureinvestigations(Mickes,2016).
4.2 RetentionInterval
Whatdojudgesandjurorsneedtoknowaboutincreasingretentionintervalthat
theydonotalreadyknow?Nosurprisethatmemorygetsworsewithtime.Butwhat
aboutreliability?Wesuspectthatjudgesandjurorshavethewrongideahere.Tomake
thepointthatjudgesandjurorsneedtoknowaboutonlysuspectidentificationaccuracy,
Wixted,Read,andLindsay(2016)reanalyzeddatafromfourretentionintervalstudies.In
eachofthestudieseithercalibrationanalysisorcorrelationcoefficientswereoriginally
reported.Retentionintervalsvariedandparticipantsweretestedeitherimmediatelyor
oneweekaftertheencodingevent(Juslinetal.,1996;Palmeretal.2013),immediatelyor
threeweeksafter(Saueretal.2010),and3monthsor6-to-9monthslater(Readetal.
1998).Remarkably,theCACresultsshowedthatregardlessofretentioninterval,
identificationsmadewithhighconfidencewerehighlyaccurate,evenaftersixmonths.In
eachcase,discriminability(asmeasuredbyd')decreasedasretentionintervalincreased.
Thus,despiteever-decreasingdiscriminabilitywithtime,highconfidenceidentifications
remainhighlyaccurate.Thisisaconsiderationjudgesandjurorsshouldtakeintoaccount.
4.3 SuggestibilityEffects
Itiswell-knownthatmemoryisreconstructive(e.g.,Bartlett,1932;Roediger&
McDermott,1995),andthatpost-eventsuggestionsareoftenincorporatedandreported
aspartofanoriginalevent(e.g.,Loftus&Palmer,1974).Thisoccurs,inpart,duetopoor
sourcemonitoring(Johnson,Hashtroudi,&Lindsay,1993).Eyewitnessesbecome
confusedbetweentheirownexperiencesandinferencestheydraw(DoIrememberthe
37
tattooordidIinferthatthegangmemberwhorobbedmehadatattoo?),betweenwhat
theyexperienceandwhataco-witnessmightreport(Didyouseethescarontherobber’s
face?),orwhattheymightseeorhearinthemedia.
BonhamandGonzález-Vallejo(2009)examinedtheeffectofmisinformationon
eyewitnessdiscriminabilityandcalibration.Participantswatchedamockcrimevideoand
thenrespondedtoquestionnairesorreadnarrativesthatcontainedcorrectinformation
andmisinformationaboutwhattranspiredinthevideo.Discriminabilitywaspoorerfor
detailsregardingthecrimewhenmisinformationwasintroduced.Likewise,the
confidence-accuracyrelationshipsufferedasafunctionofmisinformation.Inotherwords,
individualsincorrectlymaintainedhighconfidenceintheaccuracyofmemoriesabout
whichtheyhadbeenmisinformed.Unlikefactorslikecross-raceandretentioninterval,
eyewitnessesappeartobeunabletoadjusttheirconfidencetoreflectthereduced
accuracythatarisesfrommisinformation.
5
VariablesthatAffectRetrieval
5.1 CreatingFairLineups
Whatconstitutesafairlineupandhowafairlineupiscreatedhasmuchtodowith
thefillersselectedtobeinthatlineup.Infact,Woltager,Malpass,andMcQuiston(2004)
arguedthatpartofthereasonwhytherearefillersistoprotecttheinnocentsuspectfrom
beingwrongfullyidentified.
5.1.1 FillerSelection
Whatshouldthecharacteristicsofthefillersbe?Shouldthefillersinthelineup
matchtheverbaldescriptionoftheperpetratororshouldtheymatchhowthesuspect
(whomaybeinnocentorguilty)looks?Andhowcloselyshouldthefillersmatch?How
38
shouldperpetratorswithdistinctivecharacteristicsbematched?Thesearequestionsthat
havebeenthefocusofmanyinvestigations,becauseknowingtheanswerstothese
questionscanresultinthepreparationofthefairestlineupandtheresultingbestquality
eyewitnessevidence.
5.1.1.1 Description-matchedorSuspect-matched?
Wells,RydellandSeelau(1993)testedthehypothesisthatthedescriptionmatchedmethodoffillerselectionwouldresultingreaterdiscriminabilitythanthe
suspect-matchedmethodoffillerselection.Indeed,thedescription-matchedcondition
yieldedahigherd'thanthesuspect-matchedcondition(1.61vs.0.37,respectively).They
concluded,“Agoodlineupappearstobeoneinwhichalllineupmembersmatchthe
eyewitness’spre-lineupdescriptionoftheculpritbutotherwisedonotresembleeach
other.”(p.844).Althoughfollow-upresearchhasreplicatedthatfinding(e.g.,Juslinetal.
1996,Tunnicliff&Clark,2000),otherfindingssupportgreaterdiscriminabilityforthe
suspectmatchedconditions(e.g.,Lindsay,Martin&Webber,1994;Tunnicliff&Clark,
2000;Darling,Valentine,&Memon,2007).Clark,Rush,andMoreland(2013)foundno
evidencethatdescription-matchedfillerselectionisbetter,andrecommendeda
combinationofdescription-andsuspect-matchedselection,whichiswhatmanypolice
departmentsappeartodo(Wise,Safer,&Maro,2011).Moreresearchisneeded
regardinghowbesttoselectfillersforalineup,andhowsimilarthosefillersshouldbetoa
suspect.
5.1.1.2 DistinctiveFeatures
Distinctivefeatures,suchasafacetattooorscars,maybeencodedduringthe
crime,buthowthosefeaturesaresubsequentlyhandledisanissueforlawenforcement.
39
Thereareseveralpropositionsregardingdistinctivefeatureswhenassemblingalineup:to
conceal,duplicateacrossmembers,blockover,ortoleavealone.Currently,guidelines
leaveittothediscretionoftheidentificationofficerwhethertheyreplicateorconcealthe
feature(PACECodeD;TechnicalWorkingGroupforEyewitnessEvidence,1999).
Inarecentcomparisonofreplication,concealment,blockingorleavealone,data
werecollectedfrom9841participants(Colloff,Wade,&Strange,inpress).Participants
viewedvideoofacrime(oneoffourvideoswithfourdifferenttargets)inwhichtherewas
aprominentdistinctivefeature(eachtargethadadifferentdistinctivefeature).ROC
analysisrevealedthatdiscriminabilitywassimilarforthereplication,concealmentand
blockingconditions,butsignificantlylowerfortheleavealonecondition.Thisfinding
impliesthatwhenconstructingthelineup,ignoringthesuspect’sdistinctivefeatureisbad
practiceandanyotheralternativewouldbepreferred.
Theotheralternativesalsoleadtobetterreliability.Atalllevelsofconfidence,
accuracywassignificantlylowerintheleavealonecondition.Moreover,athighlevelsof
confidence(i.e.,identificationsmadewith90-100%confidence),accuracyintheleave
aloneconditionwasonlyapproximately60%,whereasintheotherconditions,accuracy
wasapproximately85%.Thisfindingimpliesthatjudgesandjurorsneedtoknowhowthe
lineupwasconstructed.Forexample,ifthedefendantwasidentifiedwithhighconfidence
fromalineupinwhichtherewasasuspectwithadistinctivefeaturethatwasnot
accountedforbyconcealing,replicatingorblocking,thenevenanidentificationmade
withhighconfidenceislesstrustworthy.
40
5.2 IdentificationProcedure/Presentation
5.2.1 Simultaneousvs.sequentiallineups
Howshouldlineupmembersbepresentedtotheeyewitness?Inasimultaneous
lineup,alllineupmembersarepresentedatonceandonlyonedecision(Whichperson,if
anyone,istheperpetrator?).Lineupmembersareviewedoneatatimeinasequential
lineup,andadecisionmayberequiredregardinglineupmember1(Isthisthe
perpetrator?)beforelineupmember2ispresented(althoughthereareseveralvariations
onhowthesequentialprocedureisconducted).Inthefirstinvestigationcomparing
simultaneousversussequentialphotolineups(Lindsay&Wells,1985),thefalseIDrate
wasmuchhigherinsimultaneouslineupsthansequentiallineups(.43vs..17,
respectively).ThecorrectIDratewasnotmuchhigherinsimultaneouslineupsthan
sequentiallineups(.58vs..50,respectively).Overtime,withtheexceptionoftwo
experiments(Lindsay,Lea,&Fulford,1991a;Lindsayetal.,1991b),thedifferencesinfalse
IDrateswerenotashighasintheoriginalstudy,andthesequentialadvantagesometimes
failedtoreplicate(e.g.,Carlson,Gronlund,&Clark,2008;Gronlund,Carlson,Dailey,&
Goodsell,2009).Othersconcludedthatthepatternofresultswasmoreconsistentwith
thesequentiallineupinducingaconservativecriterionshift(e.g.,Ebbeson&Flowe,2002;
Meissner,Tredoux,Parker,&MacLin,2005;Palmer&Brewer,2012).Despitethiscontrary
evidence,theclaimmadewasthatthesequentiallineupwassuperior(Steblay,Dysart,
Fulero,&Lindsay,2001;Steblay,Dysart,&Wells,2011).Theclaimswerebasedonhigher
diagnosticityratiosforthesequentiallineups,andbecauseoftheseclaims,sequential
lineupswererecommendedforuseoversimultaneouslineups(e.g.,Lindsay,1999;
InnocenceProject,2009;Wellsetal.,2000).Asaresult,30%ofjurisdictionsacrosstheUS
41
switchedfromusingthesimultaneouslineupstosequentiallineups(PoliceExecutive
ResearchForum,2013).
AftertheimportationofROCanalysistouseforlineupdatain2012,the
simultaneouslineuphasbeenfoundtoconsistentlyoutperformsequentiallineups(e.g.,
Carlson&Carlson,2014;Dobolyi&Dodson,2013;Gronlundetal.2012;Mickes,Flowe&
Wixted,2012).TheUSNationalAcademyofSciencescommitteedecidednottoendorse
oneprocedureovertheotherbecausetheydeemedthematterstillunresolved(National
ResearchCouncil,2014).Nonetheless,theInnocenceProjectceasedtosupportsequential
lineupsasareform(http://www.innocenceproject.org/following-the-science/;retrieved
May15,2016),althoughsomeUSstateInnocenceProjectscontinuetosupportsequential
lineups(e.g.,Minnesota:http://ipmn.org/wp-content/uploads/2016/09/MinnesotaEyewitness-Identification-Standard-Protocols.pdf,retrievedOctober18,2016).Two
recentfieldstudieshavefoundthatsimultaneouslineupsyieldbetterdiscriminability
(Amendola&Wixted,2015;Wixted,Mickes,Dunn,Clark,&W.Wells,2015),whichfurther
bolsterssupportforsimultaneouslineups.Infact,somerecentguidelineshavestatedthat
simultaneouslineupsshouldbeused.Forexample,thePennsylvaniapolicemayfollow
guidelinesthat,“…recommendthatofficerspresentwitnesseswithphotoarrays,showing
multiplephotographsofpotentialsuspects,insteadofonebyone.”
(http://www.mcall.com/news/nationworld/pennsylvania/mc-pa-new-guidelines-false-ids20160412-story.html;retrievedApril12,2016).Thus,itappearsasthoughthetideis
turninginfavorofthesimultaneouslineup,andiftheresultscontinuetoreplicate,then
morepolicymakersshouldbeginrecommendingtheuseofsimultaneouslineups.
Regardlessofthelineupprocedurethatisused,judgesandjurorsneedtoknow
aboutthereliabilityofanidentification.ThisissuehasbeenaddressedbyconductingCAC
42
analysesonexperimentsthatreportedROCanalyses.Ithasbeenshownthat
simultaneousandsequentiallineupsyieldequalreliabilityforidentificationsmadewith
highconfidence(Dobolyi&Dodson,2013;Gronlundetal.2012;Experiments1Aand1Bof
Mickes,Flowe&Wixted,2012;Weber&Brewer,2004).Thus,whilediscriminability
suffersduetosequentiallineuppresentation,reliability(atthehighestconfidencelevel)is
notdifferent.
5.2.2 ShowupProcedures
Unlikealineup,ashowupinvolvesthepresentationofasinglesuspect(not
accompaniedbyfillers)totheeyewitness.Showupshavebeencriticizedfortheir
inherentlysuggestivenaturebecausetheeyewitnessesobviouslyknowthattheperson
presentedisthepolicesuspect(Goodsell,Wetmore,Neuschatz,&Gronlund,2013;
Steblay,Dysart,Fulero,&Lindsay,2003).Despitethis,andotherpotentialdownsides,the
showupwilllikelyremainacommonwayoftestingeyewitnessmemorygiventhatitisan
easywaytotestmemorysoonafteracrimehasoccurred.
Thesimultaneousprocedureyieldsgreaterdiscriminabilitythantheshowup
procedurewhenmeasuredwithROCanalysis(Wetmore,etal.2015;Mickes,2015;
Gronlundetal.,2012).TheCACresultsdifferaswell.Showupproceduresyieldlower
reliabilityevenfortheidentificationsmadewithhighconfidencecomparedto
simultaneouslineups(Wetmore,etal.2015;Mickes,2015).Thus,thereisagrowingbody
ofempiricalevidencesuggestingthatsimultaneouslineupsaresuperiortoshowupsin
termsofbothdiscriminabilityandreliability.Thissuggeststhatlineupsshouldbeused
whenpossibleiftheresultscontinuetoreplicate.
43
5.3 BlindAdministration
Thelineupadministratormayinfluenceaneyewitnessintwoways.1)The
administratorcaninfluencewhoaneyewitnesschooses,orifaneyewitnesschooses.2)
Theadministratorcanprovidefeedbackaboutachoicethataneyewitnessmakes.The
firstsourceofinfluencemapsontodiscriminability(who)andbias(if),respectively.A
discriminabilityinfluenceresultsfromwhatGreathouseandKovera(2009)referredtoas
steering,wherebytheadministratordirectsawitnesstowardaparticularsuspect.
Alternatively,biasisaffectedifanadministratorexertsanon-specificinfluencethat
affectstherateatwhichaneyewitnesschooses.Thereisevidenceintheliteraturefor
boththesepatterns(Clark,Marshall,&Rosenthal,2009;Greathouse&Kovera,2009;Haw
&Fisher,2004).Thesecondsourceofinfluencearisesonceaneyewitnesshasmadea
choice,andtheadministratorprovidespost-IDfeedbackregardingthecorrectnessofthat
choice(e.g.,Wells&Bradfield,1998;Wright&Skagerberg,2007;seemeta-analysisby
BradfieldDouglass&Steblay,2006).Post-IDfeedbacktendstoinflateeyewitness
confidence,aswellasinflateseveralotherindicesreflectinganeyewitness’memoryfor
theperpetrator(e.g.,howlongalooktheygot,howclosetheywere).Butifwefollow
Wixtedetal.’s(2015)prescriptiontofocusonthereportsfromtheinitialidentification,
post-IDfeedbackneednotbeofgreatconcern.Butdouble-blindadministrationremains
importantsothatthereportedconfidenceassociatedwiththefirst,fairtestofan
eyewitness’memory,remainsunspoiled.
Whatdoweknowabouttheeffectsofblindadministrationondiscriminabilityand
responsebias?GreathouseandKovera(2009)variedlineuppresentation,targetpresence,
single-ordouble-blind,andbiasedorunbiasedlineupinstructions.Wefocusonthe
simultaneouslineupdata,andcomputedd'andcfromthereportedcorrectandfalseIDs
44
rates.Ifbiasedinstructions(whichimpliedthesuspectwasinthelineupandthewitness
shouldidentifyhim)wereemployed,discriminabilitywasgreaterwithdouble-blindthan
single-blindadministration(d'=2.19vs.d'=1.52,respectively),anddouble-blind
administrationinducedgreaterresponseconservatism(c=0.74vs.c=-0.32,respectively).
However,forunbiasedinstructions(thesuspectmayormaynotbeinthelineup),
discriminabilitywasslightlygreaterforsingle-blindthandouble-blindadministration(d'=
1.01vs.d'=0.70,respectively),andnodifferencearoseinresponsebias(c=0.58vs.c=
0.53,respectively).Thesedatasuggestthatcombiningdouble-blindcontrolwithunbiased
instructionsmightnotbebeneficial.
TheNationalAcademyofSciencesreport(NationalResearchCouncil,2014)
recommendsdouble-blindlineupadministration,andalonghistoryofresearchin
psychologyandmedicinesupportthemeritsofdouble-blindtesting(Rosenthal,2002).
Butthereislimitedresearchwithintheeyewitnessdomainthatcomparesdouble-to
single-blindlineupadministrationusingtarget-presentandtarget-absentlineups.Clark,
Benjamin,Wixted,Mickes,andGronlund(2015)providearecentreviewofblind
administration,andalsomentiontherelativelackofempiricaldatathatfocusesonthe
crucialindependentvariables(blindvs.non-blind,bothtarget-presentandtarget-absent,
withoutcontaminationfromfactorslikedifferenttypesofinstructions,lineup
presentationmethods).Wealsonotethatmorestudiesneedtobeconductedusing
skilledlineupadministrators(e.g.,Clarketal.,2013).Administratoreffectsareclearly
goingtobeinfluencedbytheskill(orlackthereof)oftheadministratorsinvolved
(Russano,Dickinson,Greathouse,&Kovera,2006).
45
Althoughtheadoptionofdouble-blindlineupsmightseemlikeano-brainer,the
effectsofdouble-blindlineupadministrationarecomplex.Clarketal.(2009)usedlineup
administratorswhowereblindtothepresenceorpositionofasuspect.Thelineup
administratorremainedsilentintheno-influencecondition.Inthesubtle-influence
condition,theadministratormadeinnocuousstatementslike,“Takeyourtime.”Inthe
similarity-influencecondition,theadministratoraskedifanylineupmemberresembled
theperpetrator.Forthedatainvolvingthedesignatedinnocentsuspect,discriminability
wasbetterinthesimilarityinfluence(d'=2.41)thanthenoinfluence(d'=1.67)and
subtleinfluence(d'=0.68)conditions.Clarketal.(2013)trainedadministratorstouse
techniquestoinfluencewitnesses.Theseadministratorsknewthepositionofthesuspect,
althoughthesuspectwasonlyguiltyhalfthetime,unbeknownsttotheadministrator.
Surprisingly,Clarketal.foundslightlyhigherd'intheinfluencecondition,likelybecauseit
iseasiertosteerawitnesstoaguiltysuspectthantoaninnocentsuspect.
Insum,thereisevidencethatblindadministrationcansometimesharm
discriminability,andconversely,thatnon-blindadministrationcansometimesenhance
discriminability.Ofcourse,asClarketal.(2015)argued,thatdoesnotmeanthatnonblindadministrationistoberecommended.Althoughakeygoalofeyewitnessreformsis
todeterminewhichprocedurescanenhancediscriminability,therealsoareissuesof
fairnesstoconsider(e.g.,proceduraljustice,Tyler,2003).Positiveeyewitness
identificationscannotbeconsideredasprovidingindependentevidenceofguiltiftheID
arisesfromthepressuresofalineupadministrator(alsoseeHasel&Kassin,2009).
Unfortunately,littleresearchhasexploredtheeffectsofblindlineupadministrationon
reliability,althoughClarketal.(2013)reportedthatpost-identificationconfidencewas
lowerforthoseparticipantswhowerepushedorsteered.
46
5.4 LineupandJurorInstructions
5.4.1 LineupInstructions
Numerousattemptshavebeenmadetoimproveeyewitnessevidencethroughthe
useofinstructions,withlittlesuccess.Someofthisresearchhasfocusedonthe
instructionsgiventoeyewitnesses,andotherresearchhasfocusedontheinstructions
giventojurorsregardinghowtoweigheyewitnessevidence.Webeginwithbiasedversus
unbiasedinstructionstoeyewitnesses.
5.4.1.1 Biasedvs.UnbiasedInstructions
Biasedinstructionsimplythattheperpetratorisinthelineup,unbiased
instructionsdonot(e.g.,“Theperpetratormayormaynotbepresent”).Malpassand
Devine(1981)conductedthefirststudycomparingbiasedandunbiasedinstructions,and
foundad'advantageof0.75forunbiasedinstructions.Researchthatfollowedwasin
generalagreement(e.g.,Cutler,Penrod,&Martens,1987;O’Rourke,Penrod,Cutler,&
Stuve,1989),whichculminatedinameta-analysisinwhichSteblay(1997)concludedthat
unbiasedinstructionsdecreasedchoosingfromtarget-absentlineupswithoutdecreasing
correctIDsfromtarget-presentlineups.Butmorerecentresearchhasreachedadifferent
conclusion;arecentmeta-analysisbyClark,Moreland,andGronlund(2014)showedthat
thediscriminabilityadvantageforunbiasedinstructionsaggregatedacross23studieswas
non-existent(d'=-.02).Inlightoftheseresults,itappearsthatbiasedversusunbiased
instructionsaffectonlyresponsebias:Eyewitnessaremoreconservativeafterreceiving
unbiasedinstructions.
Despiteunbiasedinstructionsnotenhancingdiscriminability,theNASreport
(NationalResearchCouncil,2014)neverthelessrecommendedthattheybeused:
47
“Witnessesshouldbeinstructedthattheperpetratormayormaynotbeinthephoto
arrayorlineup...”(p.73).Tounderstandtherationaleforthisrecommendation,onemust
considerthetradeoffsbetweenthecostsoferrors(IDaninnocentsuspect,thefailureto
IDaguiltysuspect)versusthebenefitsofcorrectdecisions(IDtheguiltysuspect,rejecta
lineupthatcontainsaninnocentsuspect).Mostpolicymakersagreethatitismore
importanttoprotecttheinnocent(limitIDsofinnocentsuspects)thanitistoimplicate
theguilty.Blackstone(1769,p.352)famouslysaidthatitis“…betterthattenguilty
personsescapethanthatoneinnocentsuffer.”IfthecostofafalseIDis10xgreaterthan
thatofamiss,eyewitnessesshouldsetaconservativecriterion,andunbiasedinstructions
shouldhelpaccomplishthat.However,itisimportanttopointoutthatthechoiceofthese
utilitiesisamatterforsocietyandpolicymakers,notforeyewitnessresearchers.Oncethe
utilitiesareagreedupon,signaldetectiontheoryprovidesthemachineryforconverting
theutilities,giventhebaseratesofguiltyandinnocentsuspectsbeingplacedintolineups,
intoanoptimalcriterionplacement.ForarecentreviewoftheseissuesseeClark,
Benjamin,Wixted,MickesandGronlund(2015).
5.4.1.2
Addinga“don’tknow”option
Anotherinstructionalchangethatcanbeofferedtoeyewitnessesistheoptionof
reportingthatthey“Don’tknow.”WeberandPerfect(2012)conductedastudythat
examinedtheinclusionofa“Don’tknow”option.Participantsviewedamock-crimevideo,
adistractingvideo,andthenatarget-presentortarget-absentshowup.Someparticipants
wererequiredtomakea“yes”or“no”decisionaboutthefaceintheshowup,other
participantswereallowedtochoose“don’tknow”iftheydesired.Collapsedover
retentioninterval(whichwas3minutesor3weeks),thoseparticipantswiththeoptionto
choose“don’tknow”madebetterdiscriminations(d'=.77)thanthoseparticipants
48
withoutthatoption(d'=.28,responsecriterionpositionwasmoreconservative,c=0.75
vs.c=0.47,respectively).Butresearchneedstobeconductedtoexaminetheeffectsof
anexplicitDon’tknowoptiononreliability.
5.4.1.3
AppearanceChangeInstructions
Researchersalsohaveexploredtheimpactofinstructingeyewitnessesthata
perpetratormayhavechangedhisappearancesincecommittingacrime.Charmanand
Wells(2007)hadparticipantsviewavideowithfourculprits;thiswasfollowedbyfour
lineups(2target-presentand2target-absent,unbiasedinstructions).Halftheparticipants
receivedappearancechangeinstructionsandhalfdidnot.Wecomputedd'andcfrom
theirdataforeachculprit,andthentookthesimpleaverageacrossculprits.Wefound
thattheappearancechangeinstructionsmadeparticipantsslightlymoreliberal,and
slightlydecreaseddiscriminability.Molinaro,Arnsdorfer,andCharman(2013)conducted
asimilarstudy,andexplicitlyvariedtheamountofappearancechange.Theyfoundthat
theappearancechangeinstructionharmeddiscriminabilityandmaderespondingmore
liberal,andthattheeffectsweresimilaracrosstargetsthatdifferedintheamountof
appearancechange.Finally,Porter,Moss,andReisberg(2014)hadparticipantsviewa
mockcrime,andusedmoreextensiveappearancechangeinstructions.Experiment2
includedanexplicitappearancechange(additionoffacialhair)forthetarget.Likethe
priorexperiments,d'waslowerfortheappearancechangecondition(especiallyforthe
Whiteculprit),althoughtherewasnoresponsebiaschange.Overall,thedatasuggestthat
theappearancechangeinstructionslightlydecreasesdiscriminability,andtendstomake
participants’responsebiasesmoreliberal.Porteretal.reportedthatparticipants
receivingtheappearancechangeinstructionswerelessconfidentintheirchoices;butan
49
assessmentofthereliabilityofeyewitnessesreceivingthisinstructionawaitsfurther
research.
5.4.2 JurorInstructions
Researchhasalsobeenconductedonhowjuryinstructionsaffecttheevaluation
ofeyewitnessevidence.Telfaire(UnitedStatesv.Telfaire,1972)instructionsdirectjurors
toconsiderfactorsthatcouldimpacttheaccuracyofaneyewitness(e.g.,viewing
conditions),butfailtoexplainhowthesefactorsimpactaccuracy.Greene(1988)revised
theTelfaireinstructions,tosimplifythem,andaddedinformationabouthowvarious
factors(i.e.,theBiggerscriteria)impacteyewitnessaccuracy.Butcomparedtocontrol
participantsthatreceivednocautionaryinstructions,theTelfaireparticipantswereno
betteratdistinguishingbetweengoodandpooreyewitnesses.Therevisedinstructions
did,however,makemockjurorsmoreskepticaloverall.Inareviewofresearchevaluating
thegoalofimprovingtheuseofeyewitnessevidencebyjurors,BornsteinandHamm
(2012,p.53)concluded“...theresearchonmodifyinginstructionsaboutwitness
identificationhasgenerallyfailedtoaccomplishthisgoal”.
NewJersey(2012)adoptedanexpanded,carefullydesigned(Schacter&Loftus,
2013)setofjuryinstructions(see
http://www.judiciary.state.nj.us/pressrel/2012/jury_instruction.pdf).Thegoalofthese
instructionswastoinformjurorsaboutthecurrentscienceofeyewitnessmemory,and
howtousethatknowledgetoassesseyewitnesstestimony.Amongotherthings,the
instructionsinformjurorsthatmemorydoesnotworklikeavideorecording,andhighlight
therisksofmakingmistakenidentifications.Papailiou,Yokum,andRobertson(2015)
assessedtheeffectivenessoftheseinstructions.Mockjurorsvieweda35-minutemurder
trial,whichincludedeitherweakorstrongeyewitnessevidence.One-halfofeachgroup
50
wasadministeredtheNewJerseyinstructions,theotherhalfwasadministeredstandard
instructions.Jurorsvotedtoconvictabout25%oftimegiventhestandardinstructions
versusabout10%ofthetimegiventheenhancedinstructions.Moreimportantly,neither
setofinstructionsaidedmockjurors’abilitytodistinguishbetweentheweakandstrong
eyewitnessevidence(despitecleardifferencesinevidencequality,seeTable1inPapailiou
etal.).
Insum,instructionstoeyewitnessesorjurorsappeartomakebothmore
conservative.IfthegoalistoreducefalseIDsofinnocentsuspects,thisiscommendable.
Butifthegoalof‘better’instructionsistoenhancethereliabilityofeyewitnessevidence,
itwouldappearthatresearcheffortsarebetterdirectedelsewhere.Moreover,if
instructionalvariationsaresimplymovingeyewitnesses’responsecriteriaaround,an
alternativeperspectiveistoconsidertheresponseconfidencereportedbyeyewitnesses.
Accordingtosignaldetectiontheory,changesinresponseconfidenceareadirect
reflectionoftheunderlyingresponsecriteria(Wixtedetal.,2015).Inotherwords,instead
ofconductingmoreresearchontheinstructionsadministeredtoeyewitnesses,more
researchshouldbeconductedonhowbesttoassesseyewitnessconfidence,which,it
appears,hasthepotentialtosignificantlyenhancetheabilitytodistinguishbetweenweak
andstrongeyewitnessevidenceifjurorscanbemadetorelyontheinitial,fairlyassessed,
identificationevidence.
6
Conclusion
Thischapterhascoveredahostofvariablesthataffecteyewitnessmemoryand
introducedanewclassificationsystemforthosevariables.Goingforwardwithresearch
oneyewitnessidentification,itwillbeimportanttoseparatelymeasurediscriminability
51
(withROCanalysis)andreliability(withCACanalysis).Moreover,thereisaneed(seealso
Clark,2008;Clark&Gronlund,2015;Gronlund,Mickes,Wixted&Clark,2015)todevelop
andtestcompetingtheoriesofdiscriminabilityandreliability,becauseeyewitness
identificationresearchgenerallyhasbeenguidedbyverballyspecifiedtheoriesor
intuition,whichmakesitisdifficulttoextractdefinitivepredictions(Bjork,1973;
Lewandowsky,1993)andslowscumulativeprogress.Aformallyspecifiedmodel,onthe
otherhand,forcesatheoreticiantobeexplicitaboutamodel’sassumptions,whichmakes
predictionstransparentandprovidesacheckonreasoningbiases(Hintzman,1991).
Someprogresshasbeenmadedevelopingformallyspecifiedexplanations.Onesuch
theory,WITNESS,wasproposedbyClark(2003).WITNESSisadirect-accessmatching
model(foranoverviewofthistypeofmodel,seeClark&Gronlund,1996)withasignaldetectionfoundation.Themodelparametersarecloselytiedtothecomponentsrelevant
toeyewitnessidentification.Themodelhasbeenusedtoexplorefillerselection,
simultaneousandsequentiallineups(Goodsell,Gronlund,&Carlson,2010),andrelative
andabsolutejudgments(Clark,Erickson,&Breneman,2011;Fife,Perry,&Gronlund,
2014).WixtedandMickes(2014)extendedasignal-detectionmodeltotheeyewitness
domain.Thetheoryproposedadiagnostic-feature-detectionhypothesistoexplainwhy
discriminabilityisgreaterfromsimultaneouslineups.Theideaisthatbyseeingallthe
lineupmembersatonce,eyewitnessescandeterminewhatfeaturestopayattentionto
andwhatfeaturesareredundantacrosslineupmembersandthereforenotdiagnostic.For
example,ifalltheindividualsinthelineupareyoungAfricanAmericanmaleswithshaved
heads,puttingattentionontheseshared(i.e.,non-diagnostic)cueswillnothelpan
eyewitnessattendtodiagnosticcuesthatareuniquetotheperpetrator.
52
Manymayfindsurprisingwhatwehaveargued,thateyewitnessidentificationscan
betrustworthy,inthepropercircumstances.Thewidely-heldalternativeviewthat
eyewitnessesarealwaysunreliable,arosebecauseoftheindisputableevidenceregarding
themalleabilityofmemory,coupledwiththelargenumberofwrongfulconvictionsdueto
faultyeyewitnessIDs.Butmeasurementerrorsinvolvinghowtoassessdiscriminability
(ROCanalysis,notdiagnosticity)andhowtoassesstherelationshipbetweenconfidence
andaccuracy(calibrationorCACanalysis,notthepoint-biserialcorrelation)contributedto
thefactthatresearchersreachedsomeprematureconclusions(forareviewsee
Gronlund,Mickes,Wixted,&Clark,2015).Butthelargesterror,arguably,wasto
recommenddiscountingeyewitnessconfidence.Ourreviewindicatesthateyewitnesses
oftenknowtocalibratetheirconfidencetoreflectthelikelihoodthattheyaremakingan
error,andthatanidentificationmadewithhighconfidenceis,inmostcircumstances
examinedtodate,muchmorelikelytobeanidentificationofaperpetrator.Conversely,
aneyewitnesswhomakesanidentificationwithlowconfidenceislikelyindicatingthatan
identificationmaynotbetrustworthy.Althoughthischapterhasnotemphasizedthelow
confidenceendoftheCACplot,theimplicationsoflowconfidenceIDsalsowarrant
carefulconsideration.AlthoughahighconfidenceIDislikelytosignalaguiltysuspect,a
lowconfidenceIDshouldleadthepolicetoconsiderthattheirsuspectisinnocent.By
ignoringconfidence,thecriminaljusticesystemhasmissedtheopportunitytoutilize
informationthatcanbeofgreatvalue.
ThisismostpainfullyevidentbyconsideringthetypicalprofileofaDNA
exonerationcaseinvolvingeyewitnessmisidentification,wheretheinitialidentifications
oftenweremadewithlowconfidence(atbest)thatlaterinflatedintoahighlyconfident
identification(seeGarrett,2011).Byignoringtheinitiallevelofconfidenceexpressedona
53
first,fair,testofmemory,andinsteadpresentingjurorswithahighlyconfidentwitnessin
court,itisnotsurprisingthatthejurorsinthesecasesreachedtheconclusionthatthe
eyewitnesseswereaccurateintheirmisidentifications.Inthosecases,hadinitial
confidencebeenmadeknown,thosewrongfullyconvictedindividualsmaynothave
suffered,andtheactualperpetratorsofthesecrimescouldhavebeenapprehended
sooner.
54
7
References
Ackerman,J.M.,Shapiro,J.R.,Neuberg,S.L.,Kenrick,D.T.,Becker,D.V.,Griskevicius,V.,
Maner,J.K.,&Schaller,M.(2006).Theyalllookthesametome(unlessthey’re
angry):Fromout-grouphomogeneitytoout-groupheterogeneity.Psychological
Science,17,836-840.
Amendola,K.,L.&Wixted,J.T.(2015)Comparingthediagnosticaccuracyofsuspect
identificationsmadebyactualeyewitnessesfromsimultaneousandsequential
lineupsinarandomizedfieldtrial.JournalofExperimentalCriminology,11,263–
284.
Alogna,V.K.,Attaya,M.K.,Aucoin,P.,Bahnik,S.,Birch,S.,Birt,A.R.,...Zwaan,R.A.
(2014).Registeredreplicationreport:Schooler&Engstler-Schooler(1990).
PerspectivesonPsychologicalScience,9,556–578.
Bartlett,F.C.(1932).Remembering:AStudyinExperimentalandSocialPsychology.
CambridgeUniversityPress.
Behrman,B.W.,&Davey,S.L.(2001).Eyewitnessidentificationinactualcriminalcases:
Anarchivalanalysis.LawandHumanBehavior,25,475-491.
Bernstein,M.J.,Young,S.G.,&Hugenberg,K.(2007).Thecross-categoryeffect:Mere
socialcategorizationissufficienttoelicitanown-groupbiasinfacerecognition.
PsychologicalScience,18,706-712.
Bjork,R.A.(1973).Whymathematicalmodels?AmericanPsychologist,28,426-433.
Blackstone,W.(1769).CommentariesontheLawsofEngland,Vol.II,BookIV.NewYork,
NY:Duyckinck,Long,Collins&Hannay,andCollins&Co.
Bonham,A.J.,González-Valleho,C.(2009).Assessmentofcalibrationforreconstructed
eye-witnessmemories.ActaPsychologica,131,34-52.
55
Bornstein,B.H.,&Hamm,J.A.(2012).Juryinstructiononwitnessidentification.Court
Review,48,48-53.
Bornstein,B.H.,Laub,C.E.,Meissner,C.A.,&Susa,K.J.(2013).Thecross-raceeffect:
Resistanttoinstructions.JournalofCriminology,2013,6pages.
BradfieldDouglass,A.,&Steblay,N.(2006).Memorydistortionineyewitnesses:Ametaanalysisofthepost-identificationfeedbackeffect.AppliedCognitivePsychology,
20,859-869.
Brewer,N.,&Wells,G.L.(2006).Theconfidence-accuracyrelationineyewitness
identification:Effectsoflineupinstructions,foilsimilarity,andtarget-absentbase
rates.JournalofExperimentalPsychology:Applied,12,11-30.
Brigham,J.C.,&Malpass,R.S.(1985).Theroleofexperienceandcontactinthe
recognitionoffacesofown-andother-racepersons.JournalofSocialIssues,41,
139–155.
Buchanan,T.W.,&Lovallo,W.R.(2001).Enhancedmemoryforemotionalmaterial
followingstress-levelcortisoltreatmentinhumans.Psychoneuroendocrinology,26,
307–317.
Carlson,C.A.,Gronlund,S.D.,&Clark,S.E.(2008).Lineupcomposition,suspectposition,
andthesequentiallineupadvantage.JournalofExperimentalPsychology:Applied.
14,118-128.
Chance,J.E.,&Goldstein,A.G.(1981).Depthofprocessinginresponsetoownandother
racefaces.PersonalityandSocialPsychologyBulletin,7,475–480.
Charman,S.D.,&Wells,G.L.(2007).Eyewitnesslineups:Istheappearance-change
instructionagoodidea?LawandHumanBehavior,31,3-22.
56
Chin,J.M.,&Schooler,J.W.(2008).Whydowordshurt?Content,process,andcriterion
shiftaccountsofverbalovershadowing.EuropeanJournalofCognitivePsychology,
20,396–413.
Christianson,S.A.&Hübinette,B.(1993).Handsup!Astudyofwitnesses’emotional
reactionsandmemoriesassociatedwithbankrobberies.AppliedCognitive
Psychology,7,365-379.
Clare,J.,&Lewandowsky,S.(2004).Verbalizingfacialmemory:Criterioneffectsinverbal
overshadowing.JournalofExperimentalPsychology:Learning,Memory,and
Cognition,30,739-755.
Clark,S.E.(2003).Amemoryanddecisionmodelforeyewitnessidentification.Applied
CognitivePsychology,17,629–654.
Clark,S.E.(2008).Theimportance(necessity)ofcomputationalmodellingforeyewitness
identificationresearch.AppliedCognitivePsychology,22,803-813.
Clark,S.(2012).Costsandbenefitsofeyewitnessidentificationreform:Psychological
scienceandpublicpolicy.PerspectivesonPsychologicalScience,7,238–259.
Clark,S.E.,Benjamin,A.S.,Wixted,J.T.,Mickes,L.,&Gronlund,S.D.(2015).Eyewitness
identificationandtheaccuracyofthecriminaljusticesystem.PolicyInsightsfrom
theBehavioralandBrainSciences,2,175-186.
Clark,S.E.,Brower,G.,Rosenthal,R.,Hicks,J.M.,&Moreland,M.B.(2013).Lineup
administratorinfluencesoneyewitnessidentificationandconfidence.Journalof
AppliedResearchinMemoryandCognition,2,158-165.
Clark,S.E.,Erickson,M.A.,&Breneman,J.(2011).Probativevalueofabsoluteand
relativejudgmentsineyewitnessidentification.LawandHumanBehavior,35,364380.
57
Clark,S.E.,&Gronlund,S.D.(1996).Globalmatchingmodelsofrecognitionmemory:
Howthemodelsmatchthedata.PsychonomicBulletinandReview,3,37-60.
Clark,S.E.,&Gronlund,S.D.(2015).Mathematicalmodelingshowsthatcompelling
storiesdonotmakeforaccuratedescriptionsofdata.J.G.W.Raaijmakers,R.
Goldstone,M.Steyvers,A.Criss,andR.M.Nosofsky(Eds.).CognitiveModelingin
PerceptionandMemory:AFestschriftforRichardM.Shiffrin.PsychologyPress,pp.
245-258.
Clark,S.E.,Marshall,T.E.,&Rosenthal,R.(2009).Lineupadministratorinfluenceson
eyewitnessidentificationdecisions.JournalofExperimentalPsychology:Applied,
15,63-75.
Clark,S.E.,Moreland,M.B.,&Gronlund,S.D.(2014).Evolutionoftheoryanddatain
eyewitnessidentificationreform.PsychonomicBulletin&Review,21,251-267.
Clark,S.E.,Rush,R.A.,&Moreland,M.B.(2013).Constructingthelineup:Law,reform,
theory,anddata.InB.Cutler(Ed.),Reformofeyewitnessidentificationprocedures.
Washington,DC:APAPublications.
Colloff,M.,Wade,K.,&Strange,D.(inpress).Unfairlineupsdon’tjustmakewitnesses
morewillingtochoosethesuspect,theyalsomakethemmorelikelytoconfuse
innocentandguiltysuspects.PsychologicalScience.
Cutler,B.L.,Penrod,S.D.,Martens,T.K.(1987).Improvingthereliabilityofeyewitness
identification:Puttingcontextintocontext.JournalofAppliedPsychology,72,629637.
Darling,S.,Valentine,T.,&Memon,A.(2007).Selectionoflineupfoilsinoperational
contexts.AppliedCognitivePsychology,22,159-169.
58
DeGutis,J.,Mercado,R.J.,Wilmer,J.,&Rosenblatt(2013).Individualdifferencesin
holisticprocessingpredicttheown-raceadvantageinrecognitionmemory.PLOS
One,http://dx.doi.org/10.1371/journal.pone.0058253.
Deffenbacher,K.A.,Bornstein,B.H.,Penrod,S.D.,&McGorty,E.K.(2004).Ametaanalyticreviewoftheeffectsofhighstressoneyewitnessmemory.Lawand
HumanBehavior,28,687-706.
Ebbesen,E.B.,&Flowe,H.D.(2002).Simultaneousv.sequentiallineups:Whatdowe
reallyknow?Retrievedfrom
http://www2.le.ac.uk/departments/psychology/ppl/hf49/SimSeq%20Submit.pdf
Fife,D.,Perry,C.,&Gronlund,S.D.(2014).Revisitingabsoluteandrelativejudgmentsin
thewitnessmodel.PsychonomicBulletin&Review,21,479-487.
Freeman,J.B.,Pauker,K.,&Sanchez,D.T.,(2016).Aperceptualpathwaytobias:
Interracialexposurereducesabruptshiftsinreal-timeraceperceptionthatpredict
mixed-racebias.PsychologicalScience,27,502-517.
Garrett,B.(2011).ConvictingtheInnocent:WhereCriminalProsecutionsGoWrong.
Cambridge,MA:HarvardUniversityPress.
Golby,A.J.,Gabrieli,J.D.E.,Chiao,J.Y.,&Eberhardt,J.L.(2001).Differentialresponsesin
thefusiformregiontosame-raceandother-racefaces.NatureNeuroscience,4,
845-8500.
Goldinger,S.D.,He,Y.,&Papesh,M.H.(2009).Deficitsincross-racefacelearning:
Insightsfromeyemovementsandpupillometry.JournalofExperimental
Psychology:Learning,Memory,andCognition,35,1105-1122.
Goodsell,C.A.,Gronlund,S.D.,&Carlson,C.A.(2010).Exploringthesequentiallineup
advantageusingWITNESS.LawandHumanBehavior,34,445-459.
59
Goodsell,C.A.,Wetmore,S.A.,Neuschatz,J.S.,&Gronlund,S.D.(2013).Showupsvs.
lineups:Areviewoftwoidentificationtechniques.B.Cutler(Ed.),Reformof
eyewitnessidentificationprocedures,APAPublications.
Greathouse,S.M.,&Kovera,M.B.(2009).Instructionbiasandlineuppresentation
moderatetheeffectsofadministratorknowledgeoneyewitnessidentification.
LawandHumanBehavior,33,70–82.
Green,E.(1988).Judge’sinstructiononeyewitnesstestimony:Evaluationandrevision.
JournalofAppliedSocialPsychology,18,353-276.
Gronlund,S.D.,Carlson,C.A.,Dailey,S.B.&Goodsell,C.A.(2009).Robustnessofthe
sequentiallineupadvantage.JournalofExperimentalPsychology:Applied,15,140152.
Gronlund,S.D.,Carlson,C.A.,Neuschatz,J.S.,Goodsell,C.A.,Wetmore,S.,Wooten,A.,
etal.(2012).Showupsversuslineups:AnevaluationusingROCanalysis.Journalof
AppliedResearchinMemoryandCognition,1,221-228.
Gronlund,S.D.,Wixted,J.T.,&Mickes,L.(2014).Evaluatingeyewitnessidentification
proceduresusingROCanalysis.CurrentDirectionsinPsychologicalScience,23,310.
Gronlund,S.D.,Mickes,L.,Wixted,J.T.,&Clark,S.E.(2015).ConductinganEyewitness
Lineup:HowtheResearchGotitWrong.B.H.Ross(Ed.)ThePsychologyof
LearningandMotivation,Volume63,(pp.1-43),AcademicPress,Waltham,MA.
Hasel,L.E.,&Kassin,S.M.(2009).Onthepresumptionofevidentiaryindependence:Can
confessionscorrupteyewitnessidentifications?PsychologicalScience,20,122-126.
Haw,R.M.&Fisher,R.P.(2004).Effectsofadministrator-witnesscontactoneyewitness
identificationaccuracy.JournalofAppliedPsychology,89,1106–1112.
60
Hintzman,D.L.(1991).Whyareformalmodelsusefulinpsychology?InW.E.Hockley,&
S.Lewandowsky(Eds.),Relatingtheoryanddata:Essaysonhumanmemoryin
honorofBennetB.Murdock(pp.39e56)Hillsdale,NJ:Erlbaum. Hoscheidt,S.M.,LaBar,K.S.,Ryan,L.Jacobs,W.J.,&Nadel,L.(2014).Encodingnegative
eventsunderstress:Highsubjectivearousalisrelatedtoaccurateemotional
memorydespitemisinformationexposure.NeurobiologyofLearningandMemory,
112,237-247.
Hourihan,K.L.,Benjamin,A.S.,&Lui,X.(2012).Across-raceeffectinmetamemory:
Predictionsoffacerecognitionaremoreaccurateformembersofourownrace.
JournalofAppliedResearchinMemory&Cognition,1,158-162.
Hugenberg,K.,Young,S.G.,Bernstein,M.J.,&Sacco,D.F.(2010).Thecategorizationindividuationmodel:Anintegrativeaccountoftheother-racerecognitiondeficit.
PsychologicalReview,117,1168-1187.
Hulse,L.M.&Memon,A.(2006).Fatalimpact?Theeffectsofemotionalarousaland
weaponpresenceonpoliceofficers'memoriesforasimulatedcrime.Legaland
CriminologicalPsychology,11,313-325.
Ihlebaek,C.,Love,T.,Eilertsen,D.E.,&Magnussen,S.(2003).Memoryforastaged
criminaleventwitnessedliveandonvideo.Memory,11,319.
InnocenceProject(2009).AsMoreStatesWeighImprovingLineups,NewInnocence
ProjectReportShowsExtentoftheProblemandEffectivenessofReform.
RetrievedMay15,2016http://www.innocenceproject.org/as-more-states-weighimproving-lineups-new-innocence-project-report-shows-extent-of-the-problemand-effectiveness-of-reform/
Johnson,K.J.,&Fredrickson,B.L.(2005).“Wealllookthesametome”Positiveemotions
61
eliminatetheown-racebiasinfacerecognition.PsychologicalScience,16,875881.
Johnson,M.K.,Hashtroudi,S.,&Lindsay,D.S.(1993).Sourcemonitoring.Psychological
Bulletin,114,3-28. Juslin,P.,Olsson,N.,&Winman,A.(1996).Calibrationanddiagnosticityofconfidencein
eyewitnessidentification:Commentsonwhatcanbeinferredfromthelow
confidence-accuracycorrelation.JournalofExperimentalPsychology:Learning,
Memory,andCognition,22,1304-1316.
Kirschbaum,C.,Pirke,K.M.,andHellhammer,D.H.1993.The“TrierSocialStressTest”—A
toolforinvestigatingpsychobiologicalstressresponsesinalaboratorysetting.
Neuropsychobiology28,76–81.
Levin,D.T.(1996).Classifyingfacesbyrace:Thestructureoffacecategories.Journalof
ExperimentalPsychology:Learning,MemoryandCognition,22,1364-1382.
Lewandowsky,S.(1993).Therewardsandhazardsofcomputersimulations.Psychological
Science,4,236-243.
Lindsay,R.C.L.,Lea,J.A.,&Fulford,J.A.(1991a).Sequentiallineuppresentation:
Techniquematters.JournalofAppliedPsychology,76,741-745.
Lindsay,R.C.L.,Lea,J.A.,Nosworthy,G.J.,Fulford,J.A.Hector,J.,LeVan,V.,&Seabrook,
C.(1991b).Biasedlineups:Sequentialpresentationreducestheproblem.Journal
ofAppliedPsychology,76,796-802.
Lindsay,R.C.L.(1999).Applyingappliedresearch:Sellingthesequentialline-up.Applied
CognitivePsychology,13,219-225.
Lindsay,R.C.L.,Martin,R.,&Webber,L.(1994).Defaultvaluesineyewitness
descriptions:Aproblemforthematch-to-descriptionlineupfoilselectionstrategy.
62
LawandHumanBehavior,18,527-541.
Lindsay,D.S.&Johnson,M.K.(1989).Theeyewitnesssuggestibilityeffectandmemory
forsource.Memory&Cognition,17,349-358.
Lindsay,R.C.L.,&Wells,G.L.(1985).Improvingeyewitnessidentificationsfromlineups:
simultaneousversussequentiallineuppresentation.JournalofAppliedPsychology,
70,556-564.
Loftus,E.F.(2005).Plantingmisinformationinthehumanmind:A30-yearinvestigation
ofthemalleabilityofmemory.Learning&Memory,12,361-366.
Lusted,L.B.(1971b).Signaldetectabilityandmedicaldecision-making.Science,171,
1217–1219.
Luus,C.A.E.,&Wells,G.L.(1991).Eyewitnessidentificationandtheselectionof
distractersforlineups.LawandHumanBehavior,15,43–57.
Maass,A.,&Kohnken,G.(1989).Eyewitnessidentification:Simulatingthe"Weapon
Effect".LawandHumanBehavior,13,397-408.
Malpass,R.S.,&Devine,P.G.(1981).Eyewitnessidentification:Lineupinstructionsand
theabsenceoftheoffender.JournalofAppliedPsychology,66,482-489.
Malpass,R.S.,&Kravitz,J.(1969).Recognitionforfacesofownandotherrace.Journalof
PersonalityandSocialPsychology,13,330–334.
Meissner,C.A.,&Brigham,J.C.(2001).Ameta-analysisoftheverbalovershadowing
effectinfaceidentification.AppliedCognitivePsychology,15,603-616.
Meissner,C.A.,Tredoux,C.G.,Parker,J.F.,&MacLin,O.H.(2005).Eyewitnessdecisions
insimultaneousandsequentiallineups:Adualprocesssignaldetectiontheory
analysis.Memory&Cognition,33,783–792.
Mickes,L.(2016).Theeffectsofverbaldescriptionsoneyewitnessmemory:Implications
63
forthereal-world.JournalofAppliedResearchinMemoryandCognition,5,270276.
Mickes,L.,Moreland,M.B.,Clark,S.E.,&Wixted,J.T.(2014).Missingtheinformation
neededtoperformROCanalysis?Thencomputed',notthediagnosticityratio.
JournalofAppliedResearchinMemoryandCognition,3,58-62.
Mickes,L.&Wixted,J.T.(2015).Ontheappliedimplicationsofthe“Verbal
OvershadowingEffect”.PerspectivesonPsychologicalScience,10,400-403.
Molinaro,P.F.,Arndorfer,A.,&Charman,S.D.(2013).Appearance-changeinstructions
effectsoneyewitnesslineupidentificationaccuracyarenotmoderatedbythe
amountofappearancechange.LawandHumanBehavior,37,432-440.
Morgan,C.A.,III,Hazlett,G.,Doran,A.,Garrett,S.,Hoyt,G.,Thomas,P.,etal.(2004).
Accuracyofeyewitnessmemoryforpersonsencounteredduringexposureto
highlyintensestress.InternationalJournalofLawandPsychiatry,27,265–279.
Morgan,C.A.,III,Southwick,S.,Steffian,G.,Hazlett,G.A.,&Loftus,E.F.(2013).
Misinformationcaninfluencememoryforrecentlyexperienced,highlystressful
events.InternationalJournalofLawandPsychiatry,36,11-17.
NationalResearchCouncil(2014)IdentifyingtheCulprit:AssessingEyewitness
Identification.
Ng,W.,&Lindsay,R.C.L.(1994).Cross-racefacialrecognition:Failureofthecontact
hypothesis.JournalofCross-CulturalPsychology,25,217–232.
OpenScienceCollaboration.(2015),Science,349,aac4716.DOI:10.1126/science.aac4716
O’Rourke,T.,Penrod,S.,Curler,B.,&Stuve,T.(1989).Theexternalvalidityofeyewitness
identificationresearch:Generalizingacrosssubjectpopulations.LawandHuman
Behavior,13,385-395.
64
Palmer,M.A.,&Brewer,N.(2012).Sequentiallineuppresentationpromotesless-biased
criterionsettingbutdoesnotimprovediscriminability.LawandHumanBehavior,
36,247–255.
Palmer,M.,Brewer,N.,Weber,N.&Nagesh,A.(2013).Theconfidence-accuracy
relationshipforeyewitnessidentificationdecisions:Effectsofexposureduration,
retentioninterval,anddividedattention.JournalofExperimentalPsychology:
Applied,19,55-71.
Papailiou,A.P.,Yokum,D.V.,&Robertson,C.T.(2015).ThenovelNewJerseyeyewitness
instructioninducesskepticismbutnotsensitivity.PLOSOne,10,
http://dx.doi.org/10.1371/journal.pone.0142695
Papesh,M.H.,&Goldinger,S.D.(2010).Amultidimensionalscalinganalysisofown-and
cross-racefacespaces.Cognition,116,283-288.
Pashler,H.,&Wagenmakers,E.-J.(2012).Editor’sintroductiontothespecialsectionon
replicabilityinpsychologyscience:Acrisisofconfidence?Perspectiveson
PsychologicalScience,7,528-530.doi:10.1177/1745691612465253
Payne,J.D.,Jackson,E.D.,Hoscheidt,S.,Ryan,L.,Jacobs,W.J.,&Nadel,L.(2007).Stress
administeredpriortoencodingimpairsneutralbutenhancesemotionallong-term
episodicmemories.Learning&Memory,14,861-868.
Payne,J.D.,Nadel,L.,Allen,J.J.,Thomas,K.G.,&JacobsW.J.(2002).Theeffectsof
experimentallyinducedstressonfalserecognition.Memory,10,1-6.
Penrod,S.&Cutler,B.(1995).Witnessconfidenceandwitnessaccuracy:Assessingtheir
forensicrelation.Psychology,PublicPolicy,andLaw,4,817-845.
Pickel,K.L.(1998).Unusualnessandthreataspossiblecasuesof"weaponfocus".
Memory,6,277-295.
65
Platz,S.J.,&Hosch,H.M.(1988).Cross-racialethniceyewitnessidentification:Afield
study.JournalofAppliedSocialPsychology,18,972–984.
PoliceExecutiveResearchForum(2013).ANationalSurveyofEyewitnessIdentification
ProceduresinLawEnforcementAgencies.http://www.policeforum.org/.
Porter,D.,Moss,A.,&Reisberg,D.(2014).Theappearance-changeinstructiondoesnot
improvelineupidentificationaccuracy.AppliedCognitivePsychology,28,151-160.
Roediger,H.L.,&McDermott,K.B.(1995).Creatingfalsememories:Rememberingwords
notpresentedinlists.JournalofExperimentalPsychology:Learning,Memory,and
Cognition,21,803-814.
Rosenthal,R.(2002).Covertcommunicationinclassrooms,clinics,courtrooms,and
cubicles.AmericanPsychologist,57,839-849.
Rotello,C.M.,Heit,E.,&Dube,C.(2015).Whenmoredatasteeruswrong:Replications
withthewrongdependentmeasureperpetuateerroneousconclusions.
PsychonomicBulletinandReview,22,944-954.
Russano,M.B.,Dickinson,J.J.,Greathouse,S.M.,&Kovera,M.B.(2006).“Whydon’tyou
takeanotherlookatnumberthree?”Investigatorknowledgeanditseffectson
eyewitnessconfidenceandidentificationdecisions.CardozoPublicLaw,Policy,and
EthicsJournal,4,355-379.
Sauer,J.,Brewer,N.,Zweck,T.,&Weber,N.(2010).Theeffectofretentionintervalonthe
confidence-accuracyrelationshipforeyewitnessidentification.LawandHuman
Behavior,34,337–347.
Sauerland,M.,Raymaeker,L.H.C.,Otgaar,H.,Memon,A.,Waltjen,T.T.,Nivo,M.,
Slegers,C.,Broers,N.J.,&Smeets,T.(2016).Stress,stress-inducedcortisol
responses,andeyewitnessidentificationperformance.BehavioralSciences&the
66
Law,34,580-594.
Schooler,J.W.,&Engstler-Schooler,T.Y.(1990).Verbalovershadowingofvisual
memories:Somethingsarebetterleftunsaid.CognitivePsychology,22,36-71.
doi:10.1016/0010-0285(90)90003-M
Seale-Carlisle,T.M.,&Mickes,L.(2015).Verbaldescriptions(cansometimes)reduce
discriminabilitybuttheconfidence-accuracyrelationshipisstrong.Poster
presentedatAnnualMeetingofthePsychonomicSociety,Chicago,2015.
Schacter,D.L.,&Loftus,E.L.(2013).Memoryandlaw:Whatcancognitiveneuroscience
contribute?NatureNeuroscience,16,119-123.
Shaw,J.I.,&Skolnick,P.(2001).Sexdifference,weaponfocus,andeyewitnessreliability.
TheJournalofSocialPsychology,134,413-420.
Shiver,E.R.,&Kugenberg,K.(2010).Power,individuation,andthecross-racerecognition
deficit.JournalofExperimentalSocialPsychology,46,767-774.
Starr,S.J.,Metz,C.E.,Lusted,L.B.,&Goodenough,D.J.(1975).Visualdetectionand
localizationofradiographicimages.Radiology,116,538-553.
Steblay,N.(1997).Socialinfluenceineyewitnessrecall:Ameta-analyticreviewoflineup
instructioneffects.LawandHumanBehavior,21,283-397.
Steblay,N.,Dysart,J.,Fulero,S.,&Lindsay,R.C.(2001).Eyewitnessaccuracyratesin
sequentialandsimultaneouslineuppresentations:Ameta-analyticcomparison.
Law&HumanBehavior,205,459-473.
Steblay,N.,Dysart,J.,Fulero,S.,&Lindsay,R.C.(2003).Eyewitnessaccuracyratesin
policeshowupandlineuppresentations:Ameta-analyticcomparison.Lawand
HumanBehavior,27,523-540.
67
Swets,J.A.(1988).Measuringtheaccuracyofdiagnosticsystems.Science,240,12851293.
Swets,J.A.,Dawes,R.M.,&Monahan.J.(2000).Psychologicalsciencecanimprove
diagnosticdecisions.PsychologicalScienceinthePublicInterest,1,1-26.
TechnicalWorkingGroupforEyewitnessEvidence.(1999).Eyewitnessevidence:Aguide
forlawenforcement[Booklet].Washington,DC:UnitedStatesDepartmentof
Justice,OfficeofJusticePrograms,NationalInstituteofJustice.
TheChristianScienceMonitor(2016).HowOhiodinershelpedpolicefindmachetewieldingattacker.http://www.csmonitor.com/USA/Justice/2016/0212/How-Ohiodiners-helped-police-find-machete-wielding-attacker.February15,2016.
Tunnicliff,J.L.,&Clark,S.E.(2000).Selectingfoilsforidentificationlineups:Matching
suspectsordescriptions.LawandHumanBehavior,24,231-258.
Tyler,T.R.,(2003).Proceduraljustice,legitimacy,andtheeffectiveruleoflaw.Crimeand
Justice,30,283-357
UnitedStatesvs.TELFAIRE,469F.2d552(D.C.Cir.1972).WashingtonPatternInstructions
–Criminal(1977).St.Paul:WestPublishingCo.
Valentine,T.(1991).Aunifiedaccountoftheeffectsofdistinctiveness,inversionandrace
onfacerecognition.QuarterlyJournalofExperimentalPsychology,43,161–204.
Valentine,T.,&Mesout,J.(2008).EyewitnessidentificationunderstressintheLondon
Dungeon.AppliedCognitivePsychology,23,151-161.
Valentine,T.,Pickering,A.,&Darling,S.(2002).Characteristicsofeyewitness
identificationthatpredicttheoutcomeofreallineups.AppliedCognitive
Psychology,17,969-993.
68
Wagstaff,G.F.,MacVeigh,J.,Scott,L.,Brunas-Wagstaff,J.,&Cole,J.(2003).Can
laboratoryfindingsoneyewitnesstestimonybegeneralizedtotherealworld?An
archivalanalysisofinfluenceofviolence,weaponpresence,andageeyewitness
accuracy.TheJournalofPsychology,137,17-28.
Weber,N.,&Brewer,N.(2004).Confidence-accuracycalibrationinabsoluteandrelative
facerecognitionjudgments.JournalofExperimentalPsychology:Applied,10,156172.
Wells,G.L.,&Bradfield,A.L.(1998)."Good,youidentifiedthesuspect":Feedbackto
eyewitnessesdistortstheirreportsofthewitnessingexperience.JournalofApplied
Psychology,83,360-376.
Weber,N.,&Brewer,N.(2004).Confidence-accuracycalibrationinabsoluteandrelative
facerecognitionjudgments.JournalofExperimentalPsychology:Applied,10,
156–172.
Weber,N.&Perfect.T.J.(2012).Improvingeyewitnessidentificationaccuracyby
screeningoutthosewhosaytheydon’tknow.LawandHumanBehavior,36,2836.
Wells,G.L.(1978).Appliedeyewitness-testimonyresearch:systemvariablesand
estimatorvariables.JournalofPersonalityandSocialPsychology,12,1546-1557.
Wells,G.L.,&Lindsay,R.C.(1980).Onestimatingthediagnosticityofeyewitness
nonidentifications.PsychologicalBulletin,88,776–784.
Wells,G.L.,Malpass,R.S.,Lindsay,R.C.L.,Fisher,R.P.,Turtle,J.W.&Fulero,S.M.
(2000).Fromthelabtothepolicestation:Asuccessfulapplicationofeyewitness
research.AmericanPsychologist,55,581-598.
69
Wells,G.L.,&Murray,D.M.(1984).Eyewitnessconfidence.InG.L.Wells&E.F.Loftus
(Eds.),Eyewitnesstestimony:Psychologicalperspectives(pp.155–170).NewYork,
NY:CambridgeUniversityPress.
Wells,G.L.,Rydell,S.M.,&Seelau,E.P.(1993).Theselectionofdistractorsfor
eyewitnesslineups.JournalofAppliedPsychology,78,835-844.
Wells,G.L.,Smalarz,L.,&Smith,A.M.(2015).ROCanalysisoflineupsdoesnotmeasure
underlyingdiscriminabilityandhaslimitedvalue.JournalofAppliedResearchin
MemoryandCognition,4,324-328.
Wells,G.L.,Smith,A.M.,&Smalarz,L.(2015).ROCanalysisoflineupsobscures
informationthatiscriticalforboththeoreticalunderstandingandapplied
purposes.JournalofAppliedResearchinMemoryandCognition,4,324-328.
Wells,W.(2014)TheHoustonPoliceDepartmenteyewitnessidentificationexperiment:
analysisandresults.Retrievedfrom:
http://www.lemitonline.org/research/projects.html
Wise,R.A.,Safer,M.A.,&Maro,C.M.(2011).WhatUSlawenforcementofficersknow
andbelieveabouteyewitnessfactors,eyewitnessinterviewsandidentification
procedures.AppliedCognitivePsychology,25,488–500.
Wixted,J.T.,&Mickes,L.(2015a).Evaluatingeyewitnessidentificationprocedures:ROC
analysisanditsmisconceptions.JournalofAppliedResearchinMemoryand
Cognition,4,318-323.
Wixted,J.T.,&Mickes,L.(2015b).ROCanalysismeasuresobjectivediscriminabilityfor
anyeyewitnessidentificationprocedure.JournalofAppliedResearchinMemory
andCognition,4,329-334.
70
Wixted,J.T.&Mickes,L.(2012).Thefieldofeyewitnessmemoryshouldabandon
probativevalueandembraceReceiverOperatingCharacteristicanalysis.
PerspectivesonPsychologicalScience,7,275-278.
Wixted,J.T.,Mickes,L.,Clark,S.E.,Gronlund,S.D.&Roediger,H.L.(2015).Confidence
judgmentsareusefulineyewitnessidentifications:Anewperspective.American
Psychologist,70,515-526.
Wixted,J.T.,Mickes,L.,Dunn,J.,Clark,S.E.,&Wells,W.(2015).Relationshipbetween
confidenceandaccuracyforeyewitnessidentificationsmadefromsimultaneous
andsequentialpolicelineups.ProceedingsoftheNationalAcademyofSciences,
113,304-309.
Wixted,J.T.,Read,D.,&Lindsay,D.S.(2016).Theeffectofretentionintervalonthe
eyewitnessidentificationconfidence-accuracyrelationship.JournalofApplied
ResearchinMemoryandCognition,
http://dx.doi.org/10.1016/j.jarmac.2016.04.006
Wogalter,M.S.,Malpass,R.S.,&Mcquiston,D.E.(2004).AnationalsurveyofUSpolice
onpreparationandconductofidentificationlineups.Psychology,Crime&Law,10,
69-82.
Wright,D.B.,Boyd,C.E.,&,Tredoux,C.G.(2001).Afieldstudyofown-racebiasinSouth
AfricaandEngland.Psychology,PublicPolicyandLaw,7,119–133.
Wright,D.B.,&Skagerberg,E.M.(2007)Postidentificationfeedbackaffectsreal
eyewitnesses.PsychologicalScience,18,172-178.
Zaragoza,M.S.,&Lane,S.M.(1994).Sourcesofmisattributionandsuggestibilityof
eyewitnesstestimony.JournalofExperimentalPsychology:Learning,Memory&
Cognition,20,934-945.
71
Table1.PossibleDecisionsandResultingOutcomes
Identified
the Suspect
Identified
a Filler
Did Not
Identify Anyone
Target
Present
Correct ID
(Hit)
Filler ID
Miss
Target
Absent
True State
Eyewitness Decision
False ID
(False
Alarm)
Filler ID
Correct
Rejection
72
Figure1.Variablesthataffecteyewitnessaccuracycategorizedbystageofmemory,and
bywhichstageofthecrime/criminalproceedingseachoccur.Dependingonthepointin
theproceedings,resultsfromdifferentanalysesareneededtoguidedecisions.
Crime
Investigation
Encoding
Storage
Retrieval
Weapon
Focus
Verbal
Overshadowing
Filler
Selection
Cross Race
Bias
Retention
Interval
Lineup
Presentation
Stress
Appearance
Change
Blind
Administration
Distinctiveness
of Perpetrator
Co-witness
Effects
Lineup
Instructions
...
...
...
Trial
Retrieval
CAC
Analysis
ROC
Analysis
Note.ROC=receiveroperatingcharacteristic;CAC=confidence-accuracycharacteristic
73
Figure2.Six-personsimultaneouslineupwithfivefillersandonesuspect.Ifthesuspectis
theperpetrator,thelineupistarget-present;ifthesuspectisinnocentthelineupistargetabsent.
Filler
Filler
Filler
Filler
Suspect
Filler
74
Figure3.HypotheticalROCdatafromtwoprocedures;ProcedureAfallsfurtherfromthe
lineofchanceperformance(dashedline),thereforeProcedureAhasbetter
discriminabilitythanProcedureB.Thesolidlinesrepresentthefitofasimplesignal
detectionmodel(asshowninthetoprightpanelofFigure6).
Target-Present Suspect ID Rate
0.80
Procedure A
Procedure B
Chance Performance
0.60
0.40
0.20
0.00
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
Target-Absent Suspect ID Rate
Procedure B
Procedure A
Target
Absent
Target
Present
Confidence Suspect IDs Filler IDs
No IDs
Suspect IDs
Filler IDs
0.257
0.386
0.143
0.343
Low
0.536
0.214
Medium
0.357
0.200
High
0.157
0.129
0.057
0.214
Low
0.079
0.393
0.103
0.514
Medium
0.036
0.357
0.046
0.443
High
0.007
0.214
0.014
0.286
0.250
0.529
Diagnosticity Ratio
6.818
2.500
pAUC
0.162
0.059
d'
1.504
0.613
No IDs
0.357
0.383
75
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
ROC
LROC
0.2
0.2
0.0
TP Correct Quadrant ID Rate
(LROC Data)
TP Quadrant ID Rate
(ROC Data)
Figure4.(TopPanel)Detection(ROC)anddetection-plus-identification(LROC)fromStarr
etal.(1975).(BottomPanel)Detection(ROC)anddetection-plus-identification(LROC)
fromExperiment1(collapsedacrossconditions)ofPalmeretal.(2013).Figurefrom
WixtedandMickes(2015a,Figure2).
0.0
0.0
0.2
0.4
0.6
0.8
1.0
TA Quadrant ID Rate
TA Suspect ID Rate
0.03
0.05
0.08
0.10
0.13
1.0
0.8
0.8
0.6
0.6
0.4
0.4
ROC
LROC
0.2
0.2
0.0
0.0
0.0
0.2
0.4
0.6
0.8
TA Filler ID Rate
TP Suspect ID Rate
(LROC Data)
TP Suspect + Filler ID Rate
(ROC Data)
0.00
1.0
1.0
76
Figure5.Demonstrationofobjectivediscriminabilityofashowupandalineupprocedure.
ROCcurvesoffair(“Innocentweak”)datafromWetmoreetal.(2015).
Showup
Lineup
Guilty
Innocent
Guilty
Innocent
61%
42%
67%
10%
1.00
Correct ID Rate
0.80
0.60
0.40
0.20
Showup
Lineup
0.00
0.00
0.10
0.20
0.30
0.40
False ID Rate
0.50
0.60
77
Figure6.Simplesignaldetectionbasedmodel.Themeans(μinnocent,μfillers)andstandard
deviationsofthefiller/innocentsuspectdistributionsare0and1,respectively.Themean
(μguilty)andstandarddeviationofthetargetdistributionaredand1,respectively.Ifthe
memorystrengthofthemostfamiliarlineupmemberexceedsthecriterion(c),an
identificationwillbemade.Unlikeinafairlineup,wherefillersandinnocentsuspectsare
drawnfromthesamedistribution(toppanelontheleft),inanunfairlineup,thefillersare
drawnfromadistributionthatdiffersfromtheinnocentsuspectdistribution(bottom
panelontheleft).Inthiscase,μfillers>μinnocentandresultsinalowerROCforunfairlineups
(bottomrightpanel).
Fair Lineup
c
No ID
µinnocent
ID
µguilty
/fillers
Memory Strength
Guilty Suspects
Innocent Suspects/Fillers
Unfair Lineup
c
No ID
1.0
ID
Correct ID Rate
0.8
µfillers µinnocent
µguilty
Memory Strength
Guilty Suspects
Fillers
Innocent Suspects
0.6
0.4
Fair
Unfair
0.2
0.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
False ID Rate
78
Figure7.Threewaystomeasurethepositivepredictivevalueofanidentification:
calibrationanalysis(topfigure),confidence-accuracycharacteristic(CAC)analysis(middle
figure),anddiagnosticityratio(bottomfigure).
1.00
Proportion Correct
0.80
0.60
0.40
0.20
0.00
20
40
60
80
100
Confidence
1.00
Proportion Correct
0.90
0.80
Procedure A
Procedure B
0.70
0.60
0.50
20
40
60
80
100
Confidence
120
Diagnosticity Ratio
100
80
60
40
20
0
20
40
60
80
100
Confidence
79