LINKAGE ANALYSIS OF SERIOUS SEXUAL OFFENCES

(1)

ȱ 8ȱ

CEPOLȱEuropeanȱPoliceȱScienceȱandȱResearchȱBulletinȱ ȱ ȱIssueȱ3ȱȬȱSummerȱȱ2010

ȱ

LASSO:ȱLINKAGEȱANALYSISȱOFȱSERIOUSȱSEXUALȱȱ OFFENCESȱȱ

AȱDECISIONȱSUPPORTȱSYSTEMȱFORȱCRIMEȱANALYSTSȱANDȱ INVESTIGATORSȱȱ

ȱ

By

ȱ

DONȱCASEY

,ȱSGT.ȱMETROPOLITANȱPOLICEȱSERVICE,ȱLONDONȱȱ

&ȱLÔNDONȱSÔUTHȱBÂNKȱU^NIVERSITY^ȱ

ȱ

PHILLIPȱBURRELLȱ

,ȱP^ROFESSORȱ^OFȱI^NTELLIGENTȱD^ECISIONȱS^YSTEMS,ȱ LONDONȱSOUTHȱBANKȱUNIVERSITYȱ

ȱ

Abstractȱ

Oneȱofȱtheȱfirstȱandȱmostȱimportantȱconsiderationsȱwhenȱinvestigatingȱaȱseriousȱsexualȱoffenceȱisȱtoȱseeȱifȱ theȱoffenceȱcanȱbeȱlinkedȱtoȱothers.ȱIfȱaȱlinkȱcanȱbeȱestablishedȱthenȱthereȱisȱaȱveryȱconsiderableȱdividendȱinȱ termsȱofȱnewȱevidenceȱandȱlinesȱofȱenquiryȱtoȱbeȱfollowed.ȱItȱalsoȱraisesȱwhatȱisȱalreadyȱaȱseriousȱincidentȱ toȱaȱhigherȱlevelȱofȱsignificanceȱwithȱaȱcorrespondingȱincreaseȱinȱtheȱresourcesȱallottedȱtoȱinvestigationȱofȱ theȱseriesȱofȱcrimes.ȱComputerisedȱdecisionȱsupportȱsystemsȱwhichȱemployȱtechniquesȱfromȱArtificialȱIntelȬ ligenceȱareȱwidelyȱusedȱinȱbusinessȱandȱfinanceȱtoȱassistȱpractitionersȱinȱarrivingȱatȱjustifiableȱconclusions.ȱ Inȱprincipleȱthisȱisȱnoȱdifferentȱfromȱtheȱactivitiesȱofȱaȱcrimeȱanalystȱorȱinvestigatorȱinȱfindingȱlikelyȱ matchesȱforȱaȱcurrentȱcrimeȱinȱtheȱoverallȱsetȱofȱcrimes.ȱȱ

ȱ

AimsȱofȱtheȱStudyȱ

TheȱaimȱofȱtheȱstudyȱisȱtoȱdevelopȱaȱcomputerisedȱdecisionȱsupportȱsystemȱthatȱcanȱbeȱusedȱbyȱcrimeȱanaȬ lystsȱandȱinvestigatorsȱtoȱsuggestȱlinksȱbetweenȱstrangerȱrapes.ȱItȱisȱintendedȱthatȱtheȱcharacteristicsȱofȱtheȱ crimeȱunderȱinvestigationȱcanȱbeȱenteredȱintoȱanȱeasyȱtoȱuseȱcomputerȱinterfaceȱandȱthatȱtheȱsystemȱwillȱ thenȱbeȱableȱtoȱsearchȱitsȱdatabaseȱofȱexistingȱcrimesȱandȱdisplayȱaȱnumberȱofȱoffencesȱthatȱhaveȱstrongȱ similarities.ȱ

Theȱdesirabilityȱofȱdevelopingȱcomputer–basedȱtoolsȱforȱlinkageȱanalysisȱhasȱbeenȱrecognisedȱbyȱtheȱleadȬ ingȱresearcherȱintoȱlinkingȱseriousȱsexualȱoffences:ȱ

ȱȱ

“TheȱultimateȱgoalȱisȱtoȱcreateȱaȱcomputerȬbasedȱscreeningȱsystemȱthatȱwillȱallowȱroutineȱandȱsystematicȱcomȬ parisonȱofȱseriousȱoffencesȱonȱaȱnationalȱbasis,ȱselectingȱcasesȱonȱtheȱbasisȱofȱtheirȱbehaviouralȱsimilarityȱthatȱareȱapȬ propriateȱforȱmoreȱdetailedȱattentionȱbyȱdetectivesȱorȱcrimeȱanalyst”ȱ

ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ Grubinȱ(2000)ȱ

ȱ

Thisȱviewpointȱisȱacknowledgedȱthroughoutȱtheȱliteratureȱandȱitȱisȱrecognisedȱthatȱtheȱconstructionȱofȱaȱ linkageȱtoolȱisȱtheȱnecessaryȱconditionȱtoȱprogressȱthisȱundertaking.ȱ

ȱ

“TheȱdevelopmentȱandȱtestȱofȱtheoriesȱandȱimplementationȱofȱfindingsȱintoȱcomputerȬbased,ȱ decisionȬsupportȱsystemsȱ…ȱhasȱtoȱbeȱtheȱproperȱbasisȱforȱanyȱprofessionalȱderivationȱofȱinferencesȱ aboutȱoffenders.”ȱȱ

ȱ ȱ ȱ ȱ ȱ Canterȱ(2000)ȱ

ItȱisȱinterestingȱthatȱhereȱCanterȱwidensȱtheȱscopeȱofȱcomputerisedȱsystemsȱtoȱincludeȱtheȱpossibilityȱofȱ inferringȱoffenderȱcharacteristics,ȱtheȱprocessȱknownȱmoreȱwidelyȱasȱȱ‘offenderȱprofiling’ȱandȱtheȱsubjectȱofȱ aȱgreatȱdealȱofȱcrimeȱliteratureȱandȱHollywoodȱoutput.ȱ

(2)

ȱ 9ȱ

Currentlyȱthereȱareȱtwoȱcomputerȱsystemsȱthatȱ dominateȱtheȱareaȱofȱcrimeȱlinkage:ȱViCAPȱ–ȱtheȱ ViolentȱCrimeȱApprehensionȱProgramȱandȱViCLAS,ȱ ViolentȱCrimeȱLinkageȱSystem.ȱViCAPȱisȱtheȱcreaȬ tionȱofȱtheȱFBIȱatȱQuanticoȱandȱhasȱbeenȱinȱexistenceȱ inȱdifferingȱformsȱsinceȱ1985.ȱViCLASȱwasȱdevelȬ opedȱbyȱtheȱRoyalȱCanadianȱMountedȱPoliceȱinȱtheȱ earlyȱ1990’sȱasȱanȱextensionȱtoȱtheȱearlierȱsystem;ȱ TheȱRCMPȱlicenseȱViCLAS,ȱforȱaȱfee,ȱandȱmaintainȱ controlȱoverȱitȱ;ȱitȱisȱusedȱinȱmanyȱEUȱjurisdictionsȱ:ȱ Belgium,ȱCzechȱRepublic,ȱFrance,ȱ Germany,ȱIreȬ land,ȱNetherlandȱandȱtheȱUnitedȱKingdom.ȱViCLASȱ isȱalsoȱusedȱinȱAustralia,ȱNewȱZealand,ȱSwitzerlandȱ andȱsomeȱstatesȱinȱtheȱUS.ȱ

Bothȱtheseȱsystemsȱwereȱdevelopedȱprimarilyȱbyȱ practitioners,ȱpsychologistsȱandȱcriminologistsȱandȱ areȱessentiallyȱrepositoriesȱofȱdataȱwhichȱdependȱ veryȱmuchȱonȱtheȱskill,ȱtrainingȱandȱexperienceȱofȱ theȱuser.ȱTheȱinfluenceȱofȱComputerȱScientistsȱhasȱ beenȱslightȱandȱthereȱhasȱbeenȱnoȱinvolvementȱbyȱ researchersȱinȱA.IȱorȱDecisionȱSupport.ȱAsȱaȱresultȱ noneȱofȱtheȱadvancesȱthatȱhaveȱbeenȱmadeȱinȱtheseȱ areasȱareȱincorporatedȱinȱeitherȱsystemȱandȱtheyȱ remainȱessentiallyȱunchangedȱinȱtheȱlastȱ20ȱ–ȱ25ȱ years.ȱ

Thereȱisȱaȱnotableȱdisparityȱbetweenȱtheȱamountȱ ofȱeffortȱthatȱpoliceȱagenciesȱinvestȱinȱgatheringȱandȱ recordingȱinformationȱthatȱrelatesȱtoȱtheseȱseriousȱ offencesȱandȱtheȱamountȱwhichȱhasȱbeenȱexpendedȱ onȱdevelopingȱtheȱcomputerȱsystemsȱontoȱwhichȱitȱ isȱentered.ȱViCAPȱandȱViCLASȱareȱpassive;ȱtheȱ workȱspentȱinȱfillingȱtheȱdatabaseȱisȱnotȱreciproȬ catedȱbyȱanyȱcorrespondingȱfunctionalityȱinȱtheȱsysȬ tem.ȱ ȱAnȱeffectiveȱcrimeȱlinkageȱdecisionȱsupportȱ systemȱshouldȱandȱcanȱassistȱtheȱuserȱinȱinvestigatȬ ingȱtheȱcrimeȱbyȱusingȱeffectiveȱcomputerȱscienceȱ technologyȱtoȱ recommendȱ answersȱ toȱtheȱquesȬ tionsȱ:ȱȱ‘Whichȱcrimesȱareȱsimilarȱtoȱthis?’ȱ,ȱ‘ȱȱHowȱ strongȱisȱtheȱsimilarity?’ȱandȱ‘Whatȱareȱtheȱfactorsȱ thatȱareȱmostȱsimilarȱandȱmostȱdissimilarȱbetweenȱ thisȱsetȱofȱcrimes’.ȱȱ

ȱ

Methodologyȱ

Fuzzyȱsetȱtheoryȱ(Zadehȱ1965)ȱȱisȱaȱwellȱestablishedȱ approachȱinȱtheȱfieldȱofȱArtificialȱIntelligenceȱthatȱ canȱdealȱwithȱimpreciseȱorȱvagueȱconceptsȱsuchȱasȱ

‘young’,ȱ‘old’,ȱ‘tall’,ȱ‘short’ȱetc.ȱTheseȱdescriptionsȱ areȱdefinedȱasȱ‘fuzzyȱsets’,ȱi.e.ȱtheyȱareȱnotȱspecificaȬ tionsȱwhichȱhaveȱaȱyesȱorȱnoȱanswer.ȱSoȱaȱsuspectȱ describedȱasȱ1.80mȱinȱheightȱdoesȱnotȱhaveȱtoȱbeȱ eitherȱ‘tall’ȱorȱ‘short’ȱbutȱcanȱbeȱaccordedȱaȱdegreeȱ ofȱbothȱqualities;ȱinȱthisȱcaseȱheȱcouldȱbeȱ0.9ȱtallȱandȱ

0.1ȱshort.ȱOrȱaȱpersonȱ35ȱyearsȱoldȱcouldȱbeȱ0.3ȱ

‘young’,ȱ0.6ȱ‘middleȬaged’ȱandȱ0.1ȱ‘old’.ȱȱThisȱtypeȱ ofȱcharacterisationȱsitsȱwellȱwithȱourȱownȱpercepȬ tionsȱofȱwhatȱareȱknownȱasȱlinguisticȱvariablesȱinȱ fuzzyȱsetȱtheoryȱandȱgiveȱaȱricherȱpictureȱofȱwhatȱ weȱseekȱtoȱdescribe.ȱMostȱofȱallȱtheyȱallotȱmeaningȬ fulȱnumbersȱtoȱtheȱtypesȱofȱdescriptionsȱwhichȱweȱ dealȱwithȱinȱdefiningȱcrimes.ȱȱ

Theȱresultȱisȱthatȱaȱcommonȱdescriptionȱofȱaȱcrimeȱ suchȱas:ȱ“AȱveryȱviolentȱattackȱonȱaȱmiddleȬagedȱwomanȱ byȱaȱyoungȱman”ȱcanȱbeȱrepresentedȱbyȱaȱnumberȱofȱ coȬordinatesȱsoȱthatȱtheȱdegreeȱofȱviolence,ȱmiddleȬ ageȱandȱyouthȱcanȱbeȱcomparedȱwithȱotherȱcrimes.ȱ ConsequentlyȱcrimesȱandȱcriminalsȱcanȱbeȱdeȬ scribedȱinȱhighlyȱdescriptiveȱtermsȱandȱproceduresȱ toȱdiscoverȱwhatȱtheȱmostȱsignificantȱdifferentiatingȱ featuresȱare,ȱusingȱmathematicallyȱandȱlogicallyȱ soundȱmethods,ȱcanȱbeȱundertaken.ȱWeȱhaveȱbeenȱ fortunateȱinȱbeingȱsuccessfulȱinȱobtainingȱdataȱonȱ 545ȱseriousȱsexualȱoffencesȱfromȱtheȱSeriousȱCrimesȱ AnalysisȱSectionȱofȱtheȱU.KȱNationalȱPoliceȱImȬ provementȱAgency.ȱWeȱhaveȱexcludedȱthoseȱofȬ fencesȱthatȱdoȱnotȱrelateȱtoȱserialȱrapes,ȱbyȱwhichȱweȱ meanȱaȱsetȱofȱrapesȱcommittedȱbyȱaȱsingleȱindividȬ ual,ȱresultingȱinȱaȱmuchȱnarrowerȱdatasetȱ(nȱ=110,ȱ developmentȱsetȱnȱ=83,ȱtestȱ=ȱ27)ȱ.ȱ

Asȱaȱstartingȱpointȱweȱhaveȱadoptedȱtheȱdimensionsȱ identifiedȱasȱsignificantȱinȱtheȱresearchȱundertakenȱ byȱGrubinȱetȱalȱ(2000)ȱinȱlinkingȱseriousȱsexualȱofȬ fencesȱthroughȱbehaviour:ȱSex,ȱControlȱandȱEscape.ȱ Sexȱcomprisesȱ13ȱvariablesȱrelatingȱtoȱtheȱsexualȱ assault;ȱControlȱhasȱ20ȱvariablesȱthatȱdefineȱtheȱconȬ trollingȱactionsȱthatȱtheȱoffenderȱusesȱtoȱsubdueȱtheȱ victimȱandȱEscapeȱ(ȱ11ȱvariablesȱ)ȱincludesȱthoseȱ actionsȱthatȱtheȱattackerȱadoptsȱtoȱensureȱthatȱheȱ leavesȱaȱminimumȱofȱevidenceȱatȱtheȱsceneȱ,ȱe.g.ȱ bindingȱandȱblindfolding.ȱ

ȱ ȱ

Resultsȱ

TheȱconsequencesȱofȱassigningȱaȱsingleȱsetȱofȱnumȬ bersȱtoȱaȱcrimeȱareȱfarȱreaching.ȱAȱgreatȱnumberȱofȱ techniquesȱcanȱbeȱemployedȱtoȱrepresentȱsimilarityȱ betweenȱcrimesȱandȱalsoȱtoȱlookȱforȱclustersȱofȱ crimes.ȱTheȱfuzzyȱcȬmeansȱalgorithmȱ(ȱBezdekȱ 1981ȱ)ȱlooksȱforȱclustersȱinȱdataȱandȱallowsȱtheȱuserȱ toȱspecifyȱtheȱnumberȱofȱinputȱdimensionsȱandȱoutȬ put.ȱTableȱ1ȱshowsȱtheȱresultsȱwhereȱtheȱthreeȱinputȱ dimensionsȱofȱ‘Sex’,’Control’ȱandȱ‘Escapeȱhaveȱbeenȱ inputȱandȱthreeȱclustersȱspecified.ȱTheȱdoubleȱlinesȱ indicateȱseriesȱboundaries,ȱi.e.ȱcrimesȱcommittedȱbyȱ

(3)

ȱ 10ȱ

theȱsameȱoffender,ȱsoȱitȱcanȱbeȱseenȱthatȱthreeȱofȱtheȱ fourȱcrimesȱinȱtheȱfirstȱseriesȱbelongȱentirelyȱinȱclusȬ terȱ‘C’ȱwhileȱbothȱcrimesȱinȱtheȱlastȱseriesȱhaveȱveryȱ highȱmembershipsȱofȱclusterȱ‘B’ȱ

ȱ ȱ

Overallȱ88%ȱofȱcrimesȱwereȱassignedȱtoȱaȱclusterȱ withȱ>ȱ80%ȱdegreeȱofȱmembershipȱandȱ15ȱofȱtheȱ28ȱ seriesȱwereȱassignedȱtoȱaȱsingleȱclusterȱatȱ80%ȱmemȬ bershipȱorȱmore.ȱ

Thisȱdemonstratesȱaȱfarȱgreaterȱdegreeȱofȱconsistencyȱ withinȱseriesȱthanȱtheȱGrubinȱstudyȱwhichȱisȱtheȱ onlyȱcomparableȱresearchȱinȱthisȱarea.ȱ

ȱ ȱ ȱ ȱ

InȱTableȱ2ȱwe’veȱdescribedȱtheȱaverageȱdistanceȱbetweenȱcrimesȱasȱtheȱvalueȱ‘close’ȱandȱthenȱmeasuredȱtheȱ degreeȱofȱclosenessȱbetweenȱeachȱcrimeȱinȱtheȱdataset.ȱTheȱfirstȱthreeȱseriesȱareȱshownȱcomprisingȱfourteenȱ crimesȱofȱlengthȱfive,ȱthreeȱandȱsixȱoffences;ȱdegreesȱofȱclosenessȱgreaterȱthanȱ0.6ȱareȱinȱbold.ȱ

ȱ

Thereȱisȱaȱstrongȱdegreeȱofȱclosenessȱ(ȱ>ȱ0.6ȱ)ȱbetweenȱȱfourȱcrimesȱinȱtheȱseriesȱ1ȱtoȱ5ȱandȱallȱofȱtheȱcrimesȱinȱ theȱsecondȱseries,ȱ6ȱtoȱ8.ȱTheȱassociationsȱinȱtheȱthirdȱseriesȱareȱ lessȱsuccessful,ȱbutȱusefulȱassociationsȱdoȱexist.ȱForȱinstanceȱ crimeȱ12ȱonlyȱhasȱonlyȱtwoȱstrongȱlinksȱbutȱtheyȱareȱbothȱwithȱ crimeȱinȱtheȱsameȱseries,ȱ11ȱandȱ14.ȱTheȱresultȱisȱtoȱdevelopȱaȱ structuredȱsearchȱstrategyȱforȱanalystsȱandȱinvestigatorsȱfromȱ oneȱcrimeȱtoȱthoseȱotherȱcrimesȱthatȱareȱlikelyȱbeȱlinkedȱtoȱit.ȱȱ ȱ

ȱ ȱ

Conclusionȱ

Theȱneedȱforȱaȱcomputerisedȱdecisionȱsupportȱsystemȱtoȱassistȱ inȱlinkingȱseriousȱcrimesȱhasȱbeenȱidentifiedȱandȱtheȱcurrentȱ systemsȱinȱuseȱshownȱtoȱbeȱinadequate.ȱEstablishedȱtechniquesȱ fromȱArtificialȱIntelligence,ȱinȱparticularȱfuzzyȱsetȱtheory,ȱcanȱ beȱappliedȱtoȱcrimeȱlinkageȱandȱhaveȱbeenȱshownȱtoȱproduceȱ promisingȱresults.ȱThisȱcouldȱbeȱfurtherȱdevelopedȱtoȱsetȱtheȱ areaȱonȱaȱsoundȱtheoreticalȱbaseȱandȱintroduceȱtheȱpossibilityȱ ofȱprofilingȱoffendersȱbyȱdiscoveringȱsimilarȱoffenderȱcharacȬ teristicsȱinȱlinkedȱcrimes.ȱ

ȱ ȱ ȱ ȱ ȱ

ȱ ȱ ȱ ȱ

C 2

C 1

C 3

SEX

CONTROL ESCAPE

Fig 1. Measuring similarity in 3 dimensions

A B C

c28 0.00 0.00 1.00

c29 0.00 0.00 1.00

c30 0.00 0.00 1.00

c31 0.00 0.39 0.60

c32 0.04 0.01 0.95

c33 0.00

0.00

1.00

c34 0.45 0.01 0.54

c35 0.00 0.00 1.00

c36 0.00 1.00 0.00

c37 0.00 0.99 0.01

c38 0.02 0.98 0.00

c39 0.03 0.97 0.01

(4)

ȱ 11ȱ

CEPOLȱEuropeanȱPoliceȱScienceȱandȱResearchȱBulletinȱ ȱ ȱIssueȱ3ȱȬȱSummerȱȱ2010 ȱ

Referencesȱ ȱ

BezdekȱJ.Cȱ(1981).ȱPatternȱRecognitionȱwithȱFuzzyȱObjectiveȱFunctionȱAlgorithms.ȱNewȱYork,ȱPlenum.ȱ ȱ ȱ

CanterȱDȱ(2000).ȱOffenderȱprofilingȱandȱcriminalȱdifferentiation.ȱLegalȱandȱCriminologicalȱPsychologyȱ5:ȱ23Ȭ46.ȱ ȱ ȱ

CanterȱD.V,ȱBennellȱC,ȱetȱal.ȱ(2003).ȱDifferentiatingȱSexȱOffences.ȱBehaviouralȱSciencesȱandȱLawȱ21.ȱ ȱ ȱ

GrubinȱD,ȱKellyȱP,ȱetȱal.ȱ(2000).ȱLinkingȱSeriousȱSexualȱAssaultȱthroughȱBehaviourȱHomeȱOfficeȱResearchȱStudyȱ215.ȱ London.ȱ

ȱ ȱ

ZadehȱLȱ(1965).ȱFuzzyȱSets.ȱInformationȱandȱControl(8):ȱ228Ȭ353.ȱ

ȱ ȱ ȱ ȱ

ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ

ȱ 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1 1.00 0.87 0.75 0.45 0.80 0.69 0.80 0.60 0.76 0.81 0.13 0.26 0.51 0.47 2 0.87 1.00 0.76 0.35 0.79 0.73 0.86 0.64 0.79 0.78 0.12 0.23 0.48 0.41 3 0.75 0.76 1.00 0.45 0.64 0.58 0.76 0.41 0.97 0.71 0.33 0.44 0.45 0.59 4 0.45 0.35 0.45 1.00 0.32 0.19 0.31 0.08 0.43 0.39 0.21 0.35 0.41 0.60 5 0.80 0.79 0.64 0.32 1.00 0.87 0.85 0.68 0.67 0.89 0.11 0.23 0.32 0.41 6 0.69 0.73 0.58 0.19 0.87 1.00 0.82 0.69 0.61 0.80 0.08 0.19 0.22 0.33 7 0.80 0.86 0.76 0.31 0.85 0.82 1.00 0.61 0.79 0.85 0.19 0.30 0.35 0.45 8 0.60 0.64 0.41 0.08 0.68 0.69 0.61 1.00 0.43 0.57 0.00 0.00 0.29 0.10 9 0.76 0.79 0.97 0.43 0.67 0.61 0.79 0.43 1.00 0.74 0.32 0.43 0.44 0.58 10 0.81 0.78 0.71 0.39 0.89 0.80 0.85 0.57 0.74 1.00 0.21 0.34 0.32 0.52 11 0.13 0.12 0.33 0.21 0.11 0.08 0.19 0.00 0.32 0.21 1.00 0.84 0.00 0.58 12 0.26 0.23 0.44 0.35 0.23 0.19 0.30 0.00 0.43 0.34 0.84 1.00 0.00 0.73 13 0.51 0.48 0.45 0.41 0.32 0.22 0.35 0.29 0.44 0.32 0.00 0.00 1.00 0.21 14 0.47 0.41 0.59 0.60 0.41 0.33 0.45 0.10 0.58 0.52 0.58 0.73 0.21 1.00

ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ ȱ

tableȱ2.ȱDegreeȱofȱȇclosenessȇȱbetweenȱcrimes

0.9

0.7

0.5

0.3

Fig 2 Closeness to the index crime