Am.J.Hum.Genet.59:497-500,1996 INVITEDEDITORIAL TheSecondNationalResearchCouncilReporton Forensic DNA Evidence BruceS.Weir PrograminStatisticalGenetics,DepartmentofStatistics,NorthCarolinaStateUniversity,Raleigh TheNationalResearchCouncilhasissuedanupdateof its1992reportontheforensicusesofDNA.This1996 reportendorsestheuseofDNA profilingforhuman identificationandacknowledgesthat"someofthestate- mentsinthe1992reporthavebeenmisinterpretedor misappliedinthecourts"(NationalResearchCouncil 1996,pp.ES-1and0-1).Inwhatislikelytobecome theirmostimportantfinding,the1996report'sauthors "foreseeatimewheneachpersoncanbeidentified uniquely(exceptforidenticaltwins),"andtheyoutline somecalculationstoshowwhenthisexpectationreason- ablycanbeexpectedtobefulfilled.

DNA profilinghasbeenusedforindividualidentifi- cationsince1985andhasbecomearoutinetoolinfo- rensicandpaternity-testinglaboratories.Whenthepro- filefromabiologicalstaindoesnotmatchthatofa specificindividual,thatpersonisexcludedasapossible contributortothestain.Whenthereisamatch,how- ever,therehasbeendebateoverthemostappropriate waytoexpressitsevidentiarystrength.Thedebatehad reachedsuchalevelby1991inthe1996reportthat theNationalResearchCouncilsetupacommitteeto examinetheissues.Theresulting1992reportalsoen- dorsedtheforensicuseofDNA profiling,anditpro- videdausefulhandbookforexplainingthenatureand utilityoftheprofiles(NationalResearchCouncil1992).

However,thesectionsconcerningthepopulation-ge- neticandstatisticalissuesinvolvedinthequantification ofmatchingprofileswereratherweak. Evenin1991,forensicDNAprofileswereoftenbased onsomanyVNTRloci,ateachofwhichsomanyalleles couldbedistinguished,thattheresultingnumberofpos- siblegenotypesfarexceededtheworldpopulationsize. Estimatingmultilocusgenotypefrequenciesforspecific profilescouldnotreston"counting"methods,which ReceivedMay22,1996;acceptedforpublicationJune25,1996.

Addressforcorrespondenceandreprints:Dr.BruceS.Weir,Depart- mentofStatistics,NorthCarolinaStateUniversity,Box8203,Raleigh, NC 27695-8203.E-mail:weir@stat.ncsu.edu Thisarticlerepresentstheopinionoftheauthorandhasnotbeen peerreviewed. © 1996byTheAmericanSocietyofHumanGenetics.Allrightsreserved. 0002-9297/96/5903-0003$02.00 simplycounthowmanytimestheprofileswereseenin populationsamples.Anappreciationfortherarityof specificprofilescouldinsteadbeobtainedbymultiplying togetherthefrequenciesoftheconstituentalleles,and theresultingextremelysmallproductswereoftenat- tacked.Neitherthe1992reportnorthe1996report pointedoutthattheseproductsdonotestimatethefre- quenciesofprofilesinthecurrentpopulation.Instead theseproductsprovidetheprobabilitiesofgenotypes thatcouldbeformedbytherandomunionofextant alleles.Inessence,thesefrequenciesrefertoinfinitepop- ulationsandneednotbelimitedtovaluessuchasthe reciprocalofthepopulationsize.

The1992reportsoughttosettleargumentsaboutthe validityoftheproductrulebysayingthatitoughtto beappliedto"ceiling"allelicfrequenciesratherthanto estimatedfrequencies.Thiswasintendedtoallowfor variationinallelicfrequenciesamonggroupswithin populations.Thisad-hocrulequicklyprovedtobeun- workable.The1996reportisunequivocal:"Weshare theviewofthosewhocriticize[theceilingprinciple]on practicalandstatisticalgroundsandwhoseenoscien- tificjustificationforitsuse"(NationalResearchCouncil 1996,pp.0-27,5-31).Similarshortshriftisgivento the"countingmethod"advocatedbythe1992report.

SomanylocicanbeusedincurrentDNA profilesthat aspecificprofilewillnotbeseeninforensicdatabases thatdonotincludetheprofiledonor.

The1996reportoffersascientificwayofaccommo- datingpopulationstructure.Theworryhasbeenthat thepeoplewhomightbeconsideredcontributorstobio- logicalsamplesinaparticularcrimemaybelongtoa groupwhoseallelefrequenciesarehigherthanthosein thegeneralpopulation.Usingestimatesofthegeneral frequenciescould,therefore,leadtoanunderestimate oftheprofilefrequencyandbeprejudicialtopeoplewho matchsimplybychance.The1996NationalResearch Councilreportpointstotheseveralpublishedstudies offorensicmarkerfrequenciesthatshowlessvariation betweensubgroupswithinracialgroupsthanbetween racialgroups.Thismeansthatquotingfrequenciesfrom differentracialgroupsprovidesanupperlimitonthe variationwithinracialgroups.

The1996reportusespublishedaccountsofVNTR 497 Am.J.Hum.Genet.59:497-500,1996 datatorecommendthatprofilefrequenciesbeassigned a"confidenceinterval"formedbybothmultiplyingand dividingthefrequencyby10.Thedatathatsuggested thisfactorweregivingprofilefrequenciesoftheorder of10-8.Unlikerulesbasedonstatisticalproceduressuch asbootstrapping,thisad-hocfactor-10ruledoesnot giveuniformlysensibleresults.Astheprofilefrequency becomeslarger,thelimitsoftheconfidenceintervals shouldbecomesmallermultiplesoftheestimate,and thefactorof10isnotappropriateforPCR-markerfre- quenciesoftheorderof10-2or10'.Nordoesthe factorof10takeaccountofsamplesize.

Nevertheless,substructurecanincreasehomozygote frequenciesinthewholepopulation,overthoseexpected fromtheHardy-Weinberglaw.Thiseffectisknownas theWahlundprincipleandisaconsequenceoftheaver- agevalue,oversubpopulations,ofthesquareofanallele frequencybeinggreaterthanthesquareoftheaverage allelefrequency.Estimatingaone-locushomozygotefre- quencyasthesquareofthewhole-populationallelefre- quencywouldthereforegiveanunderestimate.The 1996reportrecommendsuseofacorrectionfactorfor homozygoteswhentheobjectistoprovideprofilefre- quencies,wherethecorrectionfactorreflectsrealistic levelsofhumanpopulationstructure.Ifhomozygote frequenciesareincreasedbypopulationstructure,itfol- lowsthatheterozygotefrequencieswill,onaverage,be decreased.Withthislogic,the1996reportrecommends thatnocorrectionbemadeforheterozygotes.Thisis eventhough,astheyacknowledge,someheterozygotes mayalsohaveincreasedfrequencies.Ofcourse,itwill notbeknownwhetheranyparticularheterozygoteis onethathasafrequencyover-orunderestimatedbythe productrule,andaprosecutorwhofollowsthe1996 reportanddoesnotalsoadjustheterozygotefrequencies islikelytobechallenged.Thelawisconcernedabout particularcases,ratherthanaboutwhathappenson average.

Theproblemdisappearsifattentionisshiftedfrom profilefrequenciestoconditionalfrequencies.Insteadof determiningthechanceofseeingaprofileinsomeran- domperson,thequestionbecomesthechanceofseeing asecondoccurrenceoftheprofile,giventhatithasbeen seenoncealready-inthesuspect,forexample.Condi- tionalfrequenciesaregreaterthanproductsofallelic frequencies,forbothhomozygotesandheterozygotes, andtheyarisenaturallyinlikelihoodratios.

Theforensic-scienceliteratureisclearonthemerits ofphrasingthestrengthoftraceevidenceintermsof likelihoodratios,wherebytheprobabilityoftheevi- denceunderoneexplanationiscomparedtotheproba- bilityunderanalternativeexplanation.IntheDNAcon- text,thismeanstheprobabilityoffindingaspecific profileinanevidentiarystainwheniteitherissupposed tohavecomefromaspecificpersonwhosetypeisknown oritcomesfromsomeunknownperson.Thecursory rejectionofthisapproachbythe1992reportisremedied inthe1996report.Aspartofthediscussionoflikeli- hoodratios,the1996reportshowshowthelikelihood ratioreducestothereciprocaloftheprofilefrequency whensuspectandperpetratorhaveindependentfre- quencies.The1996reportembracestheratioofcondi- tionalfrequencies,usingresultsofBaldingandNichols (1994),whensuspectandperpetrator(ifdifferent)are knowntobelongtothesamesubpopulation.Because suchknowledgeislikelytobelacking,itwouldseem bettertouseanapproachthatmakesasfewassumptions aspossibleandnottorequiremembershipofindepen- dentpopulations.Thepopulation-structureparameter valuesinvolvedintheconditionalfrequenciescouldal- waysbechosentoreflecttheknowledgeavailablefor thecase.Thepointisthatthesamegeneralapproach wouldleadtoappropriateresultsinallcases.Asitis, theforensicscientistisbeingtoldtodecidewhetherto useconditionalfrequencies.

Givingcentralprominencetolikelihoodratioswould havemadethe1996report'scalculationsgivenforrela- tives,inadditiontothosegivenformembersofthesame subpopulation,appearmorenaturally.Initsdiscussion onuniqueness,the1996reportstressestheneedtocal- culatetheconditionalprobabilityofaprofile,giventhat ithasalreadybeenseen,andthisviewcouldwellhave beenadoptedthroughoutthe1996report.Althoughthe authorsofthe1996reporthavenotadoptedthegeneral useofconditionalfrequencies,theyatleasthaverecom- mendedthatresearchbeconductedintohowbestto presentevidencetothetrieroffact.Suchresearchshould beofvalue,buttherecanbelittledoubtthatthepurely population-geneticconsiderationsarebestaccommo- datedbylikelihoodratiosexpressedintermsofcondi- tionalprobabilities.

The1996reportdoesgiveastrongendorsementto theuseoflikelihoodratiosintheinterpretationofstains frommorethanonecontributor.Acarefuldiscussion isgiven,andtheresultinglogicaltreatmentemploying conditionalfrequenciesisemphasized.Wehaveshown recently(Weiretal.,inpress)thatthenaivemethodof the1992report,basedon"randommannotexcluded" thinking,actuallycanbeprejudicialtoadefendant whoseprofileisnotexcludedfromamixedstain.The 1992reportcalledforreportingonlythefrequencywith whicharandomindividualwouldbeexcludedfrom beingacontributortoamixedstain,ignoringthefact thatthestainisknowntobefrommorethanonecon- tributor.Thehurriedtreatmentofmixedstainsinthat reportwasthecauseofjustoneofthebarrierstothe presentationofsoundanalysesincourt(Weir1995),a barrierthathasbeenremovedbythe1996report.

The1996reportechoesthesentimentsofforensic scientistssuchasEvettandBuckleton(inpress),who 498 Weir:InvitedEditorial decryoveremphasisonstatisticaltestingfordepartures fromtheHardy-Weinbergfrequencies.Instead,the1996 reportassumesthatdeparturesdoexistandconsiders whethersuchdeparturesarelargeenoughtobeim- portant.Theproposedcorrectionfactorforhomozy- gotesistoensurethatcourtsarenotmisledaboutthe strengthofDNA evidence.Readerswillneedtonote thedisclaimerinthe1996report-thatis,thatthescore testdiscussedinchapter4isnotatestforHardy-Wein- bergfrequenciesandthatstudiesontheuseofexact tests(e.g.,seeMaisteandWeir1995;Zaykinetal.1995) showsatisfactorypowerfordetectingdeparturesfrom Hardy-Weinbergfrequencies,atmultiallelicloci.The staunchestcriticsoftheproductrulehavestressedthe difficultyoftestingforindependencebetweenalleles overseveralloci.However,two-locustestshavebeen foundtobehavesatisfactorily,and,withitscharacteris- ticemphasisonscientificreasonableness,the1996re- portconsidersitunlikelythattherewillbedependence atlargernumbersoflociwhenthereisnoevidencefor dependenceatoneortwoloci.Citationsaregivenof studiesthatbolsterthisviewbycomparingobserved andexpectednumbersofmatchesatseveralloci.

Althoughanoveremphasisonindependencetestingis indeedtobeavoided,the1996report'simplicationthat testingbeabandonedseemstoerrintheotherdirection.

Iftheparameterizationthatthe1996reportgivesfor possibledeparturesfromHardy-Weinbergfrequencies isused,itcanbeshownthatfindingsofsignificantdepar- turesarelikelytoindicatelargevaluesoftheseparame- ters-certainlylargerthanthevaluesof0.01or0.03 suggestedascorrectionfactorsforhomozygotes.It wouldsurelybebettertocontinuewithtesting.Al- thoughfailuretorejectthehypothesisofindependence doesnotestablishtheexistenceofHardy-Weinbergfre- quenciesanddoesnotobviatetheneedforcorrections totheproductrule,whenthetestsdorejectthehypothe- sisothermodificationsmaybenecessary.

Astatisticalrecommendationofconcernistheproce- duretobefollowedwhenasuspectisidentifiedbya searchofaDNA-profiledatabase.Suchdatabasesare nowbeingsetupinmanystates,andalreadysamples havebeencollectedfromseveralhundredthousandcon- victedoffenders.Iftheprofileinquestionhasapopula- tionfrequencyofP,thenNPcopiesareexpectedina databaseofsizeN.IftheestimatedvalueofPisP.the 1996reportadvocatesreportingNP,insteadofjustP. asamatchprobability.Thisprocedureiscalledforbe- causeNPisanupperboundontheprobabilitythatat leastoneofthedatabaseprofilesmatchestheevidence profile.However,thelikelihoodratioforassessingthe strengthofthematchstillinvolvestheprobabilityofthe matchifthepersonidentifiedisnotthesourceofthe evidenceprofile,andthishasnotbeenchangedbythe factofadatabasesearch.Indeed,asBaldingandDon- nelly(inpress)haveshown,thereisinformationinthe factthatN - 1peoplehavebeenexcludedaspossible contributors.Certainlytherewouldbenoneedforan NPcalculationifN wasthesizeofthepopulationand ifPwas1/N;findingamatchwouldbecertain,and therewouldbenoneedtoreducetheevidentiaryweight ofthematch.The1996reportacknowledgesthatmore- complicatedcalculationswillbeneededforlargedata- bases,anditcouldpreventalmostcertainargumentin courtbydetailingthecalculationstowhichitrefers.

The1996report'streatmentofhowtoquantifythe extentofpopulationstructureisbrieferthanitcould havebeen.Thereisadangerofconfusionbetweenthe quantitiesWk,definedastheproportionofapopulation inthekthsubpopulation,andthesizesnkofsamples fromthesesubpopulations.The1996reportprovides anexplanationofgeneralprinciplesratherthandetails forestimatingmeasuresofpopulationsubstructure.Use ofsubpopulationproportionsWkhasgenerallyimplied afixedpopulationapproach,insteadofarandompopu- lationapproachwherereplicationprovidesabasisfor expectationofstatistics.Itistheexpectedvaluesthat arewrittenintermsofthestructureparameters0.The distinctionbetweenfixedandrandompopulations, alongwithestimationprocedures,hasbeengivenby CockerhamandWeir(1986).Becausethenumerical consequencesofthesedifferentanalysesarenotgreat, however,theissueshouldhavenoimpactonforensic conclusions.

The1996reportdevotesachaptertolaboratoryper- formance,reflectingtheshiftinthedebatetoissuesof possiblefalsedeclarationsofamatchbetweentwopro- files.Thethoughthasbeenexpressedthatthereislittle meritinpresentingprofilefrequenciesof1in1millionif thereisa1-in-100chanceoflaboratoryerror.Acareful discussionisgiven,inchapter3ofthe1996report,of theneedforqualitycontrolandqualityassuranceand fortheaccreditationofforensiclaboratories.Thesesteps allrequireproficiencytestingandlaboratoryaudits,and the1996reportendorsesthecontinuationoftheseprac- tices.Acleardistinctionismadebetweentheresultsof proficiencytestsandtheresultsoflaboratoryperfor- manceonaspecificcase.The1996reportreiteratesthe 1992report'sstatementthatproficiency-testerrorrates shouldnotbecombinedwithprofile-frequencyesti- mates.Instead,the1996reportpointstoonewayof reducingthepossibilityoferror-byduplicateandinde- pendenttestingofsamples.Althoughthiswillnotal- waysbepossible,themeritsofsuchduplicationareobvi- ous.Lookingfurtherahead,itmaywellbethatthe contaminationissuewillberesolvedbytechnicalad- vances,suchasPCRamplification,thatwillallowtyping ofsamplestobecompleted,atacrimescene,beforea suspectisevenidentified.

Atseveralpoints,the1996reportmentionsthe"cre- 499 500 Am.J.Hum.Genet.59:497-500,1996 ativemisapplication"ofthe1992report(NationalRe- searchCouncil1996,p.0-27).A specificcitationis giventothecaseofStatev.Guevara(1993)inwhich resultswerepresentedafter"combingthroughVNTR dataonmanysubgroupstofindthelargestallelefre- quencies,takingtheupperendofaconfidenceinterval foreachsuchfrequency,ignoringlocibecauselargesam- plesindicatethatallelesforsomeotherlocusdonot occurinHardy-Weinbergproportions,andusingfixed- binfrequencieswithextremelywidebins"NationalRe- searchCouncil1996,pp.6-21).Suchprocedures"were neithercontemplatedbythe1992committeenorreflect reasonablescientificjudgments,"buttheydidcausepro- filestatisticsnottobeallowedinthatcase.Ifindthe callforscientificreasonablenesstobemostrefreshing, andIwouldliketosharethe1996report'soptimism thatadmissibilityofDNA-profilestatisticsshouldno longerbeindoubt.However,Inotethatthedefense expertintheStatev.Guevaracaseemployedthesame proceduresinaPennsylvaniaadmissibilityhearingafter readingthe1996report.

Apartfromaddressingpopulation-geneticandstatisti- calissues,the1996reportopenswithausefuldescrip- tionofthemolecularbasisofDNA typingandcloses withathoughtfuldiscussionofthelegalimplicationsof theconclusionsandrecommendationsmadeinthebody ofthe1996report.Consequently,the1996reportwill beavaluableresourcetotheforensicandlegalcommu- nities.Theauthorsofthe1996reportaretobecongrat- ulatedontheireffortstomakerecommendationsonthe basisofscientificarguments.Theyhavedonemuchto helpthepropercalculationandpresentationofDNA- profilestatistics,andthedayonwhichDNAprofilesare employedwiththesametrustasarefingerprintshas surelybeenbroughtforwardbytheirreport.

References BaldingDJ,DonnellyP.EvaluatingDNA profileevidence whenthesuspectisidentifiedthroughadatabasesearch.J ForensicSci(inpress) BaldingDJ,NicholsRA(1994)DNAprofilematchprobability calculation:howtoallowforpopulationstratification,relat- edness,databaseselectionandsinglebands.ForensicSciInt 64:125-140 CockerhamCC,WeirBS(1986)Estimationofinbreedingpa- rametersinstratifiedpopulations.AnnHumGenet50:271- 281 EvettIW,BuckletonJS.StatisticalanalysisofSTRdata.Proc 16thIntCongForensicHaemogenet(inpress) MaistePJ,WeirBS(1995)Comparisonoftestsforindepen- denceintheFBIRFLPdatabases.Genetica96:125-138 NationalResearchCouncil(1992)DNAtechnologyinforen- sicscience.NationalAcademyPress,Washington,DC (1996)TheevaluationofforensicDNAevidence.Na- tionalAcademyPress,Washington,DC StatevGuevara.K9-92-1873,MinnDistCt,DakotaCty,Jan 26,1993 WeirBS(1995)DNA statisticsintheSimpsonmatter.Nat Genet11:365-368 WeirBS,TriggsCM,StarlingL,StowellLI,WalshKAJ,Buck- letonJ.InterpretingDNAmixtures.JForensicSci(inpress) ZaykinD,ZhivotovskyL,WeirBS(1995)Exacttestsfor disequilibriawitharbitrarynumbersoflociandalleles.Ge- netica96:169-178