聯(lián)合國亞太經(jīng)社會（UNESCAP）：亞洲和太平洋地區(qū)兒童出生登記情況

SOCIALDEVELOPMENTDIVISION

BirthRegistrationinAsiaandthePacific:

AClassificationandRegressionTreesAnalysistoIdentifythe

FurthestBehindChildren

OrlandoRoman

WORKINGPAPERSERIES

SEPTEMBER2023

2

Disclaimer:TheviewsexpressedthroughtheSocialDevelopmentDivisionWorkingPaperSeriesshouldnotbereportedasrepresentingtheviewsoftheUnitedNations,butasviewsoftheauthor(s).WorkingPapersdescriberesearchinprogressbytheauthor(s)andarepublishedtoelicitcommentsforfurtherdebate.Theyareissuedwithoutformalediting.TheUnitedNationsbearsnoresponsibilityfortheavailabilityorfunctioningofURLs.Opinions,figuresandestimatessetforthinthispublicationaretheresponsibilityoftheauthorsandshouldnotnecessarilybeconsideredasreflectingtheviewsorcarryingtheendorsementoftheUnitedNations.Anyerrorsaretheresponsibilityoftheauthors.MentionoffirmnamesandcommercialproductsdoesnotimplytheendorsementoftheUnitedNations.

3

Pleasecitethispaperas:Roman,O.(2023).BirthRegistrationinAsiaandthe

Pacific:AClassificationandRegressionTreesAnalysistoIdentifytheFurthestBehindChildren.UnitedNationsESCAP,SocialDevelopmentDivision,September2023.Bangkok.

Availableat:

/kp

ESCAP/6-WP/63

Abouttheauthors:OrlandoRomanservesasaconsultantattheUnitedNationsEconomicandSocialCommissionforAsiaandthePacific.

4

TableofContents

ABSTRACT 5

1.INTRODUCTION 6

2.THEDETERMINANTSOFBIRTHREGISTRATION 8

3.DATA 9

4.METHODOLOGY 12

5.RESULTS 14

6.CONCLUSION 22

7.APPENDIX 23

REFERENCES 29

5

ABSTRACT

ThispaperinvestigatesthedeterminantsofbirthregistrationinAsiaandthePacific.DrawingondatafromtheDemographicandHealthSurveys(DHS)andtheMultipleIndicatorClusterSurveys(MICS)fromfiverepresentativecountriesintheregion,namelyCambodia,Mongolia,Pakistan,Samoa,andUzbekistan,thepaperusesClassificationandRegressionTrees(CART)toidentifythegroupsofchildrenunder5yearsofagewhoareleastlikelytohavetheirbirthbeingregistered,i.e.,thefurthestbehindgroups.Results

underscoresignificantwithinandcross-countrydisparities.Thewealthofthehouseholdwherethechildcurrentlyresides,ageofthechild,educationalattainmentofthemother,numberofchildrenunder5yearsofageinthehousehold,andsexofthechild,appeartobethemostimportantcharacteristicsinfluencingbirthregistrationratesandidentifyingthefurthestbehindgroups.ThepaperaimstoinformpolicymakingandinterventionsstrategiesintendedatimprovingbirthregistrationratesinAsiaandthePacific.

Keywords:CRVS,birthregistration,leaving

noonebehind,inequalityofopportunity.

6

1.INTRODUCTION

The2030AgendaforSustainableDevelopmentmakesacallto“l(fā)eavenoonebehind”andto“endeavortoreachthefurthestbehindfirst”inthepathtowardssustainabledevelopment.Itcrystallizestheglobalconsensusfortransitioningtowardsmoreegalitarian,cohesiveandsolidarity-basedsocieties,bypromotinginclusivedevelopmentasapathwaytoasustainablefuture.Inpracticalterms,thepledgeto“l(fā)eavenoonebehind”meansgoingbeyondassessingaverageandaggregateprogress,towardsensuringprogressforallpopulationgroupsatadisaggregatedlevel.Thisrequiresdisaggregatingdatatoidentifygroupsbeingvulnerable,excludedordiscriminatedagainst,aswellasthoseexperiencingmultipleandintersectingformsofdiscriminationandinequalities.

Birthregistrationservesasafundamentalconduitintheendeavorto“l(fā)eavenoonebehind”.Birthregistrationrefersto“thecontinuous,permanentanduniversalrecording,withinthecivilregistry,oftheoccurrenceandcharacteristicsofbirthsinaccordancewiththelegalrequirementsofa.iItisdeemedanessentialinitialstepinassessinganindividual’slegalidentityand,consequently,theiraccesstorightsandservices.Inaddition,itisafundamentalhumanrightrecognizedbytheInternationalCovenantonCivilandPoliticalRightsandtheConventionoftheRightoftheChild.1

BirthregistrationisemphasizedinSustainableDevelopmentGoal(SDG)16,target16.9,whichcalls“[b]y2030,[to]providelegalidentityforall,includingbirthregistrationThistargetistrackedbytheindicator“proportionofchildrenunder5yearsofagewhosebirthshavebeenregisteredwithacivilauthority,byage”.iiHowever,birthregistrationhasalsoadirectimpactonotherdevelopmentareas,includingnopoverty(SDG1),goodhealthandwell-being(SDG3),educationaladvancement(SDG4),accesstodecentjobs(SDG

1Inspecific,article24oftheInternationalCovenantonCivilandPoliticalRights,andarticle7oftheConventionontheRightsoftheChild.

5),reducedinequalities(SDG10),peace,justice,andstronginstitutions(SDG16),amongothers.

Theabsenceofbirthregistrationcanleadtoseverelong-termrepercussions,buildingvulnerabilitiesacrossthelifecycle.Itrenderschildrenandindividualsinvisibletogovernmentalsystemsandhinderstheirfullengagementtocontributeandbenefitfromsocietalprogress.Thosewithouttheirbirthbeingregisteredmayencountersignificantbarrierswhentryingtoaccessservicessuchashealthcareandeducation,makingthemsusceptibletomarginalizationandpoverty.Thelackofageverificationalsoexposesunregisteredindividualstohumanrightsabuses,includingunderagelabor,earlymarriage,andhumantrafficking.Theinabilitytoconfirmone'snationalitycanalsoresultinstatelessness,furthercomplicatingtheirsituation.

ThispaperidentifiesthecommoncharacteristicsaffectingtheprevalenceofbirthregistrationinAsiaandthePacific.Itusesclassificationandregressiontrees(CART)toexploreinequalitiesinbirthregistrationandidentifythosechildrenunder5yearsofagewhoareleastlikelytohavetheirbirthbeingregisteredwiththecompetentnationalauthorities.ThepaperusesDemographicandHealthSurvey(DHS)andMultipleIndicatorClusterSurvey(MICS)acrossfivecountriesinAsiaandthePacific,eachrepresentingadifferentsub-region.CARTisastraightforwardtooltooperationalizethepledgeto“l(fā)eavenoonebehind”andacceleratenationalprogresstowardsachievingtheSDGsduetoitssimplicityandinterpretability.Itusesanalgorithmtoassessallpossiblecombinationsofcharacteristicstodeterminewhichonesaccountthemostinexplainingbirthregistrationratesandidentifythefurthestbehindgroupsi.e.,thosegroupsofchildrenunder5yearsofagewiththelowestbirthregistrationrates.

7

Thispaperisstructuredasfollows:Section2offers

aliteraturereview.Section3introducesthedata.

Section4explainsCART.Section5presentsthe

findingsonbirthregistration.Section6comestoa

conclusion.Thepaperissupplementedbyan

Appendix.

8

2.THEDETERMINANTSOF

BIRTHREGISTRATION

Thescholarlyliteratureonbirthregistrationidentifiesamultitudeoffactorsinfluencingthisvitalprocess.InAsiaandthePacific,theinterplayofdemandfactorssuchassocio-economicstatus,parentaleducation,religion,ethnicity,mother’sage,andmaritalstatushasprovedtosignificantlyinfluencebirthregistrationrates.InIndia,forexample,lowerbirthregistrationratesareprevalentamongchildrenwhosemotherslackformaleducation,arenotexposedtomedia,belongtothepoorestwealthquintiles,orareidentifiedwithmarginalizedclassesorasMuslims.iiiMeanwhile,intheLaoPeople’sDemocraticRepublic,birthregistrationratesforchildrenunder5areinfluencedbyethno-geographicfactors,includingsettlementpatternsandtheirmotherhavingaminorityidentity.iv

Inadditiontoindividualcharacteristics,supply-sidefactorscanfurtheraffectthelikelihoodofabirthbeingregistered.Forinstance,inIndonesia,thecostofregistration,thedistancetotheregistrationcenter,andalackofawarenessandknowledgeabouttheregistrationprocessserveassignificantimpedimentsforchildrenunder5beingregistered.vSimilarly,inChina,bothculturalfactorsandgeographiclocationarehighlightedaskeydeterminantsofbirthregistration.vi

AnalysingstudiesbeyondAsiaandthePacificbringsforthadditionalinsights.InGhana,apositivecorrelationexistsbetweenamother’seducationlevelandhouseholdwealth,andthelikelihoodofbirthregistration.viiInNigeria,awarenessaboutbirthregistration,accesstomassmedia,alongwithmother’sage,maritalstatus,occupation,andplaceofdeliveryhavebeenidentifiedascontributorstoabirthbeingregistered.viiiAcrossSub-SaharanAfrica,reducedchancesofbirthregistrationareassociatedwithwealthandeducationstatus,limitedwomen’sautonomy,adherencetotraditionalreligions,andotherfactorssuchasprenatalcare,birthregistrationlegislation,fertilityrates,prevalenceofconflicts,andthedegreedecentralizationoftheregistrationsystem.ix

Takenasawhole,thesestudiesemphasizeamultitudeofindividual,societalandinstitutionalfactorsinfluencingbirthregistrationrates,therebyunderliningthenecessityforin-depth,country-specificanalyses.Individualcharacteristicsencompasselementssuchassocioeconomicstatus,children’sandmother’sage,parentaleducation,religion,ethnicity,andmaritalstatus.Conversely,systemicdeterminantsincludefactorssuchasdistancetoregistrationcentres,cost,availabilityofinformation,awarenessoftheregistrationprocess,andculturalnorms.

9

3.DATA

ThispaperusesDHSandMICS.ThedataspansfivecountriesinAsiaandthePacific,eachstandingasrepresentativefortheirrespectivesubregions,namelyCambodiaDHS2021-2022fromSouth-EastAsia,MongoliaMICS2018fromNorthandNorth-EastAsia,PakistanDHS2017-2018fromSouthandSouth-WestAsia,SamoaMICS2019-2020fromthePacific,andUzbekistanMICS2021-2022fromNorthandCentralAsia.

TheDHSandMICSdatasetsareselectedduetotheirconsistentcollectionofbirthregistrationdata.ThestandardDHShouseholdquestionnaireincludesaquestionaboutwhetherallchildrenunder5yearsofageareregistered.Similarly,theMICSquestionnaireasksmothersorprimarycaretakersofchildrenunder5yearsofagetoprovideinformationontheirchild’sbirthcertificateorregistrationstatuswiththerelevantcivilauthorities.Thesesurveysalsoofferotheradvantages,including:1)comparabilityacrosscountries;2)accessibilityofthedata;and3)therichsetofquestionsonhealth,demographicandbasicsocio-economicdataaboutchildrenunder5yearsofage,theirmother,andthehouseholdstheylivein.

Theanalysisofthispaperisconfinedtochildrenunder5yearsofage,astheyconstitutethereferencegroupforSDG16.9.1anddataarenotavailableforchildren5yearsofageandabove.Whetherachild’sbirthisregisteredwiththenationalauthoritiesisbasedonapositiveresponsefromeithertheirmotherorprimarycaretaker(MICS),orthehouseholdhead(DHS),toeitherofthequestions:

1)“Does(name)haveabirthcertificate?”;and

2)“Has(name)’sbirthbeenregisteredwiththeCivilRegistrationAuthority?”.

Thespecificsetofquestionsvariesaccordingtotheparticularsurvey(Table1).Thismethodoffersasimple,yeteffectivewaytoassessbirthregistrationrateswithinthesurveyedpopulations.ToillustratethedeterminantsofbirthregistrationinAsiaand

thePacific,thispaperusesSamoa’sMICS2019-2020asarepresentativecasestudy.Thechoiceofvariablesalignswiththescholarlyliterature,recognizingallpotentialdeterminantsacrossthesurveymodules,includingdatafromchildrenunder5yearsofage,women15-49yearsofage,andcharacteristicsofthehouseholdstheylivein.ForcountrieswithaDHSsurvey,theinformationisavailableinthehouseholdandhouseholdmembermodule.

TheMICSsurveymodulesareaggregatedatthehouseholdlevel,associatingeachchildunder5yearsofagewiththeirmotheroraprimarycaretakerwithinthehousehold(Table2).Thecharacteristicsselectedtoassessbirthregistrationratesarewealthofthehouseholdwherethechildrenunder5yearsofagecurrentlyreside

(belongingtothetop60orthebottom40ofthewealthdistribution)andplaceofresidence(urbanorrural).Inexaminingthecharacteristicsofthechildrenunder5yearsofage,thefollowingcharacteristicsaretakenintoaccount:numberofchildrenunder5yearsofageinthehousehold

(none(<0.5),1ormore(>0.5),2ormore(>1.5),3ormore(>2.5),fewerthan2(<1.5),fewerthan3(<2.5));ageofthechild(1,2,3,or4yearsofage);sexofthechild(boyorgirl);andwhetherthechildisanonlychild.Lastly,inevaluatingmother’scharacteristics,theconsideredvariablesare:agegroupofthemother(15-24yearsofage,25–34yearsofage,or35-49yearsofage);educationalattainmentofthemother(lowereducation,secondaryeducation,orhighereducation);maritalstatusofthemother(currentlymarried,formerlymarried,ornevermarried)andplaceofdeliveryofthechild(hospitalorhome/other).Othervariablessuchastheethnicity,religion,andlanguageofthehouseholdheadwerealsoconsidered,thoughonlyreligionconsistentlyaffecttheresultsinSamoa.Further,systemicdeterminantssuchascost,availabilityofinformation,awarenessoftheregistrationprocess,amongothers,arenotconsideredduetodataunavailability.

10

Table1:Listofcountries,variablesnames,andsurveyquestions

Country

Data

VariableName

SurveyQuestioninDHS/MICS

AnswersinDHS/MICS

Survey

Recode

Cambodia

DHS

hv140

Does(NAME)haveabirth

certificate?

IFNO,PROBE:

Has(NAME)'sbirtheverbeen

registeredwiththecivilauthority?

1=HasCertificate

2=Registered

3=Neither

8=Don'tKnow

PR

Mongolia

MICS

BR1

Does(name)haveabirthcertificate?Ifyes,ask:

MayIseeit?

1=Yes,Seen

2=Yes,NotSeen

3=No

8=DK

CH

BR2

Has(name)’sbirthbeenregisteredwiththeCivil

RegistrationAuthority?

1=Yes

2=No

8=DK

Pakistan

DHS

hv140

Does(NAME)havehis/hernameenteredontoa'bay'form?

IFYES:Does(NAME)havea

birthcertificate?

IFNO:Does(NAME)havea

birthcertificate?

1=Nameonbayformandhavebirthcertificate

2=Nameonbirthformandhavenobirthcertificate

3=Onlybirthcertificate

4=Neitherofabove

8=Don'tKnow

PR

Samoa

MICS

BR1

Does(name)haveabirthcertificate?Ifyes,ask:

MayIseeit?

1=Yes,Seen

2=Yes,NotSeen

3=No

8=DK

CH

BR2

Has(name)’sbirthbeenregisteredwiththeBirth,DeathandMarriagesOffice(BDM)?

1=Yes

2=No

8=DK

Uzbekistan

MICS

BR1

Does(name)haveabirthcertificate?Ifyes,ask:

MayIseeit?

1=Yes,Seen

2=Yes,NotSeen

3=No

8=DK

CH

BR2

Has(name)’sbirthbeenregisteredwiththeCivilRegistryOffice?

1=Yes

2=No

8=DK

Note:ESCAPelaborationbasedonlatestDHSandMICS.

11

Table2:Surveydataanddeterminantsofbirthregistration.SamoaMICS2019-2020

“ch”

childrenrecode

“wm”

womenrecode

“hl””

householdmembersrecode

“hh”

householdrecode

HH1

clusternumber

HH2

householdnumber

LN

linenumber

UF1clusternumber

UF2householdnumber

UF3

child'slinenumber

mother/caretaker'sline

UF4

number

nativelanguageofthe

UF14

respondent

UB2

ageofchild

coveredbyanyhealth

UB9

insurance

BR1birthcertificate

BR2

birthregistration

knowledgeofbirth

BR3

registration

HH6area

HL4

sex

highestlevelof

ED5A

educationattended

CAGE

ageinmonths

melev

el

mother'seducation

cinsur

ance

healthinsurance

cdisab

functionaldifficulties

ility

(age2-4years)

ethnici

ethnicityofhousehold

ty

head

religio

religionofhousehold

n

head

careta

mother'sfunctional

kerdis

disabilities(age18-49)

chwei

childrenunder5's

ght

sampleweight

winde

x5wealthindexquintile

PSU

primarysamplingunit

stratu

m

stratum

HH1

HH2

LN

WM1

WM2

WM3

WM14

WB4

WB6A

WB18

WB20

WB21

WB22

WB24

MT9

CM11

MN2

MN20

MA1

MA5

HH6

CEB

welevelinsurance

disabilitywmactivity

wreligio

n

wmemployment

ethnicity

religionwmweight

windex5

PSU

stratum

clusternumber

householdnumberlinenumber

clusternumber

householdnumberwoman'sline

number

nativelanguageoftherespondent

ageofwoman

highestlevelof

schoolattendedcoveredbyhealth

insurance

religionof

householdhead

mothertongue

ethnicgroup

occupation

everusedthe

internet

childreneverbornreceivedprenatalcare

placeofdelivery

maritalstatus

maritalstatus

area

childreneverborn

education

healthinsurancefunctional

difficulties(age18-49)

mainactivity

women'sreligionwomen's

employmentstatusethnicityof

householdhead

religionof

householdheadwoman'ssampleweight

wealthindex

quintile

primarysample

unit

stratum

HH1

clusternumber

HH2

householdnumber

HL1

linenumber

relationshiptothe

HL3

head

HL4sex

HL6

age

linenumberof

HL8

womanage15-49linenumberfor

HL10

childrenage0-4highestlevelof

ED5A

educationattended

HH6

area

ethnici

ethnicityof

ty

householdhead

religio

religionofhousehold

n

head

heleve

educationof

l

householdhead

melev

el

mother'seducation

careta

functionaldifficulties

kerdis

(age18-49years)

felevel

father'seducation

hhwei

householdsample

ght

weight

winde

x5

wealthindexquintileprimarysampling

PSU

unit

stratu

m

samplestrata

HH1

HH2

HH6

HH1

6

HH4

8

HH5

1

HC1

A

HC1

B

HC2

HC1

3

HHS

EX

HHAGE

ethni

city

religi

on

helevel

hhweightwindex5

PSU

strat

um

clusternumber

householdnumber

area

nativelanguageoftherespondent

numberofHH

members

numberofchildren

underage5

religionof

householdhead

languageof

householdheadethnicgroupofhouseholdheadinternetaccessathome

sexofhouseholdhead

ageofhouseholdhead

ethnicityof

householdhead

religionof

householdhead

educationof

householdhead

householdsample

weight

wealthindex

quintile

primarysamplingunit

samplestrata

Note:ESCAPelaborationbasedonSamoaMICS2019-2020.

12

4.METHODOLOGY

TheprimarygoalofCARTistoidentifythegroupsofchildrenunder5yearsofagewiththelowestandhighestlikelihoodoftheirbirthbeingregistered.Theindicator“birthregistration”istheresponsevariable,whilethefactorsthatcharacterizethesegroupsofchildren,theirmother,andthehouseholdstheylivein,aretheexplanatoryvariables,alsoreferredtoas“circumstances”.2Atreeisananalyticalstructurethatrepresentsgroupsofchildrenofthesamplepopulationthathavesignificantlydifferentbirthregistrationrates.

Toidentifythegroupsofchildrenunder5yearsofagewiththegreatestdifferenceinbirthregistration,CARTisconstructedforeachcountryintheformofbinaryregressiontreesusingR,anopen-sourcestatisticalsoftware.Therootnodeofthetreereferstotheentirepopulationsample.TheCARTalgorithmstartsbysearchingforthefirstsplit(or“partition”)ofthetree.Itdoessobylookingateachcircumstanceandseparatingthesampleintwogroups,sothatitbestsatisfiesacertain“splittingcriterion.”Thissplittingcriterioncanbedefinedinafewways,whiletheoneusedhereistheAnalysisofVariance,or“ANOVA.”Theformulathatrepresentsthecoreofthealgorithmisthefollowing:

SST?(SSL+SSR)

WhereSST=∑(Yi?)2isthesumofsquaresfor

theparentnode,andSSLandSSRarethesumsofsquaresfortheleftandrightchildnodes,respectively.xThesumofsquaresiscalculatedby

firstfindingthedistancebetween,thesample

mean,andtheithdatapointYi.Thisisalsoreferredtoasthedeviation.

IfdeviationsforalldatapointsY1,Y2,…,Yiare

squaredandthensummed,asin∑(Yi?)2,this

yieldsthesumofsquaresforthesedata.Thisisequivalenttochoosingthesplittomaximizethe

2Withinthefieldofmachinelearning,responseandexplanatoryvariablesaremorecommonlyreferredtoas

between-groupssumofsquaresinasimpleanalysisofvariance.

TheCARTalgorithmappliesthisANOVAtesttoeachpossiblesplit,orpartition,ofthesamplepopulationgivenbythecircumstances.Aftercomparingalloftheavailablepartitions,thealgorithmisabletoidentifythesinglepartitionthatwoulddecreasevariancethemostforthecombinedsumofsquaresofthechildnodescomparedtotheparentnode.ThisishowCARTusesANOVAtoidentifythebestpossiblesplitforeachnodedownthetree.ItshouldbenotedthatthereareothermethodologieswithinCARTumbrellalikerandomforests,baggedorboosteddecisiontrees.Inoursetting,ANOVAfitsbesttobinaryregressiontreesthathaveadiagnosticpurpose.

Inthecontextofbirthregistration,SSTwoulddescribethevarianceinthesumofsquaresfortheshareofchildrenunder5yearsofagewhohavehadtheirbirthregistered.Thepossiblepartitionsofthissample,resultinginSSLandSSR,reliesheavilyonthespecificvariablesselectedtoexplainbirthregistration.Thecharacteristicsusedintheanalysisarewealthofthehouseholdwherethechildrenunder5yearsofagecurrentlyreside(belongingtothetop60orthebottom40ofthewealthdistribution);placeofresidence(urbanorrural);numberofchildrenunder5yearsofageinthehousehold(none(<0.5),1ormore(>0.5),2ormore(>1.5),3ormore(>2.5),fewerthan2(<1.5),fewerthan3(<2.5));ageofthechild(1,2,3,or4yearsofage);sexofthechild(boyorgirl);whetherthechildisanonlychild(0,1);agegroupofthemother(15-24yearsofage,25–34yearsofage,or35-49yearsofage);educationalattainmentofthemother(lowereducation,secondaryeducation,orhighereducation);maritalstatusofthemother(currentlymarried,formerlymarried,ornevermarried)andplaceofdeliveryofthechild(hospitalorhome/other).

WithintheparametersofCART,allpartitionsarebinary;amongcircumstanceswithmultiplesplit

targetandinputattributes,respectively.

13

thresholds,suchasageofthemother,eachpossible

combinationofagegroupingisconsidered(for

example,15-24yearsofageand35-49yearsofage

couldbeasinglecohortassignedtooneofthe

nodes,with25-34yearsofageassignedtothe

other).Thismeanstheresultantnodesassociated

withSSLandSSRaremutuallyexclusiveand

complementary,andeverychildrenbelongstoone

andonlyoneofthenodesforeachpartitionofthe

treegeneration.

Theactualalgorithmthatgeneratesthenodesfor

eachpartitionworksstep-by-step,startingfrom

theentiresample(where“T”givenbySSTisthe

rootnode).Eachtimethesampleispartitioned,

newnodesaregeneratedandtheANOVAis

calculatedandcomparedtotheANOVAbeforethe

newpartition.Thisstepwiseprocessofbuilding

nodesandbranchesofatreeisalsoknownas

“recursivepartitioning.”Eachpartition(andhence

thenewpairofnodes)iskeptwhenthedecreaseof

varianceexceedsapresetthresholdalsoknownas

a“complexityparameter.”Whenthebestavailable

partitionfailstosatisfythecomplexityparameter,

orasetofotherpresetconditionscannotbe

satisfied,thealgorithmceasestomakeadditional

partitions.Thenodesthatcouldnotbefurther

partitionedarehenceforthreferredtoas“terminal

nodes.”

Inadditiontofindinggroupsofchildrenunder5

yearsofagewhosebirthisregistered,theCART

algorithm,appliedtoidentifythefurthestbehind,

requiresthateachgroupshouldhavesufficient

groupmembers.Toavoidatoosmallsub-sample

size,theanalysisceasestomakeadditional

partitionsiftheresultantnodeswouldfailto

containeither9percentofthetotalsample

populationoranabsolutevalueof49survey

respondents.Theanalysisisalsolimitedto6levels

ofnodes(encompassing5partitions),asadditional

partitionspastthatpointhavediminishingreturns

forthepurposeofidentifyingpopulationgroups.

WhenconsideringCARTmodelsincluding

ethnicity,religion,andlanguageofthehousehold

head,theresultantnodesneedtocontainatleast5

percentofthetotalsamplepopulation.This

conditionfacilitatestheCARTalgorithmtodetect

theseminoritygroups.

14

5.RESULTS

ToillustratehowCARTdetermineswhichcircumstancesaccountforthemostvariationinbirthregistration,theexampleofSamoaisusedbelow.Samoa’scivilregistrationsystemlegallyrequiresallbirths,deathsandmarriagestoberegisteredwiththeBirths,DeathsandMarriages(BDM)divisionoftheSamoaBureauofStatistics.ItismandatedthateverybirthmustbenotifiedtotheBDMofficeassoonaspracticalafterthebirth.xi

IntheMICS2019-2020,thesurveyresponseratewere3,196households,4,139womenaged15-49,and2,687mothersorcaretakersinterviewedforchildrenunder5yearsofage.SummarystatisticsfollowingSDGtarget16.9andtheindicator“proportionofchildrenunder5yearsofagewhosebirthshavebeenregisteredwithacivilauthority,byage”highlightthat67percentofchildreninSamoahavetheirbirthregisteredwithacivilauthority.ThisresultalignswithSamoa’sMICSSurveyFindingReport.xiiWhileone-thirdofch