Wellbeing inequality in a developing country from theory to practice

Second,ρisobtainedbymaximisinglfunctionwiththethresholds τi,τjwhichareequaltotheinversecumulativedistributionfunctionoftheobservedproportioninunitkrofthetable Olsson 1979: Basedonthisthe

MARTINO’BRIENUniversityofWollongong,Australia–

martinob@uow.edu.auSILVIAMENDOLIAUniversityofWollongong,Australia–smendoli@uow.edu.au

Abstract

Theemergenceofmultipleconceptofwellbeingthatcanbequantifiedhas allowedresearcherstomovebeyondanarrowfocusonincomeandconsumptio nasaprimarymeasureo f inequalityandpoverty.Althoughanalysesofmultid imensionalwellbeingareincreasinglyfeasibleduetotheavailabilityofdata,th econsumptionorincomeisstillappliedinanumberofstudies.Asaresult,thelit eratureonwellbeingremainsdeficientin twomainways:(1)the useo f inappropriateproxiesf o r wellbeing,a n d (2)ignoranceo f t h e i n t e r

d e p e n d e n c y betweendimensionsofwellbeing.Thispaperdevelopsafund amentalframeworkandappliesaprincipalcomponentanalysismethodforac alculationofthewellbeinglevelandwellbeinginequalityinVietnam.Ourresul

alsoinequality,ofwellbeingincreasedint h e period1993–1998and2002– 2008.Thischallengestheconsensusofamoderatelevel,andstabilityi n , well beinginequalityu s i n g incomeproxiedmeasures.W e arguethatempiricals tudieso f wellbeingneedtoincorporatemultipledimensionsi n additiont o di mensionali n t e r d e p e n d e n c y characteristicandthus,implement ationinthewellbeinganalysesofwellbeingu s i n g principalcomponentanal ysiscanobtaintheuniqueresultsofthelevelandinequalityofwellbeing.

Keywords:inequality;principalcomponentanalysis;Vietnam;household

wellbeing.

1 Introduction

Theconceptofinequalityhasbeenextendedinrelationtomultipledimensionsofwellbeing.Therhetoricquestion‘equalityofwhat?’,raisedbySen(1980),requestsacomprehensiveexaminationofthefacetsof,andproxiesusedformeasuresof,inequality.Theincome-

theutilitycomparisonbecomesu n b e a r a b l e int e r m s of drawin

ga completepictureofinterpersonalwellbeingdifferences(Sen1997).Althoughperceptionofwellbeingvariesamongdifferentcontextsasitdependsonsocialnormsandvalues,thedomainsofwellbeinggobeyondi n c o m e dimension.ThisideaistracedbacktoSen(1985a)’scapabilityapproachthatemphasizesnotonlywhatpeoplehave,butalsotheextenttowhichtheyarefreetodoandtobe

Ageneralperceptionofwellbeingisanythingmakingagoodlife(Deaton2013,p.24)despiten o consensuso n t h e c o n c e p t o f wellbeinghasbeenr e a c h e d ; differencesi n wellbeingachievementsneedsanassessmentof‘wellness’ofpeople’sstateofbeing(Sen2003b,p.36).Indevelopedcountries,aplethoraofstudiesfocusonwellbeingachievementso r outcomeswitha rangeo f indicatorsincluding,butn o t limitedto,

i n c o m e and

wealth,healthandfeeling,educationandsocialengagement(Deaton2013,p.24).Forinstance,OECD(2013)chooseselevenindicatorswhichrepresentsplausibledimensionsofwellbeingwhichspecifiestheeightcomponentssuggestedinStiglitzetal.

(2009).Indevelopingcountries,however,itisinsufficientlyconcernedw

ithresearchinwellbeingandinequalityinwellbeing(Cho2015)

Althoughdataavailabilityforassessmentsofinequalityinwellbeinghasemerged,seriousproblemswithempiricalanalysesremains.First

adistortedpictureo f wellbeingwhichcomprisesa varietyo f factors.Forinstance,Vietnamshowsconfusinglevelsofinequalitywithincomeandexpenditureindicatorswhichareequallyacceptedasproxiesforwellbeing(Zhuangetal.2014).In2008,therewasasignificantgapbetween

theGinicoefficientofincome(0.44)andthatofexpenditure(0.36).Additionally,t h e developmento f measureso f inequalityi n m u l t i

p l e dimensionsdisentanglest h e ambiguitieso f proxiedmethod.R a t h e r , itfurtherprovidesconfusingandinconclusiveresultscorrespondingtovaryingchoicesofparametersusedandthesequenceofaggregationacrossdimensionsintheestimations

singleindicator-40 |

ICUEH2017

ofinequality(e.g.inequalityaversion),evenwiththesamemethodsanddatasets(e.g.N i l s s o n 2010,Justino2012)

Finally,theassumptionofnon-interdependenciesbetweenvariablesisalsoviolatedasthereisplausibleevidenceofinterrelations betweeneconomic, educationandhealthi n d i c a t o r s Russiandata,f o r example,r e v e a l s a complexityo f interrelationsacrossdimensions.DecancqandLugo(2012)f

o un d that thiscorrelationstructurechangedremarkablyintheexaminedperiod(1995–

2005).Unfortunately,theycouldnotscrutinisethusfarbecauseofthepaucityofdatawhichiscommoninthemajorityofcountries.Therefore,theextentt o whicht h e indicatorweightsi n a computationo f inequalityhavenotbeenthoroughlyexamined.Apostulationofequalvariableweighti s , however,inconsistentwiththefactthatvariablesmayhaveunequalinfluencesonwellbeing,andthusonthelevelofoverallinequality

Tofillthesedeficiencies,thispaperdevelopstheanalyticalframeworkbasedonSen’sc a p a b i l i t y approachandappliesthepolychoricprincipalcomponentanalysis(PCA)toVietnamesedata.Indoingso,thisresearchcontributestotheliteratureonwellbeinginequalityintwomainways.First,itextendstheanalysisofwellbeing

byincludingthevariouswellbeingcomponents.W e considerthec o n t

r i b u t i o n o f n o n

-e c o n o m i c dim-ensionsandth-eint-eractionsofallindicatorstoanov-eralltrendinmultidimensionalinequality

Furthermore,wecomparewellbeinginequalitytrendsover

timeandacrossdifferentgeographicalareas

Theremaindero f t h e paperproceedsasf o l l o w s Then e x t sectiondiscussesthec a p a b i l i t y approach.Themethodology,dataandvariablesaredescribedinSection3.S e c t i o n 4comparesthewellbeingindexconstructionbasedonthepolychoricPCAtotheincome-

proxiedwellbeing.Section 5analysesinequalityinwellbeing.Theconclusivepartinvestigatesfurtherstepstoidentifymajorcausesofinequality

2 Thecapabilityapproach

Capabilityindicatestheindividualabilitytoobtainrealachievementsinrelationtoe x t e r n a l andinternalconditionsthatinfluencepersonaltransformingfromcommoditypossessiontopersonalwellbeing(Sen1985b).Despiteeconomicdimensionisanimportantcontributortowellbeing,itcouldnotcapturealldeterminantsofqualityoflifethat

peopleareabletodoorliveintheirfavouredways

Sen(1985b)introduceshisownapproachtomakethe individualwellbeingc o m p a r a b l e t h r o u g h a relationshipb e t w e e n t h e tw

–

Qi(Xi).Xireferst o allk i n d s o f resourcesandi s subjecttot h e personalbudgetconstraint

Thecapabilityapproachistotallyanddirectlyoperationalwithrespecttofreedomofc h o i c e s KuklysandRobeyns(2006)appreciateitsextendablecharacteristicbyaddingplausiblefunctioningssuchas‘beingeducated’and/

or‘beingemployed’.Dang(2014)supportsaconcentrationontheachievedfunctioningswhicharemoreoperationalthano n asetofcapabilitiesinthecaseofinadequateinformationonfreedomconditions

Despitedifficultiesofdatacollectionrelatingtoachievedcapabilitiesorfunctionings,i n practice,thedataandvariablesuitabilityandanalyticalmethodsassociatedwithsuchdatashouldbethoroughlyexamined(Sen2003a,p.53).Senhoweverdoesnotprovidea specificdiscussionbeyondt h a t point,butherecommendsa solutiontochoosingdimensions,i n d i c a t o r weights,andcalculationmetrict h r o u g h a democraticpublicdecision.Wellbeingevaluationsshouldconsiderwithr e s p e c ttosocialandhistoricalcontexts(Jackson2005,Qizilbash2011)

t dwellonafixedsubsetofcapabilities,butproxiesforcapabilitiesshouldbeadjustedi n favouro f r e s e a r c h perspectives.‘Thecapabilityapproachcano f t e n yielddefinitea n s w e r s ’ ofthelevelsofindividualwellbeinginsuchacase(Sen 2003a,p.46).

Finally,inequalityshouldbeassessedasa‘failureof[anddifferencesin]certainbasicc a p a b i l i t i e s’(e.g.beingnourished,beingsheltered)respectivelya s incomeandpreferencearenotaccuratemeasurementstocomparinginterpersonalwellbeing(Sen1 9 8 5 a , 2006).Thereis

nothingmathematicallywrongwiththemeasurementsofinequalityderivedfromincome,‘but[to]interpretthemasutilitycomparison…wouldbeacompletenonsequitur1’(Sen1997,p.392)

3 Methodology,data andvariables

3.1 Thepolychoricprincipalcomponentanalysis

WechoosethepolychoricPCAdevelopedinKolenikovandAngeles(2004,2009)toanalysewellbeingandinequalitybecauset h i s modifiedPCAi s superiortoi t s n a ï v e version.ThestandardPCAisoriginallyconstructedtohandlenon-discretevariables.AnapplicationofPCAtothenon-continuousdatamayhaveproblems.First,ifonebreaksac a t e g o r i c

a l variableintomorethantwodummies,PCAcouldcreatenumerousspuriousc o r r e l a t i o n s Second,atransformationfromordinalvariablestodummiescannotretaint h e ordinalfeatureofindicators.Moreimportantly,ifcategoricalvariablesaretreatedascontinuousones,aviolationintheassumptionofanormallydistributedvariableinPCAo c c u r

s analogouslyt o t h e c a s e thatdiscretevariablesareusedasindependentvariablesinOLSsincediscretevariablesdonothaveadensitybuthighskewnessandkurtosis(KolenikovandAngeles2004,2009).ThepolychoricPCAminimisesviolationo f anormal

distributionassumptionwhenappliedto discretedata ThepolychoricPCAc a n alsoassignvariousweightsfordifferentu n i t s andcategorie

so f indicatorsanddescribemorepreciselywellbeinginequality(Ward2014)

1 n o n

sequitur:‘astatementthatisnotconnectedinalogicalorclearwaytoanythingsaidbeforeit’(Merria m-Websterdictionaryn.d.).

CorrelationcoefficientsinthepolychoricPCAaredescribedinthefollowingsteps.F i r s t , two ordinalvariablesxi,xjindicateassetownership,educationaloutcomes,orhealthstatus.Theyarediscretisedindkcategories(k=1…m),anddrcategories(r=1 …

n ) respectively.Thus,thethresholdsofxi,xjaredenotedasτi,τjcorrespondin

polychoric

correlationcoefficientρ.

Second,ρisobtainedbymaximisinglfunctionwiththethresholds

τi,τjwhichareequaltotheinversecumulativedistributionfunctionoftheobservedproportioninunit(kr)ofthetable (Olsson 1979):

Basedonthistheoreticalframework,Ward(2014)resolvesthefactorloadingofvariable

xicorrespondingtothecategorydk:

o wellbeingandt h e r e f o r e , c a n beomitted.A constructiono f wellbeingindexi s asf o l l o w s :

𝑋 dkn

�ℎ= ∑𝑖=1∑d

k(j=0)βi⃓dkj.� �( i⃓dkj

)where:

basedmethodcanconsiderplausibleinterdependencyofvariables.Thefactorloadingsarec a l c u l a t e d basedonvariouscategoriesandunitsofavariable.Thischaracteristicfurthere n a b l e s ustoestimateinequalityinwellbeingwithvaryingquantitiesofasseto w n e r s h i p s, variouslevelso f educationalachievement,anda wider a n g e i n h e a

l t h status.Third,acomputationofdifferentweightsismorereasonablethananallocationo f unifiedweightforallvariables.WiththesePCAadvantages,thisstudygoesbeyondi n c o m e inequalityandanalysest h

e importanceo f differentf a c t o r s c o n t r i b u t i n g t o inequality

Thesecondroleofvariableweightsindicateswellbeinginequality.Thehighertheweights,thegreatertheshareofvariableinthetotalvarianceofthefirstcomponentofPCAandt h u s , i t showsthegapi n wellbeingdistribution.I f a v a r i a b l e hasa m i n o r standarddeviation,itisassignedasmallweightinPCAintheinequalityindex.AccordingtoMcKenzie(2005)andWard(2014),themeasurementofinequalityusingPCAisasfollows:

λisthefirsteigenvaluefromthecorrelationmatrixinPCAandalsothevarianceheldbythefirstcomponentacrossthewholepopulation

3.2 Data

ThisstudyusesdataextractedfromtheVietnamHouseholdLivingStandardSurveys(VHLSSs)whichcombinet h e retrospectivei n f o r m a t i

o n abouthouseholdst h a t haveparticipatedinthe previouswaveandaboutthefirst-

timeadditionalparticipants.A longitudinaldatasetthatiscomposedofmorethantwowavessubstantiallydecreasest h e numberofobservations,andthismightcausemeasurementerrors.Additionally,nohouseholdsinterviewedinthe1990st o o k partinthelatersurveyinthe2000s.Therefore,thispaperdoesn o t generatepaneldata,butusespooled1993–1998and2002–2008 data

SincethesizeoftheVHLSS2002isthreefoldthesizeofanyotherwave,arandomsampleo f 3 1 % o fi t s totalobservationsi s c r e a t e d wit

ha r e m a i n i n g proportiono f observationsbetweenprovincesandurban–

rural.Thistechniqueofdatacombinationc o m p u t e s uniqueweightsforordinalandcardinalvariablessothatthewellbeinglevelandinequalitycanbecomparableacrosshouseholdsinter-temporally

3.3 Variables

Twenty–oneandtwenty–

fivevariablesareusedasproxiesforwellbeinginthe1990sand2000srespectively.Regardingchanges inwellbeingstandards,thissectionpredetermineswhethervariablesused fortheperiod1993–1998remainsensibleinthef o l l o w i n g decade.Indicatorswithallcorrelationcoefficientslowerthan0.1are

consideredas‘nocontaininginformation’(MoserandFelton2007)andexcludedfromt h e model.

Assetindicators

Despitenoconsensusonastandard principle,severalguidelinesforvariablechoicesarementioned.Assetsneedtobechosencarefullyasseveralassetsmightrepre se nt prosperityinthepastbutpovertyatpresent(RutsteinandJohnson2004).Anincreasedn u m b e r

o f assetscanalsor a i s e householdcapabilities(Ward2014).Aswellbeingi s m u l t i d i m e n s i o n a l , thesevariablesmustbeplausibletoavoidbiasesinrankingindividualwellbeing

ThemagnitudeofcoefficientsonoriginalvariablesgeneratedbyPCAdependsonhowtheproportionoftotalinformationiscapturedbythecorrespondingindicators;greatercoefficientmeansthatthevariableismoreimportantcontributiontowellbeing.ThisisanadvantageofthePCA-

basedmeasurementofinequalitybecausethechoseni n d i c a t o r sindicaterelatedvariablesthatdescribewellbeingirrespectiveofwhethertheyarepresentinthemeasurement

Theextenttowhichtheroleofassetownershiphaschangedcouldbeexplainedbysocioeconomicconditions.WhentheaverageincomewasbelowUS$300inthe1990s(WorldBank2013a),Vietnamesepeoplecouldexpecttopossessaradio,oraclockase s s e n t i a l thingso f anacceptablelivingstandard.Howe ve r, theseassetsbecomelessimportanti n t h e twenty–

firstcenturyashouseholdso w n morevaluableitemst h a t generateidenticalutilities(suchascolourTVs,orwristwatches).Therefore,alistof

i n d i c a t o r s shouldberevisedovertimetoavoidanyinappropriateproxyforwellbeing

Furthermore,thisanalysisalsoconsidersthequantitiesoftheassetsownedbythehouseholds,asthesedifferencescouldbeofimportancetomeasureinequality.Ward(2014)clarifiesthataconsiderationtoassetquantitycanraisetheeffectivenessofthepolychoricPCAregardingtherankingsofhouseholdwellbeing

Educationalindicators

Educationisavitalwellbeingdeterminantbecauseknowledgeandexperiencenoto n l y reflectthehouseholdachievedfunctioningsintheeducationaldimensionitselfbutalsoinfluenceotheraspectsofwellbeing

Thisstudyselectstwovariablesasaproxyfortheeducationaldimension(butonlyo n e forthe1990s duetodataunavailability).Insteadofthehouseholdhead,thepersonwiththehighesteducationalattainmentinthehouseholdiscollected.Thereasonfort h i s isthathouseholdwellbeingcouldbeaffectedbythememberswiththehighestlevelo f educationnotonlywithintheeducationaldimensionbutalsointheeconomicandhealthaspects.InVietnam,thereisasignificantgapbetweenparentsandtheiroffspring’sschoolingbecausehouseholdheadswerelikely tohavefinished schooling e a r l y , buttheyencouragetheirchildren’sstudieseventhoughtheyareclassifiedinthepoorstratum.

Housingandhealthrelatedvariables

Thehousingvariablesusedinc lude housingcharacteristicsandhousingfacilities.Thesevariablesc a n providein fo rm a t io n o n boththequalityofaccommodationandotherconditionsrelatedtothehealthdimension.Theearlierindicatorrefersthetypeso f materialusedtohous

housingfacilitiesreflectthequalityofbasicservicesconsumedbyhouseholds(e.g.drinkingwater).Aconsiderationo f housingindicatorsi s f

o u n d i n studiesi n inequalityunderpinnedS e n ’ s capabilityapproach(KuklysandRobeyns2006,p.46)despitedifferentchoicesofhousingvariablesr e g a r d i n g wellbeing.McKenzie(2005)usesthenumberofrooms,houseownership,andt h e qualityofwallsandroofs,asproxiesforthehousingdimension;KuklysandRobeyns(2006)choosetheindicatorswhichinvestigatewhetherahouseholdhasproblemsof(water)condensation,rottingwood(windowsorfloors),keepingthehomewarmand

t h e housecapacity.MoserandFelton(2007),andWard(2014)addlightingsources,andt o i l e t typesinthisgroupandclassifythemas‘housingcapital’.Inthispaper,thehousingcharacteristicsvariablesarecombinedwithassetindicators

Additionally,housingfacilitiesareinterpretedasproxiesforthehealthdimensionbecauset h e s e variablescansignificantlyimpactindividualphysicalwellbeing.F o r example,usingsafedrinkingwatermayreducetheprobabilityofseveralinfections.Unfortunately,thistypeofinformationisunavailableinthe1990swavesandtherefore,t h e numberofsickdaysoveramonthisusedasaproxyforthefamilyhealthstatus

Refrigerato rAirconditio nerW a t e r heater Washingmac hineGascook erElectriccoo ker/

stoveElectric generator

Waterpu mpPerso nalcomp uterPrint erCamer a

Vacuumcleane r

Educational achievements

Schooling years (intheofficial universaleducational

system,fromthepre-schoolleveltograde12)oft h e mosteducated memberofahousehold.

Highesteducationalqualificationachievedbythemosteducated individualof ahousehold.H e a l t h relatedindicators

Drinking water:ordinal variable(rangingfrom1

to5)indicatesthequalityofwatersourceforthedrinking purpose.

Toilet: ordinalvariable(rangingfrom 1to5)reflectsthetypeoftoilet used.

Garbage:ordinal variable(rangingfrom1 to4)expressesthekindofrubbish

disposal.

4 The household wellbeing level: non-monetary

vis-à-vismonetaryindicator

Tovalidatethewellbeingindexandtheanalysisofwellbeinginequality,thispaperm a k e s a comparisonbetweent w o proxieso f wellbeing:n o n -

m o n e t a r y (wellbeing)i n d i c a t o r s estimatedbythepolychoricPCAandamonetaryvariable–

theconsumptionexpenditure.ThetheorybehindthiscomparisonisSen’s(1985a)capabilityapproach.Consumptionexpenditureisa c o

m m o n proxyf o r economicr a t h e r t h a n m u l t i d i m e n s i o n a l wellbeing.Senarguesthatmoneyisameansbutnotanend(outcomeo f wellbeing);thus,useofthemonetaryvariablecouldbemisleadingbecauseincome‘ g i v e s averyinadequateandbiasedviewofinequalities

’(1997,pp.384-385).Wea d v o c a t e thisargumentthatusingexpendituredataforanal

12

ysesofVietnamesepeople’swellbeingisinappropriate.Incontrast,thenon-monetaryindicatorsconsiderthelevel

ofwellbeingandt h e contributiono f variousdimensions,addingbutnotlimitedtoi n c o m e , towellbeing.

Table2describesarelationshipbetweenhouseholdexpenditureandthewellbeingindexderivedbyEq

(7).ThetwomethodologiesusedforthiscomparisonarethewithinquintilerankingconsistencyandtheSpearman-

rankcorrelationtechnique.Thepopulationineachwaveofsurveyisclassifiedinfivequintilesbasedonthehouseholdexpenditure,andthehouseholdwellbeingindicatorrespectively.Then,anidentificationm a t c

h i n g techniqueisusedtorecordthepercentmatchedwithinthesamequintilesbyt h e s e twomethods.Thesecondcolumnshowsthatthelevelofconsistencyisaround4 0 % acrossthesefourwaves.Thismeanst

h a t about6 0 % o f t h e wellbeinglevelmeasuredbyt h e

non-monetaryapproachm a y n o t bec o v e r e d byt h e expenditurevariable.Thet h i r d columnr e v e a l s therankingc o r r e l a t i o n between

t h e householdexpenditureandthewellbeingindex.Compared

withthewithin-quintilematching,theSpearman-rankcorrelationmethodi l l u s t r a t e s c l o s e r andsignificantcorrelationcoefficientsbetweenthetwoproxies2.Thistechniqueindicatesthatovertwo–

thirdsofhouseholdsareconsistentlyr a n k e d byt h e wellbeingindexandt h e householdexpenditure

Resultsofamismatchbetweenthelowestquintiledefinedbythewellbeingindexandt h e highestoneidentifiedbytheexpendituredata,an

dviceversa,arereportedinthelasttwocolumns.Thepercentofmism

atchedhouseholdsbetweentwomethodsarenegligible

2 All

resultsestimatedbytheSpearmancorrelationcoefficientsaresignificantatthe1%level.

|14

Aconsistencybetweenthehouseholdexpendituredataandthewellbeingindexinhouseholdclassification

Year Perce

ntmat ched

Spearma ncoefficie nton

Wellbeinglo westquintile/

Income

Incomelowestqui ntile / within ranking

consistency

highest quintile 3

quintile quintiles

5

2.3 7

0.00

1

0.41 Source: VHLSS 2002–2008;authors’estimation

Anotherwaytochecktherobustnesstothewellbeingindexistocomparethetrendsi n wellbeingusingthetwoproxies.AsVietnamdidnotexperienceanynotablecrisis(e.g.politicalconflictoreconomicshock)whichnegativelyaffecttheeconomic,health,oreducationaldimensions,theoverallwellbeinglevelisexpectedtoincreaseintheperiod2002–2008

Withthe pooledVHLSS2002–

2008data,valueso f wellbeingarecalculated.ThepolychoricPCAproducest h e zero-

meanaggregatewellbeingvaluesforthewholesample;hence,thewell

itsabsolutevalues.However,itisevaluatedrelativelythroughitsvariations.Table3showsrises

int h e meanofwellbeingforthewholecountryandtwoselectedregions.Thenationalwellbeinglevel,placedinthefirstcolumn,increasedspanning2002–

2008.ThistrajectoryiscompatiblewithsocioeconomicprogressinVietnamesesociety.Thistendencyisconsistentwithchangesinhouseholdrealexpenditurepresentedinthelastcolumn

Table3alsoillustrateswellbeingmovementsintwodistinctregions.Changesinw e l l b

e i n g inboththetworegionsfollowthetrajectoryofnationalwellbeing.Thise v i d e n

c e confirmsthat thewellbeingindicatorgeneratedbythepolychoricPCAisa goodproxyusedforananalysisofinequality.Thiscomparisonshowsana n a l o g o u s improvementinwellbeingfoundinWard(2014).Theseresultsfurther

3 This

fractionisestimatedbythematchingtechniquethatexpresseshowmanypercentofhouseholdscategoriseda

s highestexpenditurequintilebutasbelongtothepoorestquintileinwealthmeasuredbytheassetindic atora p p r o a c h