Thank goodness for stickiness

(1)

Contents lists available atScienceDirect

Journal

of

Health

Economics

j o u r n a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / e c o n b a s e

Thank

goodness

for

stickiness:

Unravelling

the

evolution

of

income-related

health

inequalities

before

and

after

the

Great

Recession

in

Europe

Max

Coveney

a,b,c,∗

_,

_Pilar

_{García-Gómez}

a,b,c

_,

_Eddy

_van

_Doorslaer

a,b,c,d

_,

Tom

Van

Ourti

a,b,c

a_Erasmus_School_of_Economics,_Erasmus_University_Rotterdam,_P.O._Box_1738,₃₀₀₀_DR_Rotterdam,_the_Netherlands

b_Tinbergen_Institute

c_NETSPAR,_the_Netherlands

d_Erasmus_School_of_Health_Policy_and_Management,_Erasmus_University_Rotterdam,_P.O._Box_1738,₃₀₀₀_DR_Rotterdam,_the_Netherlands

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received22May2018

Receivedinrevisedform

11November2019

Accepted14November2019

Availableonline9December2019

Keywrods: Healthinequalities Decompositionanalysis Europe Economiccrisis

a

b

s

t

r

a

c

t

TheGreatRecessioninEuropesparkedconcernsthatthecrisiswouldleadtoincreased incomerelatedhealthinequalities(IRHI).Didthiscometopass,andwhatrole,ifany,did governmenttransfersplayintheevolutionoftheseinequalities?Motivatedbythese ques-tions,thispaperseeksto(i)studytheevolutionofIRHIduringthecrisis,and(ii)decompose theseevolutionstoexaminetheseparaterolesofgovernmentversusmarkettransfers. Usingpaneldatafor7EUcountriesfrom2004to2013,weﬁndnoevidencethatIRHI per-sistentlyroseafter2008,evenincountriesmostaffectedbythecrisis.Ourdecomposition revealsthat,whilethehealthofthepoorestdidindeedworsenduringthecrisis,IRHIwere preventedfromincreasingbytherelativestickinessofoldagepensionbeneﬁtscompared tothemarketincomesofyoungergroups.AusteritymeasuresweakenedtheIRHIreducing effectofgovernmenttransfers.

Introduction

Itiswellknownthatthosewithhigherincomesenjoy longerandhealthierlivesthanthosewithlowerincomes. Suchinequalitiesarewidespreadandpersistent;theyexist acrossvirtuallyallcontexts,measuresofhealthand socio-economic status. A renewed focus on inequalities has culminatedincallsfromboththeCenterforDiseaseControl

∗ Correspondingauthorat:ErasmusSchoolofEconomics,Erasmus

Uni-versityRotterdam,P.O.Box1738,3000DRRotterdam,theNetherlands.

E-mailaddresses:coveney@ese.eur.nl(M.Coveney),

garciagomez@ese.eur.nl(P.García-Gómez),vandoorslaer@ese.eur.nl

(E.vanDoorslaer),vanourti@ese.eur.nl(T.VanOurti).

intheUSandtheEuropeanCommissiontoreduce dispar-itiesinhealth(CDC,2013;EuropeanCommission,2009a).

Partofthisrenewedfocusonhealthinequalitiesmay beattributabletotheGreatRecession.Inthewakeofthis crisis,EU policymakers expressedconcerns that socio-economicdisparitiesinhealthwouldbeexacerbated,i.e. that the health and socioeconomic status of the most vulnerablemembers of society mightbe disproportion-ally hit (European Commission, 2009a).1 _Despite _these

1 _In_2009,_for_instance,_the_European_Commission_warned_that_“the

currenteconomiccrisiscan(...)increasehealthinequalitiesthrougha

dete-riorationofsocialdeterminantsofhealth,especiallyforthosewithlower

qualiﬁcationsandsavings,andwhoare alreadyvulnerable”(European

Commission,2009b,14).

https://doi.org/10.1016/j.jhealeco.2019.102259

(2)

concerns,and thejustiﬁedalarmaboutthepotentialfor deepened disparities,comprehensive cross-country evi-denceonchangesinthedistributionofhealthbyincome before,duringandafterthecrisisislacking.2

Additionally, evidenceis also lackingon therelative importanceof(changesin)differentincome sourcesfor (changesin)income-relatedhealthinequalities(IRHI).3_We

distinguishbetweenthetwo mostimportantsourcesof income:market incomes (like wages), and government transfers(likeold-ageand unemploymentbenefits) and separatetheirinfluenceonIRHI.Changesinthesedifferent incomesourceshaveplausiblydifferingIRHIconsequences. Firstly,becausetheirdistribution acrossageand health groupsdiffers,andsecondly,becausetheytendtovaryin oppositedirectionsintimesofrecessionandgrowth.For instance,Jenkinsetal.(2012)showthatbetween2007and 2009,householdsinthecountriesmostaffectedbythe cri-sisincreasinglyreliedonincomefromsocialbenefitsand governmenttransfers,claimingthat“protectionof house-holdincomesagainstthecollapseofeconomicactivityduring theGreatRecession(GR)waslargelyprovidedbythe govern-ment”(p14).

Thedistinctionbetweenmarketandgovernment trans-fersisalsoimportantbecauseofitsimplicationsforpolicy; governments are able to manipulate transfer payments moredirectlythan,forexample,wages.Thecrisisinduced heterogeneouslabormarketeffectsacrossnearlyall Euro-peancountriesandgovernmentsrespondedwithavariety ofausteritymeasures,primarilyrelatingtounemployment andpensionbeneﬁts.Ifthereisadistinctroleoftransfer incomeforIRHIchanges,thenitisimportanttoshedlight onsomeofthe possiblyunintended IRHIconsequences thatpoliciesgoverningthesetransfersmayhavehad.

Theaimofthispaperistoaddressthesegaps inthe literature.Our contributionsarefourfold.Firstand fore-most,wepresentanoveldecompositionmethodthatseeks toidentifytheverydifferentrolesplayedbygovernment transfersandmarketearningsfortheevolutionofIRHI. Sec-ondly,weusecomparablelongitudinaldatatodocument trendsinIRHIin7Europeancountriesbetween2004and 2013–bothbeforeandaftertheﬁnancialcrisis.Third,by meansofthedecomposition,weunravelthemost impor-tantdriversofIRHItrendspre-andpost-crisis.Lastly,we providedescriptiveevidenceontherolethattheausterity measuresenactedinGreecehavehadonIRHI.

We add to theliterature using concentration index-typemeasurestocomparehealthinequalitiesbyincome acrosscountries(VanDoorslaeretal.,1997;VanDoorslaer andKoolman,2004).TheseandotherEuropean compar-ative studies report substantial pro-rich inequalities in healthinEurope.Descriptivedecompositionmethodshave

2 _{Ásgeirsdóttir}_and_{Ragnarsdóttir}₍₂₀₁₃₎_study_differences_in_IRHI_for

26Europeancountriesin2007.However,thiscross-sectionalapproach

isuninformativeabouttheevolutionofIRHIbefore,duringandafterthe

crisis.

3 _We_estimate_IRHI_using_disposable_household_income_corrected_for

householdcomposition.EstimatesofIRHImightdifferwhenother

indica-torsofsocioeconomicstatusareused.WerefertoWagstaffandWatanabe

(2003);VanOurti(2003);Lindelow(2006)andFrickandZiebarth(2013)

formoredetails.

beenusedtoexplain howsuchinequalitiesevolve over timewithincountries–thephenomenonofinteresthere. Allansonet al.(2010)develop and applythemost gen-eralsuchdecompositionbyseparatingchangesinIRHIinto thoseduetochangesintherankingvariable,andthosedue tochangesinthehealthvariable.However,thisapproach doesnot allowfor a more detailed investigationof the driversofIRHIsuchastheseparaterolesofgovernment andmarkettransfers,orhealthandincomevariationacross age,genderandplace.

Byimposingfurtherstructureaﬁnerdissectionofthe changesinIRHIispossible.Forinstance,inastudyofIRHI evolutioninEuropeinthe1990s,VanOurtietal.(2009) employanextended decompositiontoinvestigateunder what conditionsincomegrowthcanleadtoa reduction of IRHI. Baetenet al. (2013) make a further distinction betweenIRHIchanges stemmingfromincomeand non-incomefactorsinChinainthe1990sand2000s.Asimilar approachistakenbyCoveneyetal.(2016),whostudyIRHI changesinSpainbetween2004and2012.Theyﬁndthat IRHIwasprimarilydrivenbytheincomepositionofthe relativelyunhealthyelderlygroups,whotendedtomove down(up)theincome ladderin“good”(bad) economic times.

While the ﬁndings of Coveney et al. (2016) hint at thedistinctrolesplayedbygovernmenttransfersversus incomefromlabour,themethodsuseddidnotexplicitly separatetheirrolesforIRHIevolution.Motivatedbythis, we derive a newdecomposition method that explicitly incorporatesthisdistinction.Further,studyingarangeof Europeancountrieswithdifferinglevelsofexposuretothe crisisaswellasarangeofdifferenttransferpolicies pro-videsfurtherinsightsintothedeterminantsoftrendsin IRHI.

We also address a limitation faced by many previ-ousdecompositionstudieswhereanordinalSAHmeasure - often the only measure of overall health available in cross-countrypaneldatasetswithsufﬁcientlygoodincome information-mustbetransformedintoaratio-scale mea-sure.Theusualapproachistouseanintervalregression to generate a deterministic ratio-scale health measure, and to use the estimated partial associations between the covariates -including income - and theunderlying latenthealthvariabletosplitIRHIchangesstemmingfrom incomeandnon-incomefactors.However,the determin-istichealthmeasurewillnaturallymisssomevariationin theunderlyingordinalSAHhealthvariable,andchanges in the income distribution codetermine the predicted change in IRHI. Instead we propose a two-step proce-dure where we ﬁrst predict a ratio-scale measure that is conditional on the underlyingSAH category but not income,andnextuseittodecomposeIRHI.Thisovercomes both limitations and explicitly accounts for the health changes(andhowtheydifferbyincome)thatarenot cap-turedbytheregressionmodelusedinourdecomposition method.

Wedonotaimtoaddtotheliteraturethatstartedwith Ruhm(2000),linkinghealthandeconomicconditions, aim-ingtoidentifya causaleffectof thecrisisorincomeon health. Rather,ourdecompositionillustratesand unrav-elstheassociationbetweenincomeandhealth,andthus

(3)

IRHI,beforeandafterarecession.Ourapproachalsodiffers fromthecross-countrycomparisonsofMackenbachetal. (1997;2008;2018,amongothers),whichdocumentlevels andtrendsinsocio-economicinequalitiesinhealth(mostly education-andoccupation-related)foralargecrosssection ofEuropeancountries.Ourapproachoffollowingcohorts overtimehastheadditionaladvantagethatonecan distin-guishbetweenhealthy/unhealthyindividualsmovingup or downtheincome ladder, andpoorer individuals fac-ingworse/betterhealthprospectsthanricherindividuals.4

Tracing both ofthesemovementsis essential to under-standingtheevolutionofIRHI,andtoidentifyappropriate policylevers tomitigate theIRHIconsequences of eco-nomicdownturns.For example, publichealth insurance programsmayprotectthepoorfromhealthdecline,while programssuchasdisabilityinsuranceprotectthe(relative) incomesofthoseinpoorerhealth.

Ourfindings areasfollows.First,wedo notfindany lastingIRHIincreasessincethestartoftheGreat Reces-sion,evenincountriesthatwereseverelyaffectedbythe crisis.DocumentingannualIRHIchanges across7 Euro-pean countries between 2004 and 2013, we find that theseinequalitiestendedtoremainattheirpre-crisislevel in the years following theonset of therecession, even while economic conditions worsened. Before thecrisis, a time of relativelysteady economic growth inEurope, IRHIwasonaveragerelativelyflat,thoughitrose signif-icantly in somesouthern countriessuchas Greece and Italy.

Second,ourdecompositionofthesechangesrevealsthat theﬂatpost-crisistrendinIRHIcanlargelybeattributed to two factors working in opposite directions. First, in thepost-crisisperiodpoorerindividuals begantoreport worseninghealth.ThisledtoanupwardpressureonIRHI. Secondly, and simultaneously,therelativeincome posi-tion oftheelderly tendedtoimprove, astheirincomes weremainlybasedonmuchstickiergovernmenttransfers (pensions)thatwererelativelyunaffectedbythecrisis.As theelderlytendtobeinrelativelylowerhealththishad adownwardpressureonIRHI.Thecrucialdampeningrole playedbypensionshasensuredthatnosubstantialrisein IRHIhasoccurredasaresultoftheGreatRecessioninmany EUcountries.

Third,weﬁndthatdifferencesinhouseholdstructures acrosscountrieshaveIRHIimplications.Namely,theIRHI reducingeffectofpensionsislimitedincountrieswhere theyoungtendtostayathomelonger,andthussharethe pensionincomesoftheirelderlyrelatives.

Fourth,weshowthattheausteritymeasuresenactedin Greecein2010and2011–whichincludedpensionreforms -appearstohaveunderminedtheIRHIdampeningeffect ofgovernmenttransfers.Thisoccurredduetothefactthat theausteritymeasuresreducedtheredistributiveeffectof transfers,especiallypensions.

4_Kreiner_et_al.₍₂₀₁₈₎_,_for_instance,_demonstrate_that_the_income

gradi-entoflifeexpectancyinDenmarkandtheUSishalvedifaccountistaken

ofincomemobilityovertime.

Decompositionofchangesinincome-relatedhealth inequality

Our decomposition buildson methods developed by VanOurtietal.(2009),Allansonetal.(2010)andBaeten etal.(2013).Inthissectionwedescribetheapproachfor abalancedcohortofnindividualsthatweobserveatthe start(period1)andend(periodT)ofagiventimeinterval. Healthinequalitymeasurement

Tomeasure healthinequalities weusethecorrected concentrationindex(CCI)(Erreygers,2009)whichsatisﬁes theprincipleofincome-relatedhealthtransfers,the mir-rorcondition–i.e.thatSES-healthdistributionsshouldbe rankedsimilarlywhenhealthisexpressedasattainment or shortfall – and is insensitive to equal health addi-tions(absoluteinequality)(ErreygersandVanOurti,2011). Whenhealthisboundedbetween0and1,theindexcanbe writtenas: CCI (htyt)= 8 n2 n

i=1 zithit (1)

wherehtandytarethehealthandincomedistributionin

periodt=(1,...,T ),hitdescribesthehealthlevelof

indi-vidualiwhich we assumeis measuredona ratioscale andnonnegative.zit isaweightthatdependslinearlyon

the income rank of individual i in period t with indi-vidualsranked frompoor(i=1)torich (i=n),i.e. zi=

(2i−n−1) /n.Thisincomeweighttakesthevalue0for theindividualwithmedianincome,andincreaseslinearly withincomerank.

ChangesinIRHI

OurinterestliesinmonitoringchangesinIRHI.Taking thechangeintheCCIbetweenperiod1andtheﬁnalperiod T ,weobtain: CCI (hTyT)−CCI (h1y1)= 8 n2

_n

i=1 ziThiT− n

i=1 zi1hi1

(2) Eq.(2)masksthattherearetwounderlyingdriversof changesinIRHI,i.e.changesinincomeranksandchangesof healthovertime.Bothchangescanbeisolatedafter rewrit-ingEq.(2)followingthelogicusedinAllansonetal.(2010): CCI (hTyT)−CCI (h1y1) = 8 n2 n

i=1 [(ziT−zi1) hiT+zi1(hiT−hi1)] (3)

Eq.(3)showsthatthechangeofIRHIequalsthesumof health-relatedincomemobility (ﬁrstterm) and income-related health mobility (second term). Health-related incomemobilityfocusesonchangesintheincome distri-butionandmeasurestheextenttowhichthosethatmove upordowntheincomeladder(ziT−zi1)endupwithgood

healthintheﬁnalperiod(hiT).Sincehealthisnonnegative,

(4)

whenindividualswithbetterhealthclimbed(descended) theincomeladder.Instead,income-relatedhealth mobil-ityfocusesonthehealthdimensionandrevealswhether healthgainsorlosses(hiT−hi1)areconcentratedamong

initiallyricherorpoorersubgroups(zi1).Bothaspectsare

essentialtounderstandingthechangeofIRH5_.

DecompositionofIRHIchanges

However,Eq.(3)is uninformativeabouttheseparate rolesofdifferentincomesourcesforthechangeinIRHI, whichistheprimaryaimofouranalysis.Toaddressthis, incontrasttopreviousdecompositions(VanOurtietal., 2009;Allansonetal.,2010;Baetenetal.,2013),webreak totalincome(yit)downasthesumofmarketincomes(yM_it)

andgovernmenttransfers(yG

it),i.e.yit=y M it +y G it. 6_Income

weightscanthenbedeﬁned separatelyforeach source. Weights associated with total income (zit) and market

income(zM

it)aredeﬁned inthe standard waydescribed

above,whiletheincomeweightassociatedwithtransfers isdeﬁnedasthedifferencebetweenanindividual’stotal incomerankandmarketincomerank,zG

it=zit−zMit,such

thatitmeasuresthenumberofstepsontheincome lad-derthatseparatetotalfrommarketincome,andtherefore doesnotdependontherankofyG_it.Substitutingthechange intotalincomeweights(ziT−zi1)inEq.(3)bythechange

ofincome-sourcespeciﬁcincomeweightsallows describ-ingtheroleofincomere-ranking(health-relatedincome mobility)separatelyforeachincomesource:

CCI(hT|yT)−CCI(h1|y1) = 8 n2 n

i=1 [(zM_iT−zM_i1)hiT+(ziTG−z G i1)hiT+zi1(hiT−hi1)] (4)

Theﬁrsttermmeasureswhetherthoseendingupwith goodhealthinperiodT aremoreorless likelytomove upordownthemarketincomeladdercomparedtothose endingupwithworsehealth. Thesecondtermportrays theassociationbetweenthede-/increaseoftheequalizing effectofgovernmenttransfers,i.e.thechangeinthe num-berofstepsontheincomeladderbetweenmarketandtotal income,andﬁnalperiodhealth.Thethirdtermisidentical tothesecondterminEq.(2),i.e.itmeasureshealth-related totalincomemobility.7

5 _Brekke_and_Kverndokk₍₂₀₁₄₎_have_shown_that_income_{redistribution}

amongtwoequallyhealthyindividualsmightincreasetheconcentration

indexwhenhealthis,onaverage,positivelycorrelatedwithincome.This

isbecausesuchincometransfersincreasetheconcentrationofgood(bad)

healthamongtherich(poor),eventhoughtheydoreduceincome

inequal-itiesandleavethemarginalhealthdistributionunaffected.Thiseffectis

pickedupbythehealth-relatedincomemobilityterm.

6 _In_line_with_the_literature_on_the_{redistributive}_effect_of_government

transfers(Plotnick,1981;Lambert,2001),weinterpretmarketincomes

astheincomesthatwouldapplyintheabsenceofgovernmenttransfers.

7 _While _conceptually _feasible, _we _do _not _consider _the

income-source-speciﬁc contributions to income-related health mobility, i.e.

zM

i1(hiT−hi1)+zGi1(hiT−hi1),sincetheserefertotheincomepositionsin

theﬁrstperiod,andthereforearenotinformativeontheincome

dynam-icsbetweenperiod1and2.Incontrast,health-relatedincomemobility

considersthesedynamics,andhenceinformsontheevolutionofthe

WhilethisextensionoftheAllansonetal.(2010) decom-position is informative about how theseparate income sourcesinﬂuenceIRHIevolutionviare-rankingchanges,it issilentaboutthedistinctroleplayedbyfactorssuchasage, gender,regionorincomethatareassociatedwithhealth. Since–totakeoneexample–transferandmarketincomes aredistributeddifferentlyacrossvariousagegroups,such adistinctionislikelytobeimportantintracingouttheir distinctroles.

Following Van Ourti et al. (2009) and Baeten et al. (2013),weaddressthiswithasimpledescriptivemodel thatlinkshealthlinearlyandadditivelytoitsassociated factors:8 hit=˛+ (yit)+ k

j=1 xjitˇj+εit (5)

where˛isaninterceptparameter; (yit) isanon-linear

functionofincome;thereareKnon-incomevariablesxjit

withj=1,···,K (inouranalysis,theseareasetof age-sexandregiondummies),ˇjaretheassociatedparameters

reﬂectingpartialassociations,andεitcapturestheresidual

whichhasazeromean.Astheexactfunctionalformfor (.) pre-determinesthesignandmagnitudeofsomeparts ofourdecomposition,weadoptaﬂexiblefunctionalform intheempiricalapplication.

CombiningEqs.(4)and(5)showsthatthechangeinIRHI canbeexpressedasthesumof4terms:9

CCI(hT|yT)−CCI(h1|y1)= 8 n2 n

i=1 { (z_iTM−z_i1M) k

j=1 xjiTˇj

market−related incomemobility

+ (zG iT−zGi1) k

j=1 xjiTˇj

transfer−relatedincomemobility

+zi1 k

j=1 ˇj(xjiT−xji1)

ageing andmigration

+

ziT[(yiT)+εiT]

+zi1[(yi1)+εi1

] other

}

(6)

The ﬁrst 2 terms are similar to the income-source speciﬁchealth-relatedincomemobility termsinEq. (4), exceptthatincomere-rankingisweighedbynon-income relatedhealthinthelastperiod.10_Both_{market-related}_and

transfer-relatedincome mobilityare morepositive

(neg-inequalityreducingimpactofgovernmenttransfers,anditsassociation

withsecondperiodhealthlevels.

8_An_additional_assumption_is_that_there_is_no_structural_change_in_the

healthequationacrossperiods.Intheempiricalsectionwetestanddo

notrejectthishypothesis.

9_Note _that

_zM iT−z M i1

(yiT)+εiT

+

zG iT−z G i1

(yiT)+εiT

+ zi1

(yiT)+εiT− (yi1)−εi1

=ziT

(yiT)+εiT

+zi1

(yi1)+εi1

.

10_Non-income _related _health _equals _{E (h}

itx1it,···,xKit)=˛+ E

(yit)

+ k

j=1

xjitˇj+E [εit], but is simpliﬁed to

k

j=1

(5)

ative) when upwardly (transfer/market) income mobile individualshavebetter(worse)non-incomerelatedhealth inperiodT .Notethatifthenon-incomevariablesconsist ofmultiplevariablesthatenterthehealthequation addi-tively,thenthemobilitytermscompriseasummationof differentsub-terms.Thisholds,forexample,ifoneusesa setofage-sexandregiondummiesaswedo.Thisallowsus toseparatetheaggregatemobilityeffectintothe contribu-tionperage-groupandregioncategory.Summingthetotal transferandmarketmobilitytermsgivesthetotalincome mobility.

Betweenperiods,individualsmaychangebetweenage andregioncategories.Thisiscapturedbytheageingand migrationterm.Itindicateshowmobilityinnon-income relatedhealth, duetoitsassociationwithinitialincome weights,hasledtochangesinIRHI.Asageingand within-countrymigrationmayhaveconsequencesforhealth,the degreetowhichtheyareassociatedwithincomeranksmay affectIRHI.Thistermmainlyactsasacontrol,allowingus tostudychangesinIRHInetofageingandmigrationeffects. TheﬁnalothertermcapturesthechangeinIRHIthatis unrelatedtothenon-incomefactors.Sinceweincludedage, genderandregioninEq.(5),thiscorrespondstothechange inIRHIthatremainsafteraccountingforthesevariables.

It is possible tomake a further distinction between income-predictedhealthgains, (yit),andhealthresiduals,

εit.Giventhattransferandmarketincomesmaycontribute

differentlytoIRHIviathenon-linearfunctionlinkinghealth andincome,webelievethatthisdistinctionbetween (yit)

andεitisaprioriinteresting.Thisallowsustoarriveatour

ﬁnaldecomposition:11 CCI(hT|yT)−CCI(h1|y1)= 8 n2 n

i=1 {(zM iT−z M i1) k

j=1 xjiTˇj

market−relatedincomemobility

+ (zG iT−z G i1) k

j=1 xjiTˇj

transfer−relatedincomemobility

+zi1 k

j=1 ˇj(xjiT−xji1)

ageingandmigration

+zM

iT(yMiT)−zMi1(yMi1)

market−relatedinequality

+[z

iT(yiT)−zMiT(yMiT)]

−[zi1(yi1)−zMi1(yMi1)]

transfer−relatedinequality

+ε

iT(ziT

−zi1

)

errorrankchange

+ z

i1(εiT

−εi1

)

errorresidualchange

⎫

⎬

⎭

(7)

Theﬁrst3termsareidenticaltothoseinEq.(6) and the sum of the last 4 terms corresponds to the other term.Market-relatedinequalitychangemeasuresthe con-sequences for IRHI of thechange in thedistribution of marketincomes.(yM

iT)denotesthepredictedhealthlevel

inthelastperiodthatcorrespondstoyM

iT.12Theproduct

CCI is an absoluteinequality index, E [εit]=0and ˛+E

(yit)

is constant.

11_We_have_conﬁrmed_that_the₄_terms_in_equation₍₆₎_are_virtually

invariantwithrespecttoin-orexclusionoftheincometerm (yit) in

equation(5).Theresultsareavailablefromtheauthorsuponrequest.

12_Strictly_speaking_these_are_not_health_levels,_but_the_same_reasoning

asinfootnote11applies.

zM

iT(yMiT)thereforemeasuresmarket-relatedinequalityin

theincome-predictedhealthlevels.ThisissimplytheCCI for market incomerelated health inthe secondperiod. The second product in the expression is identical, but referstotheﬁrstperiod.Thedifferencebetweenthesetwo corrected concentration indices therefore captures how changesinthedistributionofmarketincomesbetweenthe twoperiodswereassociatedwithchangesin IRHI,both bytheirassociationwithhealththroughthe(.)function, andviathere-rankingofindividualsonthemarketincome scale.Foramonotonicallyincreasing(.)function, market-relatedinequalitychangewillindicaterising(falling)IRHI whentherich (poor)predominantly experienceincome improvements(deteriorations).

The next expression is the transfer-related inequal-ity change.The term

ziT (yiT)−zM_iT(yM_iT)

captures the degreetowhichtransferincomeschangetheassociation betweenincomeweightandincome-predictedhealthin the last period. In other words,this term captures the degreetowhichtheadditionoftransferstosecond-period marketincomesresultsina more (orless)equal distri-butionofincomepredictedhealthbyincomerank.13_The

secondtermmeasuresthiseffectintheﬁrstperiod.Both terms thusreﬂect whethertransferincomes resultin a moreorlessequaldistributionofincome-relatedhealth,or theextentoftheredistributiveeffectoftransferincomes in the separate periods. Their difference indicateshow this effect haschanged over time, and its consequence fortheevolutionofIRHI.Summingmarket-related inequal-itychangeandtransfer-relatedinequalitychangegivesthe changeintheCCIfortotalincome-relatedhealthbetween periods1andT .

Theresiduals inoursimpledescriptive healthmodel (Eq. 5) capturehealth that is neither accounted for by theincomefunctionnorbythesetofage-sexandregion dummies.Iftheassociationbetweenthisresidualandthe incomerankschangesbetweenthetwoperiods,thishas implicationsforIRHIwhicharecapturedbythelasttwo termsinEq.7.Theﬁrst,errorrankchange,describeshow thechangeinincome ranksbetweenthetwo periodsis associatedwithresidualhealthinthelastperiod.The sec-onderrorterm,errorresidualchange,describeswhetherthe initiallypoorversusrichexperiencethelargestchanges intheirresidualhealthlevels.Indirectly,thesumofthe lasttwoterms(ziTεiT−zi1εi1)allowsustotracethe

conse-quencesofunexplainedhealthchangesforIRHI.

Empiricalanalysis

Data

WeusetheEuropeanUnionSurveyonIncomeand Liv-ingconditions(EU-SILC),aEuropean-widesurveydesigned primarilytocollectlabourandincomerelated data.Itis wellsuitedtoouranalysisforseveralreasons.First,it pro-videsadetailedbreakdownofthesourcesofdisposable householdincome,whichiscrucialtomeasuringthe

sep-13 _For_instance,_note_that_if_government_transfers_do_not_exist_(i.e._yM

iT=

(6)

arateeffectsofgovernmenttransfersandmarketincome onIRHItrends.Secondly,individualsareaskedtoratetheir self-assessedhealth(SAH),whichisusedtoconstructour healthmeasure.

Ourselectionofcountriesisbasedondataavailability andqualityintheEU-SILC.Werequirethatcountrieshave adequateincomeandhealthdataforthewhole2004–2013 period.TableA1intheappendixprovidesanoverviewon theavailableinformationforthe29EU-SILCcountriesand theselectioncriteriausedforinclusion.Thisleavesuswith thefollowing7countries:Austria,Belgium,France,Greece, Italy,Portugal,andSpain.14_We_deﬁne_a_subset_of_these_as

crisiscountries–thosecountriesmostseverelyaffectedby thecrisis.Specifically,weapplythislabeltothecountries inoursamplethathavebeennotedbytheOECDas hav-ingsufferedworse-than-averageGDPdeclinesaswellas publicspendingcutsasaresultofthecrisis(OECD,2012); Greece,Italy,Portugal,andSpain.Ourowndataalso con-firmsthatemploymentandhouseholdincometrendsafter 2008fairedmuchworseinthelatterfourcountriesthan theremainingcountriesinoursample(seefiguresB1and B2).

TheEU-SILCisarotatingpanel.Anewrandom sam-ple(referredtoasarotationgroup)isdrawneveryyear, followedfor4yearsandthendropped.Therefore,atany point,eachcountryhas4concurrentpanelsamples.There are7rotationgroupsinourstudyperiod,i.e.2004–2007, ...,2010–2013.Weusebalanceddatafromall7rotation groupstoestimateourmodelforhealth(Eq.5).15_,16_Table₁

providesnumbersofindividuals(eachobservedfor4years) perrotationgroupand country. Duetochangesindata collectionmethods,theincomedataforFrancefrom2009 onwardsarenotcomparabletoearlierwaves.Wetherefore ignorethe2007–2010periodforFrance.17

Tableshowsforeachrotationgrouptheperiodspanned and the number of individuals observed for the whole 4yearperiodforeachcountry.

Incomemeasurement

The EU-SILC provides, per person and household, a detailedbreakdownofthecomponentsofannual

house-14 _Our_selection_criteria_is_that_a_country_is_represented_in_all₇_rotation

groups.Furthermore,althoughmanyoftheNordiccountries–Finland,

IcelandandSweden–arepresentinallrotationgroups,theiruseof

register-baseddatacollectionmethodsleadstomanymissingvaluesof

theSAHvariableraisingconcernsofattritionbias.Samplesizesinsome

ofthesecountriesaretoolowforreliableanalysis.Forinstance,thereare

only13womenabovetheageof75inthe2004sampleinIceland.

15 _We_{symmetrically}_drop_the_top_and_bottom_1%_of_total_incomes_to

removepotentialoutliers.

16 _Our_restriction_to_balanced_panels_excludes_the_possibility_of_attrition

bias.However,trendsofIRHIcomputedwhenusingalldata,notjusta

balancedpanel–areextremelysimilartothoseweﬁndhere,suggesting

attritionbiasisnotdrivingourresults.

17 _The_data_collection_method_for_certain_components_of_income_in

France,namely“interest,dividendsandproﬁtfromcapitalinvestmentsin

unincorporatedbusiness”,wentfrombeingsurvey-basedtoregister-based

in2009.TheaveragevalueofthiscomponentincreasedbyalmostD3,000,

andledtoadramaticriseinaverageincomes.Itisnotpossibleto

distin-guishbetween“real”increaseinthecomponentandinﬂationduetomore

accuratecollectionmethods.

holdincome.Weseparatetotalincomeintowhatweterm marketincomeandtransferincome.Anindividual’smarket incomeisdeﬁnedastheequivalizedvalueofdisposable householdincomebeforeallsocialtransfers,andtransfer incomeastheequivalizedvalueofthesumofallhousehold socialtransfers.18_The_income_reference_period_is_the

pre-viouscalendaryear.Table2liststheEU-SILCcomponents thatmakeuphouseholdmarketandtransferincome.

Public pensions form the largest share of transfer income, and employee income (income from work) for marketincomes.19_When_using_the_term_{“pensions”}_we_are

referringtowhatEU-SILCterms“oldage”beneﬁts.These includethecollectionofallsocialpaymentstotheelderly thataredesignedtoprovideareplacementincomewhena personretiresorhasreachedacertainage.20

Healthmeasurement

Individualsareasked:“Howisyourhealthingeneral?Is it:(1)verygood,(2)good,(3)fair,(4)bad,(5)verybad?”. WecannotdirectlyusethisvariablesincetheCCIrequires a ratio-scaledhealthmeasure (Erreygers and VanOurti, 2011).InordertotransformtheordinalSAHmeasurein EU-SILCtoaratio-scaledmeasure,weperforman inter-val regression with the threshold values imposed from externaldata.21_Covariates_included_are_age/sex_dummies,

regiondummies,asetofeducationdummies,andan indi-cator for the presence of a chronic illness. This set of predictorvariablesisparsimonious,yetisstrongly asso-ciatedwithhealth.

Weusetheintervalregressionpredictedhealthinour decomposition analysis, but conditional on the SAH cat-egory reported by therespondent. Thisimplies that the reportedSAHcategorydeterminesthethresholdswithin whichthepredictedratio-scaledhealthmeasurewilllie, while the covariates determine the exact value of the measurewithinthesethresholds.22_Conditioning_on_both

18_Household_equivalent_income_equals_household_income_divided_by

thesquarerootofthenumberofindividualslivinginthehouseholdinthe

currentperiod.

19_See_Table_A2.

20_Old_age_beneﬁts_includes_public_pension_payments,_care_allowances,

disabilitycashbeneﬁts,lumpsumpaymentsatthetimeofretirement

andothercashbeneﬁts.Itdoesnotincludeanypaymentsfromprivate

pensionplans,whichenterthemarketincomedeﬁnition.Disabilityand

otherpaymentsalsoappearasaseparatecategory,asthiscapturesthese

paymentswhentheyareaffordedtoindividualswhohavenotretired.See

theEU-SILCguidelinesdocumentationforfurtherdetails.Ourdatashows

privatepensionsarenotanimportantpartoftransfersforthesecountries.

Onaverageacrossallrotationgroupsandcountries,paymentfromprivate

plansarelessthan1%ofoldagebeneﬁts.Percountry,theaveragefraction

ofprivatepaymentstopensionpaymentsisneverhigherthan3%.

21_For_each_country_we_imposed_the_identical_thresholds_from_the

empir-icaldistribution functionofthehealth utilityindexinthe Canadian

NationalPopulationHealthSurvey1994-1995(VanDoorslaerandJones,

2003).

22_This_precludes,_for_instance,_a_situation_where_individuals_with

iden-ticalcovariatesbutdifferentlevelsofSAHareassignedidenticalhealth.

Moreformally,thepredictionconditionalontheSAHcategoryequals

E

h∗zi,aj−1≤h∗≤aj

whereh∗_irepresentsthelatenthealthvariable,zi

isavectorwithcovariatesandajwithj=1,...,5aretheSAHthresholds

imposedfromexternaldata.Considerequation(8)inVanDoorslaerand

(7)

Table1

IndividualsperrotationgroupandcountryinEU-SILCdataset.

Rotationgroup 1 2 3 4 5 6 7 Period 2004-2007 2005-2008 2006-2009 2007-2010 2008-2011 2009-2012 2010-2013 Observations Austria 2,287 1,919 1,901 1,893 1,881 2,199 2,161 Belgium 1,102 1,713 1,749 1,828 1,624 1,746 1,875 Greece 2,152 2,021 2,418 2,148 2,655 2,368 2,182 Spain 3,805 4,438 4,631 5,143 5,188 4,841 4,245 France 1,417 2,281 2,324 2,321 2,358 2,266 2,284 Italy 7,947 7,531 7,219 7,286 6,293 5,378 4,715 Portugal 1,568 1,425 1,470 1,672 1,748 2,089 2,125 Table2

Incomecomponentsoftransferandmarketincomes.

Transferincome Marketincome

• Unemploymentbeneﬁts • Grossemployeecashornearcashincome

• Old-agebeneﬁts • Companycar

• Survivorbeneﬁts • Grosscashbeneﬁtsorlossesfromself-employment

• Sicknessbeneﬁts • Pensionsreceivedfromindividualprivateplans

• Disabilitybeneﬁts • Incomefromrentalofapropertyorland

• Education-relatedallowances • Regularinter-householdcashtransfersreceived

• Family/childrenrelatedallowances • Returnsfromunincorporatedbusiness

• Socialexclusionnotelsewhereclassiﬁed • Incomereceivedbypeopleagedunder16 • Housingallowances

Minus

• Regulartaxesonwealth

• Regularinter-householdcashtransferpaid • Taxonincomeandsocialinsurancecontributions TableshowsthemakeupforourdeﬁnitionsofTransferandMarketincomesasusedintheEU-SILCsurvey.

the SAH category, as well as the individual covariates,

leadstoamoreinformativepredictionoftheunobserved

healthmeasurecomparedtoonlyconditioningon

covari-ates(VanDoorslaerandJones,2003).Inaddition,excluding incomefromthecovariatesavoidsamechanical relation-shipbetweenthepredictedhealthmeasureandincome, whichwouldpredeterminetheevolutionofIRHI.The inter-valregressionisperformedpercountry,poolingallrotation groups.Theregressionresultsforeachcountryareshown intheappendix,intableA3.23

Implementationofdecomposition

The decomposition is performed separately for each countrybut, importantly,doesnotusetheestimates of the interval regression. Instead, we estimate our sim-pleexplanatoryOLSmodelforhealth(Eq.5)bypooling allrotation groupsfor countryc,andusing theinterval regression-predicted health variable described above as thedependentvariableandascovariatesonlyage-gender, regionand income,not educationor chronicconditions (seetableA4).Thisregressionprovidesthenon-incomeand incomecoefﬁcients,andresidualsusedinthe decomposi-tion(Eq.7)forcountryc.

We then take 3 rotation groups (2004–2007, 2007–2010, 2010–2013), and calculate and decom-posethechangeintheCCIfromtheﬁrstyear(thebase

23_The_average_value_of_this_health_variable_is_shown_per_country_and

rotationgroupinﬁgureA1.

year)forcountryc.24_We_only_present_the_{decomposition}

withrespecttothelastyearoftherotationgroupbecause intermediate decompositions are similar in sign and relativemagnitudewithinrotationgroups.25 _In_order_to

allowforstatisticalinferenceonIRHIlevels,IRHIchanges and the decomposition terms, we bootstrap the entire procedure1500times,includingtheintervalregression.

Resultsanddiscussion

ThissectionﬁrstexaminesthegeneraltrendsinIRHI inthe7 countriesunderstudybetween2004and2013. Wethenseparatelystudytheroleofthemobility,health inequality,ageingandmigration,andtheerrorchangeterms inIRHIchangesbeforeandaftertheﬁnancialcrisisin2008. Finally,theroleoftheausteritymeasuresenactedinGreece onIRHIisexplored.

IRHItrendsacross7Europeancountries

Fig.1showshowIRHI,asmeasuredbytheCCIand cal-culatedusingpredictedhealth,hasevolvedbetween2004

24 _We_also_estimated_the_OLS_(and_the_underlying_interval_regression)

modelsseparatelyforeachofthethreerotationgroupsasthepartial

associationsbetweenthecovariatesandthedependentvariablesmight

havechangedduetotheGreatRecession.Theresultingdecomposition

estimates,availableuponrequestfromtheauthors,conﬁrmthatthe

assumptionofnostructuralchangeimposedinthemainresultsis

rea-sonable.

25 _An_exception_is_Greece_in_the_2010-2013_{decomposition,}_which_we

exploreinmoredetailbelow.Thefulldecompositionresultsper

(8)

Fig.1.IRHItrends:CCIineachyear.

Figureshows,foreachcountry,CCIperyear,perrotationgroup,with95%conﬁdenceinterval.NotethedifferentscaleofPortugal.Boldbarsindicate

yearsinwhichdifferenceinCCIcomparedtobaseyearisstatisticallysigniﬁcant(p<0.05).Y-axis:valueoftheCCI.SeeEq.(1)fortherelevantformula.No

comparabledataforFrancefor2007–2010(c.f.footnote18).

and2013forthe7countriesunderstudy.26_The_separate

linesrepresentthethreerotationgroupsusedtospanthe period.WhiletheconﬁdenceintervalsinFig.1are infor-mativeaboutthesamplingvariabilityoftheyearlypoint estimatesofIRHI,ourinterestliesinexaminingthechanges ofIRHIbetweendifferentperiods.Itisthereforeusefulto knowifthechangesinIRHIwithrespecttothebaseyear arestatisticallysigniﬁcant,whichisindicatedbythebold bars.27

WenotebothgeographicalandtimepatternsintheIRHI trends.IRHIevolutioninthenon-crisiscountries(Austria, Belgium,France)wasremarkablyﬂatinallthreerotations groups,instarkcomparisontothemorevolatiletrendsin thecrisiscountries(Greece,Italy,Portugal,Spain). How-ever,fornocountrydidIRHIsigniﬁcantlychangewhen comparing2013to2010.

26 _Our_focus_is_on_absolute_income_related_health_{inequalities.}

How-ever,ﬁgureA2intheappendixshowsthatrelativeincome-relatedhealth

inequalitieshaveevolvedsimilarlyoverthesameperiod.

27 _We_do_not_check_the_statistical_{signiﬁcance}_of_changes_across_rotation

groupssinceweonlyobservethesamesetofindividualsoveraperiodof

4years.

Wedistinguishbetween3differentperiodsinour analy-sisbasedontheeconomicgrowthofthedifferentcountries (seeappendixB).FollowingJenkinsetal.(2012),we con-siderthe2004–2007periodtobethepre-crisisperiod;a timeofrelativelynormalgrowthforthe7countries.We term the rotation group spanning 2007–2010 thecrisis period.Finally,thepost-crisisrotationperiod(2010–2013) iswhenconsequencesoftheGreatRecessionaremost obvi-ousin crisiscountries, whilelargeeffects forhousehold income,inequalityandemploymentareabsentforthe non-crisiscountries.

Wedonotfindparticulardifferencesinpost-crisisIRHI trendsbetweenthecrisisandnon-crisiscountries.Despite temporaryjumpsin Italy andGreece, thereis no coun-trywhereIRHIlevelsin2013significantlyexceedthoseof 2010.Thedecompositioninthenextsectionismotivated bythisfinding:why–despiteinitialconcerns–didIRHInot significantlyriseaftertheGreatRecession,noteveninthe countriesmostheavilyaffectedbythecrisis?Wewillfocus onthepre-crisis(2004–2010)andpost-crisis(2010–2013) periods,astheseperiodsencapsulateclearphasesof eco-nomicgrowthordeclineformostcountries,whilethecrisis

(9)

period(2007–2010)oftenincludesmixedperiodsofboth (seeappendixB).

Decompositionresults

Figs.2–4depicttheestimatedincomemobility, inequal-itychange,anderrorchangeterms,respectively.Theageing andmigrationtermprovestoberatherunimportant for explaining IRHIevolution (ﬁgureA5).PanelsA andBin Fig.2showtheresultsfor,respectively,thepre-crisisand post-crisisrotationgroups forallcountries.Theleftmost cluster of bars in panel A shows (in order fromleft to right)thecontributionthatmarket-relatedmobility(black), transfer-related mobility(grey) and total incomemobility (white,andthesumoftheprevioustwoterms)hadonIRHI changesinAustriabetween2004and2007.The remain-ing clusters/panelshave a similarinterpretationfor the different countries and rotation groups. In Fig. 3 each clusterofbarsshows,percountry,theeffectthat market-relatedinequalitychange,transfer-relatedinequalitychange andtotalinequalitychange(sumoftheprevioustwoterms) hadonIRHIchangeinthatrotationgroup.Similarly,Fig.4 showstherankchange,residualchange,andthetotalerror change(sumoftheprevioustwoterms).28

Boththemobilityand errortermsare muchlargerin magnitudethantheinequalityterms,andarethusthemore importantdeterminantsofIRHIchange.Weexploreeach setoftermsbelow.

Mobilityterms

Fig.2revealsthat,acrosscountriesandperiods, mar-ketmobilitytendstobepositiveandsizable.Incomparison –thoughusuallynegative–thesizeandsignoftransfer mobilityismorevaried,andthereforeitisoftenthisterm whichleadstodifferencesinthetotalmobilitytermacross country-periodcomparisons.

Recallthatthemobilitytermscanbefurthersplitinto per-age/sexgroupsandper-regioncontributions(seeEq. 7).Doingsogivesanindicationofwhichage/sexgroup’s income movementsare inﬂuencing thedirectionof the separatemobilityterms,andthereforegivesinsightinto thepatternsinFig.2.Whilewedonotrefertothesemore detailedresultsexplicitlyinthemaintext,theyunderlie muchofthefollowingdiscussion,andcanbefoundinthe appendixforeachcountry,mobilitytermandforboththe pre-andpost-crisisperiod.29

ThereasonfortheIRHIincreasingeffectofmarket mobil-ityisthatimprovementsinmarketincomesmostlyhelp the youngest, and therefore healthiest,groups to climb theincomeladder,therebyincreasinghealthdisparitiesby marketincome.Giventhevariationintransfermobility,we distinguishbetweenthefollowingpatterns.

First,onecandistinguishbetweentwotypesof peri-ods and countries: (i)those in which transfer mobility

28_Figures_A3-A4_illustrate_these_terms_for_the_rotation_group_2007-2010.

29_See_tables_A5-A11._The_results_per_region_are_supressed_as_they_are

smallandnotimportantforthedecomposition,butavailableuponrequest

fromtheauthors.

fullycompensatesfortheincreaseinIRHIcausedby mar-ketmobility, suchasinAustria,SpainandPortugal (pre-andpost-crisis),andItaly(post-crisis),and(ii)periodsand countriesinwhichtransfermobilityisclosetozero,such asinBelgiumandFrance(pre-andpost-crisis),andItaly (pre-crisis).

Second,transfermobilityispositiveinGreecepre-crisis. Furtherdecompositionofthistermsrevealsthatthiscan beattributedtohouseholdstructure.Ratherthanpensions solelybeingenjoyedbytheold,youngerpeopleinGreece alsobenefitedfromthelargeincreaseinpensionincomes between2004and 2007.Thisis duetoyoung individu-alscontinuingtoliveattheirparent’shome,andtherefore benefitingfromtheirparent’s(orgrandparent’s)influxin pensionincomeupontheretirementoftheelderly mem-bersofthehousehold.30_This_increase_in_transfer_income_for

theyoungandthejust-retired,andtotheexclusionofthe very-elderly,ledtoincreasingincomedisparitiesbetween thehealthy and theunhealthy, and therefore increased IRHI.31

Lastly,thereisaconsistentpatternforthecrisis coun-tries post-crisis, whereby transfer mobility is large and negative in the ﬁnal rotation group. In Portugal, for instance,thisterm“over-compensated”formarket mobil-ity,andledtodecreasesinIRHIbetween2010and2013. Thisisduetothe“stickiness”ofpensionsrelativetoincome fromwork –while the crisisledtoa signiﬁcant fallin theincomesoftheyoung,theincomesofelderly(and,on average,unhealthier)pensionerswerelessaffected.This generatedadropinIRHI.

Marketandtransferinequalitychange

Thesmallerassociationbetweenincomeandhealth rel-ativetotheassociationbetweenageandhealth(tableA4) impliesthat theinequalitychangetermsare small com-paredtothemobilityanderrorchangeterms(Fig.3).The totalinequalitychangetermsareverysmallandnot sta-tistically signiﬁcant.The market and transfer inequality changetermstendtobelargerthanthetotalterm,butof oppositesign, stillquantitativelyunimportantand often statisticallyinsigniﬁcant.The positive market termsare theresultofwagegrowthfortheemployedandagradual increaseofthenumberofretireesinourpanels.Bycontrast, thetransfer-relatedinequalitychangetermsarenegative sincetheredistributiveeffectoftransferswasnegativein eachyear,i.e.marketincome-relatedhealth inequalities (zM it

yM it

)werealwayslargerthantotalincome-related health inequalities (zit (yit)). The most important

gov-ernmenttransfer,intermofitsredistributiveeffects,are pensions.32

30 _This_trend_has_also_received_attention_in_the_Greek_press,_with_a₂₀₁₉

articlenotingthatmorethanhalfofGreeksbetween25and35stilllive

withtheirparents–oneofthehighestratesintheEU(GreekCityTimes,

2019).

31 _While_household_structure_plays_a_similar_{“protective”}_role_for_the

youngerinItalypre-crisis,thisismorestronglycompensatedbytheeffect

amongtheveryelderly.

32 _We_conﬁrm_this_by_repeating_our_{decomposition}_and_redeﬁning

(10)

Fig.2.Incomemobilityterms.

Figureshowsdecompositionresultsforincomemobilityterms(expressions1and2ofEq.(7),andtheirsum)for2004–2007and2010–2013rotation

groups,percountry.

Theonlyexceptiontothegeneraltrendsnotedaboveare inGreece,whichexperiencedaquantitativelyimportant decreaseinthemarketinequalitychangetermpost-crisis suchthatitcontributesnegativelytoIRHI.Wediscussthe Greekexperience in more detail in thelast part of the resultssection.

Errorterms

RecallthattheerrorrankchangetermcapturestheIRHI consequencesoftheassociationbetweenthechangesin totalincomerankbetweenthetwoperiods,andthehealth residualin thesecond period. Theerror residual change termcaptures theIRHIconsequences of theassociation betweenthechangesinthehealthresidualbetweenthe twoperiods,andthetotalincomerankintheﬁrstperiod. Toinformthefollowingdiscussionoftherankchangesterm (residualchangesterm),weexaminetheaveragechangein totalincomerank(healthresidual)foreachdecileofthe healthresidualinthesecondperiod(eachdecileofthetotal incomerankintheﬁrstperiod)forboththe2004–2007and

beneﬁts”,andattributingtheremainingtransfercomponentstomarket

income.Thefactthattheresultsremainextremelysimilarimpliesthat

pensionsarethemostimportantsocialtransferstounderstandthe

redis-tributiveeffectanditschangeovertime.Resultsavailableuponrequest.

2010–2013comparisons.Thisinformationcanbefoundin tablesA12-A18oftheappendix.

Theaggregateeffectoftheerrortermsmakesupalarge contributiontoIRHIchangeinmostperiodsandmostcrisis countries(comparethe‘total’barsinFigs.4with).While thesetermsgenerallycontributesimilarlytoIRHIchange asthemobilityterms,theconﬁdenceintervalsarewider, partlybecausetheresidualhasahighervariancethan age-gender-regionpredictedhealth.

Theconsistently negativerankchangetermindicates that individuals withhigher(lower) residualhealth, i.e. who arein better(worse)health thanpredictedbyour health modelinEq. (5),systematicallymovedown(up) the total income ranks. In other words, in addition to young,andthushealthy,individualsmovingdownthetotal incomedistribution(becausetheydonotreceivepension beneﬁts–thetransfermobilityterm),otherwisehealthy individualsalsodropdownthetotalincomeladder.This could,forinstance,meanthatformerlyhealthy,employed individuals lost theirjobsandassociated income, while unhealthierindividualsone.g.disabilitybeneﬁtsdidnot loseincomeandwentupintheincomeranks.A compar-isonoftablesA12-A18andA5-A11revealsthatthelatter totalincomerankchangesamongthehealthy, standard-izedforage,genderandregion,wereingenerallargerthan thetotalincomerankchangesforthosehealthybecause

(11)

Fig.3.Incomeinequalityterms.

Figureshowsdecompositionresultsforincomeinequalityterms(expressions4and5ofEq.(7),andtheirsum)for2004–2007and2010–2013rotation

groups,percountry.

theyareyoung,furtherstressingtheimportanceoftherank changeterms.

In contrast, theresidual changesterm is consistently positive.Thistermmeasureshowchangesinhealth not capturedbyourOLSmodel,andtheassociationwiththeir totalincomerankintheﬁrstperiod,impactsonthe evo-lutionofIRHI.Theconsistentlypositivevalueofthisterm impliesthat thisunexplainedportionofhealth tendsto worsenforindividualsinthebottomofthetotalincome dis-tribution,andviceversa.Whenthisistakenintoaccount, IRHIishigherthanitotherwisewouldhavebeen, repre-sentedbythepositiveresidualchangesterms.

Thisincreaseintheresidualchangetermisparticularly notablein2010–2013comparedtothe2004–2007 com-parisoninthe4crisiscountriesSpain,Greece,Italy,and Portugal.Thisisprimarilyduetotworeasons.First, individ-ualswithlowerincomesinthesecountriesbegantoreport worsehealth between2010and 2013,and thesehealth changeswerenotaccountedforbychangesinthe covari-atesinoursimplemodelforhealth.Secondly,especiallyin Greece,thehealthoftheinitiallyveryrichremainedhigher thanexpected,despitetheincomeeffectsofthecrisis(table A16).Wenotethatthislargeerrorresidualterminthecrisis countries(primarilyinGreeceandItaly)isoften responsi-bleforthetotalerrorcontributionbeinglargeandpositive inthe2010–2013comparison.

WhydidIRHInotriseduringtheGreatRecession?

Despitefearstothecontrary,astrikingfeatureofthe IRHItrendsinFig.1isthelackofanyconsistentsignof IRHIrisingpostcrisis.Thisholdseveninthecrisiscountries. TheaboveresultshelptoshedlightonwhyIRHIremained relativelyﬂatinthepost-crisisperiod.

Thetwomostinﬂuentialtermsinthecrisiscountries between2010and2013arethetransfermobilityand resid-ualchangesterms,particularlyinGreece,ItalyandPortugal. AsanticipatedbytheEuropeanCommission(2009a),the healthofthepoorestdoesindeedappeartohavesuffered duringthecrisis,ascapturedbytheresidualchangesterm. However,thesurprisingﬁndingoflittleIRHIchangeinthe crisiscountriespost2008appearstobedue tothefact thatthetransfermobilitytermdampenedtheIRHIincreases fromtheresidualchangesterm.Astherelativeincomesof theelderly–thoseinlowesthealth–tendedtoimprove, thisreducedIRHI.Atthesametime,self-reportedhealth (unrelatedtogender,age,region,andincome)began to systematically deteriorate among the poor, exerting an upwardpressureonIRHI.Asaresult,theIRHI-increasing effect of the residual changes term (Fig. 4) during the 2010–2013periodinGreece,Spain,ItalyandPortugalis approximatelyhalvedbythetransferincomemobilityterm (Fig.2),leadingtolittlechangeintheoveralllevelsofIRHI.

(12)

Fig.4.Errorterms.

Figureshowsdecompositionresultsforerrorterms(expressions6and7ofEq.(7),andtheirsum)for2004–2007and2010–2013rotationgroups,per

country.

Thisﬁndingpointstothecrucialroleofthetransferincome mobilityterminholdingIRHIsteadyduringtimesof cri-sis,andavoidingwhatmayhaveotherwisebeenalarge increaseinthedisparitiesinhealthbyincome.

Whileoneshouldnotconcludefromtheseﬁndingsthat theothertermsareunimportant,i.e.boththemarket mobil-ityandrankchangetermsdomatteraswell,wedoseethat ageingandmigrationandbothinequalitychangetermsare quantitativelyunimportantinallcountries,exceptGreece. Anotherimportantresultofourdecompositionisthusthat, despite income being strongly predictive of our health measure(tableA4),thein-orexclusionofincomeinthe healthmodelinEq.(5)doesnotaffectourmainﬁndings.33

Ourmoredetailedresultspointtopensionsandother old age benefits as the transfers that lead to the rela-tiveimprovementoftheelderlyduringthefinancialcrisis. Theirrelativeimmunitytomarketfluctuationswhen com-paredtowagesledtogainsontheincomeladderforthe elderlyrelativetoyoungergroups.Theexactreasonforthe decreasinghealthofthealready-poorremainsunobserved. Wenotethatitcannotbeattributedtofirst-ordereffects stemmingfromadecreaseinincome,asanysucheffects

33 _We_note_that_this_is_not_simply_a_result_of_the_second-order_income

polynomialusedintableA4.Experimentingwithotherincomefunctions,

suchasaﬂexiblesetofincomedummies,yieldedsimilarresultsforthe

errorandinequalitychangeterms.

wouldbecapturedbyourinequalitychangeterms,which provetobeunimportant.Instead,wespeculatethatthese dropsin reportedhealth may stemfrom mental health effectsofthethreatofincomeorjobloss.Usingsuicides asa proxyfor mentalhealth, thereissomeevidenceto suggestthatrisingunemploymentratesacrossEuropein 2008 wereassociated withincreased suicide rates (van GoolandPearson,2014;ToffoluttiandSuhrcke,2014).If theseeffectsweremoreconcentratedamongstthepoor, thismayexplainatleastpartofthedeteriorationinhealth thatweobserve.Furtherdiscussionisprovidedinthenext sectionfortheGreekcontext.

GreekausteritymeasuresandIRHI

The mostdrastic policychanges in this period were enactedinGreece.Inexchangefortwobailoutpackagesin 2010and2011,theGreekgovernmentintroduceda wide-rangingsetofausteritymeasures.Amongthesewerecuts in social transferssuchaspensions and unemployment beneﬁts,taxation ofpensionsabove D1400amonth by 5–10%,andfreezingmandatoryincreasesinpublic pen-sionsbetween2011and2015(OECD,2013).Wepresent theyear-by-yeardecompositionresultsbetween2010and 2013inGreece toexplore thepotentialimpactofthese austeritycutsinﬁguresA6-A8.

(13)

As mentioned above, the pattern for the inequality change term (see ﬁgure A7) for Greece between 2010 and 2013 is noticeably different from other countries, as the transfer term is positive while the market term is negative. Thedecrease in absoluteincome inequality derivingfromthelargedropinincome fromworkover thisperiodmeansthatmarketinequalitychangeis nega-tive,leadingtoreductions inIRHI.Thepositive signfor transferinequalitychangeindicatesthatthereductionin inequalitybetween2010and2013waslargerdueto mar-ketincome-relatedhealthchangesthanconsideringtotal income-relatedhealthchanges.Inotherwords:the redis-tributiveeffectoftransfersdeclinedasaresultofcutsin socialtransfersduetotheausteritymeasures,especially forpensions.

The consequences oftheausterity measuresare less obviouswhenlookingatthemobilityresultsbetween2010 and2013,thoughtheyarevisibleinthe2010–2011 com-parisonwhenthetransfermobilitytermislargeandpositive (figureA6).Theimmediateimpactwasaworseningofthe incomesoftheelderlyrelativetotheyoung,asthedropin pensionsbetween2010and2011waslargerthanthedrop inincomefromwork.Thisworsenedtherelativeincome positionofoldergroups,andincreasedIRHI.However,the transfermobilitytermswitchessigntobecomenegative between2011and2012.Thisisduetothesuddennatureof thecutintransferincomes,comparedtothemoregradual declineinmarketincomes.Whileincomeswerealready fallinginGreecebetween2010and2011,itisinthe sub-sequenttwoyearsthatthelargestfallsoccur(seefigure B2).Between2011and2013,incomesfromworkinGreece shrunksufficientlytooutweightheinitiallyIRHIincreasing effectsoftheausteritymeasures.

The error terms are the largest contributors toIRHI changeinGreeceduringthisperiod(ﬁgureA8),andwhile theyshowsimilarpatternstotheearlierrotationgroups, therearesomeidiosyncrasiesthatmaybeattributableto theausteritypolicies.Mostnotableisthattheerrorrank changes term is zero in the 2010–2011 comparison, in starkcontrasttothelater comparisons.Thismayreﬂect thesuddenshockoftheausteritypoliciesadministeredin 2010–2011,whichdisproportionatelyaffectedtheincome ranksofthose intheworsthealth.Thisappearstohave reducedtheIRHIdampeningeffectoftheerrorrankchanges term.However,aswiththetransfermobilitytermabove,the effectsoftheausteritymeasuresarewashedawayinthe subsequenttwoyearsasincomescontinuetodeclinefor thehealthiergroups.

Ofnotealsoisthelargemagnitudeoftheerrorresidual changesterms.Inthe2010–2013,thisisthetermwiththe largestcontributionofIRHIacrossallcountriesand com-parisonperiods.Thiscouldreflectthefactthatthecutsto thesocialsafetynetsandotherprograms,ontopofthe first-ordereffectsstemmingfromtheassociatedincomedrop, disproportionatelyaffectedthehealthofthealreadypoor inGreeceduringthisperiod.Tyrovolasetal.(2018)argue thatmortalityincreasesinGreeceafter2010maybedue tothesignificantcutstohealthcareexpenditureaspart oftheausteritymeasures.Thereisalsoindirectevidence thattheausteritycutsmayhaveworsenedmentalhealth, withaspikeinthesuiciderateinGreeceattheir

imple-mentation(Branasetal.,2015).While ourresultsarein linewiththeseﬁndings,ourdescriptiveanalysisisunable toattributethemtotheausteritypolicies.

Conclusion

Wemakeanumberofcontributionstotheliteratureon healthinequalities,bothintermsofmethodsand empiri-calresults.First,weadaptpreviouslyuseddecomposition methodstobetterclarifytheverydifferentrolesplayed bymarket versus transferincomes for IRHItrends.Our two-stepprocedurefurtheraccountsfortheroleplayed byhealthchangesunexplainedbyourmodel,andavoids ourhealthmeasurebeingpartlyincome-predicted.

Second,forarangeofEuropeancountries,weshowfor thefirsttimehowIRHIhasevolvedbetween2004and2013, atimeperiodthatcoversthelargestglobaleconomic con-tractioninthepost-warera.Wedocumentdistincttime andgeographictrendsinIRHI.Beforethecrisis,somecrisis countriessawIRHIrising.After2008,IRHIgenerally exhib-itedaflattrendacrossallcountries.Notably,wedonot evenfindlargeorsignificantpersistentincreasesinIRHIin thecrisisperiodforthosecountrieshithardestbytheGreat Recession.

Third,ourdecompositionmethoduncoversimportant newempiricalregularities concerningIRHIrisesorfalls. Weﬁndthatmarketincomeevolutiontendstoincrease inequalitiesinhealth, whiletherelationbetweensocial transfersandIRHIrevealsamoremixedpattern,insome casesdecreasingandinothercasesincreasingIRHI.This occursfortworeasons:(i)becausesocialtransfers–most importantlypensions–arelargely targetedatrelatively olderandotherpoorergroupswhoaretypicallyexcluded fromgainsintimesof(especiallymarket)incomegrowth; (ii)becauseinsomecountries,theyoungtendtostayand livelongerintheirparentalhouseholdandtherefore co-beneﬁtfromtheir(grand)parentpensionentitlements.

Fourth, we find indirect evidence – through resid-ualchanges inhealthnotexplainedbyincome, ageand migration –that thepoorestindividuals insomeof the crisiscountriesreportedlargerhealthlossesafterthe cri-sis.WhilethiswouldotherwisehaveincreasedIRHI,this effectislargelyneutralizedbytheprotectiveeffectof gov-ernmenttransferincomes.Thesecounterbalancingeffects explainwhyIRHInonethelessdidnotsignificantlyincrease duringthecrisis.Whilethecontributionof theresidual changestermsignalsthatthehealthofthealready-poor doesdecrease,wedonotfindevidencethatthisdecrease isdrivenbyincomelosses.Healthchangesduetoincome lossprovetoberelativelyunimportantindescribingIRHI change.Itmaybethefearoffuturelossratherthanactual lossthatcausesthisdrop.

Finally, our resultsdemonstrate that in at leastone countryausteritypoliciesbetween2010and2013havehad someeffectonIRHI:thelargereductionsinpension enti-tlementsthatwereenactedinGreeceinitiallydidincrease IRHI,andhavealsocounteredtheIRHIdampeningeffects oftransfersinlateryears.

Basedontheseempiricalﬁndings,ourresultssuggest thatgovernment transferpoliciescan anddo appear to haveasubstantialeffectonIRHI.Especiallyintimesof

(14)

cri-sis,thestickynatureofpensionsactasstabilizersthathelp toreduceIRHIbyimprovingtherelativeincomeposition oftheelderly.Inperiodsofeconomicgrowth,however,the stickinesstendstohavetheoppositeeffect,whentransfer growthtypicallylagsbehindmarketincomegrowth.

Ourﬁndings alsopoint totwo main potentialpolicy levers for governments concerned with rising levels of IRHIduringtimesofcrisis.Theﬁrstofthesearepolicies thatimprovetherelativeincomesoftheelderly,suchas moregenerouspensionschemes.Thesecondarepolicies thatpreventhealthdeteriorationforthepoor.Whilethe appropriatepolicyprescriptiondependsonthereasonsfor thesedeteriorations,ourmethodanddatadidnotallow tofurtherexploretheseunexplainedhealthchanges.We speculatethatamorerobustsocialsafetynet–whicheases thestressesonmentalhealthstemmingfrompotentialjob andhousingloss–aswellasschemeswhichensureaccess toqualityhealthcare,mayhelptostymieIRHIincreases. Finally,itisworthhighlightingthattheGreekexperience showsthatausteritymeasurescankillmuchoftheIRHI reducingeffectofpensionsduringcrises,andpotentially exacerbatethehealthdeclinesamongthepoor.

Allinall,ourﬁndingssuggestthatEuropeancountries havenotwitnessedtheriseinIRHIthatcouldbeexpected fromtheGreatRecessionthankstothestickinessof gov-ernmenttransfers,especiallypensionincomes,intheshort run:theyincreasesloweringoodtimes,butalsodecrease slowerinbad times.Ifthat stickinesscan nolongerbe affordedinthelongerrun–aswasthecaseinGreece post-crisis–then theirincomeprotectionand IRHIreducing effectmaybeeroded.

Acknowledgements

We acknowledge support from the NETSPAR pro-gramme,undertheComparativeResearchgrant.Thisstudy isbasedondatafromEurostat,EU StatisticsonIncome andLiving Conditions2004-2013.Theresponsibility for allconclusionsdrawnfromthedataliesentirelywiththe authors.We thank the editor, two anonymous review-ers,seminarparticipantsattheNetsparPensionDay2017, HESGWinter 2018Conference andtheNetspar Interna-tionalPensionWorkshop2018forusefulcommentsand suggestions.The usualcaveats apply,and all remaining errorsareourresponsibility.

AppendixA. Supplementarydata

Supplementarymaterialrelated tothis articlecanbe found, in the online version, at doi:https://doi.org/10. 1016/j.jhealeco.2019.102259.

References

Allanson,P.,Gerdtham,U.-G.,Petrie,D.,2010.Longitudinalanalysesof

income-relatedhealthinequality.J.HealthEcon.29,78–86.

Ásgeirsdóttir,T.L.,Ragnarsdóttir,D.Ó.,2013.Determinantsofrelative

andabsoluteconcentrationindices:evidencefrom26European countries.Int.J.EquityHealth12(1),53.

Baeten,S.,VanOurti,T.,VanDoorslaer,E.,2013.Risinginequalitiesin

incomeandhealthinChina:whoisleftbehind?J.HealthEcon.32 (6),1214–1229.

Branas,C.C.,Kastanaki,A.E.,Michalodimitrakis,M.,Tzougas,J.,Kranioti,

E.F.,Theodorakis,P.,Wiebe,D.J.,2015.Theimpactofeconomic

austerityandprosperityeventsonsuicideinGreece:a30-year interruptedtime-seriesanalysis.BMJOpen5(1),e005619.

Brekke,K.,Kverndokk,S.,AvailableatSSRN:2014.ImpactsofTransfers

fortheConcentrationIndex:ExplainingtheHealthEquality

Paradox?https://ssrn.com/abstract=2402705.

CDC,Foreword2013.Healthdisparitiesandinequalitiesreport-United

States,2013.MMWRSupl.62(3),1–2.

Coveney,M.,García-Gómez,P.,VanDoorslaer,E.,VanOurti,T.,2016.

HealthdisparitiesbyincomeinSpainbeforeandaftertheeconomic crisis.HealthEcon.25(S2),141–158.

Erreygers,G.,2009.Correctingtheconcentrationindex.J.HealthEcon.

28(2),504–515.

Erreygers,G.,VanOurti,T.,2011.Measuringsocioeconomicinequalityin

health,healthcareandhealthﬁnancingbymeansofrank-dependent indices:arecipeforgoodpractice.J.HealthEcon.30(4),685–694.

EuropeanCommission,2009a.SolidarityinHealth:ReducingHealth

InequalitiesintheEU.Brussels,20,2009.

EuropeanCommission,2009b.SolidarityinHealth:ReducingHealth

InequalitiesintheEU–ImpactAssessment.Brussels,20,2009.

Frick,J.R.,Ziebarth,N.R.,2013.Welfare-relatedhealthinequality:does

thechoiceofmeasurematter?Eur.J.HealthEcon.14(3),431–442.

GreekCityTimes,15May.Availableat:

https://greekcitytimes.com/2019/05/15/greek-adults-live-longer-wit h-their-parents-compared-to-other-nationalities-in-the-world/

(Accessed:23September2019).2019.GreekAdultsLiveLonger

WithTheirParentsComparedtoOtherNationalitiesintheWorld.

Jenkins,S.P.,Brandolini,A.,Micklewright,J.,Nolan,B.,2012.TheGreat

RecessionandtheDistributionofHouseholdIncome.Oxford UniversityPress.

Kreiner,C.T.,Nielsen,T.H.,Serena,B.L.,2018.Roleofincomemobilityfor

themeasurementofinequalityinlifeexpectancy.PNAS115(46), 11754–11759.

Lambert,P.L.,2001.TheDistributionandRedistributionofIncome,3rd

ed.ManchesterUniversityPress,Manchester.

Lindelow,M.,2006.Sometimesmoreequalthanothers:howhealth

inequalitiesdependonthechoiceofwelfareindicator.HealthEcon. 15,263–279.

Mackenbach,J.P.,Kunst,A.E.,Cavelaars,A.E.,Groenhof,F.,Geurts,J.J.,EU

WorkingGrouponSocioeconomicInequalitiesinHealth,1997.

SocioeconomicinequalitiesinmorbidityandmortalityinWestern Europe.Lancet349(9066),1655–1659.

Mackenbach,J.P.,Stirbu,I.,Roskam,A.J.R.,Schaap,M.M.,Menvielle,G.,

Leinsalu,M.,Kunst,A.E.,2008.Socioeconomicinequalitiesinhealth

in22Europeancountries.N.Engl.J.Med.358(23),2468–2481.

Mackenbach,J.,Valverde,J.,Artnik,B.,Bopp,M.,Brønnum-Hansend,H.,

Deboosere,P.,Kalediene,R.,Kovács,K.,Leinsaluh,M.,Martikainen,

P.,Menvielle,G.,Regidor,E.,Rychtaríková,J.,Rodriguez-Sanz,M.,

Vineis,P.,White,C.,Wojtyniak,B.,Hua,Y.,Nusselder,W.,2018.

Trendsinhealthinequalitiesin27Europeancountries.Proc.Natl. Acad.Sci.115(25),6440–6445.

OECD,2012.SocialSpendingDuringtheCrisis.OECDPublishing,Paris.

OECD,2013.PensionsataGlance2013:OECDandG20Indicators.OECD

Publishing,Paris.

Plotnick,R.,1981.Ameasureofhorizontalinequity.Rev.Econ.Stat.2,

283–288.

Ruhm,C.J.,2000.Arerecessionsgoodforyourhealth?Q.J.Econ.115(2),

617–650.

Toffolutti,V.,Suhrcke,M.,2014.Assessingtheshorttermhealthimpact

oftheGreatRecessionintheEuropeanUnion:across-countrypanel analysis.Prev.Med.64,54–62.

Tyrovolas,S.,Kassebaum,N.J.,Stergachis,A.,Abraha,H.N.,Alla,F.,

Androudi,Haro,J.M.,2018.TheburdenofdiseaseinGreece,health

loss,riskfactors,andhealthﬁnancing,2000–16:ananalysisofthe GlobalBurdenofDiseaseStudy2016.LancetPublicHealth3(8), e395–e406.

VanDoorslaer,E.,Jones,A.M.,2003.Inequalitiesinself-reportedhealth:

validationofanewapproachtomeasurement.J.HealthEcon.22(1), 61–87.

VanDoorslaer,E.,Koolman,X.,2004.Explainingthedifferencesin

income-relatedhealthinequalitiesacrossEuropeancountries. HealthEcon.13(7),609–628.

VanDoorslaer,E.,Wagstaff,A.,Bleichrodt,H.,Calonge,S.,Gerdtham,U.G.,

Gerﬁn,M.,O’Donell,O.,1997.Income-relatedinequalitiesinhealth:

someinternationalcomparisons.J.HealthEcon.16(1),93–112.

VanGool,K.,Pearson,M.,2014.Health,AusterityandEconomicCrisis:

AssessingtheShort-TermImpactinOECDCountries(No.76).OECD Publishing.

(15)

VanOurti,T.,2003.Socio-economicinequalityinill-healthamongstthe elderly:shouldoneusecurrentorpermanentincome?J.Health Econ.22(2),219–241.

VanOurti,T.,VanDoorslaer,E.,Koolman,X.,2009.Theeffectofincome

growthandinequalityonhealthinequality:theoryandempirical

evidencefromtheEuropeanPanel.J.HealthEcon.28(3), 525–539.

Wagstaff,A.,Watanabe,N.,2003.WhatdifferencedoesthechoiceofSES

Thank goodness for stickiness

Journal

of

Health

Economics

Thank

goodness

for

stickiness:

Unravelling

the

evolution

of

income-related

health

inequalities

before

and

after

the

Great

Recession

in

Europe

Max

Coveney

,

Pilar

García-Gómez

,

Eddy

van

Doorslaer

,

Tom

Van

Ourti

a

r

t

i

c

l

e

i

n

f

o

a

b

s

t

r

a

c

t

















































_,

_Pilar

_{García-Gómez}

_,

_Eddy

_van

_Doorslaer

_,