How disability insurance reforms change the consequences of health shocks on income and employment

(1)

ContentslistsavailableatScienceDirect

Journal of Health Economics

j ourna l h o m e pa 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

How disability insurance reforms change the consequences of health shocks on income and employment

^夽

Patrick Hullegie

^a,b

, Pierre Koning

^a,b,c,∗

aDepartmentofEconomics,VUAmsterdam,P.O.Box80510,1081HV,Amsterdam,theNetherlands

bTinbergenInstitute,theNetherlands

cLeidenUniversity,TinbergenInstituteandIZA,Germany

a r t i c l e i n f o

Articlehistory:

Received15November2017 Receivedinrevisedform 11September2018 Accepted11September2018 Availableonline26September2018

JELclassiﬁcations:

H53 J14 J18

Keywords:

Disabilityinsurance Healthshocks

Difference-in-differenceanalysis Workaccommodations

a b s t r a c t

ThispaperexamineswhetherDutchdisabilityinsurancereformshavehelpedorhinderedemployment opportunitiesofworkersthatarefacingunanticipatedshockstotheirhealth.Animportantcomponent ofthereformswastomakeemployersresponsibleforpayingsicknessbeneﬁtsandtostrengthentheir sicknessmonitoringobligations.Thismaystimulatepreventiveandreintegrationactivitiesbyﬁrms.

Usingadministrativedataonhospitalizations,weconcludethatbothﬁnancialincentivesandmonitoring obligationshavesubstantiallyloweredDIreceiptandincreasedtheemploymentofworkersafterahealth shock.

1. Introduction

Inthepasttwo decadestheOECDhasregularlyvoicedcon- cernoverthelabormarketpositionofpeoplewithdisabilitiesand thecostofdisabilityinsurance(DI)programs(OECD,1992,2003, 2010).Improvingthelabormarketpositionofpeoplewithdisabil- itiesisnotonlyimportantfortheirowneconomicwell-being,itis

夽 Wegratefullyacknowledgetwoanonymousrefereesfortheirconstructiveand valuablecommentstothepaper.Wealsohavebenefittedfromtheopportunity topresentthisworkatthe5thCAFEWorkshop,the11thWorldCongressofthe EconometricSociety,andinaseminarattheErasmusSchoolofEconomics,Rotter- dam.StatisticsNetherlandshasprovidedaccesstothedatathatwasusedinthis projectthrougharemoteconnectionfacility.Aspartofthedataagreement,Statis- ticsNetherlandshastherighttoreviewtheresultsofourprojectpriortotheir disseminationtoensurethattheconfidentialityofthedataisnotunintentionally compromisedandindividual-specificinformationisnotrevealed.Allerrorsareour own.∗ Correspondingauthorat:DepartmentofEconomics,VUAmsterdam,P.O.Box

80510,1081HV,Amsterdam,theNetherlands.

E-mailaddresses:p.g.j.hullegie@vu.nl(P.Hullegie),p.w.c.koning@vu.nl (P.Koning).

alsoconsideredessentialinaddressingthechallengesthatcoun- triesfaceregardingpopulationaging(OECD,2010).Recognizing theneedforreform,manycountrieshaveimplementedchangesto theirdisabilityprograms.

Theobjectiveofthispaperistoexaminewhetherthedisability reformsthatwereimplementedintheNetherlandshavehelpedor hinderedthecontinuationofworkforindividualswithhealthprob- lemsordisabilities.TheNetherlandspresentsaninterestingsetting becausethegovernmentfundamentallyreformeditsdisabilitypro- gram.Inthe1980sand1990stheDutchdisabilityprogramwas consideredtobethemostout-of-controldisabilityprogramwithin theOECD,astatussometimesreferredtoasthe“Dutchdisease.”¹To illustrate,in1990theNetherlandsspent4.7percentofitsGDPon disabilityinsurance–whichwas2.2percentagepointshigherthan

1Thephrase“Dutchdisease”originallyreferredtothewayinwhichthemanu- facturingsectorintheNetherlandswasadverselyaffectedbydiscoveriesofnatural gasinthelate1950s.Meanwhile,ithasbecomeanumbrellatermfortheproblems facedbyeconomieswithhighlevelsofenergyorothernaturalresources.Forlabor economistsitalsoreferstothesharpincreaseindisabilityrollsintheNetherlands betweenthe1960sand1980s.

https://doi.org/10.1016/j.jhealeco.2018.09.004

0/).

(2)

Norway,thesecondbiggestspenderondisabilityinsuranceinthe OECD–andmorethanthreetimesaslargeastheOECDaverage of1.3percent(OECD,2010).Duetothereformsthedisabilitypro- gramtransformedfromonethatmerely paidbeneﬁtstoonein whichemployersplayanimportantroleinreintegratingdisabled workers,andspendingdroppedtolessthantwopercentofGDPin 2010.

AnimportantcomponentoftheDutchreformswastoenhance employerincentives,whichwasdonebymakingthemresponsible forpayingsicknessbenefitsandbystrengtheningtheirsickness monitoringobligations. Especiallythelatter, asspecified inthe so-called“Gatekeeperprotocol”enactedinApril2002,iswidely consideredtobethemosteffectiveDIreformintheNetherlands (seeSection2formoredetails).Anotherreformentailedtheprolon- gationofthesickpayperiodforwhichemployersareresponsible fromonetotwo yearsin2004.Therationalebehindenhancing employerincentiveswasthattheycouldstimulatepreventiveand reintegrationactivities,andworkplaceaccommodationsforsick anddisabledworkers,therebyimprovingtheirlabormarketoppor- tunities.Inthisrespect,theDutchSurveyofWorkingConditions (NEA)of2014reports17%oftheworkingpopulationreceivingone ormoreworkplaceaccommodations(NEA,2014).Amongstthese workaccommodations,themostimportantonesincludephysical workadaptations,changesinworkingtimeortheworkingschedule andjobchangeswithinthefirm.²However,asaconsequenceofthe reformsemployersareconfrontedwithsubstantialcostswhenan employeegetssick.Thesecostsarenotonlymonetary,butalsoarise fromincreasedmonitoringobligationsandthedifficultyoftermi- natingthecontractsofworkerswithhealthproblemsordisability (OECD,2010,p.135).

While one would expect increasedemployer incentives and obligationsinthesicknessperiodthatprecedesDIclaimstolower theprobabilityoferroneousadmissionsintoDI,thispaperfocuses onemployedindividualsthatwerenotbeabletocontinueworking forsomeperiodoftimeandforwhichfirmsweresupposedtoexert preventativeandreintegrationactivities.Theideaisthusthatthese activitiesshouldstrengthenthepositionofworkerswithdisabili- ties–atleastthosethatareemployed.Inthiscontext,ouranalysis uses administrative data from hospital admissions to estimate changesintheeffectsofhealthshocksonemploymentandincome of (formerly) employedindividuals. Our data are fromhospital admissionrecordswithuniversalpopulationcoveragetodefinea suddendeteriorationofhealth(or“healthshock”)asanunsched- uledhospitaladmissionthatrequiresimmediatetreatment.With thisinformation,weutilizevariationinhealththatislessproneto measurementerrorrelativetoself-reportedhealthmeasuresand arguably(more)exogenoustolabormarketstatus.Wethusavoid identificationfromaself-reportedhealthmeasurethatispossibly endogenoustolaborforcestatus(Bound,1991;Kreider,1999)or affectedbythereformsthroughchanged(social)normsforreport- ingadisability.Wecombinethehospitaladmissionrecordswith administrativedatafromseveralothersources,whichtogetherpro- videapopulation-levelpaneldatasetwithinformationforevery personabouttheirdemographiccharacteristics,healthstatusand labormarketoutcomes.

OuranalysisismostrelatedtotheworkofGarcía-Gómezetal.

(2013), who use similar data to identify the causal impact of acute hospitalizations onemployment and income. We extend onthis workbystudyingchangesin healthshockeffects across workercohortsthatwerehospitalizedin2000and2005inorder toevaluatethepreviouslymentionedDIreforms.Ourpaperalso

2FortheU.S.,Zwerlingetal.(2003)reportﬁndingsonworkplaceaccommodations thatarebasedontheNationalHealthInterviewSurveyDisabilitySupplement(NHIS- D).Theyﬁnd12%oftherespondentstoreceiveworkplaceaccommodations.

contributestoalimitedliteraturethatevaluatespoliciesthatincen- tivizeemployerstoimprovelabormarketoutcomesofpeoplewith disabilities–seee.g.Koning(2016)forabriefoverview.Assuch, wefocusonemployerresponsibilitiesthatgoconsiderablybeyond theimpositionofworkplaceaccommodations–likee.g.the1990 AmericanswithDisabilitiesAct(ADA)intheU.S.³ –andlargely applytothesicknessperiodthatprecedesDIclaims.

Weﬁndthatthelabormarketpositionofworkerswhoexpe- rience ahealth shockhasimprovedaftertheseries ofreforms:

theyare less likely toreceive disability insurancebenefits and theyaremore likely toremainemployed.Also, wedo not find strongevidencepointingatsubstitutioneffectsintoothersocial schemesas aresultof thereform,likeincreasesin UIorsocial assistancebenefits.Based onmoreyear-to-yearcomparisonsof newworkercohortsthatfacedhealthshocks,weinferthatboth theGatekeeperreformin2002andtheextensionofthesickpay periodfromonetotwoyearsin2004havecontributed.Overall, theDIreformsimplementedbytheDutchgovernmenthavepro- tecteddisabledindividualswhoalreadyhaveajob.Ouranalysis thusshowsthatenhancingemployerincentivesmightbeafruitful waytoamoresustainablegrowthpathofDIprograms.Thiscon- firmsreviewsofe.g.AutorandDuggan(2006),Autor(2011)and KoningandLindeboom(2015)thatstresstheroleofemployersin enhancingreturn-to-workofsicklistedworkers.

Theremainderofthispaperisorganizedasfollows.Section2 detailstheinstitutionalcontextintheNetherlands.Theempirical analysisbasedontheadministrativedataispresentedinSection3.

Toputtheseﬁndingsinabroaderperspective,Section4discusses thedescriptiveanalysisbasedontheDutchLaborForceSurvey.

Finally,inSection5wediscusshowourﬁndingsmaybeuseful forcountries,suchastheU.S.,thatfacearapidandunsustainable expansionoftheDIbeneﬁciarypopulation.

2. Institutionalcontext

The provision of disability insurance in the Netherlands is mandatory and covers all employees against all income losses resultingfromimpairmentsthatoccurredonthejoborelsewhere.

Since2004workersapplyforDIclaimsaftera“waitingperiod”of twoyearsofsickness.Duringthisperiod,employersareresponsible fortheprovisionofreintegrationactivitiesandthecontinuedpay- mentofwages.Next,disabilityclaimsareassessedbytheSocial SecurityAdministration(SSA). Disabilitybeneﬁtsdependonthe

“degreeofdisability,”whichisdeﬁnedasthepercentagediffer- encebetweenpriorearningsandtheremainingpotentialearnings capacity.

Therearethreekeydifferencesbetweenthedisabilityprograms intheNetherlandsandtheUnitedStates.First,unlikeintheU.S., workersin theNetherlands mayreceivepartialdisabilitybene- fits.Hence,DutchDIbeneficiariesmaysimultaneouslyworkand receivedisabilitybenefits.Second,disabilitybenefitsonlydepend onthe“degreeofdisability”andnotthenumberofdependents and/orworkhistory.Forfullydisabledindividuals,disabilityben- efitsprovideinsurancefor70percentofthelossofincomedueto impairments.Third,healthinsurancecoverageisuniversalinthe NetherlandsandnottiedtoDIreceipt(oraperson’sjob).

3VariousempiricalstudieshaveexaminedtheconsequencesoftheADA,which intendstobandiscriminationandmandates“reasonableworkplaceaccommoda- tions.”WhileDeLeire(2000)andAcemogluandAngrist(2001)ﬁndsupportfor adverseeffectsoftheADAontheemploymentofdisabledworkers,BeegleandStock (2003)andKruseandSchur(2003)challengetheseﬁndings.BoundandWaidmann (2002)providesuggestiveevidencetherapidgrowthoftheDIprogramduringthe 1990splayedanimportantroleinexplainingthedeclineintheemploymentrateof peoplewithdisabilities.

(3)

Fig.1.DisabilityInsuranceRecipiencyandAwardRatesperAdultAged20–65.

Source:Author’scalculationsfromdataofStatisticsNetherlandswhicharepublicly availablethroughstatline.cbs.nl

ThereisgoodreasontobelievetheDutchdisabilityprogramhas beenplaguedbymoralhazardproblemsinthepast.Toillustrate, Fig.1plotsthefractionoftheworking-agepopulation(age20–65) thatreceivesDIbenefitsandthefractionthatisnewlyawardedDI benefitsfortheperiod1968–2010.Thefractionoftheworking- agepopulationreceivingDIbenefitsquicklyrosefrom2percent in1968toabout7–8percentinthemid1980s,remainedroughly constantatthisunprecedentedlevelforthenexttwodecades,and startedtodeclineatthebeginningofthe21stcentury.

Oneimportant institutional feature that gave leeway tothe sharpriseinenrollmentis thatdisabilityinsuranceincludesall medicalcontingencies.Parsons(1991)arguesthatabroaddeﬁni- tionofdisabilityrisksincreasesthelikelihoodofscreeningerrors intodisabilitydeterminations.Asitseems,applicantssuccessfully exploitedthisfeatureoftheDutchDIsystem,withtheSocialSecu- rityAdministrationprioritizingonminimizingerroneousdenials (Burkhauseretal.,2008).Inaddition,moralhazardeffectswere aggravatedbysicknessbeneﬁtsthatfullyreplacedwagesduring thewaitingperiod.Assuch,incentivestoresumeworkquicklywere limited.

AlthoughacontinuousneedforreformintheDIsystemwasfelt sincetheeighties,ittookuntil2002toattainsubstantialdecreases intheinflowrate.Priorto2002,policychangesaimedatincreasing thefinancialincentivesforemployerstoreduceDIenrollment.In particular,employersbecameresponsibleforwagepaymentofsick listedworkersin1996,andDIbenefitcostsareexperiencerated forpermanentworkerssince1998.Also,thewagepaymentperiod wasextendedtotwoyearsin2004.Whilethereisevidencethat theseincentiveshavegraduallycontributedtomorepreventative andreintegrationactivitiesofemployers,theintroductionofthe so-called“Gatekeeperprotocol”(inApril2002)isgenerallycon- sideredasthemosteffectivepolicychangethathastakenplace.

TheGatekeeperprotocolspeciﬁesallthelegalresponsibilitiesof employersandtheirsicklistedworkers.ThismeanstheSocialSecu- rityAdministrationisnolongerinvolvedintheprocessofchecking andreintegratingsickworkers,butmerelyactsasagatekeeperof theDIprogram.TheGatekeeperprotocolforcesemployerstofocus theirattentionattheonsetofasicknessperiod.Incontrastwiththe gradualimpactofincreasedemployerincentivesinthenineties,it seemsthattheprotocolhashadanalmostimmediateandpersis-

tentimpactontheDIinﬂowrate(DeJongetal.,2011;Koningand Lindeboom,2015).⁴

TounderstandwhytheGatekeeperprotocolhasbeensuccessful incurbingDIinflow,itisimportanttorealizethatitprescribesa seriesofactionsthatshouldbetakeninordertobeeligibleforDI benefits.Aftersixweeksofabsence,theemployerandemployee shouldmakeafirstassessmentonthemedicalcauseandthefunc- tionallimitations.Onthebasisofthisassessment, theydraftan accommodationandrehabilitationplan(or“reintegrationplan”) thatspecifiesthestepsthatshouldbetakentoresumeworkatthe currentornewjobandtheaccommodatedcircumstancesthatare neededforthis.Atthesametime,acasemanageroftheSocialSecu- rityAdministrationisappointedanddatesaredeterminedatwhich theplanwillbeevaluated.Theplanshouldbefinalizedintheeighth weekofabsence.Iftheworkerhasnotreturnedtoworkafterabout threemonthspriortotheendofthesicknesswaitingperiod,hefiles aDIbenefitclaim.Thecasemanagerdecideswhetherthereinte- grationeffortsoftheemployerandemployeehavebeensufficient.

Ifthisconditionismet,adoctorfromtheSocialSecurityAdminis- trationdeterminesthedegreeofdisabilityofthisworkerattheend ofthewaitingperiod.Incaseofnegligence,theemployerisheld responsibleandhastocontinueprovidingsickpayforamaximum oftwelvemonths.

InlightofthedramaticdecreaseofnewDIawardssince2002 (seealsoFig.1),KoningandLindeboom(2015)arguethatthepro- tocolhasacceleratedthecost-andriskawarenessofemployers,as wellasthespeciﬁcwaysthatareneededfordisabilityprevention.

Short-termabsenteeismcouldnolongerbeleftunnoticedbyman- agers,withtheprotocolprovidingguidancetoemployersintheir newroleinthereintegrationprocess.Asitseems,employershave becomemoreawareofthecostsoftwoyearsofcontinuedwage payments,aswellastheDIbeneﬁtcoststhatwerepassedthrough intheirDIpremiums.Notsurprisingly,however,criticismagainst theemployerincentivesandobligationshasgrownaswell.Similar toexperienceswiththeADAintheU.S.thatmandatesemployers toprovidereasonableaccommodationstoemployeeswithdisabil- ities,theadditionalresponsibilitiesmayhaveinducedemployers tohirenewworkerswithalowriskofmovingintopoorhealth, thusreducingthecostsassociatedwithsicknessordisability.

Personswhoarenotawardeddisabilityinsurancebenefitsmay insteadapplyforunemploymentinsurance(UI)benefitsiftheyare unemployed.Duringtheperiod2000–2010theUIreplacementrate was70percent.Themaximumentitlementperiodwasdependent onaperson’semploymenthistory,andrangedfromaminimum ofsixmonthstoamaximumoffiveyears.Ifapersonisnoteligi- bleforeitherdisabilityorunemploymentinsurance,hemayapply forsocialassistance.Thispayssubstantiallylowerbenefitswhich areunrelatedtopreviousearnings,andismeanstested.Withthe exceptionofareductioninthemaximumdurationfrom60to38 monthsinOctober2006,therehavebeennomajorreformsinthe UIprogramduringtheperiodofanalysis.

Finally, health insurance coverage was universal in the Netherlandsduring theperiod ofanalysis.Moreover,and more importantly,theDutchhealthinsurancesystem,bycontrasttothe DIsystem,hasneverbeenperceivedasasourceoflabormarket distortions.

4Inparticular,theDIinﬂowratedroppedfrom0.71%to0.52%oftheinsured populationoneyearafterthestartoftheGatekeeperprotocol,in2003(Statistics Netherlands,2017).Asthenumberofsicknessabsencesdidnotdecreasenoticeably duringthoseyears,itseemsthattheprotocolmainlyshortenedthelengthofsickness spellsandthusloweredtheprobabilityofDIapplications(absenceincidenceand durationdatacanberetrievedfromcbs.overheidsdata.nl/70812ned).

(4)

3. Dataandempiricalimplementation 3.1. Dataandsampleselection

Inourempiricalanalysis,weuserichadministrativedatafrom hospitaladmissionrecords,socialsecurityrecords,andthemunic- ipalityregisters,whichcanbelinkedthroughauniqueidentiﬁer for each individual.Taken together,theyprovide a population- levelpaneldatasetwithinformationforeverypersonabouttheir demographiccharacteristics,hospitaladmissions,andlabormarket outcomes.

Thehospitalregisterdatacontainsinformationonbothinpa- tientanddaycarepatientsofalmostallhospitalsintheNetherlands from1999to2005.Foreachhospitaladmissionweobserve(i)the admissionanddischargedata,(ii)whetherornotitwasanacute admission,and(iii)themaindiagnosis.WefollowGarcía-Gómez etal.(2013)andidentifyasuddendeclinein health(or“health shock”)asanunscheduledhospitalizationthatrequiresimmediate treatmentandinvolvesastayofatleastthreenights.Theadmis- sionsarerequiredtoinvolveastayofatleastthreenightsbecause unscheduledandacute hospitalizationsmayinclude lesssevere healthproblemssuchasamildheadinjury.Wealsoexcludehospi- talizationsduetopregnancyandchildbirth.Duetotheunscheduled andacutenatureofthehospitaladmissionsitisplausiblethatthey areexogenoustolabormarketoutcomes.⁵

Wedeﬁneworkerswhoexperiencedanunscheduledandacute hospitalizationofatleastthreenights,excludingthoserelatedto pregnancyandchildbirth,asthe“treatmentgroup.”Workerswho didnotexperienceahospitalizationatallformthe“controlgroup.”

Thismeansthatworkerswithtypesofhospitalizationsotherthan thoseinthetreatmentgroupareexcludedfromtheanalysis.

Treatmentandcontrolcasesarefurtherrestrictedtopersons whointheyearpriortothepotentialhealthshockwere:(i)aged 25–58,(ii)working–excludingthosewhoareondisabilitybeneﬁts intheyearoftheshock,sincetheymusthavebeenonsickness beneﬁtsintheyearbeforetheshock,and(iii)notadmittedtoa hospital.⁶ Thesesampleselectioncriteriaaresimilartotheones imposedbyGarcía-Gómezetal.(2013).FollowingBorghansetal.

(2014),wealsoexcludeallindividualswhoappearonmorethan onedisabilityschemeatthesametime (withina year;about3 percentofthesample),becauseitisnotclearwhethertheyresult fromadministrativeorcodingerrors,orwhetherthesepersonsare trulyentitledtomultipledifferentDIschemes.Byexcludingthese cases,wefocusonworkersforwhichwearesurethatthedisability reformsapply–thatis,theGatekeeperprotocolandtheextension ofthesickpayperiod.

3.2. Empiricalimplementation

ToinvestigatetheeffectsoftheDIreformsontheemployment opportunitiesofpeoplewithhealthproblemsordisability,wecom- paretheeffectofahealthshockinayearbeforethereformstothe effectofahealthshockinayearafterthereform.Ourfocusison 2000and2005asthetwoyearsinwhichahealthshockcanpoten-

5Itshouldbenotedthatmentaldisordersareunderreportedinthehealthshocks westudy.Assuch,wemissanimpairmenttype,whichisoneofthemajorcontributor toDIinﬂownowadays(seealsoTable1).Withemploymentratesthataregenerally lowerthanforanyothertypeofimpairment–seee.g.OECD(2010)–thereisa strongcaseforincreasedresearcheffortforthisparticulargroup.

6Inordertohaveasamplethathasastronglaborforceattachment,werestrictour sampletoworkersbelowtheageof59.Inthetimeperiodunderinvestigation,older workersﬁrstoptiontoretireearlywastypicallyattheageof60–seee.g.Vermeer (2013).BearinginmindthatthewaitingperiodforDIapplicationswasextendedto twoyears,theagecutoffof58thuspreventsusfrominvestigatingshockeffectsfor workersthatmayhaveoptedforearlyretirement.

tiallyoccur.Giventhesampleselectioncriteriadescribedbefore anddataavailability,2000istheearliestand2005thelatestpossi- bleyear.Moreover,andmoreimportantly,peoplewhogetdisabled in2000arenotaffectedbyintroductionoftheGatekeeperproto- col(in2002)andprolongationofﬁnancingsickpayforemployers (in2004),whereasthosewhogetdisabledin2005are.Recallthat especiallytheintroductionoftheGatekeeperprotocolisconsid- eredtobethemosteffectiveDIreformintheNetherlands,andthat itsincentivesenhancedasaresultoftheextensionofthesickness beneﬁtperiodfromonetotwoyearsin2004.Thisselectionresults inasampleof31,386unscheduledandacutehospitalizationswith astayofatleastthreenightsin2000,and27,911hospitaladmis- sionsin2005.Asweexpectthattheimpactofhealthshocksvaries byageandgender,wewillconductseparateanalysesformenaged 25–39,menaged40–58,andforwomeninthesetwoagegroups.⁷ Weconjecturethatolderpeoplearemorelikelytoexperiencea moreseriousdeteriorationofhealth,andhavefewerincentivesto investinworkresumptionduetolowerremainingworking-life expectancy(Charles,2003).

Table1showstherelativefrequencyoftheunscheduledadmis- sions in 2000 by diagnoses on the basis of the International ClassiﬁcationofDiseases9(ICD-9)forfourdemographicgroups, bothforallobservationsandforthesubsetthatenteredintoDIin 2001.Descriptivestatisticsfor2005lookverysimilarandarethere- foreomitted.Notsurprisingly,therearemanymoreunscheduled hospitalizationsamongmenandwomenaged40–58thanamong theircounterpartsaged25–39.Moreover,therearenotablediffer- encesintherelativeimportanceofcertaindiseases.Forexample, formenaged40–58,diseasesofthecirculatorysystemarethemost importantcauseofhospitalizationandaccountfor36percentof alladmissions,whereasforyoungermenthisisonly10percent.

Moreover,circulatorydiseasesaccountfor9percentofthehospital admissionsamongwomenaged25–39,whereasitaccountsfor21 percentofthehospitaladmissionamongwomenaged40–58.For thelattergroup,circulatorydiseasesarealsothemostimportant causeofhospitalizations,closelyfollowedbydiseasesofthediges- tivesystem.Bycontrast,injuriesarethemostimportantcauseofa hospitaladmissionamongmenaged25–39,accountingfor27per- centoftheadmissions.Amongallothergroups,injuriesaccountfor only14percentoftheadmissions.Theseexamplesillustratethat therearemarkeddifferencesinthecausesofhospitaladmission betweenmenandwomen,andbetweenagegroups.

Ofthesampleofunscheduledhospitaladmissionsin2000,7 percententeredintoDIin2001; thisgroupconstitutesabout4 percentofthetotalnumberofDIadmissionsinthatsameyear.

Althoughdifferencesintheregistrationofthedatarenderitdif- ficulttomakecomparisonsbetweentheimportanceofmedical conditionsinthehospitalrecordsandtheSocialSecurityAdmin- istration(SSA),twobroaderobservationsstandout.First,boththe shareofmentaldiseasesandtheshareofmusculoskeletaldiseases ofworkerswithunscheduledhospitalizationsthatwereawarded DIbenefitsaremarkedlylowerthanfortheDIinflowatlarge.In particular,theshareofawardeeswithmentalimpairmentsranges betweenabout30percentforoldermento45percentforyounger women, whereasthese sharesdonot exceed16 percentin the smallersamplewhowerehospitalizedintheyearpriortoadmis- sion in the DIprogram. Likewise, less than 8 percent of those hospitalizedin2000and receivingDIin2001wereadmittedto hospitalsduetodiseasesofthemusculoskeletalsystem,whereas itconstitutesamuchlargershareamongallDIadmissions.⁸Sec-

7TablesA.1andA.2intheappendixtothispapershowsummarystatisticsfor thesefourgroups,bothforthecohortof2000andof2005.

8Itislikelythatpartoftheworkersbeinghospitalizedwithinjuries(orpoisoning) arediagnosedashavingmusculoskeletalconditionsintheDIclaimsassessment.

(5)

Table1

Treatmentcasesbydiagnosisforfourdemographicgroupsin2000.

Men Women

Age25-39 Age40-58 Age25-39I Age40-58

All DI All DI All DI All DI

Infectiousdiseases 5.95 2.87 2.52 1.70 4.60 1.38 2.64 0.38

Neoplasms 1.72 5.46 3.90 3.71 2.68 4.84 6.57 8.76

Endocrinedisorders 1.98 1.44 1.42 0.95 1.94 2.08 1.40 2.48

Diseasesofthebloodandblood-formingorgans 0.64 0.57 0.45 0.32 1.02 2.08 0.87 0.95

Mentaldisorders 2.75 8.62 1.58 3.50 4.70 15.22 2.77 5.71

Diseasesofthenervoussystem 2.93 4.31 2.65 3.82 2.90 3.81 2.62 3.24

Diseasesofthecirculatorysystem 9.99 16.09 36.46 47.51 8.65 15.22 21.03 28.38

Diseasesoftherespiratorysystem 9.58 4.02 5.58 2.23 8.45 4.15 7.10 6.67

Diseasesofthedigestivesystem 19.58 6.61 14.56 4.88 22.37 9.69 19.06 8.19

Diseasesofthegenitourinarysystem 2.35 0.86 2.27 1.17 10.18 4.50 5.23 2.86

Diseasesoftheskin 2.89 0.29 1.29 0.42 2.04 1.04 1.30 1.90

Diseasesofthemusculoskeletalsystem 5.42 6.32 4.12 4.03 4.38 7.61 4.71 7.81

Congenitalanomalies 0.19 0.29 0.17 0.00 0.10 0.35 0.12 0.19

Symptoms,signs,andill-deﬁnedconditions 7.43 6.03 8.82 8.27 11.57 8.65 9.96 7.43

Injuryandpoisoning 26.60 36.21 14.22 17.50 14.40 19.38 14.62 15.05

Majorhealthevent 5.51 13.22 25.16 29.37 2.76 6.93 12.88 18.48

Chronicillnesses 26.26 31.61 28.04 35.10 27.48 45.67 29.62 41.33

Accidents 68.23 55.17 46.80 35.53 69.75 47.40 57.50 40.19

Observations 7,026 348 13,484 943 4,890 289 5,986 525

Notes:percentagesarecalculatedfromtreatmentcasesusedintheestimation,andarethusrestrictedtounscheduledandacutehospitaladmissionthatlastatleastthree nights,topersonsintherelevantagerange,whowereworkingandnothospitalizedinthepreviousyear.

ond,weﬁndthatthesharesofhospitalizedworkersduetodiseases ofthecirculatorysystemandthedigestivesystemareoverrepre- sentedcomparedtotheDIinﬂowatlarge.Thissuggeststhatthe healthshockswestudyarelessstronglyrelatedtoworkhazard thanthoseendinginDIadmission.Atthesametime,ofcourse,the comparisonindicatesthatmedicalconditionsthatrequirehospital admissionoftendeviatefromthosethatjustifyDIadmissionata laterstage.

Whenspecifyingtheeffectofhealthshocksonemploymentand otheroutcomemeasures,weﬁrstadoptadifference-in-difference structurethatfocusesonthedifferenceinthe“healthshock–no healthshock”outcomesbetween2005and2000.Weinterpretthis differenceinthelightofthemajorDIreformsthathavetakenplace.

Thisyieldsthefollowingmodel:

Yit=˛t+ˇtSi+tTi+ıt(Si·Ti)+_st Xi+εit,

t=1,2,3,4, (1)

whereireferstotheperson;ttothenumberofyearspassedsince theyearoftheshock;S_iindicateswhetherornotpersonihada healthshock;T_iequals1iftheyearofthehealthshockis2005, andzerootherwise.˛trepresentsthebaselineleveloftheoutcome measureattimet.Xisavectorofcovariates,includingdummiesfor ﬁveyearagegroups,nationality(Native,Nonnativenon-Western, NonnativeWestern),householdsize,municipalitysize,province, andlaborincomefromtheyearpriortothepossiblehealthshock (inquartiles).Theeffects ofthesecovariates areallowedtodif- ferbothbythetimepassedsincethepossiblehealthshock,and bywhetherornotapersonactuallyhadahealthshock(asindi- catedbythesubscriptssandton).Furthermore,thesubscriptt ontheotherparametersindicatesthatweallowtheeffectstodif- ferbythetimepassedsincethehealthshock.Insteadofestimating themodelforeacht,wepoolallobservationsandclusterstandard errorsontheindividuallevel.Theparameters(ˇ1,ˇ2,ˇ3,ˇ4)give theeffectofahealthshockin2000.Theparametersofmaininter- estare(ı1,ı2,ı3,ı4)whichgivethedifferencein“healthshock– nohealthshock”outcomesbetween2005and2000.IftheGate- keeperreform(in2002)andtheprolongationofthesickpayperiod (in2004)havebeeneffectiveinreducingtheextenttowhichill healthreducesemploymentopportunitiesandaperson’searnings

capacity,weexpecttheseparameterstohaveapositivesignfor employmentandlaborincome.

Essentially,therearetwoassumptionsthatunderlieourempir- icalstrategy.Aswefollowadifference-in-differencestrategy,the ﬁrstone isthat individuals inthetreatmentgroupand control grouphavecommontimetrendsintheabsenceofahealthshock.

Toaddressthisissue,García-Gómezetal.(2013)analyzediffer- encesinpre-treatmenttrendsofhospitalizedindividualsandtheir matchedcontrolsonsimilardataasours.Theyfindnodifferences inpretreatmentincometrendsofhospitalizedindividualsandtheir matchedcontrols.Moreover,theyshowthattheirbaselineresults donotchangeiftheyusepropensityscorematchingfortreatment andcontrolgroups. Oursecondassumptionconcernsthepossi- bilityofinteractioneffectsbetweenhealthshockeffectsandthe businesscycle.Inparticular,weassumethatannualdifferencesin themagnitudeofshockeffectscannotbeattributedtobusiness cycleeffects.Torelaxthisassumption,wewillthereforealsoesti- mateamodelthatincludestheunemploymentrateattheprovince level(provincedummiesarenolongerincluded).Specifically,we willestimatethefollowingmodeltotestfortherobustnessofour findings:

Y_it=˛t+ˇtS_i+tT_i+ıt(S_i·T_i)+_stX_i+tU_i+ϕt(S_i·U_i)+ε_it, t=1,2,3,4,

(2)

whereU_iistheunemploymentrateintheprovinceinwhichindi- vidualiislivingintheyearofthehealthshock.

Westatedearlierthattheinitialfocusinourempiricalanalysis isoncomparisonsofworkercohortsthatareobservedin2000and in2005andarefollowedinsubsequentyears.Assuch,ouraimis toidentifytheoverallimpactofpolicychangesinsickpayandDI between2000and2005.

3.3. Results

Tables 2–5 report the estimation results of our model for younger men, older men, younger women and older women, respectively.Foreachdemographicgroup,weassesstheeffectof

(6)

Table2

Regressionestimatesformen25–39.

DI UI SA Employed Earnings

(xD10,000) Effecthealthshockin2000

1yearlater 0.046*** 0.014 0.004 −0.036** −0.046

(0.007) (0.008) (0.004) (0.012) (0.078)

2yearslater 0.049*** 0.024 0.004 −0.052*** −0.131

(0.009) (0.009) (0.004) (0.014) (0.071)

3yearslater 0.039*** 0.017 0.005 −0.059*** −0.164

(0.010) (0.009) (0.004) (0.015) (0.085)

4yearslater 0.036*** 0.025 −0.001 −0.050** −0.160

(0.010) (0.010) (0.005) (0.016) (0.082)

Effecthealthshockin2005- effecthealthshockin2000

1yearlater −0.046*** −0.005 0.001 0.010 0.016

(0.003) (0.004) (0.002) (0.005) (0.029)

2yearslater −0.031*** 0.001 −0.003 0.012 −0.017

(0.004) (0.004) (0.002) (0.006) (0.032)

3yearslater −0.024*** −0.003 −0.002 0.021** 0.012

(0.004) (0.004) (0.002) (0.007) (0.041)

4yearslater −0.023*** −0.004 0.001 0.017 −0.014

(0.004) (0.005) (0.002) (0.007) (0.037)

Notes:Standarderrorsarereportedinparentheses.*,**,and***denotesigniﬁcanceatthe10,5,and1%level,respectively,andarebasedonHolmadjustedp-valuesfor multipletesting.ThedependentvariableisanindicatorforDIreceipt.Allregressionsincludedummiesforﬁveyearagegroups,nationality,householdsize,municipality size,province,andlaborincomefromtheyearpriortothepossiblehealthshock(indeciles),aswellasinteractionsofthesevariablewithanindicatorforanunscheduled andacutehospitalization.

Table3

Regressionestimatesformen40–58.

1yearlater 0.064*** 0.015** 0.000 −0.039*** −0.170

(0.005) (0.005) (0.002) (0.008) (0.073)

2yearslater 0.079*** 0.019** 0.004 −0.070*** −0.034

(0.008) (0.006) (0.002) (0.010) (0.107)

3yearslater 0.086*** 0.022** 0.004 −0.089*** −0.121

(0.008) (0.006) (0.002) (0.011) (0.082)

4yearslater 0.083*** 0.026 0.003 −0.101*** −0.201

(0.008) (0.007) (0.002) (0.012) (0.084)

1yearlater −0.064*** −0.008*** 0.001 0.010*** −0.048

(0.002) (0.003) (0.002) (0.005) (0.033)

2yearslater −0.036*** −0.005 −0.003 0.012*** −0.107

(0.003) (0.003) (0.002) (0.006) (0.038)

3yearslater −0.042*** −0.003 −0.002 0.021*** −0.037

(0.004) (0.003) (0.002) (0.007) (0.037)

4yearslater −0.044*** −0.002 0.001 0.017*** −0.022

(0.004) (0.003) (0.002) (0.007) (0.038)

healthshocksontheprobabilityofDIbenefitreceipt,Unemploy- mentInsurance(UI)benefitreceipt,SocialAssistance(SA)benefit receipt,employment andemploymentearnings.Atthis point,it shouldbestressedthattheP-valuesthatarereportedforthehealth shockdummies–denotedas(ˇ1,ˇ2,ˇ3,ˇ4)fortheeffectofahealth shockin2000(ı1,ı2,ı3,ı4)forthedifferenceinhealthshockout- comesbetween2005and2000—areadjustedformultipletesting.

Inparticular,weusetheHolmcorrectionmethodontheseeight dummiesforall(ﬁve)outcomemeasuresofinterest.

Table2showstheestimationresultsformenaged25–39.For thisgrouptheprobabilitytoreceiveDIbeneﬁtsis 4percentage pointshigherintheyearafterahealthshock,itincreasesto4.9 percentagepointsinthenextyear,andthendeclinesto3.5–4per- centagepointsinthefollowingtwoyears.Thiseffectismirrored byareductionintheemploymentprobabilityofasimilarmagni-

tude.Also,thereductioninearningslossesafterahealthshockis 3500eurosinitiallyand4400eurosafterfouryears.Theseoutcomes arecomparabletothoseobtainedbyGarcía-Gómezetal.(2013).

Finally,thereisnostatisticallysignificantevidencethatyounger mensubstitutebetweenDIbenefitsandUIorSAbenefitsaftera healthshock(seee.g.Borghansetal.,2014).

Movingtothelowerpanelofthesametable,thenextsetofesti- matesshowthedifferenceoftheeffectofahealthshockbetween 2005and2000(theıcoefficientsofmodel(1)).Thefactthatthe estimatesfortheprobabilityofDIbenefitsinthefirstyearafterthe shockarelargerthanfortheremainingperiodistobeexpected becausesince2004anillpersonneedstowaittwoyearsbefore becomingeligibletoreceiveDIbenefits.Twoyearsafterahealth shock,a managed25–39is 3.1percentage pointslesslikely to receiveDIbenefitsandthiseffectdeclinesslightlyinthenexttwo

(7)

Table4

Regressionestimatesforwomen25–39.

1yearlater 0.030*** 0.020 −0.010 −0.024 −0.076

(0.008) (0.009) (0.006) (0.015) (0.037)

2yearslater 0.045*** 0.024 −0.005 −0.014 −0.043

(0.011) (0.010) (0.006) (0.018) (0.045)

3yearslater 0.043*** 0.014 −0.004 −0.028 −0.095

(0.012) (0.010) (0.005) (0.020) (0.051)

4yearslater 0.051*** 0.019 −0.001 −0.031 −0.090

(0.013) (0.012) (0.006) (0.021) (0.056)

1yearlater −0.052*** −0.015** 0.001 0.009 0.009

(0.003) (0.005) (0.003) (0.007) (0.016)

2yearslater −0.043*** −0.008 −0.001 0.024 −0.008

(0.005) (0.005) (0.003) (0.008) (0.019)

3yearslater −0.036*** −0.000 0.003 0.016 −0.028

(0.005) (0.005) (0.003) (0.009) (0.022)

4yearslater −0.031*** −0.006 −0.001 0.018 −0.040

(0.006) (0.005) (0.003) (0.009) (0.024)

Table5

Regressionestimatesforwomen40–58.

1yearlater 0.060*** 0.010 0.003 −0.032 −0.129**

(0.008) (0.007) (0.004) (0.012) (0.037)

2yearslater 0.085*** 0.002 0.005 −0.060*** −0.145***

(0.013) (0.009) (0.004) (0.015) (0.040)

3yearslater 0.096*** −0.001 0.008 −0.076** −0.129*

(0.014) (0.009) (0.004) (0.017) (0.044)

4yearslater 0.093*** −0.001 −0.002 −0.079*** −0.016

(0.014) (0.010) (0.004) (0.017) (0.049)

1yearlater −0.078*** −0.018*** −0.000 0.008 0.008

(0.004) (0.003) (0.002) (0.005) (0.015)

2yearslater −0.071*** −0.007 −0.001 0.012 0.008

(0.005) (0.004) (0.002) (0.006) (0.018)

3yearslater −0.074*** −0.002 −0.001 0.006 0.015

(0.006) (0.004) (0.002) (0.007) (0.019)

4yearslater −0.066*** 0.000 0.001 0.002 −0.040

(0.006) (0.004) (0.002) (0.007) (0.022)

Notes:Standarderrorsarereportedinparentheses.*,**,and***denotesigniﬁcanceatthe10,5,and1%level,respectively,basedonHolmadjustedp-valuesformultiple testing.ThedependentvariableisanindicatorforDIreceipt.Allregressionsincludedummiesforﬁveyearagegroups,nationality,householdsize,municipalitysize, province,andlaborincomefromtheyearpriortothepossiblehealthshock(indeciles),aswellasinteractionsofthesevariablewithanindicatorforanunscheduledand acutehospitalization.

years.Thischangegoestogetherwithanincreaseintheemploy- mentprobabilityofyoungermen,withaneffectestimateofabout 2percentagepoints.Inpart,theseeffectsreflectthefactthatafter 2004theemployerisresponsibleforthefirsttwoyearsofsickpay, togetherwiththefactthatinourdataworkerswhoreceivesick- nessbenefitsareclassifiedasbeingemployed.Inlightofthis,it isimportanttonotethatourresultsshowthattheimprovement intherelativeemploymentprobabilitypersistsandevenincreases aftertwoyears.ThissuggeststhattheDIreformshavenotbeenat thecostsofthewell-beingofworkerswithhealthproblems,and thushavebeensuccessfulintargetingtheprogramtoyoungermen havingsubstantialhealthproblems.

Table3showstheestimationresultsformenaged40–58.Ifthis groupofworkersfacesasuddendeteriorationofhealth,initiallythe probabilityofDIreceiptincreasesby6.4percentagepoints,which

increasesto7.9percentagepointsinthesecondyear,andincreases furthertoaround8.5percentagepointsinthethirdandfourthyear.

Oldermenarealsoabout10percentagepointslesslikelytowork afterahealthshockinthefourthyearaftertheshock.Thisfinding issimilartoTrevisanandZantomio(2016)who,basedonELSAand SHAREdata,findthatafirstacutehealth shockresultsina 10- percentagepointreductioninlabormarketparticipationamong olderEuropeanmen.Ourfindingsalsosuggestthathealthshocks decreasetheamountofearningsandincreasethelikelihoodofUI benefitreceiptforoldermen.

Clearly,theeffectsonDIbeneﬁtsandemploymenteffectswe ﬁndare markedly higher for oldermenthan for theiryounger counterparts.Oneexplanationmaybethatyoungerpeoplehave strongerincentivestoreturntothelaborforcebecausetheyhave feweroptionstoreplacelostincome.Inaddition,youngerworkers

(8)

aresigniﬁcantlylesslikelytosufferfromacirculatorydiseaseor togetcancer,illnessesthatmayleadtoalonger(orpermanent) withdrawalfromthelaborforce.Thus,thenatureofthehealth shockmayalsoexplainwhytheemploymenteffectsaresmaller foryoungermenandwomen.

WenextturntothelowerpanelofTable3,whichshowshigher reductionsinthehealthshockeffectsonDIbeneﬁtreceiptthan for younger men. In particular, four years afterthe onset of a healthshocktheeffectis4.4percentagepointssmaller.Atthesame time,theemploymenteffectofhealthshocksisabout2percentage points,whichiscomparabletotheemploymenteffectforyounger men.ThisindicatesthatabouthalfofthedecreaseinDIbeneﬁt receiptamongoldermenconsistedofincreasesinworkresumption orworkcontinuation.

Finally,Tables4and5showthattheprobabilitytoreceiveDI benefitsafterahealthshockisinitiallysmallerforwomeninboth agegroupscomparedtotheirmalecounterparts.However,inthe thirdorfourthyearafterthehealthshock,theprobabilitytoreceive DIbenefitsishigheramongwomen. Theemployment effectsof healthshockstendtobesmallerforyoungwomenthan young men,while earningsresponses areinsignificant.As for thedif- ferenceinhealthshocksbetweenthecohortsof2000and2005, thereductionintheeffectsonDIreceiptare3.1and6.6percent- agepointsforyoungerandolderwomen,respectively.Theresults furtherindicateevidenceofemploymenteffectsamongyounger women,withsizeeffectsbetween1and2.5percentagepoints.For olderwomen,thedecreaseinDIreceiptisnotmirroredbyany increasesinemployment.

TheresultssofarindicatethataftertheDIreformsithasbecome less likely to enter the disability program for all demographic groups.Furthermore,individualsdidnotsubstitutetotheUIorSA programbecauseofthereforms.Eventhoughcoefﬁcientestimates arenot alwaysstatisticallysigniﬁcant ifwe adjust formultiple testing,wealsoconcludethatemployment increaseswerepro- portionaltothedecreasesinDIreceipt,exceptforolderwomen.

Asearningsdidnotincrease,theemploymentincreasesareproba-

blyassociatedwithincreasesattheextensivemargin,ratherthan hours increases for those workers remaining employed after a healthshock

3.3.1. Doeffectsdifferbytypeofhealthshock?

Next,wefurtherexaminetheroleofthenatureofthehealth shock.Inhisanalysisoftheeffectsofchangesinhealthstatuson healthinsurancecoverageandlabormarketoutcomes,McClellan (1998)makesaninterestingdistinctionbetween(i)majorhealth events,(ii)chronicillnesses,and(iii)accidents.Majorhealthevents –suchascancer,heartattack,orstroke–haveasubstantialimme- diateeffectand implylong-termfunctional limitations.Chronic illnesses–suchasdiabetes,lungdisease,arthritis,orheartfailure– generallyonlymoderatelylimitcurrentfunctioning,butmayresult in more severeimpairments due toprogression ofthe disease.

Finally,accidentshavesubstantialimmediateeffectsonfunction- ing,butarelesslikelytoresultinseverelong-termimpairments.

Table1 showsthat amongtheyoungmenand womenwho experienceahealthshock,around5percenthasamajorhealth shock.25percentisduetoachronicillness,andtheremaining 70percentareduetoaccidents.Amongmenaged40–58whoare hospitalized,25percentisduetoamajorhealthevent,28percent duetoachronicillness,and47percentduetoanaccident.Among womenaged40–58thedistributionissigniﬁcantlydifferent,with only13percentofthehospitalizationsduetoamajorhealthevent, 30percentduetoachronicillness,and57percentduetoaccidents.

Majorhealtheventsarethusconsiderablymoreimportantamong menaged25–39comparedtowomeninthatagecategory.Major healtheventsarealsomuchmoreimportantamongoldermenand womenrelativetotheiryoungercounterparts.Thedistributionis verysimilarforbothyears.

Tables6and7reporttheeffectofanunscheduledandurgent hospitalizationin2000aswellasthedifferentialeffectofthisin 2005measuredfouryearsafterwards–bothontheprobabilityto receiveDIbeneﬁtsandtheprobabilitytobeemployed,respectively.

Tostartwith,theeffectofamajorhealthshockontheprobability

Table6

RegressionestimatesfortheprobabilitytoreceiveDisabilityInsurancebeneﬁtsfouryearsafterahealthshock.

Men Women

Age25–39 Age40–58 Age25–39 Age40–58

Effectofmajorhealthshockin2000

0.109*** 0.154*** 0.165*** 0.182***

(0.023) (0.015) (0.041) (0.027)

Effectofmajorhealthshockin2005- effectofmajorhealthshockin2000

−0.044* −0.063*** −0.016 −0.098***

(0.025) (0.009) (0.047) (0.019)

Effectofonsetchronicillnessin2000

0.070*** 0.154*** 0.154*** 0.193***

(0.018) (0.015) (0.025) (0.024)

Effectofonsetchronicillnessin2005- effectofonsetchronicillnessin2000

−0.035*** −0.049*** −0.066*** −0.086***

(0.009) (0.008) (0.013) (0.013)

Effectofaccidentin2000

0.038** 0.083*** 0.065*** 0.077***

(0.017) (0.014) (0.024) (0.023)

Effectofaccidentin2005- effectofaccidentin2000

−0.016*** −0.028*** −0.018*** −0.046***

(0.005) (0.005) (0.006) (0.007)

Notes:Standarderrorsarereportedinparentheses.*,**,and***denotesigniﬁcanceatthe10,5,and1%level,respectively.Thedependentvariableisanindicatorfor employment.Allregressionsincludedummiesforﬁveyearagegroups,nationality,householdsize,municipalitysize,province,andlaborincomefromtheyearpriortothe possiblehealthshock(indeciles),aswellasinteractionsofthesevariablewithanindicatorforanunscheduledandacutehospitalization.