The Effects of Health Shocks on Labour Market Exits ...businesslaw.curtin.edu.au/wp-content/uploads/sites/5/2016/05/AJLE … · (HILDA) Survey. Our results show that health shocks

191AUSTRALIAN JOURNAL OF LABOUR ECONOMICS

Volume 13 • Number 2 • 2010 • pp 191 - 218

The Effects of Health Shocks on Labour Market Exits: Evidence from the HILDA Survey

Eugenio Zucchelli, Andrew M. Jones and Nigel Rice,TheUniversityofYorkAnthony Harris,MonashUniversity

Abstract This paper analyses the relationship between ill-health, health shocks and early labour market exits among older working individuals. We represent the transition to non-employment as a discrete-time hazard model using a stock-sample from the first six waves (2001-2006) of the Household, Income and Labour Dynamics in Australia (HILDA) Survey. Our results show that health shocks are key determinants of early exit choices. For men, negative shocks to health increase the hazard of becoming non-employed by 50 to 320 per cent, whereas for women, health shocks increase the hazard of an early exit from the labour market by 68 to 74 per cent. These findings are confirmed by both a measure of health limitations and a measure of latent health obtained using pooled ordered probit models as well as for two alternative definitions of health shocks.

JELClassification:I10,C10,C41,J14

1. Introduction Most developed countries are currently experiencing trends of declining labourforce participation, especially among working-age men, combined with an ageingpopulation (Auer andFortuny, 2000). InAustralia, despite recent rises inwomen’sparticipationrates,theoverallparticipationrateforpeopleaged15oroverisprojectedtodecreasefrom64.5to58.7percentbetween2007and2047(AustralianDepartment

Addressforcorrespondence:EugenioZucchelli,CentreforHealthEconomics,Alcuin‘A’Block,TheUniversityofYork,Heslington,York,YO105DD,UnitedKingdom.Email:[email protected]/Disclaimer:ThispaperusesunitrecorddatafromtheHousehold,IncomeandLabourDynamicsinAustralia(HILDA)Survey.TheHILDAProjectwasinitiatedandisfundedby the Australian Government Department of Families, Community Services and IndigenousAffairs (FaCSIA)and ismanagedby theMelbourne InstituteofAppliedEconomicandSocialResearch (MIAESR).Thefindings andviews reported in thispaper, however, are thoseof theauthorandshouldnotbeattributedtoeitherFaHCSIAortheMIAESR.Wewould like to thankBruceHollingsworthand theCentre forHealthEconomicsatMonashUniversity,Melbourne,fortheirsupportandcontributions.WealsowouldlikethankthemembersoftheHealthEconometricsandDataGroup(HEDG)attheUniversityofYork,UK,theeditorBoydHunter,twoanonymousrefereesandparticipantsattheSixthWorldCongressoftheInternationalHealthEconomicsAssociation(IHEA)fortheirusefulcomments.©TheCentreforLabourMarketResearch,2010

192AUSTRALIAN JOURNAL OF LABOUR ECONOMICSVOLUME 13 • NUMBER 2 • 2010

ofTreasury,2007).Thisismainlytheresultoftherapidincreaseintheproportionofindividualsaged65yearsandover.TheTreasury’spopulationprojectionsfurthershowthatwithinthenext40yearstheproportionofolderindividuals(64to84yearsold)ispredictedtomorethandoubleandthenumberoftheveryold(85andover)isexpectedtoquadruple.Asaresult,whiletherearecurrently5individualsofworking-ageforeverypersonaged65andover,by2050thisnumberisprojectedtoshrinkto2.7(AustralianDepartmentofTreasury,2010).Earlyretirementandpopulationageingpose a threat and a challenge to the sustainability of the social security systemofanyindustrialisedeconomy.Inthiscontext,understandingthedrivingforcesbehinddecisionstoexitthelabourmarketwillhelptoinformpoliciestoincentiviseworkersto remain in active employment and encourage younger retirees to return into thelabourmarket.

Thereareseveralfactorsthatcouldpotentiallyinfluenceretirementchoicesofolderworkingindividuals.Togetherwithinstitutionalfactors,suchasthegenerosityof the social security system, the introduction of early retirement options and thepresenceofdisabilitybenefitschemes(Kerkhofset al.,1999;Blundellet al.,2002),individualhealthstatusplaysamajorroleinretirementdecisions.Adeclineinhealthstatus,ceterisparibus,mayreducetheprobabilityofcontinuedworkforthreereasons(Disneyet al.,2006),poorhealthmay:raisethedisutilityofwork;reducethereturnsfromworkvialowerwagesand,byentitlingindividualstonon-wageincomethroughdisabilitybenefits,actasanincentivetoexitthelabourmarket.

While there isabundantevidenceontheimportanceoffinancial incentivesin determining retirement behaviour (Lumsdaine and Mitchell, 1999; Blundell et al., 2002; French, 2005), empirical evidence on the role of health on retirement isstill limited,especially forAustralia.Further,problemssuchasmeasurementerror(reporting bias) and the potential endogeneity of self-assessed measures of healthtogetherwiththepresenceofunobservableheterogeneityhavehamperedattemptstoreachdefiniteconclusionsonthisrelationship.Anotherimportantbutunexploredissueistherelativeroleassumedbygradualhealthdeteriorationversusunexpectedchangesinhealthorhealthshocks.Thisthemeisdirectlyrelatedtotheeconometricproblemoftheidentificationofacausaleffectofhealthonwork.Unexpectedhealthchangesandtheknowledgeoftheirtimingcouldprovidesufficientexogenousvariationtoisolatetheeffectofhealthonanindividual’slabourstatus.

Thispapercontributestotheempiricalliteraturebyassessingandquantifyingthe relative significance of gradual versus suddenhealth deterioration in early exitdecisions.To thebest of our knowledge, this is thefirst attempt to implement thiskindofanalysisusingAustralianlongitudinaldata.Werepresentthetransitiontonon-employmentasadiscrete-timehazardmodelwhichenablesustoestimatetheeffectofdifferentmeasuresofhealthandhealthshocksandanumberofsocio-economiccharacteristicsontheprobabilityofleavingtheworkforce.WeusethestocksamplingapproachofJenkins(1995)todefineoursampleofinterest.Thismethod,changingtheunitofanalysisfromtheindividualtothetimeatriskofanevent(inthiscase,retirement), allows complex sequence likelihoods to be simplified to a standardestimationforabinaryoutcome(Jenkins,1998).Inordertoovercometheproblemsrelated to measurement error (reporting bias) and endogeneity of self-assessed

193EUGENIO ZUCCHELLI, ANDREW M. JONES, NIGEL RICE AND ANTHONY HARRIS

The Effects of Health Shocks on Labour Market Exits : Evidence from the HILDA Survey

measures of health,we construct a latent health stock variablewhich is purged ofreportingbias(Bound,1991;1999).Further,wedefinehealthshocksintwoalternativeways:usinginformationontheincidenceofsuddeninjuryorillnessandlookingatthedifferencesbetweenindividual’shealthstocksovertime.

Our results, using panel data from the first six waves (2001-2006) of theHousehold, IncomeandLabourDynamics inAustralia (HILDA)survey, show thathealthplaysafundamentalroleinindividualemploymenttransitions.Forbothmenandwomen,negativeshockstohealthsignificantlyincreasethehazardofbecomingnon-employed.Apartfromageing,ill-healthandhealthshocksarequantitativelythemost important causesof early exits from the labourmarket among the individualsocioeconomic variables considered. Furthermore, estimated effects on householdtype (marital status) and composition (having own dependent children) are alsosignificantdeterminantsoftransitionstonon-employment.Ourfindingsindicatethatforwomen,livingwithapartnergreatlyenhancestheriskofanearlyexit;formen,havingdependentchildrenisassociatedwithasignificantdecreaseinthehazardofleavingthelabourforce.

2. Background Severalstudiesconcludethatill-healthisoneofthemaincausesofretirementamongolderworkers (Lindeboom,2006a).However, there is still somecontroversy in themeasurementofhealthandinmodellingtherelationshipbetweenhealthandwork.

AndersonandBurkhauser (1985)argue that self-reportedmeasuresarenotreliableandthathealthshouldbetreatedasanendogenousvariable.Takingargumentssuch as this into account,more objectivemeasures believed to be less sensitive tojustification bias or state-dependent reporting bias have been used. These includeobserved future mortality of sample respondents (Parsons, 1980; Anderson andBurkhauser, 1985), sickness absenteeism records (Burkhauser, 1979), and indicesderived from multiple indicators (Lambrinos, 1981; Bazzoli, 1985). Bound (1991)suggeststhatlaboursupplymodelsaresensitivetothemeasuresofhealthused.UsingtheU.K.RetirementHistorySurvey,Boundbuildsamodelforlaboursupply,wagesandhealth and shows that eachof the solutionsproposed in the literature leads toadifferentbias. Inparticularheargues thatwhenself-reportedmeasuresareused,healthappears toplaya larger roleandeconomic factorsa smallerone thanwhenmoreobjectivemeasuresareused.However,moreobjectivemeasures(i.e.functionallimitations) potentially lead to different biases.Objectivemeasures, unlikely to beperfectlycorrelatedwiththeaspectofhealththataffectsanindividual’scapacityforwork,willsufferfromanerrorinvariablesproblem,leadingtodownwardlybiasedestimatesoftheimpactofhealthonretirement.

Empiricalstudiesontherelationshipbetweenhealthandretirementproduceverydifferentconclusions.SticklesandTaubman(1986)andStern (1989)concludethathealthplaysamajorrolebothontheretirementdecisionandlaboursupply.Stern(1989) finds that subjective health measures have strong and independent effectson labour supply.Kerkhofs et al. (1999) estimate a retirementmodelwith a rangeofdifferenthealthconstructsandfindthat thechoiceofhealthmeasureaffects theestimateofhealthon labour supplyoutcomes.Dwyer andMitchell (1999) confirm


theseresults.Theyspecifyaretirementmodelwheretruehealthisinstrumentedwitharangeofmoreobjectiveindicators.Theirresultsshowthathealthhasastrongeffectonretirementbutthatthesizeoftheeffectvarieswiththemeasureused.Theyalsofindthatself-ratedhealthmeasuresareexogenousandthereisnoevidenceinsupportofjustificationbias.BlauandGilleskie(2001)suggestthathealth-retirementmodelsshouldavoidtheuseofasinglemeasureofhealthandthathealthshouldbetreatedasendogenous.

Morerecently,theliteraturerecognisestheimportanceofassessingtherelativesignificanceofpermanentortemporaryhealthshocksversusagradualdeteriorationofhealthinretirementdecisions.Boundet al.(1999)specifyamodelfortransitionsbetweenworkstatesandadynamicmodelforhealth,usingthreewavesoftheU.S.HealthandRetirementStudy.Inordertocorrectfortheendogeneityofself-assessedhealththeybuildalatentvariablemodelthatrelatesself-reportedmeasuresofhealthtoaseriesofphysicallimitationmeasures.Theyfindthatbothchangesinhealthandthelong-termlevelofhealthareimportantforlaboursupplydecisions.InGermany,Riphahn(1999)findsthathealthshocks,definedasasuddendropinaself-reportedmeasureofhealthsatisfaction,havesignificanteffectsonemployment,increasingtheprobabilityofleavingthelabourforce.Disneyet al.(2006)applythemethodofBoundet al.(1999)tothefirsteightwavesoftheBritishHouseholdPanelSurvey(BHPS),1991to1998.TheyfindthathealthshocksareanimportantdeterminantofretirementbehaviourintheUK.TheseresultsareconfirmedbyRobertset al.(2008),Joneset al. (2009) andGarciaGomez et al. (2010) on theBritishHousehold Panel Survey(BHPS)andbyandHaganet al.(2009)ontheEuropeanCommunityHouseholdPanel(ECHP)data.Lindeboomet al.(2006b)focusontherelationshipbetweentheonsetofdisabilityandemploymentoutcomes.Theresultsshowthathealthshocksincreasethe likelihood of an onset of disability by 138 per cent. However, health shocksare relatively rareeventsand therefore theyconclude that themajorityofobserveddisabilityratesresultfromgradualhealthdeterioration.

Research on the effects of health on labour supply of older workers inAustraliaisgrowingbutstilllimitedifcomparedtotheevidenceavailableforothercountries(especiallyUKandUS).Brazenor(2002)andWilkins(2004)usethe1998ABScross-sectionSurveyonDisability,AgeingandCarers(SDAC)toexaminetheimpactofdisabilityonearningsandemploymentstatusrespectively.Brazenorshowsthatdifferenttypesofdisabilityhaveanegativeimpactonearnings.Wilkinsfindsthatonaveragedisabilitydecreasestheprobabilityoflabourforceparticipationbyone-quarterformalesandone-fifthforfemales.CaiandKalb(2006)analysetherelationshipbetween health and labour market participation using the HILDA Survey. Theyestimateasimultaneousequationmodelforworking-ageindividualstocontrolforthepotentialendogeneityofhealth.Theirestimatesconfirmthathealthhasasignificanteffecton labour supply.Further,Laplagneet al. (2007)usedata fromHILDAandfind that both better health and education are associatedwith greater labour forceparticipation.WarrenandOguzoglu(2007)andCaiet al.(2008)alsoanalysedifferentaspectsofhealthandlaboursupplyusingtheHILDASurvey.Correspondingly,theyfind that differences in severity levels of disability explain a significant proportionofthevarianceintheparticipationratesamongdisabledindividualsandthatlower



healthstatusandhealthshocksleadtoreductionsinworkinghours.Finally,Schofieldet al.(2008)usedatafrom2003ABSSDACSurveyandfindthatamongindividualsaged45-64yearsaseriesofchronicconditionssuchasbackproblemsandarthritisarestronglyassociatedwithnon-participation.

3. Econometric Framework Duration Model for Employment Exits Oureconometricspecificationisbasedonthedurationmodelstock-samplingapproachofJenkins(1995).Followingthismethod,wecreateoursampleofinterestbyselectingonlyworkingindividualsatriskofleavingthelabourforce(agedbetween50yearsoldandtheyearpriorstateretirementage:64formenand61forwomen)inthefirstwaveof theHILDASurveyandwe follow them through the subsequent sixwavesuntil they are observed leaving thework force or are censored.Transition to non-employment is represented using a discrete-time hazardmodel. This enables us toestimatetheeffectoftwodifferentmeasuresofhealthstatus(ahealthstockmeasureandameasureofhealthlimitations)andanumberofsocio-economiccharacteristics(age,gender,education,jobstatus,maritalstatus,etc.)ontheprobabilityofleavingthelabourmarket.

Thismethod,controllingforstock-samplingandchangingtheunitofanalysisfrom the individual to the time at risk of an event (labour market exit), allows acomplexsequencelikelihoodtobesimplifiedtothemorestandardestimationforabinaryoutcome.1Weinitiallyselectonlythoseindividualswhoareworkinginwave1.Subsequently,theseindividualscanstayinthelabourforce,leavethelabourforce,or be lost to follow-up. Non-employment is considered an absorbing (permanent)state:transitionsbackinthelabourmarketarenotconsidered.UsingJenkins’(1995)notation, t=t representsthefirstobservationonthestocksample, t=1is thefirstperiodatwhichanindividualisatriskofnon-employment(age50).Attheendofthetimeperiodsomepeoplewillstillbeworking(censoreddurationdata,di=0),andsomewillhaveleftthelabourmarket(completedurationdata,di=1).Ifindividualsarelosttofollow-upbeforeleavingthelabourforcethesearealsoconsideredcensoredobservations.t=t+siistheyearwhennon-employmentoccursifsi=1andthefinalyearofourdataperiodifdi=0.Eachrespondenti,contributessi yearsofemploymentspells.Theprobabilityofleavingtheworkforceateachtprovidesinformationonthedurationdistributionandthediscrete-timehazardrateis:

(1)

whereXitisavectorofcovariateswhichmayvarywithtimeandTiisadiscreterandomvariablerepresentingthetimeatwhichlabourmarketexitisobserved.Theconditionalprobability(conditionalonnothavingleftthelabourforceatthebeginningofthetimespell)ofobservingtheeventhistoryofsomeonewithanincompletespellatinterviewis:

(2)

1FortheestimationinSTATA,seeJenkins(1998).


Theconditionalprobabilityofobservingtheeventhistoryofsomeonecompletingaspellbetweentheinitialobservation,t,andinterviewis:

(3)

Thecorresponding log-likelihoodofobserving theeventhistorydatafor thewholesampleis:

(4)

Jenkins (1995) suggests simplifying the log-likelihood by defining an indicatorvariableyit.Forthosestillworking,yit=0forallperiods;forthosewhobecomenon-employed,yit=0,forallperiodsexcepttheexitperiodwhenyit=1.Formally:

yit=1ift=t+sianddi=1,yit=0otherwise.

Using this indicator variable, the log-likelihood function can be re-expressed in asequentialbinaryresponseform:

(5)

In this way, the log-likelihood function has the same form as the ‘standard’ log-likelihoodfunctionforabinaryvariable,wheretheunitofanalysisisnowthespellperiod.2 Following Jones et al. (2009) and Hagan et al. (2009), we complete thespecificationusingacomplementarylog-loghazardfunctionforthehazardhit:

(6)

whereq(t)isthebaselinehazardmodelledasastepfunctionbyusingdummyvariablestorepresenteachyearofageatrisk.3

Health Stock and Health Shocks Health Stock Measure Therearethreemainproblemsrelatedtotheuseofself-assessedmeasuresofhealthwhenattemptingtoestimateacausaleffectofhealthonwork(AndersonandBurkhauser,1985;Bazzoli,1985;Stern,1989;Bound,1991;Boundet al.,1999;Auet al.,2005;2SeeJenkins(1995)forfurtherdetails.3Disneyet al. (2006) include initialage togetherwitha setof timedummies for timeelapsedsincethestartofthepanelintheirspecificationofthismodel.However,webelievethisisnotanappropriatemeasureofdurationdependencewhenageof labourmarketexit is theoutcomeofinterestandindividualsenterthestocksampleatdifferentages.WethusfollowJoneset al.(2009)andincludeinourspecificationasetofagedummiestorepresenttheageatriskofexitingthelabourforce.Webelievethisisamoreappropriatespecificationasitallowstheimpactofsurvivingtobedifferentforindividualsatdifferentages.Further,thisappearstobemoreconsistentwiththeoriginalformulationofthediscrete-timehazardmodelasdescribed,forexample,inJenkins(1995,1998).



Disneyet al.,2006;andBrownet al.,2010).First,self-assessedvariablesmightbeaffectedbymeasurementerrorcausedbyreportingheterogeneity:individualswiththesameunderlyinglevelofhealthmayapplydifferentthresholdswhenreportingtheirhealthstatusonacategoricalscale (LindeboomandvanDoorslaer,2004).Second,since healthmay affect productivity directly, theremight be genuine simultaneitybetween labour market and health status. Third, individuals may systematicallyoverstatetheirhealthstatustojustifybeingoutsidethelabourmarketorasameanstoobtainsocialsecuritybenefits(KerkhofsandLindeboom,1995).

In order to overcome the problems associated with measurement error ofself-assessedmeasuresofindividualhealth,wecreatealatenthealthstockvariable.FollowingtheprinciplesoutlinedbyStern(1989)andBound(1991)andsubsequentlyappliedinanumberofstudies,weestimateamodelofSAHasafunctionofmoredetailedmeasuresofphysicalhealth(self-reportedmeasuresoflimitationsinphysicalfunctioning,role-physicallimitationsandbodilypaininperformingworkandotheractivities)todefinealatenthealthstock.Wethenusethepredictedvaluesforthelatenthealthstockasourhealthvariableinthedurationmodelofemploymentexits.

The intuition behind this procedure is to use specific health measures toinstrument the endogenous and potentially error-ridden general measure of self-assessedhealth.Weconsidertheaspectofhealththataffectsanindividual’sdecisiontoretire,hit

R,tobeafunctionofasetofmorespecificmeasuresofhealth,zit:

hitR =zit b +eit, i=1,2,...,n;t=1,2,...Ti(7)

whereeitisatimevaryingerrortermuncorrelatedwithzit.Wedonotdirectlyobservehit

R butinsteadameasureofSAH, hitS.Wespecify

thelatentcounterparttohitSashit

*inthefollowingway:

hit*=hit

R +hit i=1,2,...,n;t=1,2,...Ti(8)

In(8),hitrepresentsthemeasurementerrorinthemappingofhit*tohit

R.Weassumehitisuncorrelatedwithhit

R.Substituting(7)into(8)gives:

hit*=zit b +eit +hit =zit b +nit i=1,2,...,n;t=1,2,...Ti(9)

In our model for retirement we use the predicted health stock, h^it*, purged of

measurementerror,toavoidthebiasesassociatedwithusinghit*directly.Assumingnit

isnormallydistributed,model(9)canbeestimatedasapooledorderedprobitmodelusingmaximumlikelihood.

Health Shocks It is important to establish whether transition to non-employment originates froma slowdeteriorationor froma shock (acutedeterioration) toan individual’shealth.Further, identifying health shocks offers a convenientway to eliminate a potentialsource of endogeneity bias caused by the correlation between individual-specificunobservedcharacteristicsandhealth(Disneyet al.,2006).

Wespecifyamodelforboththehealthstockvariableandameasureofhealthlimitations (arguablymoreobjective than thegeneral self-assessedhealthmeasure)


toaccountforthegradualdeteriorationinindividual’shealth.Aswespecifyhealthshocksasthelagofahealthstockvariableconditionaloninitialperiodhealth,ashockis identified throughdeviations inhealthstatusover timeandhenceeliminates theindividualeffect.Inaddition,webuildanalternativemeasureofhealthshocksbasedonself-reported informationcontained in thesurvey.Thismeasure isbasedon theresponsesfromaquestionontheoccurrenceofa‘seriousinjuryorillness’duringthetwelvemonthspriortotheinterview.Accordingly,wecreateadummyvariablewhichtakesthevalue1iftherespondentreportsaseriousinjuryorillnessintheprevioustwelvemonthsandthevalue0otherwise.4Wealsousethisvariableinthedurationmodeltogetherwiththetwogeneralhealthmeasures.

4. Data The HILDA Survey Data Wemakeuseofthefirstsixwaves(2001-2006)ofTheHousehold,IncomeandLabourDynamics inAustralia (HILDA)Survey.HILDA is ahousehold-basedpanel studywhichcollectsinformationabouteconomicandsubjectivewell-being,labourmarketdynamicsandfamilydynamics.Thedatasetcontainsabroadrangeofvariablesrelatedto individualcharacteristicsandisparticularlyinformativeoncurrentandpreviouslabourmarketactivitiesaswellasonmeasuresofindividualhealthstatus.

The first wave consists of 7682 households and 19914 individuals. Thehouseholdswere selected using amulti-stage approach (Goode andWatson, 2006).Individualinterviewswereconductedwithindividualsaged15yearsandover,butsomelimitedinformationisalsoavailableforpersonsunder15yearsold.Individualsarefollowedovertimeandthefirstwave’ssampleisautomaticallyextendedbyaddinganychildrenborntooradoptedbymembersoftheselectedhouseholdsandnewhouseholdmembersresultingfromchangesinthecompositionoftheoriginalhouseholds.

Attritionratesforthefirstthreewaves(13.2percent,9.6percentand8.4percentrespectively)areslightlyhigherthantheonesforcomparablesurveyssuchastheBritishPanelHouseholdStudy (BHPS).5According toWatsonandWooden (2004)attritionbetweenthefirstandsecondwaveisnonrandomandthere-interviewrateis lowerforpeople living inSydneyandMelbourne;aged15to24years;singleorlivinginadefactomarriage;borninanon-English-speakingcountry;AboriginalorTorresStraitIslander;livinginaflat,unitorapartment;withrelativelylowlevelsofeducation;unemployedorworkinginblue-collarorlow-skilledoccupations.WatsonandWoodenalsoconcludethatthebiasimpartedbytheselectivenessofattritionisunlikelytohavesignificantconsequences.Howeveraseriesofweightswereintroducedtocorrectforpanelattrition(GoodeandWatson,2006).

Variables Tables1and2describethevariablesusedinourmodelofemploymentexitsandthevariousmeasures of physical limitations and bodily pain used to build the healthstockmeasure.

4Thequestionon‘seriouspersonalinjuryorillness’wasaskedonlytotherespondentsfromwave2towave6,i.e.answerstothisquestionarenotavailableforwave1.5 Although Goode and Watson (2006) believe that the rates compare favourably given thecomparative waves of the BHPS were conducted 10 years earlier and it has been generallyacceptedthatresponseratestosurveyshavebeenfalling.



Table 1 - Variables Used in the Model for Labour Market Exits – Description

Variables DescriptionDependent variableLabourmarketstatus 1ifrespondentiseconomicallyinactive,0otherwiseIll-healthHealthlimitations Self-assessedhealthlimitations,1ifhealthlimitsdailyactivities, 0otherwiseSelf-AssessedHealth(SAH) Self-assessedhealth:1:excellent,2:verygood,3:good,4:fair, 5:poorHealth shocksSeriousinjuryorillness 1ifsufferedaseriousinjuryorillnessinthepast12months, 0otherwiseHousehold variablesMarital/couple 1ifmarriedorlivingtogetherwithapartner,0otherwiseSingle 1ifsingle,0otherwise(baselinecategory)Owndependentchildren 1ifrespondenthasowndependentchildren,0otherwiseNodependentchildren 1ifrespondentdoesnothaveanydependentchildren, 0otherwise(baselinecategory)Income, wealth and housing tenureLoghouseholdincome IndividualspecificequivalisedmeanlogoftotalhouseholdincomeHouseholdwealth TotalhouseholdnetwealthRentinghome 1ifrentinghome,0otherwiseOwninghome 1ifowninghomewithorwithoutamortgage,0otherwise(baseline category)Age dummiesAgedummiesforeachagecategory 1ifrespondentisaged50or51or52,etc.,0otherwise(50-64formen;50-61forwomen) (withAge50-52asbaselinecategory)EducationEducation/degrees 1ifrespondentholdsdegreeorpostdegreequalifications, 0otherwiseEducation/certificate 1ifadvanceddiplomaorcertificate,0otherwiseEducation12 1ifhighesteducationcompletedisyear12,0otherwise(baseline category)Job StatusWhitecollar1 1iflastorcurrentjobasamanager,administratororprofessional, 0otherwiseWhitecollar2 1ifclerical,salesorserviceworker,0otherwise(baselinecategory)Bluecollar 1iftradesperson,labourer,productionortransportworker, 0otherwiseGeographical variablesLivinginmajorcity 1iflivinginamajorcityarea,0otherwiseRegionalorremotearea 1iflivinginaregionalorremotearea,0otherwise(baseline category)Bornoverseas 1ifbornoverseas,0otherwiseBornAustralia 1ifborninAustralia,0otherwise(baselinecategory)


Table 2 - Specific Health Variables – Description

Variables DescriptionPhysical functioningVigorousactivities-limitedalittle 1iflimitedalittleintheabilityofperforming vigorousactivities,0otherwiseVigorousactivities-limitedalot 1iflimitedalotintheabilityofperforming vigorousactivities,0otherwiseModerateactivities-limitedalittle 1iflimitedalittleintheabilityofperforming moderateactivities,0otherwiseModerateactivities-limitedalot 1iflimitedalotintheabilityofperforming moderateactivities,0otherwiseLiftingorcarryinggroceries-limitedalittle 1iflimitedalittleintheabilityofliftingor carryinggroceries,0otherwiseLiftingorcarryinggroceries-limitedalot 1iflimitedalittleintheabilityofliftingor carryinggroceries,0otherwiseClimbingseveralflightsofstairs-limitedalittle 1iflimitedalittleintheabilityofclimbing severalflightsofstairs,0otherwiseClimbingseveralflightsofstairs-limitedalot 1iflimitedalotintheabilityofclimbing severalflightsofstairs,0otherwiseClimboneflightofstairs-limitedalittle 1iflimitedalittleintheabilityofclimbing oneflightsofstairs,0otherwiseClimboneflightofstairs-limitedalot 1iflimitedalotintheabilityofclimbingone flightsofstairs,0otherwiseBending,kneelingorstooping-limitedalittle 1iflimitedalittleintheabilityofbending, kneeling,orstooping,0otherwiseBending,kneelingorstooping-limitedalot 1iflimitedalotintheabilityofbending, kneeling,orstooping,0otherwiseWalkingonekilometre-limitedalittle 1iflimitedalittleintheabilityofwalking morethan1kilometre,0otherwiseWalkingonekilometre-limitedalot 1iflimitedalotintheabilityofwalkingmore than1kilometre,0otherwiseWalkinghalfkilometre-limitedalittle 1iflimitedalittleintheabilityofwalking halfakilometre,0otherwiseWalkinghalfkilometre-limitedalot 1iflimitedalotintheabilityofwalkinghalf akilometre,0otherwiseWalking100metres-limitedalittle 1iflimitedalittleintheabilityofwalking 100meters,0otherwiseWalking100metres-limitedalot 1iflimitedalotintheabilityofwalking100 meters,0otherwiseBathinganddressing-limitedalittle 1iflimitedalittleintheabilityofbathingor dressing,0otherwiseBathinganddressing-limitedalot 1iflimitedalotintheabilityofbathingor dressing,0otherwiseRole-physical (work and regular daily activities)Lesswork 1ifrespondentspendslesstimeworking, 0otherwiseAccomplishless 1ifrespondentaccomplisheslessthanhe wouldlike,0otherwiseLimitedinthekindofwork 1ifrespondentislimitedinthekindofwork due,0otherwiseDifficultiesworking 1ifrespondenthasdifficultiesperforming work,0otherwise



Table 2 - Specific Health Variables – Description (continued)

Variables DescriptionBodily painMildbodilypain 1ifrespondentsuffersfromverymildormild bodilypain,0otherwiseModeratebodilypain 1ifrespondentsuffersfrommoderatebodily pain,0otherwiseSeverebodilypain 1ifrespondentsuffersfromsevereorvery severebodilypain,0otherwisePaininterferesslightlywithwork 1respondent’sbodilypaininterferesslightly withwork,0otherwisePaininterferesmoderatelywithwork 1ifrespondent’sbodilypaininterferes moderatelywithwork,0otherwisePaininterferesalotwithwork 1ifrespondent’sbodilypaininterferesquitea bitorextremelywithwork,0otherwise

Labour Market Status Weuseobservedtransitionsbetweeneconomicactivityandinactivityasourmeasureof labour market exit. More specifically, our definition of economic inactivitycomprises individuals who classify themselves as retired, unpaid family workers,unpaidvolunteers, lookingafteran illpersonordisabled.Transitions fromactivitytoinactivityhavebeenusedbeforeasanoutcomemeasureinanalysingtheeffectsofhealthonretirement(Boundet al.,1999;Disneyet al.,2006).Itsuseisjustifiedbyconcernsregardingtheaccuracyofself-reportedretirementmeasureswhichisalsocomplicatedbythenotionofadisabilityrouteintoretirement.

Health Variables The HILDA Survey contains a series of health related variables both in the self-completionquestionnaire,whichcontainstheSF-36HealthandWell-BeingSurvey,and in the Person (interview) Questionnaire. In order to build the health-stockmeasure,wemakeuseof the5pointmeasureof self-assessedhealth (SAH)andaseriesofself-reportedhealthindicatorsrelatedtophysicalfunctioning,role-physicallimitations and bodily painwhich represent our specificmeasures of health (table2).The5pointmeasureofSAHisderivedfromthequestion:‘Ingeneral,wouldyousayyourhealth is: excellent; verygood; good; fair; poor’. Informationonphysicalfunctioningisderivedfromrespondents’answersonaseriesofquestionsaboutthedegreesoflimitationsinperformingasetofspecificactions,suchasclimbingflightsofstairs,liftingorcarryinggroceries,bending,kneelingorstooping,walkingdifferentdistances and bathing and dressing autonomously. We create dummy variablesfor the different degrees of each of these limitations. Role-physical functioningquestionsrelatetoproblemswithworkordailyactivitiesasaresultofphysicalhealth.Accordingly,wecreatefourdummyvariablestoreflectwhetheranindividualinthelastfourweekshadtocutdowntheamountoftimespentonworkorotheractivities;ifhehasaccomplishedlessthanhewouldlike;ifhewaslimitedinthekindofworkhewasdoingandhaddifficultiesinperformingworkorotheractivities.Wealsobuildaspecificsetofdummyvariablestodefinedifferentlevelsofbodilypainsufferedbyanindividualinthelastfourweeks(verymildormild;moderate;andsevereorvery


severebodilypain)andthedegreetowhichpaininterfereswithnormalwork(slightly;moderately;quiteabitorextremely). Inaddition,weuseanalternativemeasureofgeneralhealthbasedonhealthlimitations.Thismeasureisderivedfromthequestion:‘Does your health now limit you in these activities?’ followedby a series of dailyactivities.Wecreateadummyvariablewhichtakesavalueof1forthepresenceofanyoneofthesehealthlimitationsand0otherwise.

Income, Wealth and Housing Tenure Our income variable is the individual-specificmean of the log of total householdincome, which consists of all the sources of labour and non-labour equivalisedincome, across the 6waves of observations.As incomewill be systematically andsubstantially reduced after retirement, to ease problems related to endogeneity,weusethemeanoftheloghouseholdincomepriortoretirement.Totalhouseholdwealthisconstructedusinginformationonhouseholdnetworth.InHILDA,householdnetworthisdefinedasthedifferencebetweentotalhouseholdassetsandtotalhouseholddebtsandisprovidedinaspecialwealthmodulecollectedinwaves2and6.Tocapturetotal householdwealth prior retirement,we choose tomake use of information onhouseholdnetworthinwave2only.6Wealsoseparatelycontrolforhousingtenure.Ourretirementmodeldistinguishesbetweenindividualswhoowntheirhomeswithorwithoutamortgageandindividualswhoresideinrentedaccommodation.7

Household Variables In our model we also analyse the effect of household type (marital status) andcomposition (having dependent children) on individuals’ decisions to leave thelabourmarket.Therefore,togetherwithavariableindicatingwhetherarespondentismarriedorlivingwithapartner,themodelincludesadummyvariableindentifyingthepresenceofdependentchildren.Thesevariablesarebothlaggedoneperiodtocontrolforendogeneity.

Other Socio-economic Variables We also include other demographic, social and economic variables such as age,education,jobstatus(blueorwhitecollar),geographicalorigin(ifbornoverseas)andareaofresidence(iflivingwithinamajorcity’sarea).

Stock-sample and Descriptive Statistics Our stock-sample consists of 1564 individuals – 903men and 661 women – agedbetween50yearsoldandtheyearpriorstateretirementage(64and61yearsoldformenandwomenrespectively).8Individualsarefollowedthroughthefirstsixwavesofthe6Thatis,ourmeasureoftotalhouseholdwealthistime-invariantandusesonlywave2information.For a detailed description of the different components used to build household net worth inHILDA,seetheon-lineHILDAusermanual:http://www.melbourneinstitute.com/hilda/manual/userman_dvwealth.html.7The2008TaxReview(p.27)andarecentNATSEM(NationalCentreforSocialandEconomicModelling)researchreport(Kelly,2009)underlinethathomeownershipisasignificantfactorinretirementplanningamongAustralianindividuals.8AtthetimewhenthedatawerecollectedtheAustralianAgepensioncouldbepaidtopeopleaged65oroverformen,andaged62oroverforwomen.FordetailedandcomprehensivedescriptionsoftheAustralianretirementsystemseeWarrenandOguzoglu(2007)andKelly(2009).



HILDAsurveyuntiltheyretireorarecensored.Asweconsiderretirementanabsorbingstate,wemakeuseofinformationonlyuptothewavewherethisoccurs.Tables3to5describethetransitionsofindividualsofthestock-samplefromemploymentinwave1tootherlabourmarketstates,self-reportedretirementanddisability.Dataarepresentedtogetherandseparatelyformenandwomenandinformationonattritionanddeathisalsoprovided.Thenumberofmenandwomenwhoself-reportthemselvesasemployed(eitherasanemployeeorself-employed)rapidlydecreasesfrom1564to672betweenwave1andwave6.Also,thetotalnumberofinactiveindividualsincreasesfrom111inwave2to158inwave6.Thisrepresentsthe10percentoftheoriginalsampleof1564individuals.

Table 3 - Labour Market Status by Wave

1 2 3 4 5 6Employee 1090 836 732 621 545 474Own/Self-employed 474 364 304 259 222 198Unemployed 16 14 13 11 16Retired 60 74 83 91 101Unpaidfamilyworker 7 6 5 3 3Unpaidvolunteer 6 7 6 10 9Lookingafterillperson 1 2 7 4 4Disabled 21 24 26 33 25Attritionanddeath 253 148 143 101 89Total 1311 1163 1020 919 830Totalinactive 111 127 140 152 158Totalemployed 1564 1200 1036 877 767 672

Table 4 - Labour Market Status by Wave – Men

1 2 3 4 5 6Employee 566 440 387 325 287 246Own/Self-employed 337 256 218 186 161 145Unemployed 10 12 9 7 10Retired 33 48 45 54 57Unpaidfamilyworker 4 3 4 1 1Unpaidvolunteer 2 2 3 5 5lookingafterillperson 0 2 3 3 3Disabled 14 17 17 22 19Attritionanddeath 144 70 97 52 54Total 759 689 592 540 486Totalinactive 63 84 81 92 95Totalemployed 903 696 605 511 448 391


Table 5 - Labour Market Status by Wave – Women

1 2 3 4 5 6Employee 524 396 345 296 258 228Own/Self-employed 137 108 86 73 61 53Unemployed 6 2 4 4 6Retired 27 26 38 37 44Unpaidfamilyworker 3 3 1 2 2Unpaidvolunteer 4 5 3 5 4Lookingafterillperson 1 0 4 1 1Disabled 7 7 9 11 6Attritionanddeath 109 78 46 49 35Total 552 474 428 379 344Totalinactive 48 43 59 60 63Totalemployed 661 504 431 369 319 281

Table 6 reports descriptive statistics for all data and broken down byemploymentstatus.Thesearepresentedformenandwomenseparatelyandincludeaseriesofhealthvariables(healthlimitations;thefivecategoriesofSAHandameasureofhealthshocks,thatiswhetheranindividualhassufferedfromaseriousinjuryorillness in the previous 12months) and a set of socioeconomic characteristics (age,marital status, the presence of dependent children, household income and wealth,housing tenure, education, geographical variables and job characteristics for thoseemployed).Aclearpositiverelationshipbetweenlabourforceparticipationandhealthstatusemerges.Thatis,thebetterthehealthofthoseofworking-age,themorelikelytheyaretoremaininthelabourforce.Thisistrueforbothmenandwomen.Concerninghealthshocks,itisnotablethattheproportionofmenreportingahealthshocknearlydoublesforthegroupofnon-employedindividualscomparedtoindividualsinwork.As for the other socioeconomic characteristics, for both genders being outside thelabourmarket appears to be associatedwith a higher average age, the absence ofdependentchildren,aslightlylowerhouseholdincomeandalowerhouseholdwealth.Thedataalsoappearstorevealthepresenceofaneducationalgradient,withahigherproportion of educated individuals among the employed. Further,most individualsinthestocksamplereporthavingapartner(85.7percentofmenand70.5percentofwomen).However,thepercentageofindividualsincouplesisgreaterforthesub-sampleofnon-employedmaleindividuals.



Table 6 - Descriptive Statistics

Men Women All In work Inactive All In work InactiveHealth variablesHealthlimitations 0.269 0.227 0.589 0.208 0.213 0.342SAHexcellent 0.102 0.111 0.024 0.097 0.110 0.047SAHgood 0.366 0.386 0.238 0.402 0.400 0.288SAHVerygood 0.378 0.380 0.338 0.373 0.375 0.426SAHfair 0.125 0.107 0.270 0.113 0.104 0.181SAHpoor 0.030 0.016 0.129 0.015 0.011 0.058Healthshocks 0.101 0.085 0.194 0.081 0.082 0.118Socioeconomic characteristicsAge 57.014 56.657 59.257 55.818 56.103 57.412Married/couple 0.857 0.856 0.818 0.705 0.752 0.756Owndependentchildren 0.334 0.334 0.149 0.274 0.293 0.194Loghouseholdincome 11.153 11.201 10.739 11.062 11.118 10.871Householdwealth* 82.475 89.014 68.517 73.619 86.189 68.525Rent 0.090 0.092 0.057 0.103 0.090 0.104Education/degrees 0.236 0.233 0.182 0.238 0.239 0.184Education/certificate 0.377 0.363 0.418 0.255 0.291 0.282Education12 0.388 0.404 0.400 0.507 0.470 0.534Whitecollar1 0.474 0.536 0.000 0.419 0.530 0.000WhiteCollar2 0.202 0.236 0.000 0.287 0.291 0.000Bluecollar 0.198 0.208 0.000 0.117 0.139 0.000Livinginamajorcity 0.314 0.302 0.292 0.255 0.272 0.244Bornoverseas 0.594 0.597 0.462 0.608 0.583 0.598

Note:*householdwealthisdividedby10000.

Kaplan-Meier survival estimates of the probability of survival (not leavingthe labour force)aredisplayed infigures1 to6.Estimatesarepresentedforhealthlimitations,SAHandhealthshocksdefinedasinjuryorillness,formenandwomenseparately.Figures1and2showthatmenreportinghealthlimitationsandpoorhealthhaveagreaterprobabilityofleavingthelabourforceifcomparedtomennotreportinghealthlimitationsorreportingbetterself-assessedhealth.Similar,butsmallereffects,canbefoundforwomeninfigures4and5.Survivalestimatesformeninfigure3showtheprobabilityofnot retiringbyhealth shocks.Maleswho suffered fromahealthshockduring thepreviousyear have an increasedprobabilityof exiting the labourmarket.Oncemore,lowerprobabilitiesofretiringareassociatedwithwomenhavingsufferedfromahealthshock(figure6).


Figure 1 - Kaplan-Meier Survival Estimates of the Proportion of Men NotLeaving the Labour Force by Health Limitations

Figure 2 - Kaplan-Meier Survival Estimates of the Proportion of Men NotLeaving the Labour Force by Self-assessed Health

50

1.00

0.00

0.50

Prop

ortio

n no

t ret

ired

Age6555 60

0.75

0.25

Health_limitations = 1Health_limitations = 0

50

1.00

0.00

0.50

Prop

ortio

n no

t ret

ired

Age6555 60

0.75

0.25

SAH = ExcellentSAH = GoodSAH = Poor

SAH = Very goodSAH = Fair



Figure 3 - Kaplan-Meier Survival Estimates of the Proportion of Men NotLeaving the Labour Force by Health Shocks

Figure 4 - Kaplan-Meier Survival Estimates of the Proportion of WomenNot Leaving the Labour Force by Health Limitations

50

1.00

0.00

0.50

Prop

ortio

n no

t ret

ired

Age6555 60

0.75

0.25

Health_shocks = 1Health_shocks = 0

50

1.00

0.00

0.50

Prop

ortio

n no

t ret

ired

Age55 60

0.75

0.25

Health_limitations = 1Health_limitations = 0


Figure 5 - Kaplan-Meier Survival Estimates of the Proportion of WomenNot Leaving the Labour Force by Self-assessed Health (SAH)

Figure 6 - Kaplan-Meier Survival Estimates of the Proportion of WomenNot Retired by Health Shocks

50

1.00

0.00

0.50

Prop

ortio

n no

t ret

ired

Age55 60

0.75

0.25

SAH = ExcellentSAH = GoodSAH = Poor

SAH = Very goodSAH = Fair

50

1.00

0.00

0.50

Prop

ortio

n no

t ret

ired

Age55 60

0.75

0.25

Health_shocks = 1Health_shocks = 0



5. Results Health-stock Measure Table7presentsresultsforthelatenthealthstockobtainedbyregressingself-assessedhealth (SAH) on a set of more specific measures of health using pooled orderedprobitmodels.Thesetofhealthmeasuresusedasregressorsinthelatenthealthstockmodelincludesvariablesthatcapturedifferentdegreesoffunctionallimitations,role-physicallimitationsandvariouslevelsofbodilypain.Thesemodelswereestimatedonmen andwoman separately on data from the stock sample used for the labourmarketexitsmodels.Asexpected,bothformenandwomen,thevastmajorityoftheestimatedcoefficientsdisplaypositivesigns.Astheself-assessedhealthvariableusedis increasing in ill-health, reporting health problems is positively associated withpoorerself-ratedhealth.

Table 7 - Pooled Ordered Probit Models for SAH

Men WomenLatent health index Coef. S.E. Coef. S.E.Physical functioningVigorousactivities/limitedalittle 0.438*** (0.048) 0.335*** (0.059)Vigorousactivities/limitedalot 0.785*** (0.066) 0.536*** (0.076)Moderateactivities/limitedalittle 0.156** (0.068) 0.173** (0.074)Moderateactivities/limitedalot 0.132 (0.142) 0.0529 (0.161)Liftingorcarryinggroceries/limitedalittle 0.251*** (0.080) 0.262*** (0.079)Liftingorcarryinggroceries/limitedalot 0.442*** (0.167) 0.354** (0.179)Climbingseveralflightsofstairs/limitedalittle 0.290*** (0.053) 0.284*** (0.056)Climbingseveralflightsofstairs/limitedalot 0.695*** (0.122) 0.351*** (0.126)Climboneflightofstairs/limitedalittle 0.0266 (0.091) 0.193** (0.097)Climboneflightofstairs/limitedalot -0.0202 (0.178) -0.106 (0.200)Bending,kneelingorstopping/limitedalittle 0.0612 (0.048) -0.0856 (0.056)Bending,kneelingorstopping/limitedalot -0.247** (0.097) -0.175 (0.123)Walkingonekilometre/limitedalittle 0.246*** (0.070) 0.345*** (0.071)Walkingonekilometre/limitedalot 0.523*** (0.138) 0.463*** (0.162)Walkinghalfkilometre/limitedalittle 0.0304 (0.105) -0.177 (0.123)Walkinghalfkilometre/limitedalot -0.0684 (0.197) -0.137 (0.237)Walking100metres/limitedalittle -0.0548 (0.121) 0.0904 (0.144)Walking100metres/limitedalot 0.0792 (0.238) -0.326 (0.304)Bathinganddressing/limitedalittle 0.141 (0.101) 0.444*** (0.152)Bathinganddressing/limitedalot -0.420** (0.206) 0.182 (0.265)Role-PhysicalLesswork 0.252*** (0.085) -0.0206 (0.092)Accomplishless 0.215*** (0.070) 0.390*** (0.082)Limitedinthekindofwork -0.0689 (0.087) -0.124 (0.097)Difficultiesworking 0.208*** (0.078) 0.386*** (0.092)Bodily painMildbodilypain 0.284*** (0.048) 0.279*** (0.059)Moderatebodilypain 0.470*** (0.081) 0.428*** (0.091)Severebodilypain 0.549*** (0.130) 0.692*** (0.145)Paininterferesslightlywithwork 0.246*** (0.053) 0.158*** (0.060)Paininterferesmoderatelywithwork 0.206** (0.090) 0.134 (0.104)Paininterferesalotwithwork 0.327** (0.131) 0.291* (0.154)Observations 3552 2615Log-Likelihood 3779.61 -2709.83

Standarderrorsinparentheses;***p<0.01,**p<0.05,*p<0.1.


Formen,weobservethelargesteffects,intermsofthesizeofthecoefficients,forfunctionallimitationsrelatedtovigorousactivities,climbingseveralflightsofstairsandwalkingonekilometreaswellas forhigher levelsofbodilypain.Forwomen,largeeffectsareobservedforasimilarsetofproblemstothoseobservedformen.

Duration Analysis Results for the discrete-time hazard models of labour market exits are displayedseparatelyformenandwomenintables8a-8bandtables9a-9brespectively.Eachtablecontains results for health limitations and self-assessed latent health and show theestimatedcoefficients,standarderrorsandhazardratiosforeachvariable.Thehazardratiomeasurestheproportionaleffectsontheunderlyinghazardofleavingthelabourforceofaunitchangeinthevalueofagivenvariable.Hazardratiosarecentredaround1,allpossibledecreasesintheprobabilityofleavingthelabourmarketliebetween0and1whileallpossibleincreasesintheriskofleavingthelabourforcelieabove1.Themodelswereestimatedincorporatingunobservedheterogeneity(frailty)usingalternativelyaGammamixturedistribution(Meyer,1990)andanormaldistribution.9Neglectingunobservedheterogeneityindurationmodelsmayleadtoseriousbiases.It may lead to an overestimation of the negative duration dependence or to anunderestimationoftheeffectoftheexplanatoryvariablesonthehazard(Lancaster,1990;vandenBerg,2001).However,evidencesuggests that themisspecificationoftheunobservedheterogeneitydistributionindiscrete-timedurationmodelsdoesnotseriouslyaffecttheestimationresults:itdoesnotaffectdurationdependenceorthecovariate coefficients (Nicoletti and Rondinelli, 2006). This appears to imply thatwhile frailty must be taken into account when estimating discrete-time durationmodels,theexactspecificationislessimportant.

Table 8a - Hazard Model for Labour Market Exits – Men

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioHealth variablesHealthlimitations(0) 0.786*** (0.269) 2.195Healthlimitations(t-1) 0.159 (0.231) 1.172Latenthealth(0) 0.0518 (0.173) 1.053Latenthealth(t-1) 0.410*** (0.154) 1.506Other covariatesEducation/degrees 0.495* (0.291) 1.641 0.645** (0.315) 1.907Education/certificate 0.233 (0.217) 1.263 0.302 (0.237) 1.352Whitecollar(0) -0.169 (0.254) 0.845 -0.128 (0.272) 0.880Bluecollar(0) 0.339 (0.271) 1.403 0.301 (0.298) 1.351Loghouseholdincome(t-1) -0.872*** (0.194) 0.418 -0.727*** (0.205) 0.483Householdwealth 0.002 (0.001) 1.001 0.001 (0.001) 1.001

9Modelswereestimated inSTATAusing thepgmhaz8 routine (Jenkins,1998)which includesunobservedheterogeneityasaGammamixturedistributionaswellasusingthextcloglogcommandwhichassumesnormallydistributedunobservedheterogeneity.Resultswereverysimilar,hencewechoosetoreportestimatesfromonlyoneofthetwomodelspecifications(theonewithGammadistributedfrailty).Thecompletesetofresultsisavailableuponrequest.



Table 8a - Hazard Model for Labour Market Exits – Men (continued)

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioRentinghome(t-1) -0.171 (0.327) 0.843 -0.203 (0.371) 0.816Bornoverseas -0.296 (0.215) 0.744 -0.153 (0.227) 0.858Livinginamajorcity -0.0350 (0.197) 0.966 -0.130 (0.210) 0.878Married/couple(t-1) 0.188 (0.271) 1.206 0.169 (0.296) 1.184Dependentchildren(t-1) -0.541** (0.228) 0.582 -0.648** (0.252) 0.523Age53 0.666 (0.521) 1.947 0.809 (0.618) 2.246Age54 0.495 (0.521) 1.641 0.274 (0.645) 1.316Age55 1.095** (0.488) 2.990 1.268** (0.573) 3.552Age56 0.686 (0.526) 1.985 0.883 (0.610) 2.417Age57 0.716 (0.522) 2.046 0.715 (0.619) 2.044Age58 1.034** (0.509) 2.811 0.994 (0.615) 2.703Age59 0.503 (0.553) 1.654 0.709 (0.640) 2.033Age60 1.411*** (0.500) 4.101 1.616*** (0.593) 5.032Age61 1.671*** (0.510) 5.315 1.927*** (0.607) 6.872Age62 1.529*** (0.544) 4.616 1.506** (0.649) 4.508Age63 1.495*** (0.555) 4.460 1.649** (0.664) 5.201Age64 1.912*** (0.576) 6.764 2.012*** (0.683) 7.481Observations 2760 2340LRtestforgammavariation 12.619 8.725P-value 0.000 0.001Loglikelihood -655.305 -547.242


Table 8b - Hazard Model for Labour Market Exits – Men

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioHealth variablesHealthlimitations(0) 0.750*** (0.234) 2.116Latenthealth(0) 0.289** (0.133) 1.335Healthshocks 1.427*** (0.235) 4.167 1.360*** (0.246) 3.896Other covariatesEducation/degrees 0.494 (0.310) 1.640 0.466 (0.316) 1.594Education/certificate 0.337 (0.236) 1.401 0.320 (0.245) 1.377Whitecollar(0) -0.116 (0.276) 0.891 -0.0337 (0.284) 0.967Bluecollar(0) 0.368 (0.295) 1.444 0.330 (0.313) 1.391Loghouseholdincome(t-1) -0.867*** (0.213) 0.420 -0.711*** (0.215) 0.491Householdwealth 0.002 (0.001) 1.002 0.001 (0.001) 1.001Rentinghome(t-1) -0.466 (0.367) 0.628 -0.434 (0.383) 0.648Bornoverseas -0.292 (0.232) 0.747 -0.259 (0.239) 0.772Livinginamajorcity -0.0837 (0.211) 0.920 -0.0919 (0.216) 0.912Married/couple(t-1) 0.182 (0.298) 1.200 0.264 (0.310) 1.302Dependentchildren(t-1) -0.562** (0.248) 0.570 -0.645** (0.260) 0.525Age53 0.765 (0.567) 2.149 0.810 (0.618) 2.248Age54 0.333 (0.576) 1.395 0.224 (0.639) 1.252Age55 1.243** (0.534) 3.465 1.221** (0.580) 3.391


Table 8b - Hazard Model for Labour Market Exits – Men (continued)

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioAge56 0.707 (0.572) 2.028 0.725 (0.615) 2.065Age57 0.768 (0.574) 2.156 0.684 (0.623) 1.983Age58 1.171** (0.557) 3.226 1.051* (0.611) 2.860Age59 0.617 (0.608) 1.854 0.552 (0.660) 1.736Age60 1.582*** (0.548) 4.864 1.533** (0.599) 4.633Age61 1.876*** (0.554) 6.526 1.871*** (0.603) 6.493Age62 1.664*** (0.598) 5.282 1.439** (0.658) 4.218Age63 1.729*** (0.618) 5.635 1.723*** (0.665) 5.599Age64 2.165*** (0.630) 8.717 2.144*** (0.684) 8.536Observations 2596 2383LRtestforgammavariation 14.408 10.618P-value 0.000 0.000Loglikelihood -594.807 -550.551


Inordertoassesstheeffectofindividualhealthstatusindeterminingearlyexitdecisions,weconsiderbothageneralmeasureofhealthlimitationsandthelatenthealth stockmeasure obtained from the pooled ordered probitmodels. These twovariablesarelaggedoneperiodtoavoidproblemsofsimultaneity.Wealsoconditiononfirstperiodhealthstatus.Inthiswaytheestimatedcoefficientsoflaggedhealthcanbeinterpretedasahealthshock(table8aformen,9aforwomen).Wealsoestimatemodels for health limitations and self-assessed latent health using an alternativemeasureofhealthshockswhichidentifiesthepresenceofaseriousinjuryorillnessintheprevious12months(tables8band9b).

Formen,we observe a large, positive and highly significant effect for theinitial period health limitations variable (table 8a). Thismeans that the hazard ofleavingthelabourforceisgreaterforindividualssufferingfromphysicallimitations.Wealsoobserveapositiveandsignificantcoefficientforourmeasureoflatenthealth-stocklaggedoneperiod.Sincethemeasureisincreasingwithill-health,thisimpliesthatthehazardofanearlyexitincreasesforindividualssufferingfromahealthshock(definedintermsofadeviationfromfirstperiod,wave1,healthstock).Accordingtothecomputedquantitativeeffects(hazardratios),thepresenceoflimitationsincreasethehazardofleavingthelabourforceby122percentwhereashealthshocks(laggedhealthstockvariable),increasethesamehazardbyaround50percent(table8a).Theeffectsofthehealthandhealthshocksvariablesbecomelargerusinganalternativehealth shocks definition (table 8b). In particular, the occurrence of a health shockdefinedasseriousinjuryorillnessisassociatedwitha320percentincreaseintheprobabilityofbecomingnon-employedinthemodelforhealthlimitationsandwitha290percentincreaseinthehealthstockmodel.

Allmodelsshowthesamegradientovertheagecategorieswiththehazardofan



earlyexitbecomingpositive,largerandstatisticallysignificantasstatuarypensionageapproaches.Thequantitativeeffectsoftheagevariablesareparticularlylarge:being64yearsoldincreasesthelikelihoodofleavingthelabourforcebetweennearlysixandeighttimes(tables8aand8b).Ashouseholdincomeincreases,thehazardofretirementdecreaseswhiletheeffectsoftotalhouseholdwealthandhousingtenure(rentinghome)donotappeartobestatisticallysignificantinanyofthemodelsconsidered.

Consistently acrossmodels, early exit decisions formen are a function ofhavingdependentchildrenbutnotofmaritalstatus.Morespecifically,theestimatedcoefficientsforhavingdependentchildrenarenegativeandsignificant.Theeffectiscomparedtothebaselinecategoryofnothavingdependentchildren.Thissuggeststhatformen,havingtoprovidemaintenancetodependentchildrendecreasesthelikelihoodofleavingthelabourmarketbyaround50percent(table8aand8b).However,livingwithapartnerorbeingmarrieddoesnotseemtosignificantlyinfluencethelikelihoodofanearlyexit(coefficientsarepositivebutnotsignificant).

Wealsoobserve a gradient across educational attainment compared to thebaselinecategoryofnoqualifications:higherlevelsofeducationareassociatedwithanincreasinghazardofleavingthelabourforce(table8a).Theriskisalsorelativelylargeandpositiveforbluecollarlabourersandnegativeformanagers,administratorsand professionals, even if the effects are not significant. This is compared to thebaselineformedbyclerical,salesorserviceworkers.

Table 9a - Hazard Model for Labour Market Exits – Women

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioHealth variablesHealthlimitations(0) -0.119 (0.325) 0.888Healthlimitations(t-1) 0.555** (0.254) 1.742Latenthealth(0) -0.325 (0.234) 0.723Latenthealth(t-1) 0.516** (0.206) 1.676Other covariatesEducation/degrees -0.141 (0.308) 0.869 -0.294 (0.369) 0.745Education/certificate 0.0837 (0.251) 1.087 -0.0147 (0.314) 0.985Whitecollar(0) -0.0532 (0.252) 0.948 -0.0276 (0.308) 0.973Bluecollar(0) 0.157 (0.322) 1.171 0.0938 (0.399) 1.098Loghouseholdincome(t-1) 0.666*** (0.199) 0.514 -0.895*** (0.242) 0.408Householdwealth 0.001 (0.001) 1.001 0.001 (0.001) 1.002Rentinghome(t-1) 0.0631 (0.360) 1.065 0.154 (0.439) 1.167Bornoverseas -0.0524 (0.242) 0.949 -0.221 (0.303) 0.802Livinginamajorcity 0.373* (0.224) 1.452 0.446* (0.269) 1.561Married/couple(t-1) 0.881*** (0.262) 2.414 1.178*** (0.324) 3.249Dependentchildren(t-1) -0.003 (0.248) 0.996 -0.001 (0.295) 0.999Age53 0.420 (0.515) 1.522 0.153 (0.652) 1.165Age54 0.610 (0.478) 1.841 0.855 (0.561) 2.351Age55 1.458*** (0.440) 4.298 1.682*** (0.533) 5.375Age56 0.933* (0.491) 2.542 0.989* (0.596) 2.688Age57 1.020** (0.491) 2.772 1.315** (0.583) 3.724Age58 1.432*** (0.493) 4.185 1.597*** (0.596) 4.937Age59 1.383*** (0.526) 3.988 1.635** (0.635) 5.127


Table 9a - Hazard Model for Labour Market Exits – Women (continued)

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioAge60 1.192** (0.570) 3.295 1.696** (0.685) 5.451Age61 1.830*** (0.542) 6.236 2.124*** (0.663) 8.366Observations 1997 1652LRtestforgammavariation 6.431 9.025P-value 0.005 0.001Loglikelihood -554.944 -432.834


Table 9b - Hazard Model for Labour Market Exits – Women

Health Limitations Latent Self-assessed Health Hazard Hazard Coef. S.E. Ratio Coef. S.E. RatioHealth variablesHealthlimitations(0) 0.242 (0.289) 1.273 0.135 (0.155) 1.144Latenthealth(0)Healthshocks 0.274 (0.306) 1.315 0.106 (0.354) 1.112Other covariatesEducation/degrees -0.177 (0.320) 0.838 -0.309 (0.357) 0.734Education/certificate 0.105 (0.260) 1.110 -0.0259 (0.295) 0.974Whitecollar(0) -0.182 (0.260) 0.833 -0.158 (0.291) 0.854Bluecollar(0) 0.0316 (0.331) 1.032 -0.103 (0.375) 0.902Loghouseholdincome(t-1) -0.710*** (0.207) 0.492 -0.840*** (0.232) 0.432Householdwealth 0.002 (0.001) 1.001 -0.001 (0.001) 1.000Rentinghome(t-1) 0.141 (0.373) 1.152 0.295 (0.417) 1.343Bornoverseas -0.0596 (0.252) 0.942 -0.144 (0.285) 0.866Livinginamajorcity 0.373 (0.233) 1.453 0.363 (0.257) 1.438Married/couple(t-1) 1.128*** (0.283) 3.090 1.326*** (0.320) 3.768Dependentchildren(t-1) 0.0450 (0.261) 1.046 0.0495 (0.288) 1.051Age53 0.334 (0.569) 1.397 0.177 (0.594) 1.194Age54 0.843* (0.506) 2.324 0.765 (0.526) 2.150Age55 1.551*** (0.478) 4.717 1.483*** (0.497) 4.405Age56 1.018* (0.531) 2.766 0.857 (0.562) 2.356Age57 1.196** (0.521) 3.308 1.148** (0.548) 3.151Age58 1.646*** (0.530) 5.185 1.452*** (0.562) 4.270Age59 1.443** (0.560) 4.231 1.416** (0.587) 4.122Age60 1.422** (0.612) 4.147 1.554** (0.656) 4.728Age61 1.893*** (0.576) 6.639 1.824*** (0.619) 6.200Observations 1879 1676LRtestforgammavariation 5.6531 7.50445P-value 0.008712 0.003077Loglikelihood -506.511 -438.962




Forwomen, ill-health and health shocks haveweaker effects on early exitdecisions.Onlythecoefficientsforthelaggedmeasuresofhealth,healthlimitationsand latent health stock, are positive andhighly significant (table 9a).According tothe hazard ratios for these variables, experiencing health shocks leading to healthlimitations increase the hazard of an early exit by around 74 per cent while adeteriorationinthehealthstock(healthshock)increasethesamehazardbyaround68percent.Allotherhealthandhealthshockrelatedcoefficientsarenotstatisticallysignificant(table9a).

Contrarytomen,forwomenexitdecisionsareapositivefunctionofmaritalstatus.Beingmarriedor livingwithapartner substantially increases thehazardofleavingthelabourforce.However,havingowndependentchildrendoesnotappeartosignificantlyinfluencewomen’searlyexitsfromthelabourmarket.Ageappearstobeamongthemostimportantfactorsinwomen’searlyexitdecisions:agecategoriesfromage55onwardsarehighlysignificantforallthemodelsandthecorrespondinghazardratiosareparticularlylarge.Qualitatively,theeffectsoftheothernon-healthvariablesarethesameasinthecorrespondingmodelsformen.

6. Conclusions and Policy Implications This paper examines and quantifies the role of ill-health and health shocks indeterminingdecisionstoleavethelabourmarketamongolderworkingindividuals.We concentrate only on early exits and we use a discrete-time hazard model torepresenttransitionstonon-employmentonAustralianlongitudinaldata.Weextendearlieranalysesaccountingforthepotentialreportingbiasandendogeneityintrinsicinmeasuresofself-assessedhealthbycreatingalatenthealth-stockvariablewhichweuseasoneofourmeasuresofhealth,togetherwithameasureofhealthlimitations.The latent health index estimatesSAHas a functionofmore specificmeasures ofhealth using pooled ordered probit models. We also define health shocks in twodistinctwaysandconsidertheeffectsofanumberofsocioeconomiccharacteristicsonanindividual’searlyexitdecision.

Regardlessofthehealthvariablesused,resultsconsistentlyindicatethatill-health andhealth shocks arekeydeterminants of early exit decisions amongolderworkingindividuals.Inparticular,healthshocksgreatlyincreasetheriskofanearlyexitfromthelabourmarket.Dependingonthedefinitionofahealthshock,formennegativeshockstohealthincreasethehazardofleavingthelabourforceby50to320percent,whereasforwomenhealthshocksenhancetheriskofanearlyexitby68to74percent.

Theeffectsofhouseholdtypeandcompositionarealsoofinterestandimplyasubstantialasymmetry in labourmarketbehaviour formenandwomen.Whereashavingapartner increases theprobabilityof leaving the labourmarket forwomen,it does not significantly increase the probability formen.On the contrary, havingdependentchildrensignificantlydecreases the riskofanearlyexit fromthe labourforceformenbutdoesnotappeartoaffectwomen’sexitchoices.

Ourresultshaveimportantpolicyimplications.Theyindicatethathealthmustbetakenintoaccountwhendesigningpoliciesaimedatencouragingolderindividualstoremaininthelabourforceoryoungerretireestore-enterthelabourmarket.


References Anderson,K.H.andBurkhauser,R.V.(1985),‘TheRetirement-HealthNexus:ANew

MeasureofanOldPuzzle’, Journal of Human Resources, 20,315-330.Au, D., Crossley, T.F. and Schellhorn, M. (2005), ‘The Effect of Health Changes

and Long-term Health on theWork Activity of Older Canadians’,Health Economics, 14,999-1018.

Auer,P.andFortuny,M.(2000), ‘Ageingof theLabourForce inOECDCountries:EconomicandSocialConsequences’, ILO Employment Paper,2000/2.

AustralianDepartmentofTreasury(2007),‘IntergenerationalReport’.AustralianDepartmentofTreasury(2010),‘IntergenerationalReport’.Bazzoli,G.(1985),‘TheEarlyRetirementDecision:NewEmpiricalEvidenceonthe

InfluenceofHealth’, Journal of Human Resources, 20,214-234.Blau,D.M.(1998), ‘LabourForceDynamicsofOlderMarriedCouples’,Journal of

Labor Economics,16,595-629.Blau,D.M.andGilleskie,D.(2001),‘TheEffectofHealthonEmploymentTransitions

ofOlderMen’,InPOLACHECK,S.(ed.)Worker Well-Being in a Changing Labour Market,Amsterdam,JAIPress.

Blundell,R.,Meghir,C.andSmith,S.(2002),‘PensionIncentivesandthePatternofEarlyRetirement’, Economic Journal,112,153-170.

Bound, J. (1991) ‘Self-reportedVersusObjectiveMeasuresofHealth inRetirementModels’, Journal of Human Resources,26,106-138.

Bound, J., Schoenbaum, M., Stinebrickner, T.R. and Waidmann, T. (1999), ‘TheDynamicEffectsofHealthontheLabourForceTransitionsofOlderWorkers’,Labour Economics,6,179-202.

Brazenor,R.(2002),‘DisabilitiesandLabourMarketEarnersinAustralia’,AustralianJournal of Labour Economics, 5,319-334.

Brown, S., Roberts, J. and Taylor, K. (2010), ‘ReservationWages, LabourMarketparticipationandhealth’,Journal of the Royal Statistical Society: Series A (Statistics in Society),InPress.

Burkhauser,R.V.(1979),‘ThePensionAcceptanceDecisionofOlderWorkers’,Journal of Human Resources,14,63-75.

Burtless,G.andMoffitt,R.A.(1984),‘TheEffectsofSocialSecurityBenefitsontheLabourSupplyoftheAged’,InAaroon,H.andBurtless,G.(Eds.)Retirement and Economics Behaviour, WashingtonDC,BrookingsInstitution.

Cai,L.andKalb,G.(2006),‘HealthStatusandLabourForceParticipation:EvidencefromAustralia’,Health Economics,15,241-261.

Cai,L.,Mavromaras,K.G.andOguzoglu,U.(2008),‘TheEffectsofHealthandHealthShocksonHoursWorked’,IZA Discussion Papers,InstitutefortheStudyofLabor(IZA).

Disney,R.,Emmerson,C.andWakefield,M. (2006), ‘IllHealthandRetirement inBritain:APanelData-basedAnalysis’,Journal of Health Economics,25,621-649.

Dwyer,D.andMitchell,O.(1998),‘HealthProblemsasDeterminantsofRetirement:areSelf-ratedMeasuresEndogenous’, Journal of Health Economics,18,173-193.



French,E.(2005),‘TheEffectsofHealth,WealthandWagesonLabourSupplyandRetirementBehaviour’,The Review of Economic Studies,72,395-427.

García-Gómez,P.,Jones,A.M.andRice,N.(2010),‘HealthEffectsonLabourMarketExitsandEntries’, Labour Economics, 17,62-76.

Goode,A.andWatson,N.(2006),User Manual – Release 6.0,MelbourneInstituteofAppliedEconomicandSocialResearch,UniversityofMelbourne.

Hagan, R., Jones, A.M. and Rice, N. (2009), ‘Health and Retirement in Europe’,International Journal of Environmental Research and Public Health, 6,2676-95.

Jenkins,S.P.(1995),‘EasyEstimationMethodsforDiscrete-timeDurationModels’,Oxford Bulletin of Economics and Statistics,57,129-138.

Jenkins, S.P. (1998), ‘Discrete Ttime Proportional Hazards Regression’, STATA Technical Bulletin,39,22-32.

Jones,A.M.,Rice,N. andRoberts, J. (2009), ‘Sick ofWork orTooSick toWork?Evidence on Self-reported Health Shocks and Early Retirement from theBHPS‘, Economic Modelling,InPress,doi:10.1016/j.econmod.2009.10.001

Kelly, S. (2009), ‘Reform of the Australian Retirement Income System’,National Centre for Social and Economic Modelling (NATSEM) Research Report,Canberra.

Kerkhofs, M., Lindeboom, M. and Theeuwes, J. (1999), ‘Retirement, FinancialIncentivesandHealth’,Labour Economics,6,203-227.

Lambrinos,J.(1981)‘Health:ASourceofBiasinLabourSupplyModels’,Review of Economics and Statistics,63,206-12.

Lancaster, T. (1990), The Econometric Analysis of Transition Data, Cambridge,CambridgeUniversityPress.

Laplagne,P.,Glover,M. andShomos,A. (2007), ‘Effects ofHealth andEducationonLabourForceParticipation’,Staff Working Paper,AustralianGovernment,ProductivityCommission.

Lindeboom,M.(2006a),‘HealthandWorkofOlderWorkers‘, inJones,A.M.(ed.)Elgar Companion to Health Economics,EdwardElgar:Aldershot.

Lindeboom,M.,Llena-Nozal,A. andVanDerKlaauw,B. (2006b), ‘Disability andWork:TheRoleofHealthShocksandChildhood‘,IZA Discussion Paper,No.2096,Bonn,InstitutefortheStudyofLabor.

Lindeboom,M.andVanDoorslaer,E.(2004),‘Cut-pointShiftandIndexShiftinSelf-reportedHealth’,Journal of Health Economics, 23,1083-1099.

Lumsdaine, R.L. andMitchell, O.S. (1999), ‘New Developments in the EconomicAnalisysofRetirement’,InAshenfelter,O.C.andCard,D.(eds.)Handbook of Labor Economics,Amsterdam,Elsevier.

Meyer, B.D. (1990), ‘Unemployment Insurance and Unemployment Spells’,Econometrica,58,757-782.

Nicoletti, C. and Rondinelli, C. (2006), ‘The (mis)-specification of Discrete TimeDurationModelswithUnobservedHeterogeneity:AMonteCarloStudy’,ISER Working Paper 2006-53,InstituteforSocialandEconomicResearch(ISER).

Parsons, D. (1980), ‘The Decline inMale Labour Force Participation’, Journal of Political Economy,88,117-134.


Riphahn,R.T.(1999),‘IncomeandEmploymentEffectsofHealthShocks.ATestCasefor theGermanWelfareState’,Journal of Population Economics, 12,363-389.

Roberts, J., Rice, N. and Jones, A.M. (2008), ‘Early Retirement and Inequality inBritain and Germany: How Important is Health?’, Health, Econometrics and Data Group (HEDG) Working Papers 08/27,DepartmentofEconomics,UniversityofYork.

Schofield,D.J.,Shrestha,R.N.,Passey,M.E.,Earnest,A.andFletcher,S.L. (2008),‘ChronicDiseaseandLabourForceParticipationAmongOlderAustralians’,The Medical Journal of Australia,189,447-450.

Stern,S.(1989)‘MeasuringtheEffectofDisabilityonLabourLorceParticipation’,Journal of Human Resources,24,361-395.

Stickles,R.C. andTaubman, P. (1986), ‘AnAnalysis of theHealth andRetirementStatusoftheElderly’,Econometrica,54,1339-1356.

Stickles,R.C.andYazbeck,A.(1998),‘OntheDynamicsofDemandforLeisureandtheProductionofHealth’,Journal of Business and Economic Statistics,16,187-197.

Van Den Berg, B.J. (2001), ‘Duration Models: Specification, Identification andMultipleDurations’, inHeckman,J.J.andLeamer,E.E.(eds.) Handbook of Econometrics,Amsterdam,NorthHolland.

Warren, D. and Oguzoglu, U. (2007), ‘Retirement in Australia: A Closer Look atthe Financial Incentives’, Melbourne Institute Working Paper No. 24/07.Melbourne InstituteofAppliedandEconomicResearch,TheUniversityofMelbourne.

Watson, N. and Wooden, M. (2004), ‘Sample Attrition in the HILDA Survey’,Australian Journal of Labour Economics,7,293-308.

Wilkins,R.(2004),‘TheEffectsofDisabilityonLabourForceStatusinAustralia’,The Australian Economic Review, 37,359-382.

Documents

The Effects of Health Shocks on Labour Market Exits ...businesslaw.curtin.edu.au/wp-content/uploads/sites/5/2016/05/AJLE … · (HILDA) Survey. Our results show that health shocks