Healh Shocs Dsably and Wor Maaren Lndeboo a Ana Llena-Nozal b Bas van der Klaauw c Noveber 2007 Absrac: Ths paper focuses on he relaon beween he onse of a dsably and eployen oucoes. We develop an even-hsory odel ha ncludes accdens as a easure for healh shocs and we esae he odel usng daa fro he Brsh Naonal Chld Developen Sudy NCDS. We show ha experencng such a healh shoc ncreases he lelhood of an onse of a dsably by around 172%. However accdens are relavely rare evens and herefore he larger par of observed dsably raes resul fro gradual deeroraons n healh. The absence of a drec effec of accdens on eployen oucoes allows nsruenng he onse of dsables by he occurrences of accdens. We fnd ha onse of a dsably a age 25 causally reduces he eployen rae a age 40 wh around 14 percenage pons. The effec s sronger for ales and for low educaed worers. Our resuls ndcae ha abou wo-hrd of he assocaon beween dsably and eployen can be explaned by he causal effec of he onse of a dsably on eployen. Ths fracon s hgher for en and for lower educaed worers. a Free Unversy Aserda Tnbergen Insue HEB IZA & Nespar. b OECD c Free Unversy Aserda Tnbergen Insue & CEPR. Address: Deparen of Econocs Free Unversy Aserda De Boelelaan 1105 1081 HV Aserda he Neherlands. Eal: lndeboo@econ.vu.nl ana.llena-nozal@oecd.org blaauw@econ.vu.nl. We would le o han Ana Rue Cardoso Regna Rphahn Owen O Donell and parcpans of he IZA worshop The Older Worer Lsbon 2005 he HEA world conference Kopenhagen 2007 he European Assocaon for Labour Econoss and senar parcpans a Anwerp Unversy and he Pars School of Econocs for useful coens.
1. Inroducon A subsanal share of he worng age populaon n he ndusralzed world suffers fro a long sandng llness or dsably ha resrcs n daly acves and/or wor Dupre and Karalanen 2002. Dsably prevalence raes are already hgh a relavely young ages. For nsance n he UK around 5% of he 20-24 year old have a long sandng dsably and hs nuber ncreases o around 13% for he 40-44 year old and 28% for hose aged 55-59 Berhould 2006. Slar dsably raes are found for he US Kapeyn e al. 2006. Dsably s assocaed wh hgher benef ae up povery and lower eployen raes. Ths paper focuses on he causal effec of he onse of a long sandng llness or dsably on eployen. Ths ssue s of drec porance for polces ha nend o preven he onse of dsables and o ncrease he labor are prospecs of people wh dsables such as he UK s Dsably Dscrnaon Ac DDA or he Aercans wh Dsably Ac ADA. Delere 2000 Aceoglu and Angrs 2001 and Hochss 2005 sudy he eployen effecs of he ADA and Bell and Heuller 2005 he eployen effecs of DDA. The sze of he causal effec ndcaes he poenal effecs of such progras. Besdes he dsably an ndvdual s eployen saus also srongly depends on oher deographc and socoeconoc facors. Therefore for polcy purposes s poran o now how uch of he dfference n eployen raes of dsabled and non-dsabled worers can be ascrbed o bacground characerscs and how uch o he drec causal effec of dsables. The ssue s also of neres for he large leraure on he assocaon beween socoeconoc saus SES and healh see e.g. Sh 1999 for a survey. Even hough any econoc and epdeologcal sudes have esablshed a srong posve assocaon graden beween healh and SES lle s nown abou causal echanss. Assessng causal relaons wh observaonal daa s non-rval. No only ay here be drec effecs of healh dsably on SES and he oher way around bu also unobserved ndvdual specfc effecs can relae o boh healh and wor. Therefore ndependen varaon n healh s requred o assess s causal effec on wor. 2
We use accdens ha have caused a vs o a hospal an oupaen facly or a casualy deparen as a easure for unancpaed healh shocs. These accdens nclude for exaple raffc accdens wor place accdens hear aacs and spor nures. Such healh shocs conan new nforaon o he ndvdual and hereby provde soe unancpaed varaon. We use hs nforaon o frs assess he causal effec of an accden on he dsably saus and nex use hs as an nsruen o assess he causal effecs of a dsably on eployen oucoes. In our conex no ancpaon eans ha he exac ng of he accden s no nown n advance. Ths does no rule ou ha ndvduals ay be aware ha a soe oens he rs of experencng an accden ay be hgher han n oher perods for nsance because hs rs depends on curren eployen saus. Also we do no requre accdens o be exogenous condonal on observable characerscs. To be a b ore specfc abou he odel we consruc a dscree-e dscree-choce odel for accdens dsably and wor. The ranson raes beween dsably and wor saes can be affeced by accdens and accdens can n urn be nfluenced by he ndvdual s eployen and dsably saus. The hree endogenous varables of he odel are relaed va unobserved coponens ha rean fxed across e. Idenfcaon of hs nd of odels have been dscussed exensvely by Abbrng and Van den Berg 2003 and Hecan and Navarro 2007. Our approach relaes o a growng leraure ha uses naural experens o denfy causal relaons beween socoeconoc saus and healh. For nsance Lndahl 2005 used loery prze wnners and Snyder and Evans 2002 used changes n he socal secury law o assess he causal effec of ncoe on healh. They fnd sall effecs of ncoe on healh. Ths s n agreeen wh Case and Deaon 2005 and Sh 2007 who conclude ha he larger par of he assocaon beween healh and SES a ddle and older ages s drven by an effec of healh on SES raher han he oher way around. Møller-Danø 2004 uses a propensy score and a dfference-n-dfference achng ehod o esae he causal effec of road nures on ncoe and eployen. She fnds shor and long-run effecs of road accdens on eployen saus for en bu no for woen. Lechner and Vasquez-Alvarez 2006 and García Góez and López Ncolás 2006 use achng ehods o denfy he effecs of wor laons on eployen and ncoe. Boh papers fnd sgnfcan nagve effecs of healh on eployen and 3
ncoe. There are soe sudes n he developen leraure ha use socal experens o assess he effec of healh on socoeconoc oucoes. For nsance Mguel and Kreer 2004 used a randozed experen o evaluae a progra of a school-based reaen wh a deworng drug n Kenya. We esae our odel usng daa fro he Brsh Naonal Chld Developen Sudy NCDS. The NCDS s a longudnal sudy of around 17000 ndvduals born n Grea Bran n he wee of 3-9 March 1958. These ndvduals are followed fro brh up o he year 2000 when hey were 42 years old. A age 40 already abou 12% of he respondens face a peranen dsably and abou 29% of hese dsabled are ou of wor. The resuls show ha accdens causally ncrease he probably of he onse of a dsably wh 172%. However because accdens are rare evens he larger par of he onse of dsables coe fro a gradual deeroraon n healh. We fnd ha accdens affec an ndvdual s labor are saus only ndrecly hrough he onse of a dsably. Therefore accdens ac as nsruenal varables for he onse of a dsably. Calculaons wh he odel show ha he causal effec of he onse of a dsably a age 25 on he eployen rae a age 40 s -0.144. We furherore esae he odel for dfferen subgroups and fnd large dfferences beween ales and feales and hgh and low educaed worers. Male eployen raes a age 40 are abou 23 percenage pons reduced due o a dsably for feales hs s 12 percenage pons. Eployen raes a age 40 of low educaed worers O-level or below are reduced wh 21 percenage pons for hgh educaed worers hs s only 9 percenage pons. We show ha n he coplee saple abou wo-hrd of he assocaon beween dsably and eployen can be explaned by he causal effec of he onse of a dsably on eployen. The reanng one-hrd s heerogeney. I should however be noed ha for woen heerogeney s ore poran n explanng he assocaon whle for en and lower educaed worers he assocaon s anly explaned by he causal effec fro dsably o wor. Ths ay ply ha polces ha a drecly a he eployen effecs of he dsably can be effecve n ncreasng eployen raes for hese groups. The srucure of he paper s as follows. Secon 2 dscusses he heorecal bacground and he eprcal odel. Secon 3 nroduces he NCDS daa and repors on 4
he varables used n he eprcal par. Eprcal resuls are dscussed n Secon 4. Secon 5 concludes. 2. Theorecal bacground and he eprcal odel Healh producon odels Grossan 1972 or relaed odels e.g. Cropper 1978 Ehrlch and Chua 1990 Scles and Yazbec 1998 and Case and Deaon 2003 assue ha ndvduals nher an nal soc of healh whch deprecaes wh age and ncreases wh healh nvesens. Indvduals are raonal agens who nclude expecaons abou her healh when ang healh nvesens such as healh care consupon and wor. If healh raecores are predcable ndvduals ancpae o ha and change her behavor accordngly. So an observed change n labor are saus ha precedes a healh ranson can be he resul of ancpaed behavor raher han labor are saus causally affecng healh. An unforeseen shoc conans new nforaon o he ndvdual and hereby provdes soe unancpaed varaon n healh ha s unrelaed o wor saus. Accdens should be consdered as unancpaed healh shocs whch provde new nforaon o he ndvdual. We wll be ore specfc abou he defnon of accdens n he nex secon when we dscuss he daa. An accden ay cause he onse of a peranen dsably of chroncal condon. Here our approach dffers fro for exaple Sh 2003 and Adas e al. 2003 who use he onse of a chronc condon as a easure for healh shocs. Accdens occur a dfferen oens n lfe and herefore our odel should be dynac. A dynac odel also has he advanage ha we can subsanally relax he requreens for accdens o be vald healh shocs. Whn our dynac odel we do no resrc accdens o be exogenous. Insead we explcly odel he occurrence of an accden and allow unobservables o affec only he probably of experencng an accden he onse of dsables and labor are oucoes. Unancpaon of accdens eans ha people canno fully predc he exac ng of he occurrence of an accden. Ths allows denfyng he causal effecs of accdens on he dsably and eployen saus whou excluson resrcons or srong funconal for resrcons. Abbrng and Van den Berg 2003 and Hecan and Navarro 2007 provde exensve dscussons on 5
he denfcaon of causal effecs n even-hsory odels and dscree-e dscree-choce odels respecvely. We reurn o denfcaon ssues afer we have dscussed our eprcal odel. Eprcal specfcaon The daa conan ndvduals who were all born n he wee of 3-9 March 1958. These ndvduals are followed fro brh up o age 42. We have consruced ndvdual labor are hsores fro he oen he ndvdual leaves full-e educaon up o he end of he observaon perod. The labor are hsores conan yearly nforaon on eployen saus and healh saus. Le S l denoe he ndvdual s labor are saus on hs/her h brhday whch can eher be worng 1 or non-worng 0. Snce we only sar followng ndvduals afer leavng full-e educaon non-worng does no nclude full-e educaon. The varable S h denoes he healh saus on he ndvdual s h brhday whch can eher be dsabled 1 or non-dsabled 0. Because we only focus on peranen dsables and hus gnore shor-er laons beng dsabled s an absorbng sae. Ths ples ha S h +=1 f S h =1 for all =12. For each year we observe wheher an ndvdual experenced an accden he varable A equals 1 f an ndvdual experenced an accden a age and 0 oherwse. We observe he sequence S l T 0 +1S h T 0 +1AT 0 +1 S l T 1 S h T 1 AT 1 for an ndvdual who leaves full-e educaon a age T 0 and reans parcpang n he surveys unl age T 1. Each year he ndvdual can ove beween four dfferen saes 1 eployed/non-dsabled 2 non-eployed/non-dsabled 3 eployed/dsabled and 4 non-eployed/dsabled. To analyze he ransons beween hese saes we use a dscreee dscree-choce odel where ranson probables depend on curren and pas accdens. Accdens do no occur exogenously for exaple hose a wor ay be exposed o adverse worng condons whch ay ncrease he rs of havng an accden. We allow he probably of he occurrence of an accden n e nerval < +1] o depend on he ndvdual s labor are and dsably saus a e and also on observed 6
7 characerscs a e x and a e-nvaran unobserved coponen v a. We use a log specfcaon o odel he probably of an accden a age : exp 1 exp 1 Pr a h h l l a h h l l a h h l l a h l v s s x v s s x v x s S s S A v x s s q + + + + + + + = = = = = δ δ γ δ δ γ 1 The paraeers δ l and δ h descrbe he effecs of beng eployed and beng dsabled on he ncdence rae of accdens respecvely. The vecor of observed ndvdual characerscs x ncludes an nercep a se of socoeconocs characerscs and a fourh order polynoal n age. Recall ha because all ndvduals n he daa are born n he sae wee he polynoal n age also ncludes calendar e effecs such as busness cycle flucuaons. Accdens can affec he probably ha ndvduals ove beween dfferen dsably and eployen saes n wo ways. Frs here s an nsananeous effec whch we allow for by ncludng he ndcaor funcon A as one of he regressors n he ranson odel. Second we allow accdens o have a peranen long-run effec. Once an ndvdual has had an accden hs gh affec hs/her ranson probables n all followng years. Therefore we defne he ndcaor funcon Ā whch aes on he value 1 f he ndvdual experenced an accden pror o age and zero oherwse. 1 The ranson probables beween he dfferen dsably and wor saes are gven by: 1 1 Pr h l h l v x a A a A S S S S v x a a p = = = = = + = + =
8 whch we paraeerze agan as a log funcon + + + + + + + + = ' ' ' ' ' ' ' ' exp 1 exp v a a x v a a x v x a a p µ η β µ η β 2a f and + + + + = ' ' ' ' ' ' ' ' ' ' exp 1 1 v a a x v x a a p µ η β 2b Snce dsably s an absorbng sae hs ranson probably equals 0 f s dsabled and s non-dsabled. The paraeers η descrbe he nsananeous effecs of an accden on he dfferen ranson probables whle he paraeers µ descrbe he peranen effecs of ever havng experenced an accden. The pac of an accden can hus be dfferen for ndvduals n dfferen eployen and dsably saes. Accdens are no only relaed o he ranson probables hrough hese effecs bu we also allow for nerdependence hrough he e-nvaran unobserved heerogeney coponens. The unobserved coponen n he accden probably v a ay be relaed o he unobserved coponens n he ransons probables v. We use a rando effecs specfcaon for he unobserved heerogeney and n parcular a so-called facorloadng specfcaon o allow for correlaons beween he dfferen unobserved 1 In our odel hs ndcaor reans 1 f he ndvdual ges addonal accdens.
heerogeney ers. When esang he odel we ae wo rando facors w 1 w 2 whch boh have wo dscree ass pons a 0 and 1. The paraeers θ 1 and θ 2 denoe he probably ha he varables w 1 and w 2 ae he value 1 respecvely. The unobserved heerogeney ers follow v a = α + 3a a 1w1 α a2 w2 and v α 1w1 + α 2w2 = 3b The facor-loadng specfcaon ples ha he er v a equals 0 wh probably 1- θ 1 1-θ 2 v a =α a1 wh probably θ 1 1-θ 2 v a =α a2 wh probably 1-θ 1 θ 2 and v a =α a1 +α a2 wh probably θ 1 θ 2. The ers v are defned n a slar way. The unobserved heerogeney could for exaple pc up he dfference beween whe and blue-collar worers where he blue-collar worers gh have hgher probables of beng dsabled and non-eployed and a hgher ncdence rae of healh shocs. The paraeers α a and α deerne he sgn and he agnude of he correlaon beween he accden rae and he ranson probables. These paraeers are esaed along wh he oher odel paraeers as well as he paraeers θ 1 and θ 2. Idenfcaon of he odel The an paraeers of neres are η and µ descrbng he nsananeous and peranen effecs of an accden on dsably and wor oucoes. Abbrng and Van den Berg 2003 dscuss he denfcaon of such paraeers n even-hsory odels. Idenfcaon hnges on he assupon ha ndvduals do no ancpae he exac oen of arrval of he accden. No ancpaon n our conex eans ha ranson probables pror o he occurrence of an accden are no affeced by he fuure accden 9
and also no by he oen a whch hs accden wll occur. For exaple condonal on he observables and unobservables he ranson probables a age 25 are he sae f hs ndvdual s h by an accden a age 26 or a age 40. The accden does no have any pac on ranson probables pror o he oen a whch he ndvdual s acually h by he accden. Ths does no ply ha accdens are exogenous or ha each ndvdual has n each e perod he sae probably of experencng an accden. I raher ples ha ndvduals canno foresee he exac oen a whch an accden occurs o he. So ndvduals gh now ha n parcular perods he probably of geng an accden s hgh for exaple when hey are eployed or as hey ge older. Furherore he probably ha an accden occurs can dffer beween ndvduals based on boh observed and unobserved characerscs. The nuon behnd he denfcaon coes fro splng he daa n wo pars ranson before ndvduals are h by an accden and ransons afer ndvduals have been h by an accden. The assupon of no ancpaon ples ha we can denfy all odel paraeers excep η and µ fro daa unl he oen ndvduals are h by he accden so par of he daa. Par of he daa should hus be used o denfy he paraeers η and µ. Indeed denfcaon of η and µ coes fro coparng ranson probables before and afer he occurrence of he accden condonal on observables and unobservables whch have been denfed by par of he daa. Whereas Abbrng and Van den Berg 2003 have a connuous-e fraewor Hecan and Navarro 2006 dscuss a dscree-e analogue such as our odel. They acheve se-paraerc denfcaon for dynac dscree-choce odels under less srong assupons han Abbrng and Van den Berg 2003. An accden s defned n our daa by a queson wheher an ndvdual had been aded o a hospal or aended a hospal oupaen or casualy deparen as a resul of any nd of accden. I s clear ha hs easure sasfes he condon ha he even can no be ancpaed. I s poran o noe here ha he UK has a Naonal Healh Care syse and ha n prncple healh care s accessble for all. 2 Of course reans an 2 I s possble o buy suppleenal prvae healh nsurance. Ths prvae nsurance allows ndvduals o have ore freedo n he choce of provder o have a prvae roo when hospalzed and o have shorer wang es for specalzed care. Only abou 15% of he ndvduals have prvae nsurance. 10
ndvdual decson wheher or no o go o hospal afer an accden. For exaple an ndvdual s socoeconoc saus ay affec he probably ha he ndvdual vss he hospal afer havng experenced an accden. Such dfferences are conrolled for by he observed characerscs x and unobserved characerscs v a n he odel for accdens see equaon 1. Models equvalen o our odel have been used o esae he causal effec of acve labor are polces on he reeployen of uneployed worers. Bonnal e al. 1997 esae he causal effec of ranng progras for uneployed worers and Van den Berg e al. 2004 evaluae posng punve eporary benefs reducons on welfare recpens. The denfcaon resuls apply n our odel o he effecs of accdens on he dsably and eployen saus. Our ulae neres s n he causal effec of he onse of a dsably on eployen. Our eprcal resuls presened laer show ha accdens do no have a drec effec on eployen f he ndvdual s healh saus reans unchanged. The accden hus causes soe varaon n he dsably saus whch n urn could be used o denfy he causal effec of he onse of a dsably on eployen saus. Ths s slar o an nsruenal varable seng. We explo hs fndng n our odel sulaons o esae he causal effec of he onse of a dsably on laer eployen oucoes. Esaon Snce he odel s fully paraeerzed we can use axu lelhood o esae he paraeers. For each ndvdual we observe he sequence of realzaons s l T 0 +1 s h T 0 +1 at 0 +1 s l T 1 s h T 1 at 1. In he esaon we condon on he nal labor are saus s l T 0 +1 and healh saus s h T 0 +1 of he ndvdual.e. he oen he ndvdual leaves full-e educaon. The conrbuon o he lelhood funcon of for hs ndvdual s gven by: 11
1 w1 w2 = 0 1 q s s p l x α w + α h a1 1 T 1 1 w1 1 w w 1 2 1 w2 l = θ1 1 θ1 θ 2 1 θ 2 q sl sh x α a1w1 + α a a x α = T0 + 1 1 a a2w2 sl sh sl + 1 sh + 1 sl sh sl + 1 sh + 11 1 sl sh sl + 1 sh + 12 w + α a2 w 2 a w 2 The oal lelhood funcon s he su of he logarh of hese ndvdual lelhood conrbuons. 3. Daa Saple We use he Naonal Chld Developen Sudy NCDS a longudnal sudy ha follows abou 17000 ndvduals born n Grea Bran n he wee of 3-9 March 1958. The sudy sared as he Pernaal Moraly Survey and surveyed he econoc and obserc facors assocaed wh sllbrh and nfan oraly. Snce he frs survey n 1958 cohor ebers have been raced on sx oher occasons o onor her physcal educaonal and socal crcusances. The waves were carred ou n 1965 age 7 1969 age 11 1974 age 16 1981 age 23 1991 age 33 and 1999/2000 age 42. In addon o he an surveys nforaon abou he publc exanaons was obaned fro he schools n 1978. For he brh survey nforaon was gahered fro he oher and he edcal records. The nervews for wave 1 o 3 coverng age 7 11 and 16 were carred ou wh parens eachers and he school healh servce; whle ably ess were adnsered o he cohor ebers. The subsequen surveys ncluded nforaon on eployen and ncoe healh and healh behavor czenshp and values relaonshps parenng and housng educaon and ranng of he respondens. In waves 4 5 and 6 ndvduals were ased o rerospecvely gve nforaon on her eployen uneployen ou-of-he-labor-force and educaon/ranng perods recordng her sarng and endng daes. The NCDS s herefore hghly approprae o loo a lfe hsores. 12
Our odel descrbes ransons beween dsably and labor are saes fro he oen ha ndvduals leave full-e educaon. We herefore use a saple of 12375 ndvduals who parcpaed n he 1981-survey a age 23 and for who we can consruc labor are healh and accden hsores. 3 Case e al. 2005 nvesgaed aron fro he survey by coparng low brh wegh and faher s occupaon across dfferen NCDS waves. They dd no fnd any evdence for non-rando aron wh respec o hese varables. Furherore advsory and user suppor groups of he NCDS copared respondens and non-respondens n he laer surveys n ers of socal and econoc saus educaon healh housng and deographcs. I was found ha he dsrbuon of hese varables aong he saple survvors dd no dffer fro he orgnal saple o any grea exen NCDS User Suppor 1991. In addon he 1981 saple was copared o he UK 1981 Populaon Censuses n ers of he dsrbuons of ey varables such as aral saus gender econoc acvy gross weely pay enure and ehncy Ades 1983. The overall concluson was ha he saple appears o be represenave wh respec o hese varables. Dependen varables: labor are saus dsably saus and accdens The labor are saus s easured each year n March around he e of he brhday of he ndvduals. We dsngush wo labor are oucoes eployed and noneployed. An ndvdual s consdered o be eployed f eher he ndvdual has a fulle or par-e ob s self-eployed or on aerny leave. Also an apprenceshp schee whch s par of a ob s consdered as eployen. Curre and Hyson 1999 who use he sae daa o exane he long-run effec of low brh wegh on educaonal aanen and labor are oucoes fnd ha her eprcal resuls are no sensve o he exac defnon of eployen. In Fgures 1 and 2 we show for ales and feales a dfferen ages he eployen rae he uneployen rae and he fracon of ndvduals ou of he labor force and n full-e educaon. The fgures show ha eployen raes are hgher for ales ha a subsanal share of he feales s ou of he labor force and 3 60% of he ndvduals n our saple are presen n wave 4 age 23 5 age 33 and 6 age 42 28% only n wave 4 and 12% n waves 4 and 5. 13
ha hs fracon declnes wh age. For ales he fracon of worers ou of he labor force ncreases gradually wh age. The uneployen raes are hgher beween he ages of 22-25 hese ages concde wh a perod when he UK experenced a severe recesson. We defne a dsably/ longsandng llness fro he response o he followng queson: Do you have any longsandng llness dsably or nfry of any nd? The respondens are ased o repor wheher hey had longsandng llnesses dsables and nfres for every year a around he e of her brhday. And f so fro whch condons he responden suffered. We only consder condons ha are ncluded n he Inernaonal Classfcaon of Dseases ICD-9 produced by he World Healh Organzaon 1977. The ICD-9 s exensvely used n epdeologcal and healh anageen sudes o classfy dseases and healh probles World Healh Organzaon 2004. These nclude for nsance serous dsables ha l wor/daly acves such as eplepsy blndness deafness ulple scleross enal reardaon a congenal condon or a rauac apuaon or nernal nury see he Appendx for he coplee ls. We only consder an ndvdual as dsabled f he ndvdual repors havng he sae longsandng llness dsably or nfry for a leas hree consecuve years. 4 5 Case e al. 2005 use responses o he queson How good s your healh and fnd ha hese self repors are srongly correlaed o he queson on he prevalence of long sandng llnesses dsables and nfres. Baeal e al. 2004 show n a repor cossoned by he UK Deparen for Wor and Pensons ha age-specfc dsably 4 We added hs condon o be sure ha we have long lasng parens. For nsance f an ndvdual repors o have coplcaons of pregnancy code 11 n he ICD-9 classfcaon durng one year hen he ndvdual s no couned as peranenly dsabled. Dsables lsed n he ICD-9 code lasng ore han 3 years are consdered o be peranen. The daa confr ha recovery raes afer 3 years are very low. 5 We follow our respondens up o age 42. The exra condon ha he paren should las a leas hree years ples herefore ha n he esaon of he odel we only use he dsably and wor hsory nforaon up o he age of 40. 14
raes for eployed worers do no vary uch across surveys usng dfferen defnons for dsables. Fgure 3 shows he fracon of ndvduals wh a dsably afer age 16. Dsably raes are very slar for en and woen. A age 20 over 4% of he ndvduals n he saple has soe dsably. Ths ncreases up o abou 12% a age 40. Soe people already have long sandng dsables ha sared durng chldhood bu he aory of he dsables sared durng worng ages. In fac he slope becoes soewha seeper a older ages whch eans ha he hazard of he onse of a dsably becoes larger as people ge older. We derve our easure for an accden fro he followng queson: Have you been aded o hospal or aended a hospal oupaen or casualy deparen as a resul of any nd of accden? We observe boh he dae of he accden and he cause. 6 Men are uch ore lely o experence an accden han woen. In our saple around 77% of he en had a leas one accden durng he observaon perod whle hs was only abou 42% for woen. We follow ndvduals for ore han wo decades and herefore que a few respondens have experenced ore han one accden durng hs perod. No only he ncdence of accdens dffers beween en and woen bu also he cause dffers. Table 1 lss accdens by cause and gender. For each cause en are uch ore lely o experence an accden han woen. The os subsanal dfference n ncdence raes occurs for wor and spors-relaed accdens. Because of hs large share of wor-relaed accdens s parcularly poran n our eprcal analyses o allow he accden probably o depend on labor are saus. Fgure 4 plos annual ncdence raes by age and shows 6 The quesonnare resrcs he nuber of accdens ha can be repored o 8 n he 1981-wave and 6 n he 1991 and 1999/2000-wave. In each wave only beween 1 and 2 percen of he ndvduals acually repors hs axu. 15
ha for boh en and woen he probably of geng an accden s relavely hgh unl he d-wenes and drops subsanally aferwards. We use he annual labor are saus and dsably saus o classfy each ndvdual n each year n one of four saes: 1 wor and dsabled 2 non-wor and dsabled 3 wor and non-dsabled and 4 non-wor and non-dsabled. In Fgure 5 we show for dfferen ages he fracon of ndvduals n each sae. A all ages os ndvduals are eployed and non-dsabled. A laer ages he fracon of ndvduals n he non-wor non-dsabled sae decreases whle he fracons of ndvduals ncrease n boh dsabled saes eher wh or whou wor. The fgure shows a fall n eployen rae a ages 22-25 hs s n lne wh he observed paerns n Fgure 1 and 2. Our eprcal odel s specfed n ers of yearly ranson probables beween he four saes. Table 2 provdes for boh en and woen a suary of he yearly ransons. The able shows ha here s a hgh degree of sae dependence and ha ndvduals are uch ore lely o change labor are saus han dsably saus. The able also shows ha 16.8% of he en wh a dsably fnd a ob n he subsequen perod. Ths eans ha here rean eployen opporunes for people wh a dsably. However noe ha hs nuber should be conrased wh he 41.9% of non-dsabled worers who fnd a ob n he nex perod. For feales he dfferences beween he ob-fndng rae of dsabled and non-dsabled worers s uch saller 12.8% versus 19.3%. Bacground varables The NCDS s very rch on ndvdual characerscs. For each ndvdual we observe a range of varables ha gve nforaon on an ndvdual s nal healh socoeconoc saus and cognve ably durng chldhood. In consrucng he relevan bacground varables we follow he defnons used by Case e al. 2005 and Curre and Hyson 1999. Table 3 provdes saple eans on hese varables. For any varables here s soe e non-response. To avod losng any observaons we follow Case e al. 2005 by consrucng duy varables ha ndcae f he nforaon on a varable s ssng. So f an ndvdual s characerscs are ssng hen he varables descrbng hese 16
characerscs ae value 0 whle he duy varable ndcang ha hese varables are ssng aes value 1. For each ndvdual we observe he educaonal aanen. The level of educaon s derved by coplng an educaon varable wh caegores aggregaed o naonal vocaonal qualfcaon levels. We nclude he followng caegores: less han O-level O- level equvalen A-level equvalen and degree equvalen. The faly s socoeconoc saus s derved fro he faher s socal class a brh. The socal class corresponds o a syse used by he Brsh Regsrar General and consss of: professonal supervsory slled non-anual slled anual se-slled non-anual se-slled anual and unslled. We classfy socoeconoc saus as hgh f he faher s n a professonal supervsory slled non-anual ob; edu f he faher s n slled anual se-slled non-anual; and low f he faher s n a seslled anual and unslled ob. Followng Curre and Thoas 1999 we classfy ndvduals whose faher s nforaon s ssng by he oher s socal class. In case he socal classes of boh parens are ssng we assgn he ndvdual o low socoeconoc saus f he oher was sngle and o ssng f boh parens were presen. Low brh wegh s a duy varable for nfans wh a brh wegh below 2500 gras. There s evdence fro he epdeologcal leraure ha low brh wegh s srongly assocaed wh nfan and laer lfe oraly World Healh Organzaon 2004. Low wegh a brh can be he resul of eher preer brh before 37 wees of gesaon or resrced feal growh. In he eprcal analyses we do no ae a dsncon beween hese wo caegores. We creae a duy varable ha ndcaes f he oher soed afer he fourh onh of pregnancy. Song durng pregnancy has been found o be relaed wh cognve defcences and oher healh probles n he edcal and epdeologcal leraure see for nsance Blar e al. 1995; Coner e al. 1995; Naeye and Peers 1984; Wllas e al. 1998. Furherore we observe he oher s age a brh. Moher s age a he chld s brh can nfluence he chld s healh hrough for nsance nuronal defcences f he oher s very young or delvery coplcaons f he oher s older. Therefore n he eprcal analyses we nclude also 17
he oher s age a brh squared. We use he regon a brh o conrol for geographcal dfferences and/or dfferences n labor are condons. For each ndvdual we observe es scores on ah and socal adusen a age 7. The ah es s desgned for he NCDS and assesses arhec ably. The score ranges fro 0 o 10. Curre and Thoas 1999 show ha es scores a he age of 7 have sgnfcan pacs on laer educaon aanens and labor are oucoes. The Brsol Socal Adusen Gude s desgned o assess he chld s socal behavor n school and a hoe. The es s copleed by he eacher who nows he chld bes. Hgher scores ndcae hgher aladusen. The daa also nclude nforaon on he Souhgae Readng Tes. However ncludng hs es score n our analyses dd no prove our eprcal resuls afer ncluson of he ah score and Brsol Socal Adusen Gude. Therefore we decded no o use hs readng es score n he analyses. 4. Resuls Table 4a and 4b show he paraeer esaes of our odel. In Fgure 6 we show soe ndcaon for he f of our odel. In hs fgure we copare he observed daa o odel predcons. 7 The odel predcons anage o pc up he general lfe cycle rends n eployen and dsably. Only around age 23/24 he odel has dffcules o ach he flucuaons n observed eployen raes. We sar wh a bref dscusson of he resuls for he xng dsrbuon unobserved heerogeney. The paraeers of hs dsrbuon are repored n he lower panels of Tables 4a and 4b. Alos all paraeers of he xng dsrbuon are hghly sgnfcan ndcang he porance o allow for a sochasc relaonshp beween dsably wor and accdens. Recall fro equaons 3a and 3b ha we pose a facor-loadng specfcaon wh wo eleens ha each can ae on wo values. For each ranson we hus have four ass pons. The frs ass pon s noralzed o 0 because each odel 7 Our odel descrbes ransons beween healh and labor are saes afer havng lef full-e educaon. We have esaed a ulnoal log odel for he frs sae afer leavng full-e educaon. 18
equaon ncludes an nercep. Mos probably ass abou 42% s locaed a ass pon 2 descrbng ndvduals wh a low probably of experencng accdens. Mass pon 3 has less probably ass abou 10% and descrbes ndvduals who are os lely o ge an accden. I s no sraghforward o deec a clear paern n he effecs of he unobservables n he ranson probables. Bu copared o ass pon 1 all oher ass pons descrbe ndvduals who are less lely o ge dsabled and who are ore lely o say n eployen or ove o eployen f he healh saus reans unchanged. Table 4a shows he paraeer esaes fro he log specfcaon for he probably of geng an accden. Eployed ndvduals have abou a 41% =exp0.342-1 hgher probably of geng an accden. Recall fro Table 1 ha ndeed a subsanal share of he accdens s wor-relaed. The accden rae of an ndvdual wh a dsably s only abou 5% hgher han he accden rae for a non-dsabled person. Males are ore han hree es ore lely o ge an accden han feales. Ths s also wha we drecly observe n he daa when coparng ales and feales see Table 1. Apparenly dfferences n bacground characerscs beween ales and feales can no explan he dfferences n accdens raes. The age coeffcens ndcae ha he probably of geng an accden decreases up o age 37 and ncreases aferwards. The socoeconoc characerscs of he parens of he cohor eber all have sgnfcan pacs on he accden rae. Indvduals wh edu socoeconoc saus and whose oher soed durng pregnancy have hgher accden raes. The accden rae also ncreases wh he age of he responden s oher a brh. Indvdual characerscs are also poran. We fnd sgnfcan effecs for brh wegh regon and he es scores and educaon. Mos of hese effecs are n lne wh a pror expecaons. We fnd a soewha puzzlng effec of he ah es score a age 7. Indvduals wh a hgher es score have hgher accden raes. I should be noed ha he coeffcens purely reflec assocaons and ha one should no connec a srong causal nerpreaon o hese fndngs. Table 4b shows he paraeer esaes of a ulnoal log odel for he ransons beween dfferen labor are and dsably saes. The an paraeers of neres are he paraeers descrbng he effecs of accdens on ranson probables. 19
We allow accdens o have a drec nsananeous effec and a long-run effec. I can be seen fro he able ha he nsananeous effecs are alos always larger han he longrun effec. Furherore he nsananeous effecs of he accden are larges on he ranson raes fro he non-dsabled saes o dsabled saes. The esaed coeffcens for ransons beween eployen saes where he dsably saus reans he sae are uch saller. The long-run effecs anly show ha n he long-run accdens no only have an effec on he dsably saus bu also on he eployen saus. Inerpreng he coeffcens separaely s dffcul. To llusrae he pac of an accden we herefore consder a represenave ndvdual and perfor calculaons wh he esaed odel. 8 Suppose hs ndvdual dd no experence an accden unl age 23. The probably ha hs ndvdual s non-dsabled a hs/her 24 h brhday s 0.933 wh sandard error 0.00034 copued usng he Dela ehod. Whou experencng an accden a age 24 he probably of becong dsabled before hs/her 25 h brhday s 0.0025 s.e. 0.000062. However f he ndvdual dd experence an accden a age 24 hs probably becoes 0.0069 s.e. 0.00015. So he accden ncreases he nsananeous onse of a dsably wh around 172% s.e. 20%. The -es for sgnfcance of he pac of an accden on he onse of a dsably s hus 8.58 whch ples ha an accden no only has a very subsanal effec bu he effec s also srongly sgnfcan. No only has an accden an nsananeous effec on dsably raes bu he effec s also long lasng. More specfcally f hs ndvdual does no ge any oher accden hen wh an accden a age 24 he dsably rae a age 40 s 0.141 s.e. 0.0022. The dsably rae a age 40 equals 0.128 s.e. 0.0018 f he ndvdual never had an accden. We can conclude fro hs ha accdens are poran for he onse of dsables and ha here are boh shor-run and long-run effecs. The occurrences of accdens are relavely rare evens. Unl age 40 en experence on average abou 2.4 accdens and woen only 0.8. Model calculaons show ha accdens can only accoun for abou 12% of all dsables a age 40. Hence he larger par of long sandng dsables arses fro a gradual deeroraon n healh. Alhough we should be soewha cauous our defnon of an accden ncludes ha 8 In fac we sulae he odel for all ndvduals and copue averages over all ndvduals. 20
should lead o a hospal oupaen or casualy deparen vs. Healh evens ha dd no drecly lead o hospal vss are hus ncluded n he gradual deeroraon of healh. Ths also holds for healh evens ha ay have caused hospal care n he longer run. The dsncon beween gradual and sudden changes of healh herefore srongly depends on he defnon of he healh shoc varable. Of pre porance for hs paper s he causal effec of he onse of a dsably on eployen and for hs we requre rando varaon n he ng of accdens. Wh hs src defnon of an accden we ensure ha hs condon s sasfed. The nsananeous effec of an accden on eployen raes s neglgble. Consder agan a non-dsabled 24 year old ndvdual. We copare he suaon where hs ndvdual does no experence an accden wh he suaon where s/he does experence an accden. To ge he nsananeous effec of an accden on eployen we assue ha he accden does no lead o a dsably.e. a hs/her 25 h brhday he ndvdual s sll non-dsabled. In hs case he eployen rae a age 25 s 0.805 s.e 0.0018 wh he accden copared o 0.804 s.e. 0.0018 whou an accden. Ths dfference s hus very sall and no sgnfcan. There are herefore wo poenal ways n whch an accden can affec eployen n he long-run. Frs he odel allows for peranen effecs and second he effec can run va he onse of a dsably. We fnd ha long-run drec effecs of accdens on eployen raes are also sall. An ndvdual who receves an accden a age 24 and who does no becoe dsabled n laer years has only a 1.1 percenage pon lower eployen rae a age 40 he eployen raes a age 40 falls fro 0.890 s.e 0.0020 o 0.879 s.e. 0.0024. Now suppose ha hs accden acually leads o a dsably a age 25 hen he eployen rae a age 40 s 9 percenage pon lower he eployen rae falls fro 0.890 s.e. 0.0020 o 0.799 s.e. 0.0078. So fro he above we can conclude ha an accden causally affecs dsably raes boh n he shor and long-run. Accdens have no nsananeous and sall long-run effecs on eployen f ndvdual reans n good healh afer accden. We explo hs o esae he causal effec of he onse of a dsably on eployen. In parcular we consruc a sulaon n whch he nsananeous effec of an accden acs as 21
nsruenal varable for he onse of a dsably. Ths ples ha we urn off he peranen effec of accdens whch corresponds for exaple o he occurrence of a second accden o an ndvdual. Recall ha accdens have a subsanal effec on he onse of a dsbly and ha hs effec s srongly sgnfcan. Agan we focus on accdens a age 24. Now wh he accden as an nsruen we can copue he Wald esaor for he causal effec of he onse of a dsably a age 25 on he eployen rae a age 40 Pr S140 = 1 A24 = 1 A24 = 1 Pr S140 = 1 A24 = 0 A24 = 1 Pr S 25 = 1 A24 = 1 A24 = 1 Pr S 25 = 1 A24 = 0 A24 = 1 h h = 0.144 s.e. 0.048 The onse of a dsably a age 25 causes a sgnfcan reducon n he probably of beng eployed a age 40 by 14.4 percenage pons. Ths esaor explos he esaed causal effecs of he accdens on he dfferen ranson probables. However provdes an esaor for ndvduals a he argn where an accden rggers he onse of a dsably. We can copare hs esaor o an alernave esaor whch s derved fro ang he dfference n eployen raes beween hose who becoe dsabled a age 25 and who do no. The reducon n eployen raes a age 40 due o he onse of a dsably a age 25 s 0.205 0.00478. 9 Ths esaor s subsanally larger and uch ore precse han he Wald esaor above. Recall ha he raw daa do no show any dfferences n dsably raes beween ales and feales bu ales have uch hgher accden rss. Also he ypes of accdens dffer beween ales and feales. Ths gh also ply ha accdens have dfferen effecs for ales and feales. Therefore we esae separae odels for ales and feales. We also esae separae odels for hgh educaed A-level and Degree equvalen and low educaed O-level and below. Hgh educaed and low educaed ndvduals have dfferen ypes of obs and ay wor n dfferen secors and herefore he causal effecs of he onse of a dsably on eployen ay dffer. We do no show 9 So we use PrS l 40=1 S h 25=1-PrS l 40=1 S h 25=0. 22
he paraeer esaes for hese odels bu only show he resuls fro sulaon experens n Table 5. Fro Table 5 we can see ha here s no dfference n dsably raes a age 40 beween en and woen colun hree bu ha woen have uch lower eployen raes colun wo. Unl age 40 en experence on average 3 es as any accdens as woen colun four and for en an accden s ore lely o cause a dsably colun fve. Ths ples ha for en a uch larger share of he dsables s explaned by accdens. Low educaed ndvduals have lower eployen raes and hgher dsably raes han hgh educaed ndvduals. There s no subsanal dfference n he accden rae of hgh and low educaed worers bu for low educaed an accden s ore lely o rgger he onse of a dsably. Table 5 also shows ha he causal effec of a dsably on eployen dffers wh respec o gender and educaon colun sx. The onse of a dsably has a large and sgnfcan causal effec on eployen raes of ale and lower educaed worers. For feales and hgher educaed worers saller and nsgnfcan effecs are found. The relavely sall and nsgnfcan effec for feales ay be explaned fro dfferences n he ype of obs ha en and woen are holdng. In our analyses we do no ae a dsncon beween par-e and full-e wor. Feales are ore ofen eployed n par-e obs and ay be easer o connue worng n hese obs afer he onse of a dsably. Also feales have lower parcpaon raes and he argnal worng feale ay be dfferen fro her ale counerpar. The dfference beween low and hgh educaed ndvduals ay agan be due o he dfferen ype of obs and characerscs of he worng envronen. For nsance connung worng wh a chronc condon ay be easer for a hgh educaed worer n he servce secor han a low educaed consrucon worer. Colun seven of Table 5 repors he dfference n eployen raes beween dfferen groups of worers. Ths dfference s derved fro a drec coparson of he eployen rae beween dsabled and non-dsabled worers and hus consss of a 23
causal pac of he dsably and dfferences n bacground characerscs. 10 So hs s eployen gap erely reflecs an assocaon beween dsably and eployen oucoes. For he full saple 14.4 percenage pons of he 22.9 percenage pon dfference n eployen raes beween dsabled and non-dsabled worers s due o a causal effec. The reander s due o dfferen observed and unobserved bacground characerscs of dsabled and able boded worers. For ales vrually all of he eployen gap beween non-dsabled and dsabled worers can be explaned by he causal effec whereas for feales he dfference n eployen rae s prarly due o dfferences n bacground characerscs. Ths ples ha for ales polces drecly aed a dsabled worers such as he UK DDA or he US ADA ay be ore effecve han for feales. For feales oher facors are largely explanng he assocaon and hence polces aed a dsabled worers le for nsance worplace accoodaons ay no be effecve n ncreasng eployen raes. We fnd slar resuls for he dfferen educaon groups. For lower educaed worers he eployen dfference beween non-dsabled and dsabled worers s subsanal abou 27 percenage pons and he larger par of hs eployen dfference can be explaned by he causal effec of dsably on wor. Ths fndng corresponds o Case and Deaon 2005 who argue ha healh relaed absences fro wor are poran for he dfferences n socoeconoc oucoes le ncoe and wor and ha hs effec s sronger for worers n anual occupaons. Indeed we fnd ha for hgher educaed worers he eployen gap s uch saller only abou 11 percenage pons. The larger par of hs eployen gap can be explaned by he causal effec bu has o be noed ha he causal effec s no sgnfcanly dfferen fro zero. 5. Conclusons In hs paper we have focused on he relaon beween dsably and eployen oucoes and parcularly he causal effec of dsably on eployen. We have 10 More precse we sulae our odel for all worers n our saple and we copue he eployen gap a age 40. Ths assocaon s hus he consequence of boh he causal effec of a dsably and dfferences n observed and unobserved bacground characerscs. 24
developed an even-hsory odel ha descrbes ransons beween dsably and wor saes and we allowed he ransons raes o be affeced by healh shocs. These healh shocs are endogenous n he sense ha her arrval depends on he sae observed and unobserved characerscs ha affec he ranson raes beween healh and eployen saes. As easure of he healh shocs we have used accdens ha caused adance o a hospal hospal oupaen or casualy deparen. We have exploed he unancpaed naure of hese accdens o denfy he causal effec of such shocs on he onse of a dsably and subsequenly he causal effec of he onse of a dsably on laer eployen. The eprcal resuls show ha accdens subsanally ncrease he probably of he onse of a dsably bu accdens have no drec effec on eployen f he healh saus reans unchanged. Even hough an accden has a large effec on he onse of a dsably he fac ha accdens are rare evens causes ha only abou 12% of all dsables a age 40 can be explaned fro he occurrences of accdens. The causal effec of he onse of a dsably a age 25 on eployen raes a age 40 s n he coplee populaon -0.144. When esang separae odels for en and woen and hgh and low educaed ndvduals we found ha he onse of a dsably has a uch larger causal pac on he eployen raes of en and low educaed han her counerpars. Fnally we have copared he causal effec of he onse of a dsably on eployen o he assocaon beween dsably and eployen. The laer s he gap n eployen beween non-dsabled and dsabled worers whch ncludes no only he causal effec of dsably on eployen bu also dfferences n observed and unobserved ndvdual characerscs. Our analyses show ha whn he coplee saple abou wo-hrd of he assocaon beween dsables and eployen can be explaned fro a causal effec of dsably on eployen. Ths percenage s uch lower for woen plyng ha for woen observed and unobserved heerogeney s he prary cause for he assocaon beween dsably and eployen. For ales and lower educaed worers he larger par of he eployen gap can be explaned by he causal effec of he dsably on eployen. Ths ay ply ha for hese groups polces ha 25
a drecly a geng dsabled worers bac o wor can be effecve. 26
References Abbrng J.H. and G.J. van den Berg 2003 The nonparaerc denfcaon of reaen effecs n duraon odels. Econoerca 71 1491-1517. Aceoglu D. and J. Angrs 2001 Consequences of eployen proecon? The case of he Aercans wh Dsables Ac. Journal of Polcal Econoy 109 915-957. Adas P. M.D. Hurd D. McFadden A. Merrll and T. Rbero 2003 Healhy wealhy and wse? Tess for drec causal pahs beween healh and socoeconoc saus. Journal of Econoercs 112 3-56. Ades T. 1983 Coparng NCDS4 o he 1981 UK Census. Naonal Chldren s Bureau Naonal Chld Developen Sudes Worng Paper Fourh Follow-up: No. 11. Baeal M. T. Harres R. Brean and K. Woodfeld 2004 Revew of dsably esaes and defnons. Deparen for Wor and Pensons IAD Socal Research Dvson In-house Repors 128. Bell D. and A. Heuller 2005 The dsably dscrnaon ac n he UK: helpng or hnderng eployen aongs he dsabled?. IZA worng paper 1476. Berhould R. 2006 The eployen raes of dsabled people. Deparen for Wor and Pensons Research Repor No.298. Blar P.S. P.J. Fleng D. Bensley I. Sh C. Bacon E. Taylor J. Glodng and J. Trpp 1996 Song and he sudden nfan deah syndroe: resuls fro 1993-5 case-conrol sudy for confdenal nqury no sllbrhs and deahs n nfancy. Brsh Medcal Journal 313 195-198. Bonnal L. D. Fougère and A. Sérandon 1997 Evaluang he pac of French eployen polces on ndvdual labour are hsores. Revew of Econoc Sudes 64 683-713. 27
Case A. and A. Deaon 2005 Broen down by wor and sex: how our healh declnes. In D.A. Wse ed. Analyses n he Econocs of Agng Chcago Unversy Press. Case A. A. Ferg and C. Paxson 2005 The lasng pac of chldhood healh and crcusance. Journal of Healh Econocs 24 365-389. Coner V. I. Cornovs P. Rogar and L. Rva 1995 Wegh growh n nfans born o ohers who soed durng pregnancy. Brsh Medcal Journal 310 768-771. Cropper M.L. 1977 Healh nvesen n healh and occupaonal choce. Journal of Polcal Econoy 85 1273-1294. Curre J. and R. Hyson 1999 Is he pac of healh shocs cushoned by socoeconoc saus? The case of low brh wegh. Aercan Econoc Revew Papers and Proceedngs 89 245-250. Curre J. and B.C. Madran 1999 Healh nsurance and he labour are. In Ashenfeler O Card D ed. Handboo of Labor Econocs 3C. Aserda Elsever Norh Holland. Curre J. and D. Thoas 1999 Early es scores socoeconoc saus and fuure oucoes NBER Worng Papers 6943. Delere T. 2000 The wage and eployen effecs of he Aercans wh Dsably Ac. Journal of Huan Resources 35 693-715. Dupre D. and A. Karalanen 2002 Eployen of dsabled people n Europe Eurosa. Ehrlch I and H. Chua 1990 A odel for he deand for longevy and he value of lfe exenson. Journal of Polcal Econoy 98 761-782. García Góez P. and Á. López Ncolás 2006 Healh shocs eployen and ncoe n he Spansh labour are. Healh Econocs 15 997-1009. Grossan M. 1972 On he concep of healh capal and he deand for healh. Journal of Polcal Econoy 80 223-255. 28
Hecan J.J. and S. Navarro 2007 Dynac dscree choce and dynac reaen effecs. Journal of Econoercs 136 341-396. Hochss J. 2003 The labor are experence of worers wh dsables: he ADA and beyond. W.E. Upohn Insue for Eployen Research Kalaazoo. Kapeyn A. J.P. Sh and A. van Soes 2007 Vgnees and self-repors of wor dsably n he Uned Saes and he Neherlands. Aercan Econoc Revew 97 461-473. Lechner M. and R. Vasquez-Alvarez 2004 The effec of dsably on labor are. Oucoes n Gerany: Evdence fro Machng. CEPR worng paper 4223. Lndahl M. 2005 Esang he effec of ncoe on healh and oraly usng loery przes as exogenous source of varaon. Journal of Huan Resources 40 144-168. Mguel E. and M. Kreer 2004 Wors: denfyng pacs on educaon and healh n he presence of reaen exernales. Econoerca 72 159-217. Møller-Danø A. 2005 Road nures and long-run effecs on ncoe and eployen. Healh Econocs 14 955-970. Naeye R.L. and E.C. Peers 1984 Menal developen of chldren whose ohers soed durng pregnancy. Obserc Gynecoogy 64 601-607. NCDS User Suppor Group 1991 NCDS5 Repor. Cy Unversy: London. Scles R. C. and A. Yazbec 1998 On he deand for lesure and he producon of healh. Journal of Busness and Econoc Sascs 16 187-197. Sh J.P. 1998 Socoeconoc saus and healh. Aercan Econoc Revew Papers and Proceedngs 88 192-196. Sh J.P. 1999 Healhy bodes and hc walles: he dual relaon beween healh and econoc saus. Journal of Econoc Perspecves 13 145-166. 29
Sh J.P. 2003 Consequences and predcors of new healh evens. NBER worng paper 10063. Snyder S.E. and Evans W. N. 2002 The pac of ncoe on oraly: evdence fro he socal secury noch. NBER worng paper 9197. Van den Berg G.J. B. van der Klaauw and J.C. van Ours 2004 Punve sancons and he ranson rae fro welfare o wor. Journal of Labor Econocs 22 211-241. Wllas G.M. M. O Callaghan J.M.. Naan W. Bor M.J. Andersen D. Rchards and U. Chnlyn 1998 Maernal cgaree song and chld psycharc orbdy: A longudnal sudy. Pedarcs 102 e11. World Healh Organzaon 1977 Inernaonal Classfcaon of Dseases Nnh Revson. Geneva: WHO Publcaons. World Healh Organzaon 2004 Low brh wegh. Geneva: WHO Publcaons. 30
Table 1: Yearly ncdences of dfferen ypes of accdens Male Feale Overall 0.1199 0.0391 Road pedesran 0.0018 0.0013 Road drver 0.0179 0.0080 Worplace 0.0398 0.0072 Hoe 0.0127 0.0107 Spors 0.0338 0.0047 Oher 0.0139 0.0072 Table 2: Transon arces beween wor and dsably saes by gender Male sae n year +1 sae n year wor/ nonwor/ wor/ nonwor/ dsabled dsabled nondsabled Nondsabled wor/dsabled 95.3% 4.7% nonwor/dsabled 16.8% 83.2% wor/nondsabled 0.3% 0.1% 96.8% 2.8% nonwor/nondsabled 0.3% 0.7% 41.9% 57.2% Feale sae n year +1 sae n year wor/ nonwor/ wor/ nonwor/ dsabled dsabled nondsabled Nondsabled wor/dsabled 90.3% 9.7% nonwor/dsabled 12.8% 87.2% wor/nondsabled 0.3% 0.0% 91.7% 7.9% nonwor/nondsabled 0.1% 0.4% 19.3% 80.2% 31
Table 3: Saple ean of he ndvdual characerscs Toal Male Feale Feale 50.2% Educaon Naonal Vocaonal Qualfcaon level Below O-level equvalen 26.6% 24.6% 27.7% O-level equvalen 31.5% 27.9% 35.1% A-level equvalen 17.1% 20.9% 13.3% Degree equvalen 25.2% 26.6% 23.8% Parenal nforaon a brh ssng 7.3% 7.6% 7.1% Hgh parenal socoeconoc saus a brh 25.2% 25.5% 24.9% Medu parenal socoeconoc saus a brh 46.7% 46.0% 47.4% Low parenal socoeconoc saus a brh 20.8% 20.9% 20.6% Moher soed afer he fourh onh of pregnancy 30.5% 30.0% 31.0% Moher dd no soe afer he fourh onh of pregnancy 62.2% 62.4% 61.9% Brh nforaon ssng 5.5% 5.8% 5.3% Moher's age a brh n years 27.6 27.5 27.6 Low brh wegh less han 2500 gras 4.8% 4.1% 5.4% Noral brh wegh ore han 2500 gras 89.7% 90.1% 89.3% Born n Norh 27.1% 26.5% 27.7% Born n Mdlands 23.5% 24.3% 22.6% Born n Souh & Wales 16.3% 16.1% 16.5% Born n Scoland 10.4% 10.0% 10.7% Born n London & Souh-Eas 17.3% 17.3% 17.2% Tes scores a age 7 ssng 12.0% 12.6% 11.5% Mah 5.2 5.3 5.1 Brsol Socal Adusen Gude 8.4 9.8 7.0 Nuber of observaons 12375 32
Table 4a: Paraeer esaes of he log odel for accden probables Paraeer esaes Inercep 0.987 0.013 Beng eployed 0.342 0.014 Beng dsabled 0.048 0.004 Feale -1.203 0.010 Age dvded by 40-2.547 0.006 Age/40 2-1.190 0.007 Age/40 3 0.094 0.007 Age/40 4 1.493 0.008 O-level equvalen educaon 0.041 0.009 A-level equvalen educaon 0.171 0.011 Degree equvalen educaon 0.041 0.011 Chldhood bacground varables Low socoeconoc saus a brh -0.035 0.007 Moher soed a pregnancy 0.076 0.005 Brh nforaon ssng -0.730 0.009 Moher's age a brh dvded by 10-0.290 0.008 Moher's age a brh squared dvded by 100 0.423 0.014 Low brh wegh -0.016 0.004 Born n London or Souh-Eas 0.055 0.005 Born n Scoland -0.103 0.004 Born n Souh or Wales 0.007 0.005 Born n Norh 0.043 0.005 Born n Mdlands 0 Scores a age 7 ssng 0.066 0.008 Mah score dvded by 10 0.156 0.008 Brsol Socal Adusen Gude dvded by 10 0.051 0.005 Paraeers of he xng dsrbuon Probably 1: θ 1 θ 2 0.129 0.002 Probably 2: 1-θ 1 θ 2 0.423 0.005 Probably 3: θ 1 1-θ 2 0.105 0.001 Probably 4: 1-θ 1 1-θ 2 0.344 0.004 Locaon ass pon 1 0 Locaon ass pon 2-0.998 0.010 Locaon ass pon 3 1.204 0.013 Locaon ass pon 4 0.206 0.019 Sandard errors n parenheses See Table 4b for he nuber of observaons and he value of he log lelhood funcon 33
Table 4b: Mulnoal log wh unobserved heerogeney on ransons beween wor and dsably saes Fro Dsabled Nondsabled Wor Nonwor Wor Nonwor To Dsabled Dsabled Nondsabled Dsabled Nondsabled Nonwor Wor Wor Nonwor Nonwor Wor Nonwor Wor Inercep -2.491 0.010-1.213 0.004-6.185 0.005-9.216 0.021-4.166 0.014-3.411 0.004-4.023 0.006 2.241 0.025 Insananeous effec accden -0.074 0.004 0.462 0.003 0.862 0.004 1.658 0.006-0.019 0.010 1.021 0.005 0.863 0.004 0.045 0.033 Peranen effec of accden -0.003 0.010 0.474 0.006 0.159 0.004 0.409 0.005-0.059 0.007-0.322 0.003 0.116 0.005-0.172 0.027 Gender 0.922 0.010-0.359 0.010 0.315 0.006 1.164 0.005 1.241 0.008-1.450 0.006-0.635 0.005-1.290 0.008 Age dvded by 40-0.116 0.004 0.358 0.004 0.245 0.006 0.062 0.003 3.668 0.035-0.128 0.005-0.212 0.004-2.595 0.021 Age/40 2-0.533 0.005-0.408 0.004 0.588 0.005 0.101 0.004 0.039 0.009-0.113 0.004 0.302 0.004-1.295 0.006 Age/40 3-0.262 0.004-0.450 0.004 0.342 0.003 0.381 0.003-1.630 0.011-0.055 0.004 0.328 0.003 0.507 0.012 Age/40 4 0.017 0.004-0.623 0.004 0.219 0.004 0.676 0.006-2.429 0.010-0.035 0.003 0.543 0.005 1.513 0.009 O-level equvalen educaon -0.265 0.007-0.659 0.007-0.691 0.004 0.666 0.004 0.683 0.010 0.869 0.007-0.216 0.004-0.272 0.004 A-level equvalen educaon -0.523 0.005-0.468 0.009-1.121 0.018-0.857 0.016-0.521 0.013-0.789 0.011-0.913 0.014 0.193 0.003 Degree equvalen educaon -0.460 0.010-0.009 0.004-0.171 0.003 0.090 0.004 0.291 0.004 0.430 0.013 0.524 0.011 0.774 0.012 Parenal socoeconoc saus a brh Mssng 0.450 0.003-0.080 0.005 0.448 0.005 0.719 0.011 0.347 0.013-0.041 0.003-0.850 0.007-0.115 0.010 Hgh -0.099 0.003-0.316 0.004-0.085 0.004-0.299 0.004-0.056 0.005 0.314 0.004-0.152 0.003 0.037 0.007 Low 0.111 0.005-0.028 0.005 0.147 0.003 0.132 0.003 0.183 0.005 0.576 0.003-0.235 0.003-0.125 0.010 Moher song a pregnancy 0.123 0.003 0.007 0.004 0.159 0.004 0.358 0.003 0.122 0.010-0.315 0.003 0.184 0.003 0.006 0.009 Brh nfo ssng 0.551 0.003-0.069 0.003-0.514 0.004-0.559 0.006-0.571 0.016-0.081 0.003 0.090 0.007 0.213 0.009 Moher s age a brh /10 0.193 0.009 0.003 0.009-0.098 0.008 0.472 0.007-0.252 0.004-0.004 0.005-0.062 0.003 0.007 0.016 Moher s age a Age /10 squared -0.029 0.007-0.129 0.009 0.127 0.007-0.691 0.015 0.414 0.006 0.048 0.003 0.111 0.004 0.033 0.022 Low brh wegh -0.072 0.005-0.323 0.004 0.157 0.003-0.042 0.004-0.081 0.007 0.044 0.003-0.522 0.004-0.084 0.007 Regon of resdence a brh London 0.054 0.003-0.012 0.003-0.121 0.003-0.336 0.004-0.006 0.004-0.471 0.004 0.138 0.003-0.024 0.004 Scoland 0.217 0.003-0.005 0.004 0.044 0.003-0.275 0.003 0.236 0.007 0.603 0.005 0.167 0.003-0.076 0.004 Souh & Wales 0.165 0.004-0.094 0.003 0.179 0.003 0.117 0.003 0.030 0.003 0.364 0.004 0.070 0.003-0.030 0.005 Norh 0.374 0.004-0.101 0.003-0.056 0.003 0.215 0.003 0.211 0.004 0.217 0.004 0.257 0.003-0.055 0.004 Score a age 7 ssng -0.145 0.003-0.441 0.003 0.161 0.003-0.516 0.007 0.156 0.004-0.351 0.006-0.072 0.007 0.009 0.013 Mah score /10-0.387 0.007 0.118 0.006-0.043 0.004-0.709 0.006-0.040 0.009 0.209 0.004-0.653 0.008 0.088 0.012 Brsol socal adusen gude /10 0.209 0.004-0.185 0.004 0.055 0.004 0.180 0.003 0.140 0.013-0.329 0.004 0.101 0.003-0.088 0.009 Paraeers of he xng dsrbuon Locaon ass pon 1 0 0 0 0 0 0 0 0 Locaon ass pon 2-1.518 0.009 0.443 0.020-0.691 0.011-1.104 0.014-1.365 0.014-0.533 0.008-0.440 0.010 0.209 0.096 Locaon ass pon 3-1.003 0.006-0.073 0.034-0.065 0.005-0.962 0.009-0.785 0.016-1.021 0.008 0.007 0.004 0.983 0.033 Locaon ass pon 4-2.521 0.007 0.370 0.053-0.755 0.012-2.066 0.022-2.150 0.013-1.554 0.004-0.433 0.013 1.192 0.088 Nuber of observaons 12375 Value of he -log-lelhood 104848.05
Table 5: Sulaon experens Fracon Eployen Dsably rae Nuber of Effec accden Effec dsably Assocaon dsably rae age 40 age 40 accdens on dsably on eployen and eployen Full populaon 1.00 0.862 0.124 1.58 172% -0.144 0.048-0.229 Males 0.50 0.932 0.127 2.40 177% -0.226 0.041-0.234 Feales 0.50 0.790 0.125 0.77 157% -0.116 0.124-0.253 Low educaed 0.58 0.827 0.147 1.57 185% -0.205 0.045-0.272 Hgh educaed 0.42 0.916 0.094 1.60 169% -0.087 0.078-0.108 Noe: The eployen rae a age 40 he dsably rae a age 40 and he nuber of accdens unl age 40 are based on sulang he odel. The effec of an accden on he onse of a dsably reflecs he effec of an accden a age 24 on he dsably rae a he 25 h brhday. The effec of a dsably on eployen s he causal effec of he onse of a dsably a age 25 on he eployen rae a age 40 copued usng he Wald esaor. And he assocaon of dsably and eployen s based on sulang he odel and ang he dfference n eployen rae a age 40 of hose who were and were no dsabled a age 25 hs s hus non-causal.
Fgure 1: Labor are saes of ales. Males 1 0.9 0.8 Percenage populaon 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Age eployed uneployed ou of he labor force educaon Fgure 2: Labor are saes of feales. Feales 0.9 0.8 0.7 Percenage populaon 0.6 0.5 0.4 0.3 0.2 0.1 0 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Age eployed uneployed ou of he labor force educaon 36
Fgure 3: Dsably raes of ales and feales. 0.14 0.12 0.1 Dsably rae 0.08 0.06 0.04 0.02 0 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Age Males Feales Fgure 4: The annual ncdence raes of accdens for ales and feales. 0.25 0.2 Incdens rae 0.15 0.1 0.05 0 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Age Males Feales 37
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Fgure 5: Dsably and eployen saes. 1.0 0.9 0.8 0.7 Populaon fracon 0.6 0.5 0.4 0.3 0.2 0.1 0.0 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Age wor/dsabled nonwor/dsabled wor/nondsabled nonwor/nondsabled Fgure 6: F of he odel. 1 0.9 0.8 0.7 Populaon fracon 0.6 0.5 0.4 0.3 0.2 0.1 0 Age predced eployen predced dsably observed eployen observed dsably 38
Appendx A: Defnon of dsably We base our defnon of dsably on Cure and Madran 1999 as he enal and physcal characerscs ha eher consran noral daly acves or cause a subsanal reducon n producvy on he ob. The NCDS daa conans a se of queson on healh saus. Indvduals are ased a ages 23 33 and 42 wheher hey have a longsandng llness dsably or nfry whch ls her acves copared o people her own age. They are subsequenly requesed o docuen wheher ls her daly acves or he wor hey can do he age of he dsably onse and he ype of dsably. Dsably ypes are coded accordng o he nernaonal classfcaon of dsease ICD-9 produced by he World Healh Organzaon 1977. The ICD s exensvely used n healh sudes and s grouped no 17 broad caegores: 1. Infecons and parasc dseases e.g. uberculoss shngles herpes splex glandular fever 2. neoplass e.g. Hodgn s dsease leuea 3. endocrne nuronal and eabolc dseases and uny dsorders e.g. obesy dabees 4. dseases of he blood and blood-forng organs e.g. anea coagulaon defecs 5. enal dsorders e.g. depresson neuroc dsorders enal reardaon 6. dseases of he nervous syse and sense organs e.g. eplepsy grane blndness deafness 7. dseases of he crculaory syse e.g. hyperenson percards aorc aneurys 8. dseases of he respraory syse e.g. bronchs asha pleursy 9. dseases of he dgesve syse e.g. duodenal ulcer appendcs crrhoss of he lver 10. dseases of he genournary syse e.g. renal falure cyss nferly 11. coplcaons of pregnancy chldbrh and he puerperu e.g. sponaneous aboron eopc pregnancy 12. dseases of he sn and subcuaneous ssue e.g. eczea psorass 13. dseases of he usculoseleal syse and connecve ssue e.g. rheuaod arhrs derangeen of on 14. congenal anoales 15. ceran condons orgnang n he Pernaal perod 16. sypos sgns and ll-defned condons 17. Inury and posonng e.g. fracures sprans dslocaons rauac apuaon. 39