1 Classfcaton errors and permanent dsablty benefts n Span Serg Jménez-Martín José M. Labeaga Crstna Vlaplana Preto 1. Introducton There s a controverted debate about the effects of permanent dsablty benefts on labor market behavor. Parsons (1980) and Slade (1984) attrbute most of the decrease n the partcpaton of the workforce to the ncreasng generosty of dsablty benefts. Leonard (1979) also fnds that repercussons over labor market are consderable, whereas Haveman and Wolfe (1984a, b), Haveman et al., (1984), Bound (1989) and Haveman et al., (1991) support that the mpact s much more lmted. Other factors lke spouse contrbutons to famly ncome, the dsappearance of the stgma assocated wth early retrement and more generous early retrement benefts may explan the declne n labor partcpaton of older workers. In ths paper we study permanent dsablty benefts n Span to test whether they have beng used as an nstrument to favor pre-retrement (retrement before the early retrement age of 60). pˆ In the applcaton for dsablty benefts, we can dstngush two types of classfcaton errors: acceptaton error whch refers to ndvduals who receve a beneft but do not deserve t, and rejecton error whch ndcates ndvduals who appled for a beneft and deserve t, but t has been dened. Accordng to artcle 136 of the General Law of Socal Securty, an ndvdual deserves a beneft f after havng receved the prescrbed medcal treatment, presents anatomcal or functonal serous reductons, capable of objectve and predctably defntve determnaton, that dmnsh or annul hs labor capacty. Whle there are emprcal studes n some countres 1 provdng evdence about dsablty benefts, the lack of adequate data explans the absence of studes of ths knd n Span. The avalablty of the Dsabltes, Defcences and Health Status Survey (DDHSS from now on) carred out by the Spansh Natonal Bureau of Statstcs n 1999, whose results have recently been publshed, make t possble to conduct such an study. Unfortunately, we only know whether the respondent receves a beneft, thus beng not possble to test whether the reason why an ndvdual who deserves a dsablty beneft does not receve t: s t because he dd not apply for t, or because hs applcaton was dened? In ths sense, we are unable to emprcally dentfy the rejecton error. Fnancal help from Fundacon BBVA and project # BEC2002-04294-c02 are gratefully acknowledged. We thank workshop partcpants at Carlos III and Pompeu Fabra Unverstes. Unverstat Pompeu Fabra UNED, Madrd Unversdad Carlos III and UCAM 1 For US see, for nstance, Bound (1991), Kreder (1991), Benítez-Slva et al. (1999a); Bound et al. (1999) usng the Health and Retrement Survey. And for Europe see, Blanchet and Pele (1997), Blundell and Johnson (1997), Brugavn (1997), Pesteau and Stjns (1997), Borsch-Supan and Schnabel (1997), Palme and Svensson (1997).
2 It turns out that the fundamental ssue s to have a precse and trustworthy measure of what s understood by deservng a dsablty beneft. The dea that les behnd the concept of dsablty s not exempt from a hgh degree of subjectvty. Most studes have used self-reported health and dsablty measures (Boskn and Hurd, 1978; Gordon and Blnder, 1980; Hanoch and Hong, 1983; Berkovec and Stern, 1991). Others have argued that these ndcators are not approprate for measurng workng lmtatons (Zabalza et al., 1980; Myers, 1982; Parsons, 1982; Chrkos and Nestel, 1984; Bazzol, 1985; Bound, 1991; Bound et al., 1995; Bound and Burkhauser, 1999). The most common vew s that some survey respondents may nflate the ncdence and severty of health problems n order to ncrease the probablty of recevng a dsablty beneft (Burkhauser, 1979; Parsons, 1980 and 1982; Anderson and Burkhauser, 1985; Kerkhofs and Lndeboom, 1995). On the other hand, Nag (1979), Stern (1989), Dweyer and Mtchell (1998) and Benítez-Slva et al., (1999a) conclude that self-reported measures are good approxmatons of true dsablty status. Due to ths dsparty, we have adopted the defnton of what the Socal Securty understands for permanent dsablty, and we use ths defnton as an standard accordng to whch ndvduals determne ther level of dsablty. We agree that some ndvduals have ncentves to msreport ther dsablty status. They may try to exaggerate health problems when applyng for a beneft, although these ncentves dsappear when answerng a survey n whch confdentalty s guaranteed. Indeed, n the survey that we are gong to use, 48.75% of men and 33.92% of women who receve a dsablty beneft declare not to suffer any dsablty. We measure the acceptaton error comparng two varables: the Socal Securty awardng decson and the deservng ndcator desgned accordng to the legal text. The latter varable can be nterpreted as the decson that would take the Socal Securty n case all the nformaton was avalable 2. Usng nformaton for a set of ndvduals about whether they receve a dsablty beneft, and guessng whether they deserve t, we can estmate the jont dstrbuton of the varables recevng and deservng, from whch an acceptaton error can be derved. It s convenent to note that the acceptaton error s not the same than the error type two. The frst refers to the probablty of not deservng a beneft condtoned to recevng t, and the second refers to the probablty of recevng a beneft condtoned to not beng dsabled. 2 We face a stuaton of asymmetrc nformaton wth a completely nformed prncpal (the ndvdual) and an agent wth ncomplete nformaton (the Socal Securty). The exstence of hard-to-dagnose condtons may make t much more dffcult for Socal Securty to tag those unable for work form those that are able boded because of the mperfect nformaton about the nature of the dsablty, generatng a moral hazard problem.
3 Our results ndcate that ndvduals aged between 55 and 59, self-employers or workng n the agrculture sector have a probablty of recevng a beneft wthout deservng t sgnfcantly hgher than the rest of ndvduals. Ths confrms that dsablty benefts are beng used as an nstrument for extng the labor market for those who do not wsh to wat untl the early retrement and face the age penaltes, or those who do not have rght to early retrement because are afflated to specal regme regmes (see Boldrn et al, (1999) for a descrpton of Socal Securty rules and regmes). We also show that there are sgnfcant regonal dfferences n the probablty of recevng. Takng nto account that awardng process depends on Socal Securty Provncal Department, ths means that some departments are applyng loosely the dsablty requrements for grantng dsablty benefts. The structure of the paper s the followng: n secton we desgn a dsablty ndcator for deservng a dsablty beneft. In secton 3, we present econometrc setup for the analyss and descrbe the data. In secton 4, perform several specfcaton dagnostcs and n secton 5 we present the man estmaton results. In secton 6 we propose an alternatve mechansm for awardng dsablty benefts. Fnally, n secton 7 we conclude wth some polcy mplcatons. 2. Desgn of the dsablty ndcator In ths secton we present our ndcator for determnng whether an ndvdual deserves a permanent dsablty beneft. In the Dsabltes Questonnare from the DDHSS, ndvduals are requred to enumerate all dsabltes suffered accordng to a establshed classfcaton of 36 dsabltes 3, ndcatng also the degree of severty, expected evoluton, age when sufferng the dsablty problem, defcency that caused the dsablty, orgn and duraton. The nformaton about the relatonshp wth economc actvty s provded n the Household Questonnare. However, there are two mportant lmtatons: the degree of dsablty for permanent dsablty pensoners 4 and the possble denal of an applcaton for dsablty benefts are unknown. The second lmtaton prevent us from analyzng the rejecton error. Fnally, we only know the labor force status and the occupaton for dsablty pensoners when they declare to suffer some dsablty. The Spansh legslaton apples the professonal dsablty prncple to determne who deserves to receve a beneft. Ths prncple takes nto account three factors: the set of sequels, alments and dseases, the occupatons affected by lmtatons and the partcular effect on each ndvdual (Barba Mora, 2001). It s not possble to consder the frst factor because we do not know the occupaton for pensoners wthout 3 The concept of dsablty s lmtaton that affects daly lvng actvtes for at least one year.
4 dsabltes. As regards to the thrd factor t s mpossble to consder n a model varables such as resstance to pan and personalty, whch are genunely unobservable. In order to releve ths dsadvantage we are gong to restrct the analyss to survey respondents from 45 to 59 years old by gender. Ths means that we only requre elements of pathologcal character. The determnaton of the degree needed to establsh permanent dsablty would requre very deep medcal knowledge. Snce we do not rely on ths nformaton, we wll only look at the external symptoms of the defcences,.e., we focus on dsabltes. For each dsablty we are gong to fx a threshold of severty, forecast and orgn and we wll consder that an ndvdual deserves a dsablty beneft when satsfyng these requrements. We lke to be as exhaustve as possble, so we desgn sx dfferent crtera for deservng a dsablty beneft. The defnton of the varous crtera used are descrbed n Table 1. We dstngush aspects of the varable severty, evoluton, or orgn of the each defcency. Each crtera s represented by a bnary ndcator that takes the value one t the ndvdual satsfes the crtera and value zero otherwse. Afterwards we choose the best crtera for estmaton of the model usng exogenety and consstency tests for each of the varables. We assume that ndvduals who haven't answered the Dsabltes and Defcences Questonnares don't suffer any knd of health problem, so all crtera take the value zero. Table 1. Classfcaton of crterons for deservng a dsablty beneft Crteron 1 Crteron 2 Crteron 3 Crteron. 4 Severty Can present one of the followng Same requrement of severty that Same Same degrees: crteron 1. requrement of requrement of Serous dffculty severty that severty that Cannot do the actvty crteron 1. crteron 1. Forecast Can present one of the followng degrees: Stable, wth perspectves of mprovement Can go worse It s not possble to determne Can present one of the followng degrees: Recoverable wth restrctons Stable, wth perspectves of mprovement Can go worse It s not possble to determne Same requrement of forecast than crteron 1. Same requrement of forecast than crteron 2. Orgn All except congental and All except congental and problems n the All All problems n the chldbrth chldbrth Objectve We consder dsabltes wth a hgh degree of severty and unfavorable forecast Crteron 5 Crteron 6 Severty Can present one of the followng degrees: Same requrement of severty than crteron 5. Moderate dffculty Serous dffculty Cannot do the actvty Forecast Same requrement of forecast than crteron 1 and when severty s moderate we consder that forecast may be: Can go worse It s not possble to determne Orgn All except congental and problems n the chldbrth All Objectve We nclude degeneratve dseases (Parknson, Alzhemer) Same requrement of forecast than crteron 2, and when severty s moderate we consder that forecast may be: Can go worse It s not possble to determne. 4 The Spansh system dstngushes between great dsablty, absolute permanent dsablty, total permanent dsablty for usual professon and partal permanent dsablty for usual job.
5 3. Econometrc model We can represent Socal Securty's rule for awardng a permanent dsablty beneft as: r * = X ' β + ε r r (1) where X s a vector of characterstcs observed by the Socal Securty admnstraton, β r s the vector of weghts attrbuted to each one of these characterstcs. The term ε r can be nterpreted as some knd of nformaton known by the applcant, but unknown by the Socal Securty. Ths term can be nterpreted as a bureaucratc nose that nterferes the assgnment process. So Xβ r +ε r s the score the applcant gets accordng to Socal Securty rules usng a contnuous scale. Applcants wth a hgh score wll receve a dsablty beneft. The varable r * s a latent unobservable varable that we can represent usng a bnary ndcator: 1 f r = 0 f r r * * 0 < 0 (2) To represent f the applcant deserves a permanent dsablty measure we use a smlar equaton: d * = Z' β + ε d d (3) where Z s a set of varables that try to measure health status and β d s the correspondent vector of coeffcents or "weghts". The term ε d gathers some nformaton that s only known by the applcant. As before, the value assgned to the deservng ndcator d * s a latent unobservable varable that we can represent usng a bnary ndcator: d 1 f = 0 f d d * * 0 < 0 (4) We suppose (ε r,ε d ) that are dstrbuted as a bvarate normal wth zero mean vector, varances normalzed to one an correlaton coeffcent ρ (-1,1). We could, a pror, thnk that the set of varables X and Z have to be the same. Ths s true for varables related to dsabltes and defcences, but there are some varables that can affect the probablty of recevng although not the probablty of deservng. We specally refer to the place of resdence because t does not affect heath status but gven that the assgnment process depends on Socal Securty Provncal Departments, some dfferences n the level of exgency of dsablty requrements can emerge. As there are only four combnatons for the varables r and d, the lkelhood functon can be wrtten n terms of a multnomal dstrbuton:
6 p p p p L 11 = L( r = 1, d = 1 β r, β d, ρ, X, Z ) 10 = L( r = 1, d = 0 β r, β d, ρ, X, Z ) 01 = L( r = 0, d = 1 β r, β d, ρ, X, Z ) 00 = L( r = 0, d = 0 β r, β d, ρ, X, Z ) ( r, d β,,,, ) = r β d ρ X Z p11 p10 p01 r= 1 d= 1 r= 1 d= 0 r= 0 d= 1 r= 0 d= 0 p 00 (5) 3.1 Descrpton of the data As stated, our man data source s the Dsabltes, Defcences and Health Status Survey (DDHSS) carred out by the Spansh Natonal Bureau of Statstcs n 1999. A sample of 70.402 households and 218,185 people were ntervewed, from whch 10,484 were less than 6 years old and 207,701 were 6 or more years. We apply several flters to the ntal sample n order to get the relevant samples for the purposes of our study: Intal sample (sample A): We elmnate ndvduals below 45 and above 59. We concentrate n the range of observaton n whch dsablty benefts consttute a pathway to (defntve) ext or retrement from the labor force. The survvng sample has 19442 males and 20489 females. Restrcted ntal sample (sample B): To the ntal sample we apply the followng restrctons. We elmnate ndvduals wthout a contrbutve career and ndvduals that are workng but whose professonal stuaton s unknown. Ths leaves a sample of 18235 males and 8142 females. In these samples there are ndvduals wth and wthout dsabltes and we are gong to use them for exogenety and consstency tests 5. Sample wth dsabltes (sample B1). From sample B, we drop all ndvduals that do not answer the Dsabltes Questonnare. We elmnate 24,441 ndvduals and the sample left s 1,255 men and 681 women. From the sample of ndvduals who receve a permanent dsablty beneft and answer the Dsabltes Questonnare we elmnate 145 observatons for whch we do not know the occupaton and 255 observatons for whch we do not know relaton wth economc actvty before recevng the beneft. In ths sample, we defne a bnary varable that takes the value one f the ndvdual s recevng a contrbutve permanent dsablty beneft. Ths may not be the only relaton wth economc actvty. For example, 6.43% of men and 2.02% of women are also workng and 2.66% of men and 0.81% of women are lookng for a job. That left a sample of 984 men and 552 women. We wll perform the ratonal expectatons test wth ths fnal sample. 5 Those ndvduals who do not suffer any dsablty do not answer to the Dsabltes and Defcences Questonnare. We do not know f they have an mparment certfcate, need a caretaker, have fnshed a rehabltaton treatment or have changed of house
7 Sample wthout dsabltes (sample B0): Those observatons that are not assgned to sample B1, excludng the excluded ones, are assgned to sample B0. In Table 2 we summarze both the characterstcs of the samples used throughout the paper. Table 2. Descrpton of the samples Sample Characterstcs Gender: Sze Purposes A Indvduals aged 45-59 Men: 19442 Screenng mechansm Women: 20489 B +some requrements on economc actvty Men: 18235 Women: 8142 Exogenety and consstency test Bvarate probt B1 Ind. wth dsabltes + requrements on labor force status Men: 984 Women: 552 Ratonal expectatons test Bvarate probt B0 Ind. wthout dsabltes+ requrements on labor force status Men: 16980 Women: 7461 Prob. of recevng beng healthy 4. Dagnostcs on the dsablty ndcators 4.1 Exogenety test Before estmatng the model and usng the results wth polcy purposes, we would lke to be sure that the deservng ndcator s strctly exogenous. In other case, all the estmators would be nconsstent. We use Heckman's (1978) proposal usng a two equaton system to make such a test. The frst structural equaton represents the Socal Securty award decson and the second one ndcates whether the ndvdual deserves a dsablty beneft: r = W d = W 1 β1 + dα1 + u1 (6) 2 β 2 + u 2 where r and d are two latent contnuous varables, X s a vector of exogenous varables referred to dsabltes and defcences, and we are gong to use dfferent explanatory varables for both equatons. Then, excluded varables n one equaton serve as dentfcaton restrctons. We assume that (u 1,u 2 ) are dstrbuted accordng to a bvarate normal dstrbuton wth zero mean vector, varances normalzed to one and correlaton coeffcent ρ (-1,1). In frst nstance, we estmate the parameters usng a bvarate probt usng the restrcted ntal sample (sample B). The exogenety of the deservng equaton can be checked, after estmated through the test ρ=0. Another possblty to test exogenety of the deservng ndcator s the Lagrange multpler test. Under the null of exogenety, the model s composed by two ndependent probt equatons and bvarate probabltes and denstes concde wth the product of the correspondng margnal ones. To buld the test we follow Kefer (1982) and Greene (1993). Frst, we denote γ 1 =2r -1 and γ 2 =2d -1. In ths way, we have captured all the necessary changes of sgn to calculate the probabltes that r or d be equal to zero or one, because of a dsablty. But we suppose that the answer to all these questons s negatve because n other case they would have flled n the questonnare.
8 because γ j =1 f r =1 or d =1 and γ j =-1 f r =0 or d =0, for j=1,2. We denote δ 1 =γ 1 β 1 X 1 and δ 2 =γ 2 β 2 X 2. Then the lkelhood functon s the followng, (7) ),, ( ln ln 1 * 2 1 2 * = Φ = = n L L ρ δ δ where ρ * =γ 1 γ 2 ρ. The LM or score test s a quadratc form whch uses the frst dervatves of the unrestrcted lkelhood functon whose weghtng matrx s the nverse of the nformaton matrx for the unrestrcted lkelhood functon when both equaton are evaluated under the null. The statstc s dstrbuted accordng to χ 2 wth one degree of freedom. ( ) [ ] (9) ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( (8), ; 2 2 1 1 2 2 1 2 2 1 2 1 2 1 ' * * ' 2 ' ' * Φ Φ Φ Φ Φ Φ = = = LM L L L LM δ δ δ δ δ φ δ φ δ δ δ φ δ φ γ γ ρ β θ θ θ θ θ Table 3 reports the exogenety test by gender for the set of crtera defned. We cannot reject the null of exogenety for ndcators 3 to 6 n the case of men and for any crtera n the case of women, at standard sgnfcance levels. Table 3. Exogenety test. Restrcted ntal (Sample B) Men [N=18235] Women [N=8142] Statstc p-value Statstc p-value Crteron 1 6.6919 0.0071 2.5732 0.2698 Crteron 2 5.2417 0.0299 1.6519 0.4846 Crteron 3 2.4496 0.2936 1.2539 0.6020 Crteron 4 1.1703 0.6282 0.6004 0.8136 Crteron 5 0.9991 0.6829 0.0085 0.9976 Crteron 6 0.0172 0.9951 0.0032 0.9990 4.2 Consstency test The second dagnoss refers to consstency. We want to test whether the deservng ndcator based on self-reported dsablty status s an unbased estmator of the permanent dsablty award ndcator. [ ] ) (10 = 0 X d r E where X s a vector that ncludes dummy varables for 36 dsabltes and 27 defcences. Unbasedness can be tested through dfferent ways. Frst, we can regress by LS, (r-d) on a set of explanatory varables and test the hypothess that all coeffcents are equal to zero. Second, usng a multnomal logt model we can regress (r-d) over the same set of explanatory varables and test the same hypothess usng a
9 lkelhood rato test aganst a restrcted model whch only ncludes a constant. Fnally, we can perform a lkelhood rato test usng the results of a probt model of r-d on a set of varables aganst a restrcted probt model whch only ncludes a constant term. In all cases we use the restrcted ntal sample or sample B. In all regressons we have ncluded dummes for 36 dsabltes and 27 defcences. We have only carred consstency test for crterons 3 to 6 because crterons 1 and 2 dd not satsfy exogenety. The results of the test are reported n Table 4. For men, only crteron 6 satsfes all consstency tests, and for women both crtera 5 and 6 are vald under the results of these dagnostcs. Therefore, n what follows we use crteron 6 both for men and women. Table 4. Consstency test. Restrcted ntal sample (Sample B) Men [N=18235] Women [N=8142] Crteron 3 Crteron 4 Crteron 3 Crteron 4 Statstc pvalue Statstc pvalue Statstc pvalue Statstc pvalue OLS F(63,18171)=1.76 0.0002 F(63,18171)=1.76 0.0002 F(63,18171)=1.47 0.0097 F(63,18171)=1.57 0.0028 Probt χ 2 (63)=86.81 0.0251 χ 2 (63)=89.20 0.0166 χ 2 (63)=93.55 0.0075 χ 2 (63)=92.06 0.0099 MLogt χ 2 (63)=93.20 0.0080 χ 2 (63)=95.55 0.0051 χ 2 (63)=99.72 0.0022 χ 2 (63)=98.55 0.0028 Crteron 5 Crteron 6 Crteron 5 Crteron 6 Statstc pvalue Statstc value Statstc pvalue Statstc pvalue OLS F(63,18171)=1.36 0.0313 F(63,18171)=1.01 0.4525 F(63,18171)=1.27 0.0736 F(63,18171)=1.15 0.1920 Probt χ 2 (63)=96.66 0.0041 χ 2 (63)=73.40 0.1740 χ 2 (63)=51.22 0.8560 χ 2 (63)=52.97 0.8122 MLogt χ 2 (63)=104.47 0.0008 χ 2 (63)=81.44 0.0590 χ 2 (63)=52.97 0.8122 χ 2 (63)=63.12 0.4720 4.3 Ratonal expectatons test The results of the prevous sectons suggest that the deservng ndcator based on self-reported dsablty status s an exogenous determnant and consstent estmator of the Socal Securty award decson. However, the tests presented above are based on asymptotc propertes of the relevant test statstcs, but n small samples they mght have lttle power. Our key hypothess s that applcants have a through understandng of the award process, ncludng full knowledge of the weghts β r that Socal Securty places on varous characterstcs X, so the null hypothess that we want to test s: β r =β d. If the ratonal expectatons hypothess holds, then the deservng ndcator consttutes a vald measure of the degree of dsablty requred for recevng a dsablty beneft, and t can be used to measure the magntude of the classfcaton errors n the permanent dsablty benefts award process. Table 5 provdes the results for the sample of men and women that have declared to suffer some dsablty (sample B1). We have estmated a bvarate probt model n whch we have ncluded as explanatory varables 27 defcences and 36 dsabltes for the Socal Securty and for the applcant's equaton. For both samples we cannot reject, at standard sgnfcance levels, the null hypothess that the parameter vectors are equal.
10 Table 5. Ratonal expectatons test. Restrcted ntal sample (Sample B). Test β r=β d p-value Men [N=984] χ 2 (48)=48.10 0.4689 Women [N=552] χ 2 (48)=54.98 0.1981 In Fgures 1 to 4 we plot the estmated densty functons for the awardng and deservng decsons based on our preferred self-reported dsablty ndcator. To comp lete the pcture we present n Table 6 the man statstcs of the estmated densty functons. Note that deservng' densty functon s slghtly more skewed to the rght than the correspondng to Socal Securty. Ths means that there are more ndvduals who consder themselves as dsabled wth respect to what Socal Securty thnks about them. For women ths result s even more evdent. If we place the estmated densty functon for Securdad Socal over that of the ndvduals, we only fnd one densty functon. It could not have happened lke that, snce we mght have obtaned densty functons wth smlar means and modes, but that were totally dstngushable for havng very dfferent varances. Ths s an mplcaton of the ratonal expectatons test, because t shows that ndvduals have nternalzed the rules that Socal Securty uses n the process of awardng benefts. Fgures 1 to 4. Estmated densty functons Socal Securty (Men) Densty 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0-2.5-2 -1.5-1 -0.5 0 0.5 1 1.5 2 2.5 3 x'b Densty Socal Securty (Women) 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0-4 -3-2 -1 0 1 2 x'b Deservng ndcator (Men) Deservng ndcator (Women) Densty 0,45 0,4 0,35 0,3 0,25 0,2 0,15 0,1 0,05 0-2 -1 0 1 2 3 4 x'b Densty 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0-2 -1 0 1 2 3 4 5 x'b Table 6. Statstcs on the estmated densty functon Men [N=984] Women [N=552] S. Securty Indvduals S. Securty Indvduals Mean -0.2671 0.4206-0.8186 1.0862 Medan -0.3196 0.3195-0.8192 0.9255 St. Devaton 0.7130 0.6044 0.5623 0.9317 Maxmum 2.3946 3.0452 1.4789 4.6874 Mnmum -1.8173-1.5510-3.3142-1.1755 IQ Range 1.1120 0.7957 0.6462 1.1752
11 5. Emprcal results 5.1 A frst look at the data The analyss of descrptve data gves very nterestng result. Frst, from the sample of 984 men and 552 women, 20.05% of men and 13.99% of women who receve a beneft do not deserve t usng our preferred crtera #6. So, the acceptaton error s much larger for men than for women. There are several dsabltes that are not present among ndvduals who receve but do not deserve the beneft (dsablty for global vsual tasks, for hearng any sound, for communcatng through alternatve languages, not sealed gestures or readng-wrtng, for rememberng people/objects or nformatons/past epsodes and for executng smple or complex orders are examples) 6. Among women, n the acceptaton error area, there s nobody wth dsabltes for hearng strong sounds, lstenng the speech, mantanng body postures, washng oneself, controllng physcal needs, eatng and drnkng, dressng and undressng. On the other hand, ndvduals who receve but do not deserve usually declare to suffer osteoartcular or musculoskeletal defcences n column and upper and low extremtes, and dsabltes for movng not heavy objects and usng utensls and tools. Undoubtedly, these defcences are panful, but s hard to check how strong they are (Pérez Rueda et al., 2000). Moreover, none of these ndvduals needs a caretaker for daly actvtes. As regards those ndvduals who receve and deserve, the most common dsabltes are for gettng up and down, movng not heavy objects, movng wthout way of transport, gong n publc transport or drvng own vehcle. A look at socoeconomc characterstcs (Table B.2) reveals that there are more marred men pensoners than women, and 90% of men who do not deserve are heads of the household. Approxmately half of pensoners that deserve are between 55 and 59 years old and around 30% of those who do not deserve belong to ths age nterval. Most of male pensoners only have elementary educaton and prevously were sklled workers. Most pensoners only have elementary educaton, but whle most men were sklled workers, women only developed unsklled tasks.. In Table B.3 we report the spouse's labor force status for marred/cohabtng ndvduals n sample. If he/she s marred but s not a pensoner, the probablty of the spouse workng s hgher than f he/she s a pensoner and deserves, and twce larger than f he/she s pensoner and does not deserve. In all cases, the probablty of husband workng s twce the probablty of wfe workng. It s very nterestng to realze that f he/she s a pensoner and does not deserve the probablty that the spouse receves a permanent dsablty beneft s fve tmes than f he/she s not a pensoner. Ths result s more evdent for the case of husbands (37.5% wth respect to 9.06%).
12 Fnally, when the husband s a pensoner there s a hgher percentage of houseworker wves. Alternatvely, when the wfe s the pensoner the husband s more lkely to be n any other stuaton. 5.2 Estmaton results and forecasts In ths secton we present the results of the emprcal analyss by gender for three dfferent subsamples: (1) ndvduals wth dsabltes (sample B1); (2) ndvduals wth and wthout dsabltes (sample B); and (3) ndvduals wthout dsabltes (sample B0). 5.2.1 Indvdual wth dsabltes We have estmated a probt model for the decsons of deservng and recevng a permanent dsablty beneft (see Table C.1) restrctng the sample to ndvduals wth dsabltes. We assume that deservng precedes n tme to recevng. In order to calculate the margnal effect of each determnant we have ncluded the same set of explanatory varables n both equatons. Havng a dsablty or defcency has a postve and sgnfcant mpact on both equatons. We fnd some addtonal varables wth sgnfcatve mpact on the probabltes: age between 55 and 59, change of house because of a dsablty, mparment certfcate, caretaker, rehabltaton treatment, level of educaton, actve, unemployed and professonal occupaton. However, the marred and head of household dummes, varables related to the economc actvty of the spouse, and the place of resdence are not sgnfcant n the recevng equaton. The latter results should not cause any surprse snce these varables are supposed not to affect the health status. But nterestngly, all these determnants have sgnfcant effects n the recevng equaton, for both men and women. In terms of regonal dummes, lvng n Murca seems to guarantee the hghest probablty of recevng a dsablty beneft 7. Andaluca and Extremadura, for men, and only Andaluca, for women, are also sgnfcant although ther coeffcents are slghtly smaller than those from Murca. In order to evaluate the margnal effects we use the followng baselne: men/women less than 55 years old, not marred, college educaton, whte collar occupaton, lvng n Murca, wth only moderate severty dsabltes, wthout mparment certfcate, who has not receved treatment of rehabltaton nor changed of resdence because of a dsablty and does not need caretaker 8. The baselne probabltes of deservng are 0.324 for men and 0.211 for women. We could thnk that these probabltes are too hgh gven that ndvdual only suffers dsabltes of moderate severty. But we have to take nto account that although we have tred to consder the maxmum number of daly lvng actvtes when elaboratng the 6 See Appendx B.1. Data about defcences are avalable upon request. 7 In the estmaton of the bvarate probt model (Table C.1) we excluded the dummy varable correspondng to lvng n Murca.
13 self-reported dsablty ndcators, there are several unobserved factors such as pan or personalty that cannot be controlled for (Pérez Rueda et al., 2000). The baselne probabltes of recevng are 0.439 for men and 0.295 for women. That s, they are 40.31% and 40.29% hgher than the correspondng deservng probabltes. Moreover, men have a probablty of deservng and recevng hgher than women (53.68% and 47.61%, respectvely), but dfferences between deservng and recevng are about the same regardless the gender. Dsabltes, defcences and health varables: For men, the probablty of deservng s smaller than the probablty of recevng for defcences n upper an lower extremtes (152.61%), houseworkng (96.89%) and communcatng (53.21%). As regards women, the same happens for executng orders (165.81%), movng nsde home (61.09%) and mantanng body postures (60.02%). Dsabltes for movng nsde home, houseworkng and mantanng body postures are related to defcences n the vertebral column and upper and lower extremtes. In most cases they have an osteoartcular orgn (arthrts, rheumatsm, curvature of the spne, dsc herna) and ths make dagnoss qute dffcult because the apttude for bearng pan s nether measurable not unobservable. Age between 55 and 59 s another mportant determnant of both probabltes. The probablty of deservng ncreases 8.28% and 30.64% for men and women, respectvely. Interestngly, the probablty of recevng s hgher than the probablty of deservng (45.02% for men and 24.69% for women). Wthout any doubt, these results demonstrate that dsablty pensons have been used as an alternatve pathway to early retrement (Boldrn et al. (1999) mentoned ths use but they only show ndrect evdence). Needng a caretaker for daly lvng actvtes ncreases the deservng probablty n 92.46% for men and 40.73% for women. Ths varable s qute trustworthy because an ndvdual who needs somebody to take care of hm obvously deserves and should receve a dsablty beneft. Whether an ndvdual has changed of house because of a dsablty, ncreases the recevng probablty 53.03% for men and 66.73% for women. If he/she has an mparment certfcate, these probaltes rase 59.84% for men and 77.17% for women. It s mportant to underlne that for the case of men wth an mparment certfcate, the probablty of recevng s nearly twce the deservng one. Ths mples that the mparment certfcate, whch gves some fscal advantages, s beng used for polcy purposes. 8 Margnal effects for men are reported n Table C.2, and table of margnal effects for women are avalable on request..
14 Socoeconomc varables: Beng unemployed ncreases the probablty of deservng by 52.49% for men and 31.88% for women 9. Unemployed ndvduals may be less prone to apply for dsablty benefts because the amount of the beneft would be proportonal to the regulatory base, whch s lower n unemployment than n actvty 10. Clearly martal status and head of the household are addtonal polcy varables for beneft concesson. Whle martal status s not sgnfcant n the deservng equaton, beng marred ncreases the recevng probablty 8.65% for men and 1.66% for women. Moreover, f the ndvdual s the head of the household, the recevng probabltes ncrease 7.53% for men and 18.86% for women. Concernng varables related to the economc actvty of the spouse 11 f the ndvdual s a pensoner the probablty that the spouse s workng decreases 10.65% for men and 28.87% for women. On the other hand, there s a postve correlaton for the case of both spouses beng dsablty pensoners. If one spouse receves a dsablty beneft, the probablty that the other also receves benefts ncreases by 22.47% for men and 35.46% for women. Ths evdences some scale economes n beneft clam (when one of the spouses s a pensoner, he/she knows the awardng mechansm and t s easer for the other spouse to apply for t) and/or complementartes n lesure. The recevng probablty acheves ts maxmum when the ndvdual has elementary educaton and an unsklled job, and t decreases as the level of educaton and/or the qualty of occupaton ncreases. For both gender, the probablty of recevng ncreases more wth the educatonal level than wth the occupaton. Moreover, t seems to have a relatonshp wth the manual character of the jobs 12. Table 7 reports a cross-tabulaton of occupaton wth the estmated percentage of ndvduals wth dsabltes for usng hands and fngers. Ths knd of dsabltes are concentrated n unsklled workers, specally for women. Table 7. Manual dsabltes and professonal occupaton (Sample B1) Men [N=984] Women [N=552] Whte collar Sklled Not sklled Whte collar Sklled Not sklled Movng not heavy objects 16.32 34.21 49.47 14.81 17.28 67.90 Usng tools 11.39 33.66 54.95 13.51 14.86 71.62 Manpulatng small objects 14.60 32.12 53.28 15.32 12.61 72.07 9 The bnary varable unemployed takes the value one n two events. Frst, f at the tme of the survey the ndvdual s unemployed. Second, f the ndvdual s recevng a permanent dsablty beneft but relaton wth economc actvty before becomng pensoner was unemployed. 10 If unemployment were the consequence rather than the cause of poor health status, we would face an endogenety problem. To check what s the reason of joblesness we have used one of the questons of the DDHSS: Have you change your relaton wth economc actvty due to dsablty?. Only 10% of unemployed answered yes. 11 Several studes, for example Peracch and Welch (1994), Blau and Rphahn (1999) or Jménez et al. (1999) have observed that the spouse's labor force status affects retrement behavor of the other member of the couple. In ths paper we have proxed the spouse's labor status usng two dummy varables ndcatng whether he/she s workng or recevng a permanent dsablty beneft.
15 5.2.2 Indvduals wth and wthout dsabltes We have estmated a bvarate probt model poolng the sample of ndvduals wth and wthout dsabltes (18235 men and 8142 women, sample B) 13. In the deservng equaton we have ncluded the followng dummy varables: martal status, age between 55 and 59, head of household, professonal stuaton and occupaton, actvty sector and level of educaton. We have not ncluded regonal dummes because they do not contrbute to the lkelhood of deservng, as we have tested n the prevous models. In the award equaton we have ncluded dummes for beng marred, household sustaner, age between 55 and 59, regon and educatonal level. We do not nclude varables referred to actvty sector and occupaton n the award equaton because of the lack of nformaton about the sector for those ndvduals who receve a dsablty beneft but whch are not workng at the same tme. We only know the occupaton before recevng benefts for those who have answered the Dsabltes Questonnare. Moreover, we have not ncluded any of the 36 dsabltes because n ths sample most ndvduals dd not fll n the Dsabltes Questonnare, so we assume that they have a healthy status (94.60% of men and 93.22% of women). Anyway, we control health status by a dummy varable beng healthy that takes the value one when the ndvdual declares not to suffer any dsablty. Socoeconomc varables: Table 8 provdes nformaton about the probablty of recevng benefts for ndvduals wthout dsabltes (.e., ndvduals for whom the beng healthy ndcator s equal to one) dependng on ther occupaton and actvty sector 14. Agrculture s the actvty whch shows the hghest probablty of recevng beng healthy regardless the occupaton. The rankng of probabltes s completed, respectvely, by constructon, ndustry and servces. It seems very surprsng that the probablty n agrculture s even hgher than n constructon, whch has a hgher professonal hazard. The explanaton s that for agrculture workers the compulsory retrng age s 65. Perhaps, the acceptaton error could be reduced f early retrement was allowed. The dfference n the probablty of recevng between agrculture and servces s around 25%, regardless the occupaton. For a gven sector, the probablty of recevng decreases as the qualfcaton ncreases. For women, we have joned ndustry wth constructon n order to compute the probabltes because the sample s very small n both sectors of 12 For example: assembly lnes, seamstresses n court and confecton workshops, workmen of toys factores, replacers of supermarkets. In all these occupatons, there are many unsklled workers that become unable for dong ther jobs f they suffer any knd of manual dsablty. 13 Estmatons are avalable upon request. 14 We have related actvty sector wth the occupaton because we thnk that workng tasks have more to do wth occupaton than wth the level of educaton. Moreover, occupaton determnes not only the wage but also personal satsfacton and an ndvdual s gong to take nto account these two varables when applyng for a dsablty beneft.
16 actvty. The probabltes of recevng correspondng to women are half of those for men for any professonal occupaton and actvty sector. Table 8. Probablty of recevng beng healthy dependng on professonal occupaton and actvty sector (Sample B) Men Women Whte collar Sklled Not sklled Whte collar Sklled Not sklled Agrculture 0.1424 0.1776 0.2007 0.0719 0.1098 0.1161 Industry 0.1064 0.1553 0.1708 0.0666 0.0924 0.1067 Constructon 0.1209 0.1736 0.1887 Servces 0.1189 0.1535 0.1709 0.0569 0.0869 0.0959 Table 9 shows the probablty of recevng condtonal on beng healthy by schoolng level and professonal stuaton. For men, the self-employed have the hghest probablty of recevng beng healthy. Ths results confrm that the absence of early retrement for the self-employed ncreases the probablty of applyng for a dsablty beneft, even when healthy. Workng n the prvate or publc sector does not cause any dfferences n the probablty of recevng. For women, there are no sgnfcant dfferences n nether case. Fnally, the probablty of recevng decreases wth the level of educaton. Table 9. Manual dsabltes and professonal occupaton (Sample B) Men Women College Hgh School Elementary College Hgh School Elementary Self-employed 0.1003 0.1406 0.2045 0.0454 0.0636 0.0942 Publc Sector 0.0687 0.1183 0.1715 0.0453 0.0634 0.0939 Prvate Sector 0.0705 0.1207 0.1749 0.0455 0.0637 0.0943 5.2.3 Indvduals wthout dsabltes We have conducted a fnal test of the awardng process n the subsample whch only contans ndvduals wthout dsabltes (sample B0). Our purpose s to study the probablty of recevng a beneft condtoned on beng healthy 15. The set of explanatory varables ncludes controls for educaton, professonal occupaton, place of resdence, martal status, head of household and age between 55 and 59. We are not able to capture dfferences related to the actvty sector or professonal stuaton snce they are not avalable for non-workng pensoners. Once we have estmated the model, we compute the margnal effects wth respect to a baselne correspondng to an ndvdual of less than 55 years, not marred, not beng head of the household, wth hgh school educaton, sklled job and lvng n Murca (Table C.3). If the ndvdual s aged between 55 and 59 years old the probablty of recevng a beneft beng healthy ncreases wth respect to the baselne 117.69% for men and 166.66% for women. Ths confrms that the regulator allows healthy ndvduals to get (dsablty) benefts wthout any penaltes before the early retrement age. We also detect n ths case sgnfcant regonal dfferences n the probablty of recevng benefts. Lke n the prevous secton Murca, Andaluca and Extremadura have the hghest 15 Estmatons are avalable upon request.
17 probabltes. Ths may be because of n these regons the unemployment rate s greater (and the ncome level s lower) than the natonal average and the authortes are usng dsablty benefts as a way of artfcally decreasng the unemployment rate (and of ncreasng per capta ncome). When lookng at the quarterly unemployment rate for the perod 1980-1999 (see Fgure 5) we observe that ths s true for Andaluca and Extremadura, but not for the case Murca, whch fluctuates around the natonal mean. Alternatvely, we have represented n Fgure 6 the workng populaton n agrculture wth respect to total number of workers for ndvduals older than 55 years 16. Fgures for Andaluca and Extremadura fluctuate below and over the (decreasng) natonal mean, respectvely. But the trend for Murca s completely dfferent and fluctuates wthout pattern around the mean. Perhaps ths unexpected behavor s the explanaton for the large fgures of dsablty pensoners n Murca durng the eghtes. Fgures 5 and 6. Trends n unemployment rate and workng populaton n agrculture for Andaluca, Extremadura, Murca and Span. 35 30 Quaterly unemployment rate (%). 1980-1999 45 40 Workers agrculture/ Total workers, +55 years 25 20 15 10 5 1980.I 1981.I 1982.I 1983.I 1984.I 1985.I 1986.I 1987.I 1988.I 1989.I 1990.I 1991.I 1992.I 1993.I 1994.I 1995.I 1996.I 1997.I 1998.I 1999.I Span Andalucía Extremadura Murca Source: Actve Populaton Survey, INE. 35 30 25 20 15 10 1980.1 1981.1 1982.1 1983.1 1984.1 1985.1 1986.1 1987.1 1988.1 1989.1 1990.1 1991.1 1992.1 1993.1 1994.1 1995.1 1996.1 1997.1 1998.1 1999.1 Span Andalucía Extremadura Murca In Table 10 we report the probablty of recevng a penson when the ndvdual has declared not to have any dsablty for every combnaton of educaton and occupaton. The hghest probablty occurs when the ndvdual s an unsklled worker, whch happens for 21.95% of men and 31.82% of women 17. The probabltes for women are smaller than for men, although dfferences between both gender become smaller as the professonal qualfcaton decreases. As level of studes ncreases or occupaton decreases, the probablty of recevng a dsablty beneft wthout deservng t rases. The maxmum probablty occurs for an ndvdual wth college educaton and havng an unsklled job (0.92% of men and 1.59% of women). Ths may be explaned because these ndvduals are not havng a correct match n the labor market, so they try to ext from the labor market through permanent dsablty. Table 10. Probablty of recevng beng healthy dependng on level of studes and professonal occupaton (Sample B0) Men Women Whte collar Sklled Not sklled Whte collar Sklled Not sklled College 0.0107 0.0137 0.4063 0.0011 0.0038 0.1771 16 As we only look at workers older than 55 years old, we do not observe seasonal varatons because these ndvduals are not usually temporary workers. We have represented only untl 1999 to avod dstortons wth the fgures of mmgrants. 17 There are some empty cells because there are not any ndvdual wthout studes and beng whte collar or sklled worker, or wth elementary educaton and beng whte collar.
18 Hgh School 0.0101 0.0130 0.3989 0.0010 0.0036 0.1735 Elementary 0.0123 0.3906 0.0036 0.1685 None 0.3829 0.1705 6. An alternatve mechansm for awardng dsablty benefts In ths secton we propose a smple screenng mechansm alternatve to the Socal Securty awardng process 18 We evaluate the relatve effcency, evaluated by the acceptaton error, commtted by both methods of classfcaton. The best mechansm wll be the one provdng the mnmum number of undeserved dsablty benefts. If Socal Securty were to know the true dsablty status of the applcants, d *, could use ths nformaton n the process of awardng benefts. We have mentoned that some ndvduals have ncentves to msreport to Socal Securty, but these ncentves dsappear when they are answerng a prvate and confdental survey. Under the hypothess of ratonal expectatons, we can regress our preferred deservng ndcator (based on self-reported measures) on a set of explanatory varables X. 19 After estmaton, we compute the probablty of deservng condtonal on X, P ˆ( d X). Then, we award (A) a beneft to those ndvduals havng a probablty of deservng above a gven threshold, α [0,1]: { P( d ) α} (11) A = I X Adjustng the level of α we obtan dfferent percentages of benefts, the bgger α the smaller the number of dsablty benefts. Consder that the Socal Securty admnstraton objectve s to acheve a gven fracton of dsablty pensons, say p. Then α would be gven by the mnmum of the followng expresson: Mn α { P d X ) α} p = I ( f ( X ) dx (12) We can compute the sample analog of expresson (11) and obtan the optmal value of α, say αˆ, for whch the percentage of ndvduals whch receve benefts s equal to the Socal securty objectve, p. Fgures 7 and 8, for men and women respectvely, show the correspondng estmated (by usng a probt) densty functons. We ndcate wth a dscontnuous lne the ndex ρ that makes the rght tal equal to the 18 See Benítez Slva et al. (1999b) for an alternatve screenng mechansm. 19 In the probt model the dependent varable s the preferred crteron # 6. The set of explanatory varables ncludes: rehabltaton treatment, caretaker, mparment certfcate, change of house because of a dsablty and the 36 dsabltes mentoned n Appendx B.1. Wth respect to dsabltes we have defned a dummy varable that takes the value one f dsablty s suffered wth moderate or hgher severty and wth a forecast dfferent from recoverable. The sze of the sample s bgger than n the bvarate probt model because we do not need to drop observatons for whch some economc varables are unknown. That's why we can estmate all dsabltes separately. The results of the estmaton procedure are avalable on request.
19 probablty of recevng a dsablty beneft condtoned to deservng t 20. Ths probablty s 0.4919 for men and 0.1610 for women, and the correspondng ndexes are 0.3647 for men and 1.2455 for women, wth a standard devaton equal to 0.0016 and 1.2455, respectvely. Fgures 7 and 8. Estmated probabltes of deservng (Sample A) Men Women 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 P[receve deserve]=0.4919 alpha=0.3647 0-4 -3-2 -1 0 alpha 1 2 3 4 5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 Pr[receve deserve]=0.1610 alpha=1.2455 0-4 -3-2 -1 0 1 alpha 2 3 4 We award a dsablty beneft to those ndvduals wth a value of P ˆ( d X) greater than αˆ. Then we compare the effcency between the Socal Securty and the screenng mechansm. Our mechansm wll be better than Socal Securty f satsfes three requrements. Frst, reduces the number of ndvduals that receve wth Socal Securty but don't deserve. Second, ncludes all ndvduals that are recevng accordng to Socal Securty and deserve. And thrd, doesn't award benefts to new ndvduals who don't deserve t. Table 11 presents some summary measures obtaned from the analyss. Frst, we observe that all ndvduals who receve and deserve benefts wth Socal Securty also receve wth ths alternatve mechansm. Second, the number of ndvduals who receve but do not deserve decrease from 63.46% to 7.38% for men and from 49.18% to 9.84% for women. Thrd, ths mechansm does not assgn benefts to any ndvdual who does not deserve benefts. Fnally there are 2.77% of men and 4.33% of women who deserve but do not receve accordng to Socal Securty. If we have known whether these ndvduals have appled for a beneft n some moment we could determne whether they belong to the rejecton error. But n some crcumstances, although the applcant presents a degree of dsablty that justfes a permanent dsablty beneft, he may not satsfy the economc requrements of number of years workng and quotatons to Socal Securty. Fnally, we have reduced the number of benefts by 58.89% for men and 49.18% for women. 20 We could have used the rate of award (nº of applcatons/nº of awards), but wth the nformaton from the survey we only know f the ndvdual receves a beneft but not f he has appled for and t has been dened. The economc nterpretaton for usng the
20 Table 11. Comparson between Socal Securty and screenng mechansm Men Women Socal Securty * Total 1333 429 * Receve and deserve 487(36.53%) 218(50.82%) * Receve but not deserve 846(63.47%) 211(49.18%) Screenng mechansm * Total 542 254 * Receve now, wth SS and deserve 487(89.85%) 218(85.83%) * Receve now, wth SS and not deserve 40(7.38%) 25(9.84%) * Receve now, not wth SS and deserve 15(2.77%) 11(4.33%) * Receve now, not wth SS and not deserve 0 0 It s not strange that we have obtaned better results n the benefts awardng wth our screenng mechansm. It s smply a consequence of the use of confdental self-reported dsablty nformaton. We do not propose the dsappearance of the whole bureaucracy of the Socal Securty, just because ndvduals could be tempted to msreport dsabltes when answerng the survey wth the purpose of ncreasng the probablty of deservng. In order to avod these problems, we suggest several control measures over the applcant and the dsablty referees. For example, a medcal team should vst the home and the workng place of the applcant and study the way of dsplacement between both. Also random audts should be performed on medcal judgements, condtonng examner's wage earnngs to the concdence between hs verdct and that of the audtor. 7. Conclusons In ths paper we partally audt the permanent dsablty awardng process n Span. Frst, we have desgned a deservng ndcator usng self-reported dsablty measures from the DDHSS survey. Then, we have estmated a latent varable bvarate econometrc model for the decsons of recevng and deservng for those ndvduals who have developed contrbutve careers. The man concluson s that ndvduals aged between 55 and 59, the self-employed or workers n the agrcultural sector have a probablty of recevng a beneft wthout deservng t sgnfcantly hgher than the rest of ndvduals. Ths confrms that dsablty benefts have beng used as a pathway to retrement below the early retrement age. We have also shown that there are sgnfcant regonal dfferences and the probablty of recevng s hgher n Murca, Andaluca and Extremadura. If we take nto account that the awardng process depends on Socal Securty Provncal offces, t becomes clear that somethng abnormal s happenng n the evaluaton of the worker health status. One soluton would be to carry out more dagnoss tests and exploratons, although ths would ncrease audt expendtures. For example, n the case of the osteoartcular pathologes n whch pan plays probablty of recevng condtoned to deservng s that ndvduals wth a hgh probablty of deservng should receve a beneft.
21 a crucal role there s a trade off between the accomplshment of some expensve tests and the rsk of payng undue benefts. 8. References 1. Anderson, K., Burkhauser, R., 1985. The retrement-health nexus: a new measure of an old puzzle. Journal of Human Resources 20, 118-137. 2. Barba Mora, A., 2001. Incapacdad permanente y Segurdad Socal. Edtoral Aranzad. 3. Bazzol, G., 1985. The early retrement decson: new emprcal evdence on the nfluence of health. Journal of Human Resources 20, 214-234. 4. Berkovec, J., Stern, S., 1991. Job ext behavor of older men. Econometrca 59, 189-210. 5. Benítez-Slva, H., Hu-Man, C., Rust, J., Shedvasser, S., 1999a. An emprcal analyss of the Socal Securty Dsablty, applcaton, appeal, and award process. Labour Economcs 6, 147-178. 6. Benítez-Slva, H., Hu-Man, C., Rust, J., Shedvasser, S., 1999b. Socal Securty dsablty award process: how large s the bas n self-reported dsablty? Manuscrpt. Yale Unversty. 7. Blanchet, D., Pele, L. P., 1997. Socal Securty and retrement n France. Workng Paper NBER, nº 6214. 8. Blaud, D. M., Rphahn, R. T., 1999. Labor force transtons of older marred couples n Germany. Labor Economcs 6, 229-251. 9. Blundell,, R., Johnson, P., 1997. Pensons and retrement n the UK. Workng Paper NBER, nº 6514. 10. Boldrn, M., Jménez-Martín, S., Peracch, F., 1999. Socal Securty and Retrement n Span'', n J. Gruber and D. Wse (Eds.), Socal Securty Programs and Retrement around the World, Chcago Unversty Press for the NBER. 11. Borsch-Supan, A., Schnabel, R., 1997. Socal Securty and retrement n Germany. Workng Paper, nº 6153. 12. Boskn, M., Hurd, M., 1978. The effect of Socal Securty on early retrement. Journal of Publc Economcs 10, 361-377. 13. Bound, J., 1989. The health and earnngs of rejected dsablty nsurance applcants. Amercan Economc Revew 79, 482-503. 14. Bound, J., 1991. Self reported versus objectve measures of health n retrement models. Journal of Human Resources 26, 106-138. 15. Bound, J., Burkhauser, R. V., 1999. Economc analyss of transfer programs targeted on people wth dsabltes. In Ashenfelter, O., Card. D. (Eds.), Handbook of Labor Economcs, vol 3C, 3417-3528. Elsever, Amsterdam. 16. Bound, J., Schoenbaum, M., Wadmann, T., 1995. Dsablty status and racal dfferences n the labour force attachment. Journal of Human Resources 30, 227-267. 17. Bound, J., Schoenbaum, M., Stnebrckner, T., Wadmann, T., 1999. The dynamc effects of health on the labor force transtons of older workers. Labour Economcs 6, 179-202. 18. Brugavn, A., 1997. Socal Securty and retrement n Italy. Workng Paper NBER, nº 6155. 19. Bukhauser, R., 1979. The penson acceptance decson of older worders. Journal of Human Resources 14, 63-75. 20. Chrkos, T., Nestel, G., 1984. Economc determnants and consequences of self-reported work dsablty. Journal of Health Economcs 3, 117-136. 21. Dwyer, D., Mtchell, O., 1998. Health problems as determnants of retrement: are self-rated measures endogenous? Natonal Bureau of Economc Research. Workng Paper 6503. 22. Gordon, R., Blnder, A., 1980. Market wages, reservaton wages and retrement decsons. Journal of Publc Economcs 14, 277-308. 23. Greene, W.H., 1993. Econometrc analyss. Second Edton, Pretnce Hall, New Yersey. 24. Hanoch, G., Hong, M., 1983. Retrement, wages and labor supply of the elderly. Journal of Labor Economcs 1, 131-151. 25. Haveman, R., de Jong, P., Wolfe, B., 1991. Dsablty transfers and the work decson of older men. Quaterly Journal of Economcs 106, 131-161. 26. Haveman, R., Wolfe, B., 1984a. The declne n male labor force partcpaton: comment. Journal of Poltcal Economy 92, 532-541. 27. Haveman, R., Wolfe, B., 1984b. Dsablty transfers and early retrement: a causal relatonshp?. Journal of Publc Economcs 24, 47-66. 28. Haveman, R., Wolfe, B., Warlck, J., 1984. Dsablty transfers, early retrement, and retrenchment. In Aaron, H., Burtless, G., (Eds.), Retrement and economc behavor, 65-93. Washngton, DC; Brookngs Insttuton.
22 29. Heckman, J., 1978. Dummy endogenous varables n a smultaneous equaton system. Econometrca 46, 931-959. 30. Jménez-Martín, S., Labeaga, J.M. and Martínez-Granado, M., 1999. Workng Paper Unversdad Carlos III de Madrd. 31. Kerkhofs, M., Lndeboom, M., 1995. Subjectve health measures and state department reportng errors. Health Economcs 4, 221-235. 32. Kefer, N., 1982. Testng for dependence n multvarate probt models. Bometrca 69, 161-166. 33. Kreder, B., 1999. Latent work dsablty and reportng bas. Journal of Human Resources 34, 734-769. 34. Leonard, J. S., 1979. The Socal Securty Dsablty program and labor force partcpaton. Natonal Bureau of Economc Research. Workng Paper 392. 35. Myers, R., 1982. Why do people retre from work early? Socal Securty Bulletn, September, 10-14. 36. Nag, S., 1979. The concept and measurement of dsablty. In Dsablty Polces and Goverment Polces. New York: Praeger. 37. Palme, M., Svensson, I., 1997. Socal Securty, occupatonal pensons and retrement n Sweden. Workng Paper NBER, nº 6137. 38. Parsons, D., 1980. The declne n male labor force partcpaton. Journal of Poltcal Economy 88, 117-134. 39. Parsons, D., 1982. The male labour force decson: health, reported health and economc ncentves. Economca 49, 81-91. 40. Pérez Rueda, B., García Blázquez, M., García Blázquez Pérez, M., 2000. Fundamentos médco legales de la ncapacdad laboral permanente. Edtoral Comares, 4ª edcón. 41. Peracch, F., Welch, F., 1994. Trends n labor force transtons of older men and women. Journal of Labor Economcs 12, 210-242. 42. Pesteau P., Stjns, J. P., 1997. Socal Securty and retrement n Belgum. Workng Paper NBER, nº 6169. 43. Slade, F., 1984. Older men: dsablty nsurance and the ncentve to work. Industral Relatons 23, 260-269. 44. Stern, S., 1989. Measurng the effect of dsablty on the labor force partcpaton. Journal of Human Resources 24, 361-395. 45. Zabalza, A., Pssardes, P., Barton, M., 1980. Socal Securty and the choce between full-tme work, part-tme work and retrement. Journal of Publc Economcs 14, 245-276. Appendx A Table A. Rato between number of permanent dsablty benefts and retrement benefts 1985 1987 1989 1991 1993 1995 1997 1999 2001 Andalucía 91.9 93.6 92.4 88.6 86.1 84.1 76.5 27.2 26.5 Aragón 42.1 41.3 40.6 38.6 38.5 37.2 34.0 13.4 13.1 Asturas 56.4 54.8 53.6 51.6 51.2 50.0 45.7 18.6 18.6 Baleares 42.3 42.0 42.1 42.8 44.4 45.2 43.4 19.4 20.4 Canaras 74.6 75.1 74.1 73.2 69.5 65.4 58.5 22.9 21.9 Cantabra 43.9 43.7 43.9 43.7 43.6 42.5 39.3 14.9 14.9 C. Mancha 52.9 52.2 51.4 48.7 46.9 44.9 40.3 14.1 14.2 C. León 54.2 53.0 51.9 48.3 46.7 44.3 39.1 12.9 12.1 Cataluña 57.2 56.9 56.3 53.5 50.9 47.8 43.4 18.3 17.5 Extremadura 56.6 57.7 57.8 55.8 54.7 52.7 47.7 17.5 16.4 Galca 55.2 55.9 55.5 54.3 53.2 50.9 45.1 13.9 13.5 Madrd 40.2 38.5 33.4 37.5 36.9 34.5 31.0 13.6 12.9 Murca 150.4 153.6 157.0 152.0 148.5 138.1 116.4 28.1 25.8 Navarra 52.1 51.2 48.5 46.4 45.1 43.5 39.4 16.4 15.9 País Vasco 47.8 46.2 44.7 43.1 42.2 40.1 37.0 17.2 16.1 La Roja 65.9 65.0 63.7 60.4 59.9 57.4 50.2 17.3 16.5 Comundad Valencana 52.7 50.6 49.2 45.7 43.5 41.6 38.9 16.6 16.6 Ceuta 83.3 73.7 66.7 63.6 54.2 48.1 44.8 19.4 19.4 Mellla 100 94.1 94.1 88.9 94.7 95.0 90.5 33.3 33.3 España 59.06 58.4 56.7 55.2 53.6 51.2 46.4 17.9 17.3 Source: Mnstry of Labor and Socal Issues. Note: Snce 1997 (Law 24/1997) dsablty pensons for ndvduals aged 65+ are counted as retrement pensons.
23 Appendx B. Descrptve statstcs Table B.1. Dsabltes Men wth benefts Women wth benefts Nº of observatons Deserve Not deserve Deserve Not deserve Seeng any mage 0.98 0 0.90 0 Global vsual tasks 13.73 5.48 16.22 14.29 Detaled vsual tasks 15.03 41.11 14.41 14.29 Other vsual dsabltes 4.58 9.59 8.11 7.14 Hearng any sound 1.63 0 0 0 Hearng strong sounds 1.31 2.74 1.80 0 Lstenng the speech 9.48 9.59 8.11 0 Communcatng through speech 7.19 6.85 2.70 7.14 Communcatng through alternatve languages 0.65 0 0.90 0 Communcatng through not sealed gestures 0.65 0 0.90 0 Communcatng through readng-wrtng 9.80 0 9.01 0 Recognzng people 3.92 0 4.50 0 Rememberng nformaton 9.48 0 9.01 0 Executng smple orders 2.94 0 4.50 0 Executng complex orders 7.52 0 5.41 0 Mantanng body postures 23.53 12.33 32.43 0 Gettng up and down 33.33 26.03 39.64 14.29 Movng nsde home 19.93 12.33 23.42 7.14 Movng not heavy objects 36.93 26.03 39.64 42.86 Usng utensls and tools 31.05 20.55 32.43 35.71 Manpulatng small objects 17.65 19.18 25.23 21.43 Movng wthout transport 45.52 31.51 44.14 14.29 Movng n publc transport 41.18 20.55 44.14 14.29 Drvng own vehcle 49.67 24.66 48.65 21.43 Washng oneself 16.34 2.74 18.02 0 Controllng physcal needs 5.23 1.37 4.50 0 Dressng and undressng 16.99 5.48 11.32 0 Eatng and drnkng 4.25 1.37 4.50 0 Do the shoppng 20.26 8.22 35.14 21.43 Cookng 13.73 8.22 25.23 7.14 Washng and ronng clothes 18.63 10.96 37.84 28.57 Cleanng the house 24.18 19.18 51.35 28.57 Lookng after the famly 16.99 8.22 27.93 21.43 Relatons wth famly 2.61 1.37 4.50 7.14 Makng frends 7.84 1.37 9.91 7.14 Relatng at work 11.11 2.74 16.61 7.14 Table B.2. Socoeconomc characterstcs Wthout Wth dsablty beneft Wthout Wth dsablty beneft dsablty beneft Deserve Not deserve dsablty beneft Deserve Not deserve Workng 27.66 22.66 13.11 58.19 48.75 25 Workng but temporally absent 0.58 1.17 1.74 Lookng for frst job 0.39 0.78 1.64 0.35 Unemployed (has worked before) 5.61 7.42 6.56 13.59 13.75 Unable for workng 0.77 1.95 3.28 3.83 6.25 Contrbutve Dsablty Beneft 1.74 3.91 6.56 9.06 15 37.5 Not Contrbutve Dsablty Beneft 2.9 2.34 3.28 0.7 Retrement Beneft 0.77 1.17 11.15 16.25 12.5 Studyng 0.39 0.78 Houseworkng 58.41 57.42 63.93 Other stuatons 0.77 0.39 1.64 1.39 25 Men: (N=984) 82.43% of the sample s marred Women: (N=552) 65.28% of the sample s marred Table B.3. Spouse labor force status Men Women Receve Deserve Receve Not deserve Receve Deserve Receve Not deserve Age between 55-59 47.06 35.62 47.75 28.57 Needs caretaker 7.30 0 18.11 0 Rehabltaton treatment 30.25 28.38 35.43 35.29 Change of house 7.62 6.76 10.24 11.76 Marred 80.72 83.56 64.86 50 Man sustaner 84.97 90.41 27.03 35.71
24 Studes Wthout studes 31.05 32.77 35.13 28.57 Elementary 46.73 43.94 40.54 35.71 Hgh school 18.63 20.55 15.32 21.43 College 3.59 2.74 9.01 14.29 Profesonal occupaton (before) Unsklled 22.14 18.84 52.25 42.87 Sklled 65.44 72.46 40.54 35.71 Whte collar 12.42 8.70 7.21 21.42 Appendx C. Estmatons and forecasts Table C.1. Bvarate probt. Sample wth dsabltes Men Women Deserve Receve Deserve Receve Coef. t Coef. t Coef. t Coef. t Constant -0.8056 3.50-0.6*9-3.20-0.8503-2.20-0.9293-2.68 Dsablty: Vertebral column 0.4186 2.14 0.3044 2.02 0.7184 3.48 0.1936 2.08 Upper and low extremtes 0.0552 2.27 0.2642 2.50 0.4462 1.82 0.0813 2.39 Seeng 0.3658 2.27 0.0253 2.17 0.9053 4.01 0.0346 2.20 Hearng 0.3194 1.93 0.3654 3.54 0.4573 2.13 0.6298 2.84 Communcatng 0.2465 2.00 0.4685 2.01 0.7546 2.75 0.8165 2.45 Rememberng 0.2852 2.37 0.3842 2.95 0.3427 2.59 0.2613 2.59 Executng orders 0.3134 1.90 0.2885 2.80 0.1950 2.19 1.1224 1.93 Mantanng body postures 0.2282 2.05 0.0316 2.16 0.0706 3.30 0.2068 1.99 Movng nsde home 0.4744 2.41 0.0295 2.17 0.0291 3.13 0.1602 2.80 Usng hands and fngers 0.2162 2.49 0.2821 2.19 0.3649 2.14 0.1926 2.24 Movng outsde home 0.4109 2.87 0.3889 3.63 0.6666 3.72 0.1091 2.65 Takng care of oneself 0.5355 2.19 0.1974 3.10 0.0938 2.31 0.1865 2.36 Houseworkng 0.0405 2.18 0.5447 2.95 0.3401 1.83 0.4302 2.48 Relatng 0.8202 2.38 0.5579 2.93 0.5068 2.30 0.1732 3.05 Socoeconomc characterstcs Age between 55-59 0.0733 2.64 0.1379 2.21 0.2069 2.35 0.1332 2.90 Change of house 0.3917 2.11 0.6247 2.18 0.0516 2.15 0.5162 1.91 Imparment certfcate 0.1240 2.91 0.7150 5.77 0.2575 2.41 0.5962 3.76 Needs caretaker 0.7702 2.38 0.1811 2.60 0.2694 2.93 0.3602 2.20 Rehabltaton treatment 0.1286 2.81 0.2731 1.92 0.0582 2.34 0.0366 2.24 Spouse workng -0.0191 1.52-0.4148-2.82 0.0038 1.18-0.1079-2.13 Spouse dsablty beneft 0.0189 1.75 0.3089 2.67 0.0058 1.69 0.1699 2.55 Marred 0.0249 0.63 0.0988 2.01 0.0105 0.71 0.0141 1.98 Elementary 0.8045 2.09 1.6779 1.96 0.3394 2.63 0.9641 2.51 Hgh school 0.5977 2.23 0.6317 2.05 0.0300 2.39-0.1359-2.88 College 0.2952 2.05 0.3559 2.42-0.0893-2.27-0.6535-2.02 Actve 0.1841 2.01 0.3195-1.90-0.0569-2.24 0.2020 2.03 Unemployed 0.4425-2.93-0.2975-2.14 0.1578 2.62-1.4187-3.09 Man sustaner -0.1978-0.96 0.0860 2.43 0.0243 1.11 0.1554 2.67 Whte collar occupaton -0.1316-2.60 0.1696 3.43 0.1917 2.77 0.8429 3.42 Sklled occupaton -0.0175 3.34 0.7779 5.13 0.3394 1.94 0.9641 4.00 Autonomous Communtes Andalucía-Ceuta-Mellla 0.7624 1.39-0.0073-1.99 0.5326 1.17-0.1275-2.08 Aragón-Navarra-Roja 0.7546 1.60-0.2636-1.99 0.5235 1.21-0.4570-1.89 Asturas-Cantabra 1.0826 1.18-1.0627-2.22 0.2898 0.76-0.5114-2.46 Baleares 0.7349 1.12-1.0076-2.02 0.2621 0.61-1.1546-2.54 Canaras 0.4906 0.96-1.0566-1.91 0.5562 1.32-0.6078-2.31 Castlla La Mancha 0.9363 0.86-0.9415-3.19 0.2946 0.80-0.3446-1.97 Castlla León 0.5839 1.35-0.8752-2.04 0.3379 1.16-0.5008-1.84 Cataluña 0.6279 1.41-0.2638-2.29 0.5208 1.35-0.4553-2.06 Extremadura 1.1119 1.01-0.0242-1.90 0.8071 1.38-0.2529-1.87 Galca 0.4978 1.13-0.3957-1.99 0.2577 0.49-0.4569-2.26 Madrd 1.0609 1.06-0.9277-2.08 0.4127 1.18-0.6261-2.78 País Vasco 0.5818 1.24-0.8396-2.87 0.0953 0.23-0.7922-2.46 Comundad Valencana 0.3563 0.75-0.5315-1.84 0.3380 1.11-0.4381-2.54 Number of observatons 984 552 Rho 0.3690 0.3161 Log lkelhood -702.3350-460.4645
25 Table C.2. Margnal probabltes. Men wth dsabltes Men (1) (2) (3) (4) (5) Baselne (BL) 0.3235 0.4539-0.4031 Dsablty: Vertebral column 0.4843 47.91 0.5748 26.63-0.1868 Upper and low extremtes 0.3437 6.24 0.5589 23.13-1.5261 Seeng 0.4633 43.21 0.4639 2.20-0.0013 Hearng 0.4449 37.53 0.5985 31.86-0.3446 Communcatng 0.4163 28.69 0.6378 40.52-0.5321 Rememberng 0.4314 33.35 0.6058 33.47-0.4043 Executng orders 0.4426 31.68 0.5685 25.25-0.2845 Mantanng body postures 0.4092 26.49 0.4664 2.75-0.1398 Movng nsde home 0.5066 56.60 0.4656 2.58 0.0809 Usng hands and fngers 0.4045 25.04 0.5660 24.70-0.3993 Movng outsde home 0.4813 48.78 0.6076 33.86-0.2624 Takng care of oneself 0.5309 64.11 0.5325 17.31-0.0030 Houseworkng 0.3382 4.54 0.6659 46.71-0.9689 Relatng 0.6414 98.26 0.6708 47.79-0.0458 Socoeconomc characterstcs Age between 55-59 0.3503 8.28 0.5088 12.09-0.4502 Change of house 0.4736 46.40 0.6946 53.03-0.4666 Imparment certfcate 0.3692 14.12 0.7255 59.84-0.9643 Needs caretaker 0.6226 92.46 0.5259 15.82 0.1552 Rehabltaton treatment 0.3710 14.68 0.5518 25.57-0.4873 Spouse workng 0.3256 0.65 0.3329-28.87 0.0083 Spouse dsablty beneft 0.3394 4.91 0.6148 35.46-0.8114 Marred 0.3325 2.79 0.4932 8.65-0.4833 Elementary and unsklled 0.6244 93.01 0.8829 94.51-0.4139 Hgh school and sklled 0.4838 49.55 0.7951 75.24-0.6434 Hgh school and whte collar 0.4832 49.36 0.7847 72.92-0.6239 College and sklled 0.3654 12.95 0.6958 53.29-0.9042 Man sustaner 0.2560-20.86 0.4881 7.53-0.9066 Unemployed 0.4933 52.49 0.3396-25.18 0.3116 Autonomous Communtes Andalucía-Ceuta-Mellla 0.6197 91.56 0.4509-0.66 0.2724 Aragón-Navarra-Roja 0.6167 96.63 0.3521-24.43 0.4298 Asturas-Cantabra 0.7339 126.86 0.1193-73.72 0.8374 Baleares 0.6091 88.28 0.1306-71.23 0.7856 Canaras 0.5528 70.88 0.1205-73.45 0.7828 Castlla La Mancha 0.6838 111.37 0.1452-68.01 0.7145 Castlla León 0.5502 70.07 0.1608-64.57 0.7077 Cataluña 0.5675 66.15 0.3521-22.43 0.3796 Extremadura 0.7434 129.79 0.4443-2.11 0.4023 Galca 0.5160 59.50 0.3044-32.94 0.4185 Madrd 0.7267 124.64 0.1483-67.33 0.7959 País Vasco 0.5493 69.79 0.1697-62.61 0.6911 Comundad Valencana 0.4595 42.04 0.2586-43.03 0.4372 (1): margnal probablty of deservng; (3): margnal probablty of recevng (2) and (4): rate of change wth respect to baselne; (5)=[(1)-(3)]/(1) BL: man, younger than 55 years old, sngle, college, whte collar, lvng n Murca, sufferng only dsabltes wth moderate severty, wthout rehabltaton treatment, mparment certfcate, caretaker and has not changed of house because of a dsablty. Table C.3. Margnal probabltes. Sample wthout dsabltes Men Women [(1)-(2)]/(1) (1) (3) (2) (3) BASELINE (BL) 0.0130 0.0036 Age between 55-59 0.0283 117.69 0.0096 166.66 0.6608 Man sustaner 0.0149 14.62 0.0044 22.22 0.7047 Marred 0.0113-13.08 0.0020-44.44 0.8230 Elementary 0.0123-5.39 0.0036 0 0.7073 College 0.0137 5.38 0.0038 5.55 0.7226 Whte collar 0.0101-22.31 0.0010-72.22 0.9009 (1) and (2): margnal probablty of recevng; (3)=rate of change wth respect to BL. BL: ndvdual younger than 55 years old, sngle, not man sustaner, hgh school, sklled occupaton and lvng n Murca.