Socal Excluson and the Two-Tered Healthcare System of Brazl 1 Densard Alves Unversty of São Paulo Chrstopher Tmmns Yale Unversty Resumo No Brasl exste um sstema de saúde com dos acessos. Aqueles, com recursos, têm acesso ao sstema prvado de saúde que fornece tratamento, com qualdade quando demandado, enquanto o restante da população tem que ser atendda pelo congestonado sstema de clíncas e hosptas públcos ou atenddos pelo sstema prvado pago com recursos públcos. Dados de amostra de domcílos são utlzados para determnar quas grupos sóco-econômcos dependem do sstema públco de saúde. As tendêncas demográfcas atuas ndcam que mas e mas pessoas vão utlzar o já congestonado sstema públco de saúde nas próxmas décadas. Um modelo estlzado de escolha é estmado e seus parâmetros são usados para se obter evdêncas, a partr de smulações, sobre as conseqüêncas sobre o bem estar do aumento no congestonamento e sobre as polítcas que poderam reduzr seu mpacto, como por exemplo subsídos no uso do sstema partícula de saúde. Palavras-chave sstema de saúde, exclusão socal, máxma verossmlhança Abstract In Brazl, there exsts a two-tered system of healthcare access. Those wth suffcent means have access to a prvate system of healthcare that provdes qualty treatment on demand, whle the remander of the country reles on an overburdened system of publc clncs and hosptals. Household survey data are used to determne whch soco-demographc groups rely most on ths publc healthcare system. Current demographc trends suggest that the publc healthcare nfrastructure wll become more and more heavly used n the comng decades. A stylzed model of healthcare choce s estmated, and ts parameters are used to conduct counterfactual smulatons of the welfare mplcatons of ths ncreased 1 The authors are grateful to the Inter Amercan Development Bank and to the Nemess Project, a PRONEX program for fnancal support to ths study.
congeston, and of polces to offset t, lke prvate healthcare subsdes. Keyword Maxmum-lkelhood, healthcare, socal excluson
1 1. Introducton In Brazl, there exsts a two-tered system of healthcare. Those wth suffcent means, or whose employers provde health coverage, have access to a prvate system of healthcare that provdes qualty treatment on demand. The remander of the country, conversely, reles on a system of publc clncs and hosptals. As s the case wth most publc healthcare systems around the world, the Brazlan system s characterzed by long watng tmes, wth the practcal mplcaton that those who are forced to rely on t spend more tme beng sck and, subsequently, have a dmnshed health stock. Ths two-tered system of healthcare s a partcularly relevant concern n Brazl n lght of recent changes n the country s soco-demographc structure. In 1990, only 6.7% of Brazl s populaton was over age 60, but by 2010 ths s expected to be 9.7% and by 2030, 16.9% [World Bank (1994), (2000)]. Durng the last twenty years, famly szes amongst the poorer segments of Brazlan socety (.e., those who typcally rely most on the publc provson of healthcare), have been larger than n wealther segments of socety. Ths large populaton group has been agng, and s nearng a tme when ts healthcare needs wll grow rapdly [Cutler and Meara (1998)]. Concerns have been rased that the Brazlan publc health system wll not be up to meetng ths growng demand. In partcular, already-long watng tmes for treatment wll contnue to grow, wth the practcal mplcaton that many of the poorest segments of socety wll receve no healthcare at all. Ths mechansm of socal excluson of the poor, elderly, and rural populaton wll ncrease at the rate at whch ths segment of the Brazlan populaton s growng. Ths mechansm may have long-run feedback effects as well. Growng demand due to the ncreasng sze of the poor, elderly populaton, as well as the ncreasng cost of treatment for a lmted supply of publc health servces, wll mean that the poorest segments of Brazlan socety wll begn to lose access to healthcare. Ths wll result n a declnng health stock for the poor, renforcng ther soco-economc poston. To the extent that the poor contnue to have larger famles (e.g., as a retrement-nsurance mechansm or a source of labor for subsstence agrculture), ths wll lead to further strans on the publc healthcare system n the future and, lkely further falures. Ths paper seeks to accomplsh three tasks. Frst, we wll use an outstandng set of Brazlan household survey data to characterze whch socal groups have access to prvate healthcare. In partcular, we employ the 1998 PNAD, a broad household survey that ncludes detaled nformaton on health, healthcare consumpton, and, most mportantly, the source from whch one receves health servces. We wll argue that certan groups are systematcally dened access to prvate healthcare n an ndrect fashon. After dentfyng whch groups are subject to ths form of excluson, the second part of our research wll construct and emprcally dentfy a stylzed model of choce between alternatve sources of healthcare provson, from whch we wll be able to derve a crude measure of the welfare consequences of the ncreased healthcare congeston costs that are lkely to accompany the demographc transton we currently see n
2 Brazl. Wth these welfare conclusons, n the fnal part of our research, we wll be able to analyze the mplcatons of polcy alternatves such as prvate healthcare subsdes. Secton 2 of ths paper descrbes descrbes the household survey data we wll use for our analyss. Secton 3 outlnes our methodologcal approach for demonstratng that ndrect excluson exsts and measurng ts consequences, and reports how the mplct prce of publc healthcare vares across soco-demographc groups n Brazl. In Secton 4, we carry out two counterfactual smulatons. In the frst, we examne the mpacts of an ncrease n the mplct prce of publc healthcare, lke that whch would arse from the ncreased congeston that would accompany the predcted ncreases n healthcare demand n Brazl. In the second smulaton, we consder the mpact of a prvate healthcare subsdy beng provded by the government, makng that opton more accessble to groups who had prevously been able to afford only publc healthcare. Gven the propensty of ndvduals to swtch ther source of healthcare provson, we fnd that such a polcy mght only result n rent transfers to those segments of socety that we would not consder to be excluded. Secton 5 concludes by suggestng lmtatons to, and possble extensons of, ths research. 2. Data The 1998 PNAD s an annual household survey on soco-economc condtons of the Brazlan populaton under the responsblty o9f the Insttuto Braslero de Geografa e Estaststca (IBGE).The 1998 PNAD had a specal supplement dealng wth the health condtons of the Brazlan populaton. The supplement ncludes detaled nformaton on health. The survey covers 344,975 ndvduals and 98,166 households 2. In the analyss, however, the number of observatons wll be smaller due Use of ths survey data presents a number of advantages. Frst, t s one of the only surveys to collect data on health of the populaton n a consstent fashon. Second, data collecton on health come wth the full set of soco-economc varables on the ndvduals and on the household. Stll, the collecton of such data reles on the tranng of data enumerators to record health status. Thus, when lookng at reported llnesses, the accuracy mght not be as hgh as t would be f medcal professonals were examnng the ndvduals and reportng ther llness. The PNAD data set also presents some dsadvantages -- t s restrcted to the urban sector n the northern regon of Brazl and s, therefore, not representatve of the whole northen regon 2 In the analyss the number of household observatons s smaller due to mssng data n some varables. Used n the analyss. But even excludng mssng data from the analyss due, for nstance, to mssng data for household ncome stll permts the use of close to 90,000 observatons on households.
3 of Brazl. However, the data set covers pretty well the remanng regons of Brazl and the excluson of the rural north does not harm the representablty of the PNAD survey to the extent that the rural populaton n the north s very scarce. In our analyss we wll use the household data set nstead of the larger ndvdual sample data. The reason s that decson on health s a major decson makng process and s assumed to be a household decson. The analyss wll take care of mportant characterstcs of ndvduals wthn the household, as for nstance, takng nto account the proporton of chldren and the elderly people wthn the household because they are mportant aspects for the decson makng process on health. Table2 shows some major characterstcs of the data set. Varable Wth_HP s the proporton of the household heads who has a prvate health nsurance pad by hmself or by hs employer. It amounts to 17% of the total household heads. The remanng households use the publc health system. Varable PrvHP gves the household who pay ther own health plan. Whopay shows the proporton of people havng employer-provde prvate health nsurance. The average household ncome s R$ 962.50 monthly. The average payment of health nsurance for the 6.9% who pay ther own medcal nsurance s R$ 150.31 and ther average household ncome s R$ 2137.50 and R$ 781.19 of household per capta ncome. Income reported n Table 2 s the average household per capta ncome. Households who pays ther own health plan have per capta ncome more than twce as much as the mean household per capta ncome estmated wth the household sample data. The Metro varable gves the proporton of the household heads lvng n Brazlan metropoltan areas. Urban gves the proporton of household heads lvng n urban areas. Age60 s varable that specfyng the proporton of people n the household above 60 years of age. Varable f_knd2 gves the proporton of people bellow age 14. The proporton of the households that dd not report any knd of llness s 49.1%. Close to 51% reported some knd of llness and some of than reported as beng stroke by two or more llness at the same tme. The proporton of people lookng for any knd of medcal treatment n the last two weeks s 13.1%. Table 1 was bult usng the evaluaton of the qualty of the health system usng the perceved qualty of attendment of those 13.1% of the households who looked for medcal treatment n the last two weeks. The reported llness are self explanatory and they are the llness reported by the head of the household. The set of race-dfferentaton varables are Whte, Mxed, Black and Yellow. Yellow are manly the Asan descended household heads. The proporton of blacks s qute small, however, a large part of those people nclude n the mxed racal group are actually black people. The characterstcs of the prvate health plans acqured by 6,639 head of households are presented n the bottom part Table 2.These characterstcs are used n the estmaton of the shadow prce for publc health plans n a hedonc prce regresson wth a Heckman correcton and the results are presented n Table 5. All the attrbutes, are defned by a set of dummy varables permttng thus a complete descrpton of the type of prvate health nsurance avalable n Brazl today. The varable plcons s one when the health plan covers doctors appontment and zero otherwse.
4 The value of.9804, presented n Table 2, means that almost all health nsurance bought by head of households coves vsts to doctors. Pllst s one when the health nsurance polcy present a lst of authorzed doctors, hosptals and laboratores that can be used by the polcy holders and s zero otherwse. 92.47% of the prvate health nsurance polces present a lst of authorzed doctors, hosptals and laboratores. The value for varable plreemb ndcates that 30.99% of the health plans permt rembursement of medcal expenses when the ndvdual s attended by doctors or health centers not afflated to the health plan. Varable called plother ndcates that 81.15% of the polcy holders can be attended by doctors, hosptals and laboratores n ctes others than the one they resde. pldent ndcates that only 21.94% hold health nsurance coverng dental treatment. It s possble to observe that ths attrbute s not a wdespread characterstc of the prvate health nsurance. Plans havng ths attrbute are more expensve than the ones that do not have t. Paymore s a varable capturng the fact that some health nsurance polces mpose a lmt n total medcal expenses and anythng above ths lmt would have to be pad by the polcy holder 3. Varable plexam ndcates that 95.60% of the prvate health plan allow polcy holder to take complementary lab exams durng treatment.92.83% of the prvate polcy holders are allowed to be covered for hosptalzaton. Ths facet of the prvate health plan s ndcated by varable plnter. Varable platend ndcates that 80.54% of polcy holders are allowed to be attended by medcal servces under contract wth the heath nsurance company. Very few health plans cover the acquston of medcnes and drugs. Plmedc ndcates that ths attrbute s very specal and covers only 4.85% of the prvate health nsurance holders. Among the health nsurance holder, only 3.07% allow dental treatment. Ths aspect of the health nsurance s represented by varable odonto n Table 2. 3. Research Methodology In order to characterze ndrect excluson from prvate healthcare n Brazl, we adopt a two-pronged methodologcal approach. Frst, we employ the detaled survey nformaton n the 1998 PNAD data set, descrbed n Secton 2, to determne, generally, whch groups n Brazlan socety have access to prvate health nsurance and whch rely on publc healthcare. Beng relegated to publc healthcare s not a drect form of excluson, but rather one based on relatve prces for prvate and publc healthcare that may be dfferent for ndvduals from dfferent segments of socety. Moreover, dfferences n employment patterns for ndvduals from dfferent soco-economc groups wll nfluence ther access to employer-provded prvate health 3 It s the only attrbute among the 11 attrbutes where the dummy varable assumes the value of one when t ndcates a detrmental characterstc of the health plan.
5 nsurance. 4 3.1 Publc v. Prvate Healthcare and Brazlan Soco-Demographc Groups (3.1.1) Py ( = 1) =Φ ( X β ) ', j We expect certan groups, based on race, educaton, and locaton n Brazl, to be systematcally m ore relan t on the publc healthcare system. Frst, we nvestgate whch groups fall nto ths category wth a smple Probt regresson [Greene (2000)] of the followng form: where age y = form of healthcare coverage for ndvdual (1 = prvate, 0 = publc) X = soco-economc attrbutes of ndvdual ; these nclude - Race (Black, Asan, Mxed, Whte) - Age - Educaton s defned by years of schoolng. - Household Income - Regonal Indcators: - Percentage of Persons n the Household wth less than fourteen y ears of - Percentage of People above 60 years of Age - Household Income: the total sum of wage and other types of ncome of ndvduals lvng n the household. 4 We avod ths dffcult ssue (.e., health nsurance as an attrbute of a job for whch an ndvdual may or may not face a correspondng reducton n pay) by consderng only those ndvduals who ether buy prvate nsurance drectly (.e., those who do not receve t through an employer) or use the SUS.
6 - Mgrator s one f the ndvdual s outsde the place where he or she was born and zero otherwse. - f_knd2 s proporton of people bellow fourteen years of age lvng n the household. - Age60 s the proporton of people the age of sxty years lvng n the household. The results of ths regresson, whch are found n Table 3, correspond to general perceptons about Brazlan healthcare. Those who tend to be more relant on the publc system are less educated, male, come from the Black and Mxed racal groups and from the Northern, Center western regon of Brazl, have lower ncomes, and are old people. The presence of people above sxty years of age s hghly sgnfcant and the presence of people bellow fourteen years of age does not make a dfference n terms of makng the household gong nto the prvate health plan. Gven the results descrbed n Table 1 regardng dfferences n the qualty of healthcare across provders, ths alone could be consdered evdence of excluson of these groups. Table 4 provdes addtonal evdence along these lnes. Specfcally, t presents the results of a number of Probt regressons n whch a dummy varable ndcatng that an ndvdual has suffered from a partcular dsease (e.g., depresson, arthrts, cancer, dabetes, respratory alment, hypertenson, cardac dsease, tuberculoss, crrhoss, tendnts, and kdney dsease) s regressed on a set of ndvdual attrbutes, ncludng the form of healthcare provson (.e., SUS v. prvate) that the ndvdual uses. The dea here s that an ndvdual s health stock, whch determnes how lkely he s to suffer from any of these alments, s, n part, determned by the qualty of the healthcare he receves. An ndvdual who reles on the publc system mght, therefore, receve lower qualty care, or less care n general (f watng tmes for treatment are worse), leadng to a lower health stock and a hgher lkelhood of dsease. These results should be nterpreted wth extreme cauton, however, as we would suspect the form of healthcare provson to be smultaneously determned wth the ndvdual s health stock; e.g., an ndvdual who knows he s lkely to develop cancer mght purchase prvate health nsurance n order to guarantee hmself a hgher qualty of care. The presence of an endogenous varable n a Probt regresson can potentally lead to nconsstent estmates of all the model s parameters. The results, however, are very much consstent wth a pror expectatons about health and about the Brazlan health system Women, generally, are less lkely to suffer from almost all dseases but Crrhose. Women seems to be healther than man. Hgher ncome follows good health, meanng that poor people are more lkely to suffer from some of the llness defned for the PNAD survey. One mportant pont s that people wth prvate health plan- ether self pad or employer pad -are more lkely to suffer from some of the llness, whle healther people are less lkely to go to a prvate pad health plan The llness seems to strke more the Northeastern regons. People from the Southern and Southeastern regons are more
7 lkely to report cardac, cancer, depresson and respratory dseases. 3.2 A Model of Indvdual Healthcare Choce Whle descrbng whch elements of Brazlan socety are more lkely to rely on publcly provded healthcare, the precedng analyss does not provde any way of measurng the welfare consequences of ths ndrect form of excluson. In order to do so, we need to develop a more elaborate model that takes nto account the fact that ndvduals optmally choose what form of health nsurance to obtan n the face of market prces and a budget constrant. The second part of our emprcal analyss develops such a model. The bndng constrants on the scope of the conclusons that can be taken away from ths model come from the lack of data descrbng ndvduals full ncome endowments and actual expendture patterns on health and non-health commodtes over tme. Instead, the model takes a stylzed vew, descrbng the ndvdual s choce of health coverage as a choce between alternatve types of nsurance n a statc context. To the extent that ndvduals change health nsurance status durng the course of ther lfe, ths may bas our answers. In partcular, we assume that ndvdual chooses that form of healthcare provson n order to maxmze a utlty functon of the form: (3.2.1) α U( C, H )= C H 1 α subject to a smple budget constrant: (3.2.2) H C + P H = I C represents s consumpton of a composte numerare commodty, H represents the consumpton of effectve healthcare servces (the prce for these servces, P H s allowed to dffer by ndvdual, and to reflect the qualty of the nomnal healthcare consumed), and I represents the ndvdual s ncome. P H wll also dffer accordng to the form of healthcare provson chosen;.e., P S for SUS healthcare and P P for prvately provded healthcare. The chef source of dffculty n ths analyss s that P S s not observed (all SUS healthcare s nomnally free), but s rather only a shadow prce on SUS healthcare consumpton. Utlty maxmzaton subject to ths budget constrant yelds the followng ndrect utlty functon: (3.2.3) ( α ) H V( P, I ) = I ( 1 α ) I H P α 1 α whch dffers by whether the ndvdual chooses SUS healthcare (V(P S,I )) or prvate healthcare (V(P P,I )). Takng the optmal allocaton of ncome between composte
8 consumpton and healthcare as gven, ndvdual s choce between the two forms of healthcare provson can be modeled as a comparson of these two ndrect utlty functons. In partcular, ndvdual wll choose SUS healthcare as long as: (3.2.4) S P V( P, I ) V( P, I ) Because of the smple functonal forms that we have employed, ths bols down to P S # P P. The prce of an effectve unt of publc healthcare s not an observed magntude; nomnally, publc healthcare s free to everyone n Brazl. It has a prce, however, n the form of tme n and dsutlty of crowded watng rooms, etc... (see dscusson n Secton 2). We would expect ths prce to dffer across ndvduals accordng to ther opportunty cost of tme, preferences for cleanlness, and dsutlty of congeston;.e. dfferences for whch we hope to be able to control wth a set of observable ndvdual attrbutes (X ). The avalable data allow us to recover each ndvdual s shadow prce for an effectve unt of publc healthcare by usng nequalty (4.2.4). Once we have done so, we wll have all the tools necessary to consder the welfare mpacts of an ncrease n the congeston costs assocated wth recevng health servces from the SUS. In partcular, assumng that the ndvdual chooses the healthcare opton that maxmzes hs ndrect utlty (wth the ndvdual s percepton of the qualty dfference between publc and prvate provson factored nto that prce), prvate healthcare wll be chosen f P P # P S. P P s observed n avalable data. 5 We parameterze the natural logarthm of P S as a lnear functon of ndvdual attrbutes (X ) and an unobservable determnant (, ), whch s assumed to be dentcally and ndependently normally dstrbuted wth a unt varance and zero mean. The choce of prvate health coverage s then determned by the followng condton beng satsfed: P X β + ε.2.5) ln p p 5 In partcular, we observe n PNAD data the prce of the prvate health nsurance pad for everyone who opted for that form of coverage. We mpute prvate health nsurance premums for the rest of the sample by (1) regressng the observed prvate premums on ndvdual attrbutes and attrbutes of the polcy, (2) controllng for the selecton nto prvate healthcare provson wth a Heckman-correcton term- the results of the Heckman procedure s reported n Table 5-, and (3) fttng premums for all ndvduals for a standardzed polcy. In partcular, the standardzaton we adopt sets all of the attrbutes of the healthcare polcy to ther smplest values --.e., to gve the prce of a polcy wthout any bells or whstles. Ths creates a level playng ground for comparson of the ndvdual s decson between publc and prvate coverage.
9 whch wll be the case f: (3.2.6) ' ln P p p X Xβ ε whch occurs wth probablty 1 - Μ(ln P P - X! ). Smlarly, the probablty that ndvdual chooses publc health coverage s gven by Μ(ln P P - X! ). We can therefore wrte the lkelhood of observng all of the health coverage choces of the ndvduals n the data set (y ), gven ther observable attrbutes (X ) and prvate healthcare prce (P P ) as: (3.2.7) L( y, X, P ; β) = Π[ Φ(ln P X β)] Π[1 Φ(ln P X β)] p p ' p ' y = 0 y = 1 Ths lkelhood functon s maxmzed over the parameter vector,, usng data descrbng the decsons and attrbutes of a 10% subsample of household heads n the PNAD. The use of only household heads elmnates the correlaton n nsurance type between members of a household that exsts n the full data set. Elmnatng data wth mssng observatons for some varables, ths yelds a sample sze of N = 8267. Coeffcent estmates and standard errors are reported n Table 6. Parameter estmates generally have the expected sgn, and tend to be statstcally sgnfcant. Those who we would expect to have greater dsutlty from congeston, etc... (.e., from havng a greater opportunty cost of tme) face a hgher mputed prce for SUS healthcare. Ths s true of older and more educated ndvduals, although once ndvduals are over the age of 60 (.e., when they begn to retre), ther mputed SUS healthcare prce falls. Indvduals wth hgher ncomes face a hgher prce, also because of a greater opportunty cost of tme, and urban ndvduals face a greater cost than rural ndvduals, possbly because congeston problems are worse n the ctes. Indvduals n the South, Southeast, and Center-West regons of Brazl face hgher prces than those n the North and Northeast, and Blacks and those n the Mxed racal category face lower prces than Whtes, whle Asans face hgher prces. As a measure of model ft, we can compare the predcted health coverage decsons of ths model wth the decsons observed n the data. The model does well, correctly predctng the choces of 87% of all ndvduals. When the model fals to predct correctly, t tends to be n the case of ncorrectly forecastng the choces made by those ndvduals who opt for prvate health coverage;.e., hgh-ncome, more educated, and older (younger than 60 years) ndvduals. 4. Results and Polcy Analyss
10 4.1 Analyzng Welfare Effects of a Change n the Prce of Publc Healthcare The ntal goal of ths research was to determne whch groups n Brazlan socety would suffer the most as a result of the ncreasng congeston of the publc healthcare nfrastructure that wll lkely accompany the soco-demographc trends we are currently observng. In order to measure the welfare cost of ncreased watng tme for publc health provson, whch mght result from an ncrease n the number of elderly Brazlans relyng on the SUS wthout a correspondng ncrease n supply, we need only consder the effect on dfferent ndvduals n the sample f the prce of publc healthcare were to ncrease (e.g., by 50%), takng nto account the optmzng nsurance decson each person makes to ths prce ncrease. Many ndvduals who had chosen publc healthcare, for example, mght stck wth that choce and bear the brunt of the prce ncrease, whle others mght fnd t optmal to pay more and swtch to prvate healthcare. Those who had chosen prvate health coverage pror to the prce ncrease would experence no change n prce or dsposable ncome. We smulate the decsons of each ndvdual n the data set, backng-out the overall change n the prce of recevng healthcare he or she faces after all s sad and done. Fnally, we consder the dfference n the natural logarthms of the prces ultmately faced by each ndvdual, before and after the prce change. Ths measure provdes us wth a proportonal measure of the compensatng varaton n ncome needed to mantan the same level of utlty: (4.1) ' ( ) ' 1 H H ln P ln P = ln I ln I 1 α where P H! and I! represent the prce of healthcare provson and the accompanyng requred level of ncome needed to reach the orgnal level of utlty, after the ncrease n the prce of SUS health coverage. Note that we cannot calculate the compensatng varaton n ncome drectly, because we are unable to determne for each ndvdual. Ths results from the fact that we do not observe an ndvdual s full ncome endowment (.e., an endowment ncludng the value of avalable tme, etc...) rather, we only see the ndvdual s monetary ncome, whch s not expended at all f SUS healthcare s employed. Ths means that t s mpossble to ultmately determne whether the dfference n log prces s attrbutable to a compensatng varaton n ncome, or to heterogenety n preferences. For the followng dscusson, we assume the former. In order to quckly summarze the welfare mplcatons of an ncrease n the prce of SUS healthcare, lke that whch would accompany ncreasng congeston of that system, we regress ths proportonal measure on a vector of soco-demographc attrbutes, we are able to determne whch groups n Brazlan socety wll suffer the most. The dfference n
11 magntude of the effect across groups s somethng that we could not uncover from the smple Probt analyss descrbed n Secton 4.1, because that analyss dd not descrbe how dfferent types of ndvduals behavors would change n response to a prce change. In all, the model predcts that 6.6% of all ndvduals consumng publc healthcare pror to the prce change would swtch from publc to prvate health coverage n response to ths smulated prce ncrease. Accountng for optmzng responses s therefore mportant. The results of ths regresson appear n Table 7. Those n the South (.e., the excluded regon) fare worse than those n the rest of Brazl, especally the Center-West and Southeast. Blacks and those n the Mxed racal group fare worse than Whtes, whle Asans generally do better (owng to ther greater predsposton to have been usng prvate healthcare before the prce ncrease). Older ndvduals do better (as they are also more lkely to have been usng prvate healthcare), untl they reach the age of 60, at whch pont they generally rely more on publc healthcare and do much worse. Men generally fare worse than women, whle those wth more educaton and hgher levels of ncome do better n the face of rsng SUS prces, agan reflectng predspostons towards usng prvate health coverage. Interpretng the results of ths welfare analyss are complcated by the fact that we do not know whether the magntudes we observe are dfferences n the level of ncome requred to reach the orgnal level of utlty, or whether they smply reflect dfferences n ndvduals preferences for healthcare consumpton (.e., ). Assumng that I s the level of household ncome reported n the PNAD survey, however, we can calculate expendture shares for health coverage based on the observed prces for prvate healthcare and the mputed prces for SUS healthcare. We do so for each ndvdual n the 10% sample used above, and use them to calculate explct measures of the compensatng varaton n ncome requred to offset the ncrease n the prce of SUS healthcare smulated above. Table 8 shows how these CV measures vary wth observable soco-demographc attrbutes. The results dffer n some ways from those n Table 6. In partcular, ncreasngly educated ndvduals are worse-off, except for the hghest educaton group. Blacks, Asans, and those n the Mxed racal group all fare better than Whtes, and older ndvduals (both above and below the age of 60) fare worse. The most strkng dfference between these results and those presented n Table 6, whch calls nto queston the valdty of the assumpton about ncome, s that wealther ndvduals are made worse-off than poorer ndvduals by the ncrease n the SUS prce t s precsely these ndvduals that we would expect to be able to swtch more easly to prvate health coverage, f they were not usng t already, n response to an ncrease n the prce of publc coverage. To the extent that these ndvduals could not swtch, however, we mght expect that they would suffer most from ncreased congeston, gven ther hgher opportunty cost of tme. 4.2 Analyzng the Welfare Effects of a Prvate Healthcare Subsdy
12 The apparatus developed above also allows us to consder the mplcatons of counterfactual polces desgned to offset ncreasng congeston n the provson of publc healthcare, where a smple reduced-form analyss, lke that descrbed at the start of Secton 4.1, cannot. In partcular, we can consder the mplcaton for ndvduals optmzng choces of a prvate healthcare subsdy, desgned to expand the ndvdual s budget constrant only f the ncome s used for the purchase of prvate healthcare. The welfare mplcatons of such a polcy could then (wth better data descrbng the full ncome endowment) be compared to the mplcatons of a smple ncome subsdy that could be used for any sort of consumpton, ndcatng the value of a relatvely paternalstc polcy. In the absence of such data, we consder the welfare consequences of a smple 50% prce subsdzaton of prvate healthcare (.e., the government pays 50 cents on every 1 real spent by the ndvdual on prvate healthcare). In order to descrbe how the resultng welfare gans (agan, a proportonal measure of the compensatng varaton n ncome, assumng homogenous preferences, after the optmzng provder of healthcare s chosen) dffer across soco-demographc groups, we regress them on a vector of soco-demographc attrbutes. The results of ths regresson are descrbed n Table 9. Negatve numbers descrbe reductons n ncome that return ndvduals to ther orgnal levels of utlty (.e., ndcatng a beneft). Indvduals n the Center-West and Southeast regons seem to beneft most from ths prce subsdy, whle those n the North and Northeast beneft only margnally more than those n the South. Whtes and Asans beneft more than Blacks and those n the Mxed racal group, and those wth hgher levels of educaton beneft more than those wth less educaton. Smlarly, rcher ndvduals, women, and older ndvduals (under the age of 60) beneft more from the subsdy. Generally, these relatve benefts reflect a greater predsposton towards (or propensty to swtch to) prvate healthcare provson. The natural queston s how mght such a prvate healthcare subsdzaton polcy be targeted to beneft those ndvduals who would suffer most under an ncrease n the prce of publc care. To the extent that such subsdes, when appled broadly, seem to beneft hghncome, hgh-educaton ndvduals, they smply represent a transfer of rents, snce those ndvduals suffer less under the ncreasng prce of publc care than do the poor anyway. One possble opton would be to mplement the subsdy as part of an ncome tax collecton regme, where partcpaton crtera could be easly be establshed so as to allow the subsdy to be used only by low ncome resdents. Problems of fraud n the reportng of prvate healthcare expendtures mght make ths approach dffcult, however. Instead, we mght 6 The model predcts that approxmately 13.3% of all ndvduals consumng publc healthcare pror to the prce subsdy would swtch to prvate health coverage. Wth such a large ncrease n the demand for prvate healthcare, the government mght want to undertake polces to facltate entry by new prvate healthcare provders n addton to the subsdy, so as to avod new congeston costs.
13 focus on the results of the frst counterfactual smulaton, and target subsdy funds geographcally so as to reach those ndvduals who both lose most under the smulated ncreases n publc prces. Such s true, for example, of Black and Mxed race resdents of the South, Southeast, and Center-West, partcularly those wth low levels of educaton. Generally, these regons are vewed as beng the developed regons of Brazl, and conventonal wsdom mght suggest that more would be ganed by targetng resources to the less developed North and Northeast. The results of ths analyss, however, suggest that t s these excluded groups lvng n the developed parts of Brazl that would beneft the most from nterventon to make prvate healthcare more affordable. Fnally, dealng wth the mpacts of rsng publc healthcare costs on the elderly (a major concern gven current soco-demographc trends) by subsdzng the consumpton of prvate healthcare seems more problematc, snce those over the age of 60 are predcted to beneft less than most other groups from ths polcy. Ths arses from the model s predcton that members of ths group are not as lkely to swtch to prvate health coverage even wth the change n relatve prces. Indeed, n order to lmt the adverse effects on ths group of rsng congeston n publc healthcare consumpton wthout affectng huge rent transfers to those who are less adversely mpacted, the government wll lkely have to take steps to drectly ncrease the supply of publc healthcare provson. 5. Conclusons and Extensons The goal of ths analyss was to determne whch groups n Brazlan socety were most excluded from prvate healthcare. Prvate healthcare s generally consdered to be of a hgher qualty level; ths percepton s generally supported by the PNAD survey data. Such excluson s not of a drect form as would be racal excluson from a club, but s rather based on ndvduals facng dfferent relatve prces for publc and prvate healthcare owng to dfferences n ther observable attrbutes and preferences for healthcare consumpton. Our ntal analyss of PNAD survey data documents what s generally perceved to be the case -- that poor, rural, Black and Mxed-race Brazlans tend to rely more on publc healthcare. Ths alone would not necessarly represent a source of socal nequty, except that we expect the prce of ths form of healthcare to ncrease n the comng decades owng to the ncreasng congeston of an already-overburdened system, and we expect these groups to suffer most. In order to determne how these prce ncreases would be dstrbuted over dfferent soco-economc groups, we need a more elaborate model of optmal ndvdual decson-makng;.e., a model that allows us to determne how ndvduals would behave 7 Reachng these partcular racal groups mght be dffcult, unless the subsdes took the form of mones to establsh new prvate healthcare facltes n racally segregated neghborhoods.
under current and counterfactual relatve-prce scenaros. Operatng under constrants of data avalablty, we assumed that each ndvdual was requred to consume a sngle unt of some form of healthcare coverage (.e., publc or prvate), and that dfferences n the qualty of care across forms would be nternalzed n the prce confrontng the ndvdual. Dfferences n prce mght also arse from observable ndvdual attrbutes (.e., a drect form of dscrmnaton), or from an ndvdual s preferences for healthcare consumpton (e.g., ndvduals wth strong preferences for healthcare consumpton mght face an even hgher prce for an effectve unt of publc care than a smlar ndvdual who had weak preferences for healthcare consumpton), but avalable data do not allow us to dentfy these effects. From a smple and stylzed model of utlty maxmzaton, we were able to recover estmates of the prce of publc health coverage, and used those estmates to nfer whch socodemographc groups would suffer most from an ncrease n the congeston of the publc healthcare system. Whle the conclusons of ths analyss conform to the general perceptons regardng race, educaton, and ncome groups, they suggest that the groups most at rsk from an ncrease n the prce of SUS healthcare would be the excluded racal and educaton groups n the southern half of the country, whch, whle developed, exhbts a great deal of socal and ncome nequalty. Wth even more detaled data on the attrbutes of the alternatve forms of healthcare provson, we mght also be able to buld a more realstc hedonc model n whch ndvduals wth weak preferences for healthcare would choose to consume the type that exhbts low-levels of amentes and a low prce, whle those who derve a great deal of utlty from the consumpton of healthcare mght choose a deluxe form of healthcare provson. Ths could be mportant n predctng how dfferent ndvduals would respond to an ncrease n the congeston of the publc system, whch would ncrease watng tmes for treatment (.e., a specfc trat of the healthcare commodty). We mght fnd, for example, that certan soco-demographc groups exhbt a strong dstaste for watng tme, and they would thus tend to bear more of the burden of ncreasng congeston of the SUS. Other survey data (e.g., the 1997 PPV) provde some ndcaton of watng tme ncurred n the recept of healthcare servces, but these data exhbt many mssng observatons, and t s unclear whether they wll be approprate for such an analyss. Even wth the lmtatons and smplfcatons descrbed above, the current model s suggestve of whch groups are most lkely to suffer from the ncreasng congeston of the publc healthcare nfrastructure that s lkely to accompany current demographc trends n Brazl. From an equty perspectve, these groups are generally those about whom we are most worred, suggestng that some polcy (.e., subsdzaton of prvate healthcare or the expanson of the publc nfrastructure) must be undertaken. Whch specfc polcy response to use depends upon whch partcular group we are most tryng to help. 14
15 Table 1 Perceptons of Healthcare Qualty by Type Regon Health % Who Reason for falng to receve healthcare (for those seekng Care Receved healthcare durng prevous two weeks) Type Health 0 = SUS 1=Prvate Care Sought No Vacancy No Attendng Doctor No Attendng Expert Malfunctonng Equpment Had To Wat Too Long Other 0 93.6 31.1 53.3 4.4 4.4 4.4 4.4 North 1 97.7 0 50 0 0 25 25 0 93.8 39.5 29.7 10.3 4.9 5.4 10.3 NE 1 98.2 15.4 7.7 15.4 0 7.7 53.9 0 94.1 50 22.1 8.8 1.5 7.4 10.3 CW 1 96.8 50 40 10 0 0 0 0 95.4 43.1 28.8 13.1 5 3.8 6.3 SE 1 98.8 52.9 23.5 0 0 0 23.5 0 94.7 61.6 17.2 3 0 4 7.1 South 1 99.1 33.3 16.7 16.7 0 0 33.3 0 94.6 45.6 28.5 9.3 3.6 4.9 8 All 1 98.5 36 24 8 0 4 28
16 Table 2 Data Summary Household Heads N = 98166 Varable Mean Varable Mean Sex 0.727 Age 44.524 Educ 6.612 Age60 0.176 Income 338.591 f_knd2 0.284 Black 0.068 PrvHP 0.069 Whte 0.526 whopay 0.101 Mxed 0.400 Arthrt 0.148 Yellow 0.004 Cancer 0.004 Urban 0.831 Cardac 0.074 Metro 0.413 Crrhose 0.003 dcwest 0.109 Backache 0.312 dseast 0.342 Depress 0.078 dsouth 0.178 Dabets 0.037 dnorth 0.068 Hpert 0.197 dneast 0.302 Kdney 0.047 Wth_HP 0.170 Respr 0.042 Mgrator 0.009 Tendon 0.032 Value 79.474 Tuberc 0.002 Attend 0.131 Healthy 0.491
17 Characterstcs of Prvate Healthy Plan N=6639 plcons 0.980 plexam 0.956 pllst 0.925 plnter 0.929 plreemb 0.310 platend 0.805 plother 0.811 plmedc 0.045 pldent 0.219 odonto 0.031 paymore 0.178 Table 3 Probt Regresson N = 82900, Log Lkelhood = -16753.661 Varable Estmate Standard Error Varable Estmate Standard Error Sex 0.0051 0.0173 Dseast 0.3056 0.0345 Educ 0.1173 0.0020 Dneast 0.0823 0.0356 Income 0.0003 0.00001 Dsouth 0.0411 0.0374 Black -0.4785 0.0855 Age 0.0233 0.0008 Whte -0.1862 0.0785 Age60-0.2696 0.0290 Mxed -0.4284 0.0799 f_knd2-0.0195 0.0206 Dcwest -0.0668 0.0409 Constant -3.3594 0.0964 Table 4 Probt Regresson Determnants of Dseases
18 Dseases Arthrt Cancer Cardac Crrhose Backache Depres Varable Sex -0.3636 (29.867)* -0.0801 (-2.074)** -0.2590 (-18.250)* 0.3099 (5.621)* -0.1864 (-18.116)* -0.5768 (-42.920)* Age 0.0323 (52.069)* 0.0168 (7.935)* 0.0270 (36.155)* 0.0096 (4.333)* 0.0215 (45.530)* 0.0129 (19.772)* Educ -0.0461 (-27.706)* 0.0051 (0.978) -0.0124 (-6.447)* -0.0275 (-4.357)* -0.0368 (-29.233)* -0.0050 (-2.807)* Income -0.00005 (-4.307)* -7.73e-06 (-0.263) -0.00002 (-1.998)** -0.00003 (-0.591) -0.00006 (-7.246)* -0.00006 (-5.194)* Black 0.2166 (2.308)** 3.714 (22.035)* 0.3320 (3.214)* 0.0048 (-0.014) 0.1259 (1.796) -0.1970 (1.789) Whte 0.2207 (2.406)** 3.1977 (25.526)* 0.1884 (1.866) 0.0223 (-0.069) 0.1608 (2.357)** 0.2730 (2.536)** Mxed 0.2545 (2.764)* 3.8188 (25.907)* 0.1967 (1.938)** 0.1196 (0.367) 0.1610 (2.349)** 0.2719 (2.515)** Age60-0.1085 (-5.474)* 0.0302 (0.466) -0.0364 (-1.560) -0.2681 (-3.568)* -0.2775 (-16.240)* -0.1887 (-8.087)* f_knd2-0.0760* (-4.529) -0.1978 (-2.934)* -0.1109 (-5.304)* -0.1288 (-2.320)** -0.0064 (-0.545) -0.1018 (-5.565)* dcwest -0.3241 (-12.958)* 0.2203 (1.935)** 0.0331 (1.035) -0.0873 (-1.053) -0.1318 (-6.310)* -0.0085 (-0.284) dseast -0.5745 (-26.363)* 0.2412 (2.337)** -0.0355 (-1.268) -0.01793 (-2.450)** -0.2450 (-13.413)* -0.0581 (-2.230)** dneast -0.3596 (-17.034)* 0.0751 (0.714) -0.1752 (-6.222)* -0.2243 (-3.123)* -0.0930 (-5.168)* -0.0772 (-2.988)* dsouth -0.3832 (-16.065)* 0.3750 (3.530)* 0.0631 (2.087)** -0.1592 (-1.911)** -0.1983 (-9.918)* 0.0590 (2.094)** Wth_HP -0.0434 (-2.428)** -0.0383 (-0.708) 0.0430 (2.177)** 0.0266 (0.409) 0.0135 (1.011) 0.0002 (0.008) N 95565 95565 95565 95565 95565 95565 Log Lkelhood -32729.57-2274.64-21995.40-1875.49-55542.89-24502.90 Dseases Dabets Hpert Kdney Respr Tendon Tuberc Healthy Varable Sex -0.1779 (-10.124)* -0.2975 (-26.486)* -0.0095 (-0.563) -0.1998 (-12.069)* -0.3358 (-18.811)* 0.0935 (1.484) 0.3300 (31.543)* Age 0.0290 (30.030)* 0.0337 (60.244)* 0.0130 (16.570)* 0.0038 (4.890)* 0.01105 (12.620)* 0.0079 (2.847)* -0.0325 (-70.269)*
19 Educ -0.0023 (-0.973) 0.0150 (-10.564)* -0.0333 (-15.026)* -0.0121 (-5.638)* 0.0060 (2.678)* -0.0248 (-2.877)* 0.0296 (24.692)* Income 0.00003 (2.289)** -0.00002 (-1.885) -0.00009 (- 4.560)* -0.000005 (-0.360) 0.00004 (3.738)* -0.0002 (-1.589) 0.00004 (4.684)* Black 0.1286 (1.150) 0.3107 (4.204)* -0.0305 (-0.266) 0.2334 (1.833) 0.1738 (1.264) -0.1437 (-0.423) -0.2140 (-3.266)* Whte 0.0138 (0.127) 0.0863 (1.181) 0.0148 (0.132) 0.2103 (1.689) 0.1964 (1.468) -0.3173 (-0.957) -0.1921 (-3.024)* Mxed 0.0162 (-0.148) 0.1273 (1.732) 0.0134 (0.120) 0.2115 (1.691) 0.2212 (1.645) -0.2295 (-0.689) -0.1989 (-3.112)* Age60-0.1718 (-5.952)* -0.1925 (-10.469)* -0.1600 (-5.194)* 0.0279 (9.076)* -0.1186 (-3.850)* -0.2536 (-2.585)* 0.1571 (8.896)* f_knd2-0.1546 (-5.498)* -0.1099 (-7.545)* 0.01118 (0.594) -0.0241 (-1.157) -0.0773 (-3.214)* -0.0220 (-0.316) 0.0382 (3.452)* dcwest -0.0700 (-1.637) 0.0909 (3.653)* 0.0314 (1.057) 0.0566 (1.600) -0.1783 (-4.644)* -0.1003 (-0.727) 0.1243 (6.007)* dseast 0.0481 (1.334) 0.0794 (3.646)* -0.3460 (-12.628)* -0.0146 (-0.466) -0.2185 (-6.657)* 0.0542 (0.486) 0.2505 (13.775)* dneast -0.0585 (-1.607) -0.0076 (-0.348) -0.4078 (-14.966)* -0.1817 (-5.711)* -0.2026 (-6.229)* 0.0503 (0.461) 0.1925 (10.669)* dsouth -0.0091 (-0.231) 0.0557 (2.347)** -0.2549 (-8.443)* 0.1740 (5.222)* 0.0488 (1.409) 0.0834 (0.683) 0.1482 (7.479)* Wth_HP 0.1178 (5.065)* 0.0961 (6.489)* -0.0512 (-2.153)** -0.0048 (-0.214) 0.1397 (6.128)* -0.2064 (-1.869) -0.0677 (-5.327)* N 95565 95565 95565 95565 95565 95565 95565 Log Lkelhoo 1- Z- statstcs are n parentheses * sgnfcance level for 1% ** sgnfcance level for 5% -13340.13-40920.07-17382.01-16213.18-12853.73-1107.16-58161.94 Table 5 Heckman Procedure to Estmate the Prce of the Publc Health Servces N= 82800, Log Lkelhood = -27520.82 Varable Estmate Standard Error Varable Estmate Standard Error Sex 0.1914 0.0244 pllst 0.0280 0.0386 Educ -0.0382 0.0492 plreemb 0.1252 0.0216 Income 0.00001 0.00001 plother 0.2927 0.0268 Black -0.0087 0.0557 plcons -0.7845 0.0625 Whte -0.1418 0.0276 plexam 0.1088 0.0559 Yellow 0.1438 0.9555 plnter 0.5942 0.0413 dcwest -0.1110 0.0564 pldent -0.3128 0.0235 dseast -0.2026 0.0484 plmedc 0.0395 0.0440 dneast -0.0128 0.0492 odonto -0.0045 0.0564
20 dsouth -0.1704 0.0514 paymore -0.3670 0.0257 age -0.0001 0.0013 depen 0.1983 0.0627 Age60 0.1520 0.0391 fam_dep 0.1447 0.0620 f_knd2 0.0492 0.0278 Constant 6.6104 0.1695 platend -0.0104 0.0243 Select Sex -0.0098 0.0163 dseast 0.2822 0.0319 Educ 0.1161 0.0018 dneast 0.0626 0.0328 Income 0.0003 0.00001 dsouth 0.0653 0.0345 Black -0.0654 0.0350 age 0.0215 0.0008 Whte 0.2188 0.0174 Age60-0.1998 0.0274 Yellow 0.4010 0.0769 f_knd2 0.0790 0.0199 dcwest -0.0710 0.0377 Constant -3.5883 0.0494 Table 6 Determnants of ln P S N = 8267, Log-Lkelhood = -2640.38 Varable Estmate Standard Error Varable Estmate Standard Error Constant -0.020817 0.127874 Mxed -0.211478 0.047998 Male -0.227694 0.047326 Asan 0.273869 0.293929 Mgrator 0.150101 0.199747 Age 0.019145 2.03127 x 10-3 ED (5-8 yrs) 0.387136 0.054480 Age > 60-0.105695 0.074067 ED (9-12 yrs) 0.845308 0.058539 Income 1.99052 x 10-4 9.27403 x 10-6 ED (12+ yrs) 1.34218 0.074263 Employee -0.235912 0.053343 North -0.100590 0.094918 Self-Employed -0.255233 0.053334 Northeast -0.135340 0.067079 Domestc Worker -0.539780 0.119628 Center-West 0.140623 0.069616 Metro 1-0.025042 0.051464 Southeast 0.110980 0.054015 Metro 2-0.416106 0.048105 Black -0.406526 0.092505 Famly Sze -0.043249 0.015133 Table 6 Soco-Demographc Effects on Proportonal Measure of Compensatng Income Varaton From a 50% Increase n P S N = 8267, R-squared = 0.659412 Varable Estmate Standard Error Varable Estmate Standard Error Constant 0.419864 4.26638 x 10-3 Mxed 0.011576 1.55395 x 10-3 Male 0.014097 1.70310 x 10-3 Asan -0.076258 0.013461 Mgrator -6.19360 x 10-3 6.80662 x 10-3 Age -9.90146 x 10-4 6.74201 x 10-5
21 ED (5-8 yrs) -6.35279 x 10-3 1.67772 x 10-3 Age > 60 7.28907 x 10-3 2.72447 x 10-3 ED (9-12 yrs) -0.037660 1.98657 x 10-3 Income -2.79919 x 10-5 7.59087 x 10-7 ED (12+ yrs) -0.249227 3.41969 x 10-3 Employee 0.015117 1.95085 x 10-3 North -1.70524 x 10-3 3.28471 x 10-3 Self-Employed 0.014800 1.95308 x 10-3 Northeast -2.96403 x 10-3 2.16406 x 10-3 Domestc Worker 0.018112 3.42039 x 10-3 Center-West -8.51688 x 10-3 2.61514 x 10-3 Metro 1 8.84075 x 10-3 1.89919 x 10-3 Southeast -6.31576 x 10-3 2.00423 x 10-3 Metro 2 0.015212 1.57295 x 10-3 Black 0.017732 2.76538 x 10-3 Famly Sze 1.51898 x 10-3 4.27404 x 10-4 Table 7 Soco-Demographc Effects on Compensatng Income Varaton From a 50% Increase n P S (I Assumed to be Observed) N = 7644, R-squared = 0.402024 Varable Estmate Standard Error Varable Estmate Standard Error Constant 168.845 18.8424 Mxed -3.59553 6.77750 Male -33.5610 7.68259 Asan -22.9243 77.1354 Mgrator -50.4782 29.9692 Age.070672.299853 ED (5-8 yrs) 36.0775 7.29269 Age > 60 7.32159 12.1295 ED (9-12 yrs) 66.5361 8.97825 Income.412354 0.00705 ED (12+ yrs) -81.0515 20.8902 Employee -51.6785 8.84152 North -1.49066 14.5177 Self-Employed -47.4406 8.76669 Northeast -58.1799 9.58748 Domestc Worker -64.0738 14.9596 Center-West -32.8344 11.6064 Metro 1-21.8923 8.51166 Southeast -5.66378 8.96483 Metro 2-59.2504 6.90262 Black -14.2627 12.0434 Famly Sze 20.9211 1.87051 Table 8 Soco-Demographc Effects on Proportonal Measure of Compensatng Income Varaton From a 50% Reducton n P P N = 8267, R-squared = 0.697525
22 Varable Estmate Standard Error Varable Estmate Standard Error Constant 0.023276 7.38658 x 10-3 Mxed 0.031053 2.69043 x 10-3 Male 0.031212 2.94866 x 10-3 Asan -.144877.023305 Mgrator -7.06999 x 10-3.011785 Age -2.47009 x 10-3 1.16727 x 10-4 ED (5-8 yrs) -0.023735 2.90472 x 10-3 Age > 60 0.024978 4.71700 x 10-3 ED (9-12 yrs) -0.105878 3.43945 x 10-3 Income -4.35465 x 10-4 1.31424 x 10-6 ED (12+ yrs) -0.465490 5.92066 x 10-3 Employee 0.039820 3.37760 x 10-3 North -4.99453 x 10-3 5.68696 x 10-3 Self-Employed 0.038266 3.38146 x 10-3 Northeast -4.25265 x 10-3 3.74674 x 10-3 Domestc Worker 0.053847 5.92188 x 10-3 Center-West -0.025216 4.52770 x 10-3 Metro 1 0.017479 3.28815 x 10-3 Southeast -0.017338 3.47001 x 10-3 Metro 2 0.039686 2.72332 x 10-3 Black 0.049946 4.78783 x 10-3 Famly Sze 3.88479 x 10-3 7.39983 x 10-4 REFERENCES Alvarez, Isabel, 21 st Century, Challenges Facng the Brazlan Health Sector, A Report on the 1998 Roundtable Held n Sao Paulo, Brazl, http:\\www.amercas.org\publcatons\ Alves, Densard, Parametrc and Sem-Parametrc Modelng of Healthcare Expendture: A Household Data Analyss for the Cty of São Paulo, Texto para Dscussão no. 5/00, IPE- USP, São Paulo,SP. Cutler, Davd M. and Ellen Meara, 1998. The Medcal Costs of the Young and Old: A Forty-Year Perspectve n Davd A. Wse, edtor. Fronters n the Economcs of Agng. Chcago. Unversty of Chcago Press. De Faras, Pedro Cesar Lma, (1998), Socal Securty n Brazl: Problems and Trends, George Washngton Unversty, Insttute of Brazlan Issues, The Mnerva Program, Fall 1998. Greene, Wllan H., 2000. Econometrc Analyss, Prentce Hall Harmelng, Susan, (1999) Health Reform n Brazl, Case Study for Module 3: Reproductve Health and Health Sector Reform, World Bank Insttute. Long, James Scott, 1997. Regresson Models for Caregorcal and Lmted Dependent Varables, Sage Publcatons, London Natonal School of Publc Health, Oswaldo Cruz Foundaton, Mntsry of Health, Brazl,