Who Benefits From Social Health Insurance in Developing Countries?

Transcription

1 Who Benefits From Socia Heath Insurance in Deveoping Countries? Pau Gerter University of Caifornia at Bereey and NBER Orvie Soon University of the Phiippines, Schoo of Economics March, 2000 Abstract A major poicy issue in deveoping countries is the ac of forma insurance marets. A popuar approach to this probem is compusory socia heath insurance (SI). The movement towards SI has been motivated not ony by the desire to expand insurance coverage, but aso by fisca pressure to shift the burden of deivering and financing heath care from the pubic sector to the private. In this paper we show that SI fais to expand insurance coverage or shift the burden to the private sector because providers capture SI benefits as rent by raising price-cost margins to insured patients. As a resut the out-of-pocet costs to the insured patient are the same as to the uninsured. Our empirica resuts from the Phiippines indicate that hospitas extract 86 percent of SI benefits through price discrimination. We aso show that expanding SI actuay increased the burden on the pubic sector rather than reieving it. Keywords : Socia Insurance, Heath Insurance, and Price Discrimination. J.E.L. cassification numbers : H51, I11, and O12. Corresponding address: Pau Gerter, Haas Schoo of Business, University of Caifornia, Bereey Caifornia, or gerter@haas.bereey.edu Acnowedgements: We are indebted to Michea Ash, Rui de Figueiredo, Wiiam Evans, Ben Hermain, Maitreesh Ghata, Garric Baoc, Jeff, Hammer, Aex Herrin, and Jonathan Morduch provided hepfu comments and to Stea Aabastro, Aice Ferrer and Caros Tan who provided exceent research assistance. The authors gratefuy acnowedge financia support from grants from the Internationa Heath Poicy Program and the U.S. Nationa Institute of Aging. The usua discaimer appies.

2 1. INTRODUCTION A major poicy issue in deveoping countries is the ac of forma insurance marets. A popuar approach to this probem is mandated socia heath insurance (SI), which finances medica care through compusory payro taxes and aows beneficiaries to purchase medica care from both private and pubic providers. The movement towards SI has been motivated not ony out of the desire to expand insurance coverage, but aso to reduce financia pressure on pubic budgets (Gerter, 1998). SI is seen as a way to shift a portion of the pubic burden of deivering and financing heath care to the private sector. SI reduces the out-of-pocet price of higher quaity private care reative to ower quaity pubic care at the time of purchase, thereby providing an incentive to choose the private sector over the pubic. In this paper, we show that serious faws in the designs chosen by most SI pans significanty reduce their insurance vaue and fai to shift the burden from the pubic to the private sector. The typica SI design aows providers to capture the SI benefits as rent by raising pricecost margins to insured patients. As a resut, out-of-pocet payments at the time of iness for the insured are the same as for the uninsured, impying that SI does not provide an incentive to choose the private sector over the pubic. Hence, SI fais to expand insurance coverage or shift the burden to the private sector. These resuts are important because a arge number of countries have adopted this design, and there are strong poitica-economy reasons to beieve that countries considering SI are iey to adopt this same design. In this paper, we theoreticay and empiricay mode the SI rent extraction through price discrimination. The identification of the rent extraction, however, is not straightforward. We use a two-part tariff mode to derive optima price discrimination strategies. We show that if hospitas have maret power, they wi not ony extract a SI benefits through price discrimination, but aso use heath insurance status as information about patient type so as to expoit differences in insured and uninsured patients price easticities of demand to further price

3 discriminate. However, the two-part tariff mode has a natura hedonic price specification in which we can identify separatey the two types of price discrimination. We then use data from the Phiippines to estimate the amount of rent extraction. The Phiippines is an exceent setting for a case study for severa reasons. First, the Fiipino SI program has the typica design. Second, we can identify the price discrimination effect by using the fact that the Fiipino SI program mandates coverage ony to wage-sector empoyees and dependents. Third, since the program was introduced in 1972, there has been enough time for entry to dissipate rents if there is insufficient maret power for price discrimination. If we observe price discrimination, then this suggests that there are ong run entry barriers that aow providers an ongoing abiity to capture SI benefits. Fourth, we expoit mutipe observations on each provider to contro for seection based on heterogeneity in provider quaity. The resuts indicate that private hospitas extract 100 percent of SI benefits through increased price-cost margins and that pubic hospitas extract 70 percent. Finay, we use the estimates from the price discrimination modes to evauate who benefits from SI. We consider three issues. First, what percent of SI benefits are captured by hospitas as rent through increased price-cost margins and what percent finances patient care. Second, we examine the caim that hospitas use the profits from charging higher prices to insured patients to cross-subsidize the care of financiay indigent patients. Third, we examine the effect of expanding Medicare on pubic expenditures in the heath sector to investigate the hypothesis that SI reieves the burden on the pubic sector. The paper is organized as foows. In section 2 we describe the typica SI design, the motivation for countries to adopt SI, and the institutions that faciitate price discrimination. In section 3, we derive optima price discrimination rues. In section 4, we specify the empirica mode and describe our identification strategy. In section 5, report the estimation resuts for the Phiippines. In section 6, we present the who benefits anaysis, which is foowed by a brief concusion in section 7. 2

4 2. PRICE DISCRIMINATION UNDER SI In the post-coonia era, most ow-income countries created arge pubic heath care deivery systems modeed after the British Nationa Heath Service (NHS). These NHS-ie systems deiver medica care through pubicy operated deivery systems financed through genera tax revenues. They typicay charge at most nomina user fees in order to insure that income is not a barrier to medica care (Word Ban, 1987). However, rapid medica cost infation combined with aready boated budgets caused many governments to rethin the poicy of pubicy financed and deivered heath care (Gerter and van der gaag, 1990, Jimenez, 1994; Word Ban, 1993). A popuar aternative to the NHS-ie systems has been to introduce or expand compusory SI (Gerter, 1998). SI pans typicay finance medica care benefits for wage sector empoyees through mandatory earmared income taxes, and the benefits are used to purchase medica care from either the pubic or the private sector. 1 SI is seen as a way to shift a good portion of the pubic burden of deivering and financing heath care to the private sector. SI reduces the out-of-pocet prices of higher quaity private care reative to ower quaity pubic care at the time of purchase, thereby providing an incentive to choose the private sector over the pubic sector. In this way, SI shifts the deivery of care for wage sector famiies to the private sector, and finances their expenditures through additiona off-budget earmared income taxes, thereby reieving pressure from the genera budget. Gerter and Strum (1997) estimated that expanding SI woud reduce the demand for pubic deivered care by 33 percent in Jamaica. These resuts, however, are based on the assumption that as SI expands, private sector pric es did not increase i.e., private sector suppy is perfecty eastic. In this paper we expicity tae in to account provider response, and argue that SI actuay produces the opposite of the desired effects. 1 Administrative and compiance probems have typicay imited expansion of SI to the informa sector and, as a resut, coverage is ess than universa in most countries. 3

5 2.1 Price Discrimination There is a ong tradition of price discrimination in heath care in both deveoped and deveoping countries. Providers have often charged higher prices to the weathy and ower prices to the poor justifying the practice based on charity motives. Governments viewed this practice as in the pubic interest as it improved the access of the poor to medica care. The U.S. iterature on price discrimination in medica care marets dates bac to Kesse s (1958) wor on physician marets and most recenty as part of the hospita cost-shifting debate (Dranove and White, 1994). SI faciitates price discrimination by providing information about whom to charge higher prices. The typica SI design maximizes the amount of the insurance premium that providers can capture through raising prices to insured patients. The main features of the typica SI pan, described in Tabe 1, are first-doar coverage up to a cap beyond which individuas are iabe for the baance of the bi, insurance is imited to hospita inpatient care, providers are paid on a feefor-service basis, and providers can charge maret rates. Tabe 1: Features of SI Programs in Seected Deveoping Countries Country Financing Sources Benefits Provider payment Baance Biing Costa Rica Payro tax & saes tax 1 ST doar with capped benefits; Inpatient ony Egypt Payro tax & sin taxes 1 ST doar with capped benefits; Inpatient ony Indonesia Payro tax & genera tax revenues 1 ST doar with capped benefits; Inpatient ony Morocco Payro taxes 1 ST doar with capped benefits; Inpatient ony Korea Phiippines Thaiand Taiwan Vietnam Payro tax & genera tax revenues Payro tax & genera tax revenues Payro tax & genera tax revenues Payro tax & genera tax revenues Payro tax & genera tax revenues Fee-for-service Fee-for-service Fee-for-service Fee-for-service Yes Yes Yes Yes Co-pays Fee-for-service No 1 ST doar with capped benefits; Inpatient ony 1 ST doar with capped benefits; Inpatient ony Fee-for-service Fee-for-service Yes Yes Co-pays Fee-for-service No Co-pays; Inpatient ony Fee-for-service Yes Source: WHO Inter-regiona Consutation on Heath Insurance Reform, Seou Korea, Apri 3-7,

6 In pans with first-doar capped-benefits, SI pays the provider from the first doar of the bi up to the maximum benefit and patients must pay out of pocet any remaining difference between the price and the maximum benefits. The first-doar capped benefit is a ump sum transfer or reverse deductibe, which minimizes price distortions and the deadweight oss from mora hazard i.e. the additiona amount of heath care demanded with insurance than without it. Since the benefit is basicay a ump sum transfer, it does not distort prices at the margin, and ony distorts demand when the cap is above the amount that patient woud purchase without insurance. If the cap is sma enough so that there is no demand distortion, then the whoe benefit is avaiabe for rent capture. When there is mora hazard, some of the benefit must be used to pay for the increased production costs and deadweight oss, and therefore is unavaiabe for capture. A necessary condition for price discrimination is that providers have sufficient maret power so that potentia new entrants and other competitors cannot profitaby offer the services at a ower price. There are severa reasons to beieve that such maret power may exist. First, most countries have icensure reguations that effectivey imit physician and hospita entry. Second, hospitas have arge fixed costs and exhibit economies of scae in variabe costs. 2 Third, most ow-income countries have poor transportation systems that faciitate oca monopoies in hospitaization. Whie there are good reasons to beieve that there is enough maret power for hospitas to be abe to price discriminate, the issue can ony be resoved concusivey through empirica examination on a maret-by-maret basis. As Phips (1983) argues the best evidence of maret power is observing price discrimination. Therefore, we wi directy examine whether there is price discrimination. 2 Aba (1998) reports that fixed costs are over haf of the costs of hospitas in the Phiippines and that there are substantia economies of scae in variabe costs unti hospitas reach we over 100 beds in size. See Carey (1996) for economies of scae estimates in the US. 5

7 2.2. The Poitica-Economy of Heath Care Reform Low-income countries have imited abiities to tax, which imposes a tight budget constraint on SI benefits. SI pans face the trade-off between providing a arge number of individuas with a sma benefit or a sma number of peope with a arge benefit. This means that a arge deductibe (or copay) is required to provide uncapped benefits for arge financia riss e.g. rare catastrophic inesses such as cancer. The high deductibe ensures that benefits are avaiabe for expensive catastrophic inesses and are not used up on ess expensive high-probabiity events (e.g. fu). Because of the budget constraint, owering the deductibe woud require capping benefits. In the imit, a zero deductibe impies the owest possibe benefit cap and the east effective insurance. If capped first-doar coverage provides minima insurance, why then do we see it so widey adopted? In addition to providers, a number of powerfu poitica interest groups, each for a different reason, see this design as in their sef-interest. These same interest groups are active in countries considering SI, which suggests that it is a mode to be even more widey adopted. The most obvious interest group is the coection of internationa donors and poitica interest groups worried about the poor (Besey and Gouveia, 1994). First doar coverage aeviates the concern that sma out-of-pocet costs may deter utiization (Cornea, Joy and Stewart, 1987). Poiticians aso support first-doar coverage. Since the median voter is poor in most of these countries, first-doar coverage puts money into more voters pocets. Empoyers aso have strong incentives to support capped-benefit first-doar coverage. Typicay, SI premiums are co-financed by empoyers. Empoyers historicay have provided worers with heath benefits as a means of reducing absenteeism (Gerter and Sturm, 1997). Quic treatment of minor inesses reduces absenteeism more than the treatment of catastrophic inesses. Empoyers cap benefits since it is cheaper to fire severey i individuas who had itte chance of returning to wor and hire new wors. For simiar reasons, empoyers benefit more from the capped first-doar coverage, which is more iey to reduce absenteeism, than from catastrophic coverage with high deductibes, which is ess iey to affect worforce productivity. 6

8 3. OPTIMAL INSURANCE-BASED PRICE DISCRIMINATION In this section, we deveop a mode of optima pricing strategies when hospitas, assumed to have sufficient maret power, exercise price discrimination based on observed patient insurance status. We show that hospitas are abe to extract patients' insurance benefits by increasing price-cost margins, and use insurance status as a signa of unobserved differences in other characteristics that ead to differences in demand between insured and uninsured patients. The mode deveoped here aso has a natura hedonic pricing empirica specification and identifies the two sources of price discrimination. The empirica mode is derived in Section Demand Patients enter hospitas in order to treat an iness or injury. During treatment, they are provided with services such as medicines, ab tests, x-rays, and surgery. These services are not vaued in and of themseves, but rather for their effect on heath status. Since patients are heterogenious i.e. have different heath probems, incomes, education, tastes, etc. there is substantia variation in patients' vauations of the benefits of medica care services. For convenience we assume that there are K medica services and that patients have the foowing money-metric utiity functions: U i ( M,..., M ) O = θ H ( ) O i = θ H 1 M, (1) K where: H is the money vaue of the utiity of heath H( ) is the heath production function which is increasing and concave M is the quantity of service, = ( ) M M 1,...,M K, O is the amount the patient pays out-of-pocet to the hospita, and θ i is a preference intensity parameter. We assume that there are two types of patients. Type- patients have a ower vauation of the benefits of medica care than do type-h patients, θ h < θ. 7

9 Hospitas price their services in the form of two-part tariffs where there is a fixed fee pus a price for every unit of each service consumed. A patient's bi is B = F + K = 1 P M, (2) where F is the fixed fee and P is the unit price of M. Without insurance, the patient s out-ofpocet payment, O, is equa to B i.e. the bi charged by the hospita. Given this pricing rue, a utiity-maximizing patient wi enter the hospita if (1) is greater than the utiity from the next best aternative, which we normaize to zero, and demand each service up to the point where margina utiity equas the unit price: θ i H M * * * where = (,...,M ) M 1 K * ( M ) = P M soves (3)., (3) When patients have insurance, the payment received by the provider diverges from the patient s out-of-pocet payment. Lie most SI pans in ow-income countries, the Phiippine Medicare program provides first-doar coverage up to a maximum amount payabe to providers on a fee-for-service basis. LetR be the maximum insurance benefit. Then, assuming the hospita charges insured patients Fˆ and Pˆ, the patient's out-of-pocet payment is: K O = Fˆ + Pˆ whereas the hospita s bi is: =1 M R, (4.1) Bˆ = Fˆ + K = 1 Pˆ M. (4.2) The difference between the hospita s bi and the out-of-pocet payment is the insurance benefit. Insurance, in this case, is ie a reverse deductibe. Patients receive a ump-sum transfer after which they are at ris for the baance of the bi. The ump-sum transfer does not distort reative prices and ony distorts demand if the benefit is arger than what they woud have 8

10 purchased without the benefit. This means that an insured patient wi enter the hospita if (1) is positive, and for each service, consume the greater of the demand impied by the maximum insurance benefit, M, or * M, the eve impied by (3). Thus, if R is sufficienty ow so thatm * M, then (3) hods for insured patients as we. In the case of the Phiippines, the maximum Medicare benefit is approximatey 15 percent of the average hospita bi and ony a sma fraction of the very expensive cases that severey sew the distribution of hospita expenditures to the right. Tota demand for hospita care and each of the services depends on the distribution of patients over insurance status and preferences. Suppose that there are Y uninsured patients, of which α are type-h patients and ( 1 α ) are type- patients. Aso et Y $ be the tota number of insured patients, of which α $ are type-h patients and ( 1 α $) are type- patients. The tota number of potentia patients, then, is Y + Y $. Demands for hospitaization and hospita services are presented in Tabe 2. For a given eve of prices, the utiity of the type-h is aways bigger than the utiity of the type-. Therefore, if the prices are such that the type- have utiity ess than zero, then demand wi be zero. If the prices are such that the type-h have positive utiity but the type- do not, then ony the type-h wi enter the hospita and the demand from services wi be type-h demand. If the utiity of the type- is positive, then both types of patients wi enter the hospita and the demand from services wi be the sum of demand from the type- and type-h. Tabe 2. Demand for Hospita Care and Services Patient Type # of Patients in Hospita Demand for Hospita Services If h Uninsured 0 0 U < 0 α Y *h h M Y U &U Y ( * h ( ) M * αm + 1 α )Y U 0 α 0 < 0 Insured 0 0 ˆ U < 0 ˆα Yˆ ˆ * h α ˆM Ŷ ˆ h 0 & ˆ U U < 0 ( ˆ *h ˆ ˆ ˆ * αm + 1 α M )Ŷ ˆ h U 0 Yˆ ( ) 9

11 3.2 Hospita Behavior We consider two cases. The first is where hospitas are abe to distinguish patients by both insurance status and by intensity of preferences, and the second, the more reaistic case, is where they are ony abe to observe insurance status. We assume that hospitas are profit-maximizing organizations. However, in the Phiippines, as in most countries, there are both private proprietary and government-operated faciities. Whie the profit maximization assumption is probaby reasonabe for proprietary hospitas, it may be ess appropriate for pubic hospitas. Whie we wi use the profitmaximizing framewor to specify the empirica modes, we wi test for differences by estimating separate modes. We wi aso test the hypothesis that hospitas charge higher prices to the insured to cross-subsidize the care of financiay indigent patients Price Discrimination with Observed Patient Heterogeneity In this case, the hospita can charge separate prices to patients based on both insurance status and preference intensity. The profit function can be expressed as where h ( α ) Y + Bˆ αˆŷ + Bˆ ( ˆ ) Ŷ C( M) h Π = B αy + B 1 1 α (5) h ( ( )) ( ˆ ˆ ˆ h M = M α + M 1 α Y + M α + M ( 1 αˆ ))Ŷ, and C(M) is the cost of suppying services and is increasing and convex in M. Profits are maximized subject to the constraint that utiity is non-negative for each patient type. Profit maximization impies the participation constraints are binding i.e. utiity is set equa to zero. Setting (1) equa to zero and soving using (2) and (4.2) impies that the profit maximizing fixed fee is equa to: F Fˆ i i i K * i ( M ) i i P M = 1 * i K i i Pˆ Mˆ =1 = θ H for i = and h (6.1) ( Mˆ ) + R i = θ H for i = and h (6.2) 10

12 Notice that the fixed charge for uninsured patients is just to net consumer surpus, whereas for insured patients the fixed charge adds the insurance benefit. Maximizing (5) subject to (6) derives the first order conditions for the service prices. Using (3) and rearranging the terms aows us to express the first-order condition as: P * *h * *h C = P = Pˆ = Pˆ =. (7) M This expression impies that hospitas set unit service prices equa to margina costs. This rue is appied to a types of patients regardess of insurance status and preferences. Profits or rent are extracted ony through the fixed charge, which vary by insurance status, and by patient preference intensity. This is a version of the standard two-part tariff resut that firms set the variabe prices equa to margina cost to maximize net consumer surpus, and then extract a of net consumer surpus as profit through the fixed fee (Tiroe, 1988). Finay, we examine the difference in profits from patients with different insurance status. Substitution of (7) into (6) and both into (5), and rearranging terms yieds: i* i* ( M ) C( M ) i i Π = θ H for i =, h (8.1) ( ˆ i* M ) C( ˆ ) + R ˆ i* i i Π = θ H M for i =, h (8.2) The hospita s profits are the tota consumer surpus ess costs pus the SI benefit from insured patients. Since the services prices are equa for insured and uninsured patients, ˆP P i * = i*, service * i* demands must be equa, M i = ˆM, and a of the difference in profit is captured through the fixed fee marup. Therefore, the difference in profits from insured and uninsured patients is just the SI benefit and Bˆ *i i *i * i B = Fˆ F = R Price Discrimination with Unobserved Patient Heterogeneity Whie a patient s insurance status can easiy be identified, hospitas cannot effectivey discriminate on the basis of preference intensity. Therefore, whie the hospita can set separate prices for insured and uninsured patients, it cannot do so for type- and type-h patients. However, 11

13 if the distribution of preference types within insured and uninsured patients is nown, then insurance status provides a signa about the patient s type that can be used for further price discrimination. Therefore, differences in observed prices between insured and uninsured patients refect not ony the effort to extract the SI benefit, but aso differences in demand due to underying heterogeneity differences. Assuming that the preference distribution parameters α and $ α are nown, the hospita has two aternative pricing strategies. It coud either charge prices that maximize profits from type- patients or charge prices that maximize profits from type-h patients. If the hospita chooses the type- strategy, then both patient types enter the hospita. In this case, the hospita extracts a rent from type- and some rent from type-h. If the hospita sets prices above this point but beow the prices that maximize profits form type-h patients, then the type- do not enter and type-h do. Therefore, if the hospita is going to charge a price higher than the type- strategy, then it wi charge prices that maximize profits from the type-h patients. 3 The hospita wi choose the strategy with the greater profits. Consider first the type-h pricing strategy. In this case, the hospita nows that a of its patients are type-h. The pricing rue, then, is the same as the case when patient type is observed. It wi set the prices of the services equa to margina costs for both insured and uninsured patients and extract a consumer surpus through the fixed fees. The difference between the fixed fee charged to insured and uninsured patients is sti equa to R i.e. tota SI benefits. Now consider the type- pricing strategy. This strategy impies that the fixed fees charged both types of patients are set to extract a of the consumer surpus type- patients: F = * h * h h ( M ) P M < θ H ( M ) P M θ H (9.1) 3 F h h h ( Mˆ * ) Pˆ Mˆ + R < θ H ( Mˆ ) Pˆ ˆ + R ˆ * = θ H M. (9.2) A profit-maximizing hospita wi aways choose the type-h pricing strategy over any other set of prices stricty higher than the type- strategy. Consider for exampe the average of the type- and type-h strategies. Under both the type- and the average bi pricing strategies, the number of patients entering the hospita wi be equa to α Y + αˆ Ŷ. This being the case, and since h h h αˆ Bˆ + 1 αˆ Bˆ < Bˆ and αb + ( 1 α) B < B, profits under the type-h pricing strategy are higher. ( ) h 12

14 At these prices, the hospita does not extract a of type- h surpus through the fixed charge. However, the hospita can capture some of the remaining type- h surpus by raising the prices of services above margina cost. Increasing prices captures more type-h surpus, but owers surpus extracted from the type- through the fixed fee. The point is iustrated in figure 1, where the ines M h ( P) and ( P) M represent type-h and type- demand for service. If the hospita prices the service at margina cost (MC), then the demands wi be ines M and profits wi be 2 times the area MCFE which just the sum of the fixed fees for type-h and type- h* * M, and patients. If the hospita raises the price to P, then demands fa to M and h ' M. The hospita gains profit represented by the sashed area BECD from type-h patients, but oses profits represented the checed area ABE from type- patients. The hospita increases service prices above margina cost as ong as the margina gain in surpus extracted from type-h patients is greater than the margina cost of ost profits from type- patients. Figure 1 Price F P B C MC A E D M h ( P) M ( P) ' M * M h' M h' M M Formay, under the type- pricing strategy, the provider profit function is 13

15 Π = ( F + P M )( Y + ( 1 α ) Y ) + ( Fˆ + Pˆ M )( αˆ Yˆ + ( 1 αˆ ) Yˆ ) C( M) α. (10) Maximizing (10) subject (9) and rearranging terms yieds: P = 1 αm h α ε h ( M M ) h C M ( ) + 1 α M εˆ > 0 (11.1) Pˆ = 1 αˆmˆ αˆ h εˆ h ( Mˆ Mˆ ) h C + M ( 1 αˆ ) Mˆ εˆ > 0 (11.2) where CM is the margina cost of M andε i is type-i s price easticity of demand for M. Conditions (11.1) and (11.2) show that the price-cost margin depends on the difference in demand for services between type-h and type- patients, and a weighted average of their respective price easticities where the weights are the tota amount of the service demanded by that type. The bigger the differentia between the amount demanded by type-h and type- patients, the bigger the price-cost marup since there is more reativey surpus to capture from type-h patients. Simiar to Ramsey pricing, the more eastic demand, the ower the price-cost marups since there is more to ose through reduced fixed fees. Notice that the optima prices of the services in (11.1) and (11.2) are not a function of R, but are a function of the heterogeneity parameters. On the other hand, the difference between the optima fixed fees is just equa to R and pus the difference in consumer surpus of type- patients which is a function of the services prices, and therefore a function of the heterogeneity parameters. These resuts have three impications for the empirica wor. First, the hospita wi charge insured patients different fixed fees and different prices for the services if their are heterogeneity distributions are different. Ony in the specia case where the distributions are the same for insured and uninsured patients (i.e. α = α$ ) wi P = P $. In this case, the difference in 14

16 the fixed fees wi be equa to the SI insurance benefit. Second, the hospita price discriminates against insured patients in two ways: (i) it extracts the SI benefit R, and (ii) its uses insurance status as a signa of the differences in the heterogeneity distributions of the two patient types. Therefore, a straightforward comparison of prices wi not identify whether the hospita has captured the fu SI benefit. Third, the differences in insured and uninsured service prices is soey due to using insurance status as a signa of unobserved heterogeneity, whie the difference in the fixed fee is both the signa and extracting the SI benefit. 4. EMPIRICAL SPECIFICATION AND IDENTIFICATION Our empirica objective is to investigate whether insured patients are charged higher prices than uninsured patients for the same services. The empirica specification is a combination of the tota bi equations for uninsured and insured patients. Formay, combing (2) and (4.2) into a singe equation yieds the bi from hospita j who suppies services M i to patient i: B ij = F + ( F F ) d + P M + ( Pˆ P) i i M idi + φ j + ˆ ε (12) ij where d i equas one if the patient has insurance and zero otherwise, φ j is a hospita fixed effect and ε ij is a zero mean random disturbance that is uncorreated with d i, φ j, and M i. The specification in (12) is a hedonic pricing mode (e.g. Rosen 1974) where the tota bi is regressed against the M s i.e. the characteristics of the good, a dummy variabe indicating insurance status, and interactions between the dummy variabe and the M s. The coefficients on the M s are interpreted as the impicit prices of the good s characteristics. In equiibrium these coefficients are equa to the marup in equation (11.1). The intercept is the fixed fee. The coefficient on the dummy variabe is additiona fixed fee charged insured patients, and the coefficients on the interactions are the additiona service price charged insured patients. The addition charged insured patients is the difference between (11.2) and (11.1). 15

17 Ideay, we woud et the fixed fee and service prices vary by hospita. Hospita prices might differ because demand varies across marets and because of heterogeneity in quaity. Some hospitas are better at providing medica services because of more advance technoogy and more quaified staff, and are, therefore, abe to charge higher prices. However, since we do not have enough observations to estimate separate modes for each hospita, we fix the coefficients on the M s across hospitas and interpret them as average prices. But we do aow the fixed fee to vary by hospita by incuding a hospita fixed effect. The φ j represents the addition to the patient s bi based on provider quaity and differences in wiingness to pay. The fact that we are not abe to estimate hospita-specific modes impies that the error term is the sum of the deviation in the providers prices from the means times the amount of services purchased by the patients simiar to a random coefficients specification. This introduces heterosedasticity in the mode and, as a resut, we report White heterosedastic - consistent standard errors. A probem with the specification in (12), is that we cannot identify the extent to which prices are different due to SI rent extraction versus using insurance status as a signa of unobserved patient heterogeneity. Reca that the differences in the service prices refect ony the effect of using insurance status as a signa of the unobserved heterogeneity. Therefore, the identification probem is with the coefficient on the insurance dummy varibe. We sove the identification probem by substituting (9.1) and (9.2) evauated at the optima prices into (12): B ij = F + * * ( Ŝ S + R ) d + P M + ( Pˆ P) i i i M idi + j + θ ε (13) ij S = θ H M P M and S = θˆ H Mˆ Pˆ Mˆ, which is just * * * * * * where ( ) ( ) consumer surpus for the type- uninsured and insured patients. In a continuum of types, the first two terms of the coefficient on d i are just the difference in consumer surpuses of the margina 16

18 insured and uninsured patients who are just indifferent between entering the hospita and the next best aternative. The third term is the maximum SI benefit. The benefit has an i subscript, because it varies by diagnosis, severity of iness, and type of hospita. The specification in (13) imposes fu SI benefit extraction. To aow for ess than fu rent extraction, we mutipy R i by a coefficient that must be between zero and one. This aows us to write (13) in regression format as: B ij = α + α d + γr d + β M + λ M d + θ + ε (14) f fˆ f i i i where the parameters are given by their anaogues in (13). The coefficient on R i, γ, equas one if the hospita extracts the fu SI benefit, and equas zero if none of the benefit is extracted so that a of it is used to finance patient care. Any benefit not extracted through price discrimination pays for patient care by reducing the amount paid out of pocet at the time of care. Therefore, 1-γ measures the proportion of the benefit used to finance patient care. Price discrimination based on using insurance status as a signa of unobserved heterogeneity is captured by the coefficients on d i and on the M i d i. In addition, nonzero coefficients on R i, d i and or on the M i d i is evidence of maret power (Phips, 1974). Whie non-zero coefficients on d i and interactions with d i are evidence that hospitas have maret power and choose to price discriminate, it is not necessariy evidence that price discrimination is exercised for profit motives. Hospitas argue that they charge insured patients more to be abe to cross-subsidize the care of charity patients. To rue out the possibiity that charitabe motives drive price discrimination, we must show that the bi charged charity is greater or equa to the cost of care. We wi address this question in Section 6. The specification in (14) assumes that the heterogeneity distributions are different for insured and uninsured patients. As shown above, if the distributions are the same, then insurance status carries no usabe information about unobserved heterogeneity. In this case, the hospita i i i j ij 17

19 charges the same price for services to insured and uninsured patients. The difference in the fixed fee is just the amount of SI benefit extracted. This impies that (14) reduces to: B ij = α + γr d + β M + θ + ε. (15) f i i i j ij If the coefficients on d i and on M i d i are jointy zero, the heterogeneity distributions are the same. An additiona identification probem exists from possibe sorting based on hospita quaity. This arises from the possibiity that insured patients choose the higher quaity hospitas. In this case, the coefficients on insurance status aso capture the effect of higher quaity. Therefore, we estimate the mode using hospita fixed-effects to directy contro for the θ j as incidenta parameters. 5. PRICE DISCRIMINATION IN FILIPINO HOSPITALS In this section we use data from the Phiippines to estimate the extent to which hospitas are abe to extract Medicare insurance benefits as rent through price discrimination. The Phiippines is an exceent setting for this anaysis. We can identify the price discrimination by using the fact that the Fiipino SI program mandates coverage ony to wage sector empoyees and dependents. Since the program was introduced in 1972, there has been enough time for entry to dissipate rents if there is insufficient maret power for price discrimination. We first describe the SI program, then discuss data and measurement of variabes, and finay present the resuts. 5.1 Medicare The Fiipino SI program, Medicare, was estabished in It provides benefits for wage sector empoyees and their dependents financed through compusory payro taxes. Today, Medicare covers approximatey one-third of the popuation incuding government worers, private sector empoyees and their dependents. The program is financed by a 2.5 percent earmared payro tax 18

20 equay shared by worers and empoyers. In 1994, members contributed an average of 36 U.S. doars annuay to Medicare. Medicare provides imited coverage for inpatient hospitaization but not for outpatient services. Its inpatient benefits are first-doar coverage up to a cap, which varies with the type of care (surgica, genera medicine, maternity, pediatrics, etc.), with severity of iness (ordinary, intensive and catastrophic), with the eve of faciity (primary, secondary and tertiary), and with physician certification (genera vs. speciaist). In 1994, the average program benefit spending per caim was $85, an amount that is roughy equa to a third of the average cost of hospitaization in the private sector and a trivia portion of the costs of treating more rare catastrophic iness. 5.2 Data and Measurement The data come from survey of a random sampe of patients from132 hospitas conducted by the Phiippine Institute of Deveopment Studies for the Phiippine Department of Heath (PIDS- DOH) in Tabe 3 reports sampe sizes and descriptive statistics. The dependent variabe is the tota bi paid to the hospita. The ey independent variabes are insurance status (Medicare) and the maximum Medicare benefit for the patient (Medicare Payment). Note that most of the variance in the maximum Medicare benefit comes from individuas having Medicare or not. In the cases where they don t have Medicare, the maximum benefit is coded to zero. However, the variabe is not quite a dummy because, as discussed above, there is variation in the maximum benefit based on a number of factors. In the regression, we aso contro for ength of stay by room accommodation (charity ward, genera ward, semi-private/private room) and medica services (ab tests, radioogica exams, physician visits, surgeries, drugs and medica suppies). We aso test for the possibiity that there is measurement error in the services provided patients or that some of the services received were omitted from the data coection. This creates a 4 See Soon et a. (1997) for a detaied discussion of the sampe and data coection methods. 19

21 potentia omitted variabes probem. Insured patients may demand more services and if those services are omitted from the regression, the estimated insurance effects wi be confounded with differences in demand for the omitted services. We test for this by incuding individua characteristics that woud ead to purchasing more services in the mode. These variabes are whether the individua is a coege graduate, monthy househod per capita income, and an indicator of severity of iness (i.e. days in bed before being admitted to the hospita). If these variabes are significant predictors of the tota bi, then there are omitted services from the mode and the estimates are biased. Tabe 3: Descriptive Statistics Variabe Private Hospitas Pubic Hospitas Mean Std. Dev. Mean Std. Dev. Tota Bi Paid to Hospita (Pesos * ) 6,613 11,051 1,452 3,063 Payment Out-of- Pocet (Pesos) 5,968 10,793 1,008 2,680 Medicare Payment (Pesos) 578 1, ,347 Medicare (= 1) Charity Ward (=1) Genera Ward (=1) Private/Semi Private Room (=1) Inpatient Days Lab Tests Radioogy Exams Physician Visits Surgeries Drugs Prescriptions Suppy Orders Age Femae(=1) Monthy Famiy Per Capita Income 5,423 5,108 3,231 3,851 Income Missing (=1) Coege (=1) Days in Bed Before Admission Sampe Size: Patients Sampe Size: Hospitas * Monetary vaues are reported in 1991 Pesos. At that time the exchange rate was 25 Pesos for one US Doar. Tabe 3 shows that average bi in private hospitas was about 4.5 times the average pubic hospita bi. Severa reasons for this are immediatey apparent. Private hospitas have cose to twice as many insured patients. Their patients are better educated and have higher 20

22 incomes. These weathier insured patients consume substantiay more hospita services. Fewer patients in private hospitas are in the charity wards and more are in private rooms. In addition, patients in private hospitas have more radioogica exams and more medication. However, patients in pubic hospitas do have substantiay onger engths of stay and sighty more ab tests. Tabe 4 reports the means by insurance status. In private hospitas, the average tota bi paid hospitas is about 40 percent higher for insured patients than for the uninsured. However, the out-of-pocet payments are amost the same. Moreover, the quantity of each of the services is about the same for insured and uninsured patients. Thus, the descriptive statistics suggest that insured and uninsured patients are paying about the same out-of pocet for the same amount of services. This is consistent with price discrimination that competey extracts the SI benefit. Simiary for the pubic hospitas, the average tota bi paid the hospita for insured patient care is about 2.5 times the amount paid for the uninsured. However, the payment out-of-pocet is simiar for insured and uninsured. In the pubic case, there are more services provided to the insured than the uninsured patients. Whie these statistics are consistent with the price discrimination story, they are ess cearcut than those for private hospitas. The patients choosing pubic hospitas oo very different from those choosing private hospitas. The ey difference, as indicated in Tabe 3, is that insured, weathier and bettereducated patients tend to choose private hospitas. However, once they have made that choice, the insured patients in private hospitas oo simiar to the uninsured in private hospitas and demand simiar amounts of services as uninsured patients. Tabe 4 aso shows that, in private hospitas, insured and uninsured patients have the same income, education and severity of iness. These resuts suggest that the underying heterogeneity distributions of insured and uninsured patients may be simiar enough so that private hospitas charge the same service prices to insured and uninsured patients. This is not the case for pubic hospitas, where insured and uninsured patients oo different and demand different eves of care. 21

23 Tabe 4: Means by Insurance Status Variabe Private Hospitas Pubic Hospitas Insured Uninsured Insured Uninsured Tota Bi Paid to Hospita (Pesos) 8,134 5,784 2,535 1,079 Medicare Payment (Pesos) 1, ,734 0 Out of Pocet Payment (Pesos) 6,306 5, ,079 Charity Ward (=1) Pay Ward (=1) Private/Semi Private Room (=1) Length of Stay (days) Lab Tests Radioogica Exams Physician Visits Surgeries Drug Prescriptions Suppy Orders Age Mae (=1) Monthy Per Capita Househod Income 5,282 5,489 4,753 2,708 Income Missing (=1) Coege (=1) Days in Bed Before Admitted Sampe Size Estimation Resuts The resuts are presented Tabes 5 and 6 for private and pubic hospitas 5. For each sampe, we estimated four different modes. A of the modes incude Medicare Payment and the service variabes. Mode 1 aso incudes the Medicare dummy, interactions between Medicare and the service variabes, and patient characteristics. Mode 2 adds ony Medicare and interactions, and Mode 3 adds ony patient characteristics. Finay, Mode 4 has no added variabes. The resuts of specification tests are reported at bottom of the Tabes. Consider first tests of the hospita fixed effects. In a modes, we conducted Hausman tests to see if we coud use random effects instead of fixed effects. Random effects were rejected in a modes. Moreover, Chow tests rejected the hypothesis that the fixed effects were jointy zero in a modes. We next tested for omitted services that coud confound the insurance effect. Chow tests coud not reject the hypothesis that the coefficients on the patient characteristics were jointy zero 5 Chow tests reject poing profit and not-for-profit sampes. 22

24 in the reevant modes for both types of hospitas. Moreover, none of the individua coefficients on the patient characteristic variabes were significanty different from zero. We concude from this test that there are no omitted services that bias the estimate of the insurance effects. Finay, we tested the hypothesis that insured and uninsured patient heterogeneity distributions were the same. Chow tests coud not reject this hypothesis that the coefficient on the Medicare dummy and the interactions were jointy zero for private hospitas, but rejected it for pubic hospitas. These resuts are consistent with the descriptive pattern presented in Tabe 4. For private hospitas, this impies that insurance status provides no information about unobserved heterogeneity so that it is optima to charge insured and uninsured patients the same prices for services and raise the fixed fee by the amount of the insurance benefit that can be extracted. These specification tests impy that the data are most consistent with Mode 4 in Tabe 5 for private hospitas and Mode 2 in Tabe 6 for pubic hospitas. The coefficient on the Medicare Payment measures the extent to which hospitas are abe to capture the Medicare benefit through price discrimination. In the private hospita Mode 4, the estimated coefficient is 0.95, significanty different from zero, and not significanty different from one. This impies that private hospitas raise the fixed fee to insured patient by an amount amost equa to the insurance benefit. In this case, the hospita captures the fu benefit and Medicare finances none of the patient s care. In the pubic hospita mode 2, the estimated coefficient is 0.70, significanty different from zero, and significanty ess than one at the 0.1 eve. This impies that pubic hospitas extract ony about 70 percent of the benefit through price discrimination, and about 30 percent of the benefits are used to reduce out-of-pocet payments. The coefficients on the other variabes are consistent with common sense. Private hospitas charge positive amounts for a services except for drug prescriptions. They charge the most for surgeries and radioogica exams that are the most costy services to provide. They charge more for patients in private rooms than in genera wards, and do not charge for accommodation in charity wards. Pubic hospitas do not seem to charge for accommodation in 23

25 charity wards and for physician visits but do charge for the other services. Pubic hospitas raise the fixed fee for insured patients and charge them more than doube for radioogica exams and surgery. However, they give modest price discounts to insured patients for the other services. Tabe 5: Private Hospita Hedonic Price Modes Fixed-Effects Estimates Mode 1 Mode 2 Mode 3 Mode 4 Independent Variabes b T-Stat + b T-Stat + b T-Stat + b T-Stat + Medicare Payment (Pesos) Days in Charity Ward Days in Pay Ward Days in Private / Semi Private Room Lab Tests Radioogica Exams Physician Visits Surgeries Suppy Orders Drug Prescriptions Medicare Dummy and Interactions Medicare (=1) Days Pay Ward Medicare Days in Private Room Medicare Lab Tests Medicare Radioogica Exams Medicare Physician Visits Medicare Surgeries Medicare Suppy Orders Medicare Drug Prescriptions Medicare Patient Characteristics Age Mae (=1) Per Capita Househod Income Income Missing (=1) Coege (=1) Days in Bed Before Admitted Specification Tests Statistic P Vaue Statistic P Vaue Statistic P Vaue Statistic P Vaue F-Statistic for Patient Characteristics F-Statistic for Medicare Interactions Hausman Test for Random Effects F-Statistic for Hospita Fixed Effects Overa R-Squared Sampe Size: Patients Sampe Size: Hospitas T-Statistics are computed using White Heterosedastic Robust Standard Errors 24