HEALTH ECONOMICS Health Econ.10:457-471(2001) DOl:10.1002/hec.631 ECONOMETRICS AND HEALTH ECONOMICS EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION UNDER FAVOURABLE CONDITIONS FOR SUPPLIER-INDUCED DEMAND ARINE VAN DE VOORDE& EDDY VAN DOORSLAERB, AND ERIK SCHOKKaerta Department of Economics, University of Leuven, belgium b Department of Health Policy and Management, Erasmus University Rotterdam, Netherlands SUMMARY The effects of cost sharing on the demand for ambulatory care in experimental circumstances are well understood since the Rand Health Insurance Experiment(HIE). However, in a non-experimental real-world context supplier-induced demand of doctors might erode some of the significant negative out-of-pocket price elasticity identified in the HIE. Belgium is an interesting test case for this hypothesis because it has relatively high rates of tient cost sharing in its public health insurance system and a very high density of physicians, all remunerated fee-for-service. We have exploited the price variation generated by a substantial increase in patient co-payment ates in 1994 to estimate out-of-pocket price elasticities for three groups of users, and for three types of services sing a fixed-effects model in levels and in differences. We obtain significant out-of-pocket price elasticities for the general population in the range from -0.39 to -0.28 for GP home visits, -0.16 to -0 12 for GP offi and -0.10 for specialist visits. The estimates were generally lower and less significant for the groups of elderly and disabled. The differences we find in price responsiveness appear to be fairly robust and consistent with the hie predictions. These results suggest that -at least in the short run -non-experimental utilization effects of cost sharing are very similar to the experimental evidence, even in a situation of favourable conditions for supplier induced demand. Copyright c 2001 John Wiley Sons, Ltd KEY WORDS-Belgium; cost sharing; demand; price elasticity; supplier-induced demand INTRODUCTION Rand Health Insurance Experiment(HIE),some doubts remain regarding its external validity in The appropriate role for patient cost sharing in non-experimental real-world conditions []. The health care finance in general, and in public aim of this paper is to provide empirical evidence health insurance in particular, is not a new issue, on the effects of co-payments on the demand for and has been the subject of both extensive theo- physician services. We do so by drawing on the retical and empirical analysis and of intense policy Belgian experience in the period 1986-1995. Bel- debate for decades. Although authoritative evi- gium is an interesting country in this respect dence on the effects of patient cost sharing in because it has been making extensive use of pa experimental conditions is available since the tient out-of-pocket payments for a long time, and Correspondence to: Department of Health Policy and Management, Erasmus University Rotterdam, PO Box 1738, 3000 DR Rotterdam, Netherlands. Tel: +31 10 4088555: fax: +31 10 4089094: e-mail: vandoorslaer(@ bmg. eur. nl Received 16 October 2000 Copyright o 2001 John Wiley Sons, Ltd Accepted 8 May 2001
HEALTH ECONOMICS Health Econ. 10: 457–471 (2001) DOI: 10.1002/hec.631 ECONOMETRICS AND HEALTH ECONOMICS EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION UNDER FAVOURABLE CONDITIONS FOR SUPPLIER-INDUCED DEMAND CARINE VAN DE VOORDEa , EDDY VAN DOORSLAERb,* AND ERIK SCHOKKAERTa a Department of Economics, Uniersity of Leuen, Belgium b Department of Health Policy and Management, Erasmus Uniersity Rotterdam, Netherlands SUMMARY The effects of cost sharing on the demand for ambulatory care in experimental circumstances are well understood since the Rand Health Insurance Experiment (HIE). However, in a non-experimental real-world context, supplier-induced demand of doctors might erode some of the significant negative out-of-pocket price elasticity identified in the HIE. Belgium is an interesting test case for this hypothesis because it has relatively high rates of patient cost sharing in its public health insurance system and a very high density of physicians, all remunerated fee-for-service. We have exploited the price variation generated by a substantial increase in patient co-payment rates in 1994 to estimate out-of-pocket price elasticities for three groups of users, and for three types of services using a fixed-effects model in levels and in differences. We obtain significant out-of-pocket price elasticities for the general population in the range from −0.39 to −0.28 for GP home visits, −0.16 to −0.12 for GP office visits and −0.10 for specialist visits. The estimates were generally lower and less significant for the groups of elderly and disabled. The differences we find in price responsiveness appear to be fairly robust and consistent with the HIE predictions. These results suggest that—at least in the short run—non-experimental utilization effects of cost sharing are very similar to the experimental evidence, even in a situation of favourable conditions for supplierinduced demand. Copyright © 2001 John Wiley & Sons, Ltd. KEY WORDS — Belgium; cost sharing; demand; price elasticity; supplier-induced demand INTRODUCTION The appropriate role for patient cost sharing in health care finance in general, and in public health insurance in particular, is not a new issue, and has been the subject of both extensive theoretical and empirical analysis and of intense policy debate for decades. Although authoritative evidence on the effects of patient cost sharing in experimental conditions is available since the Rand Health Insurance Experiment (HIE), some doubts remain regarding its external validity in non-experimental real-world conditions [1]. The aim of this paper is to provide empirical evidence on the effects of co-payments on the demand for physician services. We do so by drawing on the Belgian experience in the period 1986–1995. Belgium is an interesting country in this respect, because it has been making extensive use of patient out-of-pocket payments for a long time, and * Correspondence to: Department of Health Policy and Management, Erasmus University Rotterdam, PO Box 1738, 3000 DR Rotterdam, Netherlands. Tel.: +31 10 4088555; fax: +31 10 4089094; e-mail: vandoorslaer@bmg.eur.nl Copyright © 2001 John Wiley & Sons, Ltd. Receied 16 October 2000 Accepted 8 May 2001
C. VAN DE VOORDE ET AL has gradually expanded and fine-tuned their use excluding most ambulatory care), and the general n its universal and compulsory public health in- plan for all other insured, mainly blue and white surance plan. But at the same time, Belgium is a collar workers in public and private service and country with one of the highest densities of doc- recipients of social security benefits(88% in 1995) tors in Europe, all remunerated exclusively on a covering all risks. Both plans cover the active and fee-for-service basis, thereby creating ideal condi- the non-active population, as well as their depen tions for supplier-induced responses in demand to dants. We will only consider the general plan. The exogenous shocks in income provision of medical care is predominantly pri- In an attempt to control rising public health vately organized in a government-regulated envi- are costs, the Belgian government has raised ronment. Patients have free choice of physician Ca coinsurance rates several times over this period, and direct access to GP, as well as to specialist with a fairly sharp increase on I January 1994 for care. Physician density in Belgium is extremely all insured, except for some exempt categories high by European standards. In 1995, there were (elderly, disabled, etc. )on low incomes. At the only 660 inhabitants per practising GP and 630 same time-and in order to mitigate the poten- per specialist tially harmful consequences for health care ac- Physician remuneration is on the basis of fee cess--income-related annual maximum amounts for-service with some patient co-payments. Pa- of user charges were introduced. Because not only tients pay the entire physician fee and get a fixed of care but also utilization volumes appeared to Physician fee schedules are established annually stop growing(or even to decrease)in the period but physicians can opt out of negotiated fees by 1994-1995, it is worthwhile to investigate the role formally objecting to them. In order to obtain the played by user charges in physician services uti lization rates privilege to 'overbill,, they are required to inform This paper is organized as follows. In the next schedules. There are no official estimates of the section, we briefly explain the most important features of the belgian hea Ith care and public number of specialists opting out of negotiated health insurance system, with an emphasis on the fees, but it is well known that this is most com- role of patient cost sharing. The third section mon in tertiary care like cardiac and other as paediatrics and gynaecology), and for the GP sharing, with a view to deriving some empirically opting out of the negotiated fees, it is less verifiable propositions about their likely effects. coon The reimbursement percentage differs by type section dataset that was available and the two of care and status of the insured. One group, different kinds of models used to analyse it. It which we will label "low-income WOPI, consist lso presents some testable hypotheses derived ing of widows, orphans, pensioners and invalids from the surveyed literature. The fifth section( disabled)(WOPT)with incomes below a certain presents and discusses the estimation results ob- ceiling, benefits from more generous reimburse- tained using both approaches. The sixth section ment. In 1995, for example, the price of a GP draws some conclusions from the findings. consultation (in the office) was 550 BFr(equiva lent to about 14 Euros), 30% of which is an out-of-pocket co-payment for the general popula tion, while the co-payment was only 8% for PATIENT COST SHARING IN PUBLIC HEALTH INSURANCE IN BELGIUM low-income WOPI. The price of a specialist con sultation was about 840 BFr, but had higher coinsurance rates (40% and 14%0, respectively) Compulsory health insurance covers nearly all The price of a GP home visit was slightly higher citizens in Belgium(99% in 1995)and is financed (670 BFr), with 35%0 and 8% co-payment. In primarily by income-related social insurance con- contrast to the French situation [3], reinsurance of tributions [2]. There are two main insurance these co-payments is virtually non-existent in Bel plans: one plan for the self-employed(12% of all gium. Some private insurance for co-payments is insured in 1995) covering only major risks (i.e. being offered, but mostly as a fringe benefit for Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
458 C. VAN DE VOORDE ET AL. has gradually expanded and fine-tuned their use in its universal and compulsory public health insurance plan. But at the same time, Belgium is a country with one of the highest densities of doctors in Europe, all remunerated exclusively on a fee-for-service basis, thereby creating ideal conditions for supplier-induced responses in demand to exogenous shocks in income. In an attempt to control rising public health care costs, the Belgian government has raised coinsurance rates several times over this period, with a fairly sharp increase on 1 January 1994 for all insured, except for some exempt categories (elderly, disabled, etc.) on low incomes. At the same time—and in order to mitigate the potentially harmful consequences for health care access—income-related annual maximum amounts of user charges were introduced. Because not only public health care expenditures for certain types of care but also utilization volumes appeared to stop growing (or even to decrease) in the period 1994–1995, it is worthwhile to investigate the role played by user charges in physician services utilization rates. This paper is organized as follows. In the next section, we briefly explain the most important features of the Belgian health care and public health insurance system, with an emphasis on the role of patient cost sharing. The third section briefly reviews some of the main findings of a selection of previous empirical studies on cost sharing, with a view to deriving some empirically verifiable propositions about their likely effects. The fourth section describes the time-series crosssection dataset that was available and the two different kinds of models used to analyse it. It also presents some testable hypotheses derived from the surveyed literature. The fifth section presents and discusses the estimation results obtained using both approaches. The sixth section draws some conclusions from the findings. PATIENT COST SHARING IN PUBLIC HEALTH INSURANCE IN BELGIUM Compulsory health insurance covers nearly all citizens in Belgium (99% in 1995) and is financed primarily by income-related social insurance contributions [2]. There are two main insurance plans: one plan for the self-employed (12% of all insured in 1995) covering only major risks (i.e. excluding most ambulatory care), and the general plan for all other insured, mainly blue and white collar workers in public and private service and recipients of social security benefits (88% in 1995), covering all risks. Both plans cover the active and the non-active population, as well as their dependants. We will only consider the general plan. The provision of medical care is predominantly privately organized in a government-regulated environment. Patients have free choice of physician and direct access to GP, as well as to specialist care. Physician density in Belgium is extremely high by European standards. In 1995, there were only 660 inhabitants per practising GP and 630 per specialist. Physician remuneration is on the basis of feefor-service with some patient co-payments. Patients pay the entire physician fee and get a fixed percentage reimbursed from their sickness fund. Physician fee schedules are established annually but physicians can opt out of negotiated fees by formally objecting to them. In order to obtain the privilege to ‘overbill’, they are required to inform patients of their non-adherence to the national fee schedules. There are no official estimates of the number of specialists opting out of negotiated fees, but it is well known that this is most common in tertiary care like cardiac and other surgery. For the ‘direct access’ specialisms (such as paediatrics and gynaecology), and for the GPs opting out of the negotiated fees, it is less common. The reimbursement percentage differs by type of care and status of the insured. One group, which we will label ‘low-income WOPI’, consisting of widows, orphans, pensioners and invalids (disabled) (WOPI) with incomes below a certain ceiling, benefits from more generous reimbursement. In 1995, for example, the price of a GP consultation (in the office) was 550 BFr (equivalent to about 14 Euros), 30% of which is an out-of-pocket co-payment for the general population, while the co-payment was only 8% for low-income WOPI. The price of a specialist consultation was about 840 BFr, but had higher coinsurance rates (40% and 14%, respectively). The price of a GP home visit was slightly higher (670 BFr), with 35% and 8% co-payment. In contrast to the French situation [3], reinsurance of these co-payments is virtually non-existent in Belgium. Some private insurance for co-payments is being offered, but mostly as a fringe benefit for Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION 459 ly covers the co- GP visits were delivered at the patients home in ts for hospitalizations, not for outpatient 1994 [6]. But the level and mix of consultation ecently, such private supplementary cover rates also differ substantially by socio-economic has begun to include pre-and post-hospitalization group, as we will see in Section 4 below costs as well, but this type of policy was still very rare in 1994 The Belgian health insurance system has been EVIDENCE ON PRICE SENSITIVITY OF making use of public sector cost sharing for HEALTH CARE DEMAND decades, but in the early 1990s, user charges for GP and specialist services were raised substan tially in real terms for the general non-exempt In the literature, the effects of cost sharing now well un sharing is part of a more general cost containment result of the findings that emerged from the well policy dictated by health care financing problems. known Rand Health Insurance Experiment (HIE) Policy makers do, however, realize that the rev- [7, 8]. The literature was recently reviewed in two enue-generating capacity of these measures is lim- chapters of the Handbook of Health Economics ited and are aware of potential access problems Cutler and Zeckhauser 9) conclude from a review for low-income high-users. Therefore, the 1994 of the estimates that'The demand elasticities in co-payment rate increases were accompanied by the Rand Experiment have become the standard the introduction of income-related annual stop- in the literature, and essentially all economists loss arrangements. Two types of stop-loss were accept that traditional health insurance leads to fines a post-tax income-dependent cost-sharing Zweite te moral hazard in demand(p. 584).Also ceiling. All charges in excess of the ceiling are evidence from both the experimental and non- fully reimbursed. However, as in general such experimental literature, and they reach similar conclusions. It is now considered more or less reimbursement will only occur 2-3 years later received wisdom that: (i demand elasticities for (after the taxpayers have filed the tax return), its medical care are non-zero and negative but small, mmediate effect on health care consumption is fairly negligible, and will certainly not have had -0.2, (i)the price responsiveness is not higher groups, except for ambulatory care visits, (iii) low ser payment ceiling of 15000 BFr per year for levels of cost sharing have no discernible effects eral socially weaker groups: the low-income on health status; exceptions to this rule are indi OPI, long-term unemployed and certain in viduals with low socio-economic status viduals receiving welfare benefits. In practice, health, (iv) cost sharing primarily reduces medical however, relatively few people reached the ceiling consumption for less serious symptoms, (v)there for medicines) were excluded from the de- induced by cost sharing have a larger effect on ductible [4] the use of less effective outpatient care or non Despite a much more extensive use of patient indicated hospital admissions. Decreased utiliza- cost-sharing in Belgium, the frequency of physi- tion from cost sharing appears to affect the use of cian visits, especially to a general practitioner, is appropriate care to a similar degree as inappropri- much higher than in most other European coun- ate or rarely effective care, and(vi) price respon- tries. As shown by data from the European Com- siveness generally rises with rising rates of munity Household Panel Survey of 1996, for coinsurance xample, Belgium has the highest annual rate of However, the main advantage of the rand GP visits (5.24 visits per adult per year), and the Study, that is, that it was a carefully designed fourth highest number of specialist visits(1.89 per experiment, in which only selected individual adult per year) of all EU member states [5]. from a few selected sites were randomly assigned Belgium has traditionally also had exceptionally to groups with varying rates of coinsurance, was high rates of home visits by GPs, and although also considered its most significant limitation in the rate has been declining, still over 40% of all terms of policy relevance by some commentators Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION 459 certain employees, and it only covers the copayments for hospitalizations, not for outpatient care. Recently, such private supplementary cover has begun to include pre- and post-hospitalization costs as well, but this type of policy was still very rare in 1994. The Belgian health insurance system has been making use of public sector cost sharing for decades, but in the early 1990s, user charges for GP and specialist services were raised substantially in real terms for the general non-exempt population. The increased use of patient cost sharing is part of a more general cost containment policy dictated by health care financing problems. Policy makers do, however, realize that the revenue-generating capacity of these measures is limited and are aware of potential access problems for low-income high-users. Therefore, the 1994 co-payment rate increases were accompanied by the introduction of income-related annual stoploss arrangements. Two types of stop-loss were introduced. The so-called ‘franchise fiscale’ defines a post-tax income-dependent cost-sharing ceiling. All charges in excess of the ceiling are fully reimbursed. However, as in general such reimbursement will only occur 2–3 years later (after the taxpayers have filed the tax return), its immediate effect on health care consumption is fairly negligible, and will certainly not have had an influence on our study results. The franchise sociale on the other hand, defines an immediate co-payment ceiling of 15000 BFr per year for several socially weaker groups: the low-income WOPI, long-term unemployed and certain individuals receiving welfare benefits. In practice, however, relatively few people reached the ceiling since some important types of co-payments (such as for medicines) were excluded from the deductible [4]. Despite a much more extensive use of patient cost-sharing in Belgium, the frequency of physician visits, especially to a general practitioner, is much higher than in most other European countries. As shown by data from the European Community Household Panel Survey of 1996, for example, Belgium has the highest annual rate of GP visits (5.24 visits per adult per year), and the fourth highest number of specialist visits (1.89 per adult per year) of all EU member states [5]. Belgium has traditionally also had exceptionally high rates of home visits by GPs, and although the rate has been declining, still over 40% of all GP visits were delivered at the patient’s home in 1994 [6]. But the level and mix of consultation rates also differ substantially by socio-economic group, as we will see in Section 4 below. EVIDENCE ON PRICE SENSITIVITY OF HEALTH CARE DEMAND In the literature, the effects of cost sharing now seem reasonably well understood. This is mainly a result of the findings that emerged from the wellknown Rand Health Insurance Experiment (HIE) [7,8]. The literature was recently reviewed in two chapters of the Handbook of Health Economics. Cutler and Zeckhauser [9] conclude from a review of the estimates that ‘The demand elasticities in the Rand Experiment have become the standard in the literature, and essentially all economists accept that traditional health insurance leads to moderate moral hazard in demand’ (p. 584). Also Zweifel and Manning [1] review the empirical evidence from both the experimental and nonexperimental literature, and they reach similar conclusions. It is now considered more or less received wisdom that: (i) demand elasticities for medical care are non-zero and negative, but small, typical estimates being in the range from −0.1 to −0.2, (ii) the price responsiveness is not higher for low-income groups than for high income groups, except for ambulatory care visits, (iii) low levels of cost sharing have no discernible effects on health status; exceptions to this rule are individuals with low socio-economic status and poor health, (iv) cost sharing primarily reduces medical consumption for less serious symptoms, (v) there is no evidence that reductions in consumption induced by cost sharing have a larger effect on the use of less effective outpatient care or nonindicated hospital admissions. Decreased utilization from cost sharing appears to affect the use of appropriate care to a similar degree as inappropriate or rarely effective care, and (vi) price responsiveness generally rises with rising rates of coinsurance. However, the main advantage of the Rand Study, that is, that it was a carefully designed experiment, in which only selected individuals from a few selected sites were randomly assigned to groups with varying rates of coinsurance, was also considered its most significant limitation in terms of policy relevance by some commentators: Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
460 C. VAN DE VOORDE ET AL Although the hie produced the most rigorous specialist care using cross-sectional data from the results to date on how individuals and families 1997 Belgian Health Interview Survey, and found react to cost sharing, it was from the outset that, after controlling for differences in levels of incapable of estimating the effect of cost sharing health, income, education and physician supply, on an entire population because it could not the group facing reduced co-payment rates (i.e assess how the health care system would react. the low-income WoPI) had a 35% higher proba- f, on the contrary, all physician and hospital bility of seeing a GP in a given time period. These utilization by all patients in the sites had been results suggest that co-payments in the bea s subjected to cost-sharing, and all individuals re- context do seem to have a moderating effect sponded as the hiE coinsurance groups did, then physician utilization. This suggests that doctors health care providers would have experienced a ability to preserve incomes, for example, by 15%-30% decrease in (gross)incomes/revenues. creasing the number of physician-initiated visits, supplier response, but how large, and of what the Belgian fee-for-service remuneration setting, ype? These are critical questions for policy- any reduction in utilization automatically leads to makers considering the introduction of user a proportional decrease in average phy harges. Unfortunately the HIE cannot provide comes, as doctors have very few options to in- crease the revenue per item of service, for Although Newhouse[8] argues that the magni- example, by substituting higher priced visits tude of such a supplier-induced demand response (home visits, weekend visits) for lower priced of- kely to be limited and concludes his book with fice visi the statement that 'the evidence thus seems con- Two recent examples from France which were vincing to us that a widespread increase in cost not included in the reviews have some relevance sharing would reduce demand and use, just as it for the present study. Chiappori et al. 13, 14]exam did when employed on a small scale in the Exper- ined the response in the demand for home and to examine whether some of the experimental conduct a e nd specialists in France after the introduction l0% experiment by comparing a findings can be replicated in a real world setting sample of bank and insurance employees with a which is not isolated from potential supplier re- (reinsured)control group. Their results indicate a modest effect on GP home visits but no effect on the literature for countries where coinsurance office visits to either the GP or the specialist. Thi rates approximating the rates used in the HE are finding of a very weak price-sensitivity is at tributed to the small share of the monetary price There are some earlier estimates of short-run in the total price of an office visit, as compared price elasticities for Belgium. Van Doorslaer [l] with the substantial share of time and transporta exploited the price variation induced by a reform tion costs. But even these french rates of co- of the co-payment structure for prescription drugs the early eighties to estimate significant nega- applied in the Rand HIE, and the sample used is tive own-price elasticities of the demand for(cer- highly selective. As a result, it is not yet clear tain types of) prescription drugs and cross-price what effects to expect of HIE-size coinsurance/ elasticities of the price of doctor visits. He found Co-payment rates in a real-world setting that the active populations demand for drugs was more responsive to co-payments than the non- active population, and that among the non-active, hose on low incomes DATA ESTIMATION AND TESTING Also Carrin and Van Daal [12], who estimated STRATEGY own-price and cross-price elasticities of the de mand for physiotherapy and dental care, found Our approach consists basically of a test of the that the demand of the active population was less total utilization response to the exogenous change price inelastic than that of the non-active part of in out-of-pocket prices that occurred in 1994.Our the population. Adriaenssen and De graeve [13 data do not allow us to separate the demand recently estimated demand equations for GP andand the supply response, but we can obtain an Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
460 C. VAN DE VOORDE ET AL. ‘Although the HIE produced the most rigorous results to date on how individuals and families react to cost sharing, it was from the outset incapable of estimating the effect of cost sharing on an entire population because it could not assess how the health care system would react. . . . If, on the contrary, all physician and hospital utilization by all patients in the sites had been subjected to cost-sharing, and all individuals responded as the HIE coinsurance groups did, then health care providers would have experienced a 15%-30% decrease in (gross) incomes/revenues. This would almost certainly have elicited some supplier response, but how large, and of what type? These are critical questions for policymakers considering the introduction of user charges. Unfortunately the HIE cannot provide the answers’ (Stoddart et al. [10]). Although Newhouse [8] argues that the magnitude of such a supplier-induced demand response is likely to be limited and concludes his book with the statement that ‘the evidence thus seems convincing to us that a widespread increase in cost sharing would reduce demand and use, just as it did when employed on a small scale in the Experiment’ (p. 371), it seems, nevertheless, warranted to examine whether some of the experimental findings can be replicated in a real world setting which is not isolated from potential supplier response. Surprisingly little evidence is available in the literature for countries where coinsurance rates approximating the rates used in the HIE are in effect. There are some earlier estimates of short-run price elasticities for Belgium. Van Doorslaer [11] exploited the price variation induced by a reform of the co-payment structure for prescription drugs in the early eighties to estimate significant negative own-price elasticities of the demand for (certain types of) prescription drugs and cross-price elasticities of the price of doctor visits. He found that the active population’s demand for drugs was more responsive to co-payments than the nonactive population, and that among the non-active, those on low incomes were more price responsive. Also Carrin and Van Daal [12], who estimated own-price and cross-price elasticities of the demand for physiotherapy and dental care, found that the demand of the active population was less price inelastic than that of the non-active part of the population. Adriaenssen and De Graeve [13] recently estimated demand equations for GP and specialist care using cross-sectional data from the 1997 Belgian Health Interview Survey, and found that, after controlling for differences in levels of health, income, education and physician supply, the group facing reduced co-payment rates (i.e. the low-income WOPI) had a 35% higher probability of seeing a GP in a given time period. These results suggest that co-payments in the Belgian context do seem to have a moderating effect on physician utilization. This suggests that doctors’ ability to preserve incomes, for example, by increasing the number of physician-initiated visits, is either very limited or already fully exploited. In the Belgian fee-for-service remuneration setting, any reduction in utilization automatically leads to a proportional decrease in average physician incomes, as doctors have very few options to increase the revenue per item of service, for example, by substituting higher priced visits (home visits, weekend visits) for lower priced office visits. Two recent examples from France which were not included in the reviews have some relevance for the present study. Chiappori et al. [3,14] examined the response in the demand for home and office visits to GPs and specialists in France after the introduction of a 10% co-payment rate. They conduct a natural experiment by comparing a sample of bank and insurance employees with a (reinsured) control group. Their results indicate a modest effect on GP home visits, but no effect on office visits to either the GP or the specialist. This finding of a very weak price-sensitivity is attributed to the small share of the monetary price in the total price of an office visit, as compared with the substantial share of time and transportation costs. But even these French rates of copayment are very small compared to the ones applied in the Rand HIE, and the sample used is highly selective. As a result, it is not yet clear what effects to expect of HIE-size coinsurance/ co-payment rates in a real-world setting. DATA, ESTIMATION AND TESTING STRATEGY Our approach consists basically of a test of the total utilization response to the exogenous change in out-of-pocket prices that occurred in 1994. Our data do not allow us to separate the demand and the supply response, but we can obtain an Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION estimate to see what the total utilization response normal 'official public co-payments resulting has been under relatively favourable conditions from utilization could be included, not any add for supplier-induced demand. If the utilization tional out-of-pocket payments resulting from pos- response is similar to the demand response identi- sible physician overbilling. Utilization is expressed fied in the hie, we will interpret this as implying as the average number of visits per member reim- that any supplier response-at least in the short bursed by the sickness fund. There is substantial run of 2 years--must have been relatively small. interregional variation in the Belgian system, ow ing to local demand and supply conditions, and to differences in provider behaviour. The number of GPs per 10000 inhabitants varies over the regions from a minimum of 12 to a maximum of 23(in There are about 100(non-profit)local sickness the capital). This interregional variation has re- funds ivel in five associations. In the compulsory elgium, which are grouped at the na- mained stable during the period 1986-1995 tional level insurance system, the freedom of these sickness funds is very limited, and there is no variation at Three categories of users all in the policies offered. We obtained regionally aggregated expenditure and utilization data for We do not have any further information on in- GP and specialist visits for all insured of the come or health status of the insured, but we can Federation of Christian Mutualities, the largest to some extent, stratify by income and health national association, with 4.4 million members in status by comparing price effects for three sepa 1996. This federation operates nationwide, but rate groups of insured: (i)the non-exempt (ger as a larger market share in the Dutch-speaking eral)population, which includes all salaried and region of Flanders. With about 45% of the Bel- social welfare beneficiaries and their dependants, gian population among its members, it is fairly (i) the 'other'(i.e. non-low-income) WOPI and representative of the country. Average expendi-(ii) the low-income WOPI. The administrative tures per member and average number of cons ul- data of the sickness funds make a distinction tations of GPs and specialists are close to the between these three groups and this distinction average of the Belgian population. The same is has been kept at the level of the aggregate data at true for the socio-economic background of the our disposal. Social groups(i)and(ii)are treated members: number of migrants, number of unem- identically by the insurance system, but differ in ployed, number of manual workers are all close to terms of average health. Social groups(ii) and (ii) the national average (information based on per- differ in terms of average income and, therefore, sonal communication with the sickness funds). also in terms of living circumstances. One can a There is no reason why doctors would treat pa- priori expect that their health status is not identi tients of different sickness funds differently, as all cal, but clearly, they are more similar to each are subjected to identical insurance and reim- other than to the general population. The WOPI bursement conditions group includes about one quarter of the belgian We obtained pooled data from a cross-section population. This proportion has remained rela of the 31 regional offices of the Christian Mutual- tively stable over time. However, within the ities for a period of 10 years(1986-1995). Unfor- woPI the proportion of the low-income group tunately, there was a break in the series thereafter, has been declining in the period 1986-1995 such that no more recent data could be obtained. While in 1990 it was 48.5%. it had declined to The data distinguish three categories of users: at 43. 3% in 1994 the level of these user categories the scale of the regional offices(and hence the aggregation level of our data) varies from a minimum of about Three types of utilization 2000 members to a maximum of more than 300000 members. The actual average amount of In principle, when somebody wants to consult a the co-payment per visit is expressed in constant doctor, s/he basically has a choice between three 1988 prices, and was computed by dividing the options: (i)calling a GP for a home visit, which is otal non-reimbursed expenditure by the total slightly more expensive, but has much lower time number of visits. In fact, this means that only the costs, (ii) going to see a GP in his office, and (iii) Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION 461 estimate to see what the total utilization response has been under relatively favourable conditions for supplier-induced demand. If the utilization response is similar to the demand response identified in the HIE, we will interpret this as implying that any supplier response—at least in the short run of 2 years—must have been relatively small. Data There are about 100 (non-profit) local sickness funds in Belgium, which are grouped at the national level in five associations. In the compulsory insurance system, the freedom of these sickness funds is very limited, and there is no variation at all in the policies offered. We obtained regionally aggregated expenditure and utilization data for GP and specialist visits for all insured of the Federation of Christian Mutualities, the largest national association, with 4.4 million members in 1996. This federation operates nationwide, but has a larger market share in the Dutch-speaking region of Flanders. With about 45% of the Belgian population among its members, it is fairly representative of the country. Average expenditures per member and average number of consultations of GPs and specialists are close to the average of the Belgian population. The same is true for the socio-economic background of the members: number of migrants, number of unemployed, number of manual workers are all close to the national average (information based on personal communication with the sickness funds). There is no reason why doctors would treat patients of different sickness funds differently, as all are subjected to identical insurance and reimbursement conditions. We obtained pooled data from a cross-section of the 31 regional offices of the Christian Mutualities for a period of 10 years (1986–1995). Unfortunately, there was a break in the series thereafter, such that no more recent data could be obtained. The data distinguish three categories of users: at the level of these user categories the scale of the regional offices (and hence the aggregation level of our data) varies from a minimum of about 2000 members to a maximum of more than 300000 members. The actual average amount of the co-payment per visit is expressed in constant 1988 prices, and was computed by dividing the total non-reimbursed expenditure by the total number of visits. In fact, this means that only the normal ‘official’ public co-payments resulting from utilization could be included, not any additional out-of-pocket payments resulting from possible physician overbilling. Utilization is expressed as the average number of visits per member reimbursed by the sickness fund. There is substantial interregional variation in the Belgian system, owing to local demand and supply conditions, and to differences in provider behaviour. The number of GPs per 10000 inhabitants varies over the regions from a minimum of 12 to a maximum of 23 (in the capital). This interregional variation has remained stable during the period 1986–1995. Three categories of users We do not have any further information on income or health status of the insured, but we can, to some extent, stratify by income and health status by comparing price effects for three separate groups of insured: (i) the non-exempt (general) population, which includes all salaried and social welfare beneficiaries and their dependants; (ii) the ‘other’ (i.e. non-low-income) WOPI and (iii) the low-income WOPI. The administrative data of the sickness funds make a distinction between these three groups and this distinction has been kept at the level of the aggregate data at our disposal. Social groups (i) and (ii) are treated identically by the insurance system, but differ in terms of average health. Social groups (ii) and (iii) differ in terms of average income and, therefore, also in terms of living circumstances. One can a priori expect that their health status is not identical, but clearly, they are more similar to each other than to the general population. The WOPI group includes about one quarter of the Belgian population. This proportion has remained relatively stable over time. However, within the WOPI the proportion of the low-income group has been declining in the period 1986–1995. While in 1990 it was 48.5%, it had declined to 43.3% in 1994. Three types of utilization In principle, when somebody wants to consult a doctor, s/he basically has a choice between three options: (i) calling a GP for a home visit, which is slightly more expensive, but has much lower time costs, (ii) going to see a GP in his office, and (iii) Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
C. VAN DE VOORDE ET AL going to see a specialist in his office, as specialists trated in the last 3 years of the observation pe are directly accessible without a GP referral. riod, the years prior to the change are useful in Again, the latter is somewhat more expensive estimating any underlying trends. Figures I(a) (higher fee and higher co-payment rate), but is 3(b) serve to illustrate some of the overall time possibly perceived as higher quality too. It is well trends and changes in the utilization,and the known that, for common specialities,, like, for co-payments for the three types of care for the example, gynaecologists, paediatricians, dermatol- general population (in Figure 1(a)and(b),the ogists, ophthalmologists etc, a la share of wopi (in Figure 2(a)and(b)and the low consultations in Belgium occur without a GP Income (1.e referral. In Belgium (unlike the situation in empt) WoPI (in Figure 3(a) and France [14]), all GPs deliver home visits, and (b). They show (i that the GP home visits are there are no systematic differences in the quality most prevalent among the woPl, but especia of services offered by GPs during home and office among those on low incomes, (i) that for the firs two categories of insured, a(negative)deviation from the trend occurred for all types of care immediately after the co-payment rise in 1994, (iii) that little or nothing seems to have happened for the exempt group, for whom co-payments were We have used time series data on three categories not raised, (iv) that, in contrast to the trends for of outpatient care: GP home visits, GP office most other types of utilization, the trend for GP consultations and specialist office consultations. home visits in the general population was already Although most of the price variation is concen- downward prior to the 1994 change 3,0 e-GP home visits 2.0 — GP office visits -specialist visits 1.6 GP office visits r specialist visits year Figure 1.(a) Physician utilization rates, general population; (b)co-payments, general population Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
462 C. VAN DE VOORDE ET AL. going to see a specialist in his office, as specialists are directly accessible without a GP referral. Again, the latter is somewhat more expensive (higher fee and higher co-payment rate), but is possibly perceived as higher quality too. It is well known that, for ‘common specialities’, like, for example, gynaecologists, paediatricians, dermatologists, ophthalmologists etc., a large share of consultations in Belgium occur without a GP referral. In Belgium (unlike the situation in France [14]), all GPs deliver home visits, and there are no systematic differences in the quality of services offered by GPs during home and office visits. Trends We have used time series data on three categories of outpatient care: GP home visits, GP office consultations and specialist office consultations. Although most of the price variation is concentrated in the last 3 years of the observation period, the years prior to the change are useful in estimating any underlying trends. Figures 1(a)– 3(b) serve to illustrate some of the overall time trends and changes in the utilization, and the co-payments for the three types of care for the general population (in Figure 1(a) and (b)), the WOPI (in Figure 2(a) and (b)) and the lowincome (i.e. exempt) WOPI (in Figure 3(a) and (b)). They show (i) that the GP home visits are most prevalent among the WOPI, but especially among those on low incomes, (ii) that for the first two categories of insured, a (negative) deviation from the trend occurred for all types of care immediately after the co-payment rise in 1994, (iii) that little or nothing seems to have happened for the exempt group, for whom co-payments were not raised, (iv) that, in contrast to the trends for most other types of utilization, the trend for GP home visits in the general population was already downward prior to the 1994 change. Figure 1. (a) Physician utilization rates, general population; (b) co-payments, general population Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION 5.0 4.0 GP home visits 酱一 GP office visits specialist visits 2.5 2.0 868788899099293949 210 -GP home visits 150 - GP office visits 30 110 68788899099293 year Figure 2.(a) Physician utilization rates, WOPI;(b)co-payments, WOPI Estimation strateg) possibility that the effects of other, unobserved influences on utilization unrelated to prices are Given the nature of the data, there are, in princi- indirectly controlled for. Obviously, because a ple, two options for estimating the price respon- natural experiment does not involve randomized siveness, each with its advantages and dis- controlled conditions, we can never be sure that advantages. We have pursued both. The first and the control group is homogeneous to the experi most straightforward option is to estimate the mental groups. On the contrary, it is very clear price responses as deviations from the time trend in that the low-income WoPI group constitutes a utilization, while controlling for regional hetero- very selective and protected subpopulation of the geneity. This is what we will call the levels model. socially weakest groups in Belgian society, which Its main disadvantages are that it does not really cannot be assumed to be identical to the other exploit the natural experiment type character of two groups. Their much higher utilization rates he data, and that it is rather dependent on the (cf Figures 1-3) illustrate this fact. However, w appropriate specification of the time trend. The will try to correct for this heterogeneity by allow second option is to estimate the price-sensitivity ing for different fixed effects and different trends of the two 'experimental groups'(for which an We will call this the differences model. A priori, it togenous price shock occurred) as a deviation is not obvious which of the two approaches rom the control group of low-income WOPI. By preferable, as this crucially depends on how well focusing on the utilization differences between this selective group can function as a control experimental groups and control group, there is a group. One can not be sure, for instance, that Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION 463 Figure 2. (a) Physician utilization rates, WOPI; (b) co-payments, WOPI Estimation strategy Given the nature of the data, there are, in principle, two options for estimating the price responsiveness, each with its advantages and disadvantages. We have pursued both. The first and most straightforward option is to estimate the price responses as deiations from the time trend in utilization, while controlling for regional heterogeneity. This is what we will call the leels model. Its main disadvantages are that it does not really exploit the natural experiment type character of the data, and that it is rather dependent on the appropriate specification of the time trend. The second option is to estimate the price-sensitivity of the two ‘experimental groups’ (for which an exogenous price shock occurred) as a deiation from the control group of low-income WOPI. By focusing on the utilization differences between experimental groups and control group, there is a possibility that the effects of other, unobserved influences on utilization unrelated to prices are indirectly controlled for. Obviously, because a natural experiment does not involve randomized controlled conditions, we can never be sure that the control group is homogeneous to the experimental groups. On the contrary, it is very clear that the low-income WOPI group constitutes a very selective and protected subpopulation of the socially weakest groups in Belgian society, which cannot be assumed to be identical to the other two groups. Their much higher utilization rates (cf. Figures 1–3) illustrate this fact. However, we will try to correct for this heterogeneity by allowing for different fixed effects and different trends. We will call this the differences model. A priori, it is not obvious which of the two approaches is preferable, as this crucially depends on how well this selective group can function as a control group. One can not be sure, for instance, that Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
C. VAN DE VOORDE ET AL 555555 GP home visits specialist visits GP home visits GP office visits 40 68788899992939495 Figure 3.(a) Physician utilization rates, low-income wOPI;(b)co-payments, low-income wo the low-income WoPI group's utilization be- j=l, 2, 3, we have specified the following lin haviour-or the behaviour of their doctors -has model not been affected by the media publicity the rise in co-payments to most other groups In both models, interregional variation is cap. lnq饿k= agk In paik+BkT+ R1+k(1) tured by the introduction of fixed effects. As some of the regions are quite large, these regional fixed where qok is the physician utilization rate in re- effects may not fully account for all sub-regional gion i at time I for services j for group k; Pirk is differences. More specifically, some tendencies of the out-of-pocket price for group k for services j supplier inducement may disappear through the in region i at time t; T, is a linear time variable aggregation process. Our results have to be inter- (I=, .. 10): R, is a dummy variable for region preted with this caveat in mind i(=1,……,31); Eirik is an error term with(y,k E(4nk)=0 For each k and j, we have pooled time-series Levels model and cross-sectional data for 10 years and for 31 regions in order to have 310 observations avail- To obtain estimates of the respective price elastic- able in total. We have adopted a double -logarith ities, we have first estimated a system of nine mic specification in price and quantity in order to equations, one for each of the three user groups obtain estimates of constant price elasticities(the and for each type of utilization. So, for every %k). Second, a linear time trend (estimated by the group k= 1, 2, 3 and for every type of utilization Bk coefficients)is included, as it is obvious from Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
464 C. VAN DE VOORDE ET AL. Figure 3. (a) Physician utilization rates, low-income WOPI; (b) co-payments, low-income WOPI the low-income WOPI group’s utilization behaviour—or the behaviour of their doctors—has not been affected by the media publicity around the rise in co-payments to most other groups. In both models, interregional variation is captured by the introduction of fixed effects. As some of the regions are quite large, these regional fixed effects may not fully account for all sub-regional differences. More specifically, some tendencies of supplier inducement may disappear through the aggregation process. Our results have to be interpreted with this caeat in mind. Leels model To obtain estimates of the respective price elasticities, we have first estimated a system of nine equations, one for each of the three user groups and for each type of utilization. So, for every group k=1, 2, 3 and for every type of utilization j=1, 2, 3, we have specified the following linear model ln qitjk=jk ln pitjk+jkTt+ 31 i=1 ijkRi+itjk (1) where qitjk is the physician utilization rate in region i at time t for services j for group k; pitjk is the out-of-pocket price for group k for services j in region i at time t; Tt is a linear time variable (t=1, . . . , 10); Ri is a dummy variable for region i (i=1, . . . , 31); itjk is an error term with (j, k); E(itjk)=0. For each k and j, we have pooled time-series and cross-sectional data for 10 years and for 31 regions in order to have 310 observations available in total. We have adopted a double-logarithmic specification in price and quantity in order to obtain estimates of constant price elasticities (the jk). Second, a linear time trend (estimated by the jk coefficients) is included, as it is obvious from Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION the graphs that any response in utilization be- difference between the equation of the experimen haviour has to be measured as a deviation from tal group and the control group, for k= 1, 2 the time trend until the e co- ayment change Third, we have pooled the data for all 31 regions, In gajk-In qu3=a, (In Pujk-In pa3) but have allowed for nal heterogeneity in utilization levels by estimating a fixed effects Bk-3T+∑7k-3R+4mk-3(2) model using the weighted least-squares dummy variables(WLSDV) method. The 7k estimate the where regional fixed effects. Finally, we have tested for the presence of heteroskedasticity using the White -1=/3-a test [15], and for the presence of autocorrelation Bxk-3)=Bk-B3 using the adjusted Durbin-Watson statistic for nel data [16]. Where either autocorrelation or ik-3)-70k-7y heteroskedasticity is a problem, we have used Eank-3)=fark -er Arellano's [17 method to compute heteroscedas- ticity and serial correlation consistent standard As we will automatically obtain two separate errors in panel data models for within-group esti- estimates of a,(for j=l,., 3), one from the mators of linear regression models [18]. This re- difference equation for the general population and quires re-estimation of Equation (1), using the another from the equation for high-income WOPI method of within-group deviations from time (call them %wk) for k=l, 2), we can test whether means on transformed variables and computing the homogeneity assumption is rejected by testing White's[19] robust estimates for the standard for ax=ay2 for j=l,-.,3 Testing strategy Diferences mode Estimation of Equation()provides j*k parame- This model assumes that the low-income woP each combination of k and ]), whereas estimation can be regarded as an appropriate control group, of Equation(2)provides j*(k-1)parameter vec. groups can be measured as a deviation from the tors [awey, Bok-3y Yunk-3(i=\- sog/ues of these differential utilization of experimental and control hypotheses concerning the relative groups as a function of the price differential. a parameters crucial homogeneity assumption is inescapable for 1. Estimated price elasticities are negative for all his approach to work, namely that the price groups and for all types of care elasticity in the control group is equal to that in 2. In terms of between-group differences, we ex- the experimental group, i.e. that ak=r3 Vj and pect the general population to be more price- k=1. 2. In other words we have to assume that responsive than the 'needy groups(high-and fter correction for any possible differences in low-income WOPI). One argument for this time trends and regional effects, both groups are homogeneous with respect to their price-sensitiv- or pothesis is the better average health status he general population. It is also possible te ity. It is then as if both groups are drawn from the argue that the utilization by the WOPI is less same population and only one of the groups is price elastic, because they can be persuaded ubjected to price variation. The main problem more easily by the physicians to take more with this approach is that, in general, one can treatment. our data do not allow us to distin never be certain about the validity of the homo- guish the demand and the supply effects, as we geneity assumption In our case, however, we can are only capable of estimating the overall price test whether the price elasticities of the two exper- effect. For the low-income WOPI, price-sensi- imental groups are equal tivity can hardly be measured owing to lack of The ces for the estimation procedure price variation in the period considered and 3. In terms of between-visits differences we ex pect(a)a higher price sensitivity for GP than Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
EFFECTS OF COST SHARING ON PHYSICIAN UTILIZATION 465 the graphs that any response in utilization behaviour has to be measured as a deviation from the time trend until the co-payment change. Third, we have pooled the data for all 31 regions, but have allowed for regional heterogeneity in utilization levels by estimating a fixed effects model using the weighted least-squares dummy variables (WLSDV) method. The ijk estimate the regional fixed effects. Finally, we have tested for the presence of heteroskedasticity using the White test [15], and for the presence of autocorrelation using the adjusted Durbin–Watson statistic for panel data [16]. Where either autocorrelation or heteroskedasticity is a problem, we have used Arellano’s [17] method to compute heteroscedasticity and serial correlation consistent standard errors in panel data models for within-group estimators of linear regression models [18]. This requires re-estimation of Equation (1), using the method of within-group deviations from time means on transformed variables and computing White’s [19] robust estimates for the standard errors. Differences model This model assumes that the low-income WOPI can be regarded as an appropriate control group, and that any price response of the other two groups can be measured as a deviation from the control group behaviour. We then focus on the differential utilization of experimental and control groups as a function of the price differential. A crucial homogeneity assumption is inescapable for this approach to work, namely that the price elasticity in the control group is equal to that in the experimental group, i.e. that jk=j3 j and k=1, 2. In other words, we have to assume that, after correction for any possible differences in time trends and regional effects, both groups are homogeneous with respect to their price-sensitivity. It is then as if both groups are drawn from the same population and only one of the groups is subjected to price variation. The main problem with this approach is that, in general, one can never be certain about the validity of the homogeneity assumption. In our case, however, we can test whether the price elasticities of the two experimental groups are equal. The consequences for the estimation procedure are as follows. Starting from Equation (1), and assuming that j1=j2=j3=j j, we take the difference between the equation of the experimental group and the control group, for k=1, 2 ln qitjk−ln qitj3=j(ln pitjk−ln pitj3) +j(k−3)Tt+ 31 i=1 ij(k−3)Ri+itj(k−3) (2) where j=j1=j3=j2 j(k−3)=jk−j3 ij(k−3)=ijk−ij3 itj(k−3)=itjk−itj3 As we will automatically obtain two separate estimates of j (for j=1, . . . , 3), one from the difference equation for the general population and another from the equation for high-income WOPI (call them j(k) for k=1, 2), we can test whether the homogeneity assumption is rejected by testing for j(1)=j(2) for j=1, . . . , 3. Testing strategy Estimation of Equation (1) provides j k parameter vectors [jk, jk, ijk (i=1, . . . , 31)] (one for each combination of k and j), whereas estimation of Equation (2) provides j (k−1) parameter vectors [j(k) , j(k−3), ij(k−3) (i=1, . . . , 31)]. Based on the Rand HIE results, we can formulate some hypotheses concerning the relative values of these parameters 1. Estimated price elasticities are negative for all groups and for all types of care. 2. In terms of between-group differences, we expect the general population to be more priceresponsive than the ‘needy’ groups (high- and low-income WOPI). One argument for this hypothesis is the better average health status of the general population. It is also possible to argue that the utilization by the WOPI is less price elastic, because they can be persuaded more easily by the physicians to take more treatment. Our data do not allow us to distinguish the demand and the supply effects, as we are only capable of estimating the overall price effect. For the low-income WOPI, price-sensitivity can hardly be measured owing to lack of price variation in the period considered. 3. In terms of between-visits differences, we expect (a) a higher price sensitivity for GP than Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)
C. VAN DE VOORDE ET AL for specialist care, given that specialist care Based on the robust standard errors(which are treats on average more severe conditions, and in general, larger than the standard WLs esti- (b)a higher price sensitivity for GP home mates), the estimation results indicate that hy visits than for GP office visits because the pothesis I(zero price elasticity) can be rejected for monetary price is a much smaller share of the all types of care for the general population and total price in the latter case the wopi. but not for the low-income woPi We could have estimated each of Equations (1) However, given the lack of price variation, the nd(2)separately, but in order to perform statis estimation results for this latter group are not ative, and we show them only for the tical tests of differences in the parameters across sake of completeness. For the general population Inforn equations, it is convenient to estimate (D)as a GP home visits can be seen to be much more system of nine and (2)as a system of tions. For our central exercise, we will assume price-sensitive (-0.39)than GP office visits Ix equa that e( EirikEurt'k)=0forj≠ i'and k≠k 0.16) or specialist visits (-0.10). Similarly, out-of-pocket price elasticities are negative and significant, but much lower in absolute value for he higher-income WoPI group, ranging from RESULTS -0.08 for gp home visits to -0.06 for gP office and specialist visits. The value of the F-test for the hypothesis j: a,=2=ay is 70.28, which Levels model implies that the hypothesis of equal price elastic ities for the active population and the WOPI is The WLSVD estimation results of Equation (1) very clearly rejected for the nine cases are presented in Table 1. We do not report the(279)estimates of the fixed regional effects, which are not very informative as such. Differences model Almost all of them are very significant. These fixed effects capture all time-invariant inter- The WLSDV estimates of Equation (2)are pre- regional variation. This includes the effects of sented in Table 2. Again, we do not provide the local demand and supply conditions, the age dis- results for the 31 regional dummies, but the esti- tribution of the membership of the local sickness mates of the coefficients yink-3) are all significant funds, the overall pattern of morbidity and mor- in each of the regressions. Coupled with the sig tality, the differences in income distribution, nificant estimates for two of the three time trend which can all be assumed to have been reasonably Bok-3y, this suggests that both the time trends and stable over the period 1986-1995 regional utilization differences are very differ The adjusted Durbin-Watson test statistic indi- ent for all subgroups. In general, the estimated cates the presence of positive autocorrelation in price elasticities are lower (in absolute value) than all nine equations, and the White test indicates in Table 1. The coefficients are all negative, heteroscedasticity in two of the equations for though no longer significantly different from zero specialist visits, despite the application of in the case of the high-income WOPI. If we could weighted least squares. The serial correlation may assume that the low-income WOPI's utilization also be indicative of some misspecification. Exper- behaviour provides a more accurate control for iments with alternative specifications(e.g. non-lin- unobserved variables than the time trends in lev- ear time trends) hardly changed the value of the els, then we have to conclude that failure to use djusted Durbin-Watson. Nor did the autocorre- this control group results in an overestimation of lation disappear when we introduced a separate the price sensitivity trend for each region. Dynamic specifications However, the results strongly suggest that the with lagged price variables were not feasible ow- differences in time trends and fixed effects are not ng to the limited number of post-reform observa- sufficient to correct for the heterogeneity of the tions. As autocorrelation and heteroscedasticity different groups. The estimated price elasticities affect the efficiency, but not the consistency of the are significantly different between the active pop- coefficient estimates, we decided to retain the ulation and the WOPI (F= 42.21). Therefore, the basic specification and to compute robust stan- low-income WOPI cannot be an adequate control dard errors using Arellano's method [17]. group for the two experimental groups at the Copyright a 2001 John Wiley Sons, Ltd Health Econ.10:457-471(2001)
466 C. VAN DE VOORDE ET AL. for specialist care, given that specialist care treats on average more severe conditions, and (b) a higher price sensitivity for GP home visits than for GP office visits because the monetary price is a much smaller share of the total price in the latter case. We could have estimated each of Equations (1) and (2) separately, but in order to perform statistical tests of differences in the parameters across equations, it is convenient to estimate (1) as a system of nine and (2) as a system of six equations. For our central exercise, we will assume that E(itjkitjk )=0 for jj and kk. RESULTS Leels model The WLSVD estimation results of Equation (1) for the nine cases are presented in Table 1. We do not report the (279) estimates of the fixed regional effects, which are not very informative as such. Almost all of them are very significant. These fixed effects capture all time-invariant interregional variation. This includes the effects of local demand and supply conditions, the age distribution of the membership of the local sickness funds, the overall pattern of morbidity and mortality, the differences in income distribution, which can all be assumed to have been reasonably stable over the period 1986–1995. The adjusted Durbin–Watson test statistic indicates the presence of positive autocorrelation in all nine equations, and the White test indicates heteroscedasticity in two of the equations for specialist visits, despite the application of weighted least squares. The serial correlation may also be indicative of some misspecification. Experiments with alternative specifications (e.g. non-linear time trends) hardly changed the value of the adjusted Durbin–Watson. Nor did the autocorrelation disappear when we introduced a separate trend for each region. Dynamic specifications with lagged price variables were not feasible owing to the limited number of post-reform observations. As autocorrelation and heteroscedasticity affect the efficiency, but not the consistency of the coefficient estimates, we decided to retain the basic specification and to compute robust standard errors using Arellano’s method [17]. Based on the robust standard errors (which are, in general, larger than the standard WLS estimates), the estimation results indicate that hypothesis 1 (zero price elasticity) can be rejected for all types of care for the general population and the WOPI, but not for the low-income WOPI. However, given the lack of price variation, the estimation results for this latter group are not very informative, and we show them only for the sake of completeness. For the general population, GP home visits can be seen to be much more price-sensitive (−0.39) than GP office visits (−0.16) or specialist visits (−0.10). Similarly, out-of-pocket price elasticities are negative and significant, but much lower in absolute value for the higher-income WOPI group, ranging from −0.08 for GP home visits to −0.06 for GP office and specialist visits. The value of the F-test for the hypothesis j: j1=j2=j3 is 70.28, which implies that the hypothesis of equal price elasticities for the active population and the WOPI is very clearly rejected. Differences model The WLSDV estimates of Equation (2) are presented in Table 2. Again, we do not provide the results for the 31 regional dummies, but the estimates of the coefficients ij(k−3) are all significant in each of the regressions. Coupled with the significant estimates for two of the three time trends j(k−3), this suggests that both the time trends and the regional utilization differences are very different for all subgroups. In general, the estimated price elasticities are lower (in absolute value) than in Table 1. The coefficients are all negative, though no longer significantly different from zero in the case of the high-income WOPI. If we could assume that the low-income WOPI’s utilization behaviour provides a more accurate control for unobserved variables than the time trends in levels, then we have to conclude that failure to use this control group results in an overestimation of the price sensitivity. However, the results strongly suggest that the differences in time trends and fixed effects are not sufficient to correct for the heterogeneity of the different groups. The estimated price elasticities are significantly different between the active population and the WOPI (F=42.21). Therefore, the low-income WOPI cannot be an adequate control group for the two experimental groups at the Copyright © 2001 John Wiley & Sons, Ltd. Health Econ. 10: 457–471 (2001)