WHAT DOES UNIVERSAL COVERAGE DO? THE IMPACT ON HEALTH CARE UTILIZATION AND EXPENDITURES IN THAILAND

Transcription

1 WHAT DOES UNIVERSAL COVERAGE DO? THE IMPACT ON HEALTH CARE UTILIZATION AND EXPENDITURES IN THAILAND SUPON LIMWATTANANON a,b, SVEN NEELSEN c, VIROJ TANGCHAROENSATHIEN a, PHUSIT PRAKONGSAI a, VUTHIPHAN VONGMONGKOL a, EDDY VAN DOORSLAER c,d,e, OWEN O DONNELL d,e,f a International Health Policy Program, Ministry of Public Health, Thailand b Khon Kaen University, Thailand c Institute of Health Policy and Management, Erasmus University Rotterdam, the Netherlands d Erasmus School of Economics, Erasmus University Rotterdam e Tinbergen Institute, the Netherlands f Department of Balkan, Slavic and Oriental Studies, University of Macedonia, Greece December 2012 Abstract In 2002, Thailand became the lowest income country yet to have achieved effective universal health insurance coverage. We estimate the impact of this major health reform on the utilization of public and private health care and on household medical expenditures. Identification is through comparison of changes in outcomes for groups to whom coverage was extended by the Universal Coverage Scheme (UCS) with those of public sector employees whose coverage did not change. The UCS is estimated to have reduced the probability of not being treated or being treated only by informal sector providers when sick by 3.2 percentage points from a baseline of 30%. The probability of receiving ambulatory care at a public facility was raised by 2.7 percentage points (baseline 55%) and the probability of an inpatient admission to a public hospital by 1 percentage point (baseline 5.6%). Impacts on utilization were strongest in the elderly population. UCS is estimated to have reduced the amount paid for an inpatient admission to a public hospital by 37% and to have reduced household total medical expenditure by up to 32%. JEL Classification: H42, H51, I18 Keywords: Health Financing, Health Care, Universal Coverage, Thailand Acknowledgements: The research reported in this paper was financed by the EU FP7 project entitled Health Equity and Financial Protection in Asia (HEALTH-F ). Comments received at ihea 2011 (Toronto), the Hanoi HEFPA workshop, the 2 nd Global Symposium on Health Systems Research (Beijing) and Erasmus University Rotterdam are gratefully acknowledged.

2 1. INTRODUCTION Financing of healthcare in low- and middle-income countries continues to rely heavily on direct, uninsured payments. As a result, when sick, a large part of their populations face the dire choice of either forgoing adequate care, or thrust (further) into poverty through sacrifices made to pay for it. Social health insurance coverage, if provided, tends to be limited to public sector and formal private sector employees, and perhaps their dependents. In principle, the extremely poor are often exempt from user fees at public facilities but the targeting and application of such exemptions is often weak. Extending health insurance coverage to the informal sector and nearpoor population is widely considered essential to both improving population health per se and propelling economic growth through increased labor productivity and reduced savings against health expenditure risks. Realizing this objective is, however, far from easy in the context of lowand middle-income, highly informal economies with limited capacity to collect contributions toward pre-paid health care. The Thai universal coverage reform is therefore of particular interest. In , the country introduced the fully tax-financed Universal Coverage Scheme (UCS). UCS entitled to access to public health facilities and prescribed medicines all population groups not covered by the existing schemes for public sector employees and their dependents and formal private sector employees. The only payment remaining for those covered by UCS was a flat fee of 30 Baht (equal to about 0.75 US$, removed 2006) for each patient contact, from which the poor, the elderly and children were exempted. The reform reduced the share of uninsured citizens from 29 to 5 percent within a year of its introduction. It is widely cited as the leading example of the feasibility of extending comprehensive health care coverage to vulnerable population groups within a tight budget constraint (Gottret et al., 2008). It is prominent in the 2010 World Health Report on the path to universal coverage, where it is emphasized that universal coverage has been achieved in a country that spends only 136 US$ per capita on health care (World Health Organization, 2010, p.3). Columbia, China and Mexico recently achieved almost universal health insurance coverage after introducing subsidized, premium-based schemes in 1994 (Columbia s Law 100; Tsai, 2010), 2003 (China s New Rural Cooperative Medical Scheme; Meng et al., 2012) and 2004 (Mexico s Seguro Popular; Knaul et al., 2012). Yet, around the time of reaching universal coverage (2009), all three countries spent higher shares of their GDP on healthcare (7.6, 5.1 and 6.5 percent) and more on healthcare per capita (392, 191 and 525 US$) than Thailand about 7 years after the UCS reform (4.2 percent, 160 US$) (World Health Organization, 2012). 1,2 Several studies have analyzed trends or relied on before-after comparisons to evaluate the UCS reform, observing increases in ambulatory care utilization (Damrongplasit and Melnick, 2009; Panpiemras et al., 2011) and reductions in health expenditure (Limwattananon et al., 2007; Somkotra and Lagrada, 2008; Damrongplasit and Melnick, 2009; Limwattananon et al., 2011; 1 Taiwan provides a recent example of universal coverage reform in a high-income country: with 43% of its population uninsured, and consolidating its existing social security schemes into a single-payer system, the country introduced National Health Insurance with mandatory membership in March By the end of the following year, the scheme had already achieved almost universal enrollment (Cheng and Chiang, 1997). 2 The Philippine case shows the challenge of achieving universal coverage by premium-based health insurance. The country established National Health Insurance (PhilHealth) in 1995, with the aim of achieving universal enrollment by Officially, coverage had reached 76% by 2008 but survey data show effective coverage of only 42% (Philippine Health Insurance Corporation, 2008; Department of Health, The Philippines 2010). 1

3 Evans et al., 2012). To date, there has been only one evaluation of the reform s causal impact on inpatient care utilization (Gruber et al., 2012), limitations of which we discuss in Section 3.1. In addition to presenting alternative estimates and interpretations of the causal effect of UCS on inpatient care utilization, we estimate causal effects on ambulatory care utilization and on individual and household medical expenditures for the first time. Our analysis of utilization uses data from Thailand s Health and Welfare Survey, a nationally representative repeated cross-section data set. We employ a difference-in-differences estimation for identification that compares a treatment group of informal sector workers and dependents with a control group of public sector workers and dependents that had coverage prior to the reform. Particular attention is paid to UCS effects on the pattern of health care utilization across types of public health facilities. This analysis tests for effects of two institutional reforms that accompanied the UCS coverage extension. It tests i) if the new UCS referral system that made district hospitals the primary entrance point to care was effective; and ii) if/to what degree the UCS policy s assignment of additional budgetary powers to district hospitals led to a redirection of ambulatory care delivery from health centers to district hospitals. Our empirical results indicate that UCS substantially increased health care utilization. In our full sample analysis, the reform reduced the probability that a sick individual accesses no or only informal ambulatory care by 3.2 percentage points, or 11% over the pre-reform baseline mean. This ambulatory care effect is present in the elderly, rural, urban, poor, and non-poor subsamples. UCS also increased the probability of an inpatient admission to a public facility by 1 percentage point (18%), on average. Subgroup analysis reveals statistically significant inpatient care effects in the working-age adult, elderly, urban, poor, and non-poor populations, but no effect for rural dwellers. Consistent with UCS operating almost exclusively through public providers, the increases in ambulatory and inpatient utilization came almost entirely through increases in the use of public facilities, causing no crowding out of private formal care. Furthermore, we find evidence for the new referral system with its gatekeeper role for district hospitals to be effective and for a redirection of care from health centers to district hospitals with their new budget powers: for those admitted to a public facility for ambulatory care, the full sample estimates show a 3.6 percentage point (9%) reduction in visits to provincial or university hospitals and a 6.9 percentage point, or 21% reduction in visits to health centers. Combining these estimates, we find a 10.5 percentage point (37%) increase in ambulatory visits to district hospitals. For inpatient admissions to public hospitals, our full sample estimates indicate a 5.1 percentage point (16%) increase in the probability of admission to a district as opposed to a provincial or university hospital. We analyze UCS impacts on medical expenditure with two datasets. Using the same data and identification strategy as in the utilization analysis, we estimate impacts on payments made for inpatient admissions to public hospitals. This analysis shows that UCS increased the probability of making any payment by 11.8 percentage points (35%), a reflection of the 30 Baht flat fee charged to a subset of UCS beneficiaries until At the same time, the reform reduced the amount paid by 37% in the full sample and by 62% in the working-age adult subsample which includes a particularly large share of previously uninsured individuals that gained coverage through UCS. We use data from the Thai Socioeconomic Surveys to estimate the reform s impact on households total out-of-pocket expenditure on health care. While UCS left the probability of a 2

4 household having any medical expenditure unchanged, it led to a significant increase in the probability of any expenditure for inpatient care a finding that we again ascribe to the UCS 30 Baht copay per patient contact. UCS also decreased the probability of any expenditure on medicines, in line with medicines becoming available free of charge under UCS. Among households with any medical expenditure, UCS reduced the amount spent by 31%, on average. This is substantial, considering that it refers to expenditure on all types of health care, public and private. It comprises an average reduction of 31% in spending on medicines and 33% reductions in payments for ambulatory and inpatient care, respectively. The rest of the paper is structured as follows. Section 2 describes the organization and financing of health care in Thailand before and after the introduction of the UCS. Section 3 presents our empirical strategy, data description and findings for the analysis of UCS impacts on healthcare utilization and Section 4 those for our analysis of public inpatient care expenditures in the admitted population. Section 5 comprises methods, data description, and results of the analysis of household healthcare expenditure effects. Section 6 concludes with a discussion of implications for health policy in Thailand and universal coverage further afield. 2. THE THAI HEALTHCARE SYSTEM 2.1 BEFORE UNIVERSAL COVERAGE Prior to the introduction of the UCS in 2001, the supply of health care had been increased substantially over four decades. From 1962, through a series of five-year National Economic and Social Development Plans, the national health system evolved. By the third of these plans, there was a provincial hospital in all 76 provinces and newly qualified doctors were mandated to serve in rural areas. Development of the next tier of district level hospitals began with the fourth plan in The next two decades witnessed continuous two-digit annual growth in the number of new district hospitals (mostly beds) and in the ten years prior to the UCS reform around 150 such hospitals were opened (Ministry of Public Health, various years). The number of doctors increased steadily from around 17 per 100,000 population in 1977 to around 30 in 2001 (ibid). Nursing staff increased even more dramatically. In 1977, there were less than 40 nurses per 100,000 population (ibid). Twenty-five years later, the number had more than quadrupled to around 170 (ibid). Various public health insurance schemes (Table 1) had succeeded in extending coverage to a substantial share of the population even before the introduction of the UCS. Ten years before the reform, two-thirds of Thai citizens had no formal health insurance. Just before the reform, this fraction had been reduced to less than 30% (Table 2). The single largest scheme in 2001 was the Medical Welfare Scheme (MWS) that entitled the elderly (60+), children (<12, and secondary school students), the poor, the disabled, and army veterans to comprehensive free care in public facilities upon presentation of a MWS membership card. 3 There were 19.6 million card holders in 2001 (31.5% of the population). The MWS was tax-financed and allocated regional budgets in relation to population size, income structure, mortality, as well as healthcare use and infrastructure. In 1998, the average annual budget per enrollee was just 273 Baht (~6.82 US$), raising concerns of severe underfunding (Donaldson et al., 1999; Pannarunothai, 2002). To fill the financing gap, provincial health authorities and 3 Monks and village leaders and their dependents also had MWS entitlement. 3

5 hospital staff redirected substantial funds from general budgets and other public schemes to pay for the care of MWS beneficiaries. Through such cross-subsidization, MWS expenditure per enrollee has been estimated at just under 470 Baht (11.75 US$) in 1998, around 70% greater than the official per capita budget (Donaldson et al., 1999). The second largest program prior to the UCS was the Voluntary Health Card Scheme (VHCS) with about 13 million enrollees (20.8% of the population) in Originally designed as a selffinancing mother and child care program, VHCS developed into a publically subsidized, comprehensive health insurance scheme in the 1990s. For 500 Baht (12.50 US$) per year, households could purchase a health card that entitled up to 5 household members to free care at public facilities. The 500 Baht private contribution per card was matched by a 1,000 Baht (25.00 US$) tax-financed government subsidy, resulting in a direct annual budget of 1,500 (37.50 US$) Baht per card. With over four enrollees per card on average, the VHCS s direct per enrollee budget was often insufficient to provide adequate care. Similar to the MWS, cross-subsidies thus accounted for a large share of VHCS per enrollee expenditure that amounted to about 530 Baht (13.25 US$) in 1997 (Donaldson et al., 1999). With 5.3 million enrollees (8.5% of the population) in 2001, the Civil Servant Medical Benefit Scheme (CSMBS) was the third largest program prior to the UCS. The wholly tax-financed CSMBS provides free care at public facilities for active and retired government employees, their spouses, parents, and up to three of their children below 21. Operating under an uncapped feefor-service system with a generous benefit package, CSMBS spending per capita was almost 2,500 Baht (62.50 US$) per enrollee in 1998 (Donaldson et al., 1999). Finally, the Social Security Scheme (SSS) covered 4.5 million employees of private formal sector enterprises (7.2% of the population) for care at registered private and public facilities in In April 2002, just around the time of the completion of the UCS-reform, the SSS was extended from covering only employees of enterprises with ten or more employees to covering all private formal sector employees irrespective of enterprise size. The scheme continued to not cover dependents. SSS-financing is tripartite with employers, employees and government each contributing 1.5% of gross monthly salaries (Donaldson et al., 1999). The 18 million Thais that remained without health insurance coverage in 2001 could access healthcare in two ways. First, they could purchase care out-of-pocket at private or public facilities. Second, the so-called Type B Exemption enabled public health facility staff to exercise discretion in granting free or subsidized care to apparently poor individuals with no proof of MWS membership. Type B Exemption expenditure amounted to about 2% of total public health expenditure in 1997 (Donaldson et al., 1999). Notwithstanding considerable achievements in the provision of health care and the extension of population coverage, there were structural weaknesses in the Thai healthcare system at the turn of the millennium. Lack of insurance and the underfunding of the schemes targeted on the poor (MWS) and near poor (VHCS) populations left many Thais exposed to high out-of-pocket expenditure on medical care with potentially catastrophic consequences for living standards and contributed to inequities in access to adequate care (Tangcharoensathien et al., 2000; Makinen et al., 2000; Pannarunothai, 2002; Srithamrongsawat, 2002; Siamwala, 2002; Somkotra and Lagrada, 2008). 4 Targeting of the MWS on the poor, elderly and children was far from perfect 5 and the 4 However, compared to other countries in Asia at both higher and lower levels of GDP, Thailand was performing relatively well in the provision of financial protection against medical expenditure risks and the 4

6 voluntary nature of the VHCS left it vulnerable to adverse selection (Donaldson et al., 1999). The system was also characterized by large and persistent regional disparities in both resource allocation and health outcomes. In 1999, the population to doctor ratio was almost two-fifths higher and the population to nurse ratio almost twice as high in the poor Northeast compared with the more urbanized and affluent Central region (Wibulpolprasert, 2002). While nationwide infant mortality dropped remarkably in the decades preceding the UCS reform, the rural-urban gap widened. In 1995/6, it was 85% higher in non-municipal than in municipal areas (Wibulpolprasert, 2000). Tables 1 & AFTER UNIVERSAL COVERAGE Shortly after coming into power in 2001, the populist Thai Rak Thai party embarked on an ambitious healthcare system overhaul. Starting in April 2001 with the introduction of the UCS in 6 pilot provinces, the scheme was expanded to 15 provinces in June 2001 and to the remaining 75 provinces by October of that year, leaving only the capital city uncovered. Rollout was completed in April 2002 with the inclusion of Bangkok. On the demand-side, the new UCS effectively replaced the MWS and VHCS with a gold card issued to all Thai citizens not covered by the CSMBS or SSS. The card, obtained by registering in a local provider network, gives entitlement to comprehensive inpatient and ambulatory care in public facilities, and to medicines prescribed at those facilities. 6 Fully tax-financed, the UCS initially levied a fixed charge of 30 Baht (~$0.70) per service contact on a little more than half of its enrollees. Those meeting the criteria of the former MWS were de jure exempt from this charge. In 2006, all user fees were abolished. As a result of the UCS, the percentage of the population covered by some form of health insurance jumped from 71% in 2001 to 95% in 2003, and by 2011 coverage had risen to over 98% (Table 2). On top of the coverage extension, the reform made substantial changes to the financing and architecture of the country s public healthcare system. These supply-side reforms aimed at improving the quality and efficiency of care delivery and at achieving greater regional and crossscheme equity in resource allocation. The backbone of the reformed system was the newly established Contract Unit for Primary Care (CUP). CUPs typically consist of a district hospital and a network of health centers managed by a board comprised of medical and administrative personnel that is normally chaired by the district hospital director. The CUP is the issuer of the UCS gold card that gives entitlement to care within the CUP network. Only upon referral from the CUP district hospital can UCS-patients access higher level providers, including provincial hospitals (Hughes and Leethongdee, 2007; Hughes et al., 2010). achievement of a pro-poor distribution of public health expenditure even before the UCS reform (van Doorslaer et al., 2007; O Donnell et al., 2008). 5 A survey conducted by the Ministry of Public Health in 2000 revealed that only 17% of households meeting the means test criteria were actually enrolled in the MWS (Kongsawat et al., 2000), which starkly contradicts an earlier (1998) Ministry of Public Health estimate of 90% enrollment of the poor in the scheme (Donaldson et al., 1999). 6 The benefit package includes high-cost treatments like open-heart surgery and chemotherapy. Antiretroviral treatment is part of the package since 2003 and renal replacement therapy since 2008 whereas organ transplantation remains one of the few essential treatments excluded to date. 5

7 Prior to the UCS, the Ministry of Public Health disbursed public healthcare funds to Provincial and District Health Offices through block grants based on retrospective budgeting. Money was subsequently distributed to provincial and district hospitals and health centers. Salaries that typically account for about half of Thai public health expenditure were paid directly from the Ministry. This financing mechanism was replaced by one that almost exclusively relied on capitation and that gave the district level, namely CUPs, substantial discretionary power over the local distribution of funds (Hughes and Leethongdee, 2007; Hughes et al., 2010). CUPs were given a capitation budget of 1,202 Baht (30.50 US$) (2001) per UCS-enrollee to cover ambulatory, inpatient and preventive care. With the aim of correcting geographic inequalities through the creation of financial incentives for the redeployment of health personnel to populous but understaffed regions, the UCS capitation formula initially included salary budgets. This led to large funding increases for understaffed, typically rural CUPs. At the same time, it put severe financial strain on many urban providers whose entire capitation payment could easily be consumed by salary costs at existing staffing levels. Unwilling or unable to cut staff, a majority of urban facilities successfully requested emergency funding from the government in the year of the reform. An intense lobbying effort by the medical profession eventually led to a repeal of the new financing arrangement in 2002/3 and a return to salary budgets being determined by current staffing levels. This rather undermined the original aim of achieving a more equal regional distribution of the health workforce (Hughes and Leethongdee, 2007; Hughes et al., 2010). Besides changing the financial architecture of the Thai healthcare system, the reform was accompanied by an increase in public health spending. The UCS began with a capitation of 1,202 Baht in that increased to 1,396 Baht (34.90 US$) by Total spending on the UCS in 2003 was 35% greater in real terms than total public expenditure on the MWS and VHCS schemes had been in 2001 (Table 3 8 ). The increase in expenditure per insured person was much more modest since around 23% more of the population was covered by the UCS. The magnitude of the increase in expenditure per person effectively covered depends on the proportion of the resources that were used to cross-subsidize treatment of the uninsured in the pre and post reform periods. One extreme scenario is obtained by ignoring Type B exemption and assuming no expenditure on the uninsured. In that case, UCS expenditure per person covered in 2003 was only 8.5% higher in real terms than expenditure per MWS/VHCS enrollee had been in 2001 (Table 3, row 1). The opposite extreme scenario is that in which all the nominally uninsured are assumed to have been effectively covered and their public treatment paid for from the MWS/VHCS budgets before the reform through Type B exemption and from the UCS budget after the reform. In that case, per capita spending increased by 57% (Table 3, row 2). While Type B exemption was not uncommon, it is certainly unrealistic to assume that all the uninsured were covered by this. The increase in expenditure per person effectively covered is likely to have been substantially lower than this upper bound estimate. Table 3 7 This % nominal increase in the capitation corresponds to a 10% real increase (Health Consumer Price Index adjusted, International Health Policy Program, 2011). 8 The per capita spending estimates for the UCS and CSMBS target populations in Table 3 were calculated using macro data from the Thai National Health Accounts (International Health Policy Program, 2011). They differ from the per capita spending estimates we present in column 6 of Table 1 in that they include additional administrative costs and overheads. 6

8 3. UCS IMPACT ON HEALTHCARE UTILIZATION 3.1. TREATMENT AND CONTROL GROUPS Our strategy for identifying the impact of the UCS on health care utilization is to compare before and after differences between population groups whose coverage was changed as a result of the reform and beneficiaries of the CSMBS whose coverage did not change. We use data from a nationally representative cross-sectional household survey the Health and Welfare Survey (HWS) conducted by the National Statistical Office. Treatment Group The HWS records the health insurance coverage of each household member. Our treatment group consists of all individuals not covered by one of the two formal sector employment-related schemes (CSMBS and SSS) and so targeted for coverage through the UCS. In the pre-ucs period, it comprises those covered by the MWS, the VHCS, and the uninsured. In the post-ucs period, it includes those covered by the UCS and those that remain uninsured because they are not registered with a CUP. Since the latter are part of the target population, they belong to the treatment group under the intention-to-treat principle. The impact of the UCS potentially varies with the nature and extent of pre-reform coverage. Enrollees of the MWS and VHCS programs were already entitled to free care at public facilities and their effective coverage changed only as a result of increases in the per capita subsidies. In contrast, for the previously uninsured, the reform removed the price barrier to utilization 9. Because differential pre-reform coverage is related to demographics and income levels, we present separate estimates of the impact of the UCS for children (<16), working-age adults (16-59), and the elderly (60+), and for the poor and non-poor 10. Since geographic reallocation of resources was an original aim of the reform, we also distinguish between rural and urban dwellers. Gruber et al. (2012) use the same dataset and essentially the same identification strategy to estimate the impact of the UCS on inpatient admissions, focusing on differential effects between those exempt (UCSE) and not exempt (UCSP) from the 30 Baht copay. They include pre-reform MWS enrollees in the UCSE treatment group and pre-reform VHCS enrollees and the uninsured in the UCSP treatment group. They estimate that the reform increased the probability of inpatient admission by 12% among the former group and by 8% among the latter. The greater relative impact on the exempt group is estimated to be even larger for women of childbearing age and infants. Related to this, they estimate a staggering 30% reduction in infant mortality (from a baseline of 21.5 per 1,000 births) as a result of the UCSE program. On the basis of these findings, Gruber et al. argue that the most important component of the Thai reform was not so much the extension of coverage to the previously uninsured but what they consider to be a massive increase in funding for those previously enrolled in the MWS that made their previously nominal coverage effective. This interpretation has potentially important implications for health policy beyond understanding what happened in Thailand. It suggests that health care access and health outcomes can be 9 More precisely, UCS removed financial access barriers for the previously uninsured that did not obtain their entire healthcare free of charge by the Type B exemption. 10 Broadly consistent with the official poverty rate during our observation period, we define the poor as those in the bottom tertile of per capita household income distribution. 7

9 dramatically improved by putting more resources into health systems, offering subsidized care to poor and vulnerable populations without changing entitlements and extending population coverage. While this is an interesting hypothesis placing emphasis on effective, as opposed to nominal, coverage that is somewhat at odds with the drive for universal coverage being led by the WHO, we believe that the evidence in support of it provided by the Gruber et al. analysis is weak on three counts. First, survey participants identified as MWS enrollees in the pre-reform period do not necessarily correspond to those reporting UCSE coverage in the post-reform period. This could arise from differences in the target efficiency of the MWS and UCSE programs. Similarly, those identified as uninsured or VHCS enrollees pre-reform may differ compositionally from those identified as UCSP post-reform. The 2003 HWS data show, for example, that 15% of the formerly uninsured became UCSE beneficiaries. Differences in the composition of the specific treatment groups over time potentially invalidate the difference-in-differences identification strategy. Second, there appears to be much misreporting of UCS exemption status in the HWS. While all children (and elderly) de jure became UCSE beneficiaries through the reform, 92% of the children of UCSP parents are also (falsely) reported as UCSP. Gruber et al. take reported exemption status at face value and include both elderly and children in the UCSP group. 11 Third, the claim of a dramatic fourfold increase in funding for those previously enrolled in the MWS program (Gruber et al, p.1) (from 250 Baht to 1,202 Baht) ignores both the substantial pre-ucs cross-subsidization of the MWS and the initial requirement that salary costs be paid for from the UCS capitation payment. Raising the MWS capitation to take account of cross-subsidization, perhaps by as much as 70% as discussed in the previous section, and reducing the USC capitation by removing salary costs (typically >40% of total costs in Thailand, see International Health Policy Program, 2011) would give a much lower estimate of the increase in expenditure per enrollee whose coverage switched from MWS to UCS. In fact, the increase in per capita public health resources was probably largest for the previously uninsured accessing care under the Type B exemption or forgoing public care altogether a group that forms part of Gruber et al. s UCSP treatment group. Control group Individuals belonging to the CSMBS, either as a government employee, a retiree from public sector employment or a dependent of one of those two populations, form our control group. Their insurance status did not change with the introduction of the UCS. Between 2001 and 2005 there were no changes to the CSMBS and the proportion of the population covered by it varied by just over one percentage point. The expansion of the SSS around the time of the UCS reform to cover all formal private sector employees and the resulting increase in the proportion of the population covered by it implies a change in the composition of beneficiaries. For this reason, we do not include individuals covered by SSS in the control group. 12 Those reporting coverage by private insurance, employer benefits and other insurance types (2.1% in 2001) are also dropped from the estimation sample. 11 We have estimated effects specific to UCSE and UCSP for a working-age adult subsample for which reported exemption status is arguably more accurate. These suggest a 1.1 percentage point, or 16% increase in inpatient utilization in the UCSE treatment group and a 0.8 percentage point, or 13% increase in the UCSP treatment group. So, while we also find a larger impact on the copay exempt group that had coverage prior to the UCS, the relative difference is not quite as large as that estimated by Gruber et al. (for a sample including all ages). 12 Gruber et al. (2012) do include those covered by SSS in the control group. Replication of their analysis indicates that their estimates are robust to excluding SSS from the control group. 8

10 3.2. DATA Our analysis uses one cross-section of the HWS before the introduction of the UCS (2001) and three cross-sections afterwards (2003, 2004 and 2005). The estimation sample size in 2001 (203,106) is about three times larger than that in each of the post-ucs waves, giving us an approximately equal number of observations pre- and post-reform (Table 4). The 2001 wave was conducted between April and June, before the nationwide implementation of UCS in October (except Bangkok where the UCS was implemented in April 2002). 13 Implementation began in April 2001 in six pilot provinces. With health care utilization and payments recorded for the previous two weeks (ambulatory) and one year (inpatient), recall periods for some pilot province observations span both the pre- and immediate post-reform time. UCS was, however, not effectively implemented in the pilot provinces at the time of the survey. We thus treat all 2001 data as pre-reform. 14 Analysis of health care utilization is at the individual level. The HWS records whether each household member had used each of various types of ambulatory care and whether s/he has had an inpatient admission in the last year. The information on ambulatory care is routed through a question about whether an illness that is considered not to have required hospitalization has been experienced. The recall period for this question is two weeks in 2001 and one month in We rely on year effects to take account of this inconsistency. Since this change coincides with the reform, identification of the effect of the latter relies on the assumption that the change in recall period affects the responses of treatment and control observations equally. There are no obvious reasons to doubt this. The number of different providers that a respondent reporting an illness could identify in the survey increased from two to three in Again, we must rely on year effects and the assumption that any effect of this change on reporting is the same for the treatment and control groups. For the subsample reporting illness, we test for UCS effects on three categories of ambulatory care. The first category indicates if an individual has forgone care, self-medicated (retail pharmacies, drug vendors, herbal medicines) or visited a traditional healer but has received no formal treatment from a public facility or private clinic. We refer to this category as no formal ambulatory care hereafter. The second category indicates if an individual received care from formal private providers but did not visit a public facility and the third if an individual received any care from a public provider. Because the UCS reform reduced the price and/or increased the funding of care at public facilities for the treatment group, we expect it to result in a switch from the first and second categories to the third. Separately for the treatment and control groups, Table 4 shows pre- and post-reform sample proportions reporting sickness and, conditional on reporting sickness, the proportions in each of the three categories of ambulatory care. The proportion reporting sickness rises by about 30% for both the treatment and control groups, presumably attributable to the aforementioned longer recall period in the post UCS surveys. The proportion with no formal ambulatory care is higher 13 In all other waves used, sampling was conducted in April only. We deal with this in the estimation by entering two indicators for observations sampled in May and in June. 14 Observations from the pilot provinces make up 7.2% of the 2001 sample of which only 11.5% identified themselves as UCS enrollees. Redefining these observations as being within the treatment period for the estimation of effects on ambulatory utilization, for which the recall period was two weeks, had no impact on the estimates. Our results are also robust to exclusion of observations from the pilot provinces. 9

11 in the treatment group in both periods but drops significantly, by almost three percentage points, only for this group after the reform. For both groups, utilization of private care rises and utilization of public care falls. The decrease in the probability to use public care is substantially larger for the control group (4.8 percentage points) than it is for the treatment group (1.6) and almost entirely eliminates the pre-reform difference between the two groups. For the subsample visiting a public facility, we investigate whether the pattern of utilization across levels of care changes differently between treatment and control group. We distinguish between visits to i) a health center but no other public facility, ii) a district hospital but not to a provincial or university hospital, and iii) a provincial or university hospital. Because the reform increased the budgetary powers of district hospital directors and made district hospitals the point of entry to higher levels of care, we expect both upstream (from health centers) and downstream (from provincial or university hospitals) shifts of ambulatory care towards district hospitals in the post-reform period. Patterns of utilization differ strikingly by insurance coverage (Table 4, bottom panel). Individuals in the treatment group are much more likely to use health centers and district hospitals and are much less likely use provincial or university hospitals than those covered by the CSMBS. Postreform, for both groups, the proportions using health centers and district hospitals increase and the proportion using provincial or university hospitals decreases. The increase in the probability to use district hospitals and the decrease in the probability to use provincial or university hospitals are both greater in magnitude for the treatment group, which is consistent with the UCS shifting the pattern of care toward the lower level hospitals. In the 2001 and 2003 surveys, respondents were asked whether they had any inpatient admission in the last year and, if so, to which type of hospital. From 2004, the respondent could report admissions to up to three types of hospitals. 15 For the cross-sections, we use all admissions reported and rely on the year dummy variables to absorb the effect of this change in the question. We distinguish between no admission, admission to a private but not a public hospital, and any inpatient stay in a public hospital. If the UCS reform succeeded in making inpatient care affordable to individuals who would otherwise have forgone treatment, then admissions to public facilities will have risen. In addition, if individuals who could otherwise afford inpatient treatment in private hospitals are induced by the lower relative price and better funding to switch to public hospitals, then there will be a crowding-out effect on private care. The descriptive statistics in Table 5, Panel 1 show a significant increase in the probability of an admission to a public hospital in the treatment group, but not the control group, after the UCS reform. There is a significant increase in the probability of admission to a private hospital also only in the treatment group, which is not consistent with a crowding-out effect. Furthermore, Table 5, Panel 2 shows that a much higher proportion of individuals in the treatment group use district hospitals rather than provincial or university hospitals and that, consistent with district hospitals increase in financial power and referral control over UCS patients, there is an increase in the probability to be admitted to a district hospital after the reform that is greater for the treatment group than for the control group ESTIMATION Tables 4 & 5 15 In the data, 18% of those with at least one admission report additional hospital stays. 10

12 For the sample reporting a sickness not considered to have required hospitalization, we estimate a multinomial logit (MNL) model of the three mutually exclusive categories of ambulatory care: no formal ambulatory care ( ), care at a private clinic or hospital but not at a public facility ( ) and care at a public facility ( ), where i and t are indicators of individuals and time respectively. Probabilities of these three outcomes are defined as exp Xitθ j 1. 3 exp X θ itj j Xitθ P y k 1 it k (1) We set the no formal care option as the reference ( θ 1 0 ) and define the linear index as follows, Xitθ j jdit jdit UCt Z it γ j tj, j 2,3 (2) where D it is a dummy variable indicating membership of the target treatment group (MWS/VHCS /uninsured/ucs), UCt is an indicator of the post-treatment period (>2001), tj is a year effect 16 and Zit is a vector of control variables included to increase precision and to allow for differences between the treatment and control groups in trends in observable determinants. 17 The effect of the UCS on the probability of each of the three possible outcomes for the target treatment group at the time of treatment is given by itj j α β it δt P j α it δt P P y 1 Z γ Z γ, j 1, 2,3 (3) where t Pis some point in the post-treatment period. We compute these effects by averaging over observations in the UCS target group (UCS/uninsured) within the post-ucs period ( ). We estimate analogous models for the type of public facility used by those with any ambulatory visit to a public facility (health center only (reference) vs. district hospital but no provincial or university hospital vs. provincial hospital) and for inpatient care (no admission (reference) vs. admission to a private but not a public hospital vs. admission to a public hospital). For those admitted to a public hospital, we estimate a binary logit model for admission to a district hospital as opposed to a provincial or university hospital. Adoption of the multinomial logit model implies the potentially restrictive Independence of Irrelevant Alternatives (IIA) assumption (e.g. McFadden, 1984). For example, the relative probability of seeking ambulatory care at a private clinic as opposed to forgoing or seeking only informal care or self-medicating may not be independent of the availability and quality of the third treatment option from public facilities. While the mixed logit (e.g. McFadden and Train, 2000) would be an attractive more flexible alternative, identification of this model is likely to 16 Month effects are also included to take account for the sampling undertaken in May and June (in addition to April) in 2001 only. 17 For the full sample, and separately by treatment status, Appendix Table A1 reports means and standard deviations of the control variables. The observed employment and occupation differences between treatment and control groups derive from fact that the UCS covers those outside the formal employment sector. The control group contains non-public sector workers because CSMBS coverage extends to public sector retirees and dependents of public sector workers. The 2.3% of individuals that report to be public employees not covered by the CSMBS could either be temporary public workers or the result of misreporting. Besides the occupation differences, individuals in the treatment group are on average younger, less likely to live in urban areas and less educated than individuals in the control group. 11

13 prove difficult in the absence of covariates that are specific to the care options. The nested logit (e.g. McFadden, 1984) model also succeeds in relaxing the IIA assumption but requires imposition of a hierarchy that may not be consistent with actual choices. For example, one could assume that there is a first stage decision to seek any formal sector ambulatory care followed by a second stage decision to seek care from either a public or private sector facility. But for some, perhaps poor and rural dwellers, the only feasible choice may be between no treatment and care at an inexpensive local public health center. We prefer to avoid the risk of misspecification arising from the imposition of incorrect hierarchies RESULTS Ambulatory Care Table 6 shows the estimated effects of the UCS on the probability of reporting an illness not requiring hospitalization (Panel 1), the probabilities of the three possible responses to such an illness (Panel 2), and, for the subsample utilizing care at a public facility, the probabilities of visits to each of the three public provider categories (Panel 3). In terms of the reform s impact on the probability of reporting an illness, we find a significant effect only in the poor subsample in which it amounts to a 2.1 percentage point increase, equal to an 11% increase over baseline. For the full and the other sub-samples, the absence of significant impacts on illness indicates that there is no selection bias induced by estimating the effect of the UCS on ambulatory care utilization among those reporting illness, supporting a causal interpretation of the coefficients we report in the following. We estimate that the introduction of the UCS reduced the probability of forgoing formal ambulatory care when ill by 3.2 percentage points (11%) on average, an effect primarily driven by a 2.7 percentage point (5%) increase in the probability of seeking care at a public facility. There is no evidence of the UCS resulting in the crowding out of private by public ambulatory care in the full or any sub-samples. Age-group specific analysis shows that the shift from no formal ambulatory care to treatment at public facilities is confined to the elderly, for whom the UCS increased the probability of receiving public care by 8 percentage points (13%). The elderly were in principle exempt from user fees both before (through the MWS) and after the introduction of the UCS (through the 30 Baht copay exemption applying for them). Had MWS provided effective coverage to its elderly target population, a particularly large impact among the elderly would be surprising. In fact, however, MWS targeting of the elderly was very poor with about one-third of this group not enrolled in 1998 according to Ministry of Public Health estimates (Donaldson et al., 1999), suggesting that newly won insurance coverage under the UCS may drive the elderly effect. Yet, the same Ministry of Public Health estimates suggest that in 1998 less than 60% of the MWS child and secondary school students target group was enrolled in practice. Moreover, the UCS provided coverage to many previously uninsured working-age adults. Thus, as the UCS raised insurance protection for all age groups, newly gained coverage alone cannot explain why the impact on ambulatory care is restricted to the elderly. 18 We have estimated the impact of the UCS on all utilization outcomes using the linear probability model (LPM) of binary choices that tends to be favored in the treatment effects literature. Qualitatively, and for the most part quantitatively, LPM estimates are consistent with those presented from the MNL. 12

14 Age-specific care needs provide an alternative explanation. Because the elderly typically face more and more complex and costly medical conditions, the additional financial protection provided by the UCS may have induced a stronger utilization response in this group. This may be reinforced by heterogeneity in the magnitude of the public funding increase. If the UCS predecessor schemes were especially ineffective in meeting the specific care requirements of their elderly enrollees, and if the funding increase under the UCS led to particularly large improvements in the delivery of such care, this would contribute to a stronger demand response for the elderly. Conversely, the absence of UCS impacts on working-age adults and children indicates that the previous schemes (MWS/VHCS/Type B Exemption), on average, sufficed to meet the ambulatory care requirements of these typically more healthy populations. A third possible explanation is age differences in attitudes towards welfare programs and traditional medicines. Despite its gradual extension to the entire child and elderly populations in the 1980s and 90s, the MWS continued to carry the stigma of the poor-man s program that it started out as. It seems plausible that the replacement of stigmatized MWS cards with the UCS s gold card made a greater impression on the elderly holding less positive attitudes towards public support programs than it did on their younger compatriots. Moreover, the introduction of the UCS was accompanied by a large-scale effort to champion utilization of formal medical care. It is possible, that this had a particularly large effect on the elderly with their hitherto stronger trust in traditional/informal healing methods than among the younger generations that were already more inclined towards western medicine. Testing for heterogeneous UCS impacts across living areas, we find a 4.6 percentage point (17%) reduction in the probability of not seeking formal ambulatory care in the rural sample that is driven by almost equally sized non-significant increases in the probabilities of seeking care at private and public providers. In the urban sample, the 2.5 percentage point (8%) reduction in the probability to go without formal ambulatory care exclusively derives from an increase in the utilization of public facilities. Finally, in terms of heterogeneity across socioeconomic strata, the UCS is estimated to reduce the probability of no formal care among the poor (6.7 percentage points, or 25%) entirely by raising utilization of public facilities. To a much lesser extent, the UCS reduced the probability that the non-poor go without any formal care (2.3 percentage points, or 7%) and only half of this can be attributed to a non-significant increase in the probability to resort to public providers. Again, a large part of the UCS impact appears to come through the closing of a coverage gap left by an ineffective MWS: the poor were in principle eligible to free care at public facilities before the reform, but poor MWS targeting and, possibly, inadequate care quality left many poor un- or underinsured. Among those making any visit to a public facility, we find statistically significant increases in the probability to be treated at district hospitals in the full sample and across all subsamples except children. The effect is 11 percentage points, or 38%, in the full sample and ranges between 7.5 percentage points (24%) for the elderly and 15.9 percentage points (54%) for working-age adults. With the exception of the elderly and urban subsamples, the increase in treatment at district hospitals mainly derives from reduced visits to health centers. In all but the rural and poor subsamples, which overlap considerably, we find statistically significant evidence that the increase in utilization of district hospitals also partly stems from reduced use of provincial or university hospitals. This is consistent with what one would expect from the UCS referral system that made district hospitals the point of entry to higher level care. The decrease in relative reliance on public care from health centers could reflect both the smaller relative price reduction induced by the 13