Effects of universal health insurance on health care utilization, and supply-side responses: Evidence from Japan

Journal of Public Economics 99 (2013) 1–23

Contents lists available at SciVerse ScienceDirect

Journal of Public Economics journal homepage: www.elsevier.com/locate/jpube

Effects of universal health insurance on health care utilization, and supply-side responses: Evidence from Japan☆ Ayako Kondo a,⁎, Hitoshi Shigeoka b, 1 a b

Faculty of Economics, Hosei University, 4342 Aiharacho, Machida, Tokyo, 194-0298, Japan Department of Economics, Simon Fraser University, 8888 University Drive, WMC 4653, Burnaby, BC, Canada V5A 1S6

a r t i c l e

i n f o

Article history: Received 23 February 2012 Received in revised form 13 November 2012 Accepted 18 December 2012 Available online 7 January 2013 Keywords: Universal health insurance Health care utilization Supply-side response Japan

a b s t r a c t We investigate the effects of a massive expansion in health insurance coverage on health care utilization and supply-side responses, by focusing on universal health insurance introduced in Japan in 1961. There are two major findings. First, health care utilization (measured in terms of admissions, inpatient days, and outpatient visits to hospitals) increased significantly. Second, we also find a supply response but the size of the supply response differs across service types: while the number of beds increases, effects on the number of medical institutions, physicians, and nurses are either negligible or inconclusive. Our results suggest that countries planning a large expansion in health insurance coverage would need to generate sufficient financial resources to cover the surge in health care expenditures, both in the short and long run. Our results also indicate that any slowdown in the supply-side response may constrain the ability of the health care system to meet increased demand. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Most developed countries have implemented some form of universal public health insurance system to ensure that their entire population has access to health care. Even the United States, a rare exception, is moving towards near-universal coverage through health care reform.2 Despite the prevalence of universal health care, most studies on the impact of health insurance coverage on utilization and health have been

☆ We are especially grateful to editor and two anonymous referees who significantly improved out paper. We also thank Douglas Almond, Kenneth Chay, Janet Currie, Amy Finkelstein, Michael Grossman, Chie Hanaoka, Hideki Hashimoto, Kazuo Hayakawa, Mariesa Herrmann, Takahiro Ito, Wojciech Kopczuk, Amanda Kowalski, Ilyana Kuziemko, Robin McKnight, Sayaka Nakamura, Yasuhide Nakamura, Haruko Noguchi, Seiritsu Ogura, Fumio Ohtake, Masaru Sasaki, Miguel Urquiola, Yoichi Sugita, Till von Wachter, Reed Walker, Hiroshi Yamabana, and the participants of Asian Conference of Applied Microeconomics, Kansai Labor Workshop, Applied Econometrics Workshop, NBER Japan project meeting, 22nd annual East Asian Seminar on Economics, and the seminars at Nagoya City University, University of Tokyo, Hosei University, Yokohama National University, Columbia University, IPSS, and Keio University for their helpful comments. Tomofumi Nakayama and Davaadorj Belgutei provided excellent research assistance. All errors are our own. ⁎ Corresponding author. Tel.: +81 42 783 2577; fax: +81 42 783 2611. E-mail addresses: [email protected] (A. Kondo), [email protected] (H. Shigeoka). 1 Tel.: +1 778 782 5348. 2 The Patient Protection and Affordable Care Act, passed in March 2010, imposes a mandate on individuals to either obtain coverage or pay a penalty. 0047-2727/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jpubeco.2012.12.004

limited to specific sub-populations, such as infants and children, the elderly, and the poor. 3 This paper studies the impact of a large expansion in health insurance coverage on utilization and health by examining the case of Japan, which achieved universal coverage for its entire population in 1961. We identify the effects of health insurance by exploiting regional variations in health insurance coverage prior to the full enforcement of universal coverage. In 1956, roughly one-third of the population was not covered by any form of health insurance, and the proportion of the uninsured population ranged from about zero to almost half across all prefectures. Our empirical strategy identifies changes in outcome variables in a prefecture in which the enforcement of universal coverage had a large impact, relative to a prefecture in which the impact was smaller. This study has several advantages over those in existing literature. First, since universal health insurance was achieved as early as 1961 in Japan, we can examine the impact of health insurance expansion in the long term. Since the effects of such a large policy change may emerge with lags, it is important to examine the long-term impact, in order to capture the overall implication of a large policy change. 3 Examples of studies that examine specific populations include Currie and Gruber (1996a,b), Hanratty (1996), and Chou et al. (2011), on infants and children; Finkelstein (2007), Card et al. (2008, 2009), and Chay et al. (2010), on the elderly; and Finkelstein et al. (2011), on the poor. An important exception is Kolstad and Kowalski (2012), who examine the impact of the introduction of universal health insurance in Massachusetts in 2006; however, they are unable to explore long-term effects as their data cover only the 3 years following the policy change.

2

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Second, our study of large coverage expansion enable us to examine the case of general equilibrium effects, rather than solely partial equilibrium analysis such as the RAND Health Insurance Experiment (hereafter, RAND HIE). Finally, we provide a more detailed analysis of supply-side responses to large demand shocks by investigating several outcomes that have not been explored extensively in previous studies, such as the number of physicians. 4 We obtain two key findings. First, we find that the expansion of health insurance coverage results in large increases in health care utilization, measured in terms of admissions, inpatient days, and outpatient visits to hospitals. For example, our estimates suggest that the introduction of universal health insurance increased inpatient days by 7.3% and outpatient visits by 12.6%, from 1956 to 1961. The long-term impact is even larger: the estimated increase in inpatient days and outpatient visits from 1956 to 1966 was 11.6% and 25.1%, respectively. Second, we find that supply-side responses to demand shocks differ across the types of services supplied. While the expansion of health insurance coverage did not increase the number of clinics and nurses even in the long term, the number of beds increased in response to the expansion in health insurance coverage. Our results on the number of hospitals and physicians are mixed and sensitive to the way in which we control for regional time trends. We observe a robust positive effect only on the number of beds; probably because it is less costly for existing hospitals to add beds than for new hospitals and clinics to pay large fixed costs to enter the market. The response of the number of physicians and nurses is weaker possibly because their supply is limited by the capacity of medical and nursing schools. Furthermore, we find that even the number of beds increased at a slower rate than increases in health care utilization. This paper is related to several strands of literature. The first relevant body of literature comprises studies on the effects of health insurance on utilization and expenditure. The pioneering works of the RAND HIE typically find modest effects of individual-level changes in health insurance on health care utilization and expenditure (Manning et al., 1987; Newhouse and the Insurance Experiment Group, 1993). In contrast, Finkelstein (2007) examines the impact of the introduction of Medicare in 1965, and finds a much larger effect on aggregate spending than that predicted by the RAND HIE, by virtue of individual-level changes in health insurance. Finkelstein (2007) attributes this larger effect to a shift in supply induced by market-wide changes in demand. Our study of large coverage expansion also examines the case of general equilibrium effects, rather than partial equilibrium analysis such as the RAND HIE. This paper is also related to a growing body of literature that examines the effect of a large health insurance coverage expansion on various outcomes in developing countries such as Mexico, Colombia, Thailand, and Taiwan.5 Under significant credit constraints in developing countries, health care utilization without insurance can be inefficiently low (Miller et al., 2009). Japan's per-capita gross domestic product (GDP) in 1956 was about one-quarter of that of the United States at that time. 6 Thus, our estimates may be more relevant to developing 4 For example, Finkelstein (2007) finds a large increase in hospital employment in response to the introduction of Medicare in the United States, but her data do not include most of the physicians, because physicians in the United States are not directly employed by hospitals. On the other hand, our data cover all physicians who were working at hospitals in Japan. 5 For example, see King et al. (2009) for Mexico; Miller et al. (2009) for Colombia; Cataife and Courtemanche (2011) for Brazil; Dow and Schmeer (2003) for Costa Rica; Hughes and Leethongdee (2007), Damrongplasit and Melnick (2009), and Gruber et al. (2012) for Thailand; and Chen and Jin (2010) for China. There are a considerable number of studies on Taiwan; see, for example, Chen et al. (2007), Chang (2011), and Chou et al. (2011). Studies on Taiwan also examine the effect of the introduction of universal health insurance; however, the empirical strategy of those studies mostly relies on difference-in-difference approaches, by comparing those previously covered to those recently covered. Such a strategy may not be able to capture the effects through market entry, as argued in Finkelstein (2007), unlike our case, which relies on prefecture-level hospital data. 6 Countries whose per-capita GDP is about one-quarter of that of the United States today include, for example, Chile and Turkey. Also, Japan's average life expectancy at that time was 66, whereas that of the United States was 70.

xcountries that are currently considering a massive expansion in health insurance coverage than to those of existing studies on developed countries such as the United States. 7 Our results show that countries planning to drastically expand health insurance coverage need to set aside enough financial resources to cover the anticipated surge in health care expenditures in both the short and long run. Another implication of our results is that a slow supply-side response can constrain attempts to meet the demand increase induced by large policy changes. The rest of the paper is organized as follows. Section 2 briefly reviews the history of Japan's universal health insurance system up to the 1960s, and describes our simple conceptual framework. Section 3 describes the data we use, and Section 4 presents the identification strategy. Section 5 shows the main results for utilization. Section 6 analyzes the supply-side responses to changes in demand. Section 7 concludes the paper.

2. Background 2.1. History of health care system in Japan 8 Japan's public health insurance system consists of two parallel subsystems: employment-based health insurance and the National Health Insurance (hereafter, NHI). Combining the two subsystems, Japan's health insurance program becomes one of the largest in the world today, covering nearly 120 million people, and making it almost three times as large as Medicare in the United States, which covers 43 million people (The Centers for Medicare and Medicaid Services, 2010). Employment-based health insurance is further divided into two categories: employees of large firms and government employees are covered by union-based health insurance, whereas employees of small firms are covered by government-administered health insurance. In both cases, employers must contribute around half of the insurance premiums, while the other half is deducted from employee salaries. Enrollment in the government-administered health insurance program was legally mandated for all employers with five or more employees, unless the employer had its own unionbased health insurance program. If a household head enrolls in an employment-based health insurance program, his or her dependent spouse and children are also covered by the employment-based health insurance. The NHI is a residential-based system that covers anyone who lives in the covered area and who does not have employment-based health insurance. Therefore, the NHI mainly covers employees of small firms (i.e., fewer than five employees), self-employed workers in the agricultural and retail/service sectors and their families, the unemployed, and the retired elderly. There were 4,877 municipalities in Japan as of 1955 (i.e., about 106 per prefecture on average), and each municipality is responsible for management of the NHI. An important feature of our identification strategy is that the decision to join the NHI system is left to municipalities, not individuals, and individuals living in covered municipalities cannot opt out. Both health insurance programs offer similar benefits and cover outpatient visits, admissions, diagnostic tests, and prescription drugs. However, different coinsurance rates are applied, depending on the type of insurance. Also, the rates have changed several times. When universal health insurance was achieved in Japan in 1961, the coinsurance rate for the NHI was 50% for both household heads and other family

7 Of course, the technology available at that time was quite different from that available now. However, the major causes of death in Japan around that time were not much different from the causes of death in developing countries now (e.g., pneumonia, bronchitis, gastritis, and duodenitis). 8 The discussion in this section draws heavily from Yoshihara and Wada (1999).

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

members, while that for employment-based health insurance was nearly zero for employees and 50% for family members. The coinsurance rate of the NHI for household heads was reduced to 30% in 1963, and that for other NHI enrollees was reduced to the same rate in 1968. In 1973, the coinsurance rate of employment-based health insurance for family members was also reduced to 30%.9 The history of Japan's public health insurance system dates back to the 1920s. First, in 1922, enrollment in employment-based health insurance was made mandatory for blue-collar workers in establishments with ten or more employees. In 1934, mandatory enrollment was expanded to workers in establishments with five or more employees. Then, to address the lack of health insurance amongst people excluded from employment-based health insurance, the NHI was introduced in 1938. During World War II, the wartime government rapidly expanded the NHI, and by 1944, universal health insurance had seemingly been achieved. However, in reality, coverage was far from universal: the medical system was not fully functional due to budgetary constraints as a result of the war. Furthermore, after defeat in the war, hyperinflation and other disruptions caused a serious breakdown in the health insurance system. The Japanese government, with the support of General Headquarters, began to restore the health insurance system immediately after the war. However, even in 1956, roughly one-third of the population (i.e., 30 million people) — mainly the self-employed, employees of small firms, the unemployed, and the retired elderly — was still not covered by any form of health insurance. Those without any health insurance had to bear the full cost of health care utilization. This lack of coverage was partly because a non-negligible number of municipalities had not yet rejoined the NHI system. 10 Therefore, in 1956, the Advisory Council on Social Security recommended that all municipalities should join the NHI system. Given this recommendation, the Four-Year Plan to achieve universal coverage by 1961 was proposed in 1957 by the Ministry of Health and Welfare. 11 By April 1961, all municipalities had joined the NHI, and universal health insurance was achieved. Fig. 1 shows the time series of health insurance coverage by the NHI, employment-based health insurance, and both types of insurance combined. The figure also includes a linear trend extrapolated from data prior to 1956. Two vertical lines indicate 1956, the reference year before the start of the Four-Year Plan, and 1961, the year in which universal health insurance was achieved. The number of individuals covered by both employment-based health insurance and the NHI gradually increased until the mid-1950s, and then there was a sharp increase, especially in NHI coverage, in the late 1950s. During the 4 years immediately preceding 1961, around 30% of the total population that was previously uninsured was covered by health insurance. Crowding-out of employment-based health insurance due to expansion of the NHI seems to be negligible. The insured would have likely preferred employment-based health insurance, as it offered lower coinsurance rates and the employer contributed to the premium. In theory, NHI expansion could have increased the number of self-employed workers by insuring them, who would otherwise not have been covered by employment-based health insurance. 12 Another source of crowding-out is that the expansion of the NHI could have induced firms to reduce their size to fewer than five employees, in order to be exempted from contributing to employment-based 9

The cap on the maximum out-of-pocket expenditures was not introduced until 1973. It is not clear why some municipalities did not rejoin the NHI system. Thus, we check the correlation between pre period insurance coverage rates and various potentially confounding factors in the following sections. 11 In 1959, an amendment to the National Health Insurance Act legally prescribed the mandatory participation of all municipalities in the NHI by April 1961. 12 For example, see Madrian (1994) on the job-lock effects of employment-based health insurance. 10

3

health insurance. Appendix Section A1 assesses both possibilities and finds no strong evidence of either type of crowding-out. 2.2. Institutions There are a few important institutional features of Japan's health insurance system from a supply-side perspective. First, detailed fee schedules are set by a centralized administration, and reimbursement from the health insurance system to medical providers follows these schedules strictly. 13 Until 1963, each medical institution was able to choose one schedule from two options, but it had to apply the same schedule to all patients. Thus, there was little room for each hospital or physician to charge different fees for specific types of patients, as in the United States (Cutler, 1998). Furthermore, from 1963 onwards, fee schedules have been integrated into a unified schedule that has been applied nationwide.14 Second, there was no effective legal obligation on physicians or hospitals to provide care for uninsured patients.15 Public aid for the uninsured was limited to patients quarantined with tuberculosis and other diseases specified in the Infectious Diseases Prevention Act, and those who lived on welfare. In contrast to the strict price control policy, entry and expansion of private hospitals had been left unrestricted until the upper limit on the number of beds in each region was introduced in 1985, which is out of our sample period (1950–1970). In the 1950s and 1960s, the government attempted to increase the supply of medical institutions in regions with short supplies, but the effect of this move seems to be limited. Construction of public institutions is of course guided by the government, but its impact is small compared to the increase in private hospitals. 16 Regarding private institutions, Medical Care Facilities Financing Corporation was founded in 1960 to facilitate the financing of private medical institutions. Although this institution alleviated the credit constraints of potential entrants, the decisions of whether to enter the market or not, and where to build hospitals, were still voluntary. On the other hand, the supply of physicians and nurses was constrained by the capacity of medical and nursing schools. However, their mobility was not controlled by the national government. Although medical schools had some power to control the choice of hospitals at which their alumni worked, there seemed to be no coordinated system to allocate physicians or nurses across prefectures in response to demand changes. 13 According to Ikegami (1991, 1992) and Ikegami and Campbell (1995), the national schedule is usually revised biennially by the Ministry of Health, Labor and Welfare through negotiations with the Central Social Insurance Medical Council, which includes representatives of the public, payers, and providers. 14 This stringent fee control is considered as one of the primary reasons why Japan had been able to maintain a relatively low total medical expenditures-to-GDP ratio (Ikegami and Campbell, 1995). The ratio of total medical expenditures to GDP had been slightly higher than 3 percent throughout the 1950s. Although it gradually increased during the early 1960s, it leveled off at around 4 percent in the mid-1960s until 1973, when healthcare services were made free for the elderly. 15 Article 19 of the Medical Practitioners Act stipulates that a physician cannot refuse diagnosis and treatment without legitimate reason. However, this Act was not very effective, as the lack of ability to pay the fee was considered as a legitimate reason. There was no legal obligation equivalent, for instance, to the Emergency Medical Treatment and Labor Act in the present-day United States, which mandates that hospitals must provide stabilizing care and examination for people arriving at an emergency room with a life-threatening condition, without considering whether a person is insured or has the ability to pay. 16 The ratio of public hospitals to the total number of hospitals was 33 percent in 1956; the number of public hospitals increased by only 6 percent in 1965, whereas private hospitals increased by 48 percent. Consequently, the share of public hospitals fell to 27 percent in 1965. Admittedly, however, since public hospitals tend to be larger than private ones, their share in terms of the number of beds was larger: 55 percent in 1956. Nonetheless, the rate of expansion of private hospitals was faster. The number of beds in public hospitals increased by 34 percent during the 1956-65 period, whereas that in private hospitals increased by more than 100 percent during the same period. Since we are not aware of any prefecture-level data on the number of hospitals by ownership, we are not able to examine the effect by ownership-type separately.

4

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Fig. 1. National time series of health insurance coverage rates. Note: Two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved. The dotted line is a linear trend extrapolated from data prior to 1956. Sources: Social Security Year Book (1952–1957) and Annual Report on Social Security Statistics (1958–1962).

2.3. Conceptual framework

3. Data

Fig. 2 illustrates a simple framework to interpret our estimates of the changes in healthcare utilization. The y-axis shows the before-insurance price of health care, and the x-axis shows quantity. Lines D and D' represent aggregate demand for health care before and after the implementation of the universal health insurance, respectively. Because the y-axis is the before-insurance price, the expansion of the insurance coverage is modeled by an outward shift of the demand curve rather than movement along the demand curve. Line S shows the aggregate supply of health care, whose vertical part represents the capacity constraints of health care supply.17 In the short term, the large health insurance expansion lowers the after-insurance price of health care for those who had not been covered previously. Hence, the demand curve shifts out from D to D', shifting the equilibrium from A to B. The short run response depends on sensitivity of demand to after- insurance price, and the slope of short run supply curve. If the new equilibrium is on the vertical part of the supply curve, as in Fig. 2, the capacity constraint is binding. If so, in the long term, the supply might increase in response to this large shift in demand (e.g., new entries of hospitals and clinics into the market) and it might shift out the supply curve from S to S', thereby shifting the equilibrium from B to C. This simple framework illustrates that what we estimate is the movement from one equilibrium point to another. In other words, we will not directly estimate either the elasticity of demand or the supply curve, but we will estimate the general equilibrium effects, which may include shift in supply. The new equilibrium is determined by the following two factors: the sensitivity of demand to the after-insurance price, and the long-run supply-side responses to such demand shifts. Although our research design cannot clearly distinguish between these two factors, the evolution of the estimated effects over time will give some information on these two factors, since the demand shift is expected to occur immediately after the implementation of the universal health insurance, whereas the supply response may emerge with a time lag.

Our data has been derived from various sources. Although the decision to join the NHI was made at the municipality level, municipality-level data are not available for the outcomes and explanatory variables. Thus, our unit of observation is the prefecture year. It is important to note that our analyses at the prefecture level can capture changes caused by entry and exit of medical institutions, unlike studies that rely on hospital-level data. There were 46 prefectures in Japan during our period of study, i.e. 1950–1970.18 Each prefecture had 106 municipalities on average as of 1955. We focus mainly on the period from 1950 to 1970, although some specifications use a shorter time period, due to limited availability of data pertaining to variables of interest. 19 Appendix Table A1 describes the definition, data sources, and available periods for each variable. All expenditure variables are converted to real terms at 1980 price levels, using a GDP deflator.

17 Note that, for simplicity, we do not incorporate price control by the government in the graph since it is not clear whether the price set by the government is above or below the market-clearing price shown in the graph.

3.1. Health insurance coverage rate We construct the rate of health insurance coverage for each prefecture at year 1956, the year before the implementation of the Four-Year Plan, as follows. First, the population covered by the NHI in prefecture p in 1956 (NHIp) is obtained from the Social Security Yearbook. Second, the population covered by employment-based health insurance is imputed from nationwide statistics, industry-level coverage rates, and the industry composition of each prefecture's workforce.20 Note that, due to data limitations, we need to assume that the coverage rate within

18 Amongst 47 prefectures in Japan as of now, the Okinawa prefecture has been excluded from our study as it was returned by the United States to Japan in 1973. 19 We do not extend our data beyond 1970, as some prefectures started to provide free care for the elderly in the early 1970s; this could have confounded our results. See Shigeoka (2012) for details on health care for the elderly in Japan. Also, attenuation bias caused by migration amongst prefectures would become more severe as the sample period increases. 20 Specifically, the population was divided into the following 13 categories: agriculture, forestry and hunting, fishing, mining, construction, manufacturing, wholesale and retail trades, finance and real estate, transportation and other utilities, service, government sector, unknown (employed), and unemployed.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Before Insurance Price

We define the impact of the health insurance expansion, impactp, as the proportion of the population without health insurance in prefecture p in 1956:

D’ S’

S

D

impact p ¼ 1−CovRp

B

A

C

Utilization Fig. 2. Conceptual framework.

each industry does not vary across prefectures (i.e., the variation in employment-based health insurance across prefectures is attributable solely to the variation in industry composition). 21 Then, for each year and prefecture, the coverage rate of each industry is weighted by the ratio of household heads in the industry. We use this weighted sum of industry-level coverage rates as the coverage rate of employment-based health insurance in each prefecture. Specifically, let E_CovRj denote the ratio of households covered by employment-based health insurance, amongst those with a household head who works in industry j, in 1956. Let Wpj denote the population living in prefecture p with a household head who works in industry j in 1956. Then, the imputed population covered by employment-based health insurance in 1956 in prefecture p can be written as ∑jWpj∗E_CovRj where E_CovRj is available from the Comprehensive Survey of the People on Health and Welfare.22 Wpj is calculated as linear interpolations from the 1955 and 1960 censuses. Lastly, the total population of each prefecture, popp, is taken from the Statistical Bureau's website. 23 Then CovRp, the ratio of prefecture p's population covered by any kind of health insurance in 1956, is estimated as follows 24: h i CovRp ¼ NHI p þ ∑ W pj  ECov Rj  =popp j

5

ð1Þ

21 Although some prefecture-level tables of employment-based insurance have been published, most of these tables show the location of employers, not that of employees. 22 Note that the Comprehensive Survey of the People on Health and Welfare classifies a household as being covered by an employment-based program if at least one of the household members is covered by an employment-based program. Although this is a sensible approach given that most employment-based insurance also cover spouses and children, it may also overstate the coverage rate of employment-based programs if some of the other household members are covered by the national program. Thus, as a robustness check, we tried to replace the coverage rate of employment-based program for households in the agricultural sector with zero, as most agricultural workers were self-employed in Japan at that time. The result remained virtually unchanged (available upon request). 23 These data seem to be interpolated from the Population Census by the Statistics Bureau, and the value is as of October 1. Thus, we take the average of 1955 and 1956 to derive the population as of April 1, 1956. 24 Because of measurement errors in E_CovRj, the sum of the two health insurance coverage rates slightly exceeds 100% in 4 prefectures (Iwate, Yamagata, Niigata, and Shiga). We replaced the coverage rates of these prefectures with 100%. We have confirmed that this adjustment does not change the results qualitatively.

ð2Þ

One potential concern with our calculation of impactp is the measurement errors associated with the imputation of the coverage by employment-based health insurance. Since large establishments tend to have higher coverage rates than smaller establishments in the same industry, heterogeneity in the coverage rate within each industry across prefectures may lead to an overestimation in the uninsured population for prefectures that have larger firms. To address this concern, we show robustness to an alternative measurement of insurance based on differential impact across different establishment sizes in Section 5.2. The proportion of the population without health insurance prior to universal health insurance (i.e., impactp) ranged from almost 0% in several prefectures to a high of 49%. Most of the variation in this coverage rate results from the variation in the NHI coverage rate. Indeed, the coverage rate of employment-based health insurance tends to be high in prefectures with a low total coverage rate; thus, the NHI coverage rate varies more than that of the sum of employment-based health insurance and NHI. It is critical for our identification strategy to check whether the distribution of the initial coverage rate is correlated with factors that determine healthcare utilization other than the insurance coverage. In particular, Japan experienced rapid economic growth during the period under study.25 If economic growth and health insurance expansion during the study period happen to move in the same direction, our estimates can be confounded with mean reversion of economic growth. To check this point, Panel A of Fig. 3 plots the correlation between changes in per-capita income (measured by the prefecture version of GNP) and impactp. It is difficult to know a priori whether income is positively or negatively correlated with the initial coverage rate. On the one hand, large firms might prefer to locate in growing prefectures and raise the coverage rate of employment-based health insurance. On the other hand, political pressure to restore the NHI might have been stronger in prefectures with a larger number of poor people who could not afford medical costs without the NHI. The figure suggests that the latter effect dominated the former: the increase in the coverage rate is slightly negatively correlated with the growth rate of per-capita GNP in the long term. Thus, any bias towards the estimated positive effects of health insurance expansion is likely to be downward, because the convergence of economic growth works against finding positive effects. Another question is why municipalities opted out of the NHI after the war. Panel B of Fig. 3 shows the correlation between impactp and two measures of the destruction by the war: the per-capita number of deaths (including missing) and buildings destroyed by the war. 26 The figure shows that prefectures hit harder by the war are slightly more likely to have a larger expansion in health insurance coverage. Had the municipalities hit harder by the war received more fiscal transfers for recovery during the 1950s, our estimates could have been biased. Thus, as a robustness check, we will control for the two measures of wartime destruction interacting with year fixed effects to address this concern. Fig. 4 shows the regional pattern of impactp. The figure shows substantial regional variation in the health insurance coverage rate. The 25 The average real GDP growth rate during the 1956-70 period is as high as 9.7 percent. As people became more affluent, their nutritional and sanitary conditions improved. Also, the Tuberculosis Prevention Act enacted in 1951 effectively suppressed tuberculosis, which had been one of the main causes of death in Japan until the early 1950s. 26 These two measures are the same variables used in Davis and Weinstein (2002). We are grateful for a referee's suggestion to use the war destruction measures, and thank David Weinstein for sharing the data with us.

6

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Fig. 3. Scatter plots of potentially confounding factors and health insurance coverage rate. Panel a: Per Capita GNP and Health Insurance Coverage Rate Panel b. Destruction by WWII and Health Insurance Coverage Rate. Note: Scatter plots of 46 prefectures. x-axes are impactp, the estimated changes in health insurance coverage rate from 1956 to 1961; see text for details.

proportion of the population without health insurance was relatively high in the southwest prefectures and low in the northeast prefectures. Additionally, prefectures with large populations — such as Tokyo and Osaka — tended to have low coverage rates, given the additional time needed to build a health insurance tax-collection system and to reach agreements between local governments and medical providers in cities with a larger number of physicians (Yoshihara and Wada, 1999).

3.2. Outcome and explanatory variables Our main outcome variables are divided into two categories: utilization and the supply-side response measured by capital and labor inputs. 27 Three measures of utilization are admissions, inpatient days,

27

We also examined the age-specific mortality rates in Appendix A3.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

7

Fig. 4. % of Population without any health insurance as of April 1956. Note: The proportion of the population without health insurance shown in this figure corresponds to impactp used in the regressions. See the text for the details.

and outpatient visits. Admissions represent the number of admissions to hospitals in each prefecture per calendar year. Inpatient days are the sum of the days in hospitals amongst all inpatients, while outpatient visits are visits to hospitals for reasons not requiring hospitalization. Note that these variables are limited to the utilization of hospitals (defined in Japan as medical institutions with 20 or more beds), because clinics (institutions with no more than 19 beds) are excluded from the survey.28 From a number of different sources, we also obtain the number of hospitals, clinics, beds, physicians, and nurses, in order to explore supply-side responses to expansions in health insurance coverage.

28 Unlike in the United States, direct outpatient visits to hospitals are a common practice in Japan, since there are no restrictions on the patients’ choice of medical provider. Therefore, an increase in the number of outpatient visits may simply reflect the fact that people are switching from clinics to hospitals for outpatient visits. However, almost all admissions occur in hospitals, and thus our data captures the universe of admissions and inpatient days in Japan.

Figs. 5 and 6 present the time-series patterns for each outcome variable used in this study; we compare the prefectures whose ratio of uninsured population was greater than the median (27.5%) in 1956 (i.e., high-impact prefectures) to the others (i.e., low-impact prefectures). Fig. 5 describes the utilization measures (admissions, inpatients, and outpatients) per capita. Health care utilization in high-impact prefectures seems to have started increasing more rapidly than that in low-impact prefectures, following the introduction of universal health insurance; however, the pattern is not very clear. Fig. 6 shows the supply-side variables (hospitals, clinics, beds, bed occupancy rates, physicians, and nurses); as in Fig. 5, all variables, with the exception of the bed occupancy ratio rate, increased during the sample period. The bed occupancy rate declined in the late 1950s and increased in the 1960s after the achievement of universal health insurance, possibly due to an increase in the number of inpatients. Also, high-impact prefectures had, on average, more clinics and physicians before 1956 than the low-impact ones. These two figures

8

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Fig. 5. Time series of health care utilization (weighted average by population). Note: The two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved, respectively. Low impact prefectures are prefectures whose rate of uninsured population was less than 27.5% in 1956, i.e., lower than the median. The sample covers the period from 1950 or the earliest year of available data if the data in 1950 is not available (1952 for admissions) to 1970.

underscore the importance of controlling for pre-existing differences across prefectures. Table 1 reports the summary statistics. The mean represents the weighted average of outcomes where population is used as weights, as in regression analysis. We also show the mean for 1956, the reference year for both low-impact and high-impact prefectures. In the last column, we show the coefficient of impactp from linear regression of each outcome variable over the available period on impactp without any controls. Importantly, prefectures whose initial coverage rates were lower (i.e., high-impact prefectures) tended to be more affluent, with more medical resources, lower mortality rates, and also slightly higher war-related destruction, prior to the implementation of universal coverage. This difference mostly arises from the following two factors: the larger proportion of the pre-period NHI enrollment by municipalities in the northeast prefectures, which happened to lag behind other regions in terms of industrialization and economic growth at that time, and the particularly low initial coverage rates in the two largest cities, Tokyo and Osaka. To account for these differences, we introduce dummy variables for 10 regions (defined as groups of several prefectures) interacted with year dummies in all specifications, and present results excluding Tokyo and Osaka as a robustness check.

in 1956, 1 year prior to the start of the Four-Year Plan to achieve universal coverage by 1961. The basic idea is to compare changes in outcomes in prefectures where the implementation of universal coverage led to a larger increase in health insurance coverage, to prefectures where it had a smaller effect. As discussed earlier, health insurance coverage prior to the universal health insurance may not be random. For example, income levels in 1956 and war destruction tended to be higher in prefectures with more uninsured people. Therefore, it is essential to control for unobserved components that potentially correlate with both the initial coverage rate of health insurance and health care utilization. We control for differences in the levels of the outcome variables by controlling for prefecture-level fixed effects. Furthermore, we divide the 46 prefectures into 10 regions and control for region-year effects; we also control for convergence in the growth rates by including the interaction terms of the initial value of the outcome variable and year dummies.29 The basic estimation equation is as follows:     Y pt ¼ α p  1 pref p þ δrt  1ðyeart Þ  1 pref p ∈ regionr   þκ t  Y p1956  1ðyeart Þ þ ∑ λt impactp  1ðyeart Þ t≠1956

þX pt β þ εpt

ð3Þ

4. Identification strategy Our identification strategy is akin to that of Finkelstein (2007). We exploit variations in health insurance coverage rates across prefectures

29 We divide the 46 prefectures into the following 10 regions, as defined by the Statistics Bureau: Hokkaido, Tohoku, Kitakanto-Koshin, Minamikanto, Hokuriku, Tokai, Kinki, Chugoku, Shikoku, and Kyushu.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

9

Fig. 6. Time series of per capita supply of health care (weighted average by population). Note: The two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved, respectively. Low impact prefectures are prefectures whose rate of uninsured population was less than 27.5% in 1956, i.e., lower than the median. The sample covers the period the earliest year of available data (ranging from 1951 to 1954 depending on the outcome, see Table 1) to 1970, except bed occupancy rate, which is available only up to 1966.

where subscript p indicates prefecture and t indicates year. αp represents a prefecture fixed effect; δrt represents a region-specific year effect; κt is meant to capture the differences in the growth of Y due to differences in the initial value; and impactp is the percentage of the population 1956 in prefecture p without health insurance, as defined in (2). Our parameters of interest are the λt's, which represent the coefficients of the interaction terms between year dummies and the percentage of the population without health insurance in 1956. A plot of λt's over t shows the flexibly estimated pattern over time in the changes in prefectures where the enforcement of universal coverage had a larger impact on the insurance coverage rate, relative to prefectures where it had a smaller impact. If the trend in these λt's changes

around about the 1957–1961 period — the phase-in period of universal coverage — such a change in trend is likely to be attributable to an expansion in health insurance. It is important to note that Eq. (3) does not impose any ex ante restrictions on the timing of the structural break in trend; we therefore allow the data to show when changes in the time pattern actually occur. The covariate Xpt controls for potentially confounding factors that might have been varying over time across prefectures. In our basic regression over the 1950–1970 period, we only include the log of the total population and the proportion of the population over 65 years of age, as data for many of the other control variables are not available for the years prior to 1956. As a robustness check, we restrict the sample to the 1956–1970 period and additionally include the log of real

10

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Table 1 Mean of dependent and control variables. Variable

Obs

Available period

Whole period

All prefectures in 1956

High impact prefectures in 1956

Low impact prefectures in 1956

Linear coefficient of impactp

Admission (thousands) Inpatient days (thousands) Outpatient visits (thousands) Hospitals Clinics Number of beds in hospitals Bed occupancy rate (%) Number of physicians in hospitals Number of nurses in hospitals Mortality rate: age 0−4 Mortality rate: age 5–9 Mortality rate: age 50–54 Mortality rate: age 55-59 Mortality rate: age 60–64 Population (thousands) Population over 65 (%) Real GNP per capita (in 1980 thousand yen) Real local government expenditure on health and sanitation (1980 thousand yen) Local government expenditure to revenue ratios Population density in 1955 Per capita deaths caused by WWII Per capita destroyed buildings caused by WWII

874 966 966 920 782 828 690 828 874 920 920 920 920 920 966 966 736 690

1952–1970 1950–1970 1950–1970 1951–1970 1954–1970 1951–1970 1952–1966 1953–1970 1952-70 1951–1970 1951–1970 1951–1970 1951–1970 1951–1970 1950–1970 1950–1970 1955–1970 1956–1970

148.5 7517.1 9744.5 215.4 2455.6 27619.7 82.1 1516 5884.6 8.1 0.9 8.2 12.9 20.5 3325.8 4.9 700.7 5.6

91.5 5610.1 7322.9 180.9 1911.7 19439.1 81.1 1349.7 3649.9 10.6 1.2 9.6 14.5 22.8 2939.6 3.9 378.9 1.8

118.4 7087.2 9388.6 223.3 2494.0 24420.5 81.6 1739.1 4774.8 9.9 1.1 9.4 14.2 22.3 3607.4 3.8 415.0 1.9

48.0 3224.9 3987.3 112.5 971.4 11395.3 80.2 720.9 1833.4 11.8 1.2 9.8 15.0 23.7 1861.2 4.2 320.5 1.5

1.483 81.96 103.7 2.232 28.37 281 0.044 17.47 62.36 −0.0294 0.00111 -0.00497 −0.0194 -0.0489 31.26 −0.0016 1.777 0.0169

690 46 46 46

1956–1970 1955 – –

1.03 797.1 0.004 0.031

1.02 – – –

1.03 1104.671 0.006 0.043

1.01 300.34 0.001 0.010

0.0002 12.6 0.0001 0.0007

Note: Mortality rate is the number of deaths per 1000 population. High impact prefectures are prefectures whose uninsured rate was 27.5% or higher in 1956. Low impact prefectures are prefectures whose uninsured rate was lower than 27.5% in 1956. 27.5% is the median uninsured rate in 1956.

GNP per capita, local governments' revenue-to-expenditure ratios, and the log of local governments' per-capita real expenditures on health and sanitation. Also, to control for changes in coinsurance rates applied only to the NHI in 1963 and 1968, we add interaction terms between the proportion of population covered by the NHI in the year prior to these changes, and dummy variables indicating the period after these changes. We use prefecture-specific population as a weight in all regressions, to account for substantial variations in population size. We also cluster the standard errors at the prefecture level, to allow for possible serial correlation within prefectures over time. Furthermore, following Finkelstein (2007), we take the following two approaches to account for pre-existing trends. First, we calculate changes in λt during the first 5 years following 1956 — the year in which the Four-Year Plan commenced — and take the differences with the changes in the 5 years prior to 1956; we calculate (λ61 − λ56) − (λ56 − λ51) and their estimated standard errors, to see whether they are statistically significantly distinct from zero. We also estimate (λ66 − λ61) − (λ56 − λ51); that is, we repeat the same exercise for the 1961–1966 period — the second 5-year period following the expansion – to examine the long-term effects. A drawback of this approach, however, is that it relies only on 3 years of data, and thus results can vary, depending on which of the 3 years are chosen for point-to-point comparisons. To utilize all available data efficiently, we also estimate the following deviation-from-trend model:     Y pt ¼ α p  1 pref p þ δrt  1ðyeart Þ  1 pref p ∈ regionr

5. Results on Utilization

þκ t  Y p1956  1ðyeart Þ þ γ pre  yeart  impactp

5.1. Basic results

þγmid  1ðyeart ≥ 1956Þ  ðyeart −1956Þ  impactp þγ after  1ðyeart ≥ 1961Þ  ðyeart −1961Þ  impactp þX pt β þ εpt

achievement of universal coverage in 1961. Therefore, we allow the slope to vary between the expansion period (1956–1961) and the lagged period (1961–1970). A disadvantage of this approach is that we need to impose ex-ante restrictions on the timing of trend breaks. Finally, it is important to clarify how much and in which direction migration could bias our results. During the 1950–1970 period, there were substantial inflows in working-age population given the massive migration from rural areas to cities, especially to Tokyo and Osaka, induced by the rapid industrialization and economic growth in the 50s and 60s. Since large cities tended to have low coverage rates in 1956, the prefectures that had a large increase in insurance coverage from 1956 to 1961 also had substantial population inflows during the same period. Given that younger individuals are less likely to use health care than older ones, any bias caused by inter-prefecture migration would drive estimates towards zero. In Appendix A2, we examine the relationship between changes in population and the initial insurance coverage rate. We find that, with controls for the initial level of population, changes in the insurance coverage rates are not associated with changes in population or age composition. Furthermore, as a robustness check, we present results that exclude Tokyo and Osaka from the sample. If inter-prefecture migration were to cause substantial biases, the results obtained by excluding Tokyo and Osaka should be different from the results obtained by including the two prefectures.30

ð4Þ

γpre captures any pre-existing trends that are correlated with health insurance coverage rates in 1956. γmid represents any trend breaks caused by the massive expansion in health insurance, starting in 1956; and γafter is meant to capture further trend breaks after the

Fig. 7 plots the estimated λt′s from Eq. (3) without prefecture-specific linear trends for the following three dependent variables serving as

30 It is also possible that sick people would migrate from a municipality without NHI coverage to one with NHI coverage, within the same prefecture. If so, actual changes in health insurance status might have been larger amongst healthier people, and thus the impact on health care utilization and health outcomes might have been smaller, than would have been in the case without such migration.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

11

Fig. 7. Effect of Health Insurance Coverage on Utilization. Note: Fig. 7 plots λt's, the coefficients of impactp interacted with year dummies in Eq. (3), over year t. Each graph shows the flexibly estimated pattern over time in the changes in the dependent variable given in the graph title in prefectures where the enforcement of universal coverage had a larger impact on the insurance coverage rate, relative to prefectures where it had a smaller impact. The two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved, respectively. Regressions on which these graphs are based include prefecture-fixed effects, region-specific year effects, interactions between year dummies and the value of the dependent variable as of 1956, log population and the proportion of the over 65 in the total population. Standard errors are clustered by prefecture. The sample covers the period from 1950 or the earliest year of available data if the data in 1950 is not available (1952 for admissions) to 1970.

measures of health care utilization: log of admissions, inpatient days, and outpatient visits. Because 1956 is the reference year, λ56 is set to 0 by definition. Therefore, the coefficient in each year can be interpreted as the relative change in outcomes from 1956 that would have resulted if the expansion in health insurance coverage had increased the coverage ratio by 100%, compared to a prefecture where the coverage ratio did not change. The upper left-hand graph in Fig. 7 shows the results for new admissions to hospitals. Until 1956, there was no pre-existing trend in the λt′s; at that point, the number of admissions started to increase rapidly in areas where health insurance expansion had a larger impact. The estimated λt′s for 1961, the year when the universal coverage was achieved, and 1966, 10 years after the reference year, are reported in rows (1) and (9) of Table 2. 31 The estimated λ61 and λ66 are 0.290 and 0.548, respectively. Given that roughly 28% of the total population did not have any health insurance as of 1956, these estimates imply that admissions increased by 8.5% (= exp[0.290 ∗ 0.28] − 1) over 5 years due to the enforcement of universal health insurance.

31 Henceforth, we focus mainly on λ61, which represents the changes up to the full achievement of universal health insurance, and λ66, which represents the changes during the 10 years following 1956, the reference year. The estimated coefficients and standard errors for 1950-70 are available from the authors upon request.

The long-term effect is even larger; admissions increased by 16.6% over 10 years. Inpatient days and outpatient visits show trends very similar to those of admissions: both graphs depict a sharp increase in the respective trends in the late 1950s and remain high until the late 1960s. The magnitude is larger for outpatient visits than for either admissions or inpatient days. The estimated λ61 and λ66 imply a 7.3 and 11.6% increase in inpatients days, and a 12.6% and 25.1% increase in outpatient visits, by 1961 and 1966, respectively. It is important to note that Fig. 7 shows a downward trend over time in the λt′s for inpatient days up through around 1956, and thus the estimate of the impactp on inpatient days would be higher if we accounted for the deviation from the downward trend. This is shown in the robustness checks below. There does not seem to be such a pre-existing trend for outpatient visits. 5.2. Robustness checks Table 2 presents the robustness checks for our utilization results. To save space, we only report estimates for the interaction terms of 1961 and 1966. To make the results from different specifications comparable to our basic results, rows (1) and (9) repeat the results from the basic specification. First, to check whether our results are driven by prefectures with large populations, we exclude Tokyo and Osaka, the two largest

12

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Table 2 Effects of Health Insurance Coverage Results on Utilization. Dependent variable: λ in 1961 (1) λ shown in Fig. 7 (2) Excluding Tokyo and Osaka (3) More controls (sample period: 1956–1970) (4) Adding year dummy * population Density 1955 and income 1956 (5) Adding year dummy * destruction By WWII (6) Prefecture specific linear trends (7) impactp replaced with non-coverage rate imputed from establishment size (8) impactp instrumented with emp. based insurance imputed from establishment size

λ in 1966 (9) λ shown in Fig. 7 (10) Excluding Tokyo and Osaka (11) More controls (sample period: 1956–1970) (12) Adding year dummy * population Density 1955 and income 1956 (13) Adding year dummy * destruction By WWII (14) Prefecture specific linear trends (15) impactp replaced with non-coverage rate imputed from establishment size (16) impactp instrumented with emp. based insurance imputed from establishment size

Log(admissions)

Log(inpatient days)

Log(outpatient visits)

0.290** [0.116] 0.267** [0.116] 0.279** [0.105] 0.300** [0.116] 0.228* [0.120] 0.192** [0.073] 0.263*** [0.093] 0.311** [0.119]

0.253** [0.103] 0.218** [0.108] 0.265** [0.104] 0.234** [0.107] 0.145 [0.110] 0.449*** [0.064] 0.212** [0.082] 0.259** [0.108]

0.426*** [0.116] 0.389*** [0.132] 0.412*** [0.130] 0.444*** [0.127] 0.347** [0.152] 0.409*** [0.110] 0.419*** [0.091] 0.494*** [0.120]

0.548*** [0.196] 0.459** [0.195] 0.567*** [0.188] 0.501** [0.197] 0.385* [0.195] 0.403*** [0.066] 0.508*** [0.167] 0.587*** [0.205]

0.392** [0.150] 0.302* [0.157] 0.412*** [0.149] 0.331** [0.153] 0.237 [0.153] 0.786*** [0.089] 0.331** [0.126] 0.396** [0.164]

0.800** [0.301] 0.637** [0.294] 0.884*** [0.272] 0.695** [0.290] 0.594* [0.298] 0.748*** [0.077] 0.775*** [0.252] 0.888*** [0.299]

Note: Table reports λ61 and λ61, the coefficients of impactp interacted with dummies for year 1961 and 1966, in Eq. (3) and its variants specified in each row. λt represents the effects of a 100%-point change in the insurance coverage rate on the relative changes in the dependent variable given in the column title from 1956, the baseline year, to year t. Standard errors, estimated with clustering by prefecture, are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The sample covers the period from 1950 or the earliest year of available data if the data in 1950 is not available (1952 for admissions) to 1970, unless otherwise specified. Rows 1 and 9: Baseline specification of Eq. (3). Time varying prefecture-level controls (Xpt) consist of the log of the total population and the proportion of the population over 65 years of age. Rows 2 and 10: Rows 1 and 9 excluding Tokyo and Osaka, which together comprised 15% of Japan's total population in 1956. Rows 3 and 11: Rows 1 and 9 with additional time-varying prefecture level covariates (the log of the real GNP per capita, the ratio of local governments' revenues to expenditures, local governments' per-capita real expenditures on health and sanitation, and interaction terms between the proportion of population covered by the NHI in the year prior to the change in coinsurance rates in 1963 and 1968 and dummy variables indicating the years after these changes). The sample period is limited to 1956–1970 because data on these additional variables are not available prior to 1956. Rows 4 and 12: Rows 1 and 9 with 1956 GDP per capita and 1956 population density interacted with year dummies. Rows 5 and 13: Rows 1 and 9 with measure for wartime destruction (the per-capita number of deaths and number of buildings destroyed by the WWII) interacted with year dummies. Rows 6 and 14: Rows 1 and 9 with prefecture-specific linear trends. Rows 7 and 15: Baseline specification of Eq. (3), where impactp is replaced with the alternative measure imputed by the industry-prefecture level variation in establishment size as defined in Eq. (5). Rows 8 and 16: Baseline specification of Eq. (3), where impactp is instrumented with the alternative measure imputed by the industry-prefecture level variation in establishment size as defined in Eq. (5).

prefectures, which together comprised 15% of Japan's total population in 1956. Since these two prefectures received large inflows of migrants during the study period, excluding these two prefectures also serves as an indirect test to examine whether our results are driven by the inter-prefecture migrations. Rows (2) and (10) indicate that our results are not driven by these prefectures. Second, to control for other confounding factors that may affect the outcomes, we add the following time-varying prefecture-level variables: the log of the real GNP per capita, converted to 1980 yen; the ratio of local governments' revenues to expenditures; and local governments' per-capita real expenditures on health and sanitation. Also, to control for changes in coinsurance rates applied only to the NHI in 1963 and 1968, we add interaction terms between the proportion of population covered by the NHI in the year prior to these changes and dummy variables indicating the years after

these changes. As most of our additional control variables are available only after 1956, in this specification, we limit the sample to the period from 1956 to 1970. 32 As seen in rows (3) and (11), adding these control variables does not significantly change the estimated coefficients. Third, we included the other pre-period area characteristics interacted with year dummies. Rows (4) and (12) show that our results are not affected by adding 1956 GDP per capita and 1956 population density interacted with year fixed effects, to account for the potential correlation between the war destructions and the initial insurance coverage. Rows (5) and (13) show the results that include a measure for wartime destruction (the per-capita number of deaths

32

Limiting the sample to 1956-70, in itself, has no impact on the estimated coefficients.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

and number of buildings destroyed by the WWII) interacted with year fixed effects. Although the estimates for inpatient days are no longer statistically significant, the estimates for outpatient visits and admissions remain significant at least at the 10% level. Also, the magnitudes of the estimates are not quantitatively different from those of the estimates from other specifications. Note that these specifications are much more demanding, and hence reassuring, if inclusion of these controls does not “kill” our results from main specifications. Fourth, we control for prefecture-specific linear trends in the equation (estimation) for outcome variables whose data are available dating back to at least 1952. Note that, however, we only have four to six observations prior to the base year and that the gradual change in insurance coverage took place over a 4-year period. Thus, the estimated prefecture-specific linear trend might be over-fitted; that is, it might pick up part of the effect of the policy change for our interest. Given this possibility of over-fitting, we do not include prefecture-specific linear trends in our main specification. Rows (6) and (14) show the results with prefecture-specific linear trends. Although some of the point estimates change, all λt′s remain statistically significant. So far, we have deduced each prefecture's coverage of employmentbased insurance from industry-level coverage rates and each prefecture's industry composition, assuming that the coverage rate within each industry does not vary across prefectures (i.e., the variation in employment-based health insurance across prefectures is attributable solely to the variation in industry composition). Below, we show that our omission of heterogeneity in industry-level coverage rates across prefectures does not significantly bias our results. The biggest source of heterogeneity in employment-based insurance coverage within industries across prefectures is the differences in establishment size composition of each industry across prefectures. Since establishments with five or more employees had a legal obligation to provide employment-based health insurance, the coverage rates should be higher in prefectures with larger establishments. As shown in Appendix Table A4, the coverage rates are indeed higher for industry establishments with five or more employees. Unfortunately, information on health insurance coverage by establishment-size within each industry is not available, preventing us from using both the establishment-size composition within each industry and the industry composition within each prefecture simultaneously, in order to deduce employment-based health insurance coverage more precisely. However, data on the number of employees by establishment size by each prefecture-industry cell as of 1957 are available from the Establishment Census. Assuming that all workers in establishments with 5 or more employees are all covered by employment-based insurance, we can calculate the alternative measure for impactp by using the industry-prefecture level variation in establishment size. Such an alternative estimate for impactp is defined as follows 33: "

#

impact ′p ¼ 1− NHI p þ ∑W pj  %5up est jp =popp j

ð5Þ

where %5up_estjp is the ratio of workers employed in establishments with more than five employees in industry j in each prefecture p, and Wpj is the total employment of industry j in prefecture p, which is the same as used in Eq. (1). Note that, however, as shown in Appendix Table A4, the coverage rates are lower than the proportion of workers in establishments with 33 We are grateful for a referee's suggestion to use the industry-prefecture level variation in establishment size.

13

five or more employees (i.e., workers who are supposed to be covered), probably because part-time and non-regular workers are not entitled to receive employment-based insurance. In particular, the low insurance coverage rate for the construction industry is attributable to its reliance on short-term, non-regular workers. It is also possible that some establishments evaded the legal obligation. Thus, this alternative calculation is likely to overestimate the coverage rate, and impactp′ is likely to be an underestimate of the true impact. Indeed, as shown in Appendix Fig. A1, the coverage rates based on establishment size composition tend to be higher than those based on industry-level coverage rate reported in the nation-wide statistics. Since this overestimation of employment-based insurance makes many prefectures to have a coverage rate exceeding 100%, we do not use impactp′ in our main analysis. That said, the proportion of workers in establishments with five or more employees should be a good predictor of the coverage rates of employment-based insurance in each prefecture-industry cells. Appendix Fig. A1 looks almost linear, implying that the estimates based on establishment size and the estimates based on industry composition are highly correlated. Thus, below we take two approaches to confirm the robustness of our results of either of the ways of estimating the coverage by employment-based insurance. First, rows (7) and (15) in Table 2 show results using impactp′ defined by the coverage rates based on the composition of establishment size within each industry-prefecture instead of impactp. The estimated λt′s are slightly smaller as the level of coverage rates is slightly higher, but the results are qualitatively the same. Second, we use the alternative measure as an instrument for impactp to reduce the measurement error associated with the imputation based on the industry composition. Rows (8) and (16) show that the IV estimates are almost the same as the main specifications in rows (1) and (9). In addition to the various robustness checks presented in Table 2, we show a series of further robustness checks to pre-existing trends. Specifically, we compare changes in λt during a fixed length of time following the expansion of health insurance coverage, relative to the change in λt during the same length of time before the expansion. We do not perform this test for admissions, as data for 1951 are not available. In row (1) of Table 3, we take a 5-year difference for the change in the outcome. The increase in both inpatient days and outpatient visits were statistically significant after 1956. Row (2) of Table 3 repeats the same 5-year test for 1961–1966, namely, the next 5-year period, using the same reference period (1951–1956). Table 3 Robustness of Utilization Outcomes. Dependent variable:

Log(admissions)

Log(inpatient days)

Log(outpatient visits)

(1) (λ61–λ56) − (λ56–λ51)

– – – – –0.028 [0.032] 0.098** [0.046] −0.038* [0.022]

0.462** [0.203] 0.349 [0.221] –0.048 [0.042] 0.117** [0.048] −0.044* [0.023]

0.481** [0.158] 0.353 [0.261] 0.006 [0.043] 0.085** [0.038] −0.038 [0.048]

(2) (λ66–λ61) − (λ56–λ51) (3) Slope prior to 1956 (4) (Slope in 1956–1961) to (Slope prior to 1956) (5) (Slope in 1961–1970) to (Slope prior to 1961)

Note: Standard errors, estimated with clustering by prefecture, are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The first two rows for Log(admissions) are blank because the data for 1951 are not available. Rows (1) and (2) report linear combination of λt's specified in the row headers from estimating Eq. (3), where λt represents the effects of a 100%-point change in the insurance coverage rate on the relative changes in the dependent variable given in the column title from 1956, the baseline year, to year t. Rows (3)–(5) report results from estimating Eq. (4) for each dependent variable indicated in the column titles.

14

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

None of the coefficients are statistically significant, although they are all positive. Rows (3)-(5) of Table 3 show the estimated coefficients of the two slopes in the deviation-from-trend model as Eq. (4). The slope prior to 1956 is not statistically significant and is close to zero for all three outcomes. The coefficients for the difference in the slopes both before and after 1956 (row (4)) are positive for all three utilization measures, and the changes indicated are in the same order as the estimates from other specifications. For example, the coefficient on the first slope for new admissions is interpreted as a 14.7% increase (= exp[0.098 ∗ 5 ∗ 0.28] − 1) by 1961. 34 In contrast, the estimated coefficients for the second slope (row (5)) are all negative, but the magnitude is smaller than that of the absolute value of the first slope, which is consistent with positive but flatter slopes after 1961, as shown in Fig. 7. Finally, we briefly compare our estimate of the utilization response relative to estimates from partial equilibrium analyses such as RAND HIE (Manning et al., 1987; Newhouse and the Insurance Experiment Group, 1993) and Shigeoka (2012). First, our estimates for outpatient visits, for example, are about four times larger than the increase from individual-level changes in health insurance that RAND HIE would have predicted. 35 Second, Shigeoka (2012) examines the effect of a reduction in a patient's out-of-pocket medical payment at age 70 in Japan during 1984–2008, and finds that a 70% reduction in out-of-pocket payment leads to a 10% increase in outpatient visits. This increase is similar in magnitude to that in this study in the short-term (12.6% in the 5 years following 1956) but much smaller than that in the long-term (25.1% in the 10 years following 1956). However, we need to view both comparisons with considerable caution because of many institutional differences in these two studies from our case of Japan in the 1950s.36

6. Results on Supply-Side Response Given the increase in utilization in response to the expansion of health insurance coverage in Japan, the next question is whether the supply side adequately accommodated a drastic increase in the demand for health care. Understanding this supply-side response is particularly important, since one of the major concerns regarding a massive expansion in health insurance coverage is a shortage of human capital, including physicians and nurses. 37

34 Note that the estimated coefficient provides only a one-year effect, and roughly 28 percent of Japan's total population had no health insurance coverage as of 1956. 35 Given that the coinsurance rate of the NHI in Japan was 50 percent at that time, the most comparable case in the RAND experiment HIE is the change in the coinsurance rate from 95 to 50 percent. Manning et al. (1987) showed that an individual who moved from 95 to 50 percent coinsurance would increase his or her annual number of face-to-face visits by 11 percent (i.e., from 2.73 to 3.03 visits). Therefore, the RAND HIE suggests that the effect of moving 28 percent of the population from being uninsured to being insured is tantamount to increasing the number of outpatient visits (i.e., face-to-face visits in hospitals) by 3.1 percent (11*0.28). Our estimates show that outpatient visits increased by 12.6 percent in the five years following 1956. Thus, our estimates are about four times larger than what individual-level changes in health insurance would have predicted. 36 For example, the RAND HIE took place in the United States in the 1970s, and the entire health care system, as well as people's living standards, were quite different from that in Japan in the 1950s. Moreover, RAND HIE randomly assigned insurance package at the individual level, while the policy decision to join the NHI in Japan was made at the municipality level. Also, while our paper examines the effect of health insurance per se (extensive margin), Shigeoka (2012) examines the effect of cost-sharing (intensive margin). Furthermore, Shigeoka (2012) only focuses on the elderly, who are likely to have different price responsiveness from that of the overall population in this study. 37 For example, one of the major concerns related to the Patient Protection and Affordable Care Act in the United States is the shortage of physicians (Association of American Medical Colleges, 2010).

The supply-side response is also interesting from a theoretical perspective. Finkelstein (2007) argues that a market-wide change in health insurance coverage may have larger effects than those implied by individual-level changes in health insurance coverage – especially if the expansion of health insurance coverage sufficiently increases the aggregate demand, so as to induce medical providers to incur the fixed costs associated with building new institutions. Thus, we test this hypothesis by estimating the effects of health insurance expansion on the number of medical institutions. The upper left-hand graph in Fig. 8 plots the estimated λt′s in Eq. (3), with the log of the number of hospitals as the dependent variable. The estimated λt′s for 1961, the year when the universal coverage was achieved, and 1966, 10 years after the reference year, are also reported in rows (1) and (9) of Table 4. The estimates for 1961 and 1966 are 0.229 and 0.578, respectively, and both are statistically significant at the conventional level. Therefore, this graph may lead one to believe that the hospitals increased in size in the areas where utilization had increased. However, the graph also shows a strong pre-existing trend before 1956. Indeed, as shown in rows (6) and (14) of Table 4, once prefecture-specific linear trends are included, the estimated coefficients are no longer significantly positive. Table 5 also reports that any positive effects of the number of hospitals disappear when pre-existing trends are controlled for; therefore, the positive association between the increase in health insurance coverage and the number of hospitals may not be a causal link. We repeat the same analysis for clinics; the results are shown in the upper right-hand graph in Fig. 8, as well as the second column of Table 4. As shown in the graphs, λt′s are not estimated very precisely. Moreover, none of the estimates presented in Table 4 are statistically significant. We cannot control for any pre-existing trends, because data related to clinics are available only from 1954 onwards; thus, rows (6) and (14) of Table 4 are blank and Table 5 does not contain a column for clinics. Overall, the response for the number of clinics is small. Next, we explore the other supply-side response, measured in terms of the supply of beds, physicians, and nurses. The rest of Fig. 8 shows the estimated λt′s for the following four outcomes: log of the number of beds, bed occupancy rate, log of the number of physicians, and log of the number of nurses.38 The graphs in the middle row of Fig. 8 show that the number of beds in Japan started to increase in the mid-1950s. Compared to 1956, the expansion of health insurance increased the number of beds by 3.4% by 1961 and by 10.9% by 1966. 39 The bed occupancy rate also increased significantly in the late 1950s and then declined in the early 1960s. This pattern suggests that although the number of beds increased in response to an expansion in health insurance coverage, the surge in the number of patients exceeded the increase in the supply of beds during the late 1950s. Unlike the case for the number of hospitals, we do not observe a discernible pre-existing trend for the number of beds. The third column of Table 4 and the second column of Table 5 confirm that the results for beds are robust to other specifications.

38 Because data regarding admissions, inpatient days, and outpatient visits has been obtained from hospitals only, we use the number of beds, physicians, and nurses working in hospitals, for the sake of consistency. We have confirmed that the results do not change much if we expand our data to all beds, physicians, and nurses in hospitals and clinics. 39 Note that the increase in the number of beds at that time was mainly driven by the entry and expansion of private hospitals. It is true that public hospitals also increased their supply of beds by 48 percent during the 1956-65 period; yet, the rate of increase in beds in private hospitals was in excess of 100 percent in the same period. As pointed by Ikegami (1992), there were no restrictions on the capital development of private hospitals until 1985, when a ceiling on the number of hospital beds by region was imposed. In contrast, the supply of physicians and nurses is inevitably constrained by the capacity of medical and nursing schools.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

15

Fig. 8. Effect of health insurance coverage on supply of health care. Note: Fig. 8 plots λt's, the coefficients of impactp interacted with year dummies in Eq. (3), over year t. Each graph shows the flexibly estimated pattern over time in the changes in the dependent variable given in the graph title in prefectures where the enforcement of universal coverage had a larger impact on the insurance coverage rate, relative to prefectures where it had a smaller impact. The two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved, respectively. Regressions on which these graphs are based include prefecture-fixed effects, region-specific year effects, interactions between year dummies and the value of the dependent variable as of 1956, log population and the proportion of the over 65 in the total population. Standard errors are clustered by prefecture. The sample covers the period of the earliest year of available data (ranging from 1951 to 1954 depending on the outcome; see Table 1) to 1970, except bed occupancy rate, which is available only up to 1966.

The bottom two graphs in Fig. 8 show the estimated λt′ s for the number of physicians and nurses. The graph in Fig. 8 for the number of physicians shows an increase, but at a slightly slower pace than that for beds, although the estimated λt′s are not always statistically significant. Pre-1956 data for the number of physicians are available only from 1953 onwards; thus, we do not control for prefecture-specific linear trends or pre-existing trends, and rows (6) and (14) of Table 4 are blank and Table 5 does not contain a column for physicians. The estimates for the number of nurses are noisier and are apparently weak.

To recapitulate our results: we find no robust evidence for an increase in the number of hospitals and clinics in response to an expansion of health insurance, but we do find evidence of an increase in the number of beds. The effect on the number of physicians seems to be positive and significant at the 10% level in some specifications, but estimates are noisier than that for beds, whereas the effect on the number of nurses is negligible. These various results are plausible, since it is less costly for existing hospitals to increase capacity by adding beds than for new hospitals to pay large fixed costs to enter the market. Also, not surprisingly, it is not as easy to increase the

16

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Table 4 Results on Supply of Health Care. Dependent variable:

Log(hospitals)

Log(clinics) Log(beds) BOR

Log(physicians) Log(nurses)

0.229** [0.092] 0.183* [0.092] 0.205** [0.091] 0.182* [0.091] 0.145 [0.100] −0.017 [0.060] 0.185** [0.081] 0.235** [0.107]

−0.021 [0.055] −0.031 [0.053] −0.013 [0.055] −0.024 [0.061] −0.06 [0.059] – – −0.028 [0.049] −0.035 [0.059]

0.121* [0.067] 0.085 [0.070] 0.130* [0.069] 0.122* [0.071] 0.016 [0.069] 0.075* [0.039] 0.087 [0.061] 0.11 [0.079]

0.122** [0.055] 0.115** [0.057] 0.128** [0.061] 0.099* [0.055] 0.097* [0.056] 0.351*** [0.059] 0.102** [0.042] 0.123** [0.057]

0.243* [0.128] 0.241* [0.130] 0.168 [0.118] 0.229* [0.123] 0.164 [0.134] – – 0.193 [0.115] 0.234 [0.144]

−0.132 [0.181] −0.24 [0.261] −0.102 [0.186] −0.031 [0.230] −0.105 [0.311] −0.146 [0.218] −0.123 [0.157] −0.141 [0.184]

0.578*** [0.194] (10) Excluding Tokyo and Osaka 0.509** [0.201] (11) More controls (sample period: 1956–1970) 0.622*** [0.164] (12) Adding year dummy * population 0.495*** Density 1955 and income 1956 [0.171] (13) Adding year dummy * destruction 0.448** Density 1955 and income 1956 [0.171] (14) Prefecture specific linear trends 0.145 [0.089] (15) impactp replaced with non-coverage rate imputed from establishment size 0.540*** [0.170] (16) impactp instrumented with emp. based insurance imputed from establishment size 0.649*** [0.226]

−0.085 [0.102] −0.096 [0.096] −0.083 [0.092] −0.065 [0.092] −0.132 [0.092] – – −0.087 [0.090] −0.104 [0.107]

0.368*** [0.128] 0.304** [0.135] 0.384*** [0.127] 0.329** [0.131] 0.228* [0.131] 0.299*** [0.065] 0.319** [0.128] 0.383** [0.165]

0.047 [0.051] 0.022 [0.061] 0.070 [0.061] 0.005 [0.056] 0.018 [0.056] 0.443*** [0.083] 0.019 [0.038] 0.023 [0.045]

0.387* [0.226] 0.387* [0.225] 0.372* [0.203] 0.353* [0.188] 0.353 [0.188] – – 0.389* [0.207] 0.459* [0.259]

0.142 [0.232] −0.068 [0.250] 0.182 [0.233] 0.116 [0.255] 0.09 [0.255] 0.157 [0.225] 0.119 [0.225] 0.138 [0.264]

λ in 1961 (1) λ shown in Fig. 8 (2) Excluding Tokyo and Osaka (3) More controls (sample period: 1956–1970) (4) Adding year dummy * population Density 1955 and income 1956 (5) Adding year dummy * destruction By WWII (6) Prefecture specific linear trends (7) impactp replaced with non-coverage rate imputed from establishment size (8) impactp instrumented with emp. based insurance imputed from establishment size

λ in 1966 (9) λ shown in Fig. 8

Note: Table reports λ61 and λ61, the coefficients of impactp interacted with dummies for year 1961 and 1966, in Eq. (3) and its variants specified in each row. λt represents the effects of a 100%-point change in the insurance coverage rate on the relative changes in the dependent variable given in the column title from 1956, the baseline year, to year t. Standard errors, estimated with clustering by prefecture, are presented in the brackets. BOR stands for bed occupancy rate. Standard errors, estimated with clustering by prefecture, are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The sample covers the period from 1950 or the earliest year of available data if the data in 1950 is not available (1952 for admissions) to 1970, unless otherwise specified. Rows (6) and (14) for log(clinics) and log(physicians) are left blank because available data prior to 1956 are limited to less than 4 years for these two outcomes. Rows 1 and 9: Baseline specification of Eq. (3). Time varying prefecture-level controls (Xpt) consist of the log of the total population and the proportion of the population over 65 years of age. Rows 2 and 10: Rows 1 and 9 excluding Tokyo and Osaka, which together comprised 15% of Japan's total population in 1956. Rows 3 and 11: Rows 1 and 9 with additional time-varying prefecture level covariates (the log of the real GNP per capita, the ratio of local governments' revenues to expenditures, local governments' per-capita real expenditures on health and sanitation, and interaction terms between the proportion of population covered by the NHI in the year prior to the change in coinsurance rates in 1963 and 1968 and dummy variables indicating the years after these changes). The sample period is limited to 1956–1970 because data on these additional variables are not available prior to 1956. Rows 4 and 12: Rows 1 and 9 with 1956 GDP per capita and 1956 population density interacted with year dummies. Rows 5 and 13: Rows 1 and 9 with measure for wartime destruction (the per-capita number of deaths and number of buildings destroyed by the WWII) interacted with year dummies. Rows 6 and 14: Rows 1 and 9 with prefecture-specific linear trends. Rows 7 and 15: Baseline specification of Eq. (3), where impactp is replaced with the alternative measure imputed by the industry-prefecture level variation in establishment size as defined in Eq. (5). Rows 8 and 16: Baseline specification of Eq. (3), where impactp is instrumented with the alternative measure imputed by the industry-prefecture level variation in establishment size as defined in Eq. (5).

numbers of physicians and nurses as to add beds, because the total supply of physicians and nurses is constrained by the capacities of medical and nursing schools. 40

7. Conclusion We have estimated the impact of a massive expansion in Japan's health insurance program on health care utilization and health outcomes 40 In theory, it is also possible that there was excess capacity before the expansion of health insurance coverage, or that the economies of scale enhanced efficiencies in the provision of medical services. Hence, it was not necessary to build new institutions or to hire new physicians and nurses.

in the country. We find substantial increases in health care utilization measured in terms of admissions, inpatient days, and outpatient visits to hospitals. We then investigate corresponding supply-side responses, as argued in Finkelstein (2007). While we do not find that the expansion of health insurance induced the market entry of hospitals and clinics – which would necessarily incur large fixed costs – we do find an increase in the number of beds, which may be less costly than market entry. One limitation of the current study is that we cannot address the question as to whether universal health insurance improves social welfare. Two main welfare gains are the reduction in mortality and risk protection against sudden out-of-pocket medical expenditures. In fact, we examined the effects on age-specific mortality rates, but our estimates are not precise enough to rule out either an economically meaningful improvement in mortality or no improvement at

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23 Table 5 Robustness of Supply of Health Care. Dependent variable:

Log(hospitals) Log(beds) BOR

(1) (λ61-λ56)-(λ56-λ51)

0.084 [0.195] 0.204 [0.273] 0.030 [0.031] 0.023 [0.037] −0.011

0.270 [0.199] 0.397* [0.212] −0.037 [0.035] 0.078* [0.045] −0.005

[0.037]

[0.021]

(2) (λ66-λ61)-(λ56-λ51) (3) Slope prior to 1956 (4) (Slope prior to 1956) (Slope in 1956–1961) (5) (Slope prior to 1961) − (Slope in 1961–1970)

– – – – 0.003 [0.022] 0.019 [0.020] − 0.046*** [0.016]

Log(nurses) – – – – −0.098* [0.058] 0.125 [0.081] −0.003 [0.059]

Note: BOR stands for bed occupancy rate. Standard errors, estimated with clustering by prefecture, are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. Clinics and physicians are excluded from the analyses because of the lack of pre-1956 data, and the first two rows for BOR and log(nurses) are blank because the data for 1951 are not available. Rows (1) and (2) report linear combination of λt's specified in the row headers from estimating Eq. (3), where λt represents the effects of a 100%-point change in the insurance coverage rate on the relative changes in the dependent variable given in the column title from 1956, the baseline year, to year t. Rows (3)–(5) report results from estimating Eq. (4) for each dependent variable indicated in the column titles.

all. 41 Furthermore, our results on health outcomes are limited to mortality, and thus it is possible that the introduction of universal health insurance reduced the morbidity of nonfatal diseases. Our limited data also do not allow us to explore the decline in the risk of sudden out-of-pocket medical expenditures, which is another important benefit of health insurance. 42 Rather, the key takeaway from our empirical results is that a large expansion in health insurance coverage will not only increase health care utilization immediately after expansion, but would also continue to increase healthcare utilization in the long term, regardless of whether it improves health outcomes. Therefore, countries planning to introduce universal health insurance need to anticipate a surge in health care expenditures, which might be associated with a shift in the supply curve through increased capacity. Also, our results for the supply side responses indicate that lags in an increase in supply capacity may constrain the ability of the health care system to meet the rapid increase in demand resulting from a massive expansion in coverage.

Appendix A A.1. Evidence against the crowding-out of employment-based health insurance by the NHI As explained in Section 2.1, there are two potential channels through which NHI expansion could crowd out employment-based health insurance. First, the NHI could increase the number of self-employed workers, who would otherwise not have been covered by employment-based health insurance, by insuring them. Second, the introduction of the NHI could induce firms to reduce their staff to fewer than five employees, in order to receive an exemption from making financial contributions to employment-based health insurance and to allow the NHI to cover their employees. To assess the first possibility, we calculate the ratio of self-employed workers to all individuals in the employed labor force, using 41

The results are summarized in Appendix A3. Appendix A4 shows that, at least on average, the introduction of universal health insurance did not affect out-of-pocket medical expenditures. We cannot explore the benefit of risk reduction, because data regarding the variances in individual household health care expenditures is not available. 42

17

data from the Population Censuses of 1950, 1955, and 1960. This self-employment ratio is calculated as the sum of the number of business owners without paid employees and family workers, divided by the number of all employed people who are either 15 years old or above (14 years for 1950). We exclude owners with paid employees, because they might be eligible for employment-based health insurance. Then, we regress the changes in this ratio from 1955 to 1960 on impactp, the ratio of uninsured individuals in 1956. As shown in Appendix Table A2, the ratio of uninsured individuals has no effect on the ratio of self-employed workers. Thus, we conclude that the first type of crowding-out did not occur in the case of Japan in the late 1950s. Regarding the second possibility, we obtain data on the number of establishments by size from the Establishment Census. This survey is conducted every 3 years, and we use data from 1951, 1954, 1957, 1960, 1963, and 1966. We estimate Eq. (3), with the exception that the base year (i.e., year with λ = 0) is 1957 instead of 1956. The estimated λ is shown in Appendix Table A3. Had NHI expansion induced some firms to reduce their number of employees so as to receive an exemption from contributing to employment-based health insurance, the number of establishments with one to four employees should have increased during the period from 1956 to 1961, while the number of establishments with five to nine employees should have decreased during the same period. Columns (1) and (2) of Panel A in Appendix Table A3 show that the number of establishments with one to four employees did not increase in response to NHI expansion, although the number of establishments with five to nine employees did decrease slightly. Columns (4) and (5) further show that, when looking at ratios rather than absolute numbers, establishments with one to four employees increased in the mid-1960s rather than in the late 1950s. Furthermore, the two estimates, λ63 and λ66, seem to be driven solely by Tokyo and Osaka. As shown in Panel B in Table A3, when we exclude Tokyo and Osaka, none of the λt′s remain statistically significant. Thus, column (4) of Panel B in Table A3 probably reflects the fact that Tokyo experienced a fall in the ratio of small establishments in the 1950s, and had already reached a much lower ratio than other prefectures by 1960, rather than a lagged response to NHI expansion. A.2. Migration During the period from 1950 to 1970, there were substantial inflows of working age population from rural areas to large cities, especially to Tokyo and Osaka. If population changes caused by such migrant inflows are correlated with changes in insurance coverage rates due to the implementation of universal health insurance, even after controlling for other explanatory variables, the estimated effects of the health coverage expansion may spuriously pick up the effects of population changes. One simple way to examine this concern is to estimate Eq. (3) with the log of the prefecture's total population as the dependent variable. Appendix Fig. A2 plots the estimated λt′s; prefectures with lower initial insurance coverage rates tend to have a faster increase in population, but the estimated λt′s are not statistically significant. Furthermore, the pre-existing trend seems very close to the post-1961 trends, implying that this faster growth in population seems to capture pre-existing trends before universal health insurance was implemented, and thus is not caused by the expansion of health insurance coverage. The first column of Appendix Table A5 confirms the same results. Even though the introduction of universal health insurance did not induce changes in total population, it might have changed the age-composition in the total population. For example, it is possible that young and prime age workers more often moved into larger cities to seek better employment opportunities than old, retired

18

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

people. Given that younger individuals are less likely to use health care services than older ones, any bias caused by such inter-prefecture migration would drive estimates towards zero. To assess this possibility, the second column of Table A5 presents the effect of impactp on the proportion of population older than 55 years of age in the census years (population by age is not available for inter-census years). The initial insurance coverage rate is not significantly correlated with the proportion of the old age population. Indeed, in the main text, we control for the proportion of the population over the age of 65 years in all specifications. The estimates are unaffected by controlling for the age distribution.

A.3. Results on mortality To complete the picture of the impact of expansion in Japan's health insurance coverage, we would ideally examine whether health insurance benefits the health of insured individuals. On one hand, cheaper access to health care services may improve health outcomes; on the other, if those who are not severely ill receive health care simply because of the availability of health insurance, or if the expansion of health insurance increases the volume of unnecessary treatments (i.e., an ex post moral hazard), there may be no effect on health outcomes. Therefore, the impact of health insurance on health outcomes is a priori ambiguous. As a measure of health outcomes, we compute the age-groupspecific mortality rate (i.e., number of deaths per 1,000 individuals) for the age groups of 0–4, 5–9, 50–54, 55–59, and 60–64 years. 43 We do not report the results for the age group of 10–49 years old, as the mortality rate is too low for this group. We also exclude the elderly (i.e., those aged 65 and over), to prevent our results from being confounded by the effects of welfare benefits paid to elderly persons who are not covered by the employment-based pension plan, introduced in 1961 as part of the National Pension Plan. 44 Appendix Fig. A3 presents the estimated λt′s in Eq. (3), with the mortality rates of five age groups as the dependent variables. The expansion of health insurance coverage does not seem to reduce the mortality rate amongst any of the age groups we studied. This lack of decline in mortality may be because individuals with acute, life-threatening, and treatable health conditions who might have previously sought care at hospitals, even when they lacked health insurance and consequently paid the costs incurred at their own expense. Although there was no public aid for the uninsured, mutual aid from blood relatives and the local community could have supported poor, uninsured patients. To examine such a possibility, we examine the cause-specific mortality of diseases that were considered treatable at that time, including pneumonia, bronchitis, gastritis, and duodenitis. 45 If those 43 We also examined gender-specific mortality rates, and found the results to be the same for both men and women. 44 This benefit was a bail-out measure for those who were already elderly when the National Pension Plan was enacted. The benefit was paid for disabled people aged 65 or older and for non-disabled people aged 70 years or older; it was funded by national taxes, not pension premiums. This benefit was not paid to people with other income sources, including employment-based pension benefits. Given that employment-based pensions are often provided with employment-based health insurance, the impact of this welfare benefit is likely to correlate with our measure of the impact of universal health insurance. Note that, however, the proportion of the population over 65 years of age was only 5.7%, and that over 70 was only 3.4% in Japan at that time (Census 1960). According to the Japan Statistical Year Book, the total amount of money spent on this welfare benefit was 34,607 million yen in 1961, whereas the NHI insurance benefits paid in 1961 amounted to 158,656 million yen (4.5 times as large as the welfare benefits) and total insurance benefits including employment-based insurance amounted to 347,870 million yen (10 times). Thus, the effect of this welfare transfer on the aggregate outcome is likely to be negligible. 45 At that time, hospitals could effectively treat only these short-term, acute illnesses, rather than chronic illnesses such as cancer and cardiovascular disease.

who could have been saved with appropriate treatment did not have access to health-care due to a lack of health insurance coverage, the mortality rates of these treatable diseases should have declined more in the prefectures that were more significantly affected by health insurance expansion. However, we find no statistically significant reduction in the number of deaths caused by these treatable diseases (not shown). 46 It is important to emphasize, however, that since the confidence interval for our mortality results is so large, we cannot completely rule out the possibility that our estimates include the potentially large reduction in mortality. Moreover, mortality is only one measure of health outcomes and the available data do not allow us to examine potential non-mortality health benefits from coverage expansion, such as reduced morbidity. A.4. Impact on household out-of-pocket health care expenditures Even with no improvement in health outcomes, health insurance may benefit insured individuals by reducing the risk of sudden out-of-pocket expenses, thus helping them to smooth consumption (Finkelstein and McKnight, 2008). To investigate whether, and to what extent, health insurance can reduce this risk, we need data on the distribution of out-of-pocket expenses at the individual level. However, such data are not available. Thus, in this section, we explore the effect on average out-of-pocket medical expenditures. Data pertaining to household medical out-of-pocket expenditures are taken from the National Survey of Family Income and Expenditures, which has been conducted once every 5 years since 1959. This survey is nationally representative, in that both insured and uninsured individuals are included. Each surveyed household is asked to keep track of its household budget. Therefore, data on medical expenditures consist only of out-of-pocket medical expenditures by the household, and not payments made directly from the insurance system to medical providers. In addition, medical expenditures may include the purchase of nonprescription medication at drugstores. Medical spending by households in 1959, 2 years before the achievement of universal health insurance, was 2,206 yen (in 1980 prices) per month, or 1.8% of the total household income. We examine the difference between 1959 and 1964 data to estimate the impact of health insurance on out-of-pocket expenditures, as well as the difference between 1959 and 1969 data to determine long-term effects. Specifically, we estimate the following first-difference regression: dY ¼ β0 þ β1 impact p þ β′2 dX þ εp

ðA1Þ

where X includes the same set of control variables added in rows (3) and (11) in Table 2. We use both the ratio of out-of-pocket medical expenditures to the total household expenditures, and the log of out-of-pocket medical expenditures, as dependent variables; Appendix Table A6 presents the results thereof. The estimated coefficients are small and not statistically significant. These results suggest that the growth in household out-of-pocket medical expenditures did not vary with the proportion of people recently covered by health insurance due to the introduction of universal health insurance. The finding that health insurance had almost no impact on outof-pocket medical expenditures is in stark contrast to that of studies 46 Another possibility is that the sudden increase in demand lowered the quality of health care services. Because health care utilization increased dramatically, whereas the number of physicians and nurses did not fully catch up, the expansion of health insurance coverage might have reduced the number of physicians and nurses per patient. Although we cannot directly measure the quality of medical treatment, this overcrowding effect may have lowered the overall quality of health care services.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

conducted in the United States. For example, Finkelstein and McKnight (2008) found that the introduction of Medicare in the United States resulted in a 25% decline in out-of-pocket medical expenditures. This difference may be attributable to the difference in the coinsurance

19

rate: in the case of Japan, newly covered NHI recipients still had to pay for 50% of their own health-care costs, whereas the introduction of Medicare in the United States reduced consumer costs to almost zero, with the exception of a small deductible.

Appendix Table A1 Variable Definitions and Data Sources. Variable name

Definition

Source

Admissions Inpatient days

Total number of new admissions in the calendar year. All hospitals, not including clinics. Total inpatient days (sum of days in the hospital of all patients) in the calendar year. All hospitals, not including clinics.

Outpatient visits

Total number of outpatient visits in the calendar year. All hospitals, not including clinics.

Expenditures by the NHI Number of medical claims Hospitals Clinics Age specific mortality rates

Total healthcare expenditures paid through the NHI (i.e. total healthcare expenditures excluding out-of-pocket spending). Number of claims made to the NHI by medical institutions. Number of hospitals, all kinds, as of December 31 Number of all clinics as of December 31. Total number of deaths of people in the age group divided by population of the same age group interpolated from Census. Per thousand population. Number of doctors who were working in hospitals as of December 31. Number of nurses (incl. practical nurses) who were working in hospitals as of December 31. Total number of beds in hospitals and clinics, as of December 31. Bed occupancy rate, inpatient/365/number of beds as of July 1 Population as of October 1. For years 1950, 55, 60, 65 and 70, taken from Census. Data of inter Census years are interpolated by the Statistics Bureau. Prefecture level GDP deflator in the 68SNA system with 1980 as the base year. Prefecture level GNP, deflated by prefecture GDP deflator. Local government's revenue to expenditure ratio. Sum of prefecture and municipal governments. Revenue includes transfers from the national government but excludes transfers between prefecture and municipal governments. Local government's expenditure on health and sanitation. Sum of prefecture and municipal governments.

(B) 1950– 1951:(A) 1952–1970:(B) 1950– 1951:(A) 1952–1970:(B) (I) (I) (D) (D) (E) and (F)

Physicians Nurses Beds Bed occ. rate Total population GDP deflator Real GNP per capita Fiscal rev-exp ratio Fiscal exp on health and sanitation Population by age group Population density in 1955 Per capita deaths by WWII Per capita destroyed buildings by WWII

Population by age group as of October 1. Interpolated from Census.

(D) (D) (D) (B) (E) with interpolation (G) (G) (H)

Population density of each prefecture as of October 1955

(E) with interpolation (E)

The sum of the numbers of deaths and missing divided by 1940 population

(J)

The sum of the numbers of deaths and missing divided by 1940 population

(J)

Data sources: (A) (B) (C) (D) (E) (F) (G) (H) (I) (J)

Japan Statistical Year Book, Bureau of Statistics. Hospital Report, Ministry of Health and Welfare. Annual Statistical Report of National Health Conditions, Health and Welfare Statistics Association. Survey of Medical Institutions, Ministry of Health and Welfare. Population Census, Bureau of Statistics. Vital Statistics, Ministry of Health and Welfare. Prefecture SNA in 68SNA format, available at http://www.esri.cao.go.jp/jp/sna/kenmin/68sna_s30/main.html. Annual Report on Local Public Finance Statistics, Ministry of Home Affairs. Annual Report on Social Security and Statistics, General Administrative Agency of the Cabinet. Nakamura, Takahide and Miyazaki, Masayasu, eds. Shiryo Taiheiyo Senso Higai Chosa Hokoku [Damage Survey Report of the Second World War]. Tokyo: Todai Shuppankai, 1995.

Appendix Table A2 Effect of the Coverage Expansion on the Changes in the Self-employment Ratio 1955–1960. Dependent variable: Changes in the ratio of self-employed in all employed persons from 1955 to 1960 All prefectures

Impactp defined by Eq. (2)

(1)

(2)

(3)

(4)

−0.005 [0.018]

−0.010 [0.015] 0.389*** [0.104] 46 0.19

0.001 [0.018]

−0.004 [0.150] 0.431*** [0.097] 44 0.26

Changes in Self-emp. ratio 1950–1955 Observations R2

Excl. Tokyo and Osaka

46 0.00

44 0.00

Note: Robust standard errors are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. Columns (1) and (3) report the coefficient of impactp from regressions of the changes in the ratio of self-employed in all employed people from 1955 to 1960 on impactp. Columns (2) and (4) add the changes in the self-employed ratio from 1950 to 1955 in order to control for pre-existing trends. Columns (1) and (2) use the sample of all 46 prefectures, while columns (3) and (4) repeats the same exercise without Tokyo and Osaka, which together comprised 15% of Japan's total population in 1956.

20

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Appendix Table A3 The Effect of the Health Insurance Coverage on Establishment Size. log (number of establishments with 1–4 employees)

log (number of establishments with 5–9 employees)

Log (number of all establishments)

% establishments with 1–4 employees

% establishments with 5–9 employees

A. All Prefectures λ51 −0.064 [0.087] λ54 0.034 [0.043] λ60 −0.059 [0.047] −0.043 λ63 [0.048] λ66 −0.013 [0.052] Observations 276 R-squared 0.999

0.001 [0.134] 0.046 [0.046] −0.135* [0.069] −0.115 [0.095] −0.224* [0.119] 276 0.999

0.069 [0.228] 0.029 [0.036] −0.051 [0.046] −0.097* [0.053] −0.094* [0.056] 276 0.997

−0.067 [0.118] 0.005 [0.015] 0.007 [0.015] 0.040* [0.020] 0.064** [0.027] 276 0.91

−0.012 [0.018] −0.001 [0.006] −0.006 [0.005] −0.010 [0.008] −0.020* [0.010] 276 0.988

B. Excluding Tokyo and Osaka λ51 −0.062 [0.094] λ54 0.011 [0.046] −0.017 λ60 [0.037] λ63 −0.054 [0.058] λ66 −0.005 [0.064] Observations 264 R-squared 0.997

0.063 [0.135] 0.075 [0.051] −0.049 [0.052] −0.05 [0.101] −0.081 [0.109] 264 0.998

0.129 [0.247] 0.026 [0.039] −0.016 [0.040] −0.074 [0.063] −0.032 [0.066] 264 0.992

−0.106 [0.124] −0.009 [0.013] 0.004 [0.014] 0.014 [0.013] 0.022 [0.019] 264 0.823

0.000 [0.018] 0.005 [0.006] −0.003 [0.004] −0.001 [0.007] −0.008 [0.009] 264 0.975

Note: Table A3 reports λt's, which represent the effects of a 100%-point change in the insurance coverage rate on the relative changes in the dependent variable given in the column title from 1957, the baseline year, to year t. Standard errors, estimated with clustering by prefecture, are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. Panel A use the sample of all 46 prefectures, while Panel B repeats the same exercise excluding Tokyo and Osaka, which together comprised 15% of Japan's total population in 1956, from the sample.

Appendix Table A4 Differences in Establishment-size across Industries. Average establishment size Agriculture Forestry and hunting Fishing Mining Construction Manufacturing Whole sale and retail trades Finance and real estate Transportation and other utility Service Government sector Unknown (employed) Non-employed

% workers in establishment with 5 or more employees

These three categories are assumed to be never covered by employment-based health insurance 755.2 129.2 312.9 33.6 123.5 206.8

98.1% 81.6% 90.6% 48.4% 87.1% 92.5%

39.6 53.9% Assumed to be 100% covered by employment-based insurance N/A Assumed to be never covered by employment-based health insurance

Industry-level coverage rate of employment-based insurance 11.4% 20.2% 24.0% 88.0% 37.1% 65.2% 25.5% 79.5% 87.4% 53.3% 95.2% 38.1% 8.0%

Note: Data for last column are taken from the Comprehensive Survey of the People on Health and Welfare.

Appendix Table A5 The Effect of the Coverage Expansion on Population in Census years.

λ50 λ60 λ65 λ70

log population

% population older than 55

−0.001** [0.000] 0 [0.000] 0.001 [0.001] 0.001 [0.001]

−0.001 [0.003] −0.003 [0.003] −0.007 [0.006] −0.013 [0.010]

Note: Table A5 reports λt's, which represent the effects of a 100%-point change in the insurance coverage rate on the relative changes in the dependent variable given in the column title from 1955, the baseline year, to year Standard errors, estimated with clustering by prefecture, are presented in the brackets. *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. The sample years are 1950, 1955, 1960, 1965, and 1970. The reference year is 1955.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

21

Appendix Table A6 Effect of Health Insurance Coverage on Households' Out-of-pocket Medical Expenditure.

Impactp defined by Eq. (2) Observations

Ratio of medical expenditure in household expenditure

Log(medical expenditure)

1959–1964

1959–1969

1959–1964

1959–1969

−0.002 [0.004] 46

−0.003 [0.011] 46

−0.037 [0.203] 46

−0.237 [0.481] 46

Note: Table A6 reports coefficients from estimating Eq. (A1) for the dependent variable indicated in the columns.

Appendix Fig. A1. The Relationship of the Estimate of Employer-based Health Insurance Imputed by Establishment Size and That Imputed by Industry Composition. Note: The x-axis represents employer-based health insurance imputed from industry composition, while the y-axis represents that imputed from establishment sizes. Each observation represents one of the 46 prefectures.

Appendix Fig. A2. Effect of Health Insurance Coverage on Log Population. Note: Fig. A2 plots λt's, the coefficients of impactp interacted with year dummies in Eq. (3), over year t. The graph shows the flexibly estimated pattern over time in the % changes in population in prefectures where the enforcement of universal coverage had a larger impact on the insurance coverage rate, relative to prefectures where it had a smaller impact. The two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved, respectively. Regressions on which these graphs are based include prefecture-fixed effects, region-specific year effects, and interactions between year dummies and the value of the dependent variable as of 1956. Standard errors are clustered by prefecture. The sample years are 1950, 1955, 1960, 1965, and 1970.

22

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23

Appendix Fig. A3. Effect of Health Insurance Coverage on Age-Specific Mortality Rates. Note: Fig. A3 plots λt's, the coefficients of impactp interacted with year dummies in Eq. (3), over year t. Each graph shows the flexibly estimated pattern over time in the changes in the mortality rates of the age category given in the graph title in prefectures where the enforcement of universal coverage had a larger impact on the insurance coverage rate, relative to prefectures where it had a smaller impact. Mortality rate is number of deaths per 1000 population. The two vertical lines indicate 1956, the reference year, and 1961, the year in which universal health insurance was achieved, respectively. Regressions on which these graphs are based include prefecture-fixed effects, region-specific year effects, interactions between year dummies and the value of the dependent variable as of 1956, log population and the proportion of the over 65 in the total population. Standard errors are clustered by prefecture. The sample covers the period from 1951 to 1970.

References Association of American Medical College, 2010. Physician Shortages to Worsen Without Increases in Residency Training. available online at: https://www.aamc.org/download/ 150584/data/physician_shortages_factsheet.pdf (last accessed May 31, 2011). Card, David, Dobkin, Carlos, Maestas, Nicole, 2008. The impact of nearly universal insurance coverage on health care utilization: evidence from medicare. American Economic Review 98 (5), 2242–2258. Card, David, Dobkin, Carlos, Maestas, Nicole, 2009. Does medicare save lives? Quarterly Journal of Economics 124 (2), 597–636. Cataife, Guido, Courtemanche, Charles J., 2011. Is Universal Health Care in Brazil Really Universal? NBER Working Paper, No. 17069. Chang, Simon, 2011. The effect of Taiwan's National Health Insurance on mortality of the elderly: revisited. Health Economics 21 (11), 1257–1270.

Chay, Kenneth Y., Kim, Daeho, Swaminathan, Shailender, 2010. Medicare, Hospital Utilization and Mortality: Evidence from the Program's Origins, Mimeo. Chen, Yuyu, Jin, Ginger Zhe, 2010. Does health insurance coverage lead to better health and educational outcomes? Evidence from rural China. NBER Working Paper, No. 16417. Chen, L., Yip, W., Chang, M.C., Lin, H.S., Lee, S.D., Chiu, Y.L., Lin, Y.H., 2007. The effects of Taiwan's National Health Insurance on access and health status of the elderly. Health Economics 16 (3), 223–242. Chou, S.-Y., Grossman, M., Liu, J.-T., 2011. The impact of national health insurance on birth outcomes: a natural experiment in Taiwan. NBER Working Paper, No. 16811. Currie, J., Gruber, J., 1996a. Health insurance eligibility, utilization of medical care, and child health. Quarterly Journal of Economics 111, 431–466. Currie, J., Gruber, J., 1996b. Saving babies: the efficacy and cost of recent changes in the medicaid eligibility of pregnant women. Journal of Political Economy 104, 1263–1296. Cutler, David M., 1998. Cost shifting or cost cutting? The incidence of reductions in medicare payments. Tax Policy and the Economy 12, 1–27.

A. Kondo, H. Shigeoka / Journal of Public Economics 99 (2013) 1–23 Damrongplasit, Kannika, Melnick, Glenn A., 2009. Early results From Thailand's 30 Baht health reform: something to smile about. Health Affairs 28 (3), w457–w466. Dow, W., Schmeer, K., 2003. Health insurance and child mortality in Costa Rica. Social Science & Medicine 57 (6), 975–986. Finkelstein, Amy, 2007. The aggregate effects of health insurance: evidence from the introduction of medicare. Quarterly Journal of Economics 122 (3), 1–37. Finkelstein, Amy, McKnight, Robin, 2008. What did Medicare do? The initial impact of Medicare on mortality and out of pocket medical spending. Journal of Public Economics 92 (7), 1644–1668. Finkelstein, Amy, Taubman, Sarah, Wright, Bill, Bernstein, Mira, Gruber, Jonathan, Newhouse, Joseph P., Allen, Heidi, Baicker, Katherine, The Oregon Health Study Group, 2011. The Oregon health insurance experiment: evidence from the first year. NBER Working Paper, No. 17190. Gruber, Jonathan, Hendren, Nathaniel, Townsend, Robert, 2012. Demand and reimbursement effects of healthcare reform: health care utilization and infant mortality in Thailand. NBER Working Paper, No. 17739. Hanratty, Maria, 1996. Canadian National Health Insurance and Infant Health. American Economic Review 86 (1), 276–284. Hughes, David, Leethongdee, Songkramchai, 2007. Üniversal coverage in the land of smiles: lessons from Thailand's 30 Baht health reforms. Health Affairs 26 (4), 999–1008. Ikegami, Naoki, 1991. Japanese health care: low cost through regulated fees. Health Affairs 10 (3), 87–109. Ikegami, Naoki, 1992. Japan: maintaining equity through regulated fees. Journal of Health Politics, Policy and Law 17, 689–713. Ikegami, Naoki, Campbell, J.C., 1995. Medical care in Japan. The New England Journal of Medicine 333, 1295–1299.

23

King, Gary, Gakidou, Emmanuela, Imai, Kosuke, Lakin, Jason, Ryan, T., Moore, Clayton Nall, Ravishankar, Nirmala, Vargas, Manett, Téllez-Rojo, Martha María, Hernández Ávila, Juan Eugenio, Hernandez Avila, Mauricio, Hernández Llamas, Hétor, 2009. Public policy for the poor? A randomised assessment of the Mexican Universal Health Insurance Programme. The Lancet 373 (9673), 1447–1454. Kolstad, Jonathan T., Kowalski, Amanda E., 2012. The Impact of Health Care Reform on Hospital and Preventive Care: Evidence from Massachusetts. Journal of Public Economics 96, 909–929. Madrian, Brigitte C., 1994. Employment-based health insurance and job mobility: is there evidence of job-lock? Quarterly Journal of Economics 109 (1), 27–54. Manning, Willard G., Newhouse, Joseph P., Duan, Naihua, Keeler, Emmett B., Leibowitz, Arleen, 1987. Health insurance and the demand for medical care: evidence from a randomized experiment. American Economic Review 77 (3), 251–277. Miller, Grant, Pinto, Diana M., Vera-Hernández, Marcos, 2009. High-Powered Incentives in Developing Country Health Insurance: Evidence from Colombia's Régimen Subsidiado. NBER Working Paper, No. 15456. Newhouse, Joseph, the Insurance Experiment Group, 1993. Free for All? Lessons from the RAND Health Insurance Experiment. Harvard University Press, Cambridge, MA. Shigeoka, Hitoshi, 2012. The Effect of Patient Cost-Sharing on Utilization, Health and Risk Protection, Mimeo. The Centers for Medicare and Medicaid Services, 2010. Medicare Enrollment Reports. http://www.cms.gov/MedicareEnRpts/Downloads/10All.pdf (Last accessed December 28, 2011). Yoshihara, Kenji, Wada, Masaru, 1999. History of health insurance system in Japan. (Nihon Iryo Hoken Seidoshi) Toyo Keizai Shinbum Sha (in Japanese).