Chapter 5. On Consumption Insurance Effects of the Long-term Care Insurance in Japan: Evidence from Micro Household Data

Transcription

1 Chapter 5 On Consumption Insurance Effects of the Long-term Care Insurance in Japan: Evidence from Micro Household Data Abstract Employing the micro household data of the Comprehensive Survey of the Living Conditions of the People on Health and Welfare (SLC) compiled by Ministry of Health, Labor and Welfare, this paper examines how the state of having a household member who needs long-term nursing care would result in welfare loss measured in terms of consumption, and evaluates the role of the long-term care insurance implemented in April 2000 in Japan. According to the estimation result based on the 2001 survey, when they have a disabled family member, the household consumption net of long-term care costs did not decrease so much as before the introduction of the long-term care insurance. Compared with the estimation result based on the survey conducted in 1995 and 1998, adverse effects on consumption net of long-term care costs became much weaker. These findings imply that the introduction of the nursing-care insurance system in 2000 and the moving-up of the system even before 2000 help to reduce the welfare loss associated with the state of having a disabled family member. JEL classification numbers: E21, I18 Keywords: social insurance; consumption insurance; long-term care insurance

2 1. Introduction As the Japanese population is aging, financing social security costs becomes a serious policy issue. Major growing components of social costs currently consist of benefits of health insurance and long-term care insurance. According to the projection made by the Ministry of Health, Labor and Welfare in May 2004, the ratio of long-term care insurance benefit to the National Income will increase by about 2 percentage points in the next twenty years. Cutting them is not easy because it is an in-kind benefit that provides a basic service for our life. We should be cautioned that controlling social security benefits and controlling medical and long-term care expenditure are essentially different matters. A way of shrinking social security outlays keeping the size of total spending on health care and long-term care is to narrow the coverage of social insurance. The Japanese government has raised the coinsurance rate of health insurance for workers and the elderly during the last decade. The public long-term care insurance currently covers a broad area of services, and the spending on services for those with light disability has been rapidly increasing. Since the border of long-term care services is vague, an appropriate coverage of long-term care insurance has been continuously discussed. However, if we try to solve the financial problem only by narrowing the coverage, the resulting coverage of social insurance becomes extremely low. People are then exposed to a serious risk of spending too much on health care and long-term care costs. The appropriate level of benefit should be determined by costs and benefit of insurance. Unfortunately, the social insurance s role of diversifying risks has not been closely examined in Japan. A potential role of public long-term care insurance can be identified through investigating what kind of losses a household will suffer from having a disabled or bedridden household member. There are two channels; (1) a decline of permanent income due to sacrificing the earning opportunity in order to take care of the disabled, and (2) spending on uninsured long-term care costs lowers other items of household consumption. The first channel works as follows. A person who cares a disabled family member at home might sacrifice getting a job opportunity. For example, a person who

3 lives with a bedridden parent might give up a position which might force workers to move to a different place, or, one might shorten the working hours. In any case, loss of earning opportunity for a long term would certainly decrease the permanent income. The decline of permanent income in turn lowers consumption of household. The second channel is that without having an appropriate insurance, people often have to sacrifice other items of consumption good and service, or spend their saving to pay uninsured long-term care costs. A private insurance company has a difficulty in selling an insurance plan for the long-term care services with the following two reasons. First, it would be extremely difficult to accurately calculate expenses on long-term nursing care. It would also be difficult to properly expect the future technical innovation of medical technology. Secondly, it is difficult for individuals to evaluate their own risk of needing long-term care. Individuals who underestimate the necessity would have less incentive to buy insurance. When the demand for insurance is not strong for those reasons, a private insurance company would hardly offer a plan at a reasonable price. The lack of private insurance for long-term care would be partially compensated with social welfare programs financed with general tax revenues. Before the introduction of public long-term care insurance in 2000, nursing care and social welfare services had been provided at a very low price, but the provision was rationed due to a limited budget. Since the service may not be available uniformly among those who need nursing care, the welfare programs caused distortion of resource allocation. In this paper, we aim to examine how the state of having a household member who needs long-term care will lead to welfare loss in terms of consumption. Our empirical framework has a similar spirit with a literature of consumption insurance like Cochrane (1991), Gertler and Gruber (2002), and Townsend (1994), which examined the effect of adverse health shock on consumption. Based on the Panel Study of Income Dynamics, Cochrane (1991) found that a loss of work hours due to a health problem reduced food consumption. If a household head becomes absent at workplace more than 100 days (probability of this event is 7.2 percent in the sample), consumption declines by 11 to 14 percent. From the panel data on families in rural areas in India, Townsend (1994) found that the percentage of the year when an adult male is sick has no impact on consumption. Gertler and Gruber (2002) examined panel data conducted

4 in Indonesia, and found that impairment of activities of daily living (ADL) led to a significant economic loss. The impairment of basic ADL (2 percent in the sample) decreased consumption by 59 percent. The problem of intermediate ADL (24 percent in the sample) decreased consumption by 14 percent. Our contribution here is to examine whether the public long-term care insurance implemented in 2000 helps to reduce the welfare loss in terms of consumption. Since Japan is so far one of few countries that have introduced a public long-term care insurance, evidence from Japan may be informative for other countries which scrutinize the possibility of this kind of social insurance. The data set we use is the micro data of the Comprehensive Survey of the Living Conditions of the People on Health and Welfare (Kokumin Seikatsu Kiso Chosa in Japanese, hereafter SLC), which is a nationally representative household survey. The SLC contains detailed information about long-term care needs and economic status every three years. Since the public long-term care insurance was introduced in April 2000, we use the data set before and after that event (1998 and 2001). In addition, we compare the estimation result based on the 1998 and 2001 SLC with that based on the 1995 SLC reported in Iwamoto, Kohara, and Saito (1995). A social insurance is expected to absorb the risk of long-term care needs. On the other hand, since the public long-term care insurance does not work as an income support, the loss of earnings opportunity may not be covered. In our empirical framework, this implies that the effect of long-term care need on consumption is not statistically significant after controlling for permanent income. If the effect on consumption is significant before the introduction of public long term care insurance but insignificant after that event, social insurance is found to cover an important risk that were not covered by the private sector. The remainder of this paper is organized as follows. Section 2 describes our dataset, while Section 3 reports estimation results and discusses available implications. Section 4 offers concluding remarks. The appendix conducts a similar analysis for a role of public health insurance

5 2. Data Descriptions and Basic Statistics 2.1 Data Characteristics We use the micro household data of the Comprehensive Survey of the Living Conditions of the People on Health and Welfare (hereafter, SLC) compiled by Ministry of Health, Labor, and Welfare. The SLC conducts a large-scale survey covering more than 30,000 households every three years. We employ the large-scale surveys conducted in 1998 and In addition, the estimation result based on these surveys will be compared with that based on the 1995 survey, which is reported in Iwamoto, Kohara, and Saito (2001). The SLC suits well to the purpose of our empirical investigation. First, this large-scale household survey reports several important dimensions of household characteristics concerning family structure, job status, income, financial assets, real estate, and health conditions of every family member. In regard to health conditions, the SLC surveys in detail whether a family member needs long-term care as a result of physical disability, and if it is a case, which kind of care is necessary for a disabled member. Second, given such a large-scale household survey, we can keep a reasonable number of households whose member is in need of long-term care although the occurrence probability of having severely disabled family members is extremely low among the entire households. In this regard, unlike micro datasets which survey only households with disabled members, a dataset constructed from the SLC allows us to explore differences in household behavior between a family with disabled members and without ones. Third, the choice of the 1998 and 2001 surveys is fairly convenient in evaluating possible welfare impacts of the long-term care insurance which was introduced in April, For a privacy reason, we are not allowed to have access to any detailed information concerning examinees places of residence. All we can obtain concerns in which city/town/village a surveyed household lived. Nevertheless, this information about residential places is quite useful for the following reason. There have been substantial regional differences in the extent that a municipal government (city, town, or village) provides long-term assistance for the disabled elderly, even after the central government implemented uniformly public assistance for families with members in

6 need of long-term care under the long-term care insurance law in Some municipal governments offered generous assistance for the disabled, while other local governments did not. In particular, the availability of social welfare facilities differed substantially from one city to another. We will exploit the above regional difference in estimating welfare impacts of the state of having a disabled household member. The SLC reports the annual household income of the previous year, the monthly expenditure on household consumption of May of the surveyed year, and the health condition of family members as of the first Thursday of the surveyed year. For example, the 2001 SLC reports the annual income of 2000, the monthly consumption expenditure of May, 2001, and the health condition of family members as of June 7, The 1998 SLC identifies a particular family member as a member in need of long-term care, when he/she needs nursing assistance in the following six activities: (1) washing face and teeth, (2) changing clothes, (3) eating, (4) discharging, (5) bathing, and (6) walking. Thus, the degree of need of long-term care may be measured in terms of how many activities need nursing assistance. In addition, the 1998 SLC classifies a state of a bedridden member according to levels from occasionally, sometimes through in most of time, completely. We define as a bedridden member, a member belonging to either of the two severest states. Using the 1995 SLC which follows the same definition as in the 1998 SLC, Iwamoto, Kohara, and Saito (2004) classify as a member in need of long-term care, one who needs nursing assistance in four activities or more. The 2001 SLC identifies a particular family member as a member in need of long-term care according to the same criterion as in the 1998 SLC, that is, from one disabled activity through six disabled activities. In addition, the 2001 SLC compiles the degree of need of long-term care that is authorized under the long-term care insurance law; a care manager judges a state of nursing care ranging from Level 1 (lightest) through Level 5 (most serious) by visiting an applicant for the long-term care insurance payoff. Under this classification, a member classified as Level 5 is almost equivalent to a bedridden member. We construct a dummy variable for a long-term care state which takes one if a family member older than 14 years old needs nursing assistance, or is bedridden,

7 otherwise zero. As discussed in the introduction, we evaluate the welfare impact of having a family member in need of long-term care in terms of household consumption. When a disability shock hits on one of family members, extra expenses incurred in caring the disabled may be financed from household savings, or may be compensated through giving up other consumption expenses. Some family members, in particular a spouse of a household head, may quit a current job with high income, and switch to another job with less commitment thereby sacrificing life-time income and allocating time to the care of a disabled member at home. The combined effects of these compensating activities may be eventually reflected in changes in household consumption net of long-term care costs. The SLC reports the uninsured expense on long-term nursing care including extra clothes, beds, caring tools, charged care service, and uninsured medical expenses. We thus use as a household welfare measure, the household consumption net of the expenditure on these long-term care costs. We have several remarks on the usage of consumption as a welfare measure in the context of long-term care. First, consumption expenditures reflect consumption service available from market activities, but not household activities. For example, as a result of having a disabled member, a family may more often eat at home than outside, and reduce expenditures on consumption. A decrease in food consumption for this reason may not be necessarily interpreted as welfare deterioration. Second, as a result of allocating more time to long-term care at home, a family may reduce time-consuming consumption activities such as travels and sports. Such a decline in consumption may not exactly capture an increase in disutility due to giving up leisure time. Third, household consumption may serve as a better welfare measure than household income from the perspective of a family who is forced to give up consumption by a liquidity constraint. 2.2 Basic Statistics We use a sample consisting of households whose household head (the highest income earner) is a full-time worker. A major reason for this sample selection is that we are mainly interested in a case where healthy (maybe, and younger) household members

8 support an elderly member in need of nursing care. 1 As a consequence, the number of observations is 10,498 or less for the 1998 SLC, and 13,380 or less for the 2001 SLC. Table 1 and Table 2 report basic statistics for our sample consisting of the 1998 and 2001 SLC. In regard to the 1998 SLC, the number of households with members in need of nursing care is 263 observations (2.51% of the total observations 10,498) for at least one disabled activity, 234 observations (2.23%) for at least two disabled activities, 197 observations (1.88%) for at least three disabled activities, 156 observations (1.49%) for at least four disabled activities, 124 observations (1.18%) for at least five disabled activities, and 99 observations (0.94%) for six disabled activities. In addition, the number of households with bedridden members is 724 observations (0.69%). As to the 2001 SLC, on the other hand, the number of households with members in need of nursing care is 959 observations (7.20% of the total observations 13,321) for at least one disabled activity, 777 observations (5.83%) for at least two disabled activities, 614 observations (4.61%) for at least three disabled activities, 469 observations (3.52%) for at least four disabled activities, 349 observations (2.62%) for at least five disabled activities, and 225 observations (1.69%) for six disabled activities. Based on the authorized degree of long-term care need under the long-term care insurance law, the number of households with members in need of nursing care is 1,147 observations (8.57% of the total observations 13,380) for Level 1 or higher, 850 observations (6.35%) for Level 2 or higher, 553 observations (4.13%) for Level 3 or higher, 347 observations (2.59%) for Level 4 or higher, and 175 observations (1.31%) for Level 5 (equivalent to bedridden members). Given the above difference in the share of families with disabled members between the 1998 and 2001 surveys, it may be rather difficult to compare the estimation of the 1998 survey with that of the 2001 survey. One possible reason for this substantial difference between the two surveyed years is that thanks to the introduction of the long-term care insurance in 2000, more general recognition might had been paid to the situation of disabled members within households, and consequently the 2001 survey might have recognized disabled members much more broadly than the 1998 survey. However, even in comparison between the 1995 and 1998 surveys, the former 1 Our investigation may underestimate welfare impacts of the state of having a disabled member, because our sample excludes a much more serious case in which a household head becomes in need of nursing care

9 reports smaller shares of households with disabled members than the latter. More concretely, the share of household members of four disables activities or more is only 0.81% in the 1995 SLC, but it is 1.49% in the 1998 SLC. The share of households with bedridden members is 0.38% in the 1995, but it is 0.69% in the 1998 SLC. More general recognition might have been paid as to the state of the disabled even before the long-term care insurance was implemented in 2000, or the number of members in need of nursing care might have indeed increased as the society is more and more ageing. For the above reason, we will make comparison among estimation results of 1995, 1998, and 2000 only in terms of the case of bedridden household members (the most severe case), although there is still a difference in the household share (0.38% in 1995, 0.69% in 1998, and 1.31% in 2001). In both the 1998 SLC and the 2001 SLC, the share of a family in which an adult child supports his/her disabled parents is the largest among the households with disabled members. More concretely, in the 1998 (2001) SLC, the share of households which support bedridden parents is 54.76% (55.33%), while the share of households which support bedridden spouse is 34.52% (30.96%). Table 2 compares the average of representative household characteristics such as income and consumption net of long-term care costs between household with disabled members and without ones. According to Table 2, the average income (or consumption) of the former households is not necessarily inferior to that of the latter. As long as the average level is concerned, there is no direct evidence for adverse impacts of the state of having a disabled member in terms of household income and consumption. Section 3 explores in detail how the presence of a disabled member would have welfare impacts after controlling for possible effects of household characteristics on income and consumption. 2.3 Data on Social Welfare Facilities As discussed above, the data concerning examinees residential places provided by the SLC micro data may be informative, because the availability of social welfare facilities for the disabled elderly differs from one city/town/village to another. With such facilities high available, the impact of having disabled members may be mitigated to some extent

10 As developed in the next section, we exploit regional data concerning social welfare facilities to empirically identify the above impact on income and consumption. More concretely, we construct as a proxy for the facility availability, the ratio of the number of social welfare facilities for the elderly relative to the population of those 65 years old or older city by city (town by town, or village by village). For this purpose, we collect the number of social welfare facilities for the year 1998 from the Survey of Social Welfare Facilities conducted by Ministry of Health, Labor, and Welfare. In this survey, a social welfare facility for the elderly is defined as an institution established under the Elderly Welfare Law. The average of the availability ratio is 0.14% with 0.07% standard error. For the year 2001, we collect the number of social welfare facilities from the Survey of Nursing-Care Service Facilities and Establishments again compiled by Ministry of Health, Labor, and Welfare. 2 The average ratio of these facilities relative to the population of 65 years old or older is 0.20% with 0.09% standard error. 3. Statistical Specification and Estimation Results 3.1 Statistical Specification Because all we obtain is not panel data, but cross-sectional data at a particular point of time (1998 and 2001), we can estimate not a structural form based on dynamic optimization, but a reduced-form specification to evaluate the impact of having disabled members within a family. We construct two-stage specification for this purpose. First, we estimate household income as a function of household characteristics as fixed effects, including a dummy variable associated with disabled members, and use its predicted household income as a proxy for permanent income of a household in the second stage. This first stage estimation allows us to identify impacts of having disabled members on household permanent income. By the second stage estimation, we specify a household 2 This survey covers welfare nursing-care facilities for the elderly established under the Elderly Welfare Law, medical nursing-care facilities for the elderly established with permission from a prefecture governor, and medical facilities for recuperation and nursing-care for the elderly established under the Medical Law

11 consumption net of uninsured expenditures on long-term care as a function of household characteristics including the household permanent income predicted by the first stage estimation, and a dummy variable of the presence of disabled members. There are two channels, indirect and direct, through which the state of having disabled members would yield welfare impacts in terms of consumption net of nursing care costs. By an indirect channel, a disabled state may reduce household permanent income, thereby lowering consumption. By a direct effect, on the other hand, given permanent income, consumption may decline by the state of having disabled members. One difficulty with the estimation of the above two stage specification is that the 1998 (2001) SLC reports health conditions of household members as of June of 1998 (2001), while it records the household income of the pervious year (1997 or 2000). To deal with the above timing issue, we use not only a dummy variable of currently having disabled members, but also a dummy variable of the state of having had disabled members since one year before. As the first stage estimation, a household income function is specified as: i J ln y = α x + βd, (1) j=1 j ij i where y i and x ij denote the total income and characteristics of household i. A set of variables of household characteristics { x ij } includes the number of household members in logarithm, the age of a household head and its square, the sex of a household head, married or single, the number of children relative to the number of household members, the number of dependent parents relative to the number of household members, the number of income earners, the scale of a firm for which a household head works, and a dummy variable associated with residential location. In addition, d i represents a dummy variable associated with the presence of disabled members within household i. Thus, a coefficient on d i (β ) represents the (possibly negative) effect of the state of having disabled members on permanent income. As discussed in the previous section, we construct a proxy for the availability

12 of social welfare institutions for household i who lives in a particular city/town/village, denoted by z i. In addition to the above specification, we thus adopt the following specification: J ln yi = α j xij + βd i + θd i zi, (2) j=1 A positive coefficient on the product of the facility availability variable with a dummy variable of the presence of disabled members (θ ) represents the extent that the facility availability would mitigate the impact of having disabled members on household income. As the second stage estimation, on the other hand, a function of household consumption function is specified as: J e γ (ln yi ) + λ j xij + j=1 ln ci = μd i, (3) where y ) e (ln i is the household income predicted by the first stage estimation, while c i denotes household consumption net of nursing care costs. Unlike in the first stage estimation, the SLC records the monthly household consumption of May of 1998 or 2001, and there is not any serous difference in the timing of recording health conditions (in the early June) and household consumption. A set of household characteristics adds a dummy variable of children attending school to control expenditures on education. In equation (3), a coefficient on d i (μ ) represents the direct effect of the state of having disabled members on household consumption. A possible dynamic effect such as adjustment costs or habit formation may be captured by a lower coefficient on the proxy for permanent income (γ ). Like in household income functions, we also adopt the following specification:

13 J e γ (ln yi ) + λ j xij + μd i + j=1 ln ci = ηd i zi, (4) Again, a positive coefficient on the product of the facility availability variable with a dummy variable of the presence of disabled members (η ) represents the extent that the facility availability would mitigate the impact of having disabled members on household consumption. 3.2 Estimation Results Tables 3 through 12 report the estimation results of income and consumption functions based on equations (1) through (4) using the 1998 and 2001 SLC datasets. Using the 1998 SLC, Table 3 reports the estimation result of income equation (1) and consumption equation (3), while Table 4 reports the estimation result with consideration for the facility availability based on income equation (2) and consumption equation (4). Based on the same specification, Tables 5 and 6 reports the case in which family members are disabled for at least one year. As mentioned in the previous section, the 2001 uses two definitions to classify disability conditions. The first definition is the same as that of the 1998 (determined according to the number of disabled activities), while the second is based on the degree of disability authorized under the long-term care insurance law, in which Level 5 is equivalent to bedridden. Tables 11 and 12 report the estimation results of income and consumption equations based on the first definition. 3 Tables 7 through 10 report the estimation results based on the second definition including the case in which family members are disabled for at least one year. We first discuss the estimation results of the 1998 SLC. As documented in Tables 3 and 5, in terms of household income functions, the state of having disabled members does not yield any significant impact on household income. However, once the facility availability is taken into consideration, the estimation results change in an interesting direction (see Tables 4 and 6). That is, a coefficient on a dummy variable of having disabled members is significantly negative in some cases, and the facility 3 Because the estimation results reported in Tables 11 and 12 are similar to those in Tables 7 and 8, we will not discuss the former results in this paper

14 availability mitigates the negative income impact of having disabled members. On the other hand, the state of having disabled members has significant negative effects on household consumption. For example, the state of having members with three activities or more would reduce household consumption by 11.9%, while the presence of bedridden members would lower household consumption by 16.2%. If we consider only the case in which family members are disabled for at least one year, then a coefficient on having disabled members is estimated to be smaller with less significance. This may take place due to smaller observations of households with disabled members in the latter case. Unlike in the case of income functions, the consideration of the facility availability makes these significant effects on consumption disappear (see Tables 4 and 6). The estimated marginal propensity to consume out of permanent income, which is captured byγ, is around more than 60% in all cases. The estimation results of the 2001 SLC contrast with those of the 1998 SLC in that the state of having disabled members has significantly negative impacts on not household consumption, but household income. As reported in Tables 7 and 9, in the case of Level 4 or lower, the state of having disabled members would result in 12% through 14% declines in household income, while the state of having bedridden members would reduce household income by more than 16%. Once the facility availability is considered, however, the impact of having disabled members would yield somewhat mixed effects. That is, as shown in Tables 8 and 10, the direct negative effect of having disabled members becomes weaker, while the facility availability not mitigates, but strengthen the negative impact. This result may be caused by a multi-colinearity between the facility availability and the state of having disabled members. On the other hand, the state of having disabled members has little direct impact on household consumption. The consideration of the facility availability does not change the pattern in effects on household consumption at all. Consequently, the state of having disabled members reduces household consumption only through the indirect channel. Given that a coefficient on the predicted household income is estimated to be around less than 50%, the state of having disabled members would reduce household consumption by the order of a half of the estimated impact on household income. More precisely, the state with longer than one year old bedridden members would reduce

15 household income by 16.61%, while the marginal propensity to consume out of the predicted household income is 45.61%. Accordingly, household consumption would decrease by 7.58% as a result of the indirect effect. 3.3 Discussions As discussed above, the state of having disabled members has significantly negative impacts on household welfare in terms of a decrease in household consumption in both the 1998 SLC and the 2001 SLC, but the pattern differs between the two datasets. In the 1998 SLC, the state of having disabled members has negative impacts on consumption, not on income. In the 2001 SLC, on the other hand, the state of having disabled members has negative impacts on income, but on consumption. According to Iwamoto, Kohara, and Saito (2001) that use the 1995 SLC as a main dataset, the state of having disabled members including bedridden members has significantly negative impacts on both household income and consumption. More concretely, the state of having bedridden members would reduce household income by 15.34%, and household consumption by 33.80%. Given that the marginal propensity to consume out of the predicted household income is estimated to be 22.99% (much smaller than in the 1998 and 2001 data), household consumption would decline by 37.33% as a result of the combined effect through the indirect and direct channels. Given large amounts of idiosyncratic nature of household characteristics among the relatively small number of households with disabled members, it may be rather difficult to derive a general conclusion from the above difference in the pattern in negative impacts of the state of having disabled members on household income and consumption. It is, however, possible to conclude that the consequent effect of the state of having disabled members on household welfare measured in terms of consumption has been weakened substantially over time. As mentioned above, in the case of bedridden members, household consumption would decline by 33.80% as a result of both indirect and direct channels in the year On the other hand consumption would decrease by 16.18% through only the direct channel in the year 1998, and by 7.58% by only the indirect channel in the year The pattern in diminished welfare impacts of the state of having disabled

16 members can be justified by the introduction of the long-term care insurance law in In addition, it is consistent with the fact that local governments put a long-term care system into effect by moving it forward even before The finding that the facility availability would mitigate the negative impact on income in the 1998 SLC is also supportive for the latter respect. These empirical findings imply that the introduction of the long-term insurance system and the moving-up of the system help to reduce the welfare loss associated with having a disabled family member. 4. Concluding Remarks Employing the micro household data of the Comprehensive Survey of the Living Conditions of the People on Health and Welfare (SLC) compiled by Ministry of Health, Labor and Welfare, this paper examines how the state of having a household member who needs long-term care would result in welfare loss measured in terms of consumption, and evaluates the role of the long-term care insurance introduced in April 2000 in Japan. According to the estimation result based on the 2001 survey, when they have a disabled family member, the household consumption net of long-term care costs did not decrease so much as before the introduction of the long-term care insurance. Compared with the estimation result based on the survey conducted in 1995 and 1998, adverse effects on consumption net of long-term care costs became much weaker. These findings imply that the introduction of the nursing-care insurance system and the moving-up of the system help to reduce the welfare loss associated with having a disabled family member

17 References Cochrane, J. H., 1991, A Simple Test of Consumption Insurance, Journal of Political Economy, Vol. 99, No , 1995, Time-Consistent Health Insurance, Journal of Political Economy, Vol. 103, No. 3. Cutler, D., 1993, Why Doesn t the Market Fully Insure Long-Term Care? NBER Working Paper No Currie, J., and B. C. Madrian, 1999, Health, Health Insurance and the Labor Market, in O. C. Ashenfelter and D. Card eds., Handbook of Labor Economics, Volume 3C, Elsevier. Gertler, P., and J. Gruber, 2002, Insuring Consumption Against Illness, American Economic Review, Vol. 92, No. 1. Iwamoto, Y., M. Kohara, and M. Saito, On the Welfare Loss of the State of Having Members in Need of Long-term care, Kikan Shakai Hosho Kenkyu Vol. 36, No (in Japanese) Pauly, M. V., 1990, The Rational Nonpurchase of Long-Term-Care Insurance, Journal of Political Economy, Vol. 98, No. 1. Saito, M., 1999, Dynamic Allocation and Pricing in Incomplete Markets: A Survey, Bank of Japan Monetary and Economic Studies, Vol. 17, No. 1. Strauss, J., and D. Thomas, 1998, Health, Nutrition, and Economic Development, Journal of Economic Literature, Vol. 36, No. 2. Townsend, R. M., 1994, Risk and Insurance in Village India, Econometrica, Vol. 62, No

18 Table1. Descriptive Statistics Panel A. Estimation for 1998 Data Benchmark Case (10498observations) The case including for Area Differences in LTC institutions (9825observations) Mean Standard Standard Min Max Mean deviation deviation Min Max log of family consumption log of family disposable income Number of family members Head age Head male Head married Number of kids Out of family (students) Number of Parents Out of family (hospital) Head firm scale:med Head firm scale:large City scale: metropolitan City scale: large city City scale: med city City scale: city Head age squared Number of workers LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Bedridden LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Bedridden LTC: Level1 * Ratio of LTC institutions LTC: Level2 * Ratio of LTC institutions LTC: Level3 * Ratio of LTC institutions LTC: Level4 * Ratio of LTC institutions LTC: Level5 * Ratio of LTC institutions LTC: Level6 * Ratio of LTC institutions LTC: Bedridden * Ratio of LTC LTC over 1year: Level1 * Ratio of LTC institutions LTC over 1year: Level2 * Ratio of LTC institutions LTC over 1year: Level3 * Ratio of LTC institutions LTC over 1year: Level4 * Ratio of LTC institutions LTC over 1year: Level5 * Ratio of LTC institutions LTC over 1year: Level6 * Ratio of LTC institutions LTC over 1year: Bedridden * Ratio of LTC institutions family monthly consumption (yen) family annual income (10thousand yen) Ratio of LTC institutions

19 Panel B. Descriptive Statistics for 2001 Data Specifying LTC needs over 1 year Benchmark case (13380 obs) The case including Area differences in The case including Area differences in Benchmark case (13363 obs) LTC institutions (12964 obs) LTC institutions (12948 obs) Mean Standard Standard Min Max Mean deviation deviation Min Max Mean Standard deviation Min Max Mean Standard deviation Min Max log of family consumption log of family disposable income Number of family members Head age Head male Head married Number of kids Out of family (students) Number of Parents Out of family (hospital) Head firm scale:med Head firm scale:large City scale: metropolitan City scale: large city City scale: med city City scale: city Head age squared Number of workers LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC Needs over 1year: Level LTC: Level0 * Ratio of LTC institutions LTC: Level1 * Ratio of LTC institutions LTC: Level2 * Ratio of LTC institutions LTC: Level3 * Ratio of LTC institutions LTC: Level4 * Ratio of LTC institutions LTC: Level5 * Ratio of LTC institutions LTC over 1year: Level0 * Ratio of LTC institutions LTC over 1year: Level1 * Ratio of LTC institutions LTC over 1year: Level2 * Ratio of LTC institutions LTC over 1year: Level3 * Ratio of LTC institutions LTC over 1year: Level4 * Ratio of LTC institutions LTC over 1year: Level5 * Ratio of LTC institutions family monthly consumption (yen) family annual income (10thousand yen) Ratio of LTC institutions

20 Continued: Table1-Panel B Using Disability Levels closer to the 1998 ADL levels Benchmark case (13321 obs) The case including Area differences in LTC institutions (12906 obs) Mean Standard deviation Min Max Mean Standard deviation Min Max log of family consumption log of family disposable income Number of family members Head age Head male Head married Number of kids Out of family (students) Number of Parents Out of family (hospital) Head firm scale:med Head firm scale:large City scale: metropolitan City scale: large city City scale: med city City scale: city Head age squared Number of workers Long-term Care Needs: Level Long-term Care Needs: Level Long-term Care Needs: Level Long-term Care Needs: Level Long-term Care Needs: Level Long-term Care Needs: Level Long-term Care: Level1 * long-term care institutions Long-term Care: Level2 * long-term care institutions Long-term Care: Level3 * long-term care institutions Long-term Care: Level4 * long-term care institutions Long-term Care: Level5 * long-term care institutions Long-term Care: Level6 * long-term care institutions family monthly consumption (yen) family annual income (10thousand yen) Long-term care institutions

21

22 Table Results: Benchmark Case Panel A. Long-term Care Needs: Disabled Level 1 (Shogai 1 ) Dependent log of family consumption excluding expenditure Dependent Variable: log of family disposable income LTC Needs: Level * LTC Needs: Level (0.0470) (0.0456) Predicted Income *** Number of family members *** (0.0366) (0.0290) Number of family members Head age *** (0.0380) (0.0025) Head age Head male *** (0.0004) (0.0226) Head male *** Head married *** (0.0205) (0.0271) Head married *** Number of kids *** (0.0232) (0.0543) Number of kids *** Number of Parents (0.0531) (0.0800) Number of Parents Out of family (students) *** (0.0788) (0.0509) Out of family (students) *** Out of family (hospital) * (0.0572) (0.1106) Out of family (hospital) ** Head firm size:middle *** (0.0883) (0.1086) Head firm size:middle Head firm size:big *** (0.1233) (0.1253) Head firm size:big City size: metropolitan *** (0.2591) (0.0244) City size: metropolitan *** City size: big city *** (0.0195) (0.0194) City size: big city *** City size: middle city *** (0.0182) (0.0210) City size: middle city *** City size: city (0.0188) (0.0328) City size: city Head age squared *** (0.0268) (0.0000) Constant *** Number of workers *** (0.1759) (0.0087) Constant *** (0.0879) F-stat for all the coef. = F-stat for all the coef. = Test of homoskedasticity R-squared

23 Panel B. Long-term Care Needs: Disabled Level 2 (Shogai 2 ) Dependent log of family consumption excluding expenditure Dependent Variable: log of family disposable income LTC Needs: Level ** LTC Needs: Level (0.0492) (0.0482) Predicted Income *** Number of family members *** (0.0366) (0.0290) Number of family members Head age *** (0.0381) (0.0025) Head age Head male *** (0.0004) (0.0226) Head male *** Head married *** (0.0205) (0.0271) Head married *** Number of kids *** (0.0232) (0.0543) Number of kids *** Number of Parents (0.0531) (0.0798) Number of Parents Out of family (students) *** (0.0783) (0.0509) Out of family (students) *** Out of family (hospital) * (0.0572) (0.1107) Out of family (hospital) ** Head firm size:middle *** (0.0882) (0.1085) Head firm size:middle Head firm size:big *** (0.1233) (0.1254) Head firm size:big City size: metropolitan *** (0.2591) (0.0244) City size: metropolitan *** City size: big city *** (0.0195) (0.0194) City size: big city *** City size: middle city *** (0.0181) (0.0210) City size: middle city *** City size: city (0.0188) (0.0328) City size: city Head age squared *** (0.0268) (0.0000) Constant *** Number of workers *** (0.1760) (0.0087) Constant *** (0.0879) F-stat for all the coef. = F-stat for all the coef. = Test of homoskedasticity R-squared

24 Panel C. Long-term Care Needs: Disabled Level 3 (Shogai 3 ) Dependent log of family consumption excluding expenditure Dependent Variable: log of family disposable income LTC Needs: Level ** LTC Needs: Level (0.0545) (0.0511) Predicted Income *** Number of family members *** (0.0366) (0.0290) Number of family members Head age *** (0.0380) (0.0025) Head age Head male *** (0.0004) (0.0226) Head male *** Head married *** (0.0205) (0.0271) Head married *** Number of kids *** (0.0232) (0.0543) Number of kids *** Out of family (students) *** (0.0531) (0.0509) Out of family (students) *** Number of Parents (0.0572) (0.0802) Number of Parents Out of family (hospital) * (0.0781) (0.1106) Out of family (hospital) ** Head firm size:middle *** (0.0881) (0.1088) Head firm size:middle Head firm size:big *** (0.1233) (0.1252) Head firm size:big City size: metropolitan *** (0.2592) (0.0244) City size: metropolitan *** City size: big city *** (0.0195) (0.0194) City size: big city *** City size: middle city *** (0.0181) (0.0210) City size: middle city *** City size: city (0.0188) (0.0328) City size: city Head age squared *** (0.0268) (0.0000) Constant *** Number of workers *** (0.1757) (0.0087) Constant *** (0.0879) F-stat for all the coef. = F-stat for all the coef. = Test of homoskedasticity R-squared