"What Have We Learned from the Panel Data of the Elderly?: For better life and health" (Summary)

In his opening remarks, Dr. Masahisa Fujita (President and Chief Research Officer, RIETI; Professor, Konan University; Adjunct Professor, Institute of Economic Research, Kyoto University) introduced the theme of the symposium and explained that RIETI had launched a large-scale, world-standard panel survey of middle-aged and elderly people, the Japanese Study of Aging and Retirement (JSTAR).

Dr. Arie Kapteyn (Director, RAND Labor and Population) mentioned that one of the driving forces behind the movement of having global comparative studies on aging was the Health and Retirement Study (HRS) started by the National Institute of Aging (NIA) and that the underlying purpose of the comparative studies is to learn from differences across countries in policies and their effect on outcomes. He added that RAND has been working on projects for harmonizing data and making them more accessible.

A special message from Dr. Masaru Yoshitomi (Special Adviser, RIETI), the founder of JSTAR, was delivered by his daughter. In 2004, Japan was the only industrialized country which had not yet engaged in constructing longitudinal, multidisciplinary and internationally comparable survey data necessary for analyzing and evaluating social security policies. This was the motivation behind the launch of the JSTAR project in 2005 whose remarkable success is owed to the invaluable assistance of enthusiastic participants. It is hoped that the efforts to collect vital data will continue in the future and provide an important public service both in the scientific domain and policymaking arena.

Dr. John W.R. Phillips (Behavioral and Social Research Program, National Institute on Aging (NIA)) gave a lecture on the rationale of the development and support of the HRS by NIA.

The mission of the NIA is to improve the health and well-being of older Americans by conducting high-quality research (aging processes, age-related disease, and special problems and needs of the aged) and disseminating information to interested groups so that the data can be useful for a wide array of key policy questions (Social Security, Medicare, etc.). The NIA Strategic Plan of 2007 also supports the development of longitudinal studies, data archiving and data sharing as well as the development of internationally harmonized social and behavioral data on aging to foster cross-national research.

The research model followed by the NIA is to gather considerable input from the research community, thus receiving expert advice on the direction of its scientific programs. The decision to go ahead with the Health and Retirement Study was similarly made through recommendations at numerous workshops and expert panels, such as the 1987 Ad Hoc Advisory Panel on Data Collection Priorities, as well as regular reviews. The consistent themes that came out of all of the workshops were the importance of having longitudinal, multidisciplinary data and consideration for biomarkers and administrative linkages, among others.

The NIA promotes and develops data by making use of mechanisms that feature interaction between the federal government, the data collectors, and third-party experts, reaching across to other federal agencies for administrative data linkages as well as finances. It has also started initiatives to ensure that the data is well used and harmonized through research networks and distribution mechanisms. Throughout the grant process of the NIA, there is continuous review and monitoring by third-party experts for the development of the HRS. The partnerships made with other federal agencies, such as the U.S. Social Security Administration and the Center for Medicare and Medicaid Services, have provided opportunities for doing unique research with administrative data.

Based upon the successes in cross-national research (e.g. Gruber/Wise), greater comparability of the different aging surveys for the purpose of generating more cross-national research is encouraged. Planning grants are offered for international data collection and to help create more harmonized measures and develop the studies. RAND, through support from NIA, has two programs that work to effectively distribute the HRS data: the easy-to-use RAND HRS dataset and the RAND Survey Meta Data Repository as a resource for cross-national research. Both are free to the public online.

In sum, the HRS was a result of the NIA soliciting a longitudinal aging study with significant feedback from the research community. The distribution strategy has been successful in increasing the number of scholarly users. It has also been a significant contributor to science and policy, covering many topical areas across several disciplines. For example, the HRS informed studies on the expanded Medicare Program on prescription drug benefits, the impact of the financial crisis on stocks and retirement, and the impacts on cognitive functioning after severe sepsis. The HRS is used not only in academics but also extensively throughout the federal government for policymaking. In this sense, the HRS is a public good for conducting innovative multidisciplinary research.

Dr. Phillips mentioned the Economics of Aging project which has looked at issues of late life work, retirement, and social security, as well as health care/insurance and intergenerational transfer. Regarding how the HRS data can contribute to local governments, he explained that there is potential to break down samples by individual geography, although it has not been a major focus. Relating JSTAR to HRS, Dr. Phillips commended JSTAR and predicted that JSTAR would grow and develop even more rich data.

Session Chair Dr. Hidehiko Ichimura (Faculty Fellow, RIETI; Professor of Economics at the Graduate School of Economics and the Graduate School of Public Policy, the University of Tokyo) gave the first presentation, which was on the objectives of JSTAR.

For many developed and developing countries alike, the increasing aging population is a common issue. Japan is unfortunately the leader in this respect, with an old age dependency ratio of 35% and expected to rise steadily. The main concern of such a high dependency ratio is the shortage of labor supply and deterioration of the balance on the pay-as-you-go pension system. This has led to the elderly remaining in the labor market longer, and this experience in Japan may contribute some knowledge for other countries.

Concrete numbers show, for example, that for elderly men, while the proportion has decreased compared to 1980, Japan's numbers are still higher than other G7 countries. For Japanese women (over 60) also, many remain in the labor force longer than in the G7 countries, but not Sweden. There is a trend across cohorts that men's labor force participation is declining, but for women this is increasing, and it is driving the age issue. In order to consider pension, employment, health and nursing care policies, comprehensive data collection efforts that commenced in the U.S. were propagated across the world. JSTAR is a recent addition to this family and harmonization efforts are now being led by RAND for cross-country comparisons to be made. There are panel studies all around the world, including in the U.S., Indonesia, Mexico, Europe, Korea, Japan, China, Ireland, Thailand, India and Brazil.

As in the U.S. where the HRS informs policy debates on social security reform, JSTAR hopes to play that role in Japan. The first wave in 2007 was based on a stratified random sample of five municipalities; the second wave in 2009 added two new locations; and the third wave in 2011 plans to add three more locations. This sampling design based on municipalities has the advantage of having many individuals with a uniform socioeconomic environment, namely a city. Many of the questions relate to how individuals make decisions given the environment and how the environment affects individual decisions. The sampling allows for a more flexible separating of the effect of the environment and the differences across individuals such as income level, education, etc. On the other hand, a disadvantage is that a nationally representative sample cannot be constructed easily. The aim is to obtain representativeness by increasing the number of municipalities, currently at 10. Another constraint is the lack of human resources, for which further funding is necessary.

JSTAR is largely based on the Survey of Health, Ageing and Retirement in Europe (SHARE) as is common across HRS-related surveys and has eight sections (individual characteristics and family; cognitive ability; work; health; income and consumption; grip strength; housing and assets; and medical treatment and care service usage). It differs from other studies in that it is not a nationally representative stratified random sample. In addition, JSTAR measures food intake using a questionnaire validated in Japan. For those who gave permission, it is possible to link the data to administrative records of their health and nursing care usage, as well as their medical examinations. It would be of benefit if the government allowed links to social security data and tax records.

While some argue that people, if left on their own, would not save enough for old age, Scholz, Seshadri and Khitatrakun showed in 2006 based on the HRS that people are over-saving rather than under-saving. Following this, calculating expected net lifetime wealth by cohort using income/expenditure, consumption and health information collected through JSTAR reveals that the ratio of people with a below average expected net lifetime wealth decreases with age (50 and over). Without such insights, it is simply impossible to design a sensible pension system or more generally a sensible social security system. For this, HRS type panel data provide the key.

Dr. Kapteyn of RAND gave an account of preliminary work and conclusions drawn from comparing well-being measures in a number of countries, as an example those of depression.

The motivation behind this comparative study is the aim of government policy to improve the well-being of its citizens and the obvious positive correlation between well-being and health. The objective is to determine the differences across countries and people. By using several items from the Center for Epidemiologic Studies Depression Scale (CES-D) asking respondents to report unpleasant feelings, a depression score is determined for individuals. The items chosen are those that would yield comparable data, for example demographics, major/minor health conditions, health behavior, limitations in activities of daily living (ADLs), limitations in instrumental activities, age and birth year.

The results from a multivariate analysis, i.e. keeping other things equal, depression scores from different studies were compared. The HRS shows that in the U.S. depression decreases with age. However, looking at birth year, those born earlier are more likely to be depressed. Results of the English Longitudinal Study of Ageing (ELSA) study in the UK show the same trend by age and cohort, as do the patterns from SHARE for Continental Europe though the correlations there are slighter. This exercise stresses the fact that effects of age cohorts have to be taken into account on top of age when studying depression. Even controlling for the selectivity effect, i.e. that optimistic people are more likely to live longer, essentially the effects of changes in depression by age are the same. Other demographics that influence the likelihood of depression include ethnic background, smoking, gender, work, ADLs and IADLs, marital status and income.

Controlling for these variables, one interesting finding from the HRS is that the effect on an individual's sense of depression is major for those without health insurance compared to those with health insurance. Looking at work as relating to retirement, the effects vary across the U.S., the UK and Europe. Since effects are controlled for income and wealth, etc. the differences in the importance of work on depression may be based on social factors.

In sum, the similar patterns across countries show that there is evidence for different roles played by age and cohorts. This speaks for the significance of having panel data. In addition, reduced functioning leads to more depression; money and being married are protective against depression; and females are more at risk of depression. Questions raised by differences such as work being a more important factor in the U.S. or the influence of insurance on dealing with major health shocks have to be considered when looking at policies in this field.

Dr. Ichimura asked whether there were studies looking at the linkage between depression and suicidal behavior, to which Dr. Kapteyn replied that it could be added in the JSTAR but that complications could arise in measurement and number of observations. Responding to a question from the floor, Dr. Kapteyn also explained that religion or other spiritual practices were not currently considered in the descriptive data.

Dr. Ichimura added to his presentation that JSTAR does not include explicit questions on job status (non-regular, contingent), but does include non-paid, voluntary work. He also explained that the JSTAR sample covers about 0.05% of the population and that it includes Japanese/non-Japanese alike as long as the respondents are registered in the municipalities. Furthermore, JSTAR measures social capital by number of friends in different categories. To a question about gaining permission about respondents' health conditions, Dr. Ichimura stated that building rapport is very important for the interviewer.

Dr. Axel Börsch-Supan (Director, Munich Center for the Economics of Aging (MEA) at the Max-Planck-Institute for Social Law and Social Policy (MPISOC)) spoke on the lessons for policy and the value of international comparability, using disability insurance as an example.

In order to compare differences in the effects of public and social policy on behavior, it is important to study cross-national variation through econometric analysis of survey and macro data. Some typical insights from cross-national correlations are the negative incentive effects of pension provisions on early retirement, the fallacy of the relationship between early retirement and unemployment rate, the correlation between per capita expenditure devoted to the elderly versus the young, and the effect of health care spending on health status.

To solve the causality issue, panel data and detailed micro-data are necessary because it is not easy to draw conclusions from broad macro aggregates which are very often simultaneously determined. Exogenous forces of policy effects can only be explored through micro-data and panel data as they include data over time and policy events. For example, historical experiments in Germany demonstrate the importance of using panel data to make effects, such as drop in retirement age and effect on unemployment, visible.

In Europe, SHARE has collected data from 20 European countries with a sample size of 60,000 households. It is the closest correspondence to a laboratory setting, measuring all the effects of pension systems and health care systems. SHARE has been built up to use cross-national variation among health, labor market and institutional data. The challenges in collecting data in different countries include variations in language, institutions, interpretation and methods. These are solved by using computer-assisted technologies, objective measures and common reporting styles.

An example of a similar comparative study was done on the side effects of disability insurance, for which there are huge variations in enrollment across Europe. Looking at the causes of these variations and relating them to health and other variables, it is found more probable that those who are in bad health receive disability benefits within each country. However, there seems to be a very strong effect of policy between countries in terms of lowering the number of recipients receiving disability benefits in countries where generosity is low. Thus, there is a clear distinction of health as a driver of disability benefits within each country, but it is the policy across countries that makes a difference.

In conclusion, international comparisons are powerful in detecting policy effects, but they only work when data is harmonized so as to avoid spurious effects. Obtaining panel data thus requires resources, funding, foresight, and patience.

Dr. Jinkook Lee (Senior Economist, RAND Corporation) presented some of the pilot data results from the study in India in the area of cognitive health.

The motivation behind studying India is its size, large population and huge geographic variation with each local government having different policies. As for cognitive function, it has been found to be a risk factor of physical chronic disease and vice versa. India is experiencing an epidemiological transition with non-communicable diseases increasing rapidly. However, not much is known about the cognition of older Indians.

Research has shown that in developed countries, the cognitive functioning of women is generally better than that of men. The evidence from developing countries, though much more limited, shows that women perform worse than men, due possibly to educational differences. However, the literature on India focuses only on single-city populations with the female disadvantages more pronounced in the Northern states. This North-South difference may be due to female discrimination and gender inequity.

The pilot survey of the Longitudinal Aging Study in India (LASI) was completed in 2010 covering four states and 1,546 randomly selected households. Indian census data show real gender imbalances in the North and a difference in life expectancy with the regular pattern of women living longer than men in Southern states, but not in Northern states. Based on results of cognitive tests, the LASI revealed gender differences in episodic memory as well as global cognition with women performing worse and a greater gap especially in Northern states.

The cognitive gender disparities could be explained by gender inequality (under-nutrition, education and health care), restricted social engagement and increased psychological distress. These possible risk factors of cognition were analyzed to see whether the female disadvantage persists after controlling for them. For example, there is no female disadvantage in terms of nutrition and food insecurity, but a big disadvantage in terms of education and literacy with a much higher proportion of women with no schooling in Northern. A female disadvantage is also observed in self-reported chronic diseases (but not infectious diseases); men are found to be more socially engaged; and women in Northern India are more depressed than men. Controlling for the risk factors, education and psychological distress were found to be able to explain the female disadvantage, but not others. Analyzing the covariants, the main effect of female differences observed in the Northern states is found not to be statistically significant.

In conclusion, women aged 45 and older in India do worse in cognitive tests, and more so in Northern states, and education accounts for 40%-50% of the gender disparity in cognition. This implies that greater access to education among women has a great potential to reduce gender disparities in cognition.

Dr. Lee explained that one variable to look at in the future is the extent of experiencing discrimination in early childhood, which may explain the regional differences. She also stated that marital status did not have a protective effect for cognition and depression in India. Dr. Ichimura added that looking at age and societal change over time may also be of interest. A participant further suggested adding predictive probabilities to compare North-South differences.

Dr. David Weir (Research Professor, Survey Research Center, University of Michigan; Director, Health and Retirement Study (HRS)) gave a presentation on the use of panel data like HRS for analysis of policy.

The HRS and its international network have particular relevance to policy and are beneficial for studying fundamental questions about health and processes of aging given their coverage of economics and policy participation. Aging is the policy challenge of the 21st century everywhere as aging populations create fiscal pressures with commitments to retirees exceeding current taxes from workers. Although comparatively the U.S. is not aging as rapidly, the current impasse over raising the debt limit is essentially about aging and a complete lack of what it means for public policy.

Policies for older Americans include Social Security and Medicare with the effects on the latter more unpredictable and uncontrollable. Medicare's unfunded liability, estimated at about US$4.1 trillion, is larger than that for Social Security. This is why Medicare is the biggest and most difficult aging-related policy in the U.S. To contain spending on Medicare, the best policies would be to limit spending on treatments that provide less benefit, since the U.S. spends more on health care but has a lower life expectancy. In addition, the Dartmouth atlas shows regional variation in spending within the U.S. However, high-cost medicine in areas does not seem to lead to better outcomes than low-cost medicine in other areas. Therefore, the policy prescription being pushed is to have expensive regions practice medicine more like the less expensive areas to save money without harm to health.

Looking at the distribution of expenditure for Medicare, while there are differences, the effects seem exaggerated, and what needs to be established is whether people persistently spend more than others in the same health condition. With the HRS linkages to administrative records, it is thus valuable for policy research. On the question of whether to target interventions on regional provider or individual spending variation, i.e. which has greater potential for savings, the answer would have to come from observing Medicare spending in claims data for time periods between HRS interviews and regressing over various variables. The findings show that the patient effect explains more of the residual variance than looking at the regional provider effect. Thus, targeting policies at individuals would be at least as effective as targeting providers.

Dr. James P. Smith (Distinguished Chair in Labor Markets and Demographic Studies, RAND Corporation) spoke on the comparison of the HRS and ELSA to show how they address interesting issues and puzzles.

In the background to this comparison was the motivation to determine the reasons why in terms of prevalence of various diseases the U.S. leads even the UK among those 55-64 years of age. Differences in reporting as well as conventional risk factors were considered as explanations, but they did not explain why Americans had the highest disease prevalence. Neither did less standard risk factors, such as social integration and support (behavioral/psychosocial risk factors). However, one significant explanatory risk factor were body shape indicators such as waist circumference, as opposed to body mass index (BMI) which is regarded as a poor measure of risk of disease.

Another hypothesis was to study childhood disease history. Across all age groups, Americans were more prone to childhood diseases than the English. For all kinds of adult illnesses as well, the Americans fared worse than the English. Controlling for age, gender and country differences, measuring the interaction of adult and childhood diseases, existence of diseases at very early ages seems to be a significant contributing factor for adult illnesses in the U.S. Another explanatory factor may be differential screening effects for cancer in the U.S.

Other uses of data are in mortality and disease incidence. Comparing the HRS and ELSA in these terms, it can be concluded that the U.S. has a higher prevalence and incidence of disease, but Americans outlive the English. Therefore, despite the inefficiencies in the U.S. health care system, the idea that it is not productive is incorrect since it has a lower mortality rate. However, if control measures, such as standard health behavior, marital status and work, are included, the differences in mortality between the two countries diminish greatly. Furthermore, if health status is controlled for, the mortality gradients by financial status between the two countries actually disappear. Also, comparisons over longer periods of time generate little evidence of significant wealth effects.

Session Chair Dr. Hideki Hashimoto (Professor, School of Public Health, the University of Tokyo) gave a presentation on some early results of JSTAR.

Following the end of the Second World War, Japanese life expectancy has grown to exceed that in all countries of the Organisation for Economic Co-operation and Development (OECD), although the longevity of Japanese men seems to be decelerating now. Data to be published in The Lancet on the Japanese health care system show that primary care and lifestyle modification still matter and may be the key to strong population health in Japan.

Looking at historical changes, the demographic and epidemiological analysis has detected that prolonged life expectancy of elderly men and women in Japan is mainly due to reduced mortality of stroke and heart diseases, and especially since the 1980s among those 75 years of age and older. Furthermore, the top contributors to mortality (hypertension, tobacco, inactivity and other lifestyle factors) can be treated in a primary care setting. However, these findings do not inform policy reform as they lack social or psychological backgrounds to individual behaviors.

OECD health data show that Japan spends only about 8.5% of its gross domestic product (GDP) for health, but coverage of public expenditure for health is very high. Also, using JSTAR data, it is possible to see that equity in access is high for outpatient services, but there is a poor-rich gap for dental services. Analyzing out-of-pocket payments proportionate to household income, those in their 60s are found to spend the largest amounts and even in the lowest income quartiles, this figure is about 8% of income.

With the government discussing reform plans to increase the co-payment rate for the elderly, a precautious assessment on how this affects their access and health care outcomes is necessary. In this respect, JSTAR could be a powerful contributor. For example, studying access to preventive services, such as annual checkups, one can see considerable variation depending on education and work status as well as across cities. Linked with claims data over time, expenditures for most illnesses have consistently stayed high in Japan. JSTAR additionally first introduced a food frequency questionnaire to study dietary patterns, which show, for example, that age, marital status and region make quite a difference on the intake of different types of food.

In conclusion, while the longevity trend has so far been good, there are problems in primary care, and lifestyle modification (smoking and healthy diet) should be enhanced. In addition, panel comparisons with HRS and SHARE are important to find out what makes longevity in Japan so unique.

In response to a question on individual and provider effects, Dr. Weir explained that the extent to which individual providers vary even within regions in persistent ways is not captured in HRS since the focus has been on regional clusters of provider behavior. He also added that regional variations within the U.S. are very localized and may be widening.

Dr. Smith stated that the degree of inequality in the U.S. is not the driving force of the difference compared to the UK. He speculated that one possibility is that preventive treatment of childhood illnesses is better in the UK, calling for an analysis of the medical system as a whole.

On the issue of burdens of an aging population on society or on individuals, Dr. Smith drew the conclusion that the extended period of life was also one of extended quality of life, to which Dr. Weir agreed since most of the burden has so far been physical disability. Dr. Weir added that it would be important to continue watching if there is earlier onset of cognitive decline and measuring burden of care among generations and how this is managed between state and family. With regard to taking into account environmental exposures in regions, Dr. Weir believed that this would be of interest in the future.

Dr. Robert Willis (Professor, University of Michigan) gave a presentation on his paper about the association of policy variations and survey data to explore effects on cognition.

Despite popular belief that mental exercise can stave off cognitive decline, empirical evidence for this idea is weak since the causal direction is unclear. This study uses HRS, ELSA and SHARE to investigate the idea that people could avoid cognitive decline at older ages by being in a more mentally stimulating home or work environment. A previous study using these data found remarkably strong negative correlation across countries between cognitive ability and retirement illustrated plot of cognitive performance versus labor force participation. This study finds support for the hypothesis that this is a casual effect using country-level retirement policies as instrumental variables. The rationale for this approach is that most of the cross-country variation in retirement has been shown to be a consequence of the incentive effects created by public pension, disability and tax policies, and it is unlikely that these policies have been set in response to observed age patterns of cognition in a country's population. Thus, policies provide valid instruments to remove reverse causation of cognition on retirement behavior in micro-data.

Based on such a research base, that retirement is influenced by policies, the Mental Retirement Hypothesis investigates the causal effect of retirement on cognitive status of older persons. This involves identification issues which could be misleading since retirement is a self-selected status and could lead to reverse causation. Theoretical cognitive psychology states that fluid intelligence, or the "thinking" part of ability, decreases with age, while crystallized intelligence, the "knowing" part of ability, increases. The Flynn Effect of large cohort growth in fluid abilities may be explained by aspects of human capital theory.

Thus, the two arguments to be made for the Mental Retirement Effect could be the Disengagement Lifestyle Hypothesis, i.e. increased stimulation in the workplace, or the On-the-Job Hypothesis, i.e. different incentives to invest human capital. Regressing labor force participation on cognition, the negative correlation can be illustrated clearly. From results of models using different variables, it can be concluded that early retirement has a significant negative impact on the cognitive ability of people in their early 60s which is both quantitatively important and causal. These findings are consistent with research showing that fluid intelligence is affected by human capital and show that not working at an older age reduces cognition. For Americans, there has been a reversal of the century-long trend toward early retirement, which could be good news for the cognitive capacity of the aging nation.

Session Chair Dr. Satoshi Shimizutani (Consulting Fellow, RIETI; Senior Research Fellow, Institute for International Policy Studies (IIPS)) spoke on the retirement process and social security on Japan using macro evidence of labor supply and retirement of the elderly as well as cross-sectional evidence from JSTAR.

As already stated, Japan is showing an unprecedented speed of population aging, longer life expectancies, later retirement and higher labor force participation in old age. In addition to these macro observations, it is also necessary to consider micro observations, in particular on the issue of individual decision making in retirement focusing on policy/institutional effects. In such an analysis, diversity among people and incentive mechanisms need to be kept in mind. While the relationship between social security and labor supply has been intensively studied abroad, data had been scarce in Japan until the emergence of JSTAR.

In studying retirement, three features have to be considered: (1) retirement depends on definition; (2) it may be a gradual process; and (3) it may be a joint decision. JSTAR data illustrates that there is a jump in the 60s in the non-working status, with a higher female non-working population. Furthermore, while the retirement age is concentrated on age 60, the dominant expected age of retirement is at 65. The age at which people start receiving public pension benefit (National Pension Insurance (NPI) or Employees' Pension Insurance (EPI)) is also concentrated on ages 60 or 65. It should be noted that JSTAR is the only data source to examine a forward-looking claiming decision, which has important policy implications for designing public pension policies.

Preliminary analysis using JSTAR data shows that survival probability and liquidity constraint is significantly related to claiming behavior. There are also huge discrepancies between changes using cross-section or panel data. It is also possible to track the transitions in employment status (employed to self-employed and full-time to part-time) and declining hours worked before retirement with the JSTAR data. Regression analyses reveal factors affecting probability of working for men and women, such as numbers of hours worked, spousal separation, and lifetime assets, and for men, cognitive decline.

In conclusion, JSTAR is a nice opportunity to explore the retirement "process" by longitudinal, interdisciplinary and international features. Further examination of the "process" by JSTAR is indispensable to policy evaluation and new scientific knowledge.

Dr. David Wise (John F. Stambaugh Professor of Political Economy, Harvard Kennedy School) presented on his paper dealing with preparedness for retirement by looking at levels of household assets.

The level of assets that households hold can be used as an ex post measure of retirement preparation (assets in the last year before death) as opposed to an ex ante measure (assets at the beginning of retirement). The study also emphasizes the importance of health and family pathways to end of life.

Balance sheets including the three asset categories of financial assets: home equity, social security, and defined benefit pensions, suggest that a lot of older single people are living essentially on annuities. Three different pathways to end of life were followed in the AHEAD survey: one-person households; persons who started out in 1993 in two-person households, but their spouse died in the last observed year; and persons in two-person households. In terms of evolution of assets, people who survived the longest after 1993 had the largest assets in all three family groups. Health status related highly to the drawdown of assets as well as mortality and future health events, among others.

The conclusions that can be drawn from looking at annuity income, non-annuity assets, and health before death are that a large fraction of retirees rely almost entirely on Social Security benefits (annuity wealth) for support in retirement, with no financial or housing wealth (non-annuity wealth), and those with the least wealth are in the poorest health. Taking into account that 60% of respondents say that their retirement is "very satisfying," the proportion of those "very satisfied" increases considerably in the highest percentiles in terms of health and annuity and non-annuity wealth.

In sum, wealth at death is greatest for people who remain in two-person households the longest. There is a very strong correlation between the level of assets in 1993 and the number of years a person survives after 1993. Furthermore, there is a very strong relationship between health status and wealth at death. Finally, a large percentage of people die with annuity income only, with no financial assets, and with zero housing wealth. Thus, greater financial assets (and housing wealth) would increase life satisfaction in retirement.

Responding to a question on the policy implication of his research, Dr. Willis explained that the policy thrust to have people work longer, in addition to improving fiscal balance, would be to raise their life satisfaction by keeping their minds more active.

In response to a question on the effect of education on life satisfaction, Dr. Wise stated that education is strongly related to health. He further explained that support from family was not included in the analysis, but that living arrangements in retirement would be an important consideration. On the similarity of the slopes of reduction of total wealth for all three family pathways, Dr. Wise added that the decline is estimated based on the beginning of retirement and, controlling for health status, in the previous period.

Mr. Atsushi Nakajima (Chairman, RIETI) expressed his delight at being able to host the very first international symposium on panel data for the elderly. As part of the world standard HRS family, he hoped that JSTAR will continue to produce good findings and further develop in the future.