This paper examines the effect of the 1995 implementation of National Health Insurance (NHI) on retirement behavior in Taiwan. The identification strategy is based on the fact that, in Taiwan, adult children offer significant insurance to elderly parents. The results suggest that NHI had significant effects on retirement among the elderly lacking an adequate traditional 'safety net.' NHI raised the conditional probability that a male, private sector worker over the age of 51 would retire by more than 60%. However, men with a stronger safety net in the form of adult sons were less responsive to NHI.
(JEL I18, J14, J26)
DGBAS: Directorate-General of Budget, Accounting and Statistics
FI: Farmer's Insurance
GEI: Government Employee's Insurance
LFPR: Labor Force Participation Rate
LI: Labor Insurance
NHI: National Health Insurance
OECD: Organisation for Economic Co-operation and Development
(ProQuest: ... denotes formulae omitted.)
Population aging and low fertility rate have brought a challenge to more and more countries as the growth in early retirement places an unsustainable burden on the younger generation. Between 2006 and 2050, the projected percentage of population aged 60 yr or older will grow worldwide from 11% to 22%, in North America from 17% to 27%, in Europe from 21% to 34%, in China from 11% to 31%, and in Japan from 27% to 42% (United Nations, Department of Economic and Social Affairs, Population Division, 2006). However, labor force participation among senior populations is decreasing with time. As a result, increasing the labor force participation rate (LFPR) of people aged 55-64 yr is one of the major goals of social policy within the European Union under the Lisbon and Amsterdam Treaties (Leibfritz, 2003).
Previous studies of retirement behavior mainly document that health status and postretirement income (including pensions, disability insurance, or social security) are primary determinants of one's retirement decision. However, before 1990, there was virtually no work on the impact of health insurance on labor supply and job mobility decisions (Gruber and Madrian, 2002). As medical costs continue to rise globally, the availability of postretirement health insurance has since 1990 drawn much attention in this line of research. The U.S. literature suggests that access to postretirement health insurance or pensions makes early retirement more attractive (Currie and Madrian, 1999; Gruber and Madrian, 1995, 2002; Madrian, 1994; Rogowski and Karoly, 2000). Medicare is also an important concern when workers in the United States consider the timing of their retirement (Madrian and Beaulieu, 1998; Rust and Phelan, 1997). Gruber and Madrian (1995) found that 1 yr of continuation of coverage raises the retirement hazard by 30% in the United States. They concluded that policies to provide universal health insurance coverage could lead to a large increase in the rate of early retirement. On the other hand, there has been little investigation into this issue in Asian countries where aging is as prominent as it is in Western societies. This article addresses this gap by conducting a case study for Taiwan. In Taiwan, postretirement health insurance was limited to government employees before the introduction of National Health Insurance (NHI) on March 1, 1995. Because NHI provides universal health insurance coverage for all citizens, private workers gain postretirement health insurance. This social experiment makes it feasible for economists to examine people's response to postretirement health insurance when they consider their labor force participation decisions.
Mete and Schultz (2002) conducted the first analysis of the effect of Taiwan's NHI on labor force participation among the elderly. Their study comparing public workers (who always had postretirement health insurance) to private workers (who gained postretirement health insurance) suggested that NHI had little impact. However, their work does not take into account a pension reform in the public sector that occurred at the same time. This paper develops a new identification strategy based on the strong tendency among Taiwanese adult children to provide their elderly parents with significant 'family insurance.' Specifically, the author questions whether or not NHI has had a greater impact on people with fewer adult children. The effects of health insurance on retirement decisions may differ in Taiwan compared to Western countries because the elderly depend on their children to a much greater extent. In fact, despite an increasing trend toward living independently, the majority of elderly Taiwanese citizens still live with their children. Moreover, most adult children (no matter whether they reside with their parents or not) offer substantial financial support to their parents. These facts are in line with the argument of Kotlikoff and Spivak (1981) that families can self-insure against uncertainties. Therefore, elderly citizens with a 'safety net' in the form of supportive children could be less responsive to public programs. One noteworthy social norm in Taiwan is that adult sons take more responsibility for supporting their parents than do adult daughters. Consequently, the author also attempts to examine whether prospective retirees consider the gender of their children when assessing the potential role of their adult children in securing their economic future.
This study constructs a duration data set from the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan. Drawing on estimates from a proportional hazard model, the empirical analysis obtains two findings. First, NHI raised by more than 60% the conditional probability that a male, private sector worker over the age of 51 would retire. However, men with more adult sons were less responsive to NHI than their counterparts. Second, men with more adult sons had a significantly higher conditional probability of retirement (retirement hazard) than men with fewer adult sons and women with more adult daughters had a significantly higher conditional probability of retirement than women with fewer adult daughters. This difference is particularly striking for women given social norms dictating that sons take primary responsibility of their aging parents.
Until April 2007, 30 countries in the world had implemented a universal health care system, while many others had debated whether they should introduce one. Implementing NHI in a nation without postretirement health insurance is equivalent to providing postretirement health insurance to senior citizens. This study can help policymakers to assess the extent to which NHI or similar programs would affect labor markets, especially in societies with family caretaking social norms similar to those in Taiwan. Because the financing of many welfare and social insurance programs in various countries is payroll related, the resulting increase in early retirement not only causes a decrease in aggregate output as more experienced and productive workers leave the labor force but also influences the financing of these public programs. These associated long-term effects need to be accounted for by policymakers in considering NHI or similar policies.
The rest of this paper proceeds as follows. Section II provides the background. Section III describes the data. After explaining methods in Section IV, the author presents results in Section V. Section VI concludes and provides directions for future research.
A. Trends in Elderly Labor Force Participation
As in many countries, the LFPR of the elderly population in Taiwan has decreased over the past several decades. A growing tendency toward earlier retirement contributes to this decline (Directorate-General of Budget, Accounting and Statistics [DGBAS], Executive Yuan, Taiwan, 2001a, 2001b). These trends have continued since the 1995 implementation of NHI. For example, LFPR among male workers aged 65 yr and older was 14.39% in 1995 but fell to 11.54% in 2002. LFPR among men aged 50-64 yr decreased by approximately 10 percentage points between 1995 and 2002. Despite a slight increase in LFPR for women aged 50-54 yr, women over 55 yr experienced a decline in LFPR after 1995. Because of these preexisting time trends, a before-and-after analysis is insufficient to answer the research question posed by this paper. Rather, an identification strategy based on the traditional family safety net in Taiwan will be adopted in the following analysis.
B. Dependence on Children among the Elderly
The number of adult children is not a common predictor for retirement decision in previous research focusing on Western societies. Because most Western countries have a wellestablished social insurance system and welfare programs, senior citizens have not needed to depend on their children after retirement.1
On the contrary, the elderly in Taiwan depend on their children in many ways. First, most elderly parents live with their children. Although more senior citizens have chosen to live independently in recent years, the fraction of people living with their children was still over 60% in 1996 (Kan, Park, and Chang, 2001). This rate remained as high as 75.6% among elderly Taiwanese women in 2000 (The Ministry of the Interior, Taiwan, 2002). Second, many adult children, regardless of whether they live with their parents, provide parents with financial aid. Among those over 65 yr, 47.13% of respondents said that the majority of their financial support came from their children, while 15.39% and 13.72% said that the majority came from pensions and labor income, respectively. Sixty-three percent of women over 65 yr reported that the majority of their financial support came from their children, whereas 6.2% said that support was derived from pensions and only 9.8% from labor income (The Ministry of the Interior, Taiwan, 2002). Third, children are important caregivers when parents are hospitalized.2 Among elderly males over 50 yr, 37.48% stated that their wife cared for them and 29.01% said that their children cared for them when they were hospitalized. Among women over 50 yr, 16.02% stated that their husband took care of them and 40.45% said that their children took care of them when they were hospitalized (The Ministry of the Interior, Taiwan, 2000). Finally, many children pay their parents' medical expenses.3 Among the 3,511 individuals who answered the question 'Who paid most of your medical fees last year?' in the 1989 wave of the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan, 39.82% of respondents chose 'myself,' 27.11% chose 'insurance,' and 26.63% chose 'children or children-in-law.' If we restrict the sample to respondents with children, we find that 39.87% of respondents chose 'myself,' 26.19% 'insurance,' and 28.05 % 'children or children-in-law.'
Lee, Parish, and Willis (1994) used the 1989 Taiwan Family and Women Survey to study intergenerational support in Taiwan. Their evidence shows that the loan hypothesis and the insurance hypothesis implied by the altruism model are highly consistent with their data. Repayment of parents' human capital investment in children is an important reason behind adult children's financial support. These findings explain the source of children's altruism toward their parents in Taiwan. In short, because adult children act as a safety net for the elderly, NHI coverage is expected to be more valuable to senior citizens with fewer adult children.
C. Health Insurance in Taiwan
NHI was formally implemented in Taiwan in March 1995 and provides universal health insurance coverage to all citizens. Before the implementation of NHI, an individual could obtain health insurance through one of three programs-labor insurance (LI), farmer's insurance (FI), and government employee's insurance (GEI). These insurance programs are government sponsored and tied to an individual's employment status. GEI covers government employees, LI covers private employees and a few blue-collar workers in the public sector, and FI covers farmers. There is very little private health insurance coverage in Taiwan.
In 1994, the year immediately preceding the implementation of NHI, 2.95% of the population were covered by GEI, 0.67% by retired government employees' insurance, 8.23% by FI, and 40.22% by LI. The 48% of the population without health insurance coverage were mostly children, housewives, and senior citizens. Because of the enforcement of NHI, the percentage of the population with health insurance coverage increased from 52% in 1994 to 96% in 2000. NHI consolidates the previous health insurance system through replacing maternity, injury, and illness benefits of GEI, LI, and FI programs. It also covers severe illnesses and preventative health care services. The NHI premium is payroll related, the deduction being shared by the employee, employer, and government. For outpatient visits, the out-of-pocket expense ranges from NT$50 to NT$360.4 For hospitalization, the co-payment ranges from 5% to 30% for both acute and chronic care, depending on length of stay in the hospital. In the case of severe illness or injury, no co-payment is required (Chou, Liu, and Hammitt, 2003).
Two differences between NHI and previous public health insurance programs are relevant here. First, postretirement health insurance was only available in the GEI plan before the introduction of NHI. Consequently, the previously LI-insured and FI-insured are expected to find early retirement more attractive after 1995. Second, unlike GEI and FI, LI does not provide insurance coverage for dependents. Hence, the generosity of NHI coverage may provide the previous LI-insured additional incentive to retire earlier in that NHI now covers all their dependents.
Mete and Schultz (2002) conducted the first investigation of the effect of Taiwan's NHI on labor force participation among the elderly. Drawing on 1989, 1993, and 1996 data from the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan, they used public workers as a control group and found that NHI did not contribute to a reduction in LFPRs of the elderly in 1996. However, in 1995, the Taiwanese government introduced a pension reform to the public sector, changing the original unfunded system into a fully funded system.5 This new policy might have affected the labor force participation of public workers and makes the control group (government employees) of Mete and Schultz (2002) suspect.
Two previous studies examined the effect of NHI on other outcomes. Chou and Staiger (2001) showed that NHI was associated with a decline of 4.6 percentage points in the labor force participation of women aged 20-65 yr whose husbands were paid employees. The effect was larger (6.7 percentage points) among wives of less educated men. Chou, Liu, and Hammitt (2003) documented that NHI lowered average savings by 6.9% and raised average consumption by 2.4%. The effect on savings was especially strong for the households with the least savings. However, these results must be interpreted with caution because, like the investigation conducted by Mete and Schultz (2002), these studies relied on government employees as the control group.
This paper constructs a duration data set from the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan.6 This survey was initiated in 1989 with 4,049 respondents aged 60 yr or older. Of these, 3,155, 2,669, and 2,310 individuals were reinterviewed in the 1993, 1996, and 1999 followups, respectively. Launching a second panel of individuals aged 50-66 yr in 1996 extended the survey. The completed second sample contains 2,462 individuals. For convenience, the author will denote the original panel as the 'old panel' and the new panel as the 'young panel.' Because NHI started in 1995, the empirical investigation in this paper will use the 1989, 1993, 1996, and 1999 surveys (of both the old panel and the young panel).
The Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan provides rich information about an individual's background characteristics, family structure, marital history, health status, residential history, and employment history (e.g., occupation, start date, end date, and employment status of each job after age 50). This allows the author to construct a unique duration data set characterizing a worker's job history from age 50 onward. The younger cohort's job history began in later years. Although a few work histories in the old panel go back as far as the 1950s, most people's work history began after 1960. In order to have a time period more comparable to the 1990s, when NHI was introduced, the author confines the analysis to the period 1980-1999.
Although the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan has a panel structure that facilitates the task of investigating a senior citizen's retirement behavior in a dynamic framework, this survey provides limited information about an individual's income, wealth, and consumption and neither is detailed age-varying income available. Each respondent was asked about income from jobs in 4 yr (1989, 1993, 1996, and 1999) only. Because the response rate to these questions was very low, interviewers asked people who declined to report their exact income about the 'range' of income instead. But the response rate was low even in this case. Therefore, the author does not use this information.
A. The Definition of Retirement
According to Madrian (1994), the definitions of retirement include (a) a permanent departure from the labor force, (b) a substantial reduction in the number of hours worked, (c) self-reported retirement, and (d) receipt of pension or social security benefits.
The 1989 survey of the old panel contains self-reported retirement status and age at retirement. These data come from the following questions: 'Have you ever retired from work? If so, how long ago?' Another question concerning retirement is 'When did you stop doing your last job?'8 The 1993, 1996, and 1999 surveys, however, ask only the second question, which is more ambiguous than the self-reported one. In order to identify 'retirement' more precisely, this paper uses the 1989 retirement status as a starting point for the old panel. The author traces each respondent's subsequent labor force attachment in the 1993, 1996, and 1999 follow-ups and updates his/her retirement status accordingly except for those who left the survey due to attrition. If an individual self-reported himself/herself retired in 1989 and did not resume working between 1989 and 1999, then he/she is considered to be retired and permanently out of the labor force according to Definition (a). If an individual self-reported himself/herself retired in 1989, but resumed work afterward, the individual is not categorized as 'retired' unless he/she stopped working again later on. A similar procedure is applied to the young panel.
To prevent the loss of information from deceased or lost subjects (a 'mover') in the follow-ups, this paper employs an unbalanced panel. A mover who had already retired when he/she left the survey is included in the sample. A mover who was still working when he/she left the survey is treated as a general righthand censoring-the survival time is measured from the time of entrance (entered the state of 'being at risk of retiring') to the last contact. In essence, when constructing the duration data set, the author traced each individual until he/she retired or became censored.
Individuals who had never worked and the self-employed are excluded from the sample. People over 80 yr in 1996 are also excluded from the sample to avoid the impact of outliers. In addition, the sample excludes workers who stopped working due to layoff, shutdown, or business failure so that voluntary retirement is highlighted. Because it is difficult to define retirement for farmers, they are also excluded from the sample. Due to the pension reform in the public sector starting in 1995, public employees are not included in the sample either. In sum, the following analysis focuses on private employees in nonagricultural sectors.
B. Survival Functions
A survival function derived from the above duration data set can measure the probability of a person working at least until a specific age (i.e., the survival rate). In other words, it shows the probability that a person retires after a specific age. The comparison of unconditional survival rates among various demographic groups is helpful before any regression analysis. For example, the Kaplan-Meier survival estimates of male private employees working in nonagricultural sectors (Figure 1) show that men with NHI coverage (NHI =1) have lower survival rates than men without NHI coverage (NHI = 0). This difference is obvious for most age groups in Figure 1. On the other hand, in Figure 2, the distinction between the survival rate of women with NHI coverage and the survival rate of women without NHI coverage is not as noticeable as that in the men's sample. It shows that women could respond to the policy to a less extent than men. To explore this gender difference further, the author investigates the policy effect on men and women separately in Section V.
Breaking down the sample into groups with different family safety nets gives us more insight into people's retirement behavior. In particular, the comparison between Figure 3 and Figure 4 indicates that the decrease in survival rate after NHI for observations with no adult children is of greater magnitude than that for observations with three or more adult children.
In other words, elderly employees with a weaker safety net (as in Figure 3) are more responsive to NHI than their counterparts (Figure 4). This observation motivates a new approach to identify the policy effect in this paper. Because people with different numbers of adult children alter their retirement decision in different ways in response to NHI, it has become feasible to measure the policy effect in spite of the preexisting trend toward early retirement. In light of this, the following empirical analysis will use an interaction term between the NHI indicator and the number of adult children to capture the effect of NHI.
As Hausman and Wise (1985) suggested, retirement behavior can be modeled in the following hazard framework.9 Define the age of retirement of any particular person as a continuous random variable, A, and denote a as a realization of A. Furthermore, consider a large population of people who entered the state of being at risk of retiring at 51 yr.10 Then, the probability that a person has retired by age a is given by the distribution function:
where the hazard function A(.) > 0. This specification has an intuitive interpretation that the probability of retirement goes to 1 as an individual gets older (i.e., when a gets larger).
The Cox proportional hazard model with age-varying covariates can be specified by the hazard relationship as follows:
where h^sub 0^(a) is an arbitrary and unspecified baseline hazard function; Z(a) = (z^sub 1^, z^sub 2^(a)) consists of age-invariant covariates z^sub 1^ and age-varying covariates z^sub 2^(a) and β is a column vector of regression parameters. In this framework, h(a; Z(a)) represents the 'retirement hazard' that measures the probability that a worker would retire at age a, given that this worker was still working before age a. Therefore, it is the conditional probability of retirement at age a.
The proportional hazard model is nonparametric in that it involves an unspecified baseline hazard function. Estimating a flexible model like this prevents the analysis from deriving conclusions driven by a specific parametric assumption. Consequently, the conclusions in this article will be robust and independent of the choice of functional form of the baseline hazard function. This study uses Cox's partial-likelihood estimation to estimate β. It was shown by Cox (1975) that the method used to construct this likelihood offers maximum partial likelihood estimates that are consistent and asymptotically normally distributed (Kalbfleisch and Prentice, 1980).
The explanatory variables in the model (Z(a)) consist of age-invariant variables z^sub 1^ and age-varying variables z^sub 2^(a). These are summarized in Table 1.
Inspired by the comparison between Figure 3 and Figure 4 in Section III, the following empirical analysis will include not only the policy indicator NHI^sub a^ but also the number of adult children at age a and an interaction term between these two variables. These will help test whether individuals with different numbers of adult children behaved differently before and after the introduction of the NHI program.
Table 2 provides the observation-level descriptive statistics by numbers of adult children.
The men's sample and the women's sample both show that workers with no adult children are more likely to be single persons and mainlanders.11 In order to investigate whether mainlanders have a different taste for retirement or not, the author runs a set of beforeNHI regressions.
Results in Table 3 suggest that mainlanders did not behave significantly differently from the other demographic groups before the implementation of the NHI program.12
The main spirit of the empirical model in Equation (2) is that all workers of the same age group (say, age a) 'compete' to retire given that they were still working before age a. Therefore, age is a fixed constant for each 'competition.' Table 4 shows estimates of Equation (2) for private employees in nonagricultural sectors from 1980 to 1999. The number of retirees and censored individuals are presented in each regression too. Instead of reporting coefficient estimates, the author presents the more intuitive exponential coefficients, which are the hazard ratios.
In order to investigate the gender difference shown in Figures 1 and 2, we can focus on the full sample and read the relative hazard ratio of men and women (i.e., ea in Equation (A4) of the Appendix). Table 4 shows that this relative hazard ratio is approximately 0.7 and statistically significant. In other words, the retirement hazard (the conditional probability of retirement) of males is 30% less than that of females. This is consistent with the stylized fact that men tend to stay in the labor market longer than women do. Breaking down the sample into men and women, the results show that a male employee with a spouse has a significantly lower retirement hazard than a single man. However, the spouse effect for women workers is not significant.14
Individuals with chronic conditions have a significantly higher retirement hazard rate than their counterparts. This indicates that healthier workers stay in the labor market longer. It is also expected that people with chronic conditions change their retirement behavior after they obtain eligibility for NHI. For individuals with chronic conditions at age a, the relative hazard ratio e^sup β^sub 1^^+^sup β^sub 3^^ (see Equation (A6) in the Appendix) measures their average response to the policy. The estimate based on specification  in Table 4 provides a significant ratio equal to 1.570 (z = 2.233) for all workers. The men's sample reacts significantly to the policy as well (e^sup β^sub 1^^+^sup β^sub 3^^ = 1.906 [z = 2.5291). However, the response of women (e^sup β^sub 1^^+^sup β^sub 3^^ = 1.031 [z = 0.088]) is not statistically significant. Other things being equal, the average influence of NHI on observations without chronic conditions is reflected by e^sup β^sub 1^^. It is 1.639 (z = 2.29) for the full sample, 1.953 (z = 2.44) for men, and 1.103 (z = 0.26) for women. These estimates show that the increase in the conditional probability of retirement among people with chronic conditions is less than that of their counterparts. It is possible that people with chronic illnesses received better health care and became more productive after the implementation of NHI.
The key estimates in this paper are hazard ratios for the number of adult children ac^sub a^, the policy indicator NHI^sub a^, and the interaction effect spotlighted by ac^sub a^ � NHI^sub a^. As expected, employees with more adult children have a significantly higher retirement hazard than their counterparts. Considering individuals with aca adult children at age a, e^sup β^sub 1^^+^sup β^sub 5^^ (see Equation (A10) of the Appendix) gives the ratio of after-policy retirement hazard to the before-policy retirement hazard. Other things being equal, e^sup β^sub 1^^ measures the average response of people without adult children. Take regression  in Table 4 as an example. The average increase in the conditional probability of retirement for people without adult children is significantly greater than 60% for the full sample (e^sup β^sub 1^^ = 1.639 [z = 2.29]). On the other hand, for people with one adult child in the full sample, the response to NHI (e^sup β^sub 1^^+^sup β^sub 5�1^^) is equal to 1.524 and is statistically significant (z = 2.137). For people with two adult children in the full sample, the response to NHI e^sup β^sub 1^^^sup β^sub 57times;2^^ is equal to 1.417 and is statistically significant too (z = 2.035). Evidently, the safety net's effect is smaller for people with more adult children.
Because adult sons and their families take most of the responsibility for the care of parents in Taiwan, the author also tries to test whether there exists gender bias when prospective retirees consider potential financial support from their adult children. The author replaces the predictor aca with 'the number of adult sons' and 'the number of adult daughters' to probe differences in retirement decisions caused by the gender of children. It is interesting that elderly men with more adult sons had a significantly higher retirement hazard than men with fewer adult sons and elderly women with more adult daughters had a significantly higher retirement hazard than women with fewer adult daughters. We can use specification  in Table 4 to test the presence of this gender bias. The findings show that the marginal effect (the growth in retirement hazard) caused by one additional son is 25% more than that caused by one additional daughter for male employees (see Equation (A11) in the Appendix; e^sup γ^sub 1^^/e^sup γ^sub 2^^ = 1.251 (z = 2.699) according to Table 4), while the marginal effect of one additional son is 11% less than that of one additional daughter for female employees (e^sup γ^sub 1^^/e^sup γ^sub 2^^ = 0.886 [z = -1.424]). In other words, male employees are significantly more dependent on adult sons, while women are more dependent on adult daughters.
Specification  in Table 4 suggests that male employees' response to NHI decreases significantly with the number of adult sons. In this light, the number of adult sons is relevant when men changed their retirement behavior in response to NHI. To focus on the effect of adult sons, the author employs a specification with the number of sons, but not the number of daughters, as a predictor for the men's sample. The response to NHI (the increase in retirement hazard after 1995) associated with this specification is summarized in Table 5. It is shown that the response to NHI is a decreasing function of the number of adult sons for men.
In addition, these estimates are significantly different from each other (e^sup β^sub 5^^ = 0.864 [z = - 1.84]). Men with four or more sons are expected to respond to NHI program with smaller and less significant growth in retirement hazard.16 Overall, NHI raised the conditional probability (retirement hazard) that a male worker with no adult children would retire by more than 60% (e^sup β^sub 1^^ = 1.613 [z = 1.89]). However, men with a stronger safety net in the form of adult sons are less responsive to NHI.
Comparing male employees with different numbers of adult sons, we see that people with a weak familial safety net (only one adult son) increase their retirement hazards significantly after NHI. The more adult sons a man has, the smaller and less significant his to NHI. In short, male private employees with more adult sons respond to NHI in lesser magnitude than their counterparts.17 Notice that elderly women are insensitive to NHI. Perhaps this is due to the fact that the female sample is relatively small in the data set.18
In summary, the NHI coverage is less important for individuals with more adult children. So the insurance provided by adult children offsets part of NHI's effects on retirement. For men, NHI increases their conditional probability of retirement by more than 60%. However, their response to NHI significantly decreases with the number of adult sons. This interesting phenomenon is in accord with Taiwanese social norms.
This article quantifies the impact of NHI on an individual's retirement behavior. Taking advantage of the longitudinal structure of the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan, this study builds a duration data set to evaluate how prospective retirees with varying numbers of adult children responded to NHI. The evidence suggests that NHI raised the conditional probability that a male worker in the private sector over the age of 51 would retire by more than 60%. The magnitude of the support from adult children makes parents behave differently when NHI was implemented. Particularly, elderly males who had a stronger safety net in the form of adult sons were less responsive to NHI. This could be a phenomenon unique to an East Asian country like Taiwan, where social security and health insurance systems are not well established and the elderly often depend on their adult children.
In an important review of the literature, Gruber and Madrian (2002) pointed out that there was virtually no work on the impact of health insurance on labor supply and job mobility decisions before 1990. Although more than 50 scholarly articles have been written since 1990, most of them focus on the U.S. data. The case study of Taiwan in this paper adds new evidence to this line of research. In addition, it provides an opportunity to compare the empirical evidence from a Western country with that from an Asian country. For example, Gruber and Madrian (1995) showed that 1 yr of continuation of health insurance coverage raises the retirement hazard by 30% in the United States. Their paper concluded that policies to provide universal health insurance coverage could lead to a large increase in the rate of early retirement and impact the financing of such policies. The response (the increase in retirement hazard) to NHI from the male sample with three sons in this article is also approximately 30% (e^sup β^sub 1^^+^sup β^sub 5^^�3 = 1.267 in Table 5, i.e., NHI raises the retirement hazard by 26.7%), albeit people with weaker family safety nets responded in greater magnitude to NHI. The findings in this article not only support the assertion that postretirement health insurance has an impact on retirement but also show that this impact differs across various demographic groups.
Comparing this article's estimated effects with the evidence from Medicare in the United States, it is conjectured that Medicare might have less impact on retirement if there were similar familial safety nets in the United States. For example, people with employer-provided health insurance but without employer-provided retiree health insurance might be more willing to leave the labor force before the statutory retirement age if they had additional financial support from their children. Madrian and Beaulieu (1998) documented that the lack of Medicare-dependent coverage creates an incentive for men who are covered by employer-provided health insurance and are themselves Medicare eligible to continue working until their wives are Medicare eligible as well. This effect might be less pronounced if adult children in the United States provided more financial assistance to their parents.
The evidence in this paper also presents differing gender bias among men and women. It was found that, consistent with Taiwanese social norms, male employees with more adult sons had a significantly higher conditional probability of retirement than men with fewer adult sons. However, female employees with more adult daughters had a significantly higher conditional probability of retirement than women with fewer adult daughters. Due to limited information, the author could not model fertility choice in this paper. If a suitable data set is available in the future, incorporating the joint decision of fertility and other life cycle choices into a sequential decision model would present an interesting direction for future research.
Finally, one caveat should be addressed. The safety net effect found in this study may represent an initial impact only. People may alter the degree of dependence on their children in the long run now that a national program is in place. Therefore, it is possible that the safety net effect will wane in the future.
As aging societies and increasing health care costs are becoming global phenomena, more countries are calling for reforms in welfare programs in response to changes in the social and economic environment. For example, the Organisation for Economic Co-operation and Development (OECD) stated that the aging of OECD societies in the future will require comprehensive reforms addressing the fiscal, financial, and labor market implications of aging, as well as the implications for pension, social benefits, and systems of health care and long-term care (OECD, 2007). The empirical findings in this paper could provide insight to policymakers in different countries. In developed countries, immigration is increasing and most immigrants come from societies relying on family insurance. These immigrants may respond to health insurance policies differently from people who rely less on their families for financial support. In developing countries, the family safety net is usually very important. Policymakers who propose to introduce universal health care, retiree health insurance policies, or universal pension plans in developing countries should expect to see various policy effects across different demographic groups. These issues pose both interesting directions for future study as well as ongoing concerns for policymakers. The author leaves these for future research.
Chou, S. Y., J. T. Liu, and J. K. Hammitt. 'National Health Insurance and Precautionary Saving: Evidence from Taiwan.' Journal of Public Economies, 87, 2003, 1873-94.
Chou, Y. J., and D. Staiger. 'Health Insurance and Female Labor Supply in Taiwan.' Journal of Health Economics, 20, 2001, 187-211.
Cox, D. R. 'Partial Likelihood.' Biometrika, 62, 1975, 269-76.
Currie, J., and B. C. Madrian. 'Health, Health Insurance and the Labor Market.' in The Handbook of Labor Economics, Vol. 3C, edited by O. C. Ashenfelter and D. Card. Amsterdam: North Holland, 1999, 3309-416.
Directorate-General of Budget, Accounting and Statistics, Executive Yuan, Taiwan. 'The Analysis of the Trend in Labor Force Participation Rate.' 2001a.
_____. 'The Analysis of the Employment of the Middle-aged and the Old-aged Workers in Taiwan.' 2001b.
_____. 'The Statistical Report of National Situation.' 2001c.
_____. 'The Statistical Report of National Situation, No. 170.' 2002.
Gruber, J., and B. C. Madrian. 'Health-Insurance Availability and the Retirement Decision.' American Economic Review, 85, 1995, 938-48.
_____. 'Health Insurance, Labor Supply, and Job Mobility: A Critical Review of the Literature.' NBER Working Paper Series, Working Paper No. 8817, 2002.
Hausman, J. A., and D. A. Wise. 'Social security, Health Status, and Retirement, ' in Pensions, Labor and Individual Choice, edited by D. A. Wise. Chicago: University of Chicago Press, 1985, 159-91.
Kalbfleisch, J. D., and R. L. Prentice. The Statistical Analysis of Failure Time Data. New York: John Wiley & Sons, 1980.
Kan, K., A. Park, and M. C. Chang. 'A Dynamic Model of Elderly Living Arrangements in Taiwan.' Manuscript, 2001.
Kotlikoff, L. J., and A. Spivak. 'The Family as an Incomplete Annuities Market.' Journal of Political Economy, 89, 1981, 372-91.
Lee, Y. J., W. L. Parish, and R. J. Willis. 'Sons, Daughters, and Intergenerational Support in Taiwan.' American Journal of Sociology, 99, 1994, 1010-41.
Leibfritz, W. 'Retiring Later Makes Sense.' OECD Observer. 2003. http://www.oecdobserver.org/news/fullstory.php/aid/824/Retiring_later_makes_sense.html. Accessed March 2007.
Madrian, B. C. 'The Effect of Health Insurance on Retirement.' Breakings Papers on Economic Activity, I, 1994, 181-252.
Madrian, B. C., and N. D. Beaulieu. 'Does Medicare Eligibility Affect Retirement?' in Inquiries in the Economics of Aging, edited by D. A. Wise. Chicago, IL: University of Chicago Press, 1998, 109-31.
McGarry, K. 'Caring for the Elderly: The Role of Adult Children.' in Inquiries in the Economics of Aging, edited by D. A. Wise. Chicago, IL: University of Chicago Press, 1998, 133-63.
Mete, C, and T. P. Schultz. 'Health and Labor Force Participation of the Elderly in Taiwan.' Yale University, Economic Growth Center Discussion Paper No. 846, 2002.
Organisation for Economic Co-operation and Development. 'Ageing Society.' 2007. http://www.oecd.org/topic/0, 2686,en_2649_37435_1_1_1_1_37435, 00.html. Accessed March 2007.
Rogowski, J., and L. Karoly. 'Health Insurance and Retirement Behavior: Evidence from the Health and Retirement Survey.' Journal of Health Economics, 19, 2000, 529-39.
Rust, J., and C. Phelan. 'How Social Security and Medicare Affect Retirement Behavior in a World of Incomplete Markets.' Econometrica, 65, 1997, 781-831.
The Ministry of the Interior, Taiwan. 'Summary and Analysis of the Survey of the Status of the Elderly in Taiwan.' 2000.
_____. 'The Analysis of Living Status of Elderly Women.' 2002.
United Nations, Department of Economic and Social Affairs, Population Division. 'Population Ageing 2006.' 2006. http://www.un.org/esa/populauon/publications/ageing/ageing2006chart.pdf. Accessed March 2007.
van der Klaauw, W., and K. I. Wolpin. 'Social Security and the Retirement and Savings Behavior of Low Income Households.' Manuscript, 2006.
* This study is based on the data from the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan provided by the Bureau of Health Promotion, Department of Health, R.O.C., Taiwan. The descriptions or conclusions herein do not represent the viewpoint of the Bureau. The author especially thanks the Bureau of Health Promotion at the Department of Health in Taiwan for providing data. This is a revision of a paper presented at the Pacific Rim Conference of the Western Economic Association International (WEAI), Taipei, Taiwan, January 2003, in a session organized by Jack W. Hou, California State University, Long Beach. The author is very grateful to Janet Currie, Paul Devereux, Duncan Thomas, Christopher Erickson, Yingnian Wu, Jin-Tan Liu, Jack W. Hou, Fwu-Ranq Chang, seminar participants at the University of California, Los Angeles, participants of the WEAI conference, and anonymous referees for helpful comments. The views represented here are only those of the author. Any errors are the author's responsibility.
Hsieh: Assistant Professor, Department of Economics, Northern Michigan University, 1401 Presque Isle Avenue, Marquette, MI49855. Phone 906-227-1214, Fax 906-227-2229, E-mail email@example.com