University of Southern California Dissertations and Theses

Three essays on health, aging, and retirement

(USC Thesis Other)

Three essays on health, aging, and retirement

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
THREE ESSAYS ON HEALTH, AGING, AND RETIREMENT

by

Hankyung Jun

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PUBLIC POLICY AND MANAGEMENT)

May 2022

Copyright 2022 Hankyung Jun
ii

Dedication

For Tori (Jan 10, 2012 – Nov 5, 2021)
my best friend and cham-gae (aka best dog).
Until the day we play fetch again, RIP.

iii

Acknowledgements
I have immense gratitude for the people who helped me throughout this journey. My doctoral
advisor and committee chair, Dr. Emma Aguila, provided invaluable guidance and support over
the past years. Thank you, Emma, for helping me get back on my feet when I was lost, your
encouragement and belief in my academic career helped me become a stronger person. What I
have achieved throughout the doctoral program would not have been possible without you.
I would also like to thank my committee members – Drs. Alice Chen and Soeren Mattke
– for their thoughtful guidance and mentorship throughout my time at USC. Their expertise has
greatly expanded my understanding of how to tackle challenges in the healthcare system. Alice’s
constructive feedback on my dissertation helped me improve the content tremendously. Soeren
gave me many opportunities to participate in projects where I learned valuable skills and details
of the complex U.S. healthcare system. I also want to thank my committee for mentoring me
while I applied to jobs and prepared for interviews.
To my professors and colleagues at USC, thank you for making my journey bearable and
enjoyable. I am grateful to my qualifying committee members Drs. Neeraj Sood, Julie
Zissimopoulos, Eileen Crimmins, and Jinkook Lee: thank you for all your guidance and
comments. I want to thank all my friends at USC, I miss our (pre-pandemic) gatherings and talks
at the Gateway PhD office. Julie Kim and Christine Wilson, thank you for making my doctoral
journey manageable and smoother.
I would also like to thank the Korea Foundation that supported me and my dissertation
during the last two years of the doctoral program. The fellowship helped me focus on my
research during the most critical period of my studies, and I was able to be the most productive
iv

during that time. I hope my research provides useful policy insights to Korea: a country
experiencing demographic change at an exceedingly fast pace.
Many thanks to my family in Korea. My dad, Dr. Joosung Jun, taught me to aim high and
to focus on economic intuition when conducting research. My mom, Youngah Choi, always told
me to believe in my own values and keep a strong center (jung-sim). I would also like to thank
my brother, Yunsok, who was my friend whenever I visited Korea, and my beloved dogs -
(spoiled) Nengi, (super-spoiled) Dani, (the fetch addict) Tori, and (the ultimate dog) Sari. Tori
and Sari, I’m sorry I wasn’t there to say goodbye but you two will always be in my heart.
Finally, I am extremely grateful to my significant other and best friend, Dr. Sang Kyu
Cho, who also did his doctoral studies at USC. Sang has always been there with me to celebrate
the successes and to weather the challenges. Sang helped me improve as a person and as a
researcher by offering me the best encouragement and the most honest criticisms. He provided
love, cooked endless meals, was my gym buddy and travel mate, taught me how to drive, and
helped me realize my full potential. Thank you.

v

Table of Contents
Dedication ...................................................................................................................................... ii
Acknowledgements ...................................................................................................................... iii
List of Tables ............................................................................................................................... vii
List of Figures ............................................................................................................................. viii
Abstract ......................................................................................................................................... ix
Introduction ................................................................................................................................... 1

Chapter 1. The Value of Medicare Coverage on Depressive Symptoms Among Older
Immigrants .................................................................................................................................... 4
Abstract ....................................................................................................................................... 4
1. Introduction ............................................................................................................................. 5
2. Background ........................................................................................................................... 10
3. Data ....................................................................................................................................... 14
4. Empirical Methods ................................................................................................................ 17
5. Results ................................................................................................................................... 19
6. Discussion ............................................................................................................................. 22
Appendix ................................................................................................................................... 35

Chapter 2. Health and Retirement of the Older Self-Employed: Evidence from Korea,
Mexico, and the United States .................................................................................................... 39
Abstract ..................................................................................................................................... 39
1. Introduction ........................................................................................................................... 40
2. Background ........................................................................................................................... 43
3. Data ....................................................................................................................................... 47
4. Empirical Methods ................................................................................................................ 49
5. Results ................................................................................................................................... 52
6. Discussion ............................................................................................................................. 56

Chapter 3. Social Security and Retirement in Fast-Aging Middle-Income Countries:
Evidence from Korea .................................................................................................................. 65
Abstract ..................................................................................................................................... 65
I. Introduction ............................................................................................................................ 66
vi

2. Background ........................................................................................................................... 69
3. Specification of Incentive Measures ..................................................................................... 76
4. Data and Estimation of Variables .......................................................................................... 80
5. Estimation of Retirement Models ......................................................................................... 86
6. Conclusion ............................................................................................................................. 91
Appendix ................................................................................................................................. 105

Bibliography .............................................................................................................................. 109

vii

List of Tables

Chapter One
Table 1.1. Descriptive statistics at baseline………………………………………………......…35
Table 1.2. Difference-in-difference estimation: Effect of Medicare coverage on depressive
symptoms by socioeconomic status…………………………………………...…….………......36
Table 1.3. Difference-in-difference estimation: Effect of Medicare coverage on depressive
symptoms by socioeconomic status when controlling for financial strain and healthcare use.....37
Table 1.4. Robustness checks…………………………………………………….……...……...38
Table 1.5. Difference-in-difference estimation: Effect of Medicare coverage on depressive
symptoms by race/ethnicity…………………………………………………………...….…......39
Table A.1.1. Difference-in-difference estimates for pre-treatment periods……………………..40
Table A.1.2. Coefficients of treatment status interacted with age dummies……………..………41
Table A.1.3. Difference-in-difference estimation: Effect of Medicare coverage on depressive
symptoms by socioeconomic status………………………………………………..…….……...42

Chapter Two
Table 2.1. Descriptive statistics of the self-employed by country…………………...…………..69
Table 2.2. Regression coefficients for retirement on poor self-reported health among the self-
employed by country……………………….………………………………..……….…………70
Table 2.3. Regression coefficients for retirement on depressive symptoms among the self-
employed by country……….………………………………………………………..…….……71
Table 2.4. Regression coefficients for retirement on health using a younger sample aged 50
to 69…………………………………………………………………………………….…….…72

Chapter Three
Table 3.1. Demographic trends by country…..…………………………………………...……109
Table 3.2. History of the National Pension Service ……………………………………………110
Table 3.3. Descriptive statistics of male wage workers at baseline…………………...…..……111
Table 3.4. Estimated monthly SSW and incentive measures by age………………….….….…112
Table 3.5. Probability of retirement for male wage workers…………………….………..……113
Table 3.6. Robustness check………….…………………………………….……….…………114
Table 3.7. Probability of retirement by additional sources of income………….………………115
Table A.3.1. Probability of surviving……….………………….………………….……..……116
Table A.3.2. Net replacement rates and corresponding period……….…………………...……117
Table A.3.3. Estimated incentive measures by age………….………………….…………...…118
Table A.3.4. Probability of retirement for male wage workers (full estimates) …….………..119

viii

List of Figures

Chapter One
Figure 1.1. Trends in depressive symptoms among foreign-born HRS respondents 50 to 79
years of age…………………………………………………………………………..………….33
Figure 1.2. Low socioeconomic status by race/ethnicity among foreign-born HRS respondents
50 to 79 years of age………………………………………………………………….…………34

Chapter Two
Figure 2.1. Age trends of health and retirement of the self-employed by country………….……68

Chapter Three
Figure 3.1. Earnings profiles for male wage workers………………………………...……..….107
Figure 3.2. Change in retirement probability by age……………………………..………..…..108

ix

Abstract

Population aging comes with significant challenges. Preparing financially for longer lives and
finding ways to live a healthier life has become a priority for many adults and their families.
Governments are facing fiscal challenges as the aging population strains pension systems and
affects economic growth and disease patterns. This dissertation focuses on various challenges
encountered by the demographic change. The first chapter examines the foreign-born population
in the United States and addresses the fact that many older immigrants remain uninsured. The
study finds that Medicare coverage is associated with better mental health, particularly among
immigrants with low socioeconomic status. The second chapter compares the health of older self-
employed workers in three countries – Korea, Mexico, and the United States – and shows that
workers with poor health select into retirement, but the selection is weaker in countries with less
developed social security systems. The results suggest that older workers may involuntarily stay
in the labor force due to a lack of retirement income, further exacerbating their health and creating
higher costs to society. The last chapter focuses on older wage workers in Korea, a country with
the highest elderly poverty rate and elderly suicide rate, and finds that retirement choice of older
workers in Korea are influenced by transfers from children rather than pension benefits. This
contrasts with the experience in North America and Europe, further illustrating that rapidly aging
countries, such as Korea, may not have sufficient time and resources to tackle issues associated
with population aging. These chapters provide a rich new portrait of how the aging population is
transforming the world in fundamental ways and emphasize the need for policy reforms,
particularly for marginalized populations.

1
Introduction

Population aging can lead to multiple health problems. Disadvantaged populations have poorer
health and face higher barriers in accessing healthcare. Yet there has been little research on the
drivers of health disparities and the protective factors that could reduce related inequities.
Understanding the complex interplay of age, health, and population-specific characteristics will
become increasingly relevant for policymakers, who need solid empirical evidence to allocate
scarce resources and reduce disparities in health. This dissertation aims to provide critical evidence
that could fill the gaps in our knowledge of understanding the drivers of poor health and the
protective factors that could improve the wellbeing of older adults around the world.
Chapter One examines the effect of Medicare coverage on the mental health of older
immigrants in the United States. One of the protections that individuals seek for their health is
insurance—the lack of which can affect both physical and mental health. Immigrants are more
likely to be uninsured and are already at higher risks for psychological stress; however,
understanding how health insurance affects the mental health of immigrants has been an
underexplored topic. This study applies a difference-in-difference model with propensity score
weighting to a sample of foreign-born respondents in the Health and Retirement Study (HRS). The
study finds that Medicare coverage is significantly associated with a reduction in depressive
symptoms for older immigrants, especially for those with low socioeconomic status and for
racial/ethnic minorities. The paper further demonstrates that the psychological boost can be better
explained by improved financial protection than by an increase in healthcare utilization. The
results in this chapter suggest that expanding healthcare protection to older immigrants, who

2
already have higher risks of depression, can provide psychological relief, helping reduce their risks
of other medical conditions and hence other potential medical and social costs.
Chapter Two evaluates the health effect and selection effect of retirement on older self-
employed workers in the United States, Mexico, and South Korea. Another emerging group with
unmet health needs is older self-employed workers. In countries with insecure welfare systems for
the elderly, older workers may reluctantly stay in the labor force due to financial insecurity. This
could also be true for migrant workers and first-generation immigrants in the United States. This
study use data from the HRS, the Korean Longitudinal Study of Aging (KLoSA), and the Mexican
Health and Aging Study (MHAS) to explore the effect of retirement among the self-employed. To
address the fact that workers with poor health select into retirement, I use a bivariate probit model
and compare scenarios of when there is a considerable selection bias to the case when selection is
absent. In all three countries, I find that self-employed workers select into retirement if they report
having poor health or depressive symptoms, but the selection is relatively stronger in the U.S.
compared to Korea and Mexico. The result suggests that older self-employed workers with poor
health in Korea and Mexico may involuntarily stay in the labor force due to a lack of retirement
income, where many are ineligible for public pension benefits. The study also shows that the health
of the self-employed in Korea and Mexico deteriorates rapidly with age, but retirement rates are
relatively lower than those in the U.S., further supporting the possibility of workers with poor
health not being able to retire. The fact that the self-employed in these two countries already have
relatively poorer health and lower socioeconomic status at earlier ages implies that not retiring
could further exacerbate their health and create higher costs to the society in the long term. This
chapter suggests that countries experiencing population aging at a faster speed, such as Korea and

3
Mexico, have less time to prepare and implement policy reforms for improving the welfare of the
elderly and are bound to encounter new challenges.
Chapter Three analyzes the retirement incentives of older wage workers in Korea: a country
experiencing all aspects of demographic change at an exceedingly fast pace. The rapid shift has
led to severe problems such as the drastic increase in elderly poverty and elderly suicide rates.
Using six waves of the KLoSA, this study estimates dynamic measures of retirement incentives
based on a classical labor economic model. Estimation results show that old wage workers appear
to be more responsive to the dynamic interaction between earnings and pension benefits than to
the difference between current and future benefits. The results may indicate that older workers
perceive pension benefits alone as insufficient for financing their later lives. Relatedly,
supplementary income variables such as cash transfers from children are estimated to have
significant effects. This chapter finds that retirement incentives in Korea contrast with that in North
America and Europe, where public pensions provide relatively adequate support for post-
retirement life. The study illustrates that rapidly aging countries such as Korea may not have
sufficient time and resources to tackle issues associated with population aging.
Taken together, these chapters provide a rich new portrait of how the aging population is
transforming the world in fundamental ways and emphasize the need for policy reforms,
particularly for marginalized populations who are more likely to lack health insurance and
retirement income.

4
Chapter 1
The Value of Medicare Coverage on Depressive Symptoms
Among Older Immigrants

Abstract
The immigrant population, the primary driver of US population growth, is aging and many
immigrants remain uninsured. Lack of health insurance limits access to care, aggravating the
already high level of depression for older immigrants. Yet, there is scarce evidence on how health
insurance affects their mental health. Using the Health and Retirement Study, this study examines
the effect of Medicare coverage on depressive symptoms of older immigrants in the United States.
Exploiting the fact that many of the foreign-born are not covered by Medicare after passing age
65, this study uses a difference-in-difference model with propensity score weighting. Estimation
results show that Medicare significantly reduces depressive symptoms for immigrants with low
socioeconomic status and in racial/ethnic minorities. They also show that mental health
improvements are better explained by strengthened financial protection than by an increase in
healthcare utilization. These results imply that policies that expand healthcare protection to
immigrants can lead to further health benefits and reduce existing health disparities for the aging
population.

5
1. Introduction
Health insurance matters. Lack of health insurance leads to worse health outcomes (McWilliams,
Meara, Zaslavsky, & Ayanian, 2007a). Previous research has found that uninsured adults were
more likely to delay or forgo seeking medical care (Ayanian, 2000) and receive fewer preventive
services (Hadley, 2003). They were less likely to access mental health treatment as well (Wells,
Sherbourne, Sturm, Young, & Audrey Burnam, 2002). Lack of coverage also caused financial
strain due to high healthcare costs, especially for low-income households (Galbraith, Wong, Kim,
& Newacheck, 2005), which harms psychological health. Uninsured adults in their fifties and early
sixties suffer the most, as overall health declines with age.
For most Americans, circumstances change at age 65 because of Medicare. Medicare, one
of the largest public health insurance programs in the country, improved the health of its
beneficiaries by increasing their healthcare utilization (Card, Dobkin, & Maestas, 2008; Decker &
Rapaport, 2002b; McWilliams, Meara, Zaslavsky, & Ayanian, 2007b) and protection from medical
expenditure risk (Barcellos & Jacobson, 2015; Finkelstein & McKnight, 2008). The program
boosted overall physical health among uninsured older adults with health problems (McWilliams
et al., 2007a), improved control of blood pressure and glucose and cholesterol levels among adults
with relevant chronic conditions (McWilliams, 2009; McWilliams et al., 2007a), and lowered
death rates by increasing the number of procedures performed in emergency departments (Card,
Dobkin, & Maestas, 2009). The program covering prescription drugs, Medicare Part D, also
boosted mental health (Ayyagari & Shane, 2015) by improving the use and adherence of
antidepressants (Donohue et al., 2011).
The benefits of Medicare and other health insurance programs, however, are not available
for many immigrants. In 2021, 20 percent of the foreign-born were uninsured while 7 percent of

6
the native-born were, indicating that 30 percent of the 30 million uninsured population were born
abroad (US Census Bureau, 2021). For foreign-born adults below age 65, 46 percent of
undocumented immigrants, 25 percent of lawfully present immigrants, and 10 percent of
naturalized citizens lived without health insurance (Kaiser Family Foundation, 2021). Uninsurance
rates remain particularly high among low-income immigrants (Ku & Jewers, 2013) and those who
are racial/ethnic minorities (Williams, Priest, & Anderson, 2016). Among immigrants aged 65 and
older, many lack Medicare coverage. For example, in New York City where many elderly
immigrants reside, 18 percent of the elderly population lacked Medicare in 2001 (Gray,
Scheinmann, Rosenfeld, & Finkelstein, 2006).
Recent federal policies have further restricted opportunities for the foreign-born to gain
healthcare protection. The revised public charge policies in 2019 created new barriers to obtaining
legal permanent residency for immigrants who used public benefits, which led to a drop in
Medicaid enrollment (Tolbert, Artiga, & Pham). The earlier Personal Responsibility and Work
Opportunity Act (PRWORA) prevents legal immigrants from receiving public program coverage
in their first five years after arriving in the United States, which has led to lower Medicaid
enrollment and healthcare use (Vega, Porteny, & Aguila, 2018).
This is concerning because the foreign-born population is rapidly growing and aging.
Population projections report that the foreign-born will increase from 47 million in 2021 to 69
million in 2060; the share of the foreign-born among the total population will increase from 14
percent to 17 percent during that period (US Census Bureau, 2018). During the past decade, the
foreign-born population aged 65 and older increased by 51 percent, from 5 million in 2010 to 7.5
million in 2019, which is higher than the 30 percent increase of the native-born (US Census
Bureau, 2019). As people age, they become more susceptible to disease and disability. This implies

7
that there will likely be an increase in uninsured elderly immigrants, one of the most vulnerable
groups in the country.
Lack of health insurance could have its most adverse effects on the mental health of
immigrants. Immigrants are already at higher risk than others for psychological stress, especially
if they are older, arrived at later ages, and are racial/ethnic minorities. Many have gone through a
lifetime of acculturation, with an accumulation of stressful events leading to greater mental health
risks at older ages (Lum & Vanderaa, 2010). Immigrants are also more likely to suffer from
psychological distress due to language barriers and lower levels of family support (Aroian &
Norris, 2003; Coffman & Norton, 2010; Mui, 2000). This has been particularly true for those who
arrived at older ages (Angel, Buckley, & Sakamoto, 2001), and for those in minority racial/ethnic
groups (Buckley, Angel, & Donahue, 2000; De Maio, 2010; Lum & Vanderaa, 2010; Min, Moon,
& Lubben, 2005). A recent increase in anti-immigration sentiments have exacerbated symptoms
of depression and anxiety (Becerra et al., 2020). While the foreign-born have experienced better
physical health and mortality profiles than the native-born, as shown in the immigrant health
paradox, (Markides & Rote, 2015), there is limited or mixed research on whether the same paradox
applies to mental health (Alegría et al., 2008; Oh & Koyanagi, 2021).
There is also little research regarding the effects of Medicare on the physical and,
especially, mental health of immigrants. Poor mental health can significantly diminish the quality
of life for older adults by inducing higher medical costs and greater prevalence of other medical
conditions (Katon & Ciechanowski, 2002), including dementia and Alzheimer’s disease (Byers &
Yaffe, 2011). The potential protections of Medicare can be particularly important for immigrant
populations, especially for racial/ethnic minorities where many tend to have low socioeconomic
status. Most immigrants belong to a minority racial/ethnic group; in 2019, 44 percent of the

8
foreign-born were Hispanic, 27 percent were Asian, 17 percent were non-Hispanic White, and 10
percent were Black (US Census Bureau, 2019).
This study incorporates a quasi-experimental design to examine the association of
Medicare coverage on depressive symptoms. Using the Health and Retirement Study (HRS),
which includes rich information on immigration, we use a difference-in-difference model to
compare the health of the foreign-born with and without Medicare coverage from age 65, the post-
treatment period. Applying propensity score weights further creates a comparison group with
similar characteristics and we assume most of the sample were documented immigrants. We
analyzed individuals by socioeconomic status and by race/ethnicity to identify and assess groups
at higher risk for poor mental health. In addition to assessing the effects of Medicare among
Whites, Blacks, and Hispanics, we were able to assess its effects among Asians and Pacific
Islanders, an understudied population who represent a high proportion of the foreign-born.
Our results show that Medicare coverage significantly reduces the probability of reporting
depressive symptoms, especially for immigrants with low socioeconomic status. We found the
Medicare effects were strongest among older immigrants with below-median wealth and among
immigrants without health insurance before age 65. Analyses by race/ethnicity show that, among
immigrants, Medicare coverage is related with improvements in the mental health of Hispanics,
Blacks, and Asian/Pacific Islanders, but not of Whites, even when we control for household wealth
and educational attainment. Further analyses on the possible mechanisms for the Medicare effect
find that a decrease in financial strain better explains the reduction in depressive symptoms than
an increase in healthcare utilization.
This study contributes to the literature by presenting new evidence that Medicare coverage
reduces symptoms of depression among the foreign-born population, possibly because of the

9
financial protection it offers. The use of a quasi-experimental design to assess the impacts of
overall Medicare coverage is also notable. Since the universal nature of Medicare makes it difficult
to find a meaningful counterfactual, previous research has focused on evaluating specific time
periods or subprograms such as the period after the program was introduced in 1965 (Finkelstein
& McKnight, 2008) or Medicare Part D beneficiaries (Engelhardt & Gruber, 2011). The foreign-
born population presents a counterfactual group of persons without Medicare. Furthermore, among
documented immigrants, those with and without Medicare have many similarities, and because
immigrants undergo similar experiences, they may also be similar in unobservable characteristics.
1

Expanding Medicare coverage for older immigrants can increase overall health benefits
and reduce disparities in health for the aging U.S. population. U.S. health policymakers often focus
on undocumented immigrants, yet uninsured rates for legally documented immigrants are also
high. Documented residents may be ineligible for Medicare even after making qualifying
contributions because of the substantial backlog of lawful permanent residence applications, which
are required to receive premium-free Medicare Part A. Given the growth of the foreign-born
among the older population in the United States, current policies that restrict health-care protection
for immigrants need to be reexamined.

1
The HRS also contains information on language proficiency and the age of entry to the United States, which can be
controlled for in statistical analysis.

10
2. Background
Medicare Eligibility and Health Insurance Coverage of Immigrants
Medicare is the basic source of health coverage for Americans aged 65 and over.
2
The program
has four parts. Part A covers hospital and hospice services. Part B covers physician services. Part
D covers prescription medications. Part C, also known as Medicare Advantage, is offered by
private companies as an alternative to Parts A and B and sometimes Part D. Part A premiums are
waived for beneficiaries who paid Federal Insurance Contribution Act (FICA) taxes for at least 40
credits, equivalent to 10 years. Part B, C, and D are optional and need to be purchased.
Immigrants who paid payroll taxes and became U.S. citizens or lawful permanent residents
(also known as “green card” holders) are eligible to receive premium-free Medicare Part A.
3

Immigrants would most likely lack premium-free Medicare coverage if they (1) are on the wait
list to become a permanent resident, (2) did not contribute to the payroll tax system for at least ten
years, or (3) are undocumented. The wait time for a green card can range from none to decades.
4

In 2018, 28 percent of applicants in family and employment preference categories had waited a
decade or more for their green cards, and 41 percent had waited at least five years (Bier, 2019).
Not surprisingly, Medicare coverage for older immigrants has been strongly influenced by years
of residence, employment status, and birth country (Siddharthan, 1991). In some cases, immigrant
workers may not be eligible because they did not pay their FICA taxes. Immigrants in certain

2
Younger people entitled to Social Security Disability Insurance (SSDI), with amyotrophic lateral sclerosis (a.k.a.
Lou Gehrig’s disease), or with end-stage kidney failure also qualify for the program, as do spouses of recipients.
3
Citizens and permanent residents without the required contribution history can purchase Medicare Part A for $471
per month (as of 2021) after residing in the country for five years continuously.
4
Becoming a green card holder can depend on such variables as citizenship of family members, country of origin,
and type of job.

11
occupations, such as day laborers or the self-employed in agriculture, service, and construction
industries, are less likely to contribute to Medicare through payroll taxes (Ku & Jewers, 2013).
Recent information on Medicare coverage of the older foreign-born population is scarce.
A study of hospital discharge data, found that in New York City, where immigrants comprise
around 45 percent of the city’s population age 65 and over, 16 to 18 percent of the older population
did not have access to Medicare in 2001 (Gray et al., 2006). The authors further estimated that 5
percent of the older population in Texas, 8 percent in California, and 12 percent in Florida lacked
Medicare coverage. A paper analyzing the National Health Interview Survey from 1999 to 2003
found about 80 percent of Americans were covered with Medicare at age 65 (Card et al., 2009);
we can assume a considerable share of those without such coverage were immigrants.
Immigrants younger than 65, particularly noncitizens, are also more likely to be uninsured.
The 2019 American Community Survey showed that 9 percent of naturalized immigrants, 25
percent of legally documented immigrants, and 46 percent of undocumented immigrants younger
than 65 were uninsured (Kaiser Family Foundation, 2021). The 2012 Current Population Survey
found that among noncitizen immigrant adults only 21 percent had a private insurance and 19
percent had Medicaid (Ku & Jewers, 2013). Coverage status is also related to years of residence.
A study using the 2003 Medical Expenditure Panel Survey found that only half of immigrants
arriving in the preceding ten years had insurance coverage, while two-thirds of those who had
arrived more than ten years ago were insured (Ku, 2009).

Health Impacts of Medicare
Research on the health impact of Medicare has focused on mortality and physical health outcomes.
Medicare had no discernible overall impact on mortality in its first decade (Finkelstein &

12
McKnight, 2008), but analysis of California hospital records did find it reduced death rates by 20
percent among the severely ill (Card et al., 2009). Similarly, Medicare coverage did not improve
the general health of the overall uninsured population (Polsky et al., 2009), but it did significantly
improve physical health outcomes for adults with cardiovascular disease or diabetes (McWilliams,
2009; McWilliams et al., 2007a). Little research has discussed the impact of Medicare on the
foreign-born population.
Studies analyzing the mental health impacts of Medicare have focused on specific
programs. For instance, Medicare Part D was associated with increased use of and adherence to
antidepressants in depressed older adults (Donohue et al., 2011), and eventually a reduction in
depressive symptoms (Ayyagari & Shane, 2015). In 2011, Medicare included depression screening
as a component of an annual wellness visit program, a benefit for Part B beneficiaries, but this
benefit did not increase depression screening (Pfoh, Mojtabai, Bailey, Weiner, & Dy, 2015).
Medicare Shared Savings Program Accountable Care Organizations have also not increased use
of mental health and substance abuse services (Acevedo et al., 2021).
Nevertheless, research on Medicaid and other health insurance has found that increased
coverage can boost mental health. For example, data from the Oregon Health Insurance
Experiment indicated that Medicaid coverage significantly reduced the prevalence of undiagnosed
depression and improved symptoms of depression (Baicker, Allen, Wright, Taubman, &
Finkelstein, 2018). Medicaid expansion was also associated with a reduction in the share of adults
with depression, possibly through improved access to care and medication (Fry & Sommers,
2018). At the same time, among racial/ethnic minorities, depression rates were greatly associated
with lack of health insurance (Dunlop, Song, Lyons, Manheim, & Chang, 2003).

13
Socioeconomic Status and Depression
Immigrants with lower socioeconomic status have had a higher probability of being uninsured and
higher risks of depression (Hirsch et al., 2019). Socioeconomic status of the immigrant population
has been greatly contingent on years of residence. Statistics from the National Health Interview
Survey indicated that immigrants with less than five years of residence were more likely to have
earnings below the federal poverty level than were the native-born and the foreign-born with at
least five years of residence (Bustamante, Chen, Felix Beltran, & Ortega, 2021). Immigrants who
arrived at later ages have had fewer opportunities for assimilation and have been at higher risks
for poor physical and psychological health (Angel et al., 2001; Gubernskaya, 2015).
Socioeconomic status has also been related to language proficiency and non-English speaking
immigrants have been more likely to report depressive symptoms (Gonzalez, Haan, & Hinton,
2001; Mui, 2000). Immigrant women have experienced higher levels of stress due to greater lack
of social and economic resources (Buckley et al., 2000).

Conceptual Framework and Hypotheses
Previous research showed that the immigrant population had a higher probability of being
uninsured. It also showed that health insurance coverage, in general, was associated with better
mental health. This implies that a sudden increase in coverage at age 65 could improve the mental
health of older immigrants, particularly that of those who had been uninsured, as well as those of
lower socioeconomic status who are more likely to suffer from poor mental health, and those in
racial and ethnic minorities for whom low socioeconomic status is more prevalent. Stressors
associated with being a racial/ethnic minority could further exacerbate psychological distress

14
among immigrants (Buckley et al., 2000; De Maio, 2010; Lum & Vanderaa, 2010; Min et al.,
2005).
Our consideration of prior research suggests three hypotheses, which this study explored.
These are (1) among the foreign-born, Medicare coverage reduces symptoms of depression, and
(2) the beneficial effect of fewer depressive symptoms is greater for individuals with low
socioeconomic status as well as for (3) individuals of racial or ethnic minorities.

3. Data
Data Source
This study used the HRS, a nationally representative, longitudinal study of older adults aged 50
and over in the United States. The study contains a wide range of information on demographics,
socioeconomic status, health and healthcare utilization, family, income and wealth, employment,
retirement, and immigration status. The biennial survey started in 1992 and follows respondents
until death, enrolling new cohorts as needed. We used data from the second wave in 1994 through
that in 2016 because complete information on depressive symptoms was available from the second
wave.

Outcome Variable
We used a binary variable indicating symptoms of depression. The HRS contains a shorter version
of the Center for Epidemiologic Studies Depression Scale (CESD); scores on this range from 0 to
8, with a higher score indicating poorer mental health. We created a binary variable using a
validated cutoff score of 4 or greater as an indicator of depressive symptoms (Steffick, 2000; Zivin
et al., 2010).

15

Independent Variables
Covariates included demographic variables such as age and gender. We included household size,
older immigrants living with family members have had lower risks of depression (Aroian & Norris,
2003; Mui, 2000), and four educational attainment levels (below high school, high school graduate,
some college, college graduate and above). We included binary variables of whether the
respondent had a private health insurance plan or Medicaid. Because depression has often been
associated with other chronic diseases (Moussavi et al., 2007), we included a binary variable
indicating whether the individual had been diagnosed with at least one of the following: high blood
pressure, arthritis, diabetes, heart problem, cancer, lung problem, and stroke. We also included a
dummy variable indicating whether the individual had depressive symptoms before age 65 and
whether the respondent was currently working for pay. We controlled for immigrant-specific
variables such as English language proficiency and age of entry, and missing data for these two
variables were imputed using age, sex, race/ethnicity, education, and citizenship. As previous
research did, we divided age of entry into three categories: childhood (age 14 and younger), young
adulthood (age 15 to 34), and late adulthood (age 35 and older) (Angel et al., 2001).
To further examine the possible mechanisms of the Medicare effect, we compared
estimation results when controlling for total out-of-pocket health costs, the share of out-of-pocket
health costs over household income, and the number of doctor and hospital visits. We used the
Consumer Price Index to adjust all dollar values to year 2020 values.

16
Sample Definition
We analyzed foreign-born individuals 50 to 79 years of age before and after the Medicare
eligibility “treatment” age of 65. For each wave, HRS asked respondents whether they were
covered by Medicare. We categorized individuals with Medicare coverage at age 65 and older as
the treatment group and those without it as the comparison group. Because postponing Medicare
enrollment comes with penalties, we assumed all eligible immigrants were enrolled. We only
included respondents with complete information, yielding a sample of 13,558 person-year
observations. Among the respondents, 93 percent were in the treatment group, and 7 percent were
in the comparison group.
For the main analyses, we distinguished the sample by the following socioeconomic
characteristics: educational attainment (with or without a high school diploma), household income
(above or below the $34,695 median), household wealth (above or below the $121,540 median),
and health insurance coverage before age 65 (with or without). Among the total sample, 48 percent
did not have a high school diploma, 50 percent were below the median income and wealth level,
and 39 percent did not have health insurance before age 65.
Holding socioeconomic status constant, we also analyzed our model on different
race/ethnic groups: non-Hispanic White, non-Hispanic Black, Hispanic, and Asian/Pacific
Islander. Our Asian/Pacific Islander category is based on the Other HRS race category that
includes Asians, Pacific Islanders, Alaska Natives, and Native Americans. Because we were using
the foreign-born sample, we assumed this group only included Asians and Pacific Islanders. We
also assumed the probability of undocumented immigrants participating in the survey was low;
hence the sample was most likely restricted to legally documented immigrants.

17
4. Empirical Methods
Difference-in-Difference Analysis
We used a difference-in-difference model to estimate the effect of Medicare coverage on
depressive symptoms among older immigrants. Treatment status was based on whether the
individual had Medicare. We used information before (age 50-64) and after (age 65-79) the
Medicare eligibility threshold for the pre- and post-treatment periods. To estimate the treatment
effect of Medicare, we used the following equation:

(1) 𝑌
!"
=𝛽
#
+𝛽
$
𝑃𝑜𝑠𝑡65
"
+𝛽
%
𝑇𝑟𝑒𝑎𝑡
!
+𝛽
&
𝑃𝑜𝑠𝑡65
"
×𝑇𝑟𝑒𝑎𝑡
!
+𝛽
'
𝑋
!"
+τ
"
+𝜀
!",

where 𝑌
!"
is the outcome of interest with i denoting the individual and t denoting time/age,
𝑃𝑜𝑠𝑡65
"
is a binary variable indicating post-treatment status, 𝑇𝑟𝑒𝑎𝑡
!
is an indicator of the
treatment group, 𝑋
!"
is a set of control variables as described above, τ
"
are time dummies, and 𝜀
!"

is the error term. Standard errors were corrected for heteroskedasticity and serial correlation by
clustering them at the individual level. The estimated coefficient on the interaction term of the
post-treatment period and treatment status (𝛽
&
3
) represents the average difference in means between
the treatment and comparison group.
We used a linear probability model for the binary variable of reporting depressive
symptoms. We used linear models because our interest is in the difference in conditional means
for the treatment and comparison group. If our outcomes had a very high or low mean probability,
this would pose concerns (Angrist & Pischke, 2009; Finkelstein et al., 2012). As Table 1 indicates,
however, the average values of our outcome variables were not extreme.

18
The difference-in-difference estimator is unbiased if the parallel trend assumption holds:
in the absence of treatment, the difference between the treatment and comparison group remains
fixed over time. We tested the assumption using two different methods. First, we conducted a
placebo test using two pre-treatment periods (age 50-59 and age 60-64), assuming the treatment
occurred at age 60. Results showed that the treatment effect of Medicare at age 60 was insignificant
(Table A1, appendix), indicating the parallel assumption holds. The test is based on the theory that
individuals who have a history of a parallel trend are likely to follow the same trend in the next
period (Moffitt, 1991). This test has been applied in several previous studies (Bell, Blundell, &
Van Reenen, 1999; Chen, Chu, Wang, & Zhang, 2020; Wagstaff, 2010). Next, we tested the
assumption by checking the estimates of the treatment dummy interacted with each age dummy.
With the base being the interaction term of age 60, we found that the coefficients of the treatment
status interacted with age 50 to 62 were insignificant, meaning the treatment-comparison group
difference for each age was not different to that at age 60 (Table A2, appendix). The coefficients
turned significant from age 62 with stronger coefficients from age 65, implying a change in the
treatment-comparison gap near the Medicare eligibility age.

Propensity Score Weighting
While the treatment group and comparison group shared similar underlying average
characteristics, we applied propensity score weights to make them still more comparable. An
advantage of propensity score weighting over matching is that all individuals in the sample can be
used rather than only matched cases (Guo & Fraser, 2015). This makes propensity score weighting
more suitable for analyzing smaller samples such as we used. We applied inverse probability
weights to estimate the average treatment effect on the treated. We generated the propensity scores

19
using polynomial of age and years of education, gender, Medicaid status, private insurance status,
language proficiency, and the age of entry to the United States.

Robustness Check
For robustness tests, we ran analyses using the sample without propensity score weights, using
logit models, and using the full CESD scale and a continuous variable indicating the proportion of
years depressed as alternative outcome variables. We generated the proportion of years depressed
for an individual by dividing the number of waves in which the individual reported depressive
symptoms by the total number of waves available for each respondent, calculated separately for
each pre-treatment and post-treatment period.

5. Results
Descriptive Statistics
Table 1 presents summary statistics of the pre-treatment period for the treatment group (with
Medicare coverage) and comparison group (without Medicare coverage). Before age 65,
immigrants ineligible for Medicare had characteristics similar to those who were eligible, with
propensity score weighting further reducing the differences. The weighted sample showed that for
both groups more than half were Hispanic and around half did not graduate high school, did not
speak fluent English, and entered the country in young adulthood. About 10 percent in both groups
had Medicaid and more than 60 percent has private insurance. Medicare-eligible immigrants
reported higher average out-of-pocket medical expenditures before age 65 than did those not
eligible.

20
Both groups reported similar prevalence of depressive symptoms before age 65. There were
noteworthy differences in depressive symptoms between low and high socioeconomic status, with
those of lower educational attainment, income, and wealth being at least twice as likely as others
to report depressive symptoms. Those without insurance coverage before age 65 also reported
greater prevalence of depressive symptoms.
Figure 1 shows trends in reporting of depressive symptoms by age. At each age category,
immigrants without Medicare were more likely to report depressive symptoms. This gap widened
after the Medicare eligibility age. Immigrants without Medicare saw their depressive symptoms
increase; those with Medicare coverage saw little change in their depressive symptoms.
Figure 2 presents socioeconomic characteristics by race and Hispanic origin among HRS
foreign-born respondents. Non-Hispanic Whites have been most likely to have graduated high
school, to have income and wealth above median levels, and to have been insured. Hispanics have
been the most likely to have not graduated high school, to have income and wealth below median
levels, and to be uninsured. Blacks and Asian and Pacific Islanders are between these two
extremes.

Difference-in-Difference Treatment Effect of Medicare by Socioeconomic Status
Table 2 reports the difference-in-difference treatment effect of Medicare on the probability of
reporting depressive symptoms using the propensity score weighted sample. Each column
represents a different socioeconomic status group. Medicare coverage appeared to reduce
depressive symptoms across most subgroups, but the estimated coefficients were larger and
significant for immigrants with low socioeconomic status: reductions in depressive symptoms
were greatest for those below the median wealth threshold and those who had been uninsured

21
before age 65. Among those below the median wealth threshold, for example, getting covered by
Medicare reduced the probability of reporting symptoms of depression by 16 percentage points on
average relative to those without Medicare, equivalent to a 50% reduction (-0.16/0.32). Table A3
in the appendix reports estimated coefficients of the full covariates.
Table 3 presents estimation results when further controlling for out-of-pocket health costs
(panel A), the share of out-of-pocket health costs among household income (panel B), and the
number of doctor visits and hospital visits (panel C). Compared to those in Table 2, the estimated
coefficients in panel A and panel B of Table 3, where financial covariates were added, had smaller
values and lesser statistical significance. This implies the mental health improvements may be
responsive to financial variables as Medicare provides financial protection. On the other hand, the
estimated coefficients did not change greatly when we controlled for healthcare utilization (panel
C). Details of the possible channels need to be further explored in future studies.

Robustness Checks
Robustness checks, reported in Table 4, show that results using alternative models were consistent
with our primary analyses. We reported the difference-in-difference treatment effect of Medicare
using the unweighted sample (panel A), logit models (panel B), and the CESD score and proportion
of years depressed as outcome variables (panel C). All models incorporated control variables. As
in our main analysis, the results here showed that Medicare coverage significantly improved
mental health for immigrants with low socioeconomic status, and the impact was stronger for those
below median wealth levels and those without insurance before age 65.

22
Difference-in-Difference Treatment Effect of Medicare by Race/Ethnicity
As Figure 2 depicts, high proportions of racial and ethnic minorities have low socioeconomic status.
Even when we hold socioeconomic status constant, however, Table 5 shows that the treatment
effect was statistically significant for non-White immigrants but not for White immigrants. The
percentage point reduction in depressive symptoms was the greatest for Blacks (21 percentage
points), followed by Asian/Pacific Islanders (16 percentage points) and Hispanics (13 percentage
points).

6. Discussion
This study examined the effect of Medicare coverage on depressive symptoms of older immigrants
in the United States. As we hypothesized, Medicare coverage reduced the probability of reporting
depressive symptoms across most sample groups, but the estimated effects were mostly significant
for immigrants with low socioeconomic status. The pattern was consistent across different
socioeconomic categories. We believe the coefficients were insignificant for immigrants below
median income levels because income levels tend to be low for those who have retired. We also
found that Medicare coverage improved the mental health of non-White immigrants but not that
for Whites.
The magnitude of the effect ranged from 31 to 50 percent decline in the likelihood of
reporting depressive symptoms. This is larger than the effect of a similar study using the HRS that
found that Medicare Part D reduced the probability of experiencing three or more depressive
symptoms (our study was based on four or more) by 21 percent (Ayyagari & Shane, 2015). This
study, however, was mostly focused on the native-born where the majority had Medicare Parts A
and B, with the treatment effect being based on having an additional Part D. Another study found

23
that the Oregon Medicaid experiment reduced rates of depression by 30% among individuals who
were previously uninsured (Finkelstein et al., 2012). We believe that the higher magnitude of our
study results may reflect the differences in the treatment effect for the foreign-born, compared to
the native-born, as well as other changes that occur at age 65, such as Social Security income.
5

The Medicare effect on depressive symptoms may be explained by several mechanisms.
Financial protection is a possible channel. Medicare coverage lowers out-of-pocket health
expenditures (Finkelstein & McKnight, 2008) as well as financial strain (Barcellos & Jacobson,
2015), which could provide long-lasting psychological relief. Medical-expenditure burdens are
likely to weigh higher for uninsured or low-wealth individuals. Our estimation shows that financial
covariates, such as out-of-pocket health costs or the share of out-of-pocket health costs among
income, a proxy for financial strain, interacted more significantly with the Medicare effect for
underprivileged groups such as those uninsured before age 65 and those with lower wealth levels.
Another possible explanation is an increase in healthcare utilization; however, healthcare
use did not seem to explain the Medicare effect in our sample. Medicare coverage can boost
healthcare use, leading to better treatment and management of existing health conditions and better
health in general. Previous research has shown that the onset of Medicare led to an increase in
doctor visits for uninsured adults with underlying health conditions (McWilliams et al., 2007b)
and an increase in hospital admissions related to expensive procedures (Card et al., 2008).
Medicare also increased preventive treatments, such as mammogram screening, among racial and
ethnic minorities with lower educational attainment (Decker & Rapaport, 2002a). Yet, this may
not be the case for the foreign-born. Previous research found that, even among the insured, foreign-
born adults tend to use less healthcare than the native-born, mostly as a result of language barriers,

5
Immigrants who were eligible for Medicare were also eligible for other Social Security benefits, such as old-age
pension income, while those who were ineligible for Medicare were ineligible for these other benefits as well.

24
cultural differences, and concerns that seeking care could make it harder to gain citizenship or
permanent residency (Ku & Jewers, 2013; McBride et al., 2020; Prus, Tfaily, & Lin, 2010). Our
estimates indicate that the relief in depressive symptoms was mostly a result of greater financial
protection. Future research may explore this topic further.

Policy Implications
Many immigrants do not have Medicare or other health insurance coverage. While recent discourse
on immigrant health insurance coverage is focused on undocumented residents, uninsured rates
for naturalized citizens and lawfully present residents are also high (Kaiser Family Foundation,
2021). Anti-immigration policies have further restricted the ability of legal immigrants to gain
healthcare protection and have also increased financial insecurity (Vega & Aguila, 2017; Tolbert
et al., 2019).
The decline in federal support for immigrants has shifted the burden of immigrant health
coverage to state and local safety-net providers (Bustamante et al., 2021). Illinois was the first state
to extend health coverage to low-income older immigrants not eligible for Medicaid due to
immigration status (Illinois Department of Healthcare and Family Services, 2021). Recently,
California announced an expansion of Medi-Cal coverage to low-income undocumented
immigrants over age 50 (Caiola, 2021). Many legally documented immigrants, however, live in
other states, have income levels above the Medicaid eligibility threshold, or are otherwise not
eligible for such programs.
Expanding health insurance coverage for legally documented residents may be a matter of
fairness. Immigrants have a high labor force participation rate, comprising 17.4 percent of the total
labor force in 2019 (Bureau of Labor Statistics, 2020), and many contribute to the payroll tax

25
system (Mueller, 2021). In 2009, immigrants accounted for 14.7 percent of Medicare Trust Fund
contributions but only 7.9 percent of its expenditures (Zallman, Woolhandler, Himmelstein, Bor,
& McCormick, 2013), further suggesting that many immigrants pay taxes but are ineligible for
public welfare benefits. Among tax-paying immigrants, those who lack Medicare coverage are
most likely on the long waitlist to become a lawful permanent resident.
The increase in immigrants among the elderly population provides implications for a
change in Medicare eligibility requirements. Incremental reforms, such as providing limited
Medicare access to immigrants who have paid sufficient FICA taxes but are still awaiting
permanent residency status, could be a feasible solution. Coverage options could be tailored to
years of residence, FICA tax credits, specific health conditions, or income levels. Such policy
reform would increase coverage rates for uninsured older persons and encourage participation of
immigrants in the payroll system, even bolstering the sustainability of Medicare’s financial health.
Increased Medicare coverage for uninsured older adults can yield substantial long-term
benefits as the population ages. Poor mental health at older ages can worsen management of
physical health conditions, leading to an increase in emergency room usage with higher costs
(Katon & Ciechanowski, 2002). Depression is also known to increase risks of dementia (Byers &
Yaffe, 2011). Mental illnesses have considerable indirect costs, such as productivity loss and
caregiving costs (Knapp, 2003). Older immigrants with worsening physical, mental, and cognitive
health may exhaust their personal savings and qualify for Medicaid (Borella, De Nardi, & French,
2018), creating additional burdens on state budgets.

26
Limitations
Our study has several limitations. First, we have a small sample size like other immigrant-focused
studies, with a smaller share of Asian and Pacific Islanders. Only 9 percent of our sample are Asian
and Pacific Islanders while US Census estimates show that around 27 percent of the foreign-born
population are (US Census Bureau, 2019). The share of Hispanic and Blacks in our sample are
similar to population estimates. We hope a larger dataset will be available for future research.
Second, we do not distinguish different types of Medicare, whereas the health effect of having Part
A coverage may differ from having additional coverage options. Part B enrollment will matter
since depression care is mostly office-based; however, the financial protection effect applies to
other types of care as well indicating the benefit of having at least Part A will differ from having
no Medicare coverage. Third, we do not control for cultural factors that could lead to different
perceptions towards mental health. The foreign-born population is heterogeneous, and some
immigrants may not seek mental health treatment because of the stigma associated with it or even
consider depression to be a treatable illness. We assume control variables such as language
proficiency, race/ethnicity, educational attainment, and age of arrival will control for such hidden
variation. Fourth, although we are aiming for causality, older Americans are exposed to many
changes at age 65, such as retirement or Social Security benefits, that could challenge our
estimates. We believe these factors are partially controlled by the covariates, such as income levels,
and that our estimates reflect causal inference, based on our model specification and study sample.
Lastly, our sample is most likely restricted to legally documented immigrants. We assume the
Medicare effect will also apply to undocumented residents because most such residents have low-
income levels, poor health, and are uninsured (Bustamante et al., 2021). Policy interventions aimed

27
at undocumented immigrants, however, will differ from those targeting documented residents and
need further exploration.

Conclusion
Approximately 47 million immigrants reside in the country today and this population is rapidly
aging. Older immigrants, especially those with low socioeconomic status or among racial and
ethnic minorities, face additional stressors and are prone to poorer mental health than native-born
adults. Policy reforms aimed at expanding Medicare coverage options for these immigrants can
yield long-term health benefits and reduce existing health disparities.

28
Figure 1.1. Trends in depressive symptoms among foreign-born HRS respondents 50 to 79 years
of age

24%
22%
19%
19%
20%
29%
27% 27%
31%
36%
10%
15%
20%
25%
30%
35%
40%
50-59 60-64 65-69 70-74 75-79
% reporting depressive symptoms
Age
With Medicare coverage Without Medicare coverage

29
Figure 1.2. Low socioeconomic status by race/ethnicity among foreign-born HRS respondents 50
to 79 years of age

27%
46%
70%
24%
26%
47%
69%
37%
21%
56%
71%
36%
18%
38%
54%
30%
White Black Hispanic Asian/Pacific Islander
Did not graduate high school Below median income
Below median wealth Uninsured before age 65

30
Table 1.1. Descriptive statistics at baseline
Unweighted Sample Propensity Score Weighted Sample

With Medicare
coverage
Without
Medicare
coverage
With Medicare
coverage
Without
Medicare
coverage
Male 39.3% 44.4% 39.2% 39.8%
Age 59.1 (3.6) 59.3 (3.7) 59.1 (3.6) 59.0 (3.7)
Household size 3.0 (1.7) 3.5 (1.8) 3.0 (1.7) 3.4 (1.6)
Race
White 29.9% 17.4% 29.9% 26.0%
Black 10.2% 13.2% 10.2% 12.2%
Hispanic 50.2% 63.5% 50.2% 55.6%
Asian/Pacific Islander 9.7% 5.9% 9.7% 6.3%
Education
High school and below 47.2% 56.7% 47.2% 48.3%
High school graduate 18.2% 15.1% 18.2% 12.8%
Some college 16.0% 13.2% 16.0% 17.1%
College graduate and above 18.6% 15.1% 18.6% 21.8%
Household income (2020US$) 86,843 (211,240) 64,733 (92,715) 86,843 (211,240) 81,814 (102,009)
Household wealth (2020US$)
429,544
(1,191,527)
456,117
(3,110,635)
429,544
(1,191,527)
697,113
(3,932,314)
At least one chronic condition 66.1% 62.9% 66.1% 64.0%
Depressed before age 65 45.3% 49.3% 45.3% 42.9%
Medicaid 9.5% 6.4% 9.5% 10.2%
Private insurance 61.7% 53.3% 61.7% 61.3%
Currently working 57.6% 60.9% 57.6% 58.4%
Poor English 49.0% 58.2% 49.0% 50.4%
Age of entry to the U.S.

<= 14 years old 29.3% 18.3% 29.3% 33.2%
15-34 years old 48.7% 45.2% 48.7% 47.3%
>= 35 years old 22.0% 36.5% 22.0% 19.5%
Out-of-pocket cost (2020US$) 3,617 (14,020) 2,534 (4,989) 3,617 (14,020) 2,449 (5,299)
No. of doctor visits 8.2 (16.1) 6.3 (11.5) 8.2 (16.1) 7.8 (13.8)
No. of hospital visits 0.3 (1.0) 0.2 (0.9) 0.3 (1.0) 0.2 (0.9)

Depressive symptoms by socioeconomic status
< High school grad. 0.32 (0.47) 0.39 (0.49) 0.32 (0.47) 0.39 (0.49)
≥ High school grad. 0.16 (0.36) 0.14 (0.34) 0.16 (0.36) 0.11 (0.32)
< Median income 0.35 (0.48) 0.42 (0.49) 0.35 (0.48) 0.41 (0.49)
≥ Median income 0.15 (0.36) 0.13 (0.34) 0.15 (0.36) 0.12 (0.32)
< Median wealth 0.32 (0.47) 0.34 (0.48) 0.32 (0.47) 0.33 (0.47)
≥ Median wealth 0.14 (0.35) 0.16 (0.36) 0.14 (0.35) 0.13 (0.34)
Uninsured before age 65 0.28 (0.45) 0.36 (0.48) 0.28 (0.45) 0.33 (0.47)
Insured before age 65 0.19 (0.39) 0.17 (0.38) 0.19 (0.39) 0.19 (0.39)
No. of observations 5,175 427 5,175 427
Cells indicate mean or percentages of the sample for the pre-treatment period (age 50-64). Standard errors are in
parentheses.

31
Table 1.2. Difference-in-difference estimation: Effect of Medicare coverage on depressive symptoms by socioeconomic status

Standard errors are in parentheses and clustered at the individual level. All models use the propensity-score-weighted sample and include control variables of the
following: age, gender, household size, educational attainment levels, private health insurance status, Medicaid status, chronic condition status, whether
depressed before age 65, work status, English language proficiency, and age of entry.
*p<0.10, **p<0.05, ***p<0.01.

Uninsured Insured
before age
65
before age
65
Depressive symptoms -0.06 -0.10* -0.01 -0.08 0.02 -0.16*** 0.05 -0.14** -0.03
(0.05) (0.06) (0.08) (0.05) (0.06) (0.05) (0.07) (0.06) (0.06)
No. of observations 13,558 6,474 7,084 6,828 6,730 6,648 6,910 5,242 8,316
< Median
wealth
≥ Median
wealth
Total
Education Income Wealth Health Insurance
< High
school grad.
≥ High
school grad.
< Median
income
≥ Median
income

32
Table 1.3. Difference-in-difference estimation: Effect of Medicare coverage on depressive symptoms by socioeconomic status when
controlling for financial strain and healthcare use

Standard errors are in parentheses and clustered at the individual level. All models use the propensity-score-weighted sample and include control variables of the
following: age, gender, household size, educational attainment levels, private health insurance status, Medicaid status, chronic condition status, whether
depressed before age 65, work status, English language proficiency, and age of entry.
*p<0.10, **p<0.05, ***p<0.01.

Uninsured Insured
before age
65
before age
65
Depressive symptoms -0.06 -0.08 -0.03 -0.07 0.02 -0.15*** 0.04 -0.13** -0.04
(0.05) (0.06) (0.08) (0.06) (0.06) (0.05) (0.06) (0.06) (0.06)
No. of observations 12,745 6,078 6,667 6,537 6,208 6,233 6,512 4,898 7,847
Depressive symptoms -0.04 -0.03 -0.02 -0.03 0.02 -0.13** 0.04 -0.07 -0.04
(0.05) (0.06) (0.08) (0.06) (0.06) (0.06) (0.06) (0.08) (0.06)
No. of observations 12,455 5,859 6,596 6,247 6,208 5,979 6,476 4,681 7,774
Depressive symptoms -0.08 -0.11* -0.04 -0.09 0 -0.16*** 0.02 -0.17** -0.04
(0.05) (0.06) (0.08) (0.06) (0.06) (0.06) (0.07) (0.07) (0.06)
No. of observations 12,802 5,993 6,809 6,254 6,548 6,145 6,657 4,917 7,885
Panel A. Controlling for out-of-pocket health cost
Panel B. Controlling for the share of out-of-pocket health costs over household income
Panel C. Controlling for the number of doctor visits and hospital visits
Health Insurance
< High
school grad.
≥ High
school grad.
< Median
income
≥ Median
income
< Median
wealth
≥ Median
wealth
Total
Education Income Wealth

33
Table 1.4. Robustness checks

Panel A reports the difference-in-difference treatment effect of Medicare using the unweighted sample while panels B and C reports the treatment effect using the
propensity-score-weighted sample. Panel B reports estimated odds ratios of a logit model and panel C uses the CESD scale and proportion of years depressed as
outcome variables. Standard errors are in parentheses and clustered at the individual level. All models include control variables of the following: age, gender,
household size, educational attainment levels, private health insurance status, Medicaid status, chronic condition status, whether depressed before age 65, work
status, English language proficiency, and age of entry.
*p<0.10, **p<0.05, ***p<0.01.

Uninsured Insured
before age
65
before age
65
Depressive symptoms -0.05 -0.12** 0.04 -0.06 0 -0.15*** 0.09* -0.1 -0.04
(0.04) (0.06) (0.04) (0.05) (0.05) (0.05) (0.05) (0.07) (0.05)
No. of observations 13,562 6,474 7,088 6,832 6,730 6,651 6,911 5,246 8,316
Depressive symptoms -0.46 -0.60* -0.1 -0.55* 0.19 -0.92*** 0.54 -0.94** -0.25
(0.32) (0.32) (0.72) (0.31) (0.66) (0.31) (0.72) (0.37) (0.48)
No. of observations 13,558 6,474 7,084 6,828 6,730 6,648 6,910 5,242 8,316
CESD -0.29 -0.43 -0.18 -0.33 0.11 -0.77** 0.21 -0.58 -0.28
(0.29) (0.38) (0.43) (0.36) (0.32) (0.33) (0.35) (0.36) (0.34)
Years depressed -0.07 -0.12* -0.01 -0.09 0.01 -0.15*** 0.03 -0.16*** -0.04
(0.05) (0.06) (0.08) (0.06) (0.05) (0.05) (0.06) (0.06) (0.06)
No. of observations 13,558 6,474 7,084 6,828 6,730 6,648 6,910 5,242 8,316
Panel B. Logit Model
Panel C. Alternative Outcome Variables
≥ High
school grad.
< Median
income
≥ Median
income
< Median
wealth
≥ Median
wealth
Panel A. Unweighted Sample
Total
Education Income Wealth Health Insurance
< High
school grad.

34
Table 1.5. Difference-in-difference estimation: Effect of Medicare coverage on depressive
symptoms by race/ethnicity

White Black Hispanic
Asian/Pacific
Islander
Depressive symptoms 0.06 -0.21** -0.13** -0.16***
(0.06) (0.08) (0.06) (0.05)
No. of observations 4,499 1,245 6,666 1,148
Standard errors are in parentheses and clustered at the individual level. All models use the propensity-score-
weighted sample and include control variables of the following: age, gender, household size, household wealth,
educational attainment levels, private health insurance status, Medicaid status, chronic condition status, whether
depressed before age 65, work status, English language proficiency, and age of entry.
*p<0.10, **p<0.05, ***p<0.01.

35
Appendix

Table A.1.1. Difference-in-difference estimates for pre-treatment periods

Cells indicate the difference-in-difference treatment effect of Medicare for immigrants 50 to 64 years old in the propensity-score-weighted sample, assuming the
treatment occurred at age 60 (i.e., placebo test). Standard errors are in parentheses and clustered at the individual level. All models use the propensity-score-
weighted sample and include control variables of the following: age, gender, household size, educational attainment levels, private health insurance status,
Medicaid status, chronic condition status, whether depressed before age 65, work status, English language proficiency, and age of entry.
*p<0.10, **p<0.05, ***p<0.01.

Uninsured Insured
before age
65
before age
65
Depressive symptoms -0.01 -0.01 -0.01 0 -0.03 -0.05 0.03 -0.02 0
(0.03) (0.05) (0.04) (0.05) (0.04) (0.04) (0.06) (0.05) (0.04)
No. of observations 5,602 2,601 3,001 2,225 3,377 2,807 2,795 2,633 2,969
< Median
income
≥ Median
income
< Median
wealth
≥ Median
wealth
Total
Education Income Wealth Health Insurance
< High
school grad.
≥ High
school grad.

36
Table A.1.2. Coefficients of treatment status interacted with age dummies
Depressive symptoms
D × age 50 0.04 (0.06)
D × age 51 0.03 (0.05)
D × age 52 0.03 (0.04)
D × age 53 0.01 (0.03)
D × age 54 0.02 (0.03)
D × age 55 0.04 (0.03)
D × age 56 0.05* (0.03)
D × age 57 -0.01 (0.02)
D × age 58 -0.03 (0.02)
D × age 59 -0.03 (0.02)
D × age 61 -0.03 (0.02)
D × age 62 -0.05*** (0.02)
D × age 63 -0.06*** (0.02)
D × age 64 -0.09*** (0.02)
D × age 65 -0.12*** (0.02)
D × age 66 -0.10*** (0.02)
D × age 67 -0.11*** (0.02)
D × age 68 -0.09*** (0.02)
D × age 69 -0.13*** (0.02)
D × age 70 -0.13*** (0.02)
D × age 71 -0.16*** (0.03)
D × age 72 -0.10*** (0.03)
D × age 73 -0.14*** (0.03)
D × age 74 -0.14*** (0.03)
D × age 75 -0.17*** (0.03)
D × age 76 -0.15*** (0.03)
D × age 77 -0.12*** (0.04)
D × age 78 -0.14*** (0.04)
D × age 79 -0.14*** (0.04)
No. of observations 13,562
Cells indicate coefficients of Medicare status (D) interacted with each age dummy with the base being age 60.
Standard errors are in parentheses. All models include control variables of the following: age, gender, household
size, educational attainment levels, private health insurance status, Medicaid status, chronic condition status,
whether depressed before age 65, work status, English language proficiency, and age of entry. *p<0.10, **p<0.05,
***p<0.01.

37
Table A.1.3. Difference-in-difference estimation: Effect of Medicare coverage on depressive symptoms by socioeconomic status

Uninsured Insured
before age
65
before age
65
Post 65 ´ Treatment -0.06 -0.10* -0.01 -0.08 0.02 -0.16*** 0.05 -0.14** -0.03
(0.05) (0.06) (0.08) (0.05) (0.06) (0.05) (0.07) (0.06) (0.06)
Post 65 0.02 0.04 0.00 0.03 -0.05 0.06 -0.05 0.07 -0.02
(0.04) (0.07) (0.06) (0.06) (0.06) (0.06) (0.06) (0.07) (0.05)
Treatment -0.05* -0.08* -0.01 -0.09** -0.02 -0.08** -0.03 -0.07 -0.04
(0.03) (0.04) (0.03) (0.04) (0.03) (0.04) (0.04) (0.05) (0.03)
Age 0.00 0.01 0.00 0.01 0.00** 0.01*** 0.00 0.01* 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Male -0.11*** -0.14*** -0.07** -0.17*** -0.02 -0.15*** -0.05* -0.14*** -0.09***
(0.03) (0.04) (0.03) (0.04) (0.03) (0.04) (0.03) (0.04) (0.03)
Household size 0.01 0.00 0.01 0.00 0.01 0.01 0.00 0.01 0.01
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
High school grad -0.02 -0.03 0.00 -0.05 0.06 -0.08* 0.03
(0.05) (0.06) (0.04) (0.05) (0.04) (0.04) (0.06)
Some college -0.01 0.00 0.04 0.00 -0.01 0.05 0.02 -0.02
(0.03) (0.03) (0.05) (0.03) (0.04) (0.03) (0.05) (0.04)
College grad and above -0.06 -0.05 -0.10 -0.02 -0.10** -0.01 -0.01 -0.06
(0.04) (0.03) (0.06) (0.03) (0.05) (0.03) (0.03) (0.04)
Depressed before 65 0.29*** 0.31*** 0.26*** 0.29*** 0.28*** 0.34*** 0.23*** 0.34*** 0.24***
(0.03) (0.03) (0.05) (0.03) (0.03) (0.03) (0.04) (0.03) (0.04)
Chronic condition 0.06*** 0.08*** 0.02 0.08** 0.02 0.07** 0.02 0.08** 0.03
(0.02) (0.03) (0.02) (0.03) (0.03) (0.03) (0.02) (0.03) (0.02)
Medicaid 0.08* 0.10** -0.04 0.05 0.09 0.05 -0.03 -0.01 0.15***
(0.04) (0.04) (0.06) (0.04) (0.06) (0.04) (0.07) (0.05) (0.06)
Private insurance -0.11*** -0.15*** -0.08* -0.16*** -0.02 -0.07** -0.07** -0.06** -0.12***
(0.03) (0.04) (0.04) (0.04) (0.02) (0.03) (0.03) (0.03) (0.04)
Currently working -0.01 0.05 -0.05** 0.02 -0.03 -0.02 -0.03 -0.02 0.00
(0.02) (0.03) (0.02) (0.03) (0.02) (0.03) (0.02) (0.03) (0.02)
Poor English 0.07** 0.07 0.08 0.10** 0.03 0.05 0.07* -0.01 0.13***
(0.03) (0.05) (0.05) (0.05) (0.03) (0.04) (0.04) (0.04) (0.04)
Health Insurance
< High
school grad.
≥ High
school grad.
< Median
income
≥ Median
income
< Median
wealth
≥ Median
wealth
Total
Education Income Wealth

38

Standard errors are in parentheses and clustered at the individual level. All models use the propensity-score-weighted sample and include time dummies.
*p<0.10, **p<0.05, ***p<0.01.
Age of entry: 15-34 -0.04 -0.02 -0.03 -0.03 -0.01 -0.02 -0.04 0.01 -0.06*
(0.03) (0.04) (0.05) (0.04) (0.03) (0.04) (0.04) (0.05) (0.04)
Age of entry: >= 35 -0.06 -0.03 -0.08 -0.07 0.00 -0.05 -0.04 -0.03 -0.07
(0.04) (0.05) (0.07) (0.05) (0.04) (0.04) (0.05) (0.05) (0.05)
Constant 0.03 -0.19 0.16 0.01 -0.15 -0.27 0.06 -0.18 0.06
(0.15) (0.27) (0.13) (0.23) (0.14) (0.23) (0.14) (0.28) (0.15)
No. of observations 13,558 6,474 7,084 6,828 6,730 6,648 6,910 5,242 8,316

39
Chapter 2
Health and Retirement of the Older Self-Employed:
Evidence from Korea, Mexico, and the United States

Abstract
Previous research shows that retirement improves health. Whether this applies to the self-
employed is not clear. Self-employment work differs significantly from wage and salaried work,
and retirement incentives could also vary. Retirement behaviors are also different in countries with
poor social security systems for the self-employed. Using harmonized panel datasets, this study
examines the health effect and selection effect of retirement on older self-employed workers in the
United States, Mexico, and South Korea. To address the fact that workers with poor health select
into retirement, we use a bivariate probit model and compare scenarios of a considerable selection
bias to the case when selection is absent. Results show that, in all three countries, self-employed
workers select into retirement if they report having poor health or depressive symptoms. Assuming
that the actual effect of retirement on health is positive, the selection is relatively stronger in the
U.S. compared to Korea and Mexico. The result suggests that older self-employed workers with
poor health in Korea and Mexico may involuntarily stay in the labor force due to a lack of
retirement income and other welfare benefits. The fact that the self-employed in these two
countries already had relatively poorer health and lower socioeconomic status at earlier ages
implies that not retiring could further exacerbate their health and create higher costs to the society
in the long term. Countries need to reform social security systems to support the self-employed
better, especially with the aging population.

40
1. Introduction
With the aging population, people live longer and work longer. Life expectancy has improved
dramatically across the world; in 2019, life expectancy at birth was 83.3 in South Korea, 78.9 in
the U.S., and 75.1 in Mexico (OECD, 2022a). Labor force participation rates for older adults aged
65 and older have also increased, with 35 percent, 24 percent, and 19 percent of the elderly in the
labor force in 2020, in Korea, Mexico, and the U.S., respectively (OECD, 2022b). Workers staying
longer in the labor force is a fiscally attractive option for policy makers as it can reduce social
security benefits and increase tax revenues (Baily, 2019), however, delaying retirement could
come with hidden costs if retiring from the labor force improves health.
Estimating the causal effect of retirement on health is difficult. Workers with poor health
tend to select into retirement, causing concerns for endogeneity. Earlier studies often found a
negative correlation between retirement and health (Behncke, 2012; Dave, Rashad, & Spasojevic,
2008). Yet, papers that further controlled for endogeneity, using instruments or a regression
discontinuity design, mostly found that retirement was beneficial to health (Belloni, Meschi, &
Pasini, 2016; Charles, 2004; Coe & Zamarro, 2011; Eibich, 2015; Gorry, Gorry, & Slavov, 2018;
Johnston & Lee, 2009). These papers, however, were mostly focused on wage and salaried workers,
and we do not know if the same selection bias and health effect apply to the self-employed.
Self-employment increases with age (Blanchflower, Oswald, & Stutzer, 2001). The current
demographic trend implies that there will be an increase in older self-employed workers in the
future, especially for aging countries that already have a high share of self-employed workers. For
example, Mexico and Korea are experiencing a rapid demographic change (OECD, 2022a). These
countries are also the third and sixth-highest OECD countries with the share of self-employed
workers, respectively; in 2019, 31.9 percent of workers in Mexico and 24.6 percent in Korea were

41
self-employed (OECD, 2022b). The proportion of the self-employed is expected to be higher for
older workers in these countries.
Self-employment work differs greatly from wage work and retirement effects could differ,
yet most retirement studies focus on wage and salaried workers. Self-employed workers
experience higher levels of job satisfaction (Blanchflower & Oswald, 1998; Bradley & Roberts,
2004), but also greater job-related stress from increased uncertainty and higher job demand
(Blanchflower, 2004; Jamal, 1997; Lewin-Epstein & Yuchtman-Yaar, 1991). Retirement
incentives may also differ where details on pension benefits or employer-sponsored health
insurance could influence the retirement choice of wage workers (Coile & Gruber, 2007; French
& Jones, 2011). Closing a business also comes with higher risks and loss of assets (Lewin-Epstein
& Yuchtman-Yaar, 1991), making it more difficult for the self-employed to retire.
In addition, research on the self-employed has been mostly focused on cases of North
American and European countries. The selection bias may differ across countries with different
institutional arrangements because the self-employed in many low-and-middle-income countries
lack social security benefits, making it more difficult to retire. For instance, many of the self-
employed in Korea and Mexico are not covered by public pension plans and have to rely on their
own savings (Aguila, Fonseca, & Vega, 2015; KOSIS, 2022). Analyzing the health effects of
retirement among the self-employed will be of greater importance to countries with high rates of
elderly self-employed workers.
This study examined the health effect and selection effect of retirement on self-reported
health and depressive symptoms of older self-employed workers in three different countries –
Mexico, Korea, and the U.S. For cross-country comparison, we used harmonized panel data from
the Health and Retirement Study (HRS), the Korean Longitudinal Study of Aging (KLoSA), and

42
the Mexican Health and Aging Study (MHAS). We first used a pooled logit model to estimate the
association and further controlled for the potential selection into retirement using a bivariate probit
model. The bivariate probit model assumes an error structure that allows for correlation between
the two equations of health and retirement. We compared different scenarios of the size of the
correlation to evaluate how the estimates for retirement on health changed: from no selection to a
relatively stronger selection of workers with poor health going into retirement. Although we do
not know the true value of selection in real life and are cautious to claim causality of our estimates
of retirement on health, controlling for potential correlation on unobservable variables helped us
estimate the relationship between retirement and health when the selection bias was addressed in
various magnitudes.
Our estimation results show that the negative association between retirement and health
disappeared once controlled for a relatively small value of unobserved heterogeneity, implying
that self-employed workers with poor health select into retirement. The selection effect was present
for both self-reported health and depressive symptoms. Yet, under the same assumptions of the
size of the selection, the beneficial effect of retirement was relatively stronger for mental health
than self-reported health. On average, retirement provided psychological relief even when there
were no improvements in overall health. If we assume that the true effect of retirement on health
is positive, as found in recent literature, the selection was relatively stronger in the U.S. compared
to Korea and Mexico. The results suggests that older self-employed workers in Korea and Mexico
may not be able to retire, despite having poor health, due to the lack of pension income and other
welfare benefits. The fact that health among older self-employed workers in Korea and Mexico
deteriorates rapidly with age and that labor force participation rates of the older self-employed in

43
these countries are much higher than the OECD average, further supports the possibility of workers
with poor health not being able to retire.
This study contributes to the literature by presenting new evidence that self-employed
workers with poor health select into retirement and that selection differs across countries with
different institutional arrangements for the self-employed. Unlike other self-employment studies,
we analyzed two unique countries, Korea and Mexico, with a large share of older self-employed
workers but a social security system still maturing. We found that self-employed workers in these
countries had poorer initial health and lower socioeconomic status, implying that individuals with
lower income and wealth levels select into self-employment work. The lack of safety-nets could
force older self-employed workers with poor health to stay in the labor market, further
exacerbating their health and creating greater costs in the future. With the aging population, there
is a possibility of a shift in the workforce towards self-employment for older workers, implying a
strong need for policy reform in public pension plans and other welfare systems for the self-
employed.

2. Background
Retirement and Health
Retirement affects overall health through various channels. Retiring from the workforce could be
a liberating experience that lessens anxiety and psychological distress (Drentea, 2002). Retirees
are more likely to report lower levels of work-related stress and higher levels of sleep duration and
engage in regular physical exercise (Eibich, 2015; Midanik, Soghikian, Ransom, & Tekawa, 1995).
However, if retirement happens abruptly (Szinovacz & Davey, 2004), involuntarily (Dave et al.,
2008), or due to ill-health (Vo et al., 2015), it could be associated with an increase in psychological

44
distress. Socioeconomic status also matters. A study found that retirement had health
improvements only if the worker belonged to a high socioeconomic status group (Mein, 2003).
Health, however, also influences retirement choices, causing concerns for a selection bias
(Dwyer & Mitchell, 1999). Several studies found that retiring was harmful to health. A paper
examining older workers in England found that retirement significantly increased physical health
risks and reports of poor self-assessed health (Behncke, 2012). Another study stratified the sample
to workers that did not have a health problem prior to retirement and found that retirement
worsened mental health, possibly through declines in physical activity and social interactions
(Dave et al., 2008).
Recent papers addressing the potential endogeneity showed that retirement had beneficial
health effects. Several papers used the variation in Social Security and pension rules as instruments,
either locally or across countries, and found that retirement improved mental health (Belloni et al.,
2016; Charles, 2004) and self-reported health (Coe & Zamarro, 2011) as well as life satisfaction
(Gorry et al., 2018). Studies have also applied regression discontinuity designs and found that
retiring from the labor force improved wellbeing and mental health (Eibich, 2015; Johnston & Lee,
2009).

Self-Employment and Health
Self-employment work differs from wage and salaried work in many aspects and could have
various impacts on mental and physical health. Self-employed workers in the U.S. and U.K. tend
to have higher job satisfaction than wage workers, explained by higher levels of self-efficacy
(Blanchflower & Oswald, 1998; Bradley & Roberts, 2004). Business owners also experience less

45
role ambiguity and role conflict (Tetrick, Slack, Da Silva, & Sinclair, 2000) and even engage in
healthier behaviors compared to wage workers (Yoon & Bernell, 2013).
The self-employed, however, also have higher levels of job demand that could lead to
health problems (Jamal, 1997). Many business owners report that they are constantly under strain
and tend to place more weight on work than they do on leisure (Blanchflower, 2004). Business
owners are also burdened by higher uncertainty and market fluctuations that impose significant
levels of stress (Lewin-Epstein & Yuchtman-Yaar, 1991). Self-employed workers also often lack
health insurance (Zissimopoulos & Karoly, 2007). Higher work-related stress among the self-
employed could lead to greater psychological relief and health improvements when retired.

Cross-country Differences of the Self-Employed
Self-employment work is heterogeneous and could differ in many aspects across countries. To
start with, self-employment tends to be preferred over wage work in developed countries, including
the U.S., but preference rates are much lower in other countries (Blanchflower et al., 2001;
Blanchflower & Oswald, 1998). In the U.S., self-employed men, on average, had similar or higher
levels of education and higher median earnings than non-self-employed men, and were most likely
to work in management, business, science, and arts occupations (Christnacht, Smith, & Chenevert,
2018). Work preferences may be different, however, in low-and-middle-income countries. A study
analyzing 74 developing countries found that as per capita income increased across countries, the
structure of employment shifted from agricultural work into non-agricultural self-employment and
then into wage and salaried jobs (Gindling & Newhouse, 2014).
Institutional differences may further explain the differences in work preferences across
countries. For example, the old-age pension system in the U.S. covers most of the self-employed,

46
while coverage rates of the self-employed in Korea are much lower and started to cover all small-
sized businesses from 2006.
6
In Mexico, social security is not mandatory for the self-employed,
and many workers in the informal sector are only entitled to 70 y más that provides unconditional
cash transfers to the poorest individuals aged 70 and older (Aguila et al., 2015). Health insurance
is another factor. All citizens are covered by the National Health Insurance system in Korea
regardless of work type (KOSIS, 2022). Medicare covers American self-employed workers from
age 65, but those younger than 65 need to purchase private health plans. In Mexico, contributions
to health care services are not mandatory for the self-employed, and many are covered under
Seguro Popular, which provides health services to the poor and uninsured (Aguila et al., 2015).
Immigration status within the U.S. may also be important; a study found that, unlike non-Hispanics
Whites, Mexican-born immigrants in the U.S. where not influenced by age, education levels, and
Medicare eligibility when making retirement decisions (Aguila, Lee, & Wong, 2021).

Hypotheses
Therefore, based on previous research, this study explored the following three hypotheses: (1)
older self-employed workers in Korea and Mexico depicts poorer health outcomes compared to
those in the U.S.; (2) self-employed workers with poor health select into retirement; and (3) the
selection effect is stronger in the U.S. compared to Korea and Mexico because the U.S. has a
relatively more mature social security system for the self-employed than the other two countries.

6
The program provided coverage to all workers from 2006, but actual participation rates of the self-employed were
relatively low: 58% in 2006 and 70% in 2015 (KOSIS, 2022).

47
3. Data
Data Source
This study used 12 biennial waves of the RAND HRS from 1994 to 2016, 6 biennial waves of the
Harmonized KLoSA (v.C) from 2006 to 2016, and 4 waves of the Harmonized MHAS (v.A) for
year 2001, 2003, 2012, and 2015. HRS is a nationally representative, longitudinal study of older
adults aged 50 and over in the United States and contains a wide range of information on
demographics, socioeconomic status, health, family, income and wealth, employment, and
retirement. Complete information on depressive symptoms is available from the second wave.
KLoSA and MHAS are family surveys of the HRS, and the harmonized versions contain variables
comparable to the HRS, allowing for cross-country comparison.
7

Outcome variable
To measure the overall health or wellbeing of the individual, we constructed a binary variable
indicating poor self-reported health. We assumed self-reported health was poor if the response was
either ‘poor’ in HRS and MHAS and ‘poor’ or ‘very poor’ in KLoSA. Self-reported health was
not poor, or equal to zero, if the response was either ‘fair,’ ‘good,’ ‘very good,’ or ‘excellent.’
To analyze the impact of retirement on mental health, we constructed a binary measure
indicating symptoms of depression. All three datasets contain a country-specific version of the
Center for Epidemiologic Studies Depression (CESD) scale, a list of questions asking about the
respondent’s feelings which could indicate depression and anxiety; however, the range of the scale

7
We used data from the Harmonized dataset and codebooks developed by the Gateway to Global Aging Data. For
more information, please refer to https://g2aging.org/. Variables not available in the harmonized dataset were
constructed using the original KLoSA and MHAS dataset.

48
varies across datasets. Therefore, we used previous research that validated a threshold for showing
depressive symptoms for each dataset and created a binary variable that equals one if the
respondent reports depressive symptoms and equals zero otherwise. In the HRS, the scale ranges
from 0 to 8, with a higher score indicating poorer health, and a score of 4 or higher is an indicator
of depressive symptoms (Steffick, 2000; Zivin et al., 2010). In the MHAS, the scale ranges from
0 to 9 and a score of 5 or higher is an indicator of depressive symptoms (Aguilar-Navarro, Fuentes-
Cantú, Ávila-Funes, & García-Mayo, 2007). KLoSA contains a CESD scale that ranges from 0 to
30 where the cutoff point for depressive symptoms is 10 (Ichimura et al., 2017).

Independent variables
For each dataset, we used a binary variable that equals one if the respondent is retired and equals
zero if currently working for pay. Because our sample consisted of individuals with work
experience from at least age 50, we assumed those who responded to not work for pay in later
waves were retired. Covariates included demographic factors such as age and age-squared, gender,
marital status (1= whether married or partnered, 0= otherwise), and socioeconomic factors such as
years of education and household income levels in tercile groups. Income groups were created
including wage workers and respondents that did not work. We also controlled for health insurance
status
8
that could influence both health and retirement. When the outcome was depressive
symptoms, we added self-reported health as a covariate. Employment characteristics included
weekly work hours, job type (1= white-collar job, 0= blue collar job), the type of industry
9
, and

8
Health insurance in HRS and MHAS indicates whether the respondent had any type of insurance coverage (either
public or private). In KLoSA, it indicates whether the individual had a supplementary private insurance plan because
the national health insurance system covers all Korean citizens.
9
We categorized industry by the following sectors: primary, secondary, and tertiary (Rietveld, van Kippersluis, &
Thurik, 2015). Primary includes agricultural, forestry, fishery, mining, and construction work. Secondary includes
manufacturing durable and non-durable goods. Tertiary includes the rest of the industrial categories.

49
whether it was a small-sized business.
10
Information on the industry type and business size was
not available in the harmonized versions of KLoSA and MHAS and was applied using the original
dataset.
For the endogenous model on retirement status, which is used to estimate the bivariate
probit model as described in the next section, we included binary measures on whether the
respondent received any pension benefits and whether he/she had any chronic health conditions of
the following: cancer, diabetes, heart problem, lung problem, stroke, hypertension, and arthritis.

Sample Definition
For each country, we analyzed individuals aged 50 and older with self-employment experience.
We dropped respondents without any work experience from age 50 and those with wage and
salaried work experience. Therefore, our sample included workers who were only self-employed
throughout their later lives, from at least age 50. We only included observations with complete
information. This resulted in a sample size of 13,905 person-year observations for HRS, 7,418
observations for KLoSA, and 6,987 observations for MHAS.

4. Empirical Methods
Pooled logit model
We first use a pooled logit model to estimate the association of retirement on health among the
older self-employed. The basic form of the regression is as below:

10
In HRS and KLoSA, a business was small if the number of employees was less than 10. In MHAS, a business was
assumed to be small based on the location (e.g., on the street, on a ranch/farm, in a vehicle or small stand, or a home
or store versus in a medium or large factory, farm, or other location).

50
(1) 𝑃(𝐻
!"
=1)=𝛽
#
+𝛽
$
𝑅𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡
!"
+𝛽
%
𝑋
!"
+τ
"
+𝜀
!"
,

where 𝐻
!"
is the binary measure of health (i.e., poor self-reported health or depressive symptoms)
for individual i at period t, 𝑅𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡
!"
is the binary measure of being retired, 𝑋
!"
are the control
variables as described above, τ
"
are time dummies, and 𝜀
!"
is the error term. Standard errors are
corrected for heteroskedasticity and serial correlation by clustering them at the individual level.
Analyses were conducted separately for each country.
As described in the background section, our main concern with the pooled logit model is
the possibility of endogeneity. If individuals with poor health select into retirement, the distribution
of self-employed workers who choose to work or retire will not be random, and our estimates will
be biased.

Bivariate probit model
Therefore, we apply a bivariate probit model to further correct for the potential selection bias.
Although we are cautious to claim that our estimated coefficients using the bivariate probit model
are the true causal effects of retirement on depressive symptoms, we use this model to check
whether there is a selection into retirement among the older self-employed and how the health
effect changes depending on the magnitude of the selection.
The bivariate probit model belongs to the class of Heckman’s simultaneous equation
models (Heckman, 1978), which was further expanded as a recursive model for dichotomous
choice by Maddala (Maddala, 1983). It is used when both the outcome and endogenous treatment
variable are binary, and the parameters are estimated using the maximum likelihood method. The

51
model assumes an error structure that allows for correlation between the two equations of interest
which, in our case, is:
(2) 𝐻
!"
=𝛼𝑅𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡
!"
+𝛽𝑋
!"
+𝜀
!"
,
(3) 𝑅𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡
!"
=𝛾𝑍
!"
+𝜇
!"
.

Equation (2) equals equation (1) but is written in a shorter form for simplicity. This
equation contains both exogenous and potentially endogenous variables. Equation (3) is the
regression equation of retiring where we assume 𝑍
!"
contains exogenous variables influencing the
choice of retirement. The identification of the model is based on exclusion restrictions, where at
least one of the reduced-form exogenous variables (i.e., 𝑍
!"
) must not be included in the structural
equation as explanatory variables (i.e., 𝑋
!"
) (Maddala, 1983). Although more recent studies state
that the exclusion restrictions are not strictly required (Wilde, 2000), to be conservative, we add
additional measures to equation (3) that could explain retirement choice: pension status and
whether the respondent has a chronic health condition.
The main benefit of this method is that it is possible to constrain the correlation of the error
terms of retirement and health (i.e., 𝜀
!"
and 𝜇
!"
) to generate estimates under different scenarios as
done by previous studies (Altonji, Elder, & Taber, 2005; Rietveld et al., 2015). It is assumed that
the error terms of the two models are jointly distributed as standard bivariate normal with a mean
0, variance 𝜎
)
%
and 𝜎
*
%
, and correlation coefficient of 𝜌
)*
. Any unobserved characteristics that
could influence both health and retirement status are captured through the correlated error terms.
To elaborate, r = 0 corresponds to exogeneity or no selection bias while assuming r is a
positive value indicates individuals with poor health (P(H|X)=1) are more likely to retire

52
(P(Retirement|Z)=1) or vice versa. Since we do not know the true value of r in practice, restricting
the correlation to varying values allow us to analyze the health effect of retirement when a range
of selection on unobserved variables is considered and further provides an informal way to assess
the selection effect of continued work among the older self-employed.

Sensitivity Analysis
For sensitivity analysis, we checked the estimates using a younger sample. Selection into
retirement is assumed to be larger among older workers since older age is associated with more
health problems. Therefore, we analyzed our model using a younger sample, aged 50 to 69, to see
whether the estimates were consistent with the total sample.

5. Results
Descriptive Statistics
Figure 1 presents the age trends of reporting depressive symptoms and poor self-reported health
as well as trends in retirement among the self-employed in the U.S., Korea, and Mexico. The
middle figure shows that self-employed workers in Mexico had the highest probability of reporting
depressive symptoms for each age group, followed by Korea. Self-reported health was the poorest
in Mexico in younger age groups but was exceeded by Korea as the sample aged. The average
share of reporting poor health was constant over age among the self-employed in the U.S. The rate
of increase in poor health overtime was the fastest among Korean self-employed workers. The
right-hand side figure further shows that for each age group, a relatively higher number of business

53
owners in the U.S. retired on average, followed by those in Mexico. Average retirement rates were
the lowest among self-employed workers in Korea.
Table 2 further presents descriptive statistics of the self-employed aged 50 and older for
each country. Older self-employed workers in Korea and Mexico had lower socioeconomic status
than those in the U.S. Average years of education varied greatly across countries: 12.9 years in the
U.S., 8.5 years in Korea, and 4.2 years in Mexico. Many American self-employed workers
belonged to the higher income tercile group while the opposite was shown for Mexican self-
employed workers. Similarly, 91% of business owners in the U.S. were insured, while only 65%
of self-employed workers in Mexico had any type of health insurance coverage. Differences were
also shown in the type of self-employment work. While 58% of American self-employed work
were white collar jobs, only 33% in Korea and 24% in Mexico were. In the U.S., 22% of self-
employment work was in the primary sector, while 34% and 27% were in Korea and Mexico,
respectively. Business owners in Korea and Mexico also showed a tendency to work longer weekly
hours on average than those in the U.S. For all three countries, the majority of older self-employed
workers had a small-sized business.

The Health Effect and Selection Effect of Retirement on Health
Table 2 presents regression coefficients of retirement on poor self-reported health using a pooled
logit model and a bivariate probit model. In the bivariate probit model, we constrained the
correlation coefficient (𝜌) of the error terms of retirement and poor health to the noted values to
control for potential selection. Estimates of the pooled logit model showed that for all three
countries, retirement was strongly associated with an increase in the probability of reporting poor

54
health, holding other variables constant. Odds ratios were the highest in the U.S., followed by
Korea and Mexico.
The estimated coefficients changed when we controlled for potential selection. The first
row of the bivariate probit model reported the estimated coefficients of retirement on poor health
when we assumed the correlation was zero. When we assumed selection bias was absent, the
direction of the coefficients remained positive, similar to the pooled logit model. However, when
we assumed the error terms were positively correlated, i.e., workers with poor health selected into
retirement, the coefficients became insignificant and even changed directions. If 20 to 30 percent
of the error terms were correlated, the estimated coefficients became negative in Korea and Mexico.
This suggests that a relatively small correlation between the unobserved factors of retirement and
poor health already accounts for the entire positive association. Based on past literature that found
that retirement improves health, the change in directions of the coefficients implies that the
seemingly positive association between retirement and poor health turned out to be a result of a
selection effect where workers with poorer health selected into retirement.
Interestingly, the coefficients in the U.S. required a relatively stronger correlation between
the error terms to become negative compared to Korea and Mexico. If the true health effect of
retiring is beneficial, as found in the literature, this indicates that there is a stronger selection of
workers with poor health going into retirement in the U.S. compared to Korea and Mexico. In other
words, some self-employed workers in Korea and Mexico may not be able to retire, despite having
poor health. As shown in Figure 1, the self-employed in Korea and Mexico had relatively poorer
health, but labor force participation rates for each age group were much higher than those in the
U.S., further supporting the possibility of workers with poor health not being able to retire in these
two countries.

55
Table 3 further presents the regression coefficients of retirement on depressive symptoms.
Similar to self-reported health, a small correlation between the unobserved factors of retirement
and poor health showed to account for the entire positive association, indicating a selection of
individuals with poor mental health being pushed into retirement. The selection effect was
relatively smaller for depressive symptoms than self-reported health. For instance, in the U.S., it
only required a correlation of 20 percent for the estimates to turn negative for depressive symptoms
compared to a correlation of 40 percent in self-reported health. This also suggests that the
beneficial effect of retirement could be relatively larger for mental health than self-reported health
under the same amount of selection. If we assumed that the correlation coefficient was 30 percent,
for example, the odds ratios of retirement on depressive symptoms were -0.27, -0.19, and -0.36 for
the U.S., Korea, and Mexico, respectively, while the odds ratios of retirement on poor self-reported
health were 0.00, -0.09, and -0.18, respectively.

Sensitivity Analysis
We ran regressions on younger age groups for sensitivity analysis, as presented in Table 4. Older
age is associated with more health problems; therefore, the selection bias is assumed to increase
with older age, i.e., individuals that stay in the labor force will be more likely to have good health.
Panel A reported the estimates when the outcome variable was poor self-reported health and panel
B reported results when the outcome was depressive symptoms. The overall magnitude and
direction of the coefficients using a younger age group were similar to the estimates using the total
sample.

56
6. Discussion
This study analyzed the health effect and selection effect of retirement using a sample of older
self-employed workers in three different countries and provided empirical evidence that self-
employed workers with poor health select into retirement. Our results showed that retirement and
poor health had a positive association. However, the association disappeared when a relatively
small correlation between the unobserved factors of retirement and poor health was accounted for.
Based on previous literature, retirement is known to improve health. Hence, the change in
directions of the coefficient indicates the possibility of workers with poor health selecting into
retirement.
The results indicate that maintaining initial health matters if workers prefer to stay longer
in the labor force. With the aging population creating fiscal constraints and slower economic
growth, governments will want to provide incentives to workers to postpone retirement (Baily,
2019). Our results show that improving the health of older workers is essential in encouraging
them to stay longer in the workforce.
Another aim of this study was to compare the effects across countries with different
infrastructures for the self-employed. Our results found that the overall direction of the selection
effect and health effect of retirement were similar across countries, yet the selection was relatively
stronger in the U.S. compared to Korea and Mexico, as hypothesized. The results suggest that older
self-employed workers in Korea and Mexico have lower flexibility to retire and may stay in the
labor force even with poor health. High elderly labor force participation rates further support this
argument where participation rates for those aged 65 and older were 35.3 percent in Korea and
24.0 percent in Mexico in 2020, much higher than the U.S. (19 percent), the OECD average (15
percent), and the European Union average (6 percent) (OECD, 2022b). Employment rates of the

57
older self-employed in our sample were also higher in Korea and Mexico than in the U.S., as shown
in Figure 1.
Our results further suggest that policies aimed at older workers in developed countries may
not apply to those in countries with social security systems that are in the developing stage. In less
developed countries, the primary focus should be on increasing the participation of self-employed
workers into the social security system; participation rates of the self-employed in Korea and
Mexico are much lower than those of wage workers (Aguila et al., 2015; KOSIS, 2022). Increasing
pension recipients and improving access to healthcare for the self-employed will provide
psychological and physical health benefits in the long term. With sufficient retirement income,
older workers with poor health could choose to retire and this will improve health by helping them
relieve work-related stress, engage in better sleep, and by increasing physical exercise (Eibich,
2015). Compared to wage workers, however, monitoring the participation of small-sized business
owners is more difficult since many tend to evade taxes (Joulfaian & Rider, 1998). Establishing
incentives, such as subsidies or tax credits, could encourage participation.
Another noteworthy fact is that the self-employed in Korea and Mexico had poorer health
initially, on average, compared to those in the U.S., implying that individuals with lower
socioeconomic status and worse health could be selected into self-employment. In these countries,
preferences towards self-employment work are relatively lower than wage work (Blanchflower et
al., 2001), where wage and salaried work tends to be less risky and is sometimes an indicator of
higher social status (Gindling & Newhouse, 2014). Also, a large share of self-employment work
in Korea and Mexico is based on physically draining work related to agriculture, forestry, and
fishery: 31 percent in Korea, 26 percent in Mexico, and 11 percent in the U.S. were working in
that area, according to our sample.

58
Different trends in unemployment and self-employment across countries further show that
socioeconomically less advantaged individuals could be pushed into opening their own businesses.
Many countries in North America and Europe depict a negative relationship between the rate of
self-employment and unemployment; in other words, self-employment increases when there is an
economic boom (Blanchflower & Oswald, 1998). Interestingly, the opposite trend is shown in
developing countries; this could be explained by the fact that labor market frictions in poorer
countries make wage and salaried jobs harder to find, causing high unemployment and also
promoting self-employment as an alternative (Poschke, 2019). Relatedly, a study found that a large
number of workers in Korea were pushed into self-employment during the financial crisis due to
an increase in unemployment rates rather than a voluntary transition resulting from
entrepreneurship (Kim & Cho, 2009). This further indicates that if older self-employed workers,
who already have poor health, are not able to retire due to the lack of pension income and other
welfare benefits, it will further harm their health and create greater costs to society in the long term.
Our study comes with limitations. First, depressive symptoms and self-reported health are
based on subjective evaluations rather than a clinical diagnosis and could be subject to
measurement errors. We believe cultural differences across these countries could affect the way
people respond to specific questions. For example, a study found that Latinos and Asians were
more likely to self-report poor health than non-Hispanic Whites due to different cultural
perceptions of health (Kandula, Lauderdale, & Baker, 2007; Shetterly, Baxter, Mason, & Hamman,
1996). Although limited, we tried to address the potential measurement error by using datasets
with harmonized variables. Interpreting the difference in trends, or the marginal change in health,
also shows that the health of the self-employed deteriorates over time in Korea and Mexico but
not in the U.S. Related to that, our model does not capture unobserved heterogeneity across

59
countries such as cultural or societal preferences for self-employment work, however, we try to
incorporate cross country differences in our interpretations. In addition, our study does not show
the true amount of selection of poor workers going into retirement nor the true health effect of
retiring. Our study rather focuses on providing new evidence that there is a high possibility of self-
employed workers with poor health selecting into retirement and that the selection differs across
countries. Future work could focus on analyzing the causal health effect across countries using
different methods. Finally, we do not distinguish different types of self-employed. With a limited
sample size, we were not able to run models separately for different types of work, however, we
controlled for a number of variables that could explain the variation in self-employment work,
including industry types, job types, working hours, and business size. Using a larger sample in the
future may allow such analyses.
Future work should continue to examine the effect of working and retiring among self-
employed workers in different institutional arrangements, possibly with a richer dataset. Studies
focused on low-and-middle-income countries are lacking, despite that these countries have high
shares of self-employed workers. Understanding the relationship between retirement and health
among the self-employed will help policymakers around the world to prepare for an aging society.

60
Figure 2.1. Age trends of health and retirement of the self-employed by country

The sample does not include individuals with wage work experience or those who were not working for
pay from age 50.

0 10 20 30 40 50 60 70 80 90 100
50 55 60 65 70 75 80 85
Age
Poor self-reported health(%)
0 10 20 30 40 50 60 70 80 90 100
50 55 60 65 70 75 80 85
Age
Depressive symptoms(%)
0 10 20 30 40 50 60 70 80 90 100
50 55 60 65 70 75 80 85
Age
Retirement(%)
USA Korea
Mexico

61
Table 2.1. Descriptive statistics of the self-employed by country
United States Korea Mexico
Demographic & Socioeconomic variables
Male 58% 71% 57%
Age 66.66 (9.25) 64.40 (9.16) 64.15 (9.28)
Years of education 12.92 (3.14) 8.52 (4.54) 4.22 (3.90)
Married/partnered 76% 85% 70%
Non-White 17% NA NA
Foreign-born 10% NA NA
Health insurance 91% 37% 65%
Household income

Low 22% 32% 40%
Medium 30% 33% 34%
High 48% 35% 26%
Employment variables

Retired 40% 22% 32%
Small-sized business 90% 99% 97%
White collar job 58% 33% 24%
Industry

Primary sector 22% 34% 27%
Secondary sector 6% 11% 4%
Tertiary sector 72% 55% 69%
Working hours

0-10 19% 5% 7%
11-30 35% 19% 31%
31-50 31% 39% 37%
51+ 15% 37% 25%
Health variables

Poor self-reported health 0.05 (0.21) 0.16 (0.37) 0.15 (0.36)
Depressive symptoms 0.10 (0.31) 0.20 (0.40) 0.31 (0.46)
N, person-year obs. 13,905 7,418 6,987
N, individuals 2,058 1,563 2,707
Noted are the mean or percentages with standard deviation in parenthesis. Health insurance for the U.S. and
Mexico indicates whether the respondent has any type of coverage while for Korea it indicates whether the
respondent has supplementary private insurance.
Source: RAND HRS (1994-2016), Harmonized KLoSA (2006-2016), Harmonized MHAS (2000-2015).

62
Table 2.2. Regression coefficients for retirement on poor self-reported health among the self-
employed by country
United States Korea Mexico
Pooled logit model

Retired 1.15*** (0.13) 0.74*** (0.10) 0.55*** (0.07)
N, person-year obs. 13,896 7,418 6,987

Bivariate probit model

𝜌 = 0.00 0.53*** (0.05) 0.43*** (0.04) 0.33*** (0.05)
𝜌 = 0.10 0.36*** (0.05) 0.26*** (0.04) 0.16*** (0.05)
𝜌 = 0.20 0.19*** (0.05) 0.08* (0.04) -0.01 (0.04)
𝜌 = 0.30 0.01 (0.04) -0.09** (0.04) -0.18*** (0.04)
𝜌 = 0.40 -0.16*** (0.04) -0.26*** (0.04) -0.35*** (0.04)
N, person-year obs. 13,489 7,414 6,013
Noted are the estimated coefficients for being retired on the probability of reporting poor health. The standard
errors are in parenthesis and clustered at the individual level. Bivariate probit models are estimated with the
correlation (𝜌) set to the noted values. All models include control variables of the following: age and age-
squared, gender, years of education, marital status, health insurance status, race and birth-place (for the US
only), whether it’s a small-sized business, whether it’s a white collar job, income tercile groups, industry type,
weekly work hours, and time dummies. *p<0.10, **p<0.05, ***p<0.01.

63
Table 2.3. Regression coefficients for retirement on depressive symptoms among the self-
employed by country
United States Korea Mexico
Pooled logit model

Retired 0.45*** (0.09) 0.51*** (0.09) 0.24*** (0.06)
N, person-year obs. 13,896 7,418 6,987

Bivariate probit model

𝜌 = 0.00 0.23*** (0.04) 0.30*** (0.04) 0.13*** (0.04)
𝜌 = 0.10 0.07* (0.04) 0.13*** (0.04) -0.03 (0.04)
𝜌 = 0.20 -0.10*** (0.04) -0.02 (0.05) -0.20*** (0.04)
𝜌 = 0.30 -0.27*** (0.03) -0.19*** (0.05) -0.36*** (0.04)
𝜌 = 0.40 -0.44*** (0.03) -0.36*** (0.04) -0.53*** (0.04)
N, person-year obs. 13,489 7,414 6,013
Noted are the estimated coefficients for being retired on the probability of reporting depressive symptoms when the
correlation (𝜌) is set to the noted values. The standard errors are in parenthesis and clustered at the individual level.
All models include control variables of the following: age and age-squared, gender, years of education, marital
status, health insurance status, whether report poor self-reported health, race and birth-place (for the US only),
whether it’s a small-sized business, whether it’s a white collar job, income tercile groups, industry type, weekly
work hours, and time dummies. *p<0.10, **p<0.05, ***p<0.01.

64
Table 2.4. Regression coefficients for retirement on health using a younger sample aged 50 to 69
United States Korea Mexico
Panel A. Poor self-reported health
Pooled logit model

Retired 1.06*** (0.17) 0.96*** (0.14) 0.61*** (0.09)
N, person-year obs. 8,683 5,211 5,053

Bivariate probit model
𝜌 = 0.00 0.49*** (0.06) 0.52*** (0.06) 0.37*** (0.06)
𝜌 = 0.10 0.33*** (0.06) 0.34*** (0.06) 0.21*** (0.06)
𝜌 = 0.20 0.16*** (0.06) 0.16*** (0.06) 0.04 (0.06)
𝜌 = 0.30 -0.02 (0.06) -0.01 (0.06) -0.13** (0.05)
𝜌 = 0.40 -0.19*** (0.05) -0.19*** (0.06) -0.30*** (0.05)
N, person-year obs. 8,521 5,209 4,289

Panel B. Depressive symptoms
Pooled logit model
Retired 0.49*** (0.11) 0.55*** (0.12) 0.25*** (0.08)
N, person-year obs. 7,610 3,479 3,089

Bivariate probit model
𝜌 = 0.00 0.25*** (0.05) 0.31*** (0.06) 0.13*** (0.05)
𝜌 = 0.10 0.08* (0.05) 0.14** (0.06) -0.04 (0.05)
𝜌 = 0.20 -0.08* (0.04) -0.04 (0.06) -0.20*** (0.05)
𝜌 = 0.30 -0.25*** (0.04) -0.22*** (0.06) -0.37*** (0.05)
𝜌 = 0.40 -0.42*** (0.04) -0.39*** (0.06) -0.54*** (0.05)
N, person-year obs. 7,301 3,477 2,744
Noted are the estimated coefficients for being retired on the probability of reporting poor health (panel A) and
depressive symptoms (panel B). The standard errors are in parenthesis and clustered at the individual level.
Bivariate probit models are estimated with the correlation (𝜌) set to the noted values. All models include
control variables of the following: age and age-squared, gender, years of education, marital status, health
insurance status, race and birth-place (for the US only), whether it’s a small-sized business, whether it’s a
white collar job, income tercile groups, industry type, weekly work hours, and time dummies. *p<0.10,
**p<0.05, ***p<0.01.

65
Chapter 3
Social Security and Retirement in Fast-Aging Middle-
Income Countries: Evidence from Korea
11

Abstract
Many middle-income countries have been witnessing rapid population aging, increasing pressures
on their relatively fragile public pensions. Older workers may want to delay retirement to finance
their prolonged lives, but job markets are not favorable for them. A hike in pension benefits might
increase individual financial security, but weaken pension stability. Using panel data for older
workers in Korea – a typical fast-aging middle-income country with a high rate of elderly poverty
- this study finds that the forward-looking incentive measures associated with the public pension
have a statistically significant effect on retirement probability. Older wage workers are more
responsive to the dynamic interaction between earnings and social security benefits than to the
differences between current and future benefit levels. This suggests a relatively weak role for the
pension system in supporting the elderly. The significant effect of transfers from children to
parents at retirement is particularly noteworthy, suggesting low fertility rates have made more
difficult the tasks of alleviating elderly poverty and maintaining pension stability.

11
This chapter was published in Journal of the Economics of Ageing, 2020; 17. doi: 10.1016/j.jeoa.2020.100284

66
I. Introduction
Labor market participation among older workers is affected by many variables. Previous studies
of developed countries show that the social security system can have a significant effect on
retirement, leading to a sharp decline in their labor force participation rate (LFPR) at eligibility
ages (Blau and Goodstein, 2010; Blöndal and Scarpetta, 1999; Gruber and Wise, 2004).
Will this trend repeat in middle-income countries as their public pension systems mature?
Many of these countries have seen their populations age rapidly due to low fertility rates and higher
life expectancy. This places financial pressures on pension systems, and, combined with the
prospect of decreasing family support, also adds to the risk of elderly poverty. If public support is
not adequate for financing retirement, older workers may have an incentive to remain in the labor
market. Yet the labor market in these countries is not favorable to older wage workers. This will
pose a serious policy challenge for policies seeking to maintain pension stability and minimize
elderly poverty.
A case in point is the Republic of Korea (South Korea). With low fertility rates and higher
life expectancy, it is currently the fastest aging country among all OECD members. Its old-age
dependency ratio – the number of individuals aged 65 and over per 100 people of working age –
increased from 11.2 percent in 2000 to 20.2 percent in 2016 and is projected to reach 31.7 percent
in 2025 (OECD, 2017). South Korea also has the highest poverty and suicide rates among the
elderly in OECD countries.
12
Financial concerns among older workers are reflected in their
retirement behavior. While the official retirement age (i.e., age eligible for public pension benefits)

12
While overall poverty rate was 14 percent, that for those aged 65 and older was 46 percent in 2016. Countries
following Korea in elderly suicide rates in 2011 were China, Hong Kong, and Japan (poverty rates from OECD
Pensions at a Glance Database, accessed Aug 20, 2016; suicide rates from OECD (2014)).

67
is from 60 to 65, depending on birth cohort, the effective labor market exit age
13
was 72 in 2016,
reflecting the high share of older workers who are self-employed or still in the labor force. This is
much higher than the OECD average of 65.1 and the U.S. average of 66.8 (OECD, 2017). Given
that the labor market exit age resembles the pension eligibility age for most OECD countries, it
appears that the elderly population in Korea needs to work longer to ensure financial stability in
retirement.
The Korean government faces several policy challenges here. To address elderly poverty,
it needs to strengthen public support and safety-nets for the elderly. Its largest social security
system, The National Pension Service (NPS), was created in 1988 and achieved universal coverage
in 1999. Its fund balance, however, will become exhausted as the number of beneficiaries
increases, indicating a need for reform. Making the tax-benefit scheme more sustainable requires
knowing the retirement incentives associated with the current system.
Many other middle-income countries with developing social security systems will face
similar problems. Their benefit levels may not be as sufficient as those of rich countries, thereby
posing policy challenges regarding the welfare of the elderly and the labor market structure. If
older wage workers wish to work more so as to better prepare for retirement, tension may arise
with young job seekers in a tight job market. Some older workers may have to go into self-
employment as a second-best choice, adding to the already high share of the self-employed in their
labor markets. For older self-employed workers, who tend to be relatively poor on average,
enforcement of public pension is not easy.
14
In addition, family support that used to be an essential

13
OECD defines the average effective labor market exit age as the average age of exit from the labor force for
workers aged 40 or older. Labor force withdrawals are estimated using changes in LFPRs which are calculated for
each (synthetic) cohort divided into five-year age groups.
14
While it is mandatory to contribute to the public pension program in Korea, those who do not participate (e.g.,
small business owners) have been rarely penalized.

68
source of retirement financing has been dwindling in many Asian countries. While there are many
cultural contributors to this phenomenon, sharply declining fertility rates have been a critical cause
of reduced support by children.
This study aims to contribute to the literature by providing evidence on retirement
incentives of the social security system in Korea, a typical fast-aging middle-income country. Due
to the relatively short history of the government pension program and the lack of reliable data
containing sufficient information on older adults, studies on the retirement incentives of Korea’s
main social security program have been limited. Using six waves (2006-2016) of the Korean
Longitudinal Study of Ageing (KLoSA) dataset, a relatively new database with individual-level
data for those aged 45 and over, we estimate the incentive effects of the NPS on the retirement
behavior of male workers. To compare the Korean pension situation with that in other countries,
we assess measures of the expected pension benefit and other incentive variables (i.e., accrual,
peak value, and option value) that have been used in previous cross-country comparison studies
(Gruber and Wise, 2004; Stock and Wise, 1990).
Our estimation results show that retirement incentives inherent in the NPS do have a
statistically significant effect on the probability of retirement. Older wage workers are more
responsive to utility gains from continued work due to additional earnings and increased pension
benefits than to financial gains from greater pension wealth. Furthermore, the level of pension
benefits is not statistically significant when controlling for financial and labor market
characteristics. This may indicate that older workers perceive social security benefits alone as
insufficient for financing their later life. Relatedly, supplementary income variables such as
transfers from children and government income support for retirees do have statistically significant
effects on retirement. All this suggests there will be increasing pressure on the public pension

69
system in fast-aging middle-income countries as low fertility rates lead to fewer children being
able to support their parents. Furthermore, the number of pension recipients is expected to expand
and total pension benefits paid per person will increase with life expectancy.

2. Background
Social Insurance and Retirement Incentives
Workers retire for various reasons. This may include a change in health status or cyclical economic
effects (Burtless and Quinn, 2001; Purcell, 2009). The ability to finance retirement also affects
work decisions. Once retired, an individual’s income shift from wage or business earnings to
benefits from social insurance and pension plans. While private pension plans only cover a selected
few, government programs generally provide retirement income to the whole population and can
be found in most countries.
Research on the retirement incentives of social insurance programs have shown policy
reforms changing labor force behavior. For instance, the 1983 amendments of the Social Security
system in the United States, including an increase in the full retirement age and the delayed
retirement credit, and elimination of the retirement earnings test, all influenced labor force
behaviors of older workers (Coile, 2018). Changes in the Social Security rules implemented
between 1992 to 2004 led to greater LFPR of older workers (Gustman and Steinmeier, 2009).
Studies of several nations have shown that public pension systems make it financially unattractive
to work after the entitlement age (Blöndal and Scarpetta, 1999; Gruber and Wise, 2004; Aguila,
2014).
These studies are mostly limited to developed countries. Retirement incentives may differ
in low-to-middle income societies where workers tend to work until later ages, as shown in Table

70
1. Unlike those who have built their social security systems while their population gradually
matured, most middle-income countries are developing their public safety-nets while
simultaneously experiencing rapid demographic change. These societies may not be able to
prepare and adapt their systems as developed countries did. They may even fall into a middle-
income trap
15
which will further worsen the prospects of social security (Ha and Lee, 2018).
Another crucial factor affecting older adults in middle-income countries is the rapid
decrease in fertility rates. Most fast-aging countries share a common cultural characteristic, which
is the importance of family integration. Many regions in the past, especially in Asia, had explicit
declarations that care for the elderly population is a family responsibility (Phillips, 2000). Older
parents relied on their children in later ages, which may be one reason why societies did not feel
the urge to develop social insurance systems. Many middle-income Asian countries, including
Taiwan, Hong Kong, and Singapore, however, have seen a rapid decrease in their fertility rates.
Korea’s fertility rate fell to a record low in 2018, with a birth rate of less than one child per
woman.
16
As a result, senior citizens will receive less support from their children.
Older retirees in Hong Kong, for instance, have relied greatly on transfer from children,
but many believe they can no longer depend on their children for retirement planning, and
expressed concerns about retirement due to the lack of public and private support (Lee and Law,
2004). The old-age pension program in Hong Kong – the Mandatory Provident Fund – was
initiated in 2000 and has a low replacement rate of 30-40 percent (Pai, 2006). The low replacement
rate of Taiwan’s social security system has led to informal cash transfers and to private insurance

15
In the middle-income trap, middle-income countries may get too old (i.e., the share of the working population
may decrease sharply) before their economies become fully developed, leading to long-term stagnation.
16
The total fertility rate of 2018 was announced to be 0.98, which is the lowest since records began in 1970.
Statistics Korea (http://kostat.go.kr/portal/eng/index.action). Accessed Sep 4, 2019.

71
plans becoming the most important sources of retirement income for older adults (Lin, 2012).
China is also an interesting case, where the low coverage rate of the old-age pension system and
the low benefit level of the non-contributory pension system mean that many elderly will depend
on their children for retirement income (Lin, 2012; Liu and Sun, 2016). At the same time, even
though the one-child policy
17
was relaxed in 2013, China is still seeing its fertility rates decrease,
raising questions about how much support parents can expect from children.
If public and private support are not adequate for financing retirement, as is likely to be the
case in many middle-income countries, older workers will have an incentive to remain longer in
the labor market. Yet the labor market may not welcome older workers. For example, the official
retirement age used to be around 55-60 in Korea,
18
indicating some wage workers had to exit the
labor market against their wishes or move into self-employment. To further complicate the
situation, government pensions in many of these countries are not financially stable,
19
lowering
the probability of a quick jump in the benefit level. The combination of pension instability and
limited work for older persons will pose a serious policy challenge for securing the financial
stability of the elderly population.
Korea’s case is unique because it is experiencing all aspects of demographic change in an
exceedingly fast pace. In the past two decades, the old-age dependency ratio has more than
doubled, fertility rates have decreased by more than a fourth, and life expectancy has increased by
several years, as Table 1 shows, thanks to the fast pace of modernization and industrialization.
Modernization and industrialization, however, while leading to greater leisure for older workers

17
The one-child policy was a birth planning program established in 1979 to control the size of the rapidly growing
population.
18
In Korea, many corporations dealt with high operating costs after the 1998 financial crisis by utilizing a
mandatory retirement system (Cho and Kim, 2005).
19
For instance, a report estimates the social security fund of Korea to be exhausted in 2054 (NABO, 2019). This is
mainly because the government promised higher benefits than the actuarially fair level, exploiting the fact that in the
early stages, contributors outnumber the recipients.

72
in other countries, have not done so in Korea, which has a high LFPR among older workers and
where the average age of exit from the labor force for men is 72 years old. Those remaining in the
labor force may be rural senior citizens saving for retirement (Lee, 2007; 2010), or self-employed
workers who have switched jobs after being forced from a firm around age 55 (Lee and Lee, 2013;
Cho and Kim, 2005). In contrast to most developed countries, in Korea self-employment and
agriculture accounts for a substantial portion of work among the elderly (Phang, 2011).
The biggest problem facing Korea is a high rate of elderly poverty. Older Koreans have a
median household income that is less than half that for the entire population. Their 45 percent
poverty rate is above that for other nations shown in Table 1 as well as nearly four times that for
Koreans 18 to 64 years of age, who had a poverty rate of 12.7 percent in 2017,
20
likely indicating
insufficient income and savings for Korean retirees.
Around one-fourth of income for Koreans 65 and older in 2010 was from private transfers
– mostly from their children.
21
Another study shows around 70 percent of senior citizens receive
financial transfers from their child (Kim and Cook, 2011). Income inequality of the elderly is seen
to be alleviated through private rather than public transfers, with transfers from children being
proportionally larger for lower-income parents (Kim and Cook, 2011; Kang and Lee, 2001). Given
Korea’s low birth rate, older persons in the future will not be able to rely as much on private
transfers for poverty alleviation.
Examining retirement incentives of Korea’s public pension system requires a different
perspective than that of research focused on European and North American countries. The 1997
financial crisis greatly increased elderly poverty in Korea, leading to several reforms of the public

20
OECD Statistics (https://www.oecd-ilibrary.org/statistics). Accessed Sep 4, 2019.
21
Specific income sources and their amounts for older Koreans were 28.2% from wages and business earnings,
24.5% from private transfers, 16.9% from public pensions, 15.5% from assets, 12.9% from other public transfers,
0.44% from private pensions, and 1.6% from other sources (Hwang, 2016).

73
pension system. The pension system still needs to overcome several obstacles, however, such as
increasing coverage of irregular and self-employed workers, enhancing old-age security, and
improving financial sustainability (Choi, 2006; Kwon, 2002). Lower education levels, poorer
health, living alone, residing in urban areas, and increase in public pension benefit levels have all
been associated with a higher probability of retirement for Korean workers (Son, 2010; Kwon and
Hwang, 2004). Among those in the “baby boomer” generation - those born between 1955 and 1963
– 46 percent planned to rely on public pensions and 43 percent replied to rely on savings or private
pensions for financing retirement, with the latter being of higher socioeconomic status (Baek,
2011). Private transfers play a big role in maintaining retirement income for the Korean elderly
population, especially for poorer households, but private transfers themselves are insufficient to
help the elderly escape poverty (Kwon, 2001).
There have been few attempts to use forward-looking measures for analyzing retirement
incentives of the national pension system in Korea, and many studies do not incorporate the
importance of financial transfers from children. One exception is an analysis of incentive measures
using a Korean labor panel data from 1988 to 2004 (Yuan, 2007). This paper explored the Social
Security systems of both Korea and China and provided an interesting comparison of the two
public pension schemes. It found that the incentive measures were significant, especially for the
forward-looking measures, but the effect of social security wealth itself on retirement decisions
was weak. Because the study only analyzed data through 2004, it could not incorporate the major
reform of the Korean system in 2007 and the impact of the non-contributory Basic Pension
introduced in 2014. Also, the paper assumes contributions to the pension system since 1988, not
incorporating the fact that many started to contribute from 1999 when the system became
universal. By using more recent longitudinal data from 2006 to 2016 and incorporating country-

74
specific factors (i.e., transfers from children) and more accurate assumptions on when
contributions started, we extend the research on public pensions and retirement to a middle-income
country context and provide important implications for the elderly in Korea and other middle-
income countries.

South Korea’s Social Security System
Two major programs of Korea’s Social Security system target the elderly population: the NPS and
the Basic Pension. Other services provide additional aid but are limited in benefit levels and target
population.
22

The NPS is a mandatory public pension program which was introduced in January 1988
and became available for all citizens in 1999. As shown in Table 2, its coverage over time
expanded to smaller firms, self-employed workers, and rural areas. It is a defined benefit program
23

which covers every citizen in South Korea except for public officials, military personnel, and
teachers who are covered under specific occupational pensions. There have been two major
reforms: expansion to universal coverage as well as a change in net replacement rate, eligibility
age, and contribution rates in 1999 and further reduction of the net replacement rate in 2007.
NPS provides four types of benefits: old-age pension, disability pension, survivor pension,
and lump-sum refunds, with most funds for the old-age pension.
24
Retirees can claim benefits

22
The Long-Term Care Insurance for the Elderly provides home care and institutional care for lower-income older
adults diagnosed with dementia, Parkinson’s disease, or similar afflictions. The Social-Service Electronic Voucher
program provides additional home care assistance for those who are aged 65 or older, below a health threshold, and
with income less than 160 percent of the mean level.
23
The program is partially funded, with income earners contributing to a social security trust fund which is used to
pay retirees. The fund can be invested in financial markets.
24
Among the total amount of benefits paid out in 2017, 83.5% were old-age pension benefits, 9.8% survivor
pension, 4.4% lumpsum refunds, and 1.8% disability pension benefits. Korean Statistical Information Service
(http://kosis.kr). Accessed Jan 30, 2019.

75
starting from age 60 to 65, depending on birth year; the full retirement age for our analysis sample
is age 60 or 61. Early benefits are available at a reduced amount, five years prior to the full
retirement age. Benefits depend on contribution amount and period – a minimum contribution of
10 years is required to receive lifetime monthly benefits, while 20 years is required for the full
amount. Due to the short history of the program, an exception exists for certain cohort groups who
started to contribute at a later age.
25
Delaying claims will increase benefits by increasing the total
contribution period. Participation of self-employed workers was low at the beginning of the
program, mostly due to the relatively higher burden of contribution rates and underreporting of
income, but has been gradually increasing.
Basic Pension is a non-contributory pension program enacted in July 2014. It provides a
lumpsum benefit to citizens 65 or older and below an income threshold. Around 66.3 percent of
the population 65 and older received benefits in 2018.
26
The program was enacted to alleviate
elderly poverty and is funded by general revenues. Monthly benefits in 2018 were 250,000 Korean
Won, which is around a quarter of the minimum cost of living for a single household. The
government is considering plans to increase benefit levels, but the effectiveness of the program in
alleviating poverty and reducing the inequality gap has been mixed (Hwang, 2016; Seok, 2010).

25
Those who were age 45-60 in 1988 (birth cohort 1928-1943), age 45-60 in 1995 (birth cohort 1935-1950), and age
50-60 in 1999 (birth cohort 1939-1949) can receive benefits if contributed for at least 5 years. NPS website
(http://www.nps.or.kr). Accessed Jan 30, 2019.
26
Source: Basic Pension Website (http://basicpension.mohw.go.kr). Accessed Feb 1, 2018.

76
3. Specification of Incentive Measures
Following the studies in Gruber and Wise (2004), this paper estimates three measures of retirement
incentives - the option value (OV), peak value (PV), and single-year accrual (ACC). These
variables measure the gain or loss from continued work at a potential age of retirement.
27

The incentive measures are based on the present discounted value of the stream of future
social security wealth (SSW) for each possible retirement year. We use old-age pension benefits
from the NPS for SSW. Disability pensions and lumpsum refunds are not included since we do not
have information to distinguish who is eligible. These accounted for only 1.8 percent and 4.4
percent of the total amount of benefits paid out in 2017, respectively. Survivor benefits are not
applicable to our sample of fulltime working male wageworkers. There is an additional family
benefit for family members dependent on the recipient with a modest amount of benefits.
28
This is
not included due to lack of information on income levels of all household members.
The individual will choose to retire at a period when its overall financial value is the
highest. The formula for calculating pension benefits is defined as below:

(1) 𝑃𝐵 =E2.4(𝐴+0.75𝐵)
+$
+
+1.8(𝐴+𝐵)
+%
+
+⋯+1.2(𝐴+𝐵)
+%&
+
M×N1+
#.#-.
$%
O,

where A is average monthly income of all contributors for the past three years preceding the first
benefit year,
29
B is individual average monthly income for the contribution period indexed by

27
When an individual reaches its official retirement age, he or she can choose to retire or work an additional year.
By choosing to work an additional year, that individual will earn another year of labor income but forgo a year of
social security benefits, though it may increase its future benefits by increased contributions.
28
There is an additional family benefit if a spouse, underage child, or parent over age 60 is dependent on the
receiver. For April 2016 to Mar. 2017, the family benefit is 249,600KRW annually (~18 USD monthly) for spouse,
and 166,360 KRW annually (~12 USD monthly) for children under age 19 or parents over age 60. Source: NPS
website (http://www.nps.or.kr).
29
A is a redistribution factor which is publicly noted on the NPS website (http://www.nps.or.kr).

77
inflation, P is the total number of months contributed, P1~P23 is the number of months contributed
for each specific period, and n is the number of months contributed exceeding 20 years of
contribution.
30
The replacement rate is reflected through the different proportional factors for each
corresponding period of contribution.
31
It is possible to receive early retirement benefits 5 years
prior to the full retirement age.
32

For the incentive measures, first SSW is calculated as shown below:

(2) 𝑆𝑆𝑊
"
(𝑅)=∑ 𝑝𝑟
//"
∗𝛽
(/2")
∗𝑃𝐵
4
(𝑠)
5
/64
,

where 𝑝𝑟
//"
is the probability of being alive at s conditional on surviving at age t, 𝛽 is the discount
rate, T is maximum life length, and 𝑃𝐵
4
(𝑠) is the pension benefits received at age s if one retires
at age R.
33
SSW captures the income effect of the pension reforms described in the previous
section.
Next, we calculate a single-year accrual. This measures the amount of financial gain one
may obtain from postponing retirement for a year:

(3) 𝐴𝐶𝐶
"
=𝑆𝑆𝑊
"7$
−𝑆𝑆𝑊
"
,

30
To provide an example, a worker with an average earning of $2000 who contributed from 1988 to 2009 (252
months) will receive pensions from 2010 (assuming the worker has reached its full retirement age). A for those
receiving benefits from 2010 is 1791,955 KRW (1739 USD). The estimated monthly benefit is:
!2.4(1739+0.75∗2000)
!"#
#$#
+1.8(1739+2000)
!%&
#$#
+1.5(1739+2000)
!#
#$#
0×21+
%.%$∗!#
!#
3×
!
!#
= 632 USD.
31
For instance, if a person contributed for 30 years during 1988-2017, a net replacement rate of 70% is applied to the
period 1988-1998 which has a proportional factor of 2.4, a rate of 60% is applied to the period 1999-2007 with a
proportional factor of 1.8, and so forth. Details are noted in Table A2 of the appendix.
32
For each age prior to the full retirement age, the recipient receives 94%, 88%, 82%, 76%, and 70% of the full
amount, respectively.

33
For maximum life length (T), age 100 is used. For 𝑝𝑟
!/#
, the probability of surviving at age x for males is applied
which is available in Table A1 of the appendix (Korean Statistical Information Service, http://kosis.kr. Accessed Sep
10, 2017). We use 0.97 for the discount rate as described in the next footnote.

78

where a positive value induces a worker to work another year and a negative value to retire at
period t. It is limited in that it is based on a one-year comparison where accrued pension benefits
may vary substantially in the future years.
The peak value model introduced by Coile and Gruber (2001) is a forward-looking measure
comparing current pension benefits with the maximum value of that in all future periods:

(4) 𝑃𝑉
"
=max
"
$
8"
𝑆𝑆𝑊
"9
−𝑆𝑆𝑊
"
.
In real life, it is likely that workers will calculate their optimal financial value by
considering all available options, rather than comparing current benefits to that of the following
year. Hence, this measure can provide a more realistic estimate of retirement incentives by
incorporating each possible future benefit.
The option value model by Stock and Wise (1990) further integrates the stream of lifetime
earnings, incorporating the dynamic interaction between wage income and income from public
pensions. The model compares the expected value of retirement at each possible retirement age
and evaluates the period which gives the highest value. It updates expectations of future events as
the individual ages, where the value function consists of two parts: utility from work that depends
on the wage profile and utility from retirement that depends on pension wealth until death. The
value function of an individual currently aged t who retires at future age R is shown as below:

(5) 𝑉
"
(𝑅)=∑ 𝑝𝑟
//"
∗𝛽
(/2")
∗𝑈
:
(𝑦
/
)
42$
/6"
+∑ 𝑝𝑟
//"
∗𝛽
(/2")
∗𝑈
4
(𝑆𝑆𝑊
/
(𝑅))
5
/64
,

79
where 𝑦
/
is the present discounted value of wage earnings at age s and 𝑆𝑆𝑊
/
is the present
discounted value of social security wealth at s. 𝑈
:
(𝑦
/
)= 𝑦
/
;
and 𝑈
4
(𝑆𝑆𝑊
/
(𝑅))=(𝑘∗
𝑆𝑆𝑊
/
(𝑅))
;
where 𝛾 is a risk aversion parameter and k is the disutility of labor.
34

Assuming R* to be the optimal retirement age in the future which gives the highest
expected value of retirement, the option value is the difference between the expected value of
retiring at R* from the expected value of retiring today:

(6) 𝑂𝑉
"
(𝑅
∗
)=𝐸𝑉
"
(𝑅
∗
)−𝐸𝑉
"
(𝑡),

where an individual continues to work until R* if the option value is positive and choose to retire
when the value turns negative.
All three measures have different features. The peak value and option value are forward
looking in that it takes into account the age profile of possible benefits in the future. It is assumed
individuals plan their retirement based on all future information of the expected income. Studies
show that forward-looking measures tend to provide more accurate estimates than the single-year
accrual (Gruber and Wise, 2004). The option value requires modeling of the functional form of an
individual’s utility, while the other two models tend to be more simplistic. For the option value,
choosing to work an additional year can increase utility by extending total wage earnings and
pension benefit levels, but can also decrease utility by reducing the number of years over which
benefits are received. Additionally, the option value model incorporates labor income which can
provide more accurate estimates of financial incentives on the decision to retire, however it may
also capture leisure preferences which can also impact retirement choice. In this case, the impact

34
The structural parameters are based on literature using similar methods: 𝛽=0.97, k=1.5, 𝛾=0.75 (Gruber and Wise,
2004; Aguila, 2014).

80
of retirement incentives of the social security program itself may not be correctly identified. The
peak value can help isolate the effect of pension benefits on retirement behavior by not including
labor earnings, yet it may understate the impact of financial incentives on retirement choice.

4. Data and Estimation of Variables
Data
KLoSA is a biennial panel survey of people aged 45 and over and their partners living in private
households in South Korea. The sample size is 10,254 individuals and 6,171 households at the
baseline with six waves from 2006 to 2016. A new cohort of 920 individuals, born in 1962-63,
was added in wave 5. It includes information on basic demographics, health and health care
utilization, financial and employment characteristics, and a detailed survey on interactions with
family members.
The sample used in this study consists of male wage and salaried workers aged 50 to 65 at
the baseline.
35
The sample is restricted to full-time workers - working more than 40 hours a week
- with income data in the first wave. It is assumed there is no transition from wage work to self-
employment, or vice versa. The final sample consists of 540 individuals in the first wave, totaling
a 2,727 individual-year sample. The percentage of retirees increase as the sample age.
36

35
Cohorts older than 65 in the first wave were already in their 50s in 1988, so there is a high chance they would not
have contributed to the program. Females were not included due to the small sample size of working women. Self-
employed workers were not included because of low pension participation rates. Many of the self-employed in
Korea tend to work longer than the pensionable age and underreport their income which further leads to lower
pension wealth. The self-employed also have to contribute the full amount leading to higher burdens. While it is
mandatory to participate in the public pension program, currently there is no enforcement, leading to low
participation rates. Therefore, including self-employed workers may bias the true effect of the program.
36
Portion of retirees is 12%, 18%, 27%, 31%, and 36% from wave 2 to 6, respectively.

81
Descriptive statistics of the sample are shown in Table 3. At the baseline, around a half live
in a metropolitan area and 34 percent live with at least one child, which is typical for that
generation. Around 21 percent of respondents are college graduates or above and 40 percent are
high school graduates for full-time male wage workers. Self-reported health is based on five
categories - very good, good, fair, bad, and very bad – where we assume the respondent to be in
good health if he reported his health to be ‘very good’ or ‘good.’ Mental health is measured using
a shorter 10-item version of the Center for Epidemiologic Studies Depression (CES-D) scale,
which ranges from 0 to 30 in KLoSA. A score of ten or higher indicates symptoms of depression
and anxiety (Andresen et al., 1994; Ichimura et al., 2017). 21 percent and 97 percent of the sample
reported being in good self-reported health and mental health, respectively.
Financial security at old age is a crucial factor impacting retirement decisions. Average
monthly earnings of the sample are similar to that of fulltime workers in the country.
37
A unique
fact of older Koreans is that many receive private transfers from children. A quarter of our sample
received regular financial support from their children while three quarters received occasional
transfers. Around a tenth of the sample received monthly lump sum benefits from the Basic
Pension in 2016 – a non-contributory pension program for lower-income older adults enacted in
2014. We believe the small number of recipients is due to the fact that our sample period ends in
2016, where the number of recipients may not have been as high during the first few years of the
program. Also, since this pension aims to cover the lower income or those with no income
earnings, there is a high possibility that a majority of our sample, male full-time wage workers,
were not eligible. Labor market experience is defined as age minus six and minus the number of

37
Average monthly earnings of the sample are 2330 USD (2015 real values). The average monthly income for
fulltime wage workers in Korea in August 2015 was 2,702,000 KRW, equivalent to 2388 USD. Source: Korean
Statistical Information Service (http://kosis.kr/index/index.do), cited Feb 3, 2020.

82
years in school, following Coile and Gruber (2007). Most wage and salaried workers had
occupations related to manual work, craft and machine operations, and professional jobs.

Earnings Profiles
Each individual’s earnings profile from the age they started to participate in the NPS until
retirement is needed to estimate SSW and retirement incentive measures. KLoSA only contains
income data from 2006 to 2016, so we used an additional dataset to match each worker’s full
earning history, following the method used by Aguila (2014) and Blundell, Meghir, and Smith
(2004). The Korea Labor Income Panel Survey (KLIPS) is an annual panel survey with 19 waves
from 1998 to 2016 consisting of 5,000 households age 15 years and older. Average wage profiles
of full-time male wage workers were created for each birth-cohort group using KLIPS, which is
matched to each individual in KLoSA using the following equation:

(7) 𝑌
!="
=𝜃
!
𝑌
="
,

where i is individual, g is cohort group, t is year, and Y is monthly earnings. θ
>
is an individual
fixed effect which is estimated by θ
>
=Y
>?%##@
/Y
?%##@
. The fixed effect adjusts for differences
among individuals in the same cohort group. Y
?A
is the average income from KLIPS for each cohort
group g, computed for each period t from 1998 to 2016. This data is matched to each individual’s
wage profile in KLoSA (Y
>?A
) which is available for the year of 2006, 2008, 2010, 2012, 2014, and

83
2016.
38
Earnings data are deflated to the base year of 2015 and converted to USD.
39
Average
monthly earnings for KLIPS (2234 USD), KLoSA (2330 USD), and the average population
income in 2015 (2388 USD) is comparable.
Earnings profiles for each individual from the age of year 1988 to age 70 is created. Figure
1 depicts estimated earnings profiles, averaged for five-year cohort groups for convenience. The
profiles seem close to expectations. Younger cohorts tend to have a higher level of earnings
compared to the older generation. Wage peaks around mid-to-late fifties with the average being
similar to the population level of the sample period. It is assumed future earnings decline with age
in order to reflect the reality of older workers experiencing decrease in working hours.

Social Security Wealth
The expected present discounted value of pension benefits for each individual for its potential
retirement age of 55 to 70 is computed based on equation (2). The pension eligibility age is 60 for
the birth cohort 1941-1952 and 61 for those born in 1953-1956. It is possible to receive early
retirement benefits from the age of 55 and 56, respectively.
NPS was enacted in 1988, targeting wage workers at firms with more than ten employees.
The program expanded to smaller firms with five to nine employees in 1992 and became
mandatory for the whole population in 1999. Using information on the firm size of the base year,

38
For earnings data of 2007, 2009, 2011, and 2013, the average value of the previous and following year is used.
For missing data from 1988 (when NPS was enacted) to 1998, backward estimation using a real wage growth rate of
8.49% for 1988 to 1993, and 5.395% for 1994 to 1997 is applied (Source: 1988-1992 from Ministry of Employment
and Labor (https://www.moel.go.kr/english/main.jsp) and 1993-1997 from Korean Statistical Information Service
(https://kosis.kr). Accessed Feb 1, 2017). For cohorts 1947-1956, wage data needs to be estimated forward until the
age of 70 (the latest age to retire in this study) where average growth rates of male full-time wage workers from
KLIPS is applied - wage decreases each year by 5.21% from age 60 to 65 and by 7.67% from 65 to 70 (KLIPS,
2013-2016).
39
Real values are computed using CPI from Korean Statistical Information Service (http://kosis.kr/index/index.do).
Values are converted to USD using the 2015 average exchange rate (1USD = 1131.52 KRW).

84
we are able to apply different years of contribution starting periods; 1988 for workers in firms with
more than ten employees, 1992 for those in firms with five to nine employees, and 1999 if the firm
had less than 5 workers. This led to 393 individuals contributing from 1988, 44 from 1992, and 83
from 1999. If workers contributed from 1988, they would have participated for a total of 28 years
at wave 6, and if they contributed from 1999 the years of participation will be 17 years. This fulfills
the minimum vesting requirement of 10 years. Retirement incentives will be higher for those with
longer contribution periods since the program requires 20 years of contribution to receive full
benefit amounts. We further applied information on the start year and end year of work from the
Supplementary 2007 KLoSA work-history data for more accurate contribution periods. We
assume individuals worked continuously throughout the start and end years of work. This implies
the years of contributions may be slightly overestimated which could possibly lead to issues of
measurement error, however, the median estimates of SSW of the sample closely matches that of
the whole population as shown below.
Median estimated monthly SSW for the potential retirement age of 55 to 70 is shown in
Table 4.
40
The average amount of pension benefits tends to increase in retirement age due to longer
contribution periods, but the marginal amount of change decreases, indicating lower incentives to
postpone retirement as one becomes older. The overall estimates seem to be a good match to the
actual population data. The government estimates pension benefits to be around 677 USD (767,130
KRW) if contributed for 30 years with the average income of 2,300 USD,
41
where the median
amount of benefits of our sample at age 60 is 624 USD.

40
All estimates are in real values with the base year being 2015 and converted to USD. CPI data is from the Korean
Statistical Information Service (http://kosis.kr/index/index.do). 2015 average exchange rate of USD/KRW is
1131.52.
41
NPS website (http://www.nps.or.kr). Accessed Feb 17, 2019.

85
Retirement Incentive Measures
The incentive measures described in the previous section – accrual, peak value, and option value
- is estimated using each individual’s SSW and earnings profiles where the median and standard
deviation are described in Table 4.
42

The single-year accrual incorporates differences of a single year which may be limited in
accurately measuring the dynamic changes in retirement incentives within a lifecycle. The peak
value is the difference between the current versus the maximum future SSW. The option value is
the only measure which incorporates lifetime earnings, where retirement decisions depend not only
on pension levels but also on wage profiles. The dynamic interaction between wage earnings and
pension benefits may provide more accurate estimates of retirement incentives. We assume it is
more likely that workers in Korea will react to changes in wage income rather than pension benefits
due to the relatively low levels of SSW.
For all three measures, the estimates decrease with age which indicates the cost of retiring
today declines as one age. Noteworthy is that for most of the incentive measures, there is an
incentive to delay retirement over the official retirement age of 60. Older workers may prefer to
earn wages from the labor market as long as possible. This generation may also prefer to work
longer for other reasons, such as work ethics or lower preferences for leisure.

42
The 10
th
and 90
th
percentile estimate of each variable are reported in Table A3 of the appendix.

86
5. Estimation of Retirement Models
Regression Framework
In a standard retirement model, pension wealth plays two roles in the retirement decision making
process, indicated as wealth effects and accrual effects (Coile and Gruber, 2007). Wealth effects
reflect the fact that higher SSW induces individuals to consume more leisure, and therefore retire
earlier. Accrual effects indicate that an individual’s decision to continue working is a function of
an increase in retirement wealth resulting from an additional year of work earnings. Following this
framework, the regression model is shown below:

(8) 𝑅
!"
=𝛽
#
+𝛽
$
𝑆𝑆𝑊
!"
+𝛽
%
𝐼𝑁𝐶𝐸𝑁𝑇
!"
+𝛽
&
𝑋
!"
+𝛽
'
𝐸𝐴𝑅𝑁
!"
+𝛽
-
𝐿𝐴𝐵𝑂𝑅
!"
+𝜀
!"
,

where 𝑅
!"
is a dummy variable which equals to one if an individual i retires during year t, given
that he works in the previous year, and zero otherwise. 𝑆𝑆𝑊
!"
is the expected present discounted
value of monthly pension benefits if an individual retires in year t. 𝐼𝑁𝐶𝐸𝑁𝑇
!"
is the retirement
incentive measures estimated in the previous section – accrual, peak value, and option value. 𝑋
!"

is a set of demographic characteristics including age, cohort group dummies, education levels,
residence in urban or rural area, whether living with a child, and whether in good self-reported
health and mental health. 𝐸𝐴𝑅𝑁
!"
captures differences in financial status using individual income
tercile groups. 𝐿𝐴𝐵𝑂𝑅
!"
is labor characteristics, including job type categories and labor market
experience in years and years squared. Dummy variables for age 60 and 61 are included, which is
the eligibility age for full pension benefits for our sample. Given that the dependent variable is
dichotomous, the model is estimated as a probit.

87
Retirement models may suffer from possible endogeneity, such as work preferences
affecting both retirement decisions and incentive measures. To address the source of identification
for retirement incentives, the model includes a set of controls of income tercile groups of the
average wage earnings in the model following Coile and Gruber (2007). Though imperfect, the
controls can further capture heterogeneity which could bias the estimates. We use wage earnings
to compute pension benefits and the option value, but they enter the equations nonlinearly. The
option value is computed exogenously based on specific parameters of the utility function as
described in Belloni (2008). Further incorporating the peak value provides additional
interpretations where the measure does not incorporate income and focuses only on the variation
in pension benefits.

Results
Table 5 presents estimation results of the probit regression for each incentive measure, controlling
for demographic, financial, and labor characteristics. SSW, single-year accrual, and the peak value
are measured in hundreds of US dollars, and the option value is expressed in 100 units. Following
Coile and Gruber (2007), we note in square brackets the percentage-point impact of a one-
standard-deviation increase in SSW and incentive variables.
43
For convenience, we note only
estimates of SSW and incentive measures.
44

SSW has a statistically significant and positive effect on the probability to retire in the
single year accrual model, but not in the forward-looking incentive models when controlled for all

43
Standard deviation for a 100 unit of SSW, ACC, PV, and OV is 3.83, 1.55, 2.91, and 4.96, respectively. The
results are calculated by multiplying each standard deviation to the corresponding marginal effect at the means of all
explanatory variables.
44
Estimates of all covariates are noted in Table A4 of the appendix.

88
covariates. Because financial and labor market conditions are critical in considering retirement,
their omission in specifications would make it difficult to identify social security impacts. Among
the incentive measures, only the option value model shows statistically significant estimates,
indicating greater option value will have a negative impact on retirement. The overall direction of
the result is similar to studies for different countries (Gruber and Wise, 2004; Aguila, 2014; Belloni
and Alessie, 2009). The marginal impact is slightly smaller than that in the United States. For
example, column (6) indicates a standard deviation increase in 100 units of the option value will
lower retirement by 1.4 percentage points. Although cross-country comparisons are difficult due
to different units in measurements, the impact approaches the 2.6 percentage point decrease in the
United States estimated by Coile and Gruber (2007).
45
The insignificant results of SSW may
suggest a possibility of measurement error due to incomplete data on past earnings and on when
pension contributions commenced, leading to small coefficients.
Further analysis for robustness checks is conducted as shown in Table 6. First, instead of
controlling for income groups to address possible endogeneity, the model includes the average of
both SSW and incentive measures. In comparison to the main results in Table 5, column (1), (3),
and (5) of Table 6 shows estimates slightly larger in magnitude, but comparable in the overall
direction and significance. Second, we use a logit model. Results in column (2), (4), and (6) of
Table 6 are similar to the results using a probit model.
In Table 7, the model includes terms of whether an individual is receiving income from
additional sources and their interaction with SSW, based on the option value model. The first
column includes a binary variable indicating whether the respondent received occasional cash

45
The median SSW for Korean retirees at age 60 is estimated to be 638 USD whereas for the United States it’s
99,170 USD as estimated by Coile and Gruber (2007). SSW are expressed in 100 USD in this study and 100,000
USD in Coile and Gruber (2007).

89
transfers from children, column (2) includes a variable of whether the respondent received either
occasional or regular transfers, and the last column includes a dummy variable of being a Basic
Pension recipient.
Compared to the coefficients in column (6) of Table 5, controlling for private cash transfers
magnifies the estimates of SSW and the option value substantially. This suggests that, unlike many
developed countries, transfers from children are crucial in retirement decisions for older Koreans.
Considering that the sample consists of full-time working male wage workers, we can assume that
the impact of cash transfers would be greater for part-time or female workers. Columns (1) and (2)
show that receiving cash transfers from children, either occasional or regular, are estimated to
increase the probability of retiring, possibly through improved financial security.
46
The
coefficients on the interaction terms are negative, implying recipients of transfers from children
are less influenced by pension benefits when making retirement decisions, compared to those
without such transfers. Additionally, analyzing only those in the lower income tercile groups, the
bottom 66.6 percent, showed that the coefficients for SSW were no longer significant and the
estimates for cash transfers were significant and slightly bigger in magnitude compared to that of
the whole sample. This implies the role of transfers from children in shaping retirement may be
stronger for those in poorer households. The coefficient of being a Basic Pension recipient is not
statistically significant, however, the marginal effect of SSW and the option value also increases
when controlling for the variable. The interaction term, although insignificant, is negative which
indicates that older adults who are eligible for non-contributory pension benefits tend to be less

46
For older adults aged 65 and older in Korea, a quarter of income is from cash transfers on average, mostly from
adult children (Hwang, 2016). This can be partly explained by cultural aspects like the high values put on families
and the elderly.

90
affected by the NPS. Since the Basic Pension was enacted in 2014, the impact of the non-
contributory pension may become more significant for future recipients.
Overall, the above results indicate that the benefit level and incentives embodied in the
social security system do not appear to be as strong compared to those of many developed nations,
but considering the NPS is still in its maturing stage, its incentive effects on retirement could be
more pronounced in the future.
Another fact to note is that estimates show SSW alone is not sufficient to finance post-
retirement life and workers may even be more sensitive to other sources of income. Our results
show that workers in Korea respond less to the level of pension benefits and rather to income level
groups, as shown in Table A4. Also, the option value turns out to be a better measure in explaining
retirement incentives compared to single year accrual or peak value. That is, older workers in
Korea are more responsive to the cost of retiring today as a function of earnings and social security
benefits than to the difference between current and future pension wealth. The insignificant
estimate of the single year accrual and peak value coefficient may indicate the fact that Koreans
are less influenced by increased future benefit levels when making retirement decisions. In other
words, the level of SSW is perceived as being inadequate in absolute terms.
An additional source of income critical to Korean elderly citizens is cash transfers from
children. The average monthly amount of occasional and regular transfers for this sample is 331
USD and 449 USD, respectively, which is a big part of one’s retirement income.
47
Estimation
shows that financial transfers is an influential factor affecting retirement choice, but fertility rates
in Korea and other middle-income countries have been decreasing rapidly. The culture of

47
KLoSA asks respondents the average monthly amount of regular transfers and total annual amount of occasional
transfers from all children. We noted the annual level of occasional transfers divided by 12. All values are in real
values (base 2015) and converted to USD.

91
supporting old parents in a large-family system has also been disappearing. As such, the elderly
population may not be able to rely as much on their children in the future, posing further challenges
on the role of public pensions. The presence of alternative government support is another example.
In the face of deteriorating elderly poverty, the government has been increasing the level of Basic
Pension benefits, but this may lessen incentives to resort to the main public pension service.
48

Hence, it is crucial to take into account the crowding-out effect of the Basic Pension when
designing policy reforms.

6. Conclusion
Retirement incentives inherent in a social security system have been investigated mostly in the
context of advanced countries. As summarized in Gruber and Wise (2004), decisions to work at
older ages are significantly influenced by forward-looking incentives for both public and private
pensions, although the evidence varies by data and model specifications. Similar studies for less
developed countries are rare because their public pension systems are still at maturing stages and
relevant data is lacking. As the populations for less-developed countries age, there will be a need
to expand pension benefits and to study the effect of such policy changes on retirement behavior
and pension finance.
Using six waves (2006-2016) of survey data for older workers in Korea, this study provides
evidence on the effect of the social security system on retirement behavior in a typical fast-aging
middle-income country with a high rate of elderly poverty. The study uses estimation models
developed in previous studies, including various forward-looking incentive measures for

48
Though the NPS is mandatory for most income earners, the compliance of certain groups (e.g., the self-employed,
low-income, etc.) are not as strong as the government expects. Several reports state that the Basic Pension reduce
long-term contribution incentives of the NPS (Yoon, 2014; Lee and Kim, 2013).

92
comparison purposes. While our estimated coefficients show similar directions of influence found
in previous studies, the details contrast sharply. First, among forward-looking models, only the
option value variable shows statistically significant effects. That is, workers who expect a greater
utility gain from continued work due to additional earnings and pension wealth gains are less likely
to retire than workers with a smaller gain from work. Second, supplementary income variables
such as transfers from children and extra income support for poor retirees have significant effects.
All this indicates the incentives incorporated in the current social security system do not
appear to be as strong in comparison to those of many developed countries, mainly due to the
relatively short history of the program. In making retirement decisions, older workers in Korea
appear to consider diverse factors, such as utility gains from additional wage earnings. Also
noteworthy is the significant effect on retirement of transfers from children, a common source of
retirement financing in many Asian countries. Fast-aging middle-income countries will likely face
tough policy choices for alleviating elderly poverty and maintaining pension stability. Given
decreasing fertility rates and changing family culture, income support from children will likely
drop, causing concerns about the welfare of the poor elderly.
Another issue is the potential crowding-out effects of supplementary public support
schemes. Given that the Basic Pension was introduced in 2014, the number of recipients in our
sample was insufficient for analysis, making it difficult to discern the impact of the non-
contributory pension program on retirement decisions and its relationship with the main pension
scheme. Therefore, interpretation of the insignificant estimates should be made with caution.
Nevertheless, we should keep in mind the possibility that those receiving the supplementary lump
sum pension will be less influenced by the main pension scheme in their retirement decisions. The
government needs to find an efficient way of allocating limited public resources for retirement

93
income support, minimizing the distorting effects on the incentives for older workers. This will
require ensuring that the labor market can accommodate older wage workers who wish to remain
in the labor force. The effective coverage of the main public pension also needs to be expanded so
that public resources are more effectively managed.
49

This work suggests several future directions for research. First, a weakness of our model
is the goodness of fit which may be improved in the future with a larger sample size. Figure 2
shows the hazard rates by age for actual retirement and predicted retirement using the option value
model. The predicted model does not exactly fit the actual hazard rates and only a few points lie
within the 95% confidence intervals. This may be due to the small sample size or there may be
omitted variables. The predicted model, however, does follow a similar trend with spikes at age
60 and 61, which is the eligibility age for full benefits. We believe the fit of our model can be
improved in the future with a richer sample.
Also, as Korea’s pension scheme matures, future research might compare alternative
analysis samples. The current generation of older Koreans receives lower benefits because it began
contributing to the program at later ages. Analysis of future generations that have contributed
longer and have vesting requirements may yield different results. Comparing male and female
workers will also be interesting given the increasing labor force participation rates of women over
time.
Future research might also better exploit Korea’s unique labor market characteristics. It
would be interesting to analyze the incentives for older self-employed workers once the pension
system and accompanying data have matured. The informal labor market of South Korea mostly
consists of self-employed workers with small businesses; this sector has been less explored due to

49
The Basic Pension in Korea provides benefits to 65 to 70 percent of the elderly population. This suggests it does
not just target those in poverty.

94
the lack of data. Around a quarter of employment in 2016 were self-employed, and, among the
self-employed, 29.3 percent were 60 or older.
50
The number of older self-employed workers is
expected to increase as the population ages. Some wage workers also move to self-employment.
Our study assumed full-time wage workers transit to full-time retirement, and hence did not
account for those who move to self-employment or part-time work. To assess this assumption, we
used a sample of male full-time wage workers aged 50 or older in the KLIPS to check transitions
over five years, following the method of van Solinge (2014). We found that from 2011 to 2016,
6.3 percent moved to part-time wage work and 4.9 percent to self-employment. Future work should
explore retirement incentives for those who move between work types rather than just from full-
time work to full-time retirement. NPS participation rates of self-employed workers have been low
in the past, but a government welfare program enacted in 2012 supports half of the contributions
for low-income self-employed workers, which may increase future participation rates of the self-
employed.
51

A comparative study of countries with similar pension maturity or labor market structure
would also be useful. The main source of income for most elderly in Hong Kong, Taiwan, and
mainland China is, like that in Korea, their children. By contrast, in Japan – another Asian country
with high old-age dependency ratios but a relatively mature social security system –around 80
percent of income for the elderly is from public pensions (Yashiro and Oshio, 1999). Many
countries with lower fertility rates and insecure welfare systems for the elderly may soon face
challenges similar to those in Korea. Other emerging markets, including Thailand, Vietnam, and
the Philippines, are also undergoing demographic changes and are bound to face similar pressures
on their systems for supporting the elderly (Pai, 2006). At the same time, these modernizing

50
Korean Statistical Information Service (http://kosis.kr/eng/ ). Accessed Feb 1, 2018.
51
Doori Noori program (http://insurancesupport.or.kr).

95
regions will vary in their development in ways that will affect the wellbeing of senior citizens
differently. Such settings may provide interesting cases for comparative studies.

96
Figure 3.1. Earnings profiles for male wage workers

Earnings data are deflated to the base year of 2015 and converted to USD using the 2015 average exchange rate.
Source: KLIPS (1998-2016), KLoSA (2006-2016).

500 1000 1500 2000 2500 3000
Monthly Earnings (USD)
30
32
34
36
38
40
42
44
46
48
50
52
54
56
58
60
62
64
66
68
70
Age
cohort 1: 1952-1956 cohort 2: 1947-1951 cohort 3: 1941-1946

97
Figure 3.2. Change in retirement probability by age

The predicted hazard rates by age are based on the option value model.
Source: KLoSA (2006-2016).

0 .02 .04 .06 .08
Change in Retirement Probability
55 56 57 58 59 60 61 62 63 64 65 66 67 68 69
Age
Acutal Predicted

98
Table 3.1. Demographic trends by country

Korea Japan China Mexico US France
Old-age dependency ratio (2018)
1
20.0 46.0 15.7 10.6 24.2 32.4
∆ dependency ratio (%, 2000-18) 100.8 84.9 55.8 27.8 29.4 31.9
Fertility rate (births per woman, 2017) 1.1 1.4 1.6 2.2 1.8 1.9
∆ fertility rate (%, 2000-2017) -28.3 5.2 9.0 -20.7 -14.1 1.6
Life expectancy at birth (2017) 82.6 84.1 76.4 77.3 78.5 82.5
∆ life expectancy (%, 2000-17) 8.8 3.7 6.2 4.0 2.5 4.4
GDP per capita (US$, 2018) 31,363 39,287 9,771 9,698 62,641 41,464
Elderly poverty rate (2016)
2
0.45 0.20 . 0.25 0.23 0.03
LFPR age 65+ (2018) 32.2 24.7 . 27.2 19.6 3.1
Effective labor market exit age (men,
2016)
3

72.0 70.2 . 71.6 66.8 60.0
1. Old-age dependency ratio: the proportion of people aged 65 and older per 100 working-age population (i.e., age
15-64).
2. Elderly poverty rate: the poverty threshold is set at 50% of median household disposable income in a particular
country at a particular point in time (OECD Statistics). Statistics for Japan is from 2015.
3. Effective labor market exit age: the average age of exit from the labor force for workers initially aged 40 and
over. Labor force withdrawals are estimated using changes in labor force participation rates which are calculated for
each (synthetic) cohort divided into five-year age groups (OECD Statistics).
Source: World Bank, International Comparison Program database (https://datacatalog.worldbank.org/). LFPR,
effective labor market exit age, and elderly poverty rate from OECD Statistics (https://www.oecd-
ilibrary.org/statistics). Accessed Sep 4, 2019.

99
Table 3.2. History of the National Pension Service (NPS)
Jan. 1, 1988 • Enforcement of NPS: covering workplaces with more than 10 full-time workers

Jan. 1, 1992 • Coverage expanded to workplaces with 5 to 9 full-time workers

Jul. 1, 1995 • Coverage expanded to foreigners, the self-employed, and rural areas

Apr. 1, 1999 • Universal coverage
• Net replacement rate lowered from 70% to 60%
• Retirement age increased from 60 to 65 by birth cohort
• Contribution rate increased from 6% to 9%

Jan. 1, 2007

• Net replacement rate lowered to 50% in 2008
• Net replacement rate lowered 0.5p.p. annually from 2009 with an aim to achieve
40% in 2028

Jul. 1, 2012

• Doori Noori program enacted: subsidize half of pension contributions for the lower-
income self-employed

Jul. 25, 2014

• Basic Pension enacted: a non-contributory lumpsum benefit for lower-income adults
age 65 and older
Source: NPS Website (http://www.nps.or.kr), Doori Noori Website (http://insurancesupport.or.kr), Basic Pension
Website (http://basicpension.mohw.go.kr).

100
Table 3.3. Descriptive statistics of male wage workers at baseline
Variable Mean (SD)
Demographic characteristics
Age 55.95 (4.42)
Cohort group (%)
1952-1956 35.37
1951-1947 40.93
1941-1946 23.70
Live in metropolitan area (0/1) 0.53 (0.50)
Live with at least one child (0/1) 0.34 (0.48)
Education (%)
Middle school dropout and below 18.15
Middle school graduate 21.11
High school graduate 39.81
Some college and above 20.93
Good self-reported health (0/1) 0.21 (0.41)
Good mental health (0/1) 0.97 (0.18)
Financial characteristics
Monthly earnings (2015USD) 2329.53 (1425.10)
Regular transfer from children (0/1) 0.25 (0.44)
Occasional transfer from children (0/1) 0.75 (0.43)
Basic Pension recipient (wave 6, 0/1) 0.10 (0.31)
Labor Market Characteristics
Experience (years) 38.79 (6.37)
Job type (%)
Professional work 23.88
Office work 10.82
Service and sales 3.54
Craft and machine operations 29.10
Agricultural, forestry, and fishery 1.68
Manual work 30.97
No. observations 540
Source: KLoSA (2006, 2016).

101
Table 3.4. Estimated monthly SSW and incentive measures by age
Age
SSW Accrual Peak Value Option Value
Median SD Median SD Median SD Median SD
55 383 136 45 19 469 177 1091 631
56 425 155 49 22 419 163 920 571
57 468 172 51 24 372 146 747 511
58 517 196 54 26 318 128 583 447
59 565 221 58 27 261 108 438 368
60 624 245 24 15 207 85 322 312
61 640 254 25 15 186 74 257 267
62 658 264 24 13 165 64 204 226
63 681 273 22 12 142 55 154 188
64 703 281 21 11 119 47 110 155
65 732 288 21 11 96 38 76 125
66 753 295 20 10 73 30 50 94
67 772 300 19 9 53 22 30 69
68 792 304 18 9 35 15 20 48
69 807 308 16 8 16 8 15 39
SSW are in real terms (base year 2015) and converted to USD. Each incentive measure is estimated for each
possible retirement age. SD is the standard deviation. The number of observations for each value is 540.
Source: Author calculations using KLoSA (2006-2016).

102
Table 3.5. Probability of retirement for male wage workers
Accrual Peak Value Option Value
(1) (2) (3) (4) (5) (6)
Social Security Wealth 0.089*** 0.082* 0.081*** 0.076 0.074*** 0.071
(5.64) (1.87) (4.27) (1.53) (3.36) (1.41)
[0.048] [0.004] [0.061] [0.010] [0.056] [0.008]

Incentive Measures 0.071** 0.101 -0.040 -0.00002 -0.029** -0.096***
(1.99) (1.32) (-1.02) (-0.00) (-2.09) (-2.60)
[0.015] [0.002] [-0.023] [-0.000] [-0.028] [-0.014]

Demographic Controls
Yes Yes Yes Yes Yes Yes
Financial & Labor
Controls
No Yes No Yes No Yes
SSW, accrual, and peak value are in 100 USD units; option value is expressed in units of 100. Marginal effects of
a standard deviation increase are in square brackets. Standard errors are in parenthesis and are corrected with the
Huber-White robust method for heteroscedasticity.

103
Table 3.6. Robustness check
Accrual Peak Value Option Value
(1) (2) (3) (4) (5) (6)
Social Security Wealth 0.483*** 0.174* 0.374*** 0.151 0.143 0.150
(3.07) (1.92) (2.94) (1.47) -0.64 (1.44)
[0.706] [0.004] [0.037] [0.009] [0.019] [0.008]

Incentive Measures 0.164 0.209 -0.0361 0.0431 -0.118* -0.192**
(1.35) (1.37) (-0.39) (0.32) (-1.67) (-2.38)
[0.009] [0.002] [-0.003] [0.002] [-0.020] [-0.013]

Control average SSW &
incentive measure
Yes No Yes No Yes No
Logit model
No Yes No Yes No Yes
SSW, accrual, and peak value are in 100USD units; option value is expressed in units of 100. Marginal effects of a
standard deviation increase are in square brackets. Standard errors are in parenthesis and are corrected with the
Huber-White robust method for heteroscedasticity. Column (1), (3), and (5) additionally control for average SSW
and average incentive measures instead of income groups. Column (2), (4), and (6) use a logit model.

104
Table 3.7. Probability of retirement by additional sources of income

Occasional transfers
from children
Regular/occasional
transfers
from children
Basic Pension
(1) (2) (3)
Social Security
Wealth
2.180*** 2.123*** 0.075
(3.47) (3.34) (1.44)
[0.231] [0.232] [0.013]

Option Value -0.257*** -0.226*** -0.096***
(-3.00) (-2.76) (-2.60)
[-0.035] [-0.032] [-0.021]

Supplemental
Income
18.44*** 17.84*** 0.413
(3.79) (3.60) (0.52)

Supp. Income×SSW -2.430*** -2.352*** -0.058
(-3.72) (-3.53) (-0.39)
SSW are in 100USD units, option value is expressed in units of 100. Marginal effects of a standard deviation
increase are in square brackets. Standard errors are in parenthesis and are corrected with the Huber-White robust
method for heteroscedasticity.

105
Appendix
Table A.3.1. Probability of surviving
Age/Year 1988-1994 1995-1999 2000-2004 2005-2009 2010-2015
25-34 0.99 0.99 0.99 0.99 0.99
35-39 0.98 0.99 0.99 0.99 0.99
40-44 0.97 0.98 0.98 0.99 0.99
45-49 0.96 0.97 0.97 0.98 0.98
50-54 0.94 0.95 0.96 0.97 0.98
55-59 0.92 0.93 0.95 0.96 0.97
60-64 0.88 0.90 0.92 0.94 0.95
65-69 0.82 0.85 0.87 0.90 0.92
70-74 0.73 0.77 0.80 0.84 0.87
75-79 0.62 0.66 0.69 0.74 0.78
80-84 0.48 0.53 0.56 0.60 0.64
85-90 0.35 0.38 0.40 0.44 0.47
90-94 0.22 0.24 0.25 0.28 0.30
95-100 0.11 0.13 0.14 0.15 0.16
100+ 0.00 0.00 0.00 0.00 0.00
Probability of surviving at age x, males, averaged for year and age groups.
Source: Korean Statistical Information Service (http://kosis.kr). Accessed Sep 10, 2018.

106
Table A.3.2. Net replacement rates and corresponding period
Year
Net Replacement
Rate (%)
Proportional Factor Contribution Period
1988-1998 70 2.4 P1
1999-2007 60 1.8 P2
2008 50 1.5 P3
2009 49.5 1.485 P4
2010 49 1.47 P5
2011 48.5 1.455 P6
2012 48 1.44 P7
2013 47.5 1.425 P8
2014 47 1.41 P9
2015 46.5 1.395 P10
2016 46 1.38 P11
2017 45.5 1.365 P12
2018 45 1.35 P13
2019 44.5 1.335 P14
2020 44 1.32 P15
2021 43.5 1.305 P16
2022 43 1.29 P17
2023 42.5 1.275 P18
2024 42 1.26 P19
2025 41.5 1.245 P20
2026 41 1.23 P21
2027 40.5 1.215 P22
2028 40 1.2 P23
Source: NPS Website (http://www.nps.or.kr). Accessed Sep 10, 2018.

107
Table A.3.3. Estimated incentive measures by age
Accrual Peak Value Option Value
Age 10th 50th 90th SD 10th 50th 90th SD 10th 50th 90th SD
55 24 45 75 19 244 469 703 177 587 1091 2155 631
56 24 49 79 22 204 419 640 163 462 920 1894 571
57 24 51 84 24 166 372 572 146 350 747 1625 511
58 23 54 90 26 140 318 487 128 260 583 1368 447
59 22 58 95 27 117 261 398 108 177 438 1063 368
60 -2 24 40 15 90 207 309 85 93 322 847 312
61 -1 25 40 15 89 186 272 74 62 257 681 267
62 4 24 38 13 81 165 239 64 37 204 542 226
63 3 22 36 12 71 142 205 55 18 154 413 188
64 3 21 34 11 64 119 168 47 10 110 336 155
65 6 21 34 11 55 96 135 38 4 76 250 125
66 10 20 31 10 48 73 106 30 1 50 191 94
67 9 19 28 9 37 53 79 22 0 30 136 69
68 12 18 28 9 23 35 52 15 0 20 100 48
69 11 16 26 8 11 16 26 8 0 15 63 39
Each incentive measure is estimated for each possible retirement age. SD is the standard deviation.
Source: Author calculations using KLoSA (2006-2016).

108
Table A.3.4. Probability of retirement for male wage workers (full estimates)
Accrual Peak Value Option Value
Social Security Wealth 0.0832* 0.0769 0.0670
(1.93) (1.57) (1.36)
Incentive Measures 0.0926 -0.00493 -0.0915**
(1.21) (-0.07) (-2.55)
Age 0.313*** 0.296*** 0.234***
(6.03) (4.87) (4.28)
Cohort group
1951-1947 -1.332*** -1.293*** -1.251***
(-5.87) (-5.91) (-5.74)
1941-1946 -2.944*** -2.866*** -2.625***
(-7.85) (-7.81) (-7.26)
Live in metropolitan area -0.266* -0.271* -0.311**
(-1.88) (-1.91) (-2.16)
Education
Middle school graduate 0.283 0.265 0.327
(1.10) (1.02) (1.30)
High school graduate 0.426 0.419 0.441
(1.44) (1.43) (1.52)
Some college and above -0.144 -0.11 0.0612
(-0.47) (-0.35) (0.19)
Live with at least one child -0.34 -0.338 -0.284
(-1.50) (-1.50) (-1.26)
Good self-reported health 0.0188 0.0322 0.003
(0.09) (0.15) (0.01)
Good mental health -0.0776 -0.0981 -0.0678
(-0.32) (-0.41) (-0.29)
Income group
Medium 0.486** 0.507** 0.512**
(2.29) (2.39) (2.52)
High 0.423 0.463* 0.697***
(1.52) (1.66) (2.63)
Experience 0.0676 0.0851 0.108
(0.42) (0.51) (0.60)
Experience squared -0.000205 -0.000405 -0.000627
(-0.12) (-0.22) (-0.33)
Job type
Office work 0.0396 0.0177 0.0273
(0.16) (0.07) (0.10)
Service & sales 0.223 0.226 0.101
(0.66) (0.67) (0.28)
Craft & machine operation -0.313 -0.315 -0.411*
(-1.31) (-1.31) (-1.67)
Agricultural, forestry, etc. 0.0548 0.00344 -0.103
(0.13) (0.01) (-0.23)
Manual work -0.0415 -0.0472 -0.156
(-0.18) (-0.20) (-0.643)
Constant -22.92*** -22.20*** -18.72***
(-4.81) (-4.19) (-3.48)
N 1563 1174 1174
SSW and incentive measures are in 100USD units. Standard errors are in parenthesis and are corrected with the
Huber-White robust method for heteroscedasticity. Dummy variables for age 60 and 61 are included. The
comparison base for cohort group is 1952-1956, base for education is middle school dropout and below, and for job
type it is professional work.

109
Bibliography

Acevedo, A., Mullin, B. O., Progovac, A. M., Caputi, T. L., McWilliams, J. M., & Cook, B. L.
(2021). Impact of the Medicare Shared Savings Program on utilization of mental health
and substance use services by eligibility and race/ethnicity. Health Services Research,
56(4), 581-591. doi:10.1111/1475-6773.13625

Aguila, E. (2014). Male Labor Force Participation and Social Security in Mexico. Journal of
Pension Economics and Finance 13(2), 145–71.
https://doi.org/10.1017/S1474747213000292.

Aguila, E., Fonseca, R., & Vega, A. (2015). Self-Employment, Health Insurance, and Return
Migration of Middle-Aged and Elderly Mexican Males. . In V. W., M. K., A. J., & T.-G.
F. (Eds.), Challenges of Latino Aging in the Americas: Springer, Cham.

Aguila, E., Lee, Z., & Wong, R. (2021). Migration, work, and retirement: the case of Mexican-
origin populations. Journal of Pension Economics and Finance, 1-21.
doi:10.1017/s1474747221000342

Aguilar-Navarro, S. G., Fuentes-Cantú, A., Ávila-Funes, J. A., & García-Mayo, E. J. (2007).
Validity and reliability of the screening questionnaire for geriatric depression used in the
Mexican Health and Age Study. Salud Publica de Mexico, 49(4), 256-263.

Alegría, M., Canino, G., Shrout, P. E., Woo, M., Duan, N., Vila, D., . . . Meng, X.-L. (2008).
Prevalence of Mental Illness in Immigrant and Non-Immigrant U.S. Latino Groups.
American Journal of Psychiatry, 165(3), 359-369. doi:10.1176/appi.ajp.2007.07040704

Altonji, J. G., Elder, T. E., & Taber, C. R. (2005). Selection on Observed and Unobserved
Variables: Assessing the Effectiveness of Catholic Schools. Journal of Political
Economy, 113(1), 151-184. doi:10.1086/426036

Andresen, E.M., Malmgren, J.A., Carter, W.B., Patrick, D.L., Carter, W.B., & Patrick. D.L.
(1994). Screening for Depression in Well Older Adults: Evaluation of a Short Form of
the CES-D. American Journal of Preventive Medicine 10(2), 77–84.
https://doi.org/10.1016/S0749-3797(18)30622-6.

Angel, J., Buckley, C., & Sakamoto, A. (2001). Duration or Disadvantage? Exploring Nativity,
Ethnicity, and Health in Midlife. Journal of Gerontology: SOCIAL SCIENCES, 56B(5),
S275-284.

110
Angrist, J., & Pischke, S. (2009). Mostly harmless econometrics: an empiricist’s companion.
Princeton, NJ: Princeton University Press.

Aroian, K. J., & Norris, A. E. (2003). Depression trajectories in relatively recent immigrants.
Comprehensive Psychiatry, 44(5), 420-427. doi:10.1016/s0010-440x(03)00103-2

Ayanian, J. Z. (2000). Unmet Health Needs of Uninsured Adults in the United States. JAMA,
284(16), 2061. doi:10.1001/jama.284.16.2061

Ayyagari, P., & Shane, D. M. (2015). Does prescription drug coverage improve mental health?
Evidence from Medicare Part D. Journal of Health Economics, 41, 46-58.
doi:10.1016/j.jhealeco.2015.01.006

Baek, E.Y. (2011). 베이비부머의 은퇴 준비와 준비 유형 결정요인 분석 [Determinants of
Retirement Preparation and the Types of Preparation of the Baby Boomers]. Korean
Social Security Studies 27(2), 357–83.

Baicker, K., Allen, H. L., Wright, B. J., Taubman, S. L., & Finkelstein, A. N. (2018). The Effect
of Medicaid on Management of Depression: Evidence From the Oregon Health Insurance
Experiment. The Milbank Quarterly, 96(1), 29-56. doi:10.1111/1468-0009.12311

Baily, M. N. (2019). How can policymakers improve retirement security? Retrieved from
https://www.brookings.edu/policy2020/

Barcellos, S. H., & Jacobson, M. (2015). The Effects of Medicare on Medical Expenditure Risk
and Financial Strain. American Economic Journal: Economic Policy, 7(4), 41-70.
doi:10.1257/pol.20140262

Becerra, D., Hernandez, G., Porchas, F., Castillo, J., Nguyen, V., & Perez González, R. (2020).
Immigration policies and mental health: examining the relationship between immigration
enforcement and depression, anxiety, and stress among Latino immigrants. Journal of
Ethnic & Cultural Diversity in Social Work, 29(1-3), 43-59.
doi:10.1080/15313204.2020.1731641

Behncke, S. (2012). Does retirement trigger ill health? Health Econ, 21(3), 282-300.
doi:10.1002/hec.1712

Bell, B., Blundell, R., & Van Reenen, J. (1999). Getting the Unemployed Back to Work:The
Role of Targeted Wage Subsidies. International Tax and Public Finance, 6, 339-360.

111
Belloni, M., & Alessie, R. (2009). The Importance of Financial Incentives on Retirement
Choices: New Evidence for Italy. Labour Economics 16(5), 578–88.
https://doi.org/10.1016/j.labeco.2009.01.008.

Belloni, M., Meschi, E., & Pasini, G. (2016). The Effect on Mental Health of Retiring During the
Economic Crisis. Health Economics, 25, 126-140. doi:10.1002/hec.3377

Bier, D. (2019, Jun 18). Immigration Wait Times from Quotas Have Doubled: Green Card
Backlogs Are Long, Growing, and Inequitable. Cato Institute. Retrieved from
https://www.cato.org/publications/policy-analysis/immigration-wait-times-quotas-have-
doubled-green-card-backlogs-are-long

Blanchflower, D. G. (2004). Self-employment: more may not be better. NBER Working Paper
10286.

Blanchflower, D. G., Oswald, A., & Stutzer, A. (2001). Latent entrepreneurship across nations.
European Economic Review, 45(4-6), 680-691. doi:10.1016/s0014-2921(01)00137-4

Blanchflower, D. G., & Oswald, A. J. (1998). What Makes an Entrepreneur? Journal of Labor
Economics, 16(1), 26-60. doi:10.1086/209881

Blau, D.M., & Goodstein, R.M. (2010). Can Social Security Explain Trends in Labor Force
Participation of Older Men in the United States ? The Journal of Human Resources 45(2),
328–63.

Blöndal, S., & Scarpetta, S. (1999). The Retirement Decision in OECD Countries. OECD
Economics Department Working Papers 202, 1–106.
https://doi.org/10.1787/565174210530.

Blundell, R., Meghir, C. & Smith, S. (2004). Pension Incentives and the Pattern of Retirement in
the United Kingdom. In J., Gruber & D.A., Wise (Ed.). Social Security Programs and
Retirement around the World : Micro-Estimation (pp.643–90). Chicago, IL: University of
Chicago Press.

Borella, M., De Nardi, M., & French, E. (2018). Who Receives Medicaid in Old Age? Rules and
Reality. Fiscal Studies, 39(1), 65-93. doi:10.1111/1475-5890.12145

Bradley, D. E., & Roberts, J. A. (2004). Self-Employment and Job Satisfaction: Investigating the
Role of Self-Efficacy, Depression, and Seniority. Journal of Small Business
Management, 42(1), 37-58. doi:10.1111/j.1540-627x.2004.00096.x

112
Buckley, C., Angel, J. L., & Donahue, D. (2000). Nativity and older women's health: constructed
reliance in the health and retirement study. J Women Aging, 12(3-4), 21-37.
doi:10.1300/J074v12n03_03

Burtless, Gary, and Joseph F. Quinn. (2001). Retirement Trends and Policies to Encourage Work
among Older Americans. In P.P., Budetti, et al. (Eds.). Ensuring Health and Income
Security for an Aging Workforce (pp.375–415). Kalamazoo, MI: W.E. Upjohn Institute
for Employment Research. https://doi.org/10.17848/9780880994668.ch18.

Bustamante, A. V., Chen, J., Felix Beltran, L., & Ortega, A. N. (2021). Health Policy Challenges
Posed By Shifting Demographics And Health Trends Among Immigrants To The United
States. Health Aff (Millwood), 40(7), 1028-1037. doi:10.1377/hlthaff.2021.00037

Byers, A. L., & Yaffe, K. (2011). Depression and risk of developing dementia. Nat Rev Neurol,
7(6), 323-331. doi:10.1038/nrneurol.2011.60

Caiola, S. (2021, Jul 1). California To Provide Health Insurance For Undocumented Residents
Over 50 Following Toll Of COVID-19. Capradio. Published Jul 1, 2021. Retrieved from
https://www.capradio.org/articles/2021/07/01/california-to-provide-health-insurance-for-
undocumented-residents-over-50-following-toll-of-covid-19/

Card, D., Dobkin, C., & Maestas, N. (2008). The Impact of Nearly Universal Insurance
Coverage on Health Care Utilization: Evidence from Medicare. American Economic
Review, 98(5), 2242-2258. doi:10.1257/aer.98.5.2242

Card, D., Dobkin, C., & Maestas, N. (2009). Does Medicare Save Lives?*. Quarterly Journal of
Economics, 124(2), 597-636. doi:10.1162/qjec.2009.124.2.597

Charles, K. K. (2004). Is retirement depressing?: Labor force inactivity and psychological well-
being in later life. In P. SW (Ed.), Accounting for Worker Well-being (Vol. 23, pp. 269-
299): Elsevier: Amsterdam.

Chen, Q., Chu, X., Wang, S., & Zhang, B. (2020). A Triple-Difference Approach to Re-
Evaluating the Impact of China's New Cooperative Medical Scheme on Incidences of
Chronic Diseases Among Older Adults in Rural Communities. Risk Manag Healthc
Policy, 13, 643-659. doi:10.2147/RMHP.S244021

Cho, J., & Kim, S. (2005). On Using Mandatory Retirement to Reduce Workforce in Korea.
International Economic Journal 19(2), 283–303.
https://doi.org/10.1080/10168730500080550.

113
Choi, Y.J. (2006). Transformations in Economic Security during Old Age in Korea: The
Implications for Public-Pension Reform. Ageing and Society 26(4), 549–65.
https://doi.org/10.1017/S0144686X06004879.

Christnacht, C., Smith, A., & Chenevert, R. (2018). Measuring entrepreneurship in the American
community survey: a demographic and occupational profile of self-employed workers.
Retrieved from: https://www.census.gov/library/working-papers/2018/demo/SEHSD-WP2018-
28.html

Coe, N. B., & Zamarro, G. (2011). Retirement effects on health in Europe. Journal of Health
Economics, 30(1), 77-86. doi:10.1016/j.jhealeco.2010.11.002

Coffman, M. J., & Norton, C. K. (2010). Demands of Immigration, Health Literacy, and
Depression in Recent Latino Immigrants. Home Health Care Management & Practice,
22(2), 116-122. doi:10.1177/1084822309347343

Coile, C. (2018). Working Longer in the U.S.: Trends and Explanations. NBER Working Paper
Series 24576. http://www.nber.org/papers/w24576.

Coile, C, & Gruber, J. (2001). Social Security Incentives for Retirement. In D.A., Wise (Ed.).
Themes in the Economics of Aging (pp. 311–54). Chicago, IL: University of Chicago
Press.

———. (2007). Future Social Security Entitlements and the Retirement Decision. Review of
Economics and Statistics 89(2), 234–46. https://doi.org/10.1162/rest.89.2.234.

Coverage For Immigrant Seniors. (n.d.). Illinois Department of Healthcare and Family Services. .
Retrieved from https://www.illinois.gov/hfs/Pages/CoverageForImmigrantSeniors.aspx

Dave, D., Rashad, I., & Spasojevic, J. (2008). The Effects of Retirement on Physical and Mental
Health Outcomes. Southern Economic Journal, 75(2), 497-523. doi:10.1002/j.2325-
8012.2008.tb00916.x

De Maio, F. G. (2010). Immigration as pathogenic: a systematic review of the health of
immigrants to Canada. International Journal for Equity in Health, 9(27), 1-20.

Decker, S., & Rapaport, C. (2002a). Medicare and Disparities in Women's Health. NBER
Working Paper 8761. doi:10.3386/w8761

114
Decker, S., & Rapaport, C. (2002b). Medicare and Inequalities in Health Outcomes: The Case of
Breast Cancer. Contemporary Economic Policy, 20(1), 1-11. doi:10.1093/cep/20

Donohue, J. M., Zhang, Y., Perera, S., Lave, J. R., Hanlon, J. T., Men, A., & Reynolds, C. F.
(2011). Impact of Medicare Part D on Antidepressant Treatment, Medication Choice, and
Adherence Among Older Adults With Depression. The American Journal of Geriatric
Psychiatry, 19(12), 989-997. doi:10.1097/jgp.0b013e3182051a9b

Drentea, P. (2002). Retirement and Mental Health. Journal of Aging and Health, 14(2), 167-194.
doi:10.1177/089826430201400201

Dunlop, D. D., Song, J., Lyons, J. S., Manheim, L. M., & Chang, R. W. (2003). Racial/Ethnic
Differences in Rates of Depression Among Preretirement Adults. American Journal of
Public Health, 93(11), 1945-1952. doi:10.2105/ajph.93.11.1945

Dwyer, D. S., & Mitchell, O. S. (1999). Health problems as determinants of retirement: Are self-
rated measures endogenous? Journal of Health Economics, 18(2), 173-193.
doi:10.1016/s0167-6296(98)00034-4

Eibich, P. (2015). Understanding the effect of retirement on health: Mechanisms and
heterogeneity. Journal of Health Economics, 43, 1-12.
doi:10.1016/j.jhealeco.2015.05.001

Engelhardt, G. V., & Gruber, J. (2011). Medicare Part D and the Financial Protection of the
Elderly. American Economic Journal: Economic Policy, 3(4), 77-102.
doi:10.1257/pol.3.4.77

Finkelstein, A., & McKnight, R. (2008). What did Medicare do? The initial impact of Medicare
on mortality and out of pocket medical spending. Journal of Public Economics, 92(7),
1644-1668. doi:10.1016/j.jpubeco.2007.10.005

Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse, J. P., . . . Baicker,
K. (2012). The Oregon Health Insurance Experiment: Evidence from the First Year. The
Quarterly Journal of Economics, 127(3), 1057-1106. doi:10.1093/qje/qjs020

Foreign-born workers made up 17.4 percent of labor force in 2019. (2020, May 29). The
Economics Daily, Bureau of Labor Statistics, US Department of Labor. Retrieved from
https://www.bls.gov/opub/ted/2020/foreign-born-workers-made-up-17-point-4-percent-
of-labor-force-in-2019.htm

115
French, E., & Jones, J. B. (2011). The Effects of Health Insurance and Self-Insurance on
Retirement Behavior. Econometrica, 79(3), 693-732. doi:10.3982/ecta7560

Fry, C. E., & Sommers, B. D. (2018). Effect of Medicaid Expansion on Health Insurance
Coverage and Access to Care Among Adults With Depression. Psychiatric Services,
69(11), 1146-1152. doi:10.1176/appi.ps.201800181

Galbraith, A. A., Wong, S. T., Kim, S. E., & Newacheck, P. W. (2005). Out-of-Pocket Financial
Burden for Low-Income Families with Children: Socioeconomic Disparities and Effects
of Insurance. Health Services Research, 40(6p1), 1722-1736. doi:10.1111/j.1475-
6773.2005.00421.x

Gindling, T. H., & Newhouse, D. (2014). Self-Employment in the Developing World. World
Development, 56, 313-331. doi:10.1016/j.worlddev.2013.03.003

Gonzalez, H. M., Haan, M. N., & Hinton, L. (2001). Acculturation and the Prevalence of
Depression in Older Mexican Americans: Baseline Results of the Sacramento Area
Latino Study on Aging. Journal of the American Geriatrics Society, 49, 948-953.

Gorry, A., Gorry, D., & Slavov, S. N. (2018). Does retirement improve health and life
satisfaction? Health Economics, 27(12), 2067-2086. doi:10.1002/hec.3821

Gray, B. H., Scheinmann, R., Rosenfeld, P., & Finkelstein, R. (2006). Aging without Medicare?
Evidence from New York City. Inquiry, 43(3), 211-221. doi:10.5034/inquiryjrnl_43.3.211

Gruber, J., & Wise, D.A. (2004). Social Security Programs and Retirement around the World :
Micro-Estimation. University of Chicago Press.

Gubernskaya, Z. (2015). Age at migration and self-rated health trajectories after age 50:
understanding the older immigrant health paradox. J Gerontol B Psychol Sci Soc Sci,
70(2), 279-290. doi:10.1093/geronb/gbu049

Guo, S., & Fraser, M. W. (2015). Propensity score analysis: Statistical methods and applications
(2 ed.). Thousand Oaks, CA: Sage Publications, Inc.

Gustman, A.L., & Steinmeier, T. (2009). How Changes in Social Security Affect Recent
Retirement Trends. Research on Aging 31(2), 261–90.
https://doi.org/10.1177/0164027508328312.

116
Ha, J, & Lee, S.H. (2018). Population Aging and the Possibility of a Middle-Income Trap in
Asia. Emerging Markets Finance and Trade 54(6), 1225–38.
https://doi.org/10.1080/1540496X.2018.1429263.

Hadley, J. (2003). Sicker and Poorer—The Consequences of Being Uninsured: A Review of the
Research on the Relationship between Health Insurance, Medical Care Use, Health,
Work, and Income. Medical Care Research and Review, 60(2_suppl), 3S-75S.
doi:10.1177/1077558703254101

Heckman, J. J. (1978). Dummy Endogenous Variables in a Simultaneous Equation System.
Econometrica, 46(4), 931-959.

Hirsch, J. K., Sirois, F. M., Visser, P. L., Brooks, B. D., Kohls, N., Offenbächer, M., &
Toussaint, L. (2019). Perceived stigma and health-related quality of life in the working
uninsured: Does thwarted belongingness play a role? Stigma and Health, 4(1), 1-10.
doi:10.1037/sah0000116

Health Coverage of Immigrants. (2021, Jul 15). Kaiser Family Foundation. Retrieved from
https://www.kff.org/racial-equity-and-health-policy/fact-sheet/health-coverage-of-
immigrants/

Hwang, S.J. (2016). Public Pensions as the Great Equalizer? Decomposition of Old-Age Income
Inequality in South Korea, 1998–2010. Journal of Aging and Social Policy 28(2), 81–97.
https://doi.org/10.1080/08959420.2016.1145503.

Ichimura, H, Lei, X., Lee, C., Lee, J., Park, A., & Sawada, Y. (2017). Wellbeing of the Elderly in
East Asia: China, Korea, and Japan: China, Korea, and Japan. RIETI Discussion Paper
Series 17029. http://www.rieti.go.jp/en/.

Jamal, M. (1997). Job stress, satisfaction, and mental health: An empirical examination of self-
employed and non-self-employed Canadians. Journal of Small Business Management,
35(4), 48-57.

Johnston, D. W., & Lee, W.-S. (2009). Retiring to the good life? The short-term effects of
retirement on health. Economics Letters, 103(1), 8-11. doi:10.1016/j.econlet.2009.01.015

Joulfaian, D., & Rider, M. (1998). DIFFERENTIAL TAXATION AND TAX EVASION BY
SMALL BUSINESS. National Tax Journal, 51(4), 675-687. doi:10.1086/ntj41789362

117
Kandula, N. R., Lauderdale, D. S., & Baker, D. W. (2007). Differences in Self-Reported Health
Among Asians, Latinos, and Non-Hispanic Whites: The Role of Language and Nativity.
Annals of Epidemiology, 17(3), 191-198. doi:10.1016/j.annepidem.2006.10.005

Kang, S.J., & Lee, M.W. (2001). Income Distribution and Public Transfers as Social Safety Nets
in Korea. Journal of Economic Development 26(2), 61–75.

Katon, W., & Ciechanowski, P. (2002). Impact of major depression on chronic medical illness.
Journal of Psychosomatic Research, 53(4), 859-863. doi:10.1016/s0022-3999(02)00313-
6

Kim, G., & Cho, J. (2009). Entry dynamics of self-employment in South Korea.
Entrepreneurship & Regional Development, 21(3), 303-323.
doi:10.1080/08985620802332707

Kim, E.H.W., & Cook, P.J. (2011). The Continuing Importance of Children in Relieving Elder
Poverty: Evidence from Korea. Ageing and Society 31(6), 953–76.
https://doi.org/10.1017/S0144686X10001030.

Knapp, M. (2003). Hidden costs of mental illness. British Journal of Psychiatry, 183(6), 477-
478. doi:10.1192/03-292

KOSIS. (2022). Korean Statistical Information Service. http://kosis.kr. (Accessed Oct 1, 2019).

Ku, L. (2009). Health Insurance Coverage and Medical Expenditures of Immigrants and Native-
Born Citizens in the United States. American Journal of Public Health, 99(7), 1322-1328.
doi:10.2105/ajph.2008.144733

Ku, L., & Jewers, M. (2013). Health Care for Immigrant Families: Current Policies and Issues.
Migration Policy Institute Reports. Retrieved from
https://www.migrationpolicy.org/research/health-care-immigrant-families-current-
policies-and-issues?pdf=COI-HealthCare.pdf

Kwon, H.J. (2001). Income Transfer to the Elderly in Korea and Taiwan. Journal of Social
Policy 30(1), 81–93.

Kwon, H. (2002). Welfare Reform and Future Challenges in the Republic of Korea: Beyond the
Developmental Welfare State? International Social Security Review 55(4), 23–38.
https://doi.org/10.1111/1468-246X.00136.

118
Kwon, S., & Hwang, K.S. (2004). 노년층의 은퇴 결정요인에 관한 연구 [A Study on
Determinants of the Elderly’s Retirement: Focusing on Differences by Gender and
Residential District]. Korea Gerontological Society 24(3), 69–90.

Lee, C. (2007). Long-Term Changes in the Economic Activity of Older Males in Korea.
Economic Development and Cultural Change 56(1), 99–123.
https://doi.org/10.1086/520555.

———. (2010). Labor Force Participation of Older Males in Korea: 1955 to 2005. In T., Ito &
A., Rose (Ed.). The Economic Consequences of Demographic Change in East Asia,
NBER-EASE Volume 19 (pp. 281-313). Chicago, IL: University of Chicago Press.
https://doi.org/10.7208/chicago/9780226386881.003.0009.

Lee, C., & Lee, J. (2013). Employment Status, Quality of Matching, and Retirement in Korea:
Evidence from Korean Longitudinal Study of Aging. Journal of Population Ageing 6(1–
2), 59–83. https://doi.org/10.1007/s12062-012-9080-x.

Lee, W.K.M., & Law, K.W.K. (2004). Retirement Planning and Retirement Satisfaction: The
Need for a National Retirement Program and Policy in Hong Kong. Journal of Applied
Gerontology 23(3), 212–33. https://doi.org/10.1177/0733464804268591.

Lee, Y.H., & Kim, W.S. (2013). 기획주제 : 인수위 기초연금 도입(안)에 대한 평가와 전망
[Review and Outlook on Proposal of the Presidential Transition Team on Introduction of
a Basic Pension Scheme]. Korean Social Security Studies 29(2), 1–25.

Lewin-Epstein, N., & Yuchtman-Yaar, E. (1991). Health Risks of Self-Employment. Work and
Occupations, 18(3), 291-312. doi:10.1177/0730888491018003003

Lin, H.Y. (2012). The Retirement Provisions in China, Hong Kong, Singapore and Taiwan:
Perspectives on Policy Inputs and Social Outputs. Asian Social Work and Policy Review
6(3), 163–91. https://doi.org/10.1111/j.1753-1411.2012.00069.x.

Liu, T., & Sun, L. (2016). Pension Reform in China. Journal of Aging and Social Policy 28(1),
15–28. https://doi.org/10.1080/08959420.2016.1111725.

Lum, T. Y., & Vanderaa, J. P. (2010). Health Disparities Among Immigrant and Non-immigrant
Elders: The Association of Acculturation and Education. Journal of Immigrant and
Minority Health, 12(5), 743-753. doi:10.1007/s10903-008-9225-4

Maddala, G. S. (1983). Limited Dependent and Qualitative Variables in Econometrics (1st ed.).
New York, NY: Cambridge University Press.

119

Markides, K. S., & Rote, S. (2015). Immigrant Health Paradox. Emerging Trends in the Social
and Behavioral Sciences, 1-15. doi:10.1002/9781118900772.etrds0174

McBride, K., Bacong, A. M., Reynoso, A., Benjamin, A. E., Wallace, S. P., & Kietzman, K. G.
(2020). Healthcare Decision-Making Among Dual-Eligible Immigrants: Implications
from a Study of an Integrated Medicare-Medicaid Demonstration Program in California.
Journal of Immigrant and Minority Health, 22(3), 494-502. doi:10.1007/s10903-019-
00922-5

McWilliams, J. M. (2009). Differences in Control of Cardiovascular Disease and Diabetes by
Race, Ethnicity, and Education: U.S. Trends From 1999 to 2006 and Effects of Medicare
Coverage. Annals of Internal Medicine, 150(8), 505. doi:10.7326/0003-4819-150-8-
200904210-00005

McWilliams, J. M., Meara, E., Zaslavsky, A. M., & Ayanian, J. Z. (2007a). Health of Previously
Uninsured Adults After Acquiring Medicare Coverage. JAMA, 298(24), 2886.
doi:10.1001/jama.298.24.2886

McWilliams, J. M., Meara, E., Zaslavsky, A. M., & Ayanian, J. Z. (2007b). Use of Health
Services by Previously Uninsured Medicare Beneficiaries. New England Journal of
Medicine, 357(2), 143-153. doi:10.1056/nejmsa067712

Mein, G. (2003). Is retirement good or bad for mental and physical health functioning? Whitehall
II longitudinal study of civil servants. Journal of Epidemiology & Community Health,
57(1), 46-49. doi:10.1136/jech.57.1.46

Midanik, L. T., Soghikian, K., Ransom, L. J., & Tekawa, I. S. (1995). The Effect of Retirement
on Mental Health and Health Behaviors: The Kaiser Permanente Retirement Study. The
Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 50B(1),
S59-S61. doi:10.1093/geronb/50b.1.s59

Min, J. W., Moon, A., & Lubben, J. E. (2005). Determinants of psychological distress over time
among older Korean immigrants and Non-Hispanic White elders: evidence from a two-
wave panel study. Aging Ment Health, 9(3), 210-222. doi:10.1080/13607860500090011

Moffitt, R. (1991). Program Evaluation With Nonexperimental Data. Evaluation Review, 15(3),
291-314. doi:10.1177/0193841x9101500301

120
Moussavi, S., Chatterji, S., Verdes, E., Tandon, A., Patel, V., & Ustun, B. (2007). Depression,
chronic diseases, and decrements in health: results from the World Health Surveys. The
Lancet, 370(9590), 851-858. doi:10.1016/s0140-6736(07)61415-9

Mueller, Z. (2019, May 23). Immigration 101: Do Immigrants Pay Taxes? America’s Voice.
Retrieved from https://americasvoice.org/blog/immigration-101-immigrant-taxes/

Mui, A. C. (2000). Stress, Coping, and Depression Among Elderly Korean Immigrants. Journal
of Human Behavior in the Social Environment, 3(3-4), 281-299.
doi:10.1300/j137v03n03_17

NABO. (2019). 2019~2060년 국민연금 재정전망 [Fiscal Projection of the National Pension
Service for 2019-2060]. Fiscal Estimates & Tax System Analysis Report. Vol. Aug.
2019. Seoul.

OECD. (2014). OECD Economic Surveys: Korea 2014. Paris: OECD Publishing.
https://doi.org/10.1787/eco_surveys-kor-2014-en.

———. (2017). OECD-Pensions at a Glance 2017. Paris: OECD Publishing.
https://doi.org/10.1787/pension_glance-2017-en.

———. (2022a). Life expectancy at birth (indicator). doi: 10.1787/27e0fc9d-en (Accessed Jan
31, 2022).

———. (2022b). OECD Labour Force Statistics. https://stats.oecd.org/. (Accessed Jan 31, 2022).

Oh, H., & Koyanagi, A. (2021). Immigrant status and psychotic experiences in the United States:
revisiting an immigrant paradox. Psychological Medicine, 1-3.
doi:10.1017/s0033291720005462

Pai, Y. (2006). Comparing Individual Retirement Accounts in Asia: Singapore, Thailand, Hong
Kong, and PRC. World Bank SP Discussion Paper, no. 0609.

Pfoh, E., Mojtabai, R., Bailey, J., Weiner, J. P., & Dy, S. M. (2015). Impact of Medicare Annual
Wellness Visits on Uptake of Depression Screening. Psychiatric Services, 66(11), 1207-
1212. doi:10.1176/appi.ps.201400524

Phang, H. (2011). Issues and Challenges Facing Population Ageing in Korea: Productivity,
Economic Growth, and Old-Age Income Security. Journal of Comparative Social
Welfare 27(1), 51–62. https://doi.org/10.1080/17486831.2011.532984.

121

Phillips, D.R. (2000). Ageing in the Asia-Pacific Region: Issues, Policies and Future Trends.
New York, NY: Routledge.

Polsky, D., Doshi, J. A., Escarce, J., Manning, W., Paddock, S. M., Cen, L., & Rogowski, J.
(2009). The Health Effects of Medicare for the Near-Elderly Uninsured. Health Services
Research, 44(3), 926-945. doi:10.1111/j.1475-6773.2009.00964.x

Poschke, M. (2019). Wage Employment, Unemployment and Self-Employment Across
Countries. SSRN Electronic Journal. doi:10.2139/ssrn.3401135

Prus, S. G., Tfaily, R., & Lin, Z. (2010). Comparing racial and immigrant health status and
health care access in later life in Canada and the United States. Can J Aging, 29(3), 383-
395. doi:10.1017/S0714980810000358

Purcell, P. (2009). Older Workers: Employment and Retirement Trends. Washington, DC.
http://digitalcommons.ilr.cornell.edu/key_workplace/655.

Rietveld, C. A., van Kippersluis, H., & Thurik, A. R. (2015). Self-Employment and Health:
Barriers or Benefits? Health Econ, 24(10), 1302-1313. doi:10.1002/hec.3087

Seok, S.H. (2010). 기초노령연금의 노인빈곤감소 효과 분석 [An Analysis on the Anti-
Poverty Effectiveness of the Basic Old-Age Pension System]. Journal of Welfare for the
Aged 50: 335–52.

Shetterly, S. M., Baxter, J., Mason, L. D., & Hamman, R. F. (1996). Self-rated health among
Hispanic vs non-Hispanic white adults: the San Luis Valley Health and Aging Study.
American Journal of Public Health, 86(12), 1798-1801. doi:10.2105/ajph.86.12.1798

Siddharthan, K. (1991). Health Insurance Coverage of the Immigrant Elderly. Inquiry, 29(4),
403-412.

Solinge, H. (2014). Who Opts for Self-Employment after Retirement? A Longitudinal Study in
the Netherlands. European Journal of Ageing 11(3), 261–72.
https://doi.org/10.1007/s10433-013-0303-7.

Son, J.C. (2010). 중고령자 은퇴 및 은퇴 만족도 결정요인 분석 [Determinants of Retirement
and Retirement Satisfaction in Korea]. Quarterly Journal of Labor Policy 10(2), 125–53.

122
Steffick, D. E. (2000). Documentation of Affective Functioning Measures in the Health and
Retirement Study. HRS/AHEAD Documentation Report, DR-005.

Stock, J.H., & Wise, D.A. (1990). Pensions, the Option Value of Work, and Retirement.
Econometrica 58(5), 1151. https://doi.org/10.2307/2938304.

Szinovacz, M. E., & Davey, A. (2004). Retirement Transitions and Spouse Disability: Effects on
Depressive Symptoms. The Journals of Gerontology Series B: Psychological Sciences
and Social Sciences, 59(6), S333-S342. doi:10.1093/geronb/59.6.s333

Tetrick, L. E., Slack, K. J., Da Silva, N., & Sinclair, R. R. (2000). A comparison of the stress–
strain process for business owners and nonowners: Differences in job demands,
emotional exhaustion, satisfaction, and social support. Journal of Occupational Health
Psychology, 5(4), 464-476. doi:10.1037/1076-8998.5.4.464

Tolbert, J., Artiga, S., & Pham, O. (2019, Oct 15). Impact of Shifting Immigration Policy on
Medicaid Enrollment and Utilization of Care among Health Center Patients. Kaiser
Family Foundation. Retrieved from https://www.kff.org/medicaid/issue-brief/impact-of-
shifting-immigration-policy-on-medicaid-enrollment-and-utilization-of-care-among-
health-center-patients/

US Census Bureau. (2019). American Community Survey 1-Year Estimates. Retrieved from
https://www.census.gov/topics/population/foreign-born/guidance/acs-guidance.html

US Census Bureau. (2021). Annual Social and Economic Supplement. Current Population
Survey. Retrieved from https://www.census.gov/topics/population/foreign-
born/data/tables/cps-tables.html

US Census Bureau. (2018). Population projections 2017-2060. Retrieved from
https://www.census.gov/programs-surveys/popproj.html

Vega, A., & Aguila, E. (2017). Foreign retirement income among new older immigrants in the
United States. Int Migr, 55(3), 38-56. doi:10.1111/imig.12325

Vega, A., Porteny, T., & Aguila, E. (2018). The Role of Clinics in Determining Older Recent
Immigrants' Use of Health Services. J Immigr Minor Health, 20(6), 1468-1475.
doi:10.1007/s10903-018-0693-x

Vo, K., Forder, P. M., Tavener, M., Rodgers, B., Banks, E., Bauman, A., & Byles, J. E. (2015).
Retirement, age, gender and mental health: findings from the 45 and Up Study. Aging &
Mental Health, 19(7), 647-657. doi:10.1080/13607863.2014.962002

123

Wagstaff, A. (2010). Estimating health insurance impacts under unobserved heterogeneity: the
case of Vietnam's health care fund for the poor. Health Economics, 19(2), 189-208.
doi:10.1002/hec.1466

Wells, K. B., Sherbourne, C. D., Sturm, R., Young, A. S., & Audrey Burnam, M. (2002).
Alcohol, Drug Abuse, and Mental Health Care for Uninsured and Insured Adults. Health
Services Research, 37(4), 1055-1066. doi:10.1034/j.1600-0560.2002.65.x

Wilde, J. (2000). Identification of multiple equation probit models with endogenous dummy
regressors. Economics Letters, 69(3), 309-312. doi:10.1016/s0165-1765(00)00320-7

Williams, D. R., Priest, N., & Anderson, N. B. (2016). Understanding associations among race,
socioeconomic status, and health: Patterns and prospects. Health Psychol, 35(4), 407-
411. doi:10.1037/hea0000242

Yashiro, N., & Oshio, T. (1999). Social Security and Retirement in Japan. In J., Gruber & D.A.,
Wise (Ed.). Social Security Programs and Retirement around the World (pp.239–67).
Chicago, IL: University of Chicago Press.
https://doi.org/10.7208/chicago/9780226309989.003.0008.

Yoon, J., & Bernell, S. L. (2013). The effect of self-employment on health, access to care, and
health behavior. Health, 05(12), 2116-2127. doi:10.4236/health.2013.512289

Yoon, S.J. (2014). 현안분석 1: 기초연금제도의 쟁점과 과제 [Analysis 1: Current Issues and
Tasks of the Basic Pension]. Public Finance Forum 211(0), 8–24.

Zallman, L., Woolhandler, S., Himmelstein, D., Bor, D., & McCormick, D. (2013). Immigrants
contributed an estimated $115.2 billion more to the Medicare Trust Fund than they took
out in 2002-09. Health Aff (Millwood), 32(6), 1153-1160. doi:10.1377/hlthaff.2012.1223

Zissimopoulos, J., & Karoly, L. A. (2007). Work and well-being among the self-employed at old
ages. AARP Public Policy Institute.

Zivin, K., Llewellyn, D. J., Lang, I. A., Vijan, S., Kabeto, M. U., Miller, E. M., & Langa, K. M.
(2010). Depression among older adults in the United States and England. Am J Geriatr
Psychiatry, 18(11), 1036-1044. doi:10.1097/JGP.0b013e3181dba6d2

Abstract (if available)

Abstract Population aging comes with significant challenges. Preparing financially for longer lives and finding ways to live a healthier life has become a priority for many adults and their families. Governments are facing fiscal challenges as the aging population strains pension systems and affects economic growth and disease patterns. This dissertation focuses on various challenges encountered by the demographic change. The first chapter examines the foreign-born population in the United States and addresses the fact that many older immigrants remain uninsured. The study finds that Medicare coverage is associated with better mental health, particularly among immigrants with low socioeconomic status. The second chapter compares the health of older self-employed workers in three countries – Korea, Mexico, and the United States – and shows that workers with poor health select into retirement, but the selection is weaker in countries with less developed social security systems. The results suggest that older workers may involuntarily stay in the labor force due to a lack of retirement income, further exacerbating their health and creating higher costs to society. The last chapter focuses on older wage workers in Korea, a country with the highest elderly poverty rate and elderly suicide rate, and finds that retirement choice of older workers in Korea are influenced by transfers from children rather than pension benefits. This contrasts with the experience in North America and Europe, further illustrating that rapidly aging countries, such as Korea, may not have sufficient time and resources to tackle issues associated with population aging. These chapters provide a rich new portrait of how the aging population is transforming the world in fundamental ways and emphasize the need for policy reforms, particularly for marginalized populations.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on work, retirement, and fostering longer working lives

PDF

Health care utilization and spending of the U.S. aging population

PDF

Three essays on health economics

PDF

Essays on health insurance programs and policies

PDF

Three essays on health & aging

PDF

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

PDF

Three essays on the evaluation of long-term care insurance policies

PDF

Three essays on aging, wealth, and housing tenure transitions

PDF

Uncovering hidden figures: disparities in dementia burden in an aging America

PDF

Social engagement and cognitive decline in Mexican Americans: implications for age-friendly cities

PDF

Essays on mitigation, adaptation, and resilience to urban fiscal crises

PDF

Essays on the U.S. market for substance use treatment and the impact of Medicaid policy reform

PDF

Essays on Japanese macroeconomy

PDF

Essays in opioid use and abuse

PDF

Cognitive health and self-rated memory in later life: linkages to race/ethnicity, multimorbidity, and survival

PDF

Climate change, air pollution and health: how policy can induce adaptation

PDF

Biomarkers of age-related health changes: associations with health outcomes and disparities

PDF

The role of the modern aging network: measuring innovations of Area Agencies on Aging

PDF

Mechanisms of the effect of involuntary retirement on older adults' health and mental health

PDF

Essays on macroeconomics of health and labor

Asset Metadata

Creator Jun, Hankyung (author)

Core Title Three essays on health, aging, and retirement

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Public Policy and Management

Degree Conferral Date 2022-05

Publication Date 04/18/2022

Defense Date 02/28/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics of aging,health insurance,immigrant health,Mental Health,OAI-PMH Harvest,pension,Retirement,self-employment,Social Security

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Aguila, Emma (committee chair), Chen, Alice (committee member), Mattke, Soeren (committee member)

Creator Email hankyunj@usc.edu,junhanky@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111006513

Unique identifier UC111006513

Document Type Dissertation

Format application/pdf (imt)

Rights Jun, Hankyung

Type texts

Source 20220418-usctheses-batch-928 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu