Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Three essays on health economics
(USC Thesis Other)
Three essays on health economics
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THREE ESSAYS ON HEALTH ECONOMICS
by
Yaoyao Zhu
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
(ECONOMICS)
August 2016
Copyright 2016 Yaoyao Zhu
ii
ACKNOWLEDGEMENTS
First and foremost, I would like to express my deepest gratitude to my advisor John
Strauss, for his continuous support and guidance. He introduced me to the field of De-
velopment Economics and Health Economics, and for the past years, he has patiently
guided me to conduct research and to become a better researcher. I learnt a great deal
from him with his countless hours of advice during my Ph.D. years. He has been a true
example to follow and this dissertation would not have been possible without his men-
toring and encouragement. I would also like to thank Jeff Nugent, Eileen Crimmins, Geert
Ridder and Anant Nyshadham for being part of my committee. Their comments and
feedback were very helpful.
I was also very fortunate to have the Department of Economics as a family for the
past years. Morgan Ponder, Young Miller and Fatima Perez have been very attentive to
help me solve all the problems. I thank all my friends and colleagues in the department.
Life at the University of Southern California is enjoyable with all of you around. And
among them I owe special thanks to Nazmul Ahsan, Cheng Zhou, Teresa Molina, Rakesh
Banerjee, and Ashish Sachdeva for their valuable feedback and immense support.
Last but not least, I want to thank my family for their endless love and caring. I am
grateful to my family for encouraging me in all of my pursuits and believing in me when
I doubted myself. This journey would not have been possible without the support from
them.
iii
Table of Contents
Acknowledgements ..................................................................................................................... ii
Table of Contents ........................................................................................................................ iii
List of Figures .............................................................................................................................. vi
List of Tables ............................................................................................................................... vii
Chapter 1 Introduction .................................................................................................................1
Chapter 2 Does Retirement Affect Health? Evidence from CHARLS ...................................5
2.1 Introduction ..........................................................................................................................5
2.2 Retirement and Health Study ...........................................................................................8
2.3 Retirement System in China ...........................................................................................12
2.4 Data and Measurement ...................................................................................................13
2.4.1 China Health and Retirement Longitudinal Study (CHARLS) ...........................13
2.4.2 Sample ..........................................................................................................................14
2.4.3 Measurement of Retirement......................................................................................15
2.4.4 Measurement of Health .............................................................................................16
2.4.4.1 General Health Status, Disability, Pain and Life Satisfaction ......................16
2.4.4.2 Cognitive Ability and Depressive Symptoms ...............................................16
2.4.4.3 Non-blood and Blood Biomarkers ...................................................................17
2.4.5 Descriptive Statistics .................................................................................................18
2.5 Empirical Approach .........................................................................................................21
2.6 Results ................................................................................................................................23
2.6.1 Discontinuity in Retirement Probability .................................................................23
2.6.2 Fuzzy Regression Discontinuity Results ................................................................24
2.6.3 Spousal Effects ...........................................................................................................30
2.6.4 Robustness Checks .....................................................................................................34
2.7 Conclusion and Discussion .............................................................................................35
iv
Chapter 3 Health and Labor Supply in Rural China ............................................................37
3.1 Introduction .......................................................................................................................37
3.2 Literature Review .............................................................................................................42
3.3 Framework ........................................................................................................................44
3.4 Data and Measurement ...................................................................................................45
3.4.1 Data and Sample ........................................................................................................45
3.4.2 Measurement of Labor Supply ................................................................................45
3.4.3 Measurement of Health ............................................................................................47
3.4.4 Covariates ...................................................................................................................49
3.5 Empirical Strategy ............................................................................................................51
3.6 Results ................................................................................................................................53
3.6.1 Simultaneous Decision of LFP and Hours of Work .............................................53
3.6.2 Family and Household Structures ..........................................................................58
3.6.3 Household Economic Resources .............................................................................63
3.6.4 Interactions between Health and Wealth ...............................................................64
3.7 Conclusion and Discussion .............................................................................................69
Chapter 4 The Dynamics of Health Changes and Employment Transitions ....................71
4.1 Introduction ........................................................................................................................71
4.2 Health and Labor Market Outcomes .............................................................................74
4.3 Model .................................................................................................................................78
4.3.1 Conceptual Framework ............................................................................................78
4.3.2 Empirical Specification .............................................................................................79
4.4 Data and Measurement ...................................................................................................83
4.4.1 Data ..............................................................................................................................83
4.4.1.1 China Health and Retirement Longitudinal Study (CHARLS) ....................83
4.4.4.2 Sample .................................................................................................................83
4.4.2 Measurement of Labor Market Behavior ...............................................................85
4.4.3 Measurement of Health ............................................................................................86
4.4.4 Covariates ...................................................................................................................87
4.5 Results ................................................................................................................................89
4.5.1 Health Changes and Employment Transitions .....................................................89
4.5.2 Family Composition and Employment Transitions .............................................96
v
4.5.3 The Impacts of Demographic and Socioeconomic Characteristics ....................97
4.5.4 Simulations .................................................................................................................99
4.5.4.1 Simulation Effects of Health Changes ............................................................99
4.5.4.2 Simulation Effects of Family Composition ...................................................103
4.5.5 Health Changes and Annual Work Hours ..........................................................106
4.5.6 Heterogeneity ...........................................................................................................109
4.5.6.1 Agricultural Workers v.s. Non-agricultural Workers ................................109
4.5.6.2 Interactions of Health with Education ..........................................................109
4.6 Conclusion and Discussion ...........................................................................................110
Chapter 5 Conclusion ..............................................................................................................112
Bibliography ..............................................................................................................................114
Appendix ....................................................................................................................................124
vi
List of Figures
Figure 2.1 Discontinuity in Retirement Probability ..............................................................13
Figure 2.2 Discontinuity in Retirement Probability (Formal Sector) ...................................19
Figure 3.1 Labor Force Participation (Rural Elderly) ............................................................38
Figure 3.2 Annual Work Hours (Rural Elderly) .....................................................................38
Figure 3.3 Annual Work Hours in Farming (Rural Elderly) ................................................39
Figure 4.1 Probability of Employment (Rural Elderly) .........................................................72
vii
List of Tables
Table 2.1 Summary Statistics ....................................................................................................20
Table 2.2 Discontinuity in Retirement Probability ................................................................24
Table 2.3 Impact of Retirement on Self-Reported Health, Mental Health, and
Biomarkers: Men ....................................................................................................................25
Table 2.4 Impact of Retirement on Self-Reported Health, Mental Health, and
Biomarkers: Women .............................................................................................................27
Table 2.5 First-stage Check for Spouse Retirement ...............................................................31
Table 2.6 Impact of Retirement and Spouse Retirement on Self-Reported Health,
Mental Health, and Biomarkers .........................................................................................32
Table 3.1 Summary Statistics ....................................................................................................46
Table 3.2 Health and Simultaneous Labor Supply Decision: Men .....................................54
Table 3.3 Health and Simultaneous Labor Supply Decision: Women ...............................55
Table 3.4 Health and Simultaneous Labor Supply Decision in Agriculture .....................57
Table 3.5 Effects from Family Composition ............................................................................59
Table 3.6 Effects from Family Composition (Agriculture) ....................................................61
Table 3.7 Interactions between Health and Wealth: Men ....................................................65
Table 3.8 Interactions between Health and Wealth: Women ..............................................67
Table 4.1 Summary Statistics ....................................................................................................84
Table 4.2 Employment Transitions Matrices ..........................................................................86
Table 4.3 Continued Employment Decision among Respondents Employed at
the Baseline .............................................................................................................................92
Table 4.4 Labor Force Re-entry Decision among Respondents Not Employed at
the Baseline ............................................................................................................................94
Table 4.5A Simulated Likelihood of Continued Employment w.r.t. Health
Changes .................................................................................................................................100
Table 4.5B Simulated Likelihood of Labor Market Re-entry w.r.t. Health
Changes .................................................................................................................................101
viii
Table 4.6A Simulated Likelihood of Continued Employment w.r.t. Changes in
Family Composition ............................................................................................................104
Table 4.6B Simulated Likelihood of Labor Market Re-entry w.r.t. Changes in
Family Composition ............................................................................................................105
Table 4.7 OLS Estimation of Change in Annual Hours ......................................................107
Table A2.1 Robustness Check: Discontinuity in Retirement Probability .........................124
Table A2.2 Robustness Check: Impact of Retirement (Men) ..............................................125
Table A2.3 Robustness Check: Impact of Retirement (Women) ........................................127
Table A4.1 Estimation for Initial Employment Status Equation: Continue
Employment Decision .........................................................................................................129
Table A4.2 Estimation for Initial Employment Status Equation: Labor Market
Re-entry Decision .................................................................................................................131
Table A4.3 Continued Employment Decision across Types of Jobs .................................133
Table A4.4 Estimation for Initial Employment Status Equation across Types
of Jobs .....................................................................................................................................135
Table A4.5 Employment Transition and Interactions of Health with Education ............136
Table A4.6 Continued Employment Decision among Respondents Employed
at the Baseline (Probit).........................................................................................................138
Table A4.7 Labor Market Re-entry Decision among Respondents Not Employed
at the Baseline (Probit).........................................................................................................140
1
Chapter 1
Introduction
This dissertation focuses on the health and labor supply of the aging population in China. Specif-
ically, I explore the interplay between elderly’s health and labor supply behaviors, along with
demographics, family structure and economic resources. While health may induce changes in
labor supply behaviors, labor market transitions may as well have an impact on the elderly’s
health. A thorough understanding of the dynamics between these two elements may have im-
portant policy implications; especially in a context where social policies regarding health and
employment are not yet well-established.
Chapter 2 aims to evaluate the contemporaneous impact of retirement on health and cogni-
tion for the elderly in China. In this chapter, I target on the urban residents. Unlike their rural
counterparts, urban employees in the formal sector face mandatory retirement, where men are
required to retire at age 60, female civil servants at age 55 and female workers at age 50. They are
also eligible for pension wages afterwards. To establish the causal link between retirement and
health, I exploit the properties of the mandatory retirement system in the formal sector to account
for the endogeneity issue in the retirement status. This property allows me to apply the Fuzzy
Regression Discontinuity (FRD) design since the probability of retirement jumps at the threshold
ages.
Since health is multi-dimensional, I examine the impact of retirement on a broad dimension
of health outcomes, including self-reported health, life satisfaction, cognitive ability, depressive
symptoms, non-blood and blood biomarkers. Using the China Health and Retirement Longitudi-
nal Study (CHARLS) 2011 national baseline, my estimation results suggest that retirement has
minimal impact on both men and women’s health in general. However, I do find that men are
significantly more likely to suffer from body pains and overweight problems after retirement.
Retirement also significantly reduces their life satisfaction. Health effects on women are mixed. I
find that retired women are more likely to have hypertension and high glucose levels, as well as
memory loss. However, exiting the labor force to some extent alleviates their body pains and
anemia.
2
I also examine the potential impact on an individual’s health from his/her spouse’s retire-
ment status. Rarely any sign of a spousal effect has been found. There is slight evidence suggest-
ing that individuals living with a female spouse are less likely to have overweight problems. It
also seems that women are more likely to maintain their mental status and are less likely to have
depressive symptoms if they have retired male spouses.
Results from Chapter 2 have important policy implications. Since the current mandatory re-
tirement is particularly low in China and the argument as to whether the Chinese government
should raise its retirement age has been fiercely debated for the past decade. Findings from this
chapter suggest that postponing retirement is not bad for health, and it lends support to raising
the retirement ages.
While Chapter 2 focuses on the impact of retirement on health of the aging population in
urban areas; the following two chapters center around the rural aging population and explore the
static and dynamic associations between health and employment. The rural population in China
bears some characteristics that distinguish themselves from the elderly in the other parts of the
world. First, most of them work ceaselessly as long as their health permits it. Second, a majority
of the population are involved in farming, where strenuous labor is required. For them, labor
supply heavily relies on their health status. Third, the public pension is almost absent in the rural
areas and rural elderly to some extent rely on the support from their extended family. Last, since
the ownership of the land belongs to the state and collectives in China, rural elderly are not able
to accumulate land to support themselves or to encourage intergenerational transfer, which
makes them very ill-prepared for growing old. With all these characteristics, it is interesting to
explore how the interrelation between health and employment behaves in this context and
whether the linkage between these two factors is strengthened for the rural aging population.
Chapter 3 examines the static linkage between health and the rural elderly’s labor supply
behaviors in China. Here the labor supply behaviors are characterized by both work participation
and annual hours of work. Using the CHARLS 2011 national baseline, I apply a bivariate alterna-
tive to the commonly used Tobit model, assuming that the stochastic processes for the continuous
work hours and the discrete participation switch are different. In the model, the same factors that
affect participation and work hours could have dissimilar effects on the two different decisions.
Findings from Chapter 3 suggest a strong association between health and labor supply. Spe-
cifically, individuals who reported poor health or disability in the current period are more likely
3
to leave the labor force at old ages. Both men and women with poor health work fewer hours.
These associations persist after I control for household characteristics and economic resources. I
also find differentiated health effects for old-aged adults with different levels of wealth. It seems
that wealth facilitates retirement for both elderly men and women in the rural areas.
The interrelation between health and labor supply might be better characterized as a dy-
namic process. Using two waves of data from CHARLS which were fielded in 2011 and 2013, I
examine two sets of labor market transitions: continued employment and labor market re-entry
by those who had left employment. By relating work transitions with both lagged and contem-
poraneous health, I examine how the elderly respond to changes in their health and the magni-
tude of its impact on labor supply behaviors. Both the transition process and the initial work
status are jointly estimated using Full Information Maximum Likelihood (FIML) method to ac-
count for the potential baseline self-selection bias.
For the health measurements, I include five different dimensions: self-reported general
health status, disability, depressive symptoms, biomarkers and parental longevity. Health
changes in this study are captured by changes in the first three dimensions. My results suggest
that, for both men and women, adverse changes in self-reported general health status and disa-
bility are significantly associated with lower probability of labor market participation: either in
the form of continued employment or workforce re-entry. I also observe that not just poor health,
but also declines in health, encourage the elderly to transit out of the workforce.
Family network is also found to have an important impact on the elderly’s work transitions.
It seems that having more adult daughters is significantly associated with lower work participa-
tion for individuals employed or non-employed at the baseline. When I further differentiate the
family members according to their marital status, the previous effects from adult daughters are
mainly from the married ones. In addition to sons and daughters, I also find significant associa-
tion between grandchild and the elderly’s work transitions. My results suggest different roles of
men and women in the upbringings of their grandchildren. It seems that elderly women quit their
jobs to help take care of the grandchildren, while elderly men re-enter the workforce to better
provide for them.
To estimate the magnitude of the participation responses, I conduct several simulations by
varying individuals’ health. The results indicate that the magnitudes of the adjustments in both
4
of the work transitions are large in response to health changes. There is also some evidence sug-
gesting that the participation responses of individuals to health shocks are asymmetric. In con-
trast to the “ratchet effect” that has been found in some studies of the developed countries, I find
that labor market re-entrants in rural China are much more sensitive to their health changes.
The remainder of this dissertation is organized as follows: Chapter 2 discusses the impact
of retirement on health outcomes; Chapter 3 examines the static association between health and
elderly’s labor supply behaviors; Chapter 4 further explores the dynamic association between
health changes and elderly’s employment transitions, along with the impact from family; a sum-
mary of the main findings and potential policy implications are discussed in Chapter 5.
5
Chapter 2
Does Retirement Affect Health? Evidence from CHARLS
2.1 Introduction
Unlike U.S. and some other industrialized countries, where individuals choose when to retire
1
,
people working in the formal sector
2
in China face mandatory retirement. In the formal system,
men should cease working at age 60. For women, among female civil servants, the mandatory
retirement age is 55, and for female workers in the formal system, it is 50. These age restrictions,
no matter for which group of people, is quite low when compared with others in most developed
countries. For example, in the U.S., early retirement age is 62 while the standard is 67. For most
European countries
3
, the age of eligibility for public pension is also above 65. Even if compared
to those of some developing countries
4
, the mandatory retirement ages in China are still low,
especially for women.
For the past decade, the argument as to whether or not the Chinese government should raise
its retirement age has been fiercely debated. Advocates of this increase mainly argue that this
would solve several problems brought on by the rapidly aging population. China has been expe-
riencing a health revolution for the past 50 years, and life expectancy has risen from 46 in the
1950s to currently above 76 (Wagstaff et al., 2009). At the same time, China’s population is aging
rapidly. In 2000, the proportion of individuals aged 60 and above was 10% of the total population;
however, by 2050, this is projected to be above 30% (United Nations, 2009). The fast-growing
aging population, along with the reduction in the fertility rate brought on by the family planning
policies and economic growth, has put a significant burden on both the pension system and the
younger generation. With the imperfect pension system and the embedded tradition that children
1
Exception exists for particular sectors or professions. For instance, military service, pilot and judges in
several states in the U.S. are still subject to mandatory retirement. Most of the European countries such as
France, Portugal, and Belgium have mandatory retirement age for public sector civil servants (European
Commission, 2011).
2
The formal sector in this study is defined as government, institutions, NGOs and firms.
3
For instance, retirement age for men is 62, 65, 66, and 67 in France, United Kingdom, Italy, and Norway
respectively.
4
For example, retirement age for both sexes in India is 60.
6
should support their parents when they grow old
5
, the younger generation faces unprecedented
pressures. Take the older dependency ratio
6
as an example: early in 2000, this ratio in China was
12. However, within less than 40 years, it is projected to increase dramatically to 40 (U.S. Census
Bureau, 2008). Therefore, raising retirement ages, to some degree, might alleviate the pressure on
both the pension system and the younger workers by inducing the elderly to work longer.
In addition, evidence from retirement and health studies have been cited to validate the ne-
cessity of increasing the retirement age. For example, retirement is found to significantly increase
the risk of being diagnosed with a chronic condition, increase difficulties associated with mobility
and daily activities, and it also has a negative effect on cognitive functioning (Dave et al., 2007;
Behncke, 2012; Bonsang et al., 2012; Mazzonna and Peracchi, 2012.). Therefore, in the health per-
spective of the elderly, it might be good to postpone retirement and at the same time to promote
labor force participation.
In fact, a number of the countries have already started to increase the retirement age gradu-
ally. For example, in the U.S., the normal retirement age for social security was 65, but it has now
been increased to 67. Likewise, in most European countries, retirement ages are also set to in-
crease to around 70 within the next few decades
7
.
However, it is too soon to draw a conclusion. First, empirical evidence from retirement and
health studies are not conclusive. Apart from the negative effects of retirement found from the
literature cited above, studies showing positive associations between retirement and health also
exist (Bound and Waidman, 2007; Coe and Zamarro, 2011). Second, most of the existing studies
are in the context of industrialized countries. There are few studies that examine the problem in
developing regions, particularly not in China. Countries differ considerably in cultural norms,
labor markets, health insurance and social policies; hence, there is no a priori reason to assume
that findings from the developed countries will hold for China.
5
When asked about the old-age support preferences, nearly 70% of the middle-aged and elderly in China
choose to rely on their children when they grow old. Around 20% take pension as an old age support
method. Less than 10% of respondents choose savings, commercial pension insurance and other methods
(author’s estimation using the CHARLS 2011 national baseline).
6
Older dependency ratio is the number of people aged 65 and over per 100 people aged 20 to 64.
7
For example, in the UK, the retirement age for women is to be increased gradually and equaled to the
retirement age for men in 2018. Later, the retirement age for both sexes is to be increased gradually and
reach 68 by 2046 or sooner. Similarly, in France the retirement age is to be increased gradually to 68 years
and in Germany the retirement age is to be increased gradually and reach 67 years in 2029.
7
My paper tries to fill this research gap by investigating how retirement affects health in China
using the Chinese Health and Retirement Longitudinal Study (CHARLS) national baseline. This
study contributes to the literature in several ways. First, the setting is quite different compared to
those of industrialized countries that have been studied in the previous literature. In those coun-
tries
8
, retirement is a choice. In this setting, researchers face a severe endogeneity problem since
individuals with poor health may tend to retire early (Belgrave et al., 1987). On the contrary,
individuals working in the formal sector in China face mandatory retirement, where workers are
required to retire when they reach the mandatory retirement age. This allows us to apply the
Fuzzy Regression Discontinuity (FRD) design since the probability of retirement jumps at the
threshold ages. The FRD design allows estimation of causal impacts under certain assumptions
which seem to be met in the data.
Second, although this paper is not the first study that applies FRD design to study the link
between retirement and health in China, it has the most comprehensive set of health variables
used in studies to date. Using the 2005 Census data, Lei et al. (2010) examine the effect of retire-
ment on health. However, due to data limitations, only self-reported health is used to evaluate
the impact of retirement. As is known, health is multi-dimensional, focusing only on one aspect
of health may lead to an incomplete evaluation of retirement’s impact. In Lei et al. (2011), they
further examine the effect of retirement on several more health measures using the 2002 China
Household Income Project (CHIP) urban data; however, the information on health are still limited.
Thanks to the richness of the CHARLS, in this study health can be measured in several dimen-
sions, including self-reported health status, life satisfaction, biomarkers, cognitive functioning
and depressive symptoms. Results using both self-reported and biomarker measures can help
check for problems such as justification bias
9
, it can provide a more comprehensive understand-
ing of retirement’s effect.
Third, unlike Lei et al. (2011) that focus on the urban residents as a whole, we further restrict
the sample to the urban formal sector workers who are the ones most clearly affected by the man-
datory retirement rule. In addition, we differentiate men and women, who are faced with differ-
ent retirement ages and might have different health responses towards retirement.
8
Mostly U.S. and some of the European countries. However, there are several particular sectors or profes-
sions that still have a mandatory retirement age.
9
For example, justification bias occurs when individuals use bad health as an excuse for exiting the labor
force (Chirikos and Nestel, 1984; Anderson and Burkhauser, 1985; Bazzoli, 1985; Bound, 1991).
8
Lastly, the possibility exists that the individual’s own health might be affected by his/her
spouse’s retirement status. On the one hand, if the individual has a retired spouse, he/she may
be better taken care of. On the other hand, it is also likely that because household income is lower
with the spouse’s retirement, the individual has to work even harder, possibly inducing worse
health.
In general, we find that retirement has minimal impact on men and women’s health. Some
exceptions occur for men. Our results indicate some adverse effect from retirement for men, who
are more likely to suffer from moderate or severe body pains and to be overweight. In addition,
retirement seems to critically reduce their overall life satisfaction. The effects on women, however,
are somewhat different. We find that retired women are more likely to have hypertension as well
as high glucose levels. Memory loss has also been found in this group. On the other hand, retire-
ment, to some degree, reduces their body aches and alleviates anemia symptoms. For both gen-
ders, we do not see any significant impact on depressive symptoms. There is rarely any sign of a
spousal effect as well. We find slight evidence that individuals living with a retired female spouse
are less likely to be overweight. In addition, with a retired male spouse present, women are more
likely to maintain their mental status and are less likely to develop depressive symptoms.
The rest of the paper is organized as follows. The next section discusses the existing retire-
ment and health studies. Section 2.3 gives a brief description of China’s current retirement system.
Section 2.4 describes data and provides detailed summary statistics. Section 2.5 presents the em-
pirical approach applied in the paper. Section 2.6 discusses the results, and Section 2.7 concludes.
2.2 Retirement and Health Study
In most of the retirement and health studies, an individual’s retirement status is defined accord-
ing to their work status. One is classified as retired if he/she reports not working (Rohwedder
and Willis, 2010; Coe and Zamarro, 2011; Behncke, 2012; Bonsang et al., 2012)
10
. Results from the
literature regarding the relationship between retirement and health vary. Most of the earlier work
show that retirement has some positive, or at least non-negative effects on health (Ekerdt et al.,
1983a; Ekerdt et al., 1983b; Kremer, 1985; Tuomi et al., 1991; Midanik et al., 1995; Reitzes et al.,
10
A few studies define retirement differently. For instance, in Neuman (2008), an individual is classified as
retired if they work less than 1,200 hours per year.
9
1996; Gall et al., 1997), as do some later studies (Charles, 2004; Coe and Zamarro, 2011; Fonseca
et al., 2014).
However, negative effects of retirement have also been found in numerous studies. For ex-
ample, using seven waves of the Health and Retirement Study (HRS), Dave et al. (2007) demon-
strate that complete retirement leads to a 5-16 percent increase in difficulties associated with mo-
bility and daily activities, a 5-6 percent increase in illness conditions, and a 6-9 percent decline in
mental health, over an average post-retirement period of six years. Behncke (2012) adds to the
evidence by using three waves of the English Longitudinal Study of Ageing (ELSA) data to in-
vestigate the effects of retirement on various health outcomes. Results show that retirement sig-
nificantly increases the risk of being diagnosed with a chronic condition. In particular, it raises
the risk of developing a cardiovascular disease and being diagnosed with cancer. Studies using
HRS, ELSA and Survey on Health, Ageing and Retirement in Europe (SHARE) data also suggest
a significant negative effect of retirement on cognitive functioning (Rohwedder and Willis, 2010;
Bonsang et al., 2012; Mazzonna and Peracchi, 2012).
One of the reasons that these results vary is the different health measures they use. It seems
that studies focusing on cognition have more consistent results and most of them support the
“use it or lose it” hypothesis
11
. For instance, Rohwedder and Willis (2010) address the question of
whether retirement leads to cognitive decline by estimating the effect of retirement on individuals’
episodic memory. Conducting analyses for several countries using micro data
12
, they show that
early retirement appears to have a significant negative impact on cognitive ability of the elderly.
Using the same episodic memory measure, Bonsang et al. (2012) also indicate a negative effect of
retirement on cognitive functioning in the U.S. Mazzonna and Peracchi (2012) as well confirm
that retirement causes an increase in cognitive decline by examining its impact on a set of cogni-
tion measures such as episodic memory, orientation, verbal fluency and numeracy. Results using
other health measures such as self-reported ones are more mixed. In most cases, a negative effect
from retirement on self-rated general health status, disability and body pains has been found
(Dave et al., 2007; Lei et al., 2010; Lei et al., 2011; Behncke, 2012; Calvo et al, 2012). Nonetheless, a
few researchers do find the effect retirement on self-assessed health measures to be neutral
11
More details about this hypothesis can be found in Rohwedder and Willis (2010).
12
This study uses the HRS, ELSA and SHARE data.
10
(Bound and Waidmann, 2007; Mojon-Azzi et al., 2007; Coe and Lindeboom, 2008; Coe and Za-
marro, 2011). A similar ambiguity occurs for depressive symptoms and measures of diagnosed
diseases (Dave et al., 2007; Coe and Lindeboom, 2008; Neuman, 2008; Johnston and Lee, 2009; Lee
and Smith, 2009; Behncke, 2012).
The specification of age function might be one reason that leads to different results. In retire-
ment and health studies, it is hard to account for retirement’s effect without worrying about the
confounding impact from natural aging. In addition, since retirement is highly age-related, it is
essential to specify a flexible age function to capture its non-linear effect. Most studies cited use
a simple linear or quadratic age function (Coe and Lindeboom, 2008; Neuman, 2008; Lee and
Smith, 2009; Mazzonna and Peracchi, 2010; Coe and Zamarro, 2011; Calvo et al., 2012). Bonsang
et al. (2012) is one exception. In their study, four different age specifications are examined: linear,
quadratic, cubic and quartic. They show that only the linear form is misspecified according to the
overidentification test, which confirms the non-linearity of age. They suggest that the quadratic
age function is enough for capturing the effect. However, it might still not be flexible enough to
account for the age trend when examining its effect on health along with retirement. Therefore,
we propose a linear spline function with multiple knot points, which may better capture non-
linear age effects.
Another possible reason for different results in the literature are the different methodologies
researchers apply to identify the relationship between retirement and health. The inherent prob-
lem is that retirement is a choice and individuals might decide to retire based on their prior health
status. Early studies do not account for the endogeneity of retirement; most of their conclusions
are based on simple comparisons between health of the retirees and that of the workers (Ekerdt
et al., 1983a; Ekerdt et al., 1983b and so on). Recent studies have adopted different identification
strategies to address the endogeneity problem. The most common practice is to take an instru-
mental variable approach using pension eligibility as an instrument. As is well noted in the liter-
ature, the social security benefit clearly induces individuals to quit their jobs; however, they are
not directly related to an individual’s health. Applications can be found in Charles (2004), Coe
and Zamarro (2011), Rohwedder and Willis (2011), Bonsang et al. (2012), and Mazzonna and
Peracchi (2012). Instead of utilizing age specific retirement incentives as instruments, Coe and
11
Lindeboom (2008) and Calvo et al. (2013) make use of the early retirement window offer
13
. Several
other instruments for retirement have been found in other studies. For instance, as an alternative
identification, Dave et al. (2007)
14
use spouse’s retirement status to instrument for an individual’s
retirement decision. However, spouse’s retirement status might not be a good IV since it still
suffers from the endogenous issue. Other examples include spouse’s pension eligible age, private
pension entitlement, and self-reported usual retirement age on the particular job (Neuman, 2008).
In addition to the instrumental variable approach, Bound and Waidmann (2007) tackle the
endogeneity issue by adopting a strategy similar to a regression discontinuity design and directly
estimating the effect of public pension eligibility. Johnston and Lee (2009) is the first study to our
knowledge to implement a fuzzy regression discontinuity (FRD) design to estimate the health
effect of retirement. Pooling the 1997-2005 Health Survey for England (HSE), they use age 65 as a
cutoff and confirm a discontinuity in the probability of retirement. Similarly, Lei et al. (2011) also
adopt a FRD approach to examine the effect of retirement on health in China. Exploiting the in-
stitutional feature of the mandatory retirement system, they focus only on urban men
15
and use
age 60 as a threshold. Both results from the 2005 one percent population survey and the 2002
Chinese Household Income Project Survey (CHIP) show a discontinuity in the probability of re-
tirement. Though effects of health are mixed. They find that retirement does not seem to affect
the probability of functional limitation; however, its effect on self-rated health status and the hap-
piness feeling is negative. In our study, we also apply the FRD design to identify the causal effect
of retirement on individual’s health in China.
13
“...Early retirement windows are special incentives to retire at a specific time offered by employers to employees.
These windows are often of a relatively short duration and offer enhanced retirement benefits for early retirement (for
example, offering extra years of service for a defined benefit pension plan.) These are often offered to workers in ‘career’
jobs, and are often offered by the employer as an attempt to reduce staff size....” cited from Coe and Lindeboom
(2008). More details about the early retirement window can be found in this study. They argue that the
early retirement windows is exogenous to an individual’s health and brings more exogenous variation into
retirement due to a large age span. In addition, since it is not fully predictable, it is less likely to suffer from
anticipation effect.
14
Their sample is limited to people who report that they expect to retire at the same time as their spouse
and those report to be not concerned about inadequate retirement income.
15
As mentioned in their paper, the mandatory retirement system is not applicable to rural residents. In
addition, retirement ages for women varies and the information from the dataset is limited.
12
2.3 Retirement System in China
The retirement system in China, which was formally established in 1978, is quite different from
the current retirement systems of most industrialized countries. In China, individuals with dif-
ferent types of jobs face different retirement regulations. In general, workers in the formal sector,
such as government, institutions, NGOs and firms, face mandatory retirement. Normally, male
civil servants and male workers should retire at age 60, female civil servants at age 55, and female
workers who are not civil servants at age 50 (State Council, 1978). A few exceptions apply. For
example, people who work in highly dangerous industries can take early retirement
16
. If an indi-
vidual becomes disabled and loses his/her ability to work because of job-related accidents, they
can retire at any time. Additionally, men can retire at age 50 if they completely lose the ability to
work. The same is true for women reaching age 45 (State Council, 1978). All the individuals under
the mandatory retirement system receive their pensions when they meet certain conditions
17
. On
the contrary, for those people with informal sector jobs, such as farmers and urban residents who
work in the individual households, there is no mandatory retirement. Figure 2.1 demonstrates a
more straightforward contrast between the two systems. It depicts the retirement prevalence by
age for four different groups: urban males, urban females, rural males and rural females. The
urban-rural distinction is by registration status (hukou) and does not necessarily represent the
formal-informal sector distinction; however, almost all of the formal sector jobs are in the urban
areas and farmers represent the largest population of the informal sector defined in this paper.
Therefore, Figure 2.1 can still provide some insight into the difference between the two retirement
systems. It can be seen clearly that the overall retirement prevalence for the rural group is much
lower than for that of the urban group, especially for cohorts who have passed the corresponding
mandatory retirement age.
However, processing the official retirement procedure does not necessarily mean a “non-
working” status. First, a worker can be rehired by the employer after processing retirement, even
though it is rare (Lei ei al., 2011). Second, an individual may continue to work in the informal
sector after retiring from their formal jobs. Third, due to the massive restructuring of state-owned
16
They can retire 5 years earlier to the official retirement age.
17
One of the conditions is that they have worked for more than 10 years.
13
enterprises (SOE) in the 1990s, a large amount of workers retire early without processing the of-
ficial procedure
18
. They might stop working since then, though still cannot be counted as officially
retired. Therefore, it is essential to understand the differences between these two retirement con-
cepts. In our analysis, we focus on the effect of actual retirement on health. The details of the
retirement measure will be discussed in the following section.
Figure 2.1 Discontinuity in Retirement Probability
2.4 Data and Measurement
2.4.1 China Health and Retirement Longitudinal Study (CHARLS)
The data used for analysis is from the China Health and Retirement Longitudinal Study
(CHARLS) national baseline, which was fielded between June 2011 and March 2012. CHARLS is
a study modeled after the HRS and focuses on the mid-aged and elderly aged 45 or over in China
19
.
18
They would process the retirement procedure once they have reached their normal retirement age.
19
See Zhao et al. (2014) for details.
14
The sample is nationally representative
20
and is chosen through multi-stage probability sampling.
It covers 28 provinces, 150 counties/districts and 450 villages/urban communities. The response
rate among eligible households was 80.5%
21
, which is comparable with that of other HRS baseline
surveys
22
. The final sample contains 17,705 individuals in 10,257 households. Respondents are
followed every two years, using a face-to-face computer-assisted personal interview (CAPI).
The survey contains detailed information regarding respondents’ retirement and health sta-
tus. A biomarker part is included as well. Non-blood biomarkers were collected during the main
survey by CHARLS enumerators while the blood biomarkers
23
were collected by the staff of the
Chinese Center for Disease Control and Prevention (China CDC). In total, blood samples are col-
lected for 11,847 individuals with a response rate of 67%. No selection on observable SES charac-
teristics has been detected; however, they do find that women are more likely to get their blood
samples taken and younger men are less likely to participate compared to seniors (Zhao et al.,
2014). Overall, the richness of the dataset allows for an in-depth analysis of the impact of retire-
ment on a broad set of health outcomes.
2.4.2 Sample
As previously mentioned, we are interested in a particular group of people who are subject to
mandatory retirement. Due to the institutional feature of this retirement system, the sample used
for analysis is highly selective
24
. Specifically, we include individuals who are currently or previ-
ously employed in the urban formal sector
25
and between age 45 and 75. Here, the urban-rural
20
Tibet is excluded.
21
Of the 19.5% non-response rate, 8.8% is due to refusal to respond, 8.2% is due to the inability of inter-
viewers to contact sample residents, and 2.0% to other reasons.
22
For example, response rate for the HRS baseline in 1992 is 81.6%.
23
Respondents were asked to have fasted overnight and three tubes of venous blood were collected from
them.
24
It is likely that individuals self-select into formal sector jobs. However, so far there is no means to address
this issue. Therefore, our results are confined to this particular group of people. The selected sample might
also be one reason that our results on other covariates such as education (not presented) are different from
the literature. More details of the sample will be discussed in Section 2.4.5.
25
One thing to notice is that temporary workers and casual workers are not subject to the mandatory re-
tirement system. However, due to the limitation of the data, we cannot clearly identify these people.
Though the observations should be very small.
15
distinction is made according to their registration status (hukou). These restrictions leaves us 2,047
observations, among which 1,113 are men and 934 are women.
2.4.3 Measurement of Retirement
In China, retirement can be viewed in two different ways for formal sector workers. One is the
official retirement, where individuals process retirement procedures at mandatory ages and re-
ceive pension wages. The other is the work participation status, where individuals stop working
and permanently stay out of the labor force. As mentioned in Section 2.3, having processed re-
tirement procedure does not necessarily mean not working. In addition, we are interested in how
an individual’s work status, rather than retirement procedure, affects his/her health. Therefore,
following the literature, we use the latter definition for the analysis. The official retirement meas-
ure will be examined as a robustness check.
We use a set of questions from the CHARLS to define an individual’s work status. A re-
spondent is classified as retired if he/she did not engage in agricultural work
26
for more than 10
days in the past year and did not work for at least one hour last week. All the activities such as
earning a wage, running own business and unpaid family business are considered as work; how-
ever, housework or voluntary activities are not included. Among these respondents, those who
has a job but is temporarily laid-off, or on sick leave, or in job training, but expect to return to
their jobs
27
or still receive salaries, are treated as still working. Those who have been looking for
jobs within a month are not considered as retired. Individuals who report that they had at no
point in their lifetime worked for more than 3 months are excluded from our sample
28
.
26
Agricultural work includes farming, forestry, fishing, and husbandry.
27
Questions are asked whether they expect to return to their jobs within 6 months or at a definite time in
the future.
28
Questions asked in the CHARLS are as follows: Have you worked for at least three months during your lifetime
(Works include agricultural work, earning wage work, self-employed activities, and unpaid family business work, et
al. Household chores are not considered as work in the context.). 359 respondents out of 17,705 falls into this
group. In our urban formal sector sample, no such respondent exists.
16
2.4.4 Measurement of Health
2.4.4.1 General Health Status, Disability, Pain and Life Satisfaction
I include four measures of self-reported health. In the dataset, the self-reported general health
status (GHS) is rated on a scale of 5: 1=very good, 2=good, 3=fair, 4= poor, 5=very poor. A binary
indicator is condensed as =1 if report poor or very poor health; =0 if report fair, good or very
good. A disability measure is constructed according to an individual’s ability to perform
ADLs/IADLs-related tasks
29
. Respondents are defined as disabled if they have difficulty doing
or cannot do at least one of those tasks. Pain is defined as =1 if the subject suffers from moderate
or severe body pains, =0 if he/she is not troubled by body pains or only feels a mild pain. Life
satisfaction is also rated on a scale of 5: 1=completely satisfied, 2=very satisfied, 3=somewhat
satisfied, 4=not very satisfied, 5=not at all satisfied. Similarly, a binary indicator is constructed as
=1 if reported completely or very satisfied with their current life; =0 if they reported somewhat,
not very or not at all satisfied.
2.4.4.2 Cognitive Ability and Depressive Symptoms
Two indices are created for the measurement of cognitive functioning. The first is an index of
intact mental status (0-10), which is constructed from the correct answers to ten basic questions
30
.
These ten questions are based on the Telephone Interview of Cognitive Status (TICS)
31
test and
basically are used to assess respondents’ time orientation and numeracy ability. The second indi-
cator of cognition is episodic memory. It is assessed through a test of immediate and delayed
word recall. Some researchers (Souchay et al., 2000) show that episodic memory is particularly
affected by aging, and it is among the first to deteriorate when people grow old. In the CHARLS,
29
In the health status module, respondents were asked the following questions: Because of health and memory
problems, do you have any difficulty with dressing, bathing or showering, eating, getting into or out of bed, using the
toilet, or controlling urination and defecation? Do you have any difficulty with doing household chores, preparing hot
meals, shopping for groceries, managing your money or taking medications?
30
The first set of questions ask respondent the interview date (day, month, year), day of the week, and
current season. The second set of questions ask respondent to do serial subtractions of 7 from 100.
31
See Brandt et al. (1988) for details.
17
the episodic memory task consists of memorizing a list of ten common nouns
32
. The interviewer
reads a list of ten simple nouns (e.g. dog, sky and etc.) to the respondent. The subject is then asked
to repeat as many words as possible from the word list in any order. After answering a few other
questions (around 4-5 minutes), the respondent is asked to repeat the same word list again. Fol-
lowing McArdle (2010), an index of episodic memory (0-10) is constructed as the average of the
immediate recall and delayed recall.
Another dimension which assesses mental health is also included. In the CHARLS, there is a
ten question version
33
of the Center for Epidemiologic Studies Depression Scale (CES-D) which is
used to evaluate a respondent’s depressive symptoms. Subjects are asked to rate how they have
felt and behaved during the last week on a scale of 4: rarely or none of the time (< 1 day), some
or little of the time (1-2 days), occasionally or a moderate amount of the time (3-4 days), or most
or all of the time (5-7 days). CES-D scores (0-30) are constructed following Radloff (1977). For each
negative symptom, if the respondent chooses rarely, then value 0 is assigned; if he/she chooses
most or all of the time, then value 3 is assigned. For positive symptoms, the pattern is reversed. This
results in an index ranging from 0 to 30.
2.4.4.3 Non-blood and Blood Biomarkers
Since self-reported measures might be biased, alternative indicators constructed from the
CHARLS non-blood biomarkers are inspected as well. Hypertension is measured according to
the standard definition of the World Health Organization (WHO); it equals 1 if the mean systolic
blood pressure (SBP) is 140 mm Hg or greater or if the mean diastolic blood pressure (DBP) is 90
mm Hg or greater
34
. In addition, respondents who report being diagnosed with hypertension by
a doctor are also regarded as being hypertensive. Undernourished is defined as a BMI smaller
32
It should be mentioned that, there are in total four different word lists and the CAPI program randomly
chooses one list for each respondent. This design aims to avoid a learning effect, which has been found in
the first two waves of the HRS, where all the respondents were asked to memorize the same word list.
33
CES-D questions in the CHARLS are as follows: I was bothered by things that don’t usually bother me; I had
trouble keeping my mind on what I was doing; I felt depressed; I felt everything I did was an effort; I felt hopeful about
the future; I felt fearful; My sleep was restless; I was happy; I felt lonely; I could not get going.
34
In the CHARLS National Baseline, blood pressure of each respondent is measured three times and the
mean of the measurements is used.
18
than 18.5 while overweight is defined as a BMI greater than 25. Both of the measures follow the
standard definitions of the WHO.
Measures constructed from the blood sample data
35
such as low hemoglobin levels, high
blood glucose levels and high C-reactive protein (CRP) levels are also evaluated. These blood-
biomarkers are used instead of diagnosed chronic conditions because they may capture aspects
of health unknown to the survey participants (Weir, 2008). Here, low hemoglobin level can be
regarded as an indication of malnutrition
36
. A respondent is categorized as having a low hemo-
globin level is his/her index is below 13g/dL for men or 12g/dL for women. The blood glucose
level is indicative of diabetes and pre-diabetes. An individual with a glucose level higher than
normal is more likely to develop coronary heart disease (Crimmins et al., 2008). The threshold
used in this analysis is set at 126mg/dL. C-reactive protein (CRP) is an acute phase response
protein which measures inflammation. Research has suggested that high levels of CRP are related
to the development of cardiovascular disease and cardiac events, including heart attack and
stroke (Danesh and Pepys, 2000; Ridker et al., 2000; Crimmins et al., 2008). Respondents with CRP
above 3mg/L are considered to have elevated CRP levels.
2.4.5 Descriptive Statistics
Earlier in Figure 2.1, we have seen the urban-rural distinction in the retirement system. Figure 2.2
further depicts how the mandatory retirement system affects the formal sector workers. As can
be seen, the formal sector group is divided into 3 parts according to their retirement ages: male,
female civil servants and female workers
37
. In Figure 2.2, it is not hard to notice the discontinuities
at the corresponding pension ages for both men and women. Specifically, the retirement proba-
bility jumps at age 60 for men. There is a steady increase in the retirement prevalence between
age 55-60, which might be due to those who took early retirement. Similar discontinuities can be
35
See Zhao et al. (2014).
36
Previous studies such as Thomas et al. (2011) show that the one’s work capacity becomes lower if hemo-
globin levels are below thresholds.
37
The distinction between female civil servants and worker is not clear. With the information in the
CHARLS, female civil servants can be identified if she has already processed the retirement procedure;
however, for women who are currently working, no question regarding cadre status is asked. We use ex-
pected retirement ages to differentiate this group of non-identified women. Since there is mandatory re-
tirement for the formal sector workers, individuals whose expected retirement age are 55 or above are
categorized as female civil servants in our sample.
19
found in women as well. It appears at age 55 for female civil servants and age 50 for female work-
ers.
Figure 2.2 Discontinuity in Retirement Probability (Formal Sector)
Table 2.1 provides descriptive statistics of demographic, socioeconomic and household char-
acteristics and health outcome variables for the sample, disaggregating men and women. In the
sample, 64% of the respondents have stopped working. The percentage of retirement is higher
for women. The mean age is around 59 and is similar across groups. The education level of our
sample is distinctive from the general Chinese population. On average, they have much higher
educational attainments. Among them, only 3% are illiterate, 47% have received at least some
high school education and around 15% have attended colleges
38
. Additionally, there is not much
gender differences in education for this group of respondents. It is unlike the gender differences
found in the entire sample (Lei et al., 2014b), presumably because of selection into formal sector
jobs. 92% of the respondents are currently married or cohabiting with partners. The average num-
ber of children in a family is less than 2 and the average number of household members is above
38
In the rural sample from the CHARLS baseline, 13% of rural men and about half of the rural women are
illiterate. Less than 1% of the rural residents have received some college education.
20
Table 2.1 Summary Statistics
Panel A Total Men Women
Mean Mean Mean
Demographic Character-
istics
Retirement Status 0.64 0.55 0.75
Age 58.59 59.04 58.06
Illiterate 0.03 0.02 0.06
Primary School 0.21 0.20 0.21
Middle School 0.29 0.29 0.29
High School 0.32 0.33 0.32
College and Above 0.15 0.17 0.12
Married 0.92 0.96 0.87
Household Characteris-
tics
Number of Kids 1.90 1.95 1.84
Number of HH Members 3.03 3.11 2.93
log PCE 9.15 9.11 9.21
Panel B Total Men Women
Mean Mean Mean
Self-reported Health
General Health Status: Poor 0.17 0.17 0.18
Disability 0.13 0.13 0.13
Moderate or Severe Pain 0.13 0.09 0.17
Life Satisfaction: Good 0.22 0.22 0.23
Cognition & Mental Health
Intact Mental Status (0-10) 8.06 8.00 8.14
Episodic Memory (0-10) 4.49 4.34 4.67
CESD (0-30) 5.45 4.93 6.06
Biomarkers
Hypertension 0.41 0.42 0.40
Undernourished (BMI<18.5) 0.03 0.03 0.03
Overweight (BMI>25) 0.42 0.41 0.43
Low Hemoglobin (<13g or 12g/dL) 0.09 0.07 0.12
High Glucose (Glucose>=126 mg/dL) 0.17 0.18 0.15
High C-Reactive Protein (CRP>3 mg/L) 0.18 0.18 0.19
Note: Sample is limited to urban residents currently or previously employed in the formal sector jobs. The age
range here is 45-75.
21
3. The mean log per capita household expenditure (logPCE) is 9.15, which equals to around 9,400
yuan. Panel B presents descriptive statistics for health outcomes. Almost all of the health indica-
tors are similar for men and women. Overall 17% of the respondents rate their health as poor or
very poor and 13% of them have an ADL/IADL-related disability. However, it seems that a larger
fraction of women suffer from moderate or severe body pains. Around 22% of the respondents
were completely or very satisfied with their current life. For biomarkers, 41% of the respondents
in our sample are hypertensive, according to both their reports and measured blood pressures.
9% of the individuals have low hemoglobin level and the women group in particular has a much
higher incidence rate. 17% of the sample have high glucose level and 18% have high CRP. The
percentage of undernourished is as low as 3%; however, more than 40% of the respondents are
overweight. For the cognition and mental health indicators, on average respondents answer 8
TICS questions correctly and recall less than half of the ten words. Women score slightly higher
in both dimensions. Yet, they seem to have more depressive symptoms than men according to
the CES-D indicator.
2.5 Empirical Approach
The aim of the study is to investigate the impact of retirement on self-reported health, cognition,
depressive symptoms and biomarkers. In the model, I assume that health depends on retirement
status, age, along with an error term. The equation estimated is:
𝑌 𝑖𝑗
= 𝛽 0
+ 𝛽 1
𝑅 𝑖𝑗
+ 𝑓 (𝑎𝑔𝑒
𝑖𝑗
) + 𝑢 𝑖𝑗
(2.1)
𝑌 𝑖𝑗
is a measure of health. 𝑅 𝑖𝑗
is an indicator of the respondent’s retirement status. 𝑓 (𝑎𝑔𝑒
𝑖𝑗
) is a
flexible and continuous age function. In the results, age spline function with 5 knots at age 50, 55,
60, 65 and 70 is employed to better capture the non-linear effects from aging. We also use a cubic
polynomial function of age to test the robustness of the age specification.
The parameter of interest is 𝛽 1
, which measures the effect of retirement on health. If 𝑅 𝑖𝑗
is not
correlated with the error term 𝑢 𝑖𝑗
, then OLS will lead to a consistent estimator of 𝛽 1
. However,
the assumption may not hold for the following reasons. First, reverse causality is likely to exist
between retirement and health. Health status has long been identified as a significant determinant
22
in the elderly’s retirement decisions. Various evidence has been found that decline in health status
could induce retirement (Belgrave et al., 1987; Ettner et al., 1997; Chatterji et al., 2011)
39
. To solve
this problem, the institutional feature of the mandatory retirement system is exploited by using
a Fuzzy Regression Discontinuity (FRD) design. In the formal sector, workers are required to
retire when they reach the mandatory retirement age. However, there are exceptions. In addition,
some retirees may continue to work in the informal sector after leaving their formal jobs. The FRD
design is in accordance with this institutional feature. It allows for a jump in the probability of
retirement at the threshold 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅, but the jump is not necessarily from full-time working to full-
time retirement:
lim
𝑎𝑔𝑒̃↓ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ ̅
Pr(𝑅 𝑖 = 1 | 𝑎𝑔𝑒 = 𝑎𝑔𝑒 ̃ ) ≠ lim
𝑎𝑔𝑒̃↑ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ ̅
Pr(𝑅 𝑖 = 1 | 𝑎𝑔𝑒 = 𝑎𝑔𝑒 ̃ )
(2.2)
where 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ indicates the corresponding mandatory retirement ages. Specifically, for men it is 60,
for female civil servants it is 55 and for other female formal workers it is 50. The average causal
effect of retirement on health can then be estimated as follows:
𝛽 1
𝐹𝑅𝐷
=
lim
𝑎 𝑔𝑒 ̃↓ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ ̅
𝐸 (𝑌 | 𝑎𝑔𝑒 =𝑎𝑔𝑒̃)≠ lim
𝑎𝑔𝑒̃↑ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ ̅
E(𝑌 | 𝑎𝑔𝑒 =𝑎𝑔𝑒̃)
lim
𝑎𝑔𝑒̃↓ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ ̅
𝐸 (𝑅 | 𝑎𝑔𝑒 =𝑎𝑔𝑒̃)≠ lim
𝑎𝑔𝑒̃↑ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅ ̅
E(𝑅 | 𝑎𝑔𝑒 =𝑎𝑔𝑒̃
(2.3)
We use two-stage least squares method
40
to estimate the local average treatment effect
(LATE), instrumenting retirement 𝑅 𝑖𝑗
with an indicator of pension eligibility 𝐼 (𝑎𝑔𝑒
𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅).The
empirical specifications are as follows:
𝑌 𝑖𝑗
= 𝛽 0
+ 𝛽 1
𝑅 𝑖𝑗
+ 𝑓 (𝑎𝑔𝑒
𝑖𝑗
) + 𝑋 𝑖𝑗
𝛾 + 𝑒 𝑗 + 𝑢 𝑖𝑗
(2.4)
𝑅 𝑖𝑗
= 𝛼 0
+ 𝑔 (𝑎𝑔𝑒
𝑖𝑗
) + 𝛼 1
𝐼 (𝑎𝑔𝑒
𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝑋 𝑖𝑗
𝜙 + 𝑒 𝑗 + 𝑣 𝑖𝑗
(2.5)
39
For example, Belgrave et al. (1987) show that workers in poor health and those suffering from chronic
health conditions retire earlier than those who are healthy. Ettner et al. (1997) and Chatterji et al. (2011) also
indicate that psychiatric disorders significantly reduce employment among both genders. In addition, there
might be unobserved variables that affect both retirement and health. Individuals with higher ability may
choose to invest more in human capital such as education and health, thus retire at an older age compared
to those who made less investment.
40
More details of the application can be found in Imbens and Lemieux (2008).
23
For both equations, we control for individuals’ demographic, socioeconomic and household
characteristics 𝑋 𝑖𝑗
. For the demographic variables, the highest completed education level is con-
sidered because studies have shown that schooling helps people choose healthier life-styles, thus
it contributes to better health (Kendel, 1991). Marital status is also included since marriage might
be associated with subjective well-being, as well as physical health (Kiecolt-Glaser and Newton,
2001). Expenditure is used instead of household income due to the fact that consumption expend-
itures are the better measure of the economic resources available to the family in developing
countries (Deaton, 1990; Strauss and Thomas, 1995). For household characteristics, the number of
living children (who are mostly adult) and household members are controlled for. On the one
hand, these individuals may increase the burden of the household head and spouse. On the other
hand, they could also be a source of elderly support. In addition to these covariates, community
dummies 𝑒 𝑗 are incorporated to control for factors such as health infrastructure, health prices and
local wages, which may affect respondents’ health.
It is normal to assume that model errors are correlated within certain geographical clusters
when using individual-level cross-section data. However, in this case, model errors might also be
correlated within a birth cohort. Failure to control for these within-cluster error correlations may
lead to inaccurate standard errors. Therefore, we follow Cameron et al. (2011) and apply two-way
clustering to control for both community and birth cohort.
2.6 Results
2.6.1 Discontinuity in Retirement Probability
Table 2.2 presents the first-stage results for two specifications, both of which control for a linear
spline age function, demographic and socioeconomic characteristics, as well as community fixed
effects. The latter specification also includes some household characteristics such as number of
children, number of household members and PCE to examine the potential impacts on health
from household and family. Since we limit our sample to the urban formal sector workers, the
observations for the female civil servants are quite limited. Therefore, we pool together the two
female groups, assuming that the treatment effect at the cutoffs for both groups are the same.
Pension eligible age takes the value 55 for female civil servants and 50 for female workers.
24
Aligned with Figure 2.2, results in Table 2.2 confirm the discontinuities at pension eligible
ages for both men and women. Men who pass age 60 are 23% more likely to retire. Women are
more responsive to the mandatory retirement rule. Those who pass the corresponding ages are
49% more likely to stop working. The F-statistics of the pension eligibility indicator for both gen-
ders are over 10, which further strengthens the validity of the mandatory retirement age as a
predictor for retirement decisions. Both evidence from Figure 2.2 and Table 2.2 support the ap-
plication of the FRD design in the following estimation.
Table 2.2 Discontinuity in Retirement Probability
Retirement Status
Men Women
(1) (2) (3) (4)
Pension Eligibility 0.239*** 0.226*** 0.492*** 0.493***
[0.0633] [0.0676] [0.1034] [0.1041]
HH Characteristics N Y N Y
N 1,110 1,099 792 783
R-squared 0.556 0.562 0.623 0.617
Mean 0.549 0.549 0.726 0.729
F_stat (IV) 14.270 11.170 22.650 22.460
P_value (IV) 0.001 0.002 0.000 0.000
Note: Pension eligible age for men is age 60. Pension eligible age for female civil servant is age 55 and for female
worker is age 50. A linear spline for age is included in all the specifications with five knots (50, 55, 60, 65 and 70).
Other controls contain marital status, educational attainments and community fixed effects. For both genders, two
specifications are estimated with one controlling for household characteristics and the other not. Last three rows
are mean of the dependent variable, F statistic and P value of the pension eligible age cutoff.
Standard errors are in the parenthesis. Standard errors are clustered at both community level and birth cohort
level. *** p<0.01, ** p<0.05, * p<0.1.
2.6.2 Fuzzy Regression Discontinuity Results
Table 2.3 presents the FRD estimation results for men. For each of the health indicators, two spec-
ifications are estimated with the latter one additionally controlling for household characteristics.
Panel A include the results for self-reported health measures. It seems that, after accounting for
the endogeneity issue, retirement has no contemporaneous impact on general health status
25
Table 2.3 Impact of Retirement on Self-Reported Health, Mental Health, and Biomarkers: Men
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A: Self-Reported Health
GHS: poor Disability
Moderate or
Severe Pain
Life Satisfaction:
Good
Retired -0.076 0.022 -0.004 0.083 0.399** 0.429** -0.699* -0.822**
[0.2196] [0.2355] [0.1577] [0.1494] [0.1570] [0.1761] [0.3603] [0.3939]
N 1,110 1,099 1,099 1,088 1,110 1,099 989 980
Mean 0.169 0.169 0.126 0.126 0.088 0.088 0.216 0.215
HH Charac-
teristics N Y N Y N Y N Y
Panel B: Cognition & Mental Health
Intact Mental
Status
Word Recall
Depressive
Symptoms
Retired -0.417 -0.208 -0.257 0.762 3.315 3.356
[1.1452] [1.2152] [0.9943] [1.2141] [3.1044] [3.5895]
N 1,073 1,062 991 982 1,036 1,026
Mean 7.999 7.994 4.340 4.344 4.914 4.908
HH Charac-
teristics N Y N Y N Y
Panel C: Non-Blood Biomarkers
Hypertension Under-nourished Overweight
Retired 0.134 0.181 0.165 0.121 0.803* 1.070*
[0.3053] [0.3471] [0.2002] [0.2143] [0.4807] [0.6077]
N 1,110 1,099 726 720 726 720
Mean 0.424 0.422 0.028 0.028 0.409 0.410
HH Charac-
teristics N Y N Y N Y
26
Panel D: Blood Biomarkers
Low
Hemoglobin
High Glucose High CRP
Retired -0.011 -0.039 0.181 0.214 0.149 0.097
[0.3634] [0.3580] [0.5689] [0.5644] [0.5625] [0.5260]
N 567 564 599 595 599 595
Mean 0.072 0.073 0.177 0.178 0.182 0.183
HH Charac-
teristics N Y N Y N Y
Note: 2SLS estimation results for men. A linear spline for age is included in all the specifications with five knots
(50, 55, 60, 65 and 70). Other controls contain marital status, educational attainments and community fixed effects.
For each health indicator, two specifications are estimated with one controlling for household characteristics and
the other not.
Standard errors are in the parenthesis. Standard errors are clustered at both community level and birth cohort
level. *** p<0.01, ** p<0.05, * p<0.1.
27
Table 2.4 Impact of Retirement on Self-Reported Health, Mental Health, and Biomarkers: Women
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A: Self-Reported Health
GHS: poor Disability
Moderate or
Severe Pain
Life Satisfaction:
Good
Retired 0.100 0.093 0.002 -0.001 -0.257* -0.269** 0.000 0.019
[0.0991] [0.0988] [0.1233] [0.1191] [0.1333] [0.1344] [0.1524] [0.1521]
N 792 783 783 775 790 781 741 734
Mean 0.172 0.174 0.125 0.126 0.165 0.166 0.228 0.228
HH Char-
acteristics N Y N Y N Y N Y
Panel B: Cognition & Mental Health
Intact Mental
Status
Word Recall
Depressive
Symptoms
Retired -0.013 -0.289 -1.117* -1.120* 0.875 0.817
[1.0153] [1.0223] [0.6678] [0.6709] [0.9519] [0.9704]
N 750 740 743 736 779 772
Mean 8.307 8.319 4.721 4.713 5.893 5.916
HH Char-
acteristics N Y N Y N Y
Panel C: Non-Blood Biomarkers
Hypertension Under-nourished Overweight
Retired 0.162* 0.153 0.085 0.083 0.103 0.099
[0.0976] [0.0960] [0.0610] [0.0597] [0.1882] [0.2013]
N 793 784 506 503 506 503
Mean 0.400 0.401 0.028 0.028 0.431 0.433
HH Char-
acteristics N Y N Y N Y
28
Panel D: Blood Biomarkers
Low Hemoglobin High Glucose High CRP
Retired -0.211*** -0.213*** 0.168* 0.164 0.153 0.151
[0.0699] [0.0702] [0.0987] [0.1000] [0.1250] [0.1177]
N 417 413 436 432 436 432
Mean 0.110 0.109 0.142 0.144 0.181 0.183
HH Char-
acteristics N Y N Y N Y
Note: 2SLS estimation results for women. A linear spline for age is included in all the specifications with five knots
(50, 55, 60, 65 and 70). Other controls contain marital status, educational attainments and community fixed effects.
For each health indicator, two specifications are estimated with one controlling for household characteristics and
the other not.
Standard errors are in the parenthesis. Standard errors are clustered at both community level and birth cohort
level. *** p<0.01, ** p<0.05, * p<0.1.
29
(GHS). The pattern is the same with respect to disability. One thing to notice is that retirement
does have a negative effect on body pains. According to the results presented in column (6), a
retired elderly is 43% more likely to report moderate or severe pains. Retirement also has a pro-
found adverse effect on life satisfaction. It seems that elderly respondents are 70-80% less likely
to feel completely or very satisfied about their life as a whole if they stop working.
Panel B presents results for cognitive functioning and mental health. The FRD results in col-
umn (1)-(4) show that no contemporaneous impact has been detected from retirement on cogni-
tive measures. This is consistent with the findings in Coe and Zamarro (2011) and Coe et al. (2012),
but inconsistent with a large literature in the U.S. and Europe. The phenomena here might be
explained by elderly’s involvement in social activities. Literature has shown that social activity
may help maintain the elderly’s cognitive ability by reducing the rate of cognitive decline as well
as providing resistance to mental diseases (Hu et al., 2012; Lei et al., 2014a; Wang et al., 2002).
And in fact, in our study, more than 98% of the elderly are to some extent involved in social
activities
41
. Regarding mid-aged and elderly’s mental healthiness, various studies have attempted
to establish a causal link between retirement and depressive symptoms. However, no consensus
has been reached. Our results indicate a neutral effect that resembles a set of literature (Dave et
al. 2007; Coe and Lindeboom, 2008; Neuman, 2008; Lee and Smith, 2009), showing that retirement
does not affect men’s mental healthiness, at least in the aspect of depression.
Table 2.3 Panel C & D presents results for biomarkers. The coefficients before the retirement
status are almost all positive, but not significant of standard levels, except for being overweight.
Similar to men, we find retirement has minimal effect on women’s health. In contrast to men
where we detect several significant adverse impacts, the results on women (Table 2.4) are mixed.
Table 2.4 Panel C shows some evidence that retired women are more likely to have hypertension
and high glucose level. But the effects are small and disappear after controlling for household
characteristics. There are also strong evidence indicating that retirement may help alleviate body
pains (Table 2.4 Panel A). Women are 27% more likely to report moderate or severe pains after
retirement, which might be due to the reduced workload and possibly increased exercises. The
positive effects from retirement also include the decline in anemia prevalence, which may as well
41
Social activities in our study are defined as participating in: (1) Interacting with friends; (2) Playing Ma-
jong, chess, cards, or going to community club; (3) Sport, social, or other kind of club; (4) Took part in a community-
related organization; (5) Voluntary or charity work; (6) Caring for a sick or disabled adult who does not live with you
and who did not pay you for the help; (7) Attending an educational or training course.
30
resulted from the reduced stress and better diet. One additional effect we find about women is
on cognition (Table 2.4 Panel B). In general, retirement rarely has any influence on women’s intact
mental status and depressive symptoms. However, there might be some memory impairment
since retired women seem to recall fewer words than working females.
2.6.3 Spousal Effects
In addition to the impact on health brought by the individual’s own retirement, spouse’s working
status may also affect the respondent’s health. On the one hand, there might be some family in-
come loss if the spouse stops working. This income loss, to some extent, may affect respondent’s
health. On the other hand, with a retired spouse, the respondent might be better taken care of.
Therefore, we estimate another model taking into account the spouse’s retirement status and
other spousal characteristics. The equation estimated are:
𝑌 𝑖𝑗
= 𝜋 0
+ 𝜋 1
𝑅 𝑖𝑗
+ 𝜋 2
𝑅 −𝑖𝑗
+ 𝜋 3
𝑅 𝑖𝑗
× 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝜋 4
𝑅 −𝑖𝑗
× 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝜋 5
𝐹𝑒𝑚𝑎𝑙𝑒
+𝜋 6
𝐹𝑜𝑟𝑚𝑎𝑙 + 𝑓 (𝑎𝑔𝑒
𝑖𝑗
) + 𝑓 𝑠 (𝑎𝑔𝑒
−𝑖𝑗
) + 𝐸 𝑖𝑗
𝛾 + 𝐸 −𝑖𝑗
𝜙 + 𝐹 𝑖𝑗
𝜉 + 𝑒 𝑗 + 𝑢 𝑖𝑗
(2.6)
where 𝑌 𝑖𝑗
represent the respondent’s health indicators, 𝑅 𝑖𝑗
is the respondent’s own retirement sta-
tus and 𝑅 −𝑖𝑗
is the spouse’s retirement status. We focus on the main respondents that could either
be man or woman. Therefore a gender indicator Female and its interactions with both retirement
statuses are included in the equation as well. One thing to be noted is that, the main respondents
we are targeting on are currently or previously employed in the formal sector; however, no re-
striction has been made on their spouses. Thus, one dummy variable Formal that indicates
whether the spouse is in the same group as the main respondent is specified as well. 𝑓 (𝑎𝑔𝑒
𝑖𝑗
) and
𝑓 𝑠 (𝑎𝑔𝑒
−𝑖𝑗
) are age linear splines
42
for the main respondents and their spouses accordingly. 𝐸 𝑖𝑗
and 𝐸 −𝑖𝑗
are their educational levels. 𝐹 𝑖𝑗
are household characteristics. 𝑒 𝑗 represents the commu-
nity fixed effects. Same as equation (2.5) estimated in the previous part, the two retirement indi-
cators and its interactions with gender in equation (2.6) also suffer from the endogeneity problem.
Therefore, we estimate this equation using 2SLS and instrument both the retirement statuses with
42
Both the linear splines include five knots at 50, 55, 60, 65 and 70.
31
Table 2.5 First-stage Check for Spouse Retirement
Own
Retirement
Spouse
Retirement
Own
Retirement
X Female
Spouse
Retirement
X Female
Spouse Pension Eligibility 0.119** 0.298*** -0.058*** -0.131***
[0.0513] [0.0937] [0.0206] [0.0386]
Own Pension Eligibility 0.294*** -0.075 -0.063 -0.151**
[0.1067] [0.0721] [0.0432] [0.0696]
Spouse Pension Eligibility X Female -0.113 -0.015 0.182*** 0.444***
[0.1042] [0.1078] [0.0628] [0.0705]
Own Pension Eligibility X Female 0.137 0.033 0.627*** 0.251**
[0.1188] [0.1361] [0.0945] [0.1108]
Female 0.028 -0.106 0.144* 0.110*
[0.0871] [0.0774] [0.0733] [0.0589]
N 736 736 736 736
R-squared 0.610 0.517 0.841 0.702
P_value (IVs) 0.006 0.003 0.000 0.000
Note: This sample contains the formal sector main respondents only. For a male respondent, no matter he is a
main respondent or the spouse of a main respondent, his pension eligible age is 60. For a female main respondent,
the pension eligible age is 55 or 50 depending on whether she is a civil servant or a worker. For a female spouse, if
she belongs to the formal sector, then her pension eligible age is specified the same as a female main respondent.
However, if she does not belongs to the formal sector, then her age cutoff is set to be 60. Age linear splines for the
main respondents and the spouses are included in all the specifications with five knots (50, 55, 60, 65 and 70).
Other controls contain both main respondents’ and spouses’ educational attainments, household characteristics as
well as community fixed effects. Joint significance test of the instruments are shown at the bottom.
Standard errors are in the parenthesis. Standard errors are clustered both community level and birth cohort level.
*** p<0.01, ** p<0.05, * p<0.1.
32
Table 2.6 Impact of Retirement and Spouse Retirement on Self-Reported Health, Mental Health, and Biomarkers
(1) (2) (3) (4) (5) (6) (7)
Self-Reported Health Cognition & Mental Health
GHS:
poor
Disability
Moderate or
Severe Pain
Life Satisfac-
tion: Good
Intact Mental
Status
Word
Recall
Depressive
Symptoms
Own Retire-
ment -0.201 -0.396 -0.436 -1.035 8.690* 1.649 -10.504*
[0.3729] [0.4259] [0.3943] [0.7874] [4.5219] [1.3400] [5.7987]
Own Retire-
ment X Female 0.252 0.412 0.482 1.117 -11.351*** -2.182 13.717**
[0.4194] [0.4674] [0.3664] [0.8325] [3.8828] [1.7787] [6.0207]
Spouse Retire-
ment 0.120 0.370 -0.057 0.242 -3.619 0.953 5.683
[0.1766] [0.2437] [0.1954] [0.3420] [3.8051] [1.4564] [4.9422]
Spouse Retire-
ment X Female -0.039 -0.213 -0.534 -1.102 12.867** -0.383 -16.400**
[0.4403] [0.5343] [0.4059] [0.9082] [5.2219] [2.5718] [7.2612]
N 736 728 728 669 283 283 283
Mean 0.163 0.120 0.125 0.241 8.519 4.624 5.445
33
Non-Blood Biomarkers Blood Biomarkers
Hyperten-
sion
Under-
nourished
Overweight
Low
Hemoglobin
High Glucose
High
CRP
Own Retire-
ment 0.395 -0.735 1.415 0.101 0.100 -0.023
[0.4385] [0.5530] [1.2496] [0.4354] [0.3079] [0.3333]
Own Retire-
ment X Female -0.188 0.684 -1.126 -0.484 -0.026 -0.157
[0.5131] [0.6589] [1.5251] [0.5442] [0.5142] [0.4446]
Spouse Retire-
ment -0.175 0.751 -1.970* 0.282 -0.085 -0.065
[0.2605] [0.6105] [1.1647] [0.3460] [0.3574] [0.4241]
Spouse Retire-
ment X Female 0.540 -0.608 1.355 0.535 0.469 0.103
[0.6034] [0.6913] [1.7785] [0.5618] [0.6401] [0.4900]
N 669 282 282 365 364 364
Mean 0.398 0.035 0.440 0.090 0.115 0.176
Note: This sample contains the formal sector main respondents only. Age linear splines for the main respondents and the spouses are included in all the specifi-
cations with five knots (50, 55, 60, 65 and 70). Other controls contain both main respondents’ and spouses’ educational attainments, household characteristics as
well as community fixed effects.
Standard errors are in the parenthesis. Standard errors are clustered both community level and birth cohort level. *** p<0.01, ** p<0.05, * p<0.1.
34
pension eligibilities
43
. Specifically, for the main respondents, pension eligible age is set at 60 for
men, 55 for female civil servants and 50 for female workers. For a male spouse, his pension eligible
age is 60, no matter whether or not he is in the formal sector. For a female spouse, if she is in the
formal sector, her pension eligible age is set the same as before. However, if she is not currently
or previously employed in the formal sector, then her pension eligible age is set to be 60
44
. The
interaction terms are also instrumented, with the pension eligibility interact with a gender indi-
cator
45
.
First-stage results are presented in Table 2.5. The joint test of the four instruments are signif-
icant, which confirms the validity of the 2SLS estimation. Table 2.6 presents the estimation results
for equation (2.6). Unfortunately, we rarely see any impact of spouses’ retirement on respondents’
self-reported health and biomarkers. However, there is slight evidence that individuals living
with retired female spouses are less likely to be overweight. In addition, with the presence of a
retired male spouse, women are more likely to maintain their mental status and are less likely to
develop depressive symptoms.
2.6.4 Robustness Checks
We test several specifications to see whether our results are robust. First, the age function used in
the main analyses is a linear spline with five knots. Since retirement is highly age-related and
43
The first stage equations are as follows: 𝑅 𝑖𝑗
= 𝛼 0
+ 𝛼 1
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝛼 2
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝛼 3
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥
𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛼 4
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛼 5
𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛼 6
𝐹𝑜𝑟𝑚𝑎𝑙 + 𝑔 (𝑎𝑔𝑒 𝑖𝑗
) + 𝑔 𝑠 (𝑎𝑔𝑒 −𝑖𝑗
) +
𝐸 𝑖𝑗
𝛾 1
+ 𝐸 −𝑖𝑗
𝜙 1
+ 𝐹 𝑖𝑗
𝜉 1
+ 𝑒 𝑗 + 𝑣 𝑖𝑗
, 𝑅 −𝑖𝑗
= 𝛽 0
+ 𝛽 1
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝛽 2
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝛽 3
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) ×
𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛽 4
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛽 5
𝐹𝑒 𝑚𝑎𝑙𝑒 + 𝛽 6
𝐹𝑜𝑟𝑚𝑎𝑙 + 𝑔 (𝑎𝑔𝑒 𝑖𝑗
) + 𝑔 𝑠 (𝑎𝑔𝑒 −𝑖𝑗
) + 𝐸 𝑖𝑗
𝛾 2
+
𝐸 −𝑖𝑗
𝜙 2
+ 𝐹 𝑖𝑗
𝜉 2
+ 𝑒 𝑗 + 𝜔 𝑖𝑗
.
44
One reason of using 60 as a cutoff for non-formal sector workers is that, individuals who are urban resi-
dents but not working in the formal sector can enroll in the Urban Resident Pension Insurance System,
where they can receive their pensions at age 60, for both genders.
45
𝑅 𝑖𝑗
× 𝐹𝑒𝑚𝑎𝑙𝑒 = 𝛿 0
+ 𝛿𝛼
1
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝛿 2
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝛿 3
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 +
𝛿 4
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛿 5
𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛿 6
𝐹𝑜𝑟𝑚𝑎𝑙 + 𝑚 (𝑎𝑔𝑒 𝑖𝑗
) + 𝑚 𝑠 (𝑎𝑔𝑒 −𝑖𝑗
) + 𝐸 𝑖𝑗
𝛾 3
+ 𝐸 −𝑖𝑗
𝜙 3
+ 𝐹 𝑖𝑗
𝜉 3
+
𝑒 𝑗 + 𝜑 𝑖𝑗
, 𝑅 −𝑖𝑗
× 𝐹 𝑒𝑚𝑎𝑙𝑒 = 𝜃 0
+ 𝜃 1
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝜃 2
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) + 𝜃 3
𝐼 (𝑎𝑔𝑒 𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 +
𝜃 4
𝐼 (𝑎𝑔𝑒 −𝑖𝑗
≥ 𝑎𝑔𝑒̅ ̅ ̅ ̅ ̅) × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝜃 5
𝐹𝑒𝑚𝑎𝑙𝑒 + 𝜃 6
𝐹𝑜𝑟𝑚𝑎𝑙 + 𝑛 (𝑎𝑔𝑒 𝑖𝑗
) + 𝑔𝑛
𝑠 (𝑎𝑔𝑒 −𝑖 𝑗 ) + 𝐸 𝑖𝑗
𝛾 4
+ 𝐸 −𝑖𝑗
𝜙 4
+ 𝐹 𝑖𝑗
𝜉 4
+
𝑒 𝑗 + 𝜓 𝑖𝑗
.
35
might be sensitive to age specification, we try another function using cubic ages. Second, in all
the previous specifications, we control for the community fixed effects to account for the potential
heterogeneous regional differences. In this part, we assume that there is a correlation within the
same county. A model with county fixed effect will be estimated and all the standard errors are
clustered at both county level and birth cohort level. Third, for all the “retirement” analysis, we
use the work participation definition. Here, we try another indicator: official retirement. An indi-
vidual is considered as officially retired if he/she has already processed the formal retirement
procedure. Lastly, in the analyses above, the registration system (hukou) is employed to differen-
tiate urban and rural residency. In this section, we define the urban/rural status according to the
classification methods provided by the National Bureau of Statistics
46
in China. One additional
estimation is made for men. Previously, we limit our sample to individuals aged from 45 to 75. In
this specification, we further restrict the male sample to those aged between 50 and 70 to see
whether the results still hold. All the estimations are in the appendix (Table A2.1-A2.3). Results
from robustness checks are consistent with the impacts shown in the previous tables.
2.7 Conclusion and Discussion
Researchers have been studying the impact of retirement on health for decades; however, there
is no consensus whether this effect is positive or negative. Using the China Health and Retirement
Longitudinal Study (CHARLS) national baseline, we investigate how retirement affects elderly’
health in the context of China, where this relationship has rarely been examined. Exploiting the
properties of the mandatory retirement system in the formal sector, the endogeneity issue of the
retirement status has been well accounted for. The findings of this study suggest that retirement
has minimal impact on elderly’ health as well as on cognition and depressive symptoms. How-
ever, we do find that retirement significantly increases the experiencing of moderate or severe
body pains for middle-aged and elderly men. It also severely reduces their life satisfaction. In
addition, retired men are to some extent more likely to be overweight. The impacts on women
are not strong either. For most of the health indicators examined in the analysis above, no effect
46
http://www.stats.gov.cn/english/
36
from retirement has been found. There is some evidence showing that retirement may have ad-
verse impact on women’s memory. Nonetheless, retirement also seems to help alleviate body
pains as well as low hemoglobin for women.
This study, inevitably, has its own limitations. First, due to the property of the embedded
retirement system in China, this paper mainly focuses on the formal sector workers. Results of
this highly selected sample could hardly be generalized to the other groups of people. Second, it
is possible that retirement affects the changes in health outcomes instead of the levels. In addition,
the impact of retirement on health may not be necessarily instantaneous, it is likely that the effect
occur with a lag. However, at the time of the study, CHARLS was a cross-sectional dataset. With
more waves of the CHARLS data are now available, it will be interesting to examine this impact
in future studies. One additional thing that should to be mentioned is that, retirement dates can
be fully expected for the formal sector workers. This anticipation may induce them to adjust their
behaviors before retirement, which may attenuate the potential effects and bias the estimates to-
wards zero.
Though with limitations, the findings of this paper has important policy implications. The
conclusions here lend support to raising the retirement ages by arguing that postponing retire-
ment is not bad for health. On the contrary, it might be good to the elderly. Considering the other
benefits that might be brought by raising the retirement ages (for instance, alleviating the pres-
sure on the pension system, reducing the burden on the younger generation, help maintaining a
viable labor force, and etc.), postponing retirement could be henceforth an inevitable trend.
37
Chapter 3
Health and Labor Supply in Rural China
3.1 Introduction
Aging in developing countries, especially rural areas, is much more difficult compared to devel-
oped countries. In rural regions where the formal pension is almost absent, old age support
mainly comes from two sources: elderly’s labor income and support from their families. Unfor-
tunately, the latter seems to erode over time. Historically, adult children, especially adult sons,
share the responsibility to take care of their elderly parents. However, with the one-child policy
established since 1970s, individuals have fewer sons to rely on for old-age support. In addition,
as the society becomes more commercial and individual-oriented, it is hard to say that the tradi-
tional Confucian filial piety
47
will continue to be strong in the rural areas (Benjamin et al., 2000).
This leaves the former source - labor income - extremely important when the elderly think about
“retirement” and old-age support.
In one of her books, David-Friedmann once described the situation of the Chinese farmers in
the pre-reform era as “ceaseless toil” (Benjamin et al., 2003). And it still seems to be a normal life
pattern for most of the rural elderly nowadays: they keep working until they are not physically
capable of. Using a nationally representative survey
48
in China, we try to depict this situation.
Figure 3.1 shows the labor force participation rate for rural residents aged 50 and above. For the
age 50 cohorts, it is normal to expect that the participation rate for males is higher than 90%; while
for females, it is as high as 80%. This rate only slightly goes down for the age 60 cohorts. For the
age 70 cohorts, more than 60% of men are working and around half of the women are in the labor
force. Even in their eighties, more than 20% of men and women are still involved in productive
activities. Figure 3.2 and Figure 3.3 further show the total annual hours and hours in agriculture-
related activities rural elderly worked in the past year. The patterns are similar to that found in
Figure 3.1: both male and female elderly keep working into very old ages. This life path of the
rural residents makes it essential to understand the factors that affect their labor supply behaviors.
47
In Confucian philosophy, the meaning of the filial piety includes taking care of one’s parents.
48
The China Health and Retirement Longitudinal Study (CHARLS) National Baseline. There will be de-
tailed discussion about the dataset later.
38
Figure 3.1 Labor Force Participation (Rural Elderly)
Figure 3.2 Annual Work Hours (Rural Elderly)
39
Figure 3.3 Annual Work Hours in Farming (Rural Elderly)
Since they are not only financially vulnerable but also physically vulnerable at old ages, it is par-
ticularly important to examine the role health played in the old-aged adults’ working lives.
The main objective of this paper is to examine the link between health and labor supply be-
haviors of rural elderly in China. There are several reasons that this study focuses on this group
of individuals. First of all, a majority of the population in China are rural residents. It is important
to understand their labor supply behaviors and how they respond to influencing factors like
health.
Second, huge differences exist between urban and rural areas in all aspects, and retirement
system is one of the most important dimensions among them. For the urban residents, especially
those who work in the formal sector
49
, they face mandatory retirement when they reach certain
ages. Specifically, men have to retire at age 60, female civil servants have to retirement at age 55
and female workers at age 50
50
(State Council, 1978). This makes most of the urban residents’
49
The formal sector usually refers to government, institutions, state-owned enterprises, firms and etc.
50
There are some exceptions. For example, people who work in highly dangerous industries can retire
earlier. And if an individual becomes disabled and loses his/her ability to work because of job-related
accidents, they can retire at any time. Additionally, individuals can retire earlier if they are deemed to have
completely lost the ability to work.
40
retirement decisions depend solely on the mandatory retirement system. On the contrary, in the
rural areas, there is no such system. Rural residents can choose when to stop working by them-
selves. Therefore, this allows us to examine different factors such as health that affect rural resi-
dents’ labor participation decisions as well as other labor market outcomes.
Third, there is in fact one advantage of studying this relationship in rural China. As have
been discussed in a large amount of literature, pension is always a non-negligible factor in peo-
ple’s retirement decisions and it affect individuals’ labor supply behaviors in various ways (See
Gustman and Steinmeier, 2005; Coile and Gruber, 2007). However, in rural China, this is a prob-
lem to be less worried about. For a long time, old-age support in rural China relies on all means
but formal pensions. Though New Pension System Program (NPRR) has been established since
2009 in the rural areas, the amount
51
of the pension is extremely small that it can barely cover the
expenses of a typical elderly. Therefore, it is natural to expect that at least so far, the role of pen-
sion is negligible when rural elderly make their labor supply decisions. This helps to better reveal
the relationship between health and labor supply for the old-aged adults without worrying about
the complications of pensions.
Despite the numerous things that may affect rural elderly’s labor supply behaviors, we
mainly focus on health. Because for the old-aged adults who are mostly involved in strenuous
and physically demanding works such as farming, health is an extremely important limiting fac-
tor when they decide how much effort they should make at work or whether to leave the labor
force.
Using the China Health and Retirement Longitudinal Study (CHARLS) National Baseline,
this study is able to quantify the relationship between health and labor supply of the rural elderly.
Unlike most of the previous literature, health measure in this study is multi-dimensional. We
measure health in five different dimensions - self-reported health, disability, depressive symp-
toms, biomarkers and parental longevity - in order to examine their differentiated effects. Labor
supply in this study is measured by labor force participation and hours of work. Since the major-
ity of the sample is to some degree involved in agriculture-related activities, we also consider
participation and hours of work in farming.
One thing to be mentioned is that we are not trying to establish a causal link between health
and labor supply behaviors. It is known that health is endogenous and it is hard to find a valid
51
The basic pension benefit per month per person is 55 yuan, approximately 9 U.S. dollars.
41
instrument variable from a cross-sectional dataset. Therefore, the analysis in this paper will only
focus on the associations instead of a causal relationship.
We estimate the participation decision and hours of work decision simultaneously. Our find-
ings suggest that health is no doubt one of the most important concerns when the rural elderly
make their labor supply decisions. Specifically speaking, there is a strong negative association
between poor general self-reported health status (GHS) and the elderly’s labor force participation.
Both men and women who reported poor GHS are more likely to leave the labor force at old ages.
Similar association has been found between disability and labor force participation. Individuals
with ADL/IADL-related limitations are less likely to stay working. We also find that stature and
overweight problems are important for men when they choose whether to stay in the labor force
or leave the jobs. While for women, hypertension and overweight problems are big issues which
are significantly associated with their lower probability of labor force participation as well as
reduced work hours. For both men and women, poor GHS are also associated with fewer hours
of work. All these effects from different dimensions of health persist after we controlled for all
the other covariates such as demographic characteristics, household economic conditions and
family composition. In addition, we found that health has differentiated effects on elderly’s labor
supply for individuals with different levels of wealth. It seems that wealth facilitate retirement
for both men and women in rural areas, even though the effects are small.
We also examine the impact from family, since they are indispensable parts of the old-age
support in rural China. Our results show that both men and women are significantly more likely
to stay in the labor force at old ages once they have non-adult sons in their families. No significant
association has been detected between the number of adult sons and elderly’s labor supply.
The rest of the paper is organized as follows. Section 3.2 is a brief review of the literature on
the link between health and labor market outcomes. Section 3.3 presents a labor supply model.
Data and descriptions of the variables are presented in Section 3.4. Section 3.5 is the empirical
approach and Section 3.6 provides detailed results. Section 3.7 concludes.
42
3.2 Literature Review
There have been numerous studies documenting the link between health and the labor market
behaviors in developed countries
52
. Earlier in the 70s, researchers already started to examine the
effect of health on the labor market behaviors. For example, Luft (1975) tried to examine the effect
of activity limits on labor force participation (LFP) and unemployment. Parsons (1977) studied
how self-reported health status affected individuals’ annual hours of work. Bartel and Taubman
(1979) investigated the impact of several specific chronic conditions on earnings.
In the past few decades, more and more articles started to focus on this issue and some of
them, in particular, concentrated on the elderly group, who are more vulnerable compared to the
young and middle-aged ones. Findings from these studies are quite consistent. For instance, Chi-
rikos and Nestel (1981) examined a group of older men between age 55 to age 69 and found a
negative association between health and annual hours of work. Bound et al. (1991) focused on the
respondents from the Health and Retirement Survey (HRS)
53
. They accounted the dynamic effects
of health on the transitions of older workers and indicated that health is an important determinant
of labor force participation decisions of the elderly. Several evidence from other OECD countries
also support the significant role health played in the elderly’s labor supply decisions (Cai and
Kalb, 2006; Cai and Cong, 2009; Schirle, 2010).
Health could be a more concerning factor for older people in developing countries, especially
those in the rural areas. For these individuals, work relies heavily on strength. In another word,
labor supply behaviors count on health.
In contrast to the literature from developed countries, evidence in developing countries are
comparatively limited (See Strauss and Thomas, 1998 for review). However, more studies utiliz-
ing household surveys are emerging. For instance, using a nationally representative data of 50-
64 years old (Korean Longitudinal Study of Aging), Lee and Smith (2010) found strong evidence
that depression leaded to reduced labor force participation. Benjamin et al. (2003) explored the
effects of health on labor supply of the elderly in rural China and their results indicated that
health only played a small observable role in explaining the declining pattern; however, it is quite
52
See Currie and Madrian (1999) for a detailed review.
53
Respondents are men and women born between 1931 and 1941 when the first survey was conducted in
1992.
43
significant. The role of health was further confirmed in Pang et al. (2004). By illustrating the fac-
tors that might facilitate a rural elderly’s work decision, they found that individuals with moder-
ate or severe illness were more likely to exit from the labor force. This link was also observed in
Giles et al. (2011), where they showed a very pronounced effect of health status (measured by
ADL) on work activities of China’s rural residents.
The studies mentioned above all have different ways of measuring health. Most of the studies
focused on one specific dimension of health. For instance, Luft (1975) used activity limits; Bartel
and Taubman (1979) used specific diseases such as heart disease, arthritis and so on; Bound et al.
(1999) used self-reported health status (GHS); Pang et al. (2004) used self-reported illness; and
Giles et al. (2011) used ADL-related limits. However, health is multidimensional; focusing only
on one aspect would neglect the impacts from other sides. As stated in Blau and Gilleskie (2001),
no single measure of health is adequate to explain the labor supply behaviors. To address this
issue, some researchers have tried to combine different indicators of health in their study, to ex-
amine the effects of health in a broader dimension. Cases are Blau and Gilleskie (2001), which
included GHS, difficulties with ADLs and specific conditions; Mete and Schultz (2002), which
also used GHS and ADLs; Benjamin et al. (2003), which examined GHS and BMI, though sepa-
rately.
Another issue that has been addressed by many studies is the endogeneity of health. Most of
the earlier literature treated health as exogenous or just estimated the association between health
and labor supply. Later studies experimented various ways to deal with the endogeneity problem.
Several researchers adopted an IV approach and used exogenous/objective health variables to
instrument for endogenous/subjective ones. For example, Bound et al. (1999) adopted a latent
variable approach and used functional limitation variables to instrument for the GHS. They ar-
gued that the functional limitation variables are exogenous and therefore the estimates produced
by this approach are consistent. In Mete and Schultz (2002), GHS and ADL limitations were in-
strumented by parent longevity, birthplace and childhood conditions. However, no consensus
has been arrived about a valid instrument for health. Objective health measures, to a large extent,
are less likely to be subject to issues such as rationalization; however, it is hard to argue that they
are not correlated with individuals’ labor supply behaviors. So are the instruments used in Mete
and Schultz (2002). There are also several studies trying to solve the endogeneity issue by simul-
taneous modelling approach. For instance, Blau and Gilleskie (2001) jointly estimate a model of
44
health with the employment transition models, allowing correlation between the LF and health
disturbances. Cai and Kalb (2006) also estimated the health equation and the labor force partici-
pation equation simultaneously.
Since it is too difficult to find a valid instrument from a cross-sectional dataset that is corre-
lated with health, but uncorrelated with labor outcomes; we will focus on investigating the asso-
ciations between health and labor supply behaviors for now. Later when more waves of data
comes out, we will try to extend this study in order to establish a causal link between health and
labor supply of the rural elderly.
3.3 Framework
To examine the role of health played in elderly’s labor supply decisions, we incorporate it into
the frame of the labor supply model. For simplicity, we assume the decision is made at the indi-
vidual level over one period. An individual’s objective therefore is to maximize his/her utility:
max
𝑐 ,𝐿 ,𝐻 𝑈 (𝑐 , 𝐿 , 𝐻 ; 𝑋 , 𝐹 , 𝐸 , 𝑒 1
)
(3.1)
subject to
𝑝𝑐 = 𝑤 × 𝐿 + 𝑦 (3.2)
where utility depends on consumption 𝑐 , labor supply 𝐿 , health 𝐻 and other covariates such as
demographic and socioeconomic characteristics 𝑋 , family and household structure 𝐹 , community
environment (infrastructure, public health environment and etc.) 𝐸 , and unobserved preferences
𝑒 1
. Equation (2.2) is the budget constraint, where 𝑝 is the price of consumption, 𝑤 is the wage and
𝑦 is the non-labor income.
Following Grossman (1972), a health production function is specified as follows:
𝐻 = 𝐻 (𝐼 , 𝐿 ; 𝑋 , 𝐹 , 𝐸 , 𝑒 2
) (3.3)
where health depends on health inputs 𝐼 , labor supply 𝐿 , individual’s demographic and socioec-
onomic characteristics 𝑋 , family and household structure 𝐹 , community environment 𝐸 and an
unobserved part 𝑒 2
.
Wage equation can also be specified as depending on health, individual characteristics, com-
munity environment and also an unobserved part 𝑒 3
:
w = w(H; X, E, 𝑒 3
) (3.4)
45
Solving for the optimal labor supply choice, we can get:
𝐿 𝑠 = 𝐿 (𝐻 , 𝑤 (𝐻 ; 𝑋 , 𝐸 , 𝑒 3
), 𝑋 , 𝐹 , 𝐸 , 𝑝 , 𝑦 , 𝑒 1
) (3.5)
It specifies that the rural elderly residents’ labor supply decisions is a function of health 𝐻 , wage
𝑤 , individual characteristics 𝑋 , family background 𝐹 , community environment 𝐸 , real price of
consumption goods 𝑝 , non-labor income or assets 𝑦 , as well as the individual’s unobserved pref-
erences 𝑒 1
.
3.4 Data and Measurement
3.4.1 Data and Sample
The data used for analysis is the Chinese Health and Retirement Longitudinal Survey (CHARLS)
2011 National Baseline, which has been discussed in the previous chapter section 2.4.1. Since we
focus on the elderly in rural China, the sample is limited to respondents aged 50 to 90 with a rural
hukou (registration). This results in 10,611 observations in which women represent about 51%.
3.4.2 Measurement of Labor Supply
Participation: Labor force participation measures the elderly’s decisions to work. Individuals
who were still working
54
at the time of the survey are considered as in the labor force. This group
also includes respondents who were not working when the survey took place but was searching
for new jobs during the past month
55
. Respondents who reported not working and planned to
stay out of the labor force are considered as the counterpart. From the summary statistics (Table
3.1), it can be seen that in the sample, among rural residents whose ages are between 50 and 90,
71% are still in the labor force. Participation rate is higher for men than for women.
54
Here, according to the questionnaire, work includes: 1. Engage in agricultural work (including farming, for-
estry, fishing, and husbandry for your own family or others) for more than 10 days in the past year; 2. Work (earn a
wage, run your own business and unpaid family business work, et al. Does not include doing own housework or doing
voluntary work.) for at least one hour last week; 3. Have a job but are temporarily laid-off, or on sick or other leave, or
in-job training, but expect to go back to this job at a definite time in the future or within 6 months or still receive any
salary from this job.
55
The numbers of the respondents falling into this particular group is small. In the rural sample used in
the analysis, only 24 individuals reported searching for news jobs during the last month.
46
Table 3.1. Summary Statistics
Panel A
Whole Sample Men Women
Labor Supply
Labor Force Participation 0.71 0.79 0.64
Annual Work Hours 1172.23 1432.62 931.00
Annual Work Hours (Agriculture) 832.81 920.04 750.56
Panel B
Whole Sample Men Women
Health Measures
Poor GHS 0.33 0.29 0.37
Disability 0.33 0.27 0.38
Depressive Symptoms 8.46 7.46 9.36
Leg Length 47.74 49.60 46.03
Hypertension 0.33 0.31 0.35
Overweight 0.26 0.19 0.32
Undernourished 0.09 0.09 0.09
Father is Alive 0.09 0.09 0.09
Mother is Alive 0.19 0.18 0.19
Demographic &
Socioeconomic
Characteristics
Age 62.20 62.11 62.30
Educ_Illiterate 0.37 0.18 0.55
Educ_Primary 0.43 0.53 0.35
Educ_Secondary 0.19 0.29 0.10
Married 0.85 0.90 0.81
log PCE 8.30 8.32 8.29
Leased Land (mu) 14.46 14.68 14.25
County-Level Health Insurance Reim-
bursement Rate 0.25 0.25 0.25
Family Composi-
tion
# of Adult Sons (>25 yrs) 1.47 1.37 1.57
# of Non-Adult Sons (15-25 yrs) 0.11 0.14 0.09
# of Adult Daughters (>25 yrs) 1.27 1.19 1.34
# of Non-Adult Daughters (15-25 yrs) 0.12 0.14 0.10
# of Grandchildren 4.19 3.81 4.55
Note: Age range here is 50-90.
47
Annual Hours of Work: To measure the intensity of the elderly’s labor supply, annual work hour
are constructed for respondents who were in the labor force. Basically there were in total five
categories of jobs: farming, employed farming, self-employed (non-farming), employed (non-
farming) and family business helpers. Most of the individuals held one job; however, a very small
portion of the respondents held a side job besides the main one. They can mostly often be char-
acterized as taking farming as a main job, while doing some small business or employment work
as a side job. The annual work hours
56
are calculated as the summation of all the work hours,
including in main job and side jobs. The average annual work hour is around 1,172, among which
exists a large gender difference. On average, men work more than 1,400 hours per year, while
women work less than 1,000 hours annually.
Annual Hours of Work in Agriculture-related Activities: We include another variable measur-
ing the annual work hours in agriculture-related jobs. 80% of the individuals are to some extent
engaged in farming. Hours spent in the agriculture jobs are more flexible especially compared to
the employed ones. Therefore, this variable may better reflect the respondents’ own labor supply
decisions. In addition, agriculture jobs mostly involve strenuous laboring, which makes health a
more important concerning factor. Thus it may unveil the relationship between health status and
respondents’ labor supply decisions in a better and cleaner way. The average annual hour spent
in agriculture is 832 and the pattern across age groups is similar to the previous work hour meas-
ure.
3.4.3 Measurement of Health
The concept of health is multi-dimensional, which makes it hard to measure. Currie and Madrian
(1999) summarize eight categories of the types of measures
57
in the social science literature. Indi-
cators of each category has its own advantage as well as limitation. They reflect different kinds of
56
In the questionnaire, respondents were asked “How many months did you work on [...] in the past year?”,
“How many days did you work in [...] per week on average during a normal work month in the past year?”, and “How
many hours did you usually work in [...] per day during a normal work day in the past year?”. We constructed the
annual work hours for each job and then add them together if he/she has more than one job.
57
Among which are seven types of health outcome measures: (1) self-reported health status; (2) whether
there are health limitations on the ability to work; (3) whether there are other functional limitations such as
ADLs problems; (4) the presence of chronic and acute conditions; (5) clinical assessments of such things as
mental health or alcoholism; (6) nutritional status; (7) expected or future mortality. One thing needs to be
48
information, which may have distinctive impacts on an individual’s labor market behaviors
58
.
Due to this multi-dimensionality, Strauss and Thomas (1998) argue that it is useful to examine
several indicators simultaneously. Therefore, in the following analysis, we include five dimen-
sions of health to investigate how changes in these different health aspects affect elderly’s labor
supply behaviors.
General Health Status (GHS): In CHARLS, GHS is rated on a scale of 5: 1=very good, 2=good,
3=fair, 4= poor, 5=very poor. A binary measure is created as =1 if reported poor or very poor
health; =0 if reported fair, good or very good health. In our sample, 33% of the respondents re-
ported poor health. This percentage is higher for women of whom 37% rated their health as poor
or very poor.
Disability: The disability measure is defined as the presence of any impairment in any of the
ADLs/IADLs
59
. Respondents are defined as disabled if they cannot do at least one of those tasks
or if they have great difficulty and need help from others in performing them. According to this
definition, 33% of the individuals are disabled.
Depressive Symptoms: CES-D is used as an indicator of individual’s depressive symptoms. In
the questionnaire
60
, respondents were asked to rate how they have felt and behaved during the
past week on a four-scale metric: rarely or none of the time (less than 1 day), some or little of the time
(1-2 days), occasionally or a moderate amount of the time (3-4 days) or most or all of the time (5-7 days).
The CES-D scores
61
range from 0 to 30 with higher scores representing higher levels of depressive
noticed is that, there are many other measures and dimensions such as biomarkers that are not mentioned
in this study.
58
For instance, GHS might be an overall assessment of the individual’s work capacity, while a
ADLs/IADLs indicator is a reflection of a respondent’s disability level. Stature may mirror a person’s child-
hood nutritional status that affects subsequent labor market outcomes, while depressive symptoms reflect
one’s psychological welfare which may affect work absence and productivity.
59
Difficulties which are expected to last less than three months are excluded. In the health status module,
respondents were asked the following questions: Because of health and memory problems, do you have any dif-
ficulty with dressing, bathing or showering, eating, getting into or out of bed, using the toilet, or controlling urination
and defecation? Do you have any difficulty with doing household chores, preparing hot meals, shopping for groceries,
managing your money or taking medications?
60
There is a ten question version of the Center for Epidemiologic Studies Depression Scale (CES-D). CES-
D questions in the questionnaire are as follows: DC009 I was bothered by things that don’t usually bother me;
DC010 I had trouble keeping my mind on what I was doing; DC011 I felt depressed. DC012 I felt everything I did
was an effort; DC013 I felt hopeful about the future; DC014 I felt fearful; DC015 My sleep was restless; DC016 I was
happy; DC017 I felt lonely. DC018 I could not get going.
61
Following Radloff (1977), CES-D scores (0-30) are constructed as follows: for each negative symptom, if
the respondent chooses rarely, then value 0 is assigned; if he/she chooses most or all of the time, then value
3 is assigned. For positive symptoms, the pattern is reversed.
49
symptoms. We use 10 and above as a cutoff point for high depressive symptoms (Andresen et al.,
1994). Summary statistics suggest that on average, women are more depressed than men.
Biomarkers: We include four biomarker indicators: lower leg length, hypertension, overweight
and undernourished. In the previous literature which use anthropometric measurements as
health indicators, height is commonly used as a measure of adult health. However, height of the
older individuals may be contaminated by height shrinkage from aging, which is a negative func-
tion of socioeconomic characteristics (Huang et al., 2013). To address this issue, I follow Huang
et al. (2013) and use lower leg length to predict pre-shrinkage height (See also Chumlea et al.,
1985; Zhang et al., 1998). Since lower leg length is positively correlated with attained adult height
and does not shrink as people age, we use it instead of measured height. Hypertension is meas-
ured according to the standard definition of the World Health Organization (WHO); it equals 1
if the mean systolic blood pressure (SBP) is 140 mm Hg or greater or if the mean diastolic blood
pressure (DBP) is 90 mm Hg or greater
62
. In addition, respondents who report being diagnosed
with hypertension by a doctor are also regarded as being hypertensive. Undernourished is de-
fined as a BMI smaller than 18.5 while overweight is defined as a BMI greater than 25. Both of the
measures follow the standard definitions of the WHO.
Parental Longevity: Two variables are constructed here: whether the respondent has a living re-
spondents in our sample are all aged 50 or over, the age range of their parents should be around
70 or higher. Needless to say, it is a positive indicator of health if parents of individuals have
lived to these ages. If there is intergenerational transmission between parental health and their
older adult children’s health, then this healthiness of the parents will affect the health of the re-
spondents (Kim et al., 2014). The sample shows that among the parents of the elderly respondents,
mothers are much more likely to be alive at the time of the survey. This is partly because mothers
tend to be younger than fathers and partly because female old age mortality is lower than men’s.
The possibility of having a living father or mother is similar for both men and women.
3.4.4 Covariates
We also include a set of individual characteristics. 1-year interval age dummies are used to better
capture the age gradient. Since rural elderly on average have low educational attainments, three
62
In the CHARLS, blood pressure of each respondent is measured three times and the mean of the meas-
urements is used.
50
educational groups are constructed: illiterate (reference group), have some primary education
have some secondary education. Summary statistics show that 37% of the respondents are illit-
erate, and only 19% have attended secondary schools. One thing to notice is that, there exists a
large difference between men and women. Among men, 18% are illiterate. However, for women
it is 55%. A binary marital status variable is included in which =1 if married or currently live
together, =0 otherwise
63
. It is obvious to note the percentage of married men is higher than mar-
ried women. One possibility might be that women outlive men.
It is natural to expect that the family and household structure may affect elderly’s labor sup-
ply decisions directly. For instance, if an old farmer has several adult sons, then he/she may not
need to stay in the labor force since the sons might help take care of the agriculture work. There-
fore, a set of variables are included to account for the potential impacts. Number of Adult Sons
64
and Number of Adult Daughters characterize the number of sons and daughters over 25 years old
respectively in this family, while Number of Non-Adult Sons and Number of Non-Adult Daughters
indicate the number of sons and daughters between age 15-25 accordingly. Grandchildren might
also be a concern for some of the elderly in our sample when they decide whether or not to spend
more time taking care of the grandkids. It would be ideal if we can differentiate grandkids that
are of the age that must be taken care of (for example, under 6 years old) and grandkids that are
older. However, due to the limited information on the ages of the grandchildren in the dataset,
we only include a variable accounting for the Number of Grandchildren in the family.
In addition to the demographic variables and family composition, economic variables may have im-
portant effects on labor supply decisions. Individuals with high incomes may not need to be working when
they become old; while individuals with low incomes may have to work as long as they can to support
themselves. Since in developing countries, expenditure can be measured with less noise and provides a
better measure of welfare than income, log per capital household expenditure (log PCE) is used in the anal-
ysis (Strauss and Thomas, 1995). As the majority of the elderly in rural China are involved in farming,
we also control for the amount of land leased by each household. Summary statistics show that
on average rural residents in our sample have 14 mu of land
65
. In addition to these two variables,
63
The counterpart includes never married, separated, divorced and widowed.
64
For this set of measures, we only account for individuals that are alive at the time of the survey
65
mu is the unit of land area. One mu equals to about 0.16 acre.
51
we also include a health insurance-related indicator – county-level reimbursement rate for inpa-
tient care
66
. Summary statistics show that the average reimbursement rate for inpatient care is
25%.
3.5 Empirical Strategy
As specified earlier, the rural elderly residents’ labor supply decisions is a function of health 𝐻 ,
wage 𝑤 , individual characteristics 𝑋 , family background 𝐹 , community environment 𝐸 , real price
of consumption goods 𝑝 , non-labor income or assets 𝑦 , as well as the individual’s unobserved
preferences 𝑒 1
. However, one limitation here, as well as in most of the developing countries is
that, wages may not be observable, especially in rural areas where a majority of the individuals
work on their own farms and do not earn “wage”. Therefore, we substitute the wage equation
(3.4) into (3.5), and the labor supply equation becomes:
𝐿 𝑠 = 𝐿 (𝐻 , 𝑋 , 𝐹 , 𝐸 , 𝑝 , 𝑦 , 𝑒 1
, 𝑒 3
) (3.6)
The empirical equation of this labor supply function could be written as follows:
𝐿 𝑖𝑗
𝑠 = 𝛽 0
+ 𝛽 1
𝐻 𝑖𝑗
+ 𝛽 2
𝑋 𝑖𝑗
+ 𝛽 3
𝐹 𝑖𝑗
+ 𝛽 4
𝑦 𝑖𝑗
+ 𝐸 𝑖𝑗
+ 𝘀 𝑖𝑗
(3.7)
where 𝑖 stands for individual and 𝑗 stands for community.
As mentioned above, two sets of variables are used to characterize elderly’s labor supply
behaviors: labor force participation and hours of work. In the previous literature, the labor force
participation decision and the work hour decision are usually considered separately. Standard
univariate Tobit model has traditionally been applied to account for the censoring at zero of work
hours. However, one limitation of the application of the Tobit model is that it assumes the same
stochastic process for the continuous work hours and the discrete switch at zero. Since zero work
hour of an individual might result from unemployment as well as voluntary retirement from the
labor force, Blundell and Meghir (1987) proposed to relax this assumption and allow for distinct
processes determining the 0-1 choice and the continuous work hours. In the model they depicted,
the same factors could have dissimilar effects on the two different decisions. Following their
66
Since NCMS usually does not cover the cost of outpatient visits, we only include health insurance reim-
bursement rate for inpatient visits.
52
work, we apply a bivariate alternative to the Tobit model distinguishing the two processes. Let
𝐿 2
∗
denote the hours of work and 𝐿 1
∗
be a latent variable determining the participation process.
Participation Equation:
𝐿 1
= {
1, 𝐿 1
∗
> 0
0, 𝐿 1
∗
≤ 0
(3.8)
The individual chooses to be in the labor force if the latent variable 𝐿 1
∗
is greater than zero. If 𝐿 1
∗
is
lower or equal to zero, then this individual is supposed to be out of the labor force.
Outcome Equation:
𝐿 2
= {
𝐿 2
∗
, 𝐿 1
∗
> 0
0, 𝐿 1
∗
≤ 0
(3.9)
If the individual is in the labor force, then hours of work is observed. Otherwise, it is assumed to
be missing and no one can observe that.
The empirical specification thus can be written as follows:
𝐿 1𝑖𝑗
∗
= 𝛼 0
+ 𝛼 1
𝐻 𝑖𝑗
+ 𝛼 2
𝑋 𝑖𝑗
+ 𝛼 3
𝐹 𝑖𝑗
+ 𝛼 4
𝑦 𝑖𝑗
+ 𝐸 𝑗 + 𝑣 𝑖𝑗
1
(3.10)
𝐿 2𝑖𝑗
∗
= 𝛽 0
+ 𝛽 1
𝐻 𝑖𝑗
+ 𝛽 2
𝑋 𝑖𝑗
+ 𝛽 3
𝐹 𝑖𝑗
+ 𝛽 4
𝑦 𝑖𝑗
+ 𝐸 𝑗 + 𝑣 𝑖𝑗
2
(3.11)
We assume that the error terms are joint normally distributed and homoskedastic,
(
𝑣 𝑖𝑗
1
𝑣 𝑖𝑗
2
) ~𝑁 (
0
0
,
1
𝜎 21
𝜎 12
𝜎 2
2
) (3.12)
The likelihood function of the model is given by:
ℒ = ∏ ∏ {𝑃𝑟 [𝐿 1𝑖𝑗
∗
≤ 0]}
1−𝐿 1𝑖𝑗
{𝑓 (𝐿 2𝑖𝑗
| 𝐿 1𝑖𝑗
∗
> 0) × Pr [𝐿 1𝑖𝑗
∗
> 0]}
𝐿 1𝑖𝑗
𝑚 𝑗 =1
𝑛 𝑖 =1
(3.13)
Endogeneity Issue: so far we have assumed that health can be truly measured. However, this is
not the case in reality. All the health indicators used in the analysis are, to some degree, suffer
53
from random or/and systematic measurement error (See Strauss and Thomas, 1998). For instance,
there might be reporting bias in the GHS measure, since no metric has been established for indi-
viduals to compare their health to. In addition, the way individuals evaluated their health might
be systematically correlated with their educational levels, family backgrounds, incomes and per-
haps labor supply behaviors. Similarly, it is normal to assume that an individual’s depressive
symptoms are affected by his/her socioeconomic characteristics. More objective measures such
as disability and lower leg length are less likely to suffer from systematic measurement errors;
however, they may still be contaminated by the random ones.
Many studies have attempted to deal with this endogeneity and measurement error problem
by instrumenting health measures (Bound et al., 1999; Blau and Gilleskie, 2001; Mete and Schultz,
2002; Lee and Smith, 2010). However, it is difficult to find convincing instruments that are corre-
lated with health but uncorrelated with labor supply behaviors. Some researchers use objective
health indicators to instrument for the subjective health measures; however, if the measurement
errors are systematically correlated with other variables, their coefficients will be biased as well.
One thing mentioned by Bound (1991) is that, on the one hand measurement errors in health
measures are likely to bias the health effects towards zero, on another hand endogeneity of health
indicators might exaggerate the impact of health, therefore these two effects may cancel out.
Thus, in the following analysis, we include five different health dimensions trying to capture
their different impacts without particularly dealing with the potential endogeneity issue. When
more waves of data come out, we will come back and deal with this issue then. In this paper, we
will only consider about the pattern of association between health and individual’s labor supply
decisions.
3.6 Results
3.6.1 Simultaneous Decision of LFP and Hours of Work
Table 3.2 presents the results estimating participation decision and work hour decisions for men
simultaneously. We have three sets of regressions. Within each set, first column presents the work
hour results and the second column presents the labor force participation results. Only health
indicators and demographic characteristics are included. Different dimensions of health are
added sequentially in order to investigate their variant effects. All the specification contains a
fixed community effects and all the standard errors are clustered at the community level. As is
54
Table 3.2 Health and Simultaneous Labor Supply Decisions: Men
(1) (2) (3) (4) (5) (6)
Hours LFP Hours LFP Hours LFP
Poor GHS -88.539 -0.586*** -115.421* -0.586***
[57.7234] [0.0717] [60.9043] [0.0820]
Disability -37.336 -0.572*** -35.122 -0.508***
[68.5281] [0.0846] [69.4277] [0.0877]
Depressive Symptoms -0.507 -0.002 -1.424 -0.002
[4.9325] [0.0068] [5.0790] [0.0078]
Leg Length 1.396 0.022*
[8.6958] [0.0115]
Hypertension -21.302 0.050
[53.9506] [0.0810]
Overweight -86.613 -0.223***
[68.6189] [0.0838]
Undernourished -37.082 -0.062
[90.2684] [0.1113]
Father Alive -43.512 0.229
[90.8664] [0.1605]
Mother Alive -46.765 0.053
[69.5809] [0.1123]
Educ_Primary -115.024* 0.098 -88.707 0.021 -105.776 0.008
[68.4260] [0.0779] [72.2223] [0.0879] [72.0533] [0.0978]
Educ_Secondary -81.116 0.011 -74.558 -0.121 -109.497 -0.147
[80.7886] [0.0898] [87.9301] [0.1112] [87.4298] [0.1232]
Married 333.854*** 0.475*** 313.062*** 0.439*** 301.877*** 0.443***
[82.8512] [0.0940] [88.4765] [0.1057] [91.4461] [0.1167]
N 4,375 4,375 3,941 3,941 3,424 3,424
rho -0.280 -0.280 -0.330 -0.330 -0.370 -0.370
F_stat (health) 3.420 151.770 9.200 114.080
P_value (health) 0.331 0.000 0.419 0.000
Note: All the specifications controls for age dummies and community fixed effects. Rho represents the correlation
between participation equation and work hour equation. Robust standard errors are in the parenthesis. All stand-
ard errors are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
55
Table 3.3 Health and Simultaneous Labor Supply Decisions: Women
(1) (2) (3) (4) (5) (6)
Hours LFP Hours LFP Hours LFP
Poor GHS -129.920** -0.474*** -106.024* -0.438***
[53.8965] [0.0490] [56.4270] [0.0527]
Disability -56.659 -0.305*** -21.821 -0.267***
[57.7120] [0.0585] [57.9807] [0.0629]
Depressive Symptoms 11.037** 0.003 8.656* -0.001
[4.3139] [0.0048] [4.4618] [0.0055]
Leg Length 7.874 0.000
[8.8403] [0.0095]
Hypertension -149.717*** -0.106*
[55.6831] [0.0633]
Overweight -172.691*** -0.239***
[53.0072] [0.0651]
Undernourished 62.923 0.057
[86.8402] [0.0965]
Father Alive 16.940 -0.130
[73.9919] [0.0948]
Mother Alive 69.669 0.034
[59.2233] [0.0838]
Educ_Primary 25.395 -0.020 46.169 -0.081 8.357 -0.155***
[47.3017] [0.0508] [47.4180] [0.0553] [46.9410] [0.0594]
Educ_Secondary 89.677 -0.159* 66.630 -0.248** -19.669 -0.351***
[86.8697] [0.0902] [90.0223] [0.0990] [96.7314] [0.1073]
Married 34.748 0.329*** 52.549 0.331*** 42.958 0.298***
[79.9644] [0.0685] [79.4503] [0.0766] [81.3834] [0.0850]
N 4,825 4,825 4,442 4,442 3,790 3,790
rho -0.020 -0.020 0.010 0.010 -0.050 -0.050
F_stat (health) 11.120 157.240 42.640 145.270
P_value (health) 0.011 0.000 0.000 0.000
Note: All the specifications controls for age dummies and community fixed effects. Rho represents the correlation
between participation equation and work hour equation. Robust standard errors are in the parenthesis. All stand-
ard errors are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
56
consistent with Blau et al. (1997), the largest effects come from the GHS and physical functioning.
The association between poor GHS and LFP is negative and significant. So is the link between
poor GHS and hours of work. Disability, however, only reduces individual’s labor force partici-
pation. Once they decide to stay in the labor force, it does not have much effect on how many
hours they work. One possibility might be that for these individuals, their disability are not severe
enough to stop them from working at this age. It is also possible that people select into certain
jobs that do not require these abilities. We also find a small positive effect from leg length. It seems
that men with longer legs are more likely to keep working at old ages. There is also some evidence
showing that men who are overweight are less likely to stay employed. For the other covariates,
marriage consistently has a positive and significant effect on participation and hours of work,
while education does not seem to play an important role in both decisions. It is not difficult to
note that the same health factors have differential effects on participation and hours of work when
these two decisions are considered simultaneously. Health plays an important role when indi-
viduals choose whether to stay or leave the labor force, with coefficients before the health
measures are individually and jointly significant. However, health does not seem quite essential
when evaluating its impact on hours of work. Only people with poor GHS tend to work fewer
hours and the joint test of all the health indicators shows non-significance.
Results for female are presented in Table 3.3. Similarly, poor GHS and disability are strongly
negatively correlated with women’s labor force participation; though the associations are slightly
weaker compared to men. Negative correlation has also been found between poor GHS and an-
nual hours of work. In addition, it is shown that a positive association exists between depressive
symptoms and work hours; however, this could result from a reserve relationship in which re-
spondents who work longer hours tend to be more depressed. In addition, it seems that women
who have hypertension and those who suffer from overweight problem, are much more likely to
work fewer hours. They also have a higher probability of exiting the workforce. Similar to men,
married women are more likely to stay in the labor force at old ages; however, there is no sign
that they work longer hours. Unlike the neutral role of education in men’s labor supply decisions,
it is important for women’s labor force participation. Women with some education, especially
those who have received secondary education, tend to leave the labor force at earlier ages.
Since the majority of people in the sample are to some degree involved in farming, we also examine
the effects of health on agriculture-related activities by estimating the farming participation and hours of
farm work simultaneously. Table 3.4 shows regression results for both men and women. Aligned with the
57
Table 3.4 Health and Simultaneous Labor Supply Decisions in Agriculture
Men Women
Hours LFP Hours LFP
Poor GHS 27.599 -0.563*** 60.037 -0.380***
[53.0581] [0.0763] [38.0130] [0.0551]
Disability -30.642 -0.505*** 32.396 -0.155***
[59.9308] [0.0842] [57.7787] [0.0418]
Depressive Symptoms 7.758* -0.003 10.052** 0.005
[4.6077] [0.0072] [4.6292] [0.0051]
Leg Length 3.290 0.019* -2.880 -0.005
[7.6264] [0.0110] [6.5905] [0.0073]
Hypertension -30.858 0.014 -64.824 -0.123***
[46.0348] [0.0750] [64.4599] [0.0472]
Overweight -245.730*** -0.229*** -158.839* -0.260***
[55.5778] [0.0818] [83.6008] [0.0514]
Undernourished 11.833 -0.116 67.823 0.079
[70.7365] [0.1041] [75.3393] [0.0890]
Father Alive -16.076 0.144 99.942 -0.134
[66.3217] [0.1463] [84.0289] [0.0825]
Mother Alive 2.746 0.077 -18.173 0.070
[59.4272] [0.1146] [49.9388] [0.0578]
Educ_Primary -85.758 -0.025 -79.613** -0.152**
[63.2824] [0.0918] [39.8923] [0.0695]
Educ_Secondary -275.353*** -0.199* -212.555*** -0.294***
[71.4980] [0.1185] [70.2860] [0.1100]
Married 295.802*** 0.423*** 149.351** 0.328***
[79.7202] [0.1093] [68.4660] [0.0835]
N 3,477 3,477 3,807 3,807
rho -0.290 -0.290 -0.140 -0.140
F_stat (health) 30.410 124.250 30.710 184.210
P_value (health) 0.000 0.000 0.000 0.000
Note: All the specifications controls for age dummies and community fixed effects. Rho represents the correlation
between participation equation and work hour equation. Robust standard errors are in the parenthesis. All stand-
ard errors are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
58
results in Table 3.2 and Table 3.3, both poor GHS and disability indicator significantly reduce
individual’s participation in farming. However, it seems that GHS no longer has any impact on
hours of farm work for them. One possibility might be that farming is strenuous and elderly select
into farming based on their health status. Individuals with poor health or disability may stay out
of farming and work in the non-farming businesses. It might also be that individuals who stick
with farming are those with the lowest educational attainments or skills. They have no choice but
to work on the farm and cannot take the risk of resting even if their health status deteriorates. The
depression indicators; however, is positively correlated with hours of work in farming. Never-
theless, this might be due to the fact that individuals who work longer hours on the farm are more
depressed than others. Unlike in annual hour results where education does not seem to play a
role, it consistently has a strong effect on the labor supply decisions of agriculture-related activi-
ties and the effect mainly comes from individuals with at least some secondary education.
3.6.2 Family and Household Structures
So far, we have only controlled for individual characteristics and community fixed effects. How-
ever, as discussed in some of the literature, family and household characteristics might also affect
the older adults’ labor supply decisions. Therefore, in this part, we will include family composi-
tion as well as some other household characteristics to investigate their effects and whether in-
cluding them will change the impact from health. Table 3.5 shows the regression results for both
men and women. Similarly, within each set, first column presents the work hour results and the
second column presents the participation results. For both men and women, it can be seen that
effects from poor GHS and disability on participation decision remain strong and consistent.
Regarding the family composition, one traditional thought in the context of developing coun-
tries is that individuals with more adult sons can retire earlier since they can rely on their sons
for old-age support. However, this situation seems to have changed according to our results. We
do not see much effect of having adult sons in the family; on the contrary, the largest impact of
the family structure comes from the non-adult sons. Men are more likely to stay in the labor force
if they have non-adult sons. One possible explanation is that parents have to keep working to
invest in their non-adult sons’ high school/vocational school/college education. And this could
be one kind of survival strategy to ensure their retirement in the near future once their non-adult
sons start to work and support them (Pang et al., 2004).
59
Table 3.5 Effects from Family Composition
Men Women
Hours LFP Hours LFP
Health Measures
Poor GHS -132.134** -0.585*** -115.561 -0.474***
[61.3620] [0.0937] [403.9163] [0.0710]
Disability -65.398 -0.575*** -32.207 -0.338***
[72.7603] [0.1021] [293.9752] [0.0737]
Depressive Symptoms 0.403 -0.004 9.103** -0.001
[5.4212] [0.0091] [4.3585] [0.0065]
Leg Length 3.098 0.023* 7.512 0.004
[9.0695] [0.0139] [10.1564] [0.0160]
Hypertension -27.034 -0.017 -144.646** -0.063
[58.2143] [0.0931] [69.5732] [0.0887]
Overweight -72.842 -0.250*** -184.611 -0.244***
[71.6454] [0.0968] [205.8683] [0.0904]
Undernourished -30.074 -0.058 67.246 -0.013
[91.7193] [0.1308] [91.3088] [0.1165]
Father Alive -17.266 0.294 24.921 -0.004
[94.1038] [0.1968] [77.0461] [0.1405]
Mother Alive -40.499 0.058 75.465 -0.008
[70.4702] [0.1258] [64.7246] [0.1162]
Demographic & Socioeconomic Characteristics
Educ_Primary -117.134 0.089 -1.420 -0.094
[72.9535] [0.1078] [76.2040] [0.1140]
Educ_Secondary -108.996 -0.191 -14.037 -0.155
[88.4900] [0.1419] [155.2637] [0.1292]
Married 314.421*** 0.338** 77.285 0.304***
[94.7351] [0.1389] [309.0408] [0.1135]
logPCE -0.233 -0.121** 53.811 -0.186***
[35.4772] [0.0583] [148.1331] [0.0435]
Amount of Land -0.435*** 0.000 1.475 -0.001**
[0.1384] [0.0003] [1.4999] [0.0005]
County-Level HI Reimbursement Rate -237.621 0.805 -317.025 -0.788
[809.9791] [2.0264] [1,027.5570] [1.0074]
Family Composition
# of Adult Sons 31.325 0.016 -14.207 0.013
[36.2465] [0.0552] [37.5924] [0.1026]
# of Non-Adult Sons 107.222 0.270** 71.851 0.257
[66.0210] [0.1262] [162.6456] [0.1769]
60
# of Adult Daughters -13.320 0.009 -40.189 0.001
[31.9217] [0.0615] [37.9842] [0.0763]
# of Non-Adult Daughters 70.828 -0.044 7.620 0.142
[65.4545] [0.1316] [121.9172] [0.1406]
# of Grandchildren 7.301 -0.022 17.160 -0.026
[13.1542] [0.0242] [31.3126] [0.0288]
N 2,990 2,990 3,164 3,164
rho -0.300 -0.300 0.060 0.060
F_stat (health) 9.610 92.660 32.020 116.520
P_value (health) 0.383 0.000 0.000 0.000
Note: Adult sons are defined as sons aged 25 and above. Adult daughters are defined as daughters aged 25 and above.
Non-Adult Sons are defined as sons between age 15-25. Non-Adult Daughters are defined as daughters between age
15-25. Age dummies and community fixed effects are included in all the specifications. Rho represents the correla-
tion between participation equation and work hour equation. Robust standard errors are in the parenthesis. All
standard errors are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
61
Table 3.6 Effects from Family Composition (Agriculture)
Men Women
Hours LFP Hours LFP
Health Measures
Poor GHS -4.184 -0.565*** 47.080 -0.403***
[52.1664] [0.0873] [43.1690] [0.0642]
Disability -1.914 -0.560*** 38.579 -0.242***
[68.2364] [0.0972] [51.5640] [0.0467]
Depressive Symptoms 11.084** -0.005 10.733** 0.007
[4.9603] [0.0087] [4.9990] [0.0071]
Leg Length 2.798 0.018 -0.329 -0.001
[8.2504] [0.0135] [8.0663] [0.0077]
Hypertension 2.616 -0.008 -73.818 -0.102**
[47.3396] [0.0903] [62.9264] [0.0450]
Overweight -216.762*** -0.254** -196.462** -0.264***
[58.9929] [0.0992] [79.2086] [0.0542]
Undernourished 2.366 -0.119 91.810 -0.010
[75.8824] [0.1214] [94.3827] [0.0995]
Father Alive -23.323 0.195 76.949 -0.021
[74.3237] [0.1773] [67.3182] [0.0845]
Mother Alive -35.561 0.049 -9.246 0.032
[63.5496] [0.1284] [45.7170] [0.0775]
Demographic & Socioeconomic Characteristics
Educ_Primary -92.240 0.067 -73.456 -0.089
[67.5421] [0.1012] [47.9387] [0.0815]
Educ_Secondary -218.770*** -0.201 -225.021*** -0.152
[79.8163] [0.1359] [73.2209] [0.1279]
Married 278.045*** 0.350*** 166.975** 0.304***
[83.0319] [0.1298] [67.4688] [0.0941]
log PCE -106.844*** -0.132** 29.640 -0.134***
[32.7926] [0.0569] [43.7725] [0.0338]
Amount of Land 1.516*** 0.000 1.203*** -0.000**
[0.1751] [0.0003] [0.4375] [0.0001]
County-Level HI Reimbursement Rate -307.388*** -0.639 -106.611 -0.047
[272.2306] [1.2473] [225.3611] [0.2411]
Family Composition
# of Adult Sons 6.367 0.008 15.427 0.021
[30.1767] [0.0528] [26.4347] [0.0338]
# of Non-Adult Sons 96.945 0.312** 111.593 0.228**
[72.9191] [0.1245] [68.8739] [0.0933]
62
# of Adult Daughters -16.419 -0.002 -1.931 0.005
[33.6897] [0.0581] [29.8018] [0.0452]
# of Non-Adult Daughters 41.820 -0.098 -0.602 0.127
[53.9241] [0.1231] [80.7637] [0.0863]
# of Grandchildren 11.162 -0.022 1.500 -0.029
[12.1348] [0.0229] [15.5879] [0.0201]
N 3,036 3,036 3,188 3,188
rho -0.360 -0.360 -0.130 -0.130
F_stat (health) 22.970 101.490 40.060 214.590
P_value (health) 0.006 0.000 0.000 0.000
Note: Adult sons are defined as sons aged 25 and above. Adult daughters are defined as daughters aged 25 and above.
Non-Adult Sons are defined as sons between age 15-25. Non-Adult Daughters are defined as daughters between age
15-25. Age dummies and community fixed effects are included in all the specifications. Rho represents the correla-
tion between participation equation and work hour equation. Robust standard errors are in the parenthesis. All
standard errors are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
63
The presence of a grandchild might also be a concern when the elderly make his labor supply
decisions. If the respondent has a very young grand kid, for instance, less than 6 years old; he
may need to help take care of the grand kid when the kid’s parents are out working. In this case,
we may expect the respondent to spend more time at home, either by reducing their hours of
work or leave the labor force if necessary. However, due to the limitation of the data, we can only
observe the number of grand children in this family, but cannot differentiate whether they are
young enough to be needed taken care of. Results in Table 3.5 show a slight negative association
between the number of grand kids and elderly’ labor participation; however, it is not statistically
significant.
Results (Table 3.6) examining hours of work in farming are similar when further controlling
for household and family compositions. Impacts from health on participation continue to be the
strongest, so as marital status and education. Regarding the family composition, the number of
non-adult sons is persistently associated with higher probability of work participation for both
male and female elderly.
3.6.3 Household Economic Resources
At the beginning of the analysis, it is shown that Chinese rural elderly usually keep working until
they are too old to work. One possible reason is that they cannot afford to retire at an earlier age.
If this is true, then it is normal to expect that individuals with high incomes may tend to work
less or leave the labor force early when they grow old. We try to examine this by including several
household economic measures into our analysis. The most direct way would be measurement of
household income; however, due to the noise in the income measure in developing countries, an
indicator of expenditure has been used instead. Results in Table 3.5 confirm the hypothesis. It
could be noted that for individuals with a higher per capita household expenditure, there is a
higher possibility of labor force withdrawal, for both men and women. No significant effect has
been found on hours of work for them.
Participation in farming-related jobs seems to be quite responsive to PCE. Both men and
women with higher per capita household expenditure are significantly less likely to be involved
in farming. Men tend to reduce their working hours on the farm as well.
64
As for the amount of land, for both men and women, having larger amount of leased land is
associated with more hours of work, which could reflect the demand for labor in this family.
Though the coefficients are small when compared to those of PCE.
The results from Table 3.6 also exhibit a strong association between county-level health in-
surance reimbursement rate and elderly men’s hours of work. It seems that men who live in a
county with a more generous health plan work significantly fewer hours. This might be due to a
relief of economic burden that could be brought by existing/potential inpatient services, which
is a large amount when it occurs.
3.6.4 Interactions between Health and Wealth
In addition to the household per capita expenditure mentioned above, household wealth may
also play a role in rural elderly’s labor supply decisions and may further interact with health. On
the one hand, having more household wealth may facilitate retirement when their health deteri-
orates. On the other hand, more household wealth may mitigate the links between health and
labor supply. For instance, if a rural elderly has enough wealth, then having a disability might
not be such a constraint to him/her. Perhaps he/she could operate a small business which in-
volves very little manual laboring. However, for a rural elderly with little wealth, he/she may
not have other choices but to work on the farm, where having a disability will significantly affect
their ability to work.
To examine whether there exist differentiated effects of health on labor supply for individu-
als with different levels of per capita wealth (PCW), several interactions between health indicators
and demeaned PCW have been constructed. Results including these interactions for men and
women are presented in Table 3.7 and Table 3.8 respectively.
It can been seen that the associations between health indicators and participation decision do
not change much. When it comes to the interactions, the coefficients are neither individually nor
jointly significant if we focus on work participation for men. However, when examining their’
decisions of work hours, it is found that in general wealth facilitates the links between health and
their labor supply. For example, men with an above average PCW significantly spend fewer hours
working when they have depressive symptoms. They are also likely to work fewer hours on the
farm if they have lower leg length but with a higher PCW. Differentiated health effects do exist,
however, as we can see in Table 3.7, they are very small.
65
Table 3.7 Interactions between Health and Wealth: Men
Hours LFP
Hours in
Agriculture
LFP in
Agriculture
Health Measures
Poor GHS -151.738** -0.553*** -22.818 -0.544***
[64.2295] [0.0930] [53.8312] [0.0870]
Disability -73.531 -0.540*** 25.216 -0.539***
[83.9832] [0.1112] [76.1261] [0.1030]
Depressive Symptoms -2.179 -0.005 8.250 -0.007
[5.6775] [0.0090] [5.0321] [0.0087]
Leg Length -4.218 0.027* -3.372 0.023*
[9.9116] [0.0138] [8.3825] [0.0134]
Hypertension -49.307 0.017 13.749 0.013
[59.3436] [0.0923] [49.0022] [0.0907]
Overweight -77.914 -0.269*** -208.605*** -0.258**
[72.2149] [0.0981] [59.6324] [0.1028]
Undernourished 53.052 -0.008 51.366 -0.079
[114.6161] [0.1657] [83.0411] [0.1572]
Father Alive -22.121 0.251 -30.110 0.169
[95.9737] [0.1985] [74.2346] [0.1864]
Mother Alive -21.750 0.073 -31.833 0.030
[71.7068] [0.1284] [63.1408] [0.1275]
Interactions
Poor GHS X PCW -8.163 0.006 -6.566 0.004
[7.3831] [0.0118] [6.6210] [0.0090]
Disability X PCW 4.013 0.021 22.404 0.017
[18.1094] [0.0200] [13.6864] [0.0176]
Depressive Symptoms X PCW -2.181** 0.000 -1.999*** 0.000
[0.9596] [0.0014] [0.7689] [0.0012]
Leg Length X PCW -3.009 0.003 -3.028** 0.002
[1.8844] [0.0027] [1.2748] [0.0024]
Hypertension X PCW -9.050 0.016 8.622 0.016
[9.3765] [0.0148] [6.3216] [0.0116]
Overweight X PCW 12.685 0.008 8.487 0.007
[12.0191] [0.0143] [7.3103] [0.0129]
Undernourished X PCW 36.134 0.051 21.643 0.046
[28.6319] [0.0496] [21.7807] [0.0482]
Father Alive X PCW 24.037 0.030 -0.067 0.034
[18.6859] [0.0358] [12.9259] [0.0264]
Mother Alive X PCW -10.342 -0.043* -12.837 -0.041**
66
[13.7982] [0.0224] [10.1151] [0.0205]
Demographic & Socioeconomic
Characteristics
Educ_Primary -101.822 0.090 -89.141 0.065
[72.9929] [0.1075] [66.7234] [0.1033]
Educ_Secondary -127.601 -0.232* -251.530*** -0.232*
[87.4851] [0.1370] [78.2508] [0.1347]
Married 343.370*** 0.324** 311.736*** 0.351***
[92.7148] [0.1339] [79.9904] [0.1303]
PCW (demeaned) 174.626* -0.165 156.211** -0.126
[96.3463] [0.1339] [66.5926] [0.1207]
Amount of Land -0.613*** 0.000 1.316*** 0.000
[0.1454] [0.0003] [0.1648] [0.0003]
County-Level HI Reimbursement Rate 124.011 -1.173 -303.180*** -0.765
[857.8016] [1.5492] [305.9939] [1.2945]
Family Composition
# of Adult Sons 30.188 0.028 -1.329 0.014
[36.3474] [0.0554] [29.7772] [0.0540]
# of Non-Adult Sons 99.139 0.285** 88.244 0.297**
[65.5218] [0.1263] [72.3839] [0.1242]
# of Adult Daughters -13.509 0.021 -23.708 0.000
[32.0992] [0.0602] [33.2393] [0.0582]
# of Non-Adult Daughters 83.361 -0.105 25.708 -0.140
[64.7671] [0.1303] [54.2168] [0.1233]
# of Grandchildren 5.714 -0.033 13.171 -0.029
[13.1570] [0.0236] [12.1638] [0.0228]
N 2,953 2,953 2,999 2,999
F_stat(health) 11.040 77.870 19.780 91.740
P_value(health) 0.273 0.000 0.019 0.000
F_stat2 (healthXPCW) 17.250 11.030 24.570 11.580
P_value2 (healthXPCW) 0.045 0.274 0.003 0.238
Note: PCW represents household per capita wealth. The value is demeaned in all the specifications. Adult Sons are
defined as sons aged 25 and above. Adult Daughters are defined as daughters aged 25 and above. Non-Adult Sons
are defined as sons between age 15-25. Non-Adult Daughters are defined as daughters between age 15-25. Age dum-
mies and community fixed effects are included. Robust standard errors are in the parenthesis. All standard errors
are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
67
Table 3.8 Interactions between Health and Wealth: Women
Hours LFP
Hours in
Agriculture
LFP in
Agriculture
Health Measures
Poor GHS -46.571 -0.453*** 20.698 -0.451***
[43.4323] [0.0647] [44.5721] [0.0642]
Disability 63.988 -0.236*** 64.670 -0.229***
[64.5886] [0.0534] [61.8243] [0.0528]
Depressive Symptoms 7.953* 0.004 9.200* 0.003
[4.0703] [0.0076] [4.9675] [0.0076]
Leg Length 3.915 -0.000 -3.142 0.000
[9.6182] [0.0081] [8.4221] [0.0083]
Hypertension -92.389 -0.092* -58.112 -0.093*
[71.4257] [0.0475] [65.7936] [0.0478]
Overweight 177.993*** -0.271*** -195.123** -0.265***
[51.4138] [0.0543] [75.7781] [0.0537]
Undernourished 74.940 -0.175* 99.593 -0.168
[149.3172] [0.1000] [151.6522] [0.1026]
Father Alive 16.796 0.021 107.386* 0.015
[61.7706] [0.0907] [63.6253] [0.0904]
Mother Alive 34.036 0.024 -29.684 0.031
[53.2004] [0.0784] [53.0091] [0.0781]
Interactions
Poor GHS X PCW -14.456* -0.007 -14.836** -0.007
[8.6718] [0.0078] [7.2843] [0.0078]
Disability X PCW -0.646 0.017 15.078* 0.017
[9.7169] [0.0112] [8.8206] [0.0112]
Depressive Symptoms X PCW -0.091 -0.002*** -1.401** -0.002***
[0.8694] [0.0006] [0.6796] [0.0006]
Leg Length X PCW -2.714 0.001 -2.320 0.001
[1.8060] [0.0024] [1.6113] [0.0024]
Hypertension X PCW 2.787 0.010 10.617 0.010
[6.6480] [0.0088] [7.9344] [0.0087]
Overweight X PCW 5.631 0.004 -1.069 0.003
[5.9812] [0.0071] [4.6926] [0.0069]
Undernourished X PCW 5.905 -0.064* 9.997 -0.065*
[37.9781] [0.0382] [37.0311] [0.0377]
Father Alive X PCW 12.946 0.025* 25.734** 0.027*
[13.8982] [0.0140] [12.9293] [0.0141]
Mother Alive X PCW -6.071 -0.025** -7.254 -0.025**
[12.5797] [0.0105] [13.9761] [0.0103]
68
Demographic & Socioeconomic
Characteristics
Educ_Primary -37.155 -0.093 -91.074* -0.093
[50.0527] [0.0802] [50.3908] [0.0800]
Educ_Secondary -51.191 -0.169 -236.309*** -0.177
[99.0047] [0.1282] [71.7377] [0.1275]
Married 90.760 0.291*** 161.733** 0.288***
[82.7683] [0.1003] [77.1479] [0.0997]
PCW (demeaned) 126.899 -0.037 110.237 -0.038
[79.4801] [0.1096] [71.1738] [0.1085]
Amount of Land 1.304*** -0.000 1.466*** -0.000
[0.2476] [0.0001] [0.3740] [0.0001]
County-Level HI
Reimbursement Rate -162.819 -0.003 -96.518 -0.007
[227.3370] [0.2384] [227.8300] [0.2391]
Family Composition
# of Adult Sons -8.585 0.027 12.445 0.027
[34.6391] [0.0326] [26.5724] [0.0322]
# of Non-Adult Sons 97.890 0.226*** 108.302 0.228***
[62.6484] [0.0869] [69.4894] [0.0873]
# of Adult Daughters -35.299 0.015 -0.172 0.014
[39.6612] [0.0470] [29.3011] [0.0468]
# of Non-Adult Daughters -32.079 0.121 -5.267 0.121
[76.2007] [0.0882] [81.9444] [0.0856]
# of Grandchildren 4.133 -0.031 -2.846 -0.032
[16.5901] [0.0210] [15.6916] [0.0210]
N 3,119 3,119 3,142 3,142
F_stat(health) 33.590 242.150 32.790 256.010
P_value(health) 0.000 0.000 0.000 0.000
F_stat2 (healthXPCW) 26.800 46.850 45.030 45.200
P_value2 (healthXPCW) 0.002 0.000 0.000 0.000
Note: PCW represents household per capita wealth. The value is demeaned in all the specifications. Adult Sons
are defined as sons aged 25 and above. Adult Daughters are defined as daughters aged 25 and above. Non-Adult
Sons are defined as sons between age 15-25. Non-Adult Daughters are defined as daughters between age 15-25.
Age dummies and community fixed effects are included. Robust standard errors are in the parenthesis. All
standard errors are clustered at the community level. * p<0.1, ** p<0.05, *** p<0.01.
69
Women seem to be more responsive in this case. The coefficients of the interaction terms are
jointly significant when it comes to both of the participation and work hours decision. Specifically,
women with an above average PCW significantly work fewer hours when they reported poor
health status. In addition, these women are more likely to withdraw from the labor force if they
have depressive symptoms, or are undernourished, or if their mothers are still alive.
3.7 Conclusion and Discussion
The main purpose of the paper is to evaluate the role of health in rural elderly’s labor supply
decisions, which has not been explored much in the context of developing countries, especially
China. Thanks to the richness of the CHARLS, this study is able to utilize a broad dimensions of
health measures and examine their differentiated associations with labor force participation and
hours of work.
Unlike the previous studies that either focus on participation or labor market outcomes, we
fit a model to estimate the participation decision and hours of work simultaneously. The findings
suggest that health, in particular GHS and disability, are negatively associated with elderly’s gen-
eral participation as well as participation in agriculture-related activities. Stature (measured as
leg length in the analysis) and overweight problems are also influencing factors for male respond-
ents in general labor force participation decisions. We also find a negative and significant associ-
ation between GHS and annual work hours for both men and women. These associations persist
when all the covariates such as individual’s socioeconomic status, household/family composi-
tions and household economic resources are controlled for. These links, however, are to some
degree, altered by household wealth. It seems that for both men and women, wealth facilitates
their retirement, though the effects are very small.
In addition to the effects from health, the results indicate that education and marital status
are both important in individuals’ labor supply decisions. Specifically, individuals with some
primary level education do not seem to gain an advantage over those who are illiterate. However,
respondents with at least some secondary education significantly work less on the farm or are
more likely to stay out of farming than those with lower educational attainments. A positive and
significant association between being married and labor supply can be found throughout the var-
ious specifications.
70
Another impact that should not be neglected is from family. With more non-adult sons in the
family, both men and women are more likely to stay in the labor force.
One thing that needs to be noticed is that all the associations that have been examined so far
are not causal. The endogeneity of health must be accounted for before any causal relationship
can be established. However, no consensus has been arrived about a valid instrument for health.
Hopefully when more waves of the CHARLS comes out, we will be able to account for this en-
dogeneity problem then.
71
Chapter 4
The Dynamics of Health Changes and Employment
Transitions Among the Elderly
4.1 Introduction
The link between health and labor market behavior in the context of developed countries has long
been established in the literature
67
. Many studies find that health is a major determinant of labor
market outcomes, including of older people; though no consensus has been reached on the mag-
nitude of the health effects. This interrelation between health and labor market behavior might
be different in a developing world. In a less developed country such as China, we expect health
to be a bigger factor, especially rural areas where there is no mandatory retirement age and people
work as long as their health permits.
Benjamin et al. (2003) describes the work pattern of the rural Chinese elderly as “ceaseless
toil”
68
, which means they keep working in old ages until they are not capable of working any
longer. In developed countries such as U.S., a relatively large decline in labor force participation
exists around age 60 (Blau, 1994). However, in the case of rural China, the employment rate is
substantially more continuous
69
. As highlighted in Figure 4.1
70
, employment rate is 80% for men
in their 60s and it is as high as 40% for those in their 80s. Women demonstrate a similar pattern
with a lower participation rate. Most of these men and women are involved in farming, where
strenuous labor is required. For them, labor supply heavily relies on their health status.
67
See Currie and Madrian (1999) for a detailed review.
68
This is a concept borrowed from Deborah Davis-Friedmann’s 1991 book Long lives: Chinese elderly and the
communist revolution.
69
Historically, the work patterns of the elderly in developed countries were similar to that of the current
Chinese rural elderly (See Costa, 1998).
70
Figure 4.1 shows a smoothing nonparametric bivariate relationship between age and employment rate
using the rural resident (hukou) sample with age range 50-85 from the China Health and Retirement Lon-
gitudinal Study (CHARLS) National Baseline.
72
The role of health might be further heightened with the absence of a formal pension system
in rural China. For a long time, old-age support in rural regions has relied on the elderly’s own
labor income and an informal social security arrangement (support from their extended family)
71
Figure 4.1 Probability of Employment (Rural Elderly)
However, a large reduction in family size as a result of fertility control policies started in the 1970s
significantly undermines the support elderly would be able to get from their family when they
grow old
72
. Though a new pension scheme
73
has been initiated since 2009 to formally address the
needs of the rural population, the current pension amount
74
is too low to guarantee a basic stand-
ard of living. Therefore, with the diminishing support from the private safety net and inadequate
71
In urban areas where pension system is more established and pension amount is relatively generous,
employees can count on their pension for old-age support. On the contrary, in the rural regions where
formal pension system is not well established, the support from family, especially working age adult chil-
dren is important. It is even stated in the Constitution that: ...children who have come of age have the duty to
support and assist their parents.
72
This is more of an issue in the future, since a majority of the elderly in current China were not affected
by the fertility control policies.
73
A national rural pension pilot (New Rural Pension Program) was announced in 2009 and started in late
2009 with an aim to achieve full geographic coverage no later than 2013 (State Council 2009b).
74
The basic pension level is 55 yuan (less than $10) per month, which is below the rural poverty line.
73
public pension, labor income still seems to be the major form of old-age support for elderly in
rural areas and the association between health and labor supply might be strengthened in this
context.
This study contributes to the literature by exploring the dynamics of health changes and em-
ployment transitions in rural China, where the link is yet to be established. Using two waves of
data from the China Health and Retirement Longitudinal Study (CHARLS), we model two sets
of labor market transitions: continued employment and labor force re-entry by those who had left
employment. By relating employment transitions with both lagged and contemporaneous health,
we examine how the elderly respond to changes in their health and the magnitude of its impact
on labor supply behaviors.
Our results align with those documented in the literature. Regarding continued employment
decision, a significant decline in participation is observed for both men and women who reported
poor current general health status or disability. Since lagged health is conditioned, this implies
that the deterioration in health may encourages the elderly to transit out of the workforce. Results
for the re-entry decisions further confirms the role of health changes. We show that not just poor
health, but also declines in health, help explain the elderly’s work transitions, no matter whether
it’s into or out of employment. We also find some effects from baseline biomarkers. Specifically,
men who are undernourished are more likely to exit the workforce. It is the same case for women
with overweight problems.
We also find a strong association between family composition and the elderly’s work deci-
sions. Having adult daughters, especially married ones, is significantly associated with reduced
work participation for both genders. In addition, men and women seem to play different roles in
the upbringings of their grandchildren. Our results indicate that elderly women quit jobs to help
take care of the grandchildren, while elderly men re-enter the workforce to better provide for
their grandchildren. Investigation of demographic and socioeconomic characteristics also reveal
some interesting results. While the elderly respond to the larger amount of land by staying in the
workforce, a decline in the work participation is observed for the elderly who lives in an area
with a higher health insurance reimbursement rate.
Simulations based on the model estimates show that, comparing to a “consistently healthy”
scenario, experiencing an adverse health change could lead to a 10-15 percentage points reduction
in continued employment probability. In addition, asymmetry exists between the two dynamic
74
health-employment associations. Unlike the “ratchet” effect that has been observed in other stud-
ies, labor market re-entrants in rural China are much more sensitive to their health changes. Our
simulations show that, compared to a “consistently poor health” case, an improvement in health
could result in a 11-22 percentage points increase in the labor market re-entry rate. Simulations
also reveal the distinctive importance of family composition in the elderly’s transition decisions.
For those with employment, family’s impact is trivial and health is a much more critical concern
when they decide whether or not to stay employed. For those out of employment, family is valued
as much as health in transitioning into the labor market.
The dynamic association also exists if we restrict the sample to farmers, but none exists for
non-agricultural workers. Furthermore, the extent to which health changes are associated with
employment transitions also vary slightly for individuals with different educational attainments.
However, the difference is very small.
The structure of the paper is as follows. Section 4.2 briefly reviews related literature on health
and labor market outcomes. Section 4.3 forms a conceptual framework and empirical specifica-
tions. Section 4.4 describes the dataset and measurements used for analysis. Results are discussed
in Section 4.5, followed by conclusion and discussion in Section 4.6.
4.2 Health and Labor Market Outcomes
The link between health and labor market outcomes such as labor force participation, hours of
work, and family member’s labor supply has been well documented. Most of the existing litera-
ture focus on the static relationship between health and labor market outcomes. Their results
generally suggest that health is an important determinant in an individual’s, in particular an el-
derly’s, labor market behaviors (Anderson and Burkhauser, 1984; Pitt and Rosenzweig, 1986;
Bound, 1991; Dwyer and Mitchell, 1999; Kerkhofs et al., 1999; McGarry 2004; Cai and Kalb, 2006;
Lee and Smith, 2008; Lindeboom and Kerkhofs, 2009).
However, the interrelation between health and labor supply might be better characterized as
a dynamic process
75
. For instance, using two waves of the HRS data, Blau and Gilleskie (2001)
explore this association in a dynamic context by examining the effect of health on the labor force
transitions of older men. They conclude that health is an important determinant of labor force
75
See also Au et al. (2005) and Gannon (2005).
75
behavior and has larger impact than other factors such as economic incentives. Some authors go
beyond to incorporate both current and lagged health status in the employment transition model.
Instead of health levels, they model the change in health. For example, utilizing three waves of
the HRS data, Bound et al. (1999) focus exclusively on individuals employed at wave 2 and specify
four employment alternatives
76
to examine their movements out of the labor market. By including
three waves of health conditions, they are able to distinguish the effects of persistently poor health
from those of health declines. Their results demonstrate that not only poor health, but declines in
health, help explain senior people’s retirement behaviors. Following Bound et al. (1999), Disney
et al. (2006) use the data from British Household Panel Survey and take one step further to exam-
ine the effects of health changes on transitions both in and out of economic activity. Their conclu-
sions align with Bound et al. (1999) in that, adverse health changes predict individual retirement
behavior among workers between age 50 and state pension age
77
. In addition, there is evidence
showing a “ratchet” effect, where compared to health improvement, deterioration in health has a
slightly larger impact on transition probabilities.
However, as with the literature focusing on static analysis, the dynamic studies agree that
health change is important in the elderly’s employment transition but have not reached consensus
on the magnitude of the effect. The differences in estimated health impacts may result from the
distinctive ways used to identify this interrelation, among which the two major issues are the
health measures used and the endogeneity of health.
As mentioned in Currie and Madrian (1999), the concept of health is not directly observable
and is remarkably difficult to measure
78
. A large fraction of literature rely on subjective health
measures, among which the most widely used one is the self-reported general health status (GHS).
Several studies suggest that GHS is a very good indicator of health since it well predicts subse-
quent morbidity and mortality (Thomas and Frankenberg, 2000). However, suspicions are raised
about the self-reported measures, since there is a lack of comparability across individuals without
the existence of an established metric to refer to. In addition, they may suffer from justification
76
The four different labor market statuses are: employed at the same job, employed at a new job, apply for
disability insurance (DI), and retirement (neither employed nor apply for DI).
77
At the time it was 65 for men and 60 for women.
78
It has also long been pointed out by Anderson and Burkhauser (1984) that effects from health on labor
force participation are very sensitive to the health measures used. One of the reasons that health is difficult
to measure is because of its multidimensionality.
76
bias where individuals use bad health as an excuse for exiting the labor force (Anderson and
Burkhauser, 1985; Bazzoli, 1985; Bound, 1991). In this sense, objective health measures
79
that are
less prone to the measurement errors discussed above are favored by some researchers (See Luft,
1975; Bartel and Taubman, 1979).
Intertwined with the concerns of health measures is the endogeneity problem of health. En-
dogeneity may arise if health and work affect each other simultaneously. In addition, the exist-
ence of unobserved characteristics might exacerbate the endogeneity issue by affecting both
health and labor supply.
To tackle the endogeneity problem, most studies adopt an instrumental variable approach in
which they use objective health measures to instrument for subjective ones (Stern, 1989; Kreider,
1996; Dwyer and Mitchell, 1999). In Bound et al. (1999), observed health is treated as a noisy in-
dicator of true health. They utilize all the available exogenous information, including detailed
objective health measures such as functional limitations in all three waves, to instrument for GHS,
so as to construct a time-varying health index. Therefore, with one health index in each wave for
each individual, the authors are able to capture the "health shock" and identify its impact on the
elderly’s employment transitions. This approach has the merit that it directly addresses the en-
dogeneity of health; however, it requires strong identification assumptions, that objective health
measures are exogenous
80
. In addition, since health is multi-dimensional and different aspects of
health are likely to have different effects on individual’s labor supply behaviors, a single health
index may not be adequate to capture all the impacts
81
. This has also been confirmed in Blau and
Gilleskie (2001). Their results indicate that both health limitations and GHS have strong effects
on labor force transitions. Moreover, the fit of the model further improves when more objective
health measures such as difficulties with Activities of Daily Living (ADLs) and specific chronic
health conditions are added to the specification.
Studies regarding the linkage between health and labor market behaviors in developing
world are comparatively limited, partly due to the lack of comprehensive longitudinal household
79
For instance, presence of chronic/acute conditions, functional limitations, disability and utilization of
medical care.
80
This strategy can be useful if self-reported health has measurement error uncorrelated with any meas-
urement error in objective health.
81
Strauss and Thomas (1998) suggest that multiple health measures should be used when examining this
type of linkage between health and labor market outcomes.
77
surveys
82
(See Pitt and Rosenzweig, 1986; Strauss, 1986; Strauss and Thomas, 1994; Strauss and
Thomas, 1998 for reviews of the development literature). As countries differ in various aspects,
particularly levels of economic development, health care services and social security system, the
interrelation between health and employment might be different in developing countries such as
China.
Attempts have been made by several researchers to investigate this relationship in the con-
text of China, mostly for rural areas where there is no mandatory retirement and people work as
long as their health permits. Benjamin et al. (2003) document a “ceaseless toil” working pattern
for the elderly in rural regions using three waves of the China Health and Nutrition Survey
(CHNS). They re-cast the pattern into a labor supply model and focus on the role of age and
deteriorating health
83
. Their results indicate that health only plays a small observable role in ex-
plaining the declining labor supply over the life cycle, though it was critically significant. The role
of health is further confirmed in Pang et al. (2004)
84
. By illustrating the factors that might facilitate
a rural elderly’s work decision, they find that individuals with moderate or severe illness are
more likely to exit the labor force. This association is also observed in Giles et al. (2011) where
they show a very pronounced relationship between health status (as measured by ADLs) with
the elderly’s labor force participation using the CHARLS Pilot Survey.
Unfortunately, the dynamic linkage between health and work remain mostly unexplored in
this context. As social policies in China relating to both work and health are undergoing extensive
82
In contrast to the literature from developed countries, a large amount of evidence from the developing
world are from quasi-experiments. In addition, due to lack of longitudinal survey data, health measures
are usually limited. Unlike studies based on industrialized countries that tend to use self-reported health,
functional limitations, medical conditions and utilization of medical care as the health measure; studies
focusing on developing areas usually chose nutritional status. Though nutritional status such as height and
weight are informative. They only represent one dimension of health. Since health is multi-dimensional, a
broader set of indicators are needed in order to thoroughly understand the linkage between health and
labor market behaviors. However, as more household surveys in developing countries are emerging, it is
currently possible to utilize more comprehensive health measures to examine the link and even to explore
the dynamics between health changes and employment transitions.
83
In their approach, they use physical functioning index to instrument for GHS. The physical functioning
index is constructed from a series of questions from the CHNS. Answers to these questions reflect the dif-
ficulties for specific functions associated with hearing, eyesight, use of arms, legs, etc. A single index is
created in Benjamin et al. (2003) to distill the responses.
84
Pang et al. (2004) use a data set of 1,199 households they collected in 2000 from a randomly selected,
nearly nationally representative sample of 60 villages in 6 provinces.
78
changes
85
, a thorough understanding of the dynamics between these two factors is essential.
Therefore, this study tries to fill in the gap and examine the interrelation between health changes
and employment transitions for the elderly in rural China.
4.3 Model
4.3.1 Conceptual Framework
To explore the interrelation between health dynamics and labor supply, it is useful to start with
a single individual household production model. Assume that each individual solves an inter-
temporal utility maximization problem which involves a discount factor 𝛿 (0 < 𝛿 < 1), consump-
tion 𝐶 , labor supply 𝐿 86
, health 𝐻 ∗
, observable characteristics 𝑋 , environmental factors 𝜂 , as well
as unobserved characteristics 𝜉 :
𝑚𝑎𝑥𝑈 = 𝐸 𝑡 ∑ 𝛿 𝑡 𝑢 𝑡 (𝐶 𝑡 , 1 − 𝐿 , 𝐻 𝑡 ∗
; 𝑋 𝑡 , 𝜂 𝑡 , 𝜉 𝑡 )
𝑇 𝑡 =0
(4.1)
subject to a life-time budget constraint:
(𝑝 0
𝑐 𝐶 0
∗
+ 𝑝 0
𝑛 𝑁 0
) + ∑
𝑝 𝑡 𝑐 𝐶 𝑡 ∗
+𝑝 𝑡 𝑛 𝑁 𝑡 ∏ (1+𝑟 𝑚 )
𝑡 𝑚 =1
= (𝑤 0
𝐿 0
+ 𝑦 0
) + ∑
𝑤 𝑡 𝐿 𝑡 +𝑦 𝑡 ∏ (1+𝑟 𝑚 )
𝑡 𝑚 =1
𝑇 𝑡 =1
𝑇 𝑡 =1
(4.2)
where 𝑟 represents the real rate of interest, 𝑤 is wage
87
, 𝑦 represents non-labor income, 𝑝 𝑐 and 𝑝 𝑛
are prices for non-health related consumption 𝐶 ∗
88
and health inputs 𝑁 respectively.
To incorporate health dynamics into the model, we follow Grossman (1972) and specify a
health production function where health depends on one-period lagged health 𝐻 𝑡 −1
∗
89
, current
health inputs 𝑁 𝑡 , contemporaneous labor supply 𝐿 𝑡 , a set of individual and family characteristics
𝑋 𝑡 , health environment elements 𝜂 𝑡 , and unobserved factors 𝜇 𝑡 .
85
In the last decade, New Rural Cooperative Medical Scheme (NCMS) was initiated and expanded nation-
wide for the rural residents. There have been extensive changes on insurance coverage, funding system,
management system and etc (See Barber and Yao, 2010). Similarly, New Rural Pension Program (NRPP),
implemented in 2009, has been going through considerate changes with regard to its coverage and man-
agement (See Dorfman et al., 2013).
86
Total time endowment is normalized to 1 and leisure is represented by 1 − 𝐿 .
87
Wage function is assumed as follows: 𝑤 𝑡 = 𝑤 (𝐻 𝑡 ∗
, 𝐻 𝑡 −1
∗
, 𝑋 𝑡 , 𝑒 𝑡 ).
88
Total consumption 𝐶 is the sum of non-health related consumption 𝐶 ∗
and health related consumption
𝑁 .
89
Originally, health production equation is specified as 𝐻 𝑡 ∗
= 𝐻 (𝑁 𝑡 , 𝑁 𝑡 −1
, … , 𝑁 0
; 𝐿 𝑡 , 𝐿 𝑡 −1
, … , 𝐿 0
; 𝑋 𝑡 ; 𝜂 𝑡 ; 𝜇 𝑡 ). For
simplicity, we assume that one-period lagged health is sufficient to incorporate the past information from
health inputs and labor supply (Grossman, 1972).
79
𝐻 𝑡 ∗
= 𝐻 (𝐻 𝑡 −1
∗
, 𝑁 𝑡 , 𝐿 𝑡 , 𝑋 𝑡 , 𝜂 𝑡 , 𝜇 𝑡 ) (4.3)
Assume that life-time utility is additively separable between time period and utility function
in each period is quasi-concave and increases in consumption, leisure and health. Therefore, the
optimal dynamic conditional labor supply can be derived as:
𝐿 𝑡 = 𝐿 (𝐻 𝑡 ∗
, 𝐻 𝑡 −1
∗
, 𝑤 (𝐻 𝑡 ∗
, 𝐻 𝑡 −1
∗
, 𝑋 𝑡 , 𝑒 𝑡 ), 𝑋 𝑡 , 𝜂 𝑡 , 𝑦 𝑡 , 𝘀 𝑡 ) (4.4)
4.3.2 Empirical Specification
It would be ideal if all the factors in Equation (4.4) could be considered when evaluating the in-
terrelation between health and labor supply. Unfortunately, since most elderly in rural China
work on their own farms, no wage is observable. In addition, no non-labor income will be directly
controlled for in our specification. For rural elderly, non-labor income mostly consists of transfers
and asset accumulation. The former one is undoubtedly endogenous and we try to control for it
by including the family composition, which is a significant source of transfer. The latter one is
rare. For the elderly working on the farm, land is notably their most important asset. However,
under the current land tenure system, the ownership of the land belongs to the state and the
collective, and farmers only have use rights over the land allocated to them. Therefore, no wealth
accumulation is possible through land in such an arrangement
90
. However, the amount of leased
land is still a non-negligible factor in the elderly’s labor supply decisions and will be included in
our specifications.
In addition, unlike the seniors in developed countries who may choose between a number of
employment alternatives, the exit routes for the elderly in rural China are limited: either work or
not work. We thus denote 𝐿 𝑖𝑡
𝑘 as a binary variable indicating an individual 𝑖 ’s employment status
at time 𝑡 , conditional on being in employment state 𝑘 at time 𝑡 − 1 and 𝐿 𝑖𝑡
𝑘 ∗
as its latent value.
90
Benjamin et al. (2003) mention that if the elderly in China can accumulate land, then they might use it to
directly support themselves, or to encourage inter-generational transfers from their children. However,
under the current land tenure system, this is impossible and it makes the elderly in rural areas particularly
ill-prepared for growing old.
80
𝐿 𝑖𝑡
𝑘 = {
1 (= 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 ), 𝐿 𝑖𝑡
𝑘 ∗
> 0
0 (= 𝑛𝑜𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 ), 𝐿 𝑖𝑡
𝑘 ∗
≤ 0
𝑘 = 0,1 (4.5)
With all the adjustments, following equation (4.4), the latent value 𝐿 𝑖𝑡
𝑘 ∗
is assumed to be approxi-
mated as a function of current health 𝐻 𝑖𝑡
∗
, lagged health 𝐻 𝑖 ,𝑡 −1
∗
, a set of individual demographic
characteristics, socioeconomic status, and family structures 𝑋 𝑖 91
., age function 𝑓 (𝑎𝑔𝑒
𝑖 ), province
fixed effects
92
𝜂 𝑚 and other unobserved factors 𝘀 𝑖𝑡
93
.
𝐿 𝑖𝑡
𝑘 ∗
= 𝐻 𝑖𝑡
∗
𝛽 0
𝑘 + 𝐻 𝑖 ,𝑡 −1
∗
𝛽 1
𝑘 + 𝑋 𝑖 𝛽 2
𝑘 + 𝑓 𝑘 (𝑎𝑔𝑒
𝑖 ) + 𝜂 𝑚 𝑘 + 𝘀 𝑖𝑡
𝑘 𝑘 = 0,1 (4.6)
For the simplicity of interpretation, this transition equation can also be written in the form of
baseline health and health changes: 𝐿 𝑖𝑡
𝑘 ∗
= (𝐻 𝑖𝑡
∗
−𝐻 𝑖 ,𝑡 −1
∗
)𝛽 0
𝑘 + 𝐻 𝑖 ,𝑡 −1
∗
(𝛽 1
𝑘 +𝛽 0
𝑘 ) + 𝑋 𝑖 𝛽 2
𝑘 + 𝑓 𝑘 (𝑎𝑔𝑒
𝑖 ) +
𝜂 𝑚 𝑘 + 𝘀 𝑖𝑡
𝑘 . In this case, the original coefficient of the current health indicator 𝛽 0
𝑘 is now the coeffi-
cient of the health change in our new specification, while the summation of the coefficients of
both current and lagged health in the original equation 𝛽 1
𝑘 +𝛽 0
𝑘 is the lagged health coefficient in
the new specification.
With the simplified employment alternatives, we thus focus on two sets of employment tran-
sitions. For individuals who were employed in the previous period, we examine the probability
91
Since most of these characteristics are time-invariant (individual characteristics such as gender, educa-
tion) and that time-variant characteristics (individual characteristics such as marital status, family compo-
sition and the district they live in) rarely change over the two-year period of time, we use baseline values
for all these variables.
92
The reason that province dummies are used instead of community/county dummies is that, an extremely
high proportion of elderly in rural regions are still working. If we control for the community/county fixed
effect, a large number of communities/counties will be dropped due to collinearity because all the respond-
ents in these communities/counties are employed. However, we do test a specification controlling for
county-level characteristics such as distance to the nearest health facility, distance to the nearest bus station,
and per capita income. Results are similar (not presented).
93
The probability of transiting from employment state 𝑘 in period 𝑡 − 1 to 𝑗 in period 𝑡 , conditional on 𝐻
and 𝑋 , is given by 𝑃 (𝐿 𝑖𝑡
𝑘 ∗
> 0) = ∫ 𝜙 (𝑣 )𝑑𝑣
𝐻 𝑖𝑡
∗
𝛽 0
𝑘 +𝐻 𝑖 ,𝑡 −1
∗
𝛽 1
𝑘 +𝑋 𝑖 𝛽 2
𝑘 +𝑓 𝑘 (𝑎𝑔𝑒
𝑖 )+𝜂 𝑚 𝑘 +𝜀 𝑖𝑡
𝑘 −∞
, where 𝜙 (𝑣 ) is the standard nor-
mal density 𝜙 (𝑧 ) = (2𝜋 )
−1/2
exp (−𝑧 2
/2).
81
of staying employed in the next wave. For individuals who were not employed in the prior period,
we examine the probability of re-entering the labor market
94
.
It should be noted that for either set of the employment transitions, the model is specified for
a selected sample. Suppose a person with poor health works in the previous wave and then con-
tinues to work in the next period, it could be that he/she has a preference for work regardless of
his/her health condition, or perhaps he/she cannot afford retirement, thus has no choice but stay
working in poor health. Failure to account for this potential self-selection process might attenuate
the effect of health changes on employment transitions. To tackle this problem, we jointly estimate
a reduced form equation modeling the working status of the individual in the previous wave,
allowing for correlation of the error terms between the transition and previous wave employment
equations. Let 𝐿 𝑖 ,𝑡 −1
denotes an individual’s employment status in period 𝑡 − 1 and 𝐿 𝑖 ,𝑡 −1
∗
as its
latent value.
𝐿 𝑖 ,𝑡 −1
= {
1 (= 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 ), 𝐿 𝑖 ,𝑡 −1
∗
> 0
0 (= 𝑛𝑜𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 ), 𝐿 𝑖 ,𝑡 −1
∗
≤ 0
𝑓𝑜𝑟 𝑐𝑜𝑛𝑡𝑖𝑛𝑢𝑒𝑑 𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 (4.7)
𝐿 𝑖 ,𝑡 −1
= {
1 (= non𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 ), 𝐿 𝑖 ,𝑡 −1
∗
≤ 0
0 (= 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 ), 𝐿 𝑖 ,𝑡 −1
∗
> 0
𝑓𝑜𝑟 𝑙𝑎𝑏𝑜 𝑟 𝑚𝑎𝑟𝑘𝑒𝑡 𝑟𝑒 − 𝑒𝑛𝑡𝑟𝑦 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 (4.8)
The latent value 𝐿 𝑖 ,𝑡 −1
∗
is determined by the same decision process as in the employment transi-
tion model. The equation is specified as follows
95
:
𝐿 𝑖 ,𝑡 −1
∗
= 𝐻 𝑖 ,𝑡 −1
∗
𝛼 1
+ 𝑋 𝑖 𝛼 2
+ 𝑓 (𝑎𝑔𝑒
𝑖 ) + 𝜂 𝑚 + 𝜖 𝑖 ,𝑡 −1
(4.9)
Our approach resembles Bound et al. (1999) in that we also examine the interrelation between
health and work decision in a dynamic context. By including both contemporaneous and lagged
health, we try to examine the impact of health changes on an individual’s transitions, both out of
and into employment. However, instead of using a continuous health index, we include a broad
94
We differentiate two different state transitions assuming that the factors we consider may have distinc-
tive effects on these two processes. However, we also estimate a fixed effects model which takes account of
individual fixed heterogeneity but assumes a similar process. Results show that for individuals who transit
between labor market states, there is a significant link between health changes and changes in employment
status.
95
𝐻 𝑖 ,𝑡 −2
∗
is omitted because currently we only have two waves of data for analysis.
82
set of health measures to account for multi-dimensionality. Since suitable instruments are difficult
to obtain and inappropriate instrument variables may lead to a substantial bias (Wooldridge,
2010), we do not attempt any IV approach. Certainly we are aware that the random measurement
errors in these health indicators will attenuate the estimated health effects; however, the inherent
endogeneity in health measures tend to exaggerate the coefficients. Thus the biases are in oppo-
site direction (Bound, 1991). However, one needs to be cautious in interpreting the estimated ef-
fects from each health change. With regard to the potential unobserved factors, various individual
and family characteristics have been controlled for. Moreover, province fixed effects are incorpo-
rated to account for factors such as provincial level price, health-related infrastructure and envi-
ronment. These aspects may simultaneously affect residents’ health and their employment deci-
sions.
Full information maximum likelihood (FIML) method is used to jointly estimate the equa-
tions. It is assumed that the error terms from the employment transition equation and the initial
employment status equation follow a standard bivariate normal distribution with a correlation
coefficient 𝜌 . Identification of the coefficients and the correlation 𝜌 relies on the non-linearity of
the model
96
.
As to this point, we focus on the effects from health and health changes on work participation
and employment transitions. To further explore the health effects on labor supply intensity, we
estimate another specification on hours of work. To illustrate the model, changes in working
hours from period 𝑡 − 1 to period 𝑡 , △ 𝑌 𝑡 is assumed to be a linear function of contemporaneous
health, lagged health, individual and family characteristics, age function, province fixed effects
and an unobserved error term:
△ 𝑌 𝑡 = 𝐻 𝑖𝑡
∗
𝛾 0
+ 𝐻 𝑖 ,𝑡 −1
∗
𝛾 1
+ 𝑋 𝑖 𝛾 2
+ 𝑓 (𝑎𝑔𝑒
𝑖 ) + 𝜂 𝑚 + 𝜉 𝑖𝑡
(4.10)
Since the differences in hours between periods takes into account the selection in the prior wave,
the estimation procedure employs OLS regression with a full non-selected sample, regardless of
an individual’s initial work status.
96
We also estimate the models utilizing both the non-linearity property and exclusion restrictions such as
number of male siblings, number of female siblings, father’s education and mother’s education. Results are
similar.
83
4.4 Data and Management
4.4.1 Data
4.4.1.1 China Health and Retirement Longitudinal Study
(CHARLS)
The data used for analysis comes from the first two waves of the Chinese Health and Retirement
Longitudinal Study (CHARLS). Details of the datasets have been discussed in Chapter 2 Section
2.4.1. The first wave, CHARLS national baseline, was fielded between June 2011 and March 2012.
It contains 17,705 individuals in 10,257 households, among which more than 70% of the respond-
ents are rural (rural hukou
97
). The second wave was fielded in 2013. It contains 18,648 individuals,
among which 15,684 are respondents from the 2011 baseline sample.
The study consists of an individual survey, a household survey and a community survey. It
includes detailed information about individuals’ work and health information, which allows for
an in-depth analysis of respondents’ labor supply behaviors.
4.4.1.2 Sample
Since we focus on the mid-aged and elderly in rural China and intend to examine how changes
in health affect their employment transitions, the analysis sample is restricted to respondents with
rural hukou registration, who were 50 to 90 at the baseline survey and were interviewed in both
waves. This leaves us 9,000 observations (See Table 4.1).
97
Hukou status is classified according to an individual’s registration status. If his/her registration status is
non-agriculture, then he/she is classified as urban residents; if his/her registration status is agriculture,
then he/she is classified as rural residents.
84
Table 4.1 Summary Statistics
Panel A Whole Sample Men Women
wv1: % Employed 0.74 0.82 0.67
wv1: Annual Work Hours (conditional) 1728.98 1877.66 1556.41
wv2: % Employed 0.70 0.77 0.63
wv2: Annual Work Hours (conditional) 1594.51 1741.62 1427.09
Panel B Whole Sample Men Women
Health
Measures
wv1: poor GHS 0.32 0.27 0.36
wv1: Disability 0.32 0.26 0.38
wv1: Depressive Symptoms 0.37 0.29 0.44
wv1: Limb length (cm) 47.77 49.64 46.03
wv1: Hypertension 0.32 0.31 0.34
wv1: Overweight 0.26 0.19 0.33
wv1: Undernourished 0.09 0.08 0.09
wv1: Father is Alive 0.09 0.09 0.10
wv1: Mother is Alive 0.19 0.18 0.20
Demographic &
Socioeconomic
Characteristics
wv1: Age 63.95 63.91 63.99
wv1: Educ_Illiterate 0.37 0.18 0.56
wv1: Educ_Primary 0.44 0.53 0.34
wv1: Educ_Secondary 0.19 0.29 0.10
wv1: Married 0.87 0.91 0.82
wv1: log PCE 8.27 8.29 8.25
wv1: Amount of Land (mu) 13.87 14.12 13.63
wv1: County-Level Health Insurance
Reimbursement Rate
0.25 0.25 0.25
Family
Composition
wv1: # of Adult Sons (>25 yrs) 1.44 1.33 1.53
wv1: # of Non-Adult Sons (15-25 yrs) 0.14 0.16 0.11
wv1: # of Adult Daughters (>25 yrs) 1.25 1.16 1.34
wv1: # of Non-Adult Daughters (15-25 yrs) 0.12 0.14 0.10
wv1: Whether has grandchild 0.83 0.06 0.05
Note: Summary statistics at the baseline. Age range is 50-90.
85
4.4.2 Measurement of Labor Market Behavior
Employment Transition: Before measuring an individual’s employment transition, we first need
to define their employment status in each wave. Individuals are considered as employed if they
were still working at the time of the survey. Work is defined as including: (1) Engaging in agri-
cultural work (including farming, forestry, fishing, and husbandry for own family or others) for
more than 10 days in the past year; (2) Earning a wage, running a business or participating in
unpaid family business for at least one hour last week (own housework and voluntary work are
not included); (3) Having a job but being temporarily laid-off, or on sick leave, or in job training,
but expecting to go back to this job at a definite time in the future or within 6 months. Respond-
ents who reported not working and planned to stay out of the labor force are considered as not
working
98
. It can be seen from Table 4.1 that overall, 74% of the mid-aged and elderly in our
sample are characterized as employed at the baseline, among which men has a higher participa-
tion rate (82%) than women (67%). The percentage of respondents who were employed in the
second wave is slightly lower (70%) with the same pattern that men has a relatively higher par-
ticipation rate (77%) than women (63%). We measure two sets of employment transitions (See
Table 4.2 for transition matrices). For respondents employed at the baseline, we model their tran-
sition as either stay employed or become non-employed in wave 2. Similarly, for respondents not
employed at the baseline, their employment transitions are modeled as either stay non-employed
or enter employment in the second wave.
Change in Annual Work Hours: To measure the intensity of mid-aged and elderly’s labor sup-
ply, annual work hours are constructed for respondents who were employed
99
. The annual work
hours
100
are calculated as the summation of all the work hours, including in main jobs and side
98
Individuals who were not working when the survey took place but were searching for new jobs during
the past month are considered as not employed (only several individuals fall into this category). Those who
have never worked during their lifetime are not included in the analysis.
99
Basically there are in total five categories of jobs: farming, employed farming, self-employed (non-farm-
ing), employed (non-farming), and family business helpers. Most of the respondents hold one job; however,
a small portion of them hold a side job other than the main one. They can mostly often be characterized as
taking farming as a main job, while doing some small business or employment work as a side job.
100
Respondents were asked “How many months did you work on [...] in the past year?”, “How many days did you
work in [...] per week on average during a normal work month in the past year?”, and “How many hours did you
usually work in [...] per day during a normal work day in the past year?”. We construct the annual work hours
for each job and then add them together if he/she has more than one job.
86
jobs. Mean of the annual work hours for those who were employed at the baseline is around 1,728.
Similar to the declining pattern of employment rate, the conditional annual work hours in wave
2, as well, is lower than that in the baseline
101
.
Table 4.2 Employment Transitions Matrices
Men wv2: Employed wv2: Nonemployed
wv1: Employed 71.49% 10.11%
wv1: Nonemployed 5.05% 13.35%
Women wv2: Employed wv2: Nonemployed
wv1: Employed 55.45% 12.02%
wv1: Nonemployed 7.89% 24.64%
Note: Age range is 50-90. Sample is limited to respondents with non-missing work status information at wave 1
and wave 2.
4.4.3 Measurement of Health
As mentioned in the previous chapter (Section 3.4.3), the concept of health is multi-dimensional
and hard to measure. Due to this multi-dimensionality, Strauss and Thomas (1998) argue that it
is useful to examine several indicators simultaneously. Therefore, similarly in this chapter, we
include five dimensions of health to investigate how changes in these different health aspects
affect elderly’s employment transition dynamics.
General Health Status (GHS): As has been explained in the previous chapters, in CHARLS, GHS
is rated on a scale of 5: 1=very good, 2=good, 3=fair, 4= poor, 5=very poor. A binary measure is
created as =1 if reported poor or very poor health; =0 if reported fair, good or very good health.
About 1/3 of the sample reported poor or very poor health status at the baseline. The fraction of
women reporting poor health is larger than men.
Disability: Similarly, the disability measure is defined as the presence of any impairment in any
of the ADLs/IADLs
102
. Respondents are defined as disabled if they cannot do at least one of those
101
Partly because of aging by 2 years.
102
The definitions of ADL/IADLs have been depicted in the previous chapters.
87
tasks or if they have great difficulty and need help from others in performing them. The preva-
lence rate for disability is around 32% in the sample. This percentage is higher for women than
for men.
Depressive Symptoms: CES-D is used as an indicator of individual’s depressive symptoms. As
explained in the previous chapters, we constructed a CES-D score that ranges from 0 to 30 with
higher scores representing higher levels of depressive symptoms and use 10 and above as a cutoff
point for high depressive symptoms (Andresen et al., 1994). More than 30% of the sample is above
this threshold. As with the other health measures, women seem to be at disadvantage with nearly
40% have high depressive symptoms, while about 20% of men have scores equal or over 10.
Biomarkers: We include four biomarker indicators: lower leg length, hypertension, overweight
and undernourished, which has been defined in the previous chapter Section 3.4.3.
Parental Longevity: Similarly to the previous chapter, two variables are constructed: whether the
respondent has a living biological father and whether the respondent has a living biological
mother. As indicated in the previous chapter, parental longevity may represents genetic healthi-
ness or good health behaviors. The sample shows that among the parents of the elderly respond-
ents, mothers are much more likely to be alive at the time of the survey. This is partly because
mothers tend to be younger than fathers and partly because female old age mortality is lower
than men’s. The possibility of having a living father or mother is similar for both men and women.
In the model, health changes are mainly captured by changes in GHS, disability and depres-
sive symptoms between two waves. Since biomarkers is only available from the baseline survey
and there is little variation in the two parental longevity variables, we construct them only utiliz-
ing the information from the baseline.
4.4.4 Covariates
We include a set of individual and household characteristics, including age, education, marital
status, per capital household expenditure (PCE), the amount of leased land and the median
county-level health insurance reimbursement rate. One-year interval age dummies
103
are used to
103
The upper group contains respondents aged 80 or over.
88
better capture its non-linear effect. Since rural elderly on average have low educational attain-
ments, three education groups are constructed: illiterate (reference group), have some primary
education, and have some secondary education. As one can see, more than 1/3 of the sample are
illiterate and only 19% of the respondents have ever received secondary education. One thing to
notice is the large gender difference, where more than half of women are illiterate while the rate
for men is 18%. Marital status indicator is a binary variable which takes value 1 if married or
cohabited, 0 otherwise
104
.
As indicated in the literature, economic variables also have a non-negligible effect on el-
derly’s labor supply behaviors. Individuals with higher income may transit out of the labor force
at a relatively earlier age, but still has the financial ability to support themselves. However, indi-
viduals with low income may have to work as long as they can for old-age support. In this sense,
the economic condition may play an even more critical role in rural China, where formal pension
is almost absent and old-age support heavily relies on labor income. Since in developing countries,
expenditure can be measured with less noise and provides a better measure of welfare than in-
come, we include the baseline per capita household expenditure (PCE) as an indicator for eco-
nomic conditions (Strauss and Thomas, 1998). As the majority of the elderly in rural China are
involved in farming, we also control for the amount of land leased by each household
105
. Though
farmers are only entitled to the use rights of lands and therefore not able to accumulate them, the
amount of land leased is still a critical factor in the elderly’s labor supply decisions. Since lands
that are not being actively cultivated might be subject to redistribution, rural elderly with a larger
amount of leased land has the incentive to stay working. Summary statistics show that on average
rural residents in our sample have 13.87 mu of land
106
.
In examining the interrelation between health changes and employment transitions, we
might expect some effects from health insurance. However, in rural China, most of the residents
are covered by the New Cooperative Medical Scheme (NCMS) and there is rarely any variation
in their health insurance status
107
. Therefore, we try to capture the potential impact by controlling
104
The counterpart includes never married, separated, divorced and widowed.
105
One advantage of using the land measure is that the amount of leased land is exogenous. Though Giles
and Mu (2014) mention that the amount of land may affect people’s propensity to migrate.
106
mu is the unit of land area. One mu equals to about 0.16 acre.
107
In our sample, 95% of the respondents are covered by NCMS.
89
for county-level health insurance reimbursement rate
108
at the baseline. Summary statistics show
that the average reimbursement rate for inpatient care is 25%.
Family Composition: In addition to individual and household characteristics, family members
may affect the elderly’s employment status and subsequent transition. Therefore, a set of family
composition are included in the analysis as well. These family composition include: Number of
Adult Sons
109
, Number of Adult Daughters, Number of Non-adult Sons and Number of Non-adult Daugh-
ters. Grandchildren might also be a concern for some seniors in the sample when they decide
whether or not to spend more time taking care of their grandchildren. It would be ideal if we can
differentiate grandchildren that are of the age that must be taken care of (for example, under 6
years old) and grandchildren that are older. However, due to the limited information, we only
include a variable indicating whether they have a grandchild in the family.
4.5 Results
4.5.1 Health Changes and Employment Transitions
Estimation results from the dynamic interrelation between health changes and employment tran-
sitions for individuals employed at the baseline are shown in Table 4.3 (Regression results for
initial employment status are in Table A4.1). Results for men and women are separately estimated
since their responses might be different
110
. A significant decline in work participation is observed
for both genders who report poor contemporaneous health or disability in the latest wave when
lagged health is conditioned. This lends support to the hypothesis that the deterioration in health
matters in one’s labor force transitions. Even though, conditional on the contemporaneous health,
lagged health does not seem explain the elderly’s work transitions. Lagged health is important
when we examine its relationship with work transition in a “change equation”, where we re-write
the original transition equation in the form of lagged health and health changes
111
. As mentioned
108
Since NCMS usually does not cover the cost of outpatient visits, we only include health insurance reim-
bursement rate for inpatient visits.
109
For this set of measures, we only account for children that are alive at the time of the survey. Adult
children are those over 25 years old. Non-adult children are those between age 15-25.
110
As can be seen from the summary statistics table, health and employment patterns for men and women
differ.
111
𝐿 𝑖𝑡
𝑘 ∗
= (𝐻 𝑖𝑡
∗
−𝐻 𝑖 ,𝑡 −1
∗
)𝛽 0
𝑘 + 𝐻 𝑖 ,𝑡 −1
∗
(𝛽 1
𝑘 +𝛽 0
𝑘 ) + 𝑋 𝑖 𝛽 2
𝑘 + 𝑓 𝑘 (𝑎𝑔𝑒 𝑖 ) + 𝜂 𝑚 𝑘 + 𝘀 𝑖𝑡
𝑘
90
in the previous section, in the “change equation”, the coefficient of the lagged health indicators
are in fact 𝛽 1
𝑘 + 𝛽 0
𝑘 from the original transition equation. According to Table 4.3, for both men and
women, coefficients of the lagged GHS and lagged disability are negative. In addition, our test
results of the summation of the coefficients of the lagged and current GHS/disability are signifi-
cant
112
. It confirms that not only baseline health, but also health changes, are significantly associ-
ated the elderly’s employment decision as whether to stay employed or exit the workforce. As
for the other baseline health indicators, usually we expect individuals with higher stature (longer
leg length in this case) will be more likely to stay employed. However, for rural elderly women,
it is the opposite. One possibility might be that these women have accumulated enough wealth
through laboring in the past, and therefore tend to “retire” earlier. Or perhaps these women mar-
ried into higher income households and thus do not have to work at old ages. Regarding the other
biomarkers, men who are undernourished are more likely to exit the workforce; as for women
with overweight problems. Hypertension, with a relative high prevalence among this group,
however, does not seem to be related to the elderly’s labor market transitions. As mentioned in
the previous section, having a living parent at old ages might be an indicator of genetic healthi-
ness and of good health behaviors. On the other hand, the presence of living parents may also
affect an individual’s employment behavior through the family support mechanism. The support
provided to the parents can be in the form of money or time. For men, having a living father is
positively correlated with continued employment, which could be a combined impact of inter-
generational transmission of health and necessary financial support for their parent, though the
effects are not significant. For women, the impacts of having a living father and mother are
slightly different. Similar to men, the presence of a living father is positively associated with con-
tinued employment. However, evidence from having a living mother indicates that daughter’s
role of a care-taker may have dominated the other effects.
One thing to notice is that the correlation between the error terms of the employment transi-
tion equation and the initial working status equation is around 0.6 and 0.2 for men and women
respectively. This may indicates that there could be unobserved factors such as work preference
which lead elderly to be employed in the first place and stay working in the subsequent periods.
112
Take the second specification as an example, if the transition equation is translated into the form with
baseline health and health changes, the coefficient of the lagged GHS is -0.586 for men and -0.410 for
women. Similarly, the coefficient of the lagged disability is -0.379 for men and -0.210 for women. P-values
of the significance tests are exhibited at the bottom of Table 4.3.
91
However, Wald tests of the independence of baseline participation equation and transition equa-
tion fail to reject the hypothesis, which means there might not be selection into employment at
the beginning. Simple Probit estimations based on selected sample (individuals employed at the
baseline) are presented in the appendix (Table A4.6). Results are similar to FIML estimations that
take account of the baseline selection.
In addition to evaluating the interplay between health changes and employment withdrawal,
we also examine elderly’s labor market re-entry decision for those who were not employed at the
baseline. Results are presented in Table 4.4. Similarly a decline in labor force re-entry is observed
for men and women who report poor contemporaneous health or/and those with disability. No
significant correlation is found between lagged health variables and re-entry decisions when cur-
rent health is controlled for. However, tests of the summations of lagged GHS and current GHS
for both men and women are significant, which further confirms that not only lagged health, but
also the changes in health, that associate with the elderly’s employment transition, no matter the
decision is to transit into or out of the labor market.
If the hypothesis about the positive correlation between the error terms of the initial employ-
ment status and the subsequent employment holds, then one should expect to see negative cor-
relation between the error terms in the case of labor market re-entry decision. It is indeed negative
for women; however, it is not significant. For elderly men, the correlation between the error terms
is positive and significant
113
. One possible explanation for the positive correlation between non-
participation at the baseline and labor market re-entry in the following wave is that, the previous
exits are mostly temporary. Therefore, strong motivations may exist for individuals to resume
their work, especially in a case where the previous withdrawal was due to health issue and health
has restored in the subsequent period of time.
113
Simple Probit estimation based on individuals non-employed at the baseline are in the appendix (Table
A4.7).
92
Table 4.3 Continued Employment Decision among Respondents Employed at the Baseline
Employment Transition (=1 Stay Employment; =0 Non-employment )
Men Women
Health Changes
w1: poor GHS -0.187 -0.182 -0.199 -0.069 -0.066 -0.054
[0.1164] [0.1285] [0.1358] [0.0763] [0.0724] [0.0733]
w2: poor GHS -0.393*** -0.404*** -0.401*** -0.341*** -0.344*** -0.349***
[0.0930] [0.1020] [0.1083] [0.0591] [0.0613] [0.0615]
w1: Disability -0.133 -0.115 -0.129 -0.061 -0.063 -0.056
[0.1278] [0.1327] [0.1398] [0.0895] [0.0923] [0.0924]
w2: Disability -0.266*** -0.264*** -0.259*** -0.149* -0.147* -0.151*
[0.0940] [0.0955] [0.0995] [0.0776] [0.0798] [0.0792]
w1: CESD>=10 0.190** 0.197** 0.176** 0.122 0.117 0.122
[0.0902] [0.0904] [0.0884] [0.0972] [0.0993] [0.0977]
w2: CESD>=10 -0.088 -0.087 -0.095 0.046 0.057 0.060
[0.0855] [0.0843] [0.0806] [0.0777] [0.0793] [0.0809]
Baseline Health Indicators
w1: Limb Length -0.007 -0.007 -0.009 -0.019 -0.019 -0.020
[0.0120] [0.0123] [0.0123] [0.0126] [0.0130] [0.0131]
w1: Hypertension -0.067 -0.069 -0.071 -0.095 -0.103 -0.104
[0.0665] [0.0654] [0.0598] [0.0886] [0.0908] [0.0914]
w1: Overweight -0.101 -0.083 -0.085 -0.168** -0.168** -0.169***
[0.1125] [0.1208] [0.1203] [0.0677] [0.0671] [0.0650]
w1: Undernourished -0.319*** -0.327*** -0.337*** 0.121 0.120 0.113
[0.1055] [0.1064] [0.1055] [0.1326] [0.1319] [0.1324]
w1: Father Alive 0.130 0.144 0.149 0.446*** 0.445*** 0.446***
[0.1650] [0.1678] [0.1689] [0.1588] [0.1592] [0.1625]
w1: Mother Alive 0.000 -0.009 -0.000 -0.186*** -0.189*** -0.185***
[0.0990] [0.0971] [0.1003] [0.0692] [0.0688] [0.0715]
Socioeconomic Characteristics
Educ_Primary 0.050 0.059 0.050 -0.019 -0.033 -0.031
[0.1121] [0.1209] [0.1219] [0.0901] [0.0898] [0.0918]
Educ_Secondary -0.138 -0.122 -0.147 0.254* 0.217 0.213
[0.1360] [0.1440] [0.1377] [0.1483] [0.1568] [0.1538]
Married 0.276*** 0.286*** 0.286*** 0.216** 0.224** 0.215**
[0.1047] [0.1048] [0.1056] [0.1060] [0.1064] [0.1058]
logPCE -0.127*** -0.125*** -0.129*** -0.007 -0.015 -0.013
[0.0446] [0.0444] [0.0444] [0.0765] [0.0770] [0.0754]
Land Leased 0.001*** 0.001*** 0.001*** 0.001* 0.001* 0.001*
[0.0003] [0.0003] [0.0003] [0.0006] [0.0006] [0.0006]
HI Reimbursement Rate -0.370* -0.382* -0.387* -0.582*** -0.608*** -0.604***
[0.1977] [0.1976] [0.2056] [0.1932] [0.1870] [0.1899]
93
Family Composition
# of Adult Sons 0.004 -0.058
[0.0472] [0.0436]
# of Non-adult Sons -0.002 0.004
[0.1033] [0.0993]
# of Adult Daughters -0.076** -0.048*
[0.0382] [0.0289]
# of Non-adult Daughters 0.111 0.206
[0.1357] [0.1286]
# of Married Adult Sons -0.012 -0.042
[0.0527] [0.0463]
# of Married Non-adult Sons -0.176 -0.101
[0.1366] [0.1474]
# of Married Adult Daughters -0.082** -0.053*
[0.0370] [0.0287]
# of Married Non-adult Daughters -0.026 0.252
[0.1823] [0.2092]
# of Non-married Adult Sons 0.017 -0.129**
[0.0844] [0.0530]
# of Non-married Non-adult Sons 0.068 0.040
[0.1132] [0.1026]
# of Non-married Adult Daughters 0.011 0.108
[0.1070] [0.1177]
# of Non-married Non-adult 0.211 0.174
Daughters [0.1440] [0.1549]
Whether has Grandchild -0.166 -0.088 -0.176** -0.180**
[0.1141] [0.1266] [0.0755] [0.0790]
N 2,789 2,789 2,789 2,983 2,983 2,983
Mean 0.900 0.900 0.900 0.840 0.840 0.840
rho 0.600 0.570 0.590 0.160 0.190 0.140
P_value: rho 0.187 0.258 0.273 0.608 0.531 0.631
P_value: All Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: Lagged Health 0.003 0.003 0.001 0.000 0.000 0.000
P_value: Current Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: poor GHS ß1+ß2=0 0.000 0.000 0.000 0.000 0.000 0.000
P_value: Disability ß1+ß2=0 0.000 0.000 0.000 0.005 0.008 0.014
P_value: CESD>=10 ß1+ß2=0 0.212 0.164 0.307 0.069 0.065 0.048
Note: FIML estimation of employment transition for individuals employed at the baseline. Sample is limited to rural residents
(hukou) with age range [50-90] at the baseline. Other covariates include age dummies, provincial fixed effects and urban/rural
residency. Standard errors are in the parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, **
p<0.05, * p<0.1.
94
Table 4.4 Labor Market Re-entry Decision among Respondents Not Employed at the Baseline
Employment Transition (=1 Re-enter Employment; =0 Stay Non-employ-
ment )
Men Women
Health Changes
w1: poor GHS -0.313 -0.264 -0.257 -0.147 -0.074 -0.064
[0.2086] [0.2145] [0.2067] [0.3635] [0.3872] [1.6087]
w2: poor GHS -0.624*** -0.666*** -0.644*** -0.157 -0.167 -0.173
[0.2332] [0.2366] [0.2270] [0.1427] [0.1409] [0.1617]
w1: Disability -0.083 -0.096 -0.118 -0.134 -0.128 -0.092
[0.2391] [0.2125] [0.2071] [0.3073] [0.3628] [1.1856]
w2: Disability -0.481** -0.418* -0.391* -0.466* -0.496** -0.510
[0.2232] [0.2217] [0.2268] [0.2450] [0.2217] [0.3706]
w1: CESD>=10 -0.302 -0.277 -0.258 -0.170 -0.192 -0.215
[0.1975] [0.1807] [0.1817] [0.2097] [0.1873] [0.3620]
w2: CESD>=10 0.092 0.062 0.070 0.254* 0.278* 0.226
[0.2332] [0.2100] [0.2073] [0.1543] [0.1611] [0.2126]
Baseline Health Indicators
w1: Limb Length -0.054** -0.059** -0.060** 0.040* 0.041 0.046
[0.0255] [0.0271] [0.0261] [0.0233] [0.0274] [0.0577]
w1: Hypertension -0.129 -0.142 -0.136 0.115 0.145 0.153
[0.1728] [0.1669] [0.1684] [0.1295] [0.1377] [0.3089]
w1: Overweight -0.438** -0.349* -0.366** -0.334 -0.329 -0.330
[0.1761] [0.1877] [0.1772] [0.2460] [0.2726] [0.9920]
w1: Undernourished 0.224 0.172 0.097 -0.149 -0.153 -0.215
[0.3117] [0.3082] [0.3274] [0.2380] [0.2288] [0.2441]
w1: Father Alive 0.277 0.538* 0.566* 0.252 0.240 0.206
[0.3049] [0.2965] [0.3229] [0.3347] [0.3125] [0.7615]
w1: Mother Alive 0.230 0.148 0.119 -0.037 -0.004 0.020
[0.3331] [0.3451] [0.3550] [0.1728] [0.1966] [0.2188]
Socioeconomic Characteristics
Educ_Primary 0.267 0.195 0.177 -0.058 -0.042 -0.028
[0.2185] [0.2037] [0.2001] [0.2434] [0.2573] [0.8177]
Educ_Secondary -0.142 -0.164 -0.140 0.010 -0.021 0.007
[0.3001] [0.2979] [0.2875] [0.3401] [0.3624] [1.3134]
Married 0.248 0.383* 0.383* 0.098 0.086 0.070
[0.1956] [0.2288] [0.2185] [0.2967] [0.3681] [1.3267]
logPCE -0.082 -0.055 -0.063 -0.133 -0.121 -0.125
[0.0773] [0.0817] [0.0843] [0.0960] [0.1248] [0.5496]
Land Leased 0.000 0.000 0.000 -0.000 -0.000 -0.000
[0.0004] [0.0004] [0.0004] [0.0007] [0.0008] [0.0034]
HI Reimbursement Rate 0.434 0.230 0.090 -0.145 -0.143 -0.084
[0.4829] [0.4960] [0.4959] [0.3580] [0.4315] [0.4008]
95
Family Composition
# of Adult Sons -0.140 0.019
[0.1005] [0.0768]
# of Non-adult Sons 0.285 -0.212
[0.2623] [0.5009]
# of Adult Daughters -0.201** -0.071
[0.0963] [0.0905]
# of Non-adult Daughters 0.331* 0.433*
[0.1821] [0.2444]
# of Married Adult Sons -0.138 -0.016
[0.0969] [0.2761]
# of Married Non-adult Sons 0.531 -1.320
[0.6077] [1.0164]
# of Married Adult Daughters -0.231** -0.075
[0.0953] [0.2919]
# of Married Non-adult Daugh-
ters -0.101 0.513
[0.3861] [0.5630]
# of Non-married Adult Sons -0.090 0.252
[0.2214] [0.2145]
# of Non-married Non-adult
Sons 0.330 0.088
[0.2911] [1.6374]
# of Non-married Adult
Daughters 0.022 0.053
[0.2840] [0.4793]
# of Non-married Non-adult 0.616** 0.334
Daughters [0.3061] [0.7492]
Whether has Grandchild 0.457* 0.481* 0.065 0.090
[0.2642] [0.2704] [0.3444] [0.9162]
N 2,887 2,887 2,887 2,987 2,987 2,987
Mean 0.370 0.370 0.370 0.300 0.300 0.300
rho 0.500 0.570 0.550 -0.360 -0.270 -0.140
P_value: rho 0.122 0.046 0.037 0.807 0.876 0.984
P_value: All Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: Lagged Health 0.002 0.060 0.048 0.001 0.480 0.542
P_value: Current Health 0.002 0.003 0.004 0.145 0.051 0.498
P_value: poor GHS ß1+ß2=0 0.006 0.006 0.006 0.308 0.483 0.876
P_value: Disability ß1+ß2=0 0.058 0.087 0.086 0.003 0.015 0.485
P_value: CESD>=10 ß1+ß2=0 0.396 0.349 0.433 0.701 0.674 0.969
Note: FIML estimation of employment transition for individuals not employed at the baseline. Sample is limited to rural resi-
dents (hukou) with age range [50-90] at the baseline. Other covariates include age dummies, provincial fixed effects and ur-
ban/rural residency. Standard errors are in the parenthesis. Standard errors are robust and clustered at provincial level. ***
p<0.01, ** p<0.05, * p<0.1.
96
4.5.2 Family Composition and Employment Transitions
In the discussion above, we focus on the associations between health changes and the elderly’s
employment transitions. However, in a society where filial piety is highly valued, the potential
impact from family may be non-negligible. In rural regions where public pension is almost absent
and family support has long been in place as a safety-net, most middle-aged and elderly people
still consider children as a main form of old-age support
114
.
We differentiate between adult and non-adult sons and daughters and also include an indi-
cator for the presence of grandchild in the specification in Table 4.3 (Column 2 and 4). In rural
regions where son preference used to prevail and people traditionally rely on sons for old-age
support, originally we expect that the effects will be from sons. However, results show that for
men and women employed at the baseline, having an adult daughter in the family is significantly
associated with lower probability of continued employment. No correlation has been found from
sons or non-adult daughters. It could be possible that the elderly in rural China still mainly re-
ceive support from their sons; however, as living expenses increase, the financial support from
adult sons may not be enough to support both parents. Men and women, with or without adult
sons, continue to work as they are physically able to
115
. In this sense, the role of adult daughters
in both men and women’s employment transitions seem reasonable. The additional support from
adult daughters may facilitate the elderly’s “retirement”
116
.
Further differentiation of the family members according to their marital status is specified to
examine in more detail where the effects may come from. Results in Table 4.3 column 3 and 6
show a general pattern: for both men and women, having a married child seems to be positively
associated with workforce withdrawal, while having a non-married child is positively correlated
114
In CHARLS baseline, respondents were asked: whom do they think they can rely on for old-age support?
Among children, savings, pensions, commercial pension insurance and others, more than 80% of elderly in
rural China chose children.
115
Results from the baseline estimation shows that fathers’ participation decision is not related with their
family composition. On the contrary, mothers with adult daughters are significantly less likely to be work-
ing at the baseline. However, it seems that mothers are more likely to work at the baseline if they have non-
adult sons in the family.
116
One possible explanation for the significant negative coefficient of Number of Adult Daughter could be
that elderly may drop out of the labor force in order to help take care of their daughters’ children. We
examine this possibility by re-estimating the model with an interaction term between Number of Adult
Daughter and Grandchildren. Results show that the coefficient of the interaction term is not significant, there-
fore this possibility is ruled out.
97
with continued employment. The effects from adult daughters continue to be significant and it is
mostly observed from the married ones. This is consistent with the previous result and to some
extent may indicate that part of the support to the elderly people could come from their sons-in-
law. We also find a negative association between non-married adult sons and women’s continued
employment. In addition, women with grandchildren are significantly less likely to keep working.
It could be possible that they discontinue their employment to help taking care of their grand-
children when the parents of the grandchildren are out working, which is common in the rural
areas.
For those who were already out of the labor market at the baseline, re-entry decision is also
to some extent affected by their family composition. As can be seen in Table 4.4, men with adult
daughters are less likely to re-enter the workforce. We also find that for both men and women,
having a non-adult daughter is significantly positively associated with subsequent employment.
Results that further differentiate the family members according to their marital status exhibit a
similar pattern as that in the continued employment decision. Specifically, men with married
adult daughters are less likely to re-enter the workforce. However, they tend to start work again
if they have non-married non-adult daughters. The presence of a grandchild has a different im-
pact on the elderly’s workforce re-entry decisions. Previously we observe that women with
grandchildren are more likely to exit the workforce. However, for women who were already out
of employment, whether or not having a grandchild does not affect their work choices. On the
contrary, for male elderly, the presence of a grandchild significantly increases their probability of
reemployment. This may indicate different roles of male and female elderly in the upbringings
of their grandchildren. Elderly women quit their jobs to help take care of the grandchildren, while
elderly men re-enter the workforce to better provide for their grandchildren.
4.5.3 The Impacts of Demographic and Socioeconomic Charac-
teristics
In addition to health changes and family composition, effects from demographic and socioeco-
nomic characteristics have been examined. The educational attainment of the mid-aged and el-
derly in rural China is low, with more than a third are illiterate and less than 20% have ever
98
received any secondary education (Table 4.1). It also seems to be irrelevant in senior people’s
employment transitions, either out of or into the economic activity. As can be seen in Table 4.3
and 4.4, the coefficients of the primary and secondary education dummies are almost never sig-
nificant
117
. Being married, however, is positively correlated with seniors’ continued employment.
The fact that being married is also positively associated with higher probability of work partici-
pation at the baseline suggests that married couple may have a larger family to support, thus may
need to stay working. It may also reflect a positive selection effect where healthier individuals are
more likely to get married in the first place and stay employed when they grow old. As for the
labor market re-entry decision, however, marriage does not seem to be much related.
In a developing society such as rural China, the effect from economic aspect is mixed. Higher
PCE in wave 1 is significantly associated with a lower probability of continuing employment for
senior males. This to some extent captures an income effect on labor supply. For senior females,
the association between PCE and employment transition is also negative, though small and in-
significant. Regarding the land, we find that for both genders, a larger amount of leased land is
positively associated with their continued employment probability. This might result from the
incentive that inactively cultivated farmland could be redistributed to other households. It is also
likely that labor demand is higher if they have more land. When it comes to the labor market re-
entry decision, we do not find much impact from PCE or leased land. It seems that once the el-
derly are out of employment, economic resource is not a concern when they decide whether or
not to re-enter the labor market.
As for the health insurance aspect, we find that higher reimbursement rate is significantly
correlated with lower probability of continued employment. It is possible that for the elderly who
lives in a county with a relatively generous health scheme, the economic burden of health service
cost could be reduced to some extent. Therefore, they may not need to keep working at old ages
to pay for the existing/future inpatient care services, which is a large cost to them once it occurs
118
.
Health insurance may also affect the elderly’s labor supply in an indirect way. Counties with a
higher reimbursement rate are generally wealthy and more economically developed. Residents
117
Joint test of the two education categories show no significance as well.
118
According to Lei and et al. (2015), half of the respondents in CHARLS baseline have out-of-pocket (OOP)
expenditures for inpatient care payments that equals to 20% of PCE. About 20% of respondents have OOP
expenditures greater than 40% of their PCE.
99
of those counties may have other forms of old-age support in addition to labor income, thus lead-
ing to a decline in the probability of continued employment.
4.5.4 Simulations
4.5.4.1 Simulation Effects of Health Changes
As to this point, we only know about the direction of the associations between factors of interest
and employment transitions. The interpretation of the magnitude of the estimated coefficients is
difficult. Therefore, simulations have been conducted based on the estimates to further our un-
derstanding of the interrelations between health changes, family composition and elderly’s work
transitions
119
Since the health changes in our model are captured by changes in three health di-
mensions: GHS, disability and depressive symptoms, the variations of health in the following
scenarios will only focus on these variables while original values of the other two dimensions of
health are applied. Regarding each employment transition process, we simulate the employment
probability conditional on the work status at the baseline for each observation and then take the
simple average. Results are presented in Table 4.5 and Table 4.6.
We first simulate the effects from health changes based on the estimates separately for men
and women from the second specification in Table 4.3, under the counterfactual assumptions that
(1) all the individuals in the sample are in good health at both waves
120
; (2) all the respondents
were in good health at the baseline but experienced a negative change in all three health dimen-
sions in wave 2; (3) everyone was in good health at the first wave but one of the conditions dete-
riorated in the second wave.
In the “ideal health” case where everyone is consistently healthy, 92% of senior males and
87% of senior females will continue to work conditional on being employed in the previous wave.
119
It is also interesting to investigate the simulated effects on employment transition from pure aging.
However, growing old is always accompanied by deterioration in health and it is almost impossible to
disentangle the impact from these two aspects. We try to approximate the effect from pure aging by as-
suming that all the factors are held constant (including health variables) and the only thing that changes is
respondents’ age. We simulate the conditional probability of continued employment and labor force re-
entry separately for men and women with respect to one-year and five-year increase in age. Results (not
presented) imply that pure aging may have very limited impact on elderly’s employment transitions, es-
pecially compared to those from health changes.
120
This is a scenario where all the individuals reported good GHS, no disability and low depressive symp-
tom in both waves.
100
Table 4.5A Simulated Likelihood of Continued Employment w.r.t. Health Changes
Men Women
From Employment in wv1 From Employment in wv1
To Employment in wv2
Differ-
ence
To Employment in wv2
Differ-
ence
predicted P(empl in wv2 conditional on empl in wv1)
0.8958 0.8430
predicted P(empl in wv1)
0.8874 0.8142
Simulations
"Ideal Health" P(empl in wv2 conditional on empl in
wv1)
0.9215 0.8672
"Ideal Health" P(empl in wv1)
0.9239 0.8563
Experiencing Adverse Health Shock
0.7672 -0.1543 0.7660 -0.1011
w2: poor GHS
0.8525 -0.0689 0.7899 -0.0773
w2: Disability
0.8799 -0.0416 0.8370 -0.0301
w2: CESD>=10
0.9091 -0.0123 0.8777 0.0106
Note: Simulated probability of continued employment in response to health changes for individuals employed at the baseline. Simulations are based on esti-
mates of the second specification in Table 4.3 and Table A4.1.
101
Table 4.5B Simulated Likelihood of Labor Market Re-entry w.r.t. Health Changes
Men Women
From Non-employment in
wv1
From Non-employment in
wv1
To Employment in wv2
Differ-
ence
To Employment in wv2
Differ-
ence
predicted P(empl in wv2 conditional on non-empl in
wv1))
0.3634 0.2970
predicted P(non-empl in wv1)
0.2800 0.3722
Simulations
"Bad Health" P(empl in wv2 conditional on non-empl
in wv1))
0.1045 0.1973
"Bad Health" P(non-empl in wv1)
0.4147 0.4661
Improvement in Health
0.3200 0.2154 0.3041 0.1068
w2: poor GHS
0.2287 0.1241 0.2405 0.0433
w2: Disability
0.1749 0.0704 0.3390 0.1418
w2: CESD>=10
0.0962 -0.0083 0.1366 -0.0607
Note: Simulated probability of labor force re-entry in response to health changes for individuals not employed at the baseline. Simulations are based on esti-
mates of the second specification in Table 4.4 and Table A4.2.
102
Suppose the previous assumption holds where all the individuals are in good health at the
beginning, but then experience a negative health change in the subsequent period. In this case,
both of the probabilities of staying employed for men and women reduce to 77%. Men respond
more strongly than women. Specifically, compared to the “ideal health” case, the decline of the
probability of continued employment is 15 percentage points for men, while for women, it is
about 10 percentage points.
Simulation results for each of the involving health indicators are consistent with the conclu-
sions in Section 6.1. For both men and women, the largest source of decline in continued employ-
ment probability comes from the current GHS. It alone leads to a 7-8 percentage point decrease
in the probability of staying employed. The second largest impact comes from the current disa-
bility measure, which causes 3-4 percentage point decline in the continued working probability.
We next simulate the conditional probabilities of labor market re-entry for elderly who were
not employed at the baseline. Assumptions are slightly altered. Where in the first scenario, indi-
viduals are in consistently poor health at both waves. In this setting, 10%-20% of men and women
will re-enter economic activities among those not employed at the baseline. However, if health is
assumed to improve in the subsequent wave, then there will be a large increase in re-entry rate
for both genders. Results in Table 4.5A suggest that re-entry rate for men will be 32%, which is a
22 percentage points increase compared to the “bad health” scenario. For women, the increase is
11 percentage points. Simulation for each of the binary health indicators show that for men, im-
provement in GHS leads to a 12 percentage points increase in the participation rate. For women,
amelioration in GHS and disability are both important factors in encouraging them to re-enter
the workforce, leading to an increase of 4 and 14 percentage points, respectively.
Previous literature has discussed the asymmetry of participation response of individuals to
changes in health. In general it suggests that the association between health deterioration and
labor force exits is different from the association between health improvement and labor market
re-entry. Moreover, it seems that a “ratchet” effect exists where larger health improvements are
needed to encourage reemployment (Disney et al., 2006). In the case of rural China, asymmetry
is also observed between the two dynamic associations
121
, at least for men. However, unlike the
“ratchet” effect that has been observed in other studies, male labor market re-entrants in rural
121
The asymmetry is also one of the reasons that fixed effects model, which imposes strong restrictions on
the dynamic structure and does not differentiate between types of state transitions, is not used for analy-
sis in this paper.
103
China seem to be much more sensitive to their health changes. One of the possible reasons could
be that for most re-entrants, labor force withdrawals from the previous period are temporary,
perhaps due to health deterioration. For these individuals with strong motivations to work, as
soon as their health restore, they would transit into the labor market. In addition, the fact that a
majority of the elderly in the rural areas has to rely on their labor income for old-age support may
also be an encouraging factor for them to return to work.
Previous literature has discussed the asymmetry of participation response of individuals to
changes in health. In general it suggests that the association between health deterioration and
labor force exits is different from the association between health improvement and labor market
re-entry. Moreover, it seems that a “ratchet” effect exists where larger health improvements are
needed to encourage reemployment (Disney et al., 2006). In the case of rural China, asymmetry
is also observed between the two dynamic associations
122
, at least for men. However, unlike the
“ratchet” effect that has been observed in other studies, male labor market re-entrants in rural
China seem to be much more sensitive to their health changes. One of the possible reasons could
be that for most re-entrants, labor force withdrawals from the previous period are temporary,
perhaps due to health deterioration. For these individuals with strong motivations to work, as
soon as their health restore, they would transit into the labor market. In addition, the fact that a
majority of the elderly in the rural areas has to rely on their labor income for old-age support may
also be an encouraging factor for them to return to work.
4.5.4.2 Simulation Effects of Family Composition
The employment transition probability has also been simulated with respect to changes in the
elderly’s family composition. Using the estimates of second specification in Table 4.3, first we
assume that no individual has any child or grandchild. This results in a conditional probability
based on having been working at the baseline to be 92% and 89%, for men and women respec-
tively (Table 4.6A). Then we calculated the likelihood assuming that each of the respondents has
an adult son/an adult daughter/a non-adult son/a non-adult daughter/a grandchild, the change
122
The asymmetry is also one of the reasons that fixed effects model, which imposes strong restrictions on
the dynamic structure and does not differentiate between types of state transitions, is not used for analy-
sis in this paper.
104
Table 4.6A. Simulated Likelihood of Continued Employment w.r.t. Changes in Family Composition
Men Women
From Employment in wv1 From Employment in wv1
To Employment in wv2
Differ-
ence
To Employment in wv2
Differ-
ence
predicted P(empl in wv2 conditional on empl in wv1)
0.8958 0.8430
predicted P(empl in wv1)
0.8874 0.8142
"No Child" P(empl in wv2 conditional on empl in
wv1)
0.9222 0.8923
Has an Adult Son
0.9235 0.0012 0.8831 -0.0092
Has a Non-adult Son
0.9198 -0.0025 0.8909 -0.0014
Has an Adult Daughter
0.9134 -0.0089 0.8848 -0.0075
Has a Non-adult Daughter
0.9349 0.0127 0.9216 0.0293
Has Grandchild
0.9022 -0.0201 0.8631 -0.0292
Note: Simulated probability of continued employment in response to changes in family compositions for individuals employed at the baseline. Simulations are
based on estimates of the second specification in Table 4.3 and Table A4.1.
105
Table 4.6B. Simulated Likelihood of Labor Market Re-entry w.r.t. Changes in Family Composition
Men Women
From Non-employment in
wv1
From Non-employment in
wv1
To Employment in wv2
Differ-
ence
To Employment in wv2
Differ-
ence
predicted P(empl in wv2 conditional on non-empl in
wv1)
0.3634 0.2970
predicted P(non-empl in wv1)
0.2800 0.3722
"No Child" P(empl in wv2 conditional on non-empl
in wv1)
0.6758 0.3634
Has an Adult Son
0.6326 -0.0432 0.3735 0.0101
Has a Non-adult Son
0.7544 0.0786 0.2828 -0.0806
Has an Adult Daughter
0.6139 -0.0619 0.3448 -0.0186
Has a Non-adult Daughter
0.7569 0.0811 0.4953 0.1319
Has Grandchild
0.7720 0.0962 0.3994 0.0360
Note: Simulated probability of labor force re-entry in response to changes in family compositions for individuals not employed at the baseline. Simulations are
based on estimates of the second specification in Table 4.4 and Table A4.2.
106
of continued employment rate is small, especially for men, where most of the changes are less
than 1 percentage point. The changes of re-entry probability brought by variations of family mem-
bers are larger (Table 4.6B). Compared to a childless scenario, there is a 7 percentage points in-
crease in workforce re-entry rate for men with an adult daughter. We also find that increase in re-
entry rate is as high as 8-13 percentage points for men and women with a non-adult daughter
respectively. In addition, there is a 9 percentage points increase in the probability of labor market
re-entry for men with a grandchild.
In sum, it seems that family composition has different weights when the rural elderly make
their work transitions. For those employed, family impact is trivial and health is a much more
critical concern when they decide whether or not to stay employed. For those out of employment,
both health and family are important factors in transitioning into the labor market.
4.5.5 Health Changes and Annual Work Hours
Results examining the association between health changes and changes in the intensity of the
elderly’s labor supply are presented in Table 4.7. Poor current GHS is observed to be significantly
correlated with decreasing annual work hours for both men and women. When controlling for
baseline health, individuals who report adverse current GHS on average work about 200 hours
fewer than those in good GHS. For men, our results show that declines in GHS help explains the
reduction in annual hours. For women, it seems that both lagged GHS and changes in GHS are
significantly correlated with few hours of work. Unlike the previous results where current disa-
bility measure is found to be an important factor in employment transitions, it does not seem to
affect elderly’s working hours. No effect has been detected from other health indicators either;
however, the presence of parents are still significantly correlated with women’s working hours.
Women on average work 180 more hours if their fathers are still alive; however, their working
hours reduce drastically if their mothers are alive. The correlation between family composition
and changes in work hours is not significant.
107
Table 4.7 OLS Estimation of Change in Annual Hours
Change in Annual Hours
Men Women
Health Changes
w1: poor GHS 79.205 80.391 70.847 -0.211 -3.048 -5.306
[118.3035] [117.5912] [119.1012] [61.7748] [61.1067] [61.4830]
w2: poor GHS -233.687** -232.193** -236.266*** -243.025*** -237.576*** -238.185***
[86.9960] [85.8207] [84.6356] [71.8545] [70.5194] [69.6155]
w1: Disability 81.662 80.921 76.268 30.669 30.153 35.023
[76.2973] [75.6447] [76.9176] [61.6669] [61.0927] [61.4890]
w2: Disability 9.770 8.595 11.271 36.443 36.038 33.403
[75.0828] [73.2114] [73.3862] [64.2032] [63.8877] [63.9660]
w1: CESD>=10 -7.181 -7.587 6.572 -19.941 -20.339 -21.899
[91.7370] [92.9726] [94.3872] [71.4940] [72.0029] [71.9391]
w2: CESD>=10 -13.641 -14.098 -18.223 -75.914 -74.463 -73.550
[103.7076] [102.8978] [103.0309] [89.7342] [90.1338] [91.0892]
Baseline Health Indicators
w1: Limb Length 8.380 8.292 7.530 -13.020 -13.314 -13.658
[11.8126] [11.7590] [11.2954] [10.1424] [10.2118] [10.3457]
w1: Hypertension -44.903 -44.573 -34.591 -125.990 -128.643 -135.558
[97.2056] [96.8769] [98.2279] [81.4246] [79.3243] [80.5112]
w1: Overweight 128.196 123.956 124.815 100.488 99.917 94.140
[91.5579] [90.1691] [89.8598] [63.5887] [63.8745] [64.0974]
w1: Undernourished -94.583 -95.647 -87.871 0.737 4.528 -18.711
[112.3396] [111.4479] [114.0568] [109.9617] [109.3494] [104.9369]
w1: Father Alive -49.653 -51.396 -42.407 184.681** 183.447** 171.302*
[107.5614] [110.7563] [111.9151] [86.7896] [87.5666] [90.0869]
w1: Mother Alive -68.357 -66.497 -69.548 -255.238*** -252.655*** -246.582***
[111.2110] [111.7041] [110.4642] [42.6135] [41.1861] [43.7609]
Socioeconomic Characteristics
Educ_Primary 195.392** 194.028** 198.434** -0.464 3.180 -0.072
[92.2566] [92.3418] [94.8380] [67.9799] [68.9975] [69.3804]
Educ_Secondary 110.195 111.253 112.019 44.597 43.520 25.614
[106.4265] [109.0286] [110.4369] [110.7968] [108.9262] [109.5851]
Married -62.393 -62.849 -83.197 99.738 98.462 98.911
[120.1503] [120.2204] [120.1484] [95.9151] [96.2944] [97.8362]
logPCE -37.268 -36.280 -41.142 19.925 20.094 16.142
[45.0848] [44.8612] [45.9980] [42.8102] [42.0760] [39.4261]
Land Leased 0.715*** 0.728*** 0.765*** 0.938*** 0.945*** 0.910***
[0.0857] [0.0889] [0.0863] [0.2664] [0.2821] [0.3079]
HI Reimbursement Rate 77.154 81.473 93.361 -163.067 -169.401 -155.814
[174.0179] [172.6311] [171.4748] [164.4812] [156.6696] [157.0385]
108
Family Composition
# of Adult Sons -15.324 -16.373
[37.0750] [33.8007]
# of Non-adult Sons -21.714 -95.809
[106.8897] [92.7376]
# of Adult Daughters -1.470 -3.694
[42.3287] [29.3224]
# of Non-adult Daughters -19.187 63.608
[80.1563] [81.8534]
# of Married Adult Sons 2.819 -20.038
[44.3461] [32.8362]
# of Married Non-adult Sons -193.075 -196.403
[197.6060] [223.2546]
# of Married Adult Daughters -18.985 -22.726
[45.3805] [30.1039]
# of Married Non-adult Daughters -27.170 77.884
[126.9511] [128.3231]
# of Non-married Adult Sons -98.116 -15.711
[78.2078] [68.6274]
# of Non-married Non-adult Sons 32.718 -60.768
[105.3031] [95.5511]
# of Non-married Adult Daughters 212.408** 264.684
[78.0966] [176.1111]
# of Non-married Non-adult -18.272 44.486
Daughters [115.2732] [123.5590]
Whether has Grandchild 55.066 69.908 29.573 73.495
[126.4769] [127.5769] [128.2127] [125.9569]
N 1,843 1,843 1,843 1,704 1,704 1,704
P_value: All Health 0.013 0.014 0.018 0.000 0.000 0.000
P_value: Lagged Health 0.646 0.552 0.678 0.001 0.001 0.001
P_value: Current Health 0.064 0.060 0.048 0.021 0.021 0.019
P_value: poor GHS ß1+ß2=0 0.250 0.250 0.222 0.015 0.016 0.014
P_value: Disability ß1+ß2=0 0.362 0.373 0.376 0.470 0.477 0.462
P_value: CESD>=10 ß1+ß2=0 0.861 0.854 0.921 0.339 0.345 0.349
Note: OLS estimation of annual work hours changes in response to health changes. Sample is limited to rural residents (hukou)
with age range [50-90] at the baseline. Other covariates include age dummies, provincial fixed effects and urban/rural resi-
dency. Standard errors are in the parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, ** p<0.05,
* p<0.1.
109
4.5.6 Heterogeneity
4.5.6.1 Agricultural Workers v.s. Non-agricultural Workers
A majority of the rural elderly are agricultural workers; however, there is also a small fraction of
individuals not involved in farming
123
. It is interesting to investigate whether labor market tran-
sitions differ between these two groups
124
in response to health changes. For agricultural workers,
we examine the link between health and work separately for men and women. For non-agricul-
tural workers, we pool the male and female sample since their observations are limited.
The patterns for agricultural workers shown in Table A4.3 are similar as before. For both
men and women, we find that deterioration in health, especially GHS or/and disability, is signif-
icantly associated with reduced probability of continued employment. On the contrary, for indi-
viduals who are only involved in non-farming activities, changes in health do not seem to impose
a constraint on them. Though an initial poor health level is significantly correlated with lower
work participation (Table A4.4). The different responses between farmers and non-farmers might
be explained by their job characteristics. In rural China, most of the non-farmers make a living by
managing a small business or as an employed worker. Generally, their jobs are not as physically
demanding as working on the farm, therefore a deterioration in health may not necessarily cause
them to quite the jobs. On the other hand, these jobs may not be as flexible as farming, and they
would not consider about transiting out of the workforce as long as their health condition allows
them to keep working.
4.5.6.2 Interactions of Health with Education
As a group with low educational attainment, education level does not seem to be relevant when
the elderly in rural China make their employment transitions. Though no direct impact from ed-
ucation has been found, it is possible that education affect elderly’s labor market behaviors in an
123
Only around 7% of respondents in our sample are purely involved in non-agricultural activities.
124
A respondent is categorized as an agricultural work or non-agricultural work based on his/her baseline
job. If he/she only involves in farming (either on his/her own farm or works on the others’ farms), the
respondent is regarded as an agricultural worker. If he/she does not work on the farm at all at the baseline,
we categorize them as non-agricultural workers.
110
indirect way, for instance, through its impact on health changes. To investigate this possibility, a
new specification with interaction terms of education and health has been estimated. To simplify
the interpretation, instead of using education categories, we construct an indicator measuring
years of education. This variable is then de-meaned and interacted with all the health indicators.
The interacted coefficients tell us the effects of mean of education.
Results are shown in Table A4.5, where the first two columns are estimations for continued
employment and last two columns for labor market re-entry. The interaction coefficients are
jointly significant for both men and women in their labor market transitions: either out of or into
employment. However, the interaction coefficients are small compared to the un-interacted coef-
ficients. In addition, the interacted health coefficients show the same pattern as before. Therefore,
it indicates that men and women with different educational levels may respond differently in
their labor supply decisions with respect to health changes; however, the difference is very small.
4.6 Conclusion and Discussion
This paper explores the interrelation between health changes and employment transitions of el-
derly in rural China. Using two waves of data from CHARLS, we study two labor market transi-
tions. For individuals employed at the baseline, the association between health changes and the
probability of continued employment is modeled; while for individuals who were non-employed
in the first wave, we examine their labor market re-entry decision in response to health changes.
Since there is possible selection in the first year, employment status at the baseline is controlled
for.
The results are similar to those documented in the literature regarding developed countries;
for individuals employed at the baseline, a significant decline in labor force participation is ob-
served for both men and women who reported poor current GHS or those with disability in the
latest wave when lagged health indicators are conditioned. We show that not just poor health,
but also declines in health, explain the elderly’s employment transitions. Results regarding the
labor market re-entrants are similar. We also find some effects from baseline biomarkers. It seems
that men are more likely to exit the workforce if they are undernourished, while this is the same
case for women who are overweight.
111
In rural regions where strong son preference used to prevail, associations between family
members and probability of continued employment, however, are mainly reflected in adult
daughters, especially married ones. We also observe a significant correlation between the pres-
ence of a grandchild and the elderly’s labor supply behaviors, where men and women seem to
play different roles. Examination of other covariates also reveal some interesting links. For in-
stance, both genders are more likely to stay working if they have a large amount of leased land.
In addition, we also find that the elderly are significantly more likely to exit the workforce if they
live in a region with a higher health insurance reimbursement rate.
Simulations based on the estimates further our understanding of the dynamic associations.
Compared to the “consistently healthy” case where all the individuals are assumed to be in good
health at both waves, experiencing an adverse health change leads to a decline of 10-15 percentage
points in the probability of continued employment. The adjustment of labor market re-entry prob-
ability in response to health changes is even stronger. Compared to the “consistently poor health”
case, improvement in health could result in a 11-22 percentage points increase in participation
rate for those initially non-employed.
Similar patterns can be found when we examine the associations between health changes
and changes in annual working hours. In addition, when disaggregating the sample by types of
job, the previous associations exist for agricultural workers. The extent to which health changes
are associated with employment transitions differ slightly for individuals with distinctive educa-
tional attainments; however, the difference is small.
The high proportion of working elderly and the existence of a strong dynamic association
between health and work imply that remaining in good health status or improving health is ex-
tremely important for this group, especially when pension is not adequate for old-age support
and the elderly heavily rely on labor and self-employment income for a living. However, this
interrelation might be henceforth subject to change. Future cohorts of Chinese elderly will have
fewer adult children to reply upon. In addition, since health insurance and pension schemes are
related to both the rural population’s health and labor supply, changes in these social policies
may alter the current associations between health and employment transitions. With extensive
changes undergoing in both the newly implemented medical insurance and pension scheme in
rural regions of China, it remains to be seen how the relationship between health and employ-
ment will evolve in the future.
112
Chapter 5
Conclusion
Aging is different in the urban and rural China. The urban elderly, in particular those who work
in the formal sector, share some common characteristics with the aging population in the devel-
oped countries. For example, they have proper employment, and can enjoy relatively generous
pension wages after retirement. However, they face mandatory retirement. The rural elderly, on
the other hand, face no such kind of restriction. But in the meanwhile, they have very limited
public pension support. This dissertation asks two distinctive aging-related questions with re-
spect to these two different groups. For the elderly in urban areas, we are interested in examining
the impact of retirement on health, exploiting the institutional property of the mandatory retire-
ment system. For the rural elderly, we are interested in exploring the static and dynamic associa-
tions between health and employment.
The data set used for analysis in this dissertation is the China Health and Retirement Longi-
tudinal Study (CHARLS). Chapter 2 and Chapter 3 utilize the national baseline survey which
were fielded in 2011; and Chapter 4 uses two waves of CHARLS data. We are particularly inter-
ested in the relationship between health and employment. Thanks to the richness of the data set,
it allows us to measure both of the elements in various dimensions.
Results from Chapter 2 suggest that retirement has minimal impact on both men and
women’s health in general. However, we do find that men are significantly more likely to suffer
from body pains and overweight problems after retirement, and their life satisfaction drops con-
siderately. For women, they are more likely to have hypertension and high glucose levels, as well
as memory loss. But retirement to some extent alleviates their body pains and anemia symptoms.
The static and dynamic analyses from Chapter 3 and Chapter 4 share some similar conclusions.
We find that current health, in particular self-reported health status and disability, is significantly
associated with an elderly’s labor force participation decision as well as work hours. Furthermore,
changes in self-reported health status and disability are significantly associated with the elderly’s
employment transitions, no matter the decision is to transit out of or back into the workforce.
Additional estimation results reveal that family composition, health insurance plans, amount of
113
land leased are also significantly associated with the elderly’s employment decisions. And the
association between health and employment varies with different types of jobs, educational at-
tainments and levels of income.
Implications of the results are different. For the urban formal sector workers, the argument
as whether or not the government should raise the mandatory retirement age has been debated
for a long time. Chapter 2 analyzes the issue in a health perspective. By arguing that postponing
retirement is not bad for health, it supports raising the retirement ages in the near future. As for
the rural elderly, no direct policy implication has been drawn from the analyses in Chapter 3 and
Chapter 4. However, with extensive changes currently undergoing in health insurance and pen-
sion schemes in the rural areas, the associations between health and employment established in
both studies are subject to change. It remains to be seen how the current associations will be al-
tered by the future policy changes and how the understanding of the health-work nexus may
contribute to adjusting the related social policies.
114
Bibliography
Anderson, K.; Burkhauser, V. 1984. "The importance of the measure of health in empirical
estimates of the labor supply of older men." Economics Letters 16:375-380.
Anderson, K.; Burkhauser, V. 1985. "The retirement-health nexus: a new measure of an old
puzzle." Journal of Human Resources 20:315-330.
Andresen, E.; Malmgren, J.; Carter, W.; Patrick, D. 1994. "Screening for depression in well older
adults: evaluation of a short form of the CES-D." American Journal of Preventive Medicine
10:77-84.
Au, D; Crossley, T.; Schellhorn, M. 2005. "The effect of health changes and long-term health on
the work activity of older Canadians." Health Economics 14:999-1018.
Barber, L.; Yao, L. 2010. Health insurance systems in China: a briefing note. In World Health
Report 2010.
Bartel, A.; Taubman, P. 1979. "Health and labor market success: the role of various diseases."
The Review of Economics and Statistics 61:1-8.
Bazzoli, J. 1985. "the early retirement decision: new empirical evidence on the influence of
health." Journal of Human Resources 20:214-234.
Behncke, S. 2012. "Does retirement trigger ill health? ." Health Economics 21 (3):282-300.
Belgrave, L.; Haug, M.; Gomez-Bellenge, F. 1987. "Gender and race differences in effects of
health and pension on retirement before 65." Comprehensive Gerontology 1 (3):109-117.
Benjamin, D.; Brandt, L.; Fan, J. 2003. "Ceaseless toil? health and labor supply of the elderly in
rural China." William Davidson Institute Working Paper.
Benjamin, D.; Brandt, L.; Rozelle, S. 2000. "Aging, wellbeing, and social security in rural
northern China." Population and Development Review 26:89-116.
115
Berger, M. 2004. "Health and family labor force transitions." Quarterly Journal of Business and
Economics 43:113-138.
Blau, D. 1994. "Labor force dynamics of older men." Econometrica 62 (1):117-156.
Blau, D.; Gilleskie, D. 2001. "The effect of health on employment: transitions of older men."
Worker Wellbeing in a Changing Labor Market 20:35-65.
Blundell, R.; Meghir, C. 1987. "Bivariate alternatives to the tobit model." Journal of Econometrics
34:179-200.
Bonsang, E.; Adam, S.; Perelman, S. 2012. "Does retirement affect cognitive functioning." Journal
of Health Economics 31 (3):490-501.
Bound, J. 1991. "Self-reported versus objective measures of health in retirement models." Journal
of Human Resources 26 (106-138).
Bound, J.; Schoenbaum, M.; Stinebrickner, T.; Waidmann, T. 1999. "The dynamic effects of
health on the labor force transitions of older workers." Labour Economics 6:179-202.
Bound, J.; Stinebrickner, T.; Waidmann, T. 2010. "Health, economic resources and the work
decisions of older men." Journal of Econometrics 156:106-129.
Bound, J.; Waidmann, T. 2007. "Estimating the health effects of retirement." Michigan Retirement
Research Center Research Paper UM WP 2007-168.
Brandt, J.; Spencer, M.; Folstein, M. 1988. "The telephone interview for cognitive status."
Neuropsychiatry, Nerupsychology, and Behavioral Neurology 1 (2):111-117.
Cai, F.; Giles, J.; O’keefe, P.; Wang, D. 2012. "The elderly and old age support in rural China:
challenges and prospects." World Bank Publications.
Cai, L.; Cong, C. 2009. "Effects of health and chronic diseases on labour force participation of
older working-age Austrilians." Australian Economic Papers.
116
Cai, L.; Kalb, G. 2006. "Health status and labour force participation: evidence from Australia."
Health Economics 15 (241-261).
Calvo, E.; Sarkisian, N.; Tamborini, C. R. 2012. "Causal effects of retirement timing on subjective
physical and emotional health." Journal of Gerontology Series B: Psychological Sciences and
Social Science 68 (1):73-84.
Cameron, A.; Colin, J.; Gelbach, G.; Douglas, M. 2011. "Robust inference with multi-way
clustering." Journal of Business and Economic Statistics 29 (2):238-249.
Cameron, A.; Trivedi, P. . 2005. Microeconometrics: methods and applications. Cambridge:
Cambridge University Press.
Cameron, L.; Cobb-Clark, D. 2001. "Old-age support in developing countries: labor supply,
intergenerational transfers and living arrangements." IZA Discussion Paper No. 289.
Cameron, L.; Cobb-Clark, D. 2010. "Old-age labour supply in the developing world." Applied
Economics Letters 9:10:649-652.
Charls, K. . 2004. "Is retirement depressing?: labor force inactivity and psychological well-being
in later life." Research in Labor Economics 23:269-299.
Chatterji, P.; Alegria, M.; Takeuchi, D. 2011. "Psychiatric disorders and labor market outcomes:
evidence from the National Comorbidity Survey-Replication." Journal of Health
Economics 30 (5):858-868.
Chumlea, W.; Roche, A.; Steinbaugh, M. 1985. "Estimating stature from knee height for persons
60 to 90 years of age." Journal of American Geriatrics Society 33:116-120.
Coe, N. B.; Lindeboom, M. 2008. "Does retirement kill you? evidence from early retirement
windows." IZA Discussing Paper No. 3817.
Coe, N. B.; von Gaudecker, N. M.; Lindeboom, M.; Maurer, J. 2012. "The effect of retirement on
cognitive functioning." Health Economics 21 (8):913-927.
117
Coe, N. B.; Zamarro, G. 2011. "Retirement effects on health in Europe." Journal of Health
Economics 30 (1):77-86.
Coile, C.; Gruber, J. . 2007. "Future social security entitlements and the retirement decision." The
Review of Economics and Statistics 89 (2):234-246.
Costa, D. L. 1998. "The evolution of retirement." In The Evolution of Retirement: An American
Economic History, 1880-1990. Chicago: The University of Chicago Press.
Crimmins, E.; Vasunilashorn, S.; Kim, J.; Alley, D. 2008. "Biomarkers related to aging in human
populations." Advances in Clinical Chemistry 46:161-216.
Currie, J.; Madrian, B. 1999. "Health, health insurance and the labor market." In Handbook of
Labor Economics. Amsterdam: Elsevier Science Publishers.
Danesh, J.; Pepys, M. 2000. "C-reactive protein in healthy and in sick populations." European
Heart Journal 21:1564-1565.
Dave, D.; Rashad, I.; Spasojevic, J. 2007. "The effects of retirement on physical and mental health
outcomes." Southern Economic Journal 75:497-523.
Davis-Friedmann, D. 1991. Long lives: Chinese elderly and the communist revolution. Stanford:
Stanford University Press.
Disney, R.; Emmerson, C.; Wakefield, M. 2006. "Ill health and retirement in Britain: a panel data-
based analysis." Journal of Health Economics 25:621-649.
Dorfman, M.; Wang, D.; O’Keefe, P.; Cheng, J. 2013. "China’s pension schemes for rural and
urban residents." In Matching Contributions for Pensions: A Review of International
Experience, edited by Robert; Tuesta Holzmann, David. Washington, DC: World Bank.
Doshi, J.; Cen, L.; Polsky, D. 2008. "Depression and retirement in late middle-aged U.S.
workers." Health Services Research 43 (2):693-713.
Dwyer, D. S.; Mitchell, O. S. 1999. "Health problems as determinants of retirement: are self-rated
measures endogenous." Journal of Health Economics 18:173-193.
118
Ekerdt, D.; Baden, L.; Bosse, R.; Dibbs, E. 1983a. "The effect of retirement on physical health."
American Journal of Public Health 73:779-783.
Ekerdt, D.; Bosse, R.; LoCastro, J. 1983b. "Claims that retirement improves health." Journal of
Gerontology 38:231-236.
Ettner, S.; Frank, R.; Kessler, R. 1997. "The impact of psychiatric disorders on labor market
outcomes." Industrial and Labor Relations Review 51:64-81.
Fonseca, R.; Kapteyn, A.; Lee, J.; Zamarro, G.; Feeney, K. 2014. "A longitudinal study of well-
being of older Europeans: does retirement matter." Population Ageing 7:21-41.
Gall, T.; Evans, D.; Howard, J. 1997. "The retirement adjustment process: changes in the well-
being of male retirees across time." Journal of Gerontology: Psychological Sciences 52
(B):110-117.
Gannon, B. 2005. "A dynamic analysis of disability and labour force participation in Ireland
1995-2000." Health Economics 14:925-938.
Giles, J.; Mu, R. 2014. "Village political economy, land tenure insecurity, and the rural to urban
migration decision." World Bank Policy Research Working Paper 7080.
Giles, J.; Wang, D.; Cai, W. 2011. "The labor supply and retirement behavior of China’s older
workers and elderly in comparative perspective." World Bank Policy Research Working
Paper 5853.
Giles, J.; Wang, D.; Zhao, C. 2010. "Can China’s rural elderly count on support from adult
children?" World Bank Policy Research Working Paper 5510.
Grossman, M. 1972. "On the concept of health capital and the demand for health." Journal of
Political Economy 80 (2):223-255.
Gustman, A.; Steinmeier, T. 2005. "The social security early entitlement age in a structural
model of retirement and wealth." Journal of Public Economics 89:441-463.
119
Huang, W.; Lei, X.; Ridder, G.; Strauss, J.; Zhao, Y. 2013. "Health, height, height shrinkage, and
SES at older ages: evidence from China." American Economic Journal: Applied Economics 5
(2):86-121.
Imbens, G.; Lemieux, T. 2008. "Regression discontinuity designs: a guide to practice." Journal of
Econometrics 142:615-635.
Johnston, D.; Lee, W. 2009. "Retiring to the good life? the short-term effects of retirement on
health." Economics Letters 103 (1):8-11.
Kendel, D. 1991. "Health behavior, health knowledge, and schooling." Journal of Political
Economy 99 (2):287-305.
Kerkhofs, M.; Lindeboom, M.; Theeuwes, J. 1999. "Retirement, financial incentives, and health."
Labour Economics 6:203-227.
Kiecolt-Glaser, J.; Newton, T. 2001. "Marriage and health: his and hers." Psychological Bulletin
127 (4):472-503.
Kim, Y.; Sikoki, B.; Strauss, J.; Witoelarc, F. 2015. "Intergenerational correlations of health among
older adults: empirical evidence from Indonesia." The Journal of the Economics of Ageing
6:44-56.
Kremer, Y. 1985. "The association between health and retirement: self-health assessment of
Israeli retirees." Social Science and Medicine 20 (1):61-66.
Lee, D.; Lemieux, T. 2010. "Regression discontinuity designs in economics." Journal of Economic
Literature 48 (2):281-355.
Lee, J.; Smith, J. 2009. "Work, retirement, and depression." Population Ageing 2:57-71.
Lei, X.; Smith, J.; Sun, X.; Zhao, Y. 2014a. "Gender differences in cognition in China and reasons
for change over time: evidence from CHARLS." The Journal of the Economics of Ageing
4:46-55.
120
Lei, X.; Strauss, J.; Zhang, C.; Zhao, Y. 2015. Health insurance and health care among mid-aged
and older Chinese: evidence from the National Baseline Survey of CHARLS.
Lei, X.; Sun, X.; Strauss, J.; Zhao, Y.; Yang, G.; Hu, P.; Yin, X. 2014b. "Healt outcomes and socio-
economic status among the mid-aged and elderly in China: evidence from the CHARLS
National Baseline Data." The Journal of the Economics of Ageing 3:29-43.
Lei, X.; Tan, L.; Zhao, Y. 2010. "Does retirement affect health." China Economic Quarterly 9
(4):1539-1558.
Lei, X.; Tan, L.; Zhao, Y. 2011. The impact of retirement on health: evidence from China.
Lindeboom, M.; Kerkhofs, M. 2009. "Health and work of the elderly: subjective health measures,
reporting errors and endogeneity in the relationship between health and work." Journal
of Applied Econometrics 24:1024-1046.
Luft, H. 1975. "The impact of poor health on earnings." The Review of Economics and Statistics
57:43-57.
Mazzaonna, F.; Peracchi, F. 2012. "Ageing, cognitive abilities and retirement." European
Economic Review 56:691-710.
McArdle, J. 2010. "Contemporary challenges of longitudinal measurement using HRS data." In
The SAGE Handbook of Measurement. London: SAGE Press.
McGarry, K. 2004. "Health and retirement: do changes in health affect retirement expectations?"
Journal of Human Resources 39 (3624-648).
McKee, D. 2006. "Forward thinking and family support: explaining retirement and old age labor
supply in Indonesia." CCPR Working Paper 005-06.
Mete, C.; Schultz, T. P. 2002. "Health and labor force participation of the elderly in Taiwan."
Yale U Economic Growth Center Discussion Paper 846.
121
Midanik, L.; Soghikian, K.; Ransom, L.; Tekawa, I. 1995. "The effect of retirement on mental
health and health behaviors: the Kaiser Permanent Re-tirement Study." Journal of
Gerontology: Social Sciences 50 (1):559-561.
Mojon-Azzi, S.; Sousa-Poza, A.; Widmer, R. 2007. "The effect of retirement on health: a panel
analysis using data from the Swiss Household Panel." Swiss Med Weekly 137:581-585.
Neuman, K. 2008. "Quit your job and get healthier? the effect of retirement on health." Journal of
Labor Research 29:177-201.
Pang, L.; de Brauw, A.; Rozelle, S. 2004. "Working until you drop: the elderly of rural China."
The China Journal 52:73-94.
Pitt, M.; Rosenzweig, M. 1986. "Agricultural prices, food consumption and the health and
productivity of farmers." In Agricultural Household Models: Extensions, Applications, and
Policy. Baltimore, Md.: Johns Hopkins University.
Radloff, L. 1977. "The CES-D scale: a self-report depression scale for research in the general
population." Applied Psychological Measurement 1:385-401.
Reitzes, D.; Mutran, E.; Fernandez, M. 1996. "Does retirement hurt well-being? factors
influencing self-esteem and depression among retirees and workers." The Gerontologist
36 (5):649-656.
Ridker, P.; Hennekens, C.; Buring, J.; Rifai, N. 2000. "C-reative protein and other markers of
inflammation in the prediction of cardiovascular disease in women." The New England
Journal of Medicine 342:836-843.
Rohwedder, S.; WIllis, R. 2010. "Mental retirement." Journal of Economic Perspectives 24 (1):119-
138.
StateCouncil. 1978. State council temporary measures on workers’ retirement, resignation.
edited by State Council.
122
Stern, S. 1989. "Measuring the effect of disability on labor force participation." Journal of Human
Resources 24:361-395.
Strauss, J. 1986. "Does better nutrition raise farm productivity? ." Journal of Political Economy 94
(2):297-320.
Strauss, J.; Thomas, D. 1994. "Health and labour productivity: sorting out the relationships."
Agricultural Competitiveness: Market Forces and Policy Choice: Proceedings of the
Twenty-Second In-ternational Conference of Agricultural Economists. International
Association of Agricultural Econo-mists Series.
Strauss, J.; Thomas, D. 1995. "Human resources: empirical modeling of household and family
decisions." In Handbook of Development Economics. Amsterdam: North Holland Press.
Strauss, J.; Thomas, D. 1998. "Health, nutrition, and economic development." Journal of Economic
Literature 36 (2):766-817.
Thomas, D.; Frankenberge, E. 2000. "The measurement and interpretation of health in social
surveys." RAND Working Paper.
Tuomi, K.; Jarvinen, E.; Eskelinen, L.; Ilmarinen, J.; Klockars, M. 1999. "Effect of retirement on
health and work ability among municipal employees." Scandinavian Journal of Work,
Environment and Health 17:75-81.
Wagstaff, A.; Lindelow, M.; Gao, J.; Xu, L.; Qian, J. 2009. "Extending health insurance to the
rural population: an impact evaluation of China’s New Co-operative Medical Scheme."
Journal of Health Economics 28 (1):1-19.
Weir, D. 2008. "Elastic powers: the integration of biomarkers into the Health and Retirement
Study." In Biosocial Surveys. Washington, DC: National Academies Press.
Wooldridge, J. 2010. Econometric analysis of cross-section and panel data. Cambridge, MA: MIT
Press.
123
Zhang, H.; Hsu-Hage, B.; Wahlgvis, M. 1998. "The use of knee height to estimate maximum
stature in elderly Chinese." Journal of Nutrition, Health and Aging 2:84-87.
Zhang, X.; Zhao, X.; Harris, A. . 2009. "Chronic diseases and labour force participation in
Australia." Journal of Health Economics 28:91-108.
Zhao, Y.; Crimmins, E.; Hu, P.; Ge, T.; Kim, J.; Strauss, J.; Yang, G.; Yin, X.; Wang, Y. 2014a.
China Health and Retirement Longitudinal Study 2011-2012 National Baseline blood
data users’ guide.
Zhao, Y.; Hu, P.; Smith, J.; Strauss, J.; Yang, G. 2014b. "Cohort profile: the China Health and
Retirement Longitudinal Study (CHARLS)." International Journal of Epidemiology 43:61-
68.
124
Appendix
Table A2.1 Robustness Check: Discontinuity in Retirement Probability
Age 50-70
Age Poly-
nomials
County FE
Formal
Retire-
ment
NBS sta-
tus
(1) (2) (3) (4) (5)
Men
Pension Eligi-
bility 0.238*** 0.211*** 0.232*** 0.281*** 0.316***
[0.0685] [0.0522] [0.0432] [0.0703] [0.0598]
F_stat 12.060 16.310 28.750 16.010 27.980
P_value 0.002 0.000 0.000 0.000 0.000
Women
Pension Eligi-
bility 0.499*** 0.495*** 0.560*** 0.496***
[0.0983] [0.0963] [0.0768] [0.0823]
F_stat 25.760 26.380 53.210 36.390
P_value 0.000 0.000 0.000 0.000
Note: Pension eligible age for men is age 60. Pension eligible age for female civil servant is age 55 and for female
worker is age 50. All the specifications contains an age function, marital status, education levels, household charac-
teristics (# of kids, # of HH members, log PCE) and district fixed effects. F test results of the discontinuity are also
shown. Column (1) presents the specification that further restrict the male sample to respondents between age 50-
70. Column (2) presents a specification that uses cubit age functions. Column (3) shows a specification controlling
for county fixed effects and all the standard errors are clustered at both cohort level and county level. Column (4)
uses a different definition of retirement: formal retirement procedure. Column (5) differentiate individuals into
urban and rural residents according to NBS (National Bureau of Statistics) and focuses on formal sector workers
with urban residences.
Standard errors are in the parenthesis. Standard errors are clustered at both community level and birth cohort
level. For County FE specification, standard errors are clustered at both county level and birth cohort level. ***
p<0.01, ** p<0.05, * p<0.1.
125
Table A2.2 Robustness Check: Impact of Retirement (Men)
Age 50-70
Age Poly-
nomials
County
FE
Formal
Retire-
ment
NBS Sta-
tus
(1) (2) (3) (4) (5)
GHS: good 0.028 -0.098 -0.039 0.018 0.130
[0.2876] [0.2672] [0.2315] [0.1908] [0.1283]
Disability 0.110 -0.095 0.156 0.063 -0.012
[0.1135] [0.1287] [0.1074] [0.1231] [0.1362]
Moderate or Se-
vere Pain 0.527** 0.304** 0.393*** 0.345*** 0.321***
[0.2233] [0.1439] [0.1144] [0.1212] [0.1043]
Life Satisfaction:
good -0.945* -0.731* -0.734** -0.572** -0.403*
[0.4844] [0.4226] [0.3642] [0.2842] [0.2094]
Hypertension 0.252 0.218 0.209 0.145 0.153
[0.4003] [0.3169] [0.3191] [0.2675] [0.2651]
Intact Mental Sta-
tus -0.983 -0.159 -0.213 -0.168 0.746
[1.5269] [1.2549] [0.9682] [0.9667] [1.1647]
Word Recall 0.630 0.654 0.546 0.529 0.386
[1.4799] [0.9779] [0.9360] [0.8135] [0.8193]
Depressive Symp-
toms 4.070 3.888 3.139 2.574 4.091
[4.0074] [3.4751] [2.9049] [3.1519] [2.6059]
Undernourished 0.024 -0.019 0.144 0.092 0.077
[0.2645] [0.2104] [0.1888] [0.1552] [0.0879]
Overweight 0.979 0.565 0.948 0.816** 0.640**
[0.6941] [0.5127] [0.5870] [0.4152] [0.3067]
126
Low Hemoglobin
0.066
0.186
-0.059
-0.041
0.008
[0.4306] [0.3392] [0.3637] [0.3751] [0.1731]
High Glucose 0.472 0.361 0.226 0.199 0.135
[0.8232] [0.6335] [0.4441] [0.5187] [0.2182]
High CRP -0.018 0.201 0.014 0.090 0.249
[0.6720] [0.7076] [0.4741] [0.4731] [0.2228]
Note: 2SLS estimation results for men. All the specifications contains an age function, marital status, education lev-
els, household characteristics (# of kids, # of HH members, log PCE) and district fixed effects. F test results of the
discontinuity are also shown. Column (1) presents the specification that further restrict the male sample to re-
spondents between age 50-70. Column (2) presents a specification that uses cubit age functions. Column (3) shows
a specification controlling for county fixed effects and all the standard errors are clustered at both cohort level and
county level. Column (4) uses a different definition of retirement: formal retirement procedure. Column (5) differ-
entiate individuals into urban and rural residents according to NBS (National Bureau of Statistics) and focuses on
formal sector workers with urban residences.
Standard errors are in the parenthesis. Standard errors are clustered at both community level and birth cohort
level. For County FE specification, standard errors are clustered at both county level and birth cohort level. ***
p<0.01, ** p<0.05, * p<0.1.
127
Table A2.3 Robustness Check: Effect of Retirement on Health (Women)
Age Polyno-
mials
County FE
Formal Re-
tirement
NBS Status
(1) (2) (3) (4)
GHS: good 0.099 0.093 0.085 -0.007
[0.0916] [0.0872] [0.0836] [0.0478]
Disability 0.037 -0.017 -0.003 -0.021
[0.1069] [0.1122] [0.1066] [0.0877]
Moderate or Severe
Pain -0.210* -0.286** -0.240* -0.187*
[0.1166] [0.1317] [0.1232] [0.0975]
Life Satisfaction: good -0.013 0.033 0.008 0.030
[0.1537] [0.1541] [0.1371] [0.1499]
Intact Mental Status -0.281 -0.519 -0.212 -0.412
[0.9677] [1.0461] [0.9486] [0.8911]
Word Recall -1.155* -1.197* -1.027** -0.792**
[0.6241] [0.6487] [0.4756] [0.3331]
Depressive Symptoms 0.604 0.280 0.682 -0.878
[0.9521] [0.8215] [0.7596] [1.1321]
Hypertension 0.120 0.140* 0.138* 0.089
[0.1187] [0.0833] [0.0720] [0.1486]
Undernourished 0.067 0.053 0.089* 0.093*
[0.0632] [0.0542] [0.0515] [0.0556]
Overweight 0.090 0.082 0.092 0.068
[0.2054] [0.1667] [0.1957] [0.1713]
128
Low Hemoglobin
-0.240***
-0.225***
-0.228***
-0.291***
[0.0645] [0.0770] [0.0806] [0.0751]
High Glucose 0.138 0.195** 0.198* 0.200***
[0.1105] [0.0808] [0.1070] [0.0704]
High CRP 0.156 0.088 0.149 0.085
[0.1298] [0.1210] [0.1410] [0.0748]
Note: 2SLS estimation results for women. All the specifications contains an age function, marital status, education
levels, household characteristics (# of kids, # of HH members, log PCE) and district fixed effects. F test results of
the discontinuity are also shown. Column (1) presents a specification that uses cubit age functions. Column (2)
shows a specification controlling for county fixed effects and all the standard errors are clustered at both cohort
level and county level. Column (3) uses a different definition of retirement: formal retirement procedure. Column
(4) differentiate individuals into urban and rural residents according to NBS (National Bureau of Statistics) and
focuses on formal sector workers with urban residences.
Standard errors are in the parenthesis. Standard errors are clustered at both community level and birth cohort
level. *** p<0.01, ** p<0.05, * p<0.1.
129
Table A4.1 Estimation for Initial Employment Status Equation: Continued Employment Decision
Initial Employment Status (=1 Employed; =0 Non-Employment)
Men Women
Health Indicators
w1: poor GHS -0.487*** -0.485*** -0.491*** -0.425*** -0.414*** -0.414***
[0.0594] [0.0596] [0.0604] [0.0795] [0.0790] [0.0790]
w1:Disability -0.427*** -0.429*** -0.432*** -0.255*** -0.260*** -0.259***
[0.1191] [0.1192] [0.1170] [0.0591] [0.0600] [0.0588]
w1: CESD>=10 0.032 0.027 0.017 0.093 0.086 0.082
[0.0714] [0.0726] [0.0766] [0.0710] [0.0684] [0.0685]
w1: Limb Length 0.003 0.003 0.004 -0.009 -0.009 -0.009
[0.0140] [0.0138] [0.0137] [0.0093] [0.0088] [0.0092]
w1: Hypertension 0.033 0.030 0.027 -0.072 -0.072 -0.071
[0.0765] [0.0778] [0.0741] [0.0503] [0.0511] [0.0519]
w1: Overweight -0.276*** -0.264*** -0.259*** -0.241*** -0.233*** -0.236***
[0.0790] [0.0836] [0.0824] [0.0487] [0.0478] [0.0493]
w1: Undernourished -0.050 -0.052 -0.050 -0.029 -0.040 -0.056
[0.1161] [0.1173] [0.1207] [0.0950] [0.0964] [0.0994]
w1: Father Alive 0.185 0.193 0.195 -0.121 -0.124 -0.143*
[0.1740] [0.1746] [0.1761] [0.0869] [0.0829] [0.0821]
w1: Mother Alive 0.098 0.094 0.101 -0.003 -0.010 -0.007
[0.1402] [0.1400] [0.1415] [0.0800] [0.0841] [0.0869]
Socioeconomic Characteristics
Educ_Primary 0.102 0.109 0.128* -0.099 -0.111* -0.117*
[0.0753] [0.0741] [0.0678] [0.0608] [0.0620] [0.0616]
Educ_Secondary -0.181 -0.173 -0.163 -0.187 -0.229* -0.250**
[0.1177] [0.1192] [0.1140] [0.1224] [0.1259] [0.1223]
Married 0.276*** 0.292*** 0.313*** 0.291*** 0.295*** 0.297***
[0.1014] [0.1002] [0.0990] [0.1052] [0.1022] [0.1019]
logPCE -0.059 -0.059 -0.048 -0.115*** -0.121*** -0.124***
[0.0426] [0.0424] [0.0436] [0.0282] [0.0287] [0.0291]
Land Leased -0.000*** -0.000*** -0.000*** -0.001*** -0.001*** -0.001***
[0.0001] [0.0001] [0.0001] [0.0001] [0.0001] [0.0001]
HI Reimbursement Rate -0.101 -0.114 -0.132 -0.075 -0.108 -0.101
[0.2294] [0.2298] [0.2277] [0.2028] [0.1966] [0.1951]
Family Composition
# of Adult Sons -0.026 -0.045
[0.0393] [0.0281]
# of Non-adult Sons 0.078 0.184*
[0.0667] [0.1007]
# of Adult Daughters -0.047 -0.051**
[0.0413] [0.0248]
# of Non-adult Daughters 0.027 0.088
[0.0708] [0.1060]
130
# of Married Adult Sons
-0.048
-0.057*
[0.0461] [0.0301]
# of Married Non-adult Sons -0.009 -0.104
[0.1599] [0.1563]
# of Married Adult Daughters -0.050 -0.060**
[0.0414] [0.0261]
# of Married Non-adult Daughters 0.036 -0.030
[0.1662] [0.1519]
# of Non-married Adult Sons 0.019 0.010
[0.0804] [0.0778]
# of Non-married Non-adult Sons 0.110 0.294***
[0.1112] [0.1087]
# of Non-married Adult Daughters -0.054 0.050
[0.1436] [0.0932]
# of Non-married Non-adult 0.027 0.170
Daughters [0.0851] [0.1520]
Whether has Grandchild -0.109 -0.052 -0.179* -0.128
[0.0731] [0.0817] [0.0916] [0.0920]
N 2,789 2,789 2,789 2,983 2,983 2,983
Mean 0.850 0.850 0.850 0.750 0.750 0.750
P_value: Baseline Health 0.000 0.000 0.000 0.000 0.000 0.000
Note: FIML estimation of initial employment status. Sample is limited to rural residents (hukou) with age range [50-90] at the
baseline. Other covariates include age dummies, provincial fixed effects and urban/rural residency. Standard errors are in the
parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, ** p<0.05, * p<0.1.
131
Table A4.2 Estimation for Initial Employment Status Equation: Labor Market Re-entry Decision
Initial Employment Status (=1 Non-employed; =0 Employed)
Men Women
Health Indicators
w1: poor GHS 0.444*** 0.443*** 0.444*** 0.381*** 0.360*** 0.372***
[0.0612] [0.0608] [0.0601] [0.0770] [0.0748] [0.0791]
w1: Function Limitations 0.347*** 0.353*** 0.351*** 0.248*** 0.264*** 0.253***
[0.1123] [0.1129] [0.1119] [0.0652] [0.0664] [0.0704]
w1: CESD>=10 -0.001 0.003 0.012 -0.068 -0.057 -0.050
[0.0758] [0.0774] [0.0780] [0.0796] [0.0730] [0.0908]
w1: Limb Length 0.001 0.001 0.000 0.006 0.007 0.006
[0.0134] [0.0134] [0.0136] [0.0099] [0.0098] [0.0110]
w1: Hypertension 0.015 0.020 0.019 0.041 0.046 0.043
[0.0832] [0.0840] [0.0828] [0.0551] [0.0553] [0.0672]
w1: Overweight 0.260*** 0.250*** 0.243*** 0.237*** 0.222*** 0.232***
[0.0730] [0.0771] [0.0785] [0.0530] [0.0515] [0.0718]
w1: Undernourished 0.063 0.067 0.071 -0.015 -0.001 0.008
[0.1126] [0.1146] [0.1158] [0.0863] [0.0914] [0.0902]
w1: Father Alive -0.197 -0.202 -0.208 0.087 0.089 0.116
[0.1654] [0.1658] [0.1661] [0.0764] [0.0735] [0.0733]
w1: Mother Alive -0.074 -0.070 -0.075 0.011 0.017 0.012
[0.1288] [0.1289] [0.1300] [0.0778] [0.0829] [0.0885]
Socioeconomic Characteristics
Educ_Primary -0.054 -0.063 -0.062 0.147** 0.146** 0.166**
[0.0817] [0.0810] [0.0814] [0.0630] [0.0667] [0.0676]
Educ_Secondary 0.201* 0.193* 0.191* 0.193 0.233* 0.263**
[0.1091] [0.1117] [0.1110] [0.1257] [0.1316] [0.1289]
Married -0.178* -0.203* -0.206* -0.306*** -0.335*** -0.318***
[0.1075] [0.1077] [0.1094] [0.1006] [0.0928] [0.0986]
logPCE 0.048 0.047 0.046 0.127*** 0.132*** 0.137***
[0.0441] [0.0429] [0.0425] [0.0304] [0.0318] [0.0318]
Land Leased 0.000*** 0.000*** 0.000*** 0.001*** 0.001*** 0.001***
[0.0001] [0.0001] [0.0001] [0.0001] [0.0001] [0.0001]
HI Reimbursement Rate 0.023 0.032 0.021 -0.025 0.008 -0.002
[0.2398] [0.2388] [0.2397] [0.2137] [0.2041] [0.1998]
Family Composition
# of Adult Sons 0.042 0.051*
[0.0330] [0.0302]
# of Non-adult Sons -0.070 -0.207*
[0.0727] [0.1099]
# of Adult Daughters 0.054 0.059***
[0.0419] [0.0207]
# of Non-adult Daughters 0.003 -0.172
[0.0656] [0.1315]
132
# of Married Adult Sons
0.057
0.066**
[0.0392] [0.0333]
# of Married Non-adult Sons 0.041 0.091
[0.1488] [0.1504]
# of Married Adult Daughters 0.055 0.063*
[0.0419] [0.0347]
# of Married Non-adult Daughters -0.024 -0.128
[0.1559] [0.1914]
# of Non-married Adult Sons -0.030 -0.060
[0.0799] [0.0889]
# of Non-married Non-adult Sons -0.107 -0.319***
[0.1173] [0.1222]
# of Non-married Adult Daughters 0.066 -0.097
[0.1403] [0.1054]
# of Non-married Non-adult Daughters 0.019 -0.175
[0.0767] [0.1748]
Whether has Grandchild 0.122 0.102 0.200*** 0.183**
[0.0815] [0.0920] [0.0759] [0.0873]
N 2,887 2,887 2,887 2,987 2,987 2,987
Mean 0.130 0.130 0.130 0.210 0.210 0.210
P_value: Baseline Health 0.000 0.000 0.000 0.000 0.000 0.000
Note: FIML estimation of initial employment status. Sample is limited to rural residents (hukou) with age range [50-90] at the
baseline. Other covariates include age dummies, provincial fixed effects and urban/rural residency. Standard errors are in
the parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, ** p<0.05, * p<0.1.
133
Table A4.3 Continued Employment Decision across Types of Jobs
Employment Transition (=1 Stay Employed; =0 Non-employment)
Agricultural Workers Non-agricultural Workers
Men Women Pooled Sample
w1: poor GHS -0.231** -0.051 -0.662
[0.1041] [0.0753] [0.8060]
w2: poor GHS -0.364*** -0.409*** -0.177
[0.0979] [0.0727] [0.3191]
w1: Disability -0.174 -0.071 0.146
[0.1141] [0.0997] [0.5595]
w2: Disability -0.305*** -0.121 -0.286
[0.1029] [0.0815] [0.2238]
w1: CESD>=10 0.233** 0.162* 0.059
[0.1042] [0.0975] [0.4071]
w2: CESD>=10 -0.042 0.039 -0.156
[0.0928] [0.0705] [0.1850]
w1: Limb Length -0.013 -0.020 -0.019
[0.0120] [0.0153] [0.0410]
w1: Hypertension -0.077 -0.109 0.033
[0.0670] [0.0911] [0.2886]
w1: Overweight -0.035 -0.152* -0.281
[0.1363] [0.0844] [0.2115]
w1: Undernourished -0.303*** 0.213 -0.171
[0.0989] [0.1497] [0.4664]
w1: Father Alive 0.122 0.389** 0.528
[0.1742] [0.1724] [0.7529]
w1: Mother Alive 0.002 -0.180** -0.070
[0.1437] [0.0766] [0.3904]
Female -0.936
[0.8918]
N 2,173 2,717 1,383
Mean 0.890 0.850 0.760
rho 0.660 0.170 0.210
P_value: rho 0.124 0.644 0.933
P_value: All Health 0.000 0.000 0.000
P_value: Lagged Health 0.000 0.000 0.000
P_value: Current Health 0.000 0.000 0.588
P_value: poor GHS
134
ß1+ß0=0 0.000 0.000 0.157
P_value: Disability ß1+ß0=0 0.000 0.018 0.773
P_value: CESD>=10 ß1+ß0=0 0.062 0.046 0.817
Note: FIML estimations of employment transition for agricultural workers and non-agricultural workers employed
at the baseline. All the respondents are rural residents (hukou) with age range [50-90] at the baseline. Other covariates
include age dummies, marital status, educational attainments, logPCE, land leased, county-level health insurance
reimbursement rate, family composition, provincial fixed effects and urban/rural residency. Standard errors are in
the parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, ** p<0.05, * p<0.1.
135
Table A4.4 Estimation for Initial Employment Status Equation across Types of Jobs
Initial Employment Status (=1 Employed; =0 Non-employed)
Agricultural Workers Non-agricultural Workers
VARIABLES Men Women Pooled Sample
w1: poor GHS -0.473*** -0.397*** -0.583***
[0.0633] [0.0843] [0.1117]
w1: Function Limitations -0.425*** -0.281*** -0.280**
[0.1191] [0.0622] [0.1126]
w1: CESD>=10 0.059 0.117* -0.146
[0.0732] [0.0679] [0.1159]
w1: Limb Length -0.000 -0.011 -0.014
[0.0148] [0.0085] [0.0252]
w1: Hypertension 0.045 -0.062 0.065
[0.0776] [0.0504] [0.1256]
w1: Overweight -0.312*** -0.228*** 0.041
[0.0886] [0.0478] [0.1189]
w1: Undernourished -0.039 -0.025 0.079
[0.1218] [0.1010] [0.2057]
w1: Father Alive 0.198 -0.114 -0.270
[0.1926] [0.0803] [0.1777]
w1: Mother Alive 0.086 -0.019 0.140
[0.1485] [0.0896] [0.1532]
Female -0.802***
[0.1412]
N 2,173 2,717 1,383
Mean 0.890 0.850 0.170
Test: Baseline Health 91.360 126.730 184.000
P_value: Baseline Health 0.000 0.000 0.000
Note: FIML estimations of initial employment status for agricultural workers and non-agricultural workers em-
ployed at the baseline. All the respondents are rural residents (hukou) with age range [50-90] at the baseline. Other
covariates include age dummies, marital status, educational attainments, logPCE, land leased, county-level health
insurance reimbursement rate, family composition, provincial fixed effects and urban/rural residency. Standard
errors are in the parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, ** p<0.05, *
p<0.1.
136
Table A4.5 Employment Transition and Interactions of Health with Education
Continued Em-
ployment
Labor Force Re-
entry
Men Women Men Women
w1: poor GHS -0.157 0.002 -0.185 -0.001
[0.1316] [0.0769] [0.2381] [0.2431]
w2: poor GHS -0.412*** -0.390*** -0.822*** -0.136
[0.1056] [0.0764] [0.3010] [0.1536]
w1: Disability -0.111 -0.065 -0.202 0.068
[0.1409] [0.0996] [0.2192] [0.1583]
w2: Disability -0.284*** -0.214*** -0.548*** -0.383*
[0.0941] [0.0695] [0.1698] [0.1978]
w1: CESD>=10 0.218** 0.082 -0.156 -0.216
[0.1051] [0.1267] [0.2116] [0.1799]
w2: CESD>=10 -0.135 0.069 -0.041 0.114
[0.0913] [0.0886] [0.2218] [0.1565]
w1: Limb Length -0.010 -0.023 -0.068** 0.055**
[0.0124] [0.0163] [0.0318] [0.0253]
w1: Hypertension -0.025 -0.055 -0.009 0.157
[0.0753] [0.1034] [0.1711] [0.1214]
w1: Overweight -0.135 -0.125 -0.329 -0.145
[0.1453] [0.0849] [0.2205] [0.2376]
w1: Undernourished -0.369*** 0.024 0.391 -0.219
[0.1256] [0.1679] [0.3009] [0.3177]
w1: Father Alive 0.528** 0.467*** 2.514*** 0.328
[0.2593] [0.1575] [0.8505] [0.2497]
w1: Mother Alive -0.135 -0.221*** 0.070 0.030
[0.1289] [0.0768] [0.4000] [0.1817]
w1: poor GHS X demeaned yrs of Educ -0.012 0.037** -0.042 -0.031
[0.0194] [0.0170] [0.0625] [0.0428]
w1: Disability X demeaned yrs of Educ -0.012 -0.010 0.079 0.010
[0.0234] [0.0210] [0.0528] [0.0409]
w1: CESD>=10 X demeaned yrs of Educ -0.002 -0.013 0.006 0.009
[0.0229] [0.0247] [0.0535] [0.0453]
w1: Limb Length X demeaned yrs of Educ 0.000 -0.003 0.003 0.004
[0.0037] [0.0044] [0.0085] [0.0057]
w1: Hypertension X demeaned yrs of Educ -0.021 0.025 -0.170*** 0.014
[0.0294] [0.0182] [0.0515] [0.0452]
137
w1: Overweight X demeaned yrs of Educ 0.019 0.025 -0.026 0.063**
[0.0297] [0.0240] [0.0763] [0.0313]
w1: Undernourished X demeaned yrs of Educ 0.055 -0.061 -0.102 -0.058
[0.0482] [0.0393] [0.0831] [0.0773]
w1: Father Alive X demeaned yrs of Educ -0.111** 0.038 -0.371** 0.004
[0.0437] [0.0438] [0.1467] [0.0689]
w1: Mother Alive X demeaned yrs of Educ 0.045 -0.041 0.029 0.017
[0.0292] [0.0278] [0.0777] [0.0337]
w2: poor GHS X demeaned yrs of Educ 0.000 -0.032 0.099 0.009
[0.0219] [0.0230] [0.0712] [0.0413]
w2: Disability X demeaned yrs of Educ 0.012 -0.047 0.025 0.023
[0.0204] [0.0295] [0.0578] [0.0404]
w2: CESD>=10 X demeaned yrs of Educ 0.022 0.001 0.029 -0.089**
[0.0289] [0.0244] [0.0543] [0.0375]
Years of Educ -0.027 0.158 -0.150 -0.162
[0.1855] [0.2079] [0.4315] [0.2684]
N 2,787 2,979 2,883 2,983
rho 0.570 0.090 0.620 0.420
P_value: rho 0.289 0.719 0.036 0.625
P_value: All Health 0.000 0.000 0.000 0.001
P_value: Interactions 0.010 0.000 0.000 0.000
Note: FIML estimations of employment transitions for workers employed/not employed at the baseline. All the
respondents are rural residents (hukou) with age range [50-90]. All the specifications control for individual de-
mographics, socioeconomic status and family compositions. Standard errors are in the parenthesis. Standard errors
are robust and clustered at the district level. *** p<0.01, ** p<0.05, * p<0.1.
138
Table A4.6 Continued Employment Decision among Respondents Employed at the Baseline (Probit)
Employment Transition (=1 Stay Employment; =0 Non-employment )
Men Women
Health Changes
w1: poor GHS -0.060 -0.064 -0.073 -0.049 -0.042 -0.039
[0.0823] [0.0820] [0.0827] [0.0596] [0.0594] [0.0587]
w2: poor GHS -0.435*** -0.441*** -0.443*** -0.343*** -0.347*** -0.351***
[0.0660] [0.0685] [0.0682] [0.0585] [0.0600] [0.0596]
w1: Disability -0.014 -0.003 -0.005 -0.046 -0.045 -0.043
[0.1040] [0.1047] [0.1044] [0.0854] [0.0852] [0.0853]
w2: Disability -0.292*** -0.286*** -0.288*** -0.147* -0.145* -0.149*
[0.0872] [0.0900] [0.0921] [0.0767] [0.0787] [0.0787]
w1: CESD>=10 0.197* 0.204** 0.207** 0.116 0.109 0.115
[0.1038] [0.1039] [0.1008] [0.0977] [0.0995] [0.0985]
w2: CESD>=10 -0.094 -0.093 -0.099 0.047 0.058 0.061
[0.0988] [0.0963] [0.0932] [0.0775] [0.0796] [0.0809]
Baseline Health Indicators
w1: Limb Length -0.009 -0.009 -0.010 -0.018 -0.018 -0.019
[0.0130] [0.0130] [0.0132] [0.0121] [0.0126] [0.0127]
w1: Hypertension -0.073 -0.074 -0.064 -0.088 -0.094 -0.097
[0.0778] [0.0760] [0.0737] [0.0898] [0.0930] [0.0930]
w1: Overweight -0.041 -0.029 -0.029 -0.154** -0.152** -0.157**
[0.1168] [0.1174] [0.1170] [0.0638] [0.0657] [0.0660]
w1: Undernourished -0.342*** -0.347*** -0.350*** 0.131 0.132 0.124
[0.1126] [0.1114] [0.1100] [0.1320] [0.1305] [0.1299]
w1: Father Alive 0.093 0.109 0.115 0.458*** 0.458*** 0.456***
[0.1506] [0.1511] [0.1495] [0.1674] [0.1695] [0.1712]
w1: Mother Alive -0.026 -0.033 -0.030 -0.188*** -0.193*** -0.188***
[0.0999] [0.0966] [0.0968] [0.0696] [0.0693] [0.0719]
Socioeconomic Characteristics
Educ_Primary 0.026 0.035 0.032 -0.012 -0.024 -0.025
[0.1144] [0.1210] [0.1202] [0.0934] [0.0921] [0.0925]
Educ_Secondary -0.100 -0.088 -0.098 0.283** 0.254* 0.245*
[0.1507] [0.1589] [0.1581] [0.1391] [0.1400] [0.1371]
Married 0.224* 0.233** 0.222* 0.193* 0.197* 0.193*
[0.1168] [0.1152] [0.1154] [0.1065] [0.1070] [0.1073]
logPCE -0.122** -0.120** -0.127** 0.002 -0.004 -0.005
[0.0508] [0.0498] [0.0498] [0.0678] [0.0683] [0.0677]
Land Leased 0.001*** 0.001*** 0.001*** 0.001* 0.001* 0.001**
[0.0002] [0.0002] [0.0002] [0.0005] [0.0005] [0.0006]
HI Reimbursement Rate -0.387* -0.393* -0.394* -0.505*** -0.531*** -0.527***
[0.2193] [0.2143] [0.2144] [0.1949] [0.1899] [0.1928]
139
Family Composition
# of Adult Sons 0.012 -0.048
[0.0458] [0.0452]
# of Non-adult Sons -0.019 0.013
[0.1066] [0.0969]
# of Adult Daughters -0.067* -0.045*
[0.0395] [0.0261]
# of Non-adult Daughters 0.112 0.213*
[0.1375] [0.1275]
# of Married Adult Sons 0.006 -0.032
[0.0512] [0.0472]
# of Married Non-adult Sons -0.185 -0.079
[0.1371] [0.1574]
# of Married Adult Daughters -0.075* -0.051**
[0.0418] [0.0259]
# of Married Non-adult Daughters -0.036 0.269
[0.1786] [0.2080]
# of Non-married Adult Sons 0.022 -0.124**
[0.0877] [0.0564]
# of Non-married Non-adult Sons 0.041 0.048
[0.1139] [0.0964]
# of Non-married Adult Daughters 0.029 0.112
[0.1199] [0.1142]
# of Non-married Non-adult 0.221 0.175
Daughters [0.1450] [0.1511]
Whether has Grandchild -0.158 -0.106 -0.155* -0.165**
[0.1191] [0.1246] [0.0807] [0.0813]
N 2,378 2,378 2,378 2,249 2,249 2,249
Mean 0.900 0.900 0.900 0.840 0.840 0.840
P_value: All Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: Lagged Health 0.004 0.002 0.003 0.000 0.000 0.000
P_value: Current Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: poor GHS ß1+ß2=0 0.000 0.000 0.000 0.000 0.000 0.000
P_value: Disability ß1+ß2=0 0.000 0.000 0.000 0.017 0.024 0.025
P_value: CESD>=10 ß1+ß2=0 0.289 0.230 0.230 0.084 0.073 0.054
Note: Probit estimation of employment transition for individuals employed at the baseline. Sample restricted to rural residents
(hukou) with age range [50-90] at the baseline. Other covariates include age dummies, provincial fixed effects and urban/rural
residency. Standard errors are in the parenthesis. Standard errors are robust and clustered at provincial level. *** p<0.01, **
p<0.05, * p<0.1.
140
Table A4.7 Labor Market Re-entry Decision among Respondents Not Employed at the Baseline (Probit)
Employment Transition (=1 Re-enter Employment; =0 Stay Non-employ-
ment )
Men Women
Health Changes
w1: poor GHS -0.532*** -0.520*** -0.498*** -0.054 -0.016 -0.031
[0.1677] [0.1865] [0.1783] [0.1436] [0.1479] [0.1490]
w2: poor GHS -0.696*** -0.773*** -0.739*** -0.158 -0.160 -0.165
[0.2239] [0.2414] [0.2302] [0.1285] [0.1381] [0.1369]
w1: Disability -0.238 -0.282 -0.296 -0.071 -0.088 -0.068
[0.2441] [0.2207] [0.2200] [0.1484] [0.1538] [0.1490]
w2: Disability -0.524** -0.468* -0.433* -0.504*** -0.542*** -0.531***
[0.2514] [0.2506] [0.2539] [0.1405] [0.1454] [0.1505]
w1: CESD>=10 -0.325 -0.313 -0.292 -0.187 -0.193 -0.214
[0.2053] [0.1953] [0.1947] [0.1496] [0.1461] [0.1468]
w2: CESD>=10 0.089 0.057 0.068 0.264* 0.266* 0.225
[0.2568] [0.2390] [0.2337] [0.1491] [0.1418] [0.1522]
Baseline Health Indicators
w1: Limb Length -0.058** -0.066** -0.066*** 0.042* 0.041* 0.044**
[0.0258] [0.0261] [0.0255] [0.0222] [0.0215] [0.0226]
w1: Hypertension -0.151 -0.180 -0.169 0.132 0.159 0.157
[0.1893] [0.1911] [0.1943] [0.1195] [0.1206] [0.1198]
w1: Overweight -0.600*** -0.539** -0.545** -0.280** -0.289** -0.304**
[0.1836] [0.2196] [0.2142] [0.1347] [0.1302] [0.1299]
w1: Undernourished 0.220 0.161 0.072 -0.168 -0.190 -0.234
[0.3521] [0.3613] [0.3743] [0.2472] [0.2399] [0.2321]
w1: Father Alive 0.384 0.697** 0.722* 0.289 0.271 0.214
[0.3710] [0.3520] [0.3855] [0.2519] [0.2466] [0.2565]
w1: Mother Alive 0.291 0.214 0.175 -0.031 0.001 0.029
[0.3443] [0.3744] [0.3827] [0.1930] [0.1986] [0.1998]
Socioeconomic Characteristics
Educ_Primary 0.327 0.261 0.235 -0.023 -0.007 -0.015
[0.2227] [0.2246] [0.2196] [0.1693] [0.1717] [0.1708]
Educ_Secondary -0.224 -0.254 -0.225 0.049 0.024 0.020
[0.3305] [0.3319] [0.3160] [0.2783] [0.2629] [0.2614]
Married 0.357* 0.545** 0.535** 0.038 0.057 0.060
[0.1956] [0.2339] [0.2322] [0.1627] [0.1555] [0.1514]
logPCE -0.106 -0.079 -0.087 -0.106* -0.104** -0.114**
[0.0939] [0.0996] [0.0998] [0.0540] [0.0527] [0.0541]
Land Leased 0.000 0.000 0.000 -0.000 -0.000 -0.000
[0.0004] [0.0004] [0.0004] [0.0002] [0.0002] [0.0002]
HI Reimbursement Rate 0.480 0.262 0.098 -0.223 -0.203 -0.139
[0.5147] [0.5514] [0.5558] [0.3988] [0.4269] [0.4112]
141
Family Composition
# of Adult Sons -0.171 0.027
[0.1157] [0.0516]
# of Non-adult Sons 0.363 -0.292
[0.3093] [0.3502]
# of Adult Daughters -0.252** -0.060
[0.0982] [0.0537]
# of Non-adult Daughters 0.393* 0.404*
[0.2061] [0.2142]
# of Married Adult Sons -0.172 -0.012
[0.1096] [0.0558]
# of Married Non-adult Sons 0.613 -1.347*
[0.6979] [0.6944]
# of Married Adult Daughters -0.282*** -0.070
[0.0981] [0.0600]
# of Married Non-adult Daugh-
ters -0.109 0.490
[0.4296] [0.3948]
# of Non-married Adult Sons -0.087 0.265*
[0.2287] [0.1595]
# of Non-married Non-adult
Sons 0.423 0.045
[0.3484] [0.2477]
# of Non-married Adult
Daughters 0.004 0.052
[0.3165] [0.2300]
# of Non-married Non-adult 0.725* 0.322
Daughters [0.3774] [0.3214]
Whether has Grandchild 0.424 0.457 0.092 0.119
[0.2902] [0.2914] [0.2066] [0.1995]
N 361 361 361 642 642 642
Mean 0.370 0.370 0.370 0.300 0.300 0.300
P_value: All Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: Lagged Health 0.000 0.000 0.000 0.242 0.103 0.068
P_value: Current Health 0.000 0.000 0.000 0.000 0.000 0.000
P_value: poor GHS ß1+ß2=0 0.000 0.000 0.000 0.140 0.248 0.199
P_value: Disability ß1+ß2=0 0.016 0.015 0.018 0.003 0.002 0.004
P_value: CESD>=10 ß1+ß2=0 0.370 0.321 0.397 0.727 0.725 0.962
Note: Probit estimation of employment transition for individuals not employed at the baseline. Sample restricted to rural
residents (hukou) with age range [50-90] at the baseline. Other covariates include age dummies, provincial fixed effects and
urban/rural residency. Standard errors are in the parenthesis. Standard errors are robust and clustered at provincial level. ***
p<0.01, ** p<0.05, * p<0.1.
Abstract (if available)
Abstract
This dissertation focuses on the health and labor supply of the aging population in China. Specifically, I explore the interplay between elderly’s health and labor supply behaviors, along with demographics, family structure and economic resources. While health may induce changes in labor supply behaviors, labor market transitions may as well have an impact on the elderly’s health. A thorough understanding of the dynamics between these two elements may have important policy implications
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Three essays on health & aging
PDF
Essays in health economics: evidence from Medicare
PDF
Essays on health and aging with focus on the spillover of human capital
PDF
Essays on health economics
PDF
Essays in empirical health economics
PDF
Three essays on behavioral economics approaches to understanding the implications of mental health stigma
PDF
Essays in development and experimental economics
PDF
Essays in development economics
PDF
Essays on family and labor economics
PDF
Essays on policies to create jobs, improve health, and reduce corruption
PDF
Selection and impacts of early life events on later life outcomes
PDF
Essays on family planning policies
PDF
Three essays on economics of early life health in developing countries
PDF
Hepatitis C in the post-interferon era: selected essays in health economics
PDF
Three essays in education and urban economics
PDF
Essays on human capital accumulation -- health and education
PDF
Essays on development and health economics
PDF
Essays on health insurance programs and policies
PDF
Essays on economics of education and private tutoring
PDF
Three essays in health economics
Asset Metadata
Creator
Zhu, Yaoyao
(author)
Core Title
Three essays on health economics
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
07/01/2016
Defense Date
04/22/2016
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
elderly Chinese,employment transitions,health,health changes,mandatory retirement system,OAI-PMH Harvest,Retirement
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Strauss, John A. (
committee chair
), Crimmins, Eileen (
committee member
), Nugent, Jeff (
committee member
)
Creator Email
yaoyaozh@usc.edu,zhuyaoyao0821@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-259952
Unique identifier
UC11281178
Identifier
etd-ZhuYaoyao-4499.pdf (filename),usctheses-c40-259952 (legacy record id)
Legacy Identifier
etd-ZhuYaoyao-4499.pdf
Dmrecord
259952
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Zhu, Yaoyao
Type
texts
Source
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 a...
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
Tags
elderly Chinese
employment transitions
health changes
mandatory retirement system