University of Southern California Dissertations and Theses

Essays on health economics

(USC Thesis Other)

Essays on health economics

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON HEALTH ECONOMICS
by
Younoh Kim
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2013
Copyright 2013 Younoh Kim
Acknowledgments
First and foremost, I am deeply indebted to my advisor, John Strauss for his excellent
guidance and support. I am very fortunate to have him as my advisor and have an oppor-
tunity to learn from him. He is the most patient, passionate and enthusiastic person I have
ever seen in my life. His passionate and enthusiastic lecture led me to the eld of devel-
opment economics and throughout the program, he has patiently guided me to become a
better researcher. Under his guidance, not only have I learned a lot of valuable economic
knowledge, but I have also learned how to enjoy doing research. All my success in com-
pleting the dissertation and achieving Ph.D. degree would have been impossible without
his encouragement and support.
I would also like to thank Hyungsik Roger Moon for teaching me econometrics patiently
and for providing valuable comments and feedback on my research. He has been a very
good mentor to me. I am very grateful to Jerey Nugent for being on my committee and
supporting me during all this time at USC. Without his help, I would not have had an
opportunity to study at USC.
I want to express my gratitude to Eileen Crimmins and Richard Easterlin for generously
serving on my committee and for giving me insightful feedback. I also want to thank my
co-author, Firman Witoelar. He generously provided data for my job market paper and
has been always willing to give me advice regarding Indonesian Family Life Survey. I owe
ii
special thanks to Heonjae Song, Zara Liaqat, Vlad Radoias, Fei Wang, Martin Weidner,
Jay Gwon, Young Miller, Morgan Ponder and Christopher Frias. My Ph.D. life has been
so much better and joyful with having them around.
Last but not least, I sincerely thank my parents, Namchul Kim and Seunghee Song and
my brother, Youngsoo Kim for their unconditional love and support. They always believed
in me even when I doubted myself and provided numerous encouragements. Their sacrices
allowed me to achieve everything in my life. I will always be indebted and grateful to them
for the rest of my life.
iii
Table of Contents
Acknowledgments ii
List of Tables vi
Abstract ix
Chapter 1: Introduction 1
Chapter 2: Intergenerational Correlations of Health Among Older Adults 5
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Studies of Health Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Data and Empirical Specication . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 Indonesia Family Life Survey (IFLS) . . . . . . . . . . . . . . . . . . 12
2.3.2 Empirical specication . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Intergenerational Correlations . . . . . . . . . . . . . . . . . . . . . . 20
2.4.2 Interactions with birth region . . . . . . . . . . . . . . . . . . . . . . 39
2.4.3 Respondent's height and schooling . . . . . . . . . . . . . . . . . . . 40
2.4.4 Health and SES gradients of older adults . . . . . . . . . . . . . . . 55
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Chapter 3: The Dynamics of Health and Its Determinants among Older Adults 80
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.2 Studies of Health Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4 Data and Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.5 Empirical Specication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.6.1 The Eects of Lagged Health Measures . . . . . . . . . . . . . . . . 109
3.6.2 The Eects of Other Co-variates . . . . . . . . . . . . . . . . . . . . 122
3.6.3 Interactions of Lagged Health with Age and SES . . . . . . . . . . . 123
iv
3.6.4 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Chapter 4: Conclusion 130
Bibliography 132
Appendix A 139
Appendix B 146
v
List of Tables
2.1.1 Parental SES Gradients of Health of Older Adults 1 . . . . . . . . . . . . . . . 21
2.1.2 Parental SES Gradients of Health of Older Adults 2 . . . . . . . . . . . . . . . . 23
2.1.3 Parental SES Gradients of Health of Older Adults 3 . . . . . . . . . . . . . . . . 25
2.1.4 Parental SES Gradients of Health of Older Adults 4 . . . . . . . . . . . . . . . . 27
2.1.5 Parental SES Gradients of Health of Older Adults 5 . . . . . . . . . . . . . . . . 29
2.1.6 Parental SES Gradients of Health of Older Adults 6 . . . . . . . . . . . . . . . . 31
2.2.1 Parental SES Gradients of Change in Health of Older Adults 1 . . . . . . . . . 35
2.2.2 Parental SES Gradients of Change in Health of Older Adults 2 . . . . . . . . . 37
2.3.1 Parental SES Gradients of Health of Older Adults: Interaction with Birth
Region 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.2 Parental SES Gradients of Health of Older Adults: Interaction with Birth
Region 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.3 Parental SES Gradients of Health of Older Adults: Interaction with Birth
Region 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.4 Parental SES Gradients of Health of Older Adults: Interaction with Birth
Region 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.4 Parental SES Gradients of Respondents Schooling and Height . . . . . . . . . . 53
2.5.1 Health-SES Gradients of Older Adults 1 . . . . . . . . . . . . . . . . . . . . . 56
vi
2.5.2 Health-SES Gradients of Older Adults 2 . . . . . . . . . . . . . . . . . . . . . 58
2.5.3 Health-SES Gradients of Older Adults 3 . . . . . . . . . . . . . . . . . . . . . 60
2.5.4 Health-SES Gradients of Older Adults 4 . . . . . . . . . . . . . . . . . . . . . 62
2.5.5 Health-SES Gradients of Older Adults 5 . . . . . . . . . . . . . . . . . . . . . 64
2.5.6 Health-SES Gradients of Older Adults 6 . . . . . . . . . . . . . . . . . . . . . . 66
2.5.7 Health-SES Gradients of Older Adults 7 . . . . . . . . . . . . . . . . . . . . . . 68
2.5.8 Health-SES Gradients of Older Adults 8 . . . . . . . . . . . . . . . . . . . . . . 70
2.5.9 Health-SES Gradients of Older Adults 9 . . . . . . . . . . . . . . . . . . . . . . 72
2.5.10 Health-SES Gradients of Older Adults 10 . . . . . . . . . . . . . . . . . . . . . 74
2.5.11 Health-SES Gradients of Older Adults 11 . . . . . . . . . . . . . . . . . . . . . 76
3.1.1 Transition Matrix - Hypertension . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.1.2 Transition Matrix - Low Hemoglobin . . . . . . . . . . . . . . . . . . . . . . . . 96
3.1.3 Transition Matrix - General Health Status (Poor Health = 1) . . . . . . . . . . 97
3.1.4 Transition Matrix - Underweight (BMI < 18.5) . . . . . . . . . . . . . . . . . . . 98
3.1.5 Transition Matrix - Overweight (BMI 25) . . . . . . . . . . . . . . . . . . . . 99
3.2.1 Summary Statistics 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.2.2 Summary Statistics 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.2.3 Summary Statistics 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.3.1 Dynamic Health Demand Function Estimation - Low Hemoglobin . . . . . . . 111
3.3.2 Dynamic Health Demand Function Estimation - Hemoglobin Level . . . . . . 112
3.4 Dynamic Health Demand Function Estimation - Hypertension . . . . . . . . . . 113
3.5 Dynamic Health Demand Function Estimation - Lung Capacity . . . . . . . . . 114
3.6.1 Dynamic Health Demand Function Estimation - Overweight . . . . . . . . . . . 115
3.6.2 Dynamic Health Demand Function Estimation - Underweight . . . . . . . . . . 116
3.6.3 Dynamic Health Demand Function Estimation - BMI Level . . . . . . . . . . . 117
3.7 Dynamic Health Demand Function Estimation - GHS (Poor Health) . . . . . 118
vii
3.8 Dynamic Health Demand Function Estimation - Number of Diculties with
ADLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.9.1 Disaggregated Groups 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.9.2 Disaggregated Groups 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.9.3 Disaggregated Groups 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
A.1.1 Mean and Standard Deviation of Variables . . . . . . . . . . . . . . . . .. . 141
A.1.2 Mean and Standard Deviation of Variables . . . . . . . . . . . . . . . . .. . 143
A.2.1 The Distribution of Father's Death and GHS . . . . . . . . . . . . . . . . . . 145
A.2.2 The Distribution of Mother's Death and GHS . . . . . . . . . . . . . . . . . . 145
B.1.1 FD-GMM: In the Presence of Time-Varying Measurement Errors 1 . . . . . . 147
B.1.2 FD-GMM: In the Presence of Time-Varying Measurement Errors 2 . . . . . . 147
B.2.1 Determinants of Sample Attrition 1 . . . . . . . . . . . . . . . . . . . . . . . . . 148
B.2.2 Determinants of Sample Attrition 2 . . . . . . . . . . . . . . . . . . . . . . . . . 149
B.3.1 First-stage Regression for Low Hemoglobin . . . . . . . . . . . . . . . . . . . . . 150
B.3.2 First-stage Regression for Hypertension . . . . . . . . . . . . . . . . . . . . . . . 151
viii
Abstract
This dissertation contributes to the design of better policy implications and improvements
of well-being among older adults, especially when resources are limited. This is done by
using proper econometric methods and taking advantage of the richness of the Indonesian
Family Life Survey, a panel data set containing detailed information for both respondents
and their biological parents.
Chapter 2 analyzes the health transmission from parents to their children when they
become older adults. I match the health status of older respondents to parental character-
istics (several health measures and education) to see if any correlations exist in Indonesia.
I nd that strong intergenerational correlations exist. For example, children having par-
ents with more diculties with ADLs are more likely to have the same problem as older
adults. However, surprisingly, the magnitude of correlations becomes signicantly lower
for those born in richer areas of Indonesia such as Java and Bali. This suggests that the
level of development at birth or early childhood, which may include having better health
infrastructure, substitutes for the in
uence of parental health and hence highlights the
importance of public policies that focus on community level infrastructure development in
less developed areas, in order to solve health inequality.
ix
Chapter 3 examines the determinants of chronic health conditions and explains their
persistence. I incorporate dynamics into a health demand function, nding strong cor-
relations between lagged and current health measures when nothing else is controlled.
This could represent the in
uence of lagged health or xed unobserved factors such as
genetic endowments and childhood health. To disentangle these, I estimate the in
u-
ence of lagged health by using rst-dierence two-step generalized method of moments
(FD-GMM), where the rst-dierencing removes xed unobserved factors and keeps only
lagged health. I found that it is this xed eect, representing both genetic endowments and
early life cycle including childhood health, that is most important in explaining later life
chronic conditions. The impact of past health conditioning on the xed eect, captured by
the coecients on lagged health measures, is weak, with estimated coecients relatively
close to zero. These results are robust to potential measurement errors in health and to
sample attrition. Socio-economic status also has very little in
uence on current health,
again conditioned on the xed eect and on the in
uence of lagged health. In order to
investigate if past health has dierent impacts across demographic or economic groups,
I disaggregate the sample across age, household per capita expenditure level (PCE), and
years of education. The results show that those with less education tend to show more
persistency, compared to those with higher education.
In developing countries like Indonesia, health disparities are serious issues since they
persist over generations due to the lack of proper interventions and also prevail among
disadvantaged groups such as those with less educations. My dissertation suggests more
eective and ecient ways to employ interventions and resolve health disparities, using
robust econometric technique and rich data sets.
x
Chapter 1
Introduction
According to the United Nations, a society is called Aging Society when the percentage of
population over 65 years old is more than 7 percent and Aged society when it exceeds 14
percent. Over the decades, the proportion of elderly has substantially increased and many
developed countries became either aging societies or aged societies. This phenomenon is
even more problematic in many developing countries, where aging has an unprecedentedly
rapid pace, and those countries do not seem to be ready to embrace the elderly. Their
public social security systems are not as strong as the ones in developed countries. People
mainly count on family support after retirement and so, the quality of their life is much
more sensitive to external shocks. Unfortunately, most studies on health in developing
countries have focused on childhood health, while health and quality of life for elderly
have not received much attention. This is especially troublesome since many times, due to
limited resources, priority is given solely to children oriented policy, while elderly quality
of life is assumed to be unimportant and ignored.
The objective of this dissertation is to contribute to the design of better policy impli-
cations and improvements of well-being among older adults, especially when resources are
limited. This is done by using proper econometric methods and taking advantage of the
richness of the Indonesian Family Life Survey, a panel data set containing detailed infor-
mation for both respondents and their biological parents.
In chapter 2, I analyze the health transmission from parents to their children when they
become older adults. Even if people have been exposed to infectious diseases during their
1
youth, some might develop more serious problems in the long run. For instance, those who
could not get proper care might have suered from early-life shocks throughout their life,
even as older adults. In that sense, if parental health is associated with the health of their
children, the current health disparities could persist for generations and especially so for
those with insucient access to medical care, which is common in developing countries.
Using IFLS, I match the health status of older respondents to parental characteristics (sev-
eral health measures and education) to see if any correlations exist in Indonesia. From a
public policy standpoint, it is important to identify the extent of health transmission since
the policies that impact the current generation might have additional benets for future
generations as well, and if so, this should be accounted for. I nd that strong intergenera-
tional correlations exist. For example, children having parents with more diculties with
ADLs are more likely to have the same problem as older adults. However, surprisingly,
the magnitude of correlations becomes signicantly lower for those born in richer areas of
Indonesia such as Java and Bali. This suggests that the level of development at birth or
early childhood, which may include having better health infrastructure, substitutes for the
in
uence of parental health and hence highlights the importance of public policies that
focus on community level infrastructure development in less developed areas, in order to
solve health inequality.
In chapter 3, I investigate the determinants of chronic health conditions and their
persistence, using IFLS. There has been a large and growing literature on early life origins
of health as an adult, showing strong correlations between early childhood health and
adulthood health. However, very little is known about the dynamics of health at older
ages. Once considered as diseases of the rich, chronic health conditions are now the leading
causes of death and disability in many low and middle income countries. Being more
prevalent among the elderly, chronic conditions are very closely related to their quality
of life. This is even more so in developing countries, where people age at much lower
2
income levels and do not have sucient access to medical care. For instance, Witoelar et
al.(2012) nd that hypertension is the most serious chronic disease in Indonesia as more
than half of the elderly are found to be hypertensive. However, only less than 25% of those
with hypertension are aware that they have contracted the disease since they have never
been diagnosed by doctors. Moreover, chronic conditions, dened as conditions that last 3
months or longer, are persistent and have long-lasting impacts. Using Indonesian Family
Life Survey (IFLS), I found that among those with hypertension in the past, 80% continue
to suer from it in the subsequent period, when nothing else is controlled for. The question
is what can explain this persistence in chronic conditions? Is it childhood health that has
permanent eects, which would be consistent with the early life origin hypothesis, or are
there other factors causing this persistence? Carefully analyzing these determinants could
possibly mitigate some of the negative eects associated with chronic conditions.
Six health measures known to be important for elderly health are chosen and care-
fully analyzed. They are hypertension, hemoglobin level, lung capacity, body mass index,
diculties with activities of daily living, and self reported general health status. The
possible sources causing the persistence could be genetic endowments, childhood health,
socio-economic status, health behavior, and the past history of chronic health conditions.
To disentangle these in
uences, I employ a rst-dierence two-step generalized method
of moments(FD-GMM), where the rst-dierencing removes xed eects, and keeps only
lagged health measures and time-varying characteristics. I found that it is genetic endow-
ments and early life cycle including childhood health, captured in the individual xed
eects, that explain most of the persistence in chronic health conditions. Interestingly, the
impact of past health conditions, captured in the coecients on lagged health measures, is
weak, conditional on xed eects. Socio-economic status also has very little in
uence on
current health, conditional on xed eects and lagged health status. These results suggest
that the persistence of chronic conditions are mostly explained by genetic endowments and
3
childhood health, captured in the xed eects while the past history of having the disease
does not have strong impact on continuing having the disease in the subsequent periods.
Additionally, I looked at whether past health conditions could have dierential impacts
across socio-economic or demographic status. In order to do so, I disaggregate the groups
by age, per capita household expenditure, and years of education and nd that people with
less education are less likely to recover from past bad health conditions. Given a strong
impact of childhood health and a stronger persistence of chronic conditions among less
educated elderly, my results suggest that policy makers should make investments targeting
the early stages of life. These investments should not be made only towards health, but
also towards better education in early life.
In developing countries like Indonesia, health disparities are serious issues since they
persist over generations due to the lack of proper interventions and also prevail among
disadvantaged groups such as those with less educations. My dissertation suggests more
eective and ecient ways to employ interventions and resolve health disparities, using
robust econometric technique and rich data sets.
4
Chapter 2
Intergenerational Correlations of
Health Among Older Adults
2.1 Introduction
It is widely believed that family background has a signicant in
uence on children's life.
For instance, Bowles et al. (2002) show that economic status is transmitted from parents
to ospring and moreover, the extent of intergenerational transmission of economic status
is considerably greater than what people generally thought it to be a generation ago.
The vast majority of the existent literature has focused on the relationship between
parents' education and income and the education and income of their children. Surprisingly,
however, much less work has been done on the intergenerational transmission of health.
Health is an important part of human capital. Better health makes people more productive,
and in turn may increase future earnings whereas poorer health causes low productivity,
lower happiness and more expenditure on medical care, leading to reduced income and
less opportunities for wealth accumulation. Therefore, it seems reasonable to extend our
research interest towards dimensions of health.
The main objective of this chapter is to examine the correlations of health across
generations using the Indonesia Family Life Survey (IFLS). The IFLS is a panel survey
covering 14 years from 1993 to 2007 and collects extensive information at the individual, the
household, and the community level, including indicators of economic and non-economic
5
well-being. In particular, the survey contains a rich set of information on health outcomes
of respondents, including both biomarkers and self-reports. IFLS is a well suited data
set for this study because it includes detailed information about parents even if they live
apart from their children and the information is collected either at the time of the survey
or just prior to death if they are dead. IFLS thus allows us to capture the latest health
information of each parent. These parental health variables, together with measures of
parent's education, are used as covariates to explore the intergenerational correlations of
health with health measures of older respondents, while controlling for age and birth district
of the respondent.
I take advantage of the richness of IFLS and examine several health measures of respon-
dents, including self-reports and biomarkers: a measure of self-reported general health sta-
tus; the number of measures of physical function and activities of daily living (ADLs) that
the respondent reports having diculty in conducting; the number of instrumental activ-
ities of daily living (IADLs) the respondent reports having diculty with; a measure of
cognition measured by word recall; hemoglobin; total and HDL cholesterol; hypertension;
an index of depression (the 10 question CES-D) and body mass index (BMI).
As measures of health of both parents, IFLS has information on whether they are dead
at the time of the last wave in 2007, their general health status and whether they have
diculties with any ADLs at the time of the survey or just before death.
To focus on older adults, the sample is restricted to respondents who are 50 years and
older in 2007. I use multivariate analysis in two ways. First, a cross-sectional analysis is
employed by using the information from IFLS4; this allows us to investigate the maximum
number of health outcomes. Dependent variables, in this case, are the measures of respon-
dent health status measured in 2007. Second, a simple growth model is used with changes
in a restricted number of health measures from 1993 to 2007 as outcome variables (changes
between 1997 and 2007 are used in the cases for which 1993 data are not available but 1997
6
data are). These growth or change regressions are estimated for respondents who were 50
and above in 2007 and interviewed for both 1993 (or 1997) and 2007.
Having parental health variables and schooling as right-hand side variables along with
respondent's age at baseline enables us to look at the intergenerational correlations with
the levels of health measures as well as for the changes in health. I am careful not to
interpret these relationships as necessarily causal, because there exist the usual issues
of omitted variables and possibly measurement error in parental health. Thus I cannot
identify the exact pathways that may explain these correlations. If an elderly parent is
still alive, for instance, this is an indication that parent has had good health, which may
well have indeed been transmitted to the respondent. However, many other factors may
be associated with this as well, such as a good health and nutrition environment when the
respondent was young or good health behaviors of the respondent as a child and as an
adult, which may partly have been in
uenced by health behaviors of the parent. On the
other hand, a parent having survived to 2007 also will be correlated with high levels of
SES of the parent, which may have dierent eects on respondent health. Still, given the
dearth of estimates of intergenerational correlations of health, these ndings can make a
useful rst step contribution to the literature.
The ndings suggest strong intergenerational correlations between the measures of
parental health, schooling, and the health of their adult children. For example, if par-
ents had more diculties with ADLs, their children are more likely to have the same
problem when they become older adults. Having a dead father is associated with increases
in the number of ADLs and IADLs that women report having problems with, a higher
likelihood of being underweight for women, as well as with lowered cognition for women.
Having a dead mother is correlated with a greater likelihood of having hypertension and
being underweight for both men and women, having hemoglobin level below the threshold
for men, and also with reporting poor health for women.
7
The health correlations are stronger in magnitude for the cross-sectional analysis using
the 2007 wave than are the changes between 1993 (or 1997) and 2007. This suggests that
the intergenerational in
uences are already established by the time the respondents are 36
years and over in 1993 (or 40 and over in 1997).
I also examine how these intergenerational correlations might change for respondents
born in the more developed parts of Indonesia compared to the less developed areas. Cur-
rie and Moretti (2007) have found weaker intergenerational correlations of health in the
United States for parents born in better o areas. Bhalotra and Rawlings (2009) have
found a similar result using cross-country evidence. Interestingly, these health associations
are much lower for those born in Java or Bali. These are areas of Indonesia that have
experienced the most rapid economic growth over the past 40 years (Dick et al 2002).
This suggests that being born and growing up in developed areas, which may have better
health infrastructure, substitutes for the in
uence of parental health. Finally, I examine
the relationship between concurrent SES factors of the respondent, and health, and how
those relationships change when adding the parental health and schooling factors. I nd
very similar relationships between better SES and better health as are usually found, which
largely remain after controlling for parental health and schooling.
The rest of the chapter is organized as follows. Section 2.2 provides a brief review of the
related literature. Data description and the empirical specication used are described in
section 2.3. The main regression results are discussed in section 2.4. Concluding remarks
follow in section 2.5.
8
2.2 Studies of Health Correlation
Although there are numerous studies which analyze the intergenerational correlation of
earnings, wealth or education, a limited number of studies exist that examine intergenera-
tional correlations of health. Most of this research has concentrated on the impact of early
childhood health or even fetal health on health later in life. A leading theory to explain
the association between one's health in early life and later health has been the \Barker
hypothesis". According to this theory, organ sizes or function, gene expression as well as
metabolism may adapt to a new environment in order to raise survival probabilities, when
faced with negative shocks or alterations in nutrition during very early childhood or the
fetal period. This adaptation might be benecial for short-term survival during famine but
can cause health problems later in life. (Barker et al., 1989; Fogel, 2004; Komlos, 1994)
Godfrey and Barker (2000) show that several of the major diseases of later life, includ-
ing coronary heart disease, hypertension, and type II diabetes are correlated with under-
nutrition during the fetal period. In particular, longitudinal studies of 25,000 British
subjects found evidence that birth size is highly associated with disease occurrence in later
life. People born small or disproportionate like having a big head with short arms seem to
have a higher likelihood of having coronary heart disease, high blood pressure, high choles-
terol concentrations, and abnormal glucose-insulin metabolism. This paper also suggests
that the timing of nutrition insults during pregnancy is important. For instance, people
tend to have higher risk of having hardened arteries in their mid-life if their mother had
poor nutrition during the period when arteries form in the fetus. As is well known, this
and related studies might not re
ect the true correlation of health transmission since other
factors such as one's lifestyle as an adult (smoking and drinking behaviors) or shocks in
early life (exposure to natural disasters, bad weather or economic crisis) can aect later
health as well.
9
Another theory of association between early and later health is proposed by Crim-
mins and Finch (2004). The \cohort morbidity phenotype hypothesis", a complementary
theory to the Barker hypothesis explains that in
ammation caused by infectious diseases
in early life increases the risk of later life morbidity and mortality. In particular, inva-
sion of pathogen or internal tissue injury during early childhood may induce in
ammatory
responses and the high levels of in
ammation lead to the development of atherosclero-
sis, which in turn possibly causes cardiovascular disease. They also suggest that retarded
growth in the birth period can be interpreted as a fetal adaptation to mal-nutrition, con-
sistent with the Barker hypothesis, as well as with the consequences of being exposed to
infection.
Using aggregated data by birth cohort from four northern European countries before
the 20th century, Crimmins and Finch (2006) examine the impact of early mortality on
later mortality. Specically, they regress the changes in mortality between age 70 and
74 years on the changes in mortality at dierent states of childhood simultaneously for
the same birth cohort and nd signicant correlations for all countries. According to the
authors, several infectious diseases were the main cause of infant and childhood mortality
whereas most of adult mortality was caused by chronic diseases, especially heart disease
at that time. Hence, strong correlations between early and later mortality rates can be
interpreted as the link between childhood in
ammations and heart problem in later life.
James P. Smith et al. (2009) use the China Health and Retirement Longitudinal Study
(CHARLS) to look at the correlations between childhood general health status before age
16 (retrospectively obtained), adult height (used as a measure of childhood height, and
health) and adult health outcomes. Later outcomes include GHS, the number of ADLs
and separately IADLSs that respondents report having diculty in performing, subjective
expectation of mortality, low and high BMI, cognition, depression, health behaviors such
as smoking, drinking and physical activity, and SES measures such as education, household
10
per capita expenditure and wealth. They nd that women's later health status is strongly
correlated with childhood general health, while men's BMI and mortality expectation are as
well. In an earlier study with US data, Smith (2009) using the PSID, examined the impact
of prospective childhood health on adult SES outcomes, including levels and trajectories of
education, family income, household wealth, individual earnings and labor supply. Smith's
analysis was conducted with a panel who were originally children and are now well into
adulthood. With the exception of education, Smith reports that poor childhood health has
a large eect on all outcomes with estimated eects larger when unobserved family eects
are controlled for using family xed eects.
The point is that respondents' health in very early childhood is strongly correlated
with parental characteristics, which represents an indirect link between the socio-economic
(SES) characteristics of parents and the health of their children later in life through their
children's health in early life stages. Direct evidence on the links between respondent health
as a child and health of the parents when they were children exists, but is not abundant.
Using data from British National Child Development Study in 1958, Emanuel et al. (1992)
demonstrate that infant's birth weight is positively correlated with mother's birth and non-
pregnant weight. Thomas et al. (1990) show that mother's height is positively correlated
with child survival in Brazil, controlling for mother and father schooling and household
resources. Almond and Chay (2006) use dierence in dierence regressions to compare
maternal health and birth outcomes for black and white women born in the late 1960s
to those born in the early 1960s. They suggest that due to the federal antidiscrimination
eort, black women born in the late 1960s are healthier and in turn, they are less likely
to deliver babies with low birth weight and low APGAR scores as compared to those born
earlier.
However, direct evidence on the links between health of older adults and health of
their parents when they were older is very scarce. Few papers discussed above, show that
11
intergenerational correlations exist but it is still uncertain as to whether these impact will
prevail even when their children become older adults. A recent study by Maccini and Yang
(2009), for instance, examines the in
uence of weather conditions at birth time on health,
schooling completed and socioeconomic outcomes in later life. Specically, adult outcomes
from Indonesia Family Life Survey are linked to historical rainfall data which were specied
for the individual's birth-year and birth-district. Their results nd that higher rainfall in
early life has a large positive eect on adult outcomes for women but not for men. They
do not look at the question of this chapter: whether parental health status continues to be
associated with children's health even when the children become old.
2.3 Data and Empirical Specication
2.3.1 Indonesia Family Life Survey (IFLS)
This chapter uses the data from the 1993, 1997, 2000 and 2007 waves of the Indone-
sia Family Life Survey (IFLS). This is a large-scale socio-economic survey conducted in
Indonesia which contains extensive information collected at the individual, the household
and the community levels. The survey includes not only indicators of economic but also
non-economic well-being such as consumption, income, education, assets, migration, fer-
tility, use of health care, health insurance, marriage, kinship among family members and
labor market outcomes (see Strauss et al., 2009).
IFLS ts the purpose of this chapter since it collects a rich set of information on health
outcomes including biomarkers and self reports for both respondents and their parents.
IFLS contains detailed information of parental health such as whether they had ADL
problems and they were in a poor health condition before their death if they are dead, or
at the time of the survey.
12
For the growth model, changes in health measures from 1993 to 2007 (or 1997 to 2007,
depending on health measures) are used as dependent variables and respondents who are
available in both wave 1 and wave 4 are chosen in the sample. Any longitudinal study
like this comes with a potential worry: sample attrition. Fortunately, the attrition rate
in IFLS is very low compared to other panel data sets. In particular, 7,224 households
were interviewed and detailed individual level information was collected from over 22,000
respondents in IFLS1, conducted in 1993. The re-contact rate was 93.6% of original IFLS
1 households in IFLS 4. Overall, among IFLS 1 original respondents over age 15 in 1993
who were still alive, 88% of them were re-contacted in IFLS4. Among age groups, the
highest re-contact rates (over 90%) are for those who were older than 40 years in 1993 (see
Thomas et al., forthcoming, for details).
In examining the relationship between parental health and their adult children's health,
multivariate regression is used in two ways; cross-sectional analysis and a growth model. In
both sets of estimates, parental characteristics are treated as time-invariant characteristics.
This chapter focuses on adults who are older than 50 years and it means their parents are at
least 65 years and older in the sample. Health information from the 2007 wave is available
for parents who are still alive and even if they died, IFLS collects the information as of just
before their death. If respondents do not live together with their parents, it is respondents
(adult children) who are interviewed about the health status of their biological mother and
father (see appendix A for the questions). For parents who live together with their children,
parents are directly interviewed about their health status. If they are alive in 2007, the
health information is from IFLS4 but if they died between the surveys, the information
comes from the last wave they are found alive.
1
1
For co-resident parents, if they died in 1996, the health information from IFLS2(1997) is used and if
died before 1993, IFLS 1(1993) is used. However, for non-coresident parents, their information comes from
IFLS4(2007) because their adult children are asked about biological parents' health status now or before
death.
13
Specically, dummy variables are created for being dead in 2007, diculties with ADLs
and general health status (GHS) at the survey, or just before death. The parents' death
dummy variables, one for each biological parent, are equal to 1 if the parent was dead at
the time of IFLS 4, in 2007. In the sample, only 5% of the fathers and 16% of the mothers
were still alive at the time (see table A.1 in appendix), so this dummy variable indicates a
particularly healthy parent if it is 0. I also know the date of death for many of those who
died and hence I tried dummies for death before age 60, death after age 60, and died but
age of death missing. These turned out not to be signicantly dierent from each other,
so in the main specications I use the dummies for death.
2
A dummy variable for a measure of general health status of each parent is also con-
structed. It equals 1 if the parent is reported to be in poor heath in 2007 or right before
their death if they are dead; about half for both mothers and fathers (see table A.1). For
diculties with ADLs, the dummy variable takes value 1 if the parent experienced prob-
lems with any ADL in 2007 or before they died; about one-quarter for both mothers and
fathers. The level of schooling of each parent is controlled by creating dummy variables
for each level completed: primary and junior high school and above, no schooling being
the omitted category. About 45% of fathers are reported to have had no schooling and
about 60% of mothers (see table A1). A little less than 20% of mothers are reported to
have completed primary school or more, while about 30% of fathers have.
One might imagine that health just prior to death is worse and not necessarily repre-
sentative of health earlier in life. This appears to be true in the sample. Tables A.2.1 and
A.2.2 show the distribution of the GHS variable separately for mothers and fathers who
are still alive and those dead. The distributions are dierent, worse for those parents who
2
The F-statistic for equality of the three father death coecients jointly with the separate equality of
the three mother death coecients in the poor general health regressions are 1.26 (p-value: 0.28) for women
and 0.24 (p-value: 0.91) for men. Results for other equations are similar.
14
are dead (a chi-square statistic of dierences are 40.7 for fathers and 47.5 for mothers; both
are signicant at under .01). Because of these dierences, I allowed for in the empirical
specications, interactions between the mother (or father) dead variable with the mother
(or father) GHS variable. It turns out, however, that these interactions are not jointly
signicant, so they are not reported in the main specications.
3
As mentioned above, several health measures which are known to be very important
for elderly health are used as dependent variables. The rst one captures measures of
physical functioning and activities of daily living (ADLs). It is dened as routine activities
that people tend to do every day such as eating, bathing, dressing, toileting, transferring,
and continence. Listed in the questionnaire are also physical functioning activities such
as carrying a heavy load for 20m, walking for 5km and standing from sitting from the

oor without help. In IFLS, respondents are asked whether they can do those activities
related to ADLs or physical functioning without any help or diculties. For this analysis,
each answer is recorded as 1 if respondents report that they can do them only with some
assistance or not able to do it. In the regression, the sum of the number of diculties with
ADLs is used as an outcome variable; the maximum number of ADL problems that each
respondent can have is 9. As shown in table A.1, the mean is 1.79 for women and 1.04 for
men.
The second health measure used in the analysis is instrumental activities of daily living
(IADLs). While it is not necessary for fundamental functioning, it is still required to be
able to live one's life independently. In the questionnaire, respondents answer if they can
do particular activities related to IADLs without any diculties. To shop for personal
needs, to prepare one's own meal, to take a medicine and to travel are some examples.
3
The F-statistic for the interactions of mother and father deaths with the mother or father having poor
general health in the male poor general health regression is 0.72 (p-value, 0.49). Again, results are similar
for other equations.
15
Similar to the case of ADLs, each answer is scored as 1 for those who answer that they
need help or cannot do any of those activities. Like before, the sum of these values is used
in the regression; the means are 1.0 for women and 0.55 for men (see table A.1).
General health status (GHS) is also one of the health measures examined. It is scored
as very healthy, somewhat healthy, somewhat unhealthy or unhealthy. For the analysis, a
value of 1 is scored if respondents report their health status as 'somewhat unhealthy' or
'unhealthy', 0 otherwise. 29% of women and 23% of men report being in poor health in
the sample.
The fourth health measure is body mass index (BMI, kg/m2). Following World Health
Organization (WHO) standards, dummy variables are created for being underweight if BMI
is under 18.5 and for overweight if BMI is greater or equal to 25. Increasing overweight has
become a problem for the elderly in Indonesia, especially for women (see Witoelar, Strauss
and Sikoki, forthcoming for details). Table A.1 shows that 30% of women over 50 in 2007
are overweight, and 17% of men. Yet underweight is still a problem, for 20% of men and
women.
Hypertension is measured following the standard denition of the WHO; its value is 1
for those whose systolic is greater than or equal to 140 or diastolic is greater than or equal
to 90. In IFLS4, blood pressure of each respondent is measured 3 times and the mean
of the last two measurements is used as dependent variable in the estimation. For earlier
waves of IFLS, blood pressure was measured only once. 63% of elderly women and 52% of
elderly men have hypertension in 2007 (see table A.1).
Hemoglobin levels are examined from blood spots, using the Hemocue meter, as are total
and HDL cholesterol, using the Cardiochek PA meter (non-fasting). A dummy variable is
created as equal to 1 for those whose hemoglobin level is below the threshold (for men:
16
13g/dL, for women: 12g/dL).
4
35% of women and 30% of men have low hemoglobin in
2007.
5
For total cholesterol, I create a dummy equal to 1 if the respondent has high total
cholesterol (240 mg/dL) and for HDL the dummy equals 1 if the level is low (<40 mg/dL).
High total cholesterol is 23% among women but only 11% among men, however low HDL
is a very big problem, for 39% of women and 65% of men (see table A.1).
6
As a measure of depression, respondents answered 10 questions about how they felt
during the week before. It is a self-reported depression scale from the short version of
the CES-D scale, an often used index internationally. The frequency of depression can be
chosen from 4 levels: rarely, some days (1-2 days), occasionally (3-4 days) or most of the
time (5-7 days). Following the standard way of computing CES-D, 0 is scored for those who
answered 'rarely', 1 for 'some days', 2 for 'occasionally', and 3 for 'most of the time'. Eight
out of 10 questions have a negative theme such as "I feel depressed" or "I feel lonely" and
the remaining two questions re
ect positive feelings such as "I am happy" or "I feel hopeful
about my future". For the positive questions, the scoring is reversed from 0 for 'most of
the time' to 3 for 'rarely'. The sum of all scores is used for the analysis and a higher score
on the CES-D scale indicates that respondents are more likely to have depression; means
are 4.56 for women and 3.90 for men (see tables A.1 and A.2).
4
Previous studies such as Thomas et al. (2008) show that the one's work capacity becomes lower if
hemoglobin levels are below these thresholds.
5
This is substantially lower than in 2000 and 1997, see Witoelar, Strauss and Sikoki (forthcoming).
6
Low levels of HDL are also an issue among the elderly in China (see Crimmins et al., 2011).
17
2.3.2 Empirical specication
The parental health and education variables are used as right-hand side covariates to
explore the health of their adult children, who are 50 years and older in 2007. The equation
estimated is
H
i;07
=
0
+
1
Parentalchar
i
+
2
Age
i;07
+
Birthdistrict
i
+"
i;07
(2.1)
The latest health status of respondents measured in the 4th wave of IFLS are dependent
variables in this equation.
Other covariates include dummy variables created for respondent's age: 60-64, 65-69,
70-74, 75-79 and 80 and over. These are necessary because health is highly age dependent
and parental characteristics also are correlated with respondent age since respondent age
re
ects their birth cohort. Indonesia has developed rapidly during the past few decades
which in turn led to birth cohort being highly correlated with parental characteristics.
For instance, it is more likely that parents of younger respondents had more opportuni-
ties for having higher education or faced better health infrastructure. Hence, controlling
for respondents' age helps to address the potential association between birth cohort and
parental characteristics such as health and education.
The birth district (kabupaten or kota) of the respondent is also controlled for with
dummy variables.
7
These dummies will control for contextual factors like prices, health
conditions and health infrastructure at birth that aect respondent's health during child-
hood.
8
7
I rst constructed the place of birth at the sub-district level, but there were many cells with only one
or a handful of observations, so I chose district as our level of aggregation instead. I have 99 of them.
8
Constructing these dummy variables required examining data from the migration section (MG), as well
as the control and demographic roster section (K). The migration section obtains data on sub-district and
district of birth, where one lived at age 12 and all long-distance moves since age 12. From that section one
18
In order to examine the period when these intergenerational correlations are actually
established during one's lifetime, a growth model is estimated. For number of diculties
with ADLs, general health status (GHS) and body mass index (BMI), changes in health
measures from 1993 to 2007 are constructed as outcome variables. For the other measures
such as hemoglobin and hypertension, changes between 1997 and 2007 are used, because
IFLS has collected this information only since 1997. Similar to the cross-sectional analyses,
the sample is restricted to those who are 50 years and older in 2007 and only respondents
who were interviewed in both the 1993 (or 1997) and the 2007 waves are kept in the sample.
The growth model is
H
i;07
H
i;93(97)
=
0
+
1
Parentalchar
i
+
2
Age
i;93(97)
+
Birthdistrict
i
+"
i;93(97)
(2.2)
I also investigate if there is any dierence in intergenerational correlations between
those who were born in more developed areas and in less developed areas. For this purpose
a dummy variable is created as equal to 1 for respondents who were born in Java, Jakarta,
Yogyakarta or Bali (Java dummy) and that is interacted with the parental variables in the
equation (2.1).
Furthermore, I examine the associations between parental SES markers and two impor-
tant measures of human capital accumulation: attained adult height and years of completed
schooling of their adult children.
can know if the respondent still lives in the place of birth. A diculty is involved because many districts
were divided over time and had their names changed. Since the district of birth information comes from
dierent waves of IFLS for dierent respondents, I had to convert all district codes and names into a single
year equivalent (I chose 1999) to obtain a consistent set. I had crosswalks available from the Indonesian
BPS, plus province maps showing all districts. I not only matched numeric codes, but names as well. After
deriving a consistent list of districts and codes I found that still, some had only one or a very few number of
observations, forcing me to aggregate further. I did so using the district maps to group contiguous districts.
I also checked to make sure for the binary dependent variables that none of the districts had all 0s or all
1s. This would cause a problem because the district dummies would then perfectly predict the outcomes.
I found some did and so further aggregated, again using the maps.
19
Finally, in a separate specication, respondent's education and own height are used as
covariates for their later health. These are standard proxies to represent the respondent's
own SES but they may be argued to be endogenous. Examining the association between
one's health and SES is standard and therefore, adding parental variables in this speci-
cation enables us to compare if parents' health variables still remain signicant and if
respondent's SES is signicant as well.
It should be noted that the estimates in this chapter are not necessarily causal, but
they are certainly highly suggestive.
2.4 Results
2.4.1 Intergenerational Correlations
Tables 2.1.1 - 2.1.6 present regression results from the 2007 cross-section (equation (2.1)) for
the dierent health measures. For each outcome, the coecients of parental characteristics
and respondents' age are presented, respectively for men (rst column) and women (second
column).
9
All equations include the birth district dummy variables in order to control for
any heterogeneous characteristics of communities at birth. It is possible that respondents
who were born in more developed areas had better health infrastructures or facilities and
controlling for birth place helps to address this issue. To investigate this issue further,
the Java-Bali birth province dummy is used later as an interaction term with all of the
parental characteristic variables. F-tests for groups of variables, such as the parental health
variables, are reported at the bottom of each table.
For many of the health measures, the results suggest that there exist intergenerational
correlations between the measures of parental health and schooling, and the health of their
9
Standard errors are presented in parentheses and they are adjusted for clustering at the local community
(desa or kelurahan) lived in in 2007, and are also robust to arbitrary forms of heteroskedasticity.
20
Table 2.1.1: Parental SES Gradients of Health of Older Adults 1
Poor GHS # ADL problems
Men Women Men Women
Death Father -0.00673 0.00393 -0.0397 0.208**
(0.0297) (0.0306) (0.0994) (0.102)
Mother -0.00921 0.0717*** 0.0302 0.0887
(0.0195) (0.0215) (0.0649) (0.0808)
GHS Father 0.0331* 0.0590*** 0.161** 0.0771
(0.0200) (0.0193) (0.0730) (0.0815)
Mother 0.0514** 0.0116 0.128 0.136
(0.0212) (0.0217) (0.0805) (0.0866)
ADL Father 0.0243 -0.00316 0.0231 0.215**
(0.0239) (0.0217) (0.0848) (0.0866)
Mother -0.0123 -0.0116 0.0271 -0.0951
(0.0232) (0.0224) (0.0894) (0.0919)
Father's Education At least some primary -0.0185 -0.0314 -0.132 -0.115
(0.0319) (0.0310) (0.115) (0.111)
Completed primary -0.00974 0.0140 -0.158* 0.177*
(0.0229) (0.0222) (0.0804) (0.0922)
Completed junior high school 0.000235 -0.0283 -0.149 0.223
(0.0370) (0.0459) (0.150) (0.159)
Mother's Education At least some primary -0.0198 0.0187 0.0389 -0.0273
(0.0364) (0.0370) (0.133) (0.124)
Completed primary -0.0483** -0.0302 0.0331 -0.166
(0.0241) (0.0263) (0.0803) (0.104)
Completed junior high school -0.118** -0.0418 0.110 -0.546***
(0.0464) (0.0556) (0.255) (0.209) 21
Respondent's age 60 - 65 0.0483** 0.0135 0.348*** 0.594***
(0.0222) (0.0207) (0.0899) (0.0832)
65 - 70 0.125*** 0.0645*** 0.874*** 1.087***
(0.0254) (0.0228) (0.102) (0.0941)
70 - 75 0.193*** 0.105*** 1.393*** 1.633***
(0.0278) (0.0285) (0.140) (0.123)
75 - 80 0.193*** 0.116*** 1.475*** 2.486***
(0.0346) (0.0343) (0.173) (0.180)
80 - 0.217*** 0.164*** 2.868*** 3.216***
(0.0384) (0.0342) (0.205) (0.175)
Constant 0.280*** 0.424*** 0.630 0.489*
(0.103) (0.0820) (0.390) (0.277)
R-squared 0.128 0.092 0.291 0.301
Sample size 3081 3608 3080 3605
Birth Place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's education dummy variables 0.0400 0.4593 0.5798 0.0699
Parent's health (death, GHS, ADL) dummy variables 0.0005 0.0001 0.0050 0.0003
Parent's health (death, GHS, ADL) + education dummy variables 0.0003 0.0003 0.0302 0.0001
Birth Place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
22
Table 2.1.2: Parental SES Gradients of Health of Older Adults 2
# IADL problems BMI ( < 18.5)
Men Women MEN WOMEN
Death Father 0.0510 0.121** -0.00480 0.0550**
(0.0567) (0.0572) (0.0280) (0.0215)
Mother -0.0365 0.0791* 0.0453*** 0.0295*
(0.0396) (0.0449) (0.0173) (0.0156)
GHS Father 0.113** 0.0128 -0.00121 -0.00997
(0.0471) (0.0481) (0.0193) (0.0170)
Mother 0.114** 0.0980** 0.00295 0.00648
(0.0494) (0.0486) (0.0202) (0.0168)
ADL Father 0.0227 0.0853* -0.00631 0.0189
(0.0548) (0.0492) (0.0227) (0.0186)
Mother -0.00521 -0.0363 0.00946 -0.0191
(0.0530) (0.0518) (0.0206) (0.0169)
Father's Education At least some primary -0.205*** -0.102 -0.0189 0.00356
(0.0616) (0.0662) (0.0267) (0.0321)
Completed primary -0.210*** -0.0309 -0.0160 -0.0261
(0.0514) (0.0550) (0.0211) (0.0196)
Completed junior high school -0.213** -0.113 -0.0513 -0.0894***
(0.0952) (0.0878) (0.0330) (0.0257)
Mother's Education At least some primary 0.0575 -0.0635 -0.0548* -0.0158
(0.0755) (0.0671) (0.0281) (0.0301)
Completed primary 0.0637 -0.0705 -0.0572** -0.0375*
(0.0501) (0.0613) (0.0242) (0.0197)
Completed junior high school 0.159 -0.208* -0.0769* 0.0305
(0.161) (0.110) (0.0394) (0.0402) 23
Respondent's age 60 - 65 0.180*** 0.400*** 0.0341 0.0318*
(0.0548) (0.0490) (0.0224) (0.0183)
65 - 70 0.533*** 0.673*** 0.0781*** 0.129***
(0.0635) (0.0566) (0.0232) (0.0218)
70 - 75 0.906*** 1.096*** 0.156*** 0.164***
(0.0887) (0.0765) (0.0323) (0.0263)
75 - 80 1.127*** 1.686*** 0.204*** 0.145***
(0.114) (0.103) (0.0409) (0.0313)
80 - 1.632*** 2.035*** 0.201*** 0.206***
(0.129) (0.0912) (0.0432) (0.0370)
Constant 0.309* -0.0317 0.0644 -0.0178
(0.185) (0.101) (0.0558) (0.0338)
R-squared 0.290 0.344 0.108 0.110
Sample size 3081 3605 2974 3482
Birth Place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's education dummy variables 0.0017 0.0248 0.0003 0.0001
Parent's health (death, GHS, ADL) dummy variables 0.0004 0.0014 0.1624 0.0263
Parent's health (death, GHS, ADL) + education dummy variables 0.0000 0.0000 0.0002 0.0000
Birth Place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
24
Table 2.1.3: Parental SES Gradients of Health of Older Adults 3
BMI ( 25) HB(M<13, W<12)
MEN WOMEN MEN WOMEN
Death Father 0.0448 -0.0102 0.00541 0.0113
(0.0334) (0.0370) (0.0342) (0.0349)
Mother -0.0378* -0.0636** 0.0693*** 0.0105
(0.0225) (0.0262) (0.0221) (0.0212)
GHS Father -0.00256 0.0179 -0.0216 0.0180
(0.0196) (0.0212) (0.0206) (0.0224)
Mother -0.00824 -0.0249 -0.00689 -0.0356
(0.0197) (0.0222) (0.0217) (0.0219)
ADL Father -0.0315 -0.0349 0.0212 -0.0350
(0.0222) (0.0228) (0.0258) (0.0249)
Mother 0.00991 0.0444* -0.0196 0.0465*
(0.0225) (0.0243) (0.0241) (0.0238)
Father's Education At least some primary 0.0309 0.0471 -0.0553* -0.0340
(0.0306) (0.0371) (0.0329) (0.0362)
Completed primary 0.0600** 0.101*** -0.0352 -0.0338
(0.0237) (0.0265) (0.0257) (0.0263)
Completed junior high school 0.0683 0.188*** -0.0447 -0.0776*
(0.0429) (0.0435) (0.0418) (0.0444)
Mother's Education At least some primary -0.00747 0.0561 0.110*** 0.0371
(0.0336) (0.0404) (0.0391) (0.0367)
Completed primary 0.0728*** -0.00700 -0.00638 -0.00224
(0.0262) (0.0326) (0.0294) (0.0284)
Completed junior high school 0.191*** -0.0564 -0.0332 -0.00420
(0.0689) (0.0668) (0.0575) (0.0589) 25
Respondent's age 60 - 65 -0.0273 -0.0490** 0.0389 0.0399
(0.0218) (0.0229) (0.0237) (0.0251)
65 - 70 -0.0644*** -0.101*** 0.115*** 0.0887***
(0.0195) (0.0238) (0.0264) (0.0248)
70 - 75 -0.102*** -0.151*** 0.214*** 0.119***
(0.0219) (0.0246) (0.0321) (0.0303)
75 - 80 -0.0702** -0.143*** 0.210*** 0.184***
(0.0282) (0.0290) (0.0443) (0.0377)
80 - -0.114*** -0.207*** 0.288*** 0.268***
(0.0247) (0.0284) (0.0405) (0.0387)
Constant 0.321*** 0.652*** 0.200** 0.367***
(0.105) (0.0753) (0.0789) (0.0848)
R-squared 0.089 0.127 0.126 0.096
Sample size 2974 3482 2979 3474
Birth Place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's education dummy variables 0.0000 0.0000 0.0848 0.4048
Parent's health (death, GHS, ADL) dummy variables 0.3464 0.0516 0.0364 0.4673
Parent's health (death, GHS, ADL) + education dummy variables 0.0000 0.0000 0.0150 0.4529
Birth Place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
26
Table 2.1.4: Parental SES Gradients of Health of Older Adults 4
Hypertension Cholesterol (240)
MEN WOMEN MEN WOMEN
Death Father 0.0277 0.0187 0.00514 0.0221
(0.0413) (0.0376) (0.0254) (0.0335)
Mother 0.0554** 0.0526** -0.00315 -0.00820
(0.0273) (0.0248) (0.0180) (0.0212)
GHS Father 0.0189 -0.00576 0.0160 -0.0187
(0.0249) (0.0234) (0.0159) (0.0195)
Mother -0.0123 0.0110 0.00477 0.00549
(0.0247) (0.0226) (0.0147) (0.0203)
ADL Father -0.00627 0.00864 0.00428 0.0108
(0.0287) (0.0231) (0.0185) (0.0214)
Mother -0.00119 -0.00539 -0.00552 -0.00900
(0.0277) (0.0239) (0.0171) (0.0221)
Father's Education At least some primary -0.0518 0.00581 -0.0176 -0.0169
(0.0381) (0.0377) (0.0225) (0.0318)
Completed primary 0.0304 -0.0251 0.0311* 0.0712***
(0.0297) (0.0261) (0.0188) (0.0235)
Completed junior high school 0.0448 -0.0204 0.0334 0.0602
(0.0486) (0.0502) (0.0330) (0.0406)
Mother's Education At least some primary -0.0519 0.0366 0.0467* -0.0742**
(0.0424) (0.0408) (0.0275) (0.0293)
Completed primary 0.0310 -0.00383 0.00871 -0.0387
(0.0323) (0.0325) (0.0205) (0.0290)
Completed junior high school 0.0811 0.00417 0.0109 0.0618
(0.0670) (0.0667) (0.0527) (0.0625) 27
Respondent's age 60 - 65 0.0736** 0.0741*** -0.0316** 0.0175
(0.0288) (0.0254) (0.0158) (0.0213)
65 - 70 0.114*** 0.131*** -0.0452*** -0.0113
(0.0294) (0.0249) (0.0163) (0.0208)
70 - 75 0.168*** 0.214*** -0.00662 -0.0148
(0.0331) (0.0304) (0.0207) (0.0253)
75 - 80 0.188*** 0.189*** -0.0401 -0.0130
(0.0396) (0.0320) (0.0248) (0.0302)
80 - 0.200*** 0.262*** -0.0637*** -0.0115
(0.0408) (0.0312) (0.0236) (0.0309)
Constant 0.520*** 0.545*** 0.0835 0.221**
(0.107) (0.0920) (0.0537) (0.105)
R-squared 0.082 0.091 0.098 0.113
Sample size 2985 3493 2957 3457
Birth Place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0166 0.8531
Parent's education dummy variables 0.0028 0.8191 0.1043 0.0003
Parent's health (death, GHS, ADL) dummy variables 0.4435 0.3594 0.8811 0.9638
Parent's health (death, GHS, ADL) + education dummy variables 0.0055 0.6558 0.3038 0.0032
Birth Place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
28
Table 2.1.5: Parental SES Gradients of Health of Older Adults 5
HDL (<40) Cognition
MEN WOMEN MEN WOMEN
Death Father -0.00724 -0.0337 -0.121 -0.306**
(0.0375) (0.0362) (0.132) (0.138)
Mother 0.0381 -0.0276 -0.129 -0.0391
(0.0254) (0.0246) (0.0840) (0.0837)
GHS Father 0.0111 -0.0213 -0.0360 -0.0374
(0.0243) (0.0224) (0.0830) (0.0819)
Mother -0.0558** 0.0236 -0.0751 -0.0649
(0.0228) (0.0217) (0.0826) (0.0781)
ADL Father -0.00567 0.0618** -0.0444 -0.00114
(0.0306) (0.0253) (0.0959) (0.0955)
Mother 0.0189 -0.0649*** -0.0234 0.168**
(0.0301) (0.0240) (0.0907) (0.0818)
Father's Education At least some primary 0.0228 -0.00662 0.493*** 0.408***
(0.0368) (0.0396) (0.138) (0.132)
Completed primary -0.0330 -0.0535** 0.159 0.547***
(0.0278) (0.0261) (0.0967) (0.0924)
Completed junior high school 0.0197 -0.0344 0.457*** 1.275***
(0.0489) (0.0488) (0.162) (0.166)
Mother's Education At least some primary -0.0263 0.0662 -0.123 0.135
(0.0433) (0.0434) (0.146) (0.147)
Completed primary 0.0247 -0.00629 0.347*** 0.189*
(0.0313) (0.0299) (0.107) (0.100)
Completed junior high school 0.0898 -0.0670 0.282 0.249
(0.0688) (0.0703) (0.215) (0.212) 29
Respondent's age 60 - 65 -0.0308 -0.0466** -0.392*** -0.292***
(0.0263) (0.0235) (0.0948) (0.0898)
65 - 70 -0.0538** 0.0279 -0.539*** -0.581***
(0.0268) (0.0258) (0.0942) (0.0904)
70 - 75 -0.0946*** -0.0121 -0.822*** -0.698***
(0.0318) (0.0275) (0.116) (0.126)
75 - 80 -0.0227 -0.0510 -1.158*** -1.015***
(0.0423) (0.0362) (0.150) (0.166)
80 - -0.0154 -0.0283 -1.465*** -1.330***
(0.0425) (0.0356) (0.166) (0.231)
Constant 0.686*** 0.507*** 4.740*** 4.141***
(0.0860) (0.0883) (0.223) (0.257)
R-squared 0.084 0.107 0.203 0.253
Sample size 2931 3445 2317 2309
Birth Place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0755 0.1147 0.0000 0.0000
Parent's education dummy variables 0.3484 0.0912 0.0000 0.0000
Parent's health (death, GHS, ADL) dummy variables 0.3055 0.1140 0.1166 0.0627
Parent's health (death, GHS, ADL) + education dummy variables 0.3392 0.0416 0.0000 0.0000
Birth Place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
30
Table 2.1.6: Parental SES Gradients of Health of Older Adults 6
Depression
MEN WOMEN
Death Father -0.0453 0.277
(0.277) (0.298)
Mother 0.145 0.265
(0.179) (0.197)
GHS Father 0.311** 0.108
(0.153) (0.159)
Mother 0.288* 0.561***
(0.150) (0.171)
ADL Father 0.0734 0.294
(0.197) (0.200)
Mother 0.256 -0.264
(0.183) (0.183)
Father's Education At least some primary -0.0775 -0.421
(0.243) (0.266)
Completed primary -0.470*** -0.451**
(0.160) (0.186)
Completed junior high school 0.0654 -1.103***
(0.313) (0.331)
Mother's Education At least some primary -0.617** 0.0462
(0.277) (0.337)
Completed primary -0.357** -0.296
(0.177) (0.225)
Completed junior high school -0.365 -0.228
(0.435) (0.500)
Respondent's age 60 - 65 -0.163 -0.256
(0.153) (0.163)
65 - 70 0.317* 0.143
(0.183) (0.190)
70 - 75 0.424** 0.213
(0.209) (0.227)
75 - 80 0.932*** 0.515*
(0.301) (0.288)
31
80 - 1.510*** 1.282***
(0.365) (0.341)
Constant 3.275*** 3.121***
(0.462) (0.664)
R-squared 0.110 0.107
Sample size 2830 3222
Birth Place dummy variables Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0001
Parent's education dummy variables 0.0000 0.0000
Parent's health (death, GHS, ADL) dummy variables 0.0000 0.0002
Parent's health (death, GHS, ADL) + education dummy variables 0.0000 0.0000
Birth Place dummy variables 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
32
adult children. For instance, having a father with poor general health status is associated
with increases in the number of diculties of ADLs for men. For women, if the father is
dead in 2007 or had ADL problems in 2007, or right before he died, she is more likely to
report diculties in ADLs. For poor general health status, if a father was in a poor health
condition, his children are more likely to suer from the same problem when they become
older adults regardless of their sex. Mother's poor health is positively related to men's
poor health status whereas having a dead mother is correlated with poor health for female
respondents. In the male sample, poor GHS of both parents is related to the increase in
number of diculties with IADLs. Female respondents tend to report more diculties in
IADLs if either parent were dead, if the father had ADL problems, or if the mother had
poor general health status.
Men tend to be underweight if the mother had died by 2007, while having a dead
mother or a father is positively associated with higher likelihood of being underweight for
women. For being overweight, the correlations with parental health are not signicant for
men, and are at the 10% level for women. For women parental death has the opposite signs
as for being underweight, which means that in this case, parent's being dead is negatively
associated with a woman being overweight; this is reversed for women whose mothers have
ADL problems.
Having low hemoglobin is not correlated with the parental health variables for women,
but is for men; in particular, having a dead mother is positively associated with having
low hemoglobin. Having a dead mother is associated with a higher likelihood of having
hypertension for both men and women, although all the parental health variables jointly
33
are not signicant. The parental health variables are also not jointly signicant for either
having high total cholesterol or low HDL.
10
For cognition for women, however, and depression scores for both men and women,
parental health is jointly signicant at 10% or better. Having a dead father has a nega-
tive association with the cognitive ability of women, although having a mother with ADL
problems is positively correlated. For men, higher depression scores are positively corre-
lated with parents having poor general health and for women with the mother having poor
general health.
As is generally true, higher parental schooling tends to be negatively correlated with
poorer respondent health. This is the case for poor general health for men with respect to
mother's education, for men's IADL problems with respect to father's education, or with
respect to mother's education for women. Similar results are found for having low BMI or
low hemoglobin.
Of considerable interest is the fact that the district birth dummy variables are highly
jointly signicant for all of the health variables, for both men and women. It is dicult to
tell exactly what characteristics of the birth places that are responsible for this. It could
certainly be factors such as levels of infant mortality, and thus exposure to infections and
in
ammation (eg. Crimmins and Finch, 2004), but also could be other factors associated
with economic conditions in the district at birth, such as rainfall (eg. Maccini and Yang,
2009).
In tables 2.2.1-2.2.2, the correlations between parental characteristics and the change in
health measures of their grown-up children are examined.
11
Unlike other health measures,
10
This is not to say that there is no in
uence of parental health. Had I been able to measure cholesterol
for parents before they died, for instance, that might well have been correlated with the measurement of
the respondent children.
11
Standard errors are clustered at the 1993 (or 1997) community (desa or kelurahan) of residence.
34
Table 2.2.1: Parental SES Gradients of Change in Health of Older Adults 1
GHS ADL BMI
Men Women Men Women Men Women
Death Father -0.00945 0.0191 -0.0634 0.278* 0.0845 0.0187
(0.0441) (0.0427) (0.143) (0.159) (0.238) (0.330)
Mother 0.000374 0.0532* -0.0508 -0.0268 -0.0706 -0.341
(0.0289) (0.0289) (0.0917) (0.118) (0.175) (0.229)
GHS Father 0.00394 0.0485* -0.00947 0.0996 0.0819 0.00611
(0.0265) (0.0265) (0.0968) (0.104) (0.149) (0.191)
Mother 0.0470* 0.00552 0.212** 0.177 -0.213 -0.0874
(0.0270) (0.0304) (0.0986) (0.116) (0.157) (0.178)
A D L Father 0.0436 0.0311 0.108 0.158 0.107 -0.0957
(0.0337) (0.0297) (0.111) (0.116) (0.170) (0.202)
Mother -0.00628 -0.0245 -0.145 -0.0601 0.0269 -0.0372
(0.0321) (0.0287) (0.116) (0.117) (0.186) (0.229)
Father's Education At least some primary -0.0245 -0.0232 -0.172 -0.154 0.136 0.661*
(0.0432) (0.0459) (0.161) (0.162) (0.261) (0.338)
Completed primary -0.0466 0.0680** -0.165* 0.260** 0.632*** 0.230
(0.0314) (0.0336) (0.0989) (0.125) (0.207) (0.212)
Completed junior high school -0.0155 -0.00840 -0.128 0.345* 0.0150 1.340*
(0.0491) (0.0565) (0.180) (0.188) (0.297) (0.739)
Mother's Education At least some primary 0.00966 -0.00138 0.0296 0.0584 -0.0166 -0.135
(0.0490) (0.0437) (0.168) (0.179) (0.241) (0.310)
Completed primary -0.00735 -0.0507 0.145 -0.141 -0.121 0.262
(0.0323) (0.0381) (0.105) (0.126) (0.227) (0.298)
Completed junior high school -0.113* -0.0299 0.142 -0.423* 0.138 -1.351*
(0.0666) (0.0794) (0.350) (0.226) (0.439) (0.796)
35
Respondent's age 45 - 50 -0.00340 0.0151 0.254*** 0.379*** -0.648*** -0.436**
(0.0272) (0.0285) (0.0962) (0.122) (0.177) (0.219)
50 - 55 0.0933*** 0.00666 0.712*** 0.726*** -0.847*** -0.903***
(0.0317) (0.0317) (0.128) (0.118) (0.189) (0.204)
55 - 60 0.123*** 0.0437 1.115*** 1.012*** -1.437*** -0.734***
(0.0376) (0.0362) (0.162) (0.171) (0.240) (0.260)
60 - 65 0.146*** -0.0204 1.395*** 1.825*** -0.906*** -1.553***
(0.0453) (0.0394) (0.195) (0.200) (0.222) (0.227)
65 - 0.108** 0.0567 2.159*** 1.936*** -1.152*** -1.749***
(0.0532) (0.0518) (0.213) (0.241) (0.239) (0.289)
Constant 0.293** 0.372*** 1.082* 0.128 1.188** 2.020***
(0.123) (0.100) (0.550) (0.295) (0.533) (0.739)
R-squared 0.085 0.066 0.207 0.154 0.104 0.110
Sample size 2388 2828 2380 2820 2188 2698
Birth place dummy variables Yes Yes Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0001 0.6512 0.0000 0.0000 0.0000 0.0000
Parent's education dummy variables 0.2199 0.3018 0.7286 0.1184 0.0398 0.0737
Parent's health (death, GHS, ADL) dummy variables 0.0811 0.0601 0.4129 0.0235 0.6692 0.5914
Parent's health (death, GHS, ADL) + education dummy variables 0.0858 0.1081 0.6494 0.0173 0.0820 0.1248
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Source: IFLS1, IFLS2, IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
36
Table 2.2.2: Parental SES Gradients of Change in Health of Older Adults 2
Hemoglobin Hypertension
Men Women Men Women
Death Father 0.267 -0.220 -0.0761 0.00659
(0.215) (0.167) (0.0524) (0.0466)
Mother -0.136 0.0769 0.0543* 0.0542*
(0.123) (0.103) (0.0328) (0.0299)
GHS Father 0.131 0.0526 0.0137 -0.0500*
(0.106) (0.0923) (0.0318) (0.0281)
Mother 0.123 -0.0164 -0.0223 0.0467*
(0.107) (0.0912) (0.0286) (0.0271)
A D L Father -0.223* 0.122 -0.00141 0.0443
(0.129) (0.117) (0.0358) (0.0298)
Mother 0.105 -0.0566 -0.0151 -0.0258
(0.124) (0.111) (0.0362) (0.0306)
Father's Education At least some primary 0.136 -0.338** -0.0537 -0.000736
(0.143) (0.150) (0.0460) (0.0430)
Completed primary 0.00311 -0.0408 -0.00314 -0.0674**
(0.131) (0.109) (0.0358) (0.0298)
Completed junior high school -0.256 0.0217 0.0768 -0.00759
(0.225) (0.173) (0.0612) (0.0572)
Mother's Education At least some primary 0.0816 0.235 0.00263 -0.0305
(0.177) (0.175) (0.0466) (0.0482)
Completed primary -0.00129 -0.151 -0.0223 0.00897
(0.151) (0.129) (0.0404) (0.0365)
Completed junior high school 0.582* -0.0826 -0.00561 0.0810
(0.331) (0.339) (0.0946) (0.0827)
37
Respondent's age 45 - 50 -0.175 0.0480 0.0694** -0.0271
(0.125) (0.109) (0.0314) (0.0297)
50 - 55 -0.0437 -0.364*** 0.0760** -0.0850***
(0.136) (0.121) (0.0377) (0.0315)
55 - 60 -0.225* -0.520*** 0.0689* -0.0618*
(0.135) (0.125) (0.0417) (0.0322)
60 - 65 -0.248 -0.343** 0.0498 -0.0526
(0.160) (0.134) (0.0414) (0.0388)
65 - -0.182 -0.616*** 0.0864** -0.102***
(0.177) (0.130) (0.0401) (0.0372)
Constant -0.00188 1.609*** 0.354** 0.212***
(0.355) (0.343) (0.139) (0.0753)
R-squared 0.081 0.105 0.072 0.050
Sample size 2305 2842 2340 2894
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.4139 0.0000 0.1606 0.0327
Parent's education dummy variables 0.4866 0.2132 0.5797 0.1501
Parent's health (death, GHS, ADL) dummy variables 0.1610 0.7674 0.6047 0.0911
Parent's health (death, GHS, ADL) + education dummy variables 0.3707 0.5716 0.6545 0.0604
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS1, IFLS2, IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
38
the changes from 1997 to 2007 are used for hypertension and hemoglobin because IFLS
started collecting these data only since 1997. Having a mother with poor health status is
correlated with an increase in changes of number of ADL problems for male respondents.
In the female sample, if the father was dead in 2007, the increase in the number of ADL
diculties tends to be larger. For moving into poor general health, having a mother with
poor general health is positively correlated with a deterioration for men, and having a father
with poor general health or a mother who died, is for women. For changes in BMI and
hemoglobin levels, parental health variables are not signicant for either men or women.
For changes in hypertension status, parental health variables are not jointly signicant for
men, but are, at 10% for women. For women, a mother having died of in poor general
health is positively correlated with moving from no hypertension to having hypertension,
though a father having poor general health has the opposite sign. Interestingly, for many
of the health outcome changes the parental health measures are not as a group signicantly
correlated with the changes. This is consistent with the hypothesis that the respondent-
parental health correlations, which exist in the cross-sectional data from 2007, are already
established by the time the respondents are 36 years old and over in 1993 (or depending
on the health measure, in 1997 when respondents are 40 years and over).
2.4.2 Interactions with birth region
As discussed in section 2.3, it is more likely that respondents had experienced dierent
living environments or access to health infrastructure at birth, depending on their birth
places. For instance, respondents born in more developed areas probably had better health
facilities and a better environment as compared to other areas in the IFLS sample.
In order to investigate whether a dierent level of development in the area of birth would
mitigate or exacerbate the correlation between parental health and their adult children's
health, a Java-Bali birth dummy variable is constructed as equal to 1 if respondents were
39
born in either Java (including Jakarta, east, west and central Java and Yogyakarta) or
Bali. These areas are and have been the more developed areas in our sample. I interact
this dummy with the dummies for each parental characteristic (health and schooling).
Tables 2.3.1 - 2.3.4 show that these health associations are much lower for respondents
who were born in Java or Bali. For example, having a dead father is associated with an
increase of the number of ADL diculties for women by 0.53 but this correlation almost
disappears once the interaction terms with Java are taken into account. Similar results
are shown for men and women's IADL problems. Father's death is correlated with women
having a larger number of diculties with IADLs but for those who were born in more
developed areas, this correlation is greatly reduced. The same patterns appear for the
association of a mother having poor general health with IADL or ADL problems for men;
with cognition for men and depression for women. Likewise, the association of mother's
death with low BMI for men is reduced by half for those born in Java or Bali, as is the
association of father's ADL problems with low HDL of the respondent.
These results suggest that the level of development at birth or early childhood, which
may include having better health infrastructure or facing dierent health and other prices,
substitutes for the in
uence of parental health.
2.4.3 Respondent's height and schooling
In addition to the health status at older ages, the years of completed education and the
attained height of respondents are analyzed as outcome variables. These are key human
capital outcomes that are aected by health in early life (see for example Maluccio et al,
2009) and because of that may be associated with health and schooling of parents. Height
is determined mostly in early childhood and education when respondents are young adults.
Table 2.4 shows the results of parental SES gradients of respondents' schooling and
height. It is mostly parental education which is signicantly correlated with both height
40
Table 2.3.1: Parental SES Gradients of Health of Older Adults: Interactions
with Birth Region 1
Poor GHS # ADL problems # IADL problems
Men Women Men Women Men Women
Death Father -0.00959 0.0597 -0.235 0.526*** 0.0174 0.222**
(0.0606) (0.0604) (0.236) (0.194) (0.132) (0.105)
Father * Java 0.00761 -0.0768 0.304 -0.421* 0.0495 -0.136
(0.0687) (0.0693) (0.254) (0.229) (0.147) (0.125)
Mother 0.0186 0.0529 0.139 -0.00322 -0.0510 0.0489
(0.0373) (0.0400) (0.122) (0.166) (0.0739) (0.0871)
Mother * Java -0.0367 0.0263 -0.138 0.124 0.0335 0.0419
(0.0428) (0.0471) (0.143) (0.187) (0.0898) (0.100)
GHS Father 0.0523 0.0402 0.284** 0.110 0.0899 -0.0374
(0.0380) (0.0400) (0.128) (0.166) (0.0858) (0.0962)
Father * Java -0.0266 0.0283 -0.178 -0.0522 0.0317 0.0641
(0.0439) (0.0451) (0.160) (0.190) (0.102) (0.109)
Mother 0.0600 0.0402 0.450*** 0.301** 0.312*** 0.210**
(0.0370) (0.0385) (0.144) (0.139) (0.0887) (0.0812)
Mother * Java -0.00766 -0.0422 -0.447** -0.248 -0.283*** -0.161
(0.0440) (0.0467) (0.175) (0.174) (0.105) (0.0993)
ADL Father 0.0868** 0.0747** 0.0270 0.0869 0.0653 0.110
(0.0375) (0.0376) (0.143) (0.151) (0.0865) (0.0937)
Father * Java -0.0968** -0.118*** 0.00107 0.198 -0.0673 -0.0358
(0.0480) (0.0455) (0.181) (0.186) (0.111) (0.115)
Mother -0.0209 -0.0172 -0.0146 -0.172 -0.112 -0.00906
(0.0361) (0.0387) (0.151) (0.162) (0.0794) (0.0920)
Mother * Java 0.0151 0.00856 0.0610 0.115 0.168 -0.0400 41
(0.0465) (0.0471) (0.185) (0.196) (0.104) (0.112)
Father's Education At least some primary 0.0451 -0.0295 -0.00503 -0.0153 -0.160 -0.0568
(0.0562) (0.0569) (0.213) (0.242) (0.112) (0.145)
Completed primary 0.0687 0.0402 -0.0717 0.347** -0.124 0.156
(0.0461) (0.0380) (0.170) (0.172) (0.104) (0.102)
Completed junior high school 0.0410 -0.0193 -0.104 0.644* -0.173 0.121
(0.0586) (0.0929) (0.216) (0.337) (0.144) (0.175)
At least some primary * Java -0.0941 -0.00156 -0.199 -0.140 -0.0664 -0.0579
(0.0661) (0.0658) (0.244) (0.271) (0.128) (0.163)
Completed primary * Java -0.111** -0.0436 -0.101 -0.250 -0.113 -0.281**
(0.0508) (0.0432) (0.185) (0.196) (0.116) (0.115)
Completed junior high school * java -0.0642 -0.00900 -0.0578 -0.602 -0.0748 -0.330*
(0.0753) (0.104) (0.285) (0.378) (0.182) (0.199)
Mother's Education At least some primary -0.140** 0.0333 -0.364* 0.239 -0.156 -0.0440
(0.0664) (0.0696) (0.220) (0.243) (0.127) (0.118)
Completed primary -0.104** -0.0631 -0.308* -0.225 -0.0744 -0.167
(0.0481) (0.0436) (0.170) (0.183) (0.106) (0.112)
Completed junior high school -0.222*** -0.0405 -0.278 -0.941** -0.230* -0.459**
(0.0794) (0.103) (0.242) (0.369) (0.130) (0.199)
At least some primary * Java 0.179** -0.0221 0.584** -0.374 0.304* -0.0312
(0.0774) (0.0802) (0.269) (0.283) (0.155) (0.143)
Completed primary * Java 0.0791 0.0543 0.465** 0.102 0.190 0.158
(0.0542) (0.0542) (0.191) (0.221) (0.121) (0.133)
Completed junior high school * Java 0.153 -0.00927 0.586 0.581 0.593** 0.366
(0.0966) (0.126) (0.429) (0.450) (0.254) (0.239)
Respondent's age 60 - 65 0.0501** 0.0121 0.347*** 0.605*** 0.179*** 0.402***
(0.0223) (0.0208) (0.0896) (0.0827) (0.0550) (0.0485)
42
65 - 70 0.123*** 0.0617*** 0.875*** 1.091*** 0.532*** 0.669***
(0.0253) (0.0228) (0.102) (0.0942) (0.0635) (0.0566)
70 - 75 0.190*** 0.104*** 1.380*** 1.642*** 0.901*** 1.099***
(0.0279) (0.0286) (0.139) (0.123) (0.0887) (0.0769)
75 - 80 0.189*** 0.114*** 1.476*** 2.495*** 1.125*** 1.684***
(0.0343) (0.0345) (0.173) (0.180) (0.114) (0.104)
80 - 0.219*** 0.163*** 2.866*** 3.213*** 1.631*** 2.033***
(0.0386) (0.0342) (0.205) (0.174) (0.131) (0.0913)
Constant 0.294*** 0.450*** 0.600 0.617** 0.290 0.0399
(0.103) (0.0826) (0.395) (0.286) (0.189) (0.110)
R-squared 0.133 0.095 0.298 0.304 0.294 0.346
Sample size 3081 3608 3080 3605 3081 3605
Birth place dummy variables Yes Yes Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Parents' edu dummy + parents' edu dum * java 0.0269 0.8766 0.0667 0.1283 0.0003 0.0258
Parents' death, GHS, ADL + Parents' death, GHS, ADL * java 0.0007 0.0000 0.0002 0.0004 0.0007 0.0076
Parents' death, GHS, ADL + education dum * java 0.0557 0.4028 0.0044 0.2348 0.1037 0.4298
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
43
Table 2.3.2: Parental SES Gradients of Health of Older Adults: Interactions
with Birth Region 2
BMI (<18.5) BMI (25) HB(M<13, W<12)
Men Women Men Women Men Women
Death Father -0.0385 0.0374 0.0648 0.00955 0.0606 -0.0636
(0.0563) (0.0438) (0.0498) (0.0732) (0.0540) (0.0689)
Father * Java 0.0479 0.0205 -0.0270 -0.0281 -0.0827 0.0995
(0.0650) (0.0499) (0.0653) (0.0846) (0.0684) (0.0787)
Mother 0.0718** 0.0402 -0.0277 -0.114*** 0.0692** 0.0637
(0.0317) (0.0302) (0.0387) (0.0428) (0.0339) (0.0390)
Mother * Java -0.0383 -0.0112 -0.0161 0.0699 0.00111 -0.0727
(0.0383) (0.0355) (0.0466) (0.0518) (0.0418) (0.0451)
GHS Father -0.00378 -0.0486* -0.000989 0.0260 0.00864 0.00111
(0.0344) (0.0280) (0.0353) (0.0359) (0.0340) (0.0372)
Father * Java 0.00306 0.0538 -0.00189 -0.00807 -0.0415 0.0200
(0.0412) (0.0333) (0.0411) (0.0425) (0.0420) (0.0451)
Mother 0.0157 0.0621** -0.00845 -0.00884 -0.0415 -0.0400
(0.0379) (0.0266) (0.0380) (0.0350) (0.0377) (0.0366)
Mother * Java -0.0146 -0.0811** -0.00232 -0.0249 0.0500 0.0109
(0.0442) (0.0324) (0.0434) (0.0441) (0.0450) (0.0447)
ADL Father 0.0129 0.0375 -0.0803** -0.0912** 0.00652 -0.0841**
(0.0411) (0.0296) (0.0337) (0.0390) (0.0429) (0.0350)
Father * Java -0.0315 -0.0277 0.0763* 0.0850* 0.0214 0.0726
(0.0495) (0.0377) (0.0430) (0.0472) (0.0522) (0.0464)
Mother -0.0295 -0.0540* 0.0336 0.0724* 0.0219 0.0830**
(0.0331) (0.0293) (0.0339) (0.0389) (0.0409) (0.0394)
Mother * Java 0.0613 0.0550 -0.0388 -0.0422 -0.0613 -0.0542 44
(0.0423) (0.0358) (0.0450) (0.0506) (0.0500) (0.0489)
Father's Education At least some primary 0.0567 0.0887* 0.0197 -0.0237 -0.00366 -0.0544
(0.0483) (0.0505) (0.0501) (0.0634) (0.0521) (0.0650)
Completed primary 0.0358 -0.0205 0.0502 0.101** 0.00130 -0.00821
(0.0416) (0.0334) (0.0386) (0.0445) (0.0452) (0.0461)
Completed junior high school -0.0140 -0.104*** 0.0190 0.278*** 0.00293 -0.105
(0.0528) (0.0350) (0.0594) (0.0759) (0.0627) (0.0773)
At least some primary * Java -0.111** -0.125** 0.0165 0.104 -0.0783 0.0332
(0.0563) (0.0621) (0.0601) (0.0752) (0.0655) (0.0757)
Completed primary * Java -0.0720 -0.00582 0.0138 0.00233 -0.0543 -0.0341
(0.0466) (0.0392) (0.0466) (0.0522) (0.0535) (0.0534)
Completed junior high school * Java -0.0504 0.0194 0.0768 -0.127 -0.0689 0.0363
(0.0672) (0.0460) (0.0814) (0.0915) (0.0830) (0.0944)
Mother's Education At least some primary -0.0915* -0.0624 -0.0295 0.105 0.118* 0.00460
(0.0507) (0.0514) (0.0577) (0.0734) (0.0711) (0.0643)
Completed primary -0.138*** -0.0489 0.0925** 0.0435 -0.0544 0.0269
(0.0404) (0.0336) (0.0442) (0.0556) (0.0493) (0.0450)
Completed junior high school -0.0837 -0.0249 0.0700 0.0325 -0.0497 0.00167
(0.0654) (0.0553) (0.109) (0.120) (0.0830) (0.102)
At least some primary * Java 0.0557 0.0683 0.0270 -0.0679 -0.00976 0.0451
(0.0589) (0.0614) (0.0705) (0.0891) (0.0836) (0.0772)
Completed primary * Java 0.117** 0.0161 -0.0306 -0.0777 0.0700 -0.0461
(0.0492) (0.0399) (0.0547) (0.0658) (0.0604) (0.0588)
Completed junior high school * Java 0.0125 0.0898 0.159 -0.154 0.0186 -0.00509
(0.0826) (0.0766) (0.138) (0.139) (0.112) (0.126)
Respondent's age 60 - 65 0.0326 0.0320* -0.0279 -0.0493** 0.0385 0.0415
(0.0224) (0.0183) (0.0220) (0.0229) (0.0237) (0.0253)
45
65 - 70 0.0788*** 0.130*** -0.0634*** -0.101*** 0.115*** 0.0903***
(0.0232) (0.0218) (0.0195) (0.0239) (0.0265) (0.0249)
70 - 75 0.157*** 0.165*** -0.103*** -0.152*** 0.213*** 0.121***
(0.0321) (0.0265) (0.0220) (0.0249) (0.0321) (0.0304)
75 - 80 0.205*** 0.146*** -0.0685** -0.142*** 0.210*** 0.186***
(0.0411) (0.0313) (0.0281) (0.0291) (0.0445) (0.0377)
80 - 0.204*** 0.207*** -0.114*** -0.210*** 0.289*** 0.269***
(0.0432) (0.0370) (0.0247) (0.0285) (0.0405) (0.0387)
Constant 0.0548 -0.0271 0.323*** 0.674*** 0.234*** 0.362***
(0.0592) (0.0382) (0.112) (0.0798) (0.0836) (0.0866)
R-squared 0.111 0.113 0.092 0.132 0.128 0.099
Sample size 2974 3482 2974 3482 2979 3474
Birth place dummy variables Yes Yes Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Parents' edu dum + parents' edu dum * java 0.0003 0.0001 0.0000 0.0000 0.2886 0.6435
Parents' death, GHS, ADL + Parents' death, GHS, ADL * java 0.3002 0.0367 0.4459 0.0862 0.1432 0.2700
Parents' death, GHS, ADL + education dum * java 0.3512 0.2328 0.6690 0.1144 0.8310 0.5703
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
46
Table 2.3.3: Parental SES Gradients of Health of Older Adults: Interactions
with Birth Region 3
Hypertension Cholesterol (240) HDL (<40)
Men Women Men Women Men Women
Death Father 0.0134 -0.0498 0.0644* 0.0137 0.0323 -0.126*
(0.0698) (0.0728) (0.0338) (0.0541) (0.0574) (0.0709)
Father * Java 0.0250 0.0932 -0.0847* 0.0143 -0.0580 0.125
(0.0824) (0.0839) (0.0463) (0.0667) (0.0763) (0.0827)
Mother 0.105** 0.0604 -0.0384 0.0106 0.0383 -0.0109
(0.0468) (0.0439) (0.0295) (0.0348) (0.0466) (0.0502)
Mother * Java -0.0692 -0.0103 0.0493 -0.0294 0.000411 -0.0217
(0.0562) (0.0521) (0.0358) (0.0418) (0.0553) (0.0566)
GHS Father 0.0318 -0.00139 0.00602 0.0143 -0.0152 -0.0596
(0.0414) (0.0469) (0.0258) (0.0325) (0.0402) (0.0425)
Father * Java -0.0146 -0.00606 0.0150 -0.0472 0.0344 0.0572
(0.0504) (0.0520) (0.0324) (0.0384) (0.0492) (0.0493)
Mother 0.0121 -0.00645 -0.0203 -0.0318 0.00633 0.0242
(0.0424) (0.0411) (0.0265) (0.0316) (0.0389) (0.0401)
Mother * Java -0.0368 0.0249 0.0342 0.0552 -0.0861* 0.00131
(0.0509) (0.0493) (0.0324) (0.0398) (0.0468) (0.0467)
ADL Father -0.0301 0.0155 -0.00125 -0.00693 -0.0201 0.0992**
(0.0459) (0.0422) (0.0241) (0.0331) (0.0475) (0.0435)
Father * Java 0.0353 -0.00915 0.00475 0.0258 0.0288 -0.0575
(0.0582) (0.0505) (0.0349) (0.0431) (0.0609) (0.0524)
Mother -0.0256 -0.0264 -0.00754 0.00250 -0.00470 -0.0609
(0.0462) (0.0397) (0.0237) (0.0370) (0.0416) (0.0418)
Mother * Java 0.0360 0.0292 0.00540 -0.0177 0.0286 -0.00623 47
(0.0576) (0.0498) (0.0334) (0.0452) (0.0574) (0.0508)
Father's Education At least some primary -0.119* 0.0734 -0.0725** -0.0311 0.00844 -0.129**
(0.0705) (0.0649) (0.0287) (0.0498) (0.0561) (0.0628)
Completed primary 0.0229 -0.0105 0.0166 0.0601 -0.0391 -0.119***
(0.0521) (0.0421) (0.0283) (0.0366) (0.0445) (0.0431)
Completed junior high school 0.0397 -0.0612 0.0460 0.0697 -0.0876 -0.0760
(0.0731) (0.0895) (0.0505) (0.0647) (0.0794) (0.0886)
At least some primary * Java 0.101 -0.0979 0.0809** 0.0210 0.0208 0.181**
(0.0825) (0.0780) (0.0400) (0.0625) (0.0709) (0.0761)
Completed primary * Java 0.0120 -0.0213 0.0164 0.0203 0.0112 0.0972*
(0.0628) (0.0515) (0.0361) (0.0452) (0.0540) (0.0510)
Completed junior high school * Java 0.00211 0.0600 -0.0310 -0.0160 0.184* 0.0585
(0.0965) (0.104) (0.0665) (0.0821) (0.0981) (0.107)
Mother's Education At least some primary -0.0724 0.0850 0.0197 -0.0679 -0.0620 0.174**
(0.0782) (0.0678) (0.0436) (0.0431) (0.0738) (0.0730)
Completed primary 0.0258 -0.0312 0.00646 0.0228 0.0254 0.0209
(0.0506) (0.0520) (0.0323) (0.0442) (0.0544) (0.0478)
Completed junior high school -0.0720 0.000485 -0.0224 0.0151 0.255*** -0.0139
(0.115) (0.116) (0.0762) (0.0873) (0.0902) (0.116)
At least some primary * Java 0.0243 -0.0705 0.0362 -0.00809 0.0430 -0.161*
(0.0931) (0.0834) (0.0558) (0.0571) (0.0908) (0.0888)
Completed primary * Java 0.00637 0.0427 0.00579 -0.0969* -0.00580 -0.0440
(0.0654) (0.0654) (0.0408) (0.0544) (0.0658) (0.0590)
Completed junior high school * Java 0.225 0.0166 0.0537 0.0823 -0.259** -0.0802
(0.140) (0.140) (0.106) (0.120) (0.124) (0.144)
Respondent's age 60 - 65 0.0754*** 0.0740*** -0.0303* 0.0190 -0.0302 -0.0476**
(0.0288) (0.0256) (0.0157) (0.0212) (0.0263) (0.0236)
48
65 - 70 0.115*** 0.131*** -0.0454*** -0.0112 -0.0522* 0.0274
(0.0296) (0.0250) (0.0165) (0.0209) (0.0267) (0.0258)
70 - 75 0.168*** 0.215*** -0.00712 -0.0149 -0.0930*** -0.0118
(0.0332) (0.0307) (0.0207) (0.0254) (0.0316) (0.0278)
75 - 80 0.191*** 0.189*** -0.0393 -0.0117 -0.0209 -0.0516
(0.0397) (0.0320) (0.0246) (0.0303) (0.0422) (0.0362)
80 - 0.201*** 0.261*** -0.0645*** -0.0117 -0.0170 -0.0284
(0.0407) (0.0314) (0.0235) (0.0310) (0.0424) (0.0357)
Constant 0.519*** 0.511*** 0.0880 0.225** 0.694*** 0.472***
(0.110) (0.0936) (0.0590) (0.108) (0.0917) (0.0911)
R-squared 0.085 0.093 0.102 0.115 0.087 0.111
Sample size 2985 3493 2957 3457 2931 3445
Birth place dummy variables Yes Yes Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0174 0.8289 0.0826 0.1111
Parents' edu dum + parents' edu dum * java 0.0150 0.7333 0.0271 0.0014 0.1236 0.0469
Parents' death, GHS, ADL + Parents' death, GHS, ADL * java 0.5066 0.5931 0.5586 0.9866 0.4213 0.1897
Parents' death, GHS, ADL + education dum * java 0.4825 0.7925 0.2061 0.7778 0.3549 0.1669
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
49
Table 2.3.4: Parental SES Gradients of Health of Older Adults: Interactions
with Birth Region 4
Cognition Depression
Men Women Men Women
Death Father -0.0485 -0.197 0.196 0.589
(0.227) (0.259) (0.414) (0.659)
Father * Java -0.0653 -0.187 -0.324 -0.392
(0.277) (0.296) (0.550) (0.740)
Mother -0.158 -0.0352 0.253 0.462
(0.135) (0.150) (0.336) (0.426)
Mother * Java 0.0175 0.0105 -0.138 -0.274
(0.163) (0.177) (0.388) (0.466)
GHS Father -0.0996 -0.134 0.517 0.229
(0.133) (0.142) (0.333) (0.335)
Father * Java 0.0841 0.166 -0.279 -0.192
(0.167) (0.170) (0.369) (0.377)
Mother -0.224* 0.128 0.287 0.776**
(0.124) (0.135) (0.319) (0.332)
Mother * Java 0.235 -0.306* -0.0143 -0.313
(0.161) (0.162) (0.357) (0.384)
ADL Father -0.0655 0.189 -0.157 0.445
(0.141) (0.175) (0.353) (0.409)
Father * Java 0.0463 -0.299 0.353 -0.250
(0.190) (0.207) (0.419) (0.463)
Mother 0.205 0.109 0.281 -0.475
(0.137) (0.136) (0.323) (0.379)
Mother * Java -0.398** 0.0902 -0.0550 0.354 50
(0.181) (0.168) (0.386) (0.422)
Father's Education At least some primary 0.459* 0.445* -0.187 -0.428
(0.249) (0.238) (0.396) (0.478)
Completed primary 0.0850 0.564*** -0.661** -0.640**
(0.163) (0.143) (0.332) (0.320)
Completed junior high school 0.294 1.150*** -0.110 -1.347**
(0.229) (0.309) (0.486) (0.614)
At least some primary * Java 0.0411 -0.0655 0.153 -0.00427
(0.288) (0.279) (0.501) (0.557)
Completed primary * Java 0.0952 -0.0434 0.274 0.263
(0.192) (0.176) (0.379) (0.380)
Completed junior high school * Java 0.224 0.180 0.268 0.290
(0.313) (0.358) (0.649) (0.732)
Mother's Education At least some primary -0.615** 0.313 -0.867* -0.219
(0.269) (0.261) (0.448) (0.478)
Completed primary 0.535*** 0.0871 -0.125 -0.261
(0.162) (0.141) (0.314) (0.419)
Completed junior high school 0.370 0.976*** -0.293 -1.597**
(0.426) (0.347) (0.852) (0.684)
At least some primary * Java 0.748** -0.265 0.348 0.402
(0.308) (0.316) (0.550) (0.640)
Completed primary * Java -0.278 0.184 -0.338 -0.0448
(0.199) (0.198) (0.380) (0.493)
Completed junior high school * Java -0.170 -1.163*** -0.119 2.273**
(0.494) (0.435) (0.995) (0.907)
Respondent's age 60 - 65 -0.385*** -0.299*** -0.167 -0.247
(0.0940) (0.0896) (0.154) (0.164)
51
65 - 70 -0.542*** -0.578*** 0.325* 0.138
(0.0948) (0.0915) (0.183) (0.188)
70 - 75 -0.836*** -0.695*** 0.418** 0.202
(0.117) (0.127) (0.212) (0.228)
75 - 80 -1.161*** -1.016*** 0.946*** 0.517*
(0.148) (0.167) (0.301) (0.290)
80 - -1.494*** -1.333*** 1.505*** 1.281***
(0.165) (0.230) (0.365) (0.339)
Constant 4.744*** 4.287*** 3.396*** 3.140***
(0.245) (0.267) (0.528) (0.671)
R-squared 0.211 0.260 0.111 0.111
Sample size 2317 2309 2830 3222
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0001
Parents' edu dum + parents' edu dum * java 0.0000 0.0000 0.0013 0.0000
Parents' death, GHS, ADL + Parents' death, GHS, ADL * java 0.0731 0.0778 0.0011 0.0027
Parents' death, GHS, ADL + education dum * java 0.0193 0.0863 0.9837 0.1700
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
52
Table 2.4: Parental SES Gradients of Respondents Schooling and Height
Education year Height
Men Women Men Women
Death Father -0.00707 0.0275 -0.163 -1.206**
(0.363) (0.301) (0.802) (0.480)
Mother -0.429* -0.152 -0.170 0.333
(0.227) (0.166) (0.484) (0.557)
GHS Father -0.511*** -0.215 0.688 0.550
(0.194) (0.152) (0.480) (0.508)
Mother 0.0732 -0.166 -0.203 -0.615
(0.194) (0.138) (0.448) (0.534)
ADL Father 0.136 0.0251 0.311 -0.422
(0.218) (0.174) (0.423) (0.669)
Mother -0.0633 0.275* -0.0862 -0.327
(0.221) (0.153) (0.437) (0.712)
Father's Education At least some primary 1.664*** 1.665*** 0.342 -0.655
(0.318) (0.282) (0.703) (1.015)
Completed primary 2.242*** 2.037*** 1.099* 0.463
(0.245) (0.204) (0.567) (0.480)
Completed junior high school 3.779*** 5.339*** 2.139*** -1.434
(0.423) (0.378) (0.686) (1.810)
Mother's Education At least some primary 0.980*** 1.101*** -0.891 1.569
(0.348) (0.327) (1.002) (1.189)
Completed primary 1.502*** 1.837*** -0.534 0.610
(0.266) (0.237) (0.633) (0.631)
Completed junior high school 2.905*** 3.512*** 1.824** 5.803***
(0.584) (0.496) (0.852) (1.591)
53
Respondent's age 60 - 65 -0.355* -0.507*** -1.806*** -1.678***
(0.214) (0.167) (0.484) (0.526)
65 - 70 -0.0905 -1.263*** -2.067*** -2.483***
(0.222) (0.163) (0.597) (0.506)
70 - 75 -1.397*** -1.928*** -2.982*** -4.262***
(0.254) (0.184) (0.661) (0.633)
75 - 80 -2.484*** -2.084*** -4.571*** -4.346***
(0.297) (0.200) (0.873) (0.772)
80 - -2.473*** -2.398*** -5.675*** -6.910***
(0.290) (0.195) (1.127) (0.762)
Constant 7.354*** 3.828*** 160.3*** 145.8***
(0.871) (0.436) (1.186) (3.706)
R-squared 0.337 0.458 0.110 0.096
Sample size 3081 3608 3081 3608
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's education dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's health (death, GHS, ADL) dummy variables 0.0433 0.1175 0.7408 0.1385
Parent's health (death, GHS, ADL) + education dummy variables 0.0000 0.0000 0.0002 0.0000
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
54
and years of schooling of their children. All levels of schooling of father and mother are
associated with the increase in their children's education years and the higher education,
the larger the coecient. Father's education is positively correlated with heights of men,
and mother's education with heights of both men and women, but with a stronger eect for
women, consistent with results of Thomas (1994). There are correlations between parental
schooling and respondent education. For men, having a mother who is dead is correlated
with 0.4 years less schooling and having a father with poor general health with 0.5 years
fewer schooling. For women, if the father is dead, height is on average 1.2cm less. Again,
the birth district dummies are highly signicant.
2.4.4 Health and SES gradients of older adults
Regressing one's health status on own SES is a standard specication of looking at health-
SES gradients. However, the correlation between one's health and SES might derive partly
from the in
uence of parental background. To check this, I add parental background
variables to a specication with respondent SES variables. I use own education years and
height to measure respondent's SES. Education is a standard SES measure. I would like
a measure of health during childhood to complement that, and while IFLS does not have
direct measures of child health, adult height can be used as such a variable because it
re
ects in large part height during childhood.
12
Tables 2.5.1 - 2.5.11 report both the standard model (having only respondent's own
education and height, plus age and district of birth) and the extended model (adding
parental health and education) for various health measures.
13
Results from the standard
12
Birthweight is available for part of the sample, but those for whom it is not available are not a random
sub-sample; they are people who were born at home perhaps using midwives, but who did not weight the
babies at birth.
13
See Witoelar, Strauss and Sikoki (forthcoming) for a more detailed analysis of the health-respondent
SES correlations in IFLS.
55
Table 2.5.1: Health-SES Gradients of Older Adults 1
Poor GHS
Men Women
Respondent's Edu At least some primary 0.00976 0.0199 0.0427** 0.0552***
(0.0260) (0.0260) (0.0201) (0.0204)
Completed primary -0.00933 0.00931 0.0244 0.0380
(0.0252) (0.0251) (0.0235) (0.0248)
Completed Jr. high -0.0524 -0.0313 -0.0265 -0.0106
(0.0320) (0.0326) (0.0334) (0.0347)
Completed high school -0.0930*** -0.0588** -0.0558* -0.0211
(0.0276) (0.0293) (0.0302) (0.0334)
Respondent's height 0 - median 0.00117 0.00124 0.000856 0.000882
(0.000832) (0.000910) (0.000921) (0.000945)
median -0.00166 -0.00149 0.00185 0.00203
(0.00240) (0.00245) (0.00239) (0.00236)
Respondent's age 60 - 65 0.0515** 0.0464** 0.0364* 0.0212
(0.0216) (0.0220) (0.0198) (0.0211)
65 - 70 0.139*** 0.127*** 0.0940*** 0.0758***
(0.0247) (0.0254) (0.0230) (0.0234)
70 - 75 0.207*** 0.191*** 0.152*** 0.123***
(0.0292) (0.0282) (0.0283) (0.0293)
75 - 80 0.210*** 0.190*** 0.162*** 0.131***
(0.0348) (0.0350) (0.0337) (0.0351)
80 - 0.249*** 0.214*** 0.226*** 0.189***
(0.0380) (0.0385) (0.0338) (0.0350)
Death Father -0.00577 0.00685
(0.0300) (0.0306)
Mother -0.0113 0.0726***
(0.0196) (0.0216)
GHS Father 0.0304 0.0584***
(0.0200) (0.0195)
Mother 0.0516** 0.0118
(0.0211) (0.0216)
ADL Father 0.0223 -0.00502
(0.0242) (0.0217)
56
Mother -0.0114 -0.00856
(0.0232) (0.0224)
Father's Education At least some primary -0.0132 -0.0383
(0.0322) (0.0313)
Completed primary -0.000789 0.0115
(0.0232) (0.0227)
Completed Jr. high 0.0248 -0.00792
(0.0372) (0.0464)
Mother's Education At least some primary -0.0134 0.0165
(0.0363) (0.0375)
Completed primary -0.0378 -0.0245
(0.0244) (0.0263)
Completed Jr. high -0.0987** -0.0329
(0.0466) (0.0554)
Constant 0.172 0.0935 0.369** 0.257
(0.159) (0.175) (0.156) (0.158)
R-squared 0.109 0.132 0.083 0.095
Sample size 3081 3081 3608 3608
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.3716 0.3860 0.2214 0.1799
Education dummy variables 0.0001 0.0143 0.0030 0.0100
Parent's death dummy variables 0.8102 0.0029
Parent's education dummy variables 0.4188 0.6908
Parent's GHS dummy variables 0.0002 0.0009
Parent's ADL dummy variables 0.6526 0.8291
Parent's health (death, GHS, ADL) dummy variables 0.0011 0.0001
Parent's health (death, GHS, ADL)+ edu dummy 0.0065 0.0005
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
57
Table 2.5.2: Health-SES Gradients of Older Adults 2
# ADL problems
Men Women
Respondent's Edu At least some primary -0.408*** -0.334*** -0.0588 0.00752
(0.118) (0.115) (0.0943) (0.0982)
Completed primary -0.318** -0.207* -0.0197 0.0432
(0.124) (0.122) (0.101) (0.110)
Completed Jr. high -0.486*** -0.364*** -0.0122 0.0748
(0.136) (0.136) (0.138) (0.148)
Completed high school -0.512*** -0.360*** -0.130 0.0318
(0.133) (0.136) (0.142) (0.167)
Respondent's height 0 - median 0.00458 0.00614* -0.00825*** -0.00783***
(0.00328) (0.00362) (0.00315) (0.00299)
median -0.0154* -0.0164* 0.00705 0.00986
(0.00837) (0.00890) (0.00889) (0.00854)
Respondent's age 60 - 65 0.365*** 0.335*** 0.647*** 0.593***
(0.0914) (0.0905) (0.0814) (0.0832)
65 - 70 0.917*** 0.863*** 1.156*** 1.085***
(0.102) (0.102) (0.0963) (0.0964)
70 - 75 1.437*** 1.338*** 1.762*** 1.625***
(0.146) (0.145) (0.128) (0.130)
75 - 80 1.525*** 1.404*** 2.606*** 2.475***
(0.173) (0.173) (0.182) (0.181)
80 - 3.045*** 2.805*** 3.397*** 3.196***
(0.218) (0.205) (0.182) (0.179)
Death Father -0.0423 0.200**
(0.0997) (0.101)
Mother 0.0213 0.0889
(0.0652) (0.0809)
GHS Father 0.154** 0.0814
(0.0731) (0.0811)
Mother 0.131 0.134
(0.0809) (0.0863)
ADL Father 0.0198 0.211**
(0.0848) (0.0861)
58
Mother 0.0258 -0.0969
(0.0893) (0.0920)
Father's Education At least some primary -0.0945 -0.129
(0.117) (0.114)
Completed primary -0.125 0.167*
(0.0841) (0.0932)
Completed Jr. high -0.0970 0.193
(0.149) (0.171)
Mother's Education At least some primary 0.0644 -0.0258
(0.134) (0.125)
Completed primary 0.0650 -0.170
(0.0817) (0.110)
Completed Jr. high 0.148 -0.530**
(0.260) (0.209)
Constant 0.639 -0.0190 2.167*** 1.582***
(0.652) (0.681) (0.525) (0.516)
R-squared 0.263 0.296 0.283 0.302
Sample size 3080 3080 3605 3605
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.1586 0.1350 0.0299 0.0329
Education dummy variables 0.0012 0.0160 0.8854 0.9856
Parent's death dummy variables 0.8767 0.0561
Parent's education dummy variables 0.8665 0.0863
Parent's GHS dummy variables 0.0043 0.0417
Parent's ADL dummy variables 0.8770 0.0501
Parent's health (death, GHS, ADL) dummy variables 0.0076 0.0004
Parent's health (death, GHS, ADL)+ edu dummy 0.0628 0.0001
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
59
Table 2.5.3: Health-SES Gradients of Older Adults 3
# IADL problems
Men Women
Respondent's Edu At least some primary -0.342*** -0.288*** -0.196*** -0.150***
(0.0786) (0.0734) (0.0515) (0.0528)
Completed primary -0.363*** -0.281*** -0.191*** -0.136**
(0.0792) (0.0750) (0.0634) (0.0678)
Completed Jr. high -0.500*** -0.409*** -0.254*** -0.178**
(0.0926) (0.0883) (0.0747) (0.0808)
Completed high school -0.495*** -0.383*** -0.422*** -0.306***
(0.0815) (0.0792) (0.0756) (0.0888)
Respondent's height 0 - median 0.000148 0.00123 -0.00381* -0.00355*
(0.00212) (0.00240) (0.00205) (0.00189)
median -0.00689 -0.00780 -0.00552 -0.00376
(0.00552) (0.00589) (0.00516) (0.00514)
Respondent's age 60 - 65 0.183*** 0.166*** 0.419*** 0.377***
(0.0552) (0.0550) (0.0477) (0.0492)
65 - 70 0.556*** 0.522*** 0.677*** 0.626***
(0.0626) (0.0633) (0.0564) (0.0569)
70 - 75 0.913*** 0.846*** 1.124*** 1.025***
(0.0930) (0.0920) (0.0831) (0.0818)
75 - 80 1.115*** 1.037*** 1.717*** 1.609***
(0.114) (0.113) (0.105) (0.106)
80 - 1.715*** 1.552*** 2.085*** 1.933***
(0.135) (0.129) (0.0978) (0.0954)
Death Father 0.0488 0.109*
(0.0569) (0.0574)
Mother -0.0458 0.0792*
(0.0396) (0.0447)
GHS Father 0.105** 0.0112
(0.0469) (0.0477)
Mother 0.115** 0.0893*
(0.0498) (0.0481)
ADL Father 0.0249 0.0865*
(0.0545) (0.0487)
60
Mother -0.00696 -0.0353
(0.0527) (0.0519)
Father's Education At least some primary -0.162*** -0.0642
(0.0620) (0.0682)
Completed primary -0.157*** 0.0142
(0.0523) (0.0558)
Completed Jr. high -0.146 -0.0156
(0.0940) (0.0948)
Mother's Education At least some primary 0.0767 -0.0295
(0.0762) (0.0675)
Completed primary 0.0955* -0.0379
(0.0510) (0.0632)
Completed Jr. high 0.202 -0.113
(0.163) (0.112)
Constant 0.921** 0.446 0.984*** 0.620**
(0.387) (0.420) (0.317) (0.303)
R-squared 0.259 0.298 0.326 0.348
Sample size 3081 3081 3605 3605
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.3877 0.3992 0.0178 0.0380
Education dummy variables 0.0000 0.0000 0.0000 0.0066
Parent's death dummy variables 0.3569 0.0284
Parent's education dummy variables 0.0500 0.8287
Parent's GHS dummy variables 0.0005 0.0873
Parent's ADL dummy variables 0.8982 0.2067
Parent's health (death, GHS, ADL) dummy variables 0.0005 0.0037
Parent's health (death, GHS, ADL)+ edu dummy 0.0003 0.0181
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
61
Table 2.5.4: Health-SES Gradients of Older Adults 4
BMI (<18.5)
Men Women
Respondent's Edu At least some primary -0.0468* -0.0368 -0.0281 -0.0214
(0.0266) (0.0265) (0.0210) (0.0218)
Completed primary -0.0780*** -0.0673** -0.0956*** -0.0856***
(0.0264) (0.0267) (0.0213) (0.0222)
Completed Jr. high -0.117*** -0.0974*** -0.124*** -0.112***
(0.0319) (0.0326) (0.0268) (0.0289)
Completed high school -0.192*** -0.167*** -0.135*** -0.118***
(0.0274) (0.0281) (0.0222) (0.0264)
Respondent's height 0 - median 0.00108 0.00128* 0.00159** 0.00168**
(0.000660) (0.000667) (0.000730) (0.000727)
median 0.00436* 0.00428* -0.000974 -0.00123
(0.00224) (0.00226) (0.00234) (0.00234)
Respondent's age 60 - 65 0.0444** 0.0379* 0.0374** 0.0267
(0.0223) (0.0222) (0.0183) (0.0187)
65 - 70 0.1000*** 0.0851*** 0.132*** 0.118***
(0.0223) (0.0231) (0.0217) (0.0219)
70 - 75 0.168*** 0.150*** 0.168*** 0.150***
(0.0319) (0.0326) (0.0266) (0.0266)
75 - 80 0.213*** 0.193*** 0.146*** 0.129***
(0.0412) (0.0414) (0.0310) (0.0317)
80 - 0.213*** 0.192*** 0.208*** 0.191***
(0.0424) (0.0436) (0.0386) (0.0390)
Death Father -0.00188 0.0548**
(0.0277) (0.0213)
Mother 0.0401** 0.0301*
(0.0171) (0.0155)
GHS Father -0.00758 -0.00981
(0.0193) (0.0169)
Mother 0.00468 0.00413
(0.0203) (0.0168)
ADL Father -0.00573 0.0178
(0.0225) (0.0184)
62
Mother 0.00929 -0.0156
(0.0204) (0.0171)
Father's Education At least some primary -0.00208 0.0219
(0.0271) (0.0324)
Completed primary 0.00566 -0.00319
(0.0215) (0.0204)
Completed Jr. high -0.0109 -0.0422
(0.0334) (0.0284)
Mother's Education At least some primary -0.0443 -0.00858
(0.0279) (0.0298)
Completed primary -0.0379 -0.0185
(0.0240) (0.0197)
Completed Jr. high -0.0498 0.0594
(0.0387) (0.0404)
Constant -0.0155 -0.0724 -0.136 -0.223**
(0.118) (0.123) (0.109) (0.113)
R-squared 0.116 0.122 0.112 0.118
Sample size 2974 2974 3482 3482
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.0042 0.0015 0.0746 0.0559
Education dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's death dummy variables 0.0630 0.0026
Parent's education dummy variables 0.3116 0.3233
Parent's GHS dummy variables 0.9245 0.8456
Parent's ADL dummy variables 0.9007 0.5344
Parent's health (death, GHS, ADL) dummy variables 0.2365 0.0297
Parent's health (death, GHS, ADL)+ edu dummy 0.1870 0.0499
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
63
Table 2.5.5: Health-SES Gradients of Older Adults 5
BMI (25)
Men Women
Respondent's Edu At least some primary 0.0298* 0.0258 0.0714*** 0.0599***
(0.0163) (0.0166) (0.0211) (0.0217)
Completed primary 0.0705*** 0.0593*** 0.150*** 0.133***
(0.0191) (0.0196) (0.0236) (0.0244)
Completed Jr. high 0.168*** 0.146*** 0.209*** 0.184***
(0.0319) (0.0330) (0.0398) (0.0419)
Completed high school 0.261*** 0.230*** 0.273*** 0.241***
(0.0262) (0.0282) (0.0336) (0.0391)
Respondent's height 0 - median -0.00695*** -0.00704*** -0.00481*** -0.00497***
(0.000822) (0.000840) (0.000969) (0.000940)
median 0.0127*** 0.0126*** 0.00198 0.00222
(0.00253) (0.00252) (0.00259) (0.00254)
Respondent's age 60 - 65 -0.0239 -0.0195 -0.0607*** -0.0438*
(0.0213) (0.0213) (0.0224) (0.0232)
65 - 70 -0.0821*** -0.0696*** -0.106*** -0.0861***
(0.0185) (0.0190) (0.0230) (0.0241)
70 - 75 -0.0970*** -0.0866*** -0.159*** -0.132***
(0.0210) (0.0215) (0.0252) (0.0259)
75 - 80 -0.0579** -0.0448 -0.153*** -0.120***
(0.0277) (0.0283) (0.0292) (0.0305)
80 - -0.104*** -0.0911*** -0.222*** -0.192***
(0.0240) (0.0251) (0.0276) (0.0291)
Death Father 0.0425 -0.0103
(0.0321) (0.0370)
Mother -0.0312 -0.0643**
(0.0220) (0.0258)
GHS Father 0.00720 0.0180
(0.0195) (0.0208)
Mother -0.00932 -0.0204
(0.0195) (0.0218)
ADL Father -0.0284 -0.0319
(0.0221) (0.0224)
64
Mother 0.0106 0.0365
(0.0222) (0.0240)
Father's Education At least some primary 0.0116 0.0121
(0.0303) (0.0364)
Completed primary 0.0298 0.0636**
(0.0232) (0.0273)
Completed Jr. high 0.00314 0.0964**
(0.0429) (0.0479)
Mother's Education At least some primary -0.0295 0.0447
(0.0322) (0.0391)
Completed primary 0.0395 -0.0393
(0.0258) (0.0326)
Completed Jr. high 0.132* -0.106
(0.0684) (0.0694)
Constant 1.327*** 1.313*** 1.230*** 1.306***
(0.160) (0.167) (0.157) (0.159)
R-squared 0.127 0.135 0.138 0.147
Sample size 2974 2974 3482 3482
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0001 0.0000 0.0000
Height variables 0.0000 0.0000 0.0000 0.0000
Education dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's death dummy variables 0.2398 0.0374
Parent's education dummy variables 0.0529 0.1020
Parent's GHS dummy variables 0.8838 0.5936
Parent's ADL dummy variables 0.3985 0.2390
Parent's health (death, GHS, ADL) dummy variables 0.5092 0.0729
Parent's health (death, GHS, ADL)+ edu dummy 0.0835 0.0200
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
65
Table 2.5.6: Health-SES Gradients of Older Adults 6
Hemoglobin (M:<13, W:<12)
Men Women
Respondent's Edu At least some primary -0.0732** -0.0661** 0.00221 0.00481
(0.0291) (0.0293) (0.0215) (0.0223)
Completed primary -0.0620** -0.0570* -0.0371 -0.0287
(0.0297) (0.0302) (0.0265) (0.0274)
Completed Jr. high -0.105*** -0.0908** -0.0295 -0.0188
(0.0378) (0.0378) (0.0367) (0.0382)
Completed high school -0.146*** -0.139*** -0.0945*** -0.0813**
(0.0320) (0.0330) (0.0332) (0.0369)
Respondent's height 0 - median -0.00145 -0.00113 0.00131 0.00130
(0.00116) (0.00115) (0.000810) (0.000819)
median -0.00212 -0.00253 -0.000692 -0.000552
(0.00270) (0.00269) (0.00256) (0.00255)
Respondent's age 60 - 65 0.0409* 0.0321 0.0432* 0.0398
(0.0243) (0.0240) (0.0248) (0.0254)
65 - 70 0.124*** 0.109*** 0.0882*** 0.0867***
(0.0252) (0.0263) (0.0246) (0.0254)
70 - 75 0.209*** 0.192*** 0.122*** 0.119***
(0.0318) (0.0327) (0.0297) (0.0310)
75 - 80 0.201*** 0.179*** 0.187*** 0.182***
(0.0444) (0.0444) (0.0388) (0.0392)
80 - 0.279*** 0.255*** 0.276*** 0.269***
(0.0418) (0.0419) (0.0392) (0.0402)
Death Father 0.00592 0.0126
(0.0343) (0.0349)
Mother 0.0652*** 0.0104
(0.0221) (0.0212)
GHS Father -0.0242 0.0160
(0.0205) (0.0225)
Mother -0.00608 -0.0361*
(0.0218) (0.0219)
ADL Father 0.0221 -0.0349
(0.0256) (0.0247)
66
Mother -0.0205 0.0483**
(0.0239) (0.0236)
Father's Education At least some primary -0.0410 -0.0275
(0.0331) (0.0359)
Completed primary -0.0164 -0.0253
(0.0258) (0.0271)
Completed Jr. high -0.00911 -0.0432
(0.0417) (0.0454)
Mother's Education At least some primary 0.118*** 0.0415
(0.0390) (0.0363)
Completed primary 0.00720 0.00804
(0.0297) (0.0286)
Completed Jr. high -0.00322 0.0119
(0.0574) (0.0608)
Constant 0.559*** 0.468** 0.200 0.186
(0.194) (0.196) (0.140) (0.146)
R-squared 0.124 0.133 0.095 0.098
Sample size 2979 2979 3474 3474
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.0486 0.0741 0.2324 0.2387
Education dummy variables 0.0001 0.0005 0.0272 0.1328
Parent's death dummy variables 0.0118 0.8143
Parent's education dummy variables 0.1479 0.8768
Parent's GHS dummy variables 0.3106 0.2508
Parent's ADL dummy variables 0.6146 0.1149
Parent's health (death, GHS, ADL) dummy variables 0.0474 0.4132
Parent's health (death, GHS, ADL)+ edu dummy 0.0356 0.7366
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
67
Table 2.5.7: Health-SES Gradients of Older Adults 7
Hypertension
Men Women
Respondent's Edu At least some primary -0.0145 -0.0151 -0.00383 0.00274
(0.0323) (0.0330) (0.0200) (0.0207)
Completed primary 0.00594 0.00137 0.0117 0.0249
(0.0328) (0.0347) (0.0258) (0.0274)
Completed Jr. high 0.0496 0.0371 -0.0241 -0.00426
(0.0404) (0.0428) (0.0381) (0.0400)
Completed high school 0.0596 0.0509 -0.0187 0.00307
(0.0380) (0.0421) (0.0367) (0.0385)
Respondent's height 0 - median 0.000327 0.000193 -0.00125* -0.00117
(0.00113) (0.00111) (0.000744) (0.000764)
median -0.00337 -0.00344 0.00350 0.00338
(0.00301) (0.00302) (0.00250) (0.00248)
Respondent's age 60 - 65 0.0815*** 0.0722** 0.0892*** 0.0769***
(0.0289) (0.0290) (0.0248) (0.0254)
65 - 70 0.127*** 0.110*** 0.149*** 0.135***
(0.0279) (0.0297) (0.0242) (0.0252)
70 - 75 0.190*** 0.168*** 0.234*** 0.218***
(0.0319) (0.0331) (0.0292) (0.0309)
75 - 80 0.207*** 0.188*** 0.214*** 0.191***
(0.0405) (0.0411) (0.0309) (0.0325)
80 - 0.225*** 0.200*** 0.289*** 0.265***
(0.0407) (0.0415) (0.0298) (0.0319)
Death Father 0.0264 0.0175
(0.0415) (0.0377)
Mother 0.0579** 0.0524**
(0.0274) (0.0248)
GHS Father 0.0208 -0.00436
(0.0249) (0.0235)
Mother -0.0127 0.0105
(0.0248) (0.0226)
ADL Father -0.00540 0.00732
(0.0288) (0.0231)
68
Mother -0.00190 -0.00465
(0.0278) (0.0239)
Father's Education At least some primary -0.0571 0.00168
(0.0386) (0.0382)
Completed primary 0.0211 -0.0282
(0.0315) (0.0272)
Completed Jr. high 0.0252 -0.0230
(0.0504) (0.0522)
Mother's Education At least some primary -0.0550 0.0360
(0.0422) (0.0410)
Completed primary 0.0235 -0.00407
(0.0323) (0.0325)
Completed Jr. high 0.0693 0.00709
(0.0678) (0.0662)
Constant 0.564*** 0.487** 0.774*** 0.701***
(0.198) (0.202) (0.131) (0.142)
R-squared 0.074 0.084 0.087 0.092
Sample size 2985 2985 3493 3493
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.4993 0.4662 0.1989 0.2415
Education dummy variables 0.1008 0.2859 0.8641 0.8748
Parent's death dummy variables 0.0655 0.0776
Parent's education dummy variables 0.0138 0.8228
Parent's GHS dummy variables 0.7042 0.8980
Parent's ADL dummy variables 0.9584 0.9506
Parent's health (death, GHS, ADL) dummy variables 0.3973 0.3772
Parent's health (death, GHS, ADL)+ edu dummy 0.0152 0.6863
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
69
Table 2.5.8: Health-SES Gradients of Older Adults 8
Cholesterol (240)
Men Women
Respondent's Edu At least some primary -0.0112 -0.0158 0.0272 0.0279
(0.0171) (0.0174) (0.0205) (0.0206)
Completed primary 0.0149 0.00911 0.0492** 0.0449*
(0.0171) (0.0179) (0.0225) (0.0232)
Completed Jr. high 0.0632** 0.0540* 0.0801** 0.0732*
(0.0276) (0.0285) (0.0367) (0.0377)
Completed high school 0.0539** 0.0443* 0.116*** 0.111***
(0.0219) (0.0234) (0.0326) (0.0364)
Respondent's height 0 - median 0.000667* 0.000567 -0.000987 -0.000957
(0.000379) (0.000401) (0.000811) (0.000809)
median 0.00114 0.00123 0.00273 0.00254
(0.00184) (0.00188) (0.00243) (0.00244)
Respondent's age 60 - 65 -0.0274* -0.0269* 0.0227 0.0225
(0.0159) (0.0160) (0.0210) (0.0214)
65 - 70 -0.0450*** -0.0442*** 0.000441 -0.000515
(0.0151) (0.0166) (0.0213) (0.0219)
70 - 75 -0.00292 0.00105 -0.00251 -0.0000991
(0.0205) (0.0214) (0.0247) (0.0258)
75 - 80 -0.0306 -0.0281 0.00521 0.00374
(0.0246) (0.0250) (0.0304) (0.0307)
80 - -0.0564** -0.0509** 0.00219 0.00683
(0.0226) (0.0240) (0.0329) (0.0336)
Death Father 0.00471 0.0227
(0.0254) (0.0334)
Mother -0.00134 -0.00856
(0.0179) (0.0213)
GHS Father 0.0172 -0.0174
(0.0160) (0.0196)
Mother 0.00517 0.00762
(0.0147) (0.0203)
ADL Father 0.00454 0.0103
(0.0186) (0.0213)
70
Mother -0.00568 -0.0109
(0.0171) (0.0222)
Father's Education At least some primary -0.0227 -0.0299
(0.0227) (0.0320)
Completed primary 0.0191 0.0554**
(0.0194) (0.0238)
Completed Jr. high 0.0111 0.0158
(0.0328) (0.0423)
Mother's Education At least some primary 0.0441 -0.0835***
(0.0274) (0.0294)
Completed primary 0.00160 -0.0524*
(0.0206) (0.0290)
Completed Jr. high -0.00484 0.0335
(0.0534) (0.0644)
Constant -0.0293 -0.0266 0.339** 0.322**
(0.0774) (0.0834) (0.151) (0.154)
R-squared 0.100 0.104 0.106 0.116
Sample size 2957 2957 3457 3457
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0156 0.0435 0.9302 0.9402
Height variables 0.0827 0.1534 0.4055 0.4384
Education dummy variables 0.0033 0.0160 0.0044 0.0291
Parent's death dummy variables 0.9824 0.7417
Parent's education dummy variables 0.5258 0.0016
Parent's GHS dummy variables 0.3365 0.6719
Parent's ADL dummy variables 0.9447 0.8568
Parent's health (death, GHS, ADL) dummy variables 0.8401 0.9702
Parent's health (death, GHS, ADL)+ edu dummy 0.7399 0.0188
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
71
Table 2.5.9: Health-SES Gradients of Older Adults 9
HDL (<40)
Men Women
Respondent's Edu At least some primary 0.0263 0.0319 -0.0437* -0.0462**
(0.0250) (0.0253) (0.0227) (0.0233)
Completed primary -0.0378 -0.0359 -0.0419* -0.0344
(0.0285) (0.0296) (0.0251) (0.0262)
Completed Jr. high -0.00332 -0.00369 -0.0612* -0.0403
(0.0390) (0.0401) (0.0361) (0.0375)
Completed high school -0.0613* -0.0754** -0.0768** -0.0529
(0.0342) (0.0375) (0.0350) (0.0391)
Respondent's height 0 - median 0.000796 0.000981 0.00163** 0.00168**
(0.00112) (0.00113) (0.000753) (0.000781)
median 0.00430 0.00387 0.000243 -0.0000631
(0.00277) (0.00279) (0.00274) (0.00280)
Respondent's age 60 - 65 -0.0242 -0.0260 -0.0584** -0.0477**
(0.0257) (0.0263) (0.0230) (0.0238)
65 - 70 -0.0457* -0.0460* 0.0152 0.0239
(0.0257) (0.0268) (0.0253) (0.0264)
70 - 75 -0.0798** -0.0896*** -0.0285 -0.0176
(0.0313) (0.0321) (0.0276) (0.0289)
75 - 80 -0.0168 -0.0210 -0.0707** -0.0574
(0.0420) (0.0429) (0.0357) (0.0371)
80 - 0.000617 -0.0102 -0.0499 -0.0340
(0.0428) (0.0427) (0.0363) (0.0371)
Death Father -0.00435 -0.0337
(0.0378) (0.0363)
Mother 0.0366 -0.0292
(0.0254) (0.0244)
GHS Father 0.00595 -0.0220
(0.0244) (0.0225)
Mother -0.0548** 0.0234
(0.0227) (0.0216)
ADL Father -0.00316 0.0631**
(0.0306) (0.0254)
72
Mother 0.0185 -0.0640***
(0.0299) (0.0240)
Father's Education At least some primary 0.0286 0.00434
(0.0362) (0.0397)
Completed primary -0.0208 -0.0452*
(0.0282) (0.0265)
Completed Jr. high 0.0398 -0.0170
(0.0501) (0.0511)
Mother's Education At least some primary -0.0205 0.0686
(0.0435) (0.0436)
Completed primary 0.0361 -0.00321
(0.0314) (0.0304)
Completed Jr. high 0.110 -0.0682
(0.0682) (0.0709)
Constant 0.593*** 0.538*** 0.226* 0.297**
(0.192) (0.197) (0.130) (0.141)
R-squared 0.083 0.090 0.103 0.109
Sample size 2931 2931 3445 3445
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.1669 0.1301 0.0330 0.1032
Height variables 0.0478 0.0553 0.0471 0.0533
Education dummy variables 0.0145 0.0058 0.1352 0.3499
Parent's death dummy variables 0.3538 0.2461
Parent's education dummy variables 0.2056 0.1938
Parent's GHS dummy variables 0.0329 0.4861
Parent's ADL dummy variables 0.7813 0.0140
Parent's health (death, GHS, ADL) dummy variables 0.2911 0.1069
Parent's health (death, GHS, ADL)+ edu dummy 0.2069 0.0757
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
73
Table 2.5.10: Health-SES Gradients of Older Adults 10
Cognition
Men Women
Respondent's Edu At least some primary 0.131 0.0980 0.430*** 0.366***
(0.127) (0.130) (0.0892) (0.0887)
Completed primary 0.710*** 0.674*** 0.945*** 0.830***
(0.135) (0.138) (0.0949) (0.0967)
Completed Jr. high 1.032*** 0.977*** 1.310*** 1.138***
(0.152) (0.156) (0.132) (0.138)
Completed high school 1.471*** 1.383*** 2.023*** 1.751***
(0.138) (0.146) (0.116) (0.130)
Respondent's height 0 - median -0.000844 -0.00105 -0.000669 -0.000136
(0.00350) (0.00357) (0.00324) (0.00319)
median 0.0123 0.0122 -0.00835 -0.00912
(0.0101) (0.0102) (0.00933) (0.00942)
Respondent's age 60 - 65 -0.370*** -0.342*** -0.319*** -0.281***
(0.0907) (0.0917) (0.0877) (0.0888)
65 - 70 -0.616*** -0.558*** -0.508*** -0.466***
(0.0854) (0.0889) (0.0831) (0.0872)
70 - 75 -0.768*** -0.709*** -0.563*** -0.507***
(0.117) (0.118) (0.123) (0.125)
75 - 80 -1.044*** -0.960*** -0.815*** -0.748***
(0.145) (0.147) (0.155) (0.157)
80 - -1.274*** -1.232*** -1.045*** -1.012***
(0.156) (0.159) (0.215) (0.219)
Death Father -0.100 -0.269**
(0.126) (0.132)
Mother -0.112 -0.0377
(0.0796) (0.0798)
GHS Father -0.00202 -0.0250
(0.0793) (0.0766)
Mother -0.0713 -0.0335
(0.0784) (0.0744)
ADL Father -0.0658 0.0263
(0.0914) (0.0899)
74
Mother -0.0244 0.128
(0.0885) (0.0781)
Father's Education At least some primary 0.356*** 0.205
(0.136) (0.128)
Completed primary -0.0647 0.284***
(0.0922) (0.0911)
Completed Jr. high 0.0777 0.609***
(0.157) (0.174)
Mother's Education At least some primary -0.192 0.0266
(0.139) (0.136)
Completed primary 0.182* -0.0666
(0.103) (0.0962)
Completed Jr. high -0.0341 -0.154
(0.218) (0.210)
Constant 3.892*** 4.179*** 3.570*** 3.669***
(0.580) (0.618) (0.513) (0.531)
R-squared 0.265 0.276 0.309 0.322
Sample size 2317 2317 2309 2309
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0000 0.0000 0.0000 0.0000
Height variables 0.4222 0.4437 0.5283 0.5420
Education dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's death dummy variables 0.1576 0.0957
Parent's education dummy variables 0.0478 0.0031
Parent's GHS dummy variables 0.5790 0.7681
Parent's ADL dummy variables 0.6073 0.1200
Parent's health (death, GHS, ADL) dummy variables 0.1857 0.1487
Parent's health (death, GHS, ADL)+ edu dummy 0.0181 0.0028
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
75
Table 2.5.11: Health-SES Gradients of Older Adults 11
Depression
Men Women
Respondent's Edu At least some primary -0.392* -0.328 -0.250 -0.144
(0.203) (0.203) (0.166) (0.170)
Completed primary -0.694*** -0.536*** -0.826*** -0.660***
(0.196) (0.198) (0.188) (0.198)
Completed Jr. high -0.743*** -0.570** -1.322*** -1.085***
(0.273) (0.279) (0.257) (0.279)
Completed high school -1.570*** -1.344*** -1.802*** -1.390***
(0.211) (0.224) (0.250) (0.287)
Respondent's height 0 - median 0.00575 0.00392 -0.000850 -0.000534
(0.00593) (0.00596) (0.00630) (0.00651)
median -0.00919 -0.00653 0.0175 0.0196
(0.0213) (0.0206) (0.0189) (0.0189)
Respondent's age 60 - 65 -0.134 -0.185 -0.165 -0.282*
(0.152) (0.152) (0.158) (0.163)
65 - 70 0.442** 0.327* 0.167 0.0586
(0.181) (0.182) (0.190) (0.193)
70 - 75 0.475** 0.315 0.228 0.0757
(0.209) (0.206) (0.225) (0.232)
75 - 80 0.840*** 0.741** 0.496* 0.342
(0.309) (0.302) (0.280) (0.288)
80 - 1.390*** 1.301*** 1.273*** 1.105***
(0.369) (0.367) (0.340) (0.343)
Death Father -0.0338 0.261
(0.275) (0.299)
Mother 0.112 0.266
(0.175) (0.195)
GHS Father 0.275* 0.101
(0.151) (0.160)
Mother 0.295** 0.535***
(0.148) (0.171)
ADL Father 0.0854 0.287
(0.197) (0.199)
76
Mother 0.250 -0.227
(0.182) (0.183)
Father's Education At least some primary 0.0416 -0.269
(0.244) (0.269)
Completed primary -0.302* -0.229
(0.165) (0.194)
Completed Jr. high 0.397 -0.538
(0.312) (0.351)
Mother's Education At least some primary -0.533* 0.137
(0.278) (0.333)
Completed primary -0.204 -0.0878
(0.177) (0.222)
Completed Jr. high -0.0784 0.132
(0.435) (0.504)
Constant 3.774*** 3.306*** 4.402*** 3.521***
(1.004) (1.042) (1.095) (1.167)
R-squared 0.101 0.122 0.103 0.116
Sample size 2830 2830 3222 3222
Birth place dummy variables Yes Yes Yes Yes
F-test (p-values)
Age dummy variables 0.0001 0.0006 0.0006 0.0014
Height variables 0.6249 0.8055 0.6192 0.5258
Education dummy variables 0.0000 0.0000 0.0000 0.0000
Parent's death dummy variables 0.8147 0.2003
Parent's education dummy variables 0.0150 0.6035
Parent's GHS dummy variables 0.0007 0.0006
Parent's ADL dummy variables 0.1570 0.2933
Parent's health (death, GHS, ADL) dummy variables 0.0001 0.0003
Parent's health (death, GHS, ADL)+ edu dummy 0.0000 0.0017
Birth place dummy variables 0.0000 0.0000 0.0000 0.0000
Source: IFLS4 Note: Standard errors in parenthesis. (clustered at community level)
*** signicant at 1%, ** signicant at 5%, * signicant at 10% .
Dummy variables capturing missing observations are included for father/mother's education,GHS and ADL but not reported here.
77
SES model shows that own education is strongly correlated with most all of the health
outcomes, those with higher schooling generally having better health outcomes. There are
some exceptions, notably men and women with higher education are more likely to be
overweight and have high total cholesterol. These associations may well be through diet,
though I cannot tell with the IFLS data. On the other hand, more schooling is associated
with a lower likelihood of having low HDL, which is common in Indonesia, particularly
among men. Better education is also associated with better cognition, as measured by
word recall, with lower depression scores, with a lower likelihood of being underweight, of
lower number of problems performing ADLs and IADLs, and a lower likelihood of being
in poor general health. Adult height is correlated with some of the health variables, such
as number of ADLs or IADLs a woman has problems with; taller women reporting fewer
such problems. The birth district dummies are highly signicant in all cases.
Results from the extended model show that most coecients of parental health mea-
sures remain signicant as before with the magnitudes of the coecients being very similar
to the models in tables 2.1.1 - 2.1.4, without the respondent's height and own education
variables. This is quite interesting and is consistent with parental health having impacts
over and above through respondent education and height. In contrast, the parental school-
ing coecients drop in magnitude substantially when respondent schooling dummies are
added to the specication.
Also, comparing the specications that have only respondent's SES covariates with
the ones that have both respondent's and parental SES variables, the magnitudes of the
respondent's SES coecients drop modestly when parental SES variables are added, but
remain signicant as they were to begin with.
78
2.5 Conclusion
Family background is strongly correlated with various aspects of children's life even when
they grow older. This chapter examines the dimension of family health correlations which,
despite their importance, have not been explored much due to data limitations. IFLS
provides a suitable platform to examine the intergenerational health correlations, because
it encompasses detailed information of both parents and their adult children.
The ndings suggest that there are positive intergenerational correlations between
parental health and education, and the health status of their ospring. While these cor-
relations should not be interpreted as causal, they are consistent with the types of inter-
generational correlations found for schooling and incomes. The correlations persist after
controlling for respondent SES: education and height. These health associations become
much lower, however, for respondents who were born in more developed areas such as Java
or Bali. Being born and growing up in more developed areas apparently substitutes for
in
uences inherited from parents. This result highlights the importance of programs and
policies to focus on community level infrastructure development in less developed areas.
79
Chapter 3
The Dynamics of Health and Its
Determinants among Older Adults
3.1 Introduction
There has been a large and growing literature on early life origins of health as an adult (eg.
Barker et al., 1989; Godfrey and Barker, 2000; Crimmins and Finch, 2004, 2006; Case et
al., 2005, and Smith, 2009). These papers nd strong correlations between early childhood
health and health as older adults. For example, Smith (2009) shows that a measure of
childhood health is a good predictor for the onset of diseases during adulthood. However,
very little is known about the dynamics of health at older ages.
In Indonesia, for example, hypertension is one of the most common chronic diseases.
1
Using the Indonesian Family Life Survey (IFLS), a panel survey spanning 14 years, table
3.1.1 reports the prevalence of hypertension over time among the elderly of age 50 and
above. Among those with hypertension in the past, 80% are still hypertensive when mea-
sured in the next wave of the survey, between three and seven years later. Elderly women
show an even stronger persistence in hypertension than elderly men. The correlations
between past and current hypertension are also high, 0.47 and 0.52 for men and women
respectively, again implying that it is hard to recover from hypertension, once developed.
1
Witoelar et al. (2012) nd that around 44 percent of elderly men and 53 percent of elderly women have
hypertension in Indonesia in 2007.
80
The question is why do we observe strong persistence in chronic diseases like hyper-
tension? Is it childhood health that has permanent consequences which is consistent with
the ndings of early life origins or are there some other factors explaining this persistence?
Among possible reasons that might cause persistent chronic health problems are genetic
endowments, in
uence of childhood health and nutrition, health inputs and behaviors as
an adult, and the past history of chronic conditions. Health inputs and behaviors will be
mediated by socio-economic status. This study makes a start in disentangling the separate
in
uences using longitudinal data from Indonesia.
The main contribution of this chapter is to characterize the evolution of health over
time, especially during later life. It is reasonable to describe the current health of older
adults in a dynamic framework since it is a consequence of past health inputs or behaviors.
However, incorporating dynamics into an empirical model causes potential endogeneity
problems because past health is likely to be correlated with time-invariant unobserved
characteristics. I address this problem by deriving a health demand function conditioned
on lagged health, and estimating it using a rst-dierence two-step generalized method
of moments (FD-GMM), proposed by Arellano-Bond (1991), where the rst-dierencing
removes xed unobserved factors and keeps only lagged health.
Using a panel data set from Indonesia, the Indonesian Family Life Survey (IFLS), I
nd that it is genetic endowments and early life cycle including childhood health, captured
in the individual xed eects, that play the most important roles in determining chronic
health conditions in later life. The impact of past health conditions, captured in the
coecients on lagged health measures, is weak, conditional on the xed eects. This result
is robust to measurement errors in lagged health and to sample attrition including from
selective mortality. Socio-economic status also has very little in
uence on current health,
conditional on both xed eects and lagged health. These results are consistent with the
large and growing literature on early life origins of later life health.
81
In order to investigate if past health has dierent impacts across demographic or eco-
nomic groups, I disaggregate the sample across age, per capita household expenditure level
(PCE), and years of education. The results show that those with less education tend to
show more persistence of chronic conditions, compared to those with higher education.
In terms of policy, a strong impact of early childhood health implies that investments
should be targeted towards the early stages of life. Stronger persistence of chronic con-
ditions of lower educated elderly also suggests that investments should be made not only
towards health in early life, but also, towards better education in early life. Among the
elderly, the focus should be on those with lower education attainments.
The rest of the chapter is organized as follows. Section 3.2 provides a brief review of the
related literature. A theoretical model is presented in section 3.3. Section 3.4 describes the
data and the variables used in this chapter, and section 3.5 presents the empirical model.
The main regression results are discussed in section 3.6. Concluding remarks follow in
section 3.7.
3.2 Studies of Health Dynamics
There exists some literature that estimates the degree to which children can recover from
health problems caused by nutritional deciencies at earlier ages (eg. Habicht et. al, 1995;
Adair, 1999; Mortorell, 1995, 1999; Hoddinott and Kinsey, 2001; Fedorov and Sahn, 2005;
Alderman et.al, 2006; Mani, 2012). In these studies, height is used as an indicator of
cumulative health during early childhood. For example, INCAP (Institute of Nutrition
of Central America and Panama) conducted randomized experiments in rural Guatemala
during the late 1960s and 1970s by randomly giving protein supplementation. They nd
that for children exposed to supplementation before reaching age of 3 tend to be taller and
less stunted. (eg. Habicht et. al, 1995)
82
Other studies have used panel data sets and examined the magnitudes of coecients
on lagged height for children in dynamic panel models, the \catch-up eect". Hoddinott
and Kinsey (2001), for instance, estimate the catch-up eect at 0.56. They adopt both a
two-stage least squares (2SLS) method and maternal xed eects using a sample of children
living in rural areas of Zimbabwe. Fedorov and Sahn (2005), on the other hand, estimate a
coecient on lagged height at 0.22 for children in Russia by using an Arellano-Bover (1995)
estimation. Alderman et.al (2006) employ a maternal xed eects instrumental variable
estimation (MFE-IV) which yields the catch-up coecient of 0.43 for children in Zimbabwe.
Mani (2012) also nds a partial catch-up eect, 0.23 for malnourished children in Indonesia
by using a rst-dierence GMM method (FD-GMM). While evidence of catch-up eects for
children vary some depending on countries and econometric methods, it is worth noting
that these results still suggest that children tend to recover from some, but not all, of
the height decits incurred in the past. Examining child health and its consequences is
important in the sense that undernourishment at early age may cause permanent growth
retardation and in turn, aect future well-being by lowering the human capital such as
schooling or earnings. For instance, Hoddinott et al. (2008) nd, using the INCAP data,
that children who were exposed to protein supplements between age 0 and 2 years have
a 46% increase in average wages. Therefore, studies of health and nutrition status during
early childhood and whether children could recover from past decits have received much
attention, especially in developing countries where resources are limited.
However, studies of health dynamics focusing on the elderly are very rare. Adams et
al. (2003), for example, use (ordered) probit models and look at the impact of past health
and socio-economic status on the future onsets of several diseases among the American
elderly. They control for several past health conditions as well as current incidences as
covariates and list them by following Wold causal chain such as cancer, heart disease,
stroke, degenerative and chronic conditions, and mental condition. The Wold causal chain
83
assumes that the rst component causes the second, the rst and the second cause the
third and so forth. However, this might not be perfectly true since mental conditions,
such as depression can increase the risk of having cancer or stroke
2
. Therefore, instead of
having several health measures, I analyze one health measure at a time without allowing
any contemporaneous in
uences from other co-morbidities. Adams et al. nd a strong
impact of past health on current health such as those who had an incontinence problem in
the past tend to have the same problem in the next period and the past history of having
arthritis, lung problems and the numbers of diculties of ADLs and IADLs also predict a
future incontinence problem.
Smith (2007), examining future onset of disease does not allow for contemporaneous
correlation between disease and chronic conditions but controls for several past health
measures such as self-assessed general health status and chronic conditions as right-hand
side regressors. Using the Panel Study of Income Dynamics (PSID), he nds that the future
onsets of disease are strongly correlated with past self-assessed general health status and
chronic health conditions. However, neither Adams et al.(2003) nor Smith (2007) addresses
the problem arising from the possible correlation between previous health measures and
unobserved factors like individual-specic xed eects. It is most likely that previous health
status, measured by lagged health condition, is correlated with individual time-invariant
xed eects and this will create a bias in OLS coecient estimates. Genetic endowments
and childhood health are many times unobserved in data and and so are captured in
the individual xed eect. A person with genetically stronger endowments or who had a
healthy childhood is more likely to be healthy at all times, and being healthy leads to higher
education and earnings, which further leads to better health outcomes in the subsequent
period. In this case, the impact of previous health can be over-estimated.
2
See Penninx, B. et al.(1998) and Jonas et al.(2000) for details
84
Using the National Longitudinal Surveys of Labor Market Experience (NLS), Cooksey
et al. (2012) show a strong correlation between past and current depression based on
transition matrix using CES-D data. The coecient on lagged depression is 0.86. However,
they found that when the xed eect is taken into account, this xed eect explains almost
27% of the persistence, 30% comes from lagged depression, and error terms explain the
rest. Moreover, the correlations between past and current depression do not substantially
attenuate over time, again suggesting the presence of time-invariant components.
Goldman et al. (2011), using the Health and Retirement Study, nd that past BMI
from two and four years ago are both strongly correlated with current BMI. They address
endogeneity issues by employing system GMM proposed by Arellano-Bover(1995).
3
How-
ever, they allow contemporaneous causalities by including various health conditions such as
doctor's diagnosis and self-evaluated measures. As mentioned earlier, adding many health
variables can be troublesome especially if it is not very clear in which direction that the
causality runs.
Many studies exist that examine the impact of early childhood health on own later
health (eg. Barker et al., 1989; Godfrey and Barker, 2000; Crimmins and Finch, 2004,
2006; Case et al., 2005; Smith, 2009). A leading theory is the early life origins hypothesis.
These examples include Barker's argument that organ sizes or function and gene expression
may adapt to a new environment in order to raise survival probabilities when exposed to
negative shocks during very early childhood or the fetal period (eg. Barker et al., 1989).
Crimmins and Finch (2004) argue that in
ammation caused by infectious diseases in early
life increases the risk of later life morbidity and mortality. Examples of studies that examine
the early life origins hypothesis also include Case et al. (2005), who use a 1958 British
3
The system GMM estimator can be more ecient than the FD-GMM estimator (Arellano-Bond esti-
mator, 1991) only if the variables are close to a random walk. However, in this chapter, F-statistics from
the rst-stage regression prove that lagged health measures that are used as IVs, are strongly correlated
with current health and hence, FD-GMM still gives ecient estimators.
85
birth cohort, conditioned on parental income, education and social status and show that
respondent's birthweight is strongly associated with poor general health in adulthood,
with the correlation becoming larger as people age. As another example, Smith (2009)
nds that a measure of childhood health is a good predictor for the onset of disease during
adulthood such as cancer, heart problem and hypertension. These studies, however, can be
misinterpreted since they do not account for unobserved factors such as individual specic
unobserved characteristics or time-varying shocks. The impact of early health might be
over-estimated for those with stronger genetic endowments or under-estimated for those
exposed to temporary health shocks at the time of the survey of later health status.
3.3 Model
Health is considered as one of the important components of human capital. Individuals
make decisions on health investment or health behaviors such as nutrient intakes, exercise,
and health care utilization in order to have better health and achieve both physical and
subjective well-being. Similar to agriculture or manufacturing production functions, the
health production function represents the technology available to individual or households
to convert health inputs to health outcomes. The current health status can be expressed
as the outcomes of the health production function. Given that health evolves over the life
course, it is reasonable to describe the current health in a dynamic framework, as a function
of all current and prior health inputs, individual demographic characteristics, household
background characteristics, the disease or public health environment, time-varying health
shocks, and time-invariant genetic endowments.
H
t
=h(N
t
;N
t1
;::;N
0
;I
t
;I
t1
;::;I
0
;D
t
;D
t1
;::;D
0
;
et
;
et1
;::
e0
;
e
;
ht
;
ht1
;::;
h0
;
h
)
(3.1)
86
H
t
is the current health status measured by the number of diculties of activities of
daily living (ADLs), body mass index (BMI), self-assessed general health status (GHS),
hemoglobin, hypertension, and lung capacity. N
t
is the health inputs at time t and the
health production function contains not only the current health inputs, but also the inputs
from all previous periods which may potentially have impacts on the current health sta-
tus. The model assumes that health inputs should be distinguished from consumption
goods, and to the degree that they aect health, consumption goods should be regarded
as health inputs. Individuals can not directly obtain utility from consuming health inputs.
I
t
captures infrastructure availability and the disease or public health environment in the
community where the elderly person lives at time t. D
t
denotes time-varying demographic
characteristics of the older adult such as the respondent's age.
et
re
ects time-varying
health shocks from birth to present.
e
is individual time-invariant characteristics including
the older adult's gender and genetic endowments.
ht
and
h
summarize the information
about household specic time-varying and time-invariant characteristics respectively. Fol-
lowing Grossman (1972), I assume that the one-period lagged health is sucient to capture
the information of all previous factors such as health inputs, community environments, and
other time-varying demographic characteristics from time 0 until t-1 in the sample.
4
Thus,
the current health can be re-expressed as a function of one-period lagged health status,
current period health inputs, current period community environment, individual specic
demographic characteristics, health shocks, and household specic characteristics.
H
t
=h(H
t1
;N
t
;I
t
;D
t
;
et
;
e
;
ht
;
h
) (3.2)
4
This assumption can be considered somewhat strong. However, testing this assumption is beyond the
scope of this paper.
87
Each household solves an inter-temporal utility maximization problem subject to a lifetime
budget constraint and individual's dynamic health production function at time t. The
household utility function in each period depends on consumption goods, C
t
, leisure, L
t
,
and the health status of the older adult, H
t
. The optimal choice of health inputs can be
derived from the following household utility maximization problem:
Max :U =E
t
T
X
t=0

t
u
t
[C
t
;L
t
;H
t
] (3.3)
s:t:V
T
= (
T
Y
t=0
(1 +r
t
)V
0
) +
T
X
t=0
(

Y
t=0
(1 +r

))(w
t
(T
t
L
t
) +Y
t
P
c
t
C
t
P
n
t
N
t
) (3.4)
H
t
=h(H
t1
;N
t
;I
t
;D
t
;
et
;
e
;
ht
;
h
) (3.5)
is the subjective discount factor re
ecting the extent to which household prefer future
utility to current one. P
c
t
andP
n
t
denote the prices of consumption goods and health inputs
respectively. T
t
is total time endowed to each household and w
t
is a wage rate. V
0
is a
household assets at time 0. Y
t
captures non-labor income at time t.
In order to solve this optimization problem, following Foster (1995), I assume that the
household utility function U is inter-temporally separable and each sub-utility functions,
u
t
is quasi-concave and increasing. The household is also assumed to borrow or lend freely
against its future in each time period t. Given these assumptions, the household utility
maximization problem (3.3) can be solved subject to a lifetime budget constraint (3.4)
and individual's dynamic health production function (3.5), and thus, the optimal dynamic
conditional health input demand function N

t
can be expressed as:
N

t
=f(H
t1
;I
t
;P
c
t
;P
n
t
;w
t
;D
t
;Y
t
;
et
;
e
;
ht
;
h
;E
t
(A
t+j
)) ; = 1;:::;T t (3.6)
88
and A
t
=I
t
;P
c
t
;P
n
t
;w
t
;D
t
;Y
t
;
et
;
e
;
ht
;
h
By replacing N
t
in the equation (3.5) by N

t
in equation (3.6), the dynamic condi-
tional health demand function follows as:
H

t
=h(H
t1
;I
t
;P
c
t
;P
n
t
;w
t
;D
t
;Y
t
;
et
;
e
;
ht
;
h
;E
t
(A
t+j
)) ; = 1;:::;T t (3.7)
and A =I
t
;P
c
t
;P
n
t
;w
t
;D
t
;Y
t
;
et
;
e
;
ht
;
h
This dynamic health demand function species that the current health status is a
function of one-period lagged health, H
t1
, resident community environments, I
t
, prices
of consumption goods P
c
t
and health inputs, P
n
t
, wage rates, w
t
, a set of time-varying
and time-invariant characteristics of the elderly and the household as well as household's
expectation at time t, E(A
t+j
). Each household forms expectations about the future
changes in prices, community environment, demographic characteristics, and other factors.
These expectations are used in the optimal choice of current health inputs.
3.4 Data and Variables
The data comes from the 1993, 1997, 2000 and 2007 waves of the Indonesian Family Life
Survey (IFLS). IFLS is a large-scale socio-economic survey conducted in Indonesia which
contains extensive information collected at the individual, household and community levels.
The survey includes not only indicators of economic, but also of non-economic well-being
such as consumption, expenditure, income, education assets, migration, fertility, use of
health care, health insurance, marriage, kinship among family members, and labor market
outcomes.
89
The IFLS is a continuing longitudinal survey. The rst wave, IFLS1, was conducted in
late 1993 and early 1994 and it represented 83% of the Indonesian population residing in 13
of the country's 27 provinces at the time.
5
Within each of the 13 provinces, enumeration
areas (EAs) were randomly selected for nal survey purpose. In IFLS1, 7,224 households
were interviewed and detailed individual level information was collected from over 22,000
respondents. IFLS2 sought to follow up the same sample and re-contacted 94.4% of
household interviewed in IFLS1. The re-contact rate of original IFLS1 households was
95.3 % in IFLS3 and 93.6% in IFLS4. For individuals, re-contact rates are also high. In
particular, of those individuals who were interviewed in IFLS1, 88% were followed up
in IFLS4. Among individuals surveyed in IFLS1, 2,743 had died by IFLS4. Among age
groups, the highest re-contact rates (over 90% in 2007, of original 1993 respondents) are
observed in those who were older than 40 years in 1993.
6
The attrition level is very low
in IFLS because IFLS has tracked individuals who had moved or split o from the origin
households. A potential worry associated with attrition is that there might be systematic
dierences between those who are re-interviewed and those who attrit. As Thomas et al.
(2012) show, tracking, particularly over long-distances, in IFLS has helped to mitigate
any selection bias.
7
5
These are four provinces on Sumatra (North Sumatra, West Sumatra, South Sumatra, and Lampung),
all ve of the Javanese provinces (Jakarta, West Java, Central Java, Yogyakarta, and East Java), and four
provinces covering the remaining major island groups (Bali, West Nusa Tenggara, South Kalimantan, and
South Sulawesi).
6
See Strauss et al.,(2009) and Thomas et al.,(2012), for more details
7
Thomas et al., (2012) nd that attrition is selected on not only age, location and gender but also
socio-economic characteristics in IFLS. They show that respondents who were found, but tracked over long
distances resemble more attriters who were never found, than they do respondents who were found but
never moved.
90
Sample
In order to focus on the elderly population, I restrict the sample to those respondents
who are 50 years or older in 2007 and hence, 36 or older in 1993. The sample sizes are
slightly dierent across health measures because some health measures are available for
all four waves, while others for only three, and because of missing data. Some of the
information comes from proxy interview, collected from one of the family members, if the
main respondents are not available. In this case, self-reported health such as measures
of physical functioning and ADLs or general health status can be collected but other
bio-markers such as hemoglobin or hypertension cannot. As a nal sample, a total of 2351
men and 2768 women are analyzed for the study of general health status dynamics.
Health Measures
Considering that health is a multidimensional composite, I analyze several health mea-
sures, both biomarkers and self-reports, known to be important for older adults. These
include the number of diculties with measures of physical functioning and activities of
daily living (ADLs), body mass index (BMI), self-assessed general health status (GHS),
blood hemoglobin (HB), hypertension, and lung capacity. I use these health measures
both as dependent variables and lagged independent variables.
ADLs are dened as routine activities for self-care in the individual's place of residence,
as well as outdoor environments. In IFLS, respondents were also asked to choose how easily
they could do activities related to physical functioning. Although more activities were
added in the 2007 wave, 9 activities that were originally asked in the 1993 and re-asked
in subsequent waves are used in the analysis. These are: (a)carry a heavy load (like a pail
of water) for 20 meters, (b)sweep the house
oor yard, (c)walk for 5 kilometers, (d)draw
a pail of water from a well, (e)bow, squat, kneel, (f)dress without help, (g)stand up from
sitting position in a chair without help, (h)go to the bathroom without help, (i)stand up
91
from sitting on the
oor without help. Respondents can select the degree of diculty to
be either \easily", \with diculty" or \unable to do it". To construct the variable for this
study, each answer is scored as 1 if respondents report that they can do any of those listed
activities only \with diculty" or \unable to do" it and 0, otherwise. The sum of these
scores is then used as the indicator of the number of diculties with measures of physical
functioning and ADLs.
BMI is calculated as weight (in kilograms) divided by height squared (in meters). Fol-
lowing World Health Organization standards (WHO), dummy variables are created for
being underweight if BMI is under 18.5, and for overweight if BMI is greater or equal to
25. Many papers have found that BMI has been rising for older people, especially women,
in Asia and Indonesia.
8
In all four waves of the IFLS, respondents were asked to evaluate their own health. The
question given to each respondent was \In general, how is your health?" with the following
options: \very healthy", \somewhat healthy", \somewhat unhealthy", and \unhealthy".
Those who answered \somewhat unhealthy", and \unhealthy" are coded as 1, as having
poor health, while the remaining are coded as 0, as in good health. There might be concerns
regarding how convincingly GHS works as a re
ection of a individual true level of health.
As a subjective measure, GHS may be in
uenced by factors such as the frequency of
doctor's visits or mood at the time. However, this measure has been widely used,
9
and
many studies have demonstrated that GHS can predict subsequent mortality very well in
surveys such as the HRS and ELSA.
10
8
See for example, Popkin, 1994, Monteiro et al., 2004, Strauss et al., 2004, and Witoelar et al., 2012
9
See Hurd et al. 1995, Adams et al. 2003, Contoyannis and Jones 2004
10
See Idler et al.(1997), Burstrom and Fredlund 2001, van Doorslaer and Gerdtham 2003
92
Hypertension, a chronic medical condition of having elevated blood pressure, can indi-
cate the risk of coronary heart disease, including stroke, heart attacks, and heart failure.
The blood pressure of each respondent was measured three times in the 2007 wave of IFLS
and once in the earlier waves, starting in wave 2.
11
For accuracy reasons, the average of
the last two measurements is used to construct the hypertension measure in 2007. The
dummy variable capturing the prevalence of hypertension is created following the standard
denition of the WHO; 1 for those whose systolic is greater than or equal to 140 or diastolic
is greater than or equal to 90, and 0 otherwise.
Blood hemoglobin is also one of the important health measures to be examined. It
measures anemia, but can also be raised by infections. In IFLS, hemoglobin levels were
examined, using the Hemocue meter in waves 2 through 4.
12
A dummy variable is created
to capture low hemoglobin level, as equal to 1 for those whose hemoglobin level is below
the WHO thresholds (for men: 13g/dL, for women: 12g/dL).
Finally, I examine the lung capacity of older adults. Using the Personal Best Peak
Flow Meters, it was measured three times in each wave, starting in wave 2, and the
average of three measurements is used in the analysis.
13
Health Transition Matrices
A very simple picture that shows the health dynamics is presented by the health transition
matrices in tables 3.1.1 - 3.1.5. These matrices show the degree to which previous
11
Blood pressure and pulse were measured with an Omron digital measuring device.
12
Homocue Method involves pricking a nger and collecting a drop of blood into a cuvette, then inserting
the cuvutte into the Hemocue measuring device.
13
A peak
ow meter is a device that measures how fast air comes out of the lungs when one exhales
forcefully.
93
health correlates with current health. Five health measures are studied; low hemoglobin,
hypertension, general health status, underweight and overweight.
Covariances and correlations between past and current health are computed as follows,
taking into account that these health measures are binary variables.
14
Cov (Y
t
;Y
t1
) = P (Y
t
= 1;Y
t1
= 1)P (Y
t
= 1)P (Y
t1
= 1)
=
n
11
n
11
+n
01
+n
10
+n
00

n
11
+n
01
n
11
+n
01
+n
10
+n
00
n
11
+n
10
n
11
+n
01
+n
10
+n
00
:
where n
ij
is the number of individuals whose Y
t1
= i and Y
t
= j: Among these, under-
weight and overweight show the highest correlations between outcomes in one wave and
in the prior wave. People who were underweight or overweight in the past mostly tend to
remain so. This is not very surprising since height is a stock measure of health and weight
has both stock and
ow dimensions. General health status has the lowest correlations and
especially so for the later waves. Given the nature of subjective measures, this measure
can be confounded with many other factors such as subjective mood at the time of survey
or their expectation of being healthy. Low hemoglobin has lower correlations across waves.
The common causes of low hemoglobin are nutritional deciency, especially iron, infections
from disease and other causes. Depending on changes in diet or whether a person is
recovered from certain disease or infection, this measure can vary over time. Hypertension
shows a strong correlation over waves.
The main goal of this chapter is to explain where this persistence comes from. The
strong correlations between lagged and current health measures can represent the impact
of lagged health or xed unobserved factors such as genetic endowments or childhood
health. Using rst-dierence two-step GMM, I attempt to disentangle these factors and
discuss the ndings in the result section.
14
Corr(Yt;Yt1)=Cov(Yt;Yt1)/Std.Dev(Yt)*Std.Dev(Yt1)
94
Table 3.1.1: Transition Matrix - Hypertension
2007
Men
2000
No Yes Total
No 970 400 1,370
(70.8%) (29.2%)
Yes 175 613 788
(22.21%) (77.79%)
Total 1,145 1,013 2,158
Corr(H
07
;H
00
) 0.4688
2000
Men
1997
No Yes Total
No 1,139 304 1,443
(78.93%) (21.07%)
Yes 231 484 715
(32.31%) (67.69%)
Total 1,370 788 2,158
Corr(H
00
;H
97
) 0.4558
2007
Women
2000
No Yes Total
No 1000 571 1,571
(63.65%) (36.35%)
Yes 151 931 1,082
(18.21%) (81.79%)
Total 1,197 1,456 2,653
Corr(H
07
;H
00
) 0.5197
2000
Women
1997
No Yes Total
No 1,291 349 1,640
(78.72%) (21.28%)
Yes 280 733 1,013
(27.64%) (72.36%)
Total 1,571 1,082 2,653
Corr(H
00
;H
97
) 0.5049
95
Table 3.1.2: Transition Matrix - Low Hemoglobin
2007
Men
2000
No Yes Total
No 1,217 311 1,528
(79.65%) (20.35%)
Yes 294 301 595
(49.41%) (50.59%)
Total 1,511 612 2,123
Corr(H
07
;H
00
) 0.2998
2000
Men
1997
No Yes Total
No 1,143 296 1,439
(79.43%) (20.57%)
Yes 385 299 684
(56.29%) (43.71%)
Total 1,528 595 2,123
Corr(H
00
;H
97
) 0.2408
2007
Women
2000
No Yes Total
No 1,171 374 1,545
(75.79%) (24.21%)
Yes 569 494 1,063
(53.53%) (46.47%)
Total 1,740 868 2,608
Corr(H
07
;H
00
) 0.2321
2000
Women
1997
No Yes Total
No 1,109 485 1,594
(69.57%) (30.43%)
Yes 436 578 1,014
(43.00%) (57.00%)
Total 1,545 1,063 2,608
Corr(H
00
;H
97
) 0.2636
96
Table 3.1.3: Transition Matrix - General Health Status (Poor Health = 1)
2007
Men
2000
No Yes Total
No 1,607 421 2,028
(79.24%) (20.76%)
Yes 173 150 323
(53.56%) (46.44%)
Total 1,780 571 2,351
Corr(H
07
;H
00
) 0.2061
2000
Men
1997
No Yes Total
No 1,842 208 2,050
(89.85%) (10.15%)
Yes 186 115 301
(61.79%) (38.21%)
Total 2,028 323 2,351
Corr(H
00
;H
97
) 0.2723
2007
Women
2000
No Yes Total
No 1,697 539 2,236
(75.89%) (24.11%)
Yes 302 230 532
(56.77%) (43.23%)
Total 1,999 769 2,768
Corr(H
07
;H
00
) 0.1683
2000
Women
1997
No Yes Total
No 1,929 380 2,309
(83.54%) (16.46%)
Yes 307 152 459
(66.88%) (33.12%)
Total 2,236 532 2,768
Corr(H
00
;H
97
) 0.1572
97
Table 3.1.4: Transition Matrix - Underweight (BMI < 18.5)
2007
Men
2000
No Yes Total
No 1,376 117 1,493
(92.16%) (7.84%)
Yes 104 224 328
(31.71%) (68.29%)
Total 1,480 341 1,821
Corr(H
07
;H
00
) 0.5955
2000
Men
1997
No Yes Total
No 1,423 95 1,518
(93.74%) (6.26%)
Yes 70 233 303
(23.1%) (76.9%)
Total 1,493 328 1,821
Corr(H
00
;H
97
) 0.6846
2007
Women
2000
No Yes Total
No 1,758 123 1,881
(93.46%) (6.54%)
Yes 120 290 410
(29.27%) (70.73%)
Total 1,878 413 2,291
Corr(H
07
;H
00
) 0.6401
2000
Women
1997
No Yes Total
No 1,809 110 1,919
(94.27%) (5.73%)
Yes 72 300 372
(19.35%) (80.65%)
Total 1,881 410 2,291
Corr(H
00
;H
97
) 0.7207
98
Table 3.1.5: Transition Matrix - Overweight (BMI 25)
2007
Men
2000
No Yes Total
No 1,470 116 1,586
(92.69%) (7.31%)
Yes 53 182 235
(22.55%) (77.45%)
Total 1,523 298 1,821
Corr(H
07
;H
00
) 0.6355
2000
Men
1997
No Yes Total
No 1,545 55 1,600
(96.56%) (3.44%)
Yes 41 180 221
(18.55%) (81.45%)
Total 1,586 235 1,821
Corr(H
00
;H
97
) 0.7598
2007
Women
2000
No Yes Total
No 1,477 201 1,678
(88.02%) (11.98%)
Yes 92 521 613
(15.01%) (84.99%)
Total 1,569 722 2,291
Corr(H
07
;H
00
) 0.6957
2000
Women
1997
No Yes Total
No 1,605 113 1,718
(93.42%) (6.58%)
Yes 73 500 573
(12.74%) (87.26%)
Total 1,678 613 2,291
Corr(H
00
;H
97
) 0.7893
99
Other Control Variables
A set of individual and household characteristics is controlled for in the analysis. The
set includes the age and the education level of the respondent, education interacted
with age, duration (the gap between two consecutive surveys), duration interacted with
respondent's age, and per capita household expenditure (PCE). In order to capture the
non-linear impact of age, the dummy variables are created for respondent's age with 5-year
intervals: 45-49, 50-54, 55-59, 60-64, 65 and older.
15
For the same reason, education
is also categorized as having at least some primary school, completed primary school,
completed junior high school, and completed high school or more with no schooling as
being the omitted group. The interaction term between respondent's age and the years
of education helps to understand whether the education has dierential impacts among
dierent age groups. For instance, in the case of GHS, a positive coecient of this term
would indicate that the higher the education, the lower the chance of having poor health
in subsequent periods among younger cohorts, possibly because the quality of education
has improved over time. A negative coecient would indicate that schooling makes more
dierence as people age.
IFLS is not elded annually and the periods between consecutive surveys are not
even. As noted earlier, the rst survey was conducted in 1993, and four years later, the
second survey was followed up. IFLS 3 and IFLS 4 were completed in 2000 and 2007
respectively. Following Hoddinot and Kinsey (2001), duration, the period length between
the two consecutive surveys, is included to control for uneven gaps between surveys.
16
In
the rst-dierence setting, the information on duration and the age of respondents would
15
Age 36-44 is the omitted group.
16
The duration is measured in months. For instance, if the rst survey was done in June 1993 and the
second survey was followed up in July 1997, then the duration would be 49 months.
100
be collinear and therefore, one-period lagged age is used instead in the estimation. The
regression also includes the interaction term between duration and respondent's lagged
age. This term captures age-dierential eects over time. In section 3.3, it is shown
that theoretically each household maximizes its utility assuming that they can freely
lend or borrow against future periods. However, the assumption of complete markets in
developing countries may not be realistic. Thus, I relax this assumption by including a
proxy for a household's constraint to access to credit, a one-period lagged log of household
per capita expenditure (PCE).
17 18
All control variables are treated as exogenous. The
main statistics are reported in tables 3.2.1 - 3.2.3.
Parental Characteristics
The respondents focused in this chapter are 50 years and older in 2007 which means
their parents would be at least 65 at the time. Several variables are available in IFLS
to measure the latest health of each respondent's biological parents and their schooling.
Parental schooling is treated as time-invariant since it is reasonable to assume that parents
have already completed education given their ages.
The sources of information are dierent for co-resident parents and for non-coresident
ones. For non-coresident parents, it was the adult child respondent who was interviewed
about the health status and the level of education of their biological parents. In IFLS4,
respondents were asked about the current health of their non-coresident mother and father
17
PCE is converted in real terms across both years and locations. i.e. it is converted into January 2007
Jakarta index.
18
Current PCE may be exceptionally high or low because of a sudden shock like good rainfall or an
accident to a worker preventing them from working. Current PCE and current health are correlated
because of the shock. Lagged PCE will then be correlated with the rst-dierenced shock, "t "t1. To
account for this, in some specications, log PCEt2 and log PCEt3 are used as instruments but with no
dierences in estimation results.
101
Table 3.2.1: Summary Statistics 1
MEN
1993 1997 2000 2007 Obs
Low Hemoglobin 0.32 0.28 0.29 2123
(0.467) (0.449) (0.453)
Hemoglobin level 13.65 13.73 13.77 2123
(1.781) (1.809) (1.807)
Hypertension 0.33 0.37 0.51 2158
(0.471) (0.482) (0.5)
Lung capacity 351.82 354.53 315.86 2092
(92.464) (99.798) (106.382)
Overweight 0.11 0.12 0.13 0.16 1821
(0.307) (0.327) (0.335) (0.37)
Underweight 0.16 0.17 0.18 0.19 1821
(0.367) (0.373) (0.384) (0.39)
BMI level 21.22 21.28 21.33 23.56 1821
(4.898) (3.094) (3.284) (58.844)
GHS (Poor health) 0.10 0.13 0.14 0.24 2351
(0.299) (0.334) (0.344) (0.429)
# of diculties of ADLs 0.22 0.40 0.43 1.06 2342
(0.945) (1.069) (1.08) (1.898)
Respondent's age 48.40 52.40 55.40 62.40 2351
(9.493) (9.493) (9.493) (9.493)
Log (per capita HHs expenditure) 12.72 12.73 12.84 2351
(0.748) (0.754) (0.707)
Source: IFLS 1993, 1997, 2000 and 2007
Standard deviations are reported in parenthesis.
102
Table 3.2.2: Summary Statistics 2
WOMEN
1993 1997 2000 2007 Obs
Low Hemoglobin 0.39 0.41 0.33 2608
(0.488) (0.491) (0.471)
Hemoglobin level 12.23 12.12 12.45 2608
(1.627) (1.509) (1.55)
Hypertension 0.38 0.41 0.61 2653
(0.486) (0.492) (0.488)
Lung capacity 241.32 244.27 212.60 2478
(64.496) (60.807) (68.674)
Overweight 0.20 0.25 0.27 0.32 2291
(0.4) (0.433) (0.443) (0.465)
Underweight 0.17 0.16 0.18 0.18 2291
(0.377) (0.369) (0.383) (0.384)
BMI level 21.93 22.45 22.58 27.41 2291
(3.753) (4.14) (4.415) (90.935)
GHS (Poor health) 0.13 0.17 0.19 0.28 2768
(0.337) (0.372) (0.394) (0.448)
# of diculties of ADLs 0.54 1.02 1.09 1.72 2760
(1.319) (1.522) (1.527) (2.135)
Respondent's age 48.31 52.31 55.31 62.31 2768
(9.413) (9.413) (9.413) (9.413)
Log (per capita HHs expenditure) 12.70 12.74 12.82 2768
(0.756) (0.792) (0.694)
Source: IFLS 1993, 1997, 2000 and 2007
Standard deviations are reported in parenthesis.
103
Table 3.2.3: Summary Statistics 3
Variables Description
MEN WOMEN
Mean Std. Dev Mean Std. Dev
Respondent's Edu At least some primary 0.30 (0.458) 0.30 (0.458)
Completed primary 0.29 (0.455) 0.18 (0.388)
Completed Jr. H 0.09 (0.291) 0.07 (0.247)
Completed High and more 0.16 (0.367) 0.08 (0.277)
Resp's yrs of Edu 5.58 (4.459) 3.56 (4.133)
Father's Education At least some primary school 0.09 (0.284) 0.07 (0.252)
Completed primary school 0.25 (0.43) 0.24 (0.428)
Completed junior high school and more 0.06 (0.229) 0.05 (0.22)
Mother's Education At least some primary school 0.07 (0.252) 0.05 (0.227)
Completed primary school 0.17 (0.374) 0.16 (0.365)
Completed junior high school and more 0.02 (0.14) 0.02 (0.143)
Father's Health Death =1 if dead in 2007 0.95 (0.221) 0.94 (0.233)
ADL problem = 1 if need help now or before death 0.23 (0.421) 0.23 (0.422)
GHS = 1 if unhealthy now or before death 0.50 (0.5) 0.47 (0.499)
Mother's Health Death =1 if dead in 2007 0.83 (0.377) 0.83 (0.376)
ADL problem = 1 if need help now or before death 0.25 (0.432) 0.25 (0.434)
GHS = 1 if unhealthy now or before death 0.47 (0.499) 0.44 (0.496)
Birth Cohorts born before 1932 0.12 (0.323) 0.11 (0.318)
born between 1932 and 1936 0.11 (0.314) 0.10 (0.302)
born between 1937 and 1941 0.15 (0.359) 0.16 (0.37)
born between 1942 and 1946 0.15 (0.355) 0.16 (0.366)
born between 1947 and 1951 0.22 (0.415) 0.20 (0.396)
Observations 2351 2768
Source: IFLS 1993, 1997, 2000 and 2007
104
if their parents were still alive at the time, or the latest health if they died before 2007.
19
For non-coresident parents, therefore, the health information from the 2007 wave is used
in the analysis. However, for parents who lived together with respondents at the time of
the survey, parents were directly interviewed. Hence, if parents were alive in 2007, the
information comes from IFLS4, but if they died between the surveys, a previous wave is
used.
20
More specically, dummy variables are created for being dead in 2007, having
diculties with ADLs, and poor general health status (GHS). The parents' death dummy
variables, one for each biological parent, are equal to 1 if the parent was dead at the time
of the survey in 2007. In the sample, only 6 % of fathers and 17% of mothers were still
alive at the time, so this dummy indicates a particularly healthy parent if it is 0.
21
The
GHS measures are constructed based on the response to the question \How is the health
status of your father/mother now/before his/her death". Similar to respondent's own GHS
variable, the GHS dummy variable for parents equals 1 if the respondents report the health
of their parents as \somewhat unhealthy" or \unhealthy" and 0 otherwise. For diculties
with ADLs, respondents choose \Yes" or \No" for the following question:\Now/before
death does/did your father/mother need help with basic personal needs like dressing, eating,
or bathing?", and a dummy variable is created as 1 for those who answered \Yes" and 0
19
Some may be concerned that health just prior to death may be worse and do not represent health
earlier in life. In my sample, distribution of GHS are actually dierent, worse for dead parents (a chi-square
statistic of dierences are 40.7 for fathers and 47.5 for mothers; both are signicant at under .01). To
address this problem, in the previous chapter, I compare the estimates from regressing respondent's own
health measures on parental characteristics with including the interactions between parents' death and
parents' GHS, in the same regression. These interactions are not jointly signicant, so I include only the
level health variables.
20
For example, co-resident parents, if they died in 1999 (between IFLS2 (1997) and IFLS3 (2000)), IFLS2
(1997) is used.
21
In the previous chapter on intergenerational transmission of health, I tried separate dummies for death
before age 60, death after age 60. These turned out not to be signicantly dierent from each other, so I
choose the dummies for death, as it is.
105
for \No". The levels of schooling of each parent are also constructed as dummy variables
for level completed: at least some primary school, completed primary school, and junior
high school and more with no schooling as being the omitted category.
Respondent's Birth Cohort
The respondent's birth cohort is constructed by using respondent's age in 2007.
22
The
dummy variables are created with 5-year intervals such as 55-59, 60-64, 65-69, 70-74, and
75 and older, to capture dierent experiences with respect to health, economic and social
environments.
3.5 Empirical Specication
The main goal of this chapter is to characterize the determinants of chronic diseases and
explain their persistence. The empirical framework of the dynamic conditional health
demand function can be written as follows:
H

it
=
H
H

it1
+
S
X
j=1

j
X
jit
+
R
X
j=1

j
Z
ji
+"
i
+"
h
+"
c
+"
it
(3.8)
where H

it
and H

it1
are the true health status of older adult i as measured by ADLs,
BMI, GHS, hemoglobin, hypertension, and lung capacity at time t and t-1, respectively.
However, we observe H
it
=H

it
+
i
instead, where
i
is measurement error. I assume for
22
Using respondent's age to construct the birth cohort dummy can lead to possible collinearity. However,
I use respondent's one-period lagged age as a covariate on RHS. Moreover, lagged age is dropped after
rst-dierencing, since it can be still collinear with duration and duration is measured in months, not
years.
106
now that measurement error is time-invariant.
23
After substitutingH
it
into equation (3.8),
the health demand function becomes:
H
it
=
H
H
it1
+
S
X
j=1

j
X
jit
+
R
X
j=1

j
Z
ji
+"
i
+"
h
+"
c
+"
it
+ (1)
i
(3.9)
where Xs are time-varying regressors such as respondent's age and household incomes.
Zs are time-invariant regressors which include respondent's gender and education.
24
Dif-
ferent sources of unobservables are captured by "
S
. First, "
i
captures individual specic
unobserved factors that do not change over time, such as genetic endowments and child-
hood health outcomes. Unfortunately in this chapter, it is not possible to distinguish
genetics from environment in childhood. Second, "
h
, represents time-invariant household
specic unobservables, like spousal preference towards respondent's health. "
c
captures
time-invariant community specic unobservables such as time-invariant community health
factors. Lastly, the household's future expectation, E
t
(A
t+j
) in equation (3.7) enters the
empirical specication through "
it
.
25
"
it
captures the time-varying, individual-specic
unobservables, such as individual health or income shocks, or other demographic char-
acteristics that aect current health status but are not observed by econometricians at
time period t.
It is dicult to expect a zero correlation between the error term and the explanatory
variables when it comes to the inclusion of the lagged dependent variable,H
t1
on the right
hand side.
26
One-period lagged health status is likely to be correlated with time-invariant
23
This assumption will be relaxed later and time-varying measurement error will be considered as well.
24
Unlike other papers, the education level of the respondents is treated as time-invariant since the sample
includes only elderly who are 50 years and older.
25
If expectations do not change, they will be part of "i and swept away after taking rst-dierence.
26
See Deaton(1997), Blundell and Bond(1998) and Wooldridge(2002)
107
unobservables such as "
i
;"
h
, and "
c
. For example, those born with weaker health endow-
ments are less likely to maintain good health which may lead to having lower education
and income, which further leads to poorer health outcomes in subsequent periods. In this
case, the estimates without considering the endogeneity issue, would be upward biased. The
time-invariant household-specic unobservables and community-specic unobservables also
cause correlation with the lagged health measures, creating bias in the estimated coe-
cients. First-dierences can solve this endogeneity problem as it removes all time-invariant
unobserved heterogeneity as well as any time-invariant measurement error.
H
it
=
H
H
it1
+
S
X
j=1

j
X
jit
+ "
it
(3.10)
However, even after rst dierencing, OLS estimates may still be biased because of the
correlation between H
it1
and "
it
. "
t
denotes time-varying unobservables such as time-
varying health or income shocks that not picked up by PCE. Health at time t-1 is aected
by these shocks at t-1 and hence, the endogeniety problem still remains. In this chap-
ter, equation (3.10) is estimated by general method of moments (GMM), where the rst-
dierenced health measure is instrumented with the two-period (and the three-period)
lagged health measure in accordance with Arellano and Bond (1991).
27
In addition, I use
external instruments such as parents' health and education and respondent's birth cohort.
These are dierenced out of equation (3.10) but may still be correlated with lagged changes
in health when further lags in health are not used as lagged right-hand side variables, as
they would not be in a rst-stage regression for H
it1
.
28
27
Two-period and three-period lagged health measures are used as internal IVs for ADLs, BMI and GHS,
which are available in all four waves. For the other health measures, just two-period lagged health measure
is used as an internal IV because they were collected only since IFLS2 in 1997, for three waves.
28
Place of birth dummies are also used as instruments in a separate specication along with the birth
cohort. They are constructed at the district level. The results are similar to those with using only the birth
108
A valid instrument for the rst-dierenced model should satisfy the following conditions:
(a) it should be correlated with changes in the endogenous regressor and (b) uncorrelated
with the dierenced error term in the second-stage regression.
The lagged variables of order two or higher, H
it2
;H
it3
are valid instruments in a
rst-dierenced dynamic panel model since they are correlated with H
it1
H
it2
but not
with "
it
"
it1
. This does maintain the assumption that "
it
are not serially correlated.
29
Parental characteristics are also strongly associated with the rst-dierenced lagged health
measures. For example, chapter 2 shows that having a mother with poor health status
is correlated with changes in the number of ADL problems for elderly men. For women,
having a dead mother is positively correlated with moving from no hypertension to having
hypertension. If a father has completed primary school, the adult child is less likely to
develop hypertension and less likely to have an increased the number of ADL diculties
over time.
30
Respondent's birth cohort is also used as an IV. Indonesia has developed
rapidly during the past few decades and therefore, it is more likely that younger respondents
had more opportunities for having higher education and had better health infrastructure.
3.6 Results
3.6.1 The Eects of Lagged Health Measures
The results from estimating the dynamic conditional health demand function using GMM
estimation are presented in tables 3.3 - 3.8. Results are reported separately for elderly
cohort as IV. However, they are not reported in the paper due to the collinearity concern arising from the
fact that many people still live in the same place as they were born.
29
This assumption can be tested. In the next section, the Hansen J statistic, a test of over-identifying
restrictions will be reported.
30
See Kim, Strauss, Witoelar and Sikoki, (2011) for more details.
109
men and women, since the health transitions could be dierent, depending on gender. In
all specications, an interaction term between place of residence and survey year dummy
variables are included to control for time-varying characteristics such as changes in prices
and environment at the district level. Some may worry that the place of residence dummy
variables after rst-dierencing might capture migrations, and this might be endogenously
determined. For instance, if one decided to move closer to the hospital after getting sick, the
previous unobserved health shock could be correlated with the current place of residence.
Hence, I use one-period lagged level dummy variables for the place of residence instead.
31
The rst two columns of each table report the estimates from a standard OLS regression,
without rst-dierencing. In the case of low hemoglobin level, the OLS estimates,
H
are
0.230 for men and 0.227 for women; for hypertension, 0.457 for men and 0.440 for women;
and 0.576 for men and 0.416 for women for lung capacity. The results from the OLS
regression show strong path dependence in several health outcomes. This is consistent
with the ndings of the transition matrices presented in tables 3.1.1 - 3.1.5. People seem
to maintain their health; those who were healthy tend to remain healthy while people with
health problems in the past tend to continue to carry them in the subsequent period as
well. As discussed earlier, the OLS estimates before taking rst dierence do not address
any endogeneity problems. However, it can still provide a general picture on the extent of
the bias that can arise from time-invariant and time-varying omitted variables.
Taking rst-dierences partially solves the problem by removing the omitted variables
which are invariant over time. Running an OLS regression after rst-dierencing (FD-
OLS), the coecients are estimated at -0.432 for men and -0.492 for women for low
hemoglobin level, -0.434 for men and -0.462 for women for hypertension, and -0.389 for
men and -0.357 for women for lung capacity. The negative coecients on lagged health re-
31
The equation is written as Hit =H Hit1 +
P
S
j=1
j Xjit +dist:it1 Surveyyear + "it.
110
Table 3.3.1: Dynamic Health Demand Function Estimation - Low Hemoglobin
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged Low HB 0.230*** 0.227*** -0.432*** -0.492*** 0.0289 0.0696**
(0.0164) (0.0139) (0.0196) (0.0163) (0.0337) (0.0339)
45-50 (lagged age) 0.00787 -0.0246
(0.0197) (0.0200)
50-55 0.0588** -0.00203
(0.0253) (0.0246)
55-60 0.0834*** 0.0102
(0.0314) (0.0294)
60-65 0.119*** 0.00868
(0.0389) (0.0374)
65- 0.170*** 0.00894
(0.0517) (0.0494)
At least some primary -0.0174 0.0221
(0.0274) (0.0243)
Completed primary -0.00489 0.00108
(0.0383) (0.0432)
Completed Jr. High 0.00303 0.0112
(0.0525) (0.0616)
Completed High & more -0.0277 0.0157
(0.0678) (0.0820)
Yrs of edu * lagged age -0.0000899 -0.0000713 0.000200 0.000579 0.00000556 0.000935
(0.0000905) (0.000119) (0.000858) (0.000967) (0.000937) (0.00111)
Duration 0.00475 0.00677** -0.00356 0.00505 -0.00371 0.00868*
(0.00310) (0.00339) (0.00430) (0.00423) (0.00454) (0.00518)
Duration * lagged age 0.0000240 0.0000718*** 0.0000430* 0.000106*** 0.0000368 0.0000754***
(0.0000252) (0.0000246) (0.0000221) (0.0000217) (0.0000239) (0.0000258)
Lagged log (PCE) -0.000947 -0.0166* 0.000101 -0.000718 -0.00148 -0.0243
(0.00984) (0.00927) (0.0154) (0.0136) (0.0168) (0.0161)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 2123 2608 2123 2608 2123 2608
F stat on excluded IVs 57.338 56.965
Hansen J statistics 27.560 16.490
(p-value) (0.2791) (0.8995)
Source: IFLS 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
low HB:1 for those whose hemoglobin level is below the threshold (for men: 13g/dL, for women: 12g/dL)
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, low HB
t2
, parents' health and edu, and birth cohort are used as IVs.
111
Table 3.3.2: Dynamic Health Demand Function Estimation - Hemoglobin Level
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged HB Level 0.380*** 0.337*** -0.452*** -0.448*** 0.166*** 0.157***
(0.0180) (0.0163) (0.0255) (0.0227) (0.0537) (0.0438)
45-50 (lagged age) -0.00675 0.0833
(0.0801) (0.0638)
50-55 -0.197** 0.0241
(0.0906) (0.0740)
55-60 -0.266** -0.0104
(0.116) (0.0901)
60-65 -0.417*** 0.145
(0.139) (0.114)
65- -0.542*** 0.0976
(0.187) (0.153)
At least some primary -0.00203 -0.0186
(0.100) (0.0747)
Completed primary -0.0410 0.0486
(0.147) (0.134)
Completed Jr. High -0.0823 -0.0711
(0.211) (0.187)
Completed High & more 0.0376 -0.0809
(0.260) (0.249)
Yrs of edu * lagged age 0.000257 0.000108 0.000804 -0.00415 0.00627* -0.00527
(0.000352) (0.000362) (0.00305) (0.00306) (0.00358) (0.00336)
Duration 0.0606*** 0.0575*** 0.00378 0.00908 -0.00337 0.00298
(0.0187) (0.0187) (0.0158) (0.0148) (0.0175) (0.0144)
Duration * lagged age -0.000102 -0.000268*** -0.0000922 -0.000375*** 0.0000556 -0.000321***
(0.0000910) (0.0000788) (0.0000813) (0.0000688) (0.0000899) (0.0000791)
Lagged log (PCE) 0.0793* 0.0973*** -0.0484 0.0169 -0.0187 0.0427
(0.0420) (0.0357) (0.0569) (0.0419) (0.0618) (0.0474)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 2123 2608 2123 2608 2123 2608
F stat on excluded IVs 31.386 50.241
Hansen J statistics 36.456 22.378
(p-value) (0.0495) (0.6138)
Source: IFLS 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, HB
t2
, parents' health and edu, and birth cohort are used as IVs. 112
Table 3.4: Dynamic Health Demand Function Estimation - Hypertension
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged Hypertension 0.457*** 0.440*** -0.434*** -0.462*** 0.0935** 0.184***
(0.0146) (0.0128) (0.0212) (0.0193) (0.0451) (0.0453)
45-50 (lagged age) 0.0395* 0.0763***
(0.0206) (0.0186)
50-55 0.0700*** 0.106***
(0.0252) (0.0227)
55-60 0.0897*** 0.129***
(0.0307) (0.0260)
60-65 0.131*** 0.192***
(0.0382) (0.0339)
65- 0.114** 0.251***
(0.0494) (0.0438)
At least some primary -0.0101 0.0265
(0.0274) (0.0220)
Completed primary 0.0139 0.0709*
(0.0405) (0.0385)
Completed Jr. High 0.0494 0.0751
(0.0569) (0.0563)
Completed High & more 0.0943 0.118
(0.0740) (0.0736)
Yrs of edu * lagged age -0.000101 -0.000219** -0.000335 -0.00137 -0.00112 -0.00166
(0.0000996) (0.000105) (0.000862) (0.000886) (0.000964) (0.00102)
Duration 0.00589* 0.00161 0.00837* -0.00514 0.00689 -0.00370
(0.00315) (0.00519) (0.00489) (0.00434) (0.00567) (0.00488)
Duration * lagged age 0.0000214 -0.0000297 0.0000323 -0.00000338 0.00000578 -0.0000137
(0.0000248) (0.0000219) (0.0000215) (0.0000194) (0.0000236) (0.0000223)
Lagged log (PCE) 0.00662 0.0146 0.0195 -0.00241 0.0107 -0.00217
(0.00974) (0.00901) (0.0148) (0.0128) (0.0166) (0.0160)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 2158 2653 2158 2653 2158 2653
F stat on excluded IVs 30.257 33.896
Hansen J statistics 27.901 25.551
(p-value) (0.2642) (0.4319)
Source: IFLS 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
Hypertension: 1 for those whose systolic 140 or diastolic 90
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, hypertension
t2
, parents' health and edu, and birth cohort are used as IVs.
113
Table 3.5: Dynamic Health Demand Function Estimation - Lung Capacity
OLS FD-OLS FD-GMM A
Men Women Men Women Men Women
Lagged Lung Capacity 0.576*** 0.416*** -0.389*** -0.357*** 0.117** 0.0126
(0.0157) (0.0148) (0.0217) (0.0203) (0.0557) (0.0307)
45-50 (lagged age) -3.152 -4.967**
(3.731) (2.312)
50-55 -9.439** -6.551**
(4.582) (2.762)
55-60 -22.53*** -15.77***
(5.591) (3.262)
60-65 -32.17*** -10.62***
(6.709) (3.983)
65- -31.92*** -13.86***
(8.930) (5.298)
At least some primary 12.76*** 6.454**
(4.780) (2.647)
Completed primary 20.20*** 10.72**
(7.199) (4.555)
Completed Jr. High 29.26*** 15.86**
(10.15) (6.578)
Completed High & more 49.72*** 22.10***
(13.24) (8.521)
Yrs of edu * lagged age -0.0296* 0.000797 0.0480 0.160 0.0316 0.0531
(0.0178) (0.0124) (0.149) (0.102) (0.165) (0.101)
Duration 3.377*** 1.878*** 3.996*** 0.991** 3.762*** 0.968**
(0.497) (0.403) (0.787) (0.418) (0.837) (0.428)
Duration * lagged age -0.0154*** -0.0203*** -0.0236*** -0.0185*** -0.0151*** -0.0174***
(0.00430) (0.00268) (0.00359) (0.00234) (0.00394) (0.00237)
Lagged log (PCE) 4.069** 2.256** 1.683 -0.818 3.926 -0.467
(1.620) (1.078) (2.273) (1.435) (2.462) (1.490)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 2092 2478 2092 2478 2092 2478
F stat on excluded IVs 18.264 49.918
Hansen J statistics 16.659 17.274
(p-value) (0.8628) (0.8716)
Source: IFLS 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, lung capacity
t2
, parents' health and edu, and birth cohort are used as IVs. 114
Table 3.6.1: Dynamic Health Demand Function Estimation - Overweight
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged Overweight 0.723*** 0.731*** -0.446*** -0.353*** 0.0553 0.277***
(0.0167) (0.0107) (0.0347) (0.0228) (0.0950) (0.0653)
45-50 (lagged age) 0.00757 -0.0172
(0.0107) (0.0122)
50-55 0.0127 -0.0204
(0.0128) (0.0138)
55-60 -0.00143 -0.0383**
(0.0155) (0.0170)
60-65 0.0116 -0.00970
(0.0190) (0.0207)
65- 0.0396* -0.0276
(0.0239) (0.0288)
At least some primary 0.00108 0.0111
(0.0119) (0.0130)
Completed primary 0.0165 0.0250
(0.0206) (0.0237)
Completed Jr. High 0.0239 -0.00560
(0.0318) (0.0356)
Completed High & more 0.0511 0.0172
(0.0400) (0.0449)
Yrs of edu * lagged age 0.0000185 0.0000633 0.000504 0.000172 0.000346 -0.000231
(0.0000583) (0.0000659) (0.000329) (0.000381) (0.000285) (0.000390)
Duration 0.000752 -0.00569 0.00292 -0.00442 0.00298 -0.00306
(0.00406) (0.00413) (0.00321) (0.00321) (0.00341) (0.00543)
Duration * lagged age -0.0000375*** -0.0000352** -0.0000298** -0.0000390*** -0.0000229 -0.0000297*
(0.0000141) (0.0000158) (0.0000125) (0.0000139) (0.0000145) (0.0000154)
Lagged log (PCE) 0.0100** 0.0138** 0.00348 0.0108* 0.00439 0.00301
(0.00497) (0.00561) (0.00573) (0.00602) (0.00746) (0.00729)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 1821 2291 1821 2291 1821 2291
F stat on excluded IVs 20.479 17.918
Hansen J statistics 75.11 81.85
(p-value) (0.347) (0.275)
Source: IFLS 1993, 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
over BMI = 1 for those whose BMI 25
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, OverBMI
t2
;
t3
, parents' health and edu, and birth cohort are used as IVs.
115
Table 3.6.2: Dynamic Health Demand Function Estimation - Underweight
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged Underweight 0.643*** 0.651*** -0.418*** -0.420*** 0.190*** 0.145**
(0.0157) (0.0139) (0.0271) (0.0247) (0.0713) (0.0603)
45-50 (lagged age) 0.0175 0.0210**
(0.0109) (0.00978)
50-55 0.0245 0.0382***
(0.0154) (0.0134)
55-60 0.0379* 0.0489***
(0.0200) (0.0181)
60-65 0.0351 0.0867***
(0.0258) (0.0237)
65- 0.0514 0.0715**
(0.0337) (0.0304)
At least some primary 0.00424 -0.00631
(0.0172) (0.0127)
Completed primary -0.00163 -0.0154
(0.0226) (0.0208)
Completed Jr. High 0.00587 -0.00341
(0.0295) (0.0279)
Completed High & more -0.0104 -0.0100
(0.0383) (0.0378)
Yrs of edu * lagged age -0.0000596 -0.0000525 -0.000906*** -0.000624** -0.000303 -0.000538*
(0.0000534) (0.0000578) (0.000326) (0.000286) (0.000284) (0.000277)
Duration 0.00422 -0.00212 0.000284 0.00267 0.00270 0.000620
(0.00410) (0.00348) (0.00329) (0.00260) (0.00732) (0.000946)
Duration * lagged age 0.0000202 -0.00000310 -0.00000916 -0.0000113 -0.00000229 -0.00000947
(0.0000186) (0.0000171) (0.0000160) (0.0000144) (0.0000163) (0.0000119)
Lagged log (PCE) -0.0135** -0.00540 -0.00314 0.0121** -0.000159 0.00711
(0.00580) (0.00534) (0.00727) (0.00567) (0.00835) (0.00891)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 1821 2291 1821 2291 1821 2291
F stat on excluded IVs 16.669 22.166
Hansen J statistics 53.51 73.48
(p-value) (0.940) (0.528)
Source: IFLS 1993, 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
under BMI = 1 for those whose BMI < 18.5
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, UnderBMI
t2
;
t3
, parents' health and edu, and birth cohort are used as IVs.
116
Table 3.6.3: Dynamic Health Demand Function Estimation - BMI Level
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged BMI level 0.744*** 0.553*** -0.0147 -0.436** -0.0376 0.0337
(0.284) (0.191) (0.0574) (0.222) (0.128) (0.250)
45-50 (lagged age) -2.351 -0.0900
(2.206) (0.964)
50-55 -6.015 0.615
(5.704) (2.346)
55-60 -10.62 -0.0732
(9.458) (2.163)
60-65 -14.17 -3.536
(13.07) (2.398)
65- -17.24 -0.224
(15.72) (2.224)
At least some primary -2.083 0.436
(1.305) (0.667)
Completed primary -2.438 0.703
(1.695) (3.294)
Completed Jr. High -5.110 -3.194
(4.058) (3.739)
Completed High & more -6.788 -3.668
(6.170) (4.776)
Yrs of edu * lagged age 0.0118 0.00180 0.0277 -0.0679 0.00952* -0.00100
(0.00997) (0.00699) (0.0245) (0.0446) (0.00545) (0.0152)
Duration -0.440 -0.428 -0.0490 -0.372 -0.0592 -0.240
(0.294) (0.281) (0.109) (0.531) (0.165) (0.443)
Duration * lagged age 0.0120 0.000498 0.00774 0.00118 0.000466 0.000248
(0.0116) (0.00169) (0.00738) (0.00257) (0.000459) (0.00138)
Lagged log (PCE) 0.266 0.171 0.0684 -0.287 -0.000799 0.185
(0.237) (0.495) (0.287) (1.087) (0.160) (0.381)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 1821 2291 1821 2291 1821 2291
F stat on excluded IVs 4.099 5.021
Hansen J statistics 32.28 37.17
(p-value) (0.999) (0.999)
Source: IFLS 1993, 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, BMI
t2
;
t3
, parents' health and edu, and birth cohort are used as IVs. 117
Table 3.7: Dynamic Health Demand Function Estimation - GHS (Poor Health)
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged GHS 0.232*** 0.155*** -0.502*** -0.533*** 0.0116 -0.0380
(0.0173) (0.0140) (0.0191) (0.0152) (0.071) (0.0745)
45-50 (lagged age) -0.0141 -0.0221*
(0.0126) (0.0134)
50-55 -0.00771 0.0162
(0.0165) (0.0167)
55-60 -0.000671 0.0264
(0.0213) (0.0212)
60-65 0.0182 0.0525*
(0.0270) (0.0271)
65- -0.00682 0.0799**
(0.0359) (0.0362)
At least some primary -0.000229 0.0431***
(0.0172) (0.0160)
Completed primary 0.0266 0.0546**
(0.0249) (0.0276)
Completed Jr. High 0.0193 0.0502
(0.0336) (0.0402)
Completed High & more 0.0547 0.0470
(0.0443) (0.0534)
Yrs of edu * lagged age -0.000155** -0.000162** 0.0000398 -0.000469 -0.000438 -0.00255
(0.0000609) (0.0000791) (0.000387) (0.000441) (0.0519) (0.00208)
Duration -0.0110** -0.00499 -0.00547 -0.00528 -0.0127 -0.0150
(0.00445) (0.00465) (0.00379) (0.00409) (0.237) (0.0178)
Duration * lagged age 0.0000955*** 0.0000255 0.0000844*** 0.0000187 0.0000339 -0.0000515
(0.0000198) (0.0000195) (0.0000185) (0.0000189) (0.000507) (0.0000567)
Lagged log (PCE) 0.000418 0.0108 0.00283 0.0119 0.0110 0.0271
(0.00651) (0.00664) (0.00851) (0.00859) (1.116) (0.0448)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 2351 2768 2351 2768 2351 2768
F stat on excluded IVs 48.323 68.809
Hansen J statistics 30.41 18.88
(p-value) (0.644) (0.994)
Source: IFLS 1993, 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
GHS (Poor Health) = 1 for those who evaluated own health as somewhat unhealthy, unhealthy
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, GHS
t2
;
t3
, parents' health and edu, and birth cohort are used as IVs.
118
Table 3.8: Dynamic Health Demand Function Estimation - Number of Diculties with ADLs
OLS FD-OLS FD-GMM
Men Women Men Women Men Women
Lagged ADLs 0.352*** 0.276*** -0.382*** -0.457*** 0.0641 -0.0304
(0.0312) (0.0184) (0.0336) (0.0214) (0.0403) (0.0288)
45-50 (lagged age) -0.147*** -0.0872*
(0.0373) (0.0450)
50-55 -0.261*** -0.00526
(0.0602) (0.0645)
55-60 -0.225*** 0.118
(0.0855) (0.0917)
60-65 -0.141 0.341***
(0.117) (0.121)
65- -0.158 0.688***
(0.146) (0.163)
At least some primary -0.0535 -0.0108
(0.0668) (0.0653)
Completed primary 0.106 0.000175
(0.0983) (0.114)
Completed Jr. High 0.117 0.0612
(0.126) (0.162)
Completed High & more 0.272 -0.128
(0.170) (0.216)
Yrs of edu * lagged age -0.000665*** 0.0000521 -0.00335** -0.00252 -0.000219 0.0000617
(0.000249) (0.000341) (0.00134) (0.00157) (0.00112) (0.00154)
Duration -0.0561*** -0.120*** -0.0564*** -0.0729*** -0.0524*** -0.0911***
(0.0181) (0.0187) (0.0151) (0.0143) (0.0151) (0.0209)
Duration * lagged age 0.000839*** 0.000686*** 0.000755*** 0.000547*** 0.000721*** 0.000630***
(0.0000914) (0.0000907) (0.0000764) (0.0000769) (0.0000782) (0.0000884)
Lagged log (PCE) 0.0141 0.0337 0.0125 0.00305 0.0248 0.0133
(0.0227) (0.0247) (0.0290) (0.0310) (0.0332) (0.0358)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 2342 2760 2342 2760 2342 2760
F stat on excluded IVs 42.451 57.613
Hansen J statistics 86.55 121.54
(p-value) (0.263) (0.003)
Source: IFLS 1993, 1997, 2000 and 2007 *** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors adjusted for clustering at the community level are reported in parenthesis.
Dummy variables are included to capture missing values for parent's education, GHS, ADL and household's PCE.
In FD-GMM, ADL
t2
;
t3
, parents' health and edu, and birth cohort are used as IVs. 119
present regression to the mean; those who had more serious health problems tend to recover,
while those with less severe problems in the past tend to develop more problems in the sub-
sequent period. Compared to the standard OLS estimates, the magnitude of the coecients
becomes much lower after rst-dierencing. However, rst-dierencing only corrects for
time-invariant unobservables, while bias could still remain because of time-varying omitted
variables. In order to deal with both problems at once, I adopt the rst-dierence two-step
generalized method of moments (FD-GMM).
32
The rst-dierence GMM specication (FD-
GMM) uses several instruments to identify the rst-dierenced one-period lagged health
status. These instruments are the two-period and three-period lagged health measures,
the parents' health and education, and respondent's birth cohort.
33
The columns 5-6 of
each table report the results from using the entire set of instruments. For low hemoglobin,
using the two-step GMM, the coecient on one-period lagged health,
H
, is estimated at
0.0289 for elderly men and 0.0696 for elderly women.
34
This implies that women with low
hemoglobin are only about 7 percentage points more likely than healthy women to carry
the problem in the subsequent period. Similarly, in the case of hypertension, the coecient
estimates on lagged health are 0.0935 for men and 0.184 for women. This suggests that
compared to people who did not have hypertension in the past, men with hypertension are
about 9 percentage points more likely to have hypertension in the the subsequent period
and for women, it is 18 percentage points more likely. For lung capacity, the estimates are
0.117 for men and 0.0126 for women. The magnitude of the coecients on lagged health
32
Two-step GMM uses the optimal weighting matrix and hence, it is more ecient than 2SLS. Unlike
2SLS, two-step GMM accounts for the case with error terms being heteroskedastic.
33
For hemoglobin, hypertension and lung capacity, only two-period lagged health measures can be used
as internal IVs. Three-period lagged health measures are added as internal IVs for ADLs, BMI and GHS.
34
See column 5-6, table 3.3.1
120
measures indicates the extent to which previous health has an eect on current health, con-
ditioned on xed eects. A coecient of one on one-period lagged health measure implies
the current health is determined completely by previous health. A coecient equal to zero
would indicate that past health does not have any eects on subsequent health and hence,
suggest that older adults can completely overcome previous health conditions. Given that
the coecients on lagged health status measured by hemoglobin, hypertension, and lung
capacity are relatively close to zero, previous health does not matter much in determining
current health. Two-step GMM results show similar patterns for other health measures
such as ADLs, BMI and GHS. This implies that time-invariant unobservables that were
swept away by rst-dierencing play very important roles in determining current health.
The OLS estimates on past health were upward biased because past health conditions
are most likely to be strongly correlated with time-invariant xed eects such as genetic
endowments or childhood health. Netting these in
uences out, little or no correlation is
left with lagged health.
In order to be valid, IVs should be not only strongly correlated with endogenous regres-
sors but also uncorrelated with the error term in the second-stage regression. I cannot use
the specication test for lack of serial correlation in the rst-dierence residual proposed by
Arellano-Bond (1991) because it needs at least ve survey rounds. As an alternative test,
the Hansen J statistic, a test for over-identifying restrictions, is used. This is a joint test
to check both serial correlation of the error term and the correlations between instruments
and error-terms. For low hemoglobin, the p-value of J-statistics are reported as 0.2791 for
men and 0.8995 for women. In the case of hypertension, it is 0.2642 for men and 0.4319 for
women. I found the similar results for other health measures as well. The results suggest
that the null hypothesis that all instruments are orthogonal to the error term cannot be
rejected and that there is no serial correlations between error-terms.
121
As another evidence of instrument validity, F-statistics on excluded IVs are reported
for the rst-stage regressions of changes in health measure in tables 3.3-3.8. A valid instru-
ment (IV) should be correlated with the endogenous regressor. However, if the correlation
between IVs and the endogenous variable is too weak, the results from using IVs are likely
to be inconsistent and suer from larger bias than OLS estimates.
35
Following Staiger and
Stock (1997), F-statistics from the regressions of endogenous variable on excluded IVs are
examined as a test for the relevance of IVs. In the case of low hemoglobin, it is 57.338
for men and 56.965 for women. For hypertension, F-statistics are 27.901 and 25.551 for
men and women respectively.
36
The set of IVs used in this chapter are valid and strongly
correlated with the endogenous regressor for all health measures since the F-statistics are
larger than 10, the rule of thumb values suggested by Staiger and Stock (1997).
3.6.2 The Eects of Other Co-variates
Duration is included on the right-hand side to control for the uneven gaps between two
consecutive surveys in IFLS. In the case of low hemoglobin, the OLS estimates are 0.00475
for elderly men and 0.00677 for elderly women. This implies that for every additional
month between surveys, there is a 0.6 percent increase in the chance that elderly women
will have lower hemoglobin. However, once the unobservables are taken into account, the
duration is no longer signicant in FD-GMM specication. On the other hand, duration
interacted with the age of the older adult, capturing age-dierential eects over time,
remains signicant in FD-GMM specication. For the study of low hemoglobin, a positive
coecient indicates that the longer the duration is, the higher the chance of having low
hemoglobin is in the subsequent period among older cohorts.
35
See Murray (2006) for more details
36
See tables B.3.1 and B.3.2 in appendix.
122
Per capita household expenditure (PCE) is controlled for as a socio-economic status
(SES) characteristic.
37
The OLS estimate on one-period lagged log(PCE) is 0.0138 for
elderly women in the case of being overweight. This captures the positive impact of
resources on current health as measured by BMI. The magnitude of the coecient becomes
not signicantly dierent from zero, once xed eect and the in
uence of past health are
conditioned on, which demonstrates that there is an upward bias in the OLS estimates on
log(PCE) due to the possible correlation with the time-invariant unobserved heterogene-
ity.
38
3.6.3 Interactions of Lagged Health with Age and SES
The extent to which the past health has in
uence on current health can dier across SES
or demographic groups. For example, past good health may matter less as people age.
Past bad health may matter less for better educated people if they have better knowledge
of healthy diets or less risky lifestyles and visit doctors more often when they notice any
subtle changes in their health. In addition, people with higher income are generally more
likely to utilize the health care system than those with lower incomes. In this case, their
past poor health may matter less. To get at these potential dierences, the sample is
disaggregated by age, PCE, and years of education to see if the impact of previous health
diers across groups.
Tables 3.9.1 - 3.9.3 report the results from the estimation of disaggregated groups. With
the exception of hemoglobin level, the degree of path dependence turns out to be much
stronger for those who are less educated (dened as less than completed Jr. high school).
37
The reason that I use PCE instead of household income is the potential worry about systematic measure-
ment errors in income. Bound and Krueger (1991) show that, people in the United States with low-income
tend to over-report their income in an eort to hide their situation.
38
The correlation between household specic unobservables and log(PCE) can be one of many possible
examples. For instance, the families who prefer high calorie foods may have higher household expenditure.
123
Table 3.9.1: Disaggregated Groups 1
Hemoglobin Level Low Hemoglobin Hypertension
Men Women Men Women Men Women
Coe. N Coe. N Coe. N Coe. N Coe. N Coe. N
Age in 2007 < 65 0.220*** 1371 0 .237*** 1721 0.021 1371 0.093** 1721 0.080 1401 0.182*** 1752
(0.076) (0.055) (0.041) (0.042) (0.055) (0.057)
Age in 2007 65 0.119 752 -0.079 887 0.023 752 -0.041 887 0.158 757 0.051 901
(0.086) (0.088) (0.056) (0.058) (0.099) (0.079)
Years of education < 9 0.125* 1566 0.135*** 2224 0.058 1566 0.047 2224 0.088* 1597 0.175*** 2264
(0.066) (0.047) (0.039) (0.036) (0.052) (0.048)
Years of education 9 0.428*** 557 -0.515 384 -0.061 557 -0.393 384 -0.233 561 0.332 389
(0.149) (0.821) (0.061) (1.187) (0.167) (0.668)
Lagged log(PCE) < median 0.180* 994 0.115** 1232 0.072 994 0.034 1232 0.101 1012 0.082 1237
(0.083) (0.056) (0.047) (0.046) (0.065) (0.060)
Lagged log(PCE) median 0.101 1129 0.0013 1376 -0.0027 1129 0.044 1376 0.071 1146 0.203*** 1416
(0.094) (0.069) (0.046) (0.045) (0.066) (0.063)
Table 3.9.2: Disaggregated Groups 2
BMI Level Overweight Underweight
Men Women Men Women Men Women
Coe. N Coe. N Coe. N Coe. N Coe. N Coe. N
Age in 2007 < 65 -0.055 1166 0.172 1522 -0.073 1166 0 .260*** 1522 0.151* 1166 0.120 1522
(0.039) (0.375) (0.065) (0.081) (0.085) (0.084)
Age in 2007 65 -0.319 655 -0.623 769 -0.011 655 0.235 769 0.134 655 0.017 769
(0.419) (0.567) (0.092) (0.244) (0.187) (0.143
Years of education < 9 -0.052 1357 -0.032 1951 0.166 1357 0.185** 1951 0.155** 1357 0.145** 1951
(0.13) (0.219) (0.138) (0.069) (0.077) (0.063)
Years of education 9 0.044 464 0.158 340 -0.060 464 -0.221 340 0.044 464 -0.073 340
(0.248) (0.789) (0.116) (0.342) (0.169) (0.336)
Lagged log(PCE) < median -0.102 827 -0.414 1047 -0.037 827 0.157* 1047 0.109 827 0.075 1047
(0.114) (0.765) (0.89) (0.087) (0.093) (0.075)
Lagged log(PCE) median -0.276 994 0.114 1244 -0.009 994 0.228** 1244 0.167** 994 0.093 1244
(0.414) (0.340) (0.074) (0.099) (0.085) (0.083) 124
Table 3.9.3: Disaggregated Groups 3
GHS (Poor Health) ADLs Lung Capacity
Men Women Men Women Men Women
Coe. N Coe. N Coe. N Coe. N Coe. N Coe. N
Age in 2007 < 65 -0.017 1455 -0.126 1719 -0.017 1451 -0.051* 1716 0.080 1387 -0.001 1715
(0.078) (0.086) (0.029) (0.03) (0.074) (0.038)
Age in 2007 65 0.087 896 0.006 1049 -0.002 891 -0.075 1044 0.109 705 0.070 763
(0.074) (0.118) (0.111) (0.095) (0.095) (0.066)
Years of education < 9 0.075 1754 -0.032 2355 0.059 1746 -0.022 2347 0.094 1536 0.012 2097
(0.072) (0.663) (0.044) (0.031) (0.064) (0.035)
Years of education 9 0.001 597 -0.752 413 -0.241** 596 -0.288 413 -0.039 556 0.263 381
(0.141) (0.561) (0.103) (0.567) (0.120) (0.623)
Lagged log(PCE) < median 0.001 1069 -0.011 1269 0.048 1064 -0.095*** 1264 0.098 978 0.007 1160
(0.089) (0.048) (0.045) (0.034) (0.084) (0.042)
Lagged log(PCE) median 0 .107 1244 -0.126 1499 -0.027 1278 -0.019 1496 -0.032 1114 -0.001 1318
(0.123) (0.089) (0.066) (0.047) (0.098) (0.052)
125
For instance, in the case of hypertension, the path dependence coecients are 0.088 for
men and 0.175 for women with lower years of education, while no signicant dependence is
observed for the higher educated group. As mentioned earlier, this group might not be as
knowledgeable as the higher educated group in terms of healthy behaviors and may not go
to doctors as often. I nd no signicant dierences by age group or household per capita
expenditure.
3.6.4 Robustness Checks
Time-varying Measurement Errors
In the previous section, for simplicity, I assume that measurement errors are time-invariant.
In that case, external IVs such as parents' health and education, and respondent's edu-
cation and birth cohort as well as two and three-period lagged health measures are valid
IVs.
39
However, if the health measures have time-varying measurement errors,
it
, the two-
period lagged health measure is no longer valid. The equation accounting for time-varying
measurement errors becomes
H
it
=
H
H
it1
+
S
X
j=1

j
X
jit
+
R
X
j=1

j
Z
ji
+"
i
+"
h
+"
c
+"
it

H

it1
+
it
(3.11)
Then, the rst-dierenced error term in the second stage regression, "
it

H

it1
+
it
will be correlated with two-period lagged health through the measurement error,
it2
.
One might think that there would be attenuation bias towards zero in the lagged health
coecient if time-varying measurement error is a problem. Three-period lagged health is
still uncorrelated with the rst-dierenced errors so long as the measurement errors are not
serially correlated. Hence, I re-estimate the model using three-period lagged health plus
39
Two and three-period health measures are valid under the assumption that error terms are not serially
correlated.
126
the external IVs as instruments. The results are reported in tables B.1.1 - B.1.2. Only
GHS, BMI, and the number of diculties with ADLs are examined, since four waves of
data are now required.
For most of the health measures, the coecients on lagged health again turn out to
be close to zero, or change sign and become negative; they do not become more positive.
Hence, attenuation bias caused by time-varying measurement error does not seem to be
a problem. However, it needs to be noted that the F-tests on identifying instruments do
indicate a weak instrument problem, unlike when I also use two-period lagged health.
40
Attrition and Selective Mortality in Panel Data
Another potential problem that can arise in panel data is attrition, which can result
in biased estimates, especially if attrition is correlated with endogenous regressors. In
the IFLS sample, the attrition arises from three sources; (a)mortality, (b)refusals
41
and
(c)migration.
42
Given the estimation strategy adopted in this chapter, however, the attrition resulting
from individual, household, and community specic time-invariant characteristics should
no longer be a problem. First-dierencing removes all sources of time-invariant unobserv-
ables which may correlate with attrition and with any mortality selection. The possible
remaining source of attrition is from time-varying shocks. If the shocks in 1993 are corre-
lated with those in 1997 and so forth, those who were exposed to shocks would show much
40
I try subsets of the instruments to try to address the weak instrument problem. In some cases, the
F-statistics do rise, but not by very much.
41
Compared to higher-income countries such as the United States, refusal rates in household surveys are
low in most developing countries. See Thomas et al. (2012) for details.
42
Some of the health measures such as self-reported GHS or numbers of diculties with ADLs or physical
functioning also partly come from proxy interviews, if the main respondents are not available or moved out
of the origin households and hence, cannot be physically traced down. In this case, biomarkers such as
hemoglobin, hypertension or weight cannot be measured.
127
stronger path dependence. If attrition is caused by these shocks and people with stronger
path dependence tend to leave the sample, then the estimates will be downward biased.
Following Fitzgerald et al.(1998), a linear probability model is estimated to determine
whether attrition is related to health shocks. The results are reported in tables B.2.1 and
B.2.2 in appendix. The dependent variable is attrition, coded as 1 if the individual was lost
in either the third or the fourth wave (regardless of the attrition reasons), and 0 otherwise.
The right hand side regressors include changes in respondent's health measured by GHS,
BMI, and diculties with physical functioning or ADLs from 1993 to 1997, age and years
of education, log(PCE), parent's health and education, and location variables (rural and
the place of residence dummy variables), measured in 1993. The coecients on changes
in health, proxies for initial health shocks, are jointly not signicantly dierent from zero.
This indicates that health shocks are not correlated with later attrition including death.
The results also show that, in the case of GHS attrition, both men and women with higher
PCE are less likely to stay in the survey. In addition, for men, those whose fathers have
poor GHS are more likely to leave the sample. Women who live in rural area are more
likely to be followed up while having dead fathers has opposite eects on attrition.
3.7 Conclusion
This chapter examines the determinants of chronic diseases and explains their persistence,
using a panel data set from the Indonesian Family Life Survey (IFLS). This is done by
estimating a dynamic conditional health demand function. To address the endogeneity issue
that arises from the potential correlation between lagged health measure and unobservables,
I employ a rst-dierence two-step GMM. Two and three-period lagged health measures,
parents' health and education, and respondent's birth cohort are used as IVs to identify
the rst-dierenced one-period lagged health measure.
128
The correlations of health between waves are high but they fall to small numbers
in the FD-GMM specication. The coecient on one-period lagged health measure in a
dynamic health demand function captures the degree to which previous health has an eect
on current health, conditional on xed eects such as through the in
uences of genetics
or childhood health. Given that the coecients are relatively close to zero, I nd that
there is little path dependence in health status. Socio-economic status also has very little
in
uence on current health, again conditional on xed eects and on lagged health. It is
the xed eects that evidently play the most important roles in determining chronic health
conditions in later life.
In order to investigate if past health has dierent impacts across demographic or eco-
nomic groups, I disaggregate the sample across age, per capita household expenditure level
(PCE), and years of education. The results show that those with less education tend to
show more persistence, compared to those with higher education, but with no dierences
by age or PCE.
In terms of policy, a strong impact of early childhood health implies that investments
should be targeted towards the early stages of life. Stronger persistence of disease of lower
educated elderly also suggests that investments should be made not only towards health
in early life, but also towards better education in early life. Among the elderly, the focus
should be on those with lower education attainments.
129
Chapter 4
Conclusion
Developing countries, including Indonesia, have been going through a health transition
from infectious disease to chronic disease over the past 30 years. In 1980, 72% of all deaths
were caused by infectious diseases, while in 1992, more than half of all deaths were caused
by chronic disease. Generally speaking, children are more likely to be aected by infectious
diseases while chronic diseases are more prevalent among the elderly. The recent increase
in deaths caused by chronic diseases highlights the importance of further investigation of
elderly health. Chronic disease can lower the quality of life of older adults, and especially so
for those with insucient access to medical care, which is common in developing countries.
However, due to the misperception that starvation and infectious disease are still main
issues in developing countries, the elderly are somewhat forgotten.
In order to have a better understanding of the issues pertaining to older adults in devel-
oping countries, this dissertation identies the determinants of chronic health conditions
and their persistence among the elderly in Indonesia. It has been very well documented
as early life origins hypothesis that early childhood health is very strongly correlated with
own late health as well as parents' health. The main dierence that makes my dissertation
stand apart from the existing literature is that I look at whether these correlation persists
even when people age and what are the factors causing this persistence.
The ndings of the second chapter show that the strong health correlation persists even
people age. Interestingly, however, the magnitude of the correlation drops signicantly for
people born in richer areas, meaning that they are less likely to be aected by parents'
130
negative health in
uence. People born in more developed areas would have been exposed
to better environments at birth or during early childhood. Hence, this highlights the
importance of investment on less developed areas to mitigate health inequalities in terms
of providing better health services or facilities.
The third chapter focused directly on older adults' health dynamic patterns and inves-
tigates why they are not recovering from minor chronic health conditions such as hyper-
tension, low hemoglobin levels and etc. Surprisingly, it is not that people continue to be
hypertensive or having low hemoglobin problem because they were exposed to this shock
in the past. Rather, it is genetic endowments and poor childhood health, captured in xed
eects, that makes people carry on their diseases. In addition, disadvantaged groups such
as less educated people tend to be more aected by past bad health conditions. Unfortu-
nately, due to the nature of xed eects, it is dicult to pin down which channel is more
important between genetic endowments or childhood health. However, these results still
suggest that investments should be made towards earlier stages of life and not only towards
health but also towards education.
131
Bibliography
Adair, L. (1999). Filipino children exhibit catch-up growth from age 2 to 12 years. Journal
of Nutrition, 129:1140{1148.
Adams, P., Hurd, M., McFadden, D., Merril, A., and Ribeiro, T. (2003). Heathy, wealthy,
and wise? Tests for direct causal paths between health and socioeconomic status. Jour-
nal of Econometrics, 112:3{56.
Alderman, H., J., H., and Kinsay, W. (2006). Long-term consequences of early childhood
malnutrition. Oxford Economic Papers, 58(3):450{474.
Almond, D. and Chay, K. (2006). The long-run intergenerational impact of poor infant
health:Evidence from cohorts born during the Civil Rights era. Mimeo, Department of
Economics, University of California, Berkeley.
Almond, D. and Currie, J. (2011). Killing me sofely: The fetal origins hypothesis. Journal
of Economc Perspectives, 25(3):153{172.
Arellano, M. (2003). Panel data econometrics. Oxford University Press, Oxford.
Arellano, M. and Bond, S. (1991). Some tests of specication for panel data: Monte carlo
evidence and an application of employment equations. Review of Economc Studies,
58:277{297.
Arellano, M. and Bover, O. (1995). Another look at the instrumental variable estimation
of error-components models. Journal of Econometrics, 68(1):29{51.
Association, I. P. H. (1993). Analysis of the Health Transition in Indoneisa: Implications
for Health Policy. Jakarta, Indonesia: Indonesian Public Health Association.
Barker, D. (1994). Mothers, babies and health in later life. London, BMJ Publishing Group.
Barker, D., Winter, P., Osmond, C., Margetts, B., and Simmonds, S. (1989). Weight in
infancy and death from ischaemic heart disease. Lancet 2, pages 577{580.
132
Blundell, R. and Bond, S. (1998). Initial conditions and moment restrictions in dynamic
panel data models . Journal of Econometrics, 87:115{143.
Bound, J. and Krueger, A. (1991). The extend of measurement error in longitudinal
earnings data: do two wrongs make a right. Journal of Labor Economics, 12:345{368.
Bowles, S. and Gintis, H. (2002). The inheritance of inequality. Journal of Economic
Perspectives, 16(3):3{30.
Burstrom, B. and Fredlund, P. (2001). Self rated health: Is it as good a predictor of
subsequent mortality among adults in lower as well as in higher social classes. Journal
of Epidemiology and Community Health, 55:836{840.
Cameron, N., Preece, M., and Cole, T. (2005). Catch-up growth or regression to the mean?
Recovery from stunting revisited. Journal of Human Biology, 17:412{417.
Case, A., Fertig, A., and Paxson, C. (2005). The lasting impact of childhood health and
circumstance. Journal of Health Economics, 24:365{389.
Case, A., Lubotsky, D., and Paxson, C. (2002). Economic status and health in childhood:
The origins of the gradient. American Economic Review, 92(5):1308{1334.
Case, A. and Paxson, C. (2008). Height, Health and Cognitive Function at Older Ages.
American Economic Review Papers and Proceedings, 98:463{467.
Case, A. and Paxson, C. (2009). Early Life Health and Cognitive Function at Older Ages.
American Economic Review Papers and Proceedings, 99(2):104{109.
Case, A. and Paxson, C. (2010a). Causes and Consequences of Early Life Health. Demog-
raphy, 47(supplement):S65{S85.
Case, A. and Paxson, C. (2010b). The long reach of childhoood health and circumstance:
Evidence from the Whitehall II study. The Economic Journal, 121:F183{F204.
Contoyannis, P. and Jones, A. (2004). Socio-economic status, health and lifestyle. Journal
of Health Economics, 23:965{995.
Cooksey, E., Eberwein, C., and Olsen, R. (2012). Depressivity over the Life Course. Work-
ing paper, Center for Human Resource Research, Ohio State University.
Crimmins, E. and Finch, C. (2004). In
ammatory exposure and historical changes in
human life-spans. Science, 305:1736{1739.
133
Crimmins, E. and Finch, C. (2006). Infection, in
ammation, height and longevity. Pro-
ceedings of the National Academy of Science, 103(2):498{503.
Currie, J. and Moretti, E. (2007). Biology as destiny? Short- and long-run determinants of
intergenerational transmission of birthweight. Journal of Labor Economics, 25(2):231{
263.
Deaton, A. (1997). The analysis of household surveys: A microeconometric approach in
development policy. Baltimore, Johns Hopkins University Press.
Deaton, A. and Arora, R. (2009). Life at the top: the benets of height. Economics and
Human Biology, 7(2):133{136.
Deaton, A. and Paxton, C. (1998). Aging and inequality in health and income. American
Economic Review, 88(2):248{253.
D.Goldman, Lakdawalla, D., and Zheng, Y. (2011). Food Prices and the Dynamics of Body
Weight. Economic Aspects of Obesity. Ed. Michael Grossman, Naci Mocan. Chicago:
University of Chicago Press, pages 65{90.
Dick, H., Houben, V., Linblad, J., and Thee, K. (2002). The Emergence of a National
Economy: An Economic History of Indonesia 1800-2000. University of Hawii Press.
Dohrenwend, B. (1992). Socioeconomic status and psychiatric disorders: the causation-
selection issue. Science, 255:946{952.
Elo, I. and Preston, S. (1992). Eects of early life conditions on adult mortality. Population
Index, 58(2):186{211.
Emanuel, I., Filakti, H., Alberman, E., and Evans, S. (1992). Intergenerational studies of
human birth weight from the 1958 birth cohort: Evidence for a multigenerational eect.
British Journal of Obstetrics and Gynaecology, 99(1):67{74.
F., W., Strauss, J., and Sikoki, B. (2012). Socioeconomic Success and Health in Later Life:
Evidence from the Indonesian Family Life Survey. Aging in Asia, National Academy of
Sciences Press.
Fedorov, L. and Sahn, D. (2005). Socioeconomics determinants of children's health in
Russia: A longitudinal survey. Economic Development and Cultural Change, pages
479{500.
Fitzgerald, J., Gottschalk, P., and Mott, R. (1998). An analysis of sample attrition in
panel data. Journal of Human Resources, 33(2):251{299.
134
Foster, A. (1995). Prices, credit markets, and child growth in low income rural areas.
Economic Journal, pages 551{570.
Frankenberg, E. and Karoly, L. (1995). The 1993 Indonesian Family Life Survey: Overview
and eld report. RAND. DRU-1195/1-NICHD/AID.
Frankenberg, E. and Thomas, D. (2000). The Indonesia Family Life Survey (IFLS): Study
design and results from Waves 1 and 2. RAND. DRU-2238/1-NIA/NICHD.
Godfrey, K. and Barker, D. (2000). Fetal nutrition and adult disease. American Journal
of Clinical Nutrition, 71(Suppl):1344S{1352S.
Goldman, N. (1994). Social factors and health: the causation selection issue revisited.
Proceedings of the National Academy of Science, 91:1251{1255.
Grossman, M. (1972). On the concept of human capital and demand for health. Journal
of Political Economy, 80(2):223{255.
Hoddinott, J. (2008). Eect of a nutrition intervention during early childhood on economic
productivity in Guatemalan adults. Lancet, 371:411{416.
Hoddinott, J. and Kinsey, W. (2001). Child growth in the time of drought. Oxford Bulletin
of Economics and Statistics, 63(4):409{436.
Horton, S. (1986). Child nutrition and family size in Philippines. Journal of Development
Economics, 23(1):161{176.
Hurd, M. and McGarry, K. (1995). Evaluation of subjective probability distributions in
the health and retirement study. Journal of Human Resources, 30(s):S268{S292.
Idler, E. and Benyamini, Y. (1997). Self-Rated Health and Mortality: A Review of Twenty-
Seven Community Studies. Journal of Health and Social Behavior, 38(1):21{37.
J., M., Hoddinott, J., Behrman, J., and Martorell, R. (2009). The Impact of Improving
Nutrition During Early Childhood on Education Among Guatemalan Adults. Economic
Journal, 119(537):734{763.
Jonas, B. and Mussolino, M. (2000). Symptioms of depression as a prospective risk factor
for stroke. Phychosomatic Medicine, 62(4):463{471.
Kim, Y., Sikoki, B., Strauss, J., and Witoelar, F. (2011). Intergenerational correlation
of health among older adults: Empirical evidence from Indonesia. RAND Labor and
Population Working Paper No. WR-879.
135
Lleras-Muney (2005). The Relationship Between Education and Adult Mortality in the
United States. The Review of Economic Studies, 72(1):189{221.
Luo, Y. and Waite, L. (2005). The Impact of Childhood and Adult SES on Physical, Mental,
and Cognitive Well-Being in Later Life. Journal of Gerontology, 60(2):S93{S101.
Maccini, S. and Yang, D. (2009). Under the weather: Health, schooling and economic
consequeces of early-life rainfall. American Economic Review, 99(3):1006{1026.
Mani, S. (2012). Is there complete, partial or no recovery from childhood malnutri-
tion?: Empirical evidence from Indonesia. Oxford Bulletin of Economics and Statistics,
74(5):691{715.
Marmot, M. (2012). Multi-level approaches to understanding social determinants. in Berk-
man, Lisa and Ichiro Kawachi, eds. Oxford University Press.
Martorell, R. and Habicht, J. (1986). Growth in early childhood in developing countries.
Human Growth: A Comprehensive Treatise, New York: Plenum Press.
McArdle, J. (2010). Contemporary Challenges of Longitudinal Measurement Using HRS
Data. The SAGE Handbook of Measurement. London: SAGE Press.
Monteiro, C., Moura, E., Conde, W., and Popkin, B. (2004). Socioeconomic status and
obesity in adult populations of developing countries. Bulletin of the World Health
Organization, 82(12):940{946.
Penninx, B. and Guralnik, J.and Havlik, R. (1998). Chronically depressed mood and cancer
risk in older persons. Journal of The National Cancer Institute, 90(24):1888{1893.
Popkin, B. (1994). The nutrition transition in low-income countries: An emerging crisis.
Nutrition Reviews, 52(9):285{298.
S., B. and Rawlings, S. (2009). Gradients of the intergenerational transmission of health
in developing countries. IZA, 4353.
S., B. and Rawlings, S. (2011). Intergenerational persistence of health in developing coun-
tries: The penalty of gender inequality. Journal of Public Economics, 95:286{299.
Smith, J. (1997). Demographic and economic correlates of health in old age. Demography,
34(1):159{170.
Smith, J. (2007). The impact of socioeconomic status on health over the life-course. Journal
of Human Resources, 42(4):739{764.
136
Smith, J. (2009). The Impact of childhood health on adult labor market outcomes. The
Review of Economics and Statistics, 91(3):478{489.
Smith, J., Shen, Y., Strauss, J., and Zhao, Y. (2012). The Eects of Childhood Health
on Adult Health and SES in China. Economic Development and Cultural Change,
61(1):127{156.
Staiger, D. and Stock, J. (1997). Instrumental variables regression with weak instruments.
Econometrica, 65(May):557{586.
Stock, J., Wright, J., and Yogo, M. (2002). A survey of weak instruments and weak
identication in generalized method of moments. Journal of Business and Economics
Statistics, 20(4):518{529.
Strauss, J. (2009). The Fourth Wave of the Indonesian Family Life Survey (IFLS4):
Overview and Field Report. RAND. WR-675/1-NIA/NICHD.
Strauss, J. and Thomas, D. (1995). Human resources: empirical modeling of household
and family decisions. Handbook of Development Economics, 3.
Strauss, J. and Thomas, D. (1998). Health, nutrition and economic development. Journal
of Economc Literature, 36(2):766{817.
Strauss, J. and Thomas, D. (2008). Health over the life course. Handbook of Development
Economics, 4.
Thomas, D. (1994). Like Father Like Son; Like Mother Like Daugther: Parental Resources
and Child Height. Journal of Human Resources, 29(4):950{988.
Thomas, D. and Strauss, J. (2012). Cutting the costs of attrition: Results from the
Indonesia Family Life Survey. Journal of Development Economics, 98(1):108{123.
Thomas, D., Strauss, J., and Henriques, M. (1990). Child survival, height for age, and
household characteristics in Brazil. Journal of Development Economics, 33:197{234.
van Doorslaer, E. and Gerdtham, U.-G. (2003). Does inequality in self-assessed health
predict inequality in survival by income? Evidence from Swedish data. Social Science
and Medicine, 57:1621{1629.
Venkataramani, A. (2011). The intergenerational transmission of height: Evidence from
rural Vietnam. Health Economics, 20(12):1448{1467.
137
Wooldridge, J. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge,
MIT Press.
138
Appendix A
Appendix to Chapter 2
{ Questions in IFLS {
1. General health status (GHS)
Q. In general, how is your health?
A. Very healthy { 1, Somewhat healthy { 2, Somewhat unhealthy { 3, Unhealthy { 4
2. # of diculties with ADLs and physical functioning
a) To carry a heavy load (like a pail of water) for 20 meters
b) To sweep the house
oor yard
c) To walk for 5 kilometers
d) To draw a pail of water from a well
e) To bow, squat, kneel
f) To dress without help
g) To stand up from sitting position in a chair without help
h) To go to the bathroom without help
i) To stand up from sitting on the
oor without help
139
A. 1. Easily, 3. With diculty, 5. Unable to do it
3. Cognition (the average number of correctly immediate and delayed words)
List A. (Hotel, river, tree, skin, gold, market, paper, child, king, book)
List B. (Sky, ocean,
ag, rupiah, wife, machine, house, earth, school, butter)
List C. (Mount, stone, blood, corner, shoes, letter, girl, house, valley, car)
List D. (Water, mosque, doctor, castle, re, garden, sea, village, baby, table)
4. Parent's death (Non-coresident parents)
Q. Is your father (mother) still alive?
A. Yes | 1, No | 3
5. Parent's general health status (GHS)
Q. How is the health status of your father (mother) now/before his (her) death?
A. Very healthy | 1, Somewhat healthy | 2, Somewhat unhealthy | 3
Very unhealthy | 4, Don't know | 8
6. Parent's ADL problem
Q. Now/before death does/did your father (mother) need help with basic personal needs
like dressing, eating, or bathing?
A. Yes { 1, No { 3, Unwilling to answer { 7, Don't know { 8
140
Table A.1.1: Means and Standard Deviations of Variables 1
MEN
Variables Description Obs Mean SD. Min Max
Poor GHS Somewhat unhealthy or very unhealthy = 1 3081 0.23 0.42 0 1
# of ADLs diculties The sum of the number of diculties of ADLs 3080 1.04 1.89 0 9
# of IADLs diculties The sum of the number of diculties of IADLs 3081 0.55 1.19 0 5
BMI Underweight: < 18.5 2974 0.20 0.40 0 1
Overweight: 25 2974 0.17 0.38 0 1
Hemoglobin Men: less than 13, Women: less than 12 2979 0.30 0.46 0 1
Hypertension Systolic140 or diastolic90 or doctor diagnosis 2985 0.52 0.50 0 1
Cholesterol Total Cholesterol240 2957 0.11 0.31 0 1
HDL HDL < 40 2931 0.65 0.48 0 1
Cognition The mean of correctly immediate and delayed recalled words 2317 3.56 1.58 0 9
Depression Short CES-D score 2830 3.90 3.17 0 26
Father's death Being dead in 2007 3081 0.94 0.23 0 1
Mother's death Being dead in 2007 3081 0.81 0.39 0 1
Father's poor GHS Somewhat unhealthy or very unhealthy now or before death = 1 3081 0.50 0.50 0 1
Mother's poor GHS Somewhat unhealthy or very unhealthy now or before death = 1 3081 0.47 0.50 0 1
Father's ADL problem Need help with basic personal needs now or before death = 1 3081 0.23 0.42 0 1
Mother's ADL problem Need help with basic personal needs now or before death = 1 3081 0.25 0.43 0 1
Java dummy variable Born in Jakarta, Java, Yogyakarta or Bali (developed areas) = 1 3081 0.69 0.46 0 1
Father's education At least some primary school 3081 0.09 0.29 0 1
Completed primary school 3081 0.24 0.43 0 1
Completed junior high school 3081 0.06 0.24 0 1
Mother's education At least some primary school 3081 0.07 0.25 0 1
Completed primary school 3081 0.17 0.38 0 1
Completed junior high school 3081 0.02 0.15 0 1
141
Respondent's age 50-59 3081 0.49 0.50 0 1
60-65 3081 0.15 0.36 0 1
65-70 3081 0.14 0.35 0 1
70-75 3081 0.10 0.30 0 1
75-80 3081 0.06 0.23 0 1
80- 3081 0.06 0.24 0 1
Respondent's height 3081 158.40 9.83 16 189
Respondent's edu At least some primary school 3081 0.29 0.45 0 1
Completed primary school 3081 0.29 0.45 0 1
Completed junior high school 3081 0.09 0.29 0 1
Completed senior high school 3081 0.17 0.38 0 1
142
Table A.1.2: Means and Standard Deviations of Variables 2
WOMEN
Variables Description Obs Mean SD. Min Max
Poor GHS Somewhat unhealthy or very unhealthy = 1 3608 0.29 0.45 0 1
# of ADLs diculties The sum of the number of diculties of ADLs 3605 1.79 2.17 0 9
# of IADLs diculties The sum of the number of diculties of IADLs 3605 1.01 1.32 0 5
BMI Underweight: < 18.5 3482 0.20 0.40 0 1
Overweight: >= 25 3482 0.30 0.46 0 1
Hemoglobin Men: less than 13, Women: less than 12 3474 0.35 0.48 0 1
Hypertension Systolic140 or diastolic90 or doctor diagnosis 3493 0.63 0.48 0 1
Cholesterol Total Cholesterol 240 3457 0.23 0.42 0 1
HDL HDL < 40 3445 0.39 0.49 0 1
Cognition The mean of correctly immediate and delayed recalled words 2309 3.22 1.57 0 10
Depression Short CES-D score 3222 4.56 3.59 0 27
Father's death Being dead in 2007 3608 0.95 0.23 0 1
Mother's death Being dead in 2007 3608 0.84 0.37 0 1
Father's poor GHS Somewhat unhealthy or very unhealthy now or before death = 1 3608 0.47 0.50 0 1
Mother's poor GHS Somewhat unhealthy or very unhealthy now or before death = 1 3608 0.45 0.50 0 1
Father's ADL problem Need help with basic personal needs now or before death = 1 3608 0.24 0.42 0 1
Mother's ADL problem Need help with basic personal needs now or before death = 1 3608 0.26 0.44 0 1
Java dummy variable Born in Jakarta, Java, Yogyakarta or Bali (developed areas) = 1 3608 0.68 0.47 0 1
Father's education At least some primary school 3608 0.06 0.24 0 1
Completed primary school 3608 0.23 0.42 0 1
Completed junior high school 3608 0.05 0.22 0 1
Mother's education At least some primary school 3608 0.05 0.23 0 1
Completed primary school 3608 0.15 0.35 0 1
Completed junior high school 3608 0.02 0.15 0 1
143
Respondent's age 50-59 3608 0.46 0.50 0 1
60-65 3608 0.15 0.36 0 1
65-70 3608 0.15 0.36 0 1
70-75 3608 0.10 0.30 0 1
75-80 3608 0.07 0.25 0 1
80- 3608 0.06 0.25 0 1
Respondent's height 3608 146.70 10.57 14 198
Respondent's edu At least some primary school 3608 0.29 0.46 0 1
Completed primary school 3608 0.19 0.39 0 1
Completed junior high school 3608 0.06 0.24 0 1
Completed senior high school 3608 0.08 0.28 0 1
144
Table A.2.1: The Distribution of Father's Death and GHS
Death = GHS very healthy somewhat healthy somewhat unhealthy unhealthy missing
Alive 8 251 85 10 13 367
(2.18%) (68.39%) (23.16%) (2.72%) (3.54%)
Dead 100 2395 2409 733 685 6322
(1.58%) (37.88%) (38.11%) (11.59%) (10.84%)
Total 108 2646 2494 743 698 6689
(1.61%) (39.56%) (37.29%) (11.11%) (10.44%)
Source: IFLS4
Table A.2.2: The Distribution of Mother's Death and GHS
Death = GHS very healthy somewhat healthy somewhat unhealthy unhealthy missing
Alive 25 835 237 43 24 1164
(2.15%) (71.74%) (20.36%) (3.69%) (2.06%)
Dead 63 2172 2133 640 517 5525
(1.14%) (39.31%) (38.60%) (11.58%) (9.36%)
Total 88 3007 2370 683 541 6689
(1.32%) (44.95%) (35.43%) (10.21%) (8.09%)
Source: IFLS4
145
Appendix B
Appendix to Chapter 3
146
Table B.1.1: FD-GMM: In the Presence of Time-Varying Measurement Errors 1
Overweight Underweight BMI level
Men Women Men Women Men Women
Lagged health measure -0.271 -0.0546 -0.145 -0.199 -0.589** -0.605
(0.208) (0.162) (0.254) (0.352) (0.260) (3.979)
Yrs of edu * lagged age 0.000930 0.000428 -0.000254 -0.000300 0.00339 -0.0237
(0.000654) (0.000828) (0.000456) (0.000500) (0.0211) (0.0983)
Duration -0.000265 0.00144 0.000261 0.00267 -0.0870 -0.0944
(0.00564) (0.00410) (0.00371) (0.00298) (0.253) (0.588)
Duration * lagged age -0.0000190 -0.0000303* 0.000000330 -0.00000734 -0.000203 -0.000373
(0.0000134) (0.0000165) (0.0000150) (0.0000165) (0.000734) (0.00230)
Lagged log (PCE) 0.00582 0.0174* -0.00957 0.0112 0.108 0.411
(0.00957) (0.00924) (0.0115) (0.0105) (0.393) (1.280)
Lagged dist. * year dummies Yes Yes Yes Yes Yes Yes
N 1821 2291 1821 2291 2351 2768
F stat on excluded IVs 1.263 1.464 1.360 0.920 1.450 0.786
Hansen J statistics 48.13 64.87 41.27 56.91 13.99 6.48
(p-value) (0.347) (0.053) (0.631) (0.177) (0.999) (0.999)
Table B.1.2: FD-GMM: In the Presence of Time-Varying Measurement Errors 2
General Health Status (Poor Health) Number of diculties with ADLs
Men Women Men Women
Lagged health measure -0.487 -0.466* -0.146 -0.442
(0.334) (0.248) (0.446) (0.292)
Yrs of edu * lagged age -0.00104 -0.000571 -0.00242 0.00277
(0.00166) (0.00162) (0.00306) (0.00365)
Duration -0.0193 -0.00684 -0.0627** -0.0782***
(0.0128) (0.0173) (0.0258) (0.0260)
Duration * lagged age 0.000106** 0.0000426 0.000725*** 0.000638***
(0.0000484) (0.0000335) (0.0000808) (0.0000855)
Lagged log (PCE) 0.0228 0.0205 -0.0233 -0.0414
(0.0201) (0.0257) (0.0559) (0.0511)
Lagged dist. * year dummies Yes Yes Yes Yes
N 2351 2768 2342 2760
F stat on excluded IVs 1.184 1.340 0.670 0.783
Hansen J statistics 27.50 30.42 50.80 94.76
(p-value) (0.845) (0.770) (0.560) (0.001)
147
Table B.2.1: Determinants of Sample Attrition 1
Attrition = 1 Men Women Men Women
in BMI -0.000301 0.00108
(0.000286) (0.000948)
in ADL -0.000131 -0.000147
(0.000417) (0.000739)
Age in 1993 0.0000572 -0.000272* 0.0000420 -0.000340**
(0.000171) (0.000151) (0.000183) (0.000159)
Yrs of education 0.000151 -0.0000756 -0.000818 -0.00140
(0.000717) (0.000635) (0.000888) (0.00102)
Log (PCE) 0.00418 0.00285 0.00719** 0.00544**
(0.00304) (0.00178) (0.00324) (0.00244)
Rural 0.00270 -0.0127* -0.00458 -0.0152**
(0.00732) (0.00648) (0.00747) (0.00669)
Father's education
At least some primary 0.00341 -0.00890 0.000199 -0.0120
(0.00451) (0.0102) (0.00480) (0.0106)
Completed primary 0.0101 0.00452 0.00473 0.00410
(0.00622) (0.00624) (0.00639) (0.00679)
Jr. high & more -0.00265 0.0178 0.00900 0.0283
(0.00539) (0.0152) (0.0112) (0.0179)
Mother's education
At least some primary -0.00259 -0.00565 -0.00449 0.00599
(0.00746) (0.0102) (0.00687) (0.0132)
Completed primary -0.00773 -0.00853 -0.00323 -0.000782
(0.00712) (0.00716) (0.00707) (0.00762)
Jr. high & more 0.0227 -0.0267** 0.0535 -0.0274**
(0.0243) (0.0123) (0.0350) (0.0139)
Parent's health
Father's GHS 0.00816* 0.000672 0.00920* 0.00204
(0.00483) (0.00484) (0.00532) (0.00502)
Father's ADL -0.00217 -0.00422 -0.00845 -0.00533
(0.00750) (0.00624) (0.00818) (0.00614)
Father's death -0.00498 0.00802** -0.00318 0.00880**
(0.00971) (0.00324) (0.00957) (0.00362)
Mother's GHS -0.00128 -0.00537 0.000257 -0.00720
(0.00433) (0.00463) (0.00449) (0.00470)
Mother's ADL 0.00503 0.00682 0.0109 0.00647
(0.00648) (0.00593) (0.00751) (0.00580)
Mother's death 0.00696 0.00821* 0.00579 0.00430
(0.00548) (0.00426) (0.00596) (0.00469)
Residence place in 1993 (dist.) Yes Yes Yes Yes
Sample size 2705 3089 3039 3321
Source: IFLS 1993,1997
in BMI = BMI
1997
BMI
1993
in ADLs = ADLs
1997
ADLs
1993
*** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors are reported in parenthesis.
Attrition =1 if the individual are dropped out of the sample, and 0 otherwise.
148
Table B.2.2: Determinants of Sample Attrition 2
Attrition = 1 Men Women Men Women
in BMI -0.000314 0.00108
(0.000294) (0.000948)
in ADL 0.000369 -0.000103
(0.000377) (0.000670)
in GHS -0.00126 -0.000388 -0.00650* 0.00119
(0.00375) (0.00431) (0.00350) (0.00350)
Age in 1993 -0.0000536 -0.000198 0.0000455 -0.000267*
(0.000178) (0.000164) (0.000168) (0.000156)
Yrs of education -0.000846 -0.00131 0.000149 -0.0000714
(0.000865) (0.00102) (0.000717) (0.000634)
Log (PCE) 0.00650** 0.00613** 0.00410 0.00287
(0.00301) (0.00245) (0.00305) (0.00180)
Rural -0.00666 -0.0157** 0.00279 -0.0126*
(0.00750) (0.00652) (0.00734) (0.00645)
Father's education
At least some primary -0.000298 -0.0114 0.00350 -0.00897
(0.00464) (0.0105) (0.00453) (0.0102)
Completed primary 0.00566 0.00424 0.00982 0.00451
(0.00624) (0.00731) (0.00613) (0.00625)
Jr. high & more 0.0207 0.0187 -0.00275 0.0178
(0.0152) (0.0204) (0.00536) (0.0152)
Mother's education
At least some primary -0.00605 0.00671 -0.00289 -0.00561
(0.00661) (0.0128) (0.00750) (0.0103)
Completed primary -0.00663 -0.000863 -0.00796 -0.00856
(0.00727) (0.00780) (0.00717) (0.00717)
Jr. high & more 0.0424 -0.00332 0.0224 -0.0268**
(0.0353) (0.0270)
Parent's health
Father's GHS 0.00874* 0.00305 0.00794* 0.000658
(0.00496) (0.00502) (0.00477) (0.00478)
Father's ADL -0.00699 -0.00544 -0.00205 -0.00425
(0.00778) (0.00596) (0.00749) (0.00625)
Father's death -0.00301 0.00731** -0.00489 0.00797**
(0.00915) (0.00359) (0.00972) (0.00331)
Mother's GHS 0.00149 -0.00619 -0.00128 -0.00542
(0.00432) (0.00466) (0.00432) (0.00462)
Mother's ADL 0.0104 0.00497 0.00518 0.00690
(0.00735) (0.00577) (0.00650) (0.00598)
Mother's death 0.00724 0.00430 0.00713 0.00820*
(0.00574) (0.00458) (0.00551) (0.00426)
Residence place in 1993 (dist.) Yes Yes Yes Yes
Sample size 3223 3460 2705 3089
F-test for health shocks (p-value) 0.295 0.6709
Source: IFLS 1993,1997
in BMI = BMI
1997
BMI
1993
in ADLs = ADLs
1997
ADLs
1993
in GHS = GHS
1997
GHS
1993
*** signicant at 1%, ** signicant at 5%, * signicant at 10%
Robust standard errors are reported in parenthesis.
Attrition =1 if the individual are dropped out of the sample, and 0 otherwise.
149
Table B.3.1: First-stage Regression for Low Hemoglobin
Men Women
Coef. Std. Err. Coef. Std. Err.
Excluded IVs
LowHB
t2
-0.8042*** (0.0229) -0.7500*** (0.0202)
Birth Cohorts 55 - 60 -0.0058 (0.0352) -0.0372 (0.0359)
60 - 65 0.0620 (0.0594) -0.0376 (0.0573)
65 - 70 0.0557 (0.0817) -0.0243 (0.0773)
70 - 75 0.0393 (0.1073) -0.0244 (0.102)
75 - 0.0881 (0.146) -0.0766 (0.1383)
Mother's Edu At least some primary 0.0285 (0.0443) 0.0502 (0.0437)
Completed primary -0.0092 (0.0327) -0.0062 (0.0327)
Completed Jr. High 0.0691 (0.071) -0.0624 (0.0831)
Father's Edu At least some primary 0.0160 (0.039) -0.0764* (0.0416)
Completed primary -0.0158 (0.0293) -0.0304 (0.0293)
Completed Jr. High -0.0557 (0.0538) -0.0232 (0.0602)
Parent's health Mother Death 0.0022 (0.0267) -0.0215 (0.0267)
Father Death 0.0527 (0.0384) 0.0021 (0.0404)
Mother GHS 0.0209 (0.0253) -0.0239 (0.0251)
Father GHS -0.0323 (0.0241) 0.0202 (0.0257)
Mother ADL 0.0123 (0.0283) -0.0072 (0.0277)
Father ADL 0.0092 (0.0295) 0.0110 (0.0289)
Included IVs
Education * lagged age -0.002** (0.0008) -0.0010 (0.001)
Duration 0.0008 (0.0066) -0.0051 (0.0058)
Duration * lagged age 0.0001 (0.0001) 0.0001 (0.0001)
Lagged log (PCE) 0.0042 (0.014) 0.0280** (0.0132)
Lagged dist. * year dummies Yes Yes
F statistics on excluded IVs 57.338 56.965
(p-value) (0.0000) (0.0000)
First-stage regression for low hemoglobin reported in column 5-6, table 3.3.1
150
Table B.3.2: First-stage Regression for Hypertension
Men Women
Coef. Std. Err. Coef. Std. Err.
Excluded IVs
hypertension
t2
-0.5467*** (0.0216) -0.517*** (0.0185)
Birth Cohorts 55 - 60 0.0596* (0.0355) 0.0264 (0.0313)
60 - 65 0.0294 (0.0575) 0.0404 (0.0505)
65 - 70 0.0834 (0.0795) 0.0076 (0.0676)
70 - 75 0.0787 (0.105) 0.0357 (0.089)
75 - 0.0376 (0.1401) 0.0755 (0.1228)
Mother's Edu At least some primary 0.0113 (0.0392) 0.0048 (0.0387)
Completed primary 0.0314 (0.0319) 0.0099 (0.0306)
Completed Jr. High -0.1498** (0.0765) 0.0235 (0.0813)
Father's Edu At least some primary -0.0982*** (0.0365) -0.0399 (0.038)
Completed primary -0.0489* (0.0292) -0.0299 (0.0264)
Completed Jr. High -0.0023 (0.054) -0.0567 (0.0532)
Parent's health Mother Death -0.0221 (0.028) 0.0275 (0.0228)
Father Death -0.0028 (0.0398) 0.0820** (0.0371)
Mother GHS 0.0440* (0.0256) 0.0100 (0.0217)
Father GHS -0.0027 (0.0251) -0.043* (0.022)
Mother ADL -0.0195 (0.0293) 0.0169 (0.025)
Father ADL -0.0046 (0.0299) -0.0075 (0.025)
Included IVs
Education * lagged age 0.0023*** (0.0009) 0.0002 (0.0009)
Duration -0.0021 (0.0058) -0.0032 (0.0052)
Duration * lagged age 0.0001 (0.0001) 0.0001 (0.0001)
Lagged log (PCE) 0.0051 (0.0141) -0.0045 (0.0126)
Lagged dist. * year dummies Yes Yes
F statistics on excluded IVs 30.257 33.896
(p-value) (0.0000) (0.0000)
First-stage regression for hypertension reported in column 5-6, table 3.4
151

Abstract (if available)

Abstract This dissertation contributes to the design of better policy implications and improvements of well-being among older adults, especially when resources are limited. This is done by using proper econometric methods and taking advantage of the richness of the Indonesian Family Life Survey, a panel data set containing detailed information for both respondents and their biological parents. ❧ Chapter 2 analyzes the health transmission from parents to their children when they become older adults. I match the health status of older respondents to parental characteristics (several health measures and education) to see if any correlations exist in Indonesia. I find that strong intergenerational correlations exist. For example, children having parents with more difficulties with ADLs are more likely to have the same problem as older adults. However, surprisingly, the magnitude of correlations becomes significantly lower for those born in richer areas of Indonesia such as Java and Bali. This suggests that the level of development at birth or early childhood, which may include having better health infrastructure, substitutes for the influence of parental health and hence highlights the importance of public policies that focus on community level infrastructure development in less developed areas, in order to solve health inequality. ❧ Chapter 3 examines the determinants of chronic health conditions and explains their persistence. I incorporate dynamics into a health demand function, finding strong correlations between lagged and current health measures when nothing else is controlled. This could represent the influence of lagged health or fixed unobserved factors such as genetic endowments and childhood health. To disentangle these, I estimate the influence of lagged health by using first-difference two-step generalized method of moments (FD-GMM), where the first-differencing removes fixed unobserved factors and keeps only lagged health. I found that it is this fixed effect, representing both genetic endowments and early life cycle including childhood health, that is most important in explaining later life chronic conditions. The impact of past health conditioning on the fixed effect, captured by the coefficients on lagged health measures, is weak, with estimated coefficients relatively close to zero. These results are robust to potential measurement errors in health and to sample attrition. Socio-economic status also has very little influence on current health, again conditioned on the fixed effect and on the influence of lagged health. In order to investigate if past health has different impacts across demographic or economic groups, I disaggregate the sample across age, household per capita expenditure level (PCE), and years of education. The results show that those with less education tend to show more persistence, compared to those with higher education. ❧ In developing countries like Indonesia, health disparities are serious issues since they persist over generations due to the lack of proper interventions and also prevail among disadvantaged groups such as those with less education. My dissertation suggests more effective and efficient ways to employ interventions and resolve health disparities, using robust econometric technique and rich data sets.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

Three essays on health & aging

PDF

Essays on health insurance programs and policies

PDF

Three essays on economics of early life health in developing countries

PDF

Essays on development and health economics

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Essays on human capital accumulation -- health and education

PDF

Long-term impacts of childhood adversity on health and human capital

PDF

Essays on economics of education

PDF

Selection and impacts of early life events on later life outcomes

PDF

Three essays on the statistical inference of dynamic panel models

PDF

Three essays on health economics

PDF

Essays on econometrics

PDF

Three essays in health economics

PDF

Essays on family planning policies

PDF

Essays on economics of education and private tutoring

PDF

Essays on education: from Indonesia to Los Angeles, CA

PDF

Essays on development economics

PDF

Essays on development and health economics: social media and education policy

PDF

Essays in panel data analysis

Asset Metadata

Creator Kim, Younoh (author)

Core Title Essays on health economics

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 04/15/2013

Defense Date 03/24/2013

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

Tag chronic conditions,dynamic panel data model,elderly health,OAI-PMH Harvest,path dependence

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John A. (committee chair), Moon, Hyungsik Roger (committee member), Nugent, Jeffrey B. (committee member)

Creator Email oh2211@gmail.com,younohki@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-236479

Unique identifier UC11293368

Identifier etd-KimYounoh-1547.pdf (filename),usctheses-c3-236479 (legacy record id)

Legacy Identifier etd-KimYounoh-1547.pdf

Dmrecord 236479

Document Type Dissertation

Format application/pdf (imt)

Rights Kim, Younoh

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