0 v1:
it‘nun‘

w

 

 

_ A

J$.§a3a1

 
 

a3: . gmfiwa. ma. , , a§§§u§ a. fig?

 

‘ll'l I H

 

"£858
2
QOnfl

Ewa’ioasi‘,

 

 

 

LIBRARY
Michigan State
- ' This is to certify that the
UhlverSIty dissertation entitled

 

ECONOMICCRISES AND DEMOGRAPHIC OUTCOMES:
EVIDENCE FROM INDONESIA

presented by

Pungpond Rukumnuaykit

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Economics

 

 

     
  

/22;%E:/
' r Pr'ofessor's Signature

gale A [er MI 02005

Date

 

MSU is an Affirmative Action/Equal Opportunity Institution

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateDuepGS-pts

 

ECONOMIC CRISES AND DEMOGRAPHIC OUTCOMES:
EVIDENCE FROM INDONESIA

By

Pungpond Rukumnuaykit

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

2003

 

 

 

ABSTRACT

ECONOMIC CRISES AND DEMOGRAPHIC OUTCOMES:
EVIDENCE FROM INDONESIA

By

Pungpond Rukumnuaykit

This dissertation examines the short-run impacts of the Indonesian economic
crises on different demographic outcomes of infants and women. In the first chapter, I
examine whether the recent Indonesian financial crisis and the 1997/1998 drought and
smoke haze crises had adverse effects on infants’ birthweight and mortality. In the
second chapter, I examine the effects of the 1998 economic crisis on the ages of female
first marriages and first births. This dissertation uses data from three waves of the
Indonesia Family Life Survey: IFLSl (1993), IFLSZ (1997), and IF LS3 (2000).

The methodology used in the first chapter is to compare health conditions of
newborns of different birth cohorts. The estimations of both neonatal and post-neonatal
mortality risks are carried out using multivariate regressions with socio-economic control
variables such as mother’s education, place of residence (province/community), and
gender of the child. In addition, mortality risks are estimated using hazard models to
capture the mortality risks at different ages (in months). The paper uses both
nonparametric and parametric hazard models to estimate the hazard rates. The effects of
the crises on birthweights are analyzed using multivariate regressions and comparisons of
birthweight cumulative distributions.

Estimated results on mortality outcomes show that the financial crisis had adverse

impacts on neonatal mortality in both urban and rural areas. The adverse effects of the

financial crisis on post-neonatal mortality risks were larger and more statistically
significant for urban infants than for rural infants. The drought/smoke crisis adversely
affected post-neonatal mortality risks in rural areas. The estimated results show that rural
infants born during the drought/smoke crisis experienced approximately a 4.4 percentage
points increase in their infant mortality risks (44 per 1,000 live births). The magnitude of
the effects ahnost doubled after controlling for community fixed effects. None of the
crises significantly affected birthweight. I find that the lack of evidence on the adverse
effects may be due to selection problems in reported birthweights.

In the second chapter, effects of the crisis on ages of female first marriage and
first births are estimated using hazard models. The methods used include both parametric
and non-parametric estimations of the marriage and first birth hazards, conditional and
unconditional on being married. Estimated results indicate that overall there was an
increase in the probability of getting married and a decrease in the probability of having
first births among Indonesian women during the crisis in both conditional and
unconditional analyses. These findings support the hypothesis that marriages of
individuals in a household and delaying first births may have been used as income-
smoothing mechanisms in the time of the crisis. Results from this paper are not sufficient
to draw any conclusion on why an increase in marriage probability and a delay in having
first birth took place. We speculate that women are more likely to get married during the
time of the crisis to take advantage of economies of scale and specialization in household
production and consumption. The delays of first births might be due to a consumption-
smoothing consideration or other supply factors such as the separation of spouses when

relocation of individuals occurred during the crisis.

To my parents

iv

ACKNOWLEDGEMENTS

First, I would like to express my profound gratitude to the Royal Thai
Government, specifically to the Ministry of Foreign Affairs, for granting me support to
pursue my undergraduate and graduate studies. The support has given me not only a
future career, but also opportunities to be open to higher education abroad, to work on
projects I am interested in, and to prepare myself to better serve my country and society.

I owe a debt of gratitude to many people whose help has been crucial to my
success in completing this dissertation. First of all, I have been privileged to have the
direction and unfailing guidance of an excellent mentor, my chair, John Strauss.
Professor Strauss made invaluable contributions to my thesis. He has always motivated
me to do my best work and offered timely feedback and keen insights into the
significance of my research. Professor Strauss has been like a father to me for the past
three years. His supervision and advice has helped substantially to my study and my stay
at MSU. He is the reason I feel very lucky to have come to Michigan State University. I
am also grateful to Jeff Biddle, John Goddeeris, Jeff Wooldridge, John Giles, and Andrei
Shevchenko, who have contributed insightful and critical comments and suggestions to
my research. I am deeply grateful to Albert Park, my other mentor, who introduced me
to a fascinating study of development economics. Albert also provided me with his kind
encouragement and wise guidance at crucial times. Most of all, he has never failed to

believe that I can succeed.

My heartfelt appreciation goes out to Jan Svejnar for his constant support and his
kind words of encouragement, David Neumark and Dale Belrnan for believing in me and
giving me an opportunity to teach and learn a great deal from teaching, and Thomas
J eitschko and the department of Economics at MSU for making my academic experience
go smoothly throughout the years. Thanks also to Firman, my esteemed colleague, whom
I learned a lot from. I have enjoyed the camaraderie of Firrnan, Nina, Dung, Mu Ren,
Lebohang, Elda, Elena, and Paula, whom I have benefited much from their
companionship throughout the years. Thanks them for taking me in for a wonderfirl time
at MSU. Many other people have helped me weather the emotional storms and stress
throughout my study, some of whom deserve a special mention. Thanks to Laura and
Anna for their wonderful friendship and their dependable and enduring support.
Foremost, I thank Daeng for being there for the many laughs and tears, for loving me in
all good and bad times, and for being my best fiiend. I could never ask for a better fiiend.

Finally, I owe much to my family. My sister and brothers have always offered me
their support, sympathy, and admiration, my sister and my younger brother especially,
who in many ways are my number one fans. I owe to my parents, to whom this
dissertation is dedicated, for always believing in me and encouraging me to achieve my
goals. My parents have always been there for me. My father has never failed to do what
he could to make me feel at home away from home. My mother has been my best
listener. She always listened to what I had to say no matter how long and how
uninteresting. Her steadfast emotional support has been invaluable in helping me to
focus on my academic pursuits. My parents forever inspire me with their infinite

capacity for love, joy, and faith.

vi

TABLE OF CONTENTS

LIST OF TABLE ..................................................................................... ix
LIST OF FIGURES ................................................................................... xi
CHAPTER 1 ............................................................................................ 1
Introduction ....................................................................................... 1
Conceptual Framework and Literature ........................................................ 9
A Theoretical Model ............................................................................ 16
The effect of a temporary adverse income shock on mortality risk .................. 19
Measuring Mortality Rates ................................................................. 19
Data ............................................................................................... 23
Method of Analysis ............................................................................. 28
Testing Different Measures of Mortality Rates ......................................... 28
Birthweight Reporting ..................................................................... 29
Mortality and Birthweight Statistics ..................................................... 31
Testing the Effect of Economic Crises on Birthweight and Infant Mortality ........... 35
Results ............................................................................................ 42
How did mothers with different education levels cope with the crises? ...................... 49
Hazard Models .................................................................................. 56
Parametric Estimation of Child Mortality Using Hazard Models ........................ 57
Discussions ...................................................................................... 61
Conclusion ....................................................................................... 65

vii

CHAPTER 2 .......................................................................................... 68

Introduction ...................................................................................... 68
Backgrounds .................................................................................... 73
Theoretical Framework ........................................................................ 76
The Impacts of Economic Crises on Marriage ......................................... 76
The Impacts of Economic Crises on Fertility .......................................... 81
Previous Studies ................................................................................. 86
Methodology ..................................................................................... 90
Nonparametric Estimation of Hazard of Marriage and Fertility ...................... 91
Parametric Estimation of Hazard of Marriage and Fertility .......................... 91
Data ............................................................................................... 94
Marriage Data .............................................................................. 95
Fertility Data ................................................................................ 95
Duration Data ............................................................................... 97
Identifying “Crisis” Covariate ............................................................ 99
Results .......................................................................................... 100
Results on ages of female first marriage ............................................... 100
Results on first marital pregnancies .................................................... 104
Unconditional Fertility Hazard Estimation ............................................ 107
Robustness ..................................................................................... 109
Conclusion ..................................................................................... 1 10
APPENDICES ..................................................................................... 112
BIBLIOGRAPHY ................................................................................. 155

viii

LIST OF TABLES

Table 1: Neonatal and Post-neonatal Mortality Rates Comparisons
(Per 1,000 live births) .................................................................... 113

Table 2: Proportion of births that have reported birthweight by birth location (%).......114

Table 3: Birth Location (%) ...................................................................... 114
Table 4: Birth Location Comparison (%) ...................................................... 115
Table 5: Mortality Rates (per 1,000 live births) ............................................... 116
Table 6: Birthweight Statistics ................................................................... 116
Table 7: Mother’s Education ..................................................................... 117
Table 8: Child Mortality Rates (per 1,000 live births) by Mother’s education. . . . . . . . ....1 17
Table 9: Birthweight First and Second Order Stochastic Dominance ...................... 118
Table 10: Birthweight Regressions .............................................................. 119
Table 11: Mortality Linear Probability Regressions (Urban) ................................ 120
Table 12: Mortality Logistic Regressions (Urban) ........................................... 121
Table 13: Mortality Linear Probability Regressions (Rural) ................................ 122
Table 14: Mortality Logistic Regressions (Rural) ............................................ 123
Table 15: Infant Mortality (Low Education and High Education) .......................... 124
Table 16: Neonatal Mortality (Low Education and High Education) ...................... 125
Table 17: Post-neonatal Mortality (Low Education and High Education) ................ 126
Table 18: Parametric Estimations of Mortality Hazard Rates (Urban) ..................... 127
Table 19: Parametric Estimations of Mortality Hazard Rates (Rural) ...................... 128
Table 20: Summary Statistics of Marriage Age by Birth Cohort ........................... 129

ix

Table 21: Maniage Age by Year of Marriage ................................................. 129

Table 22: Parametric Maniage Hazard Regressions (Cox Proportional Hazard) ......... 130
Table 23: Parametric Marriage Hazard Regressions (Gamma Distribution) ............... 131
Table 24: Parametric Maniage Hazard Regressions (Lognormal Distribution) ........... 132
Table 25: Parametric Fertility Hazard Regressions (Cox Proportional Hazard) .......... 133
Table 26: Parametric Fertility Hazard Regressions (Lognormal Distribution). . . . . . . 134
Table 27: Parametric Fertility Hazard Regressions (Weibull Distribution). . . . . . . . . . . .....135
Table 28: Parametric Fertility Hazard Regressions (Cox Proportional Hazard) .......... 136
Table 29: Parametric Fertility Hazard Regressions (Cox Proportional Hazard) .......... 137

Table 30: Unconditional Fertility Hazard Regressions (Cox Proportional Hazard)..... . .138

Table 31: Unconditional Fertility Hazard Regressions (Gamma Distribution). . . . . . . . ....139

LIST OF FIGURES

Figure 1: Timing of the IFLS and the Rp/U SD Exchange Rate ............................. 143
Figure 2: Food Price Index (January 1997 = 100) ............................................. 144
Figure 3: Government Health Expenditures (1993 = 100) ................................... 145
Figure 4: Rice Area Harvested (1997-2000) ................................................... 146
Figure 5: Mortality Rates (per 1,000 live births) .............................................. 147
Figure 6a-6c: Cumulative Distribution of Birthweight (Kg) ................................ 148
Figure 7a-7c: Non-parametric Hazard Rates (by Month) .................................... 149
Figure 8a-8c: Parametric Estimated Hazard Rate (Weibull Distribution) .................. 150
Figure 9: Nonparametric First-Marriage-Age Hazard by Birth Cohort ..................... 151
Figure 10: Nonparametric F irst-Marriage-Age Hazard by Education ...................... 151
Figure 11: Nonparametric F irst-Marriage-Age Hazard by Education ...................... 152
Figure 12: Nonparametric First-Marriage-Age Hazard by Residence at Age 12 ......... 152
Figure 13: Nonparametric Fertility Hazard by Marriage Cohort ............................ 153
Figure 14: Nonparametric Fertility Hazard by Education .................................... 153
Figure 15: Nonparametric Fertility Hazard by Residence .................................... 154
Figure 16: Unconditional Nonparametric Fertility Hazard by Birth Cohort ............... 154

xi

CHAPTER 1

CRISES AND CHILD HEALTH OUTCOMES: THE IMPACTS OF ECONOMIC AND
DROUGHT/SMOKE CRISES ON INFANT MORTALITY AND BIRTHWEIGHT IN

INDONESIA

Introduction

This paper examines the impacts of the recent Asian financial crisis on infant
mortality and birthweight in Indonesia. There have been a number of economic and
policy studies focusing on impacts of economic crises on finance and production.
Although some studies provide evidence of negative impacts of economic crises on real
outcomes, little is known about the impact of economic crises on child health outcomes
such as nutrition, and mortality. Often, the association between financial and production
disturbances and these outcomes are assumed (e.g. an adverse shock to production is
thought to be associated with worse child health outcomes). This paper utilizes data
from the Indonesian Family Life Survey (IF LS) to examine impacts of the crises on child
health outcomes directly. Specifically, we study the impacts of the crises on birthweight
and infant mortality.

Prior to the Asian financial crisis, Indonesian rising levels of income, education,
and public health programs had been successful in reducing infant mortality and
improving the overall health of children. The infant mortality rate was reduced by more

than half between the 1960s and the 19905: from 145 deaths per one thousand live births

in 1967 to 46 in 1997 (World Bank, 1999).l Table 1a shows neonatal mortality and post-
neonatal mortality rates2 prior to the crisis from the 1997 Demographic Household
Survey. Data from this table shows that the rates of neonatal mortality and post neonatal
mortality noticeably declined over the period of 15 years from 1982 to 1997. Indonesians
experienced higher decline in post-neonatal mortality rate (from 31.0 to 23.9 per 1,000
live births) than in neonatal mortality rate (from 28.4 to 21.8 per 1,000 live births).

The Asian financial crisis struck Indonesia in January 1998.3 Figure 1 shows that
the sustained crisis period lasted more than one year with the peak in Rupiah/USD
exchange rate in July 1998. As shown in Figure 2, the food prices in both urban and rural
areas increased more than 250 percent at the peak of the crisis. This substantial increase
in food prices is argued by Alatas (2002)4 to be a major source of impacts of the crisis felt
by Indonesians, except those in the top of the income distribution. Simulation results
from Alatas’s study indicates that the increase in food prices between February 1999 and
February 2000 accounted for approximately a 40 percent of the increase in the poverty
rate afier the crisis. According to Strauss et al (2002), food expenditures (excluding

expenditures on tobacco and alcohol) accounted for approximately 50 percent of a typical

 

l Daly, Patricia and Fadia Saadah. “Indonesia: Facing the Challenge to Reduce Maternal Mortality.” East
Asia and the Pacific Region Watching Brief. World Bank. June 1999. Issue 3.

2 Neonatal mortality is defined as death before one-month old. 0-28 and 0-30 day periods are both used
among researchers. This paper uses 0-30 days. Post-neonatal mortality is defined as death at ages 1 to 11.9
months. Infant mortality is defined as deaths at age 0 to 11.9 months.

3 Refer to figure 1.

’ Alatas, Vivi. “What Happen to Indonesia’s Poverty? A Micro Simulation Exercise Using Household
Surveys. Manuscript. World Bank. Jakarta, Indonesia. March 2002.

Indonesian’s household budget in urban areas and 57 percent in rural areas. The food
shares of household budget are higher among the poor. 5

While an increase in food prices could help net food-producers, this large increase
in the food prices during the crisis resulted in a sharp reduction of real incomes for most
of the Indonesian population, who are net food purchasers. Frankenberg, Thomas, and
Beegle in ”The Real Costs of Indonesia’s Economic Crisis: Preliminary Findings from
the IFLSZ+,” (1999) 6 reported that the proportion of households below the poverty line
rose from about 11 percentage points in 1997 to almost 20 percentage points in 1998.7

Overall government health expenditures per capita were not sustained at the peak
pre-crisis level. Figure 3 shows real government health expenditures from 1980 to 2000.
According to Leiberman et a1. (2001),8 the government per capita outlay on heath
expenditures fell significantly during the crisis by 2.9 percent and by 6.6 percent in

successive years. Then in 1999/2000, the expenditures rebounded. However, during the

 

5 Between 1997 and 1998 there was a significant increase in the food share in both rural and urban
households. According to Frankenberg, Thomas, and Beegle (1999), the increase in food shares was
concentrated among households whose per capita expenditure was below the median of the population. A
significant portion of the increase in food share can be attributed to an increase in the share of the
expenditures on staples (from 13 percent to 21 percent for urban households and from 31 percent to 39
percent in rural households). As a result of a significant increase in the share of staples, some food shares
such as that of meat are reported to decline. According to Strauss et a1. (2002), nominal wages also
increased during the financial crisis, but the increase was considerably less than the increase in food and
nonfood prices. Therefore, real wages for those that rely on market wages also declined.

6 Frankenberg E., Thomas D., and Beegle K. 1999. “The real costs of Indonesia’s Economic Crisis:
Preliminary Findings from the Indonesia Family Life Surveys.” DRU-2064-NIA/NICHD. Santa Monica,
CA: RAND.

7 Allowed for higher inflation in rural than urban areas as indicated by the price data collected in the IFLS
communities. (Frankenberg E., Thomas D., and Beegle K. 1999. “The real costs of Indonesia’s Economic
Crisis: Preliminary Findings from the Indonesia Family Life Surveys.” DRU-2064-NIA/NICHD. Santa
Monica, CA: RAND.)

8 Lieberman, 8., M. Juwono, and P. Marzoeki. “Government health expenditures in Indonesia through
December 2000: An Update.” World Bank East Asia and the Pacific Region Watching Brief. October 15,
2001, Issue 6.

crisis, Indonesia received a 278 percent increase in donor assistance, which contributed to
the sustainability of health financing and spending. This donor assistance helped dampen
the shock to the government budget, resulting in a five percent decrease in overall
outlays9.

According Frankenberg et a1. (1999), prices of both public and private healthcare
increased from late 1997 to late 1998, but public healthcare prices increased relatively
more than the prices of private healthcare. The median price of BCG immunization for
children and tetanus toxoid immunization for pregnant women rose significantly in public
facilities,‘0 but not in private facilities. There was a decrease in the quality of public
healthcare such as a reduction of the quantity of drugs given to patients and an increase in
the number of referrals to other providers. In terms of supplies, public facilities were
found to have been more affected by changes in the availability of drugs and supplies
(such as injections and bandages) while private facilities were more affected by the price
increase of these inputs. Overall, public and private facilities that provided vitamin A
declined in number. Vitamin A is essential for children under three since it reduces their
vulnerability to infectious disease. Those above the median of the income distribution
were found to shift away from public healthcare. Visits to the posyandu (healthcare post)
by children under five dropped from 46.7 percent to 27.7 percent.

In addition to the financial crisis, some rural areas in Indonesia were badly

affected by the 1997-1998 drought. The drought crisis was a consequence of climatic

 

9 Lieberman, 8., M. Juwono, and P. Marzoeki. “Government health expenditures in Indonesia through
December 2000: An Update.” World Bank East Asia and the Pacific Region Watching Brief. October 15,
2001, Issue 6.

1° The prices rose from 500 Rp. in 1997 to 750 Rp. in 1998 for Child immunization and from 500 Rp. in
1997 to 900 Rp. in 1998 for Tenanus Toxoid.

conditions identified with the El Nino Southern Oscillation (ENSO). The impacts of the
ENSO on Indonesia included postponement of monsoons, damaging rainless winds,
greater rate in infestations, disturbances in fishing patterns, scarcity of drinking water and

forest fires. ‘1

In Kalimantan and Sumatra, the drought, combined with a political
decision on land clearing for plantation concessionaires, local populations, and new
immigrants, led to fires that lasted for months between late 1997 and early 1998. The
fires affected 9.75 million hectares and over 700 million tons of carbon were emitted into
the atmosphere causing a major health hazard in Indonesia, Singapore, and Malaysia.

An apparent result of the drought crisis was a sharp drop in food production,
especially rice (Indonesia’s major food crop). Figure 4 shows that the amount of land
harvested for rice in late 1997 was smaller than usual. The harvesting cycle in late 1997
and early 1998 was delayed by as much as two months. It has been reported that some of
the worst damage was to Maluku, Nusa Tenggara, and parts of Sulawesi and

southwestern Sumatra.12

Farmers in the driest and poorest provinces in Indonesia were
reported at times to survive on one meal a day and to eat food generally reserved for
livestock.13 As a result of the drought crisis, Indonesia became a large food aid recipient

in 1998.

 

” James J. Fox. “The 1997-1998 Drought in Indonesia.” Natural Disasters and Policy Responses in Asia:
Implications for Food Security. Harvard University Asia Center. August, 1999.

'2 James J. Fox. “The 1997-1998 Drought in Indonesia.” Natural Disasters and Policy Responses in Asia:
Implications for Food Security. Harvard University Asia Center. August, 1999.

'3 James J. Fox. “The 1997-1998 Drought in Indonesia.” Natural Disasters and Policy Responses in Asia:
Implications for Food Security. Harvard University Asia Center. August, 1999.

According to Sastry in his study, “Forest Fires, Air Pollution, and Mortality in
Southeast Asia,14 smoky haze caused by a widespread series forest of fires in Indonesia
between April and November 1997 had possible short-term and long-term effects on
health. Possible problems that may have resulted in mortality include respiratory
infections and chronic conditions. Sastry uses levels of micro-particulate matter with
diameter less than 10 microns (PMio) and the mean daily visibility in kilometers as
measures of air quality to analyze the effects of smoke haze on mortality (non-traumatic,
cardiovascular, respiratory deaths) of population of different age groups (all ages, <1, 65-
74,and >74 years) in Kuala Lumpur and Kuching, Malaysia, in 1996-1997. Sastry found
significant short-term cumulative effects of smoke haze on mortality in all age groups.
Sastry claims that the displacement of deaths from the smoke haze was short-term,
however, in one segment of the Malaysian population -- those age 65-74 in Kuala
Lumpur--- there was an upward shift in mortality that lasted at least a few weeks. Sastry
suggests that an implication of his results on the short-term effects of the smoke haze in
Malaysia is that the effects in Indonesia, where the main fires took place, are likely to
have been large.

Evidence from IFLSZ+ report suggests that the short-run impacts of the financial
and the drought/smoke crises on children’s health have been small. Results from
physical assessments show no deterioration in children’s health status. There were only
negligible changes in the measurements of children older than six months in their height-

for-age and weight-for-height. Very young children were well protected from the effects

 

r4 Sastry, Narayan. “Forest Fires, Air Pollution, and Mortality in Southeast Asia.” Demography. Volume
39 number]. February 2002.

of the crisis although there is a suggestion that weight-for-height of this group of children
may have worsened (Frankenberg, Thomas, and Beegle, 1999).

This paper uses IF LS data to examine the effects the crises on neonatal mortality,
post-neonatal mortality, and birthweight. Mortality status and birthweight of children of
different cohorts (born during pre-crisis, crisis, and post-crisis periods) are examined. In
addition, detailed data on births and times of death of children allow this study to
examine mortality hazard rates at different specific ages of children.

Estimated results on mortality outcomes show that the financial crisis had adverse
impacts on neonatal mortality in both urban and rural areas. The adverse effects of the
financial crisis on post-neonatal mortality risks were larger and more statistically
significant for urban infants than for rural infants. Overall, the financial crisis increased
infant mortality risks by about 3.2 percentage points in both urban and rural areas, a very
large effect.

The drought/smoke crisis adversely affected post-neonatal mortality risks in rural
areas. The increase in the post-neonatal mortality risk is about 3.1 percentage points.
When community-fixed effects are controlled for, the drought/smoke crisis had much
larger effects. Overall, the drought/smoke crisis had no statistically significant adverse
effects on infant mortality in urban areas, while the effects in rural areas were large. Our
estimates show that rural infants born during the drought/smoke crisis experienced
approximately 4.4 percentage points increase in their infant mortality risks.

Results from hazard models confirm that the financial crisis and the drought crisis
had adverse effects on neonatal and post-neonatal mortality. The financial crisis increased

the odds of both neonatal and post-neonatal mortality for urban children more than for

rural children. As expected, in rural areas, the crisis-noncrisis differential of the mortality
risk at specific age (months) is relatively smaller for financial crisis than that of the
drought crisis.

Our findings on differential effects on infants born to mothers with different
levels of education show that in urban areas infants born to mothers with different levels
of education exhibited significantly different trends in mortality risks over time. We also
find that even though some of the crises had adverse effects on infant mortality, infants
born to mothers with different education levels did not experienced different adverse
crisis effects.

Results from the cumulative distribution comparisons of birthweights suggest that
the financial crisis also had adverse impacts on birthweight in urban areas. However,
under multivariate analyses, the adverse effect seems to disappear. None of the crises
affected birthweights in rural areas. The lack of an evidence on the adverse effects may

be due to a selection problem in reported birthweights.

Conceptual Framework and Literature

The financial crises and the drought/smoke haze crisis represent short-term
exogenous shocks to Indonesian households. Although specific biological mechanisms
by which smoke haze may affect child health outcomes are not directly estimated,15 both
crises are expected to have negative consequences on child health through resource
availability, income, prices, and environment.

According to agricultural household models (Singh, Squire, and Strauss (1986);
Hill et al. (1993)), in a country with both agriculture and manufacturing sectors, an
adverse agricultural shock not only reduces average resource availability but also affects
the distribution of resources between the two sectors, largely through its effects on the
relative prices of goods and services households consume. The direction of the effect
depends on how the price of the agricultural good, expressed in terms of the
manufactured good, responds to the shock.

With an increase in the price of agricultural goods after an adverse shock,
agricultural households who are net producers may receive a greater amount of
manufactured good for each unit of the agricultural good it does produce although each
agricultural household produces less. However, if the price of manufactured goods also
significantly increased during the crisis, these net sellers of agricultural goods could
suffer from an overall price increase.

While an increase in food prices could help net food-producers, households in
sectors not benefiting either directly or indirectly from the favorable price movement

such as net buyers of rice may face an erosion of their purchasing power as the relative

prices of goods change. As noted earlier, a large increase in food prices and an adverse
income shock during the financial crisis resulted in a sharp reduction of real incomes for
most of the Indonesian population, who are net food purchasers.

An important dimension of a reduction in purchasing power that is relevant to this
paper is through changes in the costs of raising children. These costs include both direct
and indirect (opportunity) costs. When there is a reduction in purchasing power as a
result of adverse income shocks or a rise in the price of agricultural goods, households
are likely to reduce their consumption of health inputs and family planning services,
provided that these goods and services are normal goods to the households. In addition,
the effects of the crises on consumption could be worsened if there is an increase in
healthcare costs and a reduction in availability of healthcare services, which are widely
provided by the government.

It is, however, worth noting that a drought or a financial crisis in a particular year
may have only a limited effect on the household’s resources in that year if the household
can transfer resources from another place or another time through the formal sector (such
as bank loans or crop insurance) or the informal sector (such as loans or transfers from
family or friends). However, if the shock occurs at the aggregate level, for instance a
widespread financial crisis, the ability to transfer resources could be limited.16 In this
case, we expect to observe short-run negative effects of the adverse shock on health

outcomes such as in child mortality.

 

'5 We do not estimate health production function in this paper.

'6 See Bardhan and Udry (1999) for detailed discussions.

10

A large mainstream body of research on the effects of short-term economic
fluctuations on demographic outcomes has used time-series data. The most influential
work in this area is the study by Lee (1981) of the impacts of conditions in pre-industrial
North and Western Europe using economic indices such as grain prices and weather.'7
The evidence from this study shows that post-infant mortality is positively associated
with real prices. In a similar study, Galloway (1988)18 found that a ten percent increase
in grain prices in pre-industrial Europe leads to an increase of approximately 1 percent in
mortality.l9

A number of authors have tried to study the effects of economic fluctuations on

demo a hic outcomes in less develo ed countries.20 A ma'or effort is a collective work
gr P P J

 

'7 R. Lee, “ Short-term variation: vital rates, prices and weather,” in E.A. Wrigley and RS. Schofield
(eds), The Population History of England, 1541-1871 (Cambridge, Mass: Harvard University Press, 1981),
P. Galloway, “Basic Patterns of annual variations in fertility, nuptiality, mortality, and prices in pre-
industrial Europe,” Population Studies, 42 (1988), pp. 275-302, D. Weir, “Life under pressure: France and
England, 1670-1680,” Journal of Economic History, 44 (1984), pp. 27-47. T. Richards, “Weather,
nutrition, and the economy: the analysis of short-run fluctuation in births, deaths, and marriages, France
1740-1909,” in T. Bengtsson et al. (eds.), Pre-Industrial Population Change (Stockholm: Alrnquist and
Wiksell, 1984).

'8 Galloway, P. “Basic pattern in annual variations in fertility, nuptiality, mortality, and prices in pre-
industrial Europe.” Population Studies 42(2): 275-302. 1988.

'9 Although he observed that the relationship between prices and mortality in countries that are more
developed economically is weaker.

20 J. Bravo, “Economic Crisis and mortality: short and medium-term changes in Latin America.” Paper
presented at the Conference on the Peopling of the Americas, Veracruz, Mexico (1992). J. Brovo,
“Demographic Consequences of structural adjustment in Chile.” Paper presented at the Seminar on
Demographic Consequences of Structural Adjustment in Latin America, Ouro Preto, Brazil (1992). K. Hill
and A. Palloni, “ Demographic responses to economic shocks: The Case of Latin America.” Paper present
at the Conference on the Conference on the Peopling of the Americas, Veracruz, Mexico (1992). A.
Palloni and K. Hill, “The Effects of Structural Adjustments on mortality by age and cause in Latin
America.” Center for Demography and Ecology. Working Paper 92-22, University of Wisconsin, Madison,
Wis. (1992). D. Reher and J.A. Ortega, “ Short run economic fluctuations and demographic behaviour:
some examples from twentieth century South America.” Paper presented at the Seminar on Demographic
behaviour: some examples from twentieth century South America.” Paper presented at the Seminar on
Demographic Consequences of Structural Adjustment in Latin America, Ouro Preto, Brazil (1992).

ll

1.21 In this study, first marriage, timing of first and

by the National Research Counci
second birth, and child mortality in seven countries in Sub-Saharan Africa were
examined using data from the Demographic Health Surveys (DHS) that were conducted
in each of the seven countries some time in the period between 1986 and 1990. In this
paper, various economics indicators are used as measures of economic fluctuations. The
indicators used are the gross domestic product per capita, the quantity of exports, term of
trade, and commodity prices. The authors find strong evidence of adverse effects of
economic reversals on time of first maniage and first births. 22 They, however, found no
effect of the economic reversals on child mortality net of trend, except in Ghana (in rural
areas) and Nigeria.

Palloni, Hill, and Aguirre23 employed distributed lag models and used average
GDP to study the effects of short-term economic fluctuations on marriage, marital
fertility and mortality during 1920-1990 in Latin America. They found the directions of
the response of the number of marriages are not uniform across countries. The net effects
(the sum of all lagged coefficients) are positive in all cases, except Guatemala and
Mexico, while the magnitudes of the net effect elasticities vary across countries, from

0.01 (in Mexico) to 0.67 (in Chile). Their results from the estimated responses of births

show considerable greater heterogeneity. In five out of eleven countries, the coefficients

 

2‘ Hill, K, G. Adansi-pipim, L. Assogba, A. Foster, J. Mukiza—gapere, and C. Paxson. “ Demographic
Effects of Economic Reversals in Sub-Saharan Africa. National Research Council.National Academy
Press. Washington DC. 1993.

22 Their results are strongest for the effects on first birth. The authors found a positive relation to economic
variation net of trend in all seven countries studied, except Kenya. Evidence of marriage delays as a result
of economic reversals were found in Botswana, Senegal, and Togo (especially in urban areas).

23 Palloni, A, K. Hill, and GP. Aguirre. “Economic Swings and Demographic Changes in the History of
Latin America. Population Studies, 50 (1996), pp. 105-132.

12

 

 

for lags 0 and 1 of the average GDP are positive as expected, but only lagl in Cuba is
significantly different from zero. The estimated net effects are positive in only seven
countries, and the magnitude of the net effect elasticity ranges from 0.04 (in Chile) to
0.88 (in Cuba). The effects on infant mortality rate also display considerable
heterogeneity in terms of direction of the effects. Results from only five countries are
negative, as expected, and the estimated elasticity of the net effects ranges from 0.08 (in
El Salvador) to 0.61 (in Panama). The results fi'om pooled sample estimations using
pooled sample (except Cuba) give estimated elasticity of the net effects of 0.12, 0.19, and
-0. 12 on births, marriages, and infant mortality rate respectively. An important finding as
suggested by the authors is that the effects of the lagO and lagl on infant mortality are
significant, but their absolute size are small (-0.10, and 0.00), about one-third that of the
effects on marriages and births.

Mckenzie (2002), in his study of how Mexican households coped with aggregate
shocks of the Peso crisis of 1994-1996, finds little role for inter-household transfers in
consumption smoothing in the presence of aggregate shocks. He finds that the average
transfer that households made to non-household members was reduced by 25 percent.
Although remittances to Mexican households from friends and family members in the
United States increased, on average the households received 19 percents less gifts and
donations from other Mexican households.

Another pioneer empirical study in this area is the study of child mortality of
children born or conceived during the Dutch Hunger Winter of 1944-45 by Stein et (11.24

This study uses vital registration records to compare age-specific mortality rates for

13

 

 

different cohorts. The authors found an excess mortality of children born or conceived
during the famine crisis. However, due to data limitations, the authors control for only
social status (manual/non-manual occupation).

In their study of the effects of the 1974-75 famine in Bangladesh, Razzaque et al.
(1990)25 address the above data issues using richer data from Matlab field research in a
rural area in Bangladesh. Mortality of children of different cohorts, born during famine,
post-famine, and non-famine periods, is examined using various socioeconomic controls
such as child’s gender and household economic status. Using linear logistic regression
models, they found excess mortality of children born or conceived during the famine.
However, Razzaque and his colleagues found that the effects of the famine on mortality
were not uniform. Better off household experienced little effect.

The incidence of the recent Indonesian financial crisis and the 1997/1998 drought
crisis and the availability of IFLS data allow this study to address the affects of larger
exogenous short-run shocks on demographic outcomes. This paper uses IFLS data to
examine the effects of the crises on neonatal mortality, post—neonatal mortality, and
birthweight. By using IFLS3 and earlier IF LS waves (IF LS] and IFLSZ), mortality data
on child cohorts that were born during pre-crisis, crisis, and post-crisis periods can be
examined. In addition, we can allow for the time trends that are independent of the crises
when studying these adverse effects. Since the IFLS covers respondents in both rural and

urban areas of Indonesia, the area-specific impacts of the crisis can also be studied. In

 

2’ Stein, Z., M.Susser, G. Saenger and F. Morolla, “Famine and Human Development: The Dutch Hunger
Winter of 1944-1945, New York (1975).

25 Razzaque, A., N. Alam, L. Wai, and A. Foster. “Sustained effects of the 1974-75 famine on infant and
child mortality in a rural area of Bangladesh.” Population Studies 44: 145-154. 1990.

14

 

 

addition, detailed data on births and times of death of children allow this study to

examine mortality hazard rates at different specific ages of children.

15

A Theoretical Model

The theoretical model in this paper is an adaptation of the work by Foster
(1995).26 Consider a household j. Abstracting from fertility selection, assume parents
care about the health of their surviving children, but not their individual consumption.
Denote t the beginning of period t. Let 12,-, be the health status of the child 1' at time t and
C, a vector of the household’s consumption of goods and services other than those that
are inputs in the production of children’s health. Single period household’s utility
function is assumed to be additively separable between consumption and child’s health.
The utility function is also assumed to be increasing and concave in household

consumption and child’s health status (v'(C, ) > 0, v"(C, ) < 0 and u'lhfi, )> 0, u'lhfl, )< 0).

The expected discounted utility of household j at time s is

V, =ESZT:,B"’[v(C,)+ZI: “(h,).M,,],

(=3 i=1

where E, the expectation conditional on information at time s, ,6 is the discount
factor, and I is the total number of children in the household, and the subscript j is
dropped for notational simplicity. u(h,-‘) is the utility of the household from the health of
the child i, h;. Mi, is the child 1' ’s mortality index, where M, =1 if the child is alive in
period t and M,-, = 0 if the child is not alive in period t. In every period t, the child dies

with certainty when his health, h", is below a heath threshold, his.

 

26 Foster, Andrew D. “Price, Credit Markets and Child Growth in Low-Income Rural Areas.” The
Economic Journal, Volume 105, Issue 430 (May, 1995), 551-570.

16

The production function of the child’s health status in the current period is
characterized as a function of the child’s health status in the last period, augmented by
health inputs in the last period. Let h,, be the child’s health status at time t given
characteristics at the individual level (such as age and gender), household level (such as
education of the mother), and community level (such as quality of water supply), 2,,, and
unobserved factors that affect health, (5),. Following Foster (1995), the effects of health
inputs, N,,, in the last period on the current period health status are assumed to be
proportional to a time-varying rate at which health inputs are translated into health status
gain. This rate is a function of the child’s characteristics and unobserved factors 5,,
Parents are assumed to know 2,, with full certainty. The health status of the child at time

t+1 can be characterized as

hir+l = fir (hit ;Zir ’61? )+ kit (zit ’6ir )N

i!’

 

> O and the Mortality index can be characterized as

where afi: (hi; Zr: ’ 6i! )

i!

M -— 1 if hir(hr'r-1 ’ Nit-1’ zit-l ’61-]! ) > hit. (zit ’ 81'!)
it 0 Otherwise

where 8,, is unobserved factors, unknown to the household, that affect the
mortality threshold at time t.

Assume no inter-temporal borrowing and lending, the household’s budget
constraints at time t is

I
prr +prNZNir =yr

i=1

l7

By the Law Iterated Expectation, the expected discounted utility of household j at

time s can be written as

V. air-{ran z (u(h..) probw . =11t— »] 27

(=3 i=1

Let function g” characterize mortality risk of the child i in period t;

grain Zir’ 8i!) = pr0b(Mir = 0): pr0b(hir(hir—l’ Nit—1’ zit—I’6I-1I)< hi: (Z 40817))

where g,’ < 0 and g," > 0

Suppose there is a temporary adverse income shock at period 0, either Co or M0 or
both will be reduced in order to satisfy the budget constraints at time 0. If children’s

health inputs are considered normal goods, Nro will be reduced as a result of a reduction

in income. Supposed Nio is reduced by de, then the health status of the child 1' in period

 

27Proof:
Vs =Est§sflt—S[vC(Ct)+i§1u(hit)-Mit]
=12, i 3’ SvC( C)+t§sfl" élEsiulh,,)-M,-,l

N
9)

By the Law of Iterated Expectation,

VS =Estgsflt—SV(Ct)+t§sflt—Siél ESiEt—l’hit (“(hit ).Mit )i
=Es £fl1_sv(Ct)+ En -S,1Eé s[uh(i t)-pr0b(M.t =l|t— —l)]
t=s t=s =1
=Es,§,fi“’ivl9)*.§ (..(1. W4, _,,.-.))1.

l8

1 will be reduced by k,0(z,0,6io)dNio. Furthermore, this temporary adverse income

shock will have spillover effects on future health status of the child in every period after
period 0 as the health status at any period t depends on the health status in period t-I in
the health production function. Specifically, a decrease in Mg by de will reduce the

health status of the child 1' in period t by

 

ah. ’ 6- h- :2..,5.-.
5—1;:de =[EII fi3( 2h. )]‘ki0(zioa§io)dNio'

IS

The effect of a temporary adverse income shock on mortality risk

An adverse shock to income in period 0 that results in a change of health inputs
by de will have adverse effects on the mortality risk through a reduction of the health

input by

 

 

6 prob(M 1., = 0) _ 6g" 6h" 6N“, _ 6g” 1L1 aft: (his;zis,5i3)
ayo - ahit aNio ayo aha 3:1 ah

rs

J. kiO (ziO ) 5m ) dNiO

Measuring Mortality Rates

There are two principal estimation methods to calculate mortality rates: direct
methods and indirect methods. (Spiegelman, 1955;28 Pressat, 1978”) Direct methods
calculate mortality directly using data on the date of birth of children, survival status, and

the dates of death or ages at death of deceased children.

 

28 Spiegelmarr, M. Introduction to Demography. 1955. The Society of Actuaries. Chicago. Illinois.

29 Pressat, R. Statistical Demography. Translated and Adapted by Damein A. Courtney. 1978. Methuen &
Co Ltd.

19

The direct methods require data that are usually only obtained in specifically
designed surveys with birth/pregnancy histories or from vital statistics systems. There
are three variants of direct estimation methods: a vital statistics approach, a synthetic life
table approach, and a true cohort life table approach.

“A vital statistics approach” is an approach in which the number of deaths to
children under age 12 months in a particular period is divided by the number of births in
the same period. Under “a synthetic cohort life table approach,” mortality probabilities for
small age segments based on real cohort mortality experience (e.g. 0, 1-2, 3-5, 6-11
months) are first calculated. Then these component death probabilities are combined into
the mortality rates, taking into account exposure to mortality risk of each age cohort.
Specifically, component death probability for each small age segment is calculated by
dividing the number of deaths to live-born children during specified age range and
specified time period by number of surviving children at beginning of specified age range
during the specified time period. Births and death incidences within each specific age
and specific time ranges are weighted according to exposure to mortality risk, which is
indicated by birthdate and mortality risk of interest (e. g. neonatal mortality). This
approach allows full use of the most recent data and is also specific for time periods.
However, this approach requires intensive computation and the number obtained using this
method could be influenced by the arbitrary length of age segments used.3O

Under “a true cohort life table approach,” the number of deaths to children under

age 12 months of a specific cohort of births are divided by the number of births in that

 

3° Studies done during the World Fertility Survey have shown the difference to be negligible when using
monthly segments and using 0, 1-2, 3-5, 6-11, 12-23, 24-35, 36-47, 48-59 months segments. (Demographic
Health Survey).

20

cohort. This method gives true probabilities of death but requires that all children in the
cohort must have been fully exposed to mortality risk.31 Although this method does not
take into account the most recent experience because of exposure requirement, it is the
method we chose in this paper since this method gives true probability of death while the
firll exposure problem is mitigated by using IFLS3, which includes information on children
born in the crisis. Most of these children had had firll exposure to neonatal and post-
neonatal mortality risk by the time of the interview in 2000. Moreover, the problem is less
severe because information of children born in the IF LS survey year (who had less than one
year mortality exposure) can be added using data from subsequent IF LS waves so that all
of these observations are fully exposed to infant mortality.32

An “indirect method” is used to calculate death rates when age-specific death rates
for the community are not available, but the total number of deaths is known. By this
method, the number of actual age-specific deaths in the community is multiplied by a
constant number of the age-specific death rates33 in a standard population that are usually
derived fi'om European experience, which are referred to as the “standard mortality
schedule. ” The result yields the adjusted death rates by the indirect methods.

(Spiegelman, 1955; Pressat, 1978).34

 

3' This requirement of full exposure becomes more limiting the higher the age segment of interest. For
example, to calculate under-five mortality rates, only information on children born at least five years before
the survey can be used.

32 Please refer to data section for detailed discussions. Later in this paper, monthly hazard rate estimations
are performed to avoid the problem of full exposure to infant mortality risk.

33 The number of deaths between ages x and x+n among residents in a community during a year divided by

the average number of persons between ages x and x+n living in that community during the year, multipled
by 1,000. (Spiegehnan, 1955)

21

Under indirect methods, mortality rates are calculated using only the number of
children ever born, the number of living children to women, and ages of women. The
indirect methods can utilize data that are commonly collected in censuses and many

35 Although these indirect methods can utilize data that do not contain

general surveys.
detailed information on date of birth of children and the dates of death or ages at death of
deceased children, these methods are based on an implicit assumption that the births of a
cohort of a women represents all children born in a time period.36 Another problem with
indirect methods is that the indirect methods estimate the probability of dying based on
experience that can extend over many years, resulting in an average over that period. The
methods are subject to error when there are changes in fertility and mortality trends, which
occured during 19605 -199OS in Indonesia.

Note that both direct and indirect methods could suffer from reporting error due to
the omission of deceased children. Estimation of infant mortality using direct methods
also depends on the correct reporting of age at death as under or over one year. The
heaping of deaths at age 12 months is also common, and to the extent that it causes a

transfer of deaths across the one-year boundary, infant mortality rates may be somewhat

underestimated.

 

35 “Because many late fetal deaths and neonatal deaths may be attributed to the same underlying conditions,
it has been proposed to combine the two to compute a “perinatal rate,” where the deaths are divided by
either live births alone, or the sum of live births and fetal deaths. Since there is no generally accepted
convention for computing this rate, the terms entering into it should be defined wherever it is used.”
(Spiegelnran, 1955). One can conclude that although the indirect methods can be used to extract other
mortality rates from census data, perinatal rates cannot be computed because census data give only birth
history, where only live births are recorded.

3° Documentation by the Demographic Health Survey reports that recent and on-going work shows that this

assumption may not be valid: births to women 20 to 24 (and in some cases to women 25 to 29) have more
elements of high risk of mortality than do all children born within the last five years of a survey.

22

Data

This paper uses data from the Indonesia Family Life Survey (IFLS). IFLS is a
continuing longitudinal socioeconomic and health survey that includes more than 30,000
individuals living in 7,200 households. The sample covers 321 communities in 13
provinces in Indonesia and represents about 83 percent of the Indonesia population in
199337. The first wave of IFLS was fielded in 1993 (IFLSl). The same households were
revisited in 1997 (IFLS2) and again in 2000 (IFLS3). This paper the data from these
three IFLS waves. A 25 percent sub-sample of households was re-interviewed in 1998
(IFLS2+), but the data are not used in this paper.

In IFLS surveys, special attention is paid to the measurement of health, work,
migration, marriage, child bearing, life history data on education, and economic status of
individuals and households. In each wave of IFLS, the individual and household surveys
are complemented by an extremely comprehensive community and facility survey. There
is also considerable attention placed on minimizing sample attrition in IFLS. Targeted
households and individuals who “split-off” from original households were followed if
they moved to new locations within 13 provinces of the survey areas. In each re-survey,
about 95 percent of targeted households have been re-contacted. The split-off households
added just under 1,000 households to the sample in 1997 and about 2,600 households in
2000.

The survey periods of each IFLS wave is shown in Figure 1. Data on

pregnancies, birth outcomes, infant mortality, and age of death are obtained from

 

37 Frankenberg, E., Hamilton, P., Polich, S., Suriastini, W., and D. Thomas. User’s Guide for the Indonesia
Family Life Survey. DRU2238/2-NIA-NICHD. March 2000.

23

retrospective data on pregnancy histories. Complete pregnancy histories were given by
women in the household who were between 15 and 49 years of age. These women were
asked about information of each pregnancy in detail. This information includes pre-natal
care, pregnancy outcome, birth information, post-natal care, and survival status of the
child. For prenatal-care, complete information on fiequency and type of pre-natal care of
each trimester of pregnancy was obtained. For birth information, the women were asked
detailed information about their pregnancy outcomes. In the case where there was a
miscarriage, the length of time before the pregnancy ended was reported. If the
pregnancy resulted in a live birth, information was obtained retrospectively on length of
pregnancy, birth date, place of birth, healthcare provider at birth, whether the child was
weighed at birth, and birthweight (if the child was weighed). For the child’s survival
status, mothers were asked if their children were alive at the interview date. If the child
died before the interview date, a complete history of the child’s death was obtained. This
information includes how old the child was when the child died (in days, weeks, months,
or years).

Data are obtained from IF LSl, IFLS2, and IFLS3, but are restricted to include
only children that were born between 1988 and 2000. This is because we want to restrict
the recall to five years for each wave in order to minimize recall error. This is similar to
strategies used by demographers.38

As IF L82 and IFLS3 follow the respondents of the original IF LSl, birth data are

organized as follows:

 

38 For instance, such as in the Demographic Health Surveys report infant and child mortality based on
events in the previous four or five years before the survey.

24

If the respondent (mother) was previously interviewed and reported some children
in earlier IF LS wave(s), the mother reported only children born since the last reported
child in the new IFLS wave.

If the respondent was interviewed in the previous wave, but reported no children
then, the mother reported all children born after the last interview in the new interview,
which amounts to a complete history

If the respondent was a new respondent, either because she turned 15 or older in
the new survey or because she was a new household member, a complete history was
taken.

As a result, birth data of children born in any particular year could be from any of
the IFLS waves. Information on children born during the financial crisis period (1998-
1999) can be obtained only from the IFLS3 wave, except for a few children born in
January 1998 when IF LS2 was still taking place.

In addition, observations across IF LS waves for panel respondents with preprinted
roster are carefully compared to avoid duplication, which occurred occasionally.
According to discussions of the quality of retrospective data on longitudinal surveys,39
retrospective data are of better quality if the length between the real event and the
interview date is minimized. Therefore, in the case of duplication, only the observation
from the earlier IF LS wave is included. An exception is made for 1997 and 1998 births.

Since these children were not at least one year old by IFLS2 interview date, we cannot

 

39 Beckett, M., J. DaVanzo, N. Sasu'y; C.Panis, and C. Peterson. “The quality of retrospective data: An
examination of long-term recall in a developing country.” Journal of Human Resources. Summer 2001.

This paper on reporting error provides insights into the quality of retrospective reports, particularly as it
pertains to short-term recall. Studies were reviewed which analyzed the quality of retrospective reports in

25

extract uncensored mortality information from observations that were reported in
IFLS2."0 In this case, data from IFLS3 are used where duplication occurs. Refer to
Appendix II for details.

To identify neonatal mortality and post-neonatal mortality, only live births were
included. As previously discussed, this paper uses a “true-cohort-life-table approach” to
calculate neonatal and post-neonatal mortality rates. This approach requires that the
children in our sample had full exposure to mortality risks, which, in our case, means
reaching of age.“ Instead of dropping observations that did not have one-year full
exposure to mortality risks by the interview date in IF L81 (1993) and IF L82 (1997), an
effort was made to recover these observations. For those who were younger than one-
year of age by the interview date in these earlier IFLS waves, information on their
survival is obtained from the household roster of the subsequent IF LS using identification
numbers that are consistent throughout all waves of IFLS."2 This household roster
provides information on whether the child was still living and the time of death if the
child had died. In the case where we could not track a particular child from IFLSl in
IFLSZ, we also looked for him/her in IF LS3. With a high re-survey rate of about 95
percent, we were able to recover most of the observations. As a result, we are able to add

487 children born in 1993/ 1992 and 511 children born in 1996/1997 to our sample.

 

the Malaysian Family Life Surveys (MFLS), fielded in Peninsular Malaysia in 1976 and 1988, and
conclude that many of the data quality problems found previously are present in the MFLS.

’0 Some respondents of the IFLSl997 were interviewed in January and February 1998.
‘" This full-exposure restriction is not our concern when studying monthly hazard rates later in the paper.

’2 When the child did not live in the same household in subsequent IFLS survey, survival and date of death
(if died) information was obtained from the new household.

26

In addition to data on pregnancy and survival histories, data from individual,
household, and community characteristics surveys are used to allow this study to control
for other socioeconomic characteristics such as child’s gender, mother’s education,
urban/rural areas of residence, and province of residence.

The residence of mothers we use in this paper is the residence of mother at the
time of the survey. We are aware of a general practice of taking the residence of females
at the time when they were 12 years old or residence before the first marriage in feritity
and marriage studies, but in this study of the effects on mortality, the residence of
mothers after their marriages are thought to affect healthcare provided to their children
more than pre-marital residence in terms of type and quality of the healthcare. We take
the residence of the mother at the time of the survey to be a proxy for the post-marital
residence.

The data are linked using individual and household identifications, which are

consistent throughout different surveys of each IFLS and different waves of the IF LS.

27

Method of Analysis

Testing Different Measures of Mortality Rates

The first goal of this paper is to compare mortality rates from IF LS data to the 1997
Indonesian Demographic Household Survey (DHS) data. We use a “true-cohort life table”
approach and compared the rates to mortality rates published in the Indonesian DHS 1997
publication, that uses “a synthetic cohort life table” approach. We then compare mortality
rates using DHS and IF LS data, but using the “true-cohort-live table” approach with both
datasets.

Table la, taken from the Indonesian DHS publication, shows neonatal and post-
neonatal mortality rates that are calculated using the “synthetic cohort life table” approach.
From the same dataset, table 1b shows computed neonatal mortality and post-neonatal
mortality rates from the 1997 DHS using the “true-cohort-life-table”. Data are divided
into three different periods by birth date (1982-1987, 1987-1992, 1992-1997) to match
DHS publication numbers. Mortality rates are calculated with and without individual
sampling weights provided in DHS.

Table 1b shows that across different time periods, weighted and unweighted
mortality rates are similar, and hence suggesting that the DHS observations give a good
representation of the Indonesian population. When comparing results from different
estimation approaches, weighted mortality rates are used since the DHS’s rates are
weighted. The comparison results show that different estimation approaches yield similar
rates in all three periods for both neonatal and post-neonatal mortality rates. Regardless of
methods used, we can conclude that Indonesia experienced a decrease in both neonatal and

post-neonatal mortality rates during the 10 years period prior to the economic and

28

drought/smoke haze crises. The decline is not as sharp for neonatal mortality as it is for
post-neonatal mortality.

When comparing the mortality rates of full DHS samples and the mortality rates of
DHS observations that are only from the 13 provinces surveyed in the IF LS, both neonatal
and post-neonatal mortality rates drop just by a small magnitude for both weighted and
unweighted rates. The mortality rates are smaller in the 13 IFLS provinces across all
periods. This corresponds to the provinces not surveyed being from the poorer, eastern
provinces.

Table 1c presents neonatal and post-neonatal mortality rates fi'om IFLS data using
the “true-cohort-life—table” approach. Mortality rates are calculated for the corresponding
DHS periods. When using both IF LSl and IF L82, IF LS data exhibits higher post-neonatal
mortality rates than DHS rates for 1982-1987 and 1987-1992. The IFLS post-neonatal
mortality rate is, however, lower for 1992-1997. The differences in the neonatal mortality
rates between the two datasets are smaller than the differences in the post-neonatal rates for
1982-1987 and 1987-1992, but larger for 1992-1997.

When comparing mortality rates using only IF LS for consistency check, we find
that including data from a more recent wave of IF LS gives higher rates for both neonatal
and post-neonatal mortality. The data suggest that when longer retrospective is used,

mortality rates are higher.

Birthweight Reporting

When studying the effect of the crises on a health outcome such as birthweight, a

crucial concern regarding the data is whether there is any selection problem in reported

29

outcomes. Children born at home are likely to be from rural, poorer households, and less
educated mothers. These children are likely to have lower birthweights, which are likely
to be unreported. If this selection problem is present in the birthweight distribution, then
reported numbers are biased. Table 2 shows that in Indonesia while almost all births that
took place in hospitals or community health center/delivery posts have reported
birthweight, only 51.6 percent of births that took place at home have reported
birthweight. This proportion of reported birthweight is close to the proportion of reported
birthweight from office or house of traditional midwives. Note that when a woman gave
birth at home or at family members’ house, often a traditional midwife was called in to
assist in the delivery process. Therefore, some proportion of babies that were delivered at
home could be weighed if the midwives were well trained and well-equipped with
measuring devices.

Table 3 shows proportion of birth location by different time periods. On average,
only approximately 15 percent of deliveries took place in public or private hospitals.
Health centers, village delivery posts, clinics of physician, and clinics of formally trained
midwives serve as major delivery facilities in Indonesia (26.7 percent). The proportion
of births that took place in these formal facilities increased during 1988-2000 from 21.0
percent to 26.7 percent. On average, birth deliveries that took place at home account for
more than half of all deliveries in Indonesia. Although the proportion of deliveries that
took place at home decreased substantially over time, the proportion remained relatively
high in 1998-2000 (49.3 percent). From these results, we expect to encounter problems
of reporting birthweight. This is due to a higher proportion of births that took place at

home. Unfortunately, we do not have any plausible instrument to correct for this

30

selection problem in reported birthweight. It is difficult to find a factor that affects
delivery location (and whether birthweight was reported), but does not affect birthweight
itself.

Mortality and Birthweight Statistics

The financial crisis periods are divided into two parts: crisis 1 and crisis 2 (refer
to Figure 1). Crisis] marks the period when the Rupiah rapidly devalued, the exchange
rates were extremely volatile, and food prices accelerated (January 1998 to September
1998). Crisis 2 marks the period when the exchange rates began to settle and food prices
came down. (October 1998 to June 1999). From earlier discussions, the 1997/1998
drought crisis and the smoke haze crisis covers the period from May 1997 until early
1998 in some areas. According to the timing of these different crises, the sample periods
are divided into four sub-periods: non-crisis (January 1988 to April 1997 and July 1999
to December 2000), crisis 1 (January 1998 to September 1998), crisis 2 (October 1998-
June 1999), and drought 97 (May 1997 to December 1997). By this period grouping,
crisisl also covers some of the drought/smoke haze crisis in some areas of Indonesia. 43

In addition, since crisis 1 and crisis 2 each lasted nine months, we can interpret
children who were born in crisis 2 as those who were conceived during crisis 1. Those
that were born during crisisl in rural areas that were affected by the drought could also be
roughly interpreted as those who were conceived during the drought period.

An overview of mortality rates during our periods of interests is shown in Figure 5.

Figure 5 shows mortality rates between 1988-1999. One can observe that the post-neonatal

 

’3 Although harvesting cycle in early 1998 was delayed by as much as two months, most of the effects of
the effects of the drought crisis on productions had already been felt in the second half of 1997.

31

mortality rate gradually declined until 1996. Then the rate went up in 1997, further up in
1998, and started to decline in 1999. Neonatal mortality rates were more volatile with a
downward trend until 1996. Neonatal mortality rate slightly increased in 1997, but the rate
was still lower than those of the year prior to 1996. The rates heightened in 1998, and,
similar to post-neonatal mortality rate, neonatal mortality rate started to decline in 1999.

The mortality status at the end of the first year of children born in crisis and non-
crisis periods is shown in Table 5. As we expect from Figure 5, mortality was at a
relatively high level up to 1992 and started to decline in 1993. The non-crisis periods are
divided into two periods: “1988-1992” and “1993-1997, 1999.”44

Table 5 shows that both neonatal mortality and post-neonatal mortality rates are
higher in rural areas than in urban areas in all periods. Infant mortality significantly
increased from the pre-crisis period (32.5 per 1,000 live births) during both the financial
crisis (46.2 per 1,000 live birth) and the drought/smoke haze crisis (76.1 per 1,000 live
births).

Neonatal mortality and post-neonatal mortality rates of children bom during the
financial crisis are higher than those of children born during non-crisis period of 1993-
1997 and 1999, but lower than those of the children born during 1988-1992. Overall, for
those born during the crisis, about 23 infants out of 1000 died within one month of birth
as compared to the pre-crisis rate of about 13 per 1000 live births. The percentage
increase of neonatal mortality is higher in urban areas (from10.2 to 20.5 per 1000 live
births) than in rural areas (from 15.7 to 25.4 per 1000 live births). For overall sample,

those born during the financial crisis also exhibit higher post-neonatal mortality than in

32

the pre-crisis period. However, the percentage increase in the post-neonatal mortality is
smaller than the percentage increase in neonatal mortality rate. When looking at rural and
urban samples separately, the data show that urban children experienced a large increase
in the post-neonatal mortality rate (fi'om 10.9 to 16.4 per 1000 live births) while rural
children experienced only a small increase in the rate (from 25.8 to 29.0 per 1000 live
births).

The drought/smoke haze crisis seems to have had strong negative impacts on both
neonatal mortality and post-neonatal mortality in rural areas where the largest negative
impacts are expected. Compared to the pre/post-crisis period of 1993-1997 and 1999,
Neonatal mortality increased from 15.7 to 29.0 (per 1,000 live births), while post-
neonatal mortality increased from 25.8 to 47.1 (per 1,000 live births). One can observe
that the post-neonatal mortality rate increased during the drought/smoke haze crisis so
much that it surpassed the post-neonatal mortality rate of 1988-1992 period (38.7 per
1,000 live births). This evidence suggests that the drought crisis had stronger adverse
effects on mortality than the financial crisis in rural areas. However, it is worthwhile to
recognize that the mortality rates shown in Table.4 are calculated fiom small samples,
especially the mortality rates for the crisis periods. As a result, these rates (per 1,000 live
births) are sensitive to the number of death incidences among these small samples.

Birthweight statistics are shown in Table 6. Overall, the financial crisis had a
small negative effect on birthweight. The proportion of low-birth-weight children
(children with birthweight less than 2.5 kilograms) increased from 8.1 percent during

non-crisis 1993-1997 to 8.7 percent during the financial crisis. The data also show some

 

’4 Births between July 1999 and December 2000 also belong to the non-crisis period, but they are not

33

reduction in the mean birthweight with the biggest decline among urban children (fi'om
3.17 to 3.11 kg.). The drought/smoke haze crisis appears to have a negligible adverse
effect on the mean birthweight, but the percentage of low-birth-weight children is lower
than that of the non-crisis periods in both urban and rural areas. Since we expect to
encounter selection problems in reported birthweight as discussed earlier, the estimated
effects of the crises on birthweight is subject to errors. For instance, if during a
drought/smoke crisis a high proportion of mothers shifted away from public healthcare to
in-home care or offices of traditional midwives for delivery, it is more likely that babies
were not weighed. Table 4 shows the proportions of delivery locations during pre-crisis,
crisis, and post-crisis periods. The data suggest that during both crises, there is a jump in
the proportion of deliveries that took place at home or at offices of traditional midwife
from a decreasing trend. Babies born during the crises are, therefore, less likely to be
weighed. If the unreported babies weighed less than the reported mean birthweight (e. g.
mothers who switched to deliver at home because of high costs of public healthcare were
the ones endowed with poor health), then the mean reported birthweight during the
drought/smoke crisis understimates the adverse crisis impact. On the other hand, if
unreported babies weighed more than the reported mean birthweight, (e. g. mothers who
were healthy decided to save some money by giving birth at home), then the mean
reported birthweight overestimates the adverse crisis impact. The direction of the bias is

certainly an empirical question.

 

included because those children were not yet exposed to one year mortality by the interview date.

34

Testing the Effect of Economic Crises on Birthweight and Infant

Mortality

This paper tests the hypothesis of whether crises in Indonesia have any effects on
child mortality and birthweight. Children born during the crisis may be affected
differently than those conceived during the crisis. Neonatal mortality is influenced by
conditions during pregnancy or at birth, while post-neonatal mortality is influenced more
by extemal factors during child rearing after birth“. Since conditions during pregnancy
and after births are likely to affect neonatal and post-neonatal mortality differently, the
analysis is carried out separately for neonatal and post-neonatal mortality.

Birthweight and child mortality may be affected by various socioeconomic
characteristics other than conditions affected by economic crises. In this paper, mother’s
education, urban/rural residence, geographic location (provinces and communities), and
child’s gender are used as control variables.

6 women’s education has external benefits to society.

According to Schultz,4
Higher mother’s education reduces child mortality, improves child nutrition and
schooling, and decreases fertility and population growth. Mother’s education benefits

child health in several ways. First, education helps make learning of the childcare

process more efficient, especially when the process is complex (technical efiiciency).

 

’5 A. Razzaque, N. Alam, L. Wai, and A. Foster. “ Sustained Effects of the 1974-5 Famine on Infant and
Child Mortality in a Rural Area of Bangladesh.” Population Studies. March 1990

M. Rahman, B. Wojtyniak, M. M. Rahaman and K.M.S. Aziz, “Impact of environmental sanitation and
crowding on infant mortality in rural Bangladesh.” The Lancet (1985), pp. 28-32

’6 T.P. Schultz. “Investments in the schooling and health of women and men: Quantities and returns.”
Journal of Human Resources. Fall 1993; Vol. 28, Issue. 4; pg. 694, 41.

35

Second, education helps mothers allocate household resources efficiently to improve
child’s health (allocative efiiciency). For example, mother’s education increases the
willingness to seek medical care and improves nutrition and sanitation practices.47 Third,
mothers with higher education are more likely to earn more income, and therefore can
use this additional income to consume more or better-quality child’s health inputs
(income effects). Fourth, higher education may help improve women’s bargaining power
in household resource allocation. According to Schultz (1993), women may channel
more of their income to expenditures on children than their husbands do. Improving
women’s education, therefore, could result in higher women’s bargaining power, which
in turn yields an allocation of more of the household income to expenditures on children.
When studying the effects of mother’s education on child mortality, one should
consider a possibility that education and health services are substitutes. According to a
review by Basu and Aaby (1998)48 that refers to studies by Palloni (1985)49 and
Rosenzweig and Schultz (1982),50 the dominant theoretical stance on the education-

mortality association is that the influence of personal characteristics, such as maternal

 

’7 Schultz, T. Paul. 1989. Benefits of Educating Women. Washington, DC: World Bank, Background
Papers Series, Education and Employment Division, Population and Human Resources Department.“The
Benefits of Education for Women HRO Dissemination Notes.” Human Resources Development and
Operations Policy. Number 2, March 8, 1993.

Mellington and Cameron (1999)_find that mother’s primary and secondary schooling significantly decrease

the probability of child death in both rural and urban areas in Indonesia. “Female Education and Child
Mortality in Indonesia.” Melboume- Department of Economics in its series papers number 693. 1999

‘8 Basu, AM. and Peter Aaby. The Methods and Uses of Antropological Demography. 1998. Clarendon
Press. Oxford.

‘9 Palloni, A. Health Conditions in latin America and policies for mortality change’, in J. Vallin and A.
Lopez (eds), Health Policy, Social Policy and Mortality Prospects. Liege: Ordina. 1985

36

education, on child welfare attenuates when good health services are widely available.
However, many studies give contrary examples in which access to services appears to
make little difference to education differentials (for more detailed discussions, see
Cleland and Van Ginneken, 198951). For example, Bicego and Boenna (1991) find a
stronger effect of maternal education in urban areas, where health services are assumed to
be widely available, than in rural areas.

Table 7 shows the distribution of the mean education in years and percentages of
different education levels of mothers over time. Formal education in Indonesia had been
successfully improved. From our sample, we observe a significant increase in the mean
education of mothers from 5.0 to 8.2 years during the eleven-year period. For example,
the proportion of mothers with no formal education decreased from 19.2 percent to 4.6
percent (76 percent decrease). The proportion of mothers with 1-5 years of schooling
decreased from 33.9 percent to 14.8 percent (56 percent decrease), while the proportion
of mothers with 9-11 years of schooling increased from 7.9 percent to 20.4 percent (158
percent increase).

Table 8 shows that higher mother’s education is associated with lower mortality.
The effect seems to be stronger for post-neonates than neonates. If mother’s education
contributes to child mortality as previously argued, we must control for mother’s

education when studying the effects of crises. Since the children born in the crises

 

soRosenzweig, M and T.P. Shultz. “Child mortality and fetility in Colombia: individual and community
effects.” Health Policy and Education, 2: 305. 1982.

5‘ Cleland, J. and J. Van Ginneken. 1989. “Maternal education and child survival in developing countries:
the search for pathways of influences.” Social Science and Medicine. 27: 1357-60.

37

periods are from mothers of later cohorts for whom education is higher, not controlling
for mother’s education, will underestimate the impact of the crises.

The probability of neonatal and post-neonatal mortality is estimated using linear
probability and logit regressions. Dependent variables are whether a child born alive had
neonatal mortality, whether the child had postneonatal mortality given he/she survived
neonatal mortality52 and whether a child born alive had infant mortality (either neonatal
mortality or post-neonatal mortality). Province or communities dummies are used to
control for time-invariant community-specific unobserved factors that may influence
child mortality (such as local disease patterns and public health infrastructure”). Since
the focus of this paper is to study the effects of the financial and the drought/smoke crises
on child birhtweight and mortality, controls thought to be correlated with the crises such
as household income and time-varying public health provision are excluded.

In addition to socioeconomic controls, a time trend is included in our analysis to
control for mortality trends that can be observed in Figure 5. According to the data, both
neonatal and post-neonatal mortality increased during the crises period, but the level of
these mortality rates are still not as high as that of the earlier control period (1988-1992).
Since neonatal and post-neonatal mortality had generally declined during the pre-crisis
periods from a relatively higher rate in 1988, omitting the time trend would result in an

underestimated effect of the crises.

 

52 Therefore, those born alive that experienced neonatal mortality are dropped from our sample when
studying post-neonatal mortality.

53 Pertersen, W. and R. Petersen with the collaboration of an International Panel of Demographers. 1986.

Dictionary of Demography: Terms, Concepts, and Institution. Greenwood Press. New York. Westport,
Connecticut. London.

38

As discussed earlier, the sample used includes children who were at least one year
old at the interview date. Both results fi'om the linear probability and the logistic
regression are reported. Since urban and rural areas may be affected by different crises in
different ways, regressions are performed separately for urban and rural areas.

Due to the nature of the dependent variables, the linear probability model violates
one of the Gauss-Markov assumptions. When the dependent variable is a binary variable,
its variance, conditional on the explanatory variables, depends on the explanatory
variables (unless the probability of success does not depend on any of the explanatory
variables):

Var (le) = 1300 [1-p(X)]

Where, p(x) is the probability of success, which depends on x.

As a result, there must be heteroskedasticy in a linear probability model.
Although heteroskedasticity does not cause any bias in the OLS estimators of the
coefficients, homoskedasticity is crucial for justifying the usual t and F statistics.54
Regressions robust to heteroskedasticity are, therefore, included to correct the conditional
variances.

In addition, since our period of analysis includes children born in 1988-2000,
children in our sample can be from the same mother. The outcomes of children within a
group of mother are likely to be correlated. In this paper, observations are clustered at
the mother level to allow for this type of heteroskedasticity problem in the variances of

the estimated coefficients.

 

5’ For further discussion, refer to Jeffrey M. Wooldrige. “Introductory Econometrics: A Modern
Approach” Copyright 2000. by Sourth-Westem College Publishing. USA.

39

The effects of economic crises on birthweight can be tested nonparametrically by
comparing cumulative distribution of birthweight of children born during non-crisis
period and that of children born during the crisis period. Special attention, however,
should be paid to proportion of children who have “low-birth-weight,” which is
birthweight that is less than 2.5 kilograms. Note that when testing the differences of the
distributions, data are restricted to exclude outliers. The tests include only birthweights
between 1.555 and 3.0 kilograms. Computing for the difference in the values of the
cumulative distribution at each weight point of interest was performed. The comparison
follows directly the formulation offered by Davidson and Duclos (2000).56 We will be
able to conclude that the crisis has statistically significant adverse effects on birthweight
when the cumulative distribution value of birthweight of those born during the crisis
period at all birthweights of interests (that are considered low birthweight) is
unambiguously higher than that of those born during non-crisis period. In other words,
we are testing whether the cumulative distribution of low birthweights of those born
during the crisis period first stochastically dominates that of the cumulative distribution
of low birthweights of those born during the non-crisis period.

When the test of first order-stochastic dominance fails, we test for relative
riskiness or dispersion of birthweight (second-order stochastic dominance). This is to test
whether those born during the crisis period as a group exhibit relatively higher risk of low
birthweight than those born during non-crisis period. In other words, we test whether the

probability of those born during the crisis period that have birthweight at or below a

 

55 There are few birthweights that are less than 1.5 kilograms. These observations are considered outliers.

4O

specific weight is significantly higher than that of those born during the non—crisis period.
If this is true for all low birthwieght values tested, we can conclude that the distribution
of birthweight of those born during the crisis period second-order dominates that of those
born during the crisis, and that those born during the crisis exhibit relatively higher risk
of having low birthweight. Computing for the difference in the values of the cumulative
distribution at each weight point of interest was performed using a software for
Distributative Analysis/Analyse Distributative (DAD).57 The techniques used in this
software follow directly the formulation offered by Davidson and Duclos (2000).58

In addition to nonparametric estimation of the effects of the crises on birthweight,
we include a simple parametric estimation using Ordinary Least Square (OLS)
regressions, using birthweight as a continuous variable. Explanatory variables are the
same set used in mortality regressions. Separate regressions are carried out for rural and

urban samples.

 

56 Russel Davidson and Jean-Yves Duclos. “Statistical Inference for Stochastic Dominance and for the
Measurement of Poverty and Inequality.” Econometrica v86 n6. 2000.

57 Copyright by Jean-Yves Duclos, Abdelkrim Araar, and Carl Fortin.

58 Davidson, R. and Jean-Yves Duclos. “Statistical Inference for Stochastic Dominance for the
Measurement of Poverty and Inequality.” Econometrica, V86. n6. 2000.

41

Results

Figure 6a-6c present the comparisons of the cumulative distributions of
birthweights in the crisis and the non-crisis periods. The results confirm that the
reduction in birthweight was only among urban children born in the economic crisis.
Table 9 shows the results of testing the difference of cumulative distributions using both
first and second order stochastic dominance tests. The points of testing are between 1.5
and 3.0 Kilograms. The results from these tests show that, within this range, none of the
comparisons show any significant first and second-order stochastic dominance.
However, it is hard to reject that the economic crisis had no impact on urban children as
one can observe that there is some evidence suggesting a second order stochastic
dominance in the range of 2.1-3.0 Kilograms among urban sample. In rural areas, the
results are opposite to what we expect, but none of the comparisons shows first nor
second-order dominance.

Table 10 presents the results from the multivariate OLS birthweight regressions.
The dependent variable in these regressions is birthweight in kilograms. The estimations
are carried out for urban and rural samples separately. Four specifications are presented
for each sample. The first specification is our base regression. In this specification,
explanatory variables include only the crisis dummies and the time trend. Mother’s
education dummies are added in the second specification. The last two specifications use
province and community dummies to controls for location-specific unobserved factors
respectively. Regardless of which geographic location is used, the regressions essentially

yield fixed-effect estimators.

42

The results from these regressions suggest that none of the crises had a
statistically significant adverse effects on birthweights. We find no effects even after
controlling for other factors that may affect birthweight. We observe a downward trend
in birthweight in urban areas and an upward trend in rural areas, but the time trend
variable does not appear to have statistically significant effects on birthweight. Male
children had significantly higher birthweight than female children in both urban and rural
areas. In any case, one should keep in mind that these regression results are based on
birthweights that were reported. The observations are subjects to potential selection
problems as discussed earlier.

The results from the multivariate analysis of infant mortality are shown in Tables
11 to 14. Tables 11 and 12 present results from the LPM and the Logistic regressions
using the urban sample. Tables 13 and 14 present results from the LPM and the Logit
regressions using the rural sample.59 In each set of regressions, four specifications are
presented. The first specification is our base regression. In this specification,
explanatory variables include only the crisis dummies and the time trend. Mother’s
education dummies are added in the second specification. The last two specifications use
province and community dummies to controls for location-specific unobserved factors.
Regardless of which measure of geographic location is used, the regressions essentially
yield fixed-effect estimators. In the Logit regressions, the last columns in each group of
regressions are estimates of conditional Logit models with fixed province/community

effects. To be able to identify the conditional effects, observations used in these

 

5’ None of those born during financial crisis 2 period (122 obs) experienced neonatal mortality. Since we
are interested in the estimated coefficients of the crisis dummies (fiancial and drought/smoke crises), the

43

regressions are only from those provinces/communities that experienced at least one
incidence of mortality.60 To be consistent with the Logit regressions in terms of
observations used, the last two columns of the LPM regressions limit the samples to be
the same as those in the Logit regressions. Also, when community dummies are
included, crisis dummies pick up effects that vary within communities. Then, if there is
no variation in the communities, observations belonged to these communities do not help
identifying the crisis effects.

The estimated coefficients from the linear probability models are the estimated
partial effects. For Logit regressions, the odds ratios61 are reported. As discussed earlier,
explanatory variables of interest in the neonatal mortality regressions are whether the
child was conceived during the crises, while explanatory variables of interest for post-
neonatal and infant mortality regressions are whether the child was born during the crises.
The LPM and the Logit regressions give similar results.

In urban areas, the estimated coefficients of the time trend are statistically
significant in neonatal, post-neonatal, and infant mortality regressions regardless of
whether the regressions control for other factors The estimates suggest that both
neonatal and post-neonatal mortality rates in urban areas decreased by approximately 0.2

percentage points per years (by 2 deaths per 1,000 live births per year). Infant mortality

 

“pregnancy in fiancial crisisZ” dummy variable was dropped in the Logit regressions instead of dropping
these observations. The dummy variable was also dropped in the corresponding LPM regressions.

6° The data show that none of the infants born in urban areas of Lurrrpung and Bali provinces experienced
post-neonatal mortality. Similarly, none of the infants born in rural Yogyakarta experienced post-neonatal
mortality.

6' The odd ratios, e8, show impacts in terms of Prob (mortality | x ) / Prob (survival | x).

44

declined by approximately 4 deaths per 1,000 live births per year. The estimated effects
of the time trend are significantly larger in the community fixed-effects regressions.62

In urban areas, children born during the peak of the financial crisis (crisisl)
exhibited higher odds of neonatal, post-neonatal, and infant mortality. The effects of the
crisis were more precisely estimated in the Logit regressions. Controlling for mother’s
education and province of residence does not change our point estimates by much. when
controlling for the community fixed effects, these point estimates, however, decrease in
the Logit regression and increased in the LPM regressions .

Results from the linear probability regressions show that children conceived
during the peak of the financial crisis (crisisl) had approximately 1.7 percentage points
higher probability of neonatal mortality than those conceived during the non-crisis
periods. Similarly, those born during crisisl exhibited approximately 1.6-1.7 percentage
points higher probability of post-neonatal mortality than those born during the non-crisis
periods. However, the impacts of the crisis are not statistically significant when
community fixed-effects are controlled for.

Urban children conceived during the drought/smoke haze crises exhibited higher
neonatal mortality than those conceived during the non-crisis periods. The adverse effects
are estimated to be about 2.1 percentage points in the LPM regressions. The adverse
effects were statistically significant at 10 percent level even after controlling for the
community fixed effects using Logit regression. The 1997 drought/smoke crisis did not

have any statistically significant effects on post-neonatal mortality in urban areas.

 

62 Recall that observations used in the LPM community fixed-effect regressions and the conditional Logit
regressions includes only communities with at least one incidence of mortality.

45

The results from the infant mortality regressions suggest that only the economic
crisis had a statistically significant impact on overall infant mortality in urban areas.
Similar to results from the neonatal mortality regressions, the effects of the crisis were
not statistically significant in the fixed-effect estimation for infant mortality.

In urban areas, male children had higher neonatal, post-neonatal, and infant
mortality rates. However, the estimates are not statistically significant. In addition, we
found that mother’s education did not play a significant role in reducing neonatal
mortality.63 On the other hand, the effects were felt in post-neonatal mortality. Although
in some specifications the points estimates are not statistically precise, we do observe that
children born to mothers with at least 9 years of education experienced lower probability
of post-neonatal mortality by approximately 2.8-3.5 percentage points. When controlling
for the community fixed effects, the magnitude of the effects of twelve or more years of
education more than doubled the effects in province fixed-effect regressions.64

In rural areas, both neonatal and post-neonatal mortality rates decreased over
time. The estimated coefficients of the time trend, however, suggest that rural areas
experienced a decline in these mortality rates at a slightly slower rate than the urban
areas. The community fixed-effect estimations suggest that the reduction in mortality

risks was at a faster rate in communities that experienced at least one mortality incidence,

controlling for community fixed effects.

 

63 Except at 12 or more years of education in the Logit regression.

6’ Recall that observations used in the LPM community fixed-effect regressions and the conditional Logit
regressions include only communities with at least one incidence of mortality.

46

In rural areas, the financial crisis 1 period had adverse effects on neonatal and
post-neonatal mortality. However, the estimated effects are statistically significant in
only neonatal mortality regressions. In these regressions, the estimated effects are
statistically significant at the 10-percent level even after controlling for community fixed-
effects. Rural infant mortality was affected by financial crisis 2 instead of financial crisis
1. The increase in infant mortality rate is estimated to be approximately 3.2 percentage
points. In those communities where we applied community fixed-effect estimation, the
effects are higher (4.9 percentage points) and statistically significant at the 5-percent
level.

In rural areas, the drought/smoke haze crisis exhibited statistically significant
adverse effects on post-neonatal mortality, but not on neonatal mortality. The effects of
the crisis on post-neonatal mortality are stronger than the effects of the financial crisis.
The LPM estimations show that infants born during the drought/smoke haze crisis had
approximately 2.7-3.1 percentage points higher probability of post-neonatal mortality
than those born during the non-crisis periods. The estimated effects are higher (6.3
percentage points in the LPM) and still statistically significant when controlling for
community fixed-effects in both the LPM and the Logit regressions.

Mother’s education played a greater role in reducing neonatal mortality in rural
areas than in urban areas. Our estimates from the LPM regressions suggest that mother’s
primary education is associated with approximately 1.6-1.8 percentage points lower
neonatal mortality rate. The effect of education was much stronger (4 percentage points)
using community fixed-effect estimation for those communities that experienced neonatal

mortality. If a higher level of mother’s education increases effectiveness of prenatal care,

47

these results imply that increasing mother’s education in rural areas will help reduce
neonatal mortality. An explanation of our finding that higher mother’s education did not
reduce neonatal mortality in urban areas is that urban mothers’ ability and effectiveness
in providing prenatal care could be substituted by better services and less expensive
healthcare that were more readily available in urban areas.

Similar to urban areas, education of rural mothers reduced post-neonatal
mortality. Moreover, our estimates suggest that in rural areas mother’s education started
to have statistically significant effects on post-neonatal mortality at the secondary
education level, much earlier than in urban areas.

In rural areas, we found that male infants had a higher chance of both neonatal
and post-neonatal mortality than female infants. Recall that in urban areas, a child’s
gender had no statistically significant effects on neonatal mortality. If male infants are
biologically more prone to neonatal mortality than female infants, the result from our
regressions suggest that better healthcare and more exposure to prenatal services (such as
those available in urban areas) can help overcome higher risks of neonatal mortality

among male infants.

48

How did mothers with different education levels cope with the crises?

An important question we might ask when studying the effects of the crises on
infant mortality is how households with different socio-economic backgrounds coped
with adverse short-term shocks. For instance, we ask whether poor households
responded differently to the financial crisis than rich households did. Nevertheless, since
the financial and the drought/smoke haze crises had direct impacts on household’s
income, distinguishing households by income level may result in biased estimates of the
effects of the crises as the crisis itself determines which income group a household
belonged to. To avoid this selection problem, we want to use determinant of household’
income that are not affected by the crises (in the short-run). In this paper, we experiment
with different mother’s education levels as such determinants.65

Tables 15-17 show LPM mortality regression results of children born to mothers
with 0-8 years and 9+ years of education. Table 15 shows the results from infant
mortality regressions of infants born to mothers with different levels of education. We
observe from the point estimates that in urban areas, infant mortality trends are different
between the two groups, even after controlling for community fixed effects. Those born
to mothers with lower education levels experienced a decline in infant mortality at a rate
that is twice as much as the rate for those born to mothers with higher levels (0.6 vs 0.3
percentage points per year). However, a test of the differential trend effects between

these two groups in urban areas indicates that the differential effects are not statistically

 

65 Alternatively, one can ask directly how mothers of different education levels cope with the crises.

49

significant at the 10 percent level."6 In contrast, the difference in the time trend effects
between low and high education groups is small in rural areas. The estimated effect of
the time trends for both groups is approximately 0.4 percentage points per year.

In urban areas, the point estimates of the financial crisis effects on infant mortality
suggest that the adverse effects are stronger for infants born to mothers with lower
education. However, a test of the differential effects cannot reject that the financial crisis

67 In rural areas, the effects of the

effects on the two groups are statistically the same.
financial crisis on infant mortality were slightly higher for the lower education group. A
test of the differential effects indicates no statistically significant difference between the
effects on the two groups."8

In rural areas, the drought/smoke crisis had statistically significant adverse
effects on infant mortality only for infants born to the low education group. The
magnitude of the drought/smoke effects are also much larger for those belonging to
mothers with lower education than for those belonging to mothers with higher education
(6.0 versus 1.3 percentage points). The differential effects are larger when controlling for
community fixed effects. However, the effects of the droughts/smoke crisis are not
statistically different between these two education groups.69

Table 16 shows results from neonatal mortality regressions of infants born to

mothers with low and high education. We found that in urban areas the mortality trends

 

66 The t-statistic for the differential effects is 1.50 (p = 0.13).

67 The t-statisitics for the differential effects are 1.43 (p = 0.15) for financial crisis 1 and 0.03 (p =0.98) for
financial crisis 2.

68 The t-statisitics for the differential effects are 0.36 (p =0.72) for financial crisis 1 and 0.08 (p =0.94) for
financial crisis 2.

50

of the low and the high education groups are similar. Mothers with higher education
levels experience a decline in neonatal mortality of their infants at a rate of approximately
0.28 percentage points per year, slightly higher than that of mothers with lower than six
years of education. We observe a larger differential between the two groups after
controlling for province and community fixed effects. However, the differential effects
of the time trends are still statistically insignificant.70 In rural areas, neonatal mortality
had not been statistically reduced over time regardless of the education of the mother.
When community fixed effects are controlled for, the differential effects of the time trend
between the two groups is larger, but the point estimates are still statistically
insignificant. A test of the differential effects cannot reject that the neonatal mortality
trends among the two education groups are similar.71 In addition, we observe higher and
more significant time-trend effects in urban areas than in rural areas for both education
groups.

In urban areas, financial crisis 1 had statistically significant adverse effects on
only infants from the high education group. Our point estimates suggest that infants born
to mothers with lower education were more adversely affected by the drought/smoke
crisis in their risks of neonatal mortality than infants born to mothers with low education.
However, the differential effects are not statistically significant.72 In rural areas, infants
born to mothers with higher education during the second phase of the financial crisis and

the drought/smoke crisis, however, had a lower neonatal mortality rate than those born

 

69 The t-statisitic for the differential effects is 1.50 (p = 0.13).
7" The t-statistic for the differential effects is 0.61 (p =o.54).

7‘ The t-statistic for the differential effects is 0.42 (p = 0.67).

51

during the non-crisis periods. A plausible explanation is that even though the neonatal
mortality rate for this group of infants increased during these crisis periods, the increase
in mortality risk is not so high to surpass the rates of the pre-crisis period. This
explanation is supported by the fact that the negative time trends were not statistically
significant for this group of infants. When the community-fixed effects are controlled for
(for those with 9+ years of education), the crisis dummies still have negative signs while
the estimated effect of the time trend is negative and significant. This result is puzzling.
However, our test of the differential effects of the crises suggests that there is no
statistically significant differential effects between the two education groups.73

The neonatal mortality regressions give another interesting result when we look at
infants of different genders. In urban areas, we observe a small and statistically
insignificant gender difference in the mortality rate regardless of levels of mother’s
education. In rural areas, among those born to mothers with lower education, male
infants had 1.4 percentage points higher neonatal mortality risks than female infants. The
differential effect is higher after controlling for community fixed effect. In contrast, male
infants had less neonatal mortality risk that female infants among those born to mothers
with higher education. Our test indicates that the gender differential in neonatal
mortality risk among infants born to mothers with low education is significantly different

from that among infants born to mothers with high education."

 

72 The t-statistics for the differential effects are 0.62 (p = 0.54).

73 The t-statistic for the differential effects is 0.04 (p = 0.97), 0.83 (p = 0.41), and l.00(p = 0.32 ) for
financial crisis 1, financial crisis 2, and drought/smoke crisis.

7’ The t-statistic for the differential effects is 1.79 (p = 0.07).

52

Table 17 shows results from post-neonatal mortality regressions of infants born to
mothers with low and high education. We found that in urban areas, the mortality trends
are significantly different among the low and the high education groups at the 10%
level.75 In urban areas, infants born to mothers with less than nine years of education
exhibited a much larger declining trend than those born to mothers with higher education.
The estimated effects of the time trend are estimated to be approximately 0.4-1.2
percentage points per year for lower education group. In contrast to the trend difference
in urban areas, the difference between the effects of the time trends between these two
education groups are less pronounced in rural areas. Rural infants born to mothers with
lower education experienced a faster decline in post-neonatal mortality risk. The
estimated effects of the time trend are lower in rural areas than in urban areas (-0.24
versus -0.33 percentage points per year) for lower education group. On the contrary, the
time trend effects are higher in rural areas than in urban areas for the higher education
group (-0.13 versus -0.04 percentage points per year).

In urban areas, the adverse effects of the financial and the drought/smoke crises
on post-neonatal mortality seemed larger for those born to mothers with lower level of
education than for those born to mothers with high education. However, none of the
estimated coefficients of the crisis dummies are statistically significant in either high or
low education groups. The differential effects of the crises between the two samples, in

turn, are not statistically significant.

 

75 The t-statistic for the differential trend effects is 1.67 (p = 0.09).

53

In rural areas, unlike in urban areas, the adverse effects of the financial crisis
appear to be similar for both education groups. As expected, the drought/smoke haze
crisis adversely affected children born to both groups of mothers. However, the
drought/smoke crisis had much worse effects on post-neonatal mortality for infants in
low education groups than that of infants in high education group. Our point estimates
suggest that infants belonging to mothers with less than 9 years of education tended to
suffer from the drought/smoke haze crisis much more than those belonging to mothers
with at least 9 years of education (4.3 versus 0.0 percentage points). A test of the
differential effects indicates statistically significant differential effects between the two
education groups.76

In urban areas, our estimates show discrimination in post-neonatal mortality
between male and female infants born to mothers with at least nine years of education.
The discriminating effect is large when controlling for community fixed effects (4.7
percentage points). A test of the differences in gender discrimination suggests no

7 In rural areas, the gender

differential effects between the two education groups.7
difference in post-neonatal mortality risks is statistically significant for only the low
education group. Male infants born to mothers with lower education experienced about

1.4 percentage points higher post-neonatal mortality rate than female infants. The

difference between the two education groups are, however, not statistically significant.78

 

’6 The t-statistic for the differential effects is 1.84 (p = 0.07).
’7 The t-statistic for the differential effects is 0.93 (p = 0.35).

78 The t-statistic for the differential effects is 0.71 (p = 0.48).

54

In sum, our findings indicate that infants born to mothers with different levels of
education exhibited no significantly different trends in mortality risks over time in both
urban and rural areas. We also find that even though the economic crises had adverse
effects on infant mortality, infants born to mothers with different education levels did not
experience statistically different adverse financial crisis effects in either urban or rural
areas. There were statistically significant differential effects of the drought/smoke crisis
on post-neonatal mortality between the low and the high education groups in rural areas.
Those born to mothers with lower education were more adversely affected by the

drought/smoke crisis than those born to mothers with higher education.

55

Hazard Models

The findings obtained in the previous section are based on sample observations
that include only mortality of children who were exposed to at least one year of life.
Many children born in 1999 and all children born in 2000 are excluded by this criterion.

Since we are interested in estimating the probability of dying before one month
and within 1-11 months, we can use duration models to directly extract the hazard rates
(the probability of dying in the next period given survival up to the current period)
without having to exclude those who were not yet one year old at the time of the
interview. Using the duration model is also a better way to capture the precise time of
death assuming reported dates are accurate. By using duration models, each month’s
survival status information of each child is used to estimate the mortality hazard. The
hazard rate of each month can then be obtained.

Figure 7a-7c show nonparametric estimates of discrete monthly child mortality
hazard rates from Nelson-Aalen cumulative hazard function. In urban areas, those born
during the financial crisis exhibited higher hazard rates that those born during the non-
crisis periods. Similar to results from the previous regressions, in rural areas, the drought
crisis had more overall effects on child mortality than the financial crisis. The results
from testing the equality of the survivor functions of children born in these periods,
however, show that none of crises had statistically significant adverse effects on infant
mortality. These insignificant crisis effects are inconsistent with our findings from the
OLS and logit regressions. Nevertheless, one should keep in mind that the hazard rates

used in this section to calculate the crisis effects are estimated without any controls.

56

Parametric Estimation of Child Mortality Using Hazard Models
Let T 2 0 denotes the length of time the child lived in months. T has some
distribution over the population. Consider time invariant covariates.

Let F(t; x) = conditional cdf of T where x = covariates:

F(t; x) = P(T S t; x), t 2 0

The survivor function is defined as

S(t;x) = 1 — F(t;x) = P(T > t;x)

Then, the probability of leaving the initial state in the time interval (t, t+h) is
P(t ST<t+h|T_>_t;x) forh>0

Define the hazard function as

= F(t +h,x)—F(t;x)
l—F(t;x)

 

l(t;x)=P(tST<t+h|T2t;x)=P(tST<t+h;x)/P(T.>_t;x)

If the cdf is differentiable, then the hazard function is

xi(t'x)=F(t+h’x)-F(t;x)= f(t;x) =f(t;x)
. 1—F(t;x) 1— F(t; x) 5033‘)

Then all probabilities can be computed using this hazard function. For example,

from time a to time b, a < b is

b
P(a S T < b | T 2 a;x)=1— exp[— Jl(s;x)ds]

57

Since we did not observe the end of the survival period of every child in the
sample, the survival data obtained from the survey are considered “flow” data, which are
subject to time censoring. In this case, the data are right-censored at the interview date.
For example, if the child born on January 1, 2000 was still alive on the interview date of
June 30, 2000. We only know that the child’s survival time was at least six months. We
never observed real survival time of this child. The model can be adjusted to include this
time censoring by defining censored flow data in the following way.

Observed duration t*

Define t,* as the length of time in the initial state that has a continuous conditional

density

for Ixi;6),t 20

where 0 is the vector of unknown parameters.

The observed length of time, t, in the initial state is
t, = min(t't,c,)

where c, is censoring time for individual i. In this case, the censoring time is age
of the child the interview date.

Table 18 and Table 19 show results of parametric estimations of hazard rates of
urban and rural children respectively. Hazard regressions use Weibull distribution to
allow for both positive and negative monthly hazard rates. The standard error and the 2
statistics of the hazard ratios are robust to heteroskedasticity of the variance-covariance

matrix at the mother level. In addition, we also present results of these hazard

58

regressions assuming heterogeneity in our observation. In our regressions, the
observations are assumed to have inverse-Gaussian heterogeneity (frailty).

The estimated results are similar to those obtained from previous LPM and logit
regressions. Those born during the financial and the drought/smoke haze crises appear to
have been negatively affected by the crises. The estimated effect of the financial crisis is
statistically significant at 10 percent level while the estimated effect of the drought crisis
is significant at 5 percent level. Those with mother of at least 12 years of education have
lowers estimated hazard rates than those with no education. The estimated coefficient of
the male dummy suggests boys had a higher mortality risk than girls.

In rural areas, those born during the drought/smoke haze crisis appear to have
been adversely affected. Similar to results found in the LPM and the Logit regressions,
the financial crisis did not have statistically significant effects on rural children.
Education was also more important in rural areas than in urban areas. In rural areas,
having some education, regardless of the education level, helped reduced the risk of
mortality. Similar to the results from urban areas, male children exhibited higher risk of
mortality.

The results from the hazard regression allowing for heterogeneity adjustment
confirms that the observations used are heterogeneous.79 However, the degree of the
statistically significance of the estimated coefficients are similar to those in the
regressions that assume homogeneity of the observations.

Figure 8a-8c show estimated child mortality hazard rates using parametric hazard

models. Results from hazard models confirm that the financial crisis and the drought

 

79 We reject that observation are homogeneous with p-value = 0.000 for both urban and rural observations.

59

crisis had adverse effects on neonatal and post-neonatal mortality. When comparing rural
and urban samples of the same period, rural children exhibited higher probabilities of
both neonatal mortality and post-neonatal mortality than urban children. The financial
crisis increased the odds of both neonatal and post-neonatal mortality for urban children
more than for rural children. As expected, in rural areas, the crisis-noncrisis differential
of the mortality risk at specific age (months) is relatively smaller for financial crisis than

that of the drought crisis.

60

Discussions

Several child health outcomes are examined in this paper to assess whether the
Indonesian financial and drought/smoke crises negatively impacted young children in
Indonesia. Birthweights, neonatal, post-neonatal, and infant mortality are examined in
this paper. The results from both the birthweight cumulative distribution comparison and
the OLS estimation similarly suggest that none of the crises had negative impacts on
birthweight in both urban and rural areas. Nevertheless, we realize that this evidence is
drawn from reported birthweights only. An investigation of the probability of reporting
birthweight in various delivery locations in Indonesia suggest that the birthweight
distribution drawn from reported birthweights may be biased due to unreported
birthweights of infants that were born at home or in offices of traditional midwives.
Further, biased results could come from selection problems when some mothers switched
from hospital to home delivery in the time of crisis. Unfortunately, we do not have any
plausible instrument to correct for this selection problem in reported birthweight. It is
difficult to find a factor that affects delivery location (and whether birthweight was
reported), but does not affect birthweight itself.

Unlike birthweight, our findings indicate that the financial and the drought/smoke
crises had significant adverse effects on infant mortality. The overall effects were different
in urban and rural areas. Although the financial crisis had adverse effects on neonatal

mortality in both urban and rural areas, the effects on post-neonatal mortality were felt by

61

only urban infants. Contrary to the effects of the financial crisis, the effects of the
drought/smoke crisis on post-neonatal mortality were felt by only rural infants.80

Results from multivariate regressions, nonparametric, and parametric hazard
estimations show similar effects on infant mortality. In urban areas, infant mortality was
affected by only the financial crisis. In rural areas, infant mortality was affected by both
the financial and the drought/smoke crises, but the drought/smoke crisis appeared to have
worse effects than the financial crisis.

These differential results are consistent with our expectations. Since a large
increase in the food prices and an adverse income shock during the financial crisis
resulted in a sharp reduction of real incomes for most of the Indonesian population, who
are net food purchasers, we expect to find some adverse effects on child health outcomes
among both urban and rural population. Since the smoke haze that resulted from erupted
fires during late 1997 and early 1998 affected only parts of urban areas in our sample, we
do not expect to find strong effects of the smoke haze on overall urban population.
Notice that when community fixed effects are controlled for in our analysis, the effects of
the drought/smoke crisis on neonatal mortality in urban areas lost statistical significance.
However, instead of combining the drought and the smoke/haze crisis, one could
distinguish these two crises by mapping exact locations and levels of the fire smoke haze
to each community in our sample in different periods. This task is plausible, but it would

involve elaborate work in data collection.

 

so We observed that neonatal mortality in urban areas ws affected by the drought/smoke crisis, but the level
of statistical significance is at 10%. After controlling for community fixed effects, the crisis appears to
have no effects on neonatal mortality.

62

Since the focus of this paper is to examine whether economic crises have any
effects on child mortality, controls correlated with crises such as household income and
public health provisions are excluded. Including these correlated variables in the
regression, however, may answer different questions.

In addition, regressions in this paper do not control for fertility decisions. Since
economic crises may affect fertility decisions, including fertility decisions in the
regressions might lead to endogeneity problems. Demographic theory and empirical
evidence from different countries suggest that mortality change should call forth some
fertility response (Preston, 1975). Major examples of studies that suggest mortality-
fertility link are given by Schultz (1969) and Federicksen (1966). According to Schultz,
parents try to compensate for the average incidence of death by seeking the number of
births that will give them the desired number of surviving children. Federicksen (1966)
infers from the regional birth rates and the rates of population growth in Ceylon,
Mauritius, and British Guiana that the improvement in health conditions that were
responsible for the death rates led to a subsequent reduction in birth rates. On the other
hand, some studies such as those by Adelman (1963) and Coale and Hoover (1958)
suggest no significant link between mortality and fertility. For instance, Coale and
Hoover (1958) in their classic text, Population Growth and Economic Development in
Low Income Countries, found the absence of major fertility decline in several developing
countries that had experienced a prolonged mortality decline. The direction of the
change in fertility decision during the Indonesian crises is, of course, an empirical one

that needs to be further studied.

63

Since economic crises may affect people with different socioeconomic
backgrounds differently, the study of the effects of economic crises can be extended to
focus on these differential effects. Our preliminary findings suggest that even though
some of the crises had adverse effects on infant mortality, infants born to mothers with
different education levels did not experience different adverse crisis effects.

As for infant mortality, we found that in urban areas, the financial crisis seems to
have worse effects on infant mortality for those born to mothers with lower education. In
rural areas, the effects of the financial crisis on infant mortality were slightly higher for
the lower education group. However, when community-fixed effects are accounted for,
the financial crisis effect is higher (and statistically significant) for the higher education
group. In rural areas, even though the drought/smoke crisis had adverse effects on infant
mortality for both education groups, the magnitude of the effects are larger for those
belonged to mothers with lower education. We found the differential effects between the
two groups to be larger when controlling for community fixed effects.

Even though the true underlying causes of the differential effects of the crises are
not completely explored in this paper, we view our results from the estimation of the
differential effects between different mother’s education levels as evidence of differential
effects of the crises on children with different socio-economic backgrounds. Since male
and female children exhibited different mortality rates, studying the differential effects of
the crises between male and female children may be of interest. This could be carried out
by adding interaction terms of the crisis periods and a sex dummy or by estimating

regressions separately for each group of children.

64

Conclusion

This paper examines the impacts of the recent Asian financial crisis and the
1997/98 drought and smoke haze crises on infant mortality and birthweight in Indonesia.
The paper uses data from three waves of the Indonesian Family Life Survey: IFLS]
(1993), IFLSZ (1997), and IFLS3 (2000), utilizing rich data on socio-economic
backgrounds as well as detailed information on children’s birthdates, birthweights,
mortality status at the time of interview, and ages at death if they died.

The methodology used in this paper is to compare health conditions of newborns
of different birth cohorts. Specifically, this paper examines whether those
conceived/born during the crisis periods exhibited higher risk of neonatal mortality and
post-neonatal mortality and whether their birthweights were lower than birthweights of
those born during the non-crisis periods.

The estimations of both neonatal and post-neonatal mortality risks are carried out
using multivariate regressions with socio-economic control variables such as mother’s
education, place of residence (province/community), and gender of the child. In addition,
mortality risks are estimated using hazard models to capture the mortality risks at
different age (in months). The paper uses both nonparametric and parametric hazard
models to estimates the hazard rates. The effects of the crises on birthweights are
analyzed using multivariate regressions and comparisons of birthweight cumulative
distributions. In both mortality and birthweight analyses, urban and rural samples are
analyzed separately since the economic crises could have affected mortality and

birthweight differently in rural and urban areas.

65

Estimated results on mortality outcomes show that the financial crisis had adverse
impacts on neonatal mortality in both urban and rual areas. Urban infant conceived
during the peak of the financial crisis exhibited approximately 1.7 percentage points
higher neonatal mortality risk (17 per thousand live births more) than those conceived
during non-crisis periods. The increase in neonatal mortality risk was approximately 2.2
percent for rural infants. The adverse effects of the financial crisis on post-neonatal
mortality risks were larger and more statistically significant for urban infants than for
rural infants. Overall, the financial crisis increased infant mortality risks by about 3.2
percent in both urban and rural areas.

The drought/smoke crisis adversely affected post-neonatal mortality risks in rural
areas. The increase in the post-neonatal mortality risk is about 3.1 percent. When
community fixed effects are controlled for, the drought/smoke crisis appears to have had
much larger effects. Overall, the drought/smoke crisis had no significant advese effects
on infant mortality in urban areas, while the effects in rural areas were large. Our
estimates show that rural infants born during the drought/smoke crisis experienced
approximately 4.4 percent increase in their infant mortality risks (44 per 1,000 live
births). The magnitude of the effects almost doubled after controlling for community
fixed effects.

Our findings on differential effects on infants born to mothers with different
levels of education indicates that infants born to mothers with different levels of
education exhibited no significantly different trends in mortality risks over time in both
urban and rural areas. We also find that even though the economic crises had adverse

effects on infant mortality, infants born to mothers with different education levels did not

66

experience statistically different adverse financial crisis effects in either urban or rural
areas. There were statistically significant differencial effects of the drought/smoke crisis
on post-neonatal mortality between the low and the high education groups in rural areas.
Those born to mothers with lower education were more adversely affected by the
drought/smoke crisis than those born to mothers with higher education.

Results from the cumulative distribution comparisons of birthweights suggest that
the financial crisis also had adverse impacts on birthweight in urban areas. However,
under multivariate analyses, the adverse effect seems to disappear. None of the crises
affected birthweights in rural areas. The lack of an evidence on the adverse effects
maybe due to a selection problem in reported birthweights. The data show that from
1988 to 2000, 56 percent of women gave birth at home or at a family member’s house.
For those who were born at home, only 52 percent have reported birthweights, whereas

99 percent of those born in hospitals have reported birthweights.

67

CHAPTER 2

THE EFFECTS OF THE 1998 ECONOMIC CRISIS ON AGES OF FEMALE

FIRST MARRIAGE AND FIRST BIRTHS: EVIDENCE FROM INDONESIA

Introduction

This paper utilizes data from the Indonesian Family Life Surveys (IFLS) to study
short-run effects of the 1998 Indonesian financial crisis on the ages at which females first
marry and when they have their first child. It is an extension of Rukumnuaykit (2003),
which studies the effects of the Indonesian economic crises on infant mortality and
birthweight. Previous studies on the effects of economic crisis on demographic outcomes
have not found evidence of adverse effects on infant/child mortality, but stronger
evidence has been found of the effects on fertility and marriage delays. In this paper, we
investigate ages of female first marriages and first births and argue that increasing rates
of marriage and fertility delay may have been used as consumption smoothing
mechanisms to cope with the Indonesian crisis.

The Asian financial crisis struck Indonesia in January 1998.1 Figure 1 shows that
the sustained crisis period lasted more than one year with the peak in Rupiah/USD
exchange rate in July 1998. As shown in Figure 2, food prices in both urban and rural

areas increased more than 250 percent at the peak of the crisis. Alatas (2002)2 argued

 

' Refer to figure 1.

2 Alatas, Vivi. “What Happen to Indonesia’s Poverty? “A Micro Simulation Exercise Using Household
Surveys.” Manuscript. World Bank. Jakarta, Indonesia. March 2002.

68

that this substantial increase in food prices was a major source of the impact of the crisis
felt by Indonesians, except those that belong to the top of the income distribution.
Simulation results from Alatas’s study indicates that the increase in food prices between
February 1999 and February 2000 accounted for approximately 40 percent of the increase
in poverty rate afier the crisis.

While an increase in food prices could help net food-producers, this large increase
in food prices during the crisis resulted in a sharp reduction of real incomes for most of
the Indonesian population, who are net food purchasers. The proportion of households
below the poverty line rose from about 11 percentage points in 1997 to almost 20
percentage points in 1998 (Frankenberg, Thomas, and Beegle, 1999). The household per
capita consumption declined by about 20 percent fi'om 1997 to 1998 (Frankenberg,
Thomas, and Beegle, 2003).

Results from physical assessments show no deterioration in children’s health
status in 1998. There were only negligible changes in the measurements of children older
than six months in their height-for-age and weight-for-height. Very young children were
well protected from the effects of the crisis although there is a suggestion that weight-for-
height of this group of children may have worsened (Frankenberg, Thomas, and Beegle,
1999). The fact that the short-run impacts of the crisis on child health have been small
suggests that households may have used various consumption smoothing mechanisms to
protect some certain individuals in the households from the crisis effects.

To smooth out the effects of the crisis on consumption, households may have
drawn an array of resources available to them. Past empirical evidence showed that a full

Pareto-efficient allocation of risk within local communities is rarely achieved. Generally,

69

some idiosyncratic variation still remains uninsured (Bardhan and Udry, 1999).
Moreover, when the income shocks are at an aggregate level, cross-sectional risk pooling
is not effective. In the case when insurance markets are absent or incomplete, households
often rely on alternative income and consumption smoothing mechanisms. These
alternative mechanisms are likely to take many forms. Often, the ex post mechanism for
consumption smoothing is to smooth consumption over time using saving (assets) and
credit transactions (Deaton, 1991; Bardhan and Udry, 1999). Saving and borrowing
require a surplus in other periods. The accumulation can take the form of cash, goods,
land, or livestock. Since the duration of economic crises are normally unknown,
households often face a risk of using up savings and borrowings before adverse
conditions improve. Therefore, in many cases, borrowing and saving can only provide
short-term relief.

Other consumption-smoothing strategies that households may rely on include
changing the allocation of total consumption (e.g. delay consumption of durable and
deferrable goods), changing work hours and/or types of work of household members
(Murruggarra, 1996), using the entry and exit of household members (Alamgir, 1986) or
changing location of residents of household members (Rosenzweig 1988, 1996;
Rosenzweig and Stark, 1989), and using children as a substitute for insurance (Portner,
2001; Cain 1981, 1983; Clay and Vander Haar, 1993; De V03 1985; Nugent 1985;
Thomas 1991)

The Indonesian households appeared to have smoothed their consumption by
reallocating the household’s budget. With a substantial increase in food prices, per capita

food consumption was reduced by only 9 percent while expenditures on nonfoods were

70

reduced by about a third. Households substantially reduced per capita expenditures on
“deferrable” item including clothing, furniture, and spending on ceremonies, which
declined by more than one-third. Investments in human capital (health and education
spending) were reduced by around 40 percent.

Households also expanded and relocated during the economic crisis presumably
to take advantage of household fixed consumption costs. Household size was expanding
in rural areas across the entire distribution of 1993 per capita expenditures while only
households above the median 1993 per capita expenditure gained new members in urban
areas. Evidence also suggests that some members of the poorest urban households
moved to rural areas to take advantage of low costs of living. In addition, households
seemed to be adjusting labor supply of their members in an attempt to cope with the
crisis. The number of workers and the total number of hours worked by all household
members increased in both urban and rural areas. This change reflects an increase
number of workers in the wage sector in urban areas and in family business in rural areas.
(Frankenberg, Smith, and Thomas, 2003)

Another remarkable channel the Indonesian households used to smooth
consumption is through selling previously accumulated jewelry. The ownership of
jewelry was reduced by more than 30 percent between 1997 and 19983 while the
aggregate ownership rates for all other assets have remained stable. The average value of

gold sold was approximately equivalent to four months of food consumptions in rural

 

3 The decline was more than 30 percent in rural areas and slightly less in urban areas. Jewelry was a more
common than financial assets in rural areas and most of urban areas. Households stored their wealth using
gold rather than financial assets. Those who stored gold were able to sell it for a four-to-fivefold increase

in price due to the devaluation of the Indonesian Rupiah.

71

areas and nine months of food consumption in and urban areas (Frankenberg, Smith, and
Thomas, 2003).

In this paper, we investigate ages of female first marriages and first births as yet
another potential strategy to smooth consumption. We find the effects of the crisis on
these demographic outcomes by comparing, at different ages, the odds of becoming
married for women who were exposed to the crisis and that for women who were not
exposed to the crisis. Similarly, the odds of having a first child conditional on being
married were compared. In addition, we compared the odds of having a first child
starting from 14 years of age, unconditional on being married. Our method of analysis
incorporates both parametric and non-parametric estimations of the marriage and first
births hazards. We find evidence of an increase in the probability of marriage and a
decrease in both the conditional and unconditional probabilities of a first birth when
exposed to the crisis. We argue that marriage and fertility delay may have been used as
consumption smoothing mechanisms to cope with the Indonesian crisis. Using maniage
of individuals in the households provides additional sources of economies of scales and
connection with other households to smooth consumption. However, the decision to
postpone or forgo having a child during the crisis may have been due to a dominating

income effects caused by a drastic decline in real income and a sharp increase in prices.

72

Background

Economic crises could affect individuals’ decisions on household formation and
first birth through changes in prices, employment, and wages received. In terms of
changes in employment, Frankenberg, Thomas, and Beegle (1999) found that overall the
1998 crisis had no massive short-run impacts on the unemployment rate among men.
However, the crisis had different impacts on men of different age groups. The proportion
of men who worked and men who worked for pay increased significantly for men aged
15-24 years old while the proportion of men working for pay declined for an older age
group (36-64 years old). The impacts of the crisis on employment among women were
distributed across all age ranges. Overall, there was a significant increase in the
proportion of women working from 49.2% in 1997 to 56.2% in 1998. The authors found
that this increase in proportion of women working was due to more women working as
unpaid family labor because the change in the proportion of women working for pay was
insignificant (from 36% to 37%).

In terms of wages, Frankenberg et al. found that the 1998 crisis resulted in a very
large decline of 20% to 30%4 in the median real wage of all workers between 1997 and
1998. The declines in real wages were larger for men. Urban residents seemed to be
more adversely affected.

The evidence of an increase in the proportion of women working in family
businesses and an increase in the proportion of young men working for pay suggests that
marriage rates could have increased during the crisis to utilize economies of scale and the

comparative advantage of the partner’s skills provided by cohabitation, given that men

 

‘ Depending on the deflation method used.

73

had comparative advantage working outside home compared to women.5 Also, income-
pooling may have been important for couples to smooth their consumption during the
crisis.

No evidence exists on the effects of the Indonesian economic crisis on the age of
female at marriage. Manital age, however, is known to have increased in Indonesia.
Hull and Jones (1994) reported a strong upward trend in marital age. They found that the
mean age of marriage had risen in all provinces of Indonesia throughout 19603, 19703,
and 19805. Overall, the average age at marriage of women in Indonesia was 19.3, 20.0,
and 21.1 in 1971, 1980, and 1985 respectively. In Java, the average age increased from
18.1 to 20.7 from 1964 to 1985. Similarly, data from the Demographic Health Surveys
showed that the median age at first marriage for women aged 25-49 years rose from 17.2
years in 1987 survey to 18.6 years in 1997 surveys.

It is worthwhile to note that at the same time Indonesia experienced a fertility
decline nationwide. Hull and Jones (1994) reported a 40-percent decline in the fertility
rate from the 19603 to the I980S. Overall, total fertility rate was 5.61 in 1967-70 and
3.33 in 1986-89. The authors claimed that the evidence of a major fertility decline in
Indonesia was robust to various fertility estimation techniques (see Hull, 1980; Hull and
Hull, 1984).

Similar to the crisis effects on the propensity to marry, empirical evidence of the
short-run effects of the Indonesian crisis on ages of first births is limited. As long as
fertility remains a normal good, theories of the demand for children generally lead to a
hypothesis that deterioration in income leads to a delay of having first birth due to

income effects. However, an increase in the demand for children could take place due to

 

5 Note that cohabitation between non-married couples is still rare in Indonesia.

74

substitution effects when female wages decline. If fertility can be consciously controlled,
contraception use will be adopted to control family size, provided that couple can afford
contraception.

Frankenberg et a1. (1999) reported that although there was a small change in the
type of providers from whom women obtain injections, there was no significant change in
prevalence or method mix of contraception during the first year of Indonesia’s economic
crisis. Contraceptive prevalence was estimated to increase from 56.6% in 1997 to only
57.3% in 1998. The authors concluded that the economic crisis did not result in changes
in contraceptive behaviors. They argued that this stability of contraceptive prevalence
suggests that contraception is a more appealing option than the risk of having an
additional child in the economic crisis environment for the majority of couples in

Indonesia.

75

Theoretical Framework

The Impacts of Economic Crises on Marriage

It is clear from our previous discussions that the Indonesian financial crisis
resulted in deterioration in real income and real wages. In this paper, we argue that the
combined effect of declines in real income and real wages on women’s marriage
propensity is a priori ambiguous. Specifically, we argue that changes in income and
wages induce potential gains from maniage through income pooling and specialization.
At the same time, the demand for marriage could decrease due to a decrease in the
demand for home-produced goods because of adverse income effects. A drastic decline
in wages also decreases the opportunity cost of home-produced goods, resulting in a
higher propensity to marry. On the other hand, the crisis could delay marriages through
increased in direct costs of marriage (e.g. cost of setting up new a household).
Furthermore, if the crisis adversely affects expectation of future income (probability of
low future income) and its volatility, households could use marriages of individuals in the
households as a tool to smooth future consumption, resulting in an increase in marriage
rates.

According to Becker (1973), two persons choose to marry each other if and only
if both of them are made better off from marriage than from remaining single. A couple
combines their time and market goods and services to produce “household goods,” from
which their utility is drawn. Examples of these goods include quality and quantity of
children, quality of meals, and mutual love and companionship. Since “household

goods” are produced partly by marketable commodities, each household member

76

allocates his/her time between marketable and non-marketable activities in the
“appropriate proportions.”

Single persons could gain from marriage at the time of an economic crisis because
marriage provides additional time and good and services from the spouse assuming
increasing return to scale. By pooling time and incomes, Becker argues that marriage
provides economies of scale in production of both market and household goods when the
subsititutability between the times used by each partner is imperfect. The more
complementary the inputs (the time of spouse and market goods) provided by each
spouse, the more the gain from marriage (see proof by Becker (1973)). In short, the
partner with earning capacity outside the home tends to specialize in paid employment,
while the other tends to specialize in home work. The gain from this specialization is
larger when the disparity between wage rates increases.

When the income of the household decreases, the demand for home produced
goods (e.g. quantity and quality of household works and children) decreases, provided
that these goods are normal goods. This income effect implies a marriage delay as a
result of an economic crisis. At the same time, the deterioration in female wages lowers
the shadow price of home-produced goods, resulting in an increase in the demand for
marriage. In the case when both male and female wages decrease, this substitution effect
implies that the demand for marriage decreases when the decline in male wages is higher
than that of the female wages (the female/male relative wages is higher).

Keeley (1977) incorporated search cost into the marriage theory developed by
Becker (1973). The gain from marriage depends on the combination of the

characteristics of each of the mates. According to Keeley’s marriage search model, a

77

single person enters the marriage market only if the expected gain from maniage exceeds
the costs. If the benefit exceeds the costs, the searcher decides to enter the marriage
market. Then the searcher decides for a minimum accepted offer, which is the share of
the total home produced output the searcher would receive when married. After the
searcher enters into the marriage market, the searcher accepts any offer of marriage that
equal or are higher than her acceptance wage. This theory implies that a decrease in
“single” income decreases the duration of marriage search. Therefore, based on this
search consideration, a decrease in “single” income (both current and expected future
income) as a result of an economic crisis increases the propensity to marry. On the hand,
the theory also implies that in the time of the crisis, a decrease in direct costs (monetary
and opportunity time cost) of search for partner increases the duration of search, which
results in a delay of entering into marriage. In sum, Keeley’s marriage search theory
implies that the net effects of the Indonesia crisis on the incentive to marry is ambiguous
a priori since both income and opportunity time cost (wages) declined as a result of the
crisis.

Recall that the economic crisis adversely impacted income and wages of both men
and women. The gain from marriages, then, depends on the relative income (wages) of
men and women. Holding the income of men constant, when the income of a woman (or
income of her household) deteriorates, there is a higher probability for her to enter into
the marriage market because of expected gain from marriage due to specialization is high.
The minimum offer she is willing to accept (compared to her single’s income) will be
low, resulting in shorter duration of search. At the same time, her low opportunity time

cost allows longer duration of search.

78

Holding income of women constant, when the income of men deteriorates, the
women’s expected gain fi'om marriage decreases. Women may decide to stop or delay
entering into the marriage market. Even if they do enter into the market, the duration of
search will be longer given that there is a high proportion of men who are unable to offer
the minimum accepted offer the women have set.

Contrary to results in previously discussed theories, the demand theory of
marriage also predicts a positive relationship between marriage rates and economic well-
being when taking into account changes in costs associated with marriage and resource
availability to the household. This view is widely accepted in many studies as an
explanation for marriage delay as a consequence of economic crises. According to Hill et
a1. (1993), maniage costs could be in explicit or implicit terms. Explicit costs include
costs incurred with the process of marriage (e.g. payment for a bride price and a new
home). Implicit costs, such as a “perceived need” for a certain level of wealth and
income security before marriage, may also play and important role in the marriage
decision. An economic crisis may stop or delay marriage decisions because it increases
financial constraints on the ability to set up a separate household (e. g. purchase a house
or other consumer durables) especially when borrowing opportunities against filture
income are limited. Financial constraints could result in inability of males to accumulate
enough resources to facilitate maniage (Hill et al., 1993). Palloni and Hill (1996) find
that empirically the typical pattern of nuptiality response of a crisis is an immediate drop
in the number of marriages. This drop in marriage is usually followed by a lagged
increase above and beyond expected rates in normal times. The authors claim that

although the duration of this offsetting response may depend on the severity of the crisis,

79

“the number of marriages will decrease during the first two years after the onset of the

crisis, followed by an increase, as its impact receded.” See summary results below.

 

 

Considerations [Income lower Wf lower fMlm higher
Income pooling and cost sharing Increase Increase Increase
HH demand
Income Effects Decrease

Substitution Effects Increase Decrease

Search
Probability of Entering Increase Increase Decrease
Duration of Search Decrease Increase Increase

 

 

 

 

Direct Cost of Marriage (eg. housing) Decrease

 

In addition, an economic crisis could result in an increase in the number of
marriages due to a household’ income-smoothing consideration. Kotlikoff and Spivak
(1981) and Rosenzweig and Stark (1989) offer an alternative theory that is in dissonance
with standard models of marriage. Their hypothesis is that a marital arrangement plays
an important role in a household’s ability to smooth consumption when faced with highly
variable income streams, especially when access to credit markets is limited.
Specifically, they argue that marriage arrangements, or “exchange” of individuals among
household, serve to mitigate income risk and facilitate consumption smoothing, under
conditions that there are informational costs and spatially covariant risks. Under this
consideration, implicit risk sharing arrangement among households can provide strong
economic incentives for marriage.

According to Rosenzweig and Stark’s review, empirical evidence indicates that
inter-household family transfers provide an important source of income insurance in low-
income countries. For instance, data from the Malaysian Family Life Survey (MFLS)
show that “69 percent of all women who had ever moved from one town to another did so

at the time of their marriage, with 32 percent of all moves (town to town) by women

80

accounted for by marriage” (Rosenzweig and Stark, 1989). In India, households that
experienced income shortfalls associated with variation in weather patterns relied heavily
on nonresident in-laws to provide income transfers (Caldwell, Reddy, and Caldwell 1986;
Rosenzweig 1988). Caldwell et a1. (1986) found that, in nine villages in South India, 56
percent of the relatives that provided aid during droughts were either relatives of the
head’s wife or those of the husbands of the head’s daughters. Similarly, data from ME LS
suggests that 39 percent of the values of all goods and cash transfers are for “emergency”
help and for “maintaining” household’s expenditures (Stark and Rosenzweig, 1989; Butz
and DaVanzo, 1978).

Note that using marriage as an income-smoothing tool is more likely an ex ante
consumption smoothing consideration to protect households against future income
fluctuation. Thus, under this consideration, an economic crisis could result in an increase
in marriage rates only when the crisis adversely affects the expectation of future income
and its volatility. For instance, if households (individuals) increase the degree of risk
aversion after an economic crisis takes place, there is an increase in marriage gain, which
in turn induces higher propensity to marry (for proof and simulation results, see Kotlikoff

and Spivak (1981)).

The Impacts of Economic Crises on Fertility
Economists hypothesize that fertility decisions are made by rational individuals to
maximize their well-being.6 According to Becker (1960), utility is maximized subject to

an exogenously determined budget. The demand for children is then affected by income

 

6 Part of this theory review is drawn from a review by Marvellous M. Mloyi (1992).

81

and prices. As long as fertility stays a normal good, one would expect a contraction in
income and an increase in prices to be associated with a fertility decline, or at least by a
pause in family expansion (Bertrand et al., 1993). However, demand theory also
considers other input goods such as time of the mother needed to raise the quality of
children (Mincer 1963; Becker et.al, 1973). When wages decline, the opportunity costs
of time to raise the quality of children are lower. This substitution effect induces an
increase in fertility.7

An economic crisis could also alter expectations regarding future incomes. When
fertility within maniage can be consciously controlled, conventional microeconomic
theory predicts that “couples will delay births in response to sudden declines in income.”
(Galloway 1988; Lee 1990; Palloni, Hill, and Aguirre 1996) A shift to small family size
desires is a function of both individual and community variables. Once couples desire
small family sizes, the adoption of contraception use will facilitate the realization of
small family sizes.

The reduction in family size will also result from changes in other supply
variables in addition to contraception (Mloyi, 1992). In addition to conscious
adjustments in fertility, economic crises can have unintended adverse effects on fertility

behavior. Psychological stress and declines in nutritional status associated with

 

7 In this paper, we consider only decision on the timing of first birth. Becker and Tomes (1976) argue that
when quantity-quality of children are taken into consideration, this increase in the women’s contribution to
home work (e.g. quality of children) raises the cost of an additional child, which reduces the demand for

quantity since higher quality of children would then be more expensive.

82

economic crisis tend to reduce fecundity8 and increase abstinence (Bonng and Cain
1981; Caldwell and Caldwell 1992; Kidane 1989). Fertility may also be reduced by
spousal separation due to labor migration as one or both partners search for means to
maintain consumptions, usually through employment (Galloway, 1988; Lindstrom and
Berhanu, 1999). Additionally, with the circumstances of economic hardship, nutritional
supplementation and/or termination of breastfeeding may be reduced and delayed. As a
result, the protective effect of natural contraception from breastfeeding may be retained,
even in the case of increased contraception usage. (Mloyi, 1992)

Hill et a1. (1993) emphasize the importance of distinguishing short-run and long-
run economic effects on fertility and marriage because the relationship that links
economic change and these events in the long run are complex. They claim that while
the causes of short-term variation in fertility are well-understood, “there is not a clear
consensus about the mechanisms underlying the long-run relationship between economic
change and fertility.” (Hill et al., 1993)

When taking into account long-term considerations, economic crises could induce
fertility increase by a “risk insurance approach.” Under this theory, high fertility is
considered insurance against long-term insecurity. (Cain 1981, 1983; Clay and Vander
Haar, 1993; De V05 1985; Nugent 1985; Thomas 1991). Caldwell (1976, 1978, 1982)
proposes a “wealth flow” theory, which suggests that as long as wealth flows from
children to parents, fertility could remain high evenn if the costs are high since it is
economically rational for a parent to have more children. Cain (1983) offers similar line

of reasoning. In his view, when there are limited extra family resources or institutions,

 

8 It is argued by Hill et al. that the role of nutrition sub-fecundity may be limited in the case where scarcity
is not severe. The evidence of a significant decrease in fecundity is weak. (Menken, Trussell, and

83

children can offer old-age security to parents, and can offer added insurance against
“daily-survival risks,” which may range from droughts to illness.

Portner (2001) gives a theoretical framework that explains how children can
provide insurance against uncertain future incomes. The main hypothesis of his work is
that children can act as a security asset when insurance and credit markets are either
absent or poorly functioning. His theory is based on an assumption that children are
likely to be more reliable as a means for insurance than more distant family. In Portner’s
model, children are modeled in a dynamic setting as a general insurance and saving asset.
In his model, children are costly to the household in the first period of their life, but they
provide a positive net income in subsequent periods. Children can help by working at
home or as wage labor, and older children who either have their own households or have
migrated can make transfers to their parents. Hence, “parents use children as a means to
shift income from a period with certain income to future periods with uncertain income,
thereby insuring themselves against the possibility of low income.” (Portner, 2001) His
model implies a positive relation between income and fertility in a given period holding
constant expected future income. When the future expected income is not fixed, the
model predicts a negative relation between future expected income and the number of
births since an increased probability of low future income leads to high demand for
insurance and therefore more births. This theory implies that it is not sufficient to
observe present income to determine demand for children. To completely account for the
effects of an economic crisis on fertility, one needs to also assess people’s own

expectation of future income and its variability, which are hardly observed in the data.

 

Watkins, 1981)

84

Furthermore, Portner (2001) suggests that there is very little direct evidence on whether
children serve as a substitute for insurance.

In addition to effects of economic crises on fertility through changes in income
and wages, economic crises could affect changes in fertility through the effects of
economic crisis on child mortality. Sah (1991) analyzed the effects of child mortality on
fertility decisions assuming that parents derived direct utility from the number of
surviving children. This model implied that when child mortality increased as a result of
the crisis, parents demanded more children to insure against the risk of child mortality in
order to maintain the number of children they desire.

In addition, as implied by Portner (2001) an increase in the probability of child
mortality has both substitution and income effects on the demand for children as
insurance. First, the substitution effects imply that an increase in the probability of child
mortality results in a lower return to births (i.e. more wasted resources). Secondly, The
income effects imply that a lower expected number of survivors leads to a lower expected
consumption in the future.

While the substitution effect tends to decrease the optimal number of births the
income effect tends to increase number of births. If the income effect dominates, the
optimal number of birth will increase. Moreover, the more risk averse a household is, the
more likely it is that the income effect will at one point dominate the substitution effect
when the probability of child mortality is increased. Therefore, the model is able to
illustrate the observed increase in fertility following an increase in infant and child

mortality, provided households are indeed risk averse.

85

Previous Studies

According to an empirical review by Galloway (1988), high food prices were
significantly associated with a decline in maniages in England, France, Sweden during
pro-industrial periods. Likewise, poor harvests resulted in a decline in marriages in
Croatia. Historical evidence also suggests a positive relationship between economic
well-being and fertility increase. According to the same review by Galloway (1988),
historical research showed that when grain prices increased, England, France, and
Sweden experienced fertility declines. Similarly, a rise in fertility was linked to an
increase in real wages in Sweden and in the harvest in Croatia. In countries outside
Europe, Galloway and Lee find that high prices were strongly associated with fertility
decrease in pre-World War II Bombay Presidency, slightly associated with a fertility
decrease in Taiwan, and not significantly related to changes in fertility in Japan.

Das Gupta (1995) examined fertility decline in Ludhiana District in India. She
found evidence that was parallel to that from historical Europe. Total fertility in this area
began to decline around 1940, which was before the onset of family planning
programmes and the Green Revolution in 1966. According to Das Gupta, this decline in
fertility was partly due to the expansion of irrigation system that resulted in an increase in
level of yields and a decrease in the yield fluctuation, which in turn results in increased
security against mortality risks and food shortages.

In his 1988 study, Galloway investigated the short-run effects of economic crises
on fertility, nuptiality, and mortality in Pre-industrial Europe using fluctuation of grain
prices and vital statistics. The temporal unit of analysis in his study is the calendar year.

Detrended crude birth rates and marriage rates were used as the dependent variables.

86

Detrended grain prices and detrended non-infant death rates each distributively lagged
five years were the independent variables.9 Galloway found that fertility was highly
sensitive to fluctuations of grain prices. As he expected, the largest fertility response
occurred one year after the price shock. Both the patterns and the magnitude of
elasticities of responses were similar in all areas and all periods. On average, most of the
effects of high prices on nuptiality occurred at lag 0, but there was considerable variation
between countries.

A similar study by Palloni and Hill (1996) investigated the effects of economic
swings on fertility, nuptiality, and mortality in Latin America from 1910 to 1989. In this
study, real average GDP was used as an indicator of economic well-being. The authors
estimated the effect of this indicator on number of reported births, marriages, and infant
mortality rates using local least squares with data from eleven countries. The results
showed the response of number of marriages at lag 0 was positive in seven countries
(Argentina, Chile, Cuba, Costa Rica, El Salvador, Uruguay, and Venezuela), but the
effects were statistically significant only in Chile, Uruguay, and Venezuela. The pattern
of responses lag was consistent with expectations. Results from four countries gave
negative lag 0 marriage response. The only country that had all negative responses is
Guatemala. The results on marital fertility showed greater heterogeneity in the patterns

of responses. Data from five countries indicated positive responses at lag 0 and 1 to real

 

9 Non-infant death rates are used as proxies for both adult mortality and morbidity in this paper since adult
mortality rates are not available.

87

average GDP,10 but only in Cuba (lag l) was the response statistically significant. In six
countries, the responses of lag 1 had the unexpected sign.

Hill et a1. (1993) used economic conditions such as gross domestic product per
capita, the quantity of exports, and terms of trade to study the effects of economic
reversals in Sub-Saharan Afiica on child mortality, the odds of first marriage, and timing
of first and second birth. The authors found that the effects of economic reversals on first
births are the strongest among the four demographic outcomes. They found evidence of
first-child fertility delays in all seven countries studied, except in Kenya. For first
maniage, the study revealed evidence of a positive association between economic
conditions and the odds of first marriage in only Botswana, Senegal, and Togo.

Rutenberg and Diamond (1993) examined the rapid fertility decline in Botswana
during 1981 and 1989. Based on their examination of fertility rates and employment of
traditional agricultural workers in rural areas and their reviews of relief programs timing
the drought crisis, the authors claimed that “the decline in fertility was linked to a
deterioration in social and economic conditions caused by a major drought in the early
19803 and to the increased availability of family planning services in the same period.
Fertility began to rebound in the late 19803 in response to improved conditions, which
came about as a result of a successful drought relief program. Further declines in fertility
depended on the continued success of the family planning program, particularly in the
rural areas.”

Lindstrom and Berhanu (1999) examined fertility trends between 1973 and 1989

in Ethiopia for evidence of short-term and long-term responses to famine, political

 

'0 Seven out of eleven countries have positive responses at lag 0. The authors offer an explanation that this
positive response would be expected only if there was a possibility of “anticipatory behavior, expressed as

88

events, and economic decline. Using year dummies as identifying explanatory variables
and controlling for characteristics thought to be powerful predictors of fertility (e.g. age
at the start of the interval, age squared, and place of residence in the current year), the
authors found evidence of significant short-term declines in probability of conception
during years of famine and major political and economic crisis. As for long-term effects,
the authors found that in both rural and urban areas, fertility declined in the 19803 after
increasing moderately in the 19703.

Mckenzie (2002) studied how Mexican households coped with the 1994-1996
Peso Crisis. His result showed that one strategy households used to cope with the crisis
was a reduction in fertility during the crisis, with about one in twenty households decided

to postpone or forego having a child during the crisis.

 

reduced number of conceptions or increased numbers of voluntary and spontaneous abortions.”

89

Methodology

We find the effects of the crisis on the ages of female first marriage and the
timing of first births by comparing, at same ages, the odds of becoming married for
women who were exposed to the crisis and those for women who were not exposed to the
crisis. Similarly, the odds of having first child were compared. The dependent variables
of interest are the odds of an event occurring.

1) the odds of a women becomes married in year t given that she is

unmarried at the start of year t,

2) the odds of a married woman having the first child after t years of
marriage given that she has no children at the start of year t afier
marriage. To avoid the complication of a considerable decline in the
number of births after first births and other constraints that may affect
fertility after first births (e.g. sex preference and condition of the
mother after the first birth), this fertility study limits the analysis to first
births. Note that in this conditional estimation, we take the age of
maniage as exogenous when studying the odd of having first births.

3) the odds of a woman having her first child at t years of age given that
she had no children at the start of year t. In this estimation, t starts at
age 14 years regardless of whether the women were married. This
unconditional hazard estimation takes into account the possibility that
the net effects of the crisis, estimated from the conditional fertility

hazard estimations, could be biased if the marriage hazard was affected

90

by the economic crisis. Similar to the conditional analysis, we limit the

unconditional analysis to studying only first births.

Nonparametric Estimation of Hazard of Marriage and Fertility

The nonparametric estimates of discrete yearly hazard rates are based on the
Nelson-Aalen cumulative hazard function. This cumulative hazard function produces
estimated hazard components. It is recorded at all the points at which a failure occurs
and computed as d,/n,, where d, is the number of failures occurring at time t and n, is the
population alive at t before the occurrence of the failures. In the marriage analysis, t
starts when the woman is 11 years old. In fertility analysis, since age of first pregnancy
could be greatly influenced by age of marriage, we use number of years after marriage as

the length of time in our survival analysis.

Parametric Estimation of Hazard of Marriage and Fertility

In addition to nonparametric estimations, we estimate the effects of the crisis on
marriage and fertility separately using different parametric hazard models. Let T 2 0
denotes the length of time the woman “survived” (remained single) in years. Similar to
nonparametric estimations, in marriage analysis, this T starts when the woman is 11
years. In the fertility analysis, we use number of years after marriage as the length of
time in our survival analysis.

In this paper, we estimate maximum-likelihood (cox) proportional hazards models

and hazard models that assume some distribution of survival time in the population.

91

Assume T has some distribution over the population. Consider time invariant

covariates.

Let F(t; x) = conditional cdf of T where x = covariates; province of residence,

rural/urban, village/small town/big city, women’s education:
F(t; x) = P(T S t; x), t 2 0
The survivor function is defined as
S(t;x) = 1 — F(t;x) = P(T > t;x)
Then, the probability of leaving the initial state in the time interval (t, t+h) is
P(tST<t+h|T2t;x) forh>0

Define the hazard function as

,(.,.)-P(,g,~<,.,,,2rn=P"§,§:§§§“”=F(";’_";l;3(“”-

If the c.d.f. is differentiable, then take the limit of the right-hand-side, divide by h,

as h approaches zero. The hazard function is

. _. F(t+h,x)—F(;x) 1 _ f(;x) _f(;x)
“LU-1:33;)! h t c,1—F(t;x)—l—Ft‘(t;x)—S(:;x)

Then all probabilities can be computed using this hazard function. For example,

from time a to time b, a < b is

o
P(aST<b|T2a;x)=l—exp[— Il(s;x)ds:l.

92

We estimate the hazard function based on different assumptions of the population
distribution of T. In our analyses, we estimate hazards models using Generalized Gamma,
Weibull, and Lognorrnal distributions. The standard error and the 2 statistics of the
estimated coefficients are robust to heteroskedasticity of the variance-covariance matrix
and clustering at the individual level. In addition, the estimations assume heterogeneity
in our observations. The observations are assumed to have inverse-Gaussian
heterogeneity (frailty). The hypothesis being tested is the significance of the crisis

dummy variable when the time trend is included.

93

Data

This paper uses data from the Indonesia Family Life Survey (IFLS). IFLS is a
continuing longitudinal socioeconomic and health survey that includes more than 30,000
individuals living in 7,200 households. The sample covers 321 communities in 13
provinces in Indonesia and represents about 83 percent of the Indonesia population in
1993”. The first wave of IFLS was fielded in 1993 (IF LSl). The same households were
revisited in 1997 (IF LS2) and again in 2000 (IFLS3). This paper uses the data fiom these
three IFLS waves. A 25 percent sub-sample of households was re-interviewed in 1998
(IFLS2+), but the data are not used in this paper.

In [F LS surveys, special attention is paid to the measurement of health, work,
migration, maniage, child bearing, life history data on education, and economic status of
individuals and households. In each wave of IF LS, the individual and household surveys
are complemented by an extremely comprehensive community and facility survey. There
is also considerable attention placed on minimizing sample attrition in IFLS. Targeted
households and individuals who “split-off” from original households were followed if
they moved to new locations within 13 provinces of the survey areas. In each re-survey,
about 95 percent of targeted households have been re-contacted. The split-off households
added just under 1,000 households to the sample in 1997 and about 2,600 households in

2000. The survey periods of each IFLS wave is shown in Figure 1.

 

” Frankenberg, E., Hamilton, P., Polich, S., Suriastini, W., and D. Thomas. User’s Guide for the Indonesia
Family Life Survey. DRU2238/2-NIA-NICHD. March 2000.

94

Marriage Data

Marriage data are taken from all three waves of IFLS. We start with IFLS3,
which was surveyed in 2000 to obtain complete marriage history of women in our
sample. We obtain retrospective marriage data from all women aged 15 and above at the
date of the interview. Data of ever-married women aged 15-49 years are taken from
“ever-married-women questionnaire.” Data of ever-married women aged 50+ years,
never-married women aged 50+ years, and never-married women of all ages are taken
from “adult questionnaire,” which includes all women aged at least 15 years.

From both sources, data on age at the time of the survey, birthdate, and date (or
age) of the first marriage (if ever married) are obtained. The marital history is complete
for new respondents. When the woman ever married and is a panel respondent, further
efforts are taken to track the date of the first marriage from earlier IFLS waves using the
personal identification number that is consistent across all IFLS waves. To avoid
problems related to outliers, we excluded those that reported the first married before 11
years of age. However, before the observation was taken out of the sample, we
investigate whether there was a reporting error by reconciling reported birthdate, the age

at the interview date, and the date (or age) of the first marriage.

Fertility Data

Data on pregnancies are obtained fi'om retrospective data on pregnancy histories.
Complete pregnancy histories were given by married women whose ages were between
15 and 49 years in “ever-married women questionnaire.” These women were asked about

information of each pregnancy in detail (if she ever had any). The information we use

95

includes pregnancy order, pregnancy outcome, and the date of the end of pregnancy.
Since we investigate fertility decision to have first birth, the date of the end of each
pregnancy was taken regardless of whether the pregnancy ended in a birth or a
miscarriage (Henceforth, the date of the end of each pregnancy will be called “birth
date”).

We start with IFLSl data to obtain data on number of children and pregnancy
information of each ever-married women. For women who ever had any pregnancy, only
information of her first pregnancy was extracted. In the case where the birth year of the
child is missing, we computed the birth year of the child fiom the age of mother when
pregnancy ended (if reported) and mother’s birth date/age at the time of the survey. We
encountered a small data problem that birth year of the first child is before the year the
women married for 173 cases (out of 4,637 cases). For these observations, we search for
marriage information of the women in subsequent IFLS waves, starting with IFLS2 first.
As a result, 78 observations were recovered. Further, if a woman had never given birth
by the IFLSl interview date (1993), we track them in IFLSZ and IFLS3 whether they
ever gave birth during subsequent IF LS surveys. If they did, information on her first
pregnancy was extracted.

After obtaining data from IF LSl respondents described above, we added women
who appear in IFLS2 that did not appear in our IFLSl sample. These women include
those who were at least 15 years old who became married between IFLSl and IFLS2
surveys and new respondents from split-off households. For these respondents, we
carried similar procedure to IF LS1 to obtain the data. Similar tracking procedures were

also canied out to track these new observations in IFLS3 survey. We had only 45

96

observations that birth year of the first child is before the year the women married. 17 of
these observations are recovered and corrected using IFLS3 data. Finally, we added
observations from IF LS3 that that did not appear in our sample, using similar consistency
checks described.

In the case when a woman had multiple marriages, special attention was paid to
carefully time the maniages for fertility exposure. For example, suppose a women
married at the age of 18 years, ended her marriage at the age of 21 years, married again at
the age of 25 years, and had first pregnancy at the age of 27 years, her exposure to
fertility would be only between 18-21 and 25-27 years. Note that when the information
on the date of the end of marriage was not reported, we compute this date from the
reported age of the woman when marriage ended using also her birth date/age at the time

of the survey.

Duration Data

In the conditional fertility analysis, the entry of women into the duration analysis
was simply the age when they first married. We specify the failure time to be the time
when the woman ended her first pregnancy. In other words, the women in our sample
contribute each year after their marriages to the duration analysis until the first pregnancy
took place (for those who were ever pregnant). For instance, if we take the above
example and know that the same woman ended her first pregnancy when she was 20
years old, this woman contributes to our analysis only from 18-20 years (or two years
after marriage). The exposure from 20-21 and 25—27 years old will be discarded since the

exposures were afier her first pregnancy.

97

Furthermore, since we did not observe the end of the survival period of never-
pregnant women in the sample, the survival data obtained from the survey are considered
“flow” data, which are subject to time censoring. In this case, the data are right-censored
at the interview date.12 For example, if a woman married when she was 18 years old in
the year 1990 and never had any pregnancy by the time of IFLSZ survey (1997), we only
know that the woman’s survival time was at least seven years. We never observed real
survival time of this woman. Thus, this woman only contributes first seven years after
marriage to our duration analysis.

In marriage analysis, we specify the entry age to be eleven years old. Women exit
our analysis when their first marriages took place. For those that never married, similar
procedure for censoring was applied. For instance, if a woman was born in 1973 and was
single by IFLS3 (2000) survey, this woman contributes sixteen years of exposure to
marriage, starting from when she was 11 and ended when she was 27 years old.

In addition to data on maniage and pregnancy history, data from individual,
household, and community characteristics surveys are used to allow this study to control
for other socioeconomic characteristics such as education, province of residence, and
type residence (urban/rural, village/small town/big city) at the time of exposure to
maniage and fertility.

Following a general practice in marriage analysis, the residence of women in our
marriage analysis is the residence of women at the time when they were 12 years old.
This residence information, whether the woman resided in a village, a small town, or a

big city, is assumed to be the residence before the woman married. The residence

 

'2 For those we tracked in subsequent IFLS waves, the interview date is from the last IF LS wave used in
tracking the observation.

98

information in our fertility analysis is the residence of the woman at the time of survey.
We take this to be a proxy for the post-marital residence. The residence of women after
their marriages are thought to be correlated to factors that affect fertility decisions and
fertility control, such as prices and type of available healthcare, more than pre—marital

residence.

Identifying “Crisis” Covariate

In both marriage and fertility analyses, given the time and duration of the
Indonesian economic crisis, we identify the year 1998 as the crisis year. If the exposure
period is in 1998, the crisis dummy was set to be equal to 1. Since the same woman may
be exposed to both crisis and non-crisis periods at different ages, the crisis dummy is a
time varying covariate. Hence, data are constructed so that for those that had an exposure
to the crisis year may have multiple records for different crisis dummy covariates. For
example, in the marriage analysis, a woman born in 1974, whose first marriage year was
in 1999, was not exposed to the crisis when she was 11-23 years. She was exposed to the
crisis when she was 24. Then at the age of 25 years, she married when she was not

exposed to the crisis in 1999.

99

Results

Results on ages of female first marriage

Table 20 presents summary statistics of female first marriage age by birth cohort
from 1921 to 1975. Our data include information fi'om younger females born in 1976-
1986, but their marriage-age summary is not reported here because these women had not
been fully exposed to almost all of the marriage age groups yet. Although we cannot
observe any trend in the age of female first marriage from the median age, a breakdown
of the proportions of ages of women when marriages occurred indicates an upward trend
in female marriage ages. The proportion of marriage ages between 11 and 14 years old
declined from 12.8 percent to 4.2 percent from the oldest cohort to the youngest cohort.
During the same period, the proportion of marriage ages between 23 and 25 years old
increased from 10.3 percent to 20.6 percent. An interesting trend that is consistent with
this increase in the ages of maniage is that the proportion of those who remained single
increased over time. This proportion increased from only 0.7 percent for those born in
1921 to 1940 to 7.1 percent for those born in 1966-1970.

A similar trend can be drawn from observing female marriage ages in different
calendar years. Table 21 presents the proportions of marriage ages of women in different
years from 1988 to 2000. The proportion of women whose first marriage occurred when
they were 11-14 years old declined from 4.2 percent to 1.4 percent from 1988 to 2000
while the proportion of women whose first marriage occurred when they were 31-35
years old increased from 1.4 percent to 4.1 percent. Figure 9 shows results from

nonparametric estimates of first-marriage-age hazard by three birth cohorts, 1900-1940,

100

1941-1960, and 1971-1980. From these estimates, we observe that the hazard rates of
first marriage at younger ages declined over time. For instance, the probability of getting
married at the age of 15 years (if single at the beginning of the year) is 0.12 for those
born during 1900-1940, 0.09 for those born during 1941-1960, and only 0.01 for those
born during 1971-1980. The trend is reverse for older ages, indicating an upward trend
in the ages of first maniage.

Figure 10 and 11 show nonparametric estimates of first-marriage-age hazard by
education. Figure 10 presents estimates of the hazard rates of women who had no
education, 1-5 years, and 6-8 years of education. Compared to those who had no
education, Indonesian women who had higher education had a lower likelihood of getting
married at younger ages (1 1-15 years old), but they appeared to be more likely to get
married afier 20 years of age. Even though the delay in marriage for the higher-educated
groups may be due longer time spent in school, it appears that once they finished
schooling, the probability of getting manied quickly started to surpass that of uneducated
women. More apparent evidence is shown in Figure 11. Uneducated women had a much
lower probability of getting married than those with 9-11 or 12+ years of education in
their 203. Those with 9-11 years of education had the highest probability of getting
married among the three groups in their 203, with an exception of 26 to 28 years old
where the marriage hazard was highest among those with 12+ years of education.
Interestingly, the probability of getting married for 12+ years group remains higher than
other groups in their 303.

The differences in the maniage hazard rates of women with different education

levels may be explained by theories related to human capital investment, opportunity

101

search costs, and expected income gain from marriage. However, explaining these
differences in the marriage trends is not our objective in this paper. The purpose of this
exercise is to demonstrate that education levels influence female marriage hazard rates.
Thus, ignoring education level could cause a bias in our estimations of marriage hazard
rates when studying the effects of the economic crisis on ages of female first marriage.

Figure 12 presents estimated first-marriage hazard rates at different ages by
women’ residence when whey were 12 years old. The IFLS survey stratifies these
residence data into village, small town, and big city. A clear picture emerges fi'om these
estimated hazard rates. Among these three groups, those who lived in a village had the
highest probability of getting married at every age from 11 years up to approximately 24
years old. Urban residents such as those who lived in small town or big city tended to
have a higher probability to marry later in their lives. This differential trend could be due
to different characteristics of urban and rural residence such as consumption prices,
marriage market conditions, or different characteristics of those who live in urban and
rural areas (e. g. education).

Results from parametric marriage hazard regressions using Cox-proportional
hazard assumption are presented in Table 22. In this table, we show results from six
different specifications, all with standard errors of the estimated coefficients that are
robust to heteroskedasticity and clustering at the individual level. In these regressions,
only women born between 1941 and 1986 are included to minimize recall errors.
Moreover, including only those born in a more recent periods forces the control groups
(those who were not exposed to the crisis) to be more comparable to the targeted group

(those who were exposed to the crisis) in the case when birth cohort dummies do not pick

102

up all of the time trend effects.13 The univariate regression (1) shows a positive, but
insignificant estimated coefficient of the exposure to the crisis risk. The estimated
coefficients of the crisis exposure are larger and statistically significant in all multivariate
regressions, suggesting an increase in the hazard of female marriage during the economic
crisis. Over time, women were less likely to get married. The estimated trend is
significant assuming either linear (birthyear) or non-linear (birthyear cohort dummies)
time trend. Overall, more years of education lower the probability of getting married, but
the estimated coefficients from regressions using education level dummies (regression 5
and 6) show that those with 1-8 years of education exhibited higher probability of getting
married than those with no education while those with education higher than 8 years had
lower probability of getting married. Interestingly, the effects of the time trend are much
smaller when we include education in the regressions. The estimated coefficient of the
linear time trend decreases fiom —0.016 to —0.001 when years of education is included in
the regression and to —0.005 when education level dummies are included. These results
suggest that the time trend captures also the increasing trend towards higher education of
women over time. Finally, women who resided in villages and small towns are more
likely get married than those resided in big cities.

Table 23 presents results fi'om the parametric regressions of marriage hazard
using Generalized Gamma distribution of the survival time. For robustness check, we
chose to report also the results from Lognormal distribution.14 The reported standard

errors of the estimated coefficients are robust to heteroskedasticity and clustering at the

 

‘3 We have estimates from all women (born in 1900-1986), but these estimates are not reported here. The
estimated coefficients and their significance are similar to those from women born in 1941-1856.

14 The estimated parameters in these Generalized Gamma regressions suggest that we accept the Lognormal
distribution over the Weibull distribution.

103

individual level. In addition, in all of the specifications, we assume an Inverse-Gaussian
distribution (heterogeneity) of the observations instead of assuming homogeneity of the
samples. The estimated coefficients presented here are the coefficients of the survival
risk. All results are consistent with results from regressions using the Cox-proportional
hazard assrunption except results from the univariate regressions. We find that when the
time trend and other covariates are not controlled for, women who were exposed to the
crisis risk appeared to have had lower probability to marry compared to those who were
not exposed to the crisis. When other factors are controlled for, the estimates change
sign, indicating a significantly higher probability to marry. The estimated coefficients of

the crisis-exposure dummy are statistically significant at 1 percent.

Results on first marital pregnancies

Figure 13-15 and Table 13-15 shows results from the analysis of females’ first
births after marriage, conditional on being married. Figures 13 presents nonparametric
estimates of fertility hazard rates at different time duration (years) after maniage. Our
data give similar results to data from the United States that age at first marriage had a
causal effect on the occurrence of a short first births. (Marini and Hodsdon, 1981) These
trends are also found in developing countries such as China. Feng and Quanhe (1996)
found rising age at first marriage and shortening of the interval between marriage and
first births to be two prominent features of China’s demographic transition during the
past two decades. Our results indicate that Indonesian women who married in 1986-2000

had higher probability of having first births in the first three years after marriage

 

104

 

compared to those who married in the earlier period. Recall our results on ages at first
marriage that marriage ages increased during the same period. Notice that the hazards of
first birth of women who married later in life (the younger cohort) declined faster than
that of those that married sooner. These results are consistent with results that suggest
substantial female fertility decline in their late 203. We observe that those young women
who married later in life tended to have lower probability of having birth than those
married sooner after three years of childlessness.

When looking at the effects of education on the timing of first births in Figure 13,
we observe a significant difference in the probability of having first births among women
with 0-5, 6-11, and 12+ years of education. Women with higher education levels
exhibited a higher probability of having birth in the first three years after marriage.
Evidence of the shortening of first birth intervals among higher-educated women is
consistent with our earlier findings that women with higher education tended to marry
later in life. We also observe that the probability of having first births declined
substantially after two years of childlessness among women with the highest education
(12+ years).

Figures 15 shows nonparametric fertility hazard estimates by residence. Our
results indicate that urban residents had a higher probability of having first births in the
first two years intervals than rural residents. The probability of having first births among
urban residents declined substantially after marriage, resulting in a lower probability of
having births if they stayed childlessness for at least three years after marriage. This

result could be due to delays in marriage among urban residents or other unobserved

105

conditions that are different between urban and rural settings such as quality of health
facilities.

Results from parametric marriage hazard regressions using Cox-proportional
hazard assumption are presented in Table 25. In these regressions, the standard errors of
the estimated coefficients are robust to heteroskedasticity and clustering at the individual
level. We find that married women who were exposed to the crisis exhibited lower
hazard of having first birth. The estimated coefficients are significant at 1 percent in all
specifications. The stimated coefficients of the time trend indicates that fertility declined
over time. The estimated coefficients of years of education and education-level dummies
indicate similar effects to the effects found in nonparametric estimations. Overall,
women with higher education tended to have a higher probability to give births at
different intervals after maniage than women with lower education, provided that hi gher-
educated women married later in life.

Similar results are found in parametric hazard regressions assuming Lognormal
and Weibull distributions of survival times. The estimated coefficients of survival risk
using Lognormal distribution are presented in Table 26. The estimated coefficients of
first birth hazard using Weibull distribution are present in Table 27. We find that having
exposed to the crisis risk is associated with a lower probability of having first births. The

estimates are significant at 1 percent in all specifications.

106

Unconditional Fertility Hazard Estimation

One should keep in mind that the analysis of first births in the previous section
takes ages of marriage as exogenous. The estimation of the hazard rates of first birth is
conditional on the women being married at the beginning of the time interval. Results
from these conditional fertility hazard estimations could be biased when the maniage
hazard was affected by the economic crisis. Our results show that the marriage hazard
increased and the first birth hazard, conditional on being married, decreased as a result of
the economic crisis. From these two sets of results, we cannot tell what the net crisis
effect on first birth was. In principal, we could estimate the correlated hazards by
defining a complicated likelihood function that estimate marriage and first birth hazards
simultaneously. We did not carry out this estimation in this paper.

Instead, we estimated the unconditional first birth hazard rates of women starting
at age 14 years regardless of whether the women were married. In this estimation, the
dependent variable is the odds of a woman having her first child in year t given that she
had no children at the start of year t. A similar procedure used in the marriage analysis is
applied to constructing duration data and censoring process. The hypothesis being tested
is the significance of the crisis dummy when the time trend is included.

Figure 16 compares the nonparametric unconditional fertility hazard rates, at
same ages, of women born in 1941-1960 and 1961-1986. From the estimated results, we
can see that the younger-cohort women were less likely to have their first births at
younger ages and more likely to have their first births after age 25 years compared to the

older-cohort women. Table 30 and 31 show estimates from the unconditional fertility

107

hazard regressions using a Cox proportional hazard model and a Gamma distribution of
survival time respectively. The estimated coefficients of the crisis dummy are negative15
and statistically significant, indicating that being exposed to the economic crisis lowered
the likelihood of having a first birth. This finding and our earlier finding from the
conditional fertility hazard estimates suggest that the net effect of the economic crisis

resulted in a delay of first births, even though there was a movement towards earlier

marriages.

 

'5 positive in the Gamma distribution regressions because coefficients of the “survival” time are estimated.

108

Robustness

In our analyses, we identify the crisis effects through the estimated coefficients of
the 1998 crisis exposure. In the case when birth years and marriage years do not fully
capture the time—trend effects, the significance of the crisis exposure could be due to
unobserved trend effects rather than the crisis effects. We ran a robustness check using
year 1996 dummy instead of using the 1998 crisis dummy in the marriage analysis.
Results of the parametric estimates using the Cox proportional hazard models are shown
in Table 28. The estimated coefficients of the 1996 dummy are negative and not
statistically significant, indicating that our earlier estimates are robust to this potential
problem.

In addition, since the increase in the maniage rates and the delay in first births
may not be only a short-run effect. We tested the effects of the economic crisis by
extending the crisis period to the year 1999 to include a longer-run effect. Results of the
parametric estimates using the Cox proportional hazard models are shown in Table 29.
Similar to previous results using the 1998 crisis dummy, the 1998/1999-crisis dummy is
positive and statistically significant in all specifications when the time trend is included,

suggesting that the crisis effect may not be only short-run.

109

Conclusion

This paper investigates whether there was any short-run change in the ages of
female first marriages and the timing of female first marital births as a result of the 1998
Indonesian crisis. The paper discusses channels to which an economic crisis could affect
these demographic outcomes. The discussions of various theories lead to a conclusion
that the effects of an economic crisis of female first marriage and ages of first births are
ambiguous a priori due to opposing income and substitution effects of a decline in
income and wages.

We found evidence that overall there was an increase in the probability of getting
married and a decrease in the probability of having first births among Indonesian women
during the crisis in both conditional and unconditional analyses. We argue that these
findings support the hypothesis that marriages of individuals in the household and
delaying having births may have been used as income-smoothing mechanisms in the time
of the crisis. Results from this paper are not sufficient to draw any conclusion on why an
increase in marriage probability and a delay in having first birth took place. We
speculate that women are more likely to get married during the time of the crisis to take
advantage of economies of scale and specialization in household consumption. The
delays of first births might be due to consumption-smoothing consideration or other
supply factors such as the separation of spouses when relocation of individuals occurred
during the crisis.

In the case when insurance markets are absent or incomplete, household appeared

to rely on alternative income and consumption smoothing mechanisms other than saving.

110

It is important to realize that even though changing behaviors of individuals in the
household (such as delaying births and decreasing educational investment) is a reliable
way to smooth consumption in the short run, the longer-term effects on welfare of such

actions need to be considered in order to fully assess the effects of the economic crisis.

11]

APPENDIX I

112

 

 

 

 

 

 

 

 

 

 

 

 

 

m.m~ 98 m9”:
.2 no. ..- mm... 2.. ~92
SN 3. NB“:
98 com 3m 3: an em. NB“: 2.... 8....
com 3N SN m.m~ GE
82858.. .8282 82858.. .8282 82858.. .8282 82858.. .8282
re. 3-3 8... To 32 9.882. 88>
8888. 32-83 ~83”... 3832:
88 3... 2 2.2
3m 92 EN QR 3N mm 62. .o; 82.5.. 9...
men «:3 98 SN ~.m~ 3N 682823 8239.. 3”:
tmm 2m in EN EN ...NN 682985 8.38 ......
mom am ~.~m m2 mam m.m~ 8.8.2.23 case... ..an.
8282-88 .8282 82858.. .8282 82898.. .8282
3-3 on to 3m. 9.88:. 88>
$3.82 83.32 3838.
@2252 8.8.2.82 88 mzo 2: cane
QR v.8 «tom 3N 3N min 322.5
Eumcomctumom Kan—.502 _Smcowctumoa Fume—002 _Bmcomctumoa Wane—002
3-3 as to 9.885 88>
52.3% 83-32 33-82

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

828.3... 32 2.0 .2 some

$52.5 0:.— >__Ea..._ :Emocoucn new goo—Sven he >955 29.330: 023960580 ham.— Eot Bee
$5.... 22. 83 83 282258 v.32. 22.8.5: 5282-82. .25 .5282 a some

113

 

 

m3": new .NWIH 4mg“: u8.2-Em

 

 

 

 

88 22 .....mmm ~22 8.2.8. 5.... e888. 5.; .Bo .o .2252
3 ...o ...o m...- 8.20
m2 n2 n2 a: .8582 82.882.88.32.
2... a... c... mm 88 82.8 82.32828 5.8.. 3.82.2.8
o.- 2.2 0.2 2.: 8.8.2. .6 85o .o 2....6 8.22.... ..o 2......
m2 2... o.~ m. 8.8.... .2888. .6 8:28.20
we... m2... «.8 mm... 8:9. 8.82.2. 2.2.8.882 .25
No.82 88-82 32-82 82-82 5.88.. 5....
Eco 50>
3... 5.283 5.... .n use-F
mm... 2.. .3... .22.. .858
..mm as. ...: 2.2. 82.0
...8 new 2.8 8... .8282 82.882.282.22.
Q8 QB 2.8 2.8 .82. 82.8 83.32828 5.8.. 3.5558
2.8 v.8 2.. 2.3 8.8.... .o 8....o .o 2...... 8.22.... .o 2....u
N2... 28 2.2. 5.. 8.8.... .2888. .6 88282.0
2.3 3... mm .....m 88.. .888... 2.2.8822. ..zo

 

flee... Sow-m8. 32-82 82-82 5:83 5....

Eon 50>

3o. 5.88. 2t... .5 Eadie... 8.282 2:8. 85 2...... co 5.9.2.2... .N 282

114

mum-r: 0cm NB“: 4mm”: "8.30m

 

 

 

 

2m 2m 8.. m8. o2 to... .2 .80 ..o .8252
...o .3 N... N... 3 N... m... 22.0
0.2 8.2 Q2 2.2 NE N2 ..2 .8582 82.88.228.32.
m... ..m ~.m am 0... o... o...- .ooo 82.8 82.32828 5.8.. 3.8528
No... ...2 2.2 3N QR o.e~ 8.2 8.8.... ..o 8.....o .o 226 2.8.2.... .o 256
o... 3 ~.~ .... .2 ~.~ m... 8.26.... .2288. .o 8:28.20
3.. 2.2 to... ....om v.2 m.~m ...mm 88.. .8852. 2.58822 225
88 82 85 85 R2 82 82

mate-:02 3083:990 _m_ocmc_t._ m.m_.ocoz cornea.— 5.2m

Eon 23>

 

3... 2822.28 3.83.. 5.... .e 833

115

Table 5: Mortality Rate (per 1000 live births)

 

Neonatal

Post-neonatal

 

mortality mortality Infant mortality Obs
Born during financial crisis 23.1 23.1 46.2 1040
rural 25.4 29.0 54.4 552
urban 20.5 16.4 36.9 488
Born during 97 drought period (rural) 29.0 47.1 76.1 276
Born during non-crisis period
1988-1992 28.2 33.0 61.2 4117
rural 30.6 38.7 69.3 2324
urban 25.1 25.7 50.8 1793
1993-1997, 1999 13.3 19.2 32.5 3542
rural 15.7 25.8 41.5 1976
urban 10.2 10.9 21.1 1566

 

 

 

 

 

financial crisis = January 1998 to June 1999
drought97 = May 1997 to December 1997

non-crisis = January 1988 to April 1997 and July 1999 to December 2000
Obs include children that had a chance to live >=356 days by the interview date.

Table 6: Birthweight Statistics

 

 

 

 

 

_ Mean Birth °/o of Low
. Birthweight(<2.5 Total Obs
Weight (Kg.) k 3)
Born during financial crisis 3.13 8.7 841
rural 3.17 6.9 390
urban 3.11 10.2 451
Born during 97 drought period 3.16 6.7 386
rural 3.15 8.1 186
urban 3.17 5.5 200
Born during non-crisis period
1988-1992 3.15 9.4 1568
rural 3.11 11.1 624
urban 3.17 8.3 944
1993-1997, 1999 3.18 8.1 2635
rural 3.19 8.4 1245
urban 3.17 = 7.7 1390

 

 

 

 

financial crisis = January 1998 to June 1999
drought97 = May 1997 to December 1997

non-crisis = January 1988 to April 1997 and July 1999 to December 2000

116

82.82 8.2.58
.38 28.23... a... .5 28 ommuA 8.. 3 886 m 8.. .2. 8.2.6 822.. 30

 

 

22 8.2 m... 3. 28.. +2
82 m9 .22 MN 28> :8
88 $8 m.o~ m.m~ 98> a...
2.2 as... 3.. 32 m8.» m;
82 BK ...? 2...... :8 oz
.30 £33.). 88.. 3.332 53.2,.

582888 .8882

 

cabs—gum 9.5...qu >3 $553 03. 25.." .53 v.35— >u=mtoz 230 "a 03m...

.33 26.23:. 2: >5 m>mu ommnA m2. 2 855 m no: .2: 5.2.5 .o 2230:. 25.2.. 30

 

 

So fiNm odN Ndm Q: mg... Nd mmmfi
mmm hKN mKH 04m ad” N.m BK mama
w: 2mm 55 Ndm Q3 fim 0N nmmfi
wmm flow m.mn mdm 0.3 fim MK 83
¢m> mdm 0.: «.mm 98 md NK mmmH
mam mdN mfg gem 2mm 0.: 0.0 vmma
9K NKH Tm.“ odm mNN 5.: H6 mmGH
an HR“ m.- mdm odN méH m.m NmmH
~mm mdfi fimfl Hdm NKN N3 m.m am.“
new m.mH NS mdm vém m.mH 9m 82
m¢m 0.3 mm mdN m.~m N.mH m.m mmmfi
0R m.~fl mK Ndm m.mm Ndfi o.m 83
30 Ga. Emo> +~H Ao\ov 28> 3.6 Ao\ov 28> wb Ao\ov 38> mé Ao\ov :8 oz $.85 :82 30>

 

iguana—um 9.5502 "h min...

117

......0. ..00 0:0

00.0.... 8.2.005 .8000 0.03-:00.- >0 E0058 002083.020 00200320205 0583.020 .0.
0.02800 0. .93.. 0503 00.05.50 003 0500 055 05 .00 0:030:50 .0: 00> 0050:580500

...>u._m:00:. 0:0 >t0>00 5 0:0E05000Z 05 .0. 0:0 00:05:50 0000:0000 5.. 00:05.5 30.00050.

.3003 00.030 00>>-:00n 0:0 :000.>00 .0003”. :5... 0. 0000300 0:00:50 05 5.. 500.250“.
000.0 5: 00 00230 05 00:... 00.00.05 3 :000
0000 000.0000 - 000. >30 0:0 002 .....< - 002 00:00..- " 22.9.02
000. 0050000 - 02.. >0: u 22.0 .0005
000. 0:2 - 0000 000:8 n 22.0 .0002...
080.00.... 0.0 0:0 080.00.... 0.. 00003.00 2.0.035... 000.0... 2.0002030
.0505. 00:05:50 00:00:00.0 .005 0:800 05 5.. 5030.020 020055830 05 5 00:05.50 0.0 0. Nnm
A2005. 00:05:50 8000:0000 .005 00.: 05 5.. 50350.0 02003:..an 5 00:05.50 0.0 0. Hum
.55. 550:. 05 00 05.00020 .0000. 05.5 0.0053

 

 

 

 

 

.0002
.00.... 0:0 .00.... .. 0...: ”00.000
000 ”2000002 000 ”22.0002 0000 ”22.9.02

X: ”20.5 .2200... 000 ”20.5 5000...... 000 ”22.0 .2800... 000002800 ....
.0 F00. 00.3. 80.00. 00.00 .000: 00.00- .0000. ...-.00- .000: 00.00 .0000. 00.00 0.0
.000 0 00.0 F- .0 000. 00.0- .000 F. 00.00- .0000. 00.0 F- £0.00 00. F0 .0 F .00. F000 0.0
.000 F. 00.0 F- .0000. 00.0. .000 F. 0.00- $0.00. 00.0 F- .000 F. 00.00 .0000. 00. F0 0.0
$0.00 0.. F F- .0000. 00.00- .000. 00.00- .00. F0. 00. F0- .0... F F. 00.00 20. F0. F000 0.0
.000: 00.0- .0000. 00.00- .00.... 00.00- > 8.00. 00.00- .000: 5.00 .000: 00.00 0.0
.000: 00.0. 20.00. 0F.0F- >000. 00.0? .000: 0...:- .00.0. 00.3 .000: 00.00 0.0
.000. 00.0- .00. F0. 00.0- 600. 00.0 F. .000 F. 00.0 F. >00... 00.9 80.0 F. 00.00 0.0
. F00. 00.0. 30.00. 0 F. F F- .000. 00.0 F- .000 F. 00.00- .000. 00.0 F .000 P. 00.00 0.0
.000. 00.0- .000 F. 00.0- .000. 00.0 F- .0 F. F F. 00.00. .02.... 00.9 .000 F. 00.00 0.0
.0 F0. 00.0- .000 F. 00. F .00. 0 00.0- 80.0 F. 00.00- : F0. 00.0 .000 F. 00.00 F0
.000. 00.0- .000 F. 0 F0 F- .00. F. 00.0- .000. 00.00. .000. 00.0 .00.. F. 00.00 0.0
.00. 0 00.0- .0 F0. 00.0- .000. 00.0- .000. 00.0 F- .000. F F... .000. 0F... F 0. F
.000. F00- .000. 00.0- 20.0. 00.0- .000. 00. F F- .0... F. 00.0 .000. 00..- F 0. F
.000. 00. F- .000. 00.0 F- .000. 00. F- .000. 00...- >0F. F. 00. F 80.0. 00.0 0. F
.0 ..0. F00- .00. F. 00.0- :- ..0. F00- .00. F. 00.0. .000. 00.0 .000. 00.0 0. F
80.0. 00.0 .00. F. 0 F0. 800. 00.0 .00. F. 0 F0. 300. 00.0 .000. 5.0 0. F

00v: 0:000. .0 0.500

- - 000.0 09.0 - - .030. 90.0 - - .0000. 000.0 .500 0580.0 .2...

>00. 0 n 0 .00. F u 0 >00. 0 u 0 >00. F u 0 400. 0 u 0 400. F u 0
0.2.9.02 . £ch acm3000 m_thCOZ i 9th .chmcE 90.09002 i £th .mficmcfi.
.003“ cunt:

 

 

«00:05:50 0:00:55 .005 0:0000 0:0 .050 00330 :00350 0000550 0:0 .500 05000.0 .05”.

02.05500 0000:0000 .005 0:800 0:0 00.... 20.0.5550 .0 030..

118

Table 10: Birthweight Regressions

 

 

 

 

 

Birthweight (Kg.) Urban Rural
(1) (2) (3) (4) (1) (2) (3) (4)
OLS weight weight weight weight weight weight welg' ht weight
t -0.002 -0.001 -0.001 -0.001 0.005 0.005 0.006 0.005
(0.44) (0.33) (0.25) (0.20) (1.11) (1.14) (1.29) (1.06)
preg crisis1 -0.043 -0.043 -0.049 -0.025 0.000 -0.003 -0.002 0.009
(0.97) (0.98) (1.11) (0.52) (0.01) (0.06) (0.04) (0.17)
preg crisisZ 0.032 0.031 0.028 0.046 -0.005 -0.012 -0.015 -0.013
(0.85) (0.80) (0.74) (1.05) (0.13) (0.30) (0.37) (0.27)
preg_c597 -0.064 -0.062 -0.061 -0.073 -0.01 1 -0.016 -0.023 0.01 1
(1.46) (1 .41) (1.40) (1.46) (0.27) (0.38) (0.57) (0.22)
male 0.073 0.074 0.069 0.070 0.067 0.066 0.064 0.055
(3.53)” (3.57)“ (3.37)“ (3.04)” (2.76)“ (2.76)“ (2.67)“ (2.10)"
Mother's edu
1-5 yrs -0.094 -0.082 -0.088 0.103 0.064 0.017
(1.26) (1.14) (1.12) (1.64) (1.00) (0.24)
6-8 yrs -0.129 -O.107 -0.117 0.049 0.019 -0.037
(1 .74)* (1.49) (1.53) (0.84) (0.33) (0.55)
9-11 yrs -0.106 -0.085 -O.128 0.118 0.088 0.022
(1.41) (1.19) (1.65) (1 .89)* (1.39) (0.30)
12+ yrs -0.103 -0.076 -0.102 0.056 0.012 .0951
(1.42) (1.10) (1.34) (0.93) (0.20) (0.71)
Constant 6.333 5.712 5.213 4.947 -6.818 -7.159 -7.936 -7.729
(0.86) (0.76) (0.7% (0.59) (0.76) (0.80) (0.90) (0.75)
Province dummies No No Yes No No No Yes No
Community dummies No No No Yes No No No Yes
Observations 3293 3293 3293 3293 2701 2701 2701 2701
R-squared 0.010 0.010 0.030 0.200 0.000 0.010 0.030 0.220
p-value(crises) 0.196 0.222 0.207 0.192 0.994 0.981 0.945 0.973
p-value(mothel‘s edu) 0.510 0.629 0.517 0.192 0.256 0.450

 

 

 

Absolute values of t-statistics robust to heteroskedasticity and clustering at mother level in parentheses.
* significant at 10%;“ significant at 5%; "’ significant at 1%

Observations include all Iivebirths born 1998-2000.

preg_cs1 = Pregnant in January 1998 - September 1998

preg_cs2 = Pregnant in October 1998 - June 1999
preg_cs97 = Pregnant in May 1997 - December 1997

"No-education” and ”North Sumatra" are omitted categories.

119

 

 

 

.motoucfio $3.50 2m ..wbmcsm 5.02.. 9.1. ..coawoauoéz.

km 000 .. #0 >32 n hmflwto .mm 955 .. mm 8 n Nmfito .mm «a8 .. mm :0... n wfimto

.2. ,6 9.2 non m8. 8 9 883 295 9.3 ooom - moor Eon em; .2. «.2222. ..m 83.2. 2082.820
.2; .m 28...ch ... 38 on 28....ch ......or a 285:9.» .

8935222. c. .05. .059: «a 9.20620 can 363388905: 2 «959 mermafimé ho mas-g 23.32

 

 

 

Food 83 ooo.o mood Boo ooo.o 2&8 «88 8nd :8 «8505822...
81o mood Sod goo mood mono memo ammo Sod 32o mood Sod .82832...
oooo omoo ovoo opoo oBo o~o.o o.o.o ooo.o oBo o.o.o o5.o ooo.o 8.28-3.
own. «now «now «now com 58 Eon Boo So «now «now «now 20.8388
m0> 02 02 OZ w0> 02 02 OZ m¢> 02 02 OZ mEEEDU 3:35:58
02 wm> 02 02 OZ mm> 02 02 OZ wO> 02 02 8.5530 00:39:...
...828 i888 5888 2:313 LBS 2.888 .828 .2888 2..me ...dmé ...:«8 2.888
55.. 9a.» 53. 88.8 «8.9 «68 main meow 8th >86 5:. «mum E928
....8.8 2.828 2:368 ...: 28 .25.: L98 .85.: Co. 5 81:
Ergo. o8? oven- 32o. 33. ~88- vote- moo. P- 88... 2. +2
88: 8!: .89: 88.8 8...: .89: 818 818 8.5.8
owed- 53- m E8- mood- owed- «mod- «8. P- «mod. mood- 9» F to
8m. 5 80. s :88. c .88 .8. s 8... 5 8 P. s .38 .8. 5
«8.? 28m- «Noe. wood- or 24.. God. oood- 8.. F- 88. F- m... o6
888 .88 23.8 .88 .88 8’8 8&8 88.8 8.88
t. to 82o- 2. F. F- own.” om to ~88. oohm- owed- 838- m; m;
8o to :8 «.8505.
.88 88.: Ch. 5 8m... 88.8 .8. s .2. : 8P. 5 88.8 8'. 3 Go: 8o. 5
Nov; 888 8.3 o3.o mow. r mono «Bo ~58 8 3 82o .28 onto So to 8.2.
828 $.28 828 818 888 £08 828 88.8
Boo. owmo- 81o. onto- ~88. oewd. 83. Sad. .83 220 B Eon
.3. 5 8m. 3 .88. 3 .88. : .88 .38 8o. 5 8o. 3
83.8 who; 83 wood .88 End n5.2.8 too 825 8.20 EB
88. 5 .828 2.88.8 .888 .88 E., 5 8o. 5 co. 5
83 8.3 «88 mono mood E... P on: 5: 8o to 3.20 Eon
8m. 5 ..8. 5 8o. 3 ..B. c
Kc... omod word 2 5.. 89.8 38:83
898 88.8 .88 898
onto 98o mood £88 898 ~88 8:.
8m. 5 .8». s .88. 8 L 5. c
Sod o3: $5 8: 8o :8 E20 8:.
fiiiflmo.MV iiiAhfl-MV O‘iAmmomV i.‘A°V.Vv CiAGO-NV CCA Fo.Nv ..Am F .NV CCtAmm.NV CfiAmmumv ..CAowomv $.8AmN.”v ¢§iAmmomv
mood. oono- Boo. Noto- ammo- E. to- E. to- «mud- 83- find. wand. :88. 895 .
E9... 29:. E9... E9... .9282 .9282 .9282 .9282 .9282 .9282 .9282 .9282 5......
uumOQ ..umOl uumOQ unman—

.8

88

.8

A:

 

:8

.8

Amy

E

 

.3

.8

.8

E

 

 

9.3:: m=£mmohwo~a banana...— 5354 3:3qu :— 035,—.

120

.motoooomo BEEO 2m .2955 5.02.. now ..coomosooéz.

km 000 . hm >52 u bmflmto .mm 0:2. . mm «00 u No.28 .mm 30% - mm on... u 3.3.0

6.... .0 «8.6 mon ammo. «m 9 “.3098 203 0:; Son - woo v Eon 0.0.... 55 93229.. ..m 25.0... 9.2.9.030
.2; 8 E858... ... ......“ 8 E8588 .....on E E858... .
8085223 ... .05. .859... «m 9.29820 new 3.03388222 9 .250. 3.8282. .0 mo:_m> 23.32
6032 one Etna 05 35.0508 omtooom

 

 

 

F Poo ooo.o ooo.o Koo Food ooo.o F Fno Sud ooNo 88 228E82aflm.
9to nmoo Eco nmoo «88 ano .83 ..mno nnn.o o..o.o «mod «moo 8282.2...
938 .88 958 nmoo ..ooo ooo.o ooo.o «moo nnoo ooo.o omoo omoo 8.28-”. 8.8..
Noun- 5.3...- .tnon- 888. 8...... 7 F town- ..tnnn- n tmmn- mnfl. T ..mdnn- Eden- Bonn- 805.2383
now. «now «now «now own .nnn Eon Eon 5o «8.. «now «now 208820
mm> 02 02 OZ m0> 02 02 OZ mm> 02 02 02 «50.. 5:020:00
OZ m0> 02 02 OZ m0> 02 02 OZ mm> 02 OZ mEEEDO 00:30.5.
$.88 ....on8 ....oo8 LS8 ....oo8 ....on8 .3. 8 8...: 2.5.:
mono 88o 83 ~88 52o onto 8nd ....to 3nd 9.. +3
.88 :t: LS8 .88 8o. 3 ...no8 .88 .88 858
ooo.o n58 unto mono owto mmno Fnho ooo.o ooho m... :8
8n. 5 8...: 2.88 .68 ..t: .88.: 8m... .88 .88
ooo.o ~88 ..nto «So «nod 93 onmo nnoo vnto n... o...
.F 28 8 F8 .88 ......8 .88 828 .88 888 .28
ono. F omo. F .88 nnt P no F. P .88 ~88 ooo.o no; n... 3
Sum $50505.
.88 ...n. 5 8t 8 .nm. 8 .nt8 8t 5 .no. 8 .9. 3 .88 .3. 8 .mo. 8 .3. 8
no... onn; oon. . .. 5.. on t F nnn.. an. nnn. F on E n 5.. now. P mow. F 22.
888 .68 .58 .38 888 .58 .88 .88
wood End .88 ammo 94.8 .38 8.3 888 8.20 3 ...3
.88 ...t 5 .. B. s .8... s .38 .38 5.8 .58
on... in; ovwd F Few ono. F non. F «8.. nnn.. 8.20 ...2
8t 8 2.8.8 2.8.8 ....n...8 .88 ..no. 5 .8... 5 .8... s
or tn nmod neon nan 3n. coho .8... Food 6.20 Eon
..mn. 5 .....n8 .... 3.8 ...ot8
«in moon moon Bhn 88:8...
.88 .nt8 .nt8 .nt8
38.. n3... nnn. F 8.... F 8.20 8...
8n. 5 .. 3. 5 2.88 ...oo8
«EN noon {an «own 520 8...
5.88 5888 ......tn. 5.38 2.88 ..no. : 2.8.8 .....88 .....t8 ....rn8 ....on8 $.68
ooo.o nono ooo.o ooo.o wood Boo nooo ooo.o Rod .88 8.3 and .
E9... E9... E9... E9... .9282 .9282 .9282 88:82 .8982 .9282 .8282 .9282 ..mm.
.«mOn. namol Juan. ..amoa

.8

.8

.8

.5

 

.8

.8

.8

.:

 

.8

.8

.8

.5

 

 

A5385 n:o_nnu..wo~— 0.3%.... 3:393). "N— 03w...

121

00.500000 008.50 0.0 ..ng00 5.02.. 000 ....0000300-02:

00 000 .. 00 >05. u 500.08 .00 0:2. . 00 «00 u N803 .mm .000 - mm ..0.. n 00.0.8

.0... .0 0000 000 .000. E 0. 00000.8 0.02. 0...... 080 - 000. ...00 0.0; E... 0.000.200. ..0 000.0... 00009.38
.0. E 0.000.090 $- 300 E E85090 $300. E 0.000.090 .

000055.00 ... .0>0. .050... .0 05.00020 0cm b.0_.00000.00.0.0.. 8 .0300. 00000000.. .0 00:.0> 03.002

 

 

 

 

 

 

00.8 088 ooo.o .oo.o ooo.o ooo.o . 0mo.o 0o to 00.8 :00 0.050... 08.0..
088 038 00o.o 00o.o 0008 088 00.8 00.8 00.8 0.0.0 0.0.0 0.08 0005.029...
o0o8 o0o.o o.o.o o.o.o o0o8 o0o.o o.o.o ooo.o o0o8 o.o.o ooo.o ooo.o 00.00008
0.00 8.0 00.0 8.0 0000 0000 0000 0000 .000 8.0 8.0 8.0 000002800
00> 02 02 02 00> 02 02 02 00> 02 02 02 00.5520 3755800
02 00> 02 02 02 00> 02 02 02 00> 02 ,02 00.5535 00:39.1
$.00.8 $00.08 $.0t00 $.80. ..00. .. ..00. .. ..00. .. $.00.8 $.00.8 $048 $0000 $0000
000. . . 000.0 o0t0 000.0 000.0 0 .00 000.0 0.0.0 000.0 000.0 000.0 000.0 E90000
$.0o8 $.0o8 $.00.8 $.0t0. $.0o.0. ...-.008 .008 .00... .00...
000.0. B00. 800. 000.0. 0.0.0- 000.0- 0on0- 000. .- 000. .- 0... +0.
.008. .0 t .. .00... .008. ..o. .. .00... .0 to. .008. .008.
0008. 000. .- 000.0- 000. .- . . t .- o00. .- 0008- 30.0- 808- 0... . ..0
.on. .. .00... .00... ..0. .. .00... $.00.8 .008. .008. .008.
000.0- 000. .- 03.0- 0.0.0- 000. .- 000. .- 00...- 0008- 0008- 0... 0.0
.008 .008 .008. .00... .00... .. t .. $.00.8 .. .0. .. ..00. ..
0008- 0008- 0008- 0000 000. . 000.. 000.0. 000. .- 000. .- 0... 0-.
80.... 000 0.0.00.2
$808 ...-.008 $.00.8 ...-.008 $.00.8 $.00.8 $.00.8 ...-.008 $.00.8 $80.0. $.00.8 $.00.8
000.0 0o0.0 .000 000.0 0 .00 .00. . o0t . 0 .t . 000.0 000. . 00o. . 00o. . .oo .8 0E...
$.00.8 $.00.8 $.00.8 $.00.8 $.00.8 $.00.8 $.0o.8 $.00...
000.0 000.0 000.0 000.0 000.0 8.0 000.0 000.0 .oo.x. 0.0..0 00 ...00
$.00... $.80. $.0o.0. $.00.8 ..08. 808 .008. .008.
000.0 000.0 000.0 000.0 0008 0008 0008 0008 .00.... 00.0.... 0.00
.008. ..0. .. ..0. .. .00... .00... .00... .00... .00...
000.. 000.0 .00.. 000.0 000.0 000.. 000.. 000. . .oo.... ...-.000 ...00
.008 .008. .008. .008.
0.0. .- 0 .08 0008 0008 .02... 008100...
..00. .. ..00. .. ..00. .. ..00. ..
000.0 00.0 00.0 00 ..0 .00.... .00.... 00.0
...-.008 $.00.8 ...-.0 . .0. $. .00. ..00. .. ..00. .. ..00. .. $.00.8 $.00.8 $.00.8 $.00.8 $.00.8
.008. 0008- .008- 0008- 0008- 00.8- 00.8. 0008- 0008- 00.8- 00 to. 00.8- .005...
E9... E9... E9... E9... .9282 .9882 .9282 .9282 .9282 .9282 .9282 .9008: .20..
$008 -0008 Amon— 0001
.0. .0. .0. . .. .0. .0. .0. ... .0. .0. .0. ...

 

 

 

 

20.5.... 2.0.8203. 00:39.0...— .00...-. 00:00.02 .0. 0.0....

122

$93980 3580 2a .ngaw 5.02.. 9.6 2382362..

km 000 - hm >mS. u hmflmto .8 0:3... . mm «00 u Nmfito .wm «now .. mm :03 u 33:0

.95 Fe gnu men .82 .a 9 namoaxo So; 2.3 coca - 89 E8 203 $5 29222.. __m 25.2.. mcoamzomno
.x.F a 2.85ch 3% a 2.85ch ..uoxooF 8 E855.» .

.mmmoficosmn E _m>o. .550E 5 05.833 ucm gozmmcoxmohoaoc 2 $39 3235?. F0 329 33.9.2?

:9.an woo Etna 9m 8:92:08 8:83..

 

 

 

 

 

we; owed Bod Food ooo.o ooo.o «mod 8 3 2.3 38 mnm£oEwoam>u
mood F 86 Sod m Fed 356 89o ~86 Bod ~86 god awed ooo.o Amomtovoagh
muod owed «Nod vFod mmod ooo.o mnod NFod mood omod owed «Fed nemaawi 83mm
3.33.- 83o F- 5.23 F- 93.8 F- 8.an :38. 3.8»- minus- Edna- 8.3m- 9.36- 9.3m- 265.3ij
9mm SE 85 86 RR 3.3 2.3 3.9. Feuu 85 85 86 23.9.88
«9 02 02 oz mo> oz 02 oz 2; 02 oz 02 :9: 3:86:00
02 m0> 02 02 OZ m0> 02 02 OZ mm> 02 OZ mQEEJ—u 8C_>O..n_
CCHm Pom“ CCCHNG.NV CCCA qua CCAgu PV CCC Ffiomu CCCHbN-mu Aowoov AE- Flv AwV. FV
mete «and Fund mind mmud vowd ooo.o Etc 31o 2» +~F
ENS 2%.: 81: the 83: 8!: God. :68 809
98.0 8K5 Emd momd mend wand mFoF mine 8&0 as F Tm
an. FV an. FF .ANF. FF 2w. FF .fio. FV :SFNV 23.8 8nd. 8?:
8nd ciao no; mwmd 56.0 «Bo «Ed mmmd 89o ms 9o
awe sud. fine .69 FF 6v. F. AF F. s 58va :8 F .S .fim. FV
mmmd mead v85 81F 8m. F mum. F vote 3.3 oFmd 9» m-F
:3 95522
CCCAmN.mV CCCAwm.MV CCCAO®."V CCCAE.”V CCA Pv. NV CCCAmb.NV CCCA FF. NV CCCA Foumv CCAm°.NV CCAVnuwv CCAwmnmv CCA Fv. NV
own. F 8m. F 3m. F Fan. F F Fm. F moo. F 8m. F mom. F «mm. F mom. F Rm. F 8m. F 0.9:
CCCAmo-mv CCCAg. ”V CCCA Fm.NV CCCAhonuv CCCAmw-Nv CCCAQhoNV CCAmmqu CCAgomv
«mad «and Bud 9nd Sad End «Fwd mead mate 8 Eon
L FNQ :6de Loud ... FNNV and. and. £98 Cod.
unfim 83 33 83 2. F.F SNF me. F 2%. F «mate 58
899 .3. FF 8N. FV AX”. FF 35. FF an. F. fin. FF .3. FF
Eh. F 83 mm? 30F m F3 «5. F «2.. F 5F. F 33.8 EB
:69 :2: 3:8 :VFQ
8nd mm F. F m3. F «8. F 3819a
..SFS :6de ..SFS L88
«mod «awn Stu Stu Fmato ooa
OCCAmv.mv CCCAON.MV CCCAQ r.mv CCCAmm.MV CAN“. Pv CAmm. FV CAOQ. vv CCCANN.NV CConuwv CCAOV. NV CCAhchv CCAhnowv
93. 03.0 med F Fmd Rad ~3d mvmd mFmd «Fad mmmd ommd ammo .
Eat. =5... 2ch. E35 .9282 .3282 .9282 .9282 .5282 .9282 .3882 .8282 :mfl
numOn. uamOQ numOQ Aman—
E 5 Q E E 5 Q A: i 5 5 A:

 

 

 

 

29:—Ev 2230.33— 9.33::— hEatc—z u: 039—.

123

Table 15: Infant Mortality : Low Education and High Education

 

 

 

 

Urban Mother's education = 0 - 5 years Mother's education >= 6 years
LPM (1) (2) (3) (1) (2) (3)
t (x100) -0.887 -0.735 -1 .693 -0.246 -0.234 -0.558
(3.02)“ (2.51 )" (2.59)“ (2.29)“ (2.19)" (1 .80)*
born crisis1 (x100) 6.062 3.844 8.318 2.826 2.874 5.067
(1.17) (0.78) (0.60) (1 .84)‘ (1 .86)‘ (1.19)
born crisis2 (x100) 5.417 2.635 -0.881 1.472 1.332 3.329
(1.17) (0.50) (0.12) (1.22) (1.10) (0.91)
born cs97 (x100) -2.142 -1.010 2.532 -0.277 —0.370 0.507
(1.44) (0.58) (0.57) (0.22) (0.29) (0.12)
male (x100) -0.756 -0.557 -3.042 1.418 1.440 3.063
(0.46) (0.33) (0.93) (2.42)" (2.46)” (1 .91 )*
Constant 17.731 14.689 33.864 4.921 4.669 11.173
(3.03)“ (2.51)" (2.60)“ (2.29)" (2.19)" (1 .81)’
Province dummies No Yes No No Yes No
Community dummies No No Yes No No Yes
Observations 955 955 489 3077 3077 1 102
R-squared 0.010 0.040 0.214 0.000 0.010 0.119
p-value (crisefl 0.024 0.493 0.872 0.207 0.145 0.600

 

 

 

 

 

 

 

 

 

 

Rural Mother’s education = 0 - 5 years Mother's education >= 6 years
LPM (1) L2) (3) (1) (2) (3)
t (x100) -0.413 -O.401 -0.490 -0.377 -0.387 -0.756
(2.12)” (2.08)" (1 .98)” (2.77 )*“ (2.88)“ (3.60)“
born crisist (x100) 2.798 2.586 1.747 1.579 2.043 2.248
(0.92) (0.85) (0.46) (0.98) (1 .26) (0.96)
born crisis2 (x100) 3.418 3.045 1.452 3.238 3.025 7.262
(1.10) (1.00) (0.43) (1 .70)‘ (1.60) (2.07)"
born c397 (x100) 6.129 6.747 11.173 3.447 3.428 6.683
(1 .55) (1 .72)‘ (1 .87)’ (1 .86)* (1 .86)* (2.42)“
male (x100) 3.029 2.832 3.581 1.908 1.957 2.074
(2.65)“ (2.49)“ (2.55)“ (2 .44)" (2.50)" (1 .77)‘
Constant 8.280 8.038 9.819 7.554 7.732 15.127
(2.13 ** (2.09)" (2.00)“ (2.78)” (2.88)” (3.61)“
Province dummies No Yes No No Yes No
Community dummies No No Yes No No Yes
Observations 2352 2352 1861 2748 2748 1766
R-squared 0.010 0.020 0.110 0.010 0.020 0.145
p-value (crisfl 0.272 0.367 0.290 0.140 0.128 0.027

 

 

 

 

 

 

Absolute values of t-statistics robust to heteroskedasticity and clustering at mother level in parentheses.
" significant at 10%;“ significant at 5%; *** significant at 1%
Observations include all individuals that were born 1988 - 2000 who were exposed to at least 365 days of life.

crisis1 = Jan 98 - Sept 98, crisisZ = Oct 98 - June 99, crisis97 = May 97 - Dec 97

"North Sumatra' is omitted.

124

Table 16: Neonatal Mortality : Low Education and High Education

 

 

 

 

Urban Mother's education = 0 - 5 years Mother's edumtion >= 6 years
LPM (1) (2) (3) (1) Q) Q)
t (x100) -0.224 -O.135 -0.536 -0.243 -0.243 -0.904
(1 .18) (0.71 ) (0.69) (3.29)“ (3.35)“ (3.03)“
preg crisis1 (x100) 1.945 0.829 -7.923 1.732 1.712 7.192
(0.59) (0.24) (1.16) (1 .86)’ (1 .84)‘ (1 .79)‘
preg crisisz (x100) -0.970 -1.076 -11.483 1.087 1.281 5.205
(0.73) (0.75) (0.98) (0.98) (1.16) (0.76)
preg_cs97 (x100) 2.438 2.012 -2.699 2.068 2.103 9.298
(0.64) (0.54) (0.21) (1 .78)‘ (1 .85)* (1.54)
male (x100) 0.430 0.602 2.889 0.432 0.429 1.062
(0.44) (0.62) (0.85) (0.98) (0.97) (0.55)
Constant 4.494 2.690 10.743 4.862 4.847 18.067
(1 .19) (0.71) (0.69) (3.29)” (3.35)” (3.04)”
Province dummies No Yes No No Yes No
Community dummies No No Yes No No Yes
Observations 955 955 296 3077 3077 703
R-squared 0.000 0.020 0.207 0.000 0.010 0.107
p-value (crises) 0.509 0.671 0.655 0.095 0.066 0.149

 

 

 

 

 

 

 

 

 

Pural Mother's education = 0 - 5 years Mother's education >= 6 years
LPM L1) (2) (3) (1) (2) 13)
t (x100) -0.231 -0.237 -0.495 -0.095 -0.098 -0.444
(1 .94)‘ (2.00)" (1 .96)‘ (0.86) (0.90) (1 .77)‘
preg crisis1 (x100) 1.728 1.780 2.091 1.892 1.730 6.338
(0.97) (1.01) (0.66) (1.15) (1.06) (1.47)
preg crisis2 (x100) -0.675 -0.522 0.427 -1.975 -2.109 -5.649
(0.83) (0.61) (0.19) (2.74)“ (2.74)" (2.17)“
preg_cs97 (x100) 1.808 1.718 4.151 -0.862 0584 -3.543
(0.91 ) (0.85) (1.02) (0.82) (0.56) (1.97)“
male (x100) 1 .258 1 .261 2.425 0.873 0.878 1.659
(1 .91)* (1 .91)‘ (1 .86)‘ (1.49) (1.49) (1.25)
Constant 4.611 4.738 9.886 1.917 1.956 8.908
(1 .95)‘ $01)“ U-9§X‘ (0.87) (0.91) (1 .78)‘
Province dummies No Yes No No Yes No
Community dummies No No Yes No No Yes
Observations 2352 2352 1 1 19 2748 2748 1127
R-squared 0.000 0.010 0.104 0.000 0.010 0.159
p-value (crises) 0.298 0.400 0.678 0.002 0.004 0.018

 

 

 

 

 

 

 

Absolute values of t-statistics robust to heteroskedasticity and clustering at mother level in parentheses.
* significant at 10%;“ significant at 5%; *“ significant at 1%
Observations include all individuals that were born 1988 - 2000 who were exposed to at least 365 days of life.

crisisl = Jan 98 - Sept 98. crisisZ = Oct 98 - June 99. crisis97 = May 97 - Dec 97

"North Sumatra" is omitted.

125

Table 17: Post-neonatal Mortality : Low Education and High Education

 

 

 

 

Urban Mother's education = 0 - 5 years Mother‘s education >= 6 years
LPM (1) (2) (3L (1) (2) (3)
t (x100) -0.709 —0.635 -2.113 -0.030 -0.023 -0.017
(3.00)” (2.69)“ (2.58)“ (0.38) (0.29) (0.04)
born crisis1 (x100) 4.076 2.414 17.717 1.130 1.154 -0.191
(1.05) (0.63) (0.92) (1.00) (1.00) (0.04)
born crisisz (x100) 3.801 1.747 7.489 -0.063 -0.139 -2.357
(1.10) (0.43) (0.79) (0.08) (0.18) (0.55)
born cs97 (x100) -0.596 0.400 1.598 -0.342 -0.342 -0.844
(0.54) (0.29) (0.28) (0.49) (0.49) (0.21)
male (x100) -1.189 -1.165 -7.473 0.993 1.013 4.738
(0.85) (0.80) (1 .90)* (2.64)“ (2.70)'*" (2.55)"
Constant 14.169 12.709 42.254 0.596 0.456 0.365
(3.00)"** (2.70)“ (2.59)“ (0.39) (0.29) (0.04)
Province dummies No Yes No No Yes No
Community dummies No No Yes No No Yes
Observations 930 930 304 3031 3031 589
R-squared 0.010 0.040 0.202 0.000 0.010 0.142
p-value (crises) 0.231 0.887 0.715 0.675 0.571 0.952

 

 

 

 

 

 

 

 

 

 

 

Rural Mother's education = 0 - 5 years Mothers education >= 6 years
LPM (1) (2) (3) (D (2) (3)
t (x100) -0.157 -0.147 -0.208 -0.195 .0200 -O.382
(0.92) (0.87) (0.77) (1 .93)‘ (2.00)“ (1 .85)‘
born crisis1 (x100) 1.090 1.102 -0.803 1.882 2.078 4.548
(0.42) (0.43) (0.22) (1.40) (1.53) (1.63)
born crisis2 (x100) 1.481 1.096 -0.171 0.693 0.637 1.969
(0.53) (0.40) (0.04) (0.67) (0.62) (0.71)
born cs97 (x100) 4.736 5.307 11.160 2.026 2.044 4.044
(1.44) (1.63) (1 .79)‘ (1.50) (1.51) (1.57)
male (x100) 1.870 1.676 2.643 1.094 1.141 1.343
(1 .92)* (1 .74)‘ (1 .74)‘ (1.98)“ (2.08)” (1.18)
Constant 3.159 2.952 4.198 3.894 3.993 7.642
(0.93) Q87) (0.78) (1 .94)‘ (1 .99)“ (1 .86)‘
Province dummies No Yes No No Yes No
Community dummies No No Yes No No Yes
Observations 2298 2298 1462 2680 2680 1279
R-squared 0.000 0.020 0.093 0.000 0.020 0.134
p-value (crises) 0.513 0.626 0.326 0.283 0.212 0.189

 

 

 

 

 

Absolute values of t-statistics robust to heteroskedasticity and clustering at mother level in parentheses.
" significant at 10%;” significant at 5%: ”* significant at 1%
Observations include all individuals that were born 1988 - 2000 who were exposed to at least 365 days of life.

crisis1 = Jan 98 - Sept 98, crisis2 = Oct 98 - June 99. crisi397 = May 97 - Dec 97

"North Sumatra" is omitted.

126

63.82 8: Sn .ucoaaEzmo on. s 33.9.. 26 8.5.56 8:395

.68. 852: 65 a acreage 36:88:32 9 339 es 888 2856

 

 

Sat F 6R: F 5.8 a 85:
8mm comm 2039.530
5.8m. 8.8». 685.95 8..
.....avoé Lamoé
oovd. Nmmd. a c. >
2.38.8 home
vmmé mmm... ofiE
2.83.3 2.63.8
memo 84.0 28> +NF
68.9 88.8
83 on: 98> FF-m
68.9 83.8
mend a F; 28> 3
339 >308
ammo >8. F 28> 9F
and «650E
..LEQNV £88.:
mmné mend mite oonmbcm305 Eon
.838 .838
«and F-.F mate Eocmcc Eon
83.9 38.8
memo cad 2

 

team .2855 .253:
26a EmNmI

cam. 262mm .886
2me Baum:

 

gocomeofir cmfimamw omega. 5:5

Econ. 232-0250”. wot.

 

Econ. Em~mn-o>=m_om mo;

 

323295. .3903

 

3558 v.83— Eanafl @2532 no 2338:; arson—«.3.— "3 033.

127

69..an .0: 5n .EOszamo o5 :_ 0320:. 2a 3:556 8:505

._o>o_ .059: 05 3 9533.0 8.0330me90: 8 .259. 05 82.0 Emucflw

 

 

883. 883: 8: 8 85:
88 88 888288
8.8:- 8.8:- 885.8: 8..
...888 ...:88
88.8. 88.8- a c:
...888 £828
83 83 was
588.8 588.8
8 to 2.3 28> +8
583.8 .898
88.8 88.8 28> :8
5838 $888
88.8 >88 28> 88
.898 .888
88.8 3.3 28> m;
:08 9:059:
.888 .838
YEN FNO._. mite monmbcmzec Eon
88.8 88.8
8 I 83 .88 .988: ES
888 88.8
m 8.. to; 8

 

scum E888 .888
03mm 2.85...

dorm 2885 .2538
251 Em~mI

 

aficwmofifix $8860 mm..m>:_ 53>

.Eom P5828268”. 00..

 

E5“. 232-9599“. m3

 

8239681 .3925

 

:23: 883— 6.38: 853.52 .3 2:58:38”.— utuofiauam "a 935.

128

 

 

 

 

 

 

 

 

mmm 0.0 v... P... «6.. v.3 mdm oéw V. F K coon
o3. Nd m6 N... MN? 03 on wa No K 89.
«NV 06 md fin o. 3 99 mflm NdN 8N om wag
mom md md mé o.m.. Now mNm NKN m... 5 39
x. 5 ed w... m... N. ..F Qt. ohm Nam N...” am mam?
5N Né 1N vN w. ..v Q? F. _.m mdw QM ..N Nam?
m 3 ad ad v... NNV 03 «.mm mdm N... om mag
.30 +3 910» mm- 5 omém mNAN NN-m P w Tm F v T F F mmmEms. 03:55.
3°. om< mmmEms. cm.88.). .0 50>
0”“...qu ac 23> .3 03.. mustang. "—N «Bah
84“.. 88 .8... .5... ”858
«8.. 8.9 --- --- --- AK 08 ER Emu Nd 0N m5 ..- ..an
mm: ...» --- -.. m... 53 mm? 0mm flow Qm ow 03 782.
om? we 06 md QN ...m 8.3 NmN {on v.9 or .3935?
ommm Nd md m4 0m 0.x. 5. ..r mdw Qmm m. 3 9 com .339
me? No ...N Ev md o.o.. «.3 v.8 mdm QNv om ova 7?me
cor od o.m 0.0 o.m 0.3 od QNN o.om 0.: m? omm Too?
.30 23m +3. 918 mm- rm omém mmém NN-m_. m E... v v; F 094 0
$9.55. ... :00
...8 8.85. 8< 8.8.2 5.8

 

 

 

2230 5.5 .3 uu< oust—“E .3 8.8.35 FEE—“:5 5N «Sah.

129

Table 22: Parametric Marriage Hazard Regressions (Cox Proportional Hazard)

 

 

 

 

 

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.
' significant at 10%; " signifiwnt at 5%; "" significant at 1%

Observations include all individuals that were born between 1941 - 1986.

Entry begins at age 11 years old

"No Edumtion', 'Big city”, and 'North Sumatra" are omitted categories.

130

(1) (2) (3) (41 (5) (6)
exposed to crisis risk 0.056 0.267 0.341 0.251 0.262 0.260
(1 .20) (5.48)*** (6.79)*** (5.17)” (5.39)*** (5.36)***
time trend (birth year) -0.016 -0.001 -0.005 -0.005
(15.75)*** (1.08) (4.21 )*** (4.59)***
born 1961 -1 965 -0.082
(2.52)"
born 1966 -1970 -0.225
(7.48)“
born 1971 -1 975 -0.353
(1 1.68)”
born 1976 - 1980 -0.429
(12.65)***
born 1981 -1986 -0.756
(13.40)”
years of educ -0.076
(31.1 1 )***
1-5 years 0.131 0.146
(2.91 )*** (3.14)***
6-8 years 0.053 0.086
(1.25) (1 .95)’
9-1 1 years ~0.270 -0.187
(5.84)“ (3.89)***
12+ years -0.857 -0.744
(20.65)*** (16.76)***
village 0.215
(5.47)***
small town 0.010
(0.24)
Province Dummies No No No No No Yes
Observations 1 1600 1 1600 1 1600 1 1465 1 1465 1 1465
Times at risk 1 05539 105539 1 05539 1 04321 104321 1 04321
Wald Chi (2) 1.4 249.2 322.5 1 133.5 1897.6 2074.2
Prob > Chi (2) 0.231 0.000 0.000 0.000 0.000 0.000

Table 23: Parametric Marriage Hazard Regressions (Gamma Distribution)

(survival coefficients)

 

 

 

 

(1) (2) (3) (4) (5) (6)
exposed to crisis risk 0.030 —0.037 -0.032 -0.024 -0.028 -0.032
(2.81 )*** (3.24)” (2.78)*** (2.98)*** (3.50)*** (3.80)***
time trend (birth year) 0.005 0.001 0.002 0.002
(21 .50)*** (4.10)*** (8.59)*** (9.44)***
born 1961-1965 0.027
(3.17)***
born 1966 -1970 0.077
(9.51 )***
born 1971-1975 0.116
(15.13)***
born 1976 - 1980 0.129
(18.82)”
born 1981 -1986 0.133
(16.10)***
years of educ 0.028
(55.1 1)***
1-5 years -0.005 -0.006
(0.61) (0.69)
6-8 years 0.048 0.045
(5.84)*** (5.59)**i
9-11 years 0.137 0.122
(16.04)*** (14.17)***
12+ years 0.311 0.286
(38.10)*** (33.72)***
village -0.043
(5.69)“
small town -0.015
(1 .92)*
constant 2.97 (6.06) 2.90 1.15 (0.51) (0.78)
(829.89)*** (14.43)” (569.41 )*** (2.94)*** (1.31) (1.99 **
Province Dummies No No No No No Yes
Observations 1 1600 1 1600 1 1600 1 1465 1 1465 1 1465
Times at risk 105539 105539 105539 104321 104321 104321
Wald Chi (2) 7.9 468.7 542.0 3365.9 4107.1 4321.9
Prob > Chi (2) 0.005 0.000 0.000 0.000 0.000 0.00(l_

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.
" signifimnt at 10%; " significant at 5%; “" significant at 1%

Observations include all individuals that were born between 1941 - 1986.

Entry begins at age 11 years old

We Education", “Big city. and "North Sumatra" are omitted mtegories.

131

Table 24: Parametric Marriage Hazard Regressions (Lognormal Distribution)

 

 

 

 

Change sign -
(1) (2) (3) (4) (5) (6)
exposed to crisis risk 0.029 -0.038 -0.032 -0.018 -0.021 -0.025
(2.73)“ (3.28)‘** (2.81)*** (2.06)” (2.57)“ (2.90)***
time trend (birth year) 0.005 0.001 0.002 0.002
(21.43)*** (4.22)*** (8.53)*** (9.57)”
born 1961-1965 0.026
(3.12)”
born 1966 -1970 0.077
(9.40)***
born 1971-1975 0.116
(15.11)“
born 1976 - 1980 0.129
(18.69)***
born 1981 -1986 0.132
(15.97)”
years of educ 0.028
(54.05)***
1-5 years -0.004 -0.004
(0.45) (0.44)
6-8 years 0.050 0.049
(6.02)” (5.92)”
9-11 years 0.136 0.121
(15.69)*** (13.96)“
12+ years 0.307 0.283
(37.07)*** (33.22)***
village -0.041
(5.47)***
small town -0.016
(2.00)"
constant 2.97 (6.03) 2.90 1.11 (0.50) (0.80)
(762.24)*** (14.36)*** (522.52)*** (2.84)*** (1.29) (2.07)"
Province Dummies No No No No No Yes
Observations 1 1600 1 1600 1 1600 1 1465 1 1465 1 1465
Times at risk 105539 105539 105539 104321 104321 104321
Wald Chi (2) 7.5 466.1 533.1 3277.9 4014.6 4254.4
Prob > Chi (2) 0.006 0.000 0.000 0.000 0.000 0.000

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.

' significant at 10%; " significant at 5%; “' signifiwnt at 1%
Observations include all individuals that were born between 1941 - 1986.
Entry begins at age 11 years old

"No Education", "Big city'. and ”North Sumatra" are omitted categories.

132

Table 25: Parametric Fertility Hazard Regressions (Cox Proportional Hazard)

 

 

 

(1) (2) (3) (4) (5)
exposed to crisis risk -0.584 -0.578 -0.585 -0.584 -0.587
(1 1 .22)“ (10.75)*** (10.67)“ (10.66)“ (10.73)**"
time trend (marriage year) -0.001 -0.009 -0.010 -0.009
(0.40) (5.21 )**" (5.34)*** (4.85)”
years of educ 0.04
(14.09)”
1-5 years 0.219 0.205
(4.51 )’** (4.22)”
6-8 years 0.332 0.323
(7.15)” (6.92)***
9-1 1 years 0.447 0.415
(9.10)“ (8.35)***
12+ years 0.533 0.480
(1 1 .31 )*** (9.86)***
urban 0.080
(3.58)...
Province Dummies No No No No Yes
Observations 6555 6555 6383 6383 6383
Times at risk 10306 10306 10014 10014 10014
Wald Chi (2) 125.8 125.9 330.3 326.4 420.0
Prob > Chi (2) 0.000 0.000 0.000 0.090 0.000

 

 

Absolute value of 2 statistics in parentheses
Standard errors are robust to heteroskedasticity and clustering at individual level.
' significant at 10%; “ significant at 5%; "' significant at 1%
Observations include all individuals that married after 1979.

"No Education“ and 'North Sumatra“ are omitted categories.

133

Table 26: Parametric Fertility Hazard Regressions (Lognormal Distribution)

(survival coefficients)

 

 

 

(1) L2) (3) (4) (5)
exposed to crisis risk 0.105 0.140 0.141 0.143 0.146
(7.39)*‘* (9.57)” (7.95)“ (0.81 ) (8.96)***
time trend (marriage year) -0.005 -0.001 -0.001 -0.002
(6.67)*** (2.17)” (0.32) (2.66)***
years of educ (0.01)
(9.73)“
1-5 years 0088 -0.082
(0.95) (3.63)***
6-8 years -0.122 -0.119
(0.84) (5.24)’**
9-11 years 0170 -0.158
(0.83) (6.74)***
12+ years -0.203 -0.181
(0.87) (7.85)***
urban -0.042
(4.85)***
constant 0.177 9.215 3.182 3.102 4.015
(28.23)*** (6.79)‘** (2.37)" (0.34) (2.89)“
Province Dummies No No No No Yes
Observations 6555 6555 6383 6383 6383
Times at risk 10306 10306 10014 10014 10014
Wald Chi (2) 54.7 110.1 113.4 13.1 232.0
Prob > Chi (2) 0.000 0.000 0.000 0.041 0.000

 

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.

' significant at 10%; “ significant at 5%; '“ significant at 1%
Observations include all individuals that married after 1979.
'No Education” and "North Sumatra” are omitted categories.

134

Table 27 : Parametric Fertility Hazard Regressions (Weibull Distribution)

 

 

 

 

(1) (2) (3) (4) (5)
exposed to crisis risk -0.616 -0.937 -0.943 0939 -0.943
(8.71 )*** (13.23)” (12.97)*** (12.97)“ (13.01 )**’
time trend (marriage year) 0.042 0.021 0.020 0.021
(1 0.23)“ (4.74)” (4.35)“ (4.73)*"*
years of educ 0.09
(12.74)***
1-5 years 0.607 0.575
(4.26)*** (4.06)"'**
6-8 years 0.880 0.850
(6.55)**" (6.27)“
9-11 years 1.167 1.087
(8.52)“ (7.87)***
12+ years 1.332 1.203
(9.82)*** (8.66)“
urban 0.183
(3.03)***
constant 0832 -84.290 -42.704 -40.750 -44.215
(68.95 ... (10.34)” (4.92)*** (4.56)*** (4.94)"**
Province Dummies No No No No Yes
Observations 6555 6555 6383 6383 6383
Times at risk 10306 10306 10014 10014 10014
Wald Chi (2) 75.9 232.2 364.8 367.1 432.4
Prob > Chi (2) 0.000 0.000 0.000 0.000 L 0.000

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.

‘ significant at 10%; “ significant at 5%; “' significant at 1%
Observations include all individuals that married after 1979.
“No Education“ and “North Sumatra“ are omitted categories.

135

Table 28: Unconditional Fertility Hazard Regressions (Cox Proportional Hazard)

 

t1) (2)

13)

(4)

(5)

(6)

 

exposed to crisis risk

-0.470 -0.443

-0.459

-0.430

-0.428

-0.423

(18.97)” (20.47)*** (21.36)*** (18.76)“ (18.68)*** (18.75)***

 

time trend (birth year) 0004 0.006 0.002 0.000
(2.69)*** (3.88)“ (1 .29) (0.08)
born 1961-1965 -0.119
(2.69)***
born 1966 -1970 -0.256
(6.52)***
born 1971-1975 -0.341
(8.83)***
born 1976 - 1980 -0.005
(0.1 1)
born 1981 -1986 0.232
(2.81)***
years of educ -0.059
(18.47)***
1-5 years 0.309 0.353
(5.02)*** (5.70)“
6-8 years 0.280 0.334
(4.75)*** (5.58)***
9-11 years 0.072 0.160
(1 .13) (2.45)"
12+ years -0.483 -0.369
(8.46)*** (6.16)***
urban -0.193
(6.05)”
Province Dummies No No No No No Yes
Observations 16042 16042 1 6042 1 571 7 1 5717 1 571 7
Times at risk 113309 113309 113309 111347 111347 111347
Wald Chi (2) 359.9 419.1 791.6 864.6 1105.2 1201.5
Prob > Chi (2) 0.000 0.000 0.000 0.000 0.000 0.000

 

 

Absolute value of 2 statistics in parentheses
Standard errors are robust to heteroskedasticity and clustering at individual level.

" significant at 10%; “ significant at 5%; ‘“ significant at 1%

Observations include all individuals that were born between 1941 - 1986.

Entry begins at age 14 years old

”No Education” and and ”North Sumatra“ are omitted categories.

136

Table 29: Unconditional Fertility Hazard Regressions (Gamma Distribution)
(survival coefficients)

 

 

 

(1) (2) (3) (4.) (5) t6)
exposed to crisis risk 0.087 0.062 0.075 0.040 0.036 0.036
(17.15)“** (14.09)*** (19.20)*** (11.20)*** (10.43)“ (10.37)”
time trend (birth year) 0.002 0.000 0.000 0.001
(10.71 )*“ (2.00)“ (2.63)” (4.08)”
born 1961-1965 0.032
(4.62)**"
born 1966 -1970 0.069
(9.07)”
born 1971-1975 0.111
(15.89)***
born 1976 - 1980 0.083
(15.16)*"’*
born 1981 -1986 0.023
(3.47)“
years of educ 0.023
(39.65)***
1-5 years 0010 -0.013
(1.37) (1 .74)*
6-8 years 0.024 0.019
(3.28)*** (2.44)“
9-11 years 0.079 0.068
(10.04)“ (8.17)”
12+ years 0.222 0.206
(28.75)*** (25.10)“
urban 0.02
(5.78)“
constant 3.025 -1 .103 2.969 3.605 1 .988 1 .427
(847.87)*** (2.87)*** (640.41)” (9.51)“ (5.54)*** (3.81)”
Province Dummies No No No No No Yes
Observations 16042 16042 16042 15717 15717 15717
Times at risk 113309 113309 113309 111347 111347 111347
Wald Chi (2) 294.1 284.0 986.0 1778.5 2258.4 2356.3
Prob > Chi (2) 0.000 0.000 0.000 0.000 0.000 0.000

 

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.
" significant at 10%; " significant at 5%: "" significant at 1%

Observations include all individuals that were born between 1941 - 1986.

Entry begins at age 14 years old

“No Education“ and and 'North Sumatra' are omitted categories.

137

Table 30: Parametric Marriage Hazard Regressions (Cox Proportional Hazard)

 

 

 

 

 

 

(1) (2) t3) (4) (5) (6)
Exposed to 1996 Risk -0.204 -0.040 -0.014 -0.047 -0.035 -0.041
(3.76)*** (0.73) (0.26) (0.85) (0.63) (0.74)
time trend (birth year) 0014 0.000 -0.003 -0.004
(14.57)*** (0.12) (3.04)*** (3.45)“
born 1961-1965 -0.080
(2.45)"
born 1966 -1970 -0.222
(7.35)“
born 1971-1975 -0.330
(10.97)”
born 1976 - 1980 -0.372
(11.09)***
born 1981 -1986 -0.672
(12.34)“
years of educ -0.076
(31.15)***
1-5 years 0.127 0.142
(2.83)” (3.05)“
6-8 years 0.048 0.081
(1.13) (1 .83)*
9-11 years -0.271 -0.188
(5.85)*** (3.91 )***
12+ years -0.861 -0.748
(20.75)*** (16.86)“
village 0.215
(5.47)”
small town 0.010
40.24)
Province Dummies No No No No No Yes
Observations 1 1600 1 1600 1 1600 1 1465 1 1465 1 1465
Times at risk 105552 105552 105552 104334 104334 104334
Wald Chi (2) 14.2 228.4 292.4 1112.0 1860.0 2033.9
Prob > Chi (2) 0.000 0.000 0.000 0.000 0.000 0.000

 

Absolute value of 2 statistics in parentheses

Standard errors are robust to heteroskedasticity and clustering at individual level.
' significant at 10%; " signifimnt at 5%; ... significant at 1%

Observations include all individuals that were born between 1941 - 1946.

Entry begins at age 11 years old

'No Education“. 'Big city'. and 'North Sumatra” are omitted categories.

138

Table 31: Parametric Marriage Hazard Regressions (Cox Proportional Hazard)

 

 

 

 

(1) (21 (3) (4) (5) (6)
98/99 crisis risk 0.065 0.321 0.457 0.289 0.299 0.299
(1 .94)‘ (8.69)“ (11.42)*** (7.77)“ (8.03)*** (8.05)***
time trend (birth year) -0.017 -0.003 -0.006 -0.007
(16.79)” (2.39)“ (5.45)*** (5.80)**"
born 1961-1965 -0.085
(2.63)***
born 1966 -1970 -0.232
(7.72)“
born 1971-1975 -0.386
(12.51 )***
born 1976 - 1980 -0.514
(14.35)***
born 1981 -1986 -0.922
(15.40)”
years of educ -0.076
(30.98)***
1-5 years 0.136 0.151
(3.02)“ (3.25)“
6-8 years 0.059 0.093
(1.39) (2.10)"
9-11 years 0271 -0.187
(5.86)“"’ (3.90)”
12+ years (0.85) (0.74)
(20.47)” (16.57)***
village 0.21
(5.44raa
small town 0.010
(0.23)
Province Dummies No No No No No Yes
Observations 11600 11600 11600 11465 11465 11465
Times at risk 105539 105539 105539 104321 104321 104321
Wald Chi (2) 3.8 285.5 388.3 1165.2 1933.5 2099.2
Prob > Chi (2) 0.052 0.000 0.000 0.000 0.000 0.000

 

Absolute value of 2 statistics in parentheses
Standard errors are robust to heteroskedasticity and clustering at individual level.
' significant at 10%; “ significant at 5%; "" significant at 1%

Observations include all individuals that were born between 1941 - 1986.

Entry begins at age 11 years old

'No Education“. "Big city'. and 'North Sumatra” are omitted categories.

139

APPENDIX II

140

Annendix [1: Mortality Crosswalk from the Indonesian Family Life Surveys (IFLSl-IFLS3)

 

 

 

 

 

ch00=1 panel respondent with no preprinted child roster
ch00=2 panel respondent with preprinted child roster
ch00=3 new respondent
Number of Obs
FILSZ and 3 (19952000) Neonatal Post-neonatal Total live births
All obs 59 69 3093
Number of Obs
FILS3 (1995-2000) Neonatal Post-neonatal Total live births
All obs 41 49 2034
ch00==2 or 3 29 36 1443
ch00== 12 13 591

 

 

Rates (per 1000 live births)

 

 

 

 

 

 

 

 

 

 

 

 

 

F ILS3 (19952000) Neonatal Post-neonatal
All obs 20.2 24.1
ch00==2 or 3 20.1 24.9
ch00== 20.3 22.0
Number of Obs by Year
IF L83 (1995) IFLSZ (1995)
Neonatal Post-neonatal Total live births Neonatal Post-neonatal Total live births
All obs 3 5 214 12 10 582
ch00==2 or 3 3 4 200 8 5 296
ch00== 0 1 14 4 5 286
IFLS3 (1996) lFLSZ (1996)
Neonatal Post-neonatal Total live births Neonatal Post-neonatal Total live births
All obs 4 4 271 6 10 472
ch00==2 or 3 3 1 239 0 7 253
ch00== 1 3 32 0 3 219
IFLSS (1997)
Neonatal Post-neonatal Total live births Neonatal Post-neonatal Total live births
All obs 9 1 1 297 0 0 5
ch00==2 or 3 6 9 245 0 0 5
ch00== 3 2 52 0 0 0
IFLS3 (1998) IFLSZ (1998)
Neonatal Post-neonatal Total live births Neonatal Post-neonatal Total live births
All obs 15 16 656 NA NA NA
ch00==2 or 3 10 14 390 NA NA NA
ch00== 5 2 266 NA NA NA
IFL83 (1999) IFLSZ (1999)
Neonatal Post-neonatal Total live births Neonatal Post-neonatal Total live births
All obs 10 13 596 NA NA NA
ch00==2 or 3 7 8 369 NA NA NA
ch00== 3 5 227 NA NA NA
NOTE:

Too few obs to calculate and compare mortality rates for each year

No obs in born in 2000 that had lived 365 days by the interview date of IFLSS

Similar reason for IFLSZ data with those born in 1997

( 5 obs of 1997-barns were interviewed in 1998. These have age>=365 day by the interview)

141

APPENDIX III

142

 

00.>oZ 00.3.. 00.3.). 00.>oZ 00.3.. 0038.). 00452 moi... 00.32 50462 50.3.. 50.82 00452 00.3..

F

. _ p — _ _ h . p _

 

 

_ 0

r 000.N

 

 

lllI'lllDIIl

 

Inllllitl IIIII

 

nllllltolllll

r 000..v

1 000.0

I 000.w

1 000.0.

1 000.2

I 000.:

 

 

23. 8558. amp}. 2.. as we: a: a «:5? a 2:5

000.0.

([8de

143

.aéll
535 I I

00.5..

.

8.88 8.3,. 8.5

L

F L . L p

8.: u 8.: 3:50 use... 8.5 .38... "N paw...—

woeem .3432 was:

8.88 8.5,. 3.5:

h

.

 

0

tom

1 00.

r 02

1 00m

r 0mm

1 00m

 

r 0mm

144

Figure 3: Government Health Expenditures (1993 = 100)

 

Government Health Expenditures, Overall and by Major
Component, Per Capita Constant Prices, 1993 = 100

 

 

 

 

 

 

8
O
D
d.
E.
4 .. 7 -- v __ --_ ___ __
2 “ ‘\- ‘ /‘\.__....:«-\ --_ ‘ “*—
0 l l T T l f F I i f T 1
89/90 90/91 91/92 92/93 93/94 94/95 95/96 96/97 97/98 98/99 99/00 2000
(12
mos)

 

+ GOI Funding Per Capita .... Donor Assistance Per Capita
Total Spending Per Capita

 

 

 

 

 

 

Source: Figure l: Lieberman S., M. Juwono, and P. Marzoeki. “Government Health Expenditures in
Indonesia through December 2000: An Update.” World Bank East Asia and the Pacific Region Brief.

145

330250: 3.8.. .«c 93.500 ”gem

5:252
3022.. ..o 5:05.
00.0 >42 JO aww 9.3. ...... an... >45. :7. ......s. 84... cm...

I.)

‘

 

l 000.000. ..

/ . \ r ooo.ooad

 

 

l ooo.ooo.m

OOON ill mmm w 1140'
mmm —. liar, hmmr llllwl

5.82 .3 680.32 Agassi. 8882a: 3.: 32

88:80 8.85: 82 8:. a. 2.5....

146

23>

wmo. 000. .00. N00. 000. wwm.

. — p . _ . _ . . h _ w o

 

 

22 £802
3808.800 ....... H
B
as. £3.02 m
.8980le

. . , mm

- 9.

. a.

.25... 3.. 83 :3 8.3. 0.38:. “m 2.05

147

Figure 6a-6c: Cummulative Distribution of Birthweight (Kg.)

 

 

 

 

 

 

 

 

 

 

 

 

- — — - cumweight_droughtcrisis_mral —-——— cumweight_noncrisis_mral
1 _‘ _. __ __ __
.95 ~
.9 ~
.7 ~
.5 5
.3 a
.1 ~
.05 -+
I I I I I I I I I I I
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
Baby's Weight in Kg
- — - — cumweight_fincrisis_rura|—— cumweight_noncrisis_mral — — — — cumweight_fincrisis_urban—— cumweight_noncrisis_urban
I
1 4 1 5
‘95 ~ '5 4
.9 .4 9 4
.7 "‘ 7 1
.5 " 5 _.
.3 “ 3 .4
.1 ~ ' 1 -
.05 q , , ’ ,5 _ / ,
I I T I T I I I T I I I I F I I I I I I I I
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
Baby's Weight in Kg Baby's Weight in Kg

148

Figure 7a-7e: N on-parametric Hazard Rates (by Month)

—e— born drought cn'sis_rura|—+s— born non-crisis_rural

.0225~
.02~
.01754
.015.
.0125—
.01—
.0075~
.0054
.00255
OH

 

 

l T T I
o 1 e 12
t

nonparametric hazard rate

Log-rank test for equality of survival function : p = 0.358

—e— born fin.crisis_ruraI——¢— born non-crisis_rural —e— born fin.crisis_urban + born non-crisis_urban

 

 

 

 

 

 

I 255
I
32-
.02254 75~
.02 + 15‘
.0175“
25«
.015«
.0125 — 31«
.01 ‘ 75 a
.007 ~
5 35 ~
.005«
q l
.0025~ 25
0 ~ 0~
El 35 1'2 23 <5 l 65 1:2 251
t
nonparametric hazard rate nonparametnc hazard rate

Log-rank test for equality of survival function : p = 0.823 Log-rank test for equality of survival function : p = 0.187

149

Figure 8a-8c: Parametric Estimated Hazard Rate (Weibull Distribution)

 

.01—4
.009—
.008‘ I
.007‘ ‘
.006—1.
.005~ ‘1
.0041 in
.0034
.0021

01

noncrisis_rural

— - — — droughtcrisis_rural

 

 

‘1 6 12 24

36

48 60 72

_t
Hazard Rate (Weibull Distribution)

—— noncrisis_rura|

- - - - fincrisis_rura|

.01 j
.009 -:
.008 ‘
.007 " l.

.0011
Oz

 

 

1 6 12 24 361 48 60
Hazard Rate (Weib-ull Distribution)

72

150

— noncrisis_urban -— - - - fincrisis_urban

.005 ‘

.004 n

|

l
l
l
|
l
l
l

.001 ‘

 

 

0d

 

I I I I I I I I
1 6 12 24 36 48 60 72

_t
Hazard Rate (Weibull Distribution)

1941_60

1 -5yrs

Figure 9: Nonparametric First-Marriage-Age Hazard by Birth Cohort

 

 

 

 

 

—— 1900_40 — — e — —1941_60
— — — -1971_80
.2 A
.15 z
.1 z
.05 r
0 2
l l l l I l l l
11 15 18 20 25 30 32 35
. A9?
Marriage Hazard by Birth Cohort
Figure 10: Nonparametric First-Marriage-Age Hazard by Education
no educ —-<+——1-5yrs
— — —— 6-8yrs
1
“1
2 ~ \ // \\
3w \// \ \ v/ I?‘ \
o/l \ \ / l \ t
/ I \ I ’l
'15 — II/ R\ /Q\ l \\ \ I )1 I l
I/ \ I \l \ P\\ I II ,‘ I I
I/ .’ \I \0--'/ lI‘I /\ II
“ll 6/ \/ \/,\\ I I‘
I ' \ / XI \ \ I \
1 7 I I b , \ ' /
/ , 1’ 9\ \\ /
/ \ of \ A l ‘1 ’0
/ / g V " VJ
-05 “ .. / ‘l
0’ /
/
o —«
l l l l l l V l l
11 15 18 20 25 30 32 35 40
Age

Marriage Hazard by Education

151

Figure 11: Nonparametric First-Marriage-Age Hazard by Education

9-11yrs

 

 

 

———noeduc ——e--9‘11Y|’5
—— —— 12+yrs
.2 z
.15 ~
.1 m
/
.05 ~ [I
/.
/.
P
0 _‘ $9.4” ’
l l l l l l l l
11 15 18 20 25 3O 32 35

Age
Marriage Hazard by Education

Figure 12: Nonparametric First-Marriage-Age Hazard by Residence at Age 12

 

 

village — — o — — small town
— — — -bIgCIty
/‘
[l
/ l
.2 /\
.2 \/ l
J l ’
I /
\ /\
l \
c .15~ If“ \1 // \ q
.9 / \ / \ I \' \
i p/ b ’ \ / ‘ / 9 /\
E / x \ / \ \ / \ /\/ \
U) 1— R / 9‘ / \
// \3/ \/ \ \ \b F V
? f V \ / \ '
/ / \
/
.05 ‘l J/ / \ / \l /o
7; / \/ \//
d// b
tg///
0 —I
T l l l T l l l T
11 15 18 20 25 30 32 35 40

Age
Marriage Hazard by Residence

152

educ 6-11 years

Figure 13: Nonparametric Fertility Hazard by Marriage Cohort

Married 1974-1985 ————— Married 1986-2000

 

 

 

 

m—i
_I
O

T I I I T I T I
1 2 3 4 5 6 7 8
years after marriage

Nonparametric Fertility Hazard by Marriage Cohort

Figure 14: Nonparametric Fertility Hazard by Education

 

educ 0-5 years — — e — - educ 6-11 years
— — — —— educ 12+ years

 

 

 

I I I f I I I I
1 2 3 4 5 6 7 8 9 10
years after marriage

Nonparametric Fertility Hazard by Education

—1

153

Figure 15: Nonparametric Fertility Hazard by Residence

urban residence ————— rural residence

 

 

 

 

I I I I I I I
2 3 4 5 6 7 8
years after marriage

Nonparametric Fertility Hazard by Residence

_h——I
(0“
—L
0

Figure 16: Unconditional Nonparametric Fertility Hazard by Birth Cohort

 

born 1941-1960 ————— born 1961-1986

 

 

 

I I I I I I I I I I I
15 18 20 23 25 28 30 33 35 38 40
t

Unconditional Fertility Hazard

154

BIBLIOGRAPHY

155

 

BIBLIOGRAPHY

Adelman, Irman. “An econometric analysis of population growth.” American Economic
Review. 53, no. 3 (June 1963): 314—339. '

Alatas, Vivi. “What Happen to Indonesia’s Poverty? A Micro Simulation Exercise
Using Household Surveys. Manuscript. World Bank. Jakarta, Indonesia. March
2002.

Bardhan, P. and C. Udry. Development Microeconomics. 1999. Oxford University
Press.

Basu, A.M. and Peter Aaby. The Methods and Uses of AntrOpological Demography.
1998. Clarendon Press. Oxford.

Becker, Gary. “A Theory of Marriage: Part I.” Joumal of Political Economy, Volume
81, Issue 4, Jul.-Aug., 1973, 813-846.

Beckett, M., J. DaVanzo, N. Sastry; C.Panis, and C. Peterson. “The quality of
retrospective data: An examination of long-term recall in a developing country.”
Journal of Human Resources. Summer 2001.

Boulier, Brian and Mark Rosenzweig. “Schooling, Search, and Spouse Selection:
Testing Economic Theories of Marriage and Household Behavior.” Journal of
Political Economy, Volume 92, Issue 4, Aug., 1984, 712-732.

Bravo, J. “Demographic Consequences of structural adjustment in Chile.” Paper
presented at the Seminar on Demographic Consequences of Structural Adjustment
in Latin America, Ouro Preto, Brazil (1992).

Bravo, J. “Economic Crisis and mortality: short and medium-term changes in Latin
America.” Paper presented at the Conference on the Peopling of the Americas,
Veracruz, Mexico (1992).

Butz, William P., and DaVanzo, Julie. The Malaysian Family Life Survey: Summary
Report. RAND report no. R.2351-AID. Santa Monica, Calif: RAND Corp.,
March 1978.

Caldwell, J., Reddy, P. and P. Caldwell. “Periodic High Risk as Cause of Fertility
Decline in a Changing Rural Environment: Survival Strategies in the 1980-1983
South Indian Drought.” Economic Development and Cultural Change. 34 (July
1986): 677-701.

Cleland, J. and J. Van Ginneken. 1989. “Maternal education and child survival in

developing countries: the search for pathways of influences.” Social Science and
Medicine. 27: 1357-60.

156

Coale, Ansley and Edgar Hoover. “Population Growth and Economic Development in
Low Income Counties. Princeton, NJ: Princeton Univerity Press, 1958.

Daly, Patricia and Fadia Saadah. “Indonesia: Facing the Challenge to Reduce Maternal
Mortality.” East Asia and the Pacific Region Watching Brief. World Bank. June
1999. Issue 3.

Das Gupta, M. “Fertility Decline in Panjab, India: Parallels with Historical Europe.
Population Studies 49(3): 481-500.

Davidson, R. and J ean-Yves Duclos. “Statistical Inference for Stochastic Dominance and
for the Measurement of Poverty and Inequality.” Econometrica v86 n6. 2000.

East Asia and Pacific Region. Watching Brief. “Indonesia: Undemutn'tiion in Young
Children.” World Bank. January 1999. Issue 1.

Ermisch, J. “Economic Opportunities, Marriage Squeezes, and Propensity to Marry: An
Economic Analysis of Period Marriage Rates in England and Wales.” Population
Studies, Volume 35, Issue 3, Nov., 1981, 347-356.

F edericksen, Harald. “ Determinants and consequences of mortality and fertility trends.”
Public Health Reports. 81, no. 8 (August 1966): 715-727.

Foster, Andrew D. “Price, Credit Markets and Child Growth in Low-Income Rural
Areas.” The Economic Journal, Volume 105, Issue 430 (May, 1995), 551-570.

Frankenberg E., Thomas D., and Beegle K. 1999. “The real costs of Indonesia’s
Economic Crisis: Preliminary Findings from the Indonesia Family Life Surveys.”
DRU-2064-NIA/NICHD. Santa Monica, CA: RAND.

Frankenberg, E., Hamilton, P., Polich, S., Suriastini, W., and D. Thomas. User 's Guide
for the Indonesia Family Life Survey. DRU2238/2-NIA—NICHD. March 2000.

Frankenberg, E., Smith, J. and D. Thomas. “Economic Shocks, Wealth, and Welfare.”
Journal of Human Resources, Volume 38 (Spring 2003).

Galloway, P. “Basic Patterns in Annual Variations in Fertility, Nuptiality, Mortality, and
Prices in Pre-Industrial Europe.” Population Studies, Volume 42, Issue 2 (Jul.,
1988), 275-302.

Hill, K, G. Adansi-pipim, L. Assogba, A. Foster, J. Mukiza-gapere, and C. Paxson.
“Demographic Effects of Economic Reversals in Sub-Saharan Africa.” National
Research Counci1.National Academy Press. Washington DC. 1993.

Hill, K. and A. Palloni, “ Demographic responses to economic shocks: The Case of Latin
America.” Paper present at the Conference on the Conference on the Peopling of
the Americas, Veracruz, Mexico (1992).

157

Hull, T and G. Jones. “Demographic Perspectives.” in Indonesia’s New Order: The
Dynamics of Socio-economic Transformation. Edited by H. Hill. Allen & Unwin
Pty Ltd. SRM Production Services Sdn Bhd, Malaysia. 1994.

Hull, T. “Fertility Decline in Indonesia: a review of recent evidence.” 1980. BIES
16(2), pp. 104-12.

Hull, TH. and V.J. Hull. “Population Change in Indonesia: findings of the 1980 census.”
1984. BIES. 20(3), pp. 95-119.

James J. Fox. “The 1997-1998 Drought in Indonesia.” Natural Disasters and Policy
Responses in Asia: Implications for Food Security. Harvard University Asia
Center. August, 1999.

Keeley, Michael. “Age Analysis of Age Pattern of First Marriage.” International
Economic Review, Volume 20, Issue2, Jun., 1979, 527-544.

Kotlikoff, L. and A. Spivak. “The Family as an Incomplete Annuities Market.” Journal
of Political Economy, Volume 89, Issue 2 (Apr., 1981), 372-391.

Lee, R. “ Short-term variation: vital rates, prices and weather,” in E.A. Wrigley and RS.
Schofield (eds.). The Population History of England. 1541-1871. Cambridge,
Mass: Harvard University Press. 1981.

Lieberman, 8., M. Juwono, and P. Marzoeki. “Govemment health expenditures in
Indonesia through December 2000: An Update.” World Bank East Asia and the
Pacific Region Watching Brief. October 15, 2001, Issue 6.

Lindstrom, David and Betemariam Berhanu. “The Impact of War, Famine, and
Economic Decline on Marital Fertility in Ethiopia.” Demography, Volume 36,
Issue 2 (May, 1999), 247-261.

Mckenzie, D. “How Do Households Cope with Aggregate Shocks? Evidence from the
Mexican Peso Crisis.” Manuscript. 2002.

Mellington and Cameron. “Female Education and Child Mortality in Indonesia.”
Melboume- Department of Economics in its series papers number 693. 1999

Menken, J., J. Trussell, and S. Watkins. 1981. “The nutrition fertility link: An
evaluation of the evidence.” Journal of Interdisciplinary History XI(3): 425-441.

Mloyi, M. “Crisis-Driven Fertility Decline.” Background Paper. UN Economic and
Social Council Third African Population Conference, Meeting of Experts, Dakar,
Senegal 7-10 December, 1992.

Mooney, M. and P. Hodsdon. “Effects of the Timing of Marriage and First Birth of the
Spacing of Subsequent Births.” Demography, Vol. 18, No. 4. (Nov., 1981). Pp.
529-548.

158

Palloni, A, K. Hill, and GP. Aguirre. “Economic Swings and Demographic Changes in
the History of Latin America.” Population Studies, Volume 50, Issue 1 (Mar.,
1996), 105-132.

Palloni, A. “Health Conditions in latin America and policies for mortality change”, in J.
Vallin and A. Lopez (eds.), Health Policy, Social Policy and Mortality Prospects.
Liege: Ordina. 1985

Palloni, A. and K. Hill, “The Effects of Structural Adjustments on mortality by age and
cause in Latin America.” Center for Demography and Ecology. Working Paper
92-22, University of Wisconsin, Madison, Wis. 1992.

Pertersen, W. and R. Petersen with the collaboration of an International Panel of
Demographers. Dictionary of Demography: Terms, Concepts, and Institution.
Greenwood Press. New York. Westport, Connecticut. London. 1986

Portner, Claus. “Children as Insurance.” Journal of Population Economics (2001) 14:
119-136.

Pressat, R. Statistical Demography. Translated and Adapted by Damein A. Courtney.
1978. Methuen & Co Ltd.

Preston, Samuel H. “Health Programs and Population Growth.” Population and
Development Review. Volume 1, Issue 2 (Dec. 1975), 189-199.

Rahman, M., B. Wojtyniak, M. M. Rahaman and K.M.S. Aziz, “Impact of environmental
sanitation and crowding on infant mortality in rural Bangladesh.” The Lancet
(1985), pp. 28-32.

Razzaque, A., N. Alam, L. Wai, and A. Foster. “Sustained effects of the 1974-75 famine
on infant and child mortality in a rural area of Bangladesh.” Population Studies
44: 145-154. 1990.

Reher, D. and J.A. Ortega, “Short run economic fluctuations and demographic behaviour:
some examples from twentieth century South America.” Paper presented at the
Seminar on Demographic behaviour: some examples from twentieth century
South America.” Paper presented at the Seminar on Demographic Consequences
of Structural Adjustment in Latin America, Ouro Preto, Brazil (1992).

Richards, A. and. S. Preston. “The Influence of Women’s Work Opportunities and
Marriage Rates.” Demography, Volume 12, Issue 2, May 1975, 209-222.

Richards, T. “Weather, nutrition, and the economy: the analysis of short-run fluctuation
in births, deaths, and marriages, France 1740-1909,” in T. Bengtsson et al. (eds.),
Pre-Industrial Population Change. Stockholm: Alrnquist and Wiksell. 1984.

Rosenzweig, M and T.P. Shultz. “Child mortality and fetility in Colombia: individual
and community effects.” Health Policy and Education, 2: 305. 1982.

159

Rosenzweig, M. “Risk, Implicit Contracts and the Family in Rural Areas of Low-Income
Countries.” Economic Journal. 98 (December 1988): 1148-70.

Rosenzweig, M. and Oded Stark. “Consumption Smoothing, Migration, and Marriage:
Evidence from Rural India.” Journal of Political Economy, Volume 97, Issue 4
(Aug, 1989), 905-926.

Rutenberg, N. and Ian Diamond. “Fertility in Botswana: The Recent Decline and Future
Prospects.” Demography, Vol. 30, No. 2, May 1993.

Sah, RK. “The Effects of Child Mortality Changes on Fertility Choice and Parental
Welfare.” Journal of Political Economy 99(3):582-606.

Sastry, Narayan. “Forest Fires, Air Pollution, and Mortality in Southeast Asia.”
Demography. Volume 39 numberl. February 2002.

Schultz, T. P. 1989. “Benefits of Educating Women.” Washington, DC: World Bank,
Background Papers Series, Education and Employment Division, Population and
Human Resources Department. The Benefits of Education for Women HRO

Dissemination Notes. Human Resources Development and Operations Policy.
Number 2, March 8, 1993.

Schultz, T.P. “An economic model of family planning and fertility.” Journal of Political
Economy. 77, no.2 (March/April 1969): 153-180.

Schultz, T.P. “Investments in the schooling and health of women and men: Quantities
and returns.” Journal of Human Resources. Fall 1993; Vol. 28, Iss. 4; pg. 694, 41.

Singh, 1., L. Squire and J. Strauss (eds.). 1986. Agricultural Household Models:
Extensions, Applications and Policy. Baltimore: John Hopkins University.

Spiegelman, M. Introduction to Demography. 1955. The Society of Actuaries.
Chicago. Illinois.

Stein, Z., M.Susser, G. Saenger and F. Morolla. Famine and Human Development: The
Dutch Hunger Winter of [944-1945. New York. 1975.

Strauss, J., Beegle, K., Dwiyanto, A., Herawati, Y., Pattinasarany, D., Satriawan, E.,
Sikoki, B., Sukumdi and F. Witoelar. “Indonesian Living Standards Three Years
alter the Crisis: Evidence from the Indonesia Family Life Survey.” 2003.
RAND, Santa Monica, CA.

Wang, F. and Yang Quanhe. “Age at Marriage and the First Birth Interval: The
Emerging Change in Sexual Behavior Among Young Couples in China.”
Population and Development Review, Vol. 22, No. 2. (Jun, 1996). Pp 299-320.

Weir, D. “Life under pressure: France and England, 1670-1680,” Journal of Economic
History, 44 (1984), pp. 27-47.

160

Wooldridge, J .M. Introductory Econometrics: A Modern Approach. Copyright 2000 by
Sourth-Westem College Publishing. USA.

161

 

 

“.fi . .
“v~s -v‘-—