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

 

ESE-:23) 3 2005

 

 

JAN 0 7200;

 

fi214h7

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DatoDuepGS-p. 15

_ _.___——__._.

THREE ESSAYS ON HEALTH AND MACRONUTRIENT CONSUMPTION
AMONG CHINESE ADULTS

By
ZHEHUILUO

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
THREE ESSAYS ON HEALTH AND MACRONUTRIENT CONSUMPTION
AMONG CHINESE ADULTS
By
ZHEHUILUO

The first chapter contains an in—depth analysis of the socioeconomic determinants
of adult (twenty years of age or older) body mass index (BMI) in China using the
China Health and Nutrition Survey (CHNS) 1989-1997. There is a dramatic increase
in overweight and obesity among adult Chinese in this period (for men 6% to 17% and
for women 11% to 21%). A clear theoretical framework was utilized and descriptive
and multivariate reduced form and dynamic demand function analyses were under-
taken to find a variety of factors affecting individual BMI. The key findings are that
for women the effect of education is very strong and inversely U-shaped. Productive
assets, food prices, water sources and sanitation conditions are all important determi-
nants of adult BMI in China and these factors affect men and women in different age
groups and across regions differently. With careful exposition of the socioeconomic
determinants of adult BMI in China in the 19908, the paper finds strong protective
effects of education on BMI in women, particularly in rural areas. The knowledge of
this relationship can assist public policy makers to identify target groups for improv-
ing their health status. As the Chinese economy undergoes rapid structural transition
it is extremely important to find factors that can make such transition as smooth as
possible.

The second chapter studies the associations between adult food consumption and
several socioeconomic factors such as education level, household resources, and com-
munity characteristics in the early 19903 in China. In the overall sample education

does not have significant impact on calorie intakes but does affect percent of calories '

from fat, from protein and from carbohydrates differently in different region and at
different age. The effect of productive assets is nonlinear and in inverted U—shape for
male calorie, fat and protein intakes; whereas for women more productive assets are
associated with more fat and protein intakes and more percents of calories from fat
and from protein. In rural areas the effect of productive assets is stronger than that
in the urban areas. Prices of foods, community water and sanitation conditions are
also studied. The effects of prices on calorie, fat and protein intakes and the quality
of diet measures can go in either direction. Improvements in sanitation are associated
with more energy and protein intakes in urban areas. A simple health production
function analysis on weight and BMI is carried out and both health measures in the
short period of two years can be described as a random walk process.

The third chapter presents identification of the shape of the age, cohort and time
effect profiles of male and female BMIs in 19908 in China. The analysis is used to
help pin down the model specification in the BMI demand functions later. The age
profile for women is of inverse U-shape. The year effect for men is strong. There are
not enough data to identify cohort effects due to the short length of the survey. In our
main analysis for socioeconomic determinants of BMI we will be using the five—year

cohort identification strategy.

Copyright by
ZHEHUI LUO
2003

To my family and friends

ACKNOWLEDGEMENTS

I would like to thank my advisor, Professor John Strauss, for all the time and
encouragement he gave me. Without his patient guidance this dissertation would not
have been done. I aspire to emulate his diligence in research and love for developing
countries. He taught me that many difficulties in life can be overcome by a strong
will.

I am also in great debt to my “boss”, Professor Joseph Gardiner, who throughout
the years supported me financially and emotionally. His believing in me made my
transition from a child to a scholar easier.

I thank my other committee members, Professors John Goddeeris and John Giles,
who took the time reading my drafts and made many helpful comments and sug-
gestions. I also wish to gratefully acknowledge my appreciation to Professors Jeffrey
Wooldridge and Steven Haider at MSU for suggestions made at various stages of
the research, and to Dr. Barry Popkin at CPC-UNC for providing the data used
in this dissertation. Thanks are due to professors who taught me various courses in
economics, especially Professor Warren Samuels, who benefited me greatly with his
profound knowledge and sense of history and showed me kindness beyond words.

I have been privileged to know many great people during my long years as a
graduate student at Michigan State University. Some are my teachers in my academic
pursuit and in my life, some are friends by my side through ups and downs, and some
have become like family whom I trust and love deeply. Many friends enriched my life
on campus, Facundo, Jing—I, Daiji, Ren, Pungpond and F irman, in church, Qiang,
Xiangshu, Shuangwen, Dora and many more from the Lansing Chinese Christian
Ministry. Some friends in China are still as close to my heart as ever: Sier, Zhiyong,
Fengchen, and Yongfeng.

I give special thanks to my family, my mother and my sister, for their unconditional

love and unfailing support, without which it would have been impossible for me to

vi

survive.
Finally, to James, whose humor, intelligence, good heart and love captivated me,

this dissertation is dedicated.

vii

Table of Contents

LIST OF TABLES xii

LIST OF FIGURES xvi

1 Socioeconomic Determinants of Body Mass Index of Adult Chinese

in the 19908 1
1.1 Introduction ............................ . . . . . 1
1.2 Theoretical Framework .......................... 4
1.2.1 The Model ............................. 5
1.2.2 Estimation Strategies ....................... 10
1.3 Literature Review .......................... . ., .. .. 12
1.4 Data and Descriptive Statistics ..................... 18
1.4.1 The Data ........................ , ...... 18
1.4.2 Summary Statistics ........................ 24
1.5 Reduced Form Analyses .......................... 39
1.5.1 Education ............................. 41
1.5.2 Household Resources ....................... 46
1.5.3 Marital Status ......................... .. .. . 48
1.5.4 Community Characteristics ................... ' 50
1.6 Dynamic Conditional BMI Demand ................... 71
1.6.1 With Community Dummies ................ , . . . 72

viii

1.6.2 With Community Characteristics ................ 73

1.7 Concluding Remarks ........................... 87

Adult Chinese Macronutrient Consumption and Socioeconomic De-

terminants in the Early 19908 91
2.1 Introduction ................................ 91
2.2 Literature Review ............................. 97
2.2.1 Reduced Form Demands ..................... 97
2.2.2 Descriptive Studies ........................ 98
2.2.3 Income and Nutrients ....................... 102
2.2.4 Price and Nutrients ........................ 105
2.2.5 Education and Nutrients ..................... 107
2.3 Data and Econometric Issues ...................... 108
2.3.1 The Data ............................. 108
2.3.2 Measurement Errors ............ ' .l .......... 112
2.3.31 Patterns and Trends ....................... 115
2.3.4 Estimation Methods ....................... 118
2.4 Results: Determinants of Levels ..................... 123
2.4.1 Basic Models ........................... 123
2:42 Augmented Models ........................ 128
2.5 Results: Determinants of Changes .................... 147
2.5.1 Basic Models ............ ' ............... 147

2.5.2 Augmented Models ........................ 149
2.6 Results: Production Function ...................... 154

2.7 Discussions ......................... . ....... 158

3 Age, Cohort and Year Analysis in the Socioeconomic Determinants

of Adult BMI in the 19908 178
3.1 Introduction ................................ 178
3.2 Identification of Second Differences ................... 179
3.3 Simulation Study ............................. 184
3.3.1 Models with linear and / or squared terms ............ 185

3.3.2 Models with dummy variables .................. 187

3.3.3 Profiles for y and y1 ....................... 189

3.4 BMI Profiles in CHNS .......................... 210
3.5 Estimate the Restricted Level Effects .................. 211
3.6 Concluding Remarks ............................ 221
APPENDIX ‘ 222
A Survey instruments in CHNS 223
B First Stage Regressions for Section 1.6 229

C Linear Probability Models for Undernourishment and Overweight 237

D ARI Model of Determinants of BMI 245

E Descriptive Statistics of Nutrient Intakes in CHN889-93 246

F First Stage Regressions for section 2.6 250

G The Design Matrix and Constraints Needed 255

xi

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

1.10

1.11

1.12

1.13

1.14

1.15

1.16

2.1

List of Tables

Adult BMI in CHNS 89-97 ........................
Prevalence of Adult High Blood Pressure in CHN S 1989—1997

Adult Smoking Status in CHNS 1991-1997 ...............
Basic Characteristics of Households and Individuals in CHNS 1989—97
Determinants of Adult BMI in CHNS 1989—97: Overall ........
Determinants of Adult BMI in CHNS 1989-97: Urban Areas .....
Determinants of Adult BMI in CHNS 1989-97: Rural Areas .....
Determinants of Adult BMI in CHNS 1989-97: Age 20 - 39 ......
Determinants of Adult BMI in CHNS 1989-97: Age 40 - 59 ......
Determinants of Adult BMI in CHNS 1989-97: Age 60+ .......
Adult BMI by CHNS Round 1991, 93, and 97: Overall ........

Determinants of Adult BMI in CHNS 1989-93 with Community Char-
acteristics: Overall ............................

Determinants of Adults BMI in 1993 Conditional on BMI in 1991:
Overall ...................................

Determinants of Adults BMI in 1997 Conditional on BMI in 1993:
Overall ...................................

Determinants of 1993 BMI Conditional on 1991 BMI with Changes in
Community Characteristics .......................

Determinants of 1997 BMI Conditional on 1993 BMI with Changes in
Community Characteristics .......................

Individual, Household and Community Characteristics in CHNS 89-93

xii

25

31

33

36

53

55

57

59

61

63

65

67

75

77

79

83

113

2.2
2.3

2.4
2.5
2.6

2.7
l 2.8

2.9

2.10
2.11
2.12

2.13
2.14
2.15
2.16
2.17
2.18

2.19

Average Daily Calorie, Fat, Protein and Carbohydrates Intakes: Pat-
terns and Trends in CHNS 89-93 .................... 114

Average Percent of Calorie from Fat, Protein and Carbohydrates: Pat-
terns and Trends in CHNS 89—93 .................... 117

Daily Calorie Intakes (Kcal) in CHNS 89, 91, 93 From All Food Groups 131
Daily Fat Intakes (gram) in CHNS 89, 91, 93 From All Food Groups . 133
Daily Protein Intakes (gram) in CHNS 89, 91, 93 From All Food Group8135
Percent Calorie From Fat in CHNS 89, 91, 93 From All Food Groups 137
Percent Calorie From Protein in CHN S 89, 91, 93 From All Food Group8139

Percent Calorie From Carbohydrates in CHNS 89, 91, 93 From All
Food Groups ................................ 141

Food Consumption in CHNS 89,91,93 with Community Characteris-
tics: Overall ................................ 143

Food Consumption in CHNS 89,91,93 with Community Characteristics
in Urban Areas .............................. 149

Food Consumption in CHN S 89,91,93 with Community Characteristics

in Rural Areas .............................. 154 .
Changes in Daily Calorie Intakes .................... 161
Changes in Daily Fat Intakes ...................... 163
Changes in Daily Protein Intakes .................... 165
Changes in Percent Daily Calorie From Fat ............... 167
Changes in Percent Daily Calorie From Protein ............ 169
Changes in Percent Daily Calorie From Carbohydrates ........ 171

Changes in Food Consumption in CHNS 89,91,93 with Community
Characteristics .............................. 173

2.20 A Simple Production Function Analysis on Log(weight) in 1993 and

3.1

3.2

3.3

3.4

3.5

3.6

3.7

3.8

3.9

A.1

A2

A3

A4

8.1

B2

C.1

C2

C3

Log(BMI) in 1993 .............................

Short Panel: Estimated restricted age, cohort and year effects for x

Short Panel: Estimated restricted age, cohort and year effects for x1 .
Long Panel: Estimated restricted age, cohort and year effects for x . .
Long Panel: Estimated restricted age, cohort and year effects for x1 .
Short Panel: Estimated restricted age, cohort and year effects for y

Short Panel: Estimated restricted age, cohort and year effects for y1 .
Long Panel: Estimated restricted age, cohort and year effects for y . .
Long Panel: Estimated restricted age, cohort and year effects for y1 .

Second Differences of Cohort Profiles ..................

Survey instruments for health related topics ..............
ADL type questions in CHNS 93-97 ...................
Adult BMI by Education Levels in Rural and Urban Areas ......

Community Characteristics in CHNS 1989—97 ........... ' . .

First stage regressions for Table 1.13 ..................

First stage regressions for Table 1.14 ..................

Probability Model for Overweight: Overall, Urban and Rural Areas .
Probability Model of Overweight: By Age Groups ...........

Probability Model of Undernourishment: Overall, Urban and Rural
Areas .............. g ......................

xiv

177

190

191

192

193

194

195

196

197

224

225

226

229

233

237

239

241

C4

D1

E.1

F.1

Probability Model of Undernourishment: By Age Groups ....... 243

ARl Models for BMI ........................... 245

Patterns and Trends of Individual Daily Nutrient Intakes in CHNS
1989—1993 ................................. 246

First Stage Regression for Health Production Function Analysis . . . 251

XV

. 1.1

2.1

3.1

3.2

3.3

3.4

3.5

3.6

List of Figures

Male and {Female Age Effects Identified through the specification with
age dummies, five-year cohort dummies and year dummies .......

FAO Food Balance Sheet Estimates of calories, protein and fat avail-
abilities and percent of protein and fat from vegetable and animal
products in 1961-2001. ..........................

Lowess Smoothed Male and Female BMI of Each Five- and Three-Year
Cohort vs. Age ..............................

Estimated Age and Cohort Effects for Men and Women with Arbitrar-
ily Imposed Constraints .........................

Short Panel Estimates of Age, Cohort and Year Effects for :r and 2:1.
Identifications achieved through (1) model age, cohort excludes year
effects; (2) model age, year excludes cohort effects; (3) model age, co—
hort excludes year effects; (4) model age, five-year-cohort, year effects;
and (5) model age, three-year-cohort, year effects. ...........

Long Panel Estimates of Age, Cohort and Year Effects for a: and 2:1.
Identifications achieved through (1) model age, cohort excludes year
effects; (2) model age, year excludes cohort effects; (3) model age, co-
hort excludes year effects; (4) model age, five-year—cohort, year effects;
and (5) model age, three-year-cohort, year effects. ...........

Short Panel Estimates of Age, Cohort and Year Effects for y and y1.
Identifications achieved through (1) model age, cohort excludes year
effects; (2) model age, year excludes cohort effects; (3) model age, co—
hort excludes year effects; (4) model age, five-year-cohort, year effects;
and (5) model age, three-year-cohort, year effects. ...........

Long Panel Estimates of Age, Cohort and Year Effects for y and y1.
Identifications achieved through (1) model age, cohort excludes year
effects; (2) model age, year excludes cohort effects; (3) model age, co—
hort excludes year effects; (4) model age, five-year-cohort, year effects;
and (5) model age, three-year-cohort, year effects. ...........

xvi

40

97

183

200

203

206

209

3.7

3.8

3.9

3.10

3.11

The Second Derivatives of Age and Cohort Effects for Men and Women
for BMI using two—year spaced CHNS 89, 91 and 93. * The approach
for repeated cross-sections based on means. — The approach for genuine
panel data based on individual level data. ...............

The Second Derivatives of Age and Cohort Effects for Men and Women
for BMI using four-year spaced CHNS 89, 93 and 97. * The approach
for repeated cross-sections based on means. — The approach for genuine
panel data based on individual level data. ...............

Age Effects for Men and Women, identified through (a) exact age at
interview (b) three-year cohorts, (c) five-year cohorts specifications, (d)
three—year cohorts without breaking down at certain groups into single
years, (e) five-year cohorts without breaking down at certain groups
into single years. .............................

Cohort Effects for Men and Women, identified through (a) exact age
at interview (b) three-year cohorts, (c) five—year cohorts specifications,
(d) three-year cohorts without breaking down at certain groups into
single years, (e) five—year cohorts without breaking down at certain
groups into single years. .........................

Year Effects for Men and Women, identified through (a) exact age at
interview (b) three-year cohorts, (c) five-year cohorts specifications, (d)
three-year cohorts without breaking down at certain groups into single
years, (e) five-year cohorts without breaking down at certain groups
into single years. ...................... _ .......

xvii

215

217

218

219

Chapter 1: Socioeconomic '
Determinants of Body Mass Index

of Adult Chinese in the 19908

1.1 Introduction

China, like a number of other developing countries, is experiencing a wide range
of transitions in her social, economic, and cultural structures. Since 1978 economic
reforms opened the Chinese economy to the world and revitalized a stagnant economy.
Productivity has risen and standard of living has increased substantially. Per capita
GNP increased from 170 US dollars in 1980 to 668 in 1997 (both in 1995 dollars)
and life expectancy increased from 68.3 years to 71.3 years for female and from 65.7
to 68.1 years for male during the same period ( World Development Indicators 1999).
These gains were most evident in the rural areas as a result of the implementation of
the household responsibility system (Lin 1988) as well as other reform measures such
as increases in the prices of agricultural products (Ke 1999). China’s community-
based health and sanitation programs extended services to almost all urban and rural
residents. During the 808 and 908 China has also experienced a shift in the leading
cause of death from acute infectious disease to chronic conditions, due partly to an
aging population and partly to the successful reduction in infectious disease morbidity
(Popkin et a1. 1993, Zhai et a1. 2002).

Like other countries in transition, such as Brazil, Russia, and some Asian coun-
tries, during this time, researchers found that in China shifts, desirable or undesirable,

in diet, physical activity and overweight status are among the most rapid ever docu-

mented (Popkin and Doak 1998, Popkin 1999, Popkin 2002).1 In China the dietary
pattern is moving toward one in which the proportion of energy intake from fat in-
creases each year. The intake of cereals decreased considerably during the eight-year
period of 1989 to 1997 by 127 g per person per day; the intake of vegetables decreased
by 32 g per person per day; and the intake of animal foods increased by 46.7 g and
36.8 g per person per day for urban and rural residents (Du et al. 2002). In a cohort of
prime age men and women the prevalence of overweight2 doubled for females (10.4 to
20.8%) and almost tripled in males (5.0 to 14.1%) in the same period (Bell et a1. 2001).
The increase in the level of overweight in adults is largely confined to the urban areas.
It is projected that diet-related chronic diseases, such as obesity and coronary heart
disease, will present a huge health care burden for China in the near future (Popkin,
Paeratakul, Zhai and Ge 1995a). In urban China mortality attributable to cardio-
vascular diseases increased from 86.2 per 100,000 (12.1% of total deaths) in 1957 to
214.3 per 100,000 (35.8% of all deaths) in 1990. The overall prevalence of hyperten-
sion with threshold values of 140/90 mmHg, was 12.5% in adults aged 35—64 years
in the 19908 (Reddy 2002). Increasing economic opulence brings new challenges to
the health care system which is going through a transition itself (Gu 2001, Wu 1997).
Preventive actions against the increase in obesity and diet-related non-communicable
diseases are called for. Identifying the important socioeconomic factors related to
obesity helps the designers of public health policies direct their efforts in the right
direction.

The anthropometric measure body mass index (BMI), defined by weight (kg)
divided by height squared (m2), compared with the self-reported morbidity, is an
objective measure of adult health status. To date studies on adult BMI (kg/m2)

in China, have been primarily descriptive in nature. This paper employs a clear

 

1The Bellagio Conference organized by the International Union of Nutritional Sciences Committee
on the Nutrition Transition in 2001 published a series of papers on the nutrition transition in several
developing countries Source: http://www.cpc.unc.edu/nutrition-transition/.

2Overweight is defined by body mass index (BMI) greater than or equal to 25 kg/mz.

theoretical framework to estimate socioeconomic determinants of adult BMI using
an eight-year panel data set, adding to a small number of studies that examine the
reduced form BMI demand function for adult men and women in household surveys
in developing countries. BMI demand functions provide useful information on how
exogenous factors such as prices and the community infrastructure affect health. A
dynamic or conditional BMI demand function is also estimated, which takes advantage
of the longitudinal nature of the data and avoids certain specification difficulties in
the pure reduced-form analysis (see section 1.2 for detail), although the estimates
are only partial effects of the explanatory variables in this case. The key findings
of this paper are for women the effect of education is very strong and inversely U-
shaped. The education effect is stronger in rural areas than in urban areas and
for younger generations that for older generations. Adult men in China had higher
levels of education than women, however, their educational effects on BMI were not
significant. Productive assets, prices, and environmental health conditions are all
important determinants of adult BMIs in China and these factors affect men and
women in different age groups and across regions differently. Controlling for lagged
BMI the effects of education and household resources were no longer significant in
both the OLS and 2SLS estimates.

The rest of the paper is organized as follows: the next section describes the
theory and empirical strategy. Section three reviews studies on associations between
adult BMI and socioeconomic status. Section four provides a summary of the data
and descriptive evidence for the association. Results for the reduced-form analysis are
in section five and for the conditional demand function analysis see the penultimate

section. The last section concludes.

1.2 Theoretical Framework

Three types of measurement have been used as indicators of adult health in the
literature, anthropometric measures, self-reported morbidity or general health condi-
tions, and measures of nutrient intakes. They measure different aspects of a person’s
physical condition. For example, height is an indicator of childhood nutritional and
health accumulation and BMI reflects both past and current nutritional and health
status in fat content and active tissue mass (Willett 1998). Higher BMI has been used
to diagnose obesity and lower BMI has been used as an indicator for chronic energy
deficiency, as proposed by the International Dietary Energy Consultancy Group (Ge
et a1. 1994).

Self-reported indicators of health, such as being in excellent, good, fair and poor
health, may reflect differences in perceptions of health among people in different ed-
ucational or socioeconomic groups, rather than differences in health itself. Hence the
self-assessment measures may generate biased results (Strauss and Thomas 1996).
There is no known single index that measures a person’s physical and psychologi-
cal well-being at the same time. Anthropometric measures, particularly weight and
height, are commonly accepted as measures of nutritional and health status by epi-
demiologists and nutritionists. They have also been used by economists to study
long-term trends in nutrition and mortality (Fogel 1986, Fogel 1994, Costa and Steckel
1995) and to measure standard of living over time (Whitewell and Nicholas 2001).
Whether they are good measures for health status in general is arguable. However,
they are strong markers for mortality and morbidity from certain illnesses such as
hypertension, diabetes, coronary heart disease and some cancers (NIH 1998). Based
on the Waaler curve (iso-mortality curves on height and weight plain) constructed
with data from Norwegian men aged 50-64, Fogel (1994) finds that factors associated
with improvements in height and BMI predicted the decline in mortality between

1910 and 1980 among men 65 years and older. Extreme values of BMI are signifi-

cantly associated with an increasing risk of mortality. Hence this paper uses BMI as

an indicator for adult health.

1.2.1 The Model

To understand health determination, it is necessary to specify the process by
which health is produced.3 The economic model considers a rational individual with
perfect information.4 Assume that the preferences of household members are inter-
temporally separable and that the household maximizes the present discounted value
of a common utility over health of all family members, h, = (hu, ..., hm), consumption,

2:, = (11:1,, ..., arm), and leisure, l, = ([1,, ..., (m) in time period t z

T
max fitU(ht,$t,lt;6,€h) (1.1)
t=0

t=0....,T

w.r.t {Ii-t, lita mit}i=1...n

where hit is member i’s health status in period t, 23,-; are k dimensional vectors of food
and non-food commodities consumed by member 2' = 1, ..., n, 0 = (01, ..., 6") includes
a vector of exogenous individual characteristics known to family members but not
controlled by them, such as genetic traits, and 5,, indicates unobserved household
heterogeneity in preferences or environmental factors. Single period utility is assumed
to be increasing and concave in hit (U,’, > 0, U,’,’ < 0). himxio, [1'0 and mio are given
and positive at to when a person enters adulthood. At T + 1, the person dies and
hiT+l,1‘iT+la liT+l and miT+1 are zero-

Adequate to the application of BMI, the health status at the beginning of period

t+ 1 is assumed to be influenced by health status at t, health inputs in period t,

 

3The set up in this model is similar to that in Foster (1995).

4Uncertainty is ignored in this model. There is evidence that in developing countries, for idiosyn-
cratic risks households are natural units of risk-pooling and consumption smoothing through the
accumulation and depletion of assets over time is not uncommon. Uncertainty could be important
if overweight increases the risk of mortality and the person is aware of it.

mu, such as the balance between energy expenditure (basic metabolism, working or
exercising) and intakes, medical care, or illness spell, which does not bring utility
directly. More leisure time may bring feelings of well-being and sound state of mind,
which can have a positive influence of one’s physique. Individual characteristics, 0“,
such as age,5 gender and community or environmental characteristics, 26,, may also

have a direct impact on health outcomes:

hit-+1 : fi(hit)mitalit96itathlvit) (1'2)

where 11,-, are unobservable individual, household, and community factors that affect
member i’s health. It is assumed that Bfi/Bhu > 0, and afi/Bm“ > 0. Equation
(1.2) is not a general form of health production function in that it assumes that ha
is a sufficient statistic that summarizes the effect of all past inputs and choices in
periods 1, ...,t — 1 and there is no direct lagged effects from them on hit“. A more
general health production function would allow all past inputs to be included.

A8 in: Grossman’s (1972) original work on health production, time, as in the
number of healthy days available for leisure I“ and work 13’, may also be affected by
health:6

Tit = lit +121; = Qzlhit) (1'3)

with 9: > 0.

The household maximizes utility under the standard intertemporal budget con-

 

5The solution to the maximization problem is a time-invariant policy function that determines
the control variable (my-Ll“, m“) and the state variable h“. It is conceivable as a person grows old
the effect of inputs on health (body size) varies. To incorporate such effect we can assume a health
production function that varies with age at time t, a“:

ha = fi(hu—1, "1qu 9a, 2a; viz) + git(ait)mit

6For the case where total time available for leisure and work is fixed, or not affected by health
status, labor is the same as the effective labor in Grossman’s sense. Then the equilibrium condition

I "I t m
in (1.5) will become %% = £17221? - £93117
2 m g m t

straintsz

n n 'n
At+1 = (1+ 7't+1) (At — P: E 1'21- 17;" 2: ma) + Z witlflt) 'l' yr (1-4)
i=1 i=1 i=1

AT+1 = 0 and given A0 > 0

where A, is assets, pf, p{" are vectors of prices, 20,-, is wage for individual i, and y,
is household non-wage income between t and t + 1 received at the end of period t.
There is no bequest motive and no debts in the final period T, or AT+1 = 0. The
lifetime of the household is assumed to be exogenously fixed.7 The time and budget
constraint can be collapsed to derive the full—income constraint. From the first order
conditions of maximizing (1.1) subject to (1.2) to (1.4) and 7‘, = n+1 = r we can

derive the following:

 

 

 

 

ugzn..=_1_((1+r>Pt”il_I_i€P_I"__M) (15)
U: pr‘p: fin-1 fin (1+r) '
and

fi(1+7‘)U,’,+1 = 7T it+1 (1.6)

U f, 7r it
In equilibrium in each period the household will equate marginal rate of substi-
tution between health and consumption with the shadow price of health (1.5), which
is lower than the costs for medical care because bettering current health decreases the
future medical costs and improves the healthy days or effective labor. Equation (1.6)
states that over time the marginal rate of substitution of health today and tomorrow
equates the ratio of the shadow prices.

The dynamic problem can be solved recursively and the reduced-form demand

 

7For a discussion of the length of individual lifetime as endogenous in the model of own-health
production see Grossman (2000). For simplicity in the collective household model the household’s
utility is assumed to be the utility of a dictator in the household and the lifetime of the household
becomes the lifetime of this representative and is assumed to be fixed in this model.

functions for consumption goods, health, leisure and health inputs (1:, h, l, m) are:

($,h, 1,777,): = g;,h,l (pl: Pm: wt" yia 6i, 0h: 26; via 6h) (1'7)

where p2 = (p§,p§_1,...,pg), pm 2 (p?,p?_1,...,p6"), w,- = (w,T,w,T_1,...,w,~0), y =
(yiT, y,T_1, ..., yin) and 11,36 h are unobserved individual and household heterogeneity
over the life time. As noted in Rosenzweig and Schultz (1983) without specifying
the exact form of the utility function in (1.1) a closed-form solution for the demand
equation in (1.7) cannot be obtained and the properties of these reduced-form demand
functions are “identical to those from models positing no household production of
health”.

From this clear theoretical framework we can see that estimating (1.2) by OLS
is biased — each input component is determined by and correlated to the unobserved
individual, household and community factors as shown in equation (1.7). It would
be of interest if we can identify parameters of the health production function. It is
extremely difficult to find proper instruments for all health inputs. However, in (1.7)
all the determinants, including individual, household and community characteristics
and prices, except for wage, are assumed to be orthogonal to v“, 5,. Using productive
assets as a proxy for income we estimate the relationship (1.7) using OLS, which is
consistent under standard regulation conditions, such as no measurement errors and
no omitted variables.

Little is known about the effects of underlying socioeconomic factors on adult
health in China, especially during time of transition. What factors have driven or
helped contain the dramatic increases in overweight and obesity? What is the effect
of schooling on adult health? Are there gender or regional differentials in health? An
unconditional reduced-form demand function for health provides a helpful first step

in finding answers to such important questions.

Clearly, in order to estimate the reduced-form demand function we need the
entire paths of prices and wage rates for each individual, which is not available in
any household survey and may create a saturated model problem. If we pool all the
waves together and use a vector of current prices and a dummy variable for each
community assuming all past and future prices and other unobservables within a
community follow a unique path captured by this dummy variable, then we can avoid
the omitted variable bias. Obviously one dummy variable can not stand in for all
past and future prices. We use community and year interaction dummies to capture
some variations for each community in time.

As is specified in the health production function, assuming the lagged health is
a sufficient statistic for all past information, we can also get around the need of the
entire history of prices by estimating a conditional dynamic health demand function.
Assuming the utility function is not linear in hit in (1.6) using the first order expansion
of the marginal utility of health between today and tomorrow and substituting other
control variables in shadow prices from (1.7) we can write the conditional health

demand function as:

:r m h ,
hit-+1 = C(hit) pt+ipt+a wit+a yt-l-a git-fa 6t+v th-l-s vii-fa €ht+) (18)

where Pf+ = (Pfipf—ia waif-+1), Pf: = (1733,}??4, ---,Pi’11), wit+ =(w,-T,w.-T_1,...,

wit“), yt+ = (yiT,y,-T_1, ...,y,t+1) and vi”, Eh,+ are unobserved individual and house-
hold heterogeneity from time t forward. As we can see the conditional health demand
function does not depend on all lagged prices and wage rates any more. By assum-
ing all future prices within each community could be summarized by a community

dummy variable we can estimate the following relation

:1: m h .
hit+1 = C(hitapt+1ypt+1awit+la yt+1,9u+1, 9H1, th+la Vi) (1-9)

with the composite error term V,- that includes all unobserved individual and house-
hold heterogeneity and future wage and nonwage income that is not observable to
researchers. The dynamic demand model estimates the partial effects of individual,
household and community variables conditional on past BMI.

In this paper we focus on the reduced-form and conditional demand for BMI of
adult household members. By construction, all covariates in (1.7) are exogenous, as-
suming there is no selective migration to a better health environment and no program
placement in response to poor health. The first assumption is not unreasonable in
the early 19908 in China. Though increasingly the restriction of household registry
has been relaxed, the migration from rural to urban area in China is still largely
temporary. The second assumption may be problematic in rural areas if the govern-
ment did intervene in responding to local need by investing more in health facilities

or increasing the number or the training of health professionals.

1.2.2 Estimation Strategies

The basic estimating equation takes the form hi, = Xafi + v, + 6,, where hit is
person i’s BMI in year t, X“ is a vector of exogenous variables including the individ-
ual’s age, cohort, education level and household resources, and v.- is the individual
random effect. Although we have longitudinal data we still don’t observe all past and
future prices and other households and community characteristics. When prices and
community characteristics are not explicitly controlled for, using community dummy
variables we effectively assume all past and future prices and other unobservables
within a community follow a unique path captured by this dummy variable. One
dummy variable may not stand in for all past and future prices. Thus we use com-
munity and year interaction dummies to capture some variations in each community
over time. When prices and community characteristics are explicitly controlled for,

we also include the community dummy variables for unobserved heterogeneity and

10

allow the interaction of community and year to be captured by variations in prices
and other community characteristics in different years.

Since it is established that both the upper and lower tails of the BMI distribution
are related to worse health, a natural question is why not estimating a dichotomized
outcome of being overweight or undernourished. There are two reasons. First and
most importantly, if the goal is to estimate the differential impacts of X (includ-
ing a constant) at different points or ranges of the BMI distribution, then a linear
probability model (LMP) or a probit model for a dichotomized variable does not
achieve this goal. Think of a latent variable framework. Let y“ = X b + e and y = 1
if y‘ > 0. A linear probability model estimates P(y = 1 IX) = XELPM and a
probit model estimates P(y = 1 IX) = (P(X’lfprobit). Since P(y = 1) = P(y“ >
0) = P(Xb+ e > 0) = P(e < Xb), if e ~ uniform(0, 1) then P(y = 1) = XI).
Similarly, if e ~ normal (0, 1) then P(y = 1) = <I>(X b). Hence both the linear prob-
ability model and the probit model are estimating the same latent relationship and
the direction in the estimates of BMI greater than certain cutoff point should be the
same as in the OLS estimates of BMI itself. In appendix C we provide some results
from a linear probability estimation of undernourishment and overweight.

Secondly, dichotomization loses information and results in inefficiency. The non-
linear effects of any X variables on BMI at different points or ranges of the BMI
distribution cannot be efficiently estimated. In order to identify such nonlinearity
one solution might be gleaned from the dynamic demand function with interactions
of the lagged y and the main variable of interest, such as y; = f (y{_1)g(X b) + e.

So far we have not assumed any distribution of c, in hit = Xufi + v,- + 6,. In esti-
mating the basic model, we do allow arbitrary correlations between et and e¢_1 at the
individual level and use cluster robust standard errors in all regressions for inferences.
We could also assume 6, follows an AR(1) process and estimate a linear model with

such disturbance 6; == pet_1 + u,, where |p| < 1 and u, is independent and identically

11

distributed with mean 0 and variance 0,2,. We have unbalanced panels whose observa-
tions are unequally spaced over time, i.e., not all individuals were surveyed in 1989,
1991, 1993 and 1997. We use the Baltagi and Wu (1999) approach to estimate the
model with AR(1) disturbance in which p represents autocorrelation between every

two-yearly spaced data. The results are included in Appendix D for comparisons.

1.3 Literature Review

Studies on association between adult health and socioeconomic status in the US.
concentrate on mortality or self-reported morbidity, functional limitation or general
health (Attanasio and Hoynes 2000, Smith 1998, Smith and Kington 1997). In de-
veloping countries, studies on adult health determinants are scant and sometimes
difficult to interpret due to lack of theoretical or modelling clarity (for a comprehen-
sive review see Behrman and Deolalikar 1988 and Strauss and Thomas 1998). Studies
on adult BMI, according to their nature - the questions asked and the methodologies
used to achieve those goals, can be classified into three categories: (i) descriptive
and monitoring studies; (ii) determinants of health in reduced form demand function
studies; and (iii) health production function studies. In this paper we focus on the
first two types of studies.

Obesity (defined as BMI>30) is increasingly a global health issue. According
to WHO, the prevalence of obesity has increased about 15—20 percentage points in
Western and Southern Europe and is even higher in Eastern Europe (Sundquist and
Johansson 1998). The proportion of overweight US adults (BMI>25) increased from
25.4% to 33.3% between 1976-80 and 1988-91 periods (Kuczmarski et a1. 1994).

Using a longitudinal Swedish Annual Level-of-Living Survey of men and women
aged 25-74 between 1980/81 and 1988/89, Sundquist and Johansson (1998) found

that there was a clear gradient for the attained level of education and BMI, with

12

the highest BMI for people with a low level of education. All educational groups
increased their BMI over 8 years except for men with a low education level, however,
only poorly educated females had a higher BMI adjusting for age, smoking status,
exercise level, marital status, general health status, ethnicity and time. Since most
of these variables in the multivariable analysis are endogenous , this may bias the
results. The authors thus look at the change in BMI between two surveys. They
also found that exercise habits were more changeable than smoking habits and men
and women who stopped smoking had a larger increase in BMI than never smokers.
However, for the change of BMIs they only performed univariate regressions.

Using individual level data in 1987 and 1995 Ruhm (2000) examined the vari—
ation of risky behaviors, smoking, and time-intensive health investments in physical
activity, diet and preventive medical care with the status of the economy controlling
for fixed-effects at state level. He found higher jobless rate is associated with reduced
smoking and obesity and with increased physical activity and improved diet. When
the economy starts to grow the opportunity cost of time will increase and some risky
activities may become normal goods. He found strong evidence that “health improves
when the economy temporarily deteriorates” and some evidence that “the unfavor-
able health effects of temporary upturns are partially or fully offset if the economic
growth is long-lasting.”

Realizing the long-run growth in body weight has been accompanied by almost
stable total calorie consumption and significant decreases in the relative price of
food, Lakdawalla and Philipson (2002) outlined a theory of weight management that
will predict such combined results through technological changes on both the supply
(agricultural innovation) and the demand (more sedentary production) side. Based
on their hypothesis the partial effect of the price of food on weight is negative, which
was interpreted as the supply side effect of technological change. The effect of the

strenuousness of production (S) on weight is also negative, which indicates that the

13

reduction in S due to sedentary technological change results in the increase in weight.
The relationship between income and weight can be positive, negative or inverted
U-shaped. A negative pure income effect combined with a positive indirect income
effect through changing the strenuousness of work can still produce increased weight
over time. Using the National Longitudinal Survey of Youth (N LSY) the authors
showed evidence that the partial effect of exercises is concave and occupation is ex-
ogenous with respect to weight. Individual fixed effect models of BMI on job-related
strenuousness and a one—year BMI on average level of strenuousness and strength
across all years reveal that a one-year-one-unit increase in average strenuousness low-
ers women’s BMI by 0.19 unit and the 14-year effect is about six times as large as the
one-year effect. Switching into less strenuous jobs are not preceded by increases in
BMI. People who switch into less strenuous occupations do not already weigh more
than others. Using the National Health Interview Survey 1976—94 the authors esti-
mated individual’s BMI as a function of year trend, the strength requirement of a
job (>0), job strenuousness (<0), income quartiles, education, age and age squared,
race and marital status. The main findings were: the effect of job strenuousness is
larger than other economic factors stressed as key determinants such as income and
education; income effect for men is inverse U-shaped and for women is negative.

Is obesity an inevitable price to pay for economic growth and technological in-
novation in developing countries as well or is there a unique feature that distin-
guishes them from the developed countries at a similar stage of economic development
(Popkin 2002)? Scholars in the US. and China joined their effort in finding some
answers for such questions. They describe the changing dietary and body weight
patterns among the Chinese and document the association between dietary, environ-
mental factors and obesity (Popkin et al. 1993, Ge et a1. 1994, Popkin, Paeratakul,
Zhai and Ge 1995a, Popkin, Paeratakul, Zhai and Ge 1995b, Popkin, Paeratakul and
Zhai 1995, Paeratakul et al. 1998, Stookey et al. 2000). In a cohort of prime age

14

men and women the prevalence of overweight doubled for females (10.4 to 20.8%)
and almost tripled in males (5.0 to 14.1%) in the same period (Bell et al. 2001). The
intake of cereals and vegetables decreased considerably and the intake of animal foods
. increased. These studies provide various indicators on whether the health status of
the target population is improving or deteriorating over time. There is evidence that
dietary energy and fat intakes were positively and significantly correlated with the
BMI; urban residence and higher income were correlated with lower energy intake,
higher fat intake and lower physical activity level compared to rural residence and
other income categories. These studies have a common problem which makes the
interpretation of their results difficult or misleading. They are estimates of correla-
tions between the inputs and BMI and should not be interpreted as causal effect from
a health production function. The inclusion of endogenous factors as exogenous ex-
planatory variables tends to bias the parameters in a production function estimation.

In Ge et al. (1994), using the Chinese Health and Nutrition Survey (CHNS) in
1989 (adults aged 20 to 45) and a large Chinese Nutrition Survey in 1982 (adults
of all ages), the authors found that increased income was significantly associated
with reduced BMI in the urban sample, while for the rural and overall samples the
opposite was true. Subpopulations consuming greater proportions of energy from
animal sources were more likely to be overweight. The percentage of underweight
(BMI< 18.5) was low (7 to 13%). The results were based on simple analysis of variance,
which does not establish causal relationships nor control for other factors.

In Popkin et al. (1993), again using the CHNS 89 and statistics from the Chi-
nese bureau, the authors found that per capita cereal and vegetable consumptions
increased in the early 808 and levelled off later. Per capita meat, edible oils, sugar,
eggs and fish increased each year, resulting a marked shift in the structure of the diet
to one with high proportion of energy from fat. Cross tabulations of income and ma-

jor food groups suggested that greater consumption of higher fat food was associated

15

with higher income levels.

For the elderly in China, the CHNS 91 and 93 were used (Stookey et al. 2000)
to estimate the determinants of nutrition intakes and BMI. The dependent variables
included energy, fat, protein intakes, consumption of rice, high-fat red meat, eggs,
plant oils, and BMI. The explanatory variables were age groups, sex, income tertiles,
and rural-urban residence. The prevalence of low BMI (< 18.5 or <22.0) exceeded 15%
and the prevalence of overweight (BMI > 25.0 or > 27.0) ranged between 4% and
24%. Being overweight was found to be significantly and positively correlated with
urban residence in 1991 and 1993 and higher income in 1991. The authors attempted
to correct sample selection bias due to missing values in dietary or anthropometric
measures in 1993 using such variables as household size, province dummies, and types
of roads. Their correction may not be very convincing because if they are estimating
reduced form demand functions there is no reason that their selection correction
variables should not be included in the main regression. The authors found that the
group with missing BMI or dietary information in 1991 was significantly older, from
larger families, and were less likely to live in Jiangsu province in 1991. The group
with missing values in 1993 was significantly older, from larger families, in rural areas,
and in the top income tertile. Bearing the possibility that their correction is poorly
identified we note their results below. Male and younger age groups have significantly
higher energy intakes while urban residents have significantly lower energy intakes but
higher proportion of energy from fat and protein. Increasing income was significantly
associated with greater energy, fat and protein intakes. Rice consumption declined
with older age and consumption of plant oils, high-fat red meat and eggs increased by
income tertiles. Income was positively associated with the probability of having very
low physical activity level and inversely associated with very high physical activity
level. Undernutrition was concluded to be a more pressing problem for the elderly as

BMI declined significantly with age. The coexistence of overweight and underweight

16

in the elderly suggests more future demand of health care provision.

Simple analysis of covariance was employed to find the role of diet and socioe-
conomic factors as the determinants of BMI in Popkin, Paeratakul and Zhai (1995),
using the CHNS 1989 and 1991 for adults 20 to 45 years of age. For BMI in 1991 the
covariates were total energy intakes, fat intakes of 1989, energy from sources other
than fat, household income, age, place of residence, physical activity level, and smok-
ing habits. This is an example where all covariates in a health production function
were treated as exogenous. The parameters may be under or overestimated. BMI
increased with energy and fat intakes and was negatively correlated with smoking
and physical activity levels.

In a subsequent study (Paeratakul et al. 1998), the authors employed the fixed-
effect method at individual level to estimate the effect of changes of the above variables
on the changes in BMI. They found only the change in fat intakes was positively
associated with changes in BMI in men and changes in physical activity level in women
were negatively associated with changes in BMI in women and all other factors were
no longer significant. Their corresponding cross-section analysis using OLS indicated
that both fat and energy intakes were positively associated with BMI in men and only
energy intakes were positively correlated with BMI in women. Age, urban residence
and household assets (possession of certain major durable goods) were positively
related to BMI. Education level was negatively related to BMI in women and marital
status positively related to BMI in men. The authors acknowledged the inconsistency
of the results from the OLS and the fixed-effect model yet did not give satisfactory
explanations. One source of the discrepancy may be due to the potential endogeneity
in their OLS estimation. Unobserved shocks to BMI, for example income shocks,
will affect both intakes and physical activity level resulting in endogeneity bias in all
estimates.

Among the few reduced form adult health determinants studies using anthropo-

17

metric measures, Thomas et al. (1996) estimates a BMI demand function for adults
in C6te d’Ivoire. The authors control for the joint determination of health and con-
sumption by instrumenting per capita expenditures with value of land, livestock and
assets of the household under the assumption of weak separability between leisure and
consumption. The results of this study indicate that higher food prices are associated
with lower BMI, especially among rural residents of C6te d’Ivoire. Doubling the price
of beef and fish in rural area will reduce BMI by 10% to 20% among rural dwellers.
Increasing the price of plantains, eggs and manioc would have an even larger impact
on BMI. The authors also find a significant positive impact of per capita expenditure
on BMI and find evidence to indicate that the health of men would be less affected
than that of women by declines in household resources. Education only had some
impact on BMI for urban females controlling household resources.

In this paper, taking advantage of the detailed community level price, health
facility and environment data, employing a clear theoretical model, we estimate the
reduced form and conditional BMI demand function and find potential socioeconomic
determinants. This paper fills in the gap of previous research due to lack of distinction
between endogenous and exogenous variables in the model and becomes the first paper
on Chinese adult BMI estimations with emphasis on individual as well as community

characteristics.

1.4 Data and Descriptive Statistics

1.4.1 The Data

The Chinese Health and Nutrition Survey (CHN S), conducted by the Carolina
Population Center at the University of North Carolina at Chapel Hill (CPC-UNC),
the Institute of Nutrition and Food Hygiene (INFH) and the Chinese Academy of

Preventive Medicine (CAPM), is designed “to examine the effects of health, nutri-

18

tion, and family planning policies and programs implemented by national and local
governments and to see how the social and economic transformation of Chinese soci-
ety is affecting the health and nutritional status of its population.”8 It is important
for the government to understand the impact of the economic transition on people’s
health and be prepared for increasing demand for certain health care services. The
wide range of survey instruments of the CHNS provides researchers an useful tool to
better describe and understand the associations between socioeconomic and health
status and to find out potential determinants for health and nutrition status so that
the government may adjust its program emphases and provide better care for the
people.

The survey covered 8 provinces (Guangxi, Guizhou, Henan, Hubei, Huan, J iangsu,
Liaoning, and Shandong) that vary substantially in geography, economic development,
public resources, and health indicators in 1989, 1991, 1993 and 1997. A multistage,
random cluster process was used to draw the sample. Counties in the 8 provinces
were stratified by income (low, middle, and high) and a weighted sampling scheme
was used to randomly select 4 counties (one in low, two in middle and one in high
income counties) from each province. In addition, the provincial capital and a lower
income city were selected. Villages and townships within the counties and urban and
suburban neighborhoods within the cities were selected randomly. There are about
190 primary sampling units and 3,800 some households covering approximately 16,000
individuals. For the survey design and sample structure see related articles (Popkin
et al. 1993, Stookey et al. 2000). The same household was contacted in each survey.
In 1993 an effort was made to locate the individuals who started a new household
or change to a different household in the same region. In 1997 Liaoning province was
replaced by Heilongjiang province, which is similar to Liaoning in many characteris-

tics. According to the Chinese census definition, small towns and city neighborhoods

 

8http: / /www.cpc.unc.edu / projects / china/

19

are defined as urban areas, and villages and suburbs as rural areas.

This paper focuses on adults 20 years and older. In the reduced-form BMI
demand analysis all four waves of the data are pooled together. In 1989, health and
nutrition data were only collected from preschoolers and adults age 20 to 45. The
first wave in 1989 thus includes only adults age 20 to 45 with consistent sex and
age measures from the basic module and the physical exam module.9 The number
of adult Chinese aged 20 to 45 in CHNS-89 is 6,578 and the number of individuals
with BMI measures is 5,058.10 In the second wave, the number of all adults aged
20 and above with consistent age information from the roster and from the physical
exam section11 in CHNS-91 is 10,461, among whom 8,260 have BMI measures. The
number of adults aged 20 and above with nonmissing gender12 in CHNS-93 is 9,839
and a subsample of 7,706 people have BMI measures. Finally in CHNS-97 there are
11,338 observations for adults age 20 and above with nonmissing gender information
and 8,356 adults with BMI measures. Each year after 1991 missing values in BMIs are
more likely to be men and people who are considerably younger and more educated.
It may be because at time of interview these people are more likely to be out or off
for work. For the final sample size in our analysis after accounting for all missing
values see Table 1.1.

The household survey section includes basic demographic characteristics of all

 

9In the web-based 1989 adult physical examination data set there are 5,183 observations for

adults age 20 to 49, among whom 56 people have different gender information from the physical
exam section than from the roster and 95 individuals have more than 2 years of age difference (age
calculation based on lunar calender and western calender may result in up to 2 years of difference)
between age in the physical exam section and the roster. Cleaning and correction are based on date
of birth and relationship to the household head.

10Pregnant women are excluded in all years.

11There are 330 individuals with the age difference more than 2 years. They are excluded for data
quality consideration. Because age in 1991 and forward is calculated based on date of birth it can
not be used for data cleaning again. 137 adult members in the 1989 survey died by the time of the
second round of CHNS.

l2Sequentially excluded from the original data set are those who are less than 20 years of age,
who are dead, who have missing gender information. Missing gender cannot be identified based on
relationship to the household head because in 1993 and 97 son and daughter to the household head
were given the same code in the variable.

20

household members, detailed income from different sources, time allocation at home
and economic activities, ownership of consumer durable goods and living environ-
ments.

In most of the previously reviewed papers, the focus was on the correlation
between BMI and income controlling for intakes and/or physical activities. They
are hybrid models that differ from a pure production function estimation and a pure
reduced-form demand function estimation. Incomes in developing countries have
been found to be particularly error-ridden. Improved health status may enhance
productivity and income in reverse causality. Attenuation bias and endogeneity bias
of incomes are not easily corrected. As a result this paper uses productive assets in the
main analysis.13 The nominal value of productive assets are self-reported ownership
and value of tricycles, motorcycles, tractors or walking tractors, irrigation equipment,
power threshers and water pumps. Items not available in all years are not included.14
Productive assets are discounted to the 1988 value. All price indices are based on
the China Statistical Yearbooks (SSB 1988-1998) for urban and rural areas at the
provincial level. The CHN S research team at CPC complied their own price index
based on an urban consumer basket, the SSB urban price data, and the CHNS urban
rural price ratio for each year in each province. The amount of land (mu) cultivated
in the previous year is used as an independent variable.

One important factor in health production is education. In CHNS both years
of formal education and highest level of attainment were asked. For about less than
one percent of the people years of education were top-coded at 18 years. I chose and
categorized the highest level of education by none, some primary schooling, primary

degree, middle school degree, high school diploma, and technical, vocational and

 

13Results for comparison purpose on some limited analysis of the effect of income on BMI treating
income as endogenous and instrumented by productive assets are available upon request

l“Productive assets in 1989 are calculated differently from other years. All productive assets
owned by 1989 are evaluated at last year’s purchasing prices. In the other years the total value was
self-reported current worth.

21

college or higher degree.

The health services section contains health insurance coverage of surveyed indi-
viduals, medical providers, and health facilities that the household might use. Infor-
mation on illness and uses of the health system during the previous month is asked
for some household members. Physical examinations that include weight, height and
blood pressures were given to all adults and children starting from 1991 and to adults
aged 20 to 45 in 1989. Physical functioning data were collected since 1993 for the
elderly. Three parts of the CHNS concern an interviewee’s health (Table A.1 in
Appendix A summarizes what is available year-wise in each part of the survey instru-
ments). First, in the household health and medical services module, there are ques-
tions about a person’s self-evaluation of her health status, recent illness and injury,
medical insurance and expenditure (treatment and transportation costs), diagnoses
and facilities. Secondly, in the anthropometric measurement and physical examina-
tion module, weight, height, upper arm circumference, waist and hip circumferences
(available in 1993 and 1997), and blood pressure are measured for adults (in 1989
only for adults aged 20 to 45 years old). For different age and gender groups, illness
or symptoms pertaining to them were asked. Past and current smoking and alcohol
behaviors, injury history are recorded. For the elderly (age 50 and older in 1993, and
over 55 in 1997) activities of daily living (ADL)-type questions were asked (Table A2
in Appendix A lists the 15 ADL—type questions). Thirdly, in the ever-married women
module, marriage history and birth history are surveyed in detail starting from 1993,
so is infant food record.

Internationally accepted adult BMI cutoffs are used to define underweight as
a BMI<18.5 kg/m2, overweight as a BMI Z 25 kg/m2 and obese as a BMI 2 30
kg/m2 for adults 20 years and older (the WHO standards).15 The National Institute

 

15For the elderly, Stookey et al. (2000) have used a different set of cutoffs and suggested them
to be more pertinent, which defines overweight and obesity for 60 years and older people as a BMI
_>. 22 kg/m2 and BMI Z 27 kg/mz. Recently the China Center for Disease Control and Prevention
(CCDC) proposed a cutoff for Chinese population to classify overweight as a BMI 2 24 kg/m2 and

22

of Heart, Lung and Blood clinical guidelines (NIH 1998) for identifying and treating
obesity suggests if a person’s waist circumference is greater than 88 cm (for women),
or 102 cm (for men),16 the risk of having certain illnesses are greatly increased when
they are overweight. The guideline points out that waist circumferences greater than
the cutoffs for overweight adults indicate excess fat in the abdomen out of proportion
to total body fat, which is an independent predictor of morbidity for men and women
with a BMI of 25 to 34.9 kg/m2. These cutoff points lose their incremental predictive
power in persons with a BMI Z 35 kg/mz.

The community survey section includes information on infrastructure, services,
populations in the village or neighborhood, percentage of land with poor quality, daily
wage for unskilled farmers and construction workers, percent of work force engaged
in agriculture or working out of town for more than one month, and hospital and
clinic infrastructures and personnel. All questions were answered by a knowledgeable
respondent. In each community, state ration coupon, retail and free market prices
for most commonly consumed rice, wheat, egg, pork, beef and fish were collected in
stores. Since free market prices reflect the value of each produce better they were
in the reduced-form analysis with community characteristics. Every health service
and family planning provider or facility were identified and information about person-
nel, prices and availability of services was collected. The percentages of households
within each community with certain types and sources of water and toilet facility are

aggregated from household level data.

 

obesity as a BMI 2 28 kg/m2. Since this decision is not finalized and for purpose of comparison
with other studies we still use the WHO cutoffs in this paper.

16Similarly, CCDC proposed a set of different cutoff points for women at 80 cm and for men at
85 cm.

23

1.4.2 Summary Statistics
BMI, Overweight and Undernutrition

Trends for BMIs and prevalence of being undernourished, overweight, obese and
overweight with high abdominal fat for men and women in different age, residential
and education groups are in Table 1.1.17 Over the eight-year period, 1989 to 1997,
average BMI for men and women increased by about 1 unit. To put things into
perspective, that is about 7.4 pounds of increase in weight for a men 6 foot tall.
However the overall prevalence of overweight increased from 6% to 17% in men and
from 11% to 21% in women. The percent of men and women who are obese also
increases dramatically. Using the waist circumference cutoffs in the previous session
the percent of overweight women with increasing risks is about 9% in 1997 and the
percent for men is lower, probably because the cut-off point is too high for Chinese
men. There is an improvement or decrease in undernutrition for all men (by one
percentage point) and women (by about 1.5 percentage point) over time.

Breaking the sample into different age groups we can see men and women in
prime age (40 to 59) have the highest BMIs. The prevalence of being overweight
in the younger cohort (age 20 to 39) increased from 6 to 14 percent for men and 9
to 15 percent for women in this eight-year period. Nine to 19 percent of prime age
men are overweight and the percent for women in the same age is much higher (20
percent in 1989 to 25 percent in 1997). Elderly women (aged 60 and above) during
1991 to 1997 have the same risk of being overweight as prime age women and 15 to 23
percent of the elderly men are overweight. In all age groups the prevalence of obesity
is higher for women than for men, so is the prevalence of being overweight with high

abdominal fat. The percent of men and women being undernourished decreased the

 

17Pregnant women were excluded. Having high abdominal fat is defined as having waist circum-
ference > 102 cm for male or > 88 cm for female. The sex-specific cutoffs can be used to identify
increased relative risk for the development of obesity-associated risk factors in most adults with a
BMI of 25 to 34.9 kg/m2 (NIH 1998). In 1989 data on BMI were collected for those 20 to 45 years
of age. In 1989 and 1991 data on waist circumference were not collected.

24

most for prime aged men and the elderly. The incidence of underweight is much
higher among the elderly than the other age groups due partly to shrinking. This
pattern of overweight is very similar to that found in Indonesian Living Standard
Survey (Strauss et al. 2004). Although the prevalence of overweight and obesity in
men and women is much lower than that of the United State, the increasing trend

among all age groups is certainly of concern.

Table 1.1: Adult BMI in CHNS 89-97

 

 

 

 

 

 

Male Female
1989 1991 1993 1997 1989 1991 1993 1997
All Sample

Number of observations 2196 3537 3329 3698 2354 3819 3597 ' 3882
Median BMI 21.00 21.13 21.34 21.78 21.48 21.50 21.64 22.05
Mean BMI 21.23 21.60 21.88 22.34 21.74 22.10 22.14 22.66
(2.27) (3.06) (3.59) (3.55) (2.65) (4.36) (3.71) (4.10)

% Undernourished ((18.5) 8.15 9.81 8.53 7.25 9.22 10.81 9.87 7.73
(0.27) (0.30) (0.28) (0.26) (0.29) (0.31) (0.30) (0.27)

% Overweight (225) 6.24 10.88 12.11 17.41 11.00 16.05 16.35 20.97
(0.24) (0.31) (0.33) (0.38) (0.31) (0.37) (0.37) (0.41)

% Obese (230) 0.27 1.02 1.59 2.16 0.76 2.33 1.95 2.99
(0.05) (0.10) (0.13) (0.15) (0.09) (0.15) (0.14) (0.17)

% Overweight risky “ 0.66 1.24 6.39 8.73
(0.08) (0.11) (0.24) (0.28)

Age 20—39 years

Number of observations 1791 1771 1519 1597 1939 1965 1671 1602
Median BMI 20.96 20.97 21.16 21.57 21.36 21.23 21.29 21.51
Mean BMI 21.16 21.42 21.73 22.06 21.57 21.74 21.62 22.23
(2.19) (2.83) (3.50) (3.34) (2.51) (4.61) (3.18) (4.09)

% Undernourished ((18.5) 7.43 7.57 6.65 7.01 9.59 9.36 10.23 6.62
(0.26) (0.26) (0.25) (0.26) (0.29) (0.29) (0.30) (0.25)

% Overweight (225) 5.58 7.68 9.28 13.78 9.08 11.04 10.29 14.86
(0.23) (0.27) (0.29) (0.34) (0.29) (0.31) (0.30) (0.36)

% Obese (230) 0.17 0.68 1.58 1.69 0.41 1.07 0.72 2.06
(0.04) (0.08) (0.12) (0.13) (0.06) (0.10) (0.08) (0.14)

% Overweight risky 3‘ 0.59 0.56 2.21 3.75
(0.08) (0.07) (0.15) (0.19)

 

Table continues

The differences between urban and rural areas are striking. The prevalence of
overweight or obesity in men in urban areas is more than twice of that in rural
areas. In 1997 the percent overweight is 29% for urban men and 12% for rural men.

The difference between women is also large (14 to 27% in urban women and 10 to

25

Table 1. 1 (cont’d)

 

 

 

 

 

 

Male Female

1989 1991 1993 1997 1989 1991 1993 1997

Age 40—59 years
Number of observations 405 1213 1225 1449 415 1280 1341 1542
Median BMI 21.41 21.45 21.56 22.09 22.19 22.04 22.21 22.72
Mean BMI 21.56 21.88 22.11 22.55 22.51 22.67 22.78 23.09
(2.59) (3.04) (3.63) (3.42) (3.12) (3.88) (3.95) (3.82)
% Undernourished (<18.5) 11.36 8.41 8.00 5.52 7.47 9.45 7.01 6.29
(0.32) (0.28) (0.27) (0.23) (0.26) (0.29) (0.26) (0.24)
% Overweight (225) 9.14 13.36 14.12 19.05 20.00 21.88 22.07 25.23
(0.29) (0.34) (0.35) (0.39) (0.40) (0.41) (0.41) (0.43)
% Obese (230) 0.74% 1.07 1.47 2.21 2.41 3.52 3.21 3.18
(0.09) (0.10) (0.12) (0.15) (0.15) (0.18) (0.18) (0.18)
% Overweight risky ‘ 0.16 1.04 9.25 10.44
(0.04) (0.10) (0.29) (0.31)

Age >=60 years
Number of observations 553 585 652 574 585 738
Median BMI 21.05 21.22 21.80 21.46 21.58 22.15
Mean BMI 21.59 21.78 22.51 22.03 22.17 22.69
(3.70) (3.72) (4.21) (4.33) (4.29) (4.58)
% Undernourished (<18.5) 20.07 14.53 11.66 18.82 15.38 13.14
(0.40) (0.35) (0.32) (0.39) (0.36) (0.34)
% Overweight (225) 15.73 15.21 22.70 20.21 20.51 25.34
(0.36) (0.36) (0.42) (0.40) (0.40) (0.44)
% Obese (_>_30) 1.99 1.88 3.22 4.01 2.56 4.61
(0.14) (0.14) (0.18) (0.20) (0.16) (0.21)
% Overweight risky ‘ 1.88 3.37 11.79 15.99
(0.14) (0.18) (0.32) (0.37)

Urban Areas
Number of observations 665 1223 1051 1148 732 1346 1151 1264
Median BMI 21.05 21.82 21.97 22.86 21.49 22.01 22.17 22.86
Mean BMI 21.36 22.21 22.58 23.32 21.85 22.61 22.76 23.28
(2.67) (3.47) (4.08) (3.94) (2.87) (3.81) (3.92) (4.38)
% Undernourished (<18.5) 10.83 9.65 8.47 5.84 10.38 9.66 7.65 6.57
(0.31) (0.30) (0.28) (0.23) (0.31) (0.30) (0.27) (0.25)
% Overweight (225) 9.47 17.83 20.36 28.57 14.21 22.73 23.28 26.82
(0.29) (0.38) (0.40) (0.45) (0.35) (0.42) (0.42) (0.44)
% Obese (230) 0.30 1.55 2.57 3.22 1.09 3.42 3.30 4.03
(0.05) (0.12) (0.16) (0.18) (0.10) (0.18) (0.18) (0.20)
% Overweight risky " 1.14 2.53 10.43 13.13
(0.11) (0.16) (0.31) (0.34)
Rural Areas

Number of observations 1531 2314 2278 2550 1622 2473 2446 2618
Median BMI 20.98 20.94 21.10 21.45 21.44 21.28 21.43 21.77
Mean BMI 21.18 21.28 21.56 21.89 21.69 21.82 21.86 22.36
(2.08) (2.77) (3.30) (3.26) (2.54) (4.61) (3.57) (3.92)
% Undernourished (<18.5) 6.99 9.90 8.56 7.88 8.69 11.44 10.92 8.29
(0.26) (0.30) (0.28) (0.27) (0.28) (0.32) (0.31) (0.28)
% Overweight (225) 4.83 7.22 8.30 12.39 9.56 12.41 13.08 18.14
(0.21) (0.26) (0.28) (0.33) (0.29) (0.33) (0.34) (0.39)
% Obese (230) 0.26 0.73 1.14 1.69 0.62 1.74 1.31 2.48
(0.05) (0.09) (0.11) (0.13) (0.08) (0.13) (0.11) (0.16)
% Overweight risky ‘ 0.44 0.67 4.50 6.61
(0.07) (0.08) (0.21) (0.25)

 

26

Table continues

Table 1.1 (cont’d)

 

 

 

 

 

 

Male Female

1989 1991 1993 1997 1989 1991 1993 1997

No Formal Education
Number of observations 76 392 335 299 361 1226 1094 1034
Median BMI 21.02 21.29 21.23 21.30 21.71 21.51 21.60 21.93
Mean BMI 21.46 21.71 21.77 21.99 22.08 22.06 22.09 22.54
(2.26) (3.22) (3.39) (3.86) (2.78) (3.75) (3.71) (4.58)
% Undernourished (<18.5) 3.95 10.71 9.55 11.37 6.65 13.95 12.07 11.90
(0.20) (0.31) (0.29) (0.32) (0.25) (0.35) (0.33) (0.32)
% Overweight (225) 7.89 13.27 13.13 17.73 12.47 17.94 17.64 22.05
(0.27) (0.34) (0.34) (0.38) (0.33) (0.38) (0.38) (0.41)
% Obese (230) 0.00 1.53 1.19 1.00 1.39 3.10 2.01 3.48
(0.00) (0.12) (0.11) (0.10) (0.12) (0.17) (0.14) (0.18)
% Overweight risky ‘ 0.30 0.67 8.04 11.41
(0.05) (0.08) (0.27) (0.32)

Some Primary School
Number of observations 291 624 775 548 417 584 807 606
Median BMI 21.00 20.83 21.20 21.39 21.64 21.57 21.75 22.04
Mean BMI 20.92 21.34 21.89 21.81 21.82 22.10 22.26 22.65
(2.04) (3.37) (4.01) (3.14) (2.61) (3.92) (3.79) (3.60)
% Undernourished (<18.5) 9.62 13.94 10.58 10.40 9.35 10.79 9.05 7.10
(0.30) (0.35) (0.31) (0.31) (0.29) (0.31) (0.29) (0.26)
% Overweight (225) 3.09 11.22 12.90 11.13 9.59 13.18 15.74 22.44
(0.17) (0.32) (0.34) (0.31) (0.29) (0.34) (0.36) (0.42)
% Obese (230) 0.00 0.96 2.71 1.46 1.20 2.57 2.11 3.14
(0.00) (0.10) (0.16) (0.12) (0.11) (0.16) (0.14) (0.17)
% Overweight risky “ 1.29 1.28 6.44 10.07
(0.11) (0.11) (0.25) (0.30)

Primary Degree

Number of observations 501 765 589 767 484 663 458 731
Median BMI 21.06 21.01 21.37 21.54 21.63 22.02 21.95 22.55
Mean BMI 21.26 21.41 21.65 22.10 21.86 22.61 22.70 22.85
(2.21) (2.68) (3.21) (3.60) (2.58) (6.77) (4.32) (3.40)
% Undernourished (<18.5) 8.18 10.46 8.83 8.21 7.64 7.84 6.77 5.61
(0.27) (0.31) (0.28) (0.27) (0.27) (0.27) (0.25) (0.23)
% Overweight (225) 5.79 9.41 9.00 13.56 12.40 17.04 19.00 22.98
(0.23) (0.29) (0.29) (0.34) (0.33) (0.38) (0.39) (0.42)
% Obese (230) 0.60 0.65 1.02 2.35 0.62 2.26 2.40 2.87
(0.08) (0.08) (0.10) (0.15) (0.08) (0.15) (0.15) (0.17)
% Overweight risky “ 0.51 0.78 7.21 8.89
(0.07) (0.09) (0.26) (0.28)

Lower Middle School
Number of observations 812 1058 1042 1224 689 846 808 915
Median BMI 20.85 21.05 21.20 21.66 21.34 21.36 21.48 21.82
Mean BMI 21.18 21.56 21.82 22.26 21.64 21.88 22.01 22.62
(2.31) (3.05) (3.51) (3.39) (2.65) (3.34) (3.54) (4.48)
% Undernourished (<18.5) 8.13 8.98 7.10 5.56 10.16 8.87 9.78 6.34
(0.27) (0.29) (0.26) (0.23) (0.30) (0.28) (0.30) (0.24)
% Overweight (225) 6.28 8.79 11.04 16.34 11.18 14.78 14.23 18.25
(0.24) (0.28) (0.31) (0.37) (0.32) (0.36) (0.35) (0.39)
% Obese (230) 0.37 1.32 1.34 1.80 0.44 1.54 1.98 3.06
(0.06) (0.11) (0.12) (0.13) (0.07) (0.12) (0.14) (0.17)
% Overweight risky “ 0.48 0.90 5.07 6.67
(0.07) (0.09) (0.22) (0.25)

¥

 

27

Table continues

Table 1.1 (cont’d)

 

 

 

 

Male Female
1989 1991 1993 1997 1989 1991 1993 1997
Higher Middle School

Number of observations 335 439 387 491 273 342 294 346
Median BMI 21.01 21.45 21.60 22.45 21.10 20.93 21.34 22.19
Mean BMI 21.28 21.78 22.17 22.91 21.41 21.82 21.65 22.81
(2.35) (2.95) (3.81) (3.91) (2.58) (3.73) (2.71) (4.27)

% Undernourished (<18.5) 8.06 6.38 6.98 5.70 12.82 10.82 9.18 4.91
(0.27) (0.24) (0.26) (0.23) (0.33) (0.31) (0.29) (0.22)

% Overweight (225) 7.76 12.07 13.18 24.03 9.52 14.62 15.31 19.36
(0.27) (0.33) (0.34) (0.43) (0.29) (0.35) (0.36) (0.40)

% Obese (230) 0.00 0.46 1.81 2.44 0.00 2.05 0.00 2.31
(0.00) (0.07) (0.13) (0.15) (0.00) (0.14) (0.00) (0 15)

% Overweight risky ‘ 0.52 2.04 2.38 4.91
(0.07) (0.14) (0.15) (0.22)

Tech / College+

Number of observations 181 259 201 369 130 158 136 250
Median BMI 21.50 22.14 22.21 22.88 20.81 21.09 20.93 22.14
Mean BMI 21.72 22.50 22.48 23.39 21.29 21.92 21.91 22.56
(2.45) (3.15) (3.15) (3.45) (2.68) (3.05) (3.71) (3.23)

% Undernourished (<18.5) 7.73 5.79 8.46 4.88 9.23 9.49 9.56 7.20
(0.27) (0.23) (0.28) (0.22) (0.29) (0.29) (0.30) (0.26)

% Overweight (225) 8.84 17.37 19.90 29.27 8.46 17.72 15.44 19.20
(0.28) (0.38) (0.40) (0.46) (0.28) (0.38) (0.36) (0.39)

% Obese (230) 0.00 1.16 0.50 4.61 1.54 0.63 2.94 1.60
(0.00) (0.11) (0.07) (0.21) (0.12) (0.08) (0.17) (0.13)

% Overweight risky ‘ 0.50 2.71 6.62 6.80
(0.07) (0.16) (0.25) (0.25)

 

 

Note: Standard deviations are in parentheses (a) Overweight risky or having high abdom-
inal fat is defined as having waist circumference > 102 cm for male or >88 cm for female.

Waist circumferences were not measured before 1993.

18% in rural women). Except for the year 1989 rural areas had higher prevalence of

underweight than urban areas. Even though men and women in rural areas are more

likely to engage in strenuous work they do not have higher BMIs. Undernutrition

in the rural areas is still about 8 percent although it had been decreasing over the

years. Different patterns of BMI distributions between urban and rural areas suggests

different needs in care and facilities in different regions.

The second part of Table 1.1 shows the distribution of BMI over different edu-

cation groups for men and women. The prevalence of overweight men is the highest

among the highly educated (29%) but for women it is among those with primary or

less than primary education (22%). Patterns for obesity are similar. For men the

28

percent underweight peaked at the group with some primary schooling (11%) and for
women those without any formal education had the highest prevalence in undernu-
trition (12%). This suggests higher education is protective in terms of undernutrition
but harmful in terms of overweight for men; and for women higher education may

help to reduce both under— and overweight.

Hypertension

Overweight and obesity lead to adverse metabolic effects on blood pressure and
cholesterol level. The Risk of coronary heart disease increases steadily with increasing
BMI. Adults with systolic blood pressure at Or above 140 mmHg or diastolic blood
pressure at or above 90 mmHg, or adults taking antihypertensive agents are considered
having hypertension.18 Table 1.2 summarizes the prevalence of hypertension in men
and women in our sample. In the overall sample the rates increased from 8 to 23%
in men and from 6 to 19% in women. There is a gap between men and women at
younger age but the gap closed up as people grow older. In the 20 to 39 age group
the percent of men having hypertension is twice of that of the women; in the 40 to 59
age group and the elderly the difference is about 4 percentage points. The prevalence
is higher for older' men and women than the younger ones. Less than 10 percent men
and women have high blood pressure when they are less than 40 years old, but after
age 60 half of the men and women are hypertensive.19

Rural and urban areas again have significant differences in prevalence of hyper-
tension. The percent of men and women with high blood pressure in urban areas is

almost twice as high in 1991 and 40 percent higher than that of the rural areas in

1997. Combined risks of overweight and hypertension in urban residents put more

 

18In 1989 the question about using antihypertensive agents was not asked. The definition in 1989
is hence based on blood pressure only. In 1991 to 97 blood pressures were taken three times for
accuracy. The average of these measures are used for the definition.

19There is a dip in the prevalence of hypertension in 1993 among all age groups because of a lower
diastolic average. We are not sure why this is the case.

29

stress on the health care facilities in those areas and the trend is increasing.

Across different education level the risks of hypertension is the highest for the
least educated, 39% for men and 32% for women with no formal education. Inter-
estingly education ceases to be protective at the college or higher level. There is
steady decrease in prevalence of hypertension until people with higher middle school
diploma. This pattern is the same for men and women as oppose to the different

patterns in the BMI distribution across education levels for men and women.

30

Table 1.2: Prevalence of Adult High Blood Pressure in CHNS 1989-1997

 

 

 

 

 

 

Male Female
1989 1991 1993 1997 1989 1991 1993 1997
All Sample
Number of observations 2193 3531 3326 3691 2334 3816 3592 3875
% High Blood Pressure 8.39 16.26 10.34 22.87 5.53 13.36 10.38 18.94
(0.28) (0.37) (0.30) (0.42) (0.23) (0.34) (0.31) (0.39)
Age 20-39 years
Number of observations 1788 1767 1518 1594 1925 1962 1668 1599
% High Blood Pressure 7.33 6.28 2.50 10.73 4.00 3.01 1.32 4.82
(0.26) (0.24) (0.16) (0.31) (0.20) (0.17) (0.11) (0.21)
Age 40—59 years
Number of observations 405 1211 1225 1447 409 1280 1341 1539
% High Blood Pressure 13.09 18.33 8.41 23.43 12.71 16.41 10.89 19.95
(0.34) (0.39) (0.28) (0.42) (0.33) (0.37) (0.31) (0.40)
Age >=60 years
Number of observations 553 583 650 574 583 737
% High Blood Pressure 43.58 34.82 51.38 41.99 35.16 47.49
(0.50) (0.48) (0.50) (0.49) (0.48) (0.50)
Urban Areas
Number of observations 664 1222 1049 1147 723 1346 1149 1263
% High Blood Pressure 10.54 23.16 16.11 28.42 6.09 19.17 16.01 23.67
(0.31) (0.42) (0.37) (0.45) (0.24) (0.39) (0.37) (0.43)
Rural Areas
Number of observations 1529 2309 2277 2544 1611 2470 2443 2612
% High Blood Pressure 7.46 12.60 7.69 20.36 5.28 10.20 7.74 16.65
(0.26) (0.33) (0.27) (0.40) (0.22) (0.30) (0.27) (0.37)
No Formal Education
Number of observations 76 392 334 299 357 1226 1092 1032
% High Blood Pressure 13.16 29.34 18.86 39.46 7.00 23.57 18.96 32.36
(0.34) (0.46) (0.39) (0.49) (0.26) (0.42) (0.39) (0.47)
Some Primary Education
Number of observations 291 623 774 545 414 583 805 605
% High Blood Pressure 8.25 17.66 14.99 26.61 7.49 12.01 10.31 18.84
(0.28) (0.38) (0.36) (0.44) (0.26) (0.33) (0.30) (0.39)
Primary Degree
Number of observations 501 764 589 766 479 661 458 731
% High Blood Pressure 8.38 16.75 9.00 25.20 6.47 11.80 8.30 17.65
(0.28) (0.37) (0.29) (0.43) (0.25) (0.32) (0.28) . (0.38)
Lower Middle School
Number of observations 811 1056 1041 1222 685 846 807 911
% High Blood Pressure 7.52 11.08 6.44 17.68 4.09 4.73 3.10 10.21
(0.26) (0.31) (0.25) (0.38) (0.20) (0.21) (0.17) (0.30)
Higher Middle School
Number of observations 334 437 387 491 271 342 294 346
% High Blood Pressure 9.58 13.50 7.24 17.52 2.95 4.68 3.06 8.09
(0.29) (0.34) (0.26) (0.38) (0.17) (0.21) (0.17) (0.27)
Tech/College
Number of observations 180 259 201 368 128 158 136 250
% High Blood Pressure 8.33 17.37 8.46 23.37 4.69 10.76 8.09 14.40
(0.28) (0.38) (0.28) (0.42) (0.21) (0.31) (0.27) (0.35)

 

 

Note: Adults with systolic blood pressure at or above 140 mmHg or diastolic blood pres-
sure at or above 90 mmHg, or adults taking antihypertensive agents are considered having

hypertension. In 1989 the question about using antihypertensive agents was not asked.

31

Smoking

Current and past smoking behaviors for cigarettes or pipes were asked in 1991,
93 and 97 surveys. Table 1.3 summarizes the percentage of people who at the time
of the interview smoked cigarettes or pipes. The cumulative hazards of smoking de-
pends on several factors including age at which smoking began, duration of smoking,
number of cigarettes smoked per day, and the tar and nicotine content of cigarettes.
Therefore current prevalence of smoking is a proxy for the cumulative hazards. Smok—
ing causes substantially increased risk of mortality from lung cancer, heart disease,
chronic respiratory diseases and second-hand smoking is also a health risk.

There is a slight decrease in prevalence of smoking in men but a slight increase in
women, with the overall percentage much higher in men (66%) than in women (5%).
This is generally the pattern in Indonesia and other Asia countries (Strauss et al.
2004). For men the prevalence is the highest at prime age and decreases after age 60;
for women, however, the prevalence is the highest among the elderly. The difference
between urban and rural men is significant (rural men with 6 to 10 percentage point
higher) but not significant between urban and rural women.

The effect of education on smoking is negative and graded among women with
higher educated women less likely to smoke; and among men only those with college
or higher education had a slightly lower prevalence of smoking. It seems that smoking
for men is a habit much more difficult to break. The decreased trend in smoking in
men is true in all education categories. However, the trend in women is different. At
the lower education levels (less than primary degree) the prevalence decreases in 1991
to 93 and increases in 1997; and at the higher education levels (more than middle

school diploma) the prevalence increases first and then decreases.

32

Table 1.3: Adult Smoking Status in CHNS 1991—1997

 

 

 

 

 

 

 

 

Male Female
1991 1993 1997 1991 1993 1997
All Sample
Number of observations 3537 3329 3689 3819 3597 3873
% Smoking 68.82% 66.60% 62.27% 4.58% 4.56% 4.73%
(0.46) (0.47) (0.48) (0.21) (0.21) (0.21)
Age 20-39 years
Number of observations 1771 1519 1594 1965 1671 1600
% Smoking 69.45% 65.96% 62.17% 0.87% 1.20% 2.50%
(0.46) (0.47) (0.49) (0.09) (0.11) (0.16)
Age 40-59 years
Number of observations 1213 1225 1445 1280 1341 1537
% Smoking 72.79% 71.43% 67.82% 6.88% 6.41% 4.68%
(0.45) (0.45) (0.47) (0.25) (0.25) (0.21)
Age >=60 years
Number of observations 553 585 650 574 585 736
% Smoking 58.05% 58.12% 50.15% 12.20% 9.91% 9.65%
(0.49) (0.49) (0.50) (0.33) (0.30) (0.30)
Urban Areas
Number of observations 1223 1051 1145 1346 1151 1261
% Smoking 64.76% 61.94% 54.93% 5.50% 5.47% 4.28%
(0.48) (0.49) (0.50) (0.23) (0.23) (0.20)
Rural Areas
Number of observations 2314 2278 2544 2473 2446 2612
% Smoking 70.96% 68.74% 65.57% 4.08% 4.13% 4.94%
(0.45) (0.46) (0.48) (0.20) (0.20) (0.22)
No Formal Education
Number of observations 392 335 298 1226 1094 1029
% Smoking 66.33% 65.07% 58.72% 8.08% 7.13% 7.68%
(0.47) (0.48) (0.49) (0.27) (0.26) (0.27)
Some Primary Education
Number of observations 624 775 544 584 807 604
% Smoking 71.31% 67.87% 66.91% 5.31% 4.83% 6.46%
(0.45) (0.47) (0.47) (0.22) (0.21) (0.25)
Primary Degree
Number of observations 765 589 767 663 458 731
% Smoking 71.50% 70.29% 64.67% 4.07% 4.80% 5.88%
(0.45) (0.46) (0.48) (0.20) (0.21) (0.24)
Lower Middle School
Number of observations 1058 1042 1220 846 808 913
% Smoking 69.00% 67.08% 64.26% 1.42% 2.10% 1.86%
(0.46) (0.47) (0.48) (0.12) (0.14) (0.14)
Higher Middle School
Number of observations 439 387 491 342 294 346
% Smoking 66.97% 64.08% 60.29% 0.88% 1.70% 1.16%
(0.47) (0.48) (0.49) (0.09) (0.13) (0.11)
Tech/College
Number of observations 259 201 369 158 136 250
% Smoking 61.00% 55.72% 49.32% 1.90% 2.21% 0.40%
(0.49) (0.50) (0.50) (0.14) (0.15) (0.06)

 

 

Note: Smoking status for people who at the time of the interview smoked cigarettes
or pipes was not surveyed in 1989.

33

Household Characteristics

Characteristics of household resources in the four years of the CHN S are summa-
rized in Table 1.4. This period marks China’s most noteworthy economic growth and
expansion. From 1989 to 1997 the median real per capita income in 1988 currency
increased from 424 yuan to 711 yuan and the mean went from 677 to 1039 yuan.20
The increase in rural areas is bigger in percentage in rural areas (77% in mean and
84% in median) than in urban areas (30% in mean and 60% in median). The gap
between urban and rural areas is decreasing. The median of real per capita income in
rural areas in 1989 is 35% of the median in urban areas and the percentage increases
to 40%. The mean ratios went up from 0.44 to 0.59. Although the gap is still large
it is closing.

Real productive assets are discounted to the 1988 value.21 The mean real produc-
tive assets increased from 243 yuan in 1989 to 636 yuan in 1997. Rural households
have more productive assets for engaging in agricultural activities and small busi-
nesses. The differences between rural and urban areas are getting smaller. From
1989 to 1997 productive assets increased significantly in both rural and urban areas,
from 130 to 470 yuan on average in urban and from 294 to 714 yuan in rural areas.
The amount of land cultivated each year also increased but urban households were
involved with very little farming. The amount of land farmed in the rural areas may
be a good proxy for income where there is no or little other opportunities for small
businesses or off-farm labor. Combining productive assets and the amount of land

farmed we have a proxy for the household resources.

 

20Price index is based on the China Statistical Yearbook (SSB 1988-1998) for urban and rural
areas at the provincial level. The total income in CHN S includes both earned and unearned income.
Categories surveyed in each year are: income from wage, home gardening, household farms, farming
collectives, raising livestock / poultry, household fishing, fishing collectives, household small business,
welfare subsidies and income from other sources.

21The items included each year for productive assets are tricycles, motorcycles, tractors or walking
tractors, irrigation equipment, power threshers and water pumps. Values are self-reported purchasing
values in 1989 and current worths in all other years.

34

Individual Socioeconomic Characteristics

Since in 1989 our sample includes only those who were less than 46 years old
and with higher education level than adults of all ages, when comparing the trends in
education level we look at the year from 1991 to 97. The average years of education
increased by about half a year. The overall prevalence of people with no formal
education decreased from 22 to 18 percent. Between 91 and 97 the percent of adults
with less than primary degree decreased and the percentages of all other higher levels
of education increased.

There is some gender differential in education level between men and women.
Men had 1.7 years more of education than women in 1991, 2 years more in 1993 and
2.1 years more in 1997. The percent of men in lower education levels (no formal
education, some primary schooling and primary degree) decreased and the percent of
men in other educational levels (middle school degrees, college and higher degrees)
increased. For women only the percent of no formal education decreased in prevalence.

The employment rates decreased from 82% in 1991 to 77% in 1997 due partly to
aging. The employment rates for men in all years were about 10 percentage points
higher than those for women. There is little change in percent of married men and
women over time. In our sample the increase in divorce rates is small (less than 1

percent).

Community Characteristics

Table A4 summarizes some characteristics of the communities in 1989 to 1997.
There is a dip in 1989 in the increasing trend of price of rice, wheat, beef and fish, but
not for pork which had increased the most. The price of eggs went down slightly. The
hospitals and clinics in 1997 were not surveyed and hence the reduced-form analysis
with community characteristics in the next section used only data from 1989 to 1993.

There is improvement in clean water availability in that the percent of households

35

within each community with water from water factories or underground was increasing
over time and the percent with water from an open well, spring or river etc. was
decreasing. The percent of households with in-house-flush toilet was also increasing,

suggesting improvement in sanitation environment.

Table 1.4: Basic Characteristics of Households and Individuals in CHNS 1989-97

 

 

 

 

 

1989 1991 1993 1997
All Households
Number of observations 2736 3402 3164 3512
Real PC INC Mean (yuan) 676.96 658.86 822.24 1038.55
Median 423.93 495.74 535.34 711.04
Real Productive Assets Mean (yuan) 242.73 266.11 425.37 635.60
Land Cultivated Mean (mu) 2.49 2.54 2.65 3.20
Urban Households
Number of observations 853 1114 977 1124
Real PC INC Mean (yuan) 1101.68 1014.26 1209.49 1435.06
Median 779.43 944.38 974.84 1253.26
Real Productive Assets Mean (yuan) 130.37 146.59 376.62 469.80
Land Cultivated Mean (mu) 0.26 0.22 0.20 0.15
Rural Households
Number of observations 1883 2288 2187 2388
Real PC INC Mean (yuan) 484.56 485.82 649.25 851.91
Median 274.96 322.91 346.56 503.13
Real Productive Assets Mean (yuan) 293.62 324.30 447.15 713.64
Land Cultivated Mean (mu) 3.50 3.67 3.75 4.63

 

36

Table continues

Table 1.4 (cont’d)

 

 

 

1989 1991 1993 1997
All Individuals
Number of observations 4550 7356 6926 7580
Mean Age 31.74 41.86 43.06 44.11
Mean Years of Education 7.32 6.01 6.04 6.58
% No Formal Education 9.60 22.00 20.63 17.59
% Some Primary 15.56 16.42 22.84 15.22
% Primary Diploma 21.65 19.41 15.12 19.76
% Lower Middle School 32.99 25.88 26.71 28.22
% Higher Middle School 13.36 10.62 9.83 11.04
% Technical/College 6.84 5.67 4.87 8.17
% Employed 95.65 82.02 80.06 76.86
% Married 84.15 82.22 82.49 81.56
% Divorced 0.42 0.69 0.53 0.75
% Widowed 0.29 5.76 6.11 6.12
Men

Number of observations 2196 3537 3329 3698
Mean Age 31.77 42.01 43.01 43.72
Mean Years of Education 8.15 6.97 7.11 7.71
% No Formal Education 3.46 11.08 10.06 8.09
’70 Some Primary 13.25 17.64 23.28 14.82
% Primary Diploma 22.81 21.63 17.69 20.74
% Lower Middle School 36.98 29.91 31.30 33.10
% Higher Middle School 15.26 12.41 11.63 13.28
% Technical/College 8.24 7.32 6.04 9.98
% Employed 97.71 86.45 84.40 82.51
% Married 81.33 83.04 83.24 81.64
% Divorced 0.46 0.65 0.51 0.81
% Widowed 0.18 3.22 3.00 2.81

 

Table continues

37

Table 1.4 (cont’d)

 

 

1989 1991 1993 1997
Women

Number of observations 2354 3819 3597 3882
Mean Age 31.71 41.73 43.10 44.48
Mean Years of Education 6.52 5.21 5.14 5.58
% No Formal Education 15.34 32.10 30.41 26.64
% Some Primary 17.71 15.29 22.44 15.61
% Primary Diploma 20.56 17.36 12.73 18.83
% Lower Middle School 29.27 22.15 22.46 23.57
% Higher Middle School 11.60 8.96 8.17 8.91
% Technical/College 5.52 4.14 3.78 6.44
% Employed 93.73 77.92 76.05 71.49
% Married 86.79 81.46 81.79 81.48
% Divorced 0.38 0.73 0.56 0.70
% Widowed 0.38 8.12 8.98 9.27

 

 

Note: Productive assets in 1989 are calculated differently from in other years. All
productive assets owned by 1989 is evaluated at last year’s purchasing prices. In
the other years the productive assets are the reported total value. Income and
productive assets are discounted to year 1988. Land is the total amount (1 mu=667

square meters) farmed by the household in the previous year.

38

1.5 Reduced Form Analyses

As was briefly outlined in section two, the estimation of the reduced-form BMI
demand function (1.7) uses the OLS method on data pooled over all survey years. To
account for error correlations for each individual over time the individual-level cluster
robust standard errors are used in all regressions. In all regressions age dummies and
five-year cohort dummies are included. In a separate paper (Luo 2003a) we consider
the classic identification problem in estimating age, cohort and year effects in a model.
We chose the five-year cohort dummies so that the perfect linear dependency of the
three variables no longer exists and the system can be identified. Surely there are
many other identification strategies as noted in our paper but we chose the five-year
cohorts around the early 19608’ famine and the cultural revolution period as being
one meaningful strategy. In a model including only the age dummies, five-year cohort
dummies and year dummies the estimated BMI age profiles for men and women are
shown in Figure 1.1. Both profiles are inverse U-shaped. The age effect for women
peaked around 45 to 50 and for men at a little later ages but decreasing at a slower
rate.

First, the basic specification focuses on the effects of individual education and
household covariates - productive assets and land, controlling for community level fac-
tors by community dummies or community year interaction dummies. This was done
for the whole sample, for urban and rural areas separately, and for each age group
(20—39, 40—59 and 60+) of men and women (Table 1.5 to 1.9). These stratifications
enable us to find differences between different groups of people and the rural-urban
stratification is consistent with the sampling process because the rural households
were over-sampled. As outlined in section 2 the community dummies capture the ef-
fects of unobserved past and future community level information, assuming there is
no variation over time. When such assumption seems unattainable using the com-

munity year interaction dummies partially resolves it based on a weaker assumption

39

that the variation into the future follows the same path in the past.

The marital status of the individual is added to the baseline model to examine
whether it had incremental effects controlling for other factors. The marital status
variables are not included in the basic modelling due to the concern of potential
selection biases with assortative marriage behavior. In rural areas men with higher
BMI may be considered having more ability to provide for the family and hence
are more likely to be married. For young urban women being slim could be more
attractive in both marriage market and labor market. The endogenous biases can go
both ways. Although from the descriptive statistics in the previous section there is
little change over time in marriage rates it is commonly perceived the divorce rates
are increasing in China after the reform.

—— Male Age Effect __..__ Fmale Age Effect
2.40 '7
2.00 4
1.60 a

1.20 ‘
0.80 ‘

0.40 - A
om _ [AV . 2 ii XAyflVA
414o- l\//**“xffl\[

-0.80 -*

Male/Female Age Effects
Identified with Five-year cohorts

-1.20 ‘

 

-1.60 *

 

20 26 30 35 4o 45 50 55 so 66 7o 75
age

Figure 1.1: Male and Female Age Effects Identified through the specification with

age dummies, fiveyear cohort dummies and year dummies.

40

Secondly, adding to the basic model the community characteristics including
food prices, clinic conditions, water, toilet and sanitation conditions, the augmented
model studies the joint effects of individual, household and community variables on
adult BMIs. As noted in section two, these added community variables are only
current prices and infrastructure. Community dummies are used to represent the
unique path of past and future distributions in all excluded factors. Since the ef-
fect of food prices on BMI can be either positive or negative depending on whether
income effect or substitution effect is bigger, it warrants an empirical testing to see
what the actual effects are. For communities with better health facilities, water and
sanitation conditions we’d expect better environment being beneficial assuming there
is no selective migration to a better health environment and no program placement in
response to poor health. The first assumption is not unreasonable in the early 19908
in China. Though increasingly the restriction of household registry has been relaxed,
the migration from rural to urban area in China is still largely temporary and not
primarily for health reasons. The second assumption may be problematic in rural
areas if the government did take effort in responding to local need by investing more
in health facilities or increasing the number or the training of health professionals.

All community characteristic variables are treated as exogenous.

1.5.1 Education
Whole Sample

A set of dummies for the highest degree obtained with no formal education as
reference is used to assess the nonlinear effects of education on BMI.22 In the overall
pooled sample for women (Table 1.5 column 1) having some primary education is

associated with 0.23 unit higher BMIs than those with no formal education at 0.1

 

22A linear term in completed years of education, or a linear spline of years of education are also
tried and found to fit less well than the set of dummies for highest degree obtained.

41

significant level, but the impact rises significantly when one finishes primary schooling
(0.47 unit). To put things into perspective, for a 5.5 foot woman a 0.47 unit increase in
BMI is equivalent to a 3—pound increase in weight assuming fixed height. Completing
senior high school, technical school or college is associated with a lower BMI than
having no education (-0.02 and -0.7) or lower education levels, similar to the finding of
Thomas et al. (1996) for women in C6te d’Ivoire. On average a 5.5 foot woman with
a college equivalent degree weighs 4 pounds lighter than an uneducated counterpart.
With community dummies or with community-year interaction dummies the strong
education effects remain at the same magnitude and are always jointly significant.
The nonlinear effect of education for women needs further investigation. Schooling
can affect health status either by raising the technical efficiency with which inputs are
used or by increasing the allocative efficiency of input use leading to increased incomes
(Welch 1970) or increasing women’s power and influence in household decision-making
(Behrman and Wolfe 1984). A reduced form analysis does not address the question
of how education affects the outcome, however we do observe for whom the effect of
education is the strongest.

For men education does not have any statistically significant effect for any ge-
ographic or age group, although in column 1 and 2 in Table 1.5 for men there is a
graded positive effect of education before the college level. It may suggest that there
is a different learning curve for men from women, or that BMIs of men are affected
not by his own education but by the education of the person who prepares the meal or
chooses the diet pattern. Without knowing who is the decision maker in such matters
in a household we can only hypothesize these channels. The F-statistic for testing

equality of coefficients between men and women confirms the significant differences.

42

Urban and Rural Areas

In Table 1.6, the effect of education remains mainly in the same direction for ur-
ban women except for those with some primary schooling, but none of the effects are
statistically significant. The effects for men changed shapes to high school educated
men with highest BMI followed by those not formally educated. The fact that edu-
cation in urban areas does not affect BMI significantly for men and women indicates
the most effective way of preventing overweight and obesity in urban areas may not
lie in improvement on formal education.

A closer look at Table A3 in the appendix A reviews that the patterns of BMI
distribution across different education groups for women in the whole sample are the
same for the patterns in the rural sample, but are different than that in the urban
sample (not statistically significant with a p—value from the F—statistic being 0.28).
That is why in Table 1.7 the effects of all education levels for rural women are of the
same pattern as that in the overall sample. Although the raw distribution of BMI
across different education levels for rural women indicates that the small number of
women with college equivalent education had a higher BMI than those with no formal
education, the conditional effect of college education in Table 1.7 column 1 is strong
and negative. Since the prevalence of overweight for rural females is the highest
among the no education group, the decreasing effects of higher education may be
deemed as helpful in preventing obesity.

Comparing Table 1.6 with Table 1.7 the patterns of urban and rural male edu-
cational effects are the same for people with less than high school diploma and the
effects switched signs between rural and urban men with higher educations. The ur-
ban education effects are bigger in magnitudes although none of them are significant
and none of the differences between urban and rural men based on F statistics are
significant either. However, none of these patterns showed up in the overall sam-

ple regressions. The need for stratifying the sample is partly due to the clustering

43

sampling strategy of the data collectors.

Across Age Groups

As seen in Table 1.8 for younger generations there is a strong negative graded
education effects and it starts to show at middle school level and is stronger in high
school group (-O.8 compared to -0.4 in overall sample) and in the college level (-1 vs
-0.7). To put it in perspective, a decrease of 1 unit in BMI for a 5.5-foot tall woman
would result in a weight loss of about 6 pounds.

For the prime age women (Table 1.9) the effect of education only is significant
for primary school completion and it is positive due perhaps to increased income.
The women in this age group had a much higher proportion with no formal edu-
cation (35%) as compared with the younger females (10%) and a higher proportion
with primary schooling (21%) as compared with 17 percent in the younger genera-
tion. For all other categories of education the prime age women had a much lower
percentage. Thereason that the coefficients for middle school and higher education
are not significant may be due to small sample sizes in those cells. Jointly the effect
of education is significant at 10% level when we control for community dummies, not
community-year interaction dummies.

For the elderly, education has no significant impact any more (Table 1.10), though
the effect of higher education did change sign and became positive.

Some of the differences of educational effects among women in different age
groups may be due to aging. As we have seen the age BMI profile for women peaks
in the prime aged group. The protective effects of education wear off over the years
as women age. It could also be because in the younger cohort the education effects
are close to those in the deve10ped countries if we think of education as a proxy for
income. The younger generation in our sample had a much higher education level

than older women (3 years more than the prime aged women and 6 years more than

44

the elderly). The strong and positive effects of some or complete primary education
in women aged 40 to 59 are consistent with most findings in developing countries
where increased income is associated with higher BMI. For the elderly this positive
income effect only kicks in for those very highly educated.

For prime age men and elderly men the F-statistics for testing equality of coef-
ficients between men and women suggest there is no systematic differences between
them; and for the younger cohorts the difference is significant. Instead of the graded
negative effects there is no particular pattern of the male education effects.

We do not claim our specification is complete; there can always be omitted factors
in a given reduced form analysis. Generally a fixed-effect model can be used if there
is more than one year of data. However, because education does not vary much over
time it is difficult to identify its impact with a fixed effect model.23

The strong nonlinear effect of education continues to exist after adding commu-
nity characteristics in addition to the community dummies (see the following sections).
Women with primary degrees are the ones with highest BMIs while those with tech-

nical or college degrees the lowest. The effect of education for female BMI over and

above household resources and community characteristics is apparent.

Across Years

Is the effect of education getting close to that of the developed countries? Sep—
arate regressions for men and women in each round of the survey since 1991 were
examined with age and community dummies included in all years. Heteroscedastic
robust standard errors are in parentheses. According to the hypothesis we should see
the effect of education grows more negative at each educational level as time goes
by. If such was the case it could be due to increased technical efliciency. However as

we can see in Table 1.11 it is not the case. If anything, the effects of education on

 

23The same is true when it comes to identifying the marital premium on BMI for men.

45

female BMI actually decreased over time; and the effects of male education on BMI
were positive and increased over the years, though not significant.

For men the effects are still close to that of a typical developing country and are
actually increasingly 80 though not significant statistically. For both men and women

there is room for improvement through bettering education.

1.5.2 Household Resources

Poorer households live in areas where fewer health facilities and practitioners
are available and the living and sanitary conditions are worse. If such is the case,
it is essential to control for household resources so that the estimates of community
characteristics do not reflect household characteristics. Instead of using non-wage
income24 we control household resources using log of real productive assets discounted
to 1988 value as indicators for long-run resources. Splines around the median of the
positive values of real productive assets and the amount of land cultivated (mu)
are included in all regressions. The coefficients for the lower and upper spline of
productive assets represent the level effects, not the marginal effects. Interestingly,

the effects of productive assets vary across different strata as well.

Whole Sample

For the whole sample (Table 1.5), the effect differs between men and women.
For men, more productive assets are associated with higher BMI, with larger effects
in the upper spline range and the joint effects are highly significant. For women,
higher level of productive assets are associated with higher BMI only for those with
assets less than the median; above the median the association is negative but not
significant; and the joint effects are only significant at 0.1 level. The amount of land

farmed was never significant statistically and had very small impact, though again

 

2“Real per capita income splines were also tried but not reported due to potential measurement
errors in and endogeneity bias due to income.

46

men and women have opposite signs (negative for men).

The splines of real productive assets were used to capture the non-linear effects
of household resources on BMI. Among those with more than median resources an
increase of 100 yuan in real productive assets is associated with a 0.18 unit increase
in BMI for men, so is the same increase among those with less than median resources
for women. For a 6—foot man that’s an increase of weight by 1.4 pound and for a
5.5-foot woman by 1.1 pound. A8 evident from the F statistics for testing equality
of coefficients of productive assets between men and women the difference is not
significant.

The difference between different productive assets levels could be an artifact
because of the high correlation between the two splines; or it could be capturing the
nonlinear effects of assets nicely. The fact that they are jointly significant but not all

individually suggests the former may be the case.

Urban and Rural Areas

In urban areas, the effect of productive assets and land was never significant
(Table 1.6). This is due to the small proportion of urban household owning such
assets and engaging in farming. Hence the results in the whole sample are mainly
driven by the rural sample. The rural sample has the same pattern in productive asset
effects as in the whole sample although with a larger magnitude (Table 1.7). However,

the differences between urban and rural samples are not significant statistically.

Across Age Groups

For the young and prime aged men the productive assets above the median has
a positive association with BMIs, and the amount of land farmed had different effects
(negative for age 20 to 39 and positive for age 40 to 59). For women, the younger

ones had a negative effect of higher productive assets and positive effects for lower

47

assets and the amount of land farmed (significant at 0.05 level), which is similar to
the whole sample. For the middle aged women, different from the whole sample,
higher productive assets in the above median assets range were correlated with a
significantly higher BMI and the effect of land was different as well.

For the elderly men and women having productive assets above the median was
associated with lower BMIs, though the effect was not significant. The effects of land
again display different impacts on men and women BMI, but was similar to the whole
sample.

All the above findings obtain with specifications including community or community-
year interaction dummies. So the effect of productive assets is over and above the
effect that was due to households living in better locations. The productive assets ef-
fects persist even after adding community level prices, water and sanitation variables.

There are several caveats in interpretation of these results. In a dynamic model
BMI can affect productivity and hence productive assets in later periods. There
could be selection into more strenuous vocations which pay higher wages. This would
bias the coefficients upwards. Lastly measurement errors in productive assets could
give rise to attenuation bias in the estimation. All these possible biases lead in
different directions: Without good identifying instruments it is difficult to make any
statement of the direction or magnitude of the estimates. However the association
between productive assets and BMIs is strong and consistent with findings in other

developing countries (Thomas et al. 1996).

1.5.3 Marital Status

As stated at the beginning of this section, adding marital status to the baseline
specification demands careful interpretation of results. The fact that for women across
all sample and sub-samples the coefficients of all other variables are not significantly

affected could be evidence that the effects of marriage or divorce on female BMI were

48

not by way of other individual or household characteristics. However the significant
marital effect for men did decrease the effects of education on male BMI.

Married men in all samples except for the elderly had a significantly higher BMI
than those who were single, ranging from 0.3 to 1 unit of BMI. The divorced or
separated men in all samples except for the prime aged had a lower BMI than the
singles, though the differences were not significant in any case. The widowed men
and women in the urban sample had a significantly lower BMI than the urban singles.
For women, the marital status had no significant effects on BMI, except that widows
in the urban areas had a much lower BMI than those who were single. In fact, the
single women in almost all samples had always had higher BMI.

Empirical research in the US has consistently shown that married men have sub-
stantially higher wages than otherwise similar unmarried men (?). One commonly
cited hypothesis is that marriage allows the husband to specialize in market produc-
tion and the wife specialize in home production, enabling married men to acquire
more specific human capital and earn higher wages. In rural areas men with higher
BMI may be considered having more ability to provide for the family and hence are
more likely to be married. However in China both men and women are engaged
in labor market activities, with employment rates being close to each other (Table
1.4). Hence if there is assortative mating it might exist for other traits of the partner
sought. For young urban women being slim could be more attractive in both mar-
riage market and labor market; for men being muscular impresses women more. The
endogenous biases can go both ways.

Potentially the selection biases can be corrected if we have proper instruments
for marital status but such case is rare. Since there was little variation in marital
status we cannot examine the individual fixed-effect model for assortative marriage

either.

49

1.5.4 Community Characteristics

In the first two CHN S surveys, state and free market data were collected. But by
1997, none of the communities had separate state prices so only free market stores were
visited. In all cases prices were collected for a representative basket of commodities.
We use free market prices of rice, wheat, eggs, pork, beef and fish (all for the most
commonly consumed types) to represent price levels in each community. Due to high
correlations between rice and wheat prices, and pork and beef prices only one price
from each food group was used in the model.

In 1989, 1991 and 1993, separate visits were made to obtain in—depth data in
each community for every identified health service and family planning provider or
facility. Information was collected concerning personnel, sources of funds, services
available, prices (asked separately for insured and self-pay patients), and distance
to the primary sampling units served by the facility. These were discontinued in
1997. Hence our analysis with community characteristics in addition to community
dummies only include three rounds of the survey.

The types and sources of households water and toilet facility are aggregated from
40 to 80 households in each community. They represent the percent of households
within a community with water from underground, etc. They are measures of quality
of water and sanitation in the community. Using the aggregated m'easures helps avoid
the selection biases due to migration.

In columns 1 and 3 of Table 1.12, the community dummies are not included,
but they are in columns 2 and 4 to control for unobserved community variables
that might explain where programs are located (Pitt et al. 1993). As we can see
most of the significant effects of community prices and water and sanitation qualities
disappear after adding community dummies. This might be evidence that community
unobservables should be controlled for and/ or that the variation of prices over time

is not strong before 1993. Although grain ration prices were raised twice in 1991 and

50

1992, each time by an average of 50 percent, the free market prices were not affected
as much (Ke 1999). Therefore, the food price effects on BMI can go in different
directions which is why empirical studies are important.

Before controlling for community dummies the effect of urban residence on BMI
was 0.3 for men and 0.5 for women, consistent with the findings that urban residents
have higher BMI. However, after controlling for community. unobserved heterogeneity
by the community dummies, the effect of urban residence increased to 2.4 unit of BMI
difference in men and 1.8 unit difference in women. For a 6-foot tall man the difference
in weight between urban and rural areas, other things being equal, is 1.8 pounds. For
a typical woman the difference in weight is 11 pounds. An urban area tends to have
better quality. health care infrastructure, water and sanitation conditions, as well
as other unobserved characteristics. Therefore it is again clearly important to have
community dummies.

The effect of the prices of rice and the price of fish were positive and significant
for men and women before adding community dummies. The effects of the prices of
eggs and pork for men and women were all negative and significant. Eggs and pork
constitute of high calorie intakes, hence the increase in prices may have a substitution
effect for other foods with lower calorie and result in lower BMI. Fish and rice, on
the other hand, can be thought of as substitute for red meat and wheat flour. The
increase in fish and rice prices will hence result in more consumption of less healthy
foods and increased BMI.

The percentage of households in each community obtaining drinking water from
more than five meters deep underground source was used as reference to three other
categories of water sources, namely, from open well, from spring, river, lake, rain or
snow, and from water factories. There were sign changes before and after controlling
for community dummies, which indicates the estimates without could be biased. At

this point we don’t want to draw any conclusions of the sign of the water source

51

effects. The joint effects of water sources are very significant.

The percentage of households in each community with in-house with-flush toilets
was used as reference to all other types of toilet facilities (Table 1.12). The percent of
households with in-house no-flush toilets decreased over time (Table A4) and since
it is negatively correlated with BMI it indicates a increased BMI in such neighbor-
hoods overtime. In a given year the higher the prevalence of household with such a
facility the lower the BMI is in this community. For the other toilet facility types the
negative effect is consistent with our expectation but the effects are not significant
after controlling for community dummies. What’s still significant after controlling for
community dummies are percent of households with in-house toilet but no flush.

The percentage of households in a community with no excreta around dwelling
area was used as reference to all other level of excreta. When the percent of households
with very little or some excreta around dwelling area in a community increases the
level of BMI decreases. The effects were strong for both men and women before

adding community dummies.

52

8:538 £an

 

 

 

.888 .888 :88 88.8
8:8- 888. 888- 88.8- 8388»
88.8 :88 88.8 888
8:8- :88- 888. 888- 888 3.5
88.8 :88 182.8 2.828
83- 888- 888 3.8 352
88.8 88.8 88.8 88.8 88.8 :88 88.8 :88
88.8 888 888 888 88.8- 888- 888- 888- 88.8 83
88.8 :88 88.8 88.8 ..888 ..888 ..888 ..888
888. 888- 88- 888- 888 M88 888 88.8 a as... 88 88 me:
.888 .888 .888 .888 88.8 88.8 88.8 88.8
888 888 888 888 ...88 8.8 288 ...88 a: some. 88 as 8.:
..888 ..888 ..888 ..888 A888 :28 :88 :28
888- 88- 888- 888- 888 .88 83 88.8 68:838.:-
.88.8 1:88 .888 L888 838 :28 88.8 82.8
88- 888- 8.8- 888. 88 8:8 «88 8:8 seas .85
:38 :28 :38 88.8 83.8 82.8 83.8 82.8
888 88- 88 888- 888- 888- 888 888 828 6828
..828 3:38 ..828 $8.28 :38 :28 :38 82.8
8.88 8.88 88 84.8 888- 888- 888 $88 .585
82.8 .888 .:28 .888 82.8 88.8 82.8 82.8
888 888 888 888 288- 888 88.8 888 . 5&8 eaom
3 :8 E :8 :8 :8 E :8

30 688mm mlgnlo 682 838.8,

 

 

 

88.0 888: mzmo s Em .83. so 8588.85 8.: 63.8

53

.5888 2:

0: 6:8 .8 5.5... 888% 568. one a m 6:.» 58685 85 5.5... 8_ 568. 8:... 8 H :88 605 62 woq .33
8 68666 8:73 £8 9.35605 8338: .5 5685 85 65:0: 88 856m 388 8385605 8: 23. .p

508858 655“ 0: 5:3 068: 8 9:08 8:238 93. .8

:92 mod .6

.1. 6:8 Hd as 58588 868388.38 3 6205800 85 88865 * .8585: 5 08 8088 656588 856.:
8820 .86— :080& .86556 80:00-68%8>d 65 86556 oma 8: 5285mm: :8 5 66305 0m? 8802

 

ddddd

Redd
«3de
mgdd
32d
ddddd

ddddd
ddmdd
hwfid
mmvdd
dddmd
ddddd

ddddd

3.3 d
mmudd
«dead
ddddd

86556 58838655500

ddddd 86556 36:55:00

868 6:82
53d 86556 80500
vad 86556 8.4.
3.2.0 308 98060.5
ddddd 505858

8503 6:8 :85 58:53 35305800 m0 36:8 $58: :8 58¢

 

 

 

 

ddddd ddddd ddddd dddod 86556 82.865550
ddddd ddddd ddddd ddddd 86556 36:55:00
amhmd 35d added 88d 8:88 3:82
mgdd added aged 88.0 mummd dddod Evmd ddddd 86556 80:00
38.0 ddddd good ddddd momvd ddddd 33d ddddd «85556 09..
mowed moved Rwdd omgd 38d added :ddd mdddd $88 955605
ddddd ddddd ddddd ddddd movsd mmvwd mmfid mmmwd 588:3
08s 3 6:8 85205800 588... :0“ 03.87%
ddddd named ddddd wowed ddddd 8de good ddddd 028$
$.va mos deE cod Edam wad» 3.56 3.3. :88 3.60
883388882 8 88-:- 38>»
«Ed wad mid Sud wdmd .8de comd dad 688:8-m
«new: «fix: 3de and»: dong 03.3 98an avg 8082800 mo .02
8> oz 8> oz 8> oz 8% oz 865:6 a8»....$6:5500
oz 8> oz 8% oz 8.» 02 88> 86556 36:55:00
:8 :8 :8 3 :8 88 88 a
WHO 865er mAO 862

 

 

:888 m: «3.8

54

8.55:8 3an

 

 

 

 

 

88.3 883 883 88.3 883 883 88.3 883 83-2
83: 3.8 8.32 3.3 3.32 «3 £32 2.8. 3% 3-30
563330880: go 808 32.3

8&3 32. 83 33 83 $3 32 33 853.2
33 33 33 33 58 $8. 38. $8. éosatefio 20 .oz
8? 02 R5 02 8> 02 new 02 82:83. Bazuwbasfiaoo
OZ 06> oz 8% OZ 8? OZ 06> waged—nu hfidfiaaoo
£328 128.8 328.8 1.28.8
83- 33.2- :3- $3- 33%;»
2.8.8 288 2.38 88.8
and- ES- 33- 3.03. 32mm v.32
23.8 23.8 LE8 ...238
3.3. 223- 2:3 and 35.2
28.8 23.8 28.8 38.8 83.8 28.8 838 2.3.8
833- $33- 83.3- 83.3- 33.3- wad- 33.3- 323.3. 883 33
$8.8 63.8 33.8 33.8 23.8 23.8 23.8 23.8
333- 333- was. 83.. 23. 333 Bod 33.3 am 8.0.8 Ba 3.: 3.2
23.8 23.8 23.8 $3.8 2.38 23.8 23.8 23.8
333- «83. 33. «33. 33.3 £23 «.83 was .2 983 Ba .82 ms
238 23.8 23.8 2.88 238 33.8 83.8 23.8
383- was- and- S23- 5.3- 33- ”33- 33. $330382.
23.8 23.8 2.58 28.8 83.8 33.8 83.8 233.8
22.- 33- 8%. Sad. ”33 223 and 3:. BBQ... in
23.8 83.8 88.8 2%...8 23.8 28.8 83.8 23.8
233 8.3.3 333 3.3 83- 32- 8:3- 333. 328. 0822
33.8 23.8 23.8 23.8 23.8 23.8 23.8 85.8
83.3- ~33- £3. £3 32- 83- 8.3- 32. 585
38.8 28.8 28.8 83.8 $3.8 88.8 23.8 2.38
33.3- 33.8 333- 3:3- «33. 33. 33.3- 283- 58E 88m
28 28 5 3 2.8 28 E 3

So 838% Mlqlm 832 833:5

 

 

 

822 555 “3332 mzmo a 22m :33. 3 3335862 3.2 232.

 

 

.§_Ba 2:

0... i=3 “0 :35 288% 25% one 3 m 98 23:58 on... :33... $2 25% 05 3 H 38mm 803 BE wed .wme
8 233%“. 83.? $88 038353 03888 Mo 23:88 23 ESE v.8 853m 3% 038353 Em.— vnrfi .3

496— 36 ea

.1. 98 2.0 33 3:89an b—dofimfifim E 308608 on» 833385. * .gficmaaa E v.8 £83 @3983 3350.—
283:0 18,2 288% 638296 fica8-§m.o>a 98 888833 mm.» 33 3838293 :3 5 809:2: o£< "302

 

 

 

 

 

 

333 333 333 3333 333 352
333 3:33 3323 3323 333 3323 333 £23 828833 328
£33 333 S33 3323 383 333 333 3323 828833 33
333 323 333 333 333 333 3:3 833 $3.3 93883
333 333 333 333 333 333 333 333 838:3
G080? EH5." vs 53; ~50ng U5 five EH5.— Ufld Swag £83...va mudvmoemwoo .«o ham—dag wfiauéu hem 05—97%
3333 £33 I 838% 8333338800
333 333 82883 3838800
333 333 @383 352
333 3:3 333 32.3 82883 328
333 333 3323 3333 828833 33
333 383 333 323 383 939%on
333 3333 333 333 838332
GOEOB U5 Eva Gaga: G858 $520808 mo b82358 gamma “Ow wad—gnaw
333 333 383 333 82883 331235888
333 333 333 833 83% 328800
333 333 333 883 333 2352
3333 333 333 333 333 883 333 333 85.8% 328
3333 383 333 333 833 333 2333 383 88.8% mm<
333 833 38.3 233 333 333 383 333 B33 238%on
333 233 333 333 333 333 333 333 88823
Ouvu O» 3370 mug—068000 madam”... Hem 2:qu

E 28 E 28 2.8 28 38 28

30 638% 30 632

 

8.388 3.2 333

56

82:38 0339

 

 

 

$33.8 333.8 $3.8 33.8
33- 333- 333 33 83023
833.8 33.8 33.8 93.8
333- 333- 333- :33- 833303 3855
83.8 338 1:28 1338
33- 333- 333 833 3252
333.8 333.8 333.8 833.8 88.8 33.8 88.8 33.8
833 3333 3333 3333 3333- 3333- 3333- 333- 885 353
33.8 :38 93.8 :38 I888 $38.8 1833.8 188.8
333- 3333- 333- 3333- $33 333 $33 333 3 33.8 323 as 33
333.8 1338 £338 1333.8 833.8 38.8 833.8 88.8
333 3333 £33 3333 3333 3333 :33 3333 L 383 383 Be 33
$233.8 $83.8 $83.8 133.8 33.8 83.8 83.8 83.8
:33- 3333. 333- 333- 33 333 3.33 333 $330289
:38 33.8 83.8 :38 $3.8 3.3.8 :38 A338
333- 333- 3333- 3333- 833- 333- 333- :33. gas. 335
33.8 83.8 33.8 $3.8 32.8 32.8 32.8 32.8
333- 32.8 333- 3:3- 383- 3333- 333- 833- 323 2322
183.8 1.83.8 1:38 3.83.8 82.8 $3.8 $3.8 83.8
33.3 3.3 33 3:3 323- 333- 333- 333- 55$
33.8 :28 2.2.8 82.8 $3.8 $3.8 $3.8 83.8
323 333 323 333 333- 3333- 3333- $3.8 ESE 883
3 A8 5 3 3 :8 5 3

3.5 8158 mime 832 833.3,

 

 

383 32:3 ”3333 3230 5 :3 3:3. 33 332333qu 5 333.

57

.EBS 2:

o: 53:00 :0 :05... 83.00% 0:53 05 2 a 5:3 :352: 2: :2: 000. 0:53 23 mm H 90000 :8: 30: mad .mon
o... 50:50: 30:13 3030 03:30:50.5 0338: do 5532: 2: 5:8: 0:0 «0:503 3008 03853:: ~00: 2a. .3

.526— mod :0

.1. 5:: ad 30 0:009:me 332323383 2 320508 0:: 8:022: * .885203: E 0:: 0.830 5.85:0? 33:0:

 

.8335: _0>0_ :Omu0m 3288:: tongsagéé 50:3 «28:5: 0w: 0:0 0:230:30: m0 :5 523505 81¢. 0qu

 

ooood

Road
3 Syd
Nomad
ooHNd
mvood

:ood
Sand
wfimwd
Sofid
mEmd
«mood

ooood

oomvd
mfimmd
nmfimd
vmood

«@8830 80h*hum:::::50

2333 8:89.33 383888

030.5 15.32
233 838833 :88
5.33 833833 03¢
333 33.8 9:268:
3333 8:833:

:0803 U5 :2: :003505 3:05:508 mo 5:53:00 wfiumfi :8 059d

 

 

 

 

 

 

 

ooood ooood ooood ooood «288:: :00»*m:m:::::00
ooood ooood ooood 85:85: b?::::oO
gawd Smod ooood mmmod ammod 3:333 5.3302
33d mmfiod ommmd vmfiod ommmd ooood Svmd ooood «288:: 3.8500
8de ooood ovmvd ooood omfimd ooood ovmfid ooood 0288:: 0w<
ammod moood mmmod moood «mood wdod wdod flood $0003 0330:5305
mmood «Hood flood ooood mzod movod «god good 2333035
Pan 0... 3:00 33:23:08 $3003 :8 059d
ooood good ooood 53d ooood ooood ooood ooood 05.93"
No.2: mod RAE m3: undo 8.3 3.3 Eda :33 3-50
5630050000805 Mo 3008 0:53
omfld Had omfid omfid mad mmad «mad vmfio 50:33:35
$8 $3 and $8 whom mnow whom whom 2230?:030 ...o .02
00> 02 30> 02 00> 02 00> oz «288:: :00»...m£:::::00
02 00> 02 00> 02 00> 02 00> 858:: 35:22:00
8 E E 3 3 E E E
mQO ”052:0? mAO "2.02

8388 S 333:.

58

30:8800 0308

 

 

 

 

38.8 :38 388 3.3.8
33 38.3 35.3- 22.. 32583
53.8 833.8 2.88 333.8
83- 333.3- and- 83- 333593 895::
3.3.8 328 162.8 3.828
:2- Ed- and 38.3 352
:88 ...288 588 1:88 88.8 0.888 88.8 L838
28. 38.3 23.3 ~83 «8.3. 28.- 88.3- 38.8 888 23
:38 88.8 :38 98.8 .3888 1.388 £638 £388
38.. 83.3. 38.0- :83- 83.3 38.3 Sod 38. .m :83 Ba 32 33
:38 88.8 :38 88.8 38.8 38.8 38.8 38.8
383 38.3 853 38.3 v8.3- 88. Sod- 38.3 L 38.8 Ba :2 33
1.888 £888 £888 $828 88.8 $3.8 88.8 388
:3- $3- 23- 23- Rod- «8.... 38.3 38.3- $328389
132.8 1933.8 1.3.3.8 1838 68.8 $2.8 333.8 83.8
33- Ed- 23- 33. 38.3 3.8.3 «3.3 32.3 328 .3:
388 0.888 888 0.88.8 $8.8 $5.8 65.8 83.8
22.. 23.. 83.3- 388- 32.3- 32.3- 83- 83- _8€m 28:2
$88 88.8 688 $8.8 82.8 333.8 3388 828
SS 33 32.3 23.3 32.3- 3.2.3- 33. $8.. 58:3
328 E28 82.8 32.8 $33.8 833.8 82.8 833.8
38.3- «8.3- 38.3- 83.3- 83- ~83. 38.3- 88.3 .583 @853
3 5 5 3 3 5 E 3

3:0 8380...: mulmo 832 83385

 

an - on 03. $332 3780 5 :23 2:3 :0 353880: "3.: 038.

59

.388 2:

o... 15:0 Ho 5:: 882m 25% 2: fl m 6:: 5:62: 23 :2: mm: 25% 05 E H 898 68: 32 meg .wme
8 68.66 $3.9» 88.8 95:68: 63:8: mo 562: 23 655: :8 8:29. $va 25262: 38 09H 8

49,2 mod a:

“I" 6:: fio a: .68me6 bfiosflfism fl £86580 2: 89.868 * .3558: E 08 88.5 6.86:8? 8:58
.8830 56— :ofiom 828::6 fiononzdozéc 6:: 868:6 mm: 3: 32mmtwm: z: E 6622: 82 8qu

 

886

9.3.0
5366
565.0
wooed
good

good
man“ .o
N2: .0
«3&6
wwmod
good

ooood

5‘de
mmwmd
58.0
38.0

84883: Bo»t$6::~:50

wooed 865:6 3:58:50

95:8 3:32
3.3.0 868:6 98:00
386 863:6 ow<
38.0 30me o>$o:60um
Seed :2826m

:0803 6:: :2: :838: 8:38:80 mo 32.63 wfipmma :8 0:15d

 

 

 

 

 

883 888 888 o8o.o 888% 5358888
888 o8o.o 888 888 8:58 88:88
5:8 8:8 588 88.0 8:: 352
888 888 8:3 «.83. «$3 888 886 885 8:888 ”:88
288 888 8:3 885 £88 886 8de 888 8:888 ow<
«$3 838 :33 888 £88 :85 $88 $88 383 9:26on
:88 888 888 «83 885 3:3 888 ”23 8:88:
Son 0... 3:3 358508 358... .8“ 315$
888 888 888 883 888 o8o.o 888 88.: 815-:
8.8a 8.8 5.58 8.2 3.8:. 8.”: 8.3V 8.3: is 3.30
365880888 mo “8:. 323
ES 83 ES 88 $3 83 83 Ed 838?:
SE .8» SE 585 :8 28 :8 28 88583 mo 52
8> oz 8? OZ wo> OZ 06> OZ nomad—v “dehsuhamcflaaoo
OZ 8? OZ 8? OZ 8? OZ 8.? womaaflv hug—£58800
«V E 5 3 S 5 E 3
30 "was? mac 63):

 

 

8.83 w: 03:.

60

8:868 03:8

 

 

 

$33 88.3 88.3 33.3
83o- wad- mxd 38 83°23
:23 38.3 $83 33.3
.88- 83- 88 $3 8:88 8983:
33.3 :33 £653 $38.3
3.3- 2:? ”8.: $2. 352
0.83 $8.3 $8.3 38.3 $8.3 $8.3 38.3 38.3
Sod- as? wood- wood- Sod :3 SS. :8. Baa. 83
33.3 338.3 38.3 3:83 1:83 1:83 $28.3 1:33
:8 Sod :8. ES 88. 85 N83 88 .N 6%“ Ba :2 we:
$3.3 $8.3 33.3 88.3 38.3 88.3 38.3 88.3
a8 5:. ~85 88 88. 83.- 88- :88 L 988 8a :2 we:
93.3 88.3 68.3 88.3 83.3 38.3 28.3 85.3
«8... 83- 35.3 $3- 83 $3- $3 83. $8888:-
383 $83 28.3 38.3 :83 883 38.3 88.3
and :3. $3 3.2. Sod- 33- £3. £2.- 39? ,3:
3R3 $8.3 $3.3 33.3 88.3 82.3 383 82.3
a; :3 83 88 88- 33- ”8.? 83¢. so? 28:2
1833 132.3 338.3 1:83 333 $2.3 $8.3 3:3
Es 82 £3. 83 3:3 and- SS- 38- 55$
98.3 232.3 883 L833 $3.3 $2.3 $8.3 32.3
and 83 and :3 88. 83. 83. 88- .583 88m
a: :3 E a 3 :3 3 3

mg: 638»: E 832 83:3,

 

 

 

mm - ow ww< ”3de mzmo 8 ~35 6643‘ Mo 8:88:26qu 54 2an

61

805008 05

00 7:80 00 880 000008 0880 08. E m 88 80608 0:... :05 000— 0880 05 8 H 00000 008 800 mod .83
00 008500 0038 00000.0 03000808 028000 00 8.0808 080 882 00.0 8830 000000 02003008 800 0:8 .0

._0>0_ 86 00

0... 88 H6 00 8005890 38050330 8 8065000 05 0000068 * 000038088 8 0.8 00080 8.88800 000300
000030 _0>0_ 0800A 088830 000800-80m0>a 88 008830 08 08 002000800 :0 8 800308 002 ”0002

886

886
886
886
686
mmmmd

86
886
386
$86
H806
>866

86

2.86
886
$.86
wommd

098806 8080580800

886 880830 5808800

03000 880.02
386 008820 80:00
:88 888% 00<
836 000000 03000005
$.36 0030053

00803 88 808 0002500 0.8205000 m0 3:880 @8000... 000 038d

 

 

 

 

 

866 886 886 866 00888:“. 000830808800
886 8666 886 866 008880 5808800
886 886 886 886 0330 8882
886 336 886 3.26 886 886 E86 886 008820 80:00
586 886 386 6866 886 886 886 886 008886 0w<
386 286 886 :86 886 886 586 886 30000 0>Sosvoum
886 886 886 3.86 886 886 886 886 80500an
0000 3 1880 08068000 $500... 80 028$
866 6866 886 6866 886 886 866 886 039$
madam 86m 8.8m 3.6m 8.8a 3.3 «dwfi 86H 030 80-30
3620000000805 .«0 0008 085$
«86 686 N86 836 «mm-o 8N6 H86 686 “0088008
88 86¢ 83 68v wwmv wwmv wwmv wmmv 0005080000 “‘0 .02
00> 02 00> 02 00> 02 00> 02 008830 80803888800
02 00> 02 00> 02 00> 02 00> 38880 3888800
8 E E 8 88 n3 8 3
mAO “0880M mQO “082

 

 

8088 3 2.8.

62

8:538 $388

 

 

 

$3.8 38.8 388.8 388
8:8- 288- 8888- 888- 3623
£88.: 88.8 38.: L888
883- 888- 88.7 8.5- 8883888 88838.
$3.8 $88 88.8 88.8
83.8. 8888- 888- E8- 8252
38.8 :88 88.8 :88 :88 38.8 58.8 38.8
888 888 888 888 888- 888- 888- 888- 883 855
:88 88.8 88.8 :88 888 E88 888 98.8
«88. :88- 888- :88- 888- 888- 888- 888- ‘8 888 828 as 83
88.8 338.8 88.8 138.8 88.8 88.8 38.8 5.88
828 88 828 828 888 888 888 888 L 888 Ba 32 83
88.8 $8.8 $8.8 88.8 888.8 $8.8 888.8 888.8
888 888 888 838 $88- $88- 838- 888- 88260388.
:88 88.8 38.: $8.8 $8.8 888 E88 338
888 888 888 888 888 888 888 888 .888 885
88.8 88.8 88.8 38.8 :38 88.8 :88 A888
888- 888- :88- 888- 888- 888- 888- 888- 32% 2822
83.8 88.8 83.8 :88 88.8 88.8 888.8 2.88
88.8- 888 888- 888 8:8- 888. 888- 3.88- .5828
88.8 88.8 88.8 88.8 88.8 88.8 288.8 88.8
888- 888- 888- 8:8- 888- 888- 888- 888- .583 oaom
3 A8 A8 5 3 A8 A8 3

30 8388-.— m-ao 6.82 833:5

 

 

+8 mw< ”3&me mZEO E dam £358 mo mandadbuopmm ”SA @388

63

.588 28

3 8:8 8 85 8888 25% 05 8 N 83 88:88 on» :83 8_ 25% on» 3 H 88 808 82 m3 .wme
3 88888 8:15 888 8382808 @3888 .«o 8888 2: 888 88 8838 38 8552608 88 23- .8

432 8d 8

1. 88 ad 88 8888me 8283888 8 820588 85 88.868 * .885888 8 08 828 @8888 8952
8820 18>».— Exflom .888828 fiasco-888.96 88 88838 cm.» 08 82888 :8 8 88:88 8?. 8qu

 

88d

Shad
wmmmd
mam-d
womwd
Ngwd

89d
88d
shad
mwaod
wmmwd
Mvmomd

8dd

88d
88d
ova-pd
hocwd

8888.6 898388.888

88d 88:86 8838800

25.38 18:82
wummd 88828 98:00
ode 88:86 ow<
vad 88 0388605
88d nos-8:3—

8803 v8 88 88382 $822,588 do 858:8 @588 god 8:87am

 

 

 

 

 

88888 88888 88888 8888 88888 388888868
88888 88888 88888 88888 88888 8888800
Sand vowmd 88d mvwmd 8388 1882
vammd Sand 38d dvmmd mwddd 88d 88d good 8888 fiasco
38d Named 38d 38d wind dmvmd 83d 3.de 888:8 88.
Emmd Emdd Shad 88d 88d 83d mowed dug-d 888 838805
wEwd Smwd 88d mmmwd wmmwd 33d wnvwd dmwod 80:88:3—
o8u o... 158 886588 8588 “8 2:87am
888 88.8 88d 888 88.8 88.8 coded 88.8 888d
883 2.8. 888 8888 8.38 8888 88888 8.8. 83,-. 8.20
883888882 8o 88m. 38$
$8.8 End $8.8 Sad fivd 888.8 83.8 and 88:88
82: 82: 882 8.2 $8 38 88 88 8282880 88 .oz
8% oz 8». oz 8> oz 8> 02 «8838 8.838858800
OZ 8> oz 8.». oz 8.? oz 8% 88886 588800
E E 5 3 3 E E 3
mac 8388mm- mqo "£82

 

 

8.258 83 838

64

8:888 @358

 

 

.338 888 :88 $8.8 $88 888

8.8 £8- 83. $8 $8- $8- BEES
$3.8 38.8 :38 83.8 :58 :88

8.8- «8.8 83- $8- 83- 8.8 833.8 8285
8.8.8 :88 :58 .888 .588 .328

38.? Rod. 8.88 8.8 28 and 352
88.8 38.8 :88 ......:8.8 38.8 88.8

mood- 88 wood 8.? 885 wood- 883 83
83.8 38.8 :88 :88 88.8 83.8

28 :88 «2.8 8.0 8.8 Sod ..m “33 Ba 28 m3
888 :38 :88 $8.8 88.8 :38

Sod :36 «a; 88 23. 88. a: 088 Ea 38 we:
L888 :38 .888 $38 38.8 38.8

v8.8 38. 8.3- 88 8.3- Sod 88808.8
888 88.8 33.8 83.8 35.8 88.8

$8- 8.3- 83. $8 88 wood .848 88
33.8 :88 :58 :88 :88 :88

:3- mag 23. 33 8.8 ”So 32:8 228
$88 1:88 3.888 38.8 $3.8 88.8

8:. and a? £2 83- $3.- 888
:88 183.8 :88 88.8 888 68.8

28 83 8d 8.8 83 83. . ESE 88m
3.: 8.: 38 53 88 3.: 833:5

80 6388 E 632

Sago 8 88 .8 .83 880m mZEO .3 H35 3334: ”:4 £an

 

 

 

65

Sfia 2:

3 :33 8 :23 ~3mb 853m 23 mm m was 5:88 85 ~85 mam: 25% 85 fl fl 8me “88 32 mod .waH
3 83:88 82g 833 038388 83289 We :3va 23 988 98 858m 3me 8382808 1*: 2.5.. .9
483838 38.8.“ on 8:? 8889 £ 98% 8:282 0:8 .w

40>»: mod 8 I «:8 ad a.» 889:me bfiosmsgm mm 828508 on» @3865 * .88fi88a

5 03 £88 88983 #459 3888888: $28838 ems o3 £868qu =w E coda—2: 82 6qu

 

good
mgmd
wdHcd
mwdwd
unwed

meddd
wvnmd
gmdd
dovvd
good

gadd
dwwmd
dwvmd
:3 d
good

888:6 3:88:80
253m 13934
83:96 mw<

3me m>$o=vo.~m
cassava

880.3 «Ex 88 883qu 3220508 We 3:83 wcswmu 8w mggd

 

 

 

ooood ddddd ddddd ddddd ooood ddddd 888:6 3188800
mmmdd nownd uwuwd ended Nomad mmuod magnum 18in
dddod ddddd dddod ddddd wooed ddddd 888:6 ww<
23d mmomd omnmd dmmdd :ndd ogmd 333 83828.5
33d waded vdddd 35d Shad Shad ~858va
Son 8 3:3 mucmfimuon madame» 8m 253d
33 33 3:. two ES ammo 83:3.m
mmwm 3mm 3w” wacm damn nmnm 8056,8pr .8 .02
«A? 8% mo? 8% 8% 8> 88888 3289880
Rama «2: H23 2:: n2: Haaa
mAO "mu—damn mAO 882

 

 

8.8on :4 mass

66

8:588 @388

 

 

 

88.8 88.8 88.8 88.8

«88. 88.8- 88.8- 88.8- 3,833

88.8 88.8 28.8 88.8

58. «8.8- 888- 888- 8338 8285

888.8 88.8 3.82.8 1.88.8

:88- :S- 83 88 8:38

888 88.8 88.8 88.8 88.8 88.8 88.8 88.8

88 88.8- 88 88.8- 888 .888- «88 88.8- 883 23

88.8 88.8 88.8 88.8 88.8 88.8 .888 88.8

888- 88.8- 88.8- 388- «88 88.8 88.8 888 .8 883 Ba 82 83

88.8 188.8 88.8 188.8 88.8 88.8 88.8 88.8

888 88.8 88.8 888 88.8 88.8 88.8 «88 L 83.3 Ea 32 83

$88.8 $88.8 2.88.8 1.88.8 :88 88.8 88.8 888

E8- 88.8- 38- 888- 888 88.8 :88 888 88280389

8.88 82.8 88.8 88.8 82.8 :28 82.8 88.8

88.8- 888- 88- 88.8- 88.8 88.8 83 82.8 32% Em

82.8 32.8 82.8 82.8 83.8 82.8 83.8 32.8

88.8 82.8 88.8 83.8 88.8- 888- 88.8- 8.8. 8888. 0822

$82.8 #828 1.83.8 1.828 93.8 :28 83.8 83.8

88.8 88.8 5.8 888 88.8- 88- 88.8- 8:8- 8&8

1:38 .888 383.8 ...828 83.8 82.8 83.8 33.8

88.8 83 88.8 88.8 88.8 88- 88.8 8.8.8- 388 28m
3 :8 5 3 3 :8 E 3 838:5

88 @888 mil-I8 2a:

 

 

855 ”8823338 888888 .83 8.82 888 8 88m 883. 8 3:28:88 83 @88-

67

825:8 0388

 

 

88.8 1:3.8 888.8 183.8 1888 1:38 1888 1:38

8:8 888- 3.8 888- 83 888- 888 888- 3. 85cm 583
88.8 188.8 88.8 188.8 188.8 1838 183.8 183.8

8888 $8.:- 888 88.:- 8883 8888- 88.: 8888- 8 85cm 138$
8888 18:8 8888 18:8 888 1:38 888 1:38

888 8888- 888 8888- 888 888- 888 888- “. m 85cm 38.3
88.88 8828 88888 8858 8888: 8888 8888: 8888:

88.3- 888- 888.3- 888- 8888- 88.: 888- 88.: 83 $1 8880
:888 88888 8888 8388 8888 888.38 8888 888.38

8888 888.3 88.8 888.3 88888 8888- 8888 88.8- was: 8. 256
888.5 8888 838: :88: 888.38 883.38 888.38 83.38

88.3 8888 m8: 888 83.8- 838:- 388- 88.3- 22888 .8. 3.80
888 188.8 88.8 188.8 88.8 1888 88.8 1888

8888- 8888 888- 888 8888 888 888 888 88 8o 85
188.8 188.8 .888 188.8 88.8 1888 888 188.8

888- 3.38- 888- 838- 888- 888- 888- 58. 88 mo BE
88.8 188.8 88.8 188.8 88.8 188.8 88.8 188.8

8888- 888- 8888- 888- 888- 8888- 8888- 888- mama mo 8:8
83.8 183.8 838 183.8 838 1:38 :38 18:8

888- 8888 8888- 888 8888 888 888 :88 . 8: 8o 85
1888 £3.88 1:888 1:338 1:888 1838 1888 1:38

85 8.88 88.3 88.8 888 888 8888 888 88281 :35

3 :8 5 :8 :8 :8 E :8
30 view? mac 232

 

 

8.888 3.3 888.8

68

8:25:00 233-H.

 

 

 

 

 

283. R88 288 888 885 88... 888 88.3 022.2
m: :8 2.8 8.8 8.8 8.8 8.8 :8. 83 3.30
365880880: mo each. 035$
88 388 328 838 828 E; 328 3.8. Bass-m
88 88 88 88 88 88 88 88 30 3o H2.5.2
8> Oz 8% OZ mm> OZ 8% OZ momaEdpfl-m 3558890
388 38.8 38.8 388 3:8 :38 38.8 38.8
83 83 33.0 83 83. 838 388- 858 3298 :82
38.8 $388 38.8 $388 32.8 1328 3:8 £328
.38 23- 83 28.8 383 838- 83 88o- 3238 883
38.8 1.388 38.8 138.8 32.8 132.8 3:8 $32.8
«8.8 23.8 82- :38- 83- 28.8 88.8 38.8 3228 2:: .85
38.8 135.8 38.8 333.8 38.8 338.8 38.8 138.8
3.0- 83- 888- 83- 33.8 32.8- 238- 2.2.8- 383 oz
388 ...388 38.8 ...388 338 :388 33.8 1.388
38.8 38.? 83- 83.8 88.8 83- 28- 83- 328 s 3888 E :25
38.8 38.8 38.8 38.8 33.8 38.8 33.8 38.8
28.8 883 28.? Sod 33.? 83.8 82.- 88.3- 383 @835
38.8 33.8 38.8 38.8 $38.8 $38.8 338.8 138.8
83- 38.8 83. 83- 33.2- 288- 33.2- 388- 8:: o: 3:2 2
38 38 A8 3 3.8 38 88 3
mac mans-.m- 30 222

 

3.888 a: $32.

69

.0058 80.5 0:... 2 008280 00300.8 00.83 38 000 .305 00 £00 .042 00>?

@533 38 £03 0080 A8 000080000: :8 ”8088 208-03 m530=0m 05 523 2020:0200: .80 8000008 028 .0

.8888 2:

8 8:00 80 0.25 030.0% 002% 0;... 2 m 05 0.2008 05 :23 08— 002% 05 2 H 9808 005 30“ mod .33
8 0082800 00:35 30000 02800005 03288 80 0.2008 05 0000.2 0.3 80:93 30000 02800005 38 04H. .0

._0>0_ mod 20 .1. 0:0 :52 ad ad 2805ch 53032830 0.8 85206000 0:» 8230205 * 000050080 5
03 3080 0800-08 82500 08020 322 0020A .00 38 000 A8 20008 8028800 5308800 “0.“ 205000 80:
00 38 0:3 CV 3002 .828800 80500-808853 00.0 028800 0mm 03 20008 05 E 000305 02¢ ”0802

 

 

 

 

28.8 28.8 888% 888888
388 88.0 88.8 88.8 8280
88.8 82.8 888 82.8 28. 8:2-
388 288 88.8 888 8.88 8883
888 88.8 88.8 88.8 82$
228 88.8 283 8882
88.8 28.8 888 888 888% 8280
88.8 88.8 38.8 88.8 8888 88.
88.8 88.8 88.8 :88 8884 8.2
888 88.8 888 888 88880
c0803 my: 808 #003903 3G0m0mw000 mo 520370 9:08» hem 05—57%
88.8 888 88.8 88.8 8888 8888800
E88 88.0 88.8 888 :88 833.8 88.8 3838 85883888 8880
8:8 888 888 88.8 88.8 288.8 888 888 8880
88.8 888 .388 888 88.8 880° 88.8 888.8 25 8:2.
82.8 88.8 88.8 888 888 888.8 888 888 888 883
338 88.8 83.8 88.8 888 888 :88 888 885
888 88.8 88.8 88.8 288 8882
:28 828 82.8 828 88.8 88.8 828 888 8888 8880
88.8 88.8 888. 88.8 88.8 888 88.8 8888 888% 8.4-
888 888 88.8 :88 888 88.8 :88 88.8 882 82
888 88.8 88.8 88.8 888 828 828 888 82888.
95$ 3 3:00 25065000 $3009 ham 053nm

38 38 88 38 38 38 38 38

30 38883 30 882

 

 

8.888 2.2 832-

70

1.6 Dynamic Conditional BMI Demand

The estimation of equation (1.9), the dynamic conditional BMI demand function
as outlined earlier, has both advantages and disadvantages. On the one hand we do
not need to include all past variables based on the assumption that lagged BMI is a
summary statistic for all past information. On the other hand, the estimates from
such relationship are only partial effects of the variables we are interested in.

We study the determinants of BMI in 1993 conditional on BMI in 1991 and the
determinants of BMI in 1997 conditional on BMI in 1993 controlling for age dummies,
and community dummies when community characteristics are not added explicitly.25
The lagged BMI is treated as endogenous and the model is estimated by OLS, 2SLS
and GMM methods. In 2SLS and GMM estimations the endogenous BMI is instru-
mented with previous year’s log of real productive assets splines, land cultivagad, as
well as interactions of real productive assets and land with community free market
prices of rice, eggs, pork and fish, and water, sanitation and clinic characteristics in
the previous year. The first stage regressions are included in appendix B. There is
strong evidence of heteroscedasticity in the dynamic models. Under such condition
the 2SLS is consistent and asymptotically correct inference can be made using the
Huber-White “sandwich” robust variance-covariance matrix. However, the GMM es-
timator is more efficient, although with many overidentifying restrictions the small
sample property of the GMM estimator can be poor (p.204 Wooldridge 2002). In par-
ticular, Wald tests tend to over-reject the null hypotheses (Baum et al. 2003). The
GMM and OLS estimators are included in the paper for comparisons. The specifica-
tion of interaction between lagged BMI with age groups is also tried and no particular

patterns were found.

 

25The estimation for BMI in 1997 conditional on BMI in 1991 was also carried out but not reported
here because in 1997 some community characters are not available.

71

1.6.1 With Community Dummies

In Table 1.13 the regressions are for the dependent variable adult BMIs in 1993
conditional on BMIs in 1991 and education levels and household resources as in the
baseline model. The IVs are the above stated variables in 1989. We conduct several
specification tests. The first stage P tests for the identifying instruments and the
overidentification tests26 suggested the IVs were valid instruments (see appendix B
for the first stage regressions). All first stage F tests are very significant, signaling
the correlation between the identifying IVs and the lagged BMI is not weak. The
overidentification X2 tests are heteroscedasticity—robust version based on Wooldridge
(2002 p. 123). Since each year the question on land cultivated is based on last year’s
value, here we chose to exclude it for the overidentification test. All tests can not
reject the null hypothesis that all other instruments are uncorrelated with the error
terms. The Hausman test between the OLS and the 2SLS estimates of the lagged
BMI for men suggested a significant difference at 10 percent level but cannot confirm
such difference for women.27 All three methods give positive and significant effects of
the lagged BMI on the current BMI. The estimated effect of the BMI in 1991 on the
BMI in 1993 in OLS is around 0.6 and in 2SLS and GMM about 0.8 for men, and
around 0.6 in OLS and around 0.7 in 2SLS and GMM for women.

Controlling for lagged BMI the effects of education and household resources in
1991 were no longer significant in both OLS and ZSLS estimates, except that in GMM
estimation the primary diploma in 1991 showed a strong negative effect for men. The
male marriage premium in BMI become insignificant.

The conditional age and community effects are significant for women and the age

 

26The choices of instrumental variables for the overidentification test is arbitrary. The land culti-
vated is assumed to be uncorrelated with the error terms and the overidentification test is based on
the null hypothesis that all other identifying IV s are uncorrelated with the error terms.

27The Hausman test between OLS and 2SLS estimate of the lagged BMI effect was based on the
assumption of homoscedasticity, which was clearly violated in the model. It is not surprising that
the test has such low power.

72

effect conditional on past BMI for men is not jointly significant.

In Table 1.14 the BMI in 1997 conditional on the BMI in 1993 is estimated.
The lagged BMI is instrumented by the same set of variables from 1991.28 The
first stage F tests for the identifying instruments showed signs of weak instruments.
The overidentification tests passed. The Hausman test between the OLS and ZSLS
estimates of the lagged BMI cannot reject the null hypothesis that they were the
same. Both 2SLS and GMM estimates were higher than the OLS estimate for men,
but it’s not the case for women. I

The partial effects of some primary education for men were negative and sig-
nificant in all estimates. The strong nonlinear effect of productive assets between
the higher and lower levels is much larger than those estimated in the reduced form
regressions. The effects of productive assets are of opposite signs than those in Table
1.10 for both men and women in the lower half of the assets spectrum. The marriage
premiums for men no longer exist.

The fact that we can reject BMI being exogenous in the first conditional regres-
sion (BMI 93 on BMI 91) but not the second (BMI 97 on BMI 93) suggests that
although BMI is a cumulative measure of health, values from a farther past does not
serve as a summary statistic as well as values from a more recent past. The condi—
tional dynamic model is hinged on the assumption that lagged BMI is a sufficient

statistic.

1.6.2 With Community Characteristics

In Table 1.15 and 1.16, instead of controlling for community level information
with dummy variables we use changes in community characteristics between current

and lagged period to examine the effects explicitly. Similar to the dynamic models

 

281 have also tried to instrument the BMI in 1993 with log of the real productive assets and land
in 1989 and interactions of these variables with community free market prices of rice, wheat, eggs,
pork, beef and fish, as well as water, sanitation and clinic characteristics in 1989, however, there is
strong evidence of the weak IV problem.

73

with community dummies, the first stage F -tests and the overidentification tests all
validate the instruments. The Hausman test for BMI 91 in the conditional regression
for BMI 93 rejects the null hypothesis that BMI 91 is exogenous; whereas in the
conditional model of BMI 97 on BMI 93 we can not reject the OLS and 2SLS are the
same.

Oddly, the effects of changes in prices of rice, eggs and pork in two conditional
models for men and women are of exact opposite signs. In our sample all these price
changes over time are positive. With an increasing in the price of rice, for example,
we would expect to see BMI to increase if substitution effect is stronger than income
effect. However, in our sample the increase in wheat price is actually bigger than the
increase in rice price from 1991 to 93. Therefore there is no strong substitution effect
and the coefficient for rice price changes are negative.

Also significant are some water and toilet type characters in 2SLS results and
again of opposite signs in some cases. For women only the change in in—house no—
fiush toilet percent and no toilet percent changes is negatively associated with BMI.
Overall the age dummies and water source dummies are significant for men in table
15. Although the conditional BMI specification works better with shorter lagged
period, the changes in community level information during such short period may

not be large enough to correctly identify any effect.

74

Table 1.13: Determinants of Adults BMI in 1993 Conditional on BMI in 1991: Overall

 

 

 

 

 

Male
Variables OLS ZSLS GMM OLS 2SLS GMM
BMI 91 0.573 0.843 0.803 0.571 0.837 0.802
(0.061)" (0.191)“ (0.088)“ (0.061)“ (0.192)” (0.087)**
Some primary 91 -0.100 -0.091 -0.355 -0.134 -0.117 -0.404
(0.218) (0.226) (0.227) (0.216) (0.225) (0.228)*
Primary 91 -0.225 -0.204 -0.495 -0.255 -0.228 -0.534
(0.226) (0.229) (0.220)" (0.225) (0.229) (0.221)M
Middle school 91 -0.252 -0.267 -0.384 -O.293 -0.299 -0.443
(0.215) (0.221) (0.239) (0.216) (0.222) (0.242)*
High school 91 -0.083 -0.161 -0.256 -0.121 -0.189 -0.315
(0.253) (0.267) (0.281) (0.251) (0.263) (0.283)
Tech/College+ 91 0.060 0.065 -0.343 0.025 0.037 ~0.419
(0.294) (0.297) (0.466) (0.292) (0.295) (0.474)
Log R prod assets 91 1 0.006 0.015 0.046 0.002 0.012 0.041
(0.037) (0.037) (0.029) (0.037) (0.037) (0.029)
Log R prod assets 91 2 0.044 0.026 -0.021 0.045 0.027 -0.020
(0.042) (0.041) (0.032) (0.042) (0.041) (0.031)
Land 91 0.005 -0.001 -0.016 0.003 -0.003 -0.018
(0.025) (0.026) (0.019) (0.025) (0.026) (0.019)
Married 91 0.376 0.294 0.317
(0.242) (0.266) (0.199)
Divorced Separated 91 0.108 0.164 -0.065
(0.626) (0.618) (0.583)
Widowed 91 -0.203 -0.191 -0.270
(0.408) (0.423) (0.372)
Community Dummies Yes Yes Yes Yes Yes Yes
No. of Observations 2477 2477 2477 2477 2477 2477
R-squared 0.370 0.330 0.237 0.370 0.332 0.226
First stage for BMI 91
R—squared 0.258 0.259
F-statistic for Identifying IVs 2.82 2.70
P~value 0.0000 0.0000
Test of homoscedasticity
Chi-sq stat 118.22 89.44 116.86 88.17
P-value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS: BMI 91
Chi-square statistics 3.18 3.07
P-value 0.0747 0.0798
Overidentification test
Chi-square statistics dof Chi(56) Chi(56)
Chi-square statistics 56.52 56.96
P-value 0.4555 0.4390
P-value for testing coefficients equal to zero
Education 0.7620 0.7608 0.2342 0.7064 0.7172 0.1882
Assets 0.3171 0.4992 0.2415 0.3565 0.5390 0.3186
Age dummies 0.4321 0.4164 0.3632 0.5246 0.5394 0.4484
Marital Status 0.1890 0.4746 0.1266
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

75

Table continues

Table 1.13 (cont’d)

 

 

 

 

 

Female
Variables OLS ZSLS GMM OLS 2SLS GMM
BMI 91 0.552 0.644 0.657 0.552 0.641 0.650
(0.041)** (0.106)” (0.066)" (0.041)“ (0.106)” (0.066)**
Some primary 91 0.189 0.163 0.214 0.188 0.163 0.214
(0.181) (0.185) (0.172) (0.181) (0.185) (0.172)
Primary 91 0.276 0.222 0.234 0.271 0.218 0.232
(0.199) (0.210) (0.187) (0.199) (0.210) (0.186)
Middle school 91 0.051 0.044 0.064 0.047 0.040 0.057
(0.205) (0.205) (0.181) (0.205) (0.205) (0.181)
High school 91 -0.306 -0.306 -0.158 -0.305 -0.306 -0.171
(0.256) (0.260) (0.401) (0.256) (0.259) (0.402)
Tech/College+ 91 0.343 0.432 0.510 0.358 0.443 0.487
(0.462) (0.477) (0.850) (0.469) (0.483) (0.854)
Log R prod assets 91 1 0.031 0.026 0.030 0.031 0.026 0.030
(0.039) (0.040) (0.032) (0.039) (0.040) (0.032)
Log R prod assets 91 2 -0.019 -0.014 —0.014 -0.018 -0.013 -0.013
(0.037) (0.037) (0.031) (0.037) (0.037) (0.031)
Land 91 0.014 0.015 -0.020 0.013 0.014 -0.021
(0.030) (0.030) (0.025) (0.030) (0.030) (0.025)
Married 91 0.153 0.139 0.152
(0.410) (0.422) (0.359)
Divorced Separated 91 0.685 0.644 0.681
(0.638) (0.627) (0.565)
Widowed 91 —0.146 -0.162 -0.125
(0.482) (0.489) (0.415)
Community Dummies Yes Yes Yes Yes Yes Yes
No. of Observations 2701 2701 2701 2701 2701 2701
R-squared 0.442 0.435 0.403 0.442 0.436 0.409
First stage for BMI 91
R-squared 0.238 0.238
F-statistic for Identifying IVs 3.25 3.35
P-value 0.0000 0.0000
Test of homoscedasticity
Chi-sq stat 113.30 110.84 113.71 110.62
P-value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS: BMI 91
Chi-square statistics 0.69 0.65
P-value 0.4052 0.4206
Overidentification test
Chi-square statistics dof Chi(56) Chi(56)
Chi-square statistics 39.51 41.54
P-value 0.9534 0.9252
P-value for testing coefficients equal to zero
Education 0.2554 0.3416 0.2949 0.2555 0.3374 0.2818
Assets 0.7173 0.7921 0.5794 0.7307 0.8008 0.5922
Age dummies 0.0003 0.0013 0.0001 0.0003 0.0011 0.0001
Marital Status 0.4684 0.4672 0.3977
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

Note: Also included in all regressions are age dummies. In 2SLS BMI 91 is instrumented with
real productive assets in 89, land cultivated and interactions of real productive assets and land
with community free market prices of rice, eggs, pork and fish, as well as water, sanitation and
clinic charateristics in 89. Huber/White robust standard errors are in parentheses. * indicates

the coefficient is statistically significant at 0.1 level; ** at 0.05.
76

Table 1.14: Determinants of Adults BMI in 1997 Conditional on BMI in 1993: Overall

 

 

 

 

 

Male
Variables OLS ZSLS GMM OLS ZSLS GMM
BMI 93 0.391 0.436 0.640 0.389 0.436 0.638
(0.066)” (0.144)" (0.091)“ (0.067)M (0.143)“ (0.091)"
Some primary 93 -0.476 -0.488 -0.594 -0.494 -0.507 -0.597
(0.262)"‘ (0.261)* (0.230)“ (0.261)* (0.259)* (0.228)M
Primary 93 0.128 0.134 -0.034 0.103 0.111 -0.037
(0.326) (0.327) (0.259) (0.326) (0.328) (0.259)
Middle school 93 0.290 0.295 0.104 0.253 0.259 0.089
(0.356) (0.357) (0.293) (0.354) (0.357) (0.291)
High school 93 0.208 0.200 -0.037 0.182 0.175 -0.052
(0.365) (0.364) (0.298) (0.364) (0.363) (0.297)
Tech/College+ 93 0.514 0.504 0.651 0.476 0.468 0.688
(0.464) (0.456) (0.783) (0.464) (0.454) (0.787)
Log R prod assets 93 1 -0.077 -0.075 -0.056 -0.079 -0.077 —0.058
(0.043)* (0.043)* (0.034) (0.043)* (0.043)* (0.034)*
Log R prod assets 93 2 0.092 0.089 0.075 0.093 0.090 0.075
(0.040)M (0.039)“ (0.032)” (0.040)" (0.040)” (0.031)"
Land 93 0.005 0.004 -0.006 0.004 0.004 -0.006
(0.011) (0.011) (0.010) (0.011) (0.011) (0.010)
Married 93 0.302 0.297 0.155
(0.351) (0.351) (0.304)
Divorced Separated 93 -0.288 -0.198 0.382
(0.854) (0.857) (0.800)
Widowed 93 -0.329 -0.296 0.076
(0.500) (0.498) (0.455)
Community Dummies Yes Yes Yes Yes Yes Yes
No. of Observations 1815 1815 1815 1815 1815 1815
R-squared 0.385 0.383 0.306 0.386 0.384 0.302
First stage for BMI 93
R—squared 0.259 0.261
F-statistic for Identifying IVs 1.28 1.14
P-value 0.0807 0.2296
Test of homoscedasticity
Chi-sq stat 70.55 40.69 70.89 41.68
P—value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS: BMI 93
Chi-square statistics 0.24 0.26
P-value 0.6242 0.6096
Overidentification test
Chi-square statistics dof Chi(56) Chi(56)
Chi-square statistics 65.03 63.71
P-value 0.1911 0.2235
P-value for testing coefficients equal to zero
Education 0.0056 0.0070 0.0003 0.0066 0.0080 0.0003
Assets 0.0663 0.0765 0.0605 0.0649 0.0758 0.0561
Age dummies 0.0000 0.0000 0.0047 0.0000 0.0000 0.0084
Marital Status 0.2590 0.3811 0.9290
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

77

Table continues

Table 1.14 (cont’d)

 

 

 

 

 

Female
Variables OLS 2SLS GMM OLS ZSLS GMM
BMI 93 0.481 0.399 0.624 0.483 0.393 0.592
(0.075)** (0.155)** (0.099)** (0.075)” (0.154)** (0.092)”
Some primary 93 0.196 0.237 0.024 0.165 0.210 0.021
(0.292) (0.299) (0.190) (0.287) (0.295) (0.189)
Primary 93 0.047 0.082 -0.081 0.017 0.056 -0.075
(0.316) (0.316) (0.243) (0.314) (0.313) (0.243)
Middle school 93 0.001 -0.002 0.071 -0.043 —0.045 0.055
(0.351) (0.355) (0.250) (0.347) (0.351) (0.250)
High school 93 -0.336 -0.383 -0.086 -0.350 -0.400 -0.104
(0.443) (0.459) (0.354) (0.440) (0.456) (0.351)
Tech/College+ 93 -1.023 -1.052 -0.848 —1.004 -1.031 -0.840
(0.628) (0.609)* (0.534) (0.631) (0.610)* (0.535)
Log R prod assets 93 1 -0.053 -0.047 -0.023 -0.054 -0.047 -0.027
(0.048) (0.049) (0.040) (0.048) (0.049) (0.040)
Log R prod assets 93 2 0.048 0.050 0.035 0.047 0.049 0.038
(0.038) (0.039) (0.034) (0.038) (0.038) (0.034)
Land 93 0.005 0.003 -0.005 0.005 0.003 —0.006
(0.015) (0.015) (0.013) (0.014) (0.014) (0.012)
Married 93 0.361 0.408 0.302
(0.428) (0.459) (0.408)
Divorced Separated 93 0.315 0.249 0.750
(0.846) (0.897) (0.761)
Widowed 93 -0.755 -0.680 -0.530
(0.570) (0.583) (0.489)
Community Dummies Yes Yes Yes Yes Yes Yes
No. of Observations 1948 1948 1948 1948 1948 1948
R-squared 0.352 0.348 0.3219 0.355 0.350 0.330
First stage for BMI 93
R-squared 0.246 0.247
F-statistic for Identifying IVs 1.29 1.20
P-value 0.0767 0.1437
Test of homoscedasticity
Chi-sq stat 100.69 40.43 102.23 43.82
P-value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS: BMI 93
Chi-square statistics 0.31 0.37
P-value 0.5769 0.5437
Overidentification test
Chi—square statistics dof Chi(56) Chi(56)
Chi-square statistics 46.91 48.07
P-value 0.8015 0.7656
P-value for testing coefficients equal to zero
Education 0.4125 0.3383 0.6858 0.4786 0.3865 0.7065
Assets 0.4247 0.4240 0.5843 0.4180 0.4265 0.5223
Age dummies 0.0000 0.0000 0.0000 0.0000 0.0006 0.0000
Marital Status 0.0322 0.0303 0.0167
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.6330

 

Note: Also included in all regressions are age dummies. In 2SLS BMI 93 is instrumented with
real productive assets in 1991, land cultivated and interactions of real productive assets and
land with community free market prices of rice, eggs, pork and fish, as well as water, sanitation
and clinic characteristics in 1991. Huber/White robust standard errors are in parentheses. *
indicates the coefficient is statistically significant at 0.1 level; ** at 0.05.

78

Table 1.15: Determinants of 1993 BMI Conditional on 1991 BMI with Changes in
Community Characteristics

 

 

 

 

Male
Variables OLS ZSLS GMM OLS ZSLS GMM
BMI 91 0.610 0.881 0.883 0.607 0.874 0.876
(0.062)** (0.111)” (0.049)“ (0.062)** (0.113)“ (0.050)”
Some primary 91 -0.027 0.079 0.053 —0.068 0.050 0.024
(0.214) (0.227) (0.193) (0.214) (0.229) (0.193)
Primary 91 0.011 0.099 0.034 -0.028 0.071 0.009
(0.204) (0.206) (0.166) (0.203) (0.206) (0.165)
Middle school 91 -0.039 -0.004 0.105 -0.094 -0.042 0.060
(0.191) (0.199) (0.176) (0.192) (0.201) (0.177)
High school 91 0.128 0.129 0.247 0.078 0.096 0.222
(0.231) (0.237) (0.212) (0.229) (0.234) (0.209)
Tech/College+ 91 0.365 0.344 0.387 0.326 0.319 0.367
(0.267) (0.272) (0.249) (0.265) (0.269) (0.247)
Log real prod 0.004 0.013 0.016 0.003 0.012 0.016
asset 91 1 (0.026) (0.026) (0.021) (0.026) (0.026) (0.021)
Log real prod 0.065 0.050 0.025 0.063 0.048 0.024
asset 91 2 (0.036)* (0.032) (0.022) (0.036)* (0.032) (0.022)
Land farmed in 90 -0.040 -0.033 -0.015 -0.042 —0.034 -0.015
(0.022)* (0.021) (0.016) (0.021)* (0.021) (0.015)
Married 91 0.516 0.358 0.350
(0.249)“ (0.272) (0.212)*
Divorced or 0.173 0.240 0.318
Separated 91 (0.637) (0.617) (0.582)
Widowed 91 -0.040 -0.038 -0.026
(0.390) (0.408) (0.366)
Urban residence 0.402 0.216 0.314 0.402 0.219 0.323
(0.179)” (0.200) (0.151)“ (0.179)** (0.200) (0.151)M
A Price of rice -0.010 -0.007 -0.004 -0.009 -0.006 -0.004
(0.004)" (0.003)” (0.003)* (0.004)” (0.003)“ (0.003)
A Price of eggs 0.001 0.005 -0.001 0 0.004 —0.002
(0.018) (0.020) (0.015) (0.018) (0.020) (0.015)
A Price of pork -0.017 -0.010 -0.011 -0.017 -0.010 -0.011
(0.014) (0.014) (0.011) (0.014) (0.014) (0.011)
A Price of fish -0.027 -0.013 0.003 -0.023 ~0.011 0.004
(0.027) (0.027) (0.022) (0.027) (0.027) (0.021)
A Clinic physicians -16.016 -10.186 -10.965 -16.356 -10.469 -10.156
(16.580) (16.154) (13.762) (16.629) (16.147) (13.730)
A Clinic nurses 38.155 43.631 44.652 40.777 45.346 48.568
(56.472) (53.733) (43.525) (56.516) (53.851) (43.498)
A Clinic beds -7.229 -9.625 —8.745 -7 .468 —9.753 -9.694
(17.994) (17.143) (14.216) (17.982) (17.126) (14.181)
A Water source 2 0.386 0.142 0.277 0.377 0.139 0.292
(0.341) (0.373) (0.300) (0.340) (0.372) (0.298)
A Water source 3 0.545 0.766 0.717 0.533 0.754 0.714
(0.428) (0.451)* (0.360)” (0.433) (0.456)* (0.363)"
A Water source 4 1.862 1.634 1.183 1.910 1.671 1.234
(0.464)" (0.475)“ (0.374)“ (0.465)” (0.478)M (0.372)"

 

79

table continues

Table 1.15 (cont’d)

 

 

 

 

 

Male
OLS ZSLS GMM OLS ZSLS GMM
A In house no flush -1.098 -1.042 -1.294 -1.051 -1.011 -1.270
(1.567) (1.555) (0.996) (1.577) (1.560) (0.993)
A Outside toilets 1.054 1.337 0.793 1.050 1.327 0.808
(0.734) (0.846) (0.655) (0.734) (0.846) (0.656)
A Open pit 1.520 1.798 0.995 1.507 1.783 0.993
(0.698)” (0.796)M (0.628) (0.699)" (0.797)M (0.628)
A No toilets 1.020 1.087 0.425 0.922 1.017 0.339
(1.316) (1.343) (1.121) (1.322) (1.351) (1.122)
A Very little excreta -0.274 -0.389 -0.207 -0.275 -0.388 -0.219
(0.309) (0.351) (0.246) (0.309) (0.351) (0.246)
A Some excreta -0.181 —0.067 0.225 -0.203 -0.086 0.212
(0.288) (0.322) (0.235) (0.288) (0.321) (0.234)
A Much excreta 0.503 -0.539 0.985 0.573 -0.466 1.056
(1.068) (1.366) (0.923) (1.060) (1.361) (0.911)
Community Dummies No No No No No No
Number of obs 2436 2436 2436 2436 2436 2436
R-squared 0.315 0.269 0.259 0.317 0.272 0.262
First stage for BMI 91
R-squared 0.133 0.135
F-statistic for Identifying IVs 4.64 4.96
P-value 0.0000 0.0000
Test of homoscedasticity
Chi-sq stat 101.58 37.72 98.99 34.77
P—value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS for BMI 91
Chi-square statistics 7.35 7.04
P-value 0.0067 0.0080
Overidentification test for a third instrument
Chi-square statistics dof chi(56) chi(56)
Chi-square statistics 67.52 64.50
P-value 0.1392 0.2038
P-value for testing coefficients equal to zero
Education 0.6972 0.8070 0.6175 0.6892 0.8129 0.6309
Productive Assets 0.1115 0.1455 0.1349 0.1432 0.1766 0.1381
Age dummies 0.0366 0.1213 0.0383 0.0240 0.0955 0.0139
Marital Status 0.0620 0.4210 0.2337
Prices 0.0174 0.1649 0.4128 0.0242 0.1917 0.4412
Water source 0.0006 0.0035 0.0118 0.0004 0.0027 0.0080
Toilet type 0.0820 0.0569 0.1951 0.0944 0.0631 0.2004
Excreta 0.7909 0.6854 0.2650 0.7629 0.7049 0.2441
Clinic chars 0.2669 0.4081 0.1607 0.2258 0.3747 0.1156

 

80

Table continues

Table 1.15 (cont’d)

 

 

 

Female

Variables OLS 2SLS GMM OLS ZSLS GMM
BMI 91 0.595 0.837 0.871 0.595 0.837 0.868
(0.041)M (0.074)“ (0.043)" (0.041)" (0.074)" (0.043)**
Some primary 91 0.104 0.090 0.001 0.105 0.092 -0.007
(0.170) (0.182) (0.163) (0.170) (0.182) (0.163)
Primary 91 0.308 0.166 -0.068 0.302 0.160 —0.065
(0.196) (0.216) (0.158) (0.196) (0.215) (0.158)
Middle school 91 0.150 0.123 0.106 0.141 0.114 0.095
(0.191) (0.198) (0.169) (0.190) (0.198) (0.169)
High school 91 —0.240 -0.259 -0.151 -0.234 -0.252 -0.157
(0.236) (0.257) (0.235) (0.234) (0.256) (0.234)
Tech/College+ 91 0.277 0.402 0.364 0.297 0.422 0.398
(0.323) (0.323) (0.296) (0.327) (0.328) (0.300)
Log real prod 0.018 0.004 0.026 0.018 0.004 0.026
asset 91 1 (0.033) (0.033) (0.026) (0.034) (0.033) (0.026)
Log real prod 0.002 0.004 -0.016 0.001 0.003 -0.015
asset 91 2 (0.033) (0.033) (0.027) (0.033) (0.033) (0.026)
Land farmed in 90 0.005 0.020 0.008 0.005 0.020 0.007
(0.025) (0.026) (0.020) (0.025) (0.026) (0.020)
Married 91 0.381 0.371 0.374
(0.418) (0.459) (0.399)
Divorced or 0.743 0.599 0.744
Separated 91 (0.629) (0.649) (0.599)
Widowed 91 -0.065 -0.010 0.120
(0.499) (0.532) (0.472)
Urban residence 0.487 0.266 0.204 0.482 0.263 0.197
(0.155)" (0.174) (0.148) (0.155)“ (0.174) (0.147)
A Price of rice 0.003 0.002 0.004 0.002 0.002 0.005
(0.005) (0.004) (0.003) (0.005) (0.004) (0.003)
A Price of eggs -0.017 -0.022 -0.024 -0.017 -0.022 -0.026
(0.019) (0.019) (0.016) (0.019) (0.020) (0.016)*
A Price of pork -0.011 -0.003 -0.001 -0.011 -0.003 -0.001
(0.012) (0.011) (0.009) (0.011) (0.011) (0.009)
A Price of fish -0.049 -0.041 -0.048 -0.048 -0.040 -0.046
(0.027)* (0.029) (0.024)" (0.027)* (0.029) (0.024)”
A Clinic physicians -18.546 -4.112 3.699 -19.062 -4.577 4.049
(20.299) (22.970) (17.435) (20.409) (23.112) (17.514)
A Clinic nurses -81.505 -34.404 -2.262 -84.451 -36.525 -3.803
(95.088) (106.567) (82.692) (95.471) (106.880) (82.521)
A Clinic beds 0.945 -13.465 -12.685 1.943 -12.713 -11.324
(25.309) (29.868) (23.102) (25.442) (29.989) (23.089)
A Water source 2 -0.311 -0.191 0.008 -0.316 -0.196 -0.018
(0.326) (0.332) (0.276) (0.326) (0.332) (0.277)
A Water source 3 —0.308 0.181 0.407 -0.300 0.187 0.357
(0.525) (0.539) (0.432) (0.529) (0.545) (0.436)
A Water source 4 -0.404 -0.412 0.156 -0.411 -0.419 0.106
(0.510) (0.518) (0.412) (0.511) (0.519) (0.416)

 

table continues

Table 1.15 (cont’d)

 

 

 

 

 

Female
OLS ZSLS GMM OLS ZSLS GMM
A In house no flush -0.262 -0.298 -0.279 -0.261 -0.296 -0.215
(1.019) (1.036) (0.893) (1.017) (1.035) (0.889)
A Outside toilets -0.076 0.202 0.334 -0.110 0.173 0.276
(0.610) (0.701) (0.588) (0.613) (0.704) (0.592)
A Open pit -0.186 -0.076 0.206 -0.227 -0.112 0.137
(0.693) (0.784) (0.643) (0.698) (0.789) (0.648)
A No toilets 0.464 -0.014 -0.660 0.410 -0.063 -0.810
(1.170) (1.265) (1.053) (1.166) (1.265) (1.053)
A Very little excreta —0.225 -0.237 -0.331 -0.236 -0.247 -0.331
(0.287) (0.310) (0.254) (0.285) (0.308) (0.253)
A Some excreta 0.832 0.929 0.606 0.804 0.904 0.594
(0.327)" (0.350)" (0.268)“ (0.325)M (0.348)“ (0.266)”
A Much excreta 0.656 -0.522 -0.413 0.616 -0.558 -0.385
(1.101) (1.084) (0.954) (1.105) (1.086) (0.954)
Community Dummies No No No No No No
Number of obs 2638 2638 2638 2638 2638 2638
R-squared 0.391 0.341 0.318 0.392 0.342 0.320
First stage for BMI 91
R-squared 0.124 0.125
F-statistic for Identifying IVs 6.93 7.44
P-value 0.0000 0.0000
Test homoscedasticity
Chi-sq stat 98.67 40.02 99.30 40.15
P-value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS for BMI 91
Chi-square statistics 11.61 11.60
P-value 0.0007 0.0007
Overidentification test for a third instrument
Chi-square statistics dof chi(56) chi(56)
Chi-square statistics 58.48 56.89
P-value 0.3846 0.4418
P-value for testing coefficients equal to zero
Education 0.3296 0.5643 0.6960 0.3514 0.5734 0.6665
Productive Assets 0.7171 0.9567 0.6077 0.7332 0.9634 0.5975
Age dummies 0.0027 0.0292 0.0196 0.0050 0.0299 0.0204
Marital Status 0.2493 0.4163 0.4329
Prices 0.0779 0.2752 0.0681 0.0768 0.2768 0.0582
Water source 0.7857 0.6972 0.8194 0.7807 0.6827 0.8629
Toilet type 0.8560 0.6224 0.4429 0.8436 0.6021 0.4613
Excreta 0.0093 0.0087 0.0164 0.0114 0.0097 0.0176
Clinic chars 0.5767 0.7177 0.8846 0.5730 0.7190 0.9074

 

 

Note: Also included in all regressions are age dummies for 1993. BMI 91 is instrumented by
real productive assets and land cultivated in 1989 and interactions of these variables with free
market prices for rice, eggs, pork and fish, clinic characteristics, water and sanitation variables
in 1989. Huber/ White robust standard errors are in parentheses.

82

Table 1.16: Determinants of 1997 BMI Conditional on 1993 BMI with Changes in
Community Characteristics

 

 

 

Male
Variables OLS 2SLS GMM OLS ZSLS GMM
BMI 93 0.459 0.485 0.774 0.455 0.483 0.776
(0.070)” (0.154)" (0.079)” (0.070)** (0.153)" (0.080)**
Some primary 93 -0.572 -0.571 -0.610 -0.592 -0.590 -0.617
(0.249)** (0.248)“ (0.228)** (0.247)** (0.246)“ (0.226)"
Primary 93 -0.088 -0.081 -0.244 -0.114 —0.106 —0.236
(0.329) (0.331) (0.271) (0.328) (0.331) (0.270)
Middle school 93 0.300 0.303 0.212 0.258 0.263 0.209
(0.337) (0.338) (0.291) (0.335) (0.338) (0.291)
High school 93 0.159 0.153 -0.132 0.138 0.132 -0.133
(0.333) (0.333) (0.288) (0.331) (0.331) (0.286)
Tech/College+ 93 0.500 0.491 0.280 0.464 0.456 0.272
(0.364) (0.362) (0.321) (0.362) (0.359) (0.320)
Log real prod 0.021 0.021 0.028 0.020 0.020 0.027
asset 93 1 (0.033) (0.033) (0.029) (0.033) (0.033) (0.029)
Log real prod 0.030 0.029 0.010 0.031 0.030 0.011
asset 93 2 (0.033) (0.032) (0.028) (0.033) (0.032) (0.028)
Land farmed in 92 -0.008 -0.007 -0.002 -0.008 -0.007 -0.002
(0.012) (0.012) (0.008) (0.012) (0.012) (0.008)
Married 93 0.271 0.256 -0.107
(0.310) (0.326) (0.265)
Divd Septd 93 -l.131 -1.067 -0.277
(0.959) (0.989) (0.916)
Widowed 93 -0.563 -0.547 -0.348
(0.465) (0.462) (0.432)
Urban residence 1.137 1.101 0.789 1.128 1.089 0.772
(0.224)" (0.287)" (0.212)** (0.224)” (0.286)“ (0.213)"
A Price of rice 0.015 0.014 0.006 0.014 0.014 0.006
(0.006)“ (0.006)** (0.005) (0.006)“ (0.006)" (0.005)
A Price of eggs -0.007 -0.008 -0.034 -0.006 -0.008 -0.033
(0.020) (0.022) (0.016)" (0.020) (0.022) (0.016)“
A Price of pork 0.002 0.002 0.015 0.002 0.003 0.014
(0.011) (0.012) (0.010) (0.011) (0.012) (0.010)
A Price of fish -0.007 -0.007 -0.007 -0.007 -0.007 -0.007
(0.003)M (0.003)“ (0.003)" (0.003)” (0.003)” (0.003)**
A Water source 2 0.589 0.579 0.439 0.596 0.585 0.463
(0.356)* (0.355) (0.265)* (0.355)* (0.354)* (0.262)*
A Water source 3 -0.326 -0.332 -0.100 —0.348 -0.354 -0.093
(0.363) (0.360) (0.313) (0.361) (0.358) (0.310)
A Water source 4 —0.016 -0.006 0.286 -0.006 0.005 0.309
(0.360) (0.366) (0.323) (0.362) (0.367) (0.324)

 

83

table continue

Table 1.16 (cont’d)

 

 

 

 

 

Male
OLS ZSLS GMM OLS ZSLS GMM
A In house no flush -0.527 -0.508 -0.289 -0.560 -0.538 -0.296
(0.588) (0.578) (0.485) (0.581) (0.572) (0.482)
A Outside toilets 0.455 0.418 -0.268 0.465 0.426 -0.266
(0.501) (0.543) (0.432) (0.505) (0.545) (0.435)
A Open pit cement or earth 0.791 0.758 0.149 0.793 0.758 0.150
(0.445)* (0.476) (0.386) (0.447)* (0.476) (0.388)
A No toilets 0.452 0.437 0.977 0.542 0.522 0.955
(1.381) (1.371) (1.052) (1.389) (1.379) (1.056)
A Very little excreta 0.137 0.153 0.309 0.123 0.140 0.291
(0.459) (0.466) (0.359) (0.459) (0.467) (0.360)
A Some excreta -0.581 -0.551 0.006 -0.526 —0.497 0.022
(0.458) (0.487) (0.367) (0.455) (0.478) (0.364)
A Much excreta 1.592 1.621 1.580 1.502 1.538 1.522
(1.014) (1.016) (0.802)" (1.012) (1.018) (0.797)*
Community Dummies No No No No No No
Number of obs 1777 1777 1777 1777 1777 1777
R-squared 0.295 0.294 0.201 0.297 0.296 0.1995
First stage for BMI 93
R-squared 0.141 0.146
F-statistic for Identifying IVs 2.12 2.30
P-value 0.0000 0.0000
Test of homoscedasticity
Chi-sq stat 95.80 18.77 94.28 19.03
P-value 0.0000 0.0001 0.0000 0.0001
Hausman Test between OLS and 2SLS for BMI 93
Chi-square statistics 0.09 0.10
P-value 0.7626 0.7480
Overidentification test for a third instrument
Chi-square statistics dof chi(56) chi(56)
Chi-square statistics 63.79 70.04
P-value 0.2216 0.0983
P-value for testing coefficients equal to zero
Education 0.0005 0.0005 0.0002 0.0006 0.0006 0.0002
Productive Assets 0.1989 0.2046 0.2854 0.2001 0.2052 0.2788
Age dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Marital Status 0.0550 0.1652 0.8811
Prices 0.0168 0.0217 0.0128 0.0202 0.0248 0.0137
Water source 0.2088 0.2111 0.2800 0.1809 0.1841 0.2330
Toilet type 0.0875 0.1140 0.3555 0.0844 0.1100 0.3625
Excreta 0.2323 0.2434 0.2478 0.2906 0.2948 0.2692

 

84

Table continues

Table 1.16 (cont’d)

 

 

Female

Variables OLS 2SLS GMM OLS ZSLS GMM
BMI 93 0.532 0.526 0.670 0.532 0.522 0.638
(0.077)“ (0.114)“ (0.062)** (0.076)** (0.114)“ (0.065)“
Some primary 93 0.116 0.119 0.118 0.086 0.089 0.094
(0.256) (0.262) (0.191) (0.252) (0.259) (0.188)
Primary 93 0.250 0.252 0.298 0.223 0.227 0.263
(0.273) (0.269) (0.228) (0.271) (0.267) (0.227)
Middle school 93 0.237 0.237 0.334 0.194 0.195 0.284
(0.292) (0.292) (0.235) (0.290) (0.289) (0.232)
High school 93 -0.070 -0.072 0.359 -0.102 -0.107 0.290
(0.418) (0.432) (0.353) (0.416) (0.430) (0.351)
Tech/College+ 93 -0.649 -0.650 -0.443 -0.658 -0.660 -0.472
(0.477) (0.477) (0.445) (0.480) (0.479) (0.447)
Log real prod 0.046 0.046 0.053 0.048 0.049 0.054
asset 93 1 (0.039) (0.038) (0.032)* (0.038) (0.038) (0.032)*
Log real prod -0.004 -0.003 -0.013 -0.008 —0.007 -0.013
asset 93 2 (0.031) (0.032) (0.027) (0.031) (0.032) (0.027)
Land farmed in 92 0.000 0.000 -0.011 0.000 0.000 -0.011
(0.013) (0.012) (0.010) (0.012) (0.012) (0.009)
Married 93 0.101 0.103 0.133
(0.429) (0.434) (0.408)
Divd Septd 93 -0.569 -0.574 -0.209
(0.836) (0.841) (0.736)
Widowed 93 -0.980 —0.979 -0.735
(0.560)* (0.562)* (0.506)
Urban residence 0.660 0.665 0.325 0.647 0.655 0.375
(0.237)” (0.241)” (0.185)* (0.237)“ (0.241)“ (0.186)“
A Price of rice -0.019 -0.019 -0.012 —0.019 -0.019 -0.013
(0.013) (0.013) (0.010) (0.013) (0.013) (0.010)
A Price of eggs 0.032 0.032 0.016 0.032 0.032 0.016
(0.018)* (0.018)* (0.015) (0.018)* (0.018)* (0.015)
A Price of pork 0.006 0.006 -0.001 0.007 0.007 0.002
(0.019) (0.019) (0.012) (0.019) (0.019) (0.012)
A Price of fish -0.004 -0.004 -0.003 -0.004 -0.004 -0.003
(0.002) (0.002) (0.002) (0.003) (0.003) (0.002)
A Water source 2 -0.061 -0.058 -0.102 -0.025 -0.022 -0.064
(0.307) (0.317) (0.251) (0.305) (0.315) (0.250)
A Water source 3 -0.035 -0.027 -0.117 -0.119 -0.105 -0.142
(0.478) (0.515) (0.397) (0.475) (0.513) (0.398)
A Water source 4 0.463 0.464 0.503 0.510 0.511 0.569
(0.418) (0.419) (0.344) (0.416) (0.417) (0.345)*

 

85

table continues

Table 1.16 (cont’d)

 

 

 

 

 

Female
OLS ZSLS GMM OLS ZSLS GMM
A In house no flush -1.375 -1.376 -1.378 -1.362 -1.363 -1.385
(0.655)” (0.652)M (0.568)" (0.653)M (0.650)M (0.566)“
A Outside toilets -0.005 -0.016 -0.085 -0.033 -0.050 -0.137
(0.487) (0.554) (0.454) (0.489) (0.556) (0.458)
A Open pit 0.040 0.032 -0.057 -0.007 -0.020 -0.137
(0.482) (0.522) (0.403) (0.485) (0.525) (0.409)
A No toilets -2.689 -2.700 -2.224 -2.604 -2.621 -2.179
(1.135)" (1.170)" (1.030)“ (1.153)M (1.189)** (1.051)M
A Very little excreta 0.105 0.104 0.049 0.102 0.100 0.109
(0.379) (0.379) (0.301) (0.378) (0.378) (0.303)
A Some excreta 0.556 0.558 0.487 0.602 0.605 0.565
(0.463) (0.464) (0.376) (0.465) (0.468) (0.382)
A Much excreta -0.421 -0.448 -0.006 -0.366 -0.409 -0.090
(1.284) (1.308) (0.974) (1.276) (1.303) (0.984)
Community Dummies No No No No No No
Number of obs 1901 1901 1901 1901 1901 1901
R—squared 0.292 0.292 0.2709 0.295 0.295 0.2809
First stage for BMI 93
R—squared 0.138 0.139
F-stat for Identifying IVs 2.96 2.90
P-value 0.0000 0.0000
Test homoscedasticity
Chi-sq stat 122.75 20.93 122.68 22.80
P-value 0.0000 0.0000 0.0000 0.0000
Hausman Test between OLS and 2SLS for BMI 93
Chi-square statistics 0.00 0.01
P-value 0.9495 0.9199
Overidentification test for a third instrument
Chi-square statistics dof chi(56) chi(56)
Chi-square statistics 69.29 68.77
P-value 0.1094 0.1175
P-value for testing coefficients equal to zero
Education 0.5374 0.5141 0.4293 0.5975 0.5732 0.5297
Productive Assets 0.3614 0.3212 0.2018 0.3583 0.3139 0.1855
Age dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Marital Status 0.0345 0.0338 0.0408
Prices 0.0847 0.1158 0.3383 0.1033 0.1351 0.3374
Water source 0.6190 0.6198 0.3297 0.5620 0.5653 0.2620
Toilet type 0.0805 0.0799 0.0721 0.1075 0.1057 0.0848
Excreta 0.6704 0.6716 0.6305 0.6275 0.6270 0.5331

 

 

Note: Also included in all regressions are age dummies for 1993. BMI 91 is instrumented by
real productive assets and land cultivated in 1989 and interactions of these variables with free
market prices for rice, eggs, pork and fish, clinic characteristics, water and sanitation variables
in 1989. Huber/White robust standard errors are in parentheses.

86

1.7 Concluding Remarks

Using an unique longitudinal data set of adult Chinese in the 19903 we estimate
the reduced form and the dynamic BMI demand function for men and women in
an era of transition in China. This paper fills in the gap of previous research that
lacks clear theoretical framework and becomes the first paper on Chinese adult BMI
estimations with emphasis on individual as well as community characteristics.

The key findings of this paper are for women the effect of education is very
strong and inversely U-shaped, even after controlling for community characteristics.
The education effect is stronger in rural areas than in urban areas and for younger
generations than for older generations. Adult men in China had higher levels of edu-
cation than women, however their educational effects on BMI were not significant. If
the BMI of man is a proxy for both health condition and human capital, the protec-
tive effects of education in health (negative) and the enhancing effects of education
in human capital (positive) might very well cancel each other out. For women it
seems appropriate that BMI plays only the role of a proxy for health condition be-
cause a woman’s human capital does not depend strongly on her stature. Hence the
effect of education on BMI reflects its effect on health condition only. Over time
the protective effects of education on women dissolve and the enhancing effects for
men increase (though not significant conditional on household and community vari—
ables). There is room for improvement for both men and women through increasing
education quantity and quality.

Assets, prices and environmental health conditions are all important determi-
nants of adult BMIs in China and these factors affect men and women in different
age groups and across regions differently. The effect of productive assets differs be-
tween men and women. For men, more productive assets are associated with higher
BMI, with larger effects in the upper spline range and the joint effects are highly

significant. For women, higher level of productive assets are associated with higher

87

BMI only for those with assets less than the median; above the median the associa-
tion is negative but not significant. Eggs and pork constitute of high calorie intakes,
hence the increase in prices may have a substitution effect for other foods with lower
calorie and result in lower BMI. Fish and rice, on the other hand, can be thought of
as substitute for red meat and wheat flour. The increase in fish and rice prices will
hence result in more consumption of less healthy foods and increased BMI.

When the amount of information at the community level is plenty it is critical
to choose proper number of variables to include in any model, avoiding both the
multicollinearity problem and a fully saturated model. The current food prices affect
BMI through intakes depending on which foods have higher prices; and the changes
in food prices over time affect BMI depending on which foods have higher changes
in prices. There is a trade-off between including community dummies or not when
we use current community characteristics in the model. The advantage is community
dummies capture unobserved heterogeneity at the aggregate level. The disadvantage
is any effect of community variables will have to be identified through the variation of
them over time and in a short run the change may be too little for good identification.

In the short term dynamic BMI demand function it seems to be quite reasonable
to assume the lagged BMI is a good summary statistic of all past information for
the individual. The estimated effect of lagged BMI in 2SLS is bigger than that in
OLS estimation. Controlling for lagged BMI the effects of education and household
resources in 1991 were no longer significant in both OLS and 2SLS estimates, except
that in GMM estimation the primary diploma in 1991 showed strong negative effect
for men. The male marriage premium in BMI became insignificant. The conditional
age and community effects are significant for women and the age effect conditional
on past BMI for men is not jointly significant. With an increase in the price of rice,
for example, we would expect to see BMI to increase if substitution (for wheat) effect

is stronger than income effect. However, in our sample the increase in wheat price

88

is actually bigger than the increase in rice price from 1991 to 1993. Therefore there
is no strong substitution effect and the coefficient for rice price changes are negative.
Although the conditional BMI specification works better with shorter lagged period,
the changes in community level information during such short period may not be
large enough to correctly identify any effect.

There are some limitations in this paper. First, using the productive asset in
the reduced model as a proxy for income and treating it as exogenous is not without
criticism. In a dynamic model BMI can affect productivity and therefore the holding
of productive assets in later periods. If there was selection into more strenuous
vocations with higher wages, the results will be biased upwards. Attenuation bias
could also rise with imperfect measures of productive assets and land.

Secondly the dynamic BMI demand function hinges on the assumption that
one lagged BMI is a sufficient statistic for all past information. When the time
gap between the dependent and the conditional BMI is too large this assumption is
unattainable.

Finally, the model is based on the assumption of perfect insights. Ignoring uncer-
tainty is not innocuous when there are risks other than idiosyncratic shocks. Uncer-
tainty would be important if overweight increases risk of mortality and morbidity and
people are well aware of it. The length of individual life time and the household size
are all assumed to be exogenous whereas in reality both are potentially endogenous.

With the above caveats in mind this paper provides careful exposition of the
socioeconomic determinants of adult BMI in China in the 19908 and finds strong
association between education and BMI in women and in rural areas. The knowledge
of this relationship can assist public policies to identify target groups for improving
their health status. As the Chinese economy undergoes the rapid structural transition
it is extremely important to find factors that can make such transition as smooth as

possible. Increased overweight and obesity in adults imposes high burden on the

89

health care system, thus obesity awareness education (even general education level)

and prevention measures should pay off in the long run.

90

Chapter 2: Adult Chinese
Macronutrient Consumption and

Socioeconomic Determinants in the

Early 19903

2. 1 Introduction

Alongside its rapid economic development and social and cultural transitions in
the 19903 China also experienced shifts in diet, physical activity and leading causes
of death (see Ge et al. 1992 and references in Chapter one). The dietary pattern
is moving toward one in which the proportion of energy intake from fat increases
each year. The Chinese Recommended Dietary Allowance (CRDA) of 2,400 Kcal was
established by the Chinese Nutrition Society in 1981 (Chen 1990, Ge et a1. 1991)
for adult male aged 18 to 40 and undertaking very light activity. According to the
Food and Agriculture Organization (FAO) estimates from the Food Balance Sheet
(PBS)1 in the 19803 the CRDA has been reached, per capita daily protein and fat
intakes rose dramatically in the 19803 and 19903, and both the percent of protein and
percent of fat from vegetable products are decreasing and the percents of protein and
fat from animal products are increasing (Figure 2.1). The overall calorie supply, for
instance, has increased from 1,953 in 1966 to 2,766 kcal in 1996 per person per day.

The per capita supply of meat has quadrupled, growing from 77 kcal per person per

 

1The FAO collected officially reported data on the production, trade and utilization of agricultural
commodities for countries worldwide. These databases are used to set up FBSs which provide
essential information on a country’s food system (Heilig 1999). How reliable these data are is
discussed in the next Section.

91

day in the mid-19603 to 320 kcal per person per day in the mid-19903.2 According to
a household survey of eight provinces in China during the eight-year period of 1989
to 1997 the intake of cereals decreased considerably by 127 grams per person per day;
the intake of vegetables decreased by 32 grams per person per day; and the intake
of animal foods increased by 46.7 and 36.8 grams per person per day for urban and
rural residents (Du et al. 2002). During the rapid economic growth after reforms
more health risks are resulted from the deteriorating dieting habits (Guo, Popkin and
Zhai 1999, Guo et al. 2000). It is projected that diet-related chronic diseases, such as
obesity and coronary heart disease, will present a huge health care burden for China
in the near future (Popkin, Paeratakul, Zhai and Ge 1995a).

Similar trends in nutrition transition are found in many low-income countries
(Popkin 2002, Drewnowski and Popkin 1997, Guo 1998) but no universal conclusions
concerning the roles of socioeconomic determinants on the dietary transitions exit.
Increased income expands a household’s budget constraint hence resulting in higher
quantity and quality of food consumption. Urbanization and increasing exposure to
western life styles may change tastes and food choices. Improved education raises
the allocative efficiency of food consumption, especially during times of technological
change. Higher education is also associated with increased income and investment
in human capital. Changes in prices of foods and other community characteristics
also influence food consumption through price and / or income effects. In Luo (2003 b)
we present evidence that many individual, household and community socioeconomic
determinants are significantly related with adult body mass index (BMI). It is con-

ceivable these factors should also be related with food consumption, a key input into

 

2The balance sheets can be only as good as rudimentary information for food availabilities. In
theory, food balance sheets should take into account the complete food chain, from the field to
supply in the shops. They are an accounting system that balances production, exports, imports,
overall domestic supply, use as feed or seeds, waste, inventory change, and direct use for human
consumption. Unfortunately, accurate data are not always available, especially for the amount of
waste. Missing data have to be estimated so that the sheet can be balanced. Such estimations
inevitably introduce inaccuracies. However, the FAO time series on food availability provides useful
and valuable information on general trends of dietary patterns.

92

BMI. Food consumption quantity in this paper refers to individual daily intakes of
calories, fats and proteins and food consumption quality is measured by percent of
calories from fat, from protein and from carbohydrates. Nutritional status, referring
to a person’s physical wellbeing as a result of ingestion, absorption and utilization of
nutrients, and measured by BMI in our previous study, is not the focus of this paper.

In nutrition literature, diet can be described in terms of its chemical composition,
for example, its nutrient content, or alternatively, in terms of foods or food groups.
The advantage of representing diet as the total intake of a nutrient rather than using
the contribution of only one food at a time is that such information can be directly
related to exiting knowledge of biology (Willett 1998). The advantage of studying
foods or food groups in economic analysis is that the own and cross price elasticities
are well defined and easy to interpret. However, there is huge geographic variation in
food consumption preference in China. The northern part (north of Yangzi River) of
the country produces and consumes more wheat products, while rice cultivation and
consumption dominate in the south. The average daily per capita consumption of
rice, wheat and other cereals was 25 g, 340 g and 215 g, respectively in the northern
provinces and the southern counterparts of these staples consumed was 401 g, 72
g and 42 g in 1982 (Ge et al. 1991). In addition, since nutrient intakes have been
viewed as indicators of a person’s wellbeing, they are used in this paper to find their
determinants rather than finding the determinants for several key foods.

There are generally two categories of nutrient determinant studies: the demand
for nutrient intakes and the production of nutritional status as a biological indicator
(Behrman and Deolalikar 1988), the latter being beyond the scope of this paper and
only lightly touched for completeness of documentation. In the first category, prices,
income, endowment and many economic and social factors are presumed to determine
the behavioral process. In this paper we focus on testing the effect of socioeconomic

factors on demands of total calorie, protein and fat intakes and the percent of energy

93

from protein, fat and carbohydrates. We use the China Health and Nutrition Survey
(CHNS) data in 1989, 1991 and 1993 for adults at or over 20 years of age who were
studied in our previous chapter on BMI.

For reasons stated in the earlier chapter we do not focus our attention on the
association between income and the above outcomes but instead use productive assets
as measure of household resources. In the overall sample education does not have
significant impact on calorie intakes but does affect percent of calories from fat, from
protein and from carbohydrates differently in different region and at different age.
The effect of productive assets is nonlinear and in inverted U-shape for male calorie,
fat and protein intakes; whereas for women more productive assets are associated
with more fat and protein intakes and more percents of calories from fat and from
protein. In rural areas the effect of productive assets is stronger than that in the
urban areas. At different ages the effects of assets are different. For the elderly men
calorie intakes are related with productive assets in U-shape rather than the inverted
U-shape seen for younger men. The effects of productive assets on fat and protein
intakes are significant for younger men and women but not for the elderly. Prices
of foods, community water and sanitation conditions are also studied. The effects of
prices on calorie, fat and protein intakes and the quality of diet measures can go in
either direction. The prices of rice and eggs are negatively associated with calorie,
fat, protein intakes and percents of calories from fat and protein, and are positively
related to percent of calories from carbohydrates. The price of pork is positively
associated with all intakes but the percent of calories from carbohydrates. The price
of edible oils is negatively associated with calories, fat and percent of calories from
fat in the overall sample for men and women but is positively related to the percent
of calories from protein for women. Improvements in sanitation are associated with
more energy and protein intakes in urban areas.

The basic estimates in this paper use pooled OLS method and focus on individ-

94

ual and household characteristics controlling for community year interactions. The
augmented models focus on the effects of prices, water and sanitation characteristics
at community levels. The pooled estimates imply the effects of prices and other de-
terminants over time do not change, which may be hard to maintain. During times
of transition, it is important to examine the effects of changing preferences of food
consumption caused by changes in the socioeconomic factors. By exploring the longi-
tudinal nature of the data set we are able to find out the socioeconomic determinants
of the change of people’s food consumption. Hence the changes in all outcomes be-
tween 1989 and 1991 and the changes between 1991 and 1993 are pooled together
and the OLS method was used in the basic and augmented models again to find
out the effects of the changes in prices and other community characteristics on the
changes in calorie, fat and protein intakes and the changes in percent of calories from
fat, protein and carbohydrates. The changes in prices and community characteristics
may not bear the same effect on the changes of the nutrient consumption.

Finally for completeness of documentation, a simple health production function,
as measured by weight in 1993 (or BMI in 1993), is estimated using lagged weight,
height (or lagged BMI) and current food intakes as inputs. The two stage least squares
(2SLS) estimates suggest adult weights between a short period of two years can be
modelled as a random walk process.

The paper is organized as follows. Next section reviews the literature in nutrient
demand studies, followed by description of the data and discussion of the econometric
issues in the reduced form analysis of nutrient intakes. Section IV and V present the
basic and augmented models for the levels and changes in outcomes of interest. The
penultimate section includes some discussion of the production function analysis. The

final section concludes.

95

FAO-FBS Calories Availability (Kcal)

FAO—FBS Protein Availability (gram)

—— Total Calories i(KCal)
_____ Calories: Veg rod _1 —~ Calories: Ani Prod

 

 

4"”-
,-
./
2mo~—hW#-m—wv~uum1m—wrr~mmmm vef+m_w
/,‘_\/..~'
P__—-A/
/,
7/
/
4 / /" D 'A
7 "V ..4..__ 7 \
2400 _— flAfl«i§* Aflr‘- : 'T——-" _— 1'
// / M w
.v/
P, d
f
"'
A 1’ ‘d‘ \‘V/" /
zmo — -—«————~~—« ———yv‘ *———-——-——— —¢?:‘—:v—>~¢p’— __.____ H—vw—w—mh—b—mw ———-—~—————-——~«——————~—-————~———~—~———~~——~——
V .7\\- , ,___ _ —~. —”
v ~ \‘ M
I /’ \\. v
1800-7 //’ “s-
z// ‘7
x" /
,.’ A/
//

1500-a
1200.4
eoo-a
eoo-a J,,,

 

300 —« .* — —.- _______,_.__:fl:_____‘_<_:______, W4
ma-r/
-11

T l j , f l — l
1961 1966 1971 1976 1981 1986 1991 1996
year

 

 

 

1, — — ——— Total Protein (9
————— Protein: Vegeta le Products — ., —-— Protein: Animal Products

90 *j
so 1

70 A _"Am_fl,.vfifi_.____,_._. _._ ,A-_ 7-v ,vmsgifi. ”A 7*

 

 

'I/ /" fl ’—
A _ __ /
. / __ - x
..J _ a 4*” #fig ”—7._~_—\_._____ W A—
50 ’/ f\
/ ~\ ._ fix '.
f/ \\ V " _ ,_..
9/ v
) / f \ /
/ ..I f ‘ ..—
., \ I \ /

30—« flf/

 

 

 

207—~WW~—- ~ 9g

10 4 - .x/«””

3d _/

 

 

if I _l l ._ l
1961 1966 1971 1976 1981 1986 1991 1996
year

96

 

es—— Total Fat (9
—- — — -- Fat: Vegetable Products fl- f~ Fat: Animal Products
904;
so—l1
g 70) 9 Mum—mm” murm— 1
9 l
E 604.
s l
g 501------Jr ——-——-——‘*—*- ,n,+1_1, ”'
< l
{E 4071' ~
8 ‘ - ’ ~
g 30% _ f
l r\\\d#/f_ g”, f -1 \
s 261 111111 1,- I“ 1.1 » “4+ “on”
10J - ’ZC::—~~5’ I “

 

. T ' '1— _ '—l _—‘_“—' fi '—' ' " '7" ————'_I —_ __ ' ___ " T-—
1961 1966 1971 1976 1981 1986 1991 1996
year

Figure 2.1: FAQ Food Balance Sheet Estimates of calories, protein and fat availabili-

ties and percent of protein and fat from vegetable and animal products in 1961-2001.

2.2 Literature Review

2.2.1 Reduced Form Demands

Generally the reduced-form demand function for foods can be solved recursively
from a dynamic household maximization of intertemporally separable utility under
certainty (Luo 2003b). The demand equation for consumption goods 2:, health or

nutritional status h, leisure l, and health inputs m can all be written as

($,h, 13m): = 9;,h,z(Px,Pma1111111898911,ZeWhéh) (21)

97

where pa, = (1359,1254, ...,pfi) are past and future prices for 3:, pm = (p?,p’1’~’_1,...,p6")
are prices for m, w,— = (wiT,w,-T-1, ...,wio) are wage rates from period 0 to T, y =
(yiT, y,T_1, ..., yio) are household non-wage income, and vi, éh are unobserved individ-
ual and household heterogeneity in input‘demands over the life time, such as genetic
traits, household preferences and environmental factors, are known to family mem-
bers and are assumed to be distributed independently of socioeconomic determinants
of interest. Vectors of exogenous individual (6,), household (6),), and community (zc)
characteristics are observed, such as education levels, household resources and com-
munity infrastructures. In equation (2.1) all the determinants, including individual,
household and community characteristics (19,-, 6h, 20) and prices, except for wage, are
assumed to be orthogonal to U“, 6,. Clearly in a nutritional status production func-
tion analysis all nutrient intakes and other health inputs, and leisure should be treated
as endogenous. Studies that do not clearly distinguish the two types of analyses are
bound to result in bias and fall short of empirical usefulness in policy implications
(Schultz 1984). In reviewing the literature we put all studies in perspective of the

above framework and discusses findings.

2.2.2 Descriptive Studies

Piazza (1983) estimated China’s per capita daily nutrient availability in energy,
fat and protein for the period of 1950 to 1982 based on yearly food balance sheets
(FBS). FBS estimate the quantities of food commodities available for direct human
consumption as the differences between domestic supply and all non-human food end
uses. Wang et al. (1993) improved upon the Piazza underestimates in lean years and
overestimates in bumper harvest years by including food—stock changes in the FBS.
Both studies show rapid nutritional improvements. The total per capital daily energy
availability rose from 1,614 Kcal in 1950 to 2,526 Kcal in 1981 (Piazza 1983); and the
counterparts estimated in Wang et al. (1993) are 1,752 Kcal in 1950 to 2,506 Kcal in

98

1981. The level of per capita daily energy availability peaked in 1984 at 2,863 Kcal
and started to decrease to 2,652 in 1991 (Wang et al. 1993). In addition to energy
level increases, the percentage shares of energy and protein from animal products
more than doubled during 1979-1991. Percent energy from animal product increased
from 5.1% in 1979 to 11.6% in 1991 and animal protein increased from 8.4% to 20.5%.

Estimating per capita food availability from aggregated balance sheets can be
noisy and erroneous. To estimate food available for human consumption, one needs
to incorporate measurements of production, trade, use as feed or seeds, waste and
stock changes. When much food consumption is for on—farm consumption and does
not pass through commercial channels, underreporting of production is a character-
istic of most developing reporting agricultural system (Poleman 1981). Consistency
checks between food production, consumption and trade data in India and Pakistan
indicate that output of major food grain was underestimated by about 25 percent in
the 19503 and 19603 (Evenson and Pray 1994). However, in China underreporting
may not be the problem due to exaggeration in official reports and there remains
much uncertainty on crop waste after the harvest and the magnitude of grain stocks
(Smil 1981). Overestimates may result by assuming low waste, minimal animal feed
and high grain milling rates. In addition, average per capita nutrient availability
derived from an aggregate macroeconomic balance sheet provides general trends in
consumption but is not opportune for studies at microeconomic level.

Availability is not intake either (Dowler and Seo 1985). Typically after con-
structing food balance available for consumption and converting food quantities into
nutrient content using standard food composition tables, the average per person avail-
ability is calculated where each person is weighted differently according to age and
gender. The weights may “have little to do with the metabolic processes involved.”
(Srinivasan 1992). Accurately measuring nutrient intakes is complex and difficult.

Household food inventory method for evaluating household food intakes and individ-

99

ual 24-hour recall method are becoming increasingly available in developing countries.
The household inventory method suffers from several potential biases as food avail-
ability in the national level, and leakages (wastage, food prepared for guests and given
away, and meals eaten away from home) are not captured or systematically related
to income (Strauss and Thomas 1998). Substantial intra-household variation is not
taken into account either (Behrman and Deolalikar 1988). The 24-hour recall method
corrects the above errors but may suffer from recall errors. Srinivasan (1994) does not
think the 24-hour recall estimates could be termed “habitual or long-term intakes.”
Bouis (1994) compared food demand patterns from food expenditures (availability)
and 24-hour recall surveys (intakes) in Kenya and Philippine and found that the re-
lationship between calorie and income could be overstated using the former method.

The CHNS is one of the rare household surveys that use both individual 24-
hour recall and household inventory methods over three random consecutive days to
estimate nutrient consumptions.3 The 1991 Food Composition Table (FCT 1991) for

China was utilized to calculate nutrient values for the dietary data.4 The fundamental

 

3As documented by the CHNS research team: all adults aged 20 to 45 in 1989 and all household
members in 1991 and subsequent surveys provided individual data on dietary intake, with chil-
dren’s intakes reported by mothers. The 3 consecutive days during which detailed household food
consumption data were collected were randomly allocated from Monday to Sunday and are almost
equally balanced across the 7 days of the week for each sampling unit. Household food consumption
was determined by examining changes in inventory from the beginning to the end of each day, in
combination with a weighing and measurement technique. Chinese balances with a maximum limit
of 15 kilograms and a minimum of 20 grams were used. All processed foods (including edible oils
and salt) remaining after the last meal before initiation of the survey were weighed and recorded.
All purchases, home production, and processed snack foods were recorded. Whenever foods were
brought into the household unit, they were weighed and preparation waste (e.g., spoiled rice, dis-
carded cooked meals fed to pets or animals) was estimated when weighing was not possible. At the
end of the survey, all remaining foods were again weighed and recorded. The number of household
members and visitors were recorded at each meal.
Each individual’s average daily dietary intake, calculated from the household survey, was compared
with his or her dietary intake based on 24—hour recall data. Where significant discrepancies were
found, the household and the individual in question were revisited and asked about their food con-
sumption in order to resolve these discrepancies.
All field workers were trained nutritionists who are otherwise professionally engaged in nutrition
work in their own counties and who have participated in other national surveys. Almost all inter-
viewers were graduates of post-secondary schools; many had four-year degrees. In addition, 3 days
of specific training in the collection of dietary data were provided for this survey.

4This FCT represents a significant advance over the earlier China FCT both for higher quality
chemical analyses and for improved techniques of developing average nutrient values for foods whose

100

assumption of such a calculation is that the nutrient content of a specific food is
approximately constant. Such an assumption may be acceptable for certain foods,
such as carrots, but is not without contention for others, such as selenium (Willett
1998). Uncertainty regarding the constancy of the nutrient content of food can be and
has been checked in nutritional studies but can not be avoided in household surveys
whether the inventory method or the 24-hour recall method is used.

Zhai et al. (1996) compares the individual 24-hour recall method and the house-
hold inventory weighing method using the 1991 CHNS data for all individuals includ-
ing adults and children. They separate the sample into three groups of households
—- those with guests for meals, those with away from home food consumptions, and
those with neither. The 1991 Chinese FCT consisting of raw food items and their
nutrient contents was used and a procedure to modify the 24—hour recall measure to
include edible oils and other condiments was developed. With the modification for
oils and condiments, the authors found only a 74 kcal difference between the two
methods for daily calorie intake for a reference man weighing 65 kg, age 18-45 and
undertaking light physical activities, with household inventory method having higher
estimates. The relative differences were larger for protein (3%) and fat (5%). For
households with guests and away from home consumptions the ratios are 3—5% for
protein and 4-12% for fat (household inventory method gave higher estimates in both
outcomes). Without the modification the difference is much larger. The study shows
considerable agreement of the two methods and the importance of allocating house-—
hold cooking oil used to each individual. Not surprisingly, the household inventory
and modified 24-hour recall measures of energy intake in 1991 were lower than that
in Wang et al. (1993), and the weighing inventory method estimation of fat intake is
much higher (62 vs 54 per person per day). As acknowledged by the authors, alloca-

tion of oil among household members was based on weights defined by the Chinese

 

nutrient value varies over the country in a geographic context. The UNC group has worked with
the Institute of Nutrition and Food Hygiene to update and improve this FCT.

101

recommended dietary allowances for energy, rather than based on each individual’s
proportion of the total household food consumption. An effort was made by the re—
search group in later studies to allocate total oil consumption to all members based
on their individual consumption (Guo, Popkin, Mroz and Zhai 1999), although they

did not report detailed comparison of the results from the two modification methods.

2.2.3 Income and Nutrients

There is a considerable volume of research in studying the income or expendi-
ture elasticities of calories (Behrman and Deolalikar 1988, Bouis 1994, Strauss and
Thomas 1995, Subramanian and Deaton 1996, and Behrman et al. 1997) and less so
for protein and other macronutrients. A number of studies focus on the role of income
in household food demand in China (Bhandari 1998 and Guo 1998). This paper is not
aimed at establishing such relationships for two reasons. Income measures in house—
hold surveys are fraught with measurement errors which result in attenuation bias
in estimating income elasticities. Secondly, when unobserved factors affecting nutri-
ent intakes (such as body metabolism rates) also correlate with income levels (such
as through piece rate income determined by body strength) then income should be
treated as endogenous. Otherwise there will be endogeneity bias. However proper
instruments dealing with both problems are extremely difficult to find and verify. For
completeness of documentation we review a few of such studies for China plagued
with one or two of the problems and studies in which researchers strive for a solu-
tion albeit imperfect. For other countries in nutrient and income relationships see
comprehensive reviews in Behrman and Deolalikar (1988) and Strauss and Thomas
(1998) and references within.

Using a sample of adults 18 to 50 years of age from the 1991 CHNS data Ma
and Popkin (1995) analyzes three nutrients: fat intake, calories and percentage of

calories from fat. The authors considered income, age, sex, education, smoking,

102

drinking alcohol, intensity of labor activity, family size and region of residence as
explanatory variables. The authors also tried including height, weight, number of
children under the age of seven years, and job types and excluded them based on
10% statistical significance cutoff. The authors estimated three statistical models: a
linear relationship between income and outcomes controlling for other covariates, a
model with linear and interactions of income and other covariates, and a switching
model regressing all other covariates on outcomes based on different income cut-offs
in defining high vs low income families. In the first model the authors found a positive
income effect for fat intakes and percent calorie from fat but a negative income effect
for calorie intake. In the interaction model the interaction terms are not significant.
In the switching model, where the sample is broken at 85% of income level, the two
regimens, high vs low income families, have significantly different coefficients for all
covariates and the income effects of fat and calorie intakes have different signs for
high vs low income families. The results of this paper is hard to interpret. First, it
includes choice variables that are clearly part of the biological nutrition production
process, such as smoking, drinking alcohol, intensity of labor activity. These variables
should be considered endogenous. The study is thus not a reduced form demand
function study and it is not treating factors in a production function as endogenous
either. Secondly, the switching point, i.e., the income cutoff point used to define
high vs low income families, was found using a trial and error approach based on
statistical significant differences on coefficients; and the model did not control for
potential self-selection bias. Despite the painstaking effort and the correct envision of
the nonlinear income effect and behavioral differentials between the higher and lower
income families, all main parameters of interest are biased and the direction of the
bias is unknown.

In Guo et al. (2000) the researchers present evidence of remarkable shifts in how

Chinese diets varied with income over 1989 and 1993, and are interested in finding

103

the nonlinear effects of income on six groups of key food consumption (rice, wheat
flour, coarse grains, pork, eggs and edible oils), and dietary fat intake. Using the
1989, 1991 and 1993 waves of the CHN S data for adults aged 20 to 45, a two-stage
estimation method was to correct for measurement error and endogeneity bias of
household income, while also controlling for a set of food prices, age, gender, educa-
tion level, household size, place of residence and region. The authors only mentioned
that the instruments in the first stage for per capita household income included “com-
munity information, family background variables, and household business and asset
measures” but did not give specific variables. The authors did not provide statistical
testing on the validity of the instruments they used. Individual level random-effect
models were used for each key food and for dietary fat intake to control for within
household unobservables. The model estimated nonlinear income effects through
income, income square, income and year interaction, and income square and year
interaction and the authors presented variations in elasticities through graphs. The
major findings are: income effects for low-fat, high—fiber foods (wheat flour products,
rice and coarse grain) fell from 1989 to 1993. Higher-fat foods such as pork, edible
oils and eggs became more responsive to income levels. The quantity of fat in the
diets increased significantly and seemed to be rising rapidly with increases in income.
The authors believe these changes are ”an important deterioration in the healthiness
of the Chinese diets that could burgeon as the Chinese economy continues its expan-
sion.” This paper is the only one studying Chinese food consumption treating income
as endogenous and the results are striking. Simple income changes alone can not
explain the complex shifts in Chinese adult dietary patterns in the late 19803 to the
mid-19903.

For the elderly in China, the CHN S 91 and 93 were used (Stookey et al. 2000)
to estimate the determinants of nutrition intakes. The dependent variables included

energy, fat, protein intakes, consumption of rice, high-fat red meat, eggs, and plant

104

oils. The explanatory variables were age groups, sex, income tertiles, and rural—urban
residence. Increasing income was significantly associated with greater energy, fat and
protein intakes. Rice consumption declined with older age and consumption of plant
oils, high-fat red meat and eggs increased by income tertiles.

There is a list of studies focusing on the relationship between income and dietary
transitions (Guo, Popkin and Zhai 1999 and many similar studies in the references)
that generally suffer from the same biases as the above two examples. Generally these
studies point to the relationship that higher income levels, particularly in urban areas,
are associated with consumption of a diet higher in fat.

Because the authors in Guo et al. (2000) were mostly interested in estimating
income elasticities of various foods and fat intake they did not present results on any
other socioeconomic determinants. These factors are of interest in our analysis. In
this paper, instead of studying the income effect of nutrient intakes, at individual and

household levels we focus on the effects of education and productive assets.

2.2.4 Price and Nutrients

As is seen in equation (2.1) food consumption is a function of prices of all goods.
For farm households in the special case in which household and farm-allocative deci-
sions are separable, production inputs prices affect household consumption only via
profits and can be replaced by farm profits or full income. Strauss (1984) shows the
own-price effects on calories consumption in Sierra Leone are negative when profits
are allowed to vary, although they could also be positive through income effect for
net sellers. Pitt and Rosenzweig (1985) estimates food price effects on aggregate
household consumption of foods converted into nine nutrients based on a probability
sample of 2,847 farm households in Indonesia. Only one of the four prices was found
to be unambiguously and negatively associated with the aggregate consumption of

all nutrients. Knowledge of how changes in food prices affect individual level nutrient

105

intakes is crucial to effective policy making. Here we review one paper on the effects
of prices on macronutrients in China (chapter 7 in Guo 1998).

Using the 1989, 1991 and 1993 samples of CHNS of adults aged 20—45 years, Guo,
Popkin, Mroz and Zhai (1999) studies the own- and cross-effects of six prices (rice,
wheat flour, coarse grain, pork, eggs and edible oils) on the consumption of six food
groups and on overall energy, protein, fat intakes, the proportion of the recommended
daily allowances for energy and protein, and the percentage of calories from fat. The
sample is stratified by income and the price elasticities were estimated in two parts —
the probability of consuming certain food items and the quantity consumed given it
is nonzero, similar to Pitt (1983) in an effort to incorporating corner solution at zero
consumption. Age, gender, education, household size, urban residence and regions
are included as control variables in addition to prices and income. Free market prices
were measured at the community level and when the goods were not sold in the
free market prices from the ration system were used. The measurement of intakes is
modified on account of the cooking oils based on proportion of individual meat and
vegetable consumption in total household consumption. In both the model for the
likelihood of consuming specific foods and the model for the quantity consumed, the
same set of socio-demographic factors are used. The authors found the likelihood
of consuming edible oils and rice would decrease by 16 to 20 percent respectively
for 10% increase in real prices of edible oils and rice and price changes for animal
protein foods had a large effect on reducing fat intake. Increases in the prices of
pork, eggs and edible oils are predicted to lower fat intake. Only increases in pork
prices led to reduced protein intakes. Besides the stratification of samples based on
endogenous incomes and inclusion of household size as exogenous control variable,
there may be severe bias due to omitted variables at the community level. Besides
prices the authors did not include other community characteristics such as water and

sanitation factors. Aggregate policies, not only on prices but also on program such as

106

food subsidies, could also affect food consumption. Ignoring these factors produces
downward bias in standard errors of parameters for prices and unsigned biases in
the parameters themselves. When time-invariant factors are not of primary interest
and more than one year of data are available, the fixed-effect model will remove
the time-persistent unobserved heterogeneity bias. We include community dummy
variables representing unobserved heterogeneity and find the effects of several food
prices changed signs. Since prices were measured at community level, not estimated
from household unit costs, the issue of endogeneity of prices due to quality selection is
avoided (Deaton 1988), although the prices could be measured with error, especially
due to substituting free market prices with ration prices.

In our study, specific foods or food groups are not the focus. However, better
understanding of the impacts of food prices on total calories and the composition of
energy intakes has important policy implications. It provides guidance toward shaping
the healthiness of possible targeting populations and reducing the risk factors related

to quality of diets.

2.2.5 Education and Nutrients

The interpretation of the effect of education on labor market outcomes is debated
over how much of it actually represents the benefit of human capital accumulation and
how much from unobserved individual (ability) and family background (for a com-
prehensive review see Strauss and Thomas (1995)). Similar arguments can be made
in studying the effect of education on nutrient intakes. Improved education raises the
allocative efficiency of food consumption, especially at times of technological change.
Higher education is also associated with increased income and investment in human
capital. In addition, schooling can affect preference or tastes for food composition,
although the effect may be small. More recently the causal effect of education on

earnings has been estimated through supply-side institutional features such as com—

107

pulsory schooling laws, differences in distance to schools as instrumental variables and
it is believed that such IV estimates reveals underlying heterogeneity in the returns to
human capital investments (Card 2001). In a reduced-form demand analysis we are
not concerned with estimating the causal effect of education and can not distinguish
the different interpretations of the effects of education.

Bhandari (1998) estimated the association between education and food consump-
tion in China using the 1991 CHNS. The author studied the likelihood of consuming
22 food groups and the effects of male and female household head education on
macronutrient intakes of calorie, protein and fat. The explanatory variables consid-
ered exogenous were income, age, gender, urban and province of residence, physical
activity levels, and five interactions. There was no significant effect of either male or
female education on caloric intake, a significantly negative effect of male education
on protein intake and a positive female education effect on fat intake. Since all of the
above mentioned limitations apply for this study, the results are to be interpreted
with caution. Another disadvantage of the study is the author used only one wave
of the data in which the effects of community year interactions and the changes in
prices, water and sanitation conditions can not be effectively controlled.

Since education partially reflects household resources it is necessary in our basic

and augmented analyses to control for household resources and local infrastructures.

2.3 Data and Econometric Issues

2.3.1 The Data

Three rounds of the CHNS in 1989, 1991 and 1993 are used in this study. The
focus is on adults at or above 20 years of age and those who were in the study of
determinants of BMI in an earlier paper. For detail of the sample derivation see Luo

(2003b). The 1997 nutrition data are not publicly available yet. For some individual,

108

household and community characteristics see Table 1. The percentage of women with
no formal education is much higher than the percent of men (about 30% vs 10%).
Since in 1989 our sample includes only those who were less than 46 years old and
had higher average education level than the average education of adults of all ages,
when comparing the trends in education level we look at the year from 1991 to 97
(Luo 2003 b). The average years of education increased by about half a year from 1991
to 1997. The overall prevalence of people with no formal education decreased from 22
to 18 percent. Between 91 and 97 the percent of adults with less than primary degree
decreased and the percentages of all other higher levels of education increased. There
is some gender differential in education levels between men and women. Men had
1.7 years more of education than women in 1991, 2 years more in 1993 and 2.1 years
more in 1997. The percent of men in lower education levels decreased (no formal
education by 3%, some primary schooling by 3%, and primary degree by 1%), and
the percent of men in higher education levels increased (for middle school degrees by
3%, high school by 1%, and college and above education by 2%), whereas for women
the increase in education happened in all levels and there was a significant decrease
of percent of women with no formal education.

A person is employed if she is a regular worker, contract worker, or a temporary
worker in state and collective enterprises and three—source invested enterprises.5 The
employment rates of men are also higher than that of women. This period marks
China’s most noteworthy economic growth and expansion. There is a significant
increase in household real productive assets between 1989 and 1993, all values dis-
counted to the 1988 yuan. The gap between rural and urban productive assets is
closing, with urban residents owning less such assets. There is little change in the

amount of household cultivated land.

 

5It also includes those employed by individual private businesses, and the self-employed in various
kinds of household sideline productions, small retail businesses, handicrafts, etc. These can be paid
or unpaid occupations. Farmers engaging in agricultural labor and soldiers are considered as having
a job, but students’ vacation work and housewives’ house work are not considered as being jobs.

109

The community survey section includes information on infrastructure, services,
populations in the village/ neighborhood, percentage of land with poor quality, daily
wages for unskilled farmers and construction workers, percent of work force engaged
in agriculture/ working out of town for more than one month, and hospital and clinic
infrastructure and personnel. In each community, state ration coupon, retail and
free market prices were collected in stores for the most commonly consumed items
of rice, wheat, egg, pork, beef and fish . Since free market prices reflect the value of
each produce better, they are included in the reduced—form analysis with community
characteristics. As is seen in Table 2.1, the prices of the six food groups in 1989 seem to
be too high,6 causing a dip in the generally increasing trends. This is possibly because
the price data were “cleaned” by the CHNS team before being put on the web site.7
In our paper whenever certain foods are not usually consumed in some community the
relative median prices between two similar food groups are used to help impute the
missing values. For example, when the rice price in one urban community is missing
it is imputed by the product of the ratio of provincial urban median prices of rice and
wheat and its own price of wheat. There are only a small number of communities
with such imputed price information.8

The types and sources of households water sources, toilet facilities and sanitation
(excreta) condition around the living areas are aggregated from the household level to
derive the percent of households within each community having certain resources. For

example, in Table 1 the average of percent of households in each community drawing

 

6According to Ge et al. (1992) in May 1991, the price of grain and edible oil on ration was
readjusted by a large margin for the first time since the mid-19603. The price of grain was raised
by 70 percent and the price of edible oil almost doubled. Although to lessen the impact of the rise
in costs on the population’s living standard, the government provided a subsidy to urban workers,
the free market prices in 1991 were expected to be much higher than that in 1989.

7This is possibly done by replacing missing values with the mean of the non-missing data as was
done when the CHNS research team was cleaning the income data. This imputation was not done
for other years (because the original data from other years included the coding for missing values
such as -999 but no such codes exist for 1989 data) and thus resulted in the high 1989 prices.

8For example, in 1991, free market prices of rice in 20 out of 189 communities are replaced this
way, and only one community’s free market prices of pork was missing and replaced this way via
ratio of prices of pork and beef.

110

water from an open well in 1989 is 17% and it decreased to 12% in 1993.

The three consecutive days during which detailed household food consumption
data were collected were randomly allocated from Monday to Sunday and are almost
equally balanced across the 7 days of the week for each sampling unit. Individual
dietary intake for the same 3 consecutive days was surveyed for all children aged 1 to
6 and all adults aged 20 to 45 in 1989, and for all individuals in later years.

The 1991 Food Composition Table for China was utilized to calculate nutrient
values for the dietary data. This FCT represents a significant advance over the earlier
China F CT both for higher quality chemical analyses and for improved techniques
of developing average nutrient values for foods whose nutrient value varies over the
country in a geographic context.

The outcomes of interest are calorie, fat, protein and carbohydrate intakes per
person per day and percent of calorie from fat, protein and carbohydrates. The 1991
Chinese Food Composition Table (F CT) was used to convert all 22 categories of foods
into macronutrients. The number of different foods in the FCT is 636 (for a list of all
foods see Bhandari 1998). The individual 24—hour recall measure of intakes from all
food groups is used without modification for household cooking oils (Table 2.2). The
energy content of fat, protein and carbohydrates is calculated based on the formulae
that one gram of fat yield 9 kcal of energy; one gram of protein and carbohydrates
each yield 4 kcal of energy (Whitney and Rolfes 1996). Compared with results shown
in Zhai et al. (1996) for a ”reference man” our estimate of average daily calorie and
protein intakes in 1991 for all men (2,795 g and 102 g respectively) are actually
higher than the modified measure (2,351 and 70 g) and the household inventory
measure (2,425 and 71 g). However, the fat intake for men is much lower (52 g) than
the household inventory measure (62 g) and a little lower than the modified (54 g)
measure. Compared to the average per capita daily nutrient availability measures

in Wang et al. (1993) the calorie (2,652 g) and fat (54 g) measures are close to our '

111

estimates in 1991, but the protein intake (70 g) is closer to that in Zhai et a1. (1996).

2.3.2 Measurement Errors

The modification of cooking oil and condiments affect the estimates of fat in—
take significantly. Hence our estimates of fat intake serve as a lower bound for the
actual fat intake. The other measurement problems might include the following: (1)
the non-edible part of each food is assumed to be excluded in the recall measure; (2)
wastes in food are not included; and (3) recall bias may not be random. If low-income
households waste less than high-income households then nutrient intakes will be mea—
sured with upward bias and the bias is higher for higher income groups. However,
if higher educated people consume more processed food or have better technology in
preparing food then the bias may go the other way around. If recall bias is systemat-
ically related with gender, age, income or education as in cases that women, younger
people, and/ or persons with higher income or higher education report intakes more
accurately the biases can be upward or downward.

One advantage of using the 24—hour recall measure is that away-from-home con-
sumption is included and the downward bias in the weighing method is avoided. If
people don’t know ingredients of cooked foods eaten away from home, the measure-

ment error will increase.

Measurement errors in the dependent variables can be included as one composite
error terms in the regression analysis. But if any of the above systematic errors
are correlated with explanatory variables the results will be biased. Even when the
measurement errors in the dependent variables are uncorrelated with the regressors
and resulting in unbiased estimates of coefficients, the estimated standard errors will
be larger than the standard errors from models without measurement errors. In

such cases statistically significant relationships are harder to establish. To minimize

112

.88.» $50 :3 5 $353 3830 was am? 3 82.?

wfiwmnoha 89898.28 98 83.8, 8955 8ng was £83850 .832” 558953 aoant: .8835 mama—.2»
so 3835 $035588 £20.83 03 @2522: 88: can. .03.? $3 8 @35586 was Somme o>3o=voa
Rom .cofiefiufifi 5.30:0 Sm BBEEBE go: who? men... H8 vows—88 08 ewe we was.» we 8.6 £83339: $3
5 A33 5 vwod use 32 E mods 6me E SQE meme» 8:: 2..N 888 3233820 3de was 323. 6qu

 

 

 

 

 

 

 

 

 

 

 

 

 

as :3 and assess .333 who who Es BEBEm
So So 8.: 36 83.x :3 3d 8... 63:62:89
as E... :s =63 :80 88cm .333 mod sod a; 8.3 5E
one we... we... swam «as was as 328 6:022
mm... :s 2.... :0 636m 23 :5 8s 83.5 basis
was 2.» a: tom «as ...S :s bases 86m
OS 8.... Es sea as and 85 838:5 oz
:2 a: mm; 3.855 and one So :85
m3 .2: e: 88 so 85 Es: 62a 9.4% 8.3. $8 8....
mm: a2 2: 2 Es” as.” News 2
.32 82 as: .82 $2 8.: seems
5355800 3:339:
om... «as as A25 83
82... Se: 932 €33 $83. 85 use 33 8.8 was Racism
as 8: 2w z 8.0 So so... $360289
32 $2 $2 93% 535 :s as 2... 326m in
we” as.” 9% A25 2.3 as one and 86m 2%.:
2.34 8.8” «mews. A85 sea... was was as was was 8&3 885
SE mama 83 2 was :5 25 .585 686m
82 32 $2 8.5.. 23m 2s :5 3... 8.883— oz
83. one. 2% Assv egg and was as 535
Sea. 3.2m $.38 €33 388... BE sum 3.9. Rd. 8.8 a3.
was 83 :8 z mom... 2% 38 z
82 $2 $2 €85 m2: 3.: $2 2...:
29.683“ 363%.:

 

 

saw mzmo 5 85:88.20 38.5.80 was 286.9% 4.32%.: ”3 same

113

.< 86:25.. 8m .82.. 23 .8 9.2.2.6.. $8.888 9.... 8.28588... :23... £8 :08 55.? 9.285530 .0 .38....

m5 .0...“ ARE—mam 2...... 3.2.5.. .88on vs... 30.... .22...wa 2.222.. 5...... v.8 953m .258 E... was: .8vanmfiw... £800.23...
.258 .82... 6.... Saafiwnsm 2:... £822.. .388.88=w2 823303.. £82 .33.. £562 56¢ .22. .3260... $0 we... wwm
.8260... 82.2. 8.58.9.8 .2828 .$md.8>mp 2.2.02... 5.... 8.83.8 we... 3?... 2.9 6.8.30... 8... 8.8.... no... =< A905
2an. :2889800 woo... 8:30 :5. on. no woman 2.... 8.8530380 v.8 £30.... £2 S32... .0 828...... 3...“. 2;. .302

 

 

 

 

 

 

 

 

 

 

8.8 8.8.. 8.8.. 8.8 8.8 8.8 8.8 8.8 8.... 8.88 8.88 8...... +8....8E89
8...... 888 8.8.. 8.8 ...8 2.8 2.8 8.8 8...... 8.88 8.88 :88 82.8 .8...
8.8... 8.8.. 8.8.. 8.8 8.8 8.8 8.... 8.... 8.8.. 8.88 8.8... 8.88 89.8 23...).
8.8.. 8.8.. 8.8.. ...,8 8.8 8.8 8.... 8.... 8...... 8.88 8.88 8.28 8.8.. .88....
8.8.. 8.8.. 8.8.. 8.8 8.8 8...... ...,... 8...... 2.8 8.88 8.8.8 8.8... .58.... 8.8
8.8.. 8...... 8.8.. 8.8 8.8 8.8 8.8 .88 8.8 8.88 8.....8 8.88 858...... oz
8.8.. 8.8 8.8 ...R 8.... 8.8 8...... 8.8. +8 8...
8.88 8.... 8.8.. 8.8 8.8 8.8 8...... 8.... 8.8 8.88 8.88 8...... 88.1.3
8...... 8.8.. 8.8.. 8.8 8.8 8.8 8.... 8.... 8.... 8.8... 8.88 8.8... 88 8...
8...... 8...... 8.8.. 8.8 8.8 8.8 8.... 8.... 8.8 8.88 2.88 8.8.8 .83.
8.8... 3.8.... .58 8.8 8.... 8.8 8.... 8.8 8.... 8...... 8...... 8...... 8.5
8.8.. 8.8.. 8.8.. 8.8 8.8 8.8 8...... 8.... 8.... 8.8.8 8.8.8 8...... :82...
8... 8... 8... 8... 8... 8... 8... 8... 8... 8... 8.... 8... 2.8....
.8 8.8.8.830 .8. 538.. .8. 8.. :8... 28......
8.8 8.8... 8.8... 8.8. 8.... 8.8. 8.8 ...,... 8.8 ...88 8.88 8.88 $8.638...
8.8.. 8...... 8...... ...8. 8...... 8.8. 8.8 8.8 8.8 8.88 8.88 8.88 89.8 8...
8.8.. 8.8.. 8.8.. 8...... 8...... 8.8. 8.8 8.8 8.... 8.8.... 8......8 8.88 .888 22...).
8.8.. 2.8.. 8.8.. 8.8 8.8. 8.8 3.... 8.8 8...... :88 8.2.8 8.8.8 8.8.. .38....
8.8.. 8...... 8.88 8.8 8...... 8.8. 8.8 8.... 88.. 8.88 8.88 8.88 .58.... 28.0.
8...... 2.8.. 8.8.. 8.8 8.8 8.8. 8.... 8.... 8.... 8.88 8.8.8 8.8.8 838...... oz
888 8.8... 8.8 8.8 8.... 8.... 3.88 8.88 +8 8..
......... 3...... 8.8. 8.8. 8.8. 3.8 8.8 8.8 8.... 8.88 8.88 2.88 8.... 8..
8.8.. 2.8.. 8.8.. 8...... 8.8. 8.8. 8.8 :8 8.5. 8.88 88.8 8.88 88 8..
8...... 8.8.. 8...... 8.8 8.8. 8...... 8.... 8.... 8.... 8.88 8.88 88.8 .23.
8.88 8.88 8...... 8.8. 8.8. 8.8. ...,8 8.8 8.8 8.88 888 8.88 5......
8...... 8.8.. 8.8.. .....8. 8.8. 8...... 8.8 8.... 8.8 8.88 8.88 8.88 8.96
8... 8... 8... 8... 8... 8... 8... 8... 8... 8... 8... 8... 28,.
.8. 888.3880 .8. 532.. a. a... :8... 28......

 

 

mm-mw .msz E 2...me v.8 3.38% ”8.2.8... m3...PEOQEO v.8 E395 Jam 2.280 bEQ m®8m>< ”NN 2an

114

random measurement errors, daily nutrient intakes are averages over three consecutive
days whenever available.

Random measurement errors in the explanatory variables cause the estimates
to be biased toward zero. Incomes in developing countries have been found to be
particularly error-ridden. That is one of the reasons we prefer not to use income as a
main regressor. Instead we use real productive assets and land farmed in calibrating

household resources.9

2.3.3 Patterns and Trends

Tables 2.2 and 2.3 report the general patterns and trends of average outcomes.
For the number of observations, median and standard errors of the mean in each
stratum see Appendix E. In Table 2.2 there is a decreasing trend in caloric intake for
men in urban and rural areas (98 kcal and 137 kcal decrease respectively) and women
(66 kcal in urban and 152 kcal decrease in rural areas) from 1989 to 1993. There is
an increase in calories in 1991 for more educated men and women and for younger
adults less than 40 years of age, and it is followed by a decrease across all subgroups
in 1993.

For men there is a general increasing trend in fat intake (in grams) except for a
relatively small dip for the non-educated and highest educated groups. For women
the increase is more pronounced, with the highly educated (high school diploma or
better) seeing the biggest increase. These women are also those with largest increases
in BMI (1). As some studies show a doubling of edible oil in this period (Guo, Popkin,
Mroz and Zhai 1999), the absolute level of fat intakes should be higher than those

reported in table 2. Our estimates provide a lower bound for the level and trends of

 

9Productive assets may be estimated with errors too. The items included are tricycles, motor-
cycles, tractors or walking tractors, irrigation equipment, power thrashers and water pumps. Real
productive assets are discounted to 1988 value. Values are self-reported purchasing values in 1989
and current worths in all other years. There are items excluded from this calculation because they
are not surveyed in all waves of the CHNS.

115

fat intake. Overall protein intakes have been stable, but there is a decreasing trend
among the least educated and an increasing trend among the highly educated. It is
evident that education is associated with better quality of nutrients but whether it
is through increased human capital or through shifted tastes is unknown. There is
cross the board decrease in carbohydrate intake. Simple carbohydrates are proxies
for lower quality diet.

In Table 2.3, columns 1 to 3, there is an increase in the percent of calories from fat
for men and women in all subgroups. Column 4 to 6 show increasing trends in percent
of calories from protein except for least educated men. There is a steady decrease in
percent calories from carbohydrates in all subpopulation strata. Alongside economic
development we can see decreased carbohydrates and increased fat intakes as signs
of closing the gap in eating patterns between China and developed countries. The
departure from the traditional low-fat diet in China has also been viewed as a sign of
deteriorating consumption pattern (Popkin 1999, Guo, Popkin, Mroz and Zhai 1999).

Urban residents consume less energy and carbohydrates, but more fat in all years
(Table 2.2). Calorie and carbohydrate intakes decline as education levels go up, and
for fat the reverse is generally true. There is no significant difference in protein intake
between regions. There is a increase in protein intakes among higher educated people
if the education level is less finely cut at the high end of the distribution. As people
age there is an decrease in all nutrient intakes. The relationships between percent
calorie intake from fat, protein and carbohydrates and residence, aging and education
levels are more stable the levels of intakes. Although fat intake decreases as people
grow old the percent of calorie from fat increases for men. The percents of calorie
from fat and protein for urban residents and higher educated men and women are

higher than those of rural residents and less educated.

116

.< 2:288? com :82:
.2: mo 255258 “28988 ~28 aossntummn 28:58 .mcosgvmno .«o 85:52 5m 628205 08 2202 woo“
=< .9828 mo .82 v 222» :80 885830330 98 5303 mo 88% 25 3385 mo :82 a 222% 28 mo 82%
«so 85 $38.28 2: :o 893 83.2528 2 8868:0380 98 530.5 .22 .«o 8:328 385 BB 8qu

 

 

 

 

 

 

 

 

 

 

8.8 2.8 8.8 2.: 8.: 8.2 8.8 8.8 8.2 88:882.
88 8.8 2.8 8.2 8.2 8.2 8.8 2.8 8.: .888 82
8.8 8.8 8.8 8.2 8.2 8.2 8.2 2.: 2.2 828 8822
8.8 8.8 2.8 8.2 8.2 8.2 8.: 8.2 8.2 88a 585
2.8 8.8 8.2 8.2 8.2 8.2 8.2 8.2 8.: 585 888
8.8 2.8 8.8 2.2 8.2 8.2 8.2 8.2 8.2 88882 02
2.8 8.8 8.2 8.2 8.: 8.2 +8 22
8.8 2.8 8.2 2.2 8.2 2.2 8.2 8.2 8.2 8.8 8<
8.8 8.8 8.8 8.2 8.2 8.2 2.: 2.2 8.2 88 8<
8.8 8.8 8.8 8.2 8.2 8.2 2.2 8.2 2.2 8:8
:8 8.8 8.8 8.2 8.2 8.2 8.8 8.8 8.2 82:
8.8 8.8 8.8 8.2 8.2 8.2 8.: 8.2 8.2 885
82 82 82 82 82 82 82 82 82 8888
88530380 88 588m X 8m 08
8.8 8.8 8.8 8.2 8.2 8.2 2.8 8.8 8.2 888082.
8.8 8.8 8.8 8.2 8.2 8.2 8.2 8.2 8.2 888 82
8.8 8.8 8.8 8.2 2.2 8.2 8.2 8.2 8.2 8928 8822
2.8 8.8 8.8 8.2 8.2 8.2 8.2 8.2 8.2 88a 585
8.8 8.8 2.8 8.2 2.2 8.2 8.2 8.2 3.2 .585 888
8.8 8.8 8.2 8.2 8.2 8.2 2.2 8.2 8.: 8888* oz
8.8 2.8 2.2 8.2 8.2 8.: +8 22.
8.8 8.8 8.8 8.2 8.2 8.2 8.: 8.2 8.2 8.8 8<
8.8 8.8 8.8 8.2 8.2 8.2 8.: 8.2 8.2 88 8<
8.8 8.8 8.8 2.2 8.2 8.2 8.2 8.2 8.2 852
8.8 8.8 8.8 8.2 8.2 8.2 8.8 8.8 8.2 :25
8.8 8.8 8.8 8.2 8.2 8.2 8.: 2.2 8.2 885
82 82 82 82 82 82 82 82 82 282
8adufi~n£0330 0A. Gmquunm o5 warm 05

 

 

mméw msz E 22229 was 28325 8386820380 98 53on fish Sod 2.880 m0 280qu mwduo>< ”9m 2an.

117

2.3.4 Estimation Methods

As was discussed in Chapter 1, the estimation of equation (2.1) for the reduced-
form macronutrient intakes uses OLS on data pooled over three years. To account for
heteroscedasticity and correlation of errors for each individual over time individual-
level robust standard errors are used in all regressions. In all regressions individual-
year age dummies and five-year cohort dummies are included. In a separate paper (Luo
(2003a)) we consider the identification problem in estimating age, cohort and year
effects in a model. We chose the five—year cohort dummies so that the perfect linear
dependency of the three variables no longer exists and the system can be identified.
Surely there are many other identification strategies as noted in Luo (2003a) but we
chose five-year cohorts around the early 19603’ famine and the cultural revolution
period as being one meaningful strategy. Individual-year age dummies are preferred
to using age as a continuous variable because the dummies capture the nonlinear

effects of age better.

Modelling Levels

Two sets of regressors are of interest at the individual and household levels and
at community levels. First, the basic specification focuses on the effects of indi-
vidual education and household covariates - productive assets and land, controlling
for community level factors by community-year interaction dummies. Education is
categorized as no formal education (the reference group), some primary education,
primary degree, middle school degree, high school degree, and technical institute or
college level or above. We control household resources using log of real productive
assets discounted to 1988 value as indicators for long-run resources. Splines around
the median of the positive values of real productive assets and the amount of land
cultivated (mu) are included in all regressions. The coefficients for the lower and

upper spline of productive assets represent the level effects, not the marginal effects.

118

This basic model is estimated for the whole sample, urban and rural areas sep-
arately, and for each age group (20—39, 40-59 and 60+) for men and women calorie
(Table 2.4), fat (Table 2.5), protein intake (Table 2.6) and percent calorie from fat,
protein and carbohydrates (Tables 2.7 to 2.9). These stratifications enable us to
find differences between different groups of people. The rural-urban stratification is
consistent with the presumption that rural households produce and consume at least
part of their produce and urban households are primarily consumers. The community
dummies capture the effects of unobserved time-invariant community level informa-
tion. For factors such as prices, time-invariance seems unlikely. Using community-
year interaction dummies allow the effects to differ by year, as prices, future price
expectations and other community factors change over time.

Secondly, we add to the basic model community characteristics such as prices,
water, toilet and sanitation conditions replacing community-time interaction dum-
mies. The augmented model studies the joint effects of individual, household and
community variables on all outcomes (Tables 2.10 to 2.12). In this specification we
use community dummies instead of community-year interaction dummies to capture
community level unobserved heterogeneity. It is important to control for such hetero—
geneity because infrastructure, such as the quality of water and sanitation conditions,
is likely to depend on community resources and may be non—randomly placed accord-

ing to unobserved community attributes (Pitt et al. 1993).

Modelling Changes

Estimating equation (2.1) with pooled OLS implies the effects of all socioeco—
nomic determinants are fixed over time, whereas the theory tells us we should, for
example, include all past and future prices in the model to allow different effects of
these variables. The constant effect assumption can be easily violated during the

time of rapid transition in China. Finding out the socioeconomic determinants of the

119

changes of people’s food consumption has important policy implication. Exploring
the longitudinal nature of the data set, for all outcomes, the changes between 1989
and 1991 and the changes between 1991 and 1993 are pooled together and the OLS
method was used in the basic and augmented models again to find out the effects of
the changes in prices and other community characteristics on the changes in calorie,
fat and protein intakes and the changes in percent of calories from fat, protein and
carbohydrates.

When the outcomes are changes two sets of regressors are again of interest. First,
the basic model, includes education categories, real productive assets and land culti-
vated in base year and community dummies. Secondly, the augmented model, adds
changes in food prices and community water and sanitation characteristics as explana-
tory variables in addition to the individual and household characteristics in the basic
model. Not only do the baseline levels of prices and community characteristics affect
the changes of nutrient intakes, but the changes in prices and other local factors may
also have an impact. Conceivably people respond to food price changes by substitut-
ing cheaper foods when the prices of substitution foods decrease or complementing

and consuming more food when the prices of complement foods decrease.

Modelling Production Emction

As outlined in Chapter 1 adequate to the application of BMI, the health status
at the beginning of period t + 1 is assumed to be influenced by health status at
t, health inputs in period t, m“, such as the balance between energy expenditure
(basic metabolism, working or exercising) and intakes, medical care, or illness spell,
which does not bring utility directly. More leisure time may bring feelings of well-
being and sound state of mind, which can be positive influence of one’s physique.

Individual characteristics, 9a, such as age“), gender and community or environmental

 

10The solution to the maximization problem is a time-invariant policy function that determine
the control variable (xitju, m“) and the state variable hu- It is conceivable as a person grows old

120

characteristics, 2a, may also have a direct impact on health outcomes:

hit-Fl : fi(hit1mitilitagitvthivit) (2'2)

where v“ are unobservable individual, household, and community factors that affect
member i’s health. It is assumed that Bfi/Bh“ > 0, and af./am,-, > 0. It is necessary
that Bfi/Bhit < 1.Equation (2.2) is not a general form of health production function
in that it assumes that hit is a sufficient statistic that summarizes the effect of all
past inputs and choices in periods 1, ..., t — 1 and there is no direct lagged effects from
them on hau- A more general health production function would allow all past inputs
to be included.

To estimate a health production function several foreseeable problems are con-
sidered (Behrman and Deolalikar 1988, Strauss and Thomas 1995, Cebu-Study-Team
1992, Wolpin 1997, Sickles and Taubman 1997). First the choice variables of nutrient
intake could be related to the unobserved disturbance term in the health produc—
tion. The disturbance may include unanticipated health shocks, omitted inputs and
individual heterogeneity. The correlation between nutrient intake and the error term
suggests the OLS estimates of the equation (2.2) are biased. We measured h by body
mass index in Chapter 1 and estimated the reduced-form health demand function.
Here simple health production functions, as measured by logarithm of weight in 1993
and logarithm of BMI in 1993, are estimated using lagged weight, height (or lagged
BMI), physical activity levels, and current food intakes as inputs with instrumental
variables. The identifying instruments used are education, real productive assets,
land cultivated and community characteristics including prices, water and sanitation
conditions in 1991.

The biological process of producing body weight and BMI is necessarily a balance

 

the effect of inputs on health (body size) varies. To incorporate such effect we can assume a health
production function that varias with age at time t, a“:

ha = fi(hit—l, mu, (men. law“) + git(ait)mit

121

between energy inflows and energy expenditure by metabolic processes and physical
activities aimed at maintaining good health and daily living (Srinivasan 1992). In
CHNS a set of energy expenditure questions related to occupations that have been
asked as part of the nutrition data collection since 1989.11 I categorize the energy
expenditure into two groups with light and very light activity as reference comparing
to moderate, heavy and very heavy activity levels, and instrument them with the
above set of IVs.

There are other inputs such as mu, 1,, that are omitted from the simple anal-
ysis. Omitted variable bias in OLS can be upward or downward depending on the
correlations between intakes and the omitted variables if we assume Bfi/Bmu > 0
and Bfi/Blgt > 0. When genetic endowments of an individual are excluded and are
positively related with nutrient intake for better-endowed persons are more likely to
utilize more nutrients, the estimated effects of nutrient intake on health are likely to
be overstated. The nutrient intakes and physical activities are quite possibly mea-
sured with error. The fat intake in this paper serves as lower bound of the true value
because we do not modify the oil content of cooked meals. The random measurement
error leads to attenuation bias in OLS regressions. All determinants in the reduced-
form demand function are used as identifying instruments in the production function
analysis. They serve to correct both the omitted variable biases and the attenuation

bias, although to which end they could do a better job is unknown.

 

11Very light physical activity refers to working in a sitting position, e.g., accountant, office worker,
electrical appliances repairer, watch repairer. Light physical activity normally refers to working in
a standing position, e.g., shop assistant, laboratory technician, teacher. Moderate physical activity,
e.g., student, driver, electrician, metal worker, salesman. Heavy physical activity, e.g., farmer,
dancer, steel worker, athlete. Very heavy physical activity, e.g., loader, logger, miner, stonecutter.

122

2.4 Results: Determinants of Levels

2.4.1 Basic Models

The basic results for total calories, fat and protein intakes, and the percent of
calories from fat, from protein and from carbohydrates are summarized in Tables 2.4
to 2.9. All analyses are done separately for men and women in the whole sample,
from rural and urban areas and for different age groups. Community-year interactions,
individual-year age dummies and five-year cohort dummies are also included.

Education does not have significant impacts on calorie intakes but does affect
the percent of calories from fat, from protein and from carbohydrates differently in
different region and at different ages. The effect of productive assets is nonlinear and
has an inverted U-shape for male calorie, fat and protein intakes. In rural areas the
effect of productive assets is stronger than that in the urban areas. At different ages

the effects of education and assets are different.

Total Calories (Kcal)

Results for daily calorie intake (kcal) of men and women in different strata on
the basic specification are shown in Table 2.4. The effects of education for men is
not significant in any stratum; while the effect of female education is nonlinear and
significant for the whole sample (joint F-test has p-value 0.03) and for the elderly
(age 60 and above, p—value=0.02). Overall, as female education increases the level of
calorie intakes increases first and then decreases; however for the elderly the reverse
is true. For the elderly those with the high school education consume 897 more Kcal
of calories than those without any education.

The effect of productive assets is not significant for women but is significant
for rural and younger men at 0.01 level and for the elderly at 0.1 level. Below the

median productive assets are positively correlated with calorie intake and above the

123

median the correlation is negative for rural and younger men. For the elderly the
relationship is reversed. This may be because for the young and rural men at lower
level of productive assets they complement labor input and hence increase calorie
intake whereas at higher level of productive assets they substitute for labor inputs
and hence decrease calorie demand. For the elderly the joint effect of the splines is
significant at 10% level and only the higher level spline is individually significant at
10% level. As a person gets older he may need more productive assets to complement
given labor input and hence result in higher demand for calorie.

The effect of land is only positively associated with calorie intakes for elderly
females. Community dummies are all jointly significant. The age profiles between men
and women for the urban and rural samples are different (p-value=0.004 for urban
sample and 0.008 for rural sample). Testing for differences in parameters between
urban and rural men and women only turns to be significant for age dummies in

regressions for men.

Fat Intakes (g)

Results for daily fat intakes (g) are summarized in Table 2.5. For men the effect
of education dummy coefficients are individually significant but not jointly so. For
women the effect of education is strongest in the whole and urban samples. For all
sub-samples the highest educated men and women always have the highest fat intake
(except for prime aged men and elderly women). Higher educated people are more
likely to be engaged in more sedentary jobs and require less calories. However they
consume more energy-rich fatty foods. This reflects the role of education as a proxy
for income since consumption of animal products and income levels are positively
associated as shown in Du et al. (2002) and Guo, Popkin and Zhai (1999).

The relationship between productive assets and fat intakes for men is the same

as the relationships in the calorie intake regressions. For women in the overall sample

124

the effect of productive assets is positive and for women in the rural and younger
aged sample the effects are the opposites than that found in men. Land farmed
is negatively correlated with fat intake for rural and younger men. Growing grain
products and raising animals are substitute uses of land. When more land is farmed
there is less access to animal products and hence less fat intakes. The difference in
parameters between men and women is significant for productive assets in rural and
young cohort and the effects of community dummies differ between men and women
when the sample is stratified by age. This may indicates clustering of similar-age
people in some community. There is no significant difference between urban and

rural men and women in education and productive assets effects.

Protein Intakes (g)

Results for daily protein intake (g) in Table 2.6 are similar to the results for fat
intakes. Now, the effect of land on female protein intake is negative for the rural
and prime aged samples. The effects of education, productive assets, age and cohort
dummies between urban and rural residents are not statistically different. This is
consistent with what we find in the descriptive measures in previous section. The
positive female education effects for the overall sample and the urban sample are
statistically significant at 0.05 and 0.1 levels respectively. This is consistent with what
Behrman and Wolfe (1989) finds for women in Nicaragua using sibling data and fixed-
and random-effect models in comparison with the standard estimates. The authors
argue for the support of the effect of women’s schooling in increasing nutrient intakes

rather than reflecting the effects of unobserved childhood background variables.

Percent Calories From Fat

The effects of education on the percent calories from fat (Table 2.7) is significantly

positive for men and women in all strata, except for elderly men. Recall only some

125

of these effects are significant in the model for the level of fat intakes in Table 5. If
the body has an excess of energy-yielding nutrients (fat, protein and carbohydrates),
it rearranges them into carbohydrates and fat storage compounds to be drawn upon
between meals and overnight. Health recommendations urge people to limit fat intake
to 30 percent of total calories required. Energy yielding nutrients also provide the raw
materials for building the body’s tissues and regulating its many activities. When
the percent energy from fat is too high (not just the level of fat intakes) and energy
expenditure is low, the body will gain weight (Whitney and Rolfes 1996). Higher
educated people tend to have higher percent calorie from fat. The magnitude of
the effect of education is different for different groups of people. Compared with the
uneducated the effect of high school education is the strongest for the younger men
and for the elderly women. However in Chapter 1 we find that in the overall pooled
sample for women having some primary education is associated with 0.23 unit higher
BMIs than those with no formal education at 0.1 significant level, but the impact
rises significantly when one finishes primary schooling (0.47 unit). To put things
into perspective, for a 5.5 foot woman a 0.47 unit increase in BMI is equivalent to
a 3—pound increase in weight assuming fixed height. Completing senior high school,
technical school or college is associated with a lower BMI than having no education
(-0.02 and -O.7) or lower education levels. On average a 5.5 foot woman with a college
equivalent degree weighs 4 pounds lighter than an uneducated counterpart.

The positive and increasing effects of productive assets on the percent of calories
from fat is significant for women in all samples but the elderly sample, and for men in
all samples but the urban sample. Again, land effects are generally negative except
for elderly females. The difference in cohort effects between urban and rural men
is significant, maybe indicating unobserved childhood background measures because
the rural-urban differential in availability of foods was bigger in earlier times and the

famine in the 19603 hit rural areas harder than the urban areas.

126

Percent Calories Horn Protein

Protein is a vital structural substance in all body cells. Meat is a good source
of protein, as are milk, eggs, legumes, and many grains and vegetables. When the
percent of calories from proteins is high and the total protein intakes are low, more
protein is utilized to create energy and less is metabolized as protein to provide the
vital supply to build cells.

The effects of education and productive assets on the percent of calories from
protein (Table 2.8) is more pronounced than those in the level of protein intake.
This suggests the percent calories from protein measures somewhat different aspect
of nutrient intakes than the level of protein intakes alone. Or the level measure
may be too skewed or with too much measurement errors that result in insignificant
parameters.

Beside the significant education effects in all samples except for the elderly female,
of particular interest is the difference between the effects of education in urban and
rural men (p-value=0.052). The highest educated men in urban areas consume 1.26
percent more calorie from fat than those least educated, whereas men in the same
education category in the rural areas consume about 0.4 percent more calorie from
protein than those without any formal education. The male education effects among
different age groups are stronger for the elderly and the female education effects are
stronger for the younger ones.

For both men and women higher productive assets are associated with higher
percent of calories from protein. Recall for men the productive assets are negatively
associated with protein intakes and for women the association is positive. Economi-

cally better—off women may choose higher quality foods.

127

Percent Calories From Carbohydrates

Carbohydrates are converted by the body into glucose for immediate energy and
into glycogen for reserve energy. Together glucose and glycogen provide about half of
all the energy human nerves, muscles and other body tissues use. Higher education is
related with lower carbohydrate intake for both men and women in all strata (Table
2.9). It could be because higher educated people are engaged in less labor intensive
work that require less calories, or because they have more energy from fat and protein
due to income effects.

The land effects are positive for exactly the opposite reason for negative land
effect on fat intake. The differences between urban and rural population is significant

for education and cohort effects among men.

2.4.2 Augmented Models

In addition to the above education and household resource effects we are in-
terested in the effects of different food prices and community water and sanitation
conditions on the demand for nutrients. For total calorie, fat and protein intakes we
take the logarithm of the original level and the prices used are free market prices and
are also in logarithm. For the percents of calories from fat, protein and carbohydrates
regressions, the outcomes and prices are in original levels. Tables 2.10 to 2.12 sum-
marize these results for overall, urban and rural samples. The food prices chosen in
this paper are prices of rice, pork, eggs and edible oils. As is seen in Table 2.1, there
are other food prices available in the survey. However, they can not be all included
due to high correlations between them. For example, the price of rice and the price of
wheat, the price of pork and the price of beef are highly correlated. When variables
with high correlations are included in the same model the estimates are inaccurate.
Community dummies, individual-year age dummies and five—year cohort dummies are

also included.

128

All Outcomes

The positive effects of urban residency on calories for men and women are consis-
tent with the findings that urban residents are more likely to consume and consume
more of selected food groups and are more likely to be overweight (Popkin 1999).

The elasticities of the price of rice on total calorie, fat and protein intakes, and
the percents of calories from fat and protein are negative in all sample strata; and the
elasticity of the price of rice is positive for the percent of calories from carbohydrates.
These results suggest that with an increase in rice price people consume less energy-
rich fat and protein and substitute for cheaper foods with high carbohydrates.

The effects of the price of pork are the exact opposite of the effects of the price
of rice. The effects of the prices of eggs are in the same directions as in the effects of
the price of rice. The effects of the price of edible oils are similar except that it has
a positive effect on the percent of calories from protein.

Generally there are two effects of increased food prices. For net producers of
foods increases in prices drive up income and hence demand for more or better food
consumption. For net buyers of foods increases in prices generally tighten the budget
set and reduce the demand for normal goods. However, the fact that the coefficients
in urban (Table 2.11) and rural (Table 2.12) areas are close to each other indicates
that possibly no rural households are net producers or that income effects are small.
In addition since the outcomes are not specific foods consumed but macronutrients
from all foods the two effects of prices are more difficult to disentangle.

The types and sources of households water sources, toilet facility and sanitation
(excreta) condition around the living areas are aggregated from the household level
to derive the percent of households within each community having certain resources.
Compared with communities with underground water sources (Table 2.10) a 1% in-
crease in percentage of households in a community with rain, snow or river as water

sources is associated with significant increase in calorie intake, protein and percents

129

of calories from protein and a decrease in percent calorie from carbohydrates for both
men and women. Higher percentages of household using water from snow, rain or
river indicates a poorer neighborhood. The effect on calories makes sense. There is
evidently some other mechanism through which the water condition of the community
is positively related with protein intake

Compared with individuals living in clean environments men from communities
with a higher percentage of households with little excreta around the living quarters
consume more calories, fat and protein (for women only the protein effect is strong
and positive). Men whose living environment has some excreta consume less calories,
fat, and the percent of calories from fat, but have more percents of calories from
protein and carbohydrates. The effects of excreta are stronger in urban areas than
those in rural areas. These results suggest there may be some income effects proxied
by these environmental factors.

The effects of individual and household characteristics on the percent of calories
from fat, protein and carbohydrates in all strata are strong and exhibit the same

properties as in the basic regressions.

130

8:538 2an

 

 

 

 

 

88¢ 88o 885 88.9 wand 88¢ 8.88% 382888
885 Some «£3 3% ES 23o £883 :28
as: 885 83o $85 9.85 886 usage as.
3:3 3de 58d 8:; 33s mead sea...
28¢ Emma 3.8.0 at”; 9:; $35 858:3
5803 ES :08 cooBaon $58... .8.“ moigd
886 885 88s 88d 886 886 amass. £5888
$35 £25 886 some 3.de some @883 :28
:23 ”was 835 «Sod 88s 885 «eggs s2
«Rod m8; 885 ~83 ”mono ”Sod 33$
29:. SE6 was 8mg «83 9.02 858:8
ohms 0... 3:3 mucflofiog wcsmfi Ho“ egg-m
83 was 5.3 $3 $3 and Baas-m
a: 28 $9. $3 was 88 massage no .02
we.» mo> 8% 8% 8% wo> .3526 use»... 5:58:80
38.: 88.3 $3; 833 553 85.3
£3 23 8mm $2. 83... $3 33
.333 Less: 138.8 1.825 33.8 3.33.3
N33 no.2- v2.2- 83m- 3.2. $2:- m 38.8 Ea as mo.“
68.5 3.25.3 1:88 133.5 2.3.8 1.2.33
33- eaa.mm 83m 398 :5: 33m H 383 Ba 1.2 won
38.8: 6%.me 83.8: saga $2.5 $8.3
83: was”? 25.3- o3;- Swsn- 5.3- $330389
$33: 55$ 88.3 .333 Geese 8.8.3
33 @88- mmmsm- 3.8- Esm- wgec- 628 as:
£5.me Ema-3 33.3 33.3 538 83.5
8.3.? «3.3. 8o.~m- noes.- mflsm- 5.9.- 623. 22::
games Ease sage $8.3 38.9: 333
cows.- wgé- 32.8 Eva- RBA 3.3- .585
853 A838 68.3 63.3 38.3: ages
83- mmaam- $35 SE- 85..- 23 838%.. mafia oaom
+8 ”.3 one. mm... 3.8 a? 38m :85 =§>o 8338.»
232

 

anew Boa 3 89a 8 .3 .3 mzmo s :85 mass ssao has a.“ made

131

.—Q>®— no.0 ad “5.. U5 ~0>®~ H.O

as mos-89am? 838388 88065 ... .885883 E 88 988 @8288 «$58 E>2é88m
.8gsv flosoo H8382.“ 98 83:86 ems 88 £568ng on... 5 89:95 83.. 8qu

 

 

 

 

 

 

83.: 2.88 8888 880
”an; $85 8888 mm...
885 82.0 882
888 883 888 82 858:8
888 Rh: «:8 awn-S :08?qu $588 Bu 8287mm
886 888 88.0 888 888 888 8888 $8880
888 885 $8.: 88.: 8:8 88.: 8888 880
88.0 353 «.de :88 883 $88 8888 8...
8:8 888 Ego 3.8 82.8 888 882
885 $85 888 5.3 885 £88 858:8
ohms 3 12.8 886508 9.58... .8“ mangd
88¢ 82 :2 82 22 2.3 8888
8: 88 88. 38 8:” e8 88880 8 .oz
8% 8% 8% 8% 8% 8% .3526 “8%... 5338800
$28.3 Coma 38.3 $8.: Agasa 2.8.:
8:: 83 N86 23 $8 88 83
Amt-.8 33.8 :33 82.3 38.8 38$
83.- 88 83. 83 am.” can.” m 888 88 88 84
38.5 32.8 88.8 38.3 :88 68.3
3....” ms...- 83 2w.” 2.3.- 88 H 888 88 :8 meg
$8.8: :8.me 83.3 2.8..ch $8.8 32.3
838 3.3” 85.8- 88.8- :3:- mmw8- 888389
£38.88 $8.53 833 2.8.3 88.3 138.3
8.8 88 @88- 88- 88.8. 888 888 88
:88: 1224.3 $.33 58.3 $8.3 8284.3
532 888. Bad.- 888- 8Q:- 88- 388 283
838 833 88.3 82.3 8:83 53$
@88- 83”- mmwé 83 9.3”- 8: 888
38.5 $883 88.5 833 88.3 ESQ
”8.8- 83”. 83 ”5.8. 83- 82.7 8:838 888 88
+8 8:. $8 8.... 8.8 8... 38m :85 885 83855
2&th

 

8.8on § 28g.

132

825:8 @388

 

 

 

 

 

88.8 88.8 888.8 28.8 88888 888.8 88888 88888
838.8 838 838.8 8388 883.8 88.8 888.8 :28
838 38.8 $2.8 $88 22.8 88.8 888% om<
388.8 28.8 88.8 88.8 :38 £28 883..
$88 3.8.8 «3.8.8 38.8 83.8 53 888:8
8803 98 ~88 :09.»qu $58... Mom 83.87%
88.8 888 888.8 88.8 88.8 88.8 888% 88888
$88 38.8 88.8 888.8 888 838.8 888.8 :28
388.8 22.8 388.8 3.88 838.8 82.8 888% 83
:83 888.8 888.8 888.8 $3.8 $8.8 88¢
83.8 3.38 88.8 88.8 88.8 88.8 8:888
ohms 3 153 856508 958... 3.3 “BETA
38.8 83.8 88.8 83 83 888.8 8888-:
a: $8 88 $3.. 388 .888 88888 8 .oz
mm> 8% mm.» mm? wo> wo> >583. 88%.." 3255800
88.3 88.3 182.3 1.82.3 83.3 188.3
88.8 88.8- 5.88. 88.8. 88.8 888.8- 83
88.3 888.3 888.3 888.3 83.3 888.3
83.8 38.8 88.8- 88.8- 88.8 83.8- m 88.8 88 88 83
88.3 :33 1.88.3 ......:8.3 88.3 188.3
33- 83: $3 33 38.8- 23 H 888 Ba 88 83
88.3 888.3 1.82.3 £883 883 1.88.3
88.8 88.8 82.2 88.: 83 88.8 $888888
888.3 88.3 188.3 388.3 8.88.3 1.82.3
88.” £3 83.2 83 83 83 828 .85
88.3 88.3 $833 1:23 88.3 183.3
2:8 88.8 8.8.8 3.3 a: 83 828 8822
28.3 88.3 383.3 88.3 88.3 88.:
~88 Em.” E: 3.3 88.8 83 8888
83.3 888.3 $88.3 88.3 83.3 83.3
83. 38.8 23 SS :88- :3 8288.8 888E 888
+8 83 388 83 838 83 88m 885 8826 83885
282 .

 

 

8:88 888 3. 898 m8. .8 .88 mzmo 8 883 8888 8.8 .88 ”on 288

133

._v>v— m0.0 ad ** U5 ~0>®~ H.O

an 8889.:me 185838 88863 ... .8858889 5 08 Echo c.8988 amass 357.88?"
663358 fiasco Hammgga was $8596 awn v.8 mcofimmuwm: m5 5 88305 8;. 5qu

 

 

 

 

 

 

8.8.8 888 888% 8280
288 88.8 888% 83
82.8 888 88.2
888 8803 88.8 52 8:883
80.8 3:: fig Edam: smogmn wcmumvu no.“ 833.7%
88.8 888.8 888.8 888 888.8 888.8 888.8 888888
88.8 88.8 8888 8:8 888 88.8 888% 8880
888 88.8 82.8 88.8 883 88.8 888% 83
28.8 88.8 88.8 88.8 88.8 88.8 88%
888.8 888 888 28.8 8:8 888 8:8:me
OuwN 0a 353 maQEOEvOo wcmawva no.“ @3ng
88.8 888 88.8 88.8 83 88.8 88:88
8: 88 88 88 88 88 8288880 8 .oz
8% wm> 3% mo> mm; mew 888:8 Bo»... 3158800
188.8 82.8 8.2.3 828 88.8 82.8
88.8 :3. 83- 88- 88.8- 83- 83
88.8 88.8 1.88.8 8.88.8 88.8 188.8
~88- ~88 888 ~88 88.8 88 m 888 88 88 83
88.8 88.8 88.8 88.8 88.8 88.8
83 888- N88. 888- 33 8.88 H 888 88 8.8 88
188.83 88.3 188.2 188.3 $88.8 $88.3
8.8 :88 85 8.8 888 88.8 $880888
£888: 8:8 188.3 1.88.3 $88.3 1.88.:
82.8 88.8 88.8 N88 S88 88.» 828 85
88.8 88.3 188.: 1.88.: 88.8 188.:
88.8 83- 88.8 83 88.8 83 .888 @888
88.3 88.3 82.8 $88.: 88.3 188.:
83. 88.8- 83 :3. 83 «$8 .8885
88.8 88.3 88.: 88.: 88.8 83.:
82- 83- 888- 888 888- 888 82.888 8888 088
+8 om< 88 8< 88 82 88m 835 885 8388.5
Egon

 

8.803 3 28.8

134

8:85:00 8389

 

 

 

 

 

888 888 888 8888 888 888 8888 8888800
888 8:8 888 888 838 888 888% 8880
888 888 888 888 8888 888 888.8 8<
888 888 888 888 888 E88 888
888 888 888 888 888 888 82888
8880? 8:8 88 8838: $58.. .5.“ 82.878
888 888 888 8888 888 8888 8888 8888800
888 8:8 8:88 888 888 8888 8888 8280
8888 888 :88 .8888 888 :88 888888 8<
:28 888 888 8888 888 888 888.8
888 888 888 888 888 888 82888
088 3 78:8 82205800 magma... 3“ 83.878
888 888 888 888 888 888 8888-8
8: 88 88 88 88 88 8288880 8 .oz
8% 8% 8% 8% 8% 8% 85830 H8»... 88838800
888 88.8 88.8 188.8 88 1.888
888- 888- 888- 888- 888 888- 888
88.8 88.8 88.8 88.8 88.8 88.8
88.8 888- 888- 888- 888 88- 8 888 88 88 83
88.8 888 1.88.8 .8888 888.8 188.8
888- «88 25 88.8 888 88.8 H 88.8 88 82 88
188.8 88.8 88.8 88.8 88.8 88.8
88.2 888- 888 888 888 888 $880888
88.8 88.8 88.8 88.8 88.8 88.8
88.8 88 88 888- 888 888 888 .88
88.8 83.8 83.8 88.8 82.8 88.8
888 888- 888 888 8:8 888 828m 882
82.8 88.8 88.8 88.8 82.8 88.8
888 888- 888 888- 888 83 8888
88.8 88.8 88.8 8:8 :88 88.8
888 888- 888 888 888 888 88888 88888 888
+8 8.3 888 8< 88 8< 888 885 885 858.85

882

8:80 888 3. 88.8 8 .8 .8 82:0 8 888 8888 8888 83 88 28.8

135

.38— mod 88 .1. new BR: ad
8.8 85853me 828588 88865 _.. .8838889 5 88 888 889888 8958 6867:0885
.888838 fionoo 88.8.96 8:8 88:56 own 88 828%»: on... E 8822: 818 8qu

 

 

 

 

 

 

888 888 888888 88880
8888 8888 888:8 88¢.
888 8888 $888.
8888 8895 8:8 82 888:8
893 3.2: Una can“: £09,503 wcmumvu ho“ 853>unm
8888 888 8888 888 8888 888 888888 3885880
888 88.88 8888 3.8.8 838 888 88888 88880
8888 8888 8888 E88 8888 888 888888 88
888 :88 8888 8888 8:88 888 $8
8888 88.8 8.88 888 888 8888 888:8
OuwN Ow Rag mufivmoEOOo wflmuwvu qu 03—97%
88.8 888 888 8888 8.88 888 8888.8
8: 888 888 888 888 888 8885380 88 .oz
8% 8% 8% 8% 8% 8% 8556 H8»... 3858800
:88 1.828 83.8 188.8 888.8 ..888
828 888- 838- 888- 83 8:8- 883
:88 888.8 888.8 88.8 88.8 888.8
888- 888 888 8.88 888 888 8 £888 888 88 83
88.8 888.8 88.8 88.8 888.8 888.8
:88 888 828 88.8 888 8888 8 93.8 Ba 82 83
88.5 88.8 188.8 88.8 1.88.8 188.8
8:8 888 8888 888 88.8 888 88280388
8888 88.8 83.8 88.8 88.8 88.8
8.8.8 838 :88 E88 888 888 8888. 888
88.8 888.8 88.8 888.8 82.8 888.:
8:8 888- 888 888 88.8 88.8 828. 2882
888.8 :88 88.8 188.8 88.8 1.88.:
88.8 :2 :3 88.8 888 888 .588
888.8 88.8 88.8 88.8 88.8 88.8
88.7 88.8- 8:; 888- 888 8888 888:8 58:8 meow
+8 88 8.88 88 88 88 858 825 8826 838.25
820m

 

8.888 88 288.8

136

835:8 2an

 

 

 

 

 

88.8 88.8 88.8 88.8 88.8 88.8 888% $5888
885.8 388.8 888.8 3.8.8 258 £88 888% 8280
32.8 82.8 :88 888.8 8.2.8 3.8.8 8.86% 03.
25.8 858 8.8.8 88.8 88.8 888 $82
«88.8 882.8 53.8 88.8 «58 883 8:838
G080? fig Goa cowgmfi magma» “Om azmgnnm
88.8 88.8 88.8 88.8 88.8 88.8 888% 85888
$88 88.8 23.8 88.8 88.8 383 888% 8280
$38 82.8 83.8 828.8 2.8.8 883 88:88 83.
888.8 38.8 888.8 88.8 88.8 :88 882
883.8 $8.8 88.8 88.8 88.8 88.8 858:8
OHvN Cu ~d=g EGGS—“@000 wflmamvu “Ow 9:qu
and 8.3.8 888 88.8 «88 ~58 83:38
a: $8 88. 3% gm 8% 228580 mo 52
8% 8% 8% 8% 8% 8% 888:8 88%.." 3338800
88.8 $83.8 $88.8 $888 83.8 $888
88.8 88.8- 88- :8- :88- 5.8- 23
8.8.8 828 88.8 88.8 88.8 88.8
:88- 8.3 :88- 88.8- 83 88.8 N 3&3 Ba 32 m3
888 8.3.8 $88.8 $888 82.8 88.8
”3.8. 83 £88 ”8.8 88.8- 33 H 388 Ea 82 m3
83.8 8.8.: $88.8 $88.8 $88.8 $838
83 88.8 £3 88.” 85 33 $8888.
88.: $83.8 $88.8 $88.8 $88.8 $888
23 883 Sn.” £88 888 88.8 828 .85
88.8 $838 $88.8 $838 88.8 $838
88.8 and 88.8 83 82 RS 82% @882
88.8 83.8 $888 88.8 88.8 $838
888 ES «8.8 $8 $2 8.8 .588
88.8 828 $8888 838 88.8 83.8
58. 88.8 83 88.8 23 88.8 858:8 .5828 28m
+8 mm< $8. 83 8.88 03 85m 985 :85 833:5
982

 

 

8:80 888 3. Sam .8 .8 .88 mzmo a 8 sea 288 888 ”2 838

137

.—®>®— m0.0 ad 83.. Ufld ~0>G~ H.O

um 8588::me 828338 830%.: ... .885588 5 98 £83 88988 0852 38:88qu
828888 €038 $08.96 was 8:356 mmw on.» 858883 on: 5 @2832: 814. ”302

 

 

 

 

 

 

88.8 888 8888 580
88.8 8:8 8888 8<
88.8 8:8 383
88.8 8895 :88 82 838:8.—
.oamem 3.2: GEM Edna: £98,503 wcmumou “Ow mmuzgunm
888 88.8 8888 888.8 88.8 88.8 88888 88888
88.8 88.8 58.8 88.8 88.8 38.8 8888 8080
88.8 88.8 88.8 8:8 828 88.8 888% 8<
88.8 88.8 88.8 88.8 88.8 88.8 382
888.8 88.8 888.8 888.8 28.8 88.8 888%
Bow 3 1250 3808525 958”“ “8 aggd
88.8 88 888 88.8 88.8 88.8 83:88
8: 88 88 88 88 88 852,680 8 .oz
85 88> 36> 85 mm? 8> 888:8 89»... 3238800
1888 38.8 88.8 88.8 88.8 88.8
83 88.8- 88.8. 88.8- 83. 88.8- 23
$2.3 132.3 *vaodv *Aonodv *meodv 38093
88.8 83 8:8 83 83 83 m 388 Ba 88 83
88.8 32.8 88.8 :88 8:8 888
Sod 88.8- 88.8 28.8 88.8 888 H $0.08 Ba 82 meg
188.8 1388 1388 13:8 188.8 1.888
83 8m.” 88.” H8.” «:8 8.8 8888389
138.8 :888 £388 £888 :88 $888
88.8 83” 888 8.8.“ 8: 82 3°88 85
:88 88.8 138.8 1.888 1888 3:88
83 88.8 83 83 888 88.8 .028 28:2
588 :88 138.8 188.8 888 I888
83“ 88.8 83 8: $3 82 58E
2:8 88.8 88.8 88.8 888 3.8.8
N88- 88 83 «8.8 ES- 88 888:8 583 2:8
+8 03 88 8< 88 83. 3:8 555 :85 838:3
flan—om

 

8,288 2 285

138

825:8 2an

 

 

 

 

 

3833 883 883 383.: 883 88.: 888% 888800
883 :83 8:3 883 883 883 888% :28
883 883 883 $83 3833 883 888% 8<
83 83 883 883 883 883 33.2
883 883 8:8 883 $83 833 8888
20803 Udd G08 €831.53 wcmuwmwu HO.“ mm=~w>unm
3833 883 8883 383 3833 883 888% 828800
883 883 883 883 383 883 888% :88
83 83 833 883 883 883 8:888 8<
883 883 8833 3833 883 883 $82
883 883 883 883 8883 883 8888
Sou 3 158 388888 @589 go.“ 038$
83 83 83 8:3 83 83 83:88
8: 88 88 88 88 88 88380 8 .oz
8% 8% 8% 8% 8% 8% 8883. 8.8%.." 3888800
:88 188.8 188.8 188.8 :38 188.8
833- 833- 383- 833- 83 883- 8.3
38.8 88.8 :88 1:88 888 1:88
883 833 833 883 :83 ~83 a 388 Ba 18. 83
:88 :88 138.8 38.8 :88 88.8
83.8- 83 833 383 883 833 H 288 Ba 82 83
1:88 :88 1:88 888 1:88 1:88
82 83 :83 83 88.: 83 888088
138.8 188.8 :88 138.8 1388 18:8
88.: 883 883 83 883 83 8:8 88
188.8 68.8 88.8 :28 1:88 188.8
88.: 833 383 883 S3 :83 328 22:2
1888 :88 :88 838 1:88 :88
883 833- 883- 833- 83 83 388
:88 :88 :88 83.8 :88 88.8
83 83- 83 833 83 83 858:8 58:8 oaom
+8 8< 88 8< 88 8< 8:8 :85 :85 8333/
282

 

 

8:80 88 8 Sam 8 .8 .8 880 a 5305 as“: 288 888 8.8 289

139

:96— mod p: .1. «2:: 736— ad

a: 22:85::me 32:28:: 828me ..n .88fi88a E 0:: 80:8 222:3: :258 887::882
.8353: 20:8 8z8>m 9:: 88:83. mm: 8:: 2:28:28 2: 3 8:32: 81:. 8qu

 

 

 

 

 

 

::::: 22.: 885:: :28
:2:: 2E: 888:: mm:
29:: $2: :88:
::.::: 86:3 22.: 82 888::
8:: 3:3: «2:: can; c8325 3:58: :8 83.27%
88.: 88: 88.: 88: 88: 25:: 888:: 58:88:”:
22: 22: :2:: 22: ::::: 82: 888:: :28
88: :2:: 22: 88: 28: 88: 888:: :2:
22: ::::: :2:: ::::: 22:: 88: 8:2,.
22: 2.2: 88: 88: 88: 88: 888::
0:8: 0: :38 888508 $58: :8 2:27am
2:: 3:: 2:: Km: 2:: 3:: 88:32
:2: :2: 2:: :2: :2: :2: 2:53:20 :: .oz
8% 8% 8% 8% 8% 8% 2283: 89:... 13258800
38.3 1:83 1:83 1:83 :::.3 188.3
:::.:- 3:3 :::.:- 2.9:- :2: ::::- :93
:83 188.3 ...:::.3 1:33 :83 1:83
::::- 3:: :2: :::: 2:: :2: : 38:: :8: :8: :3
82.3 83.3 :83 :83 83.3 :::.3
:2: 2:: 8:: 2:: :8: 2:: : 2:8: :8: :8: mo:
88.: 1:33 1:23 1:83 1:23 1:2:3
::::- :2: 2:: 2:: 5:: :::.: 88:838.:
$2.3 1:83 182.3 1:23 1:23 1:23
:23- :2: :2: 2:: :::.: :2: :88: :2:
:23 1:::.3 1:23 1:23 13::3 182.3
:2: 8:: 8:: 2:: 2:: 8:: :88: 2:22
Amwmdv *waHdV 132.3 132.3 *Lommdv 132.3
:2: :3: 3:: 8:: 8:: 2:: 2:82:
83.3 :23 :23 :23 8:3 ...:::.3
2:: :2: :2: :2: :2: :2: 858:8 38:: :80:
+:: 0:: 2:: 8: 2:: mm: 8:: :85 88.6 832:5
8?ng

 

5:83 :.: 28.:

140

823355 2an

 

 

 

 

 

883 88¢ 88¢ 83:. 833 3.26 9.388% £59.88
£33 3.83 835 3%,-3 833 383 835% €28
385 .53 33.3 :85 ~83 885 @883 of.
3335 83.0 «:3 52 33.0 338 332
33.5 283 :8... 333 3:3 835 858:3
5803 «Ed 58 :8383 madame... H8 833$
883 883 883 88.0 883 886 838.8% 55888
823 386 333 833 £85 53¢ 8388.6 328
$32 :86 835 23$ 383 883 838853 03.
333 flood 83:. 886 «33 88¢ 38%
883 $83 883 886 385 885 cosssvm
Chou OH 3230 mudmmoavoo Mafia—mm... .mOm 053nm
«and 353 was 82 833 333 853-m
a: NE 33. 833 gm 8% 8035030 mo .2
8% mo> 8% mo> mo> 8> .3556 So»... 3838800
383 38.3 $38.3 ......33.3 333 233.3
33. 33 £3 88 $2. 333 ES
.333 32.3 33.3 32.3 33.3 138.3
33- 33.3 38.3 83- 53. 32.3 m 338 Ea we wag
38.3 32.3 1.32.3 1333 32.3 38.3
83° 83- 333.0- :2- 333 SS- 3 3&8 Ba 32 33
1.38.3 133.: 1.38.3 $38.: 1.32.: 133.3
83.3. 83.3- 33.3- 33%- 833- :3- $328383
1.33.3 $333.: 1.32.: 3.32.3 1333 138.3
38....- 233- 38.3- .83- 53- 33.3- 628 .35
$38.3 $38.3 $38.3 1.383 $33.3 $33.3
83.3. 23. 3.3.3- 33- 33.“- 333- 62?. 23:2
33.: 38.3 338.: .3333 138.3 2333.3
«3.7 :3- 23- BS- 953- 333. .585
32.3 333.3 338.: 1.383 38.: 1333
33.3 83. 33.3. 83- 82.. mm:- cofioéo 58E 2:8
+8 $3. 33. $3. 8.8 33. 35m :25 3226 833:3
234

 

 

3:20 383 3 883 8 .5 .3 mzmo 5 836.3830 :53 2.530 385 ”3 2E-

141

88— mod 8 t." 98 32: ad as 858%:
-me 8285-QO 880%5 ... .885888 E 08 888 @8988 8:32 ~m>o_-:88m.8m88=v
8058 888.26 was 8:956 88 08 8288on 85 5 89:2: 82 8qu

 

 

 

 

 

 

88.8 88.8 88.8% :88
88.8 88.8 8:888 «3
88.8 83.8 382
:88 8595 88.8 82 858:8
888 8:: 98 93.8 Q8389 @388 3.“ 83.87am
888.8 88.8 888.8 88.8 88.8 88.8 8888 888800
8:8 «:8 88.8 28.8 38.8 88.8 8888 288
888.8 88.8 88.8 88.8 88.8 88.8 888% 83
88.8 88.8 88.8 88.8 :88 88.8 388-
,888 88.8 888.8 888.8 888.8 88.8 888me
Bow 3 8:8 8:20508 9:58... :8 mag-m
88 88.8 88 88 88.8 838 8181-8
8: 88 88 88 88 8.8 88158808 .02
8> 8% 8% 8.» 8% 8? 88:8 “8%.." 3838800
:88 188.8 :88 188.8 $88 1:88
88.8- 38 88.8 8.8 :88 88.8 83
:88 182.8 168.8 1888 1:28 1888
88.8 8.8- 88- 83- 88- 88- m 388 Ba 82 83
888 :38 $8.8 88.8 :28 :88
:3- 88.8 88.8- 888- 88.8- 88.8- H 381 88 we meg
1:88 138.: 1:38 188.: 1388 1:88
v8.8- 83- 83- 83-- 88.8- 831 8880389
1:88 1:88 1888 1:88 1388 1:88
88.2- 83. 88.8- 888- 88.”- 83”- 388. 88
:88 1:88 1:88 1888 1:88 188.8
88.8- 83- 8.8 33- 83- Eva- 388 28:2
:88 L888 1888 1388 18:8 1288
:3- 83- 83- 83- 88.:- 83- .588
68.8 88.8 88.8 88.8 :88 38.8
88 8:8- 88- 88.8- 88- 838- :2888 .588 288
+8 8< 88 8< 88 8< 1.58 :85 =85 8333/
m—dfiom

 

8.888 «mm 288

142

Table 2.10: Food Consumption in CHNS 89,91,93 with Community Characteristics:

 

 

 

Overall
Male
Calorie Fat Protein % Calorie From
Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Some primary edu 0.0091 0.0699 0.0187 0.584 0.138 -1.357
(0.0136) (0.0402)* (0.0163) (0.419) (0.129) (0.525)“
Primary 0.0094 0.0770 0.0173 0.854 0.130 -1.462
(0.0137) (0.0417)* (0.0167) (0.434)" (0.133) (0.538)"
Middle school -0.0023 0.1590 0.0217 1.805 0.356 -3.013
(0.0138) (0.0417)** (0.0168) (0.438)” (0.135)" (0551)”
High school -0.0150 0.1832 0.0191 2.231 0.508 -3.857
(0.0156) (0.0463)** (0.0189) (0.500)" (0.158)" (0.633)M
Tech/College+ -0.0049 0.1903 0.0389 2.050 0.712 -4.125
(0.0186) (0.0501)** (0.0221)* (0.580)“ (0.201)** (0.738)M
Log R prod assets 1 0.0099 0.0279 0.0132 0.172 0.040 ~0.253
(00021)" (00055)" (0.0025)** (0.069)“ (0.022)* (0.084)”
Log R prod assets 2 -0.0054 -0.0069 -0.0050 0.010 0.013 -0.095
(0.0020)** (0.0054) (0.0024)M (0.065) (0.021) (0.081)
Land 0.0024 -0.0036 -0.0001 -0.108 -0.038 0.126
(0.0012)** (0.0032) (0.0013) (0.030)” (0.010)” (0.042)"
Urban Residence 0.2154 2.3329 0.5319 23.291 3.938 -26.542
(0.0773)** (0.1873)** (0.0893)** (2.219)" (0.810)** (2.758)"
Log Price of Rice -0.0696 -0.1576 -0.0981 -0.883 -0.341 2.408
(0.0171)** (0.0490)** (0.0214)** (0.373)" (0.138)M (0.472)“
Log Price of Pork 0.0187 0.2569 0.0786 0.618 0.126 -0.849
(0.0245) (0.0719)** (0.0307)** (0.124)M (0.043)“ (0.159)M
Log Price of Eggs -0.0181 -0.0318 -0.0270 -0.034 -0.062 0.005
(0.0038)** (0.0112)** (00047)" (0.062) (0.021)M (0.077)
Log Price of Oil -0.0603 -0.1561 -0.0409 -0.375 0.056 0.262
(0.0248)** (0.0764)** (0.0306) (0.141)" (0.049) (0.167)
Water Source 2 0.0321 0.3157 0.0646 4.037 0.483 -3.937
(0.0268) (0.0857)** (0.0328)** (0.869)" (0.268)* (1.065)"
Water Source 3 0.0911 0.2712 0.1362 1.472 0.586 -2.829
(0.0377)** (0.1203)** (0.0468)** (1.289) (0.357)* (1.597)*

 

143

Table continues

Table 2.10 (cont’d)

 

 

Female
Calorie Fat Protein % Calorie From

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Some primary edu -0.0032 0.0381 0.0055 0.294 0.136 -0.349
(0.0091) (0.0297) (0.0118) (0.315) (0.102) (0.356)
Primary 0.0139 0.1503 0.0400 1.482 0.377 -1.863
(0.0101) (0.0335)** (0.0131)** (0.364)” (0.115)“ (0.414)”
Middle school -0.0202 0.1616 0.0168 2.044 0.528 -2.505
(0.0103)** (0.0329)** (0.0132) (0.365)** (0.117)** (0.412)**
High school -0.0270 0.1985 0.0165 2.818 0.637 -3.402
(0.0139)* (0.0419)** (0.0176) (0.481)** (0.152)" (0.539)“
Tech/College+ -0.0062 0.1878 0.0610 2.518 1.059 -3.482
(0.0190) (0.0510)** (0.0239)** (0.650)“ (0.214)** (0.713)**
Log R prod assets 1 0.0036 0.0143 0.0064 0.086 0.038 -0.156
(0.0019)* (0.0052)** (0.0024)** (0.064) (0.022)* (0.073)**
Log R prod assets 2 -0.0002 0.0047 0.0014 0.128 0.031 -0.146
(0.0017) (0.0050) (0.0023) (0.061)** (0.021) (0.069)“
Land 0.0019 -0.0022 -0.0009 -0.061 -0.042 0.106
(0.0010)* (0.0031) (0.0011) (0.036)* (0.009)** (0.042)“
Urban Residence 0.2494 2.4656 0.6050 24.207 4.555 -28.658
(0.0824)** (0.2005)** (0.1002)** (2.334)** (0.830)** (2.655)“
Log Price of Rice -0.0332 -0.1060 -0.0756 -0.553 -0.469 1.068
(0.0156)“ (0.0457)** (0.0201)** (0.372) (0.133)** (0.407)“
Log Price of Pork 0.0161 0.3100 0.0794 0.734 0.131 -0.824
(0.0231) (0.0685)** (0.0298)** (0.121)** (0.040)M (0.133)**
Log Price of Eggs -0.0139 -0.0306 -0.0196 -0.095 -0.026 0.107
(0.0035)** (0.0109)** (0.0045)** (0.060) (0.020) (0.066)
Log Price of Oil -0.0522 -0.1853 -0.0109 -0.467 0.122 0.392
(0.0212)** (0.0697)** (0.0276) (0.140)” (0.044)" (0.146)“
Water Source 2 0.0576 0.3074 0.0788 3.458 0.350 -3.712
(0.0238)** (0.0845)** (0.0326)** (0.867)** (0.278) (0.980)M
Water Source 3 0.0889 0.1855 0.0902 1.753 0.061 -2.370
(0.0391)** (0.1220) (0.0484)* (1.275) (0.372) (1.400)*

 

144

Table continues

Table 2.10 (cont’d)

 

 

 

 

Male
Calorie Fat Protein % Calorie From
Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Water Source 4 -0.0347 0.1699 0.0092 3.032 0.631 -2.866
(0.0244) (0.0724)** (0.0307) (0.830)” (0.261)M (0.992)"
In house no flush 0.2178 0.3453 0.1944 0.560 -0.981 3.822
(0.0632)** (0.1846)* (0.0817)** (2.121) (0.772) (2.550)
Outside toilets 0.0690 0.0294 0.0442 -1.596 -0.842 3.650
(0.0499) (0.1324) (0.0630) (1.656) (0.664) (1.924)*
Open pit 0.0355 0.0467 0.0356 -0.486 -0.484 1.893
(0.0481) (0.1297) (0.0622) (1.633) (0.657) (1.907)
No toilets 0.0142 -0.0652 0.0421 0.691 -0.111 2.661
(0.0887) (0.2548) (0.1112) (2.847) (1) (3.554)
Very little excreta 0.0785 0.1416 0.1048 0.785 0.363 -1.401
(0.0243)** (0.0683)** (0.0297)M (0.714) (0.234) (0.904)
Some excreta -0.0693 -0.2504 -0.0338 -1.822 0.603 2.121
(0.0235)** (0.0677)** (0.0289) (0.703)" (0.240)M (0.885)“
Much excreta -0.0038 0.3738 -0.0795 -1.351 -1.135 -1.099
(0.0992) (0.2838) (0.1171) (2.552) (0.817) (3.487)
Community Dummies Yes Yes Yes Yes Yes Yes
Number of obs 8415 8415 8415 8415 8415 8415
R-squared 0.2352 0.2497 0.2124 0.303 0.318 0.344
P-value for testing coefficients equal to zero
Education 0.2171 0.0176 0.8943 0.0000 0.0005 0.0000
Prod Assets 0.0000 0.0002 0.0000 0.0008 0.0054 0.0000
Age dummies 0.0001 0.0570 0.0062 0.0052 0.0201 0.0009
Cohort dummies 0.0827 0.6370 0.2266 0.0320 0.5115 0.3137
Prices 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Water source 0.0088 0.0052 0.0220 0.0000 0.0655 0.0010
Toilet type 0.0225 0.6854 0.5398 0.3035 0.6411 0.3092
Excreta 0.0000 0.0001 0.0003 0.0088 0.0270 0.0043
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

145

Table continues

Table 2.10 (cont’d)

 

 

 

 

Female
Calorie Fat Protein % Calorie From

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Water Source 4 0.0240 0.1668 0.0417 2.383 0.376 -2.965
(0.0224) (0.0709)** (0.0301) (0.800)" (0.263) (0.897)**
In house no flush 0.2497 0.5456 0.2361 3.253 -0.762 -1.617
(0.0630)** (0.1872)** (0.0828)** (2.255) (0.755) (2.479)
Outside toilets 0.1007 0.0973 0.0940 -0.225 -0.346 0.779
(0.0438)** (0.1206) (0.0560)* (1.591) (0.596) (1.741)
Open pit 0.0354 0.0156 0.0299 -0.306 -0.307 0.524
(0.0427) (0.1199) (0.0558) (1.594) (0.584) (1.754)
No toilets -0.1358 -0.2752 -0.0562 -2.925 0.628 1.485
(0.0774)* (0.2387) (0.1002) (2.958) (0.988) (3.287)
Very little excreta 0.0247 0.1039 0.0494 0.709 0.370 -1.040
(0.0222) (0.0671) (0.0279)* (0.710) (0.224)* (0.780)
Some excreta -0.0320 -0.2134 -0.0075 -1.835 0.462 1.225
(0.0209) (0.0659)** (0.0265) (0.690)M (0.228)” (0.766)
Much excreta -0.2193 0.0839 —0.0887 -0.251 1.274 -1.846
(0.0903)** (0.2800) (0.1097) (2.588) (0.821) (2.926)
Community Dummies Yes Yes Yes Yes Yes Yes
Number of obs 9162 9162 9162 9162 9162 9162
R-squared 0.2613 0.2606 0.2050 0.316 0.334 0.381

P—value for testing coefficients equal to zero

Education 0.0039 0.0000 0.0052 0.0000 0.0000 0.0000
Prod Assets 0.0357 0.0020 0.0001 0.0001 0.0002 0.0000
Age dummies 0.0004 0.0001 0.0000 0.0001 0.0000 0.0000
Cohort dummies 0.0363 0.6476 0.4957 0.4179 0.0060 0.0468
Prices 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Water source 0.0064 0.0001 0.0111 0.0004 0.4180 0.0004
Toilet type 0.0000 0.0082 0.0049 0.2430 0.7888 0.5880
Excreta 0.0177 0.0345 0.1375 0.0118 0.0621 0.0814
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

 

Also included in the models are individual-year age dummies, five-year-cohort dummies and
community dummies. Person-level robust standard errors are in parentheses. * indicates
statistical significance at 0.1 level and ** at 0.05 level.

146

2.5 Results: Determinants of Changes

During the rapid economic growth after economic reforms more health risks have
resulted from deteriorating dieting habits (Guo, Popkin and Zhai 1999, Guo et a1.
2000). Finding out the socioeconomic determinants of the change of people’s food
consumption has important policy implications. Here the changes in all outcomes
between 1989 and 1991 and the changes between 1991 and 1993 are pooled together
and OLS estimation is used in the basic and augmented models again to find out the
effects of the changes in prices and other community characteristics on the changes
in calorie, fat and protein intakes and the changes in percent of calories from fat,

protein and carbohydrates.

2.5.1 Basic Models

When the outcomes are changes in calorie, fat and protein intakes and the percent
of calories from fat, protein and carbohydrates the effects of the first set of regressors
— education, productive assets and land together with age, cohort and community
dummies — are mostly rendered insignificant, except for a few individual level variables
and the community dummies. For females in the whole sample, the rural sample and
the younger age group, the nonlinear effects of productive assets are quite strong.
Recall in Table 4 there is no such relationship present for females. For men the levels
of nutrient intakes tend to respond to the base productive asset holdings and for
women the changes of nutrient intakes are more responsive to assets. In Table 13 for
rural and younger females the effect of initial productive assets on changes in calorie
intakes become significant and are of the same direction as the effect of productive
assets on the level of calorie intakes for rural and younger men.

The land effects on changes in fat intake (Table 2.14) and changes in the percent

of calories from fat (Table 2.16) become positive for all women and rural women

147

and on changes in percent calorie from carbohydrates (Table 2.18) become negative.
This is puzzling. The effect of productive assets on changes in protein intake for
younger female becomes of the same signs as the case for effects of productive assets
on male protein level intake. Most relationships in Table 2.17 for changes in percent
calorie from protein are insignificant except for some cohort dummies and community
dummies.

Maybe the two changes measure are too close to each other, and / or taking differ—
ences on the outcomes measured with errors may exacerbate the problem and hence
make it more difficult to render significant results. When a longer panel is available
and outcomes are better measured the effects of some of the first set regressors should

become significant again.

148

Table 2.11: Food Consumption in CHNS 89,91,93 with Community Characteristics

in Urban Areas

 

 

 

 

Urban Male
Calorie Fat Protein % Calorie From

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Some primary edu -0.0105 -0.0107 0.0052 0.036 0.253 -0.945
(0.0278) (0.0748) (0.0329) (0.916) (0.285) (1.063)
Primary 0.0150 0.0878 0.0587 1.078 0.719 -2.352
(0.0270) (0.0761) (0.0325)* (0.945) (0.301)M (1.096)“
Middle school -0.0190 0.0769 0.0239 1.106 0.706 -2.391
(0.0266) (0.0751) (0.0325) (0.913) (0.298)” (1.064)"
High school -0.0251 0.1250 0.0287 1.847 0.898 -3.756
(0.0279) (0.0781) (0.0340) (0.957)* (0.317)“ (1.136)**
Tech/College+ -0.0213 0.0967 0.0490 1.168 1.167 -3.116
(0.0292) (0.0780) (0.0343) (0.971) (0.325)M (1.125)**
Log real prod assets 1 0.0058 0.0092 0.0065 0.026 0 -0.063
(0.0034)* (0.0083) (0.0041) (0.117) (0.041) (0.133)
Log real prod assets 2 -0.0005 0 -0.0022 -0.010 -0.020 -0.063
(0.0033) (0.0077) (0.0040) (0.109) (0.040) (0.124)
Land 0.0308 0.0242 0.0341 -0.152 0.042 0.594
(0.0181)* (0.0457) (0.0213) (0.493) (0.130) (0.635)
Log Price of Rice -0.0473 -0.1715 —0.1117 -1.754 -0.763 2.875
(0.0349) (0.0871)** (0.0451)** (0.669)" (0.272)” (0.778)“
Log Price of Pork 0.0624 0.3767 0.0829 0.743 0.074 -0.939
(0.0432) (0.1127)** (0.0553) (0.229)" (0.085) (0.275)"
Log Price of Eggs -0.0127 -0.0297 -0.0148 -0.045 -0.021 -0.005
(0.0072)* (0.0186) (0.0088)* (0.115) (0.043) (0.134)
Log Price of Oil -0.1477 -0.4042 -0.0722 -0.937 0.158 0.867
(0.0629)** (0.1683)** (0.0802) (0.343)** (0.122) (0.387)"
Water Source 2 0.1044 0.3974 0.1613 4.868 0.733 -6.214
(0.0848) (0.2478) (0.0958)* (2.867)* (0.806) (3.228)*
Water Source 3 0.2052 0.2330 -0.0315 -0.571 -3.781 4.148
(0.1412) (0.3567) (0.1722) (4.522) (1.711)** (5.175)

 

2.5.2 Augmented Models

Table continues

Next, in addition to the above variables we study the effects of changes in food

prices and community characteristics on the changes in outcomes. For the effects of

prices the estimates that are significant in Table 2.19 remain of same signs as the

effects of prices in the corresponding level estimation in Table 2.10. For changes of

prices, only two relationships are still significant in the whole sample, the changes

of the price of rice on the changes in male percent of calories from carbohydrates

'(negative) and female percent of calories from protein (positive).

149

Table 2.11 (cont’d)

 

 

Urban Female
Calorie Fat Protein % Calorie From

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Some primary edu -0.0011 0.0284 0.0216 -0.237 0.360 0.002
(0.0191) (0.0559) (0.0241) (0.676) (0.215)* (0.736)
Primary 0.0015 0.1186 0.0333 1.302 0.486 -1.646
(0.0208) (0.0577)** (0.0255) (0.717)* (0.229)" (0.789)“
Middle school -0.0239 0.1173 0.0170 1.813 0.670 -2.312
(0.0202) (0.0579)** (0.0245) (0.717)M (0.227)" (0.791)“
High school -0.0122 0.1922 0.0401 2.798 0.869 —3.485
(0.0230) (0.0633)** (0.0282) (0.798)M (0.267)" (0.884)”
Tech/College+ 0.0101 0.1645 0.0856 2.254 1.269 -3.348
(0.0266) (0.0714)** (0.0323)** (0.925)" (0.290)” (1.006)"
Log real prod assets 1 0.0033 0.0112 0.0036 0.154 0.001 -0.192
(0.0033) (0.0089) (0.0043) (0.121) (0.040) (0.138)
Log real prod assets 2 0.0004 0.0060 0.0025 0.125 0.027 -0.124
(0.0028) (0.0073) (0.0038) (0.101) (0.038) (0.121)
Land 0.0090 0.0130 0.0184 -0.103 0.164 -0.066
(0.0129) (0.0249) (0.0137) (0.200) (0.089)* (0.234)
Log Price of Rice -0.0155 -0.0494 -0.0345 -0.952 -0.330 1.356
(0.0306) (0.0776) (0.0389) (0.685) (0.269) (0.745)*
Log Price of Pork 0.0198 0.3341 0.0023 0.898 -0.066 -0.821
(0.0433) (0.1051)** (0.0538) (0.225)" (0.079) (0.245)"
Log Price of Eggs -0.0152 -0.0514 -0.0166 -0.144 -0.016 0.136
(0.0066)** (0.0182)** (0.0084)** (0.118) (0.042) (0.127)
Log Price of Oil -0.0878 -0.4409 0.0641 -0.946 0.439 0.742
(0.0520)* (0.1437)** (0.0698) (0.316)“ (0.115)” (0.348)"
Water Source 2 0.0628 0.4225 0.0730 5.117 0.212 -6.146
(0.0727) (0.2309)* (0.1081) (2.907)* (0.949) (3.330)*
Water Source 3 0.1207 0.3823 -0.0636 2.880 -2.850 -1.164
(0.1247) (0.3284) (0.1533) (4.222) (1.462)* (4.396)

 

Table continues

In urban areas the changes in the price of rice adversely affect the change in

protein intakes for men. In rural areas the change in percent of calories from protein

for female is negatively related with the changes in the price of rice. The changes

of the price of pork do not have any statistically significant impact on any of the

changes in outcomes. The effects of the changes in the prices of eggs remain strong

and bear the same signs as in the level estimations. The changes in the prices of oils

have negative impacts on changes in male calorie and protein intakes and changes in

female calorie intakes in the whole sample and the rural areas. In the urban areas the

impact of the change in the price of oil is positive for the change in percent of calories

from protein consumption for female. In rural females, in addition, the impact of the

150

Table 2.11 (cont’d)

 

 

 

 

Urban Male
Calorie Fat Protein % Calorie Horn

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Water Source 4 -0.0185 0.2667 —0.0184 5.538 -0.035 -4.791
(0.0458) (0.1128)** (0.0562) (1.345)“ (0.475) (1.622)“
In house no flush 0.3390 0.9510 0.2899 4.928 -0.265 0.709
(0.1495)** (0.3523)** (0.1863) (4.317) (1.742) (5.085)
Outside toilets 0.1653 0.2842 0.0773 0.044 -1.432 2.207
(0.0950)* (0.2418) (0.1082) (3.152) (1.211) (3.600)
Open pit 0.1533 0.4475 -0.0140 4.362 -2.991 -2.123
(0.1036) (0.2666)* (0.1269) (3.406) (1.279)“ (4.029)
No toilets 0.0590 0.6804 0.0279 8.667 -0.444 -9.786
(0.1418) (0.3285)** (0.1649) (3.973)" (1.548) (4.742)“
Very little excreta 0.0846 0.2525 0.0815 2.846 0.200 -5.144
(0.0692) (0.1493)* (0.0843) (1.862) (0.670) (2.225)"
Some excreta -0.3530 -0.7070 -0.2652 -4.272 1.467 3.178
(0.0768)** (0.1970)** (0.0935)** (2.168)“ (0.654)” (2.550)
Much excreta -0.9336 -0.5271 -1.6754 -10.357 —11.880 34.373
(0.3917)** (1.1049) (0.4797)** (13.187) (3.926)" (15042)”
Community Yes Yes Yes Yes Yes Yes
Number of obs 2727 2727 2727 2727 2727 2727
R—squared 0.2052 0.1917 0.2185 0.212 0.245 0.259

P-value for testing coefficients equal to zero

Education 0.3724 0.6362 0.2372 0.1475 0.0014 0.0037
Prod Assets 0.2176 0.7416 0.3072 0.9744 0.8263 0.5924
Age dummies 0.0000 0.0000 0.0000 0.0004 0.0182 0.0000
Cohort dummies 0.2508 0.0049 0.0558 0.0001 0.2528 0.0000
Prices 0.0613 0.0015 0.0276 0.0005 0.0225 0.0001
Water source 0.2814 0.1386 0.4408 0.0004 0.1027 0.0054
Toilet type 0.1254 0.1617 0.0784 0.0201 0.0180 0.0827
Excreta 0.0000 0.0003 0.0000 0.0430 0.0020 0.0035
Community 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

Table continues

change in price of oil is strong and negative for the changes in the percent of calories
from protein.

Significant effects of prices between urban and rural areas (details available upon
request) on changes in macronutrients are similar. There is no implication for policy
differentials between urban /rura1 residency although there are significant effects of
the community dummies. This could be due to several reasons. One, prices as
policy instruments may be more effective in changing people’s consumption in specific

foods, but not changes in total nutrient intakes. Two, in two years the changes in

nutrient intakes are not dramatic and are measured with errors, which results in larger

151

Table 2.11 (cont’d)

 

 

 

 

Urban Female
Calorie Fat Protein % Calorie From

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Water Source 4 0.0692 0.3628 0.0912 5.691 0.474 -6.623
(0.0394)* (0.1093)** (0.0532)* (1.289)“ (0.488) (1.477)"
In house no flush 0.3333 0.9574 0.2509 5.864 -0.890 -2.021
(0.1377)** (0.3296)** (0.1659) (4.433) (1.598) (4.636)
Outside toilets 0.1715 0.3777 0.1867 2.535 0.114 -1.979
(0.0793)** (0.2267)* (0.1004)* (3.070) (1.134) (3.353)
Open pit 0.1285 0.3945 0.0293 5.542 -1.728 -3.311
(0.0893) (0.2731) (0.1235) (3.602) (1.322) (4.142)
No toilets -0.1662 0.2695 -0.1371 4.988 0.285 -4.195
(0.1196) (0.3327) (0.1522) (4.478) (1.561) (5.057)
Very little excreta 0.0162 0.1549 -0.0076 2.960 -0.290 —2.566
(0.0634) (0.1581) (0.0811) (1.963) (0.656) (2.064)
Some excreta -0.1543 -0.4335 -0.0666 -2.321 1.649 0.049
(0.0694)** (0.1885)** (0.0808) (2.287) (0.642)** (2.445)
Much excreta -0.9837 -1.3301 —1.4313 -16.275 -8.501 29.393
(0.4562)** (1.2628) (0.5170)** (14.240) (3.823)" (15.056)*
Community Yes Yes Yes Yes Yes Yes
Number of obs 3016 3016 3016 3016 3016 3016
R—squared 0.2121 0.1893 0.1846 0.219 0.253 0.268

P-value for testing coefficients equal to zero

Education 0.5665 0.0774 0.0557 0.0024 0.0014 0.0005
Prod Assets 0.3634 0.0148 0.3173 0.0170 0.6089 0.0198
Age dummies 0.0790 0.0363 0.1718 0.0240 0.0017 0.0116
Cohort dummies 0.1250 0.1939 0.0797 0.2399 0.0446 0.0844
Prices 0.4402 0.0153 0.1704 0.0002 0.0039 0.0027
Water source 0.2248 0.0006 0.1752 0.0002 0.0892 0.0001
Toilet type 0.0059 0.2560 0.0377 0.0360 0.2486 0.4726
Excreta 0.0264 0.0466 0.0217 0.1676 0.0050 0.1496
Community 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

 

Note: Also included in the models are individual-year age dummies, five-year-cohort dum-
mies and community dummies. Person-level robust standard errors are in parentheses. *
indicates statistical significance at 0.1 level and ** at 0.05 level.

estimated standard errors and makes it harder to establish statistically significant
relationships. Three, measurement errors in the changes in prices cause the estimates
biased toward zero. Finally the authors in Guo, Popkin, Mroz and Zhai (1999) did not
control for community level heterogeneity whereas there may be nonrandom program
placement or selective migration taking place (Johnson 2003, Yaohui 1999).

Due to the stagnation of grain production and rapid increases in demand, the
Chinese government implemented measures to promote grain production in 1988,

including increased purchasing prices. These purchasing prices, which may have more

152

impact, are not available in the data set. Less than one hundred households moved in
1993 and most of them remained in the same community. The 24—hour recall estimates
of food consumption may include intakes happening in a different community with
different prices. All these caveats should be investigated further.

Changes in water sources at the community levels do not have significant im-
pact on changes in outcomes, but the effects of changes in toilet types and excreta
measures in a community remain strong. For men when the neighborhood has more
household with little excreta there is an increase in calorie, fat, protein intake and
decrease in percent calorie from carbohydrates. Similar results hold for women. There
is no statistically significant difference between the parameter estimates for men and
women. Men and women respond to environmental factors in similar fashions.

As we can see, not all determinants of the levels of nutrient intakes are significant
determinants for the changes in nutrient intakes. When designing a policy the above

results should bear some weight.

153

Table 2.12: Food Consumption in CHNS 89,91,93 with Community Characteristics

in Rural Areas

 

 

 

Rural Male
Calorie Fat Protein % Calorie From
Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Some primary edu 0.0098 0.0887 0.0150 0.721 0.061 -1.417
(0.0159) (0.0482)* (0.0190) (0.474) (0.145) (0.617)**
Primary 0.0035 0.0696 —0.0024 0.757 -0.098 -1.088
(0.0161) (0.0508) (0.0196) (0.496) (0.149) (0.635)*
Middle school 0.0012 0.1881 0.0175 2.061 0.220 -3.236
(0.0161) (0.0508)” (0.0198) (0.508)M (0.151) (0.664)”
High school -0.0179 0.2009 0.0094 2.362 0.365 -3.851
(0.0193) (0.0590)M (0.0233) (0.606)M (0.185)“ (0.792)M
Tech/College+ 0.0147 0.3127 0.0348 3.531 0.344 -6.479
(0.0299) (0.0807)** (0.0367) (0.915)" (0.356) (1.296)“
Log R prod assets 1 0.0127 0.0387 0.0171 0.260 0.055 -0.365
(0.0027)** (0.0071)** (0.0031)“ (0.086)" (0.027)M (0.108)M
Log R prod assets 2 -0.0082 -0.0139 -0.0074 -0.031 0.021 -0.048
(0.0026)** (0.0072)* (0.0031)** (0.082) (0.024) (0.105)
Land 0.0024 -0.0034 -0.0003 -0.102 -0.040 0.113
(0.0011)** (0.0033) (0.0013) (0.032)" (0.010)** (0.043)”
Log Price of Rice -0.0713 -0.1294 -0.0880 -0.223 -0.130 1.988
(0.0204)M (0.0617)** (0.0253)** (0.467) (0.162) (0.611)**
Log Price of Pork -0.0123 0.2089 0.0657 0.547 0.148 -0.773
(0.0322) (0.0989)** (0.0394)* (0.156)” (0.050)** (0.206)*"‘
Log Price of Eggs -0.0195 -0.033O -0.0310 —0.032 -0.075 0.014
(0.0046)** (0.0141)** (0.0056)** (0.074) (0.024)** (0.094)
Log Price of Oil -0.0439 -0.0913 -0.0451 -0.206 0.003 0.046
(0.0271) (0.0866) (0.0333) (0.158) (0.054) (0.191)
Water Source 2 0.0240 0.2898 0.0607 3.486 0.516 —3.341
(0.0292) (0.0949)** (0.0359)* (0.961)M (0.294)* (1.186)**
Water Source 3 0.0816 0.2600 0.1603 1.261 1.108 -3.047
(0.0401)** (0.1300)** (0.0501)** (1.373) (0.375)” (1.706)*

 

2.6 Results: Production Function

Table continues

In Chapter ‘1 we find strong persistency in a person’s health status and it is quite
reasonable to assume that lagged BMI is a good summary statistic of all past inputs
and information. Here we estimate a simple health production function with health
being measured by log weight or log BMI in 1993 (Table 2.20).

Lagged health (lagged log weight and height, or lagged log BMI in 1991), current
physical activity levels and nutrient intakes are included and treated as endogenous.

Also included in the regression are individual year age dummies. The identifying in-

154

Table 2.12 (cont’d)

 

 

Rural Female
Calorie Fat Protein % Calorie From

Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Some primary edu -0.0098 0.0263 -0.0058 0.322 0.064 -0.332
(0.0105) (0.0356) (0.0136) (0.362) (0.117) (0.411)
Primary 0.0128 0.1477 0.0369 1.417 0.340 -1.827
(0.0117) (0.0412)** (0.0154)*"‘ (0.429)” (0.133)** (0.494)"
Middle school -0.0193 0.1765 0.0174 2.056 0.499 -2.520
(0.0122) (0.0405)** (0.0159) (0.429)“ (0.138)“ (0.488)"
High school -0.0305 0.2101 0.0103 2.854 0.530 -3.382
(0.0191) (0.0600)** (0.0243) (0.657)" (0.184)” (0.738)"
Tech/College+ -0.0084 0.2823 0.0452 3.793 0.788 -4.420
(0.0317) (0.0911)** (0.0428) (1.185)** (0.429)* (1.301)“
Log R prod assets 1 0.0044 0.0173 0.0087 0.057 0.056 -0.141
(0.0023)* (0.0065)** (0.0029)** (0.077) (0.026)“ (0.087)
Log R prod assets 2 -0.0013 0.0022 -0.0003 0.135 0.026 -0.154
(0.0022) (0.0065) (0.0028) (0.076)* (0.026) (0.085)*
Land 0.0015 -0.0024 -0.0014 -0.054 -0.046 0.103
(0.0009) (0.0032) (0.0011) (0.037) (0.010)" (0.043)"
Log Price of Rice -0.0315 -0.1038 -0.0893 -0.140 -0.568 0.710
(0.0187)* (0.0580)* (0.0239)** (0.463) (0.154)" (0.509)
Log Price of Pork 0.0048 0.3062 0.1044 0.644 0.206 -0.813
(0.0285) (0.0946)** (0.0371)** (0.153)“ (0.048)“ (0.169)“
Log Price of Eggs —0.0128 -0.0221 -0.0203 -0.081 -0.030 0.090
(00042)" (0.0135) (0.0053)** (0.070) (0.023) (0.078)
Log Price of Oil -0.0429 -0.1070 -0.0273 -0.246 0.054 0.231
(0.0236)* (0.0798) (0.0304) (0.156) (0.049) (0.164)
Water Source 2 0.0449 0.2498 0.0566 2.645 0.151 -2.719
(0.0264)* (0.0944)** (0.0357) (0.961)" (0.303) (1.087)”
Water Source 3 0.0746 0.1298 0.0916 1.017 0.284 -1.884
(0.0419)* (0.1330) (0.0517)* (1.352) (0.390) (1.489)

 

struments are education dummies, real productive asset splines, land cultivated and
community characteristics including prices, water and sanitation conditions in 1991.
For details and the goodness-of-fit of the first stage regressions see Appendix F. There
is no weak instrument problem (Bound et al. 1995). The strong evidence of het-
eroscedasticity in male regressions suggests asymptotically correct inferences should
be made using the Huber-White ”sandwich” robust variance-covariance matrix. The
overidentification x2 tests are heteroscedasticity-robust, based on Wooldridge (2002).

The Wu-Hausman tests on the endogenous variables between OLS and 2SLS are

carried out.

In OLS estimates, the effects of lagged log weight for men and women on current

155

Table continues

Table 2.12 (cont’d)

 

 

 

 

Rural Male
Calorie Fat Protein % Calorie From
Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Water Source 4 -0.0385 0.0901 0.0249 1.048 0.883 -1.350
(0.0300) (0.0980) (0.0390) (1.106) (0.327)” (1.307)
In house no flush 0.1706 0.0440 0.1659 -2.670 -1.069 6.881
(0.0716)** (0.2172) (0.0954)* (2.507) (0.920) (2.986)M
Outside toilets 0.0378 -0.1742 0.0222 -3.863 -0.919 6.920
(0.0603) (0.1558) (0.0821) (1.948)" (0.839) (2.279)M
Open pit -0.0122 -0.1808 0.0118 -3.172 -0.316 5.403
(0.0567) (0.1457) (0.0774) (1.901)* (0.816) (2.219)**
No toilets -0.0964 -1.0618 0.0312 -9.855 1.345 18.776
(0.1218) (0.4210)** (0.1693) (4.419)“ (1.531) (5.601)“
Very little excreta 0.0905 0.1361 0.1215 0.551 0.444 -0.725
(0.0262)** (0.0778)* (0.0320)M (0.793) (0.254)* (1.018)
Some excreta -0.0310 -0.1620 -0.0017 -1.142 0.512 1.664
(0.0249) (0.0735)** (0.0308) (0.758) (0.264)* (0.969)*
Much excreta 0.0514 0.3447 0.0038 -2.357 -0.642 -1.724
(0.1039) (0.2996) (0.1224) (2.652) (0.846) (3.666)
Community Yes Yes Yes Yes Yes Yes
Number of obs 5688 5688 5688 5688 5688 5688
R-squared 0.2355 0.2410 0.2300 0.282 0.292 0.319
P-value for testing coefficients equal to zero
Education 0.4886 0.0019 0.7070 0.0000 0.0175 0.0000
Prod Assets 0.0000 0.0001 0.0000 0.0001 0.0004 0.0000
Age dummies 0.0030 0.3526 0.0660 0.1572 0.0965 0.3658
Cohort dummies 0.1665 0.4307 0.4466 0.1901 0.8196 0.6653
Prices 0.0000 0.0091 0.0000 0.0141 0.0010 0.0004
Water source 0.0344 0.0105 0.0151 0.0032 0.0077 0.0197
Toilet type 0.0439 0.6214 0.5829 0.1957 0.1476 0.0063
Excreta 0.0006 0.0285 0.0041 0.1794 0.1255 0.1303
Community 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

Table continues

weight are significantly less than 1, the effects of moderate or heavy physical activities
are negative, and the joint effects of all nutrient intakes are significant for men at
0.05 level and for women at 0.1 level. In the 2SLS estimates, however, the effect
of lagged log weight for men is not significantly different from 1, and for women it
is only marginally significant at 0.1 level. Conditional on lagged weight and height,
current physical activities and nutrient inputs do not have any significant impact.
The coefficients for nutrient inputs are generally larger in 2SLS regressions, indicating
attenuation biases in OLS regressions.

The fact that the coefficients for lagged weight are one suggests that in a short pe-

156

riod of two years from 1991 to 1993, adult weight can be modelled as a random walk
process. The Wu—Hausman F -statistics indicates significant differences on endoge-
nous variables between OLS and 2SLS estimates. The overidentification tests lead
to the conclusion of validity of instruments assuming that eight of all the identifying
instruments are exogenous.

The results for log BMI in 1993 are very similar to the results for log weight.

BMI can also be characterized as a random walk process in a short period of time.

157

2.7 Discussions

Using an unique longitudinal data set of adult Chinese in the early 19903 we esti-
mate the reduced demand function for calorie, fat and protein intakes and for percent
of calories from fat, protein and carbohydrates. We also study the determinants for
the changes in these outcomes. The socioeconomic factors that are significantly re-
lated to nutrient consumption may not be related in the same way to the changes in
nutrient consumption.

The theoretical framework of this paper is the same as in Luo (2003b) and similar
to Foster (1995). The key findings of this paper are that education does not have
significant impact on calorie intakes but does affect percent of calories from fat, from
protein and from carbohydrates differently in different region and at different age.
The effect of productive assets is nonlinear and in inverted U-shape for male calorie,
fat and protein intakes. In rural areas the effect of productive assets is stronger than
that in the urban areas. At different ages the effects of education and assets are dif-
ferent. The effects of prices on calorie, fat and protein intakes and the quality of diet
measures can go either way. Improvements in sanitation are associated with more
energy and protein intakes in urban areas. Changes in prices and community charac—
teristics are related with the changes in nutrient consumption in different ways. The
difference between urban and rural areas does not seem strong. There is a trade-off
between including community dummies or not when we use current community char-
acteristics in the model. The advantage is community dummies capture unobserved
heterogeneity at the aggregate level. The disadvantage is any effect of community
variables will have to be identified through the variation of them over time and in a
short run the change may be too little for good identification.

There are some limitations in this paper. First, using the productive asset in the
reduced model as a proxy for income and treating it as exogenous is not without crit-

icism. In a dynamic model nutrient intakes can affect productivity and therefore the

158

holding of productive assets in later periods. If there was selection into more stren-
uous vocations with higher wages, the results will be biased upwards. Attenuation
bias could also rise with imperfect measures of productive assets and land.

The model is based on the assumption of perfect insights. Ignoring uncertainty
is not innocuous when there are risks other than idiosyncratic shocks. Uncertainty
would be important if overweight increases risk of mortality and morbidity and people
are well aware of it. The length of individual life time and the household size are all
assumed to be exogenous whereas in reality both are potentially endogenous.

Food prices may be endogenous in the augmented models. It was plausible that
food prices modify household income constraint and the estimated price effect may
be biased.

The paper focus on macronutrients instead of specific food consumptions. Effec-
tive policies may be a little difficult to deduct from here due to the ambiguity of the
effects of many of the determinants.

The corner solutions and loss to follow-up in the interview of three-day dietary
recall are not considered in this paper.

Keeping these caveats in mind this paper finds that certain individual macronu-
trient intakes are strongly correlated with education levels and household resources.
Food consumption is very sensitive to food prices and the direction can go either way.
In designing food price or subsidy programs the local government should start with

those less sensitive factors and proceed with care.

159

Table 2.12 (cont’d)

 

 

 

 

Rural Female
Calorie Fat Protein ‘70 Calorie From
Dependent Var Log(kcal) Log(g) Log(g) Fat Protein Carbo
Water Source 4 -0.0034 0.0578 0.0129 0.256 0.300 -0.801
(0.0285) (0.0956) (0.0379) (1.063) (0.323) (1.175)
In house no flush 0.2081 0.2600 0.2113 0.352 -0.914 0.775
(0.0738)** (0.2281) (0.0995)** (2.689) (0.884) (2.995)
Outside toilets 0.0654 -0.1243 0.0271 -2.682 -0.981 3.715
(0.0550) (0.1417) (0.0710) (1.849) (0.712) (2.041)*
Open pit -0.0037 -0.1876 -0.0210 -2.909 -0.638 3.309
(0.0526) (0.1349) (0.0682) (1.825) (0.683) (2.010)*
No toilets -0.1042 -1.0122 -0.0059 -14.044 0.778 9.995
(0.1168) (0.3646)** (0.1503) (4.172)" (1.351) (4.705)"
Very little excreta 0.0393 0.1206 0.0723 0.459 0.505 -0.923
(0.0240) (0.0757) (0.0301)** (0.777) (0.240)" (0.863)
Some excreta -0.0089 -0.1554 0.0080 -1.582 0.349 1.174
(0.0220) (0.0711)** (0.0285) (0.718)** (0.248) (0.810)
Much excreta -0.1827 0.1212 -0.0106 -0.056 1.907 -3.024
(0.0924)** (0.2903) (0.1127) (2.645) (0.845)“ (3.014)
Community Yes Yes Yes Yes Yes Yes
Number of obs 6146 6146 6146 6146 6146 6146
R-squared 0.2635 0.2517 0.2335 0.288 0.306 0.336
P-value for testing coefficients equal to zero
Education 0.0266 0.0001 0.0768 0.0000 0.0015 0.0000
Prod Assets 0.0757 0.0586 0.0001 0.0037 0.0002 0.0000
Age dummies 0.0004 0.0232 0.0000 0.8066 0.0000 0.0000
Cohort dummies 0.0724 0.7805 0.3657 0.7650 0.0869 0.4639
Prices 0.0002 0.0039 0.0000 0.0010 0.0000 0.0001
Water source 0.0299 0.0055 0.0486 0.0296 0.7995 0.0690
Toilet type 0.0001 0.0044 0.0486 0.0208 0.4874 0.2257
Excreta 0.0674 0.2190 0.1194 0.0788 0.0230 0.1273
Community 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

 

Note: Also included in the models are individual-year age dummies, five-year-cohort dum-
mies and community dummies. Person-level robust standard errors are in parentheses. *
indicates statistical significance at 0.1 level and ** at 0.05 level. The percent of households
with the following water sources: (1) underground, (2) open well, (3) spring, river lake,
rain or snow, and (4) water factory. Omitted is the first source. The omitted category for
excreta is percent of no excreta.

160

mvflflmua—OU Omfidrfi

 

 

 

 

 

888 888 88.0 3.38 338 £88 858% 855800
888 835 88.8 :88 :88 3.8.8 858% 8880
58.8 393 so”; 888 $88 85.8 858% a?
£88 858 $88 888 5.88 883 888..
833 «was 8.58 838 $88 $88 5583
:oEoB 98 .88 :853 $583 5: 83.87am
888 888 888 888 888 888 858% 85888
NE; Ema gas 3st ~83 828 858% 8280
$88 $88 8:8 8...; 888 838 858% 8.4.
38... $58 888 as; 88.5 888 888-<-
wwofio 288 888 8:3 $88 388 5583
Dawn 0... 3:30 wadomoavoo manna... HOH 2:97AM
was 2.2 memo ammo 8&8 53 88:88
am 82 5mm 88 88 on? 828580 8 .oz
8% 8% 84% we? mo> 8% 388:: 80%... 3238800
$8.8: 35.8: 88.: can: 85.3: $8.:
88.:- 83 55.. EE. 88.: £3- 53
some: :88: $3: 82.3: :28: 8%.“:
88.2. «new- 8.3- 23- «8.8- 888. m 888 58 88 was
5:: 88.5 38.3 :83: :88: 888:
89%. «8.2 5.2 $2: 883 88...: H 888 58 .88 won
:58: 88.3.: 888...: 25.3: 33.2: amen:
88.2.”. 88.2 was: an? $5.3 88.2 88280389
ASE-m: $3.3: $8.5: $3.3: $3.2”: $38:
5.8: $33 «8.5 «SS 332 $88.. 588 .5:
28.8: 83...“: 28.5: 83.0: 68.8: :38:
832- 832 83: 83. 8883 3.28 .888 8:52
83.8: 88.8: 38.3.: A83: 38.5: 38.8:
232 :88. 2.2m SQ:- mmosm 83 .555
$38.8: 82.3: 33%: 88.8: 32.8: :35:
5.3m- 832 «8.8- 838- ~38 38.5- 5588 %885 .58
+8 83 $5 mm... 8.8 8.6. 88m 885 58.6 838.5»

882

 

 

 

8888 8580 38: 8 8550 83 8389

161

.858— mod

8 1. 8:8 :86— ~.o um wocdoficmfi 588538 88285 * .885839 E v.8 £88 @8988 «958
687.8885 .81556 floss Box-95 8:8 88556 88 08 9828me8 2: E 88305 8?- 8qu

 

 

 

 

 

 

$88 882 858% 8880
”$3 888 858% 88
$38 88:. 8884
888 5885 $88 82 5588
808 #9:: U3 dawn-u: G®®>$®£ wflmumwu no.“ 82:qu
888 588 588 888 888 858 858% 85588
838 $88 883 8.88 888 888 858% 8280
883 8.58 888 888 $38 883 858% 8<
:83 8de $88 82:. 858 83 888
at; «88 $88 882 $88 $88 5558
OuON Cu ~d5~v® 33205000 wcmuwva HOW 9:qu
52 ©de 83 83 23 82. 58:88
88 on: 38 8mm 28 88. 80888880 8 .oz
8% 8% 8% 8% 8% 8% 855:6 8%... 3258800
$8.8: $8.3 88.: $8.3 88.8: 88.3
838- 88.8. 83 «88 88.“ 83 53
A83: 888: 8.28.5 38.3 388: 188.8
88.5 :5:- 828. ~59- 53m- 8::- m 888 5_8 88 meg
38.8: C: 88:: $3: 83: 88.8
88.8- 888- 88.3 88.8 83 88.8 H 888 58 88 83
£8.85 288.8: 88.8: 83.8: 82.2: 88.8:
”388. 8588 88%- 88.8- 838 88.8 88502889
3888: 88.5: $8..on 25%: $3.5: 38.8:
882- 88.8 83. 888- 88.8 23 858 .85
38.8: 88.8: 8°.an 83.3 38.8: 38.8:
$888 88.3. 88.2 28.8 33. 85: 858 25:2
:38: 88.”: 88.8: 88.8: 8:.me 23::
8888- 88.8.- 98.8 85,..- 838 88.8- 888
38.8: 28.8: 68.8: $8.8: $2.8: 88.5
8.32- 98.8.. «8.: 88- 8.88. «a; 5555 888E 888
+8 84 38- 8< 818 88 888 .88: 885 83885
2588b

 

8.88: 2.». 289

162

8.53:8 2an

 

 

 

 

 

88.8 888.8 88.8 88.8 883 883 $888 E2888
«88 88.8 88.8 28.8 888 53.8 @888 :25
53 88.8 88.8 288 88.8 88.8 83.8% 03.
:83 88.8 838 88.8 88.8 888 3§<
32.8 88.8 8E8 88.8 :88 88.8 8883
G080? Uflm five :83”.va wflfiuau MOM 8=B>nnm
88.8 888.8 888.8 88.8 88.8 888.8 888% @8888
8:8 883 82.8 :88 28.8 8:8 8.88% ”.880
88.8 88.8 83.8 888 $88 88.8 8E88 .3.
$8.8 $8.8 58.8 88.8 338 E88 332
:88 83.8 88.8 58.8 88.8 88.8 838:3
OuON Cu 353 muflvmommvoo wcmuau qu 9;qu
v88 88 53 88 83 Ed 88:38
am 82 $8 88 82 88 maosatofio 8 .oz
8N» 8> we? _ mo> 8% 8% _ A8839 3?»... zficsgoo
38.: $8.8 38.8 28.8 $3.3 $8.8
~84 83 80.8 83 :3 3:8 23
88.3 88.3 $8.8 82.3 3.8.: 88.8
88.“ 88.8. 88.8- 8nd- ES 83.8- a m 338 8a 22 m3
$8.3 98.: $5.8 $8.3 $2.3 83.8
23- 83 83 «88 88.8- 83 L 383 Ba as 84
......38.3 88.8 $8.: 88.: :85 38.3
3.3- 88.3- 8; 83- £2 23- +owm=ooE$e
38.5 68.5 $8.8 38.3 88.: 33$
8.8.3 83 “8..“ 83- m8.” 85- 328 Em
83.“: $85 $33 98.3 38.: 38.3
:32 88.8- 83 88.”- 23 83- .8an 283
38.8: 38.8 $8.8 38$ $85 83.3
83 83- «8.8 83- $3 821 ESE
88.3 8:3 88.8 38$ $8.5 $8.3
88.2- 88.8- E; 88.? 83 83.- 838:8 ESE 28a
+8 om< 88 a? 88 m.3. 35m 985 .1226 853:5

2&2

 

$33 3m ban a $8.20 83 28.8

163

496— mod

3 t." 98 ~96— Hd um wagoflcwmm 388.388 8883?: * .8838083 5 28 88.8 @8988- 8959
567:8qu 6628826 80:00 398.96 and 83:56 own 03 8888:?“ may 5 vows—05 em; 5qu

 

 

 

 

 

 

88.8 88.8 888% :28
88.8 28.8 85.8% 88
88.8 38.8 8.82.
88.8 8803 88.8 8: 858%m
mamas 19:: Ufld Gaga: Gggvfi wdmawOu up“ maid-87%
88.8 888.8 88.8 88.8 888.8 88.8 888% @2888
8:8 88.8 888 88.8 888.8 82.8 888% :28
3.8.: 82.8 :88 88.8 :88 85.8 888% 8E
28.8 88.8 88.8 88.8 88.8 88.8 8884.
88.8 88.8 88.8 8:8 88.8 83.8 858:5
OHQN O» 3360 mafimmoamonv wamummu ho.“ Qua—gum
88.8 82 83 88 Sad 83 88:88
8” 83 88 88 £2 88 828880 8 .oz
8% 8% 8.». mo> mo.» 8% 2356 39m... 5328800
288.: 88.3 88.3 $88.3 ...883 1.88.3
83- 88.8 82.8 «8.8 83” 28.8 83
83.3 88.3 88.3 88.3 88.3 88.3
38.8- 83 Ed- 88.8- 88.8 ”8.8. . a 888 Ba 82 83
8%.: 88.3 88.3 88.3 22.3 88.3
22.- 58.8- 83- .888- 88.2- 28.8- .2 888 83 32 82
88.3 88.3 88.3 83.3 2.8.3 88.3
88.8- :3. 88.” 2.8.8 82.. «88 8880888
28.5 88.3 88.3 88.3 88.3 88.3
858. 83- 88.8- 23- 888- «8.2- 328 85
88.23 88.3 82.3 83.3 88.3 83.3
88.8 83- 88.8- 88.8 83- 88.8- 388 28:2
188.3 82.3 28.3 83.3 82.3 88.3
82.8- 282- 82..” ES 88.2- 88.8 58:8
2.8.3 88.3 83.3 88.3 88.3 28.3
88.8 83.2- 83- 83.- 8w.m- 88.2- 82888 SEE 888
+8 8:. 88 83. 88 83. 85m 535 885 833.25
03:85

 

8.883 EN 8388

164

8:588 2nt

 

 

 

 

2.85 88° 885 $53 $85 $23 fiaaé 582.800
8:5 886 3:3 83o ammo gas @883 :28
«$3 88.: ammo ”82. 83c 35.9 85.8% ow<
$85 23o E3... ”5; 83d god 382
8:8 megs 33o 3w; 2:3 386 858:8
G080? mug five G8E0£ wfiaméu uOu £25qu
886 886 885 88d 885 885 8.88% @5888
Sad «.83 88¢ 93:. :23 :86 82:83. :28
293 $3 ”82. 385 £86 885 8.88% a3
83° 88: 3mg «wwwd $38 885 $3.2
$3.0 88d 236 as; ~83 88d 858:8
08s 3 3:3 356508 maxim... .8“ 2:97am
83. and memo 83 fig :2. 853$
3m 82 $3 88 82 on? 223283 mo .02
8». 8? m5 8% 8% 8% 33:6 80»... 5828800
2%.: 28.8 $3.8 28.8 A399 568
m5? $3- 82.. Bed. 33 08.? was
$5.3 $8.: :38 68.8 gm: 88.8
35 «85. «and- mag- 53. ”as. u N 3me Ba 38 wag
and £53 338 $8.8 $2.5 83.8
$3- ammo- «Ed Rod as: 2.2. L 383 Ea 32 m3
$8.3 83.8 385 €53 38.8 633
302- $3- 83y 2.3 23 93.9 +mmm=ooE89
85va $3.8 $5.8 Agni 38.8 58.9
$3 2...: £2. 23. «m3 8% 328 :me
88.3 58.3 33.3 $3.3 $8.8 gas
83- £3 £3 can.” 83 83 323. 22::
$me 82.3 33.9 88.3 3.3.3 88.3
8a.... ”and- $3 33 83 83 .585
385 $5.3 $8.8 $3.3 $3.8 93.3
2?. :3 $3 $3 33 2.? 838:? ESE 28m
+8 ow< $3 a? 3-8 03. 35m :85 $85 833:5
234

 

 

 

8.35 £38m .33 a 33:0 53 made

165

496— mod

58 3.. «:8 3%: ad ad 898%?me 1838883 85865 ... .885538 E 38 £88 BEE-3m $399
367:8qu 888828 .tonoo Box-96 98 833:6 ems m8 mnemguwmu 05 E 89:05 Om?- ”302

 

 

 

 

 

88.8 8:8 888% 5880
88.8 82.8 888% 88:.
88.8 88.8 882
28.8 883 £88 8: 858%m
mama :23 98 555 c8383 $58» now 8297a
8888 88.8 88.8 88.8 88.8 88.8 888% 88880
88.8 8:8 :88 888.8 88.8 88.8 888% 880
88.8 88.8 88.8 88.8 88.8 88.8 888% 83
88.8 88.8 828 8.2.8 88.8 88.8 883.
88.8 88.8 888.8 38.8 88.8 28.8 858%m
OHON Cu #653 BGQMUEOOU mdmuwou HO.“ 2:qu
858 88.8 88.8 88.8 88.8 83 8888.8
88 on: 888 88 22 88 8585880 8 .oz
36> we.» 8> 36> mo> 88> 88856 88%.." hamcsaaoo
88.8 88.8 88.8 88.8 88.8 88.8
88.8 88.8- 88.8 88.8 33- 888 83
88.8 88.8 188.8 88.8 88.8 83.8
88.: 888- 82- 88.8- 888- 88.8- .. 8 888 88 88 83
88.8 88.8 238.8 88.8 88.8 83.8
$3. 88.8- 33 82.8 88.8 8:8 L 888 88 82 83
88.88 :88 8:8 88.88 83.8 88.8
88.8- 83 888- 88.5.. 55- 82.:- 8880888
88.88 88.8 888.8 :88 88.8 88.8
88.8- 88.8 :5- 888- :3- 83- 888 85
83.88 88.8 83.8 88.8 82.8 88.8
:2: 88.8 88.8- 83 88.8- 88.8- 88 2%52
82.38 83.8 82.8 88.8 88.8 88.8
88.:- Sqm- 85 88.:- 83- 88.:- .588
88.8 83.8 88.8 88.8 88.8 88.8
88.8- 83 88.8 83 :3- 88.8- 88838 88:8 8:8
+8 83 88. 83 8888 8< 88m :85 :85 838:3

@7255

8.88 3.8 8.88

166

8:588 0388

 

 

 

 

 

 

88.8 888.8 88.8 888.8 888.8 88.8 8888 888880
28.8 888.8 88.8 888.8 888.8 883 888:: 880
888.8 888.8 88.8 888.8 823 88.8 8888 88
888.8 888.8 888.8 888.8 888.8 88.8 :88
383 888.8 88.8 8.83 88.8 883 858:8
£0803 fig Q08 $008503 mad-age and w0=~w>anm
88.8 888.8 88.8 88.8 88.8 88.8 8888 888.80
888 E88 883 8.8.8 883 888.8 8888 880
883 888.8 888.8 82.: 883 888.8 8888 :8
88.8 888.8 8.8.8 883 888.8 888.8 888
82.8 883 88.8 888.8 883 888.8 858:8
OHON Ou 3560 mud0m0mm000 wamuméu no.“ 03—97%
88.8 88.8 83 883 83 883 88:88
88 82 88 8888 88: 888. 8:58:80 8 .oz
8% 8% 8% 8> 8% 8.» 858:8 .80»... 3838800
888 82.8 88.8 88.8 82.8 88.8
88.8 88.8 :88- 888 88.8 88.8 83
82.8 88.8 82.8 82.8 888.8 88.8
88.8 83 883 88.8- .83 88.8 8 888 88 8: 83
88.8 88.8 82.8 88.8 88.8 82.8
88.8- 83- 88.8- 388- :3- 8.8.8- H 888 88 8: 83
...888 888.8 88.8 88.8 888.8 88.8
888- 888- 83: 88.8- 83- 888.:- 8880889
88.88 888.8 888.8 88.8 :88 888.8
88.: 8:; 83.8 23- 83 88.8- 8:8 88
88.8 88.8 88.8 88.8 88.8 88.8
88.8 88.8- 888 :3- 888 83- 8:8 28:2
88.8 88.8 888.8 888.8 83.8 88.8
88.8- 88.:- 83 83- 88.8- 83- 8:88
88.8 88.8 888.8 88.8 888.8 888
28. 88.8- 888- 8888- 83 88.8- 8888 .888 288
+88 88 888. 88 88.88 08 8:8 :85 :85 83888,

2:2

 

 

8h 88m 2880 38Q 888m 5 mama-8:0 6H.m 8389

167

.56— mod

an I 98 W52 ad 3 monmofiqwmm $059.36 88863 * .88ficoaa 5 2w $8.20 Edvasm “£50.—
_o>o_.:ofiom .8_88=v 28:8 Boméé can 828826 emu was 982.86%?“ 05 E 39:05 82 ”302

 

 

 

 

 

 

Emma 38¢ £838 328
.53 «33 828826 mm<
8.85 :83 39.2
:83 88o? 838 82 8:853
macaw 3.2; mug Edam: £82503 main—8a “Om 8=~w>um
885 888 885 888 886 885 8§a€ 382.800
853 888 3:3 Sand 88¢ 82$ @28st 8930
Eng 2%? $88 835 388 8de 828826 $2.
838 8.3.0 83¢ $86 «$2 8.8.0 382
28.2. 338 $33 $33 $38 28:. 838:3
Emu 3 335 3220580 wcsmou 8m mggd
was and mag 83 22 a; 83:38
8” 83 3mm 8mm 22 ...me 823530 2o 52
8% $5 mew mo> 8> 8> 295% Box... 3235800
83.8 232.8 38.8 $38.8 $288 $2.88
:3- £3 83 $3 83 88 23
28.8 28.8 22.8 22.8 2.5.8 228
£28- mm; 28.0. $3.- and 88.? a 39.8 Ba 32 m3
238 2.3.8 228 22.8 238 32.8
and £3- wad- ES- 82.. 8.3- 2 383 Ea we woa
23.8 28.8 88”.: 2%.: 2.2.: 88.:
Si”. 23- 33 33 $3 $2 $330289
33.8 23.8 23.8 88.8 88.: 23.8
88.8 2.8%- 83.. 82.2- ”mm? 83- 328 in
2.2.8 8%.: 2.88 28.8 2.2.: 63.8
23. 22.- 82- «mod- 33. 33- 32% @832
£838 88.: 2.38 28.8 23.: 838
83- 85.8 $3 33 $3- «mad 585
88.8 28.8 85.8 28.8 38.: 2.8.8
8“.“ and- 83- ”mod 33. 23- 838:? ESE 88m
+8 my? $8. $3 $2 ow< 33m :85 zego 833:5
o—mth

 

8.288 8a «3.2.

168

8:55:00 2an

 

 

 

 

 

885 885 885 835 as? $85 amaasn $2588
386 came 3.35 mag 2:3. 835 865% 5:8
3:6 £93 883 ~23. 883 8:5 8:883. a?
£33 Ego 83d :23 ”id 383 $82.
223 S33 88¢ ma; 8on $33 838:3
G039? Us G08 flggvfl wdmawmu uOm 8=~d>nam
2.85 88d 885 885 885 885 8:886 @2588
3:3 $8": 88.: 885 £36 885 85.8% :25
£85 £85 82d 88d 2:3 385 86.8% .3
83.0 :83 $23 326 Sm; £25 303.
£83 38d 885 £86 825 .88: 858:8
OHQN Cu qua—GO manomoewoo wdmuflwu HO.“ osgunm
$3. £2 82. 32 23 E3 88:3»:
3” 82 $3 88 82 88. 282520 .o .02
8> 8% 8> mm> 8> 8.? >883: .25»... 5858800
:28 338 38.8 :88 62.8 :38
3a.? god- 83. 33- $3- 33. 33
:88 38.8 :88 83.8 :28 83.8
38¢ 23.. Sod- 83. 35.8 god- m 3&3 Ba we we:
:38 88.8 $8.8 $8.8 $8.8 :38
83. ES- Sod- 93.8 wood Sod- : 30.0.3 Ba 32 we:
88.: 88.8 :38 $88 $8.8 83.8
£3- 33. 83 wood. 88.? 8.3- +omo=ooE8e
838 :88 :38 82.8 :38 :58
wmvd a; 83 23 2:3 88¢ 328 in
:28 L388 :38 68.8 33.8 :38
33. S3. 33 83: 83 was _8€m 22:2
888 :88 :38 88.8 83.8 :38
v98 32. 33 23 58. $3. SEE
3&8 :38 :88 :38 88.8 :38
£3 £3 33 £3 88¢ 82 8:835 mafia 2:8
+8 3:. $8. om< 83m 03: 35m 555 =§>o 85ams>
232

 

 

538m 88m 2860 EEG unmoumm E mowfifio ”5d 2nt

169

.68: mod

3 .1. «Ed :52 ad ad mocaoflawfi 3052.88 8886.: ... .3588: a. PE 828 33:88 9859
6.67:8qu 6888:: €038 8989a «28 835:6 own 08 £588:me m5 E 8:32: ofl< 5qu

 

 

 

 

 

 

88... 88... 8.88% 880
88... 88... 8.88% mm...
88... 88... 888..
88... 8803 :8... 82 8.88.8.
€080? US: G08 :83umfl wflmumoa mo“ wozgunm
88... 88.8 888 88... 888... 88... 8.88% 8.8888
2.8... 28... «8.... 88... 88... 82... 8.88% .880
:8... 82... 88... 88... 88... 88... 8.88% mm...
:8... 88... 88... 83... 88... 88... 882
88... 28... 88... 88... 88... 28... 8.88%:
05: 3 158 8.5.0508 $58» :8 2:87am
28 2:... 2:... .8... 8.8 :2... .8888:
8: 8... 88 88 28 88 88.8880 8 .oz
3% 8% 86% 88> 8% 8% 88888 .89»... 3888800
:28 :88 :88 :88 :88 :88
.8... 8.....- .8... 88... 8.8- 8.8 .88..
:88 :88 :88 :88 :88 :88
2.... .88- 8.....- 8.....- 2.8 .88- m 888 88 :8 8..
:28 :88 :88 :88 :88 :88
8.8- 8.8- .8... 8.8 .88- 8.....- . 88.8 88 .88 8..
:88 .._:8.8 :88 :88 :88 :88
8.....- 8.....- 2m... .88- :88- 8.....- 88.8088:
:88 :88 :88 :88 .888 :88
.88- :8... 8:8- 8:8- .8..- 8.....- 82.8 8.8
:88 :88 :88 :88 :88 :28
8.8- :8 8:8- 8.... 83. 8.8- :88 8.8.2
:38 :88 :88 :88 :38 :28
.8..- 88...- :8... 8.....- .28- 8.8- 888.5
:88 :88 :88 :88 :88 :28
.8... .28- 88... ...8... .88 8.....- 8.82.8 88.8 888
+8 8.... 88 mm... 88 8... 88¢ 835 .885 8.8.8.,

2850,.“

 

8.888 2.: 8:8

170

83.5.80 8an

 

 

 

 

 

88... ..8..... ..8..... 8...... 88... 8...... 8.88...V 88:88....
8...... 88... :8... ..8... 8:... 88... 8.88.... 82.00
3...... 8...... 88.. :8.... :8...... 8...... 8.88.... 8..
88.. :8... 88... 88... 88... .8... 88.8
22... 8...... 8...... 88... .2... 8...... 82.88...
8803 8.... .88 :838: 8.38.. .8 838.7%
88... ..8..... 88.. 88... ..8.... 88... 8.88.... 8.8888
:8... 8...... 8:... .88... 8...... 8...... 8.88.8 89....
88... 8...... 88... 88.. 88... 88... 8.88.8 8...
8...... :8... 8:... 88... :8.... .8... 8......
28... 8...... 8...... 88.. 8...... 88... 8.88....
08: 0... 8:8 8:88.:on mew-.8. ..8 wig-m
8...... 8m... :8... 8:... 2:... .8... 88.8....
.8 ..8. 88 88 8.... 8.... 82.8.8.5 .o ...z
8% 8% 8% 8% 8% 8% >883. .88... 38:35:00
:8.... :2... :2... :8.... :8... :8....
:8... 8....- ..8... .8..- .m.... .8...- .8...
:8.... :8.... :8... :2... :8.... :8....
8.....- 8.... 8.... ..8... .8...- 8..... m 8.8.. .5... .8. 8..
:8.... :8.... :8.... :2... :8.... :8...
8.... 2.... 8.... ..8... 8.... 8.... . 8...... .5... .8. 8..
::8.... 1.88:. :3... .88. :88. I:2...
.8... 8:... 8....- ..8: :8: 8...: 88.8038...
:8.... .88. :8... :8... :88. :8...
8.....- 88- :8...- ..8... 8:...- 88... .8... 8.:
:8... :8... :8... :8... :8... :8....
:2..- .8... .2...- 88 .8...- .8... .8... 2.8....
:8... :8... :8... :8... :88. :8....
8.... 8.: 2.... 8... 8.... ..8. .88....
:8.... :8... :8... :8... :88. :8....
:8: .8... 8...... 8.... 8.... 8.... 8.88.... .88.... .88
+8 8.: 8.... 8... 8.8 8... .83. 8...... :8...... 838:5
..8...

 

 

 

88.830380 808 8.880 EEG 2.88.. E 898:0 ”mad 8an

171

49,3 mod

3 ..z.” 98 823 ad ...a 852%:me Eosmsgm 330:3: * .885839 E 03 Echo “~36:an amzne
357:8me 663326 9838 8958;“ 98 83356 ems 08 mcofimfiwmu 23 E vows—2: 82. ”8oz

 

 

 

 

 

 

as; ~23 8:258 828
2mg 83¢ 883% $3
$8.0 £88 $82
88.0 8895 885 82 858:3

macaw Eh: cam 535 5.0393 wfiuwou 5w 83w>¢

888 888 888 888 885 888 fleeé £58800

8:; $38 «.23 $35 ”Sod 886 85.8% :28

$88 on: :55 Same 835 $85 £883. “3

885 8mg 8a; 836 835 :85 3&3

888 823 $85 $88 $53 ammo 838:8

ohms o» Bade mfiBBEmOo mesmva 8% “53.7%

$3. 83 $3 38 3g 88 8335

m8” 83 3% 8mm 22 £8 223530 3 .oz

8% 8% mm; m; 8% 8% @556 89%... 3258800

$8.8 3:8 2:8 $38.8 3.33.8 $388

$3 Sad. 33. 33- $3- 88- 83

88.8 38.8 32.8 33.8 88.8 23.8

83 33. 8.3 83 $3- god a 388 8a 32 m3

888 28.8 $2.8 $2.8 $3.8 $2.8

mag- wand 83 £3 32 88 H 38% SE 32 mag

:53 88.8 83.8 3%.: $8.: 38.:

8% Sn.“ 2?. ES- 83- $3- +omo=ooE8e

33.8 28.8 88.: 32.: A838 :38

83: 83 Ed £3 83 $3 328 @m

38.8 83.: $58 838 as: 83.8

83- ”8.8 and 23. 83 was 32% 22:2

...L§.8 82.: 63.8 38.8 83.: 88.8

83 £3- 84;- 53- 83 fig- 585

38.8 83.: $88 $58 38.: 33.8

83- So... 33. $3- god «8.? 838:8 SEE 088

+8 mm< 3.8 ow< 3.8 03. 35m :85 fisso 833:5

29:0,.“

 

8.888 mg 298

172

822328 2nt

 

 

.888 23.8 88.8 88.8 8:88 83.38
85. 2.3- 82.8 32. 38.2- 2.883 m 85% 835<
88.8 88.8 .288 28.8 88.8 88.2.8
83- 82.8- ”2.” 8.3 8.8.8 852 m 85% 68>?
288 888 2.8.8 .288 8828 1.88.88
82.8- 28.8- 8.2 88.2- 85.8 88.8”- :0 mo 854
$2.28 2.8.8 :888 5.88.8 .1828 .3888
83 88.8- :3- £3- 83. 2.2.8..- mmwm 8o 8:24
28.8 28.8 238 83.8 88.8 88.88
2.2.8 88.8 838. 3.9.8 8.2.8- 8.8.8. ii We 854
$288 88.8 88.8 88.8 28.8 8388
85 and- 5.8. 8.2.8 .83. 88.8- 8.3 mo 8&4
88.8 88.8 88.8 88.88 2.8.88 88.38
28.8- we...“ 3:. 33... 33 2.2.83 88288 5&2 3:5
888 2.88 88.8 838 882.8 83.8
2.8.8- 88.8- 8.8.8 88.8- 8.2.8 mag. 83 3:5
228 2.2.8.8 838 23.8 88.8 8888
2.8.8 88.8- $8.8 88.8. 88.8- 28.2- m $88 8.5 28.. 32 3:5
8:8 838 82.8 88.8 88.8 288.88
82.8 88.8- 88.8- 8.2.8 82.8 88.8 2 383 Ba 32 meg 325
3.888 83.8 88.8 2.88 88.8 88.8
8.8.“ :3. 8.8.8- 88.2- 88.8 83: $330889 38.:
2.8.8 288 28.8 28.8 88.8 888.88
~28 88.8 82.8- 82. 33- 38.8 628 85 3:5
888 8288 8:8 82.8 2.2.8 83.28
2.88 $8.8 82.8. :3. 5.8- 2.8.3” 328. 22.3 2.23
88.8 82.88 2.88 83.8 8E8 28.28
38.8 38.8 83- £3 and. $4.8 S85 :53
88.8 83.8 88.8 83.8 28.8 83.3
83 23 88.8- 88.8 83. 2.8.8- :8 58% meow 323

9:8 5805 an 28 88 288 83.33» 22.823

892 28:6 84 53224 54 28:54
2.22

 

 

wompmtmuoflwno 3328800 :33 3.54% mZEO 5 203885200 woorm E mowgno ”ad @388

173

8:538 @388

 

 

:88 888.8 888.8 88.8 888.8 88888:
8888 88.8 888- 828: 8888 88.8: 8 8:58 5834
1:8,: :38 £888.: 888 28888 ...888888
8888- 8:8 888 $88 888: 8888: 8 8:58 8834
:88 888.8 888.8 88.8 88.8 1.8888
8888- :88 888- 888- 888- 88888- 80 mo 855
18:8 :88 $82.8 $888 1888.8 188.8
8888 888- 8888- 23. 88:- 888:- mmwm mo 854
888.8 88.8 888.8 888.8 888.8 88888
58 288 828- 8:8- 888- 38.:- 888 88 85d
:88 £888 :38 888.8 :888 :33
8:88 888- 8888 8888- 88.: 8888 8p: 88 854
88.8 $82.8 1888.8 8.88888 8838 83.8.8
:88- 883- 8888: 88888- 8838 8.8888- 85888 935 38::
1888.8 888.8 188.8 :38 1.888 83.8
8888- 8.88 888 888 8888 8:8- 8:3 385
83.8 88.8 82.8 $88.88 88.8 ...888
828 8888- 8888- 888- 8888- 8888:- 8 $38 888 82 83 383
83.8 88.8 :88 88.8 88.8 82.8
888 8888- 8.88- 828 £88- 888 H 3883 888 .82 83 385
.8888 888.8 888.8 888.8 8,888.8 8888
888:- 888- 888.: 8.88- 8888 8288 $828038“. 385
888.8 888.8 88.8 885.8 88.8 88888
88.: 888- 888:- 83- 8888- 8888- 828 885 38E
88.8 8888 888.8 88.8 888.8 88.88
828 8:8- 88:8- 8888- 8888- 8888- 328 28882 383
888.8 8888 888.8 83.8 888.8 8888
8:8- 388 :88- 8888- 883- 3888- .585 385
888.8 8888 888.8 888.8 888.8 88888
8888- 888 828 88.: :88- 888: 888 83888 8888 E85

330 58on an 88 88 888 8383» 858885

888 8:28 84 83854 884 ocoaoq
gnaw-m

 

8.888 28 838.8

174

835:8 2an

 

 

 

 

888 885 888 88.0 888 888 828888 388888
88¢ 8:; £38 886 888 28¢ 820me
883 33o 238 288 «Ed 2:3 25 8.894
:28 33o owned ”888 238 8:3 888 §§><
good 883 $38 888 ago 888 8854
£93 885 ago 388 8&8 Sand 888% 8880
M88 38¢ 823 :8; 33¢ 59% «888% 8E
832 $38 ”83 888 Bad 388 882 BE
to; 383 2:3 328 Some 888 8:883
OumN 8 353 macmwomwwOo wcmuwmwu no.“ 25.87%

38 £8 83 8:. 53 33 Bangm
$3. $8 $3. $9. $2. 3% go Ho 85882
mm? wm> wo> 8% 8% 8% womEafig 832238800
25.3 8%.: 23.8 62.8: 835 $9.38
83- 8%.? 83- SS :8 83:. 8288 6324
88.: 388.8 33.: 38.3 833 368$
33 £3 83- £3- 3%. 3.32- 8288 0802
133.: $388.8 10.88.: $65.3 $35.8 :Emg:
£3. 33 88.». 388 $52 98.8” 8888 2:: .53
$8.8 $5.8 83.3 88.5 82.8 88488
a.” was $3 58 8&2- SE2- 88:8 24
83.3 $3.3 $8.3 $8.3 58.3 333
EN.“ 83.. $38 38.0- :3- $5.3m- 88 58G
$2.3 :3: 83.3 8mg: :28: 38.39
38.... £3- 88o- 83 88.8- 3.3 828... caged
$33 Lind 38$ $23.3 $33 8.928me
93.? $3 83.” 88.8.. 25.: 88.88 888 8. $82 84
9%.: 33.8 38.: :58 and :83:
$3 $8.? on: SE «a: 9.3: v 880m 8834

880 88on 88 E E :88: 8388> 828389

82$ 8880 084 88054 8% ocoaod
282

 

8.803 2.8 833.

175

.82: mod 3 .8.- 98 ~96— Hd aw ougommcwfi
8285.88 8883?: ... .8855qu E was macho «8.8::me 8258 _m>£-:0m$m 828838 538.888
98 85:56 to:8-B§.o>a £28838 mmw Sow-EzEsz 38 £258 05 E «.2832: 83.. $qu

 

 

 

 

 

888 88.8 888.8 888 888 888 «8888. 88.8880
888 88.8 888 888 888 888 820me
888 88.8 828 888 88.8 888 25 828-84
228 $38 888 888 88.8 883 888 §§><
888 88.8 888 288 288 88.8 8858
888 28.8 88.8 888 888 88.8 8888. 8880
888 88.8 :88 828 8.8.8 882 888% 83
888 R88 883 888 8:8 88.8 83.3. 85
888 8.88 828 8.88 8:8 388 8388.8
OHQN Do 3:30 macvmoamouu wag-«Bu uOm 05~w>nam

83 88.8 888 888 888 83 8888-8
88. $8. 88. 88. 88 82. so 8 85882
mm> mm.” wo> 8? 8% 8% gag—3Q kumGSEEOM-v
88.8 .88.: 88.8 883: 88.8: 83.88
88.8- 88.8 888 88.: 3.2 88.83- 8288 8:24
88.: $888 88.: $88.88 88.8 88.88
888 888 88.: BS- 888- 888 8288 88.84
$88.: $888 88.: $88.8 88.8 88.88
88.8- 83 83.8 5.8 83 2.3.8- .8288 2:: .882
88.8 88.: 88.8 883: 88.88 88.88
88.8 8.3 5.2.- 88.2- 88.:- 838- 80:8 28
88.8 88.: 88.8 :88: 8:8: 88.88
8.5- 888- 83 82.. 88.8 88.88 ”.8 8864
88.8 88.: 88.8 88.8: 88.8: 88.88
83.- 888 8.8.8. 88.: 88.88 8:88. 8.83 88883
88.8 88.: 88.8 $82.8: $88.88 $88.88
88. 83 8.2 88.8. 3.8.8 88.88 888 8 888 84
:8: 88.8 88.: 88.8 $88.8 88.8:
82.8- 8.3- 888 E8 88.2 82: 8 88cm 8834

0230 582% ”Em 3 Am: 2.883 m03£3> «Gavan-Ewan

808 8880 84 83054 8.2 88804
Edgar.“

 

8.88: 8.8 288

176

.38: mod 88 I. :32 H6 3 8885886 38858838 8 856588 on... 8883?: ... .m 598%; com £8638on
owcpm 83m 05 mo aM-«Oémmavoow 9: 8m .53 5 8:28:58 853888 28 8883 .833 $55.2: 83888885 388888 98
five-83:3 :5: 93 968.8 ozuosvoa .82 503838 :33 8888:sz 98 .msoaowovao 98 8x35 8.85:: €8.50 98 £92
.9358 8293: 32.30 42m 98 Emmy: Jammy: 83.3 828828 $8 .89» 1533?: 2.8 888882 :8 5 39:2: 9.2 ”302

 

 

 

177

 

3.28 828 8888 8:8 83.8
8.8 :8 8.8 2.2 838-30
888 comuwofiaaoEugO
888 888 888 888 83,8
$8 8.8 8.8 8.8 aséswm
85.38.» maoaomovco mo «85 5883.83-35
888 888 388 888 8:8 888 88.8 888 83.
828 8:8 888 888 888 8888 8888 888 8835
888 888 888 888 888 :88 888 $88 W38 :8 .5 2863 83
356880 :0 £88 8.“ 838.7m
888 888 888 288 88.8 888 888 888 8383.8
88 38 38 38 88 88 88 88 3° 88 .8.832
$688: 3.88: $888: 88.: $888: :88: 888: $8.:
888 888 888- $8....- 888 888- 88.? 88.8 88828
88.8: :88: 388: 388: 888: $88: 888: $88:
88 8.8. H88 88- 888- 888 888. 388 58on 528 8:280 8
:88: 888: :88: 888: :88: $88: :88: 888:
888- 888- 388- 888 888 888 888 888 88 888 8:280 8
88.: 888: 888: 388: $8: 388: $888: 388:
888 888 888- 888 H88 H88- 8888 888- Aaa: 58on
888: A888: 888: A888: 888: 888: 888: L888:
S88 88- 8888- 888- 888 88- 888- 888- 838: 88
:88: :88: :88: :88: $288: 888: $388: 888:
888 888- 888 888- 888- 888 888- S88 8888: 8:25
$388: 888: $388: $88: $888: 888: $888: $88:
888- 888 :88- 888 :88- 888 288- 888 E .852 s 38882
$988: $288: :88: 888: $888: $828:
888 H88 H88 888- 888 88.8 88 Em 83
$3.88: .2888 $888: $88:
888 3.8 8.88 888- 88 233 83
$688: $288: $888: £888:
28 888 888 :88 38 B85 83
mac 2 mac 2 So 2 mac 2
88 Em 83 285.8 88 2883 83 88.88% 8888 82m 83 282 88 £883 83 282

 

 

33 E 925ch 6:8 mama 5 3:32:30: no 863.23- :osoqfim ”83260.5 295m < ”om-m 8388

Chapter 3: Age, Cohort and Year
Analysis in the Socioeconomic
Determinants of Adult BMI in the
1990s

3. 1 Introduction

The age-period-cohort (APC) analysis in the medical research literature aims at
determining whether the phenomenon under examination is an origin-related process
(cohort effect dominant), an aging process or a function of time (instantaneous pro-
cess). In the study of body mass index (BMI) there are reasons to believe the three
effects may all be important. In different phases in life BMI reflects different aspects
of body growth. Before reaching maturity BMI is most responsive to increases in
height and for the elderly shrinking becomes the main factor for reduced BMI. Do
men and women have distinctive life-cycle BMI profiles? For the people born dur-
ing the severe famine in the late 19503 and early 19608 in China, did the impact
carry on to their adulthood or have they made up the difference over time? Year
(or time, period) effects may arise from improved living standards or/and medical
technologies that reduce chronicle diseases or even the changing ideas of what is fash-
ionable. However, because of the linear dependency of age, period and cohort, the
level effects of the three factors can not be identified without further normalization or
exclusion assumptions based on prior information. Convenient normalizations may
lead to erroneous interpretation of the data.

In the upper panel of Figure (3.1) the Lowess smoothed BMI of each five-year

178

cohort shows that at age less than 55 men born of a five-year later cohort have higher
BMIs and the differences are larger for the 35 to 50 age group than for the younger
cohorts. However, the cohort differences for women were not as pronounced. For
women the inverse U shape of BMI versus age is more evident than for men. The
lower panel of Figure (3.1) depicts each three-year cohort BMI against age. As we can
see the cohort differences for men become less significant and so does it for women. In
the upper panel for the five-year-cohort graphs two cohorts are of particular interest.
The July 1957 - June 1962 (”x”) and July 1967 - June 1972 cohort (”o”), the former
spans the famine years and the latter represents the first half of the cultural revolution.
Correspondingly in the lower panel the two three-year cohorts are the July 1957 - June
1960 (”x”) and July 1969 - June 1971 cohort (”o” ). It seems that the impact of famine
is more severe on women although later in life there seems to be over-compensation

after age 35.

3.2 Identification of Second Differences

Deaton (1997) suggests that when data are plentiful it is reasonable to use dummy
variables for all three sets of effects to allow the data to choose any pattern (page
124). For individual 2' in cohort c at time t, denote his or her age as a. Assume the

following additive model for variable y,- :
yi,c,a,t = ’U. + ac + 78a + 7t + €i,c,a,t (31)

Suppose or. ranges from age a1 ~ (1,4, t E t1 ~ tT and c E t -— 0. takes values
from c1 ~ Cp. To avoid the trivial linear dependencies associated with using dummy
variables, the usual normalizing constraints are either {a1 = 0, 61 = 0, '71 = 0} or
{flac = 0, £16,, = 0, 27317, = 0} (Heckman and Robb 1985, Holford 1983).

It has been pointed out in the specification above yet another constraint is neces-

179

sary due to the linear relationship between c, a and t (Holford 1983, and the citations
within). However, as shown in the appendix, provided that age and cohort categories
are consecutive, the number of constraints needed to identify all three effects depend
on the spacing of the periods in which the data were taken. For annual vital or preva-
lence data, the number of constraints necessary is one; for biannual survey data, the
number of constraints needed is two; for data taken every four years, the number of
constraints needed is four; and for our data taken in 1989, 91, 93 and 97 the number
of constraints necessary is two. If we leave it to the computer to arbitrarily impose
two constraints the results are meaningless (Figure 3.2).

McKenzie (2002) provides an approach to identify the second derivatives (or
second differences) of these effects without any normalizing conditions. The changes
in slopes effectively provide information on changes in growth rates and the convexity
or concavity of the curves. The Mckenzie approach applies to both repeated cross-
sections and genuine panel data. Below we illustrate how the method works in the
latter case since our data are of panel structure and McKenzie (2002) laid out the
former case. In the former case replace the individual information by corresponding
group means and the method follows. We can use both the individual and group
observations to see how sensitive the two approaches are to the data. Assume data
are collected annually.

Within each individual let Aty, E ytcm — y,,c,a_1,t_1. First time-differencing y,-

eliminates the cohort effect:

Atyi = (ac + fig + 7t + 5i,c,a,t) _ (ac + lag—1 + 7t—1 + €i,c,a—l,t—1) (3.2)

: Ba _ fia—1 + 7t — 7t—1 + 6i,c,a,t _ 6i,c,a—1,t—l

:60 — fia—l + 7t — 7t—1 + Atéi

180

For another individual from a one-year earlier cohort

yj,c—1,a+1,t = U + Ole—1 + 130“ + 7; + €j,c—1,a+1,t (3.3)

the first difference of y,- is

Acyj E yj,c—l,a+1,t-yj,c—1,a,t—1 (3.4)

= 1304-1— fig + 7t — flit—14. Atéj

Subtracting (3.2) from (3.4) eliminates the time effects and gives

AcAtyiJ = (fia+1 — fia + 71‘ 7t-1+ Atej) - (fia " fla—l + ”it ‘ 7t—1 + Atéz‘)
= (180+1— :80) _ (:80 _ [Ba—1)+ Atej _ Atei

a: + Acme, (3.5)

where 6: denotes the second difference of age effects, or the difference in the slope
between age a + 1 and a from the slope between age a and a — 1. Structure the data
as in (3.5) and we can estimate 6:. Hence the changes in the slopes of age effects can
be identified without any normalizing assumption.

For the individual i in (3.1) at time period t + 1 we have

yi,c,a+1,t+1 = u + ac + fia-H + 7t+l + €i,c,a+1,t+l (36)

Subtracting (3.1) from (3.6) gives

At+1yi = (ac + flu“ + 714.1 + 5i,c,a+1,t+1) —‘ (ac + 50 + 7; + fi,c,a,t)

= fia+1 — '30 + 7t+1 “ 7t + At+1€i (3'7)

181

Denote the difference between (3.7) and (3.4) as AaAty,,,-.

AaAtyi,j = (flan " 5a + 7t+1 “ 7: + At+1€i) — (3m — 18a + 7t " ”Vt—1 + Atfj)
= (7t+1 - 7t) - (7t _ 7t—1)+ At+15i - Atej
63‘ + Agata-,- (3.8)

where 6? denotes the second difference of time effects around t. Structuring the data
as in (3.8) allows us to estimate (LT.

Subtracting (3.3) from (3.1) gives

Acyig' = (ac + fia + 7t + 6i,c,a,t) _' (ac—l + [Ba-+1 + 7: + 6j,c—l,a-+—1,t)

= ac — aC—l + fla — [Ba-+1 + ACEiJ (3'9)

Do the same for another pair of observations at time t + 1, ym,c,a+1,t+1 = u + are +

fla-H + 7t+1 + Em,c,a+1,t+l and yn,c,a+1,t+1 = C + ac + fla+1 + 7t+l + €n,c,a+1,t+1 and we

get

Acymm = ac+1 - ac + Ba — [Ba-+1 + A05mm (310)
Subtracting (3.9) from (3.10) eliminates the age effects and gives
AcAcyi,j,m,n = (O’c+1 “ ac + fia _ 1804-1 + A65mm) _ (ac " ac—l + Ba " 1804-1 + Aceiu’)

= (ac-+1 — ac) " (ac _ ac—l) + Acemm _ AceiJ

($5 + AcAcei,j,m,n (3.11)

where (if denotes the second difference of cohort effects around 0, and 0 takes values
from 1915 to 1976. Constructing the data as in (3.11) we can estimate (Sf.

When only repeated cross section data are available, the pseudo-panel version of

182

equation (3.1) is

gent = u + ac + fia + 7t + Ec,a,t

and all the above manipulations follow through and 5", (ST and 6C can be estimated.

Since we have true longitudinal data, we can apply both the genuine panel approach

based on individual data and the pseudo-panel approach based on group means. The

Mckenzie method works the best when the data are evenly spaced in time. Hence we

perform the analysis using two-year-apart data in 1989, 91 and 93 and four-year-apart

data in 1989, 93 and 97, resulting in the second difference estimators across two-year

and four-year intervals.

x-Cohort 1957-1961 , + Cohort 1962-1 966. o-Cohort 1967-1 971

Male BMI Five Year Cohorts
24 1

I
23 4 i' a I," \\ / ‘1':
. / , ‘ \ .’
22 < ‘1. "1' ’1’. ." Ir’l‘l ,‘l ‘2’ {/4 '\
4' \f" l
21 < ':/I J
20 ‘

 

19

 

2‘02'53'03154045»5b3§6’0§3f075
90

Female BMI Five Year Cohorts

24 / X
/
23 i/ /::I’ JJ/‘/ 5‘5“. 1:]; I} A

6 'w
1,1

 

 

io2'550d5404f5’05‘566357b7é
98

x-Cohort 1957-1959, + Cohort 1963-1 965. o—Cohort 1969—1971

Male BMI Three Year Cohorts
24 +

23 -- / .. f} /‘\ .l/ A
*r‘u/ 1);"! ‘\ , /
22 4 ”:2, 4 /’/I/,"// ‘ I r
A [£4 U, / if“ ",‘ /:l/H 5., ”v
x”, ’1'. '1’}:
I .' ax |
< ,1 .' ‘/ '
21 f-Jift' V
20 ‘

 

19 « T
202530354045505560657075
ge

 

Female BMI Three Year Cohor
24 ‘ / \
23 ‘ /// \Jofl /,( /
22. J/ r.
. t/ /

21 i ifpf‘

2o 4

19 A 1’ T T T I I 1' T I T *

202530354o?505560657o75
ge

 

Figure 3.1: Lowess Smoothed Male and Female BMI of Each Five- and Three-Year

Cohort vs. Age

183

—-— Male 6 effect — Female age effect —~— Male cohort efleet— Female cohort efl

 

 

 

 

 

3.07 -With Perfect oliearity 8.44 ‘With Perfect Colinearity
j lllllllllllllllllz :1“th .
lll' lllmm'" «WWW/WWW)

Figure 3.2: Estimated Age and Cohort Effects for Men and Women with Arbitrarily

Imposed Constraints

3.3 Simulation Study

To answer the question ”can the second derivatives help selecting correct esti-
mation strategy identifying the age, year and cohort effects”, we do the following
simulation exercise.

Create a data set with 2500 (or 10,000 for the long panel) observations, with
age ranging 20 to 59 in year 1980 and same people being surveyed for the next four
(or nineteen for the long panel) years till 1984 (or 1999 for the long panel). The
first example is a short five-year panel data and the second example is a relatively
longer 20-year panel. The cohort is from 1921 to 1960. Generate an outcome variable
with linear but no higher order age, period and cohort effects such that a: = 0.5a +
0.3c + 0.7t + e, where a = age — 20, c = cohort — 1921, t = year — 1980, and
6 ~ N (0, 1). Maintaining age and year effects but changing the cohort effect we get
another outcome variable x1 = 0.5a — 0.3c+ 0.7t + e in order to compare the impact of
this change on estimations of age and year effects. Generate an outcome variable with
quadratics of age, period and cohort effects but no interactions amongst them such

that y = 0.5a — 0.04a2 + 0.3c — 0.026” + 0.7t -— 0.011.L2 + e, where 6 ~ N(0, 1). Generate

184

another variable with different cohort effects but the same age and period effects such
that y1 = 0.5a — 0.04a2 + 0.5c - 0.0462 + 0.7t — 0.01t2 + 6, where 6 ~ N(0,1). Set
Oy/Ba = 0, the inflection point for age profile of y is at age 26.25; so is it for y1.Set
By/Bc = 0, the inflection point for cohort profile of y is at cohort 1928.5; and1927 for
3/1-

We can estimate models with the following constraints: (1) linear cohort and
year effects, (2) linear age and year effects, (3) linear age and cohort effects, (4) linear
cohort and year effects with quadratic age, cohort and year effects, (5) linear age
and year effects with quadratic age, cohort and year effects, (6) linear age and cohort
effects with quadratic age, cohort and year effects, (7) linear cohort and year effects
with quadratic cohort and year effects, (8) linear age and year effects with quadratic
age and year effects, and (9) linear age and cohort effects with quadratic age and
cohort effects. None of these models are correct specifications according to the data
generating process. Tables 3.1 to 3.4 summarize the regression results for the above

specifications for a: and 2:1; and Tables 3.5 to 3.8 for y and y1.

3.3.1 Models with linear and/ or squared terms
Estimates for a: and :51

When the data generating process is linear in age, period and cohort, clearly the
linear dependency a E t — c requires one of the linear effects be excluded from the
model. However, when, for example in Table 3.1 for :c = 0.5a + 0.30 + 0.7t + 6, age is
left out, the estimated coefficient for cohort should reflect the combined effect of age
and cohort, i.e. a: = 0.5(t — c) + 0.3c + 0.7t + e = —0.2c + 1.2t + e. In column 1 of
Table 3.1 we can see that is in fact the case. Similarly, when cohort effect is excluded
from the model the coefficients for age and year effects are also combined effects. It
is impossible to disentangle individual effects from such estimators. Table 3.2 of the

short panel for :51 displays the same feature. In Table 3.3 and Table 3.4 when the

185

panel gets longer the point estimates for the combined effects are more accurate.

Using the McKenzie approach we can estimate the second derivatives of age,
regressing the corresponding second differences of the outcome variables on age as a
continuous variable, as seen at the bottom of Tables 3.1 to 3.4. None of the second
derivatives of age, period or cohort are significant, suggesting the model should not
include higher order terms. This is confirmed by columns 4 to 9 in Tables 3.1 to
3.4. When the second derivatives estimated without normalization assumption indi-
cate there is no higher order effects the square terms should not be included in the
modelling.

The difference between :5 and x1 lies only in the cohort effect (0.3 for x and -0.3
for 3:1). As we can see in column 3, 6 and 9 models excluding year effect estimate the
combined age and year effects consistently; and in column 2, 5, and 8 models excluding
cohort effect give rise to the combined age and cohort effects that are affected by the

coefficients of c in the data generating processes.

Estimates for y and y1

When the data generating process is with linear and second order effects of age,
period and cohort factors and the model is specified as in columns 4 to 6 from Tables
3.5 to 3.8, we can see the second order effects are correctly estimated. However,
the linear terms are still combined effects. For example in column 6 then the linear
year effect is excluded the estimated model in effect is y = 0.5a — 0.04a2 + 0.30 —
0.02c2 + 0.7t — 0.01t2 + 6 = 0.5a — 0.04a2 + 0.3c — 0.0202 + 0.7(a + c) — 0.01t2 + 6 =
1.2a + c — 0.04a2 — 0.02c2 — 0.01t2 + 6. As we can see all combined linear effects (age
and year, cohort and year) are fairly accurately estimated.

When linear age effect is set to zero, there is no reason to believe the second
order age effect should be different than zero. Hence in column 7 both linear and

squared terms of age effects are set to zero. In such case, the estimated model in

186

effect is y = 0.5a—0.04a2+0.3c—0.0262+0.7t—0.01t2+6 = 0.5(t-c) —0.04(t—c)2+
0.3c — 0.0202 + 0.7t — 0.01t2 + 6 = —0.2c + 1.2t — 0.06c2 — 0.05t2 + 0.04tc. As we can see
in the long panel case (Table 3.7 column7) the coefficients for c2 and t2 are correctly
estimated; the estimator for coefficient of c (= 3.677) is actually the combined effect
of c and tc evaluated at t = 97; and the estimator for coefficient of t (= —0.361) is
the combined effect of t and ct evaluated at c = 39.

Using the McKenzie approach we estimate the second derivatives of age, cohort
and year effects. The t—statistic strongly suggests the second order age effect should
not be zero (p=.000), the second order cohort effect is also nonzero (p=.038), and the
second order year effect is not different from zero (p=.338). At least the McKenzie
approach instructs us to include both squared terms of age and cohort.

The difference between y and y1 is in the cohort parameters. The fact that the
estimated age coefficients between Tables 3.7 and 3.8 in columns 3, 6 and 9 stay close
but the estimated age coefficients between Tables 3.7 and 3.8 in column 2, 5 and 8
change dramatically indicates the two data generating processes vary mainly in the

cohort factor.

3.3.2 Models with dummy variables

Clearly models with only linear terms or even linear and squared terms are
not estimating parameters we are interested in. When the data generating process
is defined as above for y and y1, it is quite impossible to separately estimate the
coefficient for a, c or t. However, in a model with no interactions where y = 51a +
62a2 + 63c + [34c2 + ,65t -l- ,Bfit2 + 6, if we set By/Ba E 0, the inflection point for
age profile of y is at age a such that B, + 262a = 0. We can estimate regressions
with dummy variables that include: (1) age and cohort dummies, (2) age and year
dummies, (3) cohort and year dummies, (4) age, five-year-cohort and year dummies, or

(5) age, three-year—cohort, and year dummies. All these models get around the exact

187

linear dependency between age, period and cohort through different identification
assumptions. In model (1) year effect is assumed to be zero, the estimated age and
cohort profiles are combined profiles; and similar assumptions are made for cohort
and age in model (2) and (3). In model (4) it is assumed every five-year-cohort has the
same cohort effect and in model (4) it is so for every three-year cohort. After plotting
out the estimated age profile, we can find the inflection point and the relationship
between 61 and 32. For example, if we can approximately eyeball the inflection point
of age at 6 years after the beginning of all estimated age periods, we can say 61 + 12
62 = 0.At the mean time we can use the McKenzie approach to estimate 62. Hence

we will have an estimate of ,81.

Profiles for :1: and x1

Throughout Figure 3.3 to 3.6 the small circled lines are the true profiles from
the known data generating processes.

Figure 3.3 displays on the left the age, cohort, and year effects for x each identified
through four of the above five different identifications and on the right for 2:1 both for
the short panel. Figure 3.4 corresponds to the longer panel for :L‘ and 1:1. For the age
profiles in Figure 3.3 and 3.4, we can see when the model (short dashed line) includes
age and cohort dummies the age profile (with slope about 1.2) is the combined age
and year effects and is not affected by the coefficients on c in the data generating
processes (0.3 for x and -.03 for $1). So is the case for year effects (lower panel dotted
line) identified through models excluding age dummies because these year profiles are
combined year and age effects that are not affected by the coefficients on c in the data
generating processes.

When the age profile is identified by models including cohort dummies, be they
one-year, three-year or five-year cohort dummies, the age profile is underestimated

on the left for a: and overestimated on the right for an. AS we can see the model with

188

age, three-year-cohort and year dummies is closest to the true profiles of age, cohort

and year.

3.3.3 Profiles for y and y1

Figure 3.5 and 3.6 show the results from the dummy-variable models for y and y1.
Again we can see all models are estimating combined profiles instead of the individual
profiles that we are interested in. The five-year and three-year cohort specifications
give rise to very similar profiles and are generally closer to the true profiles from the
known data generating processes.

One basic assumption in age, period, cohort analysis without interaction terms
is that the age effect at any given age across different birth cohorts are the same.
Notice in our model, particularly in the short panel, the age effects for younger ages
are identified through later birth cohorts and for older ages through earlier birth
cohorts. If the assumption does not hold the estimated age profile is biased.

As seen in Tables 3.5 to 3.9 the estimated second derivative of age using the
McKenzie approach without any normalization assumption is -.002; and from the
age profile identified through the three-year—cohort specification in Figure 3.6 the
inflection point is at 12 years. Therefore we can estimate from setting By/Ba E
61 + 262a = 0, where :62 = -0.02 and a = 12, hence 31 = 0.48 which is very close to
the true linear age effect in the known data generating process of y.

In conclusion, when estimating age, period and cohort effects in general we should
(1) avoid using simple linear specifications, (2) when interested in the age profile use
three-year or five-year cohort specifications, and (3) interpret the combined effect

correctly.

189

 

 

 

8.8.135»:
88.1: .8883 88”.”- so» no 8.828 288 83888
88".: A883 868.“. :88 8 88268 8.8% 388st

88.1: :88 «88.. 8a 8 $8258 .888 assess:

 

 

 

moooé moooé voooé voooé voooé voooé No8: No8; moHoA MmE qum
2.89: ovéoov «somov moovom modvom moovoo moovwo mwovwo mwovoo Hound
oomm oomm oomw oonw oomm oomm oomw 83 83 30 .«o non—8:2
:3”me *Lwndov *LvmeS *Lomdzv *medzv **$m.N:V 1,mede *mewmv *medmv
wmémoa- No.83- omémofi- :owa 2.33. 5.83. onovoH- onovoa- 3.33- 3328
:88 :88 :88 :88 :88
mood mood mooo mood mooo o 330%
:88 1:88 o :88 :88 :88
88. vooo; Hooo. Hooo. 88. o 838:8
:88 1:83 :88 :88 :88
Hood- woooi Hood. Hood- Hood- o duéwc
188.8 1888 1:88 1888 1:88 1388
wood v24 wmoo mmHg wood «34 No an?»
1888 1:8.8 1:8.8 1:88 1:88 1:88
ouoo ammo- wmoo wood. wood wofio- no €028
1888 1:88 1888 1:88 1:88 1:88
oomA goo meA wmmo NwHA wofio mo owe
oHcmhcoz ofldmauons oHcmowm $565000
onus?» oHfioaoo oHowa oHEo» oHfioaoo oHomd onus.» oneness oHcma Rom
o w n o w v o m H
muafismcogooofi wafidbmaoOEooE muefibmcogmvog

 

 

 

 

83$on fionoo one .vmoToooH 89” .moom mow—HE cw< w + two + £6 + and H a
x 8m 8oon 88% one 8038 some 88ng ocucfismm ”BEE 83m Ho 2an

190

 

8.8u:8§o2

mow“: .Aooooo moo-8:3- .89» mo 83858:: 9:88 woe-oasmm
88.”: :88: 8.08.: 89:8 8 82:28:: 883 83858:

88.": :88 :88. 08 8 83:36?

8:89: Eagumml

 

 

 

 

88.: 88: 88.: 88.: 88.: 88.: 88.: 88.: 88.: mm: :8:
8.888 8.88... 8.888 8.88: 8.88: 8.88: 8+2: 8+2: 8+2: 381...:
88 88 88 88 88 88 88 88 88 Bo 8 388:2
ISM-ma 1.38.63 *LcmdoHv 3.8.sz :3.sz :8.sz 1.3de .113de *Lovdmv
8.8::- 88.:.:..:.. 8.8:- 8:8- 8:8- 8:8- 88::- w:.8m:- 8.88- 23:88
:88 :88 :88 :88 :88
88.8 888 88.8 88.8 88.8 8.8- 838.:
:88 :83 :88 :88 :88
88.- 818.8 88.- :88- 88.- 8.8- 8:888
:88 :83 :88 :88 :88
88. :88. 88. :88. 88. 8.8- 888
188.8 188.8 1:8.8 1:88 1:88 1:88
:88 8:: 88.8 8:: 8:8 8:: :8 39:
1:88 1:8.8 1:88 1:8.8 1:88 1:88
8:8 8:8- 83 8:8- 8:8 8:8- :8- 89:8
1888 1:88 1:88 1:88 1:88 1:88
8:: 83 8:: 8:8 8:: 83 8: 8.:
ofldmuda "amicaoo oHdmmwd BGQBESU
oil-39m one-8:8 oHoma oHBom onfioaoo oHoma on: one-8:8 oflvmd Bum
8 w h 8 m : m a :
Eflmduumfloofifioz manqauumdoofifioz magma-mamGOU—omvoz

 

 

 

 

88:8: 82:8 93 88:88:88 .88 888 8:: w + 8.8 + 8.8 I 8.8 n E.
:x :8 muomwm :8» U5: fionoo .mwd 8.68.82 Euwfimamm Ava-mm fionm ”ma 2an

191

 

«osmnfiavfioz

88.”: .2888 88:8- :8: 8 888:8 888 8888:
88H: .2883 8.83- 888 8 839:8 883 8888::
88.“: .383 ~88. ”8 :o 852:8 808m 8888::

 

 

 

 

8:: 8:: 8:: 8:: 8:: 8:: :8: :8: :8: mm: :8:
8+8: 8+8: 8+8: 8+8: 8+8: 8+8: 8+8.“ 8+8.“ 8+8: 3%»:
88: 88: 88: 88:: 88:: 88: 888: 88: 88: 8o :o :85:
£88 1:88: 1:38 £8.28 £8.58 :8:: 1:88 1:88 1:88
8.88:- 8.8::- 888- 8.88:- 888- 8.88:- 8.3:. 888- 8:8:- 8388

A888 883 388 A888 383

888. 888.- 888. 888. 888. :8- 88:
888 L883 :83 A888 :83
:88.- 88.- N88.- 888- 888.- 8:.- 8:88
588.8 1888.8 .888 8 .8888 :88 8
:88- 88.- 88- :88.- :8::- 8.o- :88

1:88 £88 £888 3:88 1888 1:888

: 8a.: :8: 82 :8: :8: 8 :3:
138.8 388.8 188.8 $88.8 $38.8 $28.88
«8.: 8:8- :8: 83- :8: 8:8- :8 :28
138.8 2.88.8 198.8 188.8 $38.8 $38.88
8:: 82 88.: 88 :8: 88 no 8::
any-.82" ovafionoo oHduéma £56880
ONE?» oflfiosoo oflmma oflamwz onfioaoo cwam 0H8?» oHfioas oHema Ham
8 w 8 w m- + m m :

macahamfloofimvoz mucmduumdooflmvoz mudmduumGOOBmvoz

 

 

 

 

83.53 88:8 98 5236mm“ Hag .wpém 8&5: ww< u + «No + m6 + and H H

x :8 803% 80% 8:8 €38 6? 88358.: kuwfismm ”888% wqoq ”m.” @389

192

 

mm.mHHA:Hv=doE

88.": .888 8883- :8: 8 852:8 888 8888::
m5”: .388 8.38.»- fionoo 8 85:55:. 8:88 88858:

88.": .3888 888.8 8.: .8 888:8 888 8855mm.-

 

 

 

 

88: 88.: 88.: 88.: 88.: 88.: 88.: 88.: 88.: :8:): :8::
8+3; 8+8.” 8+88 8+83 8+8: 8+83 8+88 8+88 8+88 33m»:
88: 88: 88: 88: 88: 88: 888: 88: 88: 8o 8 88:7:
$23.3 *medfiv *me.v3 1.33953 *LVONS *LVOSS *Lhwfiv 3.?de 1.?de
8.8::- 888- 8.88- 8.88- 8:8- 8.38- 8.38- 8.38- 838- 5:88

3.83 3.83 38:: A888 383

8:88.- 88.- :88- 88.- :88- :8- 88.:
388.8 383 :83 388.8 383
A88. 888.- :88. :88. :88. 8.8- 8:88
388.8 388.8 383 388.8 38::
888.- A888.. 388.- 388.- 388- 8.8- :88

Inc-8.8 13.8.8 33.5.8 338.3 "1.38.8 1.38.8

8:8 83.: :88 83.: 38.8 8:: 3 :8:
£388 £388 .1388 2.388 ......38.8 3.38.8
8888 88.8- 38.8 388- 88.8 :88- :8- :28
$38.8 3.38.8 1.38.8 1.388 #388 1.388
83.: 8:8 83.: 388 8:: :88 8.8 :8
sung-38m oflguonoo Guam-m 3585000
"-88% oflfionoo New-m snag H9838 onw-a ell-.88» Oil-p.838 oflwwd 80m
8 w 3 8 m :. m 3 :

33:88:08: 38:88:25: 3588888:

 

 

 

 

coma-Hug 98:8 6:8 dag-A53 .89» .wu-om 8mg: mw< u + £5 + end I umd H :H

C: :8 308mm :88» «:8 98:8 .www @3258: woe-883mm Ava-mm waoq afim $388

193

 

88:-H3882
888.: .8888 888.- 88.: 8 882.8 8:88 8888
EH: .8888 888.- 8888 8 882.8 8888. 8888
88.8 .288 88.- 88 8 882.8 8:88 8888

 

 

 

 

 

:88: 888.: 888.: 88.: 88.: 88: 888:. 888: 888: 88:): 83:
8.88:8. 8.88:8 8.888 8.888 8.888 8.888 8.3: 8.2: :85: 888-..:
8888 888 88 888 888 888 8888 88 888 88 8 88882
£88.88 £88.88 18.888 £3.88 138:8 1:388 88.88 88.88 88.88
8.88:- 8.8:.- 88888- 8.888- 88888- 8888- 88.888- 8888- 8888- 838:8

:88 1.888 .888 28:88 .888

888.8- 88.8- 888- 88.8- 88.8- 88- 88»
1.3oo.8 1.8 1.8.38.8 1.38.8 1.38.8
88.8- 8888- 88.8- 88.8- 8888- 8.8- 8:888
......3oo.8 1.8 1.38.8 1.38.8 ......30o.8
88.8- 88.8- 88.8- 8:88- 8:88- :88. 888..w

1.38.8 8.va98 1.3mo.8 **C.mo.8 AnoHdv 8.1.3688

8:8 888. 88.: 88.: 88:8 888- 8 88
£88.88 .1388 1:888 .1888 8:8 88:88
88.8 88.8 88.: :88- 88:.8 83.8 8.8 8888
1.288 3.888 1.988 1.8888 3.88:8 .1888
88:: 88.: 88.: :88 :88- 838- 8.8 888
ofldmawmz oHuflSno: Guam-8 $58880
oHHmoz oflfioaonv oflmwa OMB?» ofluhfioo chow-8 oflag oHfionoo oHom8 15m
8 8 :- 8 8 :. 8 8 :

muflmduumdoufiofioz mafifiuumflOO—Omuoz madmwuugOnV—mfioz

 

 

 

 

 

.8 :8 88:88 :88.» 8:8 8:28 dwa 3:888:88: 8888:88m ”8:88 fioam ”9m :83-8H.

88:-:8: 8888 8.8 .888: 88 .888 88...... 88. u + «:88 I :8 + .888 n 98.8 + «88.8 I 88.8 u 8

194

 

angucefioz
Sang .3888 «.88.- 3% mo 823:8 28$ Esasmm
Song .3888 389. tons mo 832558 383. Biefim

 

 

 

 

Song A83 «8.- a? mo 8>s§cmc 8:83 BEBE?
woSA 884 38.3 8:: 8:: 88.3 mood awed awed mag aon
afimmawm mmémgm 5.32m whmmmuv wkmmmhv wkmmmvv omém 86m 86m 383$
83 comm comm comm comm comm 8mm 8mm 8mm Bo mo 838:2
*Lcmdmv :83.me8 #35608 *Lmodzv $36.28 *Lmofizv 1.3938 “$393.8 *Lwcdfimv
8.3%. $63. «mafia- wouflvm- $.33- 8.03m- wwfiw» $sz 3.5» 33950

88.8 ...Lomodv 3330.8 *Lmfiodv $230.8

:3 wood. mag. 83. m3? 8.? 33%
1.38.8 1.8 1309.8 3.38.8 “1.38.8
mmod- owed- mmod- mmod- awed- 36- 338:8
1.38.8 :8 338.8 .33098 .33098
Sod- Rod- 8.9a- Sod- owed- 3o- 88%

188.8 1388 1:88 #838 £338 1:28

Ewe- 83. 83 ES 88¢. 83- 3 8%
1.888 328.8 1:88 :38 182.8 188.8
83.3 3.8m mama mvod- movd- med md fiasco
*Lomodv 3.3.8.8 :Gmodv 3mo.8 ...Lhmfidv **C.Ho.8
oHUwon oHdwfigoo oHcmoww 388580
CHE?» cutono“. oflmwd OHS?» oflfiononv oHomw H39» oflfionoo Hume” Bum
a m N. c n v m m H

muSmuumsoQovoE maEaSmaoOEvoZ macfiuumaogovoz

 

 

 

 

 

82-82 toss 23 $332 3% .88 8mg ow< u + «:3. I ”3 + «Sod I 93 + «so... .. and n s
Tm SM 309% Hum» 98 fionoo .mww $30232 «5‘38“me ”Edam fienm ”9m Each

195

 

8.818552

wanna A83 H88.- 38 mo azafisc 888 Baas-Hm
88."; .888 88c.- t28 8 859:8 388 8:85

8o."AH .888 m8.- owa «o gmgfic 88$ Began-ml

 

 

 

 

3.34 mvmod 38w 3.2: 3.84 28A 3.36 out; @536 MmE aon
mo+mc.w mo+vm.m $.8me 3.8-ems mo+oms 3+2“; 8.89: Samoa Sécowa madam-m
89: 88H 88H 88H 88H 803 89: 88“ 89: Bo mo 8:852
2.808 .....§.5 £5.88 £8.28 $88.5 $98.3 3.8 A338 3.38
3.33. wfiamfi- 5.83- $.52. «@33- «@53- hmwm- nmdm- Emu-mm- £8838

:88 ......a8.8 :8 :8 :8

Sod- owed- 08.0- Sod- Sod- Hod. 380»
:on .1.on .3..on 18v :on
Bod- owed- owed- owed- omod- mod. Ewe-8:8
:on :on .1.on ,3..on .3on
85.8- 888- 88.8- 85.8- 888- 8.8.. as»:

*Lgodv *Lomodv 1.38.8 338.8 3.38.8 3.88.8

cmcd End- 256 coma nmod wow;- ud Bo»
*Lmoodv 3328.8 338.8 138.3 1328.8 8.2898
«and Re.” 823 8nd. mmod v33 md p.838
£388 £88 £88 1.88.8 1.38.8 1.388
83 83 83 83 88.7 33- no ”ma
OHS-89A any-€048 oflvuéww 356580
on; cue-8:8 ofloww OMB?» cflfiozonv oHow-m OMB?” Guiana“. enema 33H
m m N. o n v m m a

BguugoOmeoz 3:85309032 Eafiufiaogovoz

 

 

 

 

88782 €28 93 .8288 :2 .88 88:: om< u + «no.8 u :8 + «08.8 I one + «68.8 I 8.8 n a

x 8m muommm .39» EB fiasco 6? ©3358." woumfifimm ”Edam was ”Nam £an

196

 

8H8 A88 «8.. 0w0 .«0 $383.58 80008 vegagumnml

8.81.8.3»:
manna .888 8.38.8- 3% 3 35233 €83 8338
8o.HQ .388 88.- 80:8 .«0 83833208 @088 8.8082qu

 

 

 

 

3384 2.85 836 H88; H84 H84 88.: v8.2 v8.2 ”82 800mm
8+8...“ 8.833 8.8%: mo+3.m 8158 8+3...“ 8.88 8.83 8.88 madam-m
883 88H 882 883 08H 8H 8H 8H 8H 30 .«0 358:2
:88 $2.088 133.88 :88: £8.28 1.8.28 3.8.38 3.8.38 $3.38
8.83- 348$ 8.88- 888? 8.8mm- 8.88. 8.83 8.83 8.83 u.3808

1.38.8 1.38.8 1.28 1.8 1.8

~88 8.8- 28- 28. 28.8- 88- 83%
1.8 1.8 1.on 1.8 1.8
88.8- 88.0- 88.8- 88.8- 88.0- 8.8- 880:8
1.8 1.8 1.98 1.8 1.8
88- 3.88- 08.8- 88.8- 88.8- 8.8- 033

1.1.3898 1.3398 1.3898 ......38.8 1.:«98 1.3Ho.8

88.8- 88.8- 834 884 Sumac- 88.7 No 80>
1.38.8 1.38.8 1.38.8 1.38.8 1.3228 138.8
88.8 3.35 8H.“ Sod. >86. 85.8 md 80:8
338.8 .3388 1:88 338.8 338.8 2888
$3 and 28 to... 28- 83- .8 33
oHdeom oHdmfiogoo oflgww 3868000
8H8?» onfionoo oflowd OHS?» oflfionoo cummd oHBQn oflfionoo oflwmd 33m
8 3 N. w n v m m H

macfibmcogovoz 80858090802 30858092002

 

 

 

 

882-32 3.38 83 .8282 3% .88 8mm... em... 3 + «:88 I :8 + «38.8 I 3.8 + 3.38.8 I 88 u s
Tm Rom B880 .80» 800 80:00 .08 88:88 89.25me 2085 wnoq ”33 030R.

197

Age Effects

Age Effects

50 '1

40“

30 W

20 “

10 —«

 

l//
.J/l
'/’r
;'/’
/
1/
./l
./
1/’
{/1
,//
l// M
; 143-”3' -
(re/v
,/ (gr—0’”
.. ”'6‘ d
’/ «5' ‘ — —
. ,fgr‘8>
M7
- *Y‘ '
’e/U‘” / J
’/ .4, ”8 ,. —~ " . f—fi‘
,/ _ .- ‘“"' P”. ’ _' ,‘l
AM“ - " 1,, ——- f“
// I?” W ‘1‘. ‘
/’ M .- fifl "'
’ - V __, ._ ”, ”-«t
/ ’45 / 8 ...,-a"
., __ _ I...
~ M , 7.: »~‘*
»’ ‘.»-*’:’M ,.. :::-‘~~ T/V» b,

,’ ,_ ’ {'53”

.— .&* —_.—$-'

. I" M

 

50 8

4o

30 —.

20*

10 ~

 

r/1
J j/ 6"?”13/8 '
/ / ’x/ / “re/ear“
/, A)” .13
, ,/ 8
/’ 4””) “8"3’8/
”’1’ /,,.;.. fi’p’k
/ ’1 [[[[
J:
// / 1/ A WW
,// / 0M
// - *3" -
,./’ Tr”,
/ r/ ”M
I”, / I ‘1
, /_/e."€"M
534—6”

 

 

20 25 30 3‘5 45 43 50‘ 5'5 60
age
Model x1= 0.5a-0.3c+0.7t

198

65

Cohort Effects

Cohort Effects

35 I_‘ /'/,J«‘
.r"//‘

(‘III
.“/’I
,r/
25 * -/'/
[1"
//’/
.2 /‘
v>’/‘/’/
///r
;'//
15 — If”
14‘/
.«//
// ‘ y
// ‘ P4k4fvfi f
.u’” AOm-‘L *
// 't" —&r"4.7
/ 4}4epfif* 4) V
5 —‘ // WWJH'
/ .t 9~Qarfi
H ##3## 44"
av— , :_ _—_- "L” m --.J.g__——_—_e_—- 1.x:— -.—_2-—_:‘: IL; :44“;—- ;—— 1:7L‘;—_

 

T j T T -
1920 1930 19'40 1950 1960
cohort

Model X: 0.53+O.3C+0.7t

35—

25*

 

15 r AMI
” #4,,
5 e H,
_/’/
\(;:€*; 2'1“; ”LT—7" — —' uL——-— ~— 3.: :____ _—,“: -— '— :— -:____ .___. “w i : —
t 3 ‘v—f._,: A
V *thfi‘flw
——< r t We.
—5 . 4% -
‘ WM-"
€“‘1f—~t¢__‘ . ‘mfi—e—‘H .
t" r ,
Met
*‘Cc—

 

 

T T _ #—
1920 1930 19'40 19150 19330
cohort

Model x1 = 0.5a—0.30+0.7t

199

 

 

 

5 ._.
4 ~ [I/
m ,»~”
3: . a ' ' / ’/ f,»/
m ' // I A///"
a / /r [’[y l /
>03 2 J 4% / - 1’ ” ,,/*’
, (xi/’4 ///r/’
. - //l, u **** ’// I//
. r/f/fl’ J #, I ,
1 J {a ///,
.3///
o l
'f—(n‘_w’ _‘ l “ml"'~' .__. w—_ ' f
1981 1982 1983 1984
year

Model x= 0.5a+0.3c+0.7t

 

 

 

 

5 _4
4 A
m
*5 3 ~
0 ,/‘~"
t /,
L” I// ’r’/
a ////
o __. /’8”
>- 2 J ,,,/'
1 A /’////x qw‘r‘fir 1"” 55” fax ” M
G’”/ /fl¢:—f! fl(
5'3"»! " ’
”/9!
o _t
l1 l l — — l -_
1981 1982 1983 1984

year
Model x1 = O.5a—O.3c+0.7t

Figure 3.3: Short Panel Estimates of Age, Cohort and Year Effects for :1: and x1.
Identifications achieved through (1) model age, cohort excludes year effects; (2) model
age, year excludes cohort effects; (3) model age, cohort excludes year effects; (4) model

age, five-year-cohort, year effects; and (5) model age, three—year-cohort, year effects.

200

Age Effects

Age Effects

40 ' /

sol

 

 

,4:-
,£r*-"
,r 7 a ,
(, ._< ,5
2 ’ twat?
."1 ‘._7‘:r '
A/ '7": ‘fit
‘* A,
,I’
49’
- (6:?
*fi, -kt'
,1 . ‘ '7" ‘
,‘U '
~1 , ~ We
,' 7. , r’ v 7”_ -
fir ( __. I '
kw "a ,— "v i,".F-—‘ W‘
ajV" ,1 7
.~ 41:" r 4 g 7
3fl.) —-* n -
.— __4- A _5
__ «3 kg
‘2’ ’l ;' ~1"J ‘
.. _ -1-
I r‘ V" i ...u- “I:
- __.—'5 ’ » fl...-
0 I. r _l'-’
L» 1 e—- -v — -— ————-———-7———————— ————--——.—— V T T—

20 25 30 — 35 4o 15 5'0 55 6'0 65
age
Model X: O.5a+0.3c+0.7t

4o

,}
, 74")“
, l“,
30 d , ,c
,. .4
,1”, ’f“
, ‘7‘ F“
, ' ‘ ’A
, ¢:" ‘
7,/
.y’,
1"" 1
4 ,1 1+"
/ t '
’/ ,/ fl Vl~raJ
‘44 u’ /l' , {V‘r'r
I A ”93+! ‘

 

.(">' ‘ v— ~77 ">1
". , ... ‘ #—.7 i
10 m ,/ - 23’”
x V -.r-

 

_ l 'l l'

f i " ' T I T T
20 25 so 35 4o 45 so 55 so 65
age
Model x1= 0.5a-0.3c+0.7t

201

Cohort Effects

Cohort Effects

 

 

 

35 a ,2 ”/1
25 9 ,. f
15 ~ r‘ ”i AM.
5 1 ,,,,, /__’_+:F;# M we” W" “W" M “New My
_5 — — _w ‘ — a # — f . i f m 1
-15 *—
_25 —1
1‘6§6_f_'— 135.0%“ 19110 *J—fo‘éo “A. JTQ—CC
cohort
Model X: O.5a+0.3c+0.7t
35 —
25 ——
15 ,9 ,H If”, ff,”
5 -9 J, “Jr/,2 if“; f/Ir/rr‘ "” ”J [ft]
_5 a } f if? :V “W” ”“322 swank” ”or-*«wfixwfiflfik V
-15 A E i W
-25
l _ _ __

 

 

 

 

1 _ _ _ __ . ‘1»
1930 19110 19150 1960
cohort

Model x1 = 0.5a—O.30+0.7t

202

20*

15*

-/’12
a?"

Year Effects
’2

l ' :r‘i’l'r"
"I,
1o—~ «w' xv
V .1/ , .,// I

 

 

 

 

 

! fl ' T E‘— —_'—.T _— w—‘—T M
1 981 1986 1991 1996 2001
year
Model X: 0.5a+0.30+0.7t
l
20 J
15 -
F3 -.
a: .l/«rre/
LLI [M
:13 10 ~   , ’fl/ze/fl “““““

 

 

 

19Ta1 1986 1991 19S6 I 2001
year
Model x1 = 0.5a-0.3c+0.7t

Figure 3.4: Long Panel Estimates of Age, Cohort and Year Effects for a: and $1.
Identifications achieved through (1) model age, cohort excludes year effects; (2) model
age, year excludes cohort effects; (3) model age, cohort excludes year effects; (4) model

age, five-year-cohort, year effects; and (5) model age, three-year-cohort, year effects.

203

Age Effects

Age Effects

354
25—

15 "1

5 1 _, ..r '— \ \-‘"\\..\
-“, ...H 9—4) 0— - ‘“x_
U - v —— 3“H‘_'# ———————-——— —— —— ——\—— ~ ~ - —— —‘ A—‘ — —- # - m I) r
‘ta.
“fife.“ _ \ ~. ml .K‘k-
51 ea \-\\
— ' Tax. - -~ ' \
‘~)L.. \ “ \ ‘~‘ ‘\
j.‘ , ~-, _ \ \

‘3 ,\
‘Jp
iflA,‘
-1 5 _ {‘7
iffy»
'7‘» \
ff. ‘ , ‘
, \
1| . ‘ ~
25~ a u *
- , , 2 \
v‘r \ ‘
.‘ ‘ , \
\.
-35 ‘

Aw~

 

 

 

 

.____ __.—*— —

T

20 25 30 35 40 i5 50 55 60 65
age
Model y=O.5a-O.04asq+0.3c-O.020sq+0.7t-O.01 tsq

_T" — l T T f l

 

 

/ / a H a \_ _A\
”/— — I — 7A ‘ x ‘\
15~ //’ *5» ‘5
\,
r\ \
/ ,/’*"‘/ _ “h x““~~
5 A / /// \“x \
P ¢+—4— ‘3 \\-\
k “~,
_u 1. "v" F” Jaw}; ,__ — _-_____ ,A #_ _‘E — -7 i,
_ 17”“
\‘
‘5 —-l 6 Q \
“6“,
1338
X
‘9‘
‘51
J

 

l l l T l l l W

25 25 30 35 4o 45 50 55 a) 1%
age
Model y1 =0.5a-0.04asq+0.5c—0.04csq+0.7t—0.01 tsq

204

Cohort Effects

Cohort Effects

35<l
25—1

15*

ff \
Z, \
5 ~ ’// r,— — I ’_ _' _- \ — ~—.
f, __.. .,_A / . __ _ ___.- —— -———~ _- w
,/"' .- —— ~-- ,1 ; ~—— \\_._. - .\
F .67“ {3:} foflff}- 9“)“ 4; .- _ ~42“ __.—~—
** -<; 9 1.; 6 ~
' ‘c3\_ \
‘ 913‘
5 __1 ‘fi—N‘}
(3k

-15 J '1

—25 j

 

—35 r
1

 

 

1920 1930 1920 1950
cohort

Model y=0.5a-0.04asq+0.3c—0.0205q+0.7t-0.01tsq

351

 

 

_ _ /
— _ —_‘ r5 _ a. ‘V‘ ‘ g _
-- *"“ ,. _— _ ..—_‘ H‘ \
If} _ __ ”a" \\ — W
5 H 4.53 \ —
gr . \‘ ‘ _fl _—
4 _ A“ - — -— \ \ V l
3 45» H 4}”4} H ‘6 6% \‘\c_ k —-
., .c _* \\
k t}‘ A \
\I\& H \
\9
5 f“
_ 6‘
-
l J ‘3 \ \
‘\
We; \

\
_4 ‘K,
—1 5 a
N1
"‘9
it
\
-25-— a
t r,
k I“

-35 '5

 

 

1920 1950 1920 1950
cohort

Model y1 =0.5a-0.04asq+0.5c—0.04csq+0.7t-0.01tsq

205

 

 

 

 

 

 

3 _,
2 r ,,,/ ,w/“fi ”
,//
1 a //,,
£2 C *———"””_”‘—“”";:A:;:::
8 , -. -2 -— —— L—ngwflfflfg, Ea
t :_.:_._:_:..r-———--——-
m 0 .1
<7:
G)
>—
-1 .1
-2 -
-3 l
1981 1982 1983 1984
year
Model y=0.5a-0.04asq+0.3c—0.02csq+0.7t-0.01 tsq
3 —+
2 - fl’f’f/m/ ,9 I"
1 a /W/, MM“
59. CV”
O
o
3:
Lu 0 9
a QTQ ‘v M- _-
m \x ,_ .5 5 __ k
> \\\;f ~ ~ _ a
AP I A "‘M=""~*- 1::‘j-~‘c.‘_‘
-2 ~ 1
-3 a
r , 1 1
1981 1982 1933 1984
year

Model y1 =0.5a—0.04asq+0.50—0.04csq+0.7t-0.01 tsq

Figure 3.5: Short Panel Estimates of Age, Cohort and Year Effects for y and y1.
Identifications achieved through (1) model age, cohort excludes year effects; (2) model
age, year excludes cohort effects; (3) model age, cohort excludes year effects; (4) model

age, five-year-cohort, year effects; and'(5) model age, three-year—cohort, year effects.

206

Age Effects

Age Effects

-35 a
—45-1

 

-25 l

 

L.

"‘ f'
'— 4F‘: “31:“... 3 _; 3.;.3 _ __._ 3” r __ ~3wufi_
' ~ ‘943 -
‘ "ii. a.
“"459 - \
EN
‘fik *1
c \‘u .
a 313. _
YJ‘. \
0%,
“:1 x
‘V‘Ik \
\E;\
Bk,
‘3.
f“
‘5),
"f1
0
U
"a
,5
,0

 

 

20

j
25

l

30

i l “—TFFF—F _l—h—"Wgé‘ l

..T_
35 4O 45 50 60 65

age
Model y=0.5a—0.04asq+0.3c-0.020sq+0.7t—0.01 tsq

 

 

 

 

-/ \."‘\
. , r 2”)!“ \\
.-I #:17/ \
'j / //—— \\—\“ \
l.» ‘/—-"”P "\\\ ‘ \‘
.. - 0" 0 NEW“ 9»
r _~: _ 5’ {‘filgtzzgflf _. ._ >~ N; ——
,. J,‘ 3‘ ,
:rq‘f‘m‘ \\ \ V3.
V R ~
NL‘W“
j\
Y‘L\
it“: \
J‘» - \
\ti. ’. \\
“A.
‘9“ \
‘3‘ \
“SK
\‘TK \
CE \
‘Q.
‘91
KAN
~{"\
‘51
\G \
5,
\Qg
'53
' a

 

—1"

20

l

25

l

30

l l l l ’- 1

35 40 45 50 55

age
Model y1 =0.5a-0.04asq+0.5c-0.04csq+0.7t—0.01 tsq

207

Cohort Effects

Cohort Effects

es-J
55-7
45-
35-
25..
15-—

5 —l [ff-#14"; “f/ \__ ; 4 :;;A~::f _ __ A_-___
_d__..A,-_.y:z:‘_—‘ _2': . “ — ‘1 . ‘— R».;_ ‘
4:7 '4" L64? -—4£¢—-{+ 0 +3 —+} 43.“. ~H.€, - ' P
(Y _V v _t’ ‘Q—+)__9__“fi> ..
\ ; “+4 ~ _
” W‘W ( _
' :’~ ~ ‘»$““%— A.
-5 - ju-$"~“‘E‘~<.g
' -—V‘L<c, ,
fl‘wt -{
”\fx

.45~ “fie-
-25*e

-35 -l
r ‘ l ‘ f
1950 1950 1940 1950 1960
cohon

Model y=0.Sa—0.04asq+0.30-0.0Zqu+O.7t—0.01 tsq

 

 

 

 

15 u_/, _ _ \s‘w-
— _ :’_'_ _/ _— .\
. V ._ .—.- f ‘4 Aa— ‘1 _
”’1‘“ x
sj ;:;M- ~\\\; \~n-
__-—«’7’73' ~ R — - ~\ , " ,
Wrg*+j’"°:’—t’_3—fi t‘""“ '3' “9 *+‘)«_+} ~ \“‘\ _.
‘i'k «_ ~(kfia \\\\
‘Q- A \~
-5 -1 we. \ \
K‘R‘Mp \ K
_ K‘s; \\
-15 3 '9 \
R\ \
ft \

 

 

l T __
1950 1950 1920 1950 1960
cohort -

Model y1 =0.5a—0.04asq+0.Sc-0.04csq+0.7t-0.01tsq

208

Year Effects

Year Effects

-10—J
-15 -'1

-20 r

—25—:

 

 

 

-30 4

 

T—- .____.__3_,___~T__

‘ . r ' ‘ T 1“
1981 1 986 1 991 1996 2001

year

Model y=0.5a-0.04asq+0.3c-0.0chq+0.7t—0.01tsq

4'7?" "“ 4:)

~~a ‘

_103 A _ \_\

-15-a

—20 ‘1

-25 x

-30 *

 

1951 1956 1951 19b6 2061
year

Model y1 =0.Sa—0.04asq+0.5c—0.04osq+0.7t—0.01 tsq

Figure 3.6: Long Panel Estimates of Age, Cohort and Year Effects for y and y1.

Identifications achieved through (1) model age, cohort excludes year effects; (2) model

age, year excludes cohort effects; (3) model age, cohort excludes year effects; (4) model

age, five-year—cohort, year effects; and (5) model age, three-year-cohort, year effects.

209

3.4 BMI Profiles in CHNS

Figure 3.7 shows the OLS estimates of the second differences of the age and
cohort effects using 1989, 1991 and 1993 data assuming the data generating process
3.1 is true for all individuals. The asterisks are from the approach based on repeated
cross-sections, or the mean approach, and the connected lines are based on the genuine
panel data approach, or the level approach.

The two approaches both indicate much variation in the slopes of the age and
cohort effects after age 55. The standard errors based on the mean approach need
to be corrected. No inference is based on the mean approach without correcting the
standard errors. The standard errors in the level approach are valid provided that
shawl (u, ac, flu, 7,) ~ i.2'.d.(0, 02). The F tests of the null hypothesis that all changes
in slopes equal to zero or a constant based on the individual level data are valid.
The F-statistic for zero changes in male age effects is 1.71 (p=.0005), for female age
effects is 1.17 (p=.1724), for changes in male cohort effects is 3.03 (p=.0000) and
for changes in female cohort effects is 1.57 (p=.0026). This suggests that the age
and cohort profiles are not quadratic. The growth rate in BMI is relatively constant
around age 45.

Since the identification of the age and cohort effects are more reliable for those
aged 20 to 45, we test the null hypothesis that the changes in slopes of the age effects
in this range equal to zero or a constant and the the null hypothesis that the changes
in slopes of the cohort effects in the range of 1944 to 1973 equal to zero or a constant.
The F-statistic for changes in the age range for male age effects is 1.25 (p=.1881), for
female is 0.55 (p=.9620), and the F-statistic for changes in the cohort range for male
cohort effects is 1.62 (p=.0191), and for female is 1.59 (p=.0233). As we can see we
can not reject the null of zero changes in the slopes of age effects in this range for
men nor women. However, the changes in the slopes of the cohort effects from 1944

to 1973 are different from zero.

210

Using 1989, 1993 and 1997 data the changes in the slope of age and cohort effects
have similar shape as in the two-year spaced case. Figure 3.8 indicates less variation
in some ranges of age and cohort profiles for men and women. Table 9 shows the
regression results of equation (3.11) with two—year spaced and four-year spaced data.
Overall, there is less significant changes in five-year differences. This can be viewed
as evidence that we could regroup our data into five or three-year cohorts, which
essentially imposes the restriction that within each cohort group the cohort effects

are the same.

3.5 Estimate the Restricted Level Effects

As demonstrated before and verified by the software in our application we need
two meaningful constraints to identify all age, period and cohort effects. By grouping
our data into five- or three-year cohorts we are in effect imposing more constraints
than necessary. The design matrices are full rank in both cases. This is the first
identification strategy.

The second strategy is to maintain some cohorts as single-year cohorts and group
others, based on the changes in slopes in Table 3.9. The threeyear cohorts are July
1906 - June 1909, July 1909 - June 1912, ..., July 1972 - June 1975, and the last two
years July 1975 - June 1977. Horn Table 3.9, we can see the changes from 1927 to
1931, 1933 to 36, 36 to 39 are significantly different from zero for men and changes
from 1939 to 42 and 63 to 66 are not zero for women. We can break these cohort groups
down to each original birth cohort year. Doing so we still have enough restrictions to
identify the system but do not impose what might be wrong constraints. Similarly,
for the five-year cohorts specification the cohort groups are July 1906 - June 1912,
July 1912 - June 1917, ..., July 1967 - June 1972, and July 1972 to June 1977. We
break the groups 1952 - 1957 and 1957 - 1962 down to each original birth cohort

211

Table 3.9: Second Differences of Cohort Profiles

 

 

 

'fim-year Spaced Data.

Four-year Spaced Data

 

1989, 91, 93 1989, 93, 97
Differences Male Female Differences Male Female

1915-1914-(1913—1912) 4.249 2.247 1915—1914-(1911-1910) 8.710 1.537
(2.416)‘ (2.177) (3.331)“ (3.601)
1916-1915-(1914-1913) -0.772 -2.787 1916-1915—(1912-1911) -0.033 -4.652
(2.237) (2.309) (3.846) (3.601)
1917-1916-(1915-1914) -3.077 -0.426 1917-1916—(1913—1912) -3.387 2.028
(2.237) (2.309) (3.846) (3.601)
1918-1917-(1916—1915) 0.925 3.490 1918-1917-(1914-1913) 3.942 2.180
(1.581) (2.177) (2.720) (2.401)
1919—1918-(1917-1916) 0.798 -2.792 1919-1918-(1915-1914) -5.021 -0.476
(1.708) (2.177) (2.720)* (2.172)
1920-1919-(1918-1917) -0.672 2.121 1920-1919—(1916-1915) 0.591 2.715
(1.708) (1.686) (2.355) (2.547)
1921-1920—(1919—1918) 0.487 -1.437 1921-1920-(1917-1916) 1.958 -4.764
(1.708) (1.686) (1.848) (2.547)‘
1922-1921-(1920-1919) -0.293 0.988 1922-1921-(1918-1917) 0.615 0.380
(1.708) (1.633) (1.848) (2.401)
1923-1922-(1921-1920) 0.622 -1.286 1923-1922-(1919—1918) -0.382 -2.023
(1.528) (1.633) (1.923) (2.401)
1924-1923-(1922-1921) -1.791 0.400 1924-1923-(1920-1919) -3.301 0.007
(1.528) (1.425) (1.923)‘ (1.747)
1925-1924-(1923-1922) 1.485 3.250 1925-1924-(1921-1920) 2.590 -1.116
(1.234) (1.425)” (1.923) (1.572)
1926-1925-(1924-1923) 1.707 -2.454 1926-1925-(1922-1921) -0.034 -1.156
(1.118) (1.234)“ (1.923) (1.801)
1927-1926-(1925—1924) -0.641 -1.488 1927-1926—(1923—1922) -2.661 0.558
(1.099) (1.234) (1.528)‘I (1.801)
1928-1927-(1926—1925) -0. 767 2.053 1928-1927-( 1924-1923) -0.905 -0.782
(1.080) (1.137)* (1.332) (1.572)
1929-1928-(1927—1926) 1.443 -0.926 1929—1928-(1925-1924) -1.284 1.161
(0.854)* (1.104) (1.332) (1.572)
1930—1929-(1928-1927) -3.656 -0.895 1930—1929-(1926-1925) -0.637 -0.745
(1.015)“l (1.104) (1.259) (1.273)
1931-1930-(1929-1928) 0.799 0.206 1931-1930—(1927-1926) 0.384 1.892
(1.015) (1.074) (1.081) (1.273)
1932-1931-(1930—1929) 2.351 1.657 1932-1931-(1928-1927) -1.514 -0.263
(0.986)" (1.074) (1.081) (1.254)
1933-1932-(1931-1930) 1.162 -1.010 1933-1932-(1929-1928) 0.613 -1.262
(0.986) (1.020) (0.982) (1.074)
1934-1933-(1932-1931) -4.993 0.814 1934-1933-(1930-1929) 1.874 1.222
(1.015)" (1.046) (1.143) (1.139)
1935-1934-(1933-1932) 2.276 0.927 1935-1934-(1931-1930) 0.317 -0.077
(1.015)" (1.046) (1.143) (1.184)
1936-1935-(1934—1933) 3.519 -1.396 1936-1935-(1932-1931) -3.243 -0.839
(0.960)” (0.933) (1.067)" (1.184)
1937-1936-(1935-1934) -3.037 -1.916 1937-1936-(1933-1932) 1.255 0.903
(0.924)” (0.933)" (1.004) (1.098)
1938-1937-(1936-1935) -1.688 2.088 1938-1937-(1934—1933) 1.503 -1.361
(0.924)‘ (0.953)" (1.143) (1.019)
1939—1938-(1937-1936) 2.546 1.047 1939-1938-(1935-1934) -l.437 -0.548
(0.854)" (0.953) (1.143) (0.980)
1940-1939-(1938-1937) -0.617 0.358 1940-1939-(1936-1935) ~1.576 0.990
(1) (0.973) (1.081) (1.153)
1941-1940-(1939-1938) -1.012 0.933 1941-1940-(1937-1936) 1.114 2.234
(1) (0.973) (1.004) (1.153)‘
1942—1941-(1940—1939) 0.701 -1.873 1942-1941-(1938-1937) 0.426 -0.263
(1.015) (0.984)"l (0.962) (1.029)
1943-1942-(1941-1940) 1.363 -2.024 1943-1942-(1939—1938) -1.131 -1.436
(1.015) (0.984)" (0.962) (1.029)
1944-1943-(1942-1941) -0.250 2.343 1944-1943-(1940-1939) 2.272 0.548
(0.863) (0.943)" (1.126)" (1.074)

 

212

Table continua

Table 3.9 (cont’d)

 

Two-year Spaced Data

Four-year Spaced Data

 

 

1989, 91, 93 1989, 93, 97

Differences Male Female Differences Male Female
1945-1944-(1943-1942) -0.168 0.083 1945-1944-(1941-1940) -0.739 -1.081
(0.845) (0.943) (1.126) (1.074)
1946-1945-(1944-1943) -0.934 —0.256 1946-1945-(1942-1941) 0.007 1.062
(0.723) (0.798) (0.907) (0.980)
1947-1946-(1945—1944) -0.072 -0.206 1947-1946-(1943-1942) -0.104 1.275
(0.631) (0.692) (0.882) (0.980)
1948-1947-(1946—1945) -0.032 -0.191 1948-1947-(1944-1943) -0.545 -2.054
(0.523) (0.562) (0.814) (0.873)"
1949—1948-(1947-1946) -0.017 -0.322 1949-1948—(1945—1944) 0.194 0.524
(0.523) (0.530) (0.710) (0.759)
1950-1949—(1948—1947) -0.085 0.223 1950-1949-(1946-1945) 0.372 -0.609
(0.529) (0.521) (0.561) (0.618)
1951-1950—(1949—1948) 0.832 0.182 1951-1950—(1947—1946) 0.644 0.184
(0.529) (0.502) (0.561) (0.596)
1952-1951-(1950-1949) -0.349 —0.168 1952-1951-(1948—1947) -0.717 -0.533
(0.459) (0.498) (0.587) (0.590)
1953—1952—(1951-1950) -0.500 -l.021 1953—1952-(1949—1948) -0.009 -0.606
(0.459) (0.474)“ (0.587) (0.556)
1954-1953-(1952-1951) 0.355 0.541 1954-1953-(1950—1949) -0.391 0.616
(0.462) (0.466) (0.533) (0.556)
1955-1954-(1953—1952) 0.867 0.866 1955-1954-(1951-1950) -0.225 -0.052
(0.462)‘ (0.466)* (0.533) (0.530)
1956-1955-(1954-1953) -0.937 -0.342 1956-1955-(1952-1951) 0.369 0.237
(0.482)"I (0.450) (0.559) (0.523)
1957-1956—(1955-1954) 0.119 -0.123 1957-1956—(1953—1952) -0.159 1.674
(0.482) (0.459) (0.559) (0.524)”
1958-1957-(1956—1955) -0.175 -0.546 1958-1957-(1954-1953) -0.130 -1.438
(0.478) (0.533) (0.542) (0.629)"
1959-1958-(1957-1956) 0.010 0.286 1959—1958-(1955-1954) -0.137 0.231
(0.478) (0.533) (0.542) (0.629)
1960-1959—(1958—1957) 0.431 0.883 1960-1959-(1956—1955) 0.791 -0.600
(0.620) (0.617) (0.691) (0.660)
1961-1960-(1959-1958) -0.429 -0.208 1961-1960-(1957-1956) -0.580 -0.550
(0.631) (0.617) (0.702) (0.660)
1962-1961-(1960-1959) 1.433 0.930 1962-1961419531957) 1.583 1.696
(0.634)" (0.586) (0.702)" (0.586)"
1963-1962-(1961-1960) -0.883 -0.696 1963-1962-(1959—1958) -0.947 -1.487
(0.614) (0.586) (0.544)* (0.590)”
1964-1963-(1962-1961) -1.289 -0.604 1964-1963-(1960—1959) 0.256 1.037
(0.554)" (0.494) (0.702) (0.647)
1965-1964—(1963-1962) 0.815 1.092 1965-1964-(1961-1960) 0.581 0.292
(0.453)‘ (0.484)" (0.714) (0.647)
1966-1965(1964-1963) -0.010 -1.588 1966-1965-(1962-1961) -1.288 —1.613
(0.515) (0.589)" (0.653)” (0.613)”
1967-1966—(1965-1964) -0.757 0.751 1967-1966-(1963-1962) 0.726 0.690
(0.545) (0.589) (0.627) (0.604)
1968-1967-(1966-1965) 0.885 -0.406 1968-1967-(1964-1963) 0.882 -0.691
(0.545) (0.550) (0.582) (0.602)
1969-1968-(1967-1966) 0.801 0.226 1969-1968-(1965-1964) -0.473 -0.237
(0.502) (0.550) (0.525) (0.611)
1970-1969.(1968-1967) -1.354 0.615 1970—1969-(1966-1965) -0.211 0.685
(0.502)” (0.510) (0.555) (0.644)
1971-1970—(1969-1968) 0.117 -1.051 1971-1970-(1967-1966) 0.383 0.303
(0.638) (0.667) (0.624) (0.658)
1972-1971-(1970-1969) 0.969 -0.902 1972-1971-(1968—1967) -0.371 -0.056
(0.758) (0.829) (0.624) (0.647)
1973-1972-(1971-1970) -3.588 0.642 1973-1972-(1969—1968) -0.734 -1. 124
(1.291)" (1.425) (0.656) (0.791)
Number of obs 4284 4930 Number of obs 4287 4690
R-squared 0.0446 0.0205 R-squared 0.0203 0.019

 

213

because they are significantly different from zero as a group (Table 3.9).

Yet another way of breaking the perfect collinearity between age, period and
cohort is, instead of defining age as period minus cohort (July to the next June
cohort), to use the exact age at interview time (interview date — birth date). In this
case for those who were born before July the exact linear dependency between age,

period and cohort (July to the next June cohort) breaks down and the system can be

identified.

 

 

 

 

 

 

 

 

Men
to
‘6 4 -
:13
m 3 —
0
O
< 2 ~ K
g 1 - A [\
X
o x x x x x
3% xx K x K x
0 -1 x
o 1 x l
‘O
s 2 1 .
0
.2 .3 -
I l I I I T I I T
20 25 30 35 40 45 50 55 60 65 70 75
Age
Women
a:
‘6 4 -1
£3
Lu 3 a
g, x
< 2 4
‘6
m 1 —
0
.2
3 ° 7
m - .1
o 1
E -2 a
O
8
a) '3 ‘

 

 

214

Men

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1'9

2: a— l'

2 2~

8 1- “ x " a A

‘3' 01 AMY/l /‘ A AAxxVflugflfibA l‘

g .fv .11 11w 17m .V

g -2-1 v f V

3 ~3—

3 -4-

§ -5*

0’ 1 9'50 1 9'30 19'40 1 9'50 1 9'60 1 9'70 1 9330
Cohort

21 Women

.“3 4‘

(.2 ii A A A. A l: a as- x

3 0. Fl Hflll xllkfiCiA/K “PerlU‘

.9 ,1. 1‘] [Hill If x LIV" \JfllU

g 7" ‘V V V x v v VT

-: .2-

g 1

3 .4-

8 -5n

‘0 1 9'20 1930 1 9'40 1 9'50 1 9'60 1970 1 9130
Cohort

Figure 3.7: The Second Derivatives of Age and Cohort Effects for Men and Women
for BMI using two-year spaced CHNS 89, 91 and 93. * The approach for repeated
cross-sections based on means. — The approach for genuine panel data based on

individual level data.

215

Men

 

..V “'U

 

 

 

 

.4

 

4—

1
.2-1
.3-1
-4-1

2085 out. _o mo>=a>too ucooow

Age

 

 

60

 

 

 

X
55

 

45
Age

‘K
35

 

 

 

Women

 

_ a — _ _
2 4| 0 4| 2

— q
4 3

.3—4
.4—

Eootm oo< .o mo>=u>too vacuum

216

Men

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9

o

a:

3:

m

r: 3‘

.2 2.. x

o

2 ,_ J A L- ..

o

3 °‘ frl'T'l ' . . "

.2 x x x

é“ -1" V V g x

" . 4

8 2

'0 _3—4

C

° 1

0 .

:5 4 I I I I I I r
1920 1930 1940 1950 1960 1970 1980

Cohon

% Women

o 3“

a:

2‘ l A ll

0

2 . .1 ll

0 1" it II x

O x

.. ,_ 12,111 11.11.411.1le

l NIH/N \l WV“.

:- '1‘ '

.. l v ,. V 'v 1 l"

.2 x

a -2*

D x

u 31

C

° J

o u

:5 4 T I I I I I *I—
1920 1930 1940 1950 1960 1970 1980

Cohon

Figure 3.8: The Second Derivatives of Age and Cohort Effects for Men and Women
for BMI using four-year spaced CHN S 89, 93 and 97. * The approach for repeated
cross-sections based on means. — The approach for genuine panel data based on

individual level data.

217

 

   

 
 

 
 
  
 

..
s

O
O
a“
o
c

0“ '
"‘ .‘...' "‘ O
W \V." -;~.-- ~
0 Q ‘ .
\\

 

 

 

l
.0 ‘0
Q C V we,
.3. We’vsls-45vq V!
\/\ 1'" I
V \\ \ \ P
i " V’\ \ I \
N. I \"' \A\/
I I I l I ‘T‘
30 40 50 60 7O 80
Age
a
5M
m...
01
2°“
2'?“
9°1-
I
80

 

 

— iden age at interview
iden five-year cohorts
— — five.year cohorts no break

 

- — - - iden three-year cohorts
—-— three-year cohorts no break

 

 

Figure 3.9: Age Effects for Men and Women, identified through (a) exact age at

interview (b) three-year cohorts, (c) five-year cohorts specifications, ((1) three-year

cohorts without breaking down at certain groups into single years, (e) five-year cohorts

without breaking down at certain groups into single years.

218

 

     
  

      

 

 

 

 

 

09-1
§°‘
131:1 .’ —.\.’.\0_u——-o—I"—".\o-—'~"\o‘
8N' ~\"‘~——
2 “
'2" ° ' “My-:3»... --------- ~~o"'~-"“

“I..." \¢.:.::———————-————
‘1‘
I I I I I I I r
1910 1920 1930 1940 1950 1960 1970 1980
Cohort

3’1
[Llo-
tv-
2 - '

N.
84..
2 I
E"
If I I I f if I I

1910 1920 1930 1940 1950 1960 1970 1980
Cohort

 

iden age at interview
--------- iden five-year cohorts
— — five-year cohorts no break

 

— — - - iden three-year cohorts
— -— three-year cohorts no break

 

 

Figure 3.10: Cohort Effects for Men and Women, identified through (a) exact age

at interview (b) three-year cohorts, (c) five-year cohorts specifications, (d) three-year

cohorts without breaking down at certain groups into single years, (e) five-year cohorts

without breaking down at certain groups into single years.

219

 

 

 

 

  
 
 

 

 

 

 

 

“2-
9’1
5
O
)-
o“?-
a
2
o
T I I I f
1990 1992 1994 1996 1998
Year
'- 4 —— iden age at interview
gum —-—- identhree-yearcohorts _/-/
E --------- iden five~year cohorts
g“? ' —-— three-year cohorts no break
>- v .. —— five-year cohorts no break ’
.9 ”’
a x’
094‘
u. __ ’
—————— ’
o4 ---------- ””
1990 1992 1994 1996 1998

Figure 3.11: Year Effects for Men and Women, identified through (a) exact age at
interview (b) three-year cohorts, (c) five-year cohorts specifications, ((1) three-year
cohorts without breaking down at certain groups into single years, (e) five-year cohorts

without breaking down at certain groups into single years.

Figures 3.9, 3.10 and 3.11 are estimated age, period and cohort profiles for men
and women with the above five different identifying strategies. The most meaningful
profiles are based on the five-year cohorts strategies.

For men, the increasing trend by age (Figure 3.9) is not as clear as that for
women. The female BMI-age profile peaks around 50 and then starts to decline.
Shrinking is faster for female than for male. The hunch back shaped age profile for
women is consistent with What we find in figure 1.

The period effect for men (Figure 3.10) is linear and increasing over time at a
larger magnitude than for women. The strong linear year effect for men suggests as

we pool the data from all four years of survey in the main analysis for BMI it is

220

important to control for the year effect.

The cohort profile (Figure 3.11) for men is quite flat, which seems to be at odd
with figure 1. However, this may not be due to the failure of the model but because
we have a relatively short panel that does not allow us to fully estimate the cohort
effect. As we can see from figure 1, especially after we group our data to five-year
cohorts, at each age there are only two to three cohort groups available to identify
any cohort effect. If we have a longer panel the fitting will be improved.

There is a decreasing trend in female cohort profile and two dips around the

revolution and the pre-famine years.

3.6 Concluding Remarks

This APC exercise helps us to better understand the BMI distribution against
age, cohort and year for men and women. The age profile for women is inversely
U-shaped. The year effect for men is strong. There are not enough data to identify
cohort effects due to the short length of the survey, for any given cohort there are only
data for a short span of age. In our main analysis for socioeconomic determinants of

BMI we will be using the five-year cohort identification strategy.

221

APPENDIX

222

Appendix A: Survey instruments

in CHNS

The China Health and Nutrition Survey (CHN S) was designed to examine the
effects of health, nutrition, and family planning policies and programs implemented by
national and local governments and to see how the social and economic transformation
of Chinese society is affecting the health and nutritional status of its population.
The impact on nutrition and health behaviors and outcomes is gauged by measuring
changes in community organizations and programs as well as by measuring changes
in sets of household and individual economic, demographic, and social factors (CPC
online).

The survey instruments were designed by an interdisciplinary group of social
scientists and biomedical researchers with extensive experience in survey research on
these topics.

The health services section contains health insurance coverage of surveyed indi-
viduals, medical providers, and health facilities that the household might use. Infor-
mation on illness and uses of the health system during the previous month is asked
for some household members. Physical examinations that include weight, height and
blood pressures were given to all adults and children starting from 1991 and to adults
aged 20 to 45 in 1989. Physical functioning data were collected since 1993 for the

elderly.

223

.533 $3 :3 8:0 HE: .9353 3-33 :_ 8:5 853 boa—.588 0.8? 3:585 boo—n 0:335 “Ed ozoummms

.9325 3.32 :_ 8:3... 8:: “58:32: me? So“ :6? 383.9»
.925ch 250823355” 96: 32% $3 5 nv 0... cm own 9:33 12:01
4858:: 9.8. cox? 953 2382: 22:8

:5 $3 5 329:? Rama b5 33 :3 av o» 3 vows ciao.“ 203 8 .33 E 35:33 Eons woe—ma 0.83 .620 93 3 2d of» 23:88 03::ch

.898 8:5 no .298 533088 .298 go: we cocoa 03 muoBm:<a
:oo: .8 £8.“ 45%

55:83 ma cocoa 08 20393. £va 59:: 235:: 888 E393: 5 “3933 mm .8585: 2:3 2: 53 5 39:5 33 E 032E?» co: m_ Ema 3:9...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

\. \. \, \, AME 8:5 05.58:: woo—n 0:835

\. \ \. \. 3 am :25 0:583 6003 0239mm

\. \. A83 8:235:86 $35

\ \. AEov eggflasoho :5

x x x x a Asa; as 5% 385

\ \. \. \, A83 8:238:88 :Ec Saab

x x x \ 3.5V .59:

x x x x a as; 253

\. \. \ X o SEEMEQ NAKED .mmémb .5-NHD 6:0.

\ \. \ «$95: 8 26 833505 31% :_ 825E€ 5:5 9395 3.95 .8435 .3-an

\, X \. $85: 8 2% 89>:qu 38v :_ 852555 was .259: 8:: pawn 3-va .3va .HmémD
\. \. S \. $363 “8:88 93 E wonzgamo: whee 133:: 30E .mmE

x x x \ 326s use a: 5 83338.. was .39: 3o: .32

\. \. \ «805: m1: 3 26 85353 ESE: use >58 3 05.2: gap .358 Rom .832

\. \. \, \. n was—83 58 ama— 23 :33? 33.5 .8 56:5 2: mm? 29% 30m .32

\, \. \, a New: 5% 2:8: 550 we 55 3 $8.388 5133 59m 2:88p :oz 2:03 30m .32

 

bag” was: 2:3 awe“

nausea—human

868 @838 :23: com 35:5me racism ”H..<. wish.

224

0503:8003 00800.50 000:: 0030:: :08 0:00.» 03: E 3:030: :06 0:0:0EQ .0300 m: 03 D 00.» m:

«.33 o: 8398 5:3 0: 5030: 0:: .:0>03o: .33 :H .0870 0: ED 00.» m: :03 o: 830:0 :0:3 0:. 5030: :30 0:: .33 5 ..00000500 0:0 w:E::=0
E 530030 020: :0.» 00.. 0:0 m3: 0:0 Eb ...w..w::::oam 30:33 0:030 30: 0 338:0 5 3:00:30 030: :0.» 00.. 0:0 3030000 033 0:0 83..

.09 0.:00 :0 .302 0002 ..Ggofimfl 080m 3:30:39 02 00 00000 0:03 0:02:20 .33 :H on 0.:00 :0 iguana 30> $030533 080m

.2353: 03:5 < 500858: 02 00 00000 0:03 E0330 .ma3 :: 5:08.50 0:0 3 0:0 «:3 5 w:_0oO 000388 .003 000:9: m :00: 0:: E 005:0:mm0
.3058 .33 :: 0:0000: 0::0::0500:0 0:: 0... w::0:0000 .33 :3 mm :0>0 0:0 33 E on :0>0 0:09:08 3:0 09:00 0:03 30300:: 0:3 JG“:

 

 

 

ES :5 $2.8 2: 0%: .2
25 cm: .32 :5: 0:58.00 .3
:5 05 :28: .8 053:: .0:
$5 20 .0530 0:
$5 20 2.0.8.0 a 05.3. .2
85 NS .08.: :85: .580 magnum .2
85 :5 08202:: n 0505 .m
35 80 was: so: a 0.58:0 .0
85 E: .: ...oaofium 28 0.58:0 0.
$5 $0 09. 05.230 .0
$5 80 .059: m a: mafia ...,
85 no: :58. a $28 05:03 .0
$5 :00 $.88. 8.: 0:303 .0
$5 $0 $38.22 a 0:303 0
$5 80 $3082: a 0555: .:

 

30: 80: as 8:32.50 96.: so: on
3-3 mzmo a 2.2.8.8 2:... :0... 0.... 2.0:.

 

 

225

Table A3: Adult BMI by Education Levels in Rural and Urban Areas

 

 

 

 

 

 

 

Male Female
All Urban Rural All Urban Rural
No Formal Education
Number of observations 1126 261 865 3779 936 2843
Median BMI 21.25 22.76 20.94 21.64 22.88 21.33
Mean BMI 21.79 23.11 21.39 22.17 23.27 21.81
(3.40) (3.49) (3.27) (3.92) (4.41) (3.67)
% Undernourished (<18.5) 10.04% 3.83% 11.91% 12.60% 11.00% 13.12%
(0.30) (0.19) (0.32) (0.33) (0.31) (0.34)
% Overweight (225) 14.21% 28.35% 9.94% 18.26% 30.02% 14.39%
(0.35) (0.45) (0.30) (0.39) (0.46) (0.35)
% Obese (230) 1.24% 1.92% 1.04% 2.73% 5.13% 1.93%
(0.11) (0.14) (0.10) (0.16) (0.22) (0.14)
% Overweight high risk 0.36% 0.38% 0.35% 5.48% 11.32% 3.55%
(0.06) (0.06) (0.06) (0.23) (0.32) (0.19)
Some Primary Schooling
Number of observations 2256 525 1731 2426 563 1863
Median BMI 21.12 22.10 20.94 21.75 22.60 21.50
Mean BMI 21.58 22.63 21.26 22.24 23.04 21.99
(3.43) (4.31) (3.05) (3.60) (3.60) (3.57)
% Undernourished (<18.5) 11.44% 10.67% 11.67% 9.03% 7.99% 9.34%
(0.32) (0.31) (0.32) (0.29) (0.27) (0.29)
% Overweight (225) 10.64% 21.90% 7.22% 15.70% 25.58% 12.72%
(0.31) (0.41) (0.26) (0.36) (0.44) (0.33)
% Obese (230) 1.55% 2.86% 1.16% 2.31% 4.44% 1.66%
(0.12) (0.17) (0.11) (0.15) (0.21) (0.13)
% Overweight high risk 0.75% 1.90% 0.40% 4.66% 9.41% 3.22%
(0.09) (0.14) (0.06) (0.21) (0.29) (0.18)
Primary School
Number of observations 2640 567 2073 2355 573 1782
Median BMI 21.26 21.83 21.12 22.00 22.86 21.78
Mean BMI 21.63 22.30 21.45 22.53 23.17 22.33
(3.03) (3.59) (2.84) (4.65) (3.76) (4.89)
% Undernourished (<18.5) 9.09% 11.11% 8.54% 6.88% 6.28% 7.07%
(0.29) (0.31) (0.28) (0.25) (0.24) (0.26)
% Overweight (225) 9.77% 19.75% 7.04% 18.17% 25.31% 15.88%
(0.30) (0.40) (0.26) (0.39) (0.44) (0.37)
% Obese (230) 1.21% 2.29% 0.92% 2.12% 3.32% 1.74%
(0.11) (0.15) (0.10) (0.14) (0.18) (0.13)
% Overweight high risk 0.34% 1.23% 0.10% 4.16% 7.16% 3.20%
(0.06) (0.11) (0.03) (0.20) (0.26) (0.18)

 

226

Table continues

Table A.3 (cont’d)

 

 

 

Male Female
All Urban Rural All Urban Rural
Middle School
Number of observations 4173 1242 2931 3290 1227 2063
Median BMI 21.23 21.62 21.13 21.48 21.91 21.35
Mean BMI 21.76 22.16 21.59 22.06 22.53 21.78
(3.20) (3.58) (3.00) (3.63) (4.10) (3.29)
% Undernourished (<18.5) 7.36% 9.10% 6.62% 8.69% 7.66% 9.31%
(0.26) (0.29) (0.25) (0.28) (0.27) (0.29)
% Overweight (225) 11.10% 16.75% 8.70% 14.71% 20.29% 11.39%
(0.31) (0.37) (0.28) (0.35) (0.40) (0.32)
% Obese (230) 1.29% 1.53% 1.19% 1.82% 2.69% 1.31%
(0.11) (0.12) (0.11) (0.13) (0.16) (0.11)
% Overweight high risk 0.38% 0.32% 0.41% 3.10% 4.48% 2.28%
(0.06) (0.06) (0.06) (0.17) (0.21) (0.15)
High School
Number of observations 1655 761 894 1260 713 547
Median BMI 21.62 21.97 21.28 21.44 21.55 21.24
Mean BMI 22.10 22.54 21.73 21.96 22.01 21.89
(3.42) (4.03) (2.75) (3.50) (3.34) (3.69)
% Undernourished (<18.5) 6.65% 7.88% 5.59% 9.29% 9.40% 9.14%
(0.25) (0.27) (0.23) (0.29) (0.29) (0.29)
% Overweight (225) 14.98% 19.58% 11.07% 14.92% 15.85% 13.71%
(0.36) (0.40) (0.31) (0.36) (0.37) (0.34)
% Obese (230) 1.27% 1.97% 0.67% 1.19% 1.12% 1.28%
(0.11) (0.14) (0.08) (0.11) (0.11) (0.11)
% Overweight high risk 0.73% 1.18% 0.34% 1.90% 1.82% 2.01%
(0.08) (0.11) (0.06) (0.14) (0.13) (0.14)
Technical School / College
Number of observations 1015 754 261 680 526 154
Median BMI 22.31 22.29 22.40 21.43 21.44 21.40
Mean BMI 22.67 22.71 22.55 22.01 22.07 21.83
(3.21) (3.25) (3.07) (3.22) (3.35) (2.73)
% Undernourished (<18.5) 6.50% 6.63% 6.13% 8.68% 8.56% 9.09%
(0.25) (0.25) (0.24) (0.28) (0.28) (0.29)
% Overweight (225) 20.59% 22.15% 16.09% 15.88% 16.54% 13.64%
(0.40) (0.42) (0.37) (0.37) (0.37) (0.34)
% Obese (230) 2.07% 2.39% 1.15% 1.62% 2.09% 0.00%
(0.14) (0.15) (0.11) (0.13) (0.14) (0.00)
% Overweight high risk 1.08% 1.33% 0.38% 3.82% 3.61% 4.55%
(0.10) (0.11) (0.06) (0.19) (0.19) (0.21)

 

 

Note: Standard deviations are in parentheses. Having high risk or high abdominal
fat is defined as having waist circumference > 102 cm for male or >88 cm for female.
Waist circumferences were not measured before 1993.

227

5523 Nov m_ 8.58 8ng 958925: mafia: ago—038:0: no 250qu omega 2: 385888 a 5333
$3 E 613358 3m ._o>o_ 20595: Bob anmuwww 08 QR 23. 52522: go: 393 985856 £53 98 3:93: 2: 33 E 6qu

 

 

 

 

5o 2:. So 3o 85 3o 85 85 8.0 86 2:. So 3288 6:2 as
w; 85 25 8o 2:. :d and 3° So was 8d 85 3238 meow &
a3 86 «so 33 85 as “no NS was and Se and sees 23: 50> .5
So as 36 :8 5o 93 mg S... was as a; and Seuss oz .5
mod 85 So So So :3 8d 85 2:. 3o 86 35 some 550 a
8d So So So So So as so So So So 2:. E8 02 0a
o; a; $6 mg as and a; 3o ”3 who as 93 ea 58 6:8. a
so «as 35 25 «no one a; :3 33 mod and :d 8E8 Ea. &
35 3o 23 3o 85 3o 85 2:. 3o 85 8d 8.: £3838? E69 as
as Ed was 85 was was 85 :3 Ed 3d woo «3 554225 8:09 .5
N3 33 as and as as as who Gd as a; 33 532a 5:3 .x.
85 :3 8.0 So So So So So So Ed 2:. 3o 52 .93 5% .323 $55 as
25 So So as 85 «S as :3 Ed and mod w; =03 88 as? .5
3o 3... :8 ”no 25 and a; as was $3 mad as @8885 8:3 g
:3 So So 2:. So So So cod 2; we 8: 8333 £32 256
8d mg o; :5 «we 86 So 85 85 a; 8: m8; 256
cod So 2:. 2:. Ed 3o 85 So as m: 8: was: 256
3o 25 23 mod Se :5 8d 25 mod ...2 8i 2325a 256
8.2 8.... 5: as new com 23 m: was HE a: 2k 3 mo BE
8.2 3.2 8.2 a; was as 3s a: 8s 3... a: was “81:0 SE
2.3 2.2 3.: 88 3.» was :3 Es 3.... Rs 2.... mg 88 Ho 85
5.... new as we". 2:. 3.4. mm.” «.3. 8...” 3.... 88 a? ammo a0 85
a.“ new 3.“ :3 a3. 43 ...3 2: NE a: a? a: as? we 85
3.... NZ ”3 m3 3.” 84. m3 :4 84 SA a: e: 8: ea 85
8.x 3.3.. 3.3 23 32 8.2 a; a: 9:. as 8.8 $2 owes 383 858528
32 NS: 2.2 as :s 3% «.5. was. 8.". as as was was: Sic; 85%
8.8 an: :3 8.: $2 8.2 8.2 8.2 mm: 3.8 2.3 8.2 :38 we as x83 .x.
saw... was was. SS 8.2 Ed. 2.3 was 3:. :3 new 8.9. $33254 a as
9me 32m :38 32a «.88 SE mama 39a 33“ 3:: main 2.88 833$
_eé :85 3. 35m 535 3 35m :85 3 35m 935 5.
aaafi mama Haas awafi

 

 

3-me mzmo s 8288855 585888 “3. 2.3.

228

Appendix B: First Stage

Regressions for Section 1.6

These are the first stage regressions in Table 1.13 where BMI in 1991 is instru-

mented with information from 1989.

Table B.1: First stage regressions for Table 1.13

 

 

 

Variables Male Female
Some primary 91 0.030 -0.003 0.295 0.295
(0.237) (0.238) (0.229) (0.230)
Primary 91 -0.029 -0.056 0.610 0.610
(0.239) (0.240) (0.247)" (0.247)**
Middle school 91 0.109 0.072 0.091 0.095
(0.245) (0.246) (0.263) (0.264)
High school 91 0.345 0.310 0.046 0.050
(0.289) (0.290) (0.345) (0.345)
Tech/College+ 91 0.022 -0.008 -0.969 -0.955
(0.337) (0.338) (0.470)“ (0.473)“
Log real prod asset 91 1a -0.053 -0.056 0.059 0.059
(0.052) (0.053) (0.060) (0.060)
Log real prod asset 91 2‘ 0.079 0.079 -0.101 -0.101
(0.052) (0.052) (0.060)* (0.060)*
Land farmed in 90 0.046 0.044 -0.007 -0.007
(0.033) (0.033) (0.040) (0.040)
Married 91 0.329 0.081
(0.267) (0.497)
Divorced Separated 91 -0.304 0.309
(0.874) (0.988)
Widowed 91 0.022 0.130
(0.477) (0.585)
Log R Productive Asset spline 1 in 89 -0.122 -0.131 -2.033 -2.035
(1.240) (1.240) (1.508) (1.510)
Log R Productive Asset spline 2 in 89 0.461 0.474 0.714 0.717
(1.189) (1.190) (1.421) (1.423)
Land farmed in 88 -1.106 -1.050 -0.678 -0.680
(0.835) (0.838) (1.296) (1.296)

 

 

229

Table continues

Table B.1 (cont’d)

 

 

 

Male Female
Price Rice*Log R Prod 51 89 -0.017 -0.011 0.064 0.066
(0.237) (0.237) (0.262) (0.262)
Price Rice*Log R Prod 32 89 0.005 0.004 0.201 0.199
(0.242) (0.242) (0.259) (0.259)
Price Rice*Land 88 0.095 0.092 -0.058 -0.057
(0.098) (0.098) (0.115) (0.115)
Price Egg*Log R Prod 81 89 -0.003 -0.004 -0.049 -0.049
(0.043) (0.043) (0.052) (0.053)
Price Egg*Log R Prod 52 89 -0.042 -0.041 0.041 0.041
(0.046) (0.046) (0.056) (0.056)
Price Egg*Land 88 -0.007 -0.006 0.007 0.007
(0.023) (0.023) (0.030) (0.030)
Price Pork*Log R Prod 81 89 -0.050 -0.050 0.111 0.111
(0.067) (0.067) (0.079) (0.080)
Price Pork*Log R Prod 32 89 0.074 0.073 -0.102 -0.103
(0.075) (0.075) (0.086) (0.086)
Price Pork*Land 88 0.013 0.011 0.028 0.028
(0.046) (0.046) (0.054) (0.054)
Price Fish*Log R Prod 81 89 0.032 0.032 -0.001 -0.001
(0.023) (0.023) (0.026) (0.026)
Price Fish*Log R Prod 52 89 -0.027 -0.027 0.019 0.019
(0.024) (0.024) (0.028) (0.028)
Price Fish*Land 88 0.016 0.016 0.005 0.005
(0.014) (0.014) (0.013) (0.013)
Clinic Physician*Log R Prod 81 89 -54.763 55.600 -73.328 -73.211
(40.758) (40.768) (49.170) (49.215)
Clinic Physician*Log R Prod 52 89 58.932 59.749 56.078 56.089
(40.179) (40.191) (49.316) (49.374)
Clinic Physician*Land 88 6.544 10.576 -12.828 -13.092
(43.261) (43.361) (51.190) (51.243)
Clinic Nurse*Log R Prod 81 89 91.537 87.161 342.405 343.472
(214.023) (214.068) (262.008) (262.194)
Clinic Nurse*Log R Prod s2 89 -255.278 -247.959 -296.318 -296.527
(236.251) (236.321) (292.011) (292.189)
Clinic Nurse*Land 88 1277.351 1470.020 -488.022 -489.179
(3070.487) (3074.360) (644.643) (645.153)
Clinic Bed*Log R Prod 81 89 -109.046 -106.985 -4.284 -4.665
(107.524) (107.548) (130.627) (130.712)
Clinic Bed*Log R Prod 52 89 176.883 172.719 12.951 12.986
(118.294) (118.347) (146.623) (146.718)
Clinic Bed*Land 88 8.459 1.093 24.826 25.516
(86.225) (86.399) (102.190) (102.311)

 

230

Table continues

Table 8.1 (cont’d)

 

 

Male Female

Water Source 2*Log R Prod s1 89 -0.354 -0.354 -0.430 -0.427
(0.302) (0.303) (0.323) (0.323)

Water Source 2*Log R Prod 32 89 0.168 0.163 0.213 0.211
(0.286) (0.286) (0.308) (0.309)

Water Source 2*Land 88 0.095 0.094 -0.103 -0.103
(0.093) (0.093) (0.097) (0.097)

Water Source 3*Log R Prod 81 89 0.113 0.109 -0.200 -0.202
(0.377) (0.377) (0.450) (0.451)

Water Source 3*Log R Prod 82 89 0.021 0.021 0.291 0.293
(0.367) (0.367) (0.454) (0.455)

Water Source 3*Land 88 0.120 0.115 0.088 0.085
(0.211) (0.211) (0.197) (0.198)

Water Source 4*Log R Prod 81 89 0.064 0.070 0.219 0.219
(0.204) (0.204) (0.242) (0.242)

Water Source 4*Log R Prod 52 89 -0.014 -0.022 -0.269 -0.270
(0.220) (0.220) (0.249) (0.250)

Water Source 4*Land 88 -0.348 -0.360 -0.117 -0.116
(0.162)** (0.162)** (0.212) (0.213)

In house no flush*Log R Prod 81 89 0.845 0.857 1.436 1.434
(1.040) (1.041) (1.269) (1.270)

In house no flush*Log R Prod 52 89 -0.805 -0.824 -0.042 -0.037
(1.008) (1.009) (1.227) (1.228)

In house no flush*Land 88 0.875 0.822 0.294 0.295
(0.802) (0.804) (1.252) (1.254)

Outside toilets*Log R Prod 81 89 0.271 0.260 1.438 1.437
(0.955) (0.956) (1.192) (1.193)

Outside toiletsI*Log R Prod 32 89 -0.449 -0.441 -0.359 —0.357
(0.922) (0.922) (1.141) (1.142)

Outside toilets*Land 88 0.695 0.642 0.582 0.583
(0.808) (0.810) (1.254) (1.255)

Open pit cement or earth*Log R Prod 31 89 0.389 0.398 1.665 1.661
(0.899) (0.899) (1.114) (1.115).

Open pit cement or earth*Log R Prod 32 89 -0.614 -0.627 -0.717 -0.713
(0.869) (0.870) (1.065) (1.066)

Open pit cement or earth*Land 88 0.706 0.663 0.459 0.460
(0.783) (0.785) (1.250) (1.251)

No toilets*Log R Prod 81 89 1 1.043 3.477 3.482
(1.477) (1.478) (1.831)* (1.833)*

No toilets*Log R Prod 82 89 -1.440 -1.459 -2.415 -2.416
(1.423) (1.424) (1.792) (1.793)

 

 

231

Table continues

Table 3.1 (cont’d)

 

 

 

Male Female

No t0ilets*Land 88 1.283 1.276 0.106 0.107
(1) (1.001) (1.322) (1.322)

Other toilets*Log R Prod s1 89 -0.152 —0.158 -1.712 -1.719
(1.288) (1.289) (1.522) (1.523)

Other toilets*Log R Prod s2 89 0.050 0.045 2.427 2.430
(1.229) (1.229) (1.465)* (1.466)*

Other toilets*Land 88 0.511 0.465 0.696 0.699
(0.848) (0.849) (1.311) (1.312)

Little excreta*Log R Prod $1 89 -0.091 -0.107 0.102 0.103
(0.283) (0.283) (0.341) (0.341)

Little excreta*Log R Prod 52 89 0.051 0.062 -0.213 -0.215
(0.305) (0.305) (0.361) (0.361)

Little excreta*Land 88 0.046 0.045 0.209 0.208
(0.158) (0.158) (0.187) (0.187)

Some excreta*Log R Prod 31 89 -0.214 -0.207 0.093 0.090
(0.315) (0.316) (0.419) (0.419)

Some excreta*Log R Prod s2 89 0.244 0.236 0.098 0.100
(0.320) (0.320) (0.430) (0.431)

Sorne excreta*Land 88 0.102 0.098 0.238 0.238
(0.129) (0.129) (0.154) (0.154)

No excreta*Log R Prod s1 89 0.381 0.366 2.746 2.758
(1.287) (1.288) (1.664)* (1.665)*

No excreta*Log R Prod s2 89 -0.716 -0.699 -2.958 -2.969
(1.313) (1.313) (1.592)* (1.593)*

No excreta*Land 88 0.175 0.180 -0.689 -0.690
(0.380) (0.381) (0.482) (0.482)

Number of obs 2477 2477 2701 2701
R-squared 0.2578 0.2587 0.2379 0.2379

 

 

 

The percent of households with the following water sources: (1)
underground, (2) open well, (3) spring, river lake, rain or snow, and
(4) water factory. Omitted is the first source. The inhouse with

flush toilet type is omitted. The no excreta category is omitted.

232

mented with information in 1991.

These are the first stage regressions in Table 1.14 where 1993 BMI was instru-

Table B.2: First stage regressions for Table 1.14

 

 

 

Var Male Female
Some primary 93 0.252 0.246 0.523 0.533
(0.322) (0.322) (0.253)" (0.253)“
Primary 93 —0.303 -0.318 0.449 0.460
(0.359) (0.360) (0.326) (0.327)
Middle school 93 -0.194 -0.215 0.007 0.026
(0.353) (0.354) (0.316) (0.317)
High school 93 0.118 0.113 -0.474 -0.465
(0.419) (0.420) (0.431) (0.431)
Tech/College+ 93 0.220 0.179 -0.254 -0.227
(0.530) (0.530) (0.665) (0.666)
Log real prod asset 93 1" -0.038 -0.041 0.090 0.096
(0.059) (0.059) (0.058) (0.058)
Log real prod asset 93 213 0.072 0.073 0.007 0.001
(0.056) (0.056) (0.054) (0.055)
Land farmed in 92 0.027 0.027 -0.015 -0.014
(0.021) (0.021) (0.023) (0.024)
Married 93 0.077 0.332
(0.428) (0.701)
Divorced Separated 93 -1.973 -1.110
(1.328) (1.530)
Widowed 93 -0.636 0.647
(0.667) (0.798)
Log real prod asset 91 1a 0.959 0.925 -1.620 -1.589
(1.045) (1.045) (0.990) (0.991)
Log real prod asset 91 2“ -1.107 -1.062 0.586 0.568
(0.946) (0.946) (0.921) (0.922)
Land farmed in 90 -1.383 -1.394 -2.134 -2.266
(2.304) (2.303) (2.560) (2.562)

 

 

233

Table continues

Table 8.2 (cont’d)

 

 

 

Male Female
Price Rice*Log R Prod sl 91 0.109 0.109 0.086 0.079
(0.249) (0.249) (0.224) (0.224)
Price Rice*Log R Prod 32 91 -0.305 -0.310 -0.017 -0.017
(0.301) (0.300) (0.286) (0.286)
Price Rice*Land 90 -0.426 -0.449 0.353 0.346
(0.313) (0.313) (0.335) (0.335)
Price Egg*Log R Prod 51 91 -0.038 -0.037 0.041 0.040
(0.037) (0.037) (0.035) (0.035)
Price Egg*Log R Prod 82 91 0.064 0.063 -0.020 -0.019
(0.037)* (0.037)* (0.039) (0.039)
Price Egg*Land 90 0.020 0.022 -0.029 —0.027
(0.030) (0.030) (0.032) (0.032)
Price Pork*Log R Prod 51 91 0.007 0.008 0.014 0.014
(0.013) (0.013) (0.012) (0.012)
Price Pork*Log R Prod 82 91 -0.010 -0.011 -0.010 -0.010
(0.012) (0.012) (0.012) (0.012)
Price Pork*Land 90 -0.005 -0.005 0.007 0.007
(0.011) (0.011) (0.010) (0.010)
Price Fish*Log R Prod 31 91 -0.007 -0.005 0.104 0.100
(0.060) (0.060) (0.056)* (0.056)*
Price Fish*Log R Prod s2 91 0.036 0.032 -0.031 -0.028
(0.056) (0.057) (0.054) (0.054)
Price Fish*Land 90 -0.054 -0.055 -0.006 -0.005
(0.038) (0.038) (0.039) (0.039)
Clinic Physician*Log R Prod s1 91 106.470 104.397 31.799 -32.189
(64.158)* (64.177) (56.539) (56.568)
Clinic Physician*Log R Prod 82 91 -64.291 -63.582 1.969 2.593
(59.262) (59.269) (53.180) (53.200)
Clinic Physician*Land 90 30.506 25.657 -4.420 -4.605
(45.891) (45.968) (51.507) (51.565)
Clinic Nurse*Log R Prod 51 91 -5205.468 —5241.722 -971.784 -1003.204
(1269822)“ (1269582)” (606.475) (608.499)*
Clinic Nurse*Log R Prod 32 91 4192.287 4219.063 668.101 691.569
(1064981)" (1064.810)** (533.770) (535.244)
Clinic Nurse*Land 90 0.344 0.875 —1071.045 —1066.944
(1234.162) (1233.638) (913.902) (914.324)
Clinic Bed*Log R Prod s1 91 -192.314 -178.749 40.569 34.565
(258.593) (258.717) (164.804) (164.896)
Clinic Bed*Log R Prod s2 91 174.245 164.440 -11.620 -6.370
(225.482) (225.569) (145.094) (145.175)
Clinic Bed*Land 90 71.949 71.989 62.266 60.736
(64.611) (64.636) (62.866) (62.899)

 

234

Table continues

Table 8.2 (cont’d)

 

 

Male Female
Water Source 2*Log R Prod 51 91 0.091 0.090 -0.429 —0.443
(0.355) (0.355) (0.325) (0.325)
Water Source 2*Log R Prod 32 91 0.028 0.025 0.062 0.077
(0.354) (0.354) (0.307) (0.307)
Water Source 2*Land 90 -0.084 -0.080 0.093 0.094
(0.162) (0.162) (0.177) (0.177)
Water Source 3*Log R Prod s1 91 0.239 0.258 0.296 0.310
(0.779) (0.779) (0.592) (0.592)
Water Source 3*Log R Prod s2 91 -0.192 -0.209 -0.455 -0.474
(0.665) (0.665) (0.535) (0.535)
Water Source 3*Land 90 -0.594 -0.582 0.262 0.276
(0.279)** (0.279)** (0.275) (0.277)
Water Source 4*Log R Prod 81 91 -0.451 -0.446 -0.148 -0.156
(0.318) (0.318) (0.300) (0.300)
Water Source 4*Log R Prod 82 91 0.585 0.588 -0.057 —0.052
(0.302)* (0.302)* (0.281) (0.281)
Water Source 4*Land 90 0.031 0.033 -0.072 -0.068
(0.159) (0.159) (0.170) (0.170)
In house no flush*Log R Prod 81 91 -1.462 -1.435 1.475 1.481
(1.017) (1.018) (0.969) (0.970)
In house no flush*Log R Prod 82 91 1.210 1.185 -0.705 —0.707
(1.281) (1.282) (0.931) (0.932)
In house no flush*Land 90 1.875 1.942 1.557 1.705
(2.421) (2.420) (2.670) (2.673)
Outside toilets*Log R Prod s1 91 -1.069 -1.058 0.812 0.827
(0.725) (0.724) (0.721) (0.721)
Outside toiletsI*Log R Prod 52 91 1.020 1.012 —0.236 -0.253
(0.676) (0.676) (0.682) (0.683)
Outside toilets*Land 90 2.109 2.150 2.176 2.309
(2.265) (2.264) (2.523) (2.526)
Open pit cement or earth*Log R Prod s1 91 -1.201 -1.192 0.721 0.727
(0.806) (0.806) (0.771) (0.772)
Open pit cement or earth*Log R Prod 82 91 1.263 1.250 -0.225 -0.233
(0.742)* (0.742)* (0.714) (0.714)
Open pit cement or earth*Land 90 2.122 2.172 1.802 1.932
(2.295) (2.295) (2.556) (2.558)
No toilets*Log R Prod $1 91 -0.474 -0.455 0.230 0.324
(1.605) (1.608) (1.281) (1.288)
No toilets*Log R Prod 82 91 -0.356 -0.379 0.182 0.124
(1.643) (1.648) (1.361) (1.371)
No toilets*Land 90 4.220 4.262 0.041 0.117
(2.657) (2.656) (2.930) (2.934)

 

 

235

Table continues

Table 8.2 (cont’d)

 

 

 

Male Female
Other toilets*Log R Prod 31 91 -2.283 -2.344 0.521 0.512
(1.425) (1.425)* (1.274) (1.275)
Other toilets*Log R Prod 82 91 2.056 2.105 0.454 0.458
(1.372) (1.374) (1.246) (1.247)
Other toilets*Land 90 2.439 2.472 2.494 2.636
(2.421) (2.420) (2.675) (2.677)
Little excreta*Log R Prod 81 91 0.829 0.826 0.179 0.178
(0.455)* (0.455)* (0.411) (0.411)
Little excreta*Log R Prod 82 91 -0.825 -0.808 -0.061 -0.063
(0.405)" (0.406)“ (0.366) (0.366)
Little excreta*Land 90 -0.063 -0.072 -0.193 -0.203
(0.210) (0.210) (0.206) (0.206)
Some excreta*Log R Prod s1 91 0.191 0.193 -0.449 -0.445
(0.362) (0.362) (0.364) (0.365)
Some excreta*Log R Prod 82 91 -0.233 -0.230 0.422 0.416
(0.368) (0.368) (0.358) (0.358)
Some excreta*Land 90 0.158 0.117 0.045 0.046
(0.243) (0.244) (0.235) (0.236)
No excreta*Log R Prod 31 91 -5.475 -5.386 2.813 2.791
(3.943) (3.944) (3.167) (3.168)
No excreta*Log R Prod 82 91 3.331 3.272 -2.423 -2.431
(3.297) (3.298) (2.811) (2.812)
No excreta*Land 90 0.324 0.352 0.747 0.902
(1.681) (1.682) (1.695) (1.701)
Number of obs 1815 1815 1948 1948
R-squared 0.2585 0.2606 0.2462 0.2472

 

 

The percent of households with the following water sources: ( 1)
underground, (2) open well, (3) spring, river lake, rain or snow, and
(4) water factory. Omitted is the first source. The inhouse with
flush toilet type is omitted. The no excreta category is omitted. For
first stage regressions for Table 1.15 and Table 1.16 please contact

me

236

Appendix C: Linear Probability

Models for Undernourishment and

Overweight

Table C.1: Probability Model for Overweight: Overall, Urban and Rural Areas

 

 

 

 

 

Male

Variables Overall Urban Rural
Some primary education -0.003 —0.006 -0.027 -0.032 -0.0004 -0.001
(0.014) (0.014) (0.040) (0.039) (0.014) (0.014)
Primary -0.008 -0.012 -0.035 -0.040 -0.009 -0.010
(0.015) (0.015) (0.040) (0.040) (0.015) (0.015)
Middle school 0.001 -0.003 -0.050 -0.055 0.006 0.004
(0.015) (0.015) (0.040) (0.040) (0.015) (0.015)
High school 0.020 0.016 -0.012 -0.017 0.022 0.020
(0.018) (0.018) (0.042) (0.042) (0.020) (0.020)
Tech/College+ -0.006 -0.009 -0.042 -0.047 -0.005 -0.007
(0.023) (0.023) (0.044) (0.044) (0.030) (0.030)
Log real prod assets 18‘ 0 -0.001 0.004 0.003 -0.002 -0.002
(0.002) (0.002) (0.005) (0.005) (0.002) (0.002)
Log real prod assets 2 a 0.005 0.005 0.003 0.003 0.005 0.005
(0.002)" (0.002)” (0.005) (0.005) (0.002)** (0.002)”
Land -0.001 -0.001 -0.006 -0.006 -0.001 -0.001
(0.001)" (0.001)** (0.007) (0.007) (0.001)* (0.001)*
Married 0.030 0.056 0.006
(0.011)** (0.024)” (0.012)
Divorced Separated -0.018 0.015 -0.061
(0.034) (0.073) (0.029)”
Widowed -0.017 -0.046 -0.021
(0.022) (0.054) (0.022)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 12760 12760 4087 4087 8673 8673
R-squared 0.179 0.180 0.192 0.194 0.151 0.151

P-value for testing coefficients equal to zero

Education 0.3828 0.3673 0.4839 0.4212 0.4059 0.4199
Assets 0.0196 0.0213 0.1717 0.1849 0.0416 0.0404
Age dummies 0.8750 0.8563 0.9648 0.9601 0.5659 0.5596
Cohort dummies 0.5106 0.5191 0.2658 0.2694 0.6820 0.6824
Marital Status 0.0019 0.0127 0.0567
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

237

Table continues

Table C. 1 (cont’d)

 

 

 

 

Female

Variables Overall Urban Rural
Some primary education 0.006 0.005 -0.016 -0.021 0.002 0.002
(0.013) (0.013) (0.030) (0.030) (0.014) (0.014)
Primary 0.022 0.021 -0.030 —0.037 0.025 0.024
(0.015) (0.015) (0.032) (0.032) (0.017) (0.017)
Middle school -0.001 -0.001 -0.007 -0.014 -0.008 -0.008
(0.016) (0.016) (0.033) (0.034) (0.017) (0.017)
High school -0.025 -0.024 -0.030 -0.037 -0.010 -0.009
(0.020) (0.020) (0.038) (0.038) (0.024) (0.024)
Tech/College+ -0.051 -0.051 -0.036 -0.043 -0.092 -0.089
(0.025)" (0.026)M (0.039) (0.040) (0.044)“ (0.044)”
Log real prod assets 1“ 0.002 0.002 0.005 0.005 0.001 0.001
(0.002) (0.002) (0.005) (0.005) (0.003) (0.003)
Log real prod assets 2 a -0.001 -0.001 -0.003 -0.003 0.001 0.001
(0.002) (0.002) (0.005) (0.005) (0.003) (0.003)
Land 0.001 0.001 -0.008 -0.008 0 0
(0.001) (0.001) (0.008) (0.008) (0.001) (0.001)
Married 0.018 -0.012 0.018
(0.014) (0.027) (0.016)
Divorced Separated 0.026 0.001 0.004
(0.051) (0.069) (0.070)
Widowed -0.012 -0.079 0.012
(0.023) (0.041)* (0.027)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 13652 13652 4493 4493 9159 9159
R—squared 0.167 0.167 0.191 0.192 0.152 0.152

P-value for testing coefficients equal to zero

Education 0.0190 0.0284 0.7922 0.7249 0.0423 0.0530
Assets 0.4221 0.4321 0.6578 0.6616 0.5516 0.5652
Age dummies 0.0817 0.0860 0.1730 0.1943 0.5504 0.5268
Cohort dummies 0.0148 0.0154 0.0744 0.0714 0.1238 0.1306
Marital Status 0.2056 0.1902 0.7229
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

 

Note: Also included in the regressions are age dummies and five-year cohort dum-
mies. Person level cluster robust standard errors are in parentheses. * indicates
statistical significance at 0.1 level and ** at 0.05 level.

238

Table C.2: Probability Model of Overweight: By Age Groups

 

 

 

 

 

Male

Variables Age 20—39 Age 40-59 Age 60+
Some primary education 0.001 -0.004 -0.024 -0.027 -0.022 -0.022
(0.026) (0.026) (0.023) (0.023) (0.034) (0.034)
Primary -0.012 -0.018 -0.024 -0.027 -0.015 -0.015
(0.026) (0.026) (0.024) (0.024) (0.040) (0.040)
Middle school -0.007 —0.013 -0.015 -0.019 -0.084 -0.084
(0.025) (0.025) (0.026) (0.026) (0.047)* (0.047)*
High school 0.014 0.008 0.006 0.003 -0.014 -0.011
(0.028) (0.028) (0.035) (0.035) (0.071) (0.071)
Tech/College+ -0.011 -0.016 0.008 0.002 -0.133 -0.134
(0.032) (0.032) (0.042) (0.042) (0.066)” (0.066)”
Log real prod assets 1‘3L -0.001 -0.001 -0.003 -0.003 0.008 0.008
(0.002) (0.002) (0.004) (0.004) (0.009) (0.009)
Log real prod assets 2 a 0.006 0.006 0.010 0.010 -0.014 -0.013
(0.003)“ (0.003)** (0.004)** (0.004)** (0.009) (0.009)
Land -0.001 -0.001 -0.003 -0.003 -0.002 -0.002
(0.001) (0.001) (0.002)** (0.002)M (0.002) (0.002)
Married 0.029 0.042 -0.041
(0.012)** (0.041) (0.071)
Divorced Separated -0.002 0.028 -0.142
(0.044) (0.093) (0.103)
Widowed -0.031 -0.017 -0.054
(0.026) (0.048) (0.077)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 6678 6678 4292 4292 1790 1790
R-squared 0.197 0.198 0.291 0.292 0.359 0.359

P-value for testing coefficients equal to zero

Education 0.5130 0.4869 0.7246 0.6880 0.3056 0.2993
Assets 0.0221 0.0271 0.0115 0.0118 0.2615 0.2931
Age dummies 0.5810 0.3719 0.7032 0.6745 0.6329 0.6341
Cohort dummies 0.2476 0.2674 0.2590 0.2754 0.5621 0.5740
Marital Status 0.0050 0.1691 0.5386
Community dummies 0.9595 0.9697 0.0000 0.0000 0.0000 0.0000

 

239

Table continues

Table C.2 (cont’d)

 

 

 

 

Female

Variables Age 20-39 Age 40-59 Age 60+
Some primary education —0.016 -0.016 0.010 0.009 -0.013 -0.014
(0.019) (0.019) (0.020) (0.020) (0.041) (0.041)
Primary -0.016 -0.016 0.029 0.028 -0.099 -0.104
(0.021) (0.021) (0.025) (0.025) (0.053)* (0.054)*
Middle school —0.029 -0.029 0.006 0.005 -0.110 -0.116
(0.020) (0.020) (0.030) (0.030) (0.083) (0.083)
High school -0.064 -0.063 0.099 0.099 0.177 0.183
(0.024)“ (0.024)** (0.049)” (0.049)" (0.179) (0.178)
Tech/College+ -0.074 -0.072 0.007 0.006 0.047 0.039
(0.030)** (0.030)“ (0.055) (0.055) (0.138) (0.138)
Log real prod assets 1a 0 0 0 0 0.015 0.015
(0.003) (0.003) (0.005) (0.005) (0.008)* (0.008)*
Log real prod assets 2 a -0.002 -0.002 0.008 0.008 -0.015 -0.014
(0.003) (0.003) (0.005)* (0.005)* (0.008)* (0.008)*
Land 0.003 0.003 -0.003 -0.003 0.002 0.002
(0.001)“ (0.001)** (0.002) (0.002) (0.003) (0.003)
Married 0.017 0.022 -0.037
(0.014) (0.058) (0.088)
Divorced Separated -0.037 -0.028 0.015
(0.076) (0.100) (0.143)
Widowed 0.156 -0.014 -0.064
(0.099) (0.065) (0.089)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 7177 7177 4578 4578 1897 1897
R-squared 0.180 0.181 0.284 0.284 0.356 0.356

P-value for testing coefficients equal to zero

Education 0.0272 0.0413 0.3465 0.3435 0.2562 0.2189
Assets 0.6709 0.6564 0.0499 0.0573 0.1341 0.1538
Age dummies 0.0512 0.0442 0.5502 0.5524 0.6055 0.6125
Cohort dummies 0.0092 0.0088 0.5018 0.4928 0.1057 0.0972
Marital Status 0.2459 0.5738 0.6732
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

 

Note: Also included in the regressions are age dummies and five-year cohort dum-
mies. Person level cluster robust standard errors are in parentheses. * indicates
statistical significance at 0.1 level and ** at 0.05 level.

240

Table C.3: Probability Model of Undernourishment: Overall, Urban and Rural Areas

 

 

 

 

Male
Variables Overall Urban Rural
Some primary education 0.034 0.036 0.061 0.063 0.041 0.042
(0.014)“ (0.014)" (0.020)** (0.020)** (0.017)” (0.017)"
Primary 0.024 0.027 0.054 0.057 0.034 0.036
(0.014)* (0.013)” (0.022)“ (0.021)** (0.017)" (0.017)"
Middle school 0.018 0.022 0.037 0.040 0.031 0.033
(0.014) (0.014) (0.019)* (0.019)" (0.017)* (0.017)*
High school 0.020 0.024 0.028 0.031 0.030 0.033
(0.015) (0.015) (0.021) (0.021) (0.020) (0.020)*
Tech/College+ 0.015 0.018 0.023 0.026 0.041 0.043
(0.018) (0.017) (0.022) (0.022) (0.024)* (0.024)*
Log real prod assets la 0.000 0.000 0.002 0.002 -0.002 -0.002
(0.002) (0.002) (0.003) (0.003) (0.002) (0.002)
Log real prod assets 2 a -0.001 -0.001 -0.002 -0.002 0.000 0.000
(0.002) (0.002) (0.003) (0.003) (0.002) (0.002)
Land 0.000 0.000 -0.002 -0.002 0.000 0.000
(0.001) (0.001) (0.003) (0.003) (0.001) (0.001)
Married -0.018 -0.016 -0.007
(0.013) (0.023) (0.015)
Divorced Separated 0.093 0.037 0.149
(0.051)* (0.081) (0.067)”
Widowed 0.028 0.066 0.013
(0.031) (0.054) (0.039)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 12760 12760 4087 4087 8673 8673
R-squared 0.12 0.13 0.14 0.15 0.138 0.140
P-value for testing coefficients equal to zero
Education 0.1990 0.1596 0.0480 0.0305 0.2672 0.2481
Assets 0.4417 0.4441 0.7186 0.6842 0.3596 0.3623
Age dummies 0.5040 0.6021 0.0042 0.0016 0.2572 0.2900
Cohort dummies 0.1535 0.1795 0.2674 0.2954 0.6166 0.6243
Marital Status 0.0302 0.2733 0.1184
Community dummies 0.0004 0.0003 0.0494 0.0561 0.0011 0.0010

 

241

Table continues

Table C.3 (cont’d)

 

 

 

 

Female

Variables Overall Urban Rural
Some primary education -0.018 —0.017 -0.017 -0.015 -0.010 -0.010
(0.010)* (0.010)* (0.019) (0.019) (0.011) (0.011)
Primary -0.038 -0.037 -0.038 -0.035 -0.030 -0.030
(0.010)“ (0.010)M (0.019)* (0.019)* (0.012)** (0.012)**
Middle school -0.018 -0.018 -0.045 -0.043 -0.006 -0.006
(0.011)* (0.011)* (0.019)" (0.019)** (0.013) (0.013)
High school -0.002 -0.002 -0.038 -0.036 0.007 0.007
(0.013) (0.013) (0.021)* (0.021)* (0.018) (0.018)
Tech / College+ -0.002 -0.002 -0.033 -0.031 -0.005 -0.005
(0.017) (0.017) (0.022) (0.022) (0.030) (0.030)
Log real prod assets 1“ 0.000 0.000 -0.003 -0.003 0.001 0.001
(0.002) (0.002) (0.003) (0.003) (0.002) (0.002)
Log real prod assets 2 a 0.000 0.000 0.003 0.003 0.000 0.000
(0.002) (0.002) (0.003) (0.003) (0.002) (0.002)
Land 0.000 0.000 -0.003 -0.004 0.000 0.000
(0.001) (0.001) (0.004) (0.004) (0.001) (0.001)
Married -0.012 -0.012 -0.002
(0.016) (0.029) (0.019)
Divorced Separated 0.021 0.021 0.044
(0.036) (0.046) (0.068)
Widowed 0.015 0.017 0.012
(0.023) (0.037) (0.030)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 13652 13652 4493 4493 9159 9159
R-squared 0.12 0.125 0.113 0.11 0.144 0.144

P-value for testing coefficients equal to zero

Education 0.0014 0.0021 0.2239 0.2743 0.0566 0.0594
Assets 0.9680 0.9631 0.5520 0.5579 0.8849 0.8798
Age dummies 0.1619 0.1850 0.1648 0.1649 0.3480 0.3510
Cohort dummies 0.5156 0.5233 0.4815 0.4799 0.3047 0.3106
Marital Status 0.2978 0.5250 0.8205
Community dummies 0.0000 0.0000 0.0176 0.0184 0.0000 0.0000

 

 

 

Note: Also included in the regressions are age dummies and five-year cohort dum-
mies. Person level cluster robust standard errors are in parentheses. * indicates

statistical significance at 0.1 level and ** at 0.05 level.

242

Table C.4: Probability Model of Undernourishment: By Age Groups

 

 

 

 

Male
Variables Age 20—39 Age 40—59 Age 60+
Some primary education -0.010 -0.008 0.037 0.045 0.066 0.065
(0.030) (0.030) (0.020)* (0.020)“ (0.028)" (0.028)"
Primary -0.021 -0.020 0.050 0.057 0.021 0.020
(0.031) (0.031) (0.019)” (0.019)" (0.032) (0.032)
Middle school -0.020 -0.018 0.024 0.034 0.036 0.037
(0.030) (0.030) (0.019) (0.019)* (0.035) (0.035)
High school -0.030 -0.028 0.034 0.042 0.001 -0.004
(0.031) (0.031) (0.025) (0.025)* (0.053) (0.053)
Tech/College+ —0.044 -0.042 0.023 0.036 0.026 0.026
(0.035) (0.035) (0.028) (0.029) (0.045) (0.045)
Log real prod assets 1“ 0.000 0.000 -0.002 -0.002 0.005 0.005
(0.002) (0.002) (0.002) (0.002) (0.007) (0.007)
Log real prod assets 2 a -0.002 -0.002 -0.002 -0.002 -0.002 -0.003
(0.002) (0.002) (0.002) (0.002) (0.008) (0.008)
Land 0.001 0.001 -0.001 -0.001 0.008 0.008
(0.001) (0.001) (0.001) (0.001) (0.004)M (0.004)"
Married -0.002 -0.070 0.043
(0.014) (0.038)* (0.066)
Divorced Separated 0.044 -0.020 0.293
(0.066) (0.087) (0.162)*
Widowed -0.010 0.097 0.028
(0.026) (0.066) (0.073)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 6678 6678 4292 4292 1790 1790
R-squared 0.162 0.162 0.224 0.232 0.413 0.417
P-value for testing coefficients equal to zero
Education 0.5802 0.5848 0.1526 0.0836 0.3047 0.2960
Assets 0.5008 0.5076 0.1157 0.1277 0.7938 0.7907
Age dummies 0.7063 0.7658 0.6773 0.6583 0.6089 0.6328
Cohort dummies 0.6464 0.6479 0.0986 0.0885 0.8436 0.8418
Marital Status 0.8886 0.0140 0.3030
Community dummies 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

 

243

Table continues

Table C.4 (cont’d)

 

 

 

 

Female

Variables Age 20-39 Age 40—59 Age 60+
Some primary education -0.026 -0.026 0.002 0.003 0.020 0.021
(0.016)* (0.016)* (0.013) (0.013) (0.031) (0.031)
Primary -0.030 -0.030 -0.032 -0.031 -0.021 -0.021
(0.016)* (0.016)* (0.014)“ (0.014)“ (0.034) (0.034)
Middle school 0026 -0.026 0.003 0.004 -0.018 -0.019
(0.016)* (0.016)* (0.017) (0.017) (0.037) (0.038)
High school -0.010 -0.011 -0.044 -0.044 -0.110 -0.113
(0.019) (0.019) (0.022)** (0.022)** (0.062)* (0.062)*
Tech/College+ 0.003 0.001 -0.064 -0.064 -0.065 -0.063
(0.024) (0.024) (0.023)" (0.023)“ (0.049) (0.050)
Log real prod assets 1at -0.001 -0.001 -0.002 -0.002 0.001 0.001
(0.002) (0.002) (0.003) (0.003) (0.006) (0.006)
Log real prod assets 2 a 0.002 0.002 0.000 0.000 -0.003 -0.003
(0.002) (0.002) (0.003) (0.003) (0.005) (0.005)
Land -0.001 -0.001 0.000 0.000 -0.001 -0.001
(0.001) (0.001) (0.001) (0.001) (0.004) (0.004)
Married -0.016 0.027 0.012
(0.018) (0.030) (0.054)
Divorced Separated 0.007 0.014 0.072
(0.057) (0.045) (0.111)
Widowed -0.003 0.056 0.021
(0.063) (0.040) (0.056)
Community *year dummy Yes Yes Yes Yes Yes Yes
No. of Observations 7177 7177 4578 4578 1897 1897
R—squared 0.146 0.147 0.222 0.222 0.388 0.389

P-value for testing coefficients equal to zero

Education 0.2009 0.2306 0.0037 0.0037 0.3050 0.2979
Assets 0.5701 0.5425 0.5372 0.5489 0.8340 0.8604
Age dummies 0.2308 0.1956 0.6336 0.6656 0.2185 0.1975
Cohort dummies 0.8244 0.8265 0.0493 0.0528 0.3416 0.3656
Marital Status 0.7889 0.5292 0.9071
Community dummies 0.9775 0.9729 0.0000 0.0000 0.0000 0.0000

 

Note: Also included in the regressions are age dummies and five-year cohort dum-
mies. Person level cluster robust standard errors are in parentheses. * indicates
statistical significance at 0.1 level and ** at 0.05 level.

244

 

Appendix D: AR1 Model of

Determinants of BMI

Table D1: AR1 Models for BMI

 

 

 

Male Female
Some primary schooling 0.066 0.043 0.058 0.054
(0.119) (0.119) (0.105) (0.105)
Primary -0.013 -0.044 0.270 0.264
(0.124) (0.124) (0.119)M (0.119)“
Middle school 0.003 -0.036 -0.018 -0.024
(0.127) (0.127) (0.129) (0.129)
High school 0.131 0.098 -0.345 -0.357
(0.148) (0.148) (0.167)** (0.167)"
Tech / College+ -0.035 -0.059 -0.493 -0.515
(0.170) (0.170) (0.211)“ (0.212)”
Log real prod asset 1 0.023 0.021 0.044 0.045
(0.017) (0.017) (0.019)** (0.019)"
Log real prod asset 2 -0.002 0 -0.021 -0.021
(0.017) (0.017) (0.019) (0.019)
Land farmed -0.002 -0.002 0.002 0.002
(0.007) (0.007) (0.009) (0.009)
Married 0.388 -0.196
(0.118)“ (0.169)
Divorced Separated -0.349 -0.222
(0.350) (0.425)
Widowed -0.005 -0.409
(0.239) (0.235)*
Number of obs 12760 12760 13652 13652
Community*year dummies Yes Yes Yes Yes
Wald chi2 2549.10 2571.00 2032.62 2036.13
Rho estimated autocorrelation coefficient 0.434 0.434 0.495 0.495
fraction of variance due to 11, 0.295 0.293 0.444 0.443
modified Bhargava et al. Durbin-Watson 1.438 1.438 1.361 1.361
Baltagi-Wu LBI 2.365 2.365 2.293 2.294

 

 

245

Appendix E: Descriptive Statistics
of Nutrient Intakes in CHNS89-93

Table E.1: Patterns and Trends of Individual Daily Nutrient Intakes in CHNS 1989-

1993

 

 

 

Male Female
1989 1991 1993 1989 —19_91'_ 1993
Overall

No. Obs 2074 3413 3208 2242 3752 3476

Calorie (kcal) Median 2724.65 2726.46 2603.09 2359.46 2330.36 2245.11
Mean 2816.94 2794.99 2688.56 2441.09 2403.25 2313.70

(std) (870.10) (816.16) (815.98) (712.18) (705.54) (672.66)

Protein (g) Median 95.20 97.18 95.11 82.65 84.85 84.72
Mean 101.18 102.06 100.44 88.94 89.48 88.23

(std) (39.56) (36.57) (36.73) (34.83) (32.32) (32.04)

Fat (g) Median 38.40 44.01 45.69 31.62 36.16 39.06
Mean 47.65 51.76 53.86 40.57 44.31 45.87

(std) (38.79) (39.65) (43.01) (34.04) (36.43) (36.48)

Carbohydrates (g) Median 462.01 443.60 410.88 408.96 389.65 368.87
Mean 482.57 462.36 434.09 427.91 408.72 385.06

(std) (159.05) (151.25) (151.14) (135.80) (134.18) (128.35)

% Calorie from fat Median 13.51 15.59 17.14 12.43 15.11 16.80
Mean 14.61 16.15 17.45 14.39 16.01 17.28

(std) (9.62) (10.09) (10.72) (9.80) (10.32) (10.76)

% Calorie from protein Median 13.87 14.13 14.34 14.02 14.34 14.60
Mean 14.37 14.70 15.04 14.57 14.98 14.60

(std) (3.33) (3.41) (3.59) (3.56) (3.47) 14.60

% Calorie from carbohydrates Median 70.43 67.27 65.79 71.95 68.91 67.26
Mean 69.42 66.97 65.38 70.70 68.52 67.06

(std) (12.54) (13.03) (13.61) (11.72) (12.17) (12.49)

Urban Areas

No. Obs 633 1194 1015 694 1332 1110

Calorie (kcal) Median 2520.74 2523.25 2443.17 2151.21 2134.54 2101.66
Mean 2598.40 2577.65 2500.23 2214.54 2214.52 2149.33

(std) (763.02) (728.01) (749.07) (609.99) (648.88) (626.39)

Protein (g) Median 98.45 98.43 96.30 84.01 85.85 85.47
Mean 101.84 103.34 101.92 89.26 90.22 89.52

(std) (36.75) (36.76) (37.63) (33.15) (32.42) (31.79)

Fat (g) Median 49.65 53.69 57.57 40.23 45.76 49.88
Mean 55.43 59.57 63.10 46.41 51.63 54.51

(std) (38.62) (37.77) (39.38) (31.72) (36.20) (35.81)

Carbohydrates (g) Median 398.33 379.71 352.77 345.49 326.90 315.30
Mean 410.51 392.87 368.06 357.74 343.44 323.73

(std) (112.62) (116.13) (115.63) (102.17) (104.42) (99.56)

% Calorie from fat Median 18.14 19.74 21.95 17.70 19.86 21.64
Mean 18.27 20.01 21.86 18.09 20.02 21.92

(std) (8.92) (9.55) (9.39) (9.36) (9.72) (9.84)

% Calorie from protein Median 15.25 15.62 15.89 15.81 15.85 16.24
Mean 15.63 16.06 16.33 16.02 16.33 16.70

(std) (3.16) (3.37) (3.63) (3.44) (3.45) (3.68)

% Calorie from carbohydrates Median 64.51 62.25 60.08 64.97 63.21 61.00
Mean 64.49 61.98 59.93 65.50 62.99 61.11

(std) (11.26) (11.48) (11.47) (10.85) (10.99) (10.96)

 

246

Table continues

 

 

 

 

Table continued Male Female
1989 1991 1993 1989 1991 1993
Age 20—39
No. Obs 1686 1682 1435 1838 1924 1579
Calorie (kcal) Median 2719.27 2830.43 2699.58 2356.48 2423.88 2336.42
Mean 2815.52 2913.59 2802.60 2439.61 2509.81 2405.95
(std) (871.61) (843.48) (804.05) (711.81) (711.15) (677.22)
Protein (g) Median 95.47 100.73 99.02 83.21 88.84 88.42
Mean 101.75 106.31 104.92 89.51 93.41 92.01
(std) (40.27) (37.79) (37.08) (35.47) (32.83) (32.18)
Fat (g) Median 38.46 45.54 47.45 32.74 37.66 40.82
Mean 47.57 53.17 55.92 41.36 46.52 47.50
No. Obs (38.86) (40.52) (42.85) (34.31) (38.34) (36.95)
Carbohydrates (g) Median 462.40 467.36 440.39 406.71 408.85 384.56
Mean 482.25 486.19 457.07 425.49 426.99 400.89
(std) (160.28) (154.46) (149.80) (135.35) (135.42) (133.01)
% Calorie from fat Median 13.56 15.55 16.98 13.14 15.18 16.70
Mean 14.61 15.88 17.43 14.69 16.10 17.34
No. Obs (9.66) (9.74) (10.83) (9.92) (10.25) (10.76)
% Calorie from protein Median 13.92 14.11 14.38 14.07 14.32 14.57
Mean 14.45 14.69 15.04 14.66 14.98 15.39
(std) (3.35) (3.38) (3.49) (3.61) (3.50) (3.71)
% Calorie from carbohydrates Median 70.38 67.69 66.22 71.33 68.98 67.34
Mean 69.38 67.57 65.97 70.36 68.54 66.99
(std) (12.62) (12.71) (13.26) (11.82) (11.94) (12.48)
Age 40—59

No. Obs 388 1185 1198 404 1262 1327
Calorie (kcal) Median 2773.12 2747.12 2635.58 2371.04 2347.64 2290.76
Mean 2823.10 2811.20 2727.54 2447.86 2431.95 2349.42
(std) (864.63) (764.96) (812.15) (714.74) (674.80) (651.01)
Protein (g) Median 93.44 97.00 95.97 80.75 84.62 85.33
Mean 98.74 100.86 101.26 86.34 89.05 88.67
(std) (36.27) (33.80) (36.17) (31.69) (30.98) (31.34)
Fat (g) Median 37.44 43.63 44.89 26.32 35.69 38.57
Mean 47.99 51.62 53.30 36.98 43.58 45.99
(std) (38.53) (39.45) (43.17) (32.60) (35.67) (37.53)
Carbohydrates (g) Median 457.26 444.53 417.61 423.05 395.59 382.56
Mean 483.99 464.77 441.14 438.93 417.21 393.03
(std) (153.74) (145.96) (152.21) (137.46) (131.00) (123.65)
% Calorie from fat Median 13.31 15.54 16.70 10.60 14.55 16.49
Mean 14.61 16.02 17.04 13.07 15.55 16.96
(std) (9.42) (10.23) (10.52) (9.16) (10.29) (10.87)
% Calorie from protein Median 13.62 13.95 14.23 13.92 14.15 14.54
Mean 14.03 14.46 14.98 14.17 14.72 15.15
(std) (3.25) (3.29) (3.64) (3.28) (3.35) (3.63)
% Calorie from carbohydrates Median 70.45 67.58 65.91 74.41 69.72 67.73
Mean 69.58 66.95 65.50 72.24 69.17 67.53
(std) (12.20) (13.14) (13.79) (11.10) (12.17) (12.51)

 

247

Table continues on next page

 

 

 

 

Table continued Male Female
1989 1991 1993 1989 1991 1993
No Formal Education
No. Obs 75 378 327 350 1196 1073
Calorie (kcal) Median 3129.63 2569.97 2468.10 2635.49 2280.42 2187.89
Mean 3133.97 2618.60 2602.30 2687.51 2344.65 2275.60
(std) (983.27) (814.04) (952.81) (768.87) (720.97) (705.67)
Protein (g) Median 102.55 88.04 83.30 82.65 79.36 77.27
Mean 108.18 91.27 89.60 89.25 83.38 81.92
(std) (45.26) (32.37) (35.96) (35.19) (29.38) (30.73)
Fat (g) Median 26.17 29.44 30.64 19.93 26.63 27.57
Mean 41.56 40.86 42.34 33.95 37.54 37.27
(std) (40.34) (35.33) (46.19) (35.84) (34.05) (33.77)
Carbohydrates (g) Median 563.48 445.64 410.53 489.65 400.62 381.45
Mean 576.47 460.10 449.20 503.91 415.31 400.99
(std) (181.29) (157.02) (191.60) (142.95) (142.14) (139.24)
% Calorie from fat Median 8.56 11.35 12.07 6.89 11.38 11.46
Mean 11.26 13.66 14.18 10.47 13.94 14.32
(std) (8.96) (10.18) (10.85) (8.97) (10.40) (10.34)
% Calorie from protein Median 13.34 13.51 13.35 12.88 13.72 13.70
Mean 13.65 14.07 13.90 13.20 14.37 14.44
(std) (3.16) (3.11) (3.11) (2.97) (3.25) (3.28)
% Calorie from carbohydrates Median 76.24 72.31 71.89 79.36 72.90 72.76
Mean 74.54 70.87 69.70 75.98 71.13 70.83
(std) (11.26) (12.48) (13.91) (10.71) (12.26) (12.05)
Some Primary Schooling

No. Obs 272 595 758 390 575 779
Calorie (kcal) Median 2938.05 2662.15 2587.53 2390.16 2370.51 2314.79
Mean 3001.04 2786.70 2657.82 2475.20 2467.40 2378.96
(std) (907.35) (854.34) (813.28) (678.10) (718.48) (686.44)
Protein (g) Median 95.12 92.99 90.93 76.85 84.87 85.06
Mean 101.46 97.74 96.40 84.13 88.59 88.36
(std) (38.33) (35.21) (34.91) (30.19) (31.60) (31.96)
Fat (g) Median 32.30 39.72 42.66 22.29 34.03 38.48
Mean 43.22 47.62 51.52 32.12 41.43 45.14
No. Obs (35.33) (39.38) (44.38) (28.09) (34.15) (35.62)
Carbohydrates (g) Median 508.08 450.64 409.88 436.28 408.89 393.44
Mean 538.34 471.13 431.77 460.59 431.49 403.12
(std) (178.73) (154.53) (147.59) (142.66) (142.02) (134.13)
% Calorie from fat Median 9.91 13.88 16.67 8.34 13.86 16.26
Mean 12.74 14.85 16.84 11.48 14.67 16.63
No. Obs (9.27) (9.89) (10.71) (8.65) (9.72) (10.72)
% Calorie from protein Median 13.10 13.49 13.82 13.03 13.77 14.27
Mean 13.56 14.16 14.60 13.70 14.45 14.94
(std) (2.89) (3.33) (3.45) (3.40) (3.30) (3.58)
% Calorie from carbohydrates Median 74.53 69.20 66.68 76.86 70.67 68.52
Mean 72.16 68.53 65.95 74.54 70.36 68.13
(std) (11.68) (12.71) (13.80) (10.47) (11.57) (12.32)

 

248

Table continues

 

 

 

 

Table continued Male Female
1989 1991 1993 1989 1991 1993
Middle School
No. Obs 766 1024 996 659 832 764
Calorie (kcal) Median 2683.71 2772.08 2659.74 2303.54 2316.80 2272.34
Mean 2808.16 2854.06 2735.91 2360. 76 2406.16 2330.88
(std) (884.38) (818.02) (759.66) (667.80) (685.69) (659.49)
Protein (g) Median 95.99 98.88 98.27 84.73 87.38 89.08
Mean 101.46 104.48 104.52 89.08 92.78 91.92
(std) (38.22) (36.54) (37.35) (32.42) (33.18) (32.12)
Fat (g) Median 35.78 44.70 50.88 35.36 40.38 45.32
Mean 47.12 52.29 57.46 43.72 47.66 50.51
No. Obs (40.10) (39.55) (43.16) (32.72) (36.40) (35.18)
Carbohydrates (g) Median 462.73 453.89 419.27 381.28 382.05 363.01
Mean 483.29 471.68 437.13 401.04 399.26 375.39
(std) (164.50) (154.54) (141.31) (124.54) (127.60) (121.65)
% Calorie from fat Median 13.04 15.29 18.04 15.28 16.52 18.39
Mean 14.41 16.03 18.34 16.10 17.19 18.95
No. Obs (9.68) (9.81) (10.79) (9.76) (10.13) (10.27)
% Calorie from protein Median 14.00 14.12 14.58 14.54 14.93 15.20
Mean 14.47 14.74 15.33 15.07 15.47 15.84
(std) (3.28) (3.31) (3.62) (3.47) (3.57) (3.63)
% Calorie from carbohydrates Median 71.36 67.82 65.20 69.12 67.13 65.18
Mean 69.70 66.87 64.57 68.55 66.96 64.92
(std) (12.64) (13.17) (13.58) (11.40) (11.81) (11.84)
High School
No. Obs 324 422 368 263 336 280
Calorie (kcal) Median 2639.68 2749.21 2644.53 2178.72 2209.19 2241.90
Mean 2721.10 2801.24 2720.45 2290.17 2307.26 2322.62
(std) (837.45) (791.93) (757.54) (697.17) (625.33) (689.44)
Protein (g) Median 97.04 100.74 101.98 83.48 87.08 93.63
Mean 102.69 107.12 107.11 91.13 92.11 97.07
(std) (40.94) (39.69) (34.86) (39.97) (32.31) (32.93)
Fat (g) Median 46.22 53.43 52.89 38.21 46.26 50.35
Mean 52.28 59.36 59.85 46.98 53.88 59.72
(std) (39.29) (42.48) (37.97) (37.51) (35.28) (48.25)
Carbohydrates (g) Median 431.03 426.71 407.58 374.73 348.21 336.61
Mean 444.89 442.88 421.43 373.93 360.67 347.69
(std) (137.50) (130.79) (135.51) (104.61) (101.91) (105.51)
% Calorie from fat Median 16.54 18.38 19.49 17.19 19.84 21.72
Mean 16.48 18.26 19.37 17.30 20.18 21.89
(std) (9.49) (9.97) (9.77) (9.97) (9.44) (11.01)
% Calorie from protein Median 14.67 14.88 15.42 15.45 15.39 16.31
Mean 15.06 15.32 15.91 15.68 16.03 16.85
(std) (3.17) (3.49) (3.58) (3.52) (3.59) (3.83)
% Calorie from carbohydrates Median 67.05 64.04 62.91 66.38 63.37 60.66
Mean 66.63 64.38 62.62 66.74 63.28 61.00
(std) (12.12) (12.76) (12.82) (11.73) (10.77) (12.33)

 

 

Note: The daily intake of energy, fat, protein and carbohydrates contents are based on the 1991 China Food
Composition Tables (FCT). The energy content of fat, protein and carbohydrates is calculated based on the
formulae that one gram of fat yields 9 kcal of energy; one gram of protein and carbohydrate each yields 4

kcal of energy.

249

Appendix F: First Stage

Regressions for section 2.6

A simple health production function analysis was carried out in chapter two
in which the lagged health (lagged log weight and height, or lagged log BMI in
1991), current physical activity levels and nutrient intakes are included and treated
as endogenous. Also included in the regression are individual year age dummies.
The identifying instruments are education dummies, real productive asset splines,
land cultivated and community characteristics including prices, water and sanitation
conditions in 1991.

The first stage regressions for these endogenous variables on all exogenous factors

are listed here.

250

8:58:00 @388

 

 

...memdv **AmHm.oV 3.3:.de 1.2533 ...ngdv 8.1.36.8 1.3288 *Lmoodv 1.3ng

88- :8: 888 ...88- :8:- 88 8.88- 888- 88- : 8558 88.5
1:88 1:28 1:88 1:88 :88 :88 1:88.88 1:88 1:8.8

...88- :88- 8.88- ...88- 88 88- 88 88- 88 80 8 82.:
1:88 1:88 :38 1:88 :88 :88 1:88 :88 1:88

..88 :88 888 888 888- 88 88- 888 88- .88 so 8:5
:88 1:88 1:88 1:88 1:88 :88 :88 :88 :88

:88- 88- :88- 888- 8:8- 88- 88- 888 88- 85.: a5 85
1:88 1:88 1:88 1:58 :88 :88 :88 :88 :88

8.: 8: 8.2 88: 88 88- 88 88 88 8.8 8 8F:
1:88 1:88 1:88 :88 1:88 1:88.88 :88 :88 :88

:8- 88- 88.:- 8:8- 8:8 288 8 888 888 8:3
:88 :28 :88 :88 :88 :88 :88 1:88 1:8.8

888. 8:8- 88- 83.8- 888- 88 88 88 88 8 888 8.8 88. 8.:
1:88 1:28 1:38 1:88 :88 :88 :88 :88 :88

...88 :88 8.2 88 ...88 88- 8 888 888 H 888 8.8 _8. 88
1:88 1.8.8 :88 :8.8 :88 1:88 :88 1:88 1:88

88 8:: 88 88- 8.:- 8...8- 88 888 88 88:50:88.
1:38 1:88 1:88 :38 :88 1:88 :88 1:88 1:88

88.: :88 88 8:8 88- 8:8- 88 ...88 88 35.8 88
1:88 1:88 1:88 :88 :88 1:88 :88 1:88.88 1:88

888 888 83. 88.: 888- 888- 88 88 88 _88... 5882
:88 :88 :88 :88 :88 :88 :88 1:88 :88

8:8 888 8:: 88- 888- 88- 88- 88 :88 888:8
:88 1:88 :88 :88 :88 :88 :88 1:88 :88

8:8 8.: 8...: 88- 88- 88- 888- 88 888 88888 8388 8.58
588.: so: 8.: so: :8 :8 :8: 88 88 88 88 88

8:250 .8 88:0 .5 8.8 588.: 888.0 88288. Em 8: .888 83 £883 83
8.2

 

 

88.3898. eczema compassem 588m 88 commmouwom owgm swarm ”Hum @389

251

8:58:00 @388

 

 

888 888 8888 8888.8 8888 888 888 888 888 888.8
83: 8:: 8.8 8.: 8.: 8.8 8.: 8.8 8.8 8588888.:
888 888 8888 8888 8888 888 888 888 888 888.:
88 8.: 8.: 8.: 8.: 8.: 8.: 8.: 8.: :8:
8:8 :88 £88 :88 888 83 888 E3 83 8888-:
88 88 88 88 88 88 88 88 88 88 8 8:882
1:88 1:88 1:88 1:88 1:88 1:88 1:88 1:8.8 1:8.8

:88 8:8 88:: :88: 8:8 58 88.: 3.3 :8:. 3888
15:8 1:88 1:888 1:838 :88 :88 1:88 1:88 1:8.8

83- 8:8- 8:8- :88- :88- 8:8 :88- :88- 8:8- 8:88 8:2
1:88 :88 :88 38:88 1:88 1:88 1:88 1:88 1:88

8:8- 8:8 8:8 :8:- 88: 83 :88- 888- :88- 8:88 88:
1:88 1:88 1:88 1:38 1:88 1:88 1:88 1:88 1:88

8:8- :8:- :88- 8:8- 8:8- 8:3 888- 888- :88- 8888 288 :8>
:88 :88 :888 1:88: :88 :88 1:88 1:88 1:88

:8:- 83- :88- :88- :88- 83- 83- :88- 888- :88: oz
1:88 1:88 :88 :88 :88 1:88 1:88 1:88 1:88

83- 8:8- :8:- 8::- 888 888 888- 888- 83- 88 88o
:88 :88 :88 :88 :88 :88 :88 :88 1:88

83- 83 :88 8:3 :88 888 888- :88- 888- 8:88 8880
:38 :88 :88 1:88 :88 1:88 1:88 :88 1:88

8:8 83- 8:8- 3:88 :3 8:8 :88- :88- :88- :8: 8 88: 8
1:88 1:88 1:88 :88 1:88 :88 1:88 1:8.8 1:8.8

:88 :8: :R: 88.:- 8:.:- :88- :88- 888- 888- v 8:5: 883
1:88 1:88 1:88 5:8 :88 :88 1:88 1:88 1:88

8:8- :8: :88: :88- E8- 888 :88- 888- 888- : 8:5: 883
582: as: 8: as: :8 :8 :8: 8: :8: 88 88 :8

8:25 :8 882:0 8 8:: 588: 83:0 8838:. Em :3 8:88 :3 8:83 :3
2:2

 

8.888 8.: 28:8

252

825:8 2an

 

 

38.8 38.8 38.8 138.8 1.388 138.8 138.8 1.38.8 1.38.8
83- S3 382 83- 32- 83 83- 23- 83- N 85cm 855
1.388 38.8 1.1.33.8 1.388 38.8 33.8 138.8 1.388 38.8
G3- 83- 33- 83- 83- 383 883 283- 33 __o 8 83
1388 £888 388 1.388 38.8 138.8 1.38.8 1.388 138.8
83 :3 3.3 33 83- 83 883- 283- 383- 18m 8 125
38.8 388 888 3:8 388 38.8 38.8 38.8 38.8
83- 83- 83- 83- 33- 83- 283- 883 83- fiom 8 85
1.38.8 1388 138.8 1.1.38.8 38.8 38.8 38.8 1.38.8 1.388
83 83 83 83 83 33- 33 83 83 8.8 8 85
138.8 138.8 1.388 38.8 138.8 138.8 38.8 38.8 :88
83- 83- 83- 33 E3 83 o 83 83 23
138.8 1.32.8 1.388 138.8 38.8 38.8 38.8 38.8 38.8
83- 23- 83- 83- 33- 283- 283- 83 83 m 28% 8a 32 33
138.8 1.328 £38 1388 38.8 38.8 138.8 8 1.388
83 83 83 83 83 283- 383 283 883 2 283 Ea 32 82
138.8 1.33.8 1.38.8 38.8 33.8 138.8 38.8 138.8 38.8
3; 83 3.2 83 8:2- 83- 83- 33 ~83 $880389
138.8 1.388 1.33.8 1.38.8 38.8 1.388 38.8 1.38.8 1.388
83 83 832 33 R3- 83- 383 33 83 _83 .85
1.33.8 1388 138.8 38.8 1.33.8 1388 388 1.38.8 1.388
3.3 83 83.3 882 83- 83- 883 383 83 3223. 2222
133.8 138.8 138.8 :88 138.8 138.8 38.8 ...388 1.388
83 a: 83 83- 32- 833- 33 83 33 .535
1.388 123.8 138.8 33.8 38.8 138.8 ...388 38.8 1.388
:3 83 :3 282 E3- 83- 33 S3 :3 858:8 383 28m
23on 80¢ 3 3E E 3 :3 88 83 82 82 82

8:28 8 8:28 8 8m 28on 8:28 32822 22m 82 £82 33 £33 82

$380,”.—

 

3888 E 232.

253

 

 

 

88.8 88.8 88.8 88.8 88.8 88.8 88.8 88.8 88.8 21.5.8
8.8. 8.8” «3: 8.8 8.3 8.8- 3; 88.2 8.2 8=E§§E
88.8 88.8 88.8 88.8 88.8 88.8 88.8 88.8 88.8 2.5.8
8.: 88.8 v? 8....” 8.8 8.8 83- 8: 2.8 88m,
88.8 883 88.8 38.8 82.8 23. 88.8 $3 «2.8 Essa-m
888 888 888 .888 888 888 888 888 8% Bo mo H22:2
*ngdv 333.3 *Lowwsv *Lmvwbv :Cbmév *Lowodv *meodv *Lwoodv *Lomodv

$8.: «8.2 88.8.. 888.88 88.8 3.8 88.8 88.... 38.8 28880
1.88.: $82.8 $88.8: :8:-.88 238.8 82.8 $888 1888 1888.8

5.8- 82.2- 88.5- 88.88- 838- 3:8 8:8- 83- 88.8- 3288 632
188.8 88.8 88.8 88.8 £888 $88.8 188.8 888 1.88.8

8.82- 88.8- 23 83. 88.8 83 $8.8- 88.8- 88.8. 3228 2:8
£888 188.8 1888.8 1.88.8 88.8 188.8 188.8 88.8 388.8

88.7 88.8. 88.2- 88.:- 888. 82.8 88.8- 38.8- 88.8- 8238 2:: bo>
189.8 888.8 8828 $82.88 82.8 82.8 138.8 1.88.8 1.838

88.....- 83 $8.: 5.8. 32. .888 «:8. 88.8- 88.8- 38:8 02
188.8 2888.8 1838 88.8 88.8 $888 ...888 188.8 188.8

88.8- 88.8. 888- 882.- :88- 32 88.8. 888- 88.8- E 580
88.8 8888.8 88.8 83.8 888.8 188.8 88.8 ...888 88.8

8.83- 33- 88.8. 33. 33. 82.8 88.8 88.8- 88.8- 3088 @835
888 1.83.8 82.8 88.8 88.8 188.8 88.8 88.8 88.8

88.8 88.8- 23- 883- 8%.? 38.8 38.8- 888- 88.8- as: o: 3:2. a
......88.8 3.888 188.8 88.8 188.8 3.888 88.8 188.8 188.8

888.8 88.8 8?, 38.8- 23- 88.8- 288- 33.- 888- v 880m 855
1.88.8 1.88.8 A388 1.828 __.-83.8 888 1.888 188.8 1.888

.58- 8.3 883 :88- 83- 8.8.8 :88- 888- 88.8- m 85cm as?
58on Sam 88 aem 88 88 888 88 82 82 82 83

8:080 8 8:28 8 am 58on 8:25 83352 cam 83 £82 83 £82» 83
oEEmm

 

8.888 E 288.

254

Appendix G: The Design Matrix

and Constraints Needed

1 . Consecutive Years

Example ((31.1). Age: 20, 21, 22, 23; Year: 1989, 1990, 1991; Cohort: 1966, 67,
68, 69, 70, 71

 

 

P CO cl 02 03 c4 c5 06 a1 a2 a3 a4 t1 t2 t3 i
1 0 0 O 1 O O 1 0 O O 1 O O
1 0 0 1 O 0 O 0 1 0 O 1 O 0
1 O 1 O O O O 0 O 1 O 1 0 0
1 1 O 0 0 0 0 O 0 0 1 1 0 0
1 0 0 0 O 1 0 1 0 0 O 0 1 0
X = 1 0 O 0 1 0 0 O 1 O 0 0 1 0
1 O 0 1 O 0 0 O 0 1 0 O 1 0
1 0 1 0 O 0 0 O 0 0 1 0 1 O
1 0 0 O 0 0 1 1 0 0 0 0 0 1
1 0 0 O O 1 O 0 1 0 0 O 0 1
1 0 0 O 1 0 O 0 0 1 O 0 O 1
_ 1 0 0 1 0 O O 0 O 0 1 O O 1 J

Rank(X)=10, k=1+5+3+2=11
Linear dependency:
t3 = 0.502 + c3 +1.5c4 + 205 + 2.566 + 0.5a2 + a3 +1.5a4 — O.5t2 — 1.5

Hence the number of constraints necessary is one after setting

{01:0afi120171:0}°

255

Example (GL2). Age: 20, 21, 22, 23, 24; Year: 1989, 1990, 1991; Cohort: 1965,
66, 67, 68, 69, 70, 71

CO c1 c2 c3 c4 c5 c6 07 a1 a2 a3 a4 a5 t1 t2 t3 -
1 O 0 O 0 1 0 0 1 0 O 0 0 1 0 0
1 0 O 0 1 0 0 0 0 1 0 0 0 1 O O
1 0 0 1 0 O 0 0 O O 1 0 0 1 0 O
1 0 1 O O O 0 0 0 0 0 1 0 1 0 0
1 1 0 0 0 O 0 0 0 0 0 O 1 1 0 0
1 0 O 0 O 0 1 0 1 0 O O 0 O 1 0
X = 1 O 0 O 0 1 0 0 0 1 O 0 O O 1 0
1 0 0 O 1 O O 0 0 0 1 0 0 0 1 0
1 0 0 1 0 0 0 0 0 0 0 1 0 0 1 O
1 O 1 0 0 0 0 0 0 O 0 0 1 O 1 0
1 0 O 0 0 0 O 1 1 0 O 0 O 0 O 1
1 0 0 0 0 O 1 0 0 1 0 0 0 O 0 1
1 0 0 0 0 1 O 0 0 O 1 0 O 0 0 1
1 0 0 O 1 O O 0 O 0 O 1 O 0 0 1
b 1 O 0 1 O 0 0 O 0 O O 0 1 0 O 1 _

 

 

Rank(X)=12, k=1+6+4+2=13
Linear Dependency:
t3 = 0.5c2 + c3 + 1.504 + 2C5 + 2.566 + 307 + 0.5a2 + a3 + 1.5a4 + 2a5 — O.5t2 — 2

Hence the number of constraints necessary is one after setting

{01:0, 31:0, 71:0}.

256

2. Biannual Data

Example (G21). Age: 20, 21, 22, 23; Year: 1989, 1991, 1993; Cohort: 1966, 67,
68, 69, 70, 71, 72, 73

 

 

P 00 c1 c2 c3 c4 c5 c6 c7 c8 a1 a2 a3 a4 t1 t2 t3 -
1 O O 0 1 O O 0 0 1 0 O 0 1 0 O
1 0 O 1 0 0 0 O 0 0 1 0 O 1 O O
1 0 1 0 0 0 0 0 0 O 0 1 O 1 O 0
1 1 0 0 O 0 0 0 0 O O 0 1 1 O 0
1 0 0 0 O 0 1 0 0 1 0 O 0 0 1 0
X = 1 0 O 0 0 1 0 0 0 0 1 O 0 0 1 0
1 O 0 0 1 0 0 O 0 0 0 1 O 0 1 0
1 0 O 1 0 0 0 O 0 O O 0 1 O 1 0
1 0 0 0 0 0 0 O 1 1 0 O 0 0 0 1
1 O 0 0 0 O O 1 O 0 1 0 0 0 0 1
1 0 O 0 0 0 1 O 0 O O 1 O 0 O 1
_ 1 0 0 0 O 1 0 0 0 0 0 0 1 0 O 1 _

Rank(X)=11, k=1+7+3+2=13

Linear dependency:

a4=1—c2—c4—06—08—a2

t3 = —O.5*c2+0.5*c3+c5+0.5*c6+1.5*c7+08—0.5*a2+0.5*a3—O.5*t2

Hence the number of constraints necessary is two after setting

{a1=0, 51:0, 71:0}.

257

Bibliography
Attanasio, O. P. and Hoynes, H. W. (2000), ‘Differential mortality and wealth accu—
mulation’, THE JOURNAL OF HUMAN RESOURCES 35, 1-29.

Baltagi, B. H. and Wu, P. X. (1999), ‘Unequally spaced panel data regressions with
ar(1) disturbances’, Econometric Theory 15(6), 814—23.

Baum, C. F., Schaffer, M. E. and Stillman, S. (2003), ‘Instrumental variables and
gmm: estimation and testing’, The Stata Journal 3(1), 1—31.

Behrman, J. R. and Deolalikar, A. B. (1988), Health and nutrition, in H. Chenery and
T. N. Srinivasan, eds, ‘Handbook of Development Economics’, Vol. 1, Elsevier
Science Publishers B.V.

Behrman, J. R., Foster, A. and Rosenzweig, M. R. (1997), ‘The dynamics of agricul-
tural production and the calorie-income relationship: evidence from pakistan’,
Journal of Econometrics 77, 187—207.

Behrman, J. R. and Wolfe, B. L. (1984), ‘More evidence on nutrition demand: income
seems overrated and women’s schooling underemphasized’, Journal of Develop-
ment Economics 14(1-2), 105—28.

Behrman, J. R. and Wolfe, B. L. (1989), ‘Does more schooling make women better
nourished and healthier?’, Journal of Human Resources 24(4), 644—63.

Bell, A. C., Ge, K. and Popkin, B. M. (2001), ‘Weight gain and its predictors in
chinese adults’, International Journal of Obesity 25, 1079—86.

Bhandari, R. (1998), Education and food consumption behavior in China: household
analysis and policy implications, PhD thesis, North Carolina State University,
A Bell & Howell Company, Ann Arbor, MI.

Bouis, H. E. (1994), ‘The effect of income on demand for food in poor countries:
are our food consumption databases giving us reliable estimates?’, Journal of
Development Economics 44(1), 199—226.

Bound, J ., Jaeger, D. A. and Baker, R. M. (1995), ‘Problems with instrumental vari-
ables estimation when the correlation between the instruments and endogenous
explanatory variable is weak’, Journal of the American Statistical Association

90, 443-50.

Card, D. (2001), ‘Estimating the return to schooling: progress on some persistent
econometric problems’, Econometrica 69(5), 1127—60.

Cebu-Study-Team (1992), ‘A child health production function estimated from longi-
tudinal data’, Journal of Development Economics 38(2), 323—51.

Chen, X. (1990), ‘Dietary standards and goals for the people’s republic of china with
special emphasis on nutritional requirements of chinese infants’, Progress in Food
and Nutrition Science 14, 83—92.

258

Costa, D. L. and Steckel, R. H. (1995), Long-term trends in health, welfare, and
economic growth in the united states. National Bureau of Economic Research,
Historical Paper no. 76.

Deaton, A. (1988), ‘Quality, quantity and spatial variation in price’, American Eco-
nomic Review 78(3), 418-30.

Deaton, A. (1997), The Analysis of Household Survey: a Microeconometric Approach
to Development Policy, The Johns Hopkins University Press: Baltimore.

Dowler, E. A. and Seo, Y. O. (1985), ‘Assessment of energy intake: estimates of food
supply vs measurement of food consumption’, Food Policy 10(3), 278—88.

Drewnowski, A. and Popkin, B. M. (1997), ‘The nutrition transition: new trends in
the global diet’, Nutrition Reviews 55(2), 31.

Du, 8., Lu, B., Zhai, F. and Popkin, B. M. (2002), ‘A new stage of the nutrition
transition in china’, Public Health Nutrition 5(1A), 169—74.

Evenson, R. E. and Pray, C. E. (1994), ‘Measuring food production: with reference
to south asia’, Journal of Development Economics 44(1), 173—97.

Fogel, R. W. (1986), Nutrition and the decline in mortality since 1700: some prelim—
inary findings, in S. L. Engerman and R. E. Gallman, eds, ‘Long—Term Factors
in American Economic Growth’, The University of Chicago Press.

Fogel, R. W. (1994), ‘Economic growth, population theory and physiology: the bear-
ing of long-term processes on the making of economic policy’, The American
Economic Review 84(3), 369—95.

Foster, A. D. (1995), ‘Prices, credit markets and child growth in low-income rural
areas’, The Economic Journal 105, 551—70.

Ge, K., Chen, C. and Shen, T. (1991), ‘Food consumption and nutritional status in
china: Achievements, problems and policy implications, part i’, Food, Nutrition
and Agriculture 1(2/3), 54—61.

Ge, K., Chen, C., Shen, T. and Zhang, S. (1992), ‘Food consumption and nutritional
status in china: Achievements, problems and policy implications, part ii’, Food,
Nutrition and Agriculture 2(4), 10—17.

Ge, K., Weisell, R., Guo, X., Cheng, L., Ma, H., Zhai, F. and Popkin, B. M. (1994),
‘The body mass index of chinese adults in the 19803’, European Journal of Clin-
ical Nutrition 48(Suppl 3), 8148—54.

Grossman, M. (1972), ‘On the concept of health capital and the demand for health’,
The Journal of Political Economy 80, 223—55.

259

Grossman, M. (2000), The human capital model, in A. J. Culyer and J. P. Newhouse,
eds, ‘Handbook of Health Economics’, Vol. 1a, Elsevier Science Publishers B.V.,
chapter 7.

Cu, E. (2001), ‘Market transition and the transformation of the health care systems
in urban china’, Policy Studies 22(3-4), 197—215.

Guo, X. (1998), The impact of household income and food prices on food consumption
and dietary fat intake in China, 1989-1993, PhD thesis, University of North
Carolina, UMI Dissertation Services: A Bell & Howell Company, Ann Arbor,
MI.

Guo, X., Mroz, T. A., Popkin, B. M. and Zhai, F. (2000), ‘Structural change in the
impact of income on food consumption in china, 1989-1993’, Economic Develop-
ment and Cultural Change 48, 737—60.

Guo, X., Popkin, B. M., Mroz, T. A. and Zhai, F. (1999), ‘Food price policy can sig-
nificantly reduce the negative effects of the nutrition transition in china’, Journal
of Nutrition 129, 994—1001.

Guo, X., Popkin, B. M. and Zhai, F. (1999), ‘Patterns of change in food consumption
and dietary fat intake in chinese adults, 1989-1993’, Food and Nutrition Bulletin
20(3), 344—53.

Heckman, J. and Robb, R. (1985), Using longitudinal data to estimate age, period
and cohort effects in earnings equation, in W. M. Mason and S. E. Fienberg, eds,
‘Cohort Analysis in Social Research: Beyond the Identification Problem’, New
York: Springer-Verlag, pp. 137—50.

Heilig, G. K. (1999), China Food: Can China Feed Itself ?, IIASA, Laxenburg.

Holford, T. R. (1983), ‘The estimation of age, period and cohort effects for vital rates’,
Biometrics 39, 311—24.

Johnson, D. G. (2003), ‘Provincial migration in china in the 19903’, China Economic
Review 14, 22—31.

Ke, B. (1999), Policy and institutional change for agriculture in china: production,
consumption and trade implications, in G. H. Peters and J. won Braun, eds,
‘Food Security, Diversificaiton and Resource Management: Refocusing the Role
of Agriculture’. Proceedings of the twenty—thrid international conference of agri-
cultural economists.

Kuczmarski, R. J., Flegal, K. M., Campbell, S. M. and Johnson, C. L. (1994), ‘In-
creasing prevalence of overweight among us adults: the national health and nu-
trition examination surveys’, The Journal of the American Medical Association

272, 205-11.

260

Lakdawalla, D. and Philipson, T. (2002), The growth of obesity and technological
change: a theoretical and empirical examination, working paper 8946, National
Bureau of Economic Research.

Lin, J. Y. (1988), The household responsibility system in China’s rural reform, in
A. Maunder and A. Valds, eds, ‘Agriculture and Governments In An Interde-
pedent World’. Proceedings of the 20th international conference of agricultural
economists.

Luo, Z. (2003a), Age, cohort and year analysis in the socioeconomic determinants of
adult BMI in the 1990.9, Michigan State University, Department of Economics.

Luo, Z. (2003b), Socioeconomic Determinants of Body Mass Index of Adult Chinese
in the 19903, Michigan State University, Department of Economics.

Ma, H. and Popkin, B. M. (1995), ‘Income and food-consumption behaviour in china:
a structural-shift analysis’, Food and Nutrition Bulletin 16(2), 155—65.

McKenzie, D. J. (2002), Disentangling age, cohort, and time effects in the additive
model, working paper 009, Stanford University Economics Department.

NIH (1998), Clinical Guidelines On The Identification, Evaluation, and Treatment of
Overweight and Obesity in Adults. NIH Publication No 98-4083.

Paeratakul, S., P0pkin, B. M., Ge, K., Adair, L. S. and Stevens, J. (1998), ‘Changes
in diet and physical activeity affet the body mass index of chinese adults’, In-
ternational Journal of Obesity 22, 424—31.

Piazza, A. (1983), Tfends in food and nutrient availability in china, 1950-81, Working
Paper 60, World Bank, World Bank, Washington, DC.

Pitt, M. M. (1983), ‘Food preferences and nutrition in rural bangladesh’, The Review
of Economics and Statistics 65(1), 105—14.

Pitt, M. M. and Rosenzweig, M. R. (1985), ‘Health and nutrient consumption across
and within farm households’, The Review of Economics and Statistics 67, 212—
23.

Pitt, M. M., Rosenzweig, M. R. and Gibbons, D. M. (1993), ‘The determinants and
consequences of the placement of govenment programs in indonesia’, World Bank
Economic Review 7(3), 319—48.

Poleman, T. T. (1981), ‘A reappraisal of the extent of world hunger’, Food Policy
6(4), 236—52.

Popkin, B. M. (1999), ‘Urbanization, lifestyle changes and the nutrition transition’,
World Development 27 (11), 1905—16.

Popkin, B. M. (2002), ‘The shift in stages of the nutrition transition in the developing
world differs from past experiences!’, Public Health Nutrition 5(1A), 205—14.

261

Popkin, B. M. and Doak, C. (1998), ‘The obesity epidemic is a worldwide phe—
nomenon’, Nutrition Reviews 56(4), 106—14.

Popkin, B. M., Ge, K., Zhai, F., Guo, X., Ma, H. and Zohoori, N. (1993), ‘The
nutrition transition in china: a cross-sectional analysis’, European Journal of
Clinical Nutrition 47, 333—46.

Popkin, B. M., Paeratakul, S., Zhai, F. and Ge, K. (1995a), ‘Dietary and environ-
mental correlates of obesity in a population study in china’, Obesity Research
3(Suppl 2), 8135—43.

Popkin, B. M., Paeratakul, S., Zhai, F. and Ge, K. (1995b), ‘A review of dietary
and environmental correlates of obesity with emphasis on developing countries’,
Obesity Research 3(Suppl 2), 8145—53.

Popkin, B. M., Paeratakul, S. and Zhai, K. G. F. (1995), ‘Body weight patterns
among the chinese: results from the 1989 and 1991 china health and nutrition
surveys’, American Journal of Public Health 85(5), 690—4.

Reddy, K. S. (2002), ‘Cardiovascular diseases in the developing countries: dimensions,
determinants, dynamics and directions for public health action’, Public Health
Nutrition 5(1A), 231—7.

Rosenzweig, M. R. and Schultz, T. P. (1983), ‘Estimating a household production
function: heterogeneity, the demand for health inputs, and their effects on birth
weight’, The Journal of Political Economy 91(5), 723—46.

Ruhm, C. (2000), ‘Are recessions good for your health?’, The Quarterly Journal of
Economics 115(2), 617—50.

Schultz, T. P. (1984), ‘Studying the impact of household economoic and com-
munity variables on child mortality’, Population and Development Review
10(suppl), 215—35.

Sickles, R. C. and Taubman, P. (1997), Mortality and morbidity among adults and
the elderly, Elsevier Science B. V.

Smil, V. (1981), ‘China’s food: availability, requirements, composition, and
prospects’, Food Policy 6(2), 67—77.

Smith, J. P. (1998), ‘Socioeconomic status and health’, The American Economic
Review 88(2), 192-96.

Smith, J. P. and Kington, R. (1997), ‘Demographic and economic correlates of health
in old age’, Demography 34(1), 159—70.

Srinivasan, T. N. (1992), Undernutrition: concepts, measurements, and policy impli-
cations, Clarendon Press, Oxford, chapter 4, pp. 97—120.

262

Srinivasan, T. N. (1994), ‘Data base for development analysis: an overview’, Journal
of Development Economics 44(1), 3—27.

SSB (1988-1998), China Statistical Yearbook, State Statistics Bureau, China Statisti—
cal Publishing House.

Stookey, J. D., Zhai, F., Zohoori, N. and Popkin, B. M. (2000), ‘Nutrition of elderly
people in china’, Asia Pacific Journal of Clinical Nutrition 9(4), 243—51.

Strauss, J. (1984), ‘Joint determination of food consumption and production in ru-
ral sierra leone, estimation of a household-firm model’, Journal of Development
Economics 14(1-2), 77—103.

Strauss, J ., Beegle, K., Dwiyanto, A., Herawati, Y., Pattinasarany, D., Satriawan, E.,
Sikoki, B., Sukamdi and Witoelar, F. (2004), Indonesian living standards three
years after the crisis: evidence from the Indonesia Family Life Survey, Institute
of South East Asia Study, Singapore.

Strauss, J. and Thomas, D. (1995), Human resources: empirical modeling of house-
hold and family decisions, in J. Behrman and T. N. Srinivasan, eds, ‘Handbook
of Development Economics’, Vol. 3, Elsevier Science Publishers B.V.

Strauss, J. and Thomas, D. (1996), ‘Measurement and mismeasurement of social
indicators’, The American Economic Review 86(2), 30—34.

Strauss, J. and Thomas, D. (1998), ‘Health, nutrition, and economic development’,
JOURNAL OF ECONOMIC LITERATURE 36(2), 766—817.

Subramanian, S. and Deaton, A. (1996), ‘The demand for food and calories’, Journal
of Political Economy 104(1), 133—62.

Sundquist, J. and Johansson, S.-E. (1998), ‘The influence of socioeconomic status,
ethnicity and lifestyle on body mass index in a longitudinal study’, International
Journal of Epidemiology 27, 57—63.

Thomas, D., Lavy, V. and Strauss, J. (1996), ‘Public policy and anthropometric
outcomes in the Céte d’Ivoire’, Journal of Public Economics 61, 155—92.

Wang, Q., Jensen, H. H. and Johnson, S. R. (1993), ‘China’s nutrient availability and
sources, 1950-91’, Food Policy pp. 403—13.

Welch, F. (1970), ‘Education in production’, Journal of Political Economy 78(1), 35—
59.

Whitewell, G. and Nicholas, S. (2001), ‘Weight and welfare of australians, 1890-1940’,
Austrilian Economic History Review 41(2), 159—75.

Whitney, E. N. and Rolfes, S. R. (1996), nderstanding Nutrition, 7 edn, West Pub-
lishing Company.

263

Willett, W. (1998), Nutritional Epidemiology, New York: Oxford University Press.

Wolpin, K. (1997), Determinants and consequences of the mortality and health of
infants and children, in M. R. Rosenzweig and O. Stark, eds, ‘Handbook of
Population and Family Economics’, Elsevier Science B. V.

Wooldridge, J. (2002), Econometric Analysis of Cross Section and Panel Data, The
MIT Press.

World Development Indicators (1999), World Bank.

Wu, Y. (1997), ‘China’s health care sector in transition: resources, demand and
reforms’, Health Policy 39, 137—52.

Yaohui, Z. (1999), ‘Leaving the countryside: rural-to—urban migration decisions in
china’, American Economic Review 89(2), 281—86.

Zhai, F., Fu, D., Du, 8., Ge, K., Chen, C. and Popkin, B. M. (1996), ‘Evaluation
of the 24—hour individual recall method in china’, Food and Nutrition Bulletin
17(2), 154—61.

Zhai, F., Fu, D., Du, S., Ge, K., Chen, C. and Popkin, B. M. (2002), ‘What is china
doing in policy-making to push back in the negative aspects of the nutrition
transition’, Public Health Nutrition 5(1A), 269—73.

264