v.
.4 1.} . JV?! 5».
.1 . ‘p. .5.
\‘11?.h: . y t 142. :33”.

5*. 5:4! .
n
a.

I. ‘3]:
.2... “(2.6.3:
v. A...” "“Wwi ‘ ‘ .
I a :V: . ‘ . .u..\..... w
r ‘ l: ;. ..
1. .. .l

E

$1.35.;
.3

J 65.} I...

5 Si: . .

.‘w . fififimfinfi‘ a.

1“,? gunman?” wwimm‘,

s . LL?

r

a.
x:

.
L...

x 1...
~ 1 :1. f4
0

:4...) .1.-
5.3:. 1 s
o.
3.9.:
23.».

35.;
2. at.

w... v.1:
flmuvfihuflu $4.5. _
Oil... hm“; fi

. A.
A :IvI‘

it v. .3;
an. in

rum...

1
l!

..!\5$u..i
ii D0)

 

 

 

 

may“

I
V :3:

 

 

 

3 LIBRARY
59" ’0 Michigan State
University

This is to certify that the
dissertation entitled

THE IMPACT OF AIDS-RELATED PRIME-AGE MORTALITY
ON RURAL FARM HOUSEHOLDS: PANEL SURVEY
EVIDENCE FROM ZAMBIA

presented by

ANTONY CHAPOTO

has been accepted towards fuifillment
of the requirements for the

DOCTORAL degree in AGRICULTURAL ECONOMICS

 

 

4%
Major PMfi Signature '
JI’QA‘ % 20‘): 75> 0 C

Date

 

MSU is an Affirmative Action/Equal Opportunity Institution

.--.--.-a----o-.---.-.‘._ _

9*-'---I-v-o---l.-----q---.-.-.-.-.-.---u-.—.-

o.-.--.—.—.-.-.-..--u-I-g

 

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

N06 1) @2009

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/CIRC/DateDue.indd-p.1

THE IMPACT OF AIDS-RELATED PRIME-AGE MORTALITY ON RURAL
FARM HOUSEHOLDS: PANEL SURVEY EVIDENCE FROM ZAMBIA

By

Antony Chapoto

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Department of Agricultural Economics

2006

 

mo:

358

[hm
th: 5

mi

 

 

 

ABSTRACT

THE IMPACT OF AIDS-RELATED PRIME-AGE MORTALITY ON RURAL
FARM HOUSEHOLDS: PANEL SURVEY EVIDENCE FROM ZAMBIA

By

Antony Chapoto

Using nationally representative panel data of rural farm households surveyed in
2001 and 2004, this dissertation is divided into two parts. First, the study determines the
ex-ante socioeconomic characteristics of prime-age adults who die between the ages of
15 and 59 years of disease-related causes and second, estimates the impact of prime-age
mortality on rural farm households’ composition, farm and crop production, livestock
assets and off-farm income.

Given the dramatic rise in prime-age mortality in Zambia and other African
countries due to HIV/AIDS, there is a critical need to better understand the pathways
through with the disease in transmitted. The first essay of this dissertation is dedicated to
this end. Results from several probit models show that single women are 2 to 5 times
more likely to die of disease-related causes as women who are the heads or spouses of
their households. There is no clear relationship between educational attainment and
probability of dying; both well educated and poorly educated men and women should
continue to be targeted for HIV/AIDS behavior change campaigns. Also, social factors
driving the spread of AIDS are considerably more complex than simply poverty-based
explanations, although poverty may certainly contribute to risky behavior and poor health

which are important pathways by which the disease is spread.

 

(A.
k .

i7
SIL

ho:

SUE.

 

 

The second essay of this dissertation determines the impacts of prime-age
mortality on rural households in Zambia. Using prior death in the household (which is
found to be a predictor of future mortality in the household), age dummies, and drought
shocks as instruments for prime-age death between 2001 and 2004, the Hausman-Wu chi
square test for endogeneity shows that prime-age mortality variables are endogenous for
pooled ordinary least squares (OLS) models. Differencing the time-invariant unobserved
household characteristics largely addressed the endogeneity problem. Based on these
difference models, the second essay highlights the following: (1) in response to the
death of a male household head, poorer households have substantially greater difficulties
in coping than non-poor households, which are likely to almost fully restore household
size to pre-death levels; (2) in contrast to the conventional wisdom that afflicted
households cope with the‘reduction in family size by switching to labor-saving crops
such as roots and tubers, the findings indicate no clear pattern of crop shifts to labor-
saving crops. The death of non-spouse females in the household is actually associated
with a 5% decline in area under roots and tubers; (3) wealth status does not seem to
exacerbate the impact of mortality on area under cultivation. Instead, land cultivated and
area under cereals decline more in relatively non-poor households than among poor
households after the death of a male head of household; (4) afflicted households liquidate
small animals to cope with the impact of prime-age (PA) death and the value of cattle
assets appears to suffer greatly from the death of a PA male head of household. These
findings provide important information that may assist the Zambian government, donors,
and development planners in developing specific policies or interventions to mitigate the

impacts of AIDS on vulnerable households.

To my mother"

*I would like dedicate this work to my mother who passed away on August 5, 2002.
Unfortunately, she could not live to see the fruits of her love and sacrifice during my
academic life. I know she would have loved to see me smile and say thank you for

making my dream come true. May her soul rest in peace.

iv

 

 

la}
0?}
prc
II a
al‘l

All

Lil

Eia

Ou

ol'

 

ACKNOWLEDGEMENTS

I would like to thank my major professor and dissertation supervisor, Dr. Thom
Jayne, for mentoring me through various publications, providing various outreach
opportunities for me in Zambia, and for the detailed comments and suggestions he
provided on this and other work. He fostered a stress-flee working relationship which
was crucial to the completion of this dissertation. I would also like to extend my
appreciation to other committee members: Dr. John Giles, Dr. Gretchen Bierbeck, Dr.
Allan Schmid, Dr. Mike Weber and Dr. John Strauss for being there whenever needed.

I am also indebted to my colleagues in the department of agricultural economics,
Lilian Kirimi, Mary Mathenge, Cathrine Ragassa, Fernardo Balsevich, Lesiba Bopape,
Elan Satriawan, Ricardo Labarta, just to mention a few, for their support throughout my
graduate studies at Michigan State University. My gratitude also goes to Nicole Mason
for proof reading this manuscript as well as making valuable suggestions and comments.

I also would like to express my deepest thanks and respect to my wife, Tendayi
Chapoto (nee Chitiyo) for her unwavering support during the writing of this dissertation.
Our children, Tafadzwa Antony Jr. and Ashley Tinotenda, have been my greatest source
of inspiration for finishing this dissertation.

My sincere appreciation goes to my father, Mr. Dominic Chapoto and my brothers
and sisters for their support and care over the years.

Finally, this work would not have been possible without the financial support
from the USAID/Zambia mission to the Food Security Research Proj est/Zambia and

Food Security 111.

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. ix
LIST OF FIGURES ......................................................................................................... xiii
CHAPTER 1
INTRODUCTION .............................................................................................................. 1
CHAPTER 2
CHARACTERISTICS OF INDIVIDUALS AFFLICTED BY AIDS-RELATED
MORTALITY IN ZAMBIA ............................................................................ 6
2.1 . Introduction ...................................................................................................... 6
2.2. Data and Methodology ..................................................................................... 7
2.2.1. Data and attrition ..................................................................................... 7
2.2.2. Estimation strategies and variables ....................................................... 18
2.3. Results ................................................................. ............. 21
2.3.1. Adult mortality and HIV prevalence in Zambia ................................... 21
2.3.2. Descriptive analysis .............................................................................. 23
2.3.3. Attributes of Deceased Prime-Age Individuals .................................... 25
2.3.3.1. Relationship of deceased to head of household ................................ 33
2.3.3.2. Age groups ........................................................................................ 33
2.3.3.3. Education, mobility, income and household wealth indicators ........ 34
2.4. Summary of findings ...................................................................................... 39
CHAPTER 3

IMPACT OF HIV/AIDS-RELATED MORTALITY ON RURAL FARM
HOUSEHOLDS IN ZAMBIA: IMPLICATIONS FOR POVERTY

REDUCTION STRATEGIES ........................................................................ 44
3. 1 . Introduction ................................................................................................... 44
3.2. Review of empirical studies ........................................................................... 46

vi

3.2.1. Empirical limitations of prior studies ................................................... 46
3.2.2. Effects on household composition and labor availability ..................... 48
3.2.3. Effects on agricultural production and cropping patterns ..................... 50
3.2.4. Effects on assets and non-farm income ................................................ 52
3.3. Data and methods ........................................................................................... 53
3.3.1. Data ....................................................................................................... 54
3.3.2. Attrition ................................................................................................. 56
3.3.3. Relationship between Adult mortality and HIV /AIDS ......................... 60
3.3.4. Conceptual framework .......................................................................... 63
3.3.5. Econometric model ............................................................................... 69
3.3.6. Empirical model and estimation strategy .............................................. 72
3.3.7. Econometric issues ................................................................................ 75
3.3.7.1. Attrition bias ..................................................................................... 75
3.3.7.2. Identification of impact of death ....................................................... 79
3.4. Results ............................................................................................................ 87
3.4.1. First-stage regression models ................................................................ 87
3.3.2. Is prime-age death endogenous? ........................................................... 88
3.4.3. Impact of prime-age death on household composition ......................... 91
3.4.4. Impact of PA death on farm and crop production ............................... 103
3.4.5. Impact of PA mortality on value of livestock ..................................... 122
3.4.6. Impact of PA mortality on off-farm income ....................................... 125
3.4.7. Does the impact differ if the sample excludes homecoming sick prime-
age household members who died between 2001 and 2004? ............. 130
3.5 Summary of findings .................................................................................... 134

vii

CHAPTER 4

CONCLUSION AND POLICY RECOMMENDATIONS ............................................ 142
APPENDICES ................................................................................................................ 1 5 5
REFERENCES ............................................................................................................... 204

viii

LIST OF TABLES

Table 2.1. Relationship between household size, attrition, dissolution, and prime-age
mortality in 2001-2004. .................................................................................... 9
Table 2.2. Prevalence of prime-age (PA) mortality by province, rural Zambia between
2001 and 2004. ............................................................................................... l 1
Table 2.3. Household characteristics stratified by attrition status ................................... 13
Table 2.4. Individual-level re-interview model (Probit) ................................................... 16

Table 2.5. Descriptive statistics among prime-age adults who died due to illness in

Table 2.6

Table 2.7.

Table 2.8.

Table 3.1.

Table 3.2.

Table 3.3.

Table 3.4.

Table 3.5.

Table 3.6.

2001-2004, and remaining prime-age adults in the sample .......................... 24

Probit models of PA mortality in 2001—2004 by gender and wealth status
(Province-fixed effects models): Corrected for attrition. ............................... 27

Probit models of PA mortality in 2001-2004 by gender and wealth status,
(Village-fixed effects models): Corrected for attrition .................................. 30

Simulations of the probability of mortality based on specific individual and
household attributes. ...................................................................................... 37

Prevalence of prime-age (PA) mortality by province, rural Zambia between

2001 and 2004. ............................................................................................... 57
Characteristics of non-afflicted and afflicted households ............................... 64
Household-level re-interview model (Probit) ................................................. 78

F irst-stage F —statistic for significance of identifying instruments, Pseudo R2
and % correctly predicted ............................................................................... 89

Summary table of Hausman Wu Chi-square test and Sargan N*R square test
for overidentification for pooled and differenced samples. ........................... 90

The impact of PA mortality on household composition by gender and position
of the deceased ............................................................................................... 93

Table 3.7. The impact of PA mortality on household composition by initial household

pre-death characteristics. ................................................................................ 97

Table 3.8. The impact of PA mortality on household composition by gender and position

in household by poverty status ..................................................................... 100

ix

Table 3.9. The impact of PA mortality on cultivated land by gender and position in

household ..................................................................................................... 107
Table 3.10. The impact of PA mortality on cultivated land by gender, position and initial
household pre-death characteristics ............................................................. 110

Table 3.1 l. The impact of PA mortality on cultivated land by gender and position in
household by poverty status ......................................................................... 1 13

Table 3.12. The impact of PA mortality on gross value of output and gross output per
hectare by gender and position in household ............................................... l 16

Table 3.13. The impact of PA mortality on crop output and output per hectare. by gender
and position in household ............................................................................. l 19

Table 3.14. The impact of PA mortality on assets and off-farm income by gender and
position in household ................................................................................... 124

Table 3.15. The impact of PA mortality on assets and off-farm income by initial
household pre-death characteristics ............................................................. 126

Table 3.16. The impact of PA mortality on assets and off-farm income by gender and
position in household by poverty status ....................................................... 129

Table 3.17. The impact of PA mortality on household composition by gender and
position of the deceased: Restricted sample ................................................. 133

Table A2.]. Descriptive Statistics of variables used in the analysis .............................. 156

Table A22. Probit model results of PA mortality between 2001-2004 by gender (village-
fixed effects) ................................................................................................. 158

Table A.2.3. Probit Models of prime-age mortality in 2001-2004 by gender (province-
fixed effects) ................................................................................................. 161

Table A24. Probit Models of PA mortality in 2001-2004 by gender and poverty
(income) status (province-fixed effects models): Corrected for attrition ..... 164

Table A25. Probit Models of PA mortality in 2001-2004 by gender and poverty
(income) status, (village-fixed effects models): Corrected for attrition ....... 167

Table A26. Probit Models of PA mortality in 2001-2004 by gender and wealth status
(Province-fixed effects models): Not corrected for attrition ........................ 170

Table A27. Probit Models of PA mortality in 2001-2004 by gender and wealth status,
(Village-fixed effects models): Not corrected for attrition .......................... 173

Table A3. 1. Mean values by year and whether a household is afflicted by gender of
deceased or non-afflicted (l) ........................................................................ 177

Table A3.2. Mean values by year and whether a household is afflicted (by gender and

status of deceased) or non-afflicted (2) ........................................................ 178
Table A3.3. Descriptive statistics: right hand variables of impact models ..................... 179
Table A3.4. Correlation matrix: First stage regressions variables .................................. 180
Table A3.5. Correlation matrix: Impact models ............................................................. 181
Table A3.6. First stage regression results: Pooled probit .............................................. 182

Table A3.7. First stage probit results: for use in the 2nd stage differenced models. ...... 184

Table A3.8. Tests of endogeneity and overidentifying restrictions of PA mortality on land
cultivated and area under cereals ................................................................. 188

Table A3.9. Tests of endogeneity and overidentifying restrictions of PA mortality on area
under roots and tubers and other crops ........................................................ 189

Table A310. Tests of endogeneity and overidentifying restrictions of PA mortality on
values of gross output and gross output per hectare .................................... 190

Table A3.] 1. Tests of endogeneity and overidentifying restrictions of PA mortality on
values of cattle and small animals ................................................................ 191

Table A3. 12. Tests of endogeneity and overidentifying restrictions of PA mortality on
values of household size and off-farm income ............................................. 192

Table A313. Tests of endogeneity and overidentifying restrictions of PA mortality on
number of males and females ....................................................................... 193

Table A314. Tests of endogeneity and overidentifying restrictions of PA mortality on
number of boys and girls .............................................................................. 194

Table A3.15. The impact of PA mortality on HH composition by gender and position in
HH by poverty status: Not corrected for attrition ........................................ 195

Table A316. The impact of PA mortality on cultivated land by gender and position in
household: Not corrected for attrition .......................................................... 196

Table A317. The impact of PA mortality on cultivated land by gender and position in
HH by poverty status: Not corrected for attrition ........................................ 197

xi

Table A3.18. The impact of PA mortality on gross value of output and gross output per
hectare by gender and position in household: Not corrected for attrition... 198

Table A3.19. The impact of PA mortality on assets and off-farm income by gender and
position: Not corrected for attrition .............................................................. 199

Table A320. The impact of PA mortality on assets and off-farm income by gender and
position by poverty status: Not corrected for attrition ................................. 200

Table A321. The impact of PA mortality on cultivated land by gender and position in
household: Restricted sample ....................................................................... 201

Table A322. The impact of PA mortality on assets and off-farm income by gender and
position in household: Restricted sample ..................................................... 202

Table A323. The impact of PA mortality on gross value of output and gross output per
hectare by gender and position in household: Restricted ............................. 203

xii

LIST OF FIGURES
Figure 3.1. Potential Pathways by which HIV/AIDS prime-age morbidity and mortality
affects rural farm household ........................................................................... 68

Figure A21. Distribution of illness-related prime-age mortality by gender and age-
groups between 2001 and 2004 .................................................................... 157

Figure A3.1. Zambia’s HIV Prevalence Rates, by Province, 2001-2002. ..................... 176

xiii

 

 

 

 

(I)

”T“

in

CV

 

 

CHAPTER 1
INTRODUCTION

It is now widely accepted that policy makers and development planners in Sub-
Saharan Afiica can no longer afford to ignore the potential economic impacts of
HIV/AIDS. Yet, more than 20 years have passed since the onset of the disease in Africa
and there remains a dearth of quantitative information available on the economic impacts
of the disease on rural farm households and their responses.1 Decisions need to be made

soon rather than later on how to respond to the disease. However, due to limited

 

resources, there is a gap between desired and available levels of funding and human
resources for HIV prevention (e. g., vaccines, behavior change), treatment (e.g., ARV
therapies), and mitigating the impacts of AIDS (e. g., social and economic programs to
protect the living standards of afflicted households and hard-hit communities). Moreover,
every dollar invested in AIDS prevention, treatment, and mitigation cannot be used to
promote basic education, improved agricultural technology, the development of
infrastructure and markets, and other long-term investments necessary to raising living

standards. Therefore, governments and international organizations need solid guidance

.4 ""‘-a .. a...

on the cost-effectiveness of alternative kinds of investments to simultaneously combat the
AIDS pandemic and the chronic poverty also prevalent in the region. Combating the I
interrelated problems of HIV/AIDS and poverty requires a better empirical understanding

of how communities, households, and individuals respond to the disease, how their

responses differ according to various circumstances, the binding constraints that must be

 

To quote Scrcchitano and Whitlock (2002, p. 2), “the internatlonal community IS now at the stage where
technicians will have to shift from merely describing the effects of HIV/AIDS on agricultural production,
to actually measuring it.”

 

 

 

TC

pr

bf

 

 

 

relieved in order to allow for recovery, and the factors affecting resistance and
resilience.2

Southern Africa is home to less than 2% of the world’s population yet it has about
30% of the world’s people living with HIV/AIDS (UNAIDS, 2003). There are seven
countries in the world where HIV-prevalence rates for prime-age adults (aged 15-49)
exceed 20%, all of which are in Southern Africa. Zambia, which is the country of focus
in this study, is one of these seven countries.

Zambia’s estimated HIV prevalence rate is 21 .6%,3 making it the sixth most
afflicted country in the world. A nationally representative Demographic and Health
Survey (DHS) in 2001—2002 found that almost 16% people aged 15-49 (who agreed to be
tested) were HIV -positive. Estimates based on antenatal clinic-based surveillance data
reported higher HIV/AIDS prevalence rates of up to 22% (ZDHS, 2002). While HIV
prevalence rates vary geographically from 8% percent in Northern Province to 22% in
Lusaka Province, it is reasonable to assume that neither rural nor urban communities will
be spared from the potential negative economic effects of HIV/AIDS.

There are a growing number of studies in Afiica attempting to measure the
economic impact of HIV/AIDS on rural households’ livelihood and welfare. These
studies are constrained by the absence of longitudinal comprehensive data from farm
households, hence most empirical studies resort to making conclusions based on cross-

sectional data analysis. Also, a major difficulty in measuring the impact of adult

 

Resistance refers to the ability to av01d contracting the disease, while resrlzence refers to the ability to
withstand its effects, once it has manifested in communities (Loevinsohn and Gillespie, 2003).

3 These estimates are acknowledged to be potentially overstated, because (1) they are based on blood tests
of women visiting antenatal clinics located mainly in urban areas, which are considered to have higher
prevalence rates than in rural areas; and (2) the antenatal data does not include men, who are likely to have
lower rates of HIV infection than women (UNAIDS, 2003; Chin, 2003).

. T-fi

 

 

 

 

1111'

11::

r:-

 

 

 

 

 

mortality, especially mortality attributable to AIDS, is that it is influenced by behavioral
choices rather than by random events. Individuals and households incurring adult
mortality are more likely to display certain characteristics. For example, during the early
years of the epidemic in sub-Saharan Africa, evidence suggested that men and women
with higher education and income were more likely to contract HIV than others because
they were more likely to have numerous sexual partners (Ainsworth and Semali 1998;
Gregson, Waddell, and Chandiwana 2001).4 If prime-age mortality remains correlated
with individual and household characteristics such as social status, education, and wealth
— which are also important determinants of incomes and other welfare indicators — failure
to control for these characteristics may generate biased estimates of the impact of adult
mortality on household welfare. The few longitudinal empirical studies measuring the
impact of adult mortality from AIDS on agriculture and rural farm households’ welfare
acknowledge that illness and the death of prime-age adults especially mortality
attributable to AIDS may be endogenous to outcomes but do not explicitly test and/or
attempt to correct for this potential problem.

Using nationally representative two-year panel survey data of 5,420 households
from Zambian small- and medium-scale rural farm households this dissertation measures
the economic impact of AIDS-related prime-age mortality on rural household livelihoods.

In particular, this dissertation has three specific objectives: first, to determine the socio-

 

As information about HIV transmission spreads, it is believed that educated people are more likely to

change their behavior in ways that reduce their vulnerability to the disease compared to less educated
people.

 

 

economic characteristics of rural individuals and households afflicted5 by illness-related
mortality in Zambia; second, to estimate the effects of AIDS-related prime-age adult
mortality on indicators of rural households’ welfare, specifically agricultural production,
cropping patterns, value of livestock and off-farm income; and third, to provide
information on the effectiveness of alternative policy levers for mitigating the effects of
HIV/AIDS on rural farm households.

The analysis of the socio-economic characteristics of rural individuals and
households afflicted by illness-related death in Zambia will help us understand the
vectors of HIV transmission. There may have been one main transmission vector in early
years related to income and mobility, but it is possible that additional transmission
vectors have become important. Therefore, this needs to be clarified in order to
understand how to effectively combat the disease. In addition to providing us with
potential evidence supporting the likely endogeneity of prime-age death, the findings will
also help policy makers and development agencies to re-evaluate current policy
interventions to address the HIV /AIDS crisis and poverty through effective poverty
reduction initiatives to prevent new infection and provide appropriate treatment and care
that targets the most vulnerable households and individuals. The second part of the
dissertation, adopts and extends the empirical model of Yarnano and Jayne (2004) in two
ways: (1) I control for household characteristics and factors that may influence the

severity of the impact on rural households, such as initial (pre-death) wealth status,

 

5 This paper follows the taxonomy convention proposed by Barnett and Whiteside (2002): “Afflicted”
households are those that have incurred a prime-age death in their households; households that have not
directly suffered a death but are nevertheless affected by the impacts of death in the broader community are
referred to in this study as “affected.” Households not directly suffering a death may be non-afflicted, but
it is doubtful that there are any non-affected households in hard-hit communities of Eastern and Southern
Africa.

landholding size and effective dependency ratio; and (2) I test for the possibility that
prime-age adult mortality in the household is endogenous, something that has not been
done in previous studies. The findings from this analysis provide important information
that will assist in developing specific mitigation or interventions to counteract the
negative impacts of the disease to the vulnerable households by government, donors and
development planners as well as stimulate more debate in the research community.

The rest of this dissertation is organized as follows: Chapter 2 examines the
socio-economic characteristics of individuals afflicted by illness related mortality
(objective 1), while the third chapter determines the impacts of prime-age mortality on
household composition, land cultivated and crop mix, farm production, livestock assets
and off-farm income (objective 2). Finally, chapter 4 presents the conclusion and policy

recommendations.

CHAPTER 2

CHARACTERISTICS OF INDIVIDUALS AFFLICTED BY AIDS-RELATED
MORTALITY IN ZAMBIA

2.1. Introduction

Campaigns to prevent the spread of HIV/AIDS require accurate knowledge of the
characteristics of those most likely to contract the disease. Studies conducted in Sub-
Saharan Afiica during the 19803 generally found a positive correlation between
socioeconomic characteristics such as education, income, and wealth and subsequent
contraction of HIV (see Ainsworth and Semali, 1998, Gregson, Waddell, and
Chandiwana, 2001). However, as the disease has progressed, the relationship between
socioeconomic status and HIV contraction may have changed in many areas of Sub-
Saharan Africa, although there is little hard evidence to support this. For example, it is
increasingly believed that poverty forces some household members to adopt more risky
behaviors that contribute to HIV infection, which could mean that AIDS-related mortality
is disproportionately affecting relatively poor households. This chapter seeks to
determine the ex ante socioeconomic characteristics of individuals who die from illness
between 15 and 59 years of age (hereafter called “prime age” mortality), using nationally
representative panel data on 18,821 individuals in 5,420 households surveyed in 2001 and
2004 in rural Zambia. The findings from this chapter will help policy makers and
development agencies better understand current transmission pathways of HIV /AIDS,

which should help in the formulation of AIDS prevention and mitigation strategies.

Several probit models of disease-related mortality of prime—age (PA) individuals
in rural Zambia between May 2001 and May 2004 were estimated, using nationally
representative rural household survey data collected by the Government of Zambia. The
results of these models are used to report the probabilities of mortality over a three-year
period for a range of individual profiles that differ according to their gender, level of
income, education, months residing away from home, distance to district town, and other
individual and household characteristics.

The remainder of this chapter is organized as follows: Section 2.2 describes the
data, issues related to sample attrition between the 2001 and 2004 surveys, and estimation
methods. Estimation results and their interpretation are presented in Section 2.3. Section
2.4 discusses the conclusions of the chapter and implications for donor and government

ICSpOI'lSC.

2.2. Data and Methodology

2.2.1. Data and attrition

The study uses nationally representative longitudinal data on 18,821 prime—age
individuals (15-5 9 years of age) in 6,922 households in 393 standard enumeration areas
(SEAS)6 in Zambia surveyed in May 2001 and May 2004. The survey was carried out by
the Central Statistical Office (C80) in conjunction with the Ministry of Agriculture and

Cooperatives (MACO) and Michigan State University’s Food Security Research Project.

 

6 . . . . .

“Standard enumeration areas” (SEAS) are the lowest geographic sampling unit in the Central Statistical
Office’s sampling framework for its annual Post Harvest Surveys. Each SEA contains roughly 150 to 200
rural households.

For sampling procedures, see Megill, 2004. Of the 6,922 households interviewed in
2001, 5,420 (78.3%) were re-interviewed in May 2004. If attrition caused by
enumerators not re-visiting several SEAS in 2004 that were included in the 2001 survey is
excluded, the re-interview rate rises to 88.7%. And if attrition caused by adult household
members being away from home during the enumeration period and those refusing to be
interviewed is excluded, the re-interview rate rises to 94.5%.

Table 2.1 presents the relationship between household attrition, dissolution, and
household size in 2001. The findings show that the percentage of households “attriting”
is inversely related to household size (column C). While 8.4% of the households sampled
in 2001 contained either one or two members, these households accounted for over 12%
of the cases of attrition and 18% of the cases of household dissolution. In contrast,
65.5% of the sample contained households with 5 or more members and among these
households only 47% of attrition due to dissolution is observed. (columns C and D). In
addition, Table 2.1, column F shows that dissolution was a more important cause of
household attrition among smaller households than among larger households. By
contrast, larger households were more likely to incur a prime-age adult death (column G
and H). This is because the probability that a household will incur a prime-age adult
death is positively correlated with the number of adult members in the household.

Table 2.2 presents basic information on the households surveyed, re-interview rates,
and prevalence of disease-related mortality over the 2001-2004 period. Of the 5,420

households successfully re-interviewed, 362 of these households (6.3%) had at least one

6d message 5:28": Eases

58v ommefitm one 55 0.68 @0905 mEonomno: £0
0 Edd—OURH Edd—OOH...“ CPS—COD

m Eugen! Q 5.5.500" m :53an

”382

38 use 88 Assam Easeaaam Be 883% 3:5 35.: a8 $38308 ”88m

 

 

 

 

New cam now _ Name 30,—.
v.2 E. 03 EN mm a: mew SN
as mm 5.2 . wd S on gm a
w.w 3c NNN 06 em we coo m
we ow 1mm v6 mm cg omn n
we we wgm ow we SN vmm c
5.0 mm v.5 we we cam 33 m
5‘ cm 33 in on mom 2.4: v
06 mm mam ms mm 03 SN. m
we 9 vém N: me w: mwm m
wd H 3% find em we 09 fl
$3 398:: $0 go 558:: 528:: 598:: 52852
“Ev 06¢ DE he as Q as g
Eczema—o: mange
BBBZBEE mo 3:588 $3833: 298% aosBOmmE 9 33-88 0383
o\o we @2838 <m weak—=2: mo .x. Son mo .x. use manage E wfiawfim SON E 3%
<m math—ea mEondmsom ma wag—036 mm mEZommE mEonomsom mEonomsom m22~o§dfl Eonomsom
mEonomsom mEonomsom manomsom

 

.vchugcu E mum—3.8:. owed—Ea can iota—8mg £33.33 .05 22.02::— 5953 Eamneuflom .H.~ 03:.

 

 

('9

 

 

disease-related prime-age (PA) death over the three-year period.7 Of these 362
households incurring a prime-age disease-related death, 53 of them suffered multiple
prime-age deaths, with 45 households experiencing 2 deaths, 6 households experiencing
3 deaths, and 2 households experiencing 4 prime-age deaths. Survey design problems
made it difficult to determine the relationships between the deceased in households
suffering multiple deaths.

Using the World Health Organization (WHO) standard algorithm for diagnosis of
HIV infection in the absence of blood tests, 24% of the deaths due to diseases are
estimated to be AIDS related.8 However, because information was not collected on all
of the WHO minor symptoms, it is likely that our classification of AIDS and non-AIDS
deaths’ underestimates the percentage of deaths related to AIDS. Therefore, in this
chapter the analysis is confined to correlates of prime-age mortality due to disease in
general.

There were 419 prime-age deaths recorded in the sample, 21 of which were due to
accidents or homicide and 398 cases of illness-related PA mortality. Of the 398 cases of

illness—related PA mortality, 165 (41%) were men and 233 (59%) were women (see

 

This only includes households With mdrvrduals in both the first and second surveys. A small number of
recorded deaths were due to violence or accidents; these were excluded from the analysis.

8 Following the standard WHO algorithm, the major clinical symptoms for which data was collected are:
chronic diarrhea for more than one month, prolonged fever (intermittent and constant) for more than one
month, and weight loss of more than 10% of body weight. The minor signs were prolonged cough,
prolonged difficulty in breathing, prolonged pneumonia, thrush in the mouth and ‘rash’, which is
considered to be an indicator of generalized pruritic dematitis if it occurred in combination with two major
signs. A review of literature on verbal autopsies and lay diagnoses shows that there is no "ideal" method of
measuring AIDS-specific mortality in a Zambian population-based sample. Therefore, we cannot get a

" gold standard" diagnosis on a true population basis, since the validation of verbal autopsy studies in
literature are flawed (the validation sample come from a clinical sample and therefore are not likely to be
representative of the population) (Birbeck, 2004).

10

mat/E ma Ounce—o 2 Snow mo 0280
MS? can 88558.“ E 8ng .23» 20202 amnesia Ocegaou awfufitmufiufl 5:888 56$ 30 22323.20: $N.m One Oozommfi mm BEBE
o\0\..m .<mm no So @268 .xNO— .2332 one; o\oN.O doBBEBEeH “on $5.3 05 .52 .5283 can :2: Son no.“ 3-2 meme mm “.263 2 owaeatmd 8202

Amoco: 22:5 39an 28 2028.22 x0388 8an connfioaow demon 3902 no >863 .385 SE5
559218 cocoa: Ewfia .222 aim mafia: .529: one 55 808 8O nwsoo 322205 2me 838 one “we": an One 3:58 a .85 208 8.8 3:56 oEoEo GE:
mo venom team a E 5303 boon («o o\oO~ 55 228on $3 Emmi; Emma BEE N EonSEmwa—o 033 0 ANN nonoom 83 @8303 06 .«o 8% mic—amaze >2 moi:

$255 «we 05 E “on one? :5 VOON can SON 23502 O06 98 2023902 05 BEE, on? 252339: wE—SEE

 

 

 

 

 

 

 

2. 8.8V _2 21.3 8 228.2 88 8.32.: 2: 8.8 $8 883 8% N88 :89
88 2.82 8 8.5 Q 8.58.2 a. 8.28.2 2 8.3 8 8.3V 28 2: Bees
NM 8.5 2 8+3 2 8.28.: cm 2.28.2 mm 8.8 2 28b 8% 5 505:8
88 8.5 8 8.2: N 88:? 8 88:8 8 8.8 2 8.88 38 N: Euamsafioz
mm 2.3 2 8.3 3 28:2: S 8.288 R 2.2 mm 8.3 RS 32 8882
on 888 e 8.8 8 8.28.2 2 82:2 8 8.9 2 8.3 22 EN flash
2.. 8.3 S 88$ : 2.2:.2 R 2.3; 2 8.3 2. 8.85 as Re 2832
mm 8.3 R 8.8: 2 8.23.2 2. 8.2.34.2 8 8.8 K 8.38 cm: 82 Beam
em 8.5 e 8.3 v 8.28.: 2 8.32.: e 2.2 2 28$ N8 888 202880
3. 8.48 2 8.2V 2 22:2 8 8.2:.2 R 8.8 S. 8.88 8.3 v: .850
1:28 >1- 2x; 328:: Ox; 528:: 0 MS? 82:32 0%)? 33:5: 2x; 828:: 2x; #58:: 528::
88 E 8 8O 8O 8O 8O 8O
2 2w; SEE
owe me fine one owne— nouaoammmflo 0:“th 032 VOON

3-2 206323 an O5 2me o A OON E anew 216m VOON S 2OON

8886 Bob 338 020 on? 3 owmdadn one .23 a . E 332?..85 0023on

mi? 2168 . mass: 8 one . _ 832383 mEonomsom

228v . 8 55 8c 8:28 -2 mEonomzom
we own 5:52 v 33% owaegm
wean—BOD: @3235 . Bod mEonomsom

 

VOON van SON SEE—m Eagoaaam O5 OOONROE racism 285m «mom namEOUgOmU U350m

.VOON can SON 562502 «BEEN ~93.— .oofi>oa t3 «3:558 88% omdeEa mo 852?on .N.N 033.

ll

 

 

 

(IQ

‘70

:3‘

"1

 

 

figure A2.1 for the distribution of PA illness-related mortality by gender and age-
groups).9 Of the 18,821 prime-age adults recorded in 2001, 36% had left the sample
between 2001 and 2004 for causes other than death, e. g., moving to another location,
getting married and starting another household elsewhere. Non-afflicted individuals are
defined as those remaining in the sample and alive at the time of the second interview in
May 2004.

Excluded from this analysis are 211 prime-age individuals who joined the
household after the 2001 survey and died between 2001 and 2004. Strictly speaking, the
relevant sample is composed of prime-aged adults who were residents of sampled
households in 2001. Including individuals joining sampled households later might
overestimate the prevalence of prime-aged mortality, as indicated in Table 2.2, Columns
D and B. Other studies have found that a high proportion of HIV -positive individuals
returned to their rural families to receive terminal care after becoming ill (e. g., Kitange et

al., 1996).

Comparison of means between reinterviewed and attrited households

To test for possible bias in results due to household attrition, the mean levels of
control variables measured in May 2001 are compared for households that were re-
interviewed versus those that attrited. The means of many variables differ statistically
between re-interviewed and attrited households (Table 2.3). For example, households not
re-interviewed had slightly younger household heads (43 years vs. 45 years), smaller

household sizes with fewer children age 5 and below, fewer boys and girls age 6 to 14,

 

9 . .
After weighting, women accounted for 61% of PA mortality.

12

Table 2.3. Household characteristics stratified by attrition status.

 

 

 

Re-interviewed . Noim’ .
N=5420 lnteerewed leference

Household attributes in 2000 N=1502

Mean Std. dev Mean Std. dev Mean t-stat
Age of household head (years) 44.71 15.04 42.50 15.04 2.21” 5.72
Mean education of head and spouse(years) 5.78 3.22 5.86 3.68 -0.07 -0.27
Household size (number) 5.91 3.01 5.17 2.63 0.73" 9.58
Children 5 and under (number) 0.93 0.93 0.83 0.91 0.09" 4.08
Boys 6 to 14 (number) 1.47 1.34 1.30 1.21 0.17” 5.31
Girls 6 to 14 (number) 1.57 1.35 1.39 1.22 0.18" 6.08
Prime-age male 15 to 59 (number) 1.26 0.99 1.07 0.87 0.19” 6.71
Prime-age female 15 to 59 (number) 1.33 0.88 1.20 0.78 0.14“ 6.52
Elderly Males age 60 and above (number) 0.14 0.35 0.10 0.31 0.04” 4.06
Elderly Females age 60 and above (number) 0.11 0.33 0.10 0.30 0.02+ 1.84
Households with chronically ill adult (number) 1.27 0.46 1.31 0.50 -0.04 -0.58
Prime-age death between 1996-2000(=1) 0.10 0.30 0.10 0.30 0.00 -8
Landholding size (ha) 2.80 2.82 2.45 2.69 0.35“ 5.44
Land cultivated (ha) 1.49 1.38 1.25 1.20 0.24" 7.19
Total household income ('000 ka) 1843 3962 1819 3571 23.91 1.10
Value of assets ('000 ka) 901 2793 550 1751 351.26” 5.76
Productive assets ('000 ka)b 108 399 53 238 55.02" 6.23
Distance to nearest tarmac/main road (km) 25.32 35.49 24.93 33.39 0.39 0.58
Distance to nearest district town (km) 34.48 22.57 36.00 23.77 -1.52 -l.78

 

Source: CSO/MACO/FSRP Post Harvest Survey 1999/00 and Supplemental Surveys, 2001 and 2004

Notes:

** indicates 1 percent significance level; *indicates 5 percent significance level; + indicates 10 percent

significance level.

a t cannot be computed because the standard deviations of both groups are 0.

b . .
Productlve assets are the sum of the value of farm equipment (scotch carts, barrows and ploughs) and

livestock.

fewer prime-age males and females and elderly males, slightly smaller landholdings, less
farm equipment and animals, and slightly higher rates of chronically ill adults in 2001.
This is not surprising given the data presented in Table 2.1 showing that attriting
households were smaller to start with in 2001. Systematic differences between attritors
and non attritors, coupled with a high attrition rate, may cause concern about inference
with this data. Also, if the attrited households suffered a higher incidence of PA
mortality between 2001 and 2004, there would be attrition bias when estimating the ex
ante socioeconomic characteristics of individuals who died of AIDS-related causes.10 So
one should be worried about the possibility of systematic attrition leading to selection
bias.

In order to deal with potential attrition bias, the inverse probability weighting
(IPW) method is adopted, which assumes that the probability of being re—interviewed as a
flmction of observables information is the same as the probability of being re-interviewed
as a function of observables, plus unobservables that are only observable for non-attrited
observations (see Wooldridge, 2002).11 In general, the IPW method works well if the
observations on observed variables are strong predictors of non-attrition and if the
observations on unobserved variables are not strong predictors of non-attrition. Interview
quality variables are used to predict interview; in particular, 59 enumeration teams are
used to predict re—interview. Each enumeration team was headed by a supervisor who

was authorized to decide how much effort enumerators make to contact designated

0 Available evidence on attrition rates in longitudinal surveys in developing countries range from 5 to 30
percent for two rounds (see Alderman, et a1, 2001; Thomas, Frankenberg, and Smith, 2001, Yamano and
lame, 2004). For a discussion of IPW see Wooldridge, 2002.

The literature addressing the detection and correction of selection bias is extensive, and a complete
review of this literature is beyond the scope of this paper. Overviews of sample selection models can be
found in Fitzgerald, Gottschalk, and Moffit (1998), and Alderman et al ( 2001).

14

households afier not finding a valid respondent at home after the first visit. The re-

interview model is specified as follows:

PT0b(Rkht =1) = f (H1 Vr— jrlhk,2000aX h,2000rEhtaP) (2-1)

Where ka is one if individual (k) is in a household (h) that is re-interviewed at time t,
conditional on being interviewed in the previous survey, and zero otherwise; HIVH is
the district HIV-prevalence rate at the nearest surveillance site in 1999; Ihkgooo is a vector
of individual characteristics in 2000; Xh2000 is a set of household characteristics in the
2001 survey including landholding, productive assets, demographic characteristics
(number of children ages 5 and under, number of prime age males and females),
ownership of various assets; Em is a set of 59 enumeration team dummies; and P is a set
of 9 provincial dummies. Note that all of the variables in (2.1) are observable even for
individuals in households that were not re-interviewed in 2004.

Equation (2.1) is estimated with Probit for attrition between the 2001 and 2004
surveys, obtaining predicted probabilities (Przom). Then, the inverse probability (l/Przoor)
is computed, and applied to the probit models described in the following section for
estimating prime-age (PA) mortality.

Table 2.4 shows that single or previously married individuals were less likely to
be re-interviewed than married individuals. However, other factors constant, the results
also show that prime-age male members were 4.7 percent more likely to remain in the
households between the first and second survey compared to females. This may be
because in most parts of Zambia which are patrilineal, females are more likely to leave

their parents’ home when they marry compared to men who may marry and still live with

15

Table 2.4. Individual-level re-interview model (Probit’)

 

1=Individuals contained in 2001 and 2001 surveys,

 

 

 

 

Attributes 0=individuals contained only in 2001
dy/dx z P>z
Individual characteristics in 2000
Gender(1=male, 0=female) 0.047 6.69 0.000
Never married (=1) -0. 172 -14.28 0.000
Previously married (=1) -0. 139 -9. 13 0.000
Age group (=1)
Age 20-24 -0.076 -5.92 0.000
Age 25-29 -0.021 -1.42 0.155
Age 30-34 0.021 1.36 0.173
Age 35-39 0.037 2.02 i 0.043
Age 40-44 0.090 4.70 0.000
Age 45-49 0.103 4.56 0.000
Age 50-54 0.134 6.64 0.000
Age 55-59 0.124 5.38 0.000
Years of education (=1)
1-3 years 0.023 1.48 0.139
4-6 years 0.020 1.53 0.125
7 years 0.000 . -0.02 0.985
8 years and above -0.007 -0.42 0.673
Salary wage income (=1) -0.044 -3.15 0.002
F orrnal/Informal business activity (=1) 0.033 2.50 0.012
Months away from home (number) -0.013 -4.51 0.000
Household characteristics in 2000
Children 5 years and under (number) 0.010 1.77 0.077
Children ages 6 to 11 (number) 0.002 0.81 0.420
Prior PA death from diseases in 1996-2000 (=1) -0.038 -2.25 0.024
Landholding size (Ha) 0.002 1.01 0.312
Drafi animals and farm equipment (ka) 0.002 2.31 0.021
Community characteristics
District HIV prevalence rate in 1999 -0.003 -1.73 0.084
District on the line of rail (=1 , 0 otherwise) -0,030 -1,16 0244
Distance to the nearest tarmac road (km) -0.000 -0.09 0.927
Distance to the district Town/Boma (km) -0.000 —0.82 0.410
Enumeration team dummies includedb Yes
Joint test for individual characteristics (X2)c 1477.83 [0.000]
Joint test for household characteristics (X2)c 20.38 [0.002]
Joint test for conununity variables (x2) 4.99 [0.288]
Joint test for team effects (X2)c 1074.45 [0.000]
fiedicted probability of positive outcomes at f 0.81
_1_\1 umber of prime-age adults 18817

 

Source: CSOfMACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004

NOtes: Absolute z—scores, calculated using heteroskedasticity robust standard errors clustered for
mdividuals. 3 Estimated coefficients are marginal changes in probability. bEnumeration teams are included

but not reported in the table. cJoint test for individual and household characteristics, and enumeration team
effects are significant at 1 percent significance level.

16

 

 

 

 

 

their
the p
1610c
incre
3W3}
11161;

in in

mar
edur
m0:

bet:

rel."

 

 

their parents. Older individuals (age 30 and above) were more likely to be contained in
the panel compared to younger members (age 29 and below) who are more likely to
relocate. Generally, there exists a stronger positive association with reinterview as age
increases. Also, individuals with salary and wage income, and who spend more time
away from home in 2000 were less likely to be contained in both surveys. This is an
indication that individuals with these characteristics are more mobile and less likely to be
in households re-interviewed.

In contrast, individuals who had formal or informal business income were more
likely to be contained in both surveys. Although the coefficients on years of education
are not significant even at the 10 percent level of significance, it would appear that the
marginal probability of remaining in the sample decreases as an individual’s years of
education increases. This suggests that individuals with more years of education are
more likely to be contained only in the first survey and may have moved elsewhere for
better prospects.

Turning to household characteristics, results in Table 2.4 show that individuals in
households experiencing adult death between 1996 and 2000 were less likely to be re-
interviewed compared to individuals in households experiencing no death during the
same period. Landholding size and productive assets are positively associated with
reinterview. Individuals in households with many children were more likely to be re-
interviewed.

The lagged HIV prevalence variable is negatively associated with re-interview
and statistically significant at the 10 percent level. This may suggest that AIDS

exacerbates attrition in standard household surveys. Households suffering from adult

17

 

 

I101

attrl

.9
.J

 

l0...

mortality due to AIDS may have moved away or dissolved, although the lagged HIV
prevalence rate may be picking up the effects of other spatial factors correlated with
district-level attrition rates, such as migration and mobility. Other community
characteristics such as distance of household to the nearest tarmac road or to the district
town appear to reduce the probability of being re-interviewed although this effect is
statistically insignificant at 10 percent. This is because enumerators may be less likely to
attempt to re-visit households in remote or relatively inaccessible locations. The
enumeration team dummies are also jointly significant; suggesting that differences in
enumeration team effort are a strong predictor of re-interview. In any case, the results in
Table 2.4 suggest the importance of controlling for attrition, as is done in the remainder
of the analysis. However, the magnitudes of the results between models corrected for
attrition (Tables 2.6 and 2.7) versus not-corrected for attrition (tables A26 and A27) do
not differ significantly, which suggests that, at least in this particular national sample,

attrition bias does not create major problems for statistical inference.

2.2.2. Estimation strategies and variables

In order to examine the relationship between socioeconomic characteristics and
the probability of PA death, all individuals in households interviewed in 2001 were used,
and it was determined whether they died between 2001 and 2004. Probit regressions
were run for a dichotomous (0/ 1) dependent variable: whether the person died of disease-
related causes equals one, zero otherwise. The base model for the analysis is as follows:

Pr0b(Ait =1) = g (1 iZOOO’X 1120005111 V: C) (2-2)

_j,

18

 

 

uh:

ZETC

inte
era;
heat
1101:
the 1
low:
(set

refe-

div)

fl'ij

21:0

 

 

where A is a binary variable that equals one if individual 1' died between 2001 and 2004,
zero otherwise, 112000 is a set of individual characteristics in 2000, X10000 is a set of
household characteristics in 2000, HIVH- is the lagged district HIV-prevalence rate in
1999, and C is a set of community variables including 393 village dummies. Because
initial 2000 conditions associated with subsequent mortality are being measured, all of
the variables are observable even for individuals in households that were not re-
interviewed in 2004 but contained in 2001 survey. The vector of individual
characteristics include: relationship of the deceased to the person who was household
head in 2000, marital status, age, years of education, and months residing away from
home. Ages are entered as five-year age groups up to the age of 59, with ages 15 to 19 as
the reference group. Years of schooling are also included in dummy variable form for
lower primary (one to three years), upper primary (three to six years), completed primary
(seven years), and secondary and higher schooling (eight years and above), with the
reference group being those with no formal schooling. Months away from home are
divided into three binary variables: 0, 1, and 2 or more months away during the 2000/01
survey season. Individuals who died in 2001 are excluded in computing months away
from home variables because 86 (22% of total prime-age deaths) of those who died in
2001 were at home all the time in 2000, suggesting that these individuals were already
chronically ill and were more likely to be at home throughout the year.

Household characteristics include: landholding size, value of productive assets
(farm equipment and farm animals) and ownership of durable assets (housing quality,

radio, motor vehicles and water source).

19

 

 

inte'
WOI

dis:

 

 

Community variables include: distance of the village from the nearest tarmac
road and district town, whether the district is located on the line of rail (proxies for
degree of interaction between local residents and extent of contact with outsiders passing
through the area).

The inclusion of quadratic terms of landholding size, productive assets, distance
of village from the nearest tarmac road and district town are tested for because their
marginal effect on the probability of being afflicted may be non-linear. However,
specification tests rejected the non-linearity hypotheses in all cases so no quadratic terms
are included in the reported model results.12 Potential regional differences in factors
associated with prime-age adult mortality in terms of wealth and income were also tested
for, but there was very limited evidence of this through specification tests,13 so the pooled
national sample stratified by gender and assets and/or income status is used.

Equation (2.2) is estimated with Probit using the inverse probabilities from the re-
interview model as weights. These models are run separately for prime-aged men and
women, and for individuals in the top versus bottom half of the 2001 income and assets
distribution in order to understand whether the socioeconomic correlates of adult

mortality vary by gender and/or assets/income. As a robustness check to examine the

 

12 In addition to the linear terms, the quadratic terms of landholding size, productive assets, distance of
village from the nearest tarmac road and district town are added to the models and a test of joint
significance is done in order to determine whether to include the quadratic term in addition to the linear
term.

'3 In order to test for regional wealth and/or income differences, I rank household wealth distribution
(value of productive assets-farm equipment and livestock) into terciles then run two sets of models having
two wealth dummies (top and bottom wealth distribution with the middle group as the reference), eight
provincial dummies, interaction terms of wealth distribution dummies and provincial dummies and other
household and community variables in the second model. The joint test of significance of wealth and
provincial dummies was rejected even at 20% level of significance. A similar approach is done with
household income (sum of value of productive assets, gross value of crop output, formal/informal business
income, and non-farm income) but the test for joint significance showed no regional difference by income
status as well.

20

 

 

 

"a
(L;

I-J
'1’)

Illif

fror.

 

rat;

 

 

impact of initial poverty levels on the correlates of prime-age mortality, marginal
probabilities are reported for models stratified by value of assets and household income.
Results by income stratification are reported in Tables A24 and A25. Also, presented in
Appendix Tables A22 and A23 are village- and province-fixed effects models (stratified
by gender but pooled across income and asset groups) with household income and value
of assets dummies as explanatory variables. The results indicate that neither household

income nor assets were statistically related to individuals’ probability of death.

2.3. Results
2.3.1. Adult mortality and HIV prevalence in Zambia

This section begins by investigating the correlation between prime-age mortality
rates from the Zambia panel household survey data and district HIV prevalence rates
from antenatal clinics as reported in Zambia’s Demographic Health Survey (CSO, MoH
and Macro International, 2003).14 A strong relationship between prime-age mortality and
HIV prevalence rates would suggest that a large proportion of prime-age mortality
observed in our household data is indeed due to AIDS-related causes.
Figure 2.1 presents a scatter plot of provincial HIV prevalence and rural adult mortality
rates from our provincially representative household data. The strength of these
correlations is notable, especially considering that the provincial HIV prevalence rate is
not disaggregated by urban/rural classification. The Pearson correlation coefficient of
0.84 suggests that the adult mortality rates observed in our survey data is closely

associated with HIV-prevalence.

 

l4 . . . .

Natlonal estimates of HIV prevalence 1n sub-Saharan Afrlca are almost exclusively based upon surveys
of antenatal clinics, the majority of which are located in urban areas. The Zambia Demographic Health
Survey figures are derived from blood sample testing of a randomly selected national sample of PA adults.

21

Flgl

and

 

QOCNI—CCN emu-Gk hum—3:0:- o—fluufl .383:

N0

SC;

 

 

sur

1

p11 .

i i

Figure 2.1. Correlation between Provincial adult mortality rates from C80 2001

and 2004 household survey data and 2001 HIV + prevalence rates, Zambia.

28

 

26.

24 Lusaka

   
   
   
  

22 . O Copperbelt

20 . 0 Southern

18-

15 ~ Western 0 Eastern

14 -
O Luapula

12.

Rural adult mortality rate, 2001-2004

10 . 0 Northern

0 Northwestern

 

 

 

l I I T l 1 T I l

6 8 10 12 l4 16 18 20 22 24 26
Urban/rural Adult HIV prevalence, 2001

Notes: Pearson correlation coefficient is 0.84.

Sources: Adult mortality rates (age 15-49) derived from the 2001 and 2004 household
surveys. HIV+ prevalence rates are from 2001 Sentinel Surveillance Site information
published by the Ministry of Health and ZDHS respectively.

22

 

 

 

Ail

di:

 

AR

1

 

El. ',

 

 

2.3.2. Descriptive analysis
Table 2.5 presents characteristics of the prime-aged individuals who died in sampled
households. The following features are discernible: First, more women die from prime-
aged disease (and most likely, from AIDS) than men. The first row of Table 2.5 reports
absolute numbers of prime-aged men and women having died compared to individuals
remaining in the sample. Afier weighting the results to the national level, results indicate
that 61% of the illness-related prime-age deaths in Zambia’s small- and medium-scale
farm sector between 2001 and 2004 were women. These results are also consistent with
emerging evidence that a higher proportion of women are dying of AIDS than men in
Southern Africa (UNAIDS, 2003). Women’s mortality rates are expected to be
somewhat higher than men’s in low-income countries even in the absence of HIV
because of matemal-related mortality, but these figures count only illness-related deaths.
An important question is whether this 61% finding is explained by the physiological
differences between men and women’s susceptibility to contracting the disease,15 or
whether it also reflects gender differences in the use of ARV therapy. Because the use of
ARV therapy was known to be extremely low during the survey period (less than 1% of
all HIV -positive individuals), it is likely that physiological difference is the primary
explanation for this finding.

In addition, prime-age female mortality is occurring predominantly among single
women in the younger age groups (see figure A2.1). These results are consistent with the

findings of a five-country study by Mather et al. (2004). Moreover, men and women

 

’5 Because of women’s greater surface area where infected blood can be exchanged during sexual activity,
the risk of HIV transmission from an infected male to a susceptible female is 2-4 times higher than the risk
of HIV transmission from an infected female to a susceptible male (Chin, 2003, drawing from Gray et a1,
2001 based on findings from Rakai, Uganda).

23

 

 

Tab‘

Attri

N113:
N173
lndii
Reid.

(.
0:-
"A

Mei;

 

COT

Table 2.5. Descriptive statistics among prime-age”I adults who died due to illness in

2001-2004, and remaining prime-age adults in the sample

 

 

 

PA adult deaths PA adults
due to illness in remaining in the
Attributes 2001 —2004 sample
Male Female Male Female
Number prime-age adults (unweighted) 165 233 5735 5851
Number prime-age adults (weighted) 17,801 27,730 659,478 677,593
Individual Characteristics in 2000
Relationship of deceased to the HH head in 2000 (%)
Head/spouse 55.5 54.6 61.5 76.4
Others(sons and daughters, uncles etc.) 44.5 45.4 38.5 23.6
Marital Status in 2000 (%)
Single 43.5 55.4 38.3 32.5
Married 56.5 44.6 61.7 67.5
Age (years) 35.6 33.5 31.2 32.7
Completed Schooling (%)
No formal education 7.5 21.9 9.8 22.9
1-3 years 11.6 16.7 10.6 16.1
4-6 years 20.9 22.7 25.4 25.2
7 years and above 60.1 38.7 54.2 35.7
Salaried/wage employment (%) 11.3 3 14.2 4.1
Informal/formal Business activities (%) 21.2 13.2 18.5 13.3
Months spent away from home (median) 0.4 0.5 0.4 0.3
Household Characteristics
Female headed HH in 2000 (%) 17.7 32.6 12.3 22.8
Prior death of adults between 1996-2003 (% of HH)b 68.4 65.5 6.0 5.9
landholding size in 2000 (Hectares) 3.4 2.7 3.0 2.9
Draft animals and equipment (000’ ka) 641.9 588.1 729.2 664.4
Per Capita Household income quartiles in 2000 (%)
Poor (bottom 50%) 47.6 56.1 49.0 50.7
Non-poor (top 50%) 52.4 43.9 51.0 49.3
Community Characteristics in 2000
Distance to the nearest town (km) 31.4 33.1 33.4 34.4
Distance to the nearest tarmac road (km) 23.7 22.9 24.1 24

 

Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004

Notes: aPrime-age is defined as ages 15-59 for both men and women. bRefers to other adults ages 15 to 59
in household who died up to 8 years before the individual under analysis.

24

who died between 2001 and 2004 were somewhat more likely to be better educated

(Table 2.5).

2.3.3. Attributes of Deceased Prime-Age Individuals

This section presents results from probit models estimated to determine the
attributes of deceased prime-age individuals in rural Zambia. Two sets of models were
estimated: provincial- and village-fixed effects. Provincial fixed effects models allow us
to examine the effects of variables measured at the district level, e. g., lagged HIV
prevalence rates, and indicators of market access (distance to the nearest town, distance
to the nearest tarmac road, and district on the line of rail). These models also provide
more accurate estimates of probability of death over the three-year survey interval
because the full sample is utilized.16 All findings pertaining to probability of mortality
are derived from these models. By contrast, the advantage of village-fixed effects models
is that they control for inter-village differences in the attributes of mortality and thus may
provide a more accurate indication of the importance of household-level and individual-
level correlates of mortality within communities. Both sets of results are reported so that
the reader can examine the robustness of these findings.

The probabilities of PA death for men and women were roughly 0.6% and 1.1%,
respectively (see Table A2.1). However, the probability of death for relatively non-poor

men was 0.9% and 0.6% (depending on whether the sample is stratified in terms of 2001

 

16 By contrast, when estimating models with village dummies, the estimation program automatically drops
those households where no within-village variation in the dependent variable exists, which restricts the
sample somewhat. Of the 393 villages in the sample, 118 villages experienced no prime-age disease-
related mortality among their households. Estimating probability of mortality from such models will
generate upwardly biased probabilities because of the many cases dropped of individuals residing in
villages where there were no recorded disease-related deaths over the survey interval.

25

 

 

ill-c;

0.5“

p007

m0.

 

 

ho‘l

resi

I111}

pit

at:
1"

 

income or asset levels), whereas for relatively poor men, the death probabilities were
0.5% and 0.8%. There was little variation in probability of death between poor and non-
poor women.

Table 2.6 and 2.7 presents the results of the province-level and village-level probit
models stratified by assets distribution for some initial individual and household
characteristics in 2001 that greatly affected the probability of dying. For both relatively
poor and non-poor women, being married and/or the head or spouse of the household
significantly reduces the likelihood of death. Among men, the effects of being married
in 2001 on the probability of mortality between 2001 and 2004 are weak but still
negative. In the village-fixed effects models, relatively poor men who are heads of
households are significantly less likely to die than other poor men.

Other variables that affect the probability of death are whether the individual
resided at home throughout the year, whether the household experienced prior prime-age
mortality, and education (for non-poor women mainly). Variables that had little effect or
ambiguous effects on the probability of dying included whether the individual was
engaged in formal or informal business activities, landholding size of the household, and
community indicators of proximity to towns and markets.

To aid in understanding the magnitude of the impact of these variables on death
probabilities, the model results reported in Table 2.6 and Table A2.4 (Columns B, D, F
and H) were used to compute estimated probabilities of dying over the 3-year period for
20 different individual “profiles.” These simulations are reported in Table 2.8 and

discussed in remainder of this paper.

26

 

 

 

Lei —uU.UU-—.—C.v ”Am—033:. w.u.0.:0 COX—t. -.UU—_._>C-.A_.v v1.3.3: £230)? 1:4“ Lazy—.33 >.-£ 3330‘... .CCh C. >~£IIC~C (.1 T.II.:T;:z.|.i:i£.l.4~ii.leli.4N 2. ~
. t 7. In 1 A ! A ..

F

 

 

 

 

 

 

 

$33 3.3 83% G3 833 SW3 83.3 3.13
33 :3 :33 :23 23 33 :33 :33 am-mm om<
3.3 5.3 5.3 3.3 $33 333 3.3 $33
:83 $3 :83 :33 33 83 :33 :83 43m 83
333 § .3 3.3 2 3.3 3.3 3.3 9:3 :3 .3
83 S3 :33 2.8.83 .83 .. _ «.3 :R3 :33 an? 83
3.3 3.3 3.3 3.3 833 3.3 3.3 3.3
sue—36 ammo-o aavvod 1330.0 *hNod MNod .30de aaomod vvtov ow<
$33 3.3 3.3 3.3 :33 2: .3 3.3 893
:33 :83 :83 :333 :33 :23 .33 :33 332 om<
333 3.3 5.3 5.3 . 3.3 3.3 3.3 $33
.33 :23 :23 :33 3.3.3 :33 :23 .23 43% owe.
5.3 3.3 3.3 3.3 83.3 c n. 3 803 3.3
:33 2.38.0 3.3.3 :93 23 23 .33 258.0 own-e. om<
3.3 3.3 633 £33 3.3 3.3 3.3 3.3
B3 83 83 83 83 83 L _3 :23 V38 ow<
Guy coca E 2.3..» om<
£33 $33 $33 233 c 3.3 33 9:3 853
83 83- 83 33 83- 83- 33- 83- :3 8E9: 8.62
3.3 3.3 $33 3.3 24.3 5.3 3.3 833
+33- +23- 83- 83- .23- .: :3- 83- 83. :n3 BEE 38650
3.3 $33 833 633 $33 63 2:3 3. 3
33- :83- .83 83 83- 83- 33- 33- cuv 883333
82 a 83388.25 3.2%.:
3 33 E 3 83 G3 83 93
0?th 2m:
Ae\oOw no.5 coon-do: Homo-d: Afoow 89:23 803 Homm<

 

Aumrfiofio 3 #1133334. own-0&3 commencem—

 

48353

8.3 320880 ”Am—2308 Boo-mo cone-3:383 £38 £333 33 Howaom >9 ”scam-SON E 5:858 <m no .2032: 3905 9N 033-

27

 

$3 :33 38 E8 88 3.8 333 3.8
83- 83- 83 S3 83- 83- S3- 83- :3 mace 2: one 8 8:85
3.3 8.3 3.8 3.8 :33 5.8 5.8 3.8

.83 :83 83 S3 83 383 83- 83- 88 a as 88355 >5 835

833-38239 33.3.5.5

8.3 3.3 5.3 84.3 3.8 3.8 5.8 833

:83- ...83- 83- 83- 83- 53- 83- 83- @2388 one 3838335
63.8 - £33 - €33 - E; 3 -

:_ _ 3 - :83 - :33 - :33 - :13 .88- 88 a 8% <3 53
833 8.8 8.8 5.8 E .8 2 33 8.3 so. 3
83 83 83 83 83- 83- .83- 83- :3 26882 868838

88 5 88:38.20 22.88:

3.3 - 3.3 - 3.3 - 83.8 - :3 88 8 been

:33- - B3- - .33- - 83- - sea 5: anon seed 25
3.8 - 3.3 - 3.3 - 3.3 - :3 88 a so» be

.833- - :33- - :83- - :23- - seaweeds been a 3283
$8 - 3.8 - E8 - 5.3 -
83 - 83- - N83- - 83 - s23 88 a erase aoaoam
:43 - G33 - 3.3 - 3.3 -
83- - ~83 - 83- - .83 - :3 88 a 8:83 omen, .83
3.3 :33 3.3 53 3.3 3.3 5.8 3.3

:83- 83- 83 83 33 :3 83 83 26% as an“; m
84.3 :33 83 3.3 3.8 8.8 :33 8.3
583- 83- :3 33 M83- 83- .33 :23 an; a
8.3 3.3 3.3 3.3 3.8 3.8 8.8 3.8

.583- ...83- .83 :83 83 83 83 83 sea 3
3.3 8.3 E3 E3 33 5.3 3.8 3.8

.383- 383- .83 .333 383 383 83 83 as» m;

:8 88 5 85335 .e 38>

.8338 3 2.3.

28

.82? 308 :05 33 833:3, 383598 :3 wagon.“ .8328 3828 3.386 Bot mac Ho DEQBSQ 22.3 33335 53:: 322335 05

88.35 88» w 8 a: 336 on? 288:0: B an 2 2 8mm $163 550 9 flown-m3 3.02383 38585 ~30? Ho EEO-33 .35— oonaomawmm E383 m 32865
* :96— uunaomawmm 3:8th _ 38868 .3... 32233.30: 38 323333 $030 38953 3358 36333833880: wEm: 3323—3 .muuoom-N 82033.3 ”380 Z

.voom 3:3 Bow Sufism EnoEoEQ=m can oooQamE 535m 3333 ”mom ngOUEOmU H885m

 

 

 

:8 :33 3 8 38 :8 E8 383 383 83333833 33 53632
. . . . . . . . 33828 3838-3833
:3 3 33 3 83 3 83 3 33 3 33 3 83 3 333 3 833 35833 833393 322323
mu? me? we? we? mo> we? we? moxw flown—OE mvgfl 3853,on
3.3 3.3 63.3 33.3 28.3 $8.3 $3.3 5.3
.383 83.3 333.3 333.3 383 38.3 333.3 333.3 353 383333 58% 23 2 83335
5.3 :33 3.3 83.3 3.3 33.3 3 .3 33.3
333.3 333.3 333.3 333.3 38.3 833- 38.3- 333.3- 353 332 38:3 3338 23 2 8535

 

333.538 3.3 233.3

29

 

 

 

 

8m. 8 5.: 3.8 3.8 83.: Saw 3.8 $8
23 23 :33 :§3 :23 :3 2:63 :83 3-3 ow<
6:8 838 and 3.8 83.8 c 38 3.8 838

.33 .33 :33 .Lfid 83- 83- :«30 :33 3.9. of.
3.8 SW8 3.8 and 3.8 6&8 5.8 c ".8
wood :56 .3wm—d 3.336 *LNod *omod Egaofid {*hOMd 91m? ow<
3.8 3.: 5.8 5.8 an: 3.8 3.8 3.8

:23 .. _ N3 :33 1.38.0 83 83 :33 :3 .o 318. a?
3.8 83 3.8 3.8 3.: 3.: 3.8 3.8

:33 :33 :33 £33 .83 83 :33 :a3 32 ow<
and 3.8 8&8 3.8 3.8 838 3.8 Gnu

*VNOd *ONod “gamed signed .IRLQO scum Mod {#506 1.5006 vmém ow<
3.8 3.8 3.8 3.8 5.8 3.8 5.8 5.8

:23 :33 :23 :53 83 83 :23 :33 2-2 a?
5.: 3.8 5.8 628 5.8 8:8 2 08 5.8
83 :83 S3 S3 S3- 83- : S3 :33 E3 0?.

:uv :93 E «98..» uw<

8:8 2 _ .8 328 5.8 3.8 3.8 E .8 3.: .
S3 83- 83 33 83 83- 83- 83- at Brae 552
an: an: 838 3.8 £28 5.: A2.: and

83- 83- S3- 83- 83- S3. 83- 83- say BEE $856
3.: E: 3.8 5.8 8&8 5.8 838 3.:

33- 33- 83- 83- 83- S3- .83- 83. Al uaamasm

82 5 83:39:26 ESE—.5
d8 68 Q no 88 3 J8 3
Bacon 3m:
30% mob 58-82 ax? 8288 88

 

Aomgufio o Juv~16< omméfiflm commooofl

.5553 3..

332.80 A2358 38:0 Bé¢ma=g .253: 5.33 2:. 38.5w 3 38-83 a £338 E .3 .2255 Eta .3 as;

30

 

88.8 88. : 888 2 n8 2 m. : 88.: 8.8 88.8
83 83 83- 83- 83- 888- 83- 888- Cu: 3535 8888 3588
A8,: 8.: 88.8 8.8 8.: 3.8 5.: 88.:
83- 83- :83- :83- 83- 83- 83 83 A3588 2% 852888385
888 - 8.8 - E8: - 28.8 -
:83 - :888 - :8 8 8 - :888 - 2...: D88- 82 a 588 <8 58
88.8 5.8 8.8 38 A88 58 88.8 3.8
83 83 83 83- 83 83 888- 83- en: 2238: 888888
. 88 5 88838.35 2288:
8.8 - A88 - 8.8 - A88 - an: 88 a use:
:83- - 83- - :83- - 83- - 888 8% 80% 888 25
8.8 - :88 - 88.: - 88.8 - an: 88 a so» 28
:888- - :888- - :888- - :888- - 52828 2:2 a 8238
A88 - A8: - Gm: - 88.: -
83 - 83- - 83- - 83 - 9:": 88 a 8&8 ”858m
68.: - 88.: - 3. : - 68.8 -
83- - 83 - 83- - :83 - an: 88 a 2:85 8:, 828
8.8 § .8 68.8 8.8 A8: 8.8 8.8 8.:
:888- :888- 83 83 83 83 83 83 26% 83 £8 8
8.8 3.: 5.8 88.8 8.8 A3.: A88 A88 _
.83- 83- 83 888 83- 888- .88 :83 £8 8
A28 8.: an : A8: 3.8 88.8 28.8 2 8.:
:83- .888- :3 :3 _88- 88- 888 83 £8 3.
88.: 88.: an: 5.: 8.8 8.8 5.: 68.:
88- :83- 83 23 888- 888- 83 88 ”a: 3

an: 88 a 888:3. 8 28>

 

82:82.8 2 2.3.

31

29383 Sun: 2333?: 05

ego-«on 88% w 2 a: :3: on? 223:0: E mm 8 2 moms $168 550 9 flown-m". 805388 88885 EEO-«E Ho Egon; 20>“: oogomawwm Boob: m 8.868

.. :26— ooaaomiwmm Evoke: _ 88868: «20586:. How c2833 895 @8888 7458 365833880: mam: “888328 .mouoow-N 8203888 8802

88.8 85 88 82:8 388288 85 8882 8&8 ES: 288 Emfiouggo 5288

 

VmNN

 

$8 88 82 82 88 2M: 22 8085308 58:52
mo-w 8% 8% 8% 80> 8%. 8% 8.» 3:205 8:556 own—:8»
88: 88.: G3: 988: 88: 83: 88: 888:

83- 888- 83 83 83- 83- 83- 888- 2": aces: :95 858m
6:: 38: 33: G3: 288: 88: 88.: §.:

83 83- 83 83 288 83 888 83 Cures 85 258 8% 8§§m
88: 63: :3: 83: 88.: 8.: G3: 888:

83- 83- 83 888- 83 888 :u: REE: 38 8:88

 

:83 :83

835:8 8 28..-

32

2.3.3.1. Relationship of deceased to head of household

Single and relatively young women are much more likely to die than married women.
Individual profiles 1 and 2 in Table 2.8 are identical in every respect except for their
marital status. According to the probit model results, women fitting the “profile 2”
category have a 1.09% likelihood of dying over a 3-year period, compared to 0.45% for
women fitting the “profile 1” category. Other attributes constant, single women are about
2.4 times more likely to die of disease-related causes than married women. Married men
are also less likely to die from disease-related causes than single men. Comparing profiles
‘ . 12 and 1‘3flin Table 2. §,fisingle men with the specified characteristics are 59% more likely to

die than married men with otherwise similar characteristics.
2.3.3.2. Age groups

Age is one of the more important determinants of death due to illness as shown by the
marginal probabilities in tables 2.6 and 2.7 (where the 15-19 year old age group dummy is
omitted). The marginal probability of dying from disease rises steeply from age 15,
peaking between ages 30 and 39 for poor females, and 45 to 59 years of age for men,
regardless of their poverty status. This finding confirms previous findings showing that
females are more likely to die at an earlier age than their male counterparts. The predicted
probability of dying from diseases for women residing in relatively high-asset households
rises from age 15, peaks between ages 50 and 54, and then declines. However, among
women in relatively poor families, the probability of mortality peaks in the 30-34 year age
range. The predicted probability of disease-related mortality for a relatively poor woman

in the 30-34 year age range is roughly twice as high as a poor woman in the 20-24 year age

33

range. The probability of mortality for a relatively non-poor woman in the 30-34 year age

range is only 1.3 times higher than a non-poor woman in the 20-24 year age range.

2.3.3.3. Education, mobility, income and household wealth indicators

Unlike earlier studies in Sub-Saharan Africa that generally found a positive
correlation between education and HIV-related deaths (e.g., Ainsworth and Semali, 1998,
Hargreaves, eta12002, Gregson, et a1 2001, Glynn, et a1 2004), the results in Table 2.6 and
2.7 show a much weaker relationship between educational attainment and the probability of
mortality from disease. There is no statistically significant relationship for men in
household w1th either low- or hi gh-i—ncome and/or assets. " This appears—to be in contrast to
earlier findings showing that highly educated men were the most likely to die of disease-
related death (Ainsworth and Semali, 1998). Among women, the findings generally
indicate a negative relationship between educational attainment and the probability of
disease-related death, especially for relatively non-poor women. The results for non-poor
women do not show a clear relationship between educational attainment and probability of
dying. These findings are consistent with de Walque (2004), who found that over time,
susceptibility to HIV/AIDS in Uganda declined for relatively well-educated people more so
than for poorly educated people, as information regarding precautionary measures spread.
One apparent implication of this finding is that well educated and (especially) poorly
educated men and women should continue to be targeted for HIV/AIDS education
campaigns.

The results are somewhat consistent with findings in the 19803 and early 19908

indicating that prime-age mortality is more likely to affect men in the upper income

34

brackets. This is shown in Table 2.8 by comparing profiles 13 vs.16, which are identical in
all respects except for assets and/or income. Non-poor men with the attributes shown in
profile 13 are 1.3 times more likely to die of disease-related deaths compared to men in the
bottom half of the assets distribution (non-poor men have a probability of mortality over a
three-year period of 2.09% compared to 1.58% for men in the bottom half of the assets
distribution). Although poverty might be expected to raise the probability of infection of
sexually transmitted diseases and HIV since men with low incomes may be less able to
afford condoms or STD treatment, our findings indicate that the influence of high
economic and social status tends to predominate for men. As shown in table 2.6, women in
the lower and upper half of the asset distribution are equally likely to die of disease-related
causes, with the probability of mortality over the 3-year period being roughly 1.0%
regardless of their households’ income or asset levels. This may vary somewhat according
to age group. For example, comparing profiles 7 vs. 9, women in the bottom half of the
asset distribution (with the particular characteristics specified for these profiles) have a
probability of mortality over a three year period of 1.25% compared to 0.66% for women in
the top half of the assets distribution. If the age group of profiles 7 and 9 is changed from
20-25 to 35-39 and all other characteristics kept the same, the opposite result is obtained
where women in the bottom half of the asset distribution have a probability of mortality
over a three year period of 2.2 % compared to 3.2% for women in the top half of the assets
distribution. This finding highlights the sensitivity of the relationship between poverty and
probability of death to age group.

Women from relatively poor households who have some form of formal/informal

business income are less likely to die of disease-related causes than poor women who did

35

not have any formal/informal business activity (profiles 9 v3.10). This finding seems to
support Epstein (2002, 2003) who contends that female members in poorer household with
few employment opportunities are more likely to engage in riskier sexual activities for
economic reasons exposing themselves to HIV infection. So efforts to provide greater
income-earning opportunities for poor women may make at least a modest contribution to
reducing female PA mortality. However, Epstein’s argument is contradicted by our finding
that women from relatively non-poor households having some formal/informal business
income are 10% more likely to die of disease related causes than women with similar
characteristics not having business income. Non-poor women with businesses are more

* likely to "spend more time away from home and have more social interactiOns than pbor'
women with and without businesses. Other things equal, working women with their own
income sources may be less vulnerable (along the lines of Epstein’s argument), but
working may also involve being outside the village or working away from home more,
which may in turn increase certain risk factors. Recent research demonstrates that relative
economic disadvantage is found to significantly increase the likelihood of a variety of
unsafe sexual behaviors and experiences in KwaZulu-Natal Province, South Africa
(Hallman, 2004). However, the findings from rural Zambia provide mixed evidence,
which calls into question the view that poverty leading to risky behavior is the major
pathway through which the disease is spread, although this may certainly be one of many
pathways. Among rural prime-aged Zambian women, there appears to be no clear
relationship between income and asset levels, access to business income, and probability of

dying.

36

,. 620530: :05 .«o 885%
98 mug: 9:3 mm wows—EEm 3m 5an ES :2: wot-RE £28388 05 mo momentum com 6N 033. 5 £82: :07.me“: no woman 388:5 nouflafima

voom wad doom «Autism Baoaoaaam was oooQaaE hour-Sm Evian “mom namEOUEOmU H385m

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.88 08.2 > 02 0z o as» «M 8-8 85 BE“: 29:8 8
88; O8; 2 oz oz o an“; 8 8-8 85 Bus: 283 e
8.: 08.2 > 02 oz o $8 8 318 85 cans: 222 M:
88¢ 886 2 oz 02 o as; ”N 318 85 88:3 2“: t
8:; 08w.— 2 oz 02 o as: mm 8.8 33 85m 222 2
88.8 868 2 02 a; c was.» 8 as? 85 8:8 222 2
083 888 2 02 we» 8 8% mm as? 85 88 222 E
88; 88.8 2 02 oz o as; 8 218 85 8:5 252 2
886 88.9 2 oz 02 o as; 8 918 85 cog: 0:: 2
$8.». .83. z a; 02 8m 3% 3 8-8 33 85m 28E :
88¢ x; 2 z 9% oz O was: 2 8-8 33 85m 28$ 2
88.. 88.“ 2 oz 02 o 238 3 8-8 33 85m 283 a
8:: 88¢ z we» 02 o v.38 2 8-8 85 8.8 88$ 8
886 88d 2 oz 02 o $8 3 8-8 85 8:5 ego-a \-
886 885 2 oz 8». o as» 3 8-8 8%: 85m 28$ 0
88.0 886 2 02 a; o as; 3 8-8 85 BE“: 228 m
8%; 88d 2 oz 02 o ans 3. 8-8 32 85m 222 v
883 885 2 oz 02 o e8» 3. 8-8 32 corn: 2“: 8
8:8 88; 2 02 02 o 88 o-.. 8-8 32 88 25:8 8
$86 886 2 oz 02 o as.» 3. 8-8 32 cups: 2558 _
05005 muvmm<
88 8.85
coca -33 fiance-non mmofimsm among: 80> Ba @5on 9.on 858m 03on
Ho» m E 3:838 E :38 8553 o «3 080: Bed cognac-”m om< 0885 552 8650 _§E>€E
we a 88695 mo 58¢ Sum hmgom bfimm $3M m5=02

 

 

.muuantta 2232::— ES 12.239: 953% .8 13.3 3:33.: we baa—Sea 05 be .383385 .m.~ 288,—.

37

The results show that irrespective of poverty status males and women living two or more
months away from home per year in 2001/2001 period are more likely to die of disease-
related causes between 2002 and 2004 compared to men and women of the same
characteristics but who spent all the time at home. For example, comparing profiles 14
and 15, non-poor men who spend 2 and more months away from home in 2000/01 have a
probability of mortality over a three year period of 7.8%, whilst males of the same
characteristics who spend all their time at home and did not die in 2001 had a probability
of mortality over the same period of 3.6%. In contrast, comparing profiles 10 and 11,
poor women living 2 and more months away from home are 4 times more likely to die of
disease related deaths than women of the same characteristics but resided at home
through out the year.

Finally, the probit results show that the prior death of at least one adult in the
household over the past 8 years is the single most important factor influencing the
probability that a prime-aged individual will die due to illness.17 Irrespective of
income/assets status, men and women experiencing a prior death of a prime-age person in
their household are 14 to 16 times more likely to die of disease-related causes than the
average prime-age individual. This is shown by comparing Table 2.8 profiles 17 and 18
for men and profiles 19 and 20 for women. The probability that men and women with the
profiles shown in rows 18 and 20 would die over a 3-year period is 12.8%and 12.2%. In
this way, AIDS differs from other kinds of diseases (e.g., malaria), which does not

appreciably raise the likelihood of subsequent death in the family after one member

 

l7 . . .
Respondents 1n the 2001 survey were asked about prior deaths in the household back to 1996, which
respondents in the 2004 survey were asked about deaths experienced in the household since the 2001

survey. A binary variable equaling one if the household experienced a death between 1996-2004 period is
computed.

38

contracts the disease. To the extent that the death of two prime-age members from the
same household within a few years of each other causes extreme hardships on remaining
members, especially for children, the implication of this finding is that special programs
to target and support AIDS-afflicted households are likely to become an important
component of poverty reduction strategies, especially in areas hard-hit by AIDS such as
most of eastern and southern Africa.

Household variables that appeared to be largely unrelated to the probability of an
individual dying from disease include several indicators of rural wealth such as
landholding size and livestock assets. As reported in Table 2.6, indicators of market
access, such as the village’s distance to the nearest tarmac road or district town were
largely unrelated to the probability of an individual dying from disease. This indicates
that the disease has moved far into the interior of rural Zambia, such that proximity to
towns and highways that initially were the main locations where the disease was
transmitted no longer has a significant bearing on the probability of death. District-level
HIV prevalence rates are correlated strongly only with the probability of death among
women in the non-poor groups. This is perhaps not surprising considering that HIV
prevalence rates are derived from blood tests of women (not men) who visit antenatal
clinics in peri-urban and urban areas who are more likely to be non-poor than most

women contained in this sample.

2.4. Summary of findings

This study has identified important ex ante socio-economic conditions of

individuals and households in rural Zambia who die between the ages of 15 and 59 years

39

of disease-related causes, using nationally representative panel data on 18,821 individuals
surveyed in 2001 and 2004 in rural Zambia. The findings of the study can help policy-
makers and development agencies better understand current transmission pathways of
HIV/AIDS, which should help in the formulation of up-to-date AIDS prevention and
mitigation strategies.

Overall, the probability that a prime-aged (i.e., 15-59 year) woman would die of a
disease was roughly 1.0 percent over the 3-year period, while the comparable probability
for men was 0.6 percent. Just over 60% of the prime-age deaths observed in this
nationally-representative rural sample were women, supporting other findings that
women are being disproportionately afflicted by the disease.

Consistent with findings in the 19803 and early 19903, men in the upper half of
the assets distribution are more likely to die of disease-related causes than men residing
in poor households. In contrast, women in the lower half of income/assets distribution
are equally likely to die of disease-related causes as women residing in the upper half of
assets/income distribution. An emerging strand of the social science literature on
HIV/AIDS in Africa stresses the relationship between poverty, risky sexual behavior, and
subsequent contraction of the disease. It has been argued that single women unable to
sustain themselves through wage labor or agriculture are more likely to resort to
transactional sex for survival. If this is an important social pathway contributing to the
spread of the disease in Africa, then one might expect to find a relationship over time
between household- and individual-level indicators of poverty, especially for single
women, and subsequent chronic illness and death. The results also indicate that relatively

poor women who have some form of formal/informal business income are less likely to

40

die of disease-related death than women with same characteristics and no formal/informal
business activity. This finding suggests that efforts to provide greater income-eaming
opportunities for poor women may make at least a modest contribution to reducing
female prime-age mortality. This relationship does not hold, however, for relatively non-
poor females. And 47.2 (45.0) percent of the women dying of disease-related causes over
the 3-year survey period came from households in the top half of the asset (income)
distribution. These findings suggest that the social factors driving the spread of AIDS are
considerably more complex than simply poverty-based explanations, although poverty
may certainly contribute to risky behavior and poor health which are important pathways
by which the'disease is spread.

Single women and men in poor households are twice as likely to die of disease-
related causes as poor women and men who are the heads or spouses of their households.
Single women and men in relatively non poor households are 3.7 and 4.5 times more
likely to suffer a disease-related death compared to married women and men who are the
heads or spouses of their households. Irrespective of gender, individuals who spend
several months or more away from home are 2 to 10 times more likely to die of disease-
related causes in succeeding years than individuals with similar socio-economic attributes
but who reside at home all year. It is possible that the creation of business opportunities
that involve men and women spending more time away from home for extended periods
may exacerbate the AIDS problem in rural Zambia and negate the positive effects of
greater financial independence for women, unless progress is made in public health and
educational campaigns to promote the use of condoms, other forms of safe sex, and

prevention interventions.

41

Years of formal education was found to be largely unrelated to vulnerability to
death for men. For women, the evidence is not robust, but the data tend to show that
educational attainment reduces somewhat women’s vulnerability to disease-related death,
especially for non-poor women. This result may indicate that public health information is
indeed working for the more educated strata of rural Zambian society, because earlier
studies in the region found that HIV rates were much higher for relatively well educated
men and women (Ainsworth and Semali, 1998). This finding suggests that education
coupled with public health campaigns may be an important empowerment tool for
women and may help to reduce the risk of HIV contraction among women. Also,
HIV/AIDS education campaigns should still target both the literate and illiterate because
men of any education level have roughly the same risk of contracting HIV.

Most importantly, the prior death of a prime-aged person in the household
substantially increases the probability of another prime-aged member dying. Irrespective
of income status, prime-aged men and women experiencing a prior death in their
household are 23.0 and 18.1 times more likely to die of disease-related causes than men
and women in households with no prime-age deaths in the past 8 years. The predicted
probability of death was 12.4% and 16.3% for men and women experiencing a prior
disease—related death in their household in the past 8 years versus 0.54% and 0.90% for
men and women not experiencing a prior prime-aged death. Of the 362 households
experiencing prime-age mortality between 2001 and 2004, 15% of them suffered multiple
prime-age deaths. In this way, AIDS differs from other kinds of diseases (e.g., malaria),
which do not appreciably raise the likelihood of subsequent death in the family after one

member contracts the disease. To the extent that the death of two prime-age members

42

fi'om the same household within a few years of each other causes extreme hardships on
remaining members, especially for children, the implication of this finding is that
programs and strategies to support the care and education of orphans and children in
AIDS-afflicted households may need to become a critical component of poverty
reduction strategies in areas hard-hit by AIDS, such as most of eastern and southern
Afiica. More research is necessary to understand the longer-term impacts of the disease
on household behavior and welfare, and to develop programs that can mitigate the
adverse consequences. At this point in time, the research community still knows very
little about the cost-effectiveness of alternative ways of mitigating the impacts of AIDS,
but a solid understanding of the socio-economic factors associated with the disease is
likely to help considerably in designing appropriate risk messages and prevention

strategies.

43

CHAPTER 3

IMPACT OF HIV/AIDS-RELATED MORTALITY ON RURAL FARM
HOUSEHOLDS IN ZAMBIA: IMPLICATIONS FOR POVERTY
REDUCTION STRATEGIES

3.1. Introduction

At this point in time, little is known about whether adult mortality due to AIDS
causes different or more severe shocks to household welfare than mortality of adult
members due to other causes. Most micro-level HIV/AIDS impact studies in the
literature hypothesize about the impacts of the epidemic but rarely quantify them.18
These studies are constrained by the absence of micro-level information on how
households respond to HIV /AIDS and the subsequent impacts on agricultural production,
productive assets, non-farm (off-farm) income and any other key indicators of household
welfare. The behavioral responses factored into macroeconomic-level models of the
impacts of HIV /AIDS on economic growth are largely assumed rather than derived from
micro-level empirical findings. Not surprisingly, their predicted effects on economic
growth and development differ substantially. For example, Cuddington (1993) estimates
that an HIV prevalence of 10 percent implies a reduction in economic growth of less than
one percent. By contrast, Sachs et al. (2001) calculate that the 2.2 million AIDS-related

deaths in 1999 reduced Africa’s gross domestic product growth rate by 35 percent. The

 

l8 .

For example see Haslwrmmer, 1994; FAO, 2003; UNAIDS, 1999; Barnett et al., 1995; Du Guemy, J.
1999; Drinkwater, 1993; Mutangadura, 1999; Topouzis, 2000; Stokes, 2003; SAFAIDS, 1998; Kwaramba,
1997; Pitayanon, et al., 1997; Tibaijuka, 1998; Rugalema, 1998.

44

wide variation in these predictions is exacerbated by the paucity of quantitative micro-
level information on how households respond to HIV/AIDS.

Using comprehensive rural farm household longitudinal data from Zambia, this
paper measures the impacts of prime-age (PA) adult morbidity and mortality on crop
production and cropping patterns, household size, livestock and non-farm income. The
paper adopts and extends the counterfactual (difference-in-difference) approach used by
Yamano and Jayne (2004), one of the few quantitative assessments to date using large-
sarnple household survey data. This study goes beyond existing studies by controlling for
initial (pre-death) household conditions that may influence the severity of the impacts of
adult mortality and by treating mortality as potentially endogenous. In particular, the
study determines the impact of prime-age mortality conditional on initial poverty status,
landholding size, effective dependency ratio, and the gender and position of the deceased
person. Moreover, the possibility that PA death in the household is endogenous is taken
into account by conceptualizing the measurement of effects of prime—age adult death on
rural agricultural households’ welfare as a two stage process: first, by examining the
characteristics of afflicted households; and second, conditional on being afflicted,
determining the effects of morbidity and mortality on indicators of household welfare
both prior to and afier mortality. The findings from this chapter provide important
information that may assist governments, donors, and development planners in
developing specific policies or interventions to mitigate the impacts of the disease on
vulnerable households.

The remainder of the chapter is organized as follows: Section 3.2 briefly reviews

the literature on HIV /AIDS’ impacts on rural household behavior and welfare in Sub-

45

Saharan Africa, and highlights key methodological issues involved in analyzing these
issues. Section 3.3 describes the data and methods used in this chapter. Results and a

summary of findings are presented in Sections 3.4 and 3.5 respectively.

3.2. Review of empirical studies

This section reviews some of the household-level studies measuring the impact of
premature PA mortality in rural farm households in sub-Saharan Africa. First, the
section begins by looking at some of the data and methodological challenges to be
overcome in empirical measurement of impacts of AIDS. Second, the remainder of the
section reviews the available literature on the effects of mortality and morbidity on
household composition, crop production, livestock and off-farm income in deve10ping

countries.

3.2.1. Empirical limitations of prior studies

There are a growing number of studies in Africa attempting to provide micro-
level information on the impacts of HIV/AIDS on rural households and their responses
but there is still modest quantitative information on the effects of HIV/AIDS-related
mortality. Most of these studies are faced with four major limitations. First, the few
available micro-level studies of the effects of HIV/AIDS on rural households are almost
always drawn from specific geographic sites purposively chosen because they were
known to have high HIV infection rates, such as Rakai in Uganda and Kagera in

Tanzania (Barnett and Blaikie, 1992; Barnett et al., 1995; Tibaijuka, 1997; World Bank,

46

1999; Lundberg, Over, and Mujinja, 2000). While providing valuable insights into how
afflicted households respond to the disease, such studies are limited in their ability to
extrapolate to understand national level impacts. The paucity of nationally representative
micro-level information remains a critical limitation on the generation of more reliable
macro-level projections on the effects of HIV /AIDS.

Second, there are only a few longitudinal studies that examine the effects of
disease-related mortality on afflicted households. Cross-sectional surveys cannot
adequately measure the dynamic effects of mortality or control for unobserved
heterogeneity, which are undoubtedly important in this context. Cross-sectional studies
do not allow us to measure effects of mortality on outcomes since there is no information
prior to the death event; such studies only allow us to compare ex post outcomes of
afflicted versus non-afflicted households, although this reveals very little about impacts
of mortality. Furthermore, for studies with no controls, it is unclear if any observed
changes in household welfare for the period before and after death can be attributed to
morbidity and mortality apart from other shocks or initial conditions affecting afflicted
and non-afflicted households alike.

Third, a major difficulty in measuring the impact of adult mortality, especially
mortality attributable to AIDS, is that it is influenced by behavioral choices rather than by
random events. The few longitudinal empirical studies measuring the impact of adult
mortality from AIDS on agriculture and rural farm households’ welfare acknowledge that
the death of prime-age adults, especially mortality attributable to AIDS, may be
endogenous to outcomes but nevertheless treat mortality as exogenous without testing for

endogeneity (e. g. Ainsworth and Dayton, 2000; Beegle, 2003; Booysen, 2003; Yamano

47

and Jayne, 2004;). However, with longitudinal data, the endogeneity issue, while still
important, is not as critical as with cross-sectional data because fixed effects and/or
difference-in difference models can be estimated to control for time-invariant individual
and household characteristics. Nevertheless, time-varying unobserved heterogeneity
remains a major problem even with longitudinal data, which may influence both the
dependent variables of interest as well as the probability of mortality in the household.
Thus, there is the need to test for endogeneity and, if present, explore other methods that
may control for endogeneity of PA mortality due to illness.

Fourth, almost all of the quantitative micro-level studies to date have measured
the effects of mortality in afflicted households compared to non-afflicted households.
Yet, especially in hard-hit areas, if non-afflicted households are likely to be indirectly
affected by the mortality occurring around them, non-afflicted households may not be a
valid control group. This situation, in which a minority of households incurs a shock, but
the shock is spread across households in a community presents methodological
challenges for estimating the full effects of the shock using household survey data. These
four major limitations of existing studies should be kept in mind in the following review
of empirical findings on the impact of premature HIV/AIDS-related death on farm

households.

3.2.2. Effects on household composition and labor availability

The most immediate impact of HIV/AIDS-related illness and death is on the
human capital base, principally in terms of the availability and allocation of labor

(Rugalema, 1999; Topouzis and du Guemy 1999). At the household level, labor input

48

diminishes as a prime-age adult succumbs to protracted illness and the labor of other
household members and extended family members is diverted to care. The death of any
productive member of the household constitutes a permanent loss of labor for agricultural
and off-farm employment and other social and home care activities, although caregiving
labor will no longer be diverted to caring for the patient after his or her death (White and
Robinson, 2000). This labor shock can also cause shocks to the household’s capital
resources as income streams are lost, and as medical and funeral expenses rise. The
death of a male head can also be accompanied by loss of land for the remaining
household members.

Some studies have shown that households experiencing premature PA mortality
due to HIV/AIDS-related causes adopt a mixture of coping strategies including: an
increased in— and out-migration of household members, an increased rate of fostering, and
higher rates of remarriage for surviving spouses (Ntozi, 1997; Urassa et al, 2001).
However, the way in which households adjust to internal labor supply shocks varies
according to the resources of the households. For example, better-off households may be
able to hire workers or attract additional members to at least partially offset the loss of
another (see Yamano and Jayne, 2004; Beegle, 2003; Ainsworth, Ghosh, and Semali,
1995, Donovan et al, 2003, Mather et al., 2005). Yamano and Jayne (2004) show that
households suffering the death of the head-of-household or spouse were largely unable to
replace the labor lost through the death, whereas households suffering the death of
another adult were able to attract new household members. Without such adaptation,
either through altered structures or roles, and/or through external assistance, families and

households may become non-functioning social and productive units and ultimately

49

dissolve (Hosegood et al., 2004). However, in most studies, the vast majority of
households (with two or more members) suffering the death of an adult remain intact and

do not dissolve (Mather et al., 2005).

3.2.3. Effects on agricultural production and cropping patterns

Existing research has presented mixed evidence on the impacts of mortality on
agricultural households and consequently there is no clear consensus on appropriate
programmatic responses, especially considering the opportunity cost of scarce
government and donor resources. Some qualitative studies provide evidence of reduction
in area cultivated, shifts to less labor-intensive crops, and reduced weeding (Barnett et al,
1995 ; Topouzis and du Guemy, 1999; Topouzis, 2000; Harvey, 2004). The view that
emerges here is that mitigation policy should prioritize labor-saving technology.
However, other research in Uganda finds no significant change in agricultural production
induced by labor shortages (Barnett and Blaike, 1992). Also, Barnett et a1. (1995)
conclude from case study research in Uganda, Tanzania, and Zambia that the effects of
adult mortality on rural livelihoods may vary considerably within and across countries
given numerous factors such as the rate of HIV prevalence, population densities, the
nature of the cropping system, and the size of the local labor market.

Beegle (2003), using panel data from Kagera, Tanzania, finds that although cash
cropping is temporarily scaled back following a male death and wage income falls,
afflicted households do not shift towards subsistence crops. Putting these findings in
context, Beegle (2003) also notes that the areas of highest AIDS-related mortality in

Tanzania (such as Kagera) are in the Lake Victoria basin, an area with high population

50

density and, thus, a large labor supply and relatively high labor/ land ratios. However, her
study lacks measures of aggregate output and does not draw conclusions on changes in
total crop production or the composition of crop production. Also, the study evaluated
short-tenn effects during the early years of the epidemic (1991-1993) and the extent to
which these findings hold over the longer run is uncertain.

A study in Kenya by Yamano and Jayne (2004) found that rural households
suffering a PA death between 1997 and 2000 generally experienced a decline in
agricultural output relative to non-afflicted households, but the magnitude and statistical
significance of this finding was a function of the gender, age, and position in the
household of the deceased person as well as the household’s initial level of wealth prior
to incurring the shock. They found no evidence for significant losses in cultivated land
and net crop output among households in the top half of the wealth distribution. These
results suggest that the effects of mortality are highly context-specific and should not be
over-generalized. Once again, however, the three-year time frame over which effects
were measured raises questions about the potential longer-term impacts.

Barnett et a1. (1995) find little evidence of impact of HIV/AIDS in Tanzania but
in Uganda they find some discernible evidence where poor households shift to
subsistence crops over the period 1989-1993. More recent work by Dorward (2003) using
a non-linear programming model and a household typology in Malawi to predict input
and output responses to various shocks, such as price, drought, and adult illness, shows
that responses to adult illness such as reduced area cultivated and outcomes such as lower
yields vary considerably by characteristics of the household, such as percentage loss in

household labor, income and asset levels.

51

Other studies have noted that the recent shift in area cultivated from maize to root
tubers in much of southern Africa may be reflecting labor shortages and small farmers’
attempts to shift to less labor intensive crops (Barnett, 1994; FAO, 1993; FAO, 1995;
FAO, 2004; FASAZ, 2003; Shah, 2002). It is possible that the AIDS epidemic has
contributed to these shifts, but one has to acknowledge that such shifts may also be due to
major changes in agricultural policy, such as market reform programs that eliminated
pan-territorial price supports for maize and also reduced fertilizer subsidies (used
primarily on maize) in much of eastern and southern Africa, resulting in a shift of
household incentives from growing maize to tubers (Jayne et al., 2005). The failure to
take account of such policy changes may result in mis-attributing the shifts in cropping

patterns to AIDS-related causes.

3.2.4. Effects on assets and non-farm income

Farm households are known to rely on remittance and non-farm income as a
primary means to afford assets such as oxen, scotch carts, ploughs, and fertilizer, which
are used to capitalize farm production (Reardon et al., 1995). Unfortunately, such
sources of income are often at risk among AIDS-afflicted households, particularly those
that were already asset poor and vulnerable (Donovan et al., 2003; Mushati et al., 2003).
Morbidity and death of a household member tighten cash constraints on agricultural
production as medical and funeral expenses rise and care giving by other members
further reduces household income-eaming potential. Topouzis and du Guemy (1999)
note that households respond initially by disposal of assets that are reversible, including

liquidating savings, seeking remittances from the extended family and borrowing from

52

 

 

dc‘

tha

311C

DEC
C011

P011

 

 

informal or formal sources of credit. If necessary, the sale or disposal of productive
assets typically follows use of these sources of support but may jeopardize a household’s
future livelihood (Stokes, 2003). This is supported by evidence from Kenya which shows
that households first attempt to dispose of small animals and other assets with the least
impact on long-term production potential. Cattle and productive farm equipment are sold
in response to severe cash requirements after incurring a death in the family (Y arnano and
Jayne, 2004). Such ex post coping strategies are costly in the short term and may cause
households not to recover from impacts of death even in the long-term.

From this review it can be hypothesized that the effects of HIV/AIDS-related
deaths are heterogeneous, that the magnitude and significance of the effects will be
largely conditioned by the gender and household position of the deceased individual, and
that effects may depend on household-specific characteristics, such as initial vulnerability
and poverty prior to the onset of illness and a household’s ability to attract new members.
If this hypothesis receives more empirical support from ongoing studies, then it will be
necessary to move away from generalized conclusions about the main factors
constraining afflicted households’ ability to recover and begin formulating appropriate
policy and programmatic responses based on the specific characteristics of the region, the
regional economy, the localized farming system, the profitability and riskiness of

alternative crops, and available resources.

3.3. Data and methods

This section presents data and methods used in measuring the impact of PA

mortality on household composition, land cultivated, crop production, livestock assets

53

 

 

h:

C0
St

“I

ref:
infc
retr
bas
den.

SKI

 

and off-farm income. First, the section begins by discussing the data and attrition issues
associated with any longitudinal survey data. Second, the conceptual framework from
which the hypotheses are generated is discussed. Third, the remainder of the section

presents the estimation models and identification issues.

3.3.1. Data

The data comes from nationally representative longitudinal data on 5,420
households in 393 standard enumeration areas (SEAS)19 in Zambia surveyed in May 2001
and May 2004. The survey was carried out by the Central Statistical Office (C80) in
conjunction with the Ministry of Agriculture and Cooperatives (MACO) and Michigan
State University’s Food Security Research Project. The 1999/2000 nationally
representative Post Harvest Survey (PHS), which surveyed about 7,500 households, was
the base for the Supplemental Survey (SS) of May/June 2001. The SS covered the same
reference period as the PHS of 1999/00 crop and marketing year, but collected additional
information on non-farm income, adult and child mortality information including
retrospective questions on mortality in the household over the previous five years, and
basic socio-economic information on all individuals listed in the 1999/00 PHS
demographic roster. Because of missing information on some households, the valid
sample was reduced to 6,922 households. A follow-up survey of the same 6,922
households surveyed in SS 2001 was revisited in May/June 2004 and a total of 5,420

households were re-interviewed. Enumerators revisiting these households asked for the

 

l9 . . . . . . .
“Standard enumeration areas” (SEAS) are the lowest geographic sampling umt 1n the Central Statistical

Office’s sampling framework for its annual Post Harvest Surveys. Each SEA contains roughly 15 to 20
rural households.

54

whereabouts of the members included in the demographic roster of the initial survey, and
recorded cases of death and illness, departure, and new arrival of individual members.
The 1999/00 PHS sampling frame was based on information and cartographic
data from the 1990 Zambia Census of Population and Households. The census
questionnaire included a question on whether the household engaged in agricultural
activities (crop growing, livestock and poultry raising, and fish farming), as well as check
items to identify the specific crops grown and animals raised by the household.
Households were included in the sample only if they were found to cultivate crops or
raise livestock. The reason for excluding the non-agricultural households was to improve
the efficiency of the sampling frame for crop and livestock production and other
agricultural characteristics.20
Zambia is divided into nine provinces, which are further divided into 70 districts
(see map of Zambia in appendix, Figure A3.1). For the Census enumeration, a
cartographic operation was conducted to define census supervisory areas (CSAs), which
were further divided into standard enumeration areas (SEAS).2' A stratified three-stage
sample design was used. The CSAs were primary sampling units selected with
probability proportional to size (PPS) at the first stage, where the measure of size was
based on the total number of households in the CSA. At the second sampling stage, one

SEA was selected with PPS within each sample CSA. This resulted in a similar

dispersion of the sample and probabilities of selection as if the SEAS had been selected

 

20
Although the rural households of landless farm laborers and those engaged in other economic activities

are of analytical interest, they are best studied through other surveys, such as the Living Conditions
Monitoring Survey (Megill, 2004).

The SEA is the smallest area with well defined boundaries identified on census sketch maps and each
SEA was covered by an individual enumerator for the census data collection.

55

 

 

511,"

PH

fan

m0

3.3

 

directly at the first sampling stage. Within each selected SEA, all households were listed
and stratified by size for selecting the sample households at the last sampling stage.
Households were classified into small- and medium-scale farming households,
defined as those cultivating areas less than 5 hectares and between 5 and 20 hectares,
respectively. Households cultivating more than 20 hectares were classified as large-scale
farmers and were not included in this survey. Initial village listings of all households
were generated to prepare the sample frames. The percentage of households who
engaged in neither crop nor animal production on their land was found to be low, less
than 4%. Landlessness is somewhat higher in areas closer to towns, where a higher
proportion of households are engaged exclusively in non-farm activities. Since smaller
households vastly outnumber the larger ones, the survey over-sampled the medium-scale
farming households in order to ensure adequate inclusion of the larger households in the
survey. A weighting procedure was formulated in order for the sample estimates from the
PHS and SS surveys to be representative of the population of small to medium scale
farmers. These sampling weights were multiplied with sample descriptive estimates. For

more details about survey design and sampling procedures see Megill (2004).

3.3.2. Attrition

In a study such as this, which addresses the impacts of mortality using
longitudinal data, treatment of attrition is particularly important. Adult mortality may
contribute to household dissolution, which contributes to attrition and may lead to under-

reporting of mortality rates and biased measurement of the socio-economic impacts of

56

 

.2: -l: .. l . ,I. JilIIIi
«\CCA. _ .Ub:..>C._L >9. u>~£f£LCZ~ A<Ly .4...T.ZZ....L;C UL: if

38 5.53...-....,._..:._:x .:..... i... .. .
. P.

 

Iin.l.llII

.. « ..........,..

 

 

 

Amoco: ~38»— vow5a5 85 82588 x2985 580: BEBE—ow
.558 5&5: mo .0063 .585 5:8.» 5:08 .3 855.8: Ema .58 Sufi mafia: .5808 28 55 808 5.“ nwzoo 805555 nwi 588 one 5.5— 5 85
3808 a .58 808 5m 5558 08°85 58% mo venom team a 8 Emmy? 363 no °\oo~ 58 558m .52 Ew53v 8me 588 N ”coumogmma—o 0:30 63.50%
05 no 8% £88me .3— mafia media: 8 58.5w 8 55¢ .«o 0880“. .v 855m 8 538m 25 88w 58:85 053 85 .85 838 85 53:8 $5208

5 53m 538 55358. .3 588 553 8856 5500 A2525“ :5 :5 080: 85 :558888: .5 wontowano 053 o\o~.m 85 #52086 8 5888 053
$56 .<mm .«o :5 5.68 $2: $838 053 o\o~.o 4530358868 «0: .XRAN 05 Mo; 80803 85 58 Son .8.“ 3-2 5mm .5 menace .04 05-0859 ”562

38 Ba Son .535 388285 9... 88.32 35m .35.: mom 5883258 seam

 

 

 

 

 

 

 

3.3.3 58 S. cm 5 :m 42 55 8% 2% .28.
ASE .m 9.5 MN . S 3. w. 9.5 mwm £5 Essa
3.5 a. 2.8 mm o a. x mm 2.3 o8 NS 8058.0.
2.5 o 3.3 m - mm s 2 6.on EM 3. 8882.82
8.9V S 3.3 S. 4 mm 3 a. 8.3 RE .82 8852
8.5 3 ads h _ R a w 3.3 .2 EN 8.83
29¢ 3 A35 2 4 mm. a E 5.5 as A: 2:83
8.3 2 93 mm a 3 K 8 3.33 8: :2 835
ad. 2 2.3 a M Ma 2 N. cams N: am .3880
$.va S 2...: 2 v 8 4m «m 3.9: E v: .280
Afov 5388 gov 5888 52888 538:: 598:8 59888 3.95385: 59852
A5 A3 3 E 3 3 3w 3
8me 588
“Mme“? 0:238:83 on? E 0350 ”£85 282 2a: 88
5.88 080 808%: 8 a -wwfiwwwwmdm 8 53035:: 88>on
agave 33.8mm? 38:55 Snow 5 8:5 owmwwmmwmrw “WNW—Mm“. 53:35:

 

33:88: 532828-“: 23> Nme 8 £38 o>umto5a

 

doom 85 Bow 5535p m58mN :88 .8838 .3 «N5858 2% omeéEtmE coco—«>05 ._.m 2an

57

mortality. The impacts of adult mortality would most likely be different for households
that dissolve as opposed to those that remain intact.

Table 3.1 presents basic information on the households surveyed, re-interview
rates, and prevalence of disease—related mortality over the 2001—2004 period. Of the
5,420 households successfully re-interviewed, 571 households had at least one prime-age
death in the sample, of which 547 of these households had at least one disease-related
prime-age (PA) death over the three-year period, 30 households had prime-age deaths
due to accidents or homicide, and 6 households had deaths due to both causes. Of the
5,420 households that were re-interviewed in 2004, 78 households did not appear to be
the same households interviewed in 2001 so are excluded from this analysis. Of the
remaining 5,342 households, 542 households incurred at least one prime-age disease-
related death, 52 (9.6%) of them suffered multiple prime-age deaths, with 44 households
experiencing 2 deaths, 6 households experiencing 3 deaths and 2 households
experiencing 4 prime-age deaths. Of those households experiencing multiple prime-age
deaths, 15 households experienced more than one male death and 16 households had
more than one female death. Out of a total of 571 prime-age deaths, 211 (37.0%) prime-
age individuals joined the household after the 2001 survey and died between 2001 and
2004. This is evidence that a high proportion of HIV-positive individuals returned to
their rural families to receive terminal care after becoming ill.22

Longitudinal data often provide an understanding of the dynamic behavior of

individual households not possible with cross-sectional data. However, one major

detracting feature of panel data surveys is that they are almost always affected by some

 

22 . . . . . . . . .
Other studies have found that a high proportion of HIV-posmve indiViduals returned to their rural
families to receive terminal care after becoming ill (e.g., Kitange et al., 1996).

58

 

 

 

 

 

0

TC‘

ho.

fev

 

 

level of attrition over time. Attrition is a result of a number of factors such as changes in
population (6. g. dissolutions due to death or relocation), non-contact (qualified
respondents away from home), death and/or relocation, non-contact (qualified
respondents away from home), non-response and refusals. Non-random sample attrition
may create selection biases.

As discussed in Chapter 2, the 2004 sample was affected by attrition. Of the
6,922 households interviewed in 2001, 5,420 (78.3%) were re-interviewed in May 2004.
Therefore, these data only measure the effects of mortality and illness on households that
remained intact over the 3-year survey interval.

To test for possible bias in results due to household attrition, the mean levels of
control variables measured in May 2001 are compared for households that were re-
interviewed versus those that attrited. The means of many variables differ statistically
between re-interviewed and attrited households (Table 2.3). For example, households not
re-interviewed had slightly younger household heads (43 years vs. 45 years), smaller
household sizes with fewer children age 5 and below, fewer boys and girls age 6 to 14,
fewer prime-age male and female and elderly males, slightly smaller landholdings, less
farm equipment and animals, and slightly higher rates of chronically ill adults in 2001.
This is not surprising given the data presented in Chapter 2, Table 2.1 showing that
attriting households were smaller to start with in 2001. Systematic differences between
attritors and non-attritors, coupled with a high attrition rate, may cause concern about
inference with this data. Also, if the attrited households suffered a higher incidence of

PA mortality between 2001 and 2004, there would be attrition bias when estimating the

59

 

 

 

 

't)

he

304

m

 

 

ex ante socioeconomic characteristics of individuals who died of AIDS-related causes.23

So one should be worried about the possibility of systematic attrition leading to selection
bias. In order to deal with potential attrition bias, the inverse prob ability weighting

method is adopted as discussed in detail in section 3.3.7.1.

3.3.3. Relationship between Adult mortality and HIV/AIDS

While not all disease-related mortality can be attributed to AIDS in any given country or
region, recent epidemiological studies demonstrate that in Eastern and Southern Africa,
HIV has become the leading cause of disease-related death among adults between 15 and
59 years of age (Ainsworth and Semali, 1998; UNAIDS/WHO, 1998; Ngom and Clark,
2003; World Bank, 1999). Given the difficulty and cost of obtaining reliable estimates of
AIDS-related mortality, some studies have used a combination of serological surveys to
track the HIV status of sampled adults over time (Urassa etal., 2001); and/or “verbal
autopsies/lay diagnosis” in which relatives and caregivers of the deceased are interviewed
to record information regarding signs and symptoms of the terminal illness, all of which
help to reduce the probability of incorrect diagnosis (Garenne et al., 2000; Urassa et al.,
2001; Doctor and Weinreb, 2003; Araya et. al., 2004). The diagnostic indicators of the
lay diagnoses are then used to provide estimates of the share of AIDS-attributable
mortality based on an “expert algorithm”.

For example, used the World Health Organization (WHO) algorithm to estimate

the share of AIDS-related deaths and this estimate with the

 

Available evidence on attrition rates in longitudinal surveys in developing countries range from 5 to 30

percent for 2 rounds (see Alderman, et al, 2001; Yamano and Jayne, 2004). For a discussion of IPW see
Wooldridge, 2002.

60

expected number. The expected number was derived from a comparison of adult
mortality in this sample with the pre-AIDS mortality levels measured in the 1987
Malawi census, giving an ‘excess mortality factor’ that can be considered to be AIDS
related. It was calculated that 74.9% of observed deaths in 1998—2001 would be from
AIDS and according to the WHO algorithm, 75.5% of the VA deaths was found to be
AIDS-related deaths. 24 Araya et. al., 2004 used two “gold standards’ to validate the lay
diagnosis information: the hospital discharge diagnosis of causes of death obtained by
surveillance of hospital deaths (including autopsy results) and the physician review of the
VAs. Independent of the ‘gold standard’, they find that the lay diagonis such as lung
disease and cold had a specificity of about 90% and a combined sensitivity of about 55%
in determining AIDS mortality and the sensitivity increased to 60-65% when diarrhea,
TB, herpes zozer and mental or nerve problem are included.

These studies concluded that even in the presence of a reluctance to talk to
HIV/AIDS, lay diagnosis of causes of death can be used for monitoring AIDS mortality.
However, an identification of ‘AIDS’ deaths’ through VAs based diagnosis lacks
sensitivity to the extent that certain illnesses will be missed and specificity to the extent
that any non-HIV tuberculosis or cancer will also fit the criteria (Doctor and Weinreb,
2003). Also, a review of literature on verbal autopsies and lay diagnoses shows that there
is no generally accepted ideal method of estimating AIDS-specific mortality in a
Zambian population-based sample. Therefore it was not possible to get a " gold standard"

diagnosis on a true population basis, since the validation of verbal autopsy studies in the

 

24 WHO classification: 2 major signs (weight loss greater than 10% of body weight in a short period of
time, chronic diarrhea for more than a month) and at least one minor sign (persistent cough for more than
one month, itching skin rash, fungal infection of mouth and/or throat, history of herpes zoster, generalized
herpes simplex infection and enlarged lymph nodes).

61

literature is flawed (the validation samples come from clinical samples and therefore are
not likely to be representative of the population) (Gretchen Birbeck, personal
communication).25

In the survey, we collected information from respondents about symptoms leading
to death of the deceased in an effort to try to explore the potential differences between
defining the death as adult mortality due to illness in general and defining it as prime-age
mortality when the cause is predicted as AIDS-related. If the reason for cause of death
was given as ‘disease,’ follow up questions on the symptoms leading to death (lay.
diagnosis) were asked for each adult who died between 2001 and 2003. The lay
diagnosis (LD) questions were included to aid in developing an algorithm for prime-age
mortality where the cause is predicted as AIDS-related. The major clinical symptoms
for which data were collected are chronic diarrhea, prolonged fever (intermittent and
constant), and sudden weight loss of more than 10% of body weight. The minor signs
were prolonged cough, prolonged difficulty in breathing, prolonged pneumonia, and
thrush in the mouth and ‘rash’, which is considered to be an indicator of generalized
pruritic dermatitis if it occurred in combination with two major signs.

Using the World Health Organization (WHO) standard algorithm for diagnosis of
HIV infection in the absence of blood tests, 28.4% of the disease-related deaths in our
sample are estimated to be AIDS-related (columns g and h , Table 3.1). However,
because questions were asked about only 3 of the 6 “minor symptoms” as specified by
WHO, it is likely that our classification of AIDS and non-AIDS deaths underestimates

the percentage of deaths related to AIDS. Therefore, as in chapter 2, the decision was

 

25 . . .
Gretchen Birbeck IS a professor in the department of Neurology & Epidemiology at Michigan State
University.

62

 

 

 

 

 

 

made to confine the analysis to the impacts of prime-age mortality due to disease in
general.

Table 3.2 summarizes the number of afflicted and non afflicted households.
Among the afflicted households, death is disaggregated by age, gender and position of
the deceased in the household. The majority of the deaths due to illness (63.4 %) are as a
result of mortality of other non-core male and female household members (non-core

referring to members other than the household head and spouse).

3.3.4. Conceptual framework

It is widely accepted that HIV/AIDS will affect many aspects of the rural economy in the
hardest-hit countries of Africa, although the particular pathways, impacts and magnitudes
remain unclear and continue to be debated. Using the household as the unit of analysis”,
one can try to understand the mechanism by which rural farm households are afflicted by
AIDS-related illness and death by developing a framework that views the household as a
system of interactions between the household supply of labor to off-farm activities and

farm production and the retum flow to the household of food and cash to sustain their

 

26 The paper uses “household” as the basic unit of analysis. However, there is no universal definition of a
household so the definition of a household adopted in the survey and in this study follows the concept of a
rural farm household by the United Nations: "T he concept of household is based on the arrangements
made by persons, individually or in groups, for providing themselves with food or other essentials for
living. A household may be either (a) a one-person household, that is to say, a person who makes provision
for his or her own food or other essentials for living without combining with any other person to form part
of a multi-person household, or (b) a multi-person household, that is to say, a group of two or more
persons living together who make common provision for food or other essentials for living. The persons in
the group may pool their incomes and may, to a greater or lesser extent, have a common budget; they may

be related or unrelated persons or constitute a combination of persons both related and unrelated " (UN,
1998)

63

Table 3.2. Characteristics of non-afflicted and afflicted' households

 

 

 

Category Poor Non Poor Total
(A) (B) (C)
------------ number -~-~-------~

Non Afflicted (no PA death or chronic illness) 2122 2085 4207
Household with chronically ill PA adults and no deaths’ 300 280 580
Prime-age mortality (ages 15 to 59)

Male heads 53 38 91

Female Heads or spouse 61 61 122

Other males 75 92 167

Other females 76 126 202
Elderly mortality (ages 60 and above)

Elderly male 60 68 128

Elderly female 41 49 90
Number of householdsa 2,675 2,667 5,342

 

Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004

Notes: aAfflicted households are those in which a person between the ages of 15-59 (i.e., “prime-aged” or
PA individuals) died or has been chronically ill during the period 2001-2004. The number of households is
the sum of households suffering a death (excluding elderly deaths’), households with chronically ill PA
members but no death, and non-afflicted households less the number of households incurring more than
one death between 2001 and 2004. There are 27 households with more than one prime-age death of which
12 are in the poor category and 15 in the non-poor category.

64

daily livelihoods (F A0, 1995). Inter-household dynamics, and the complex links
between individual households and extended families, both within and between
communities are also very important since they tend to interact with household
production and consumption. Also, households’ allocation of resources are based on a
number of interacting factors, such as household characteristics; biophysical factors,
tastes and preferences, institutional, regional community characteristics and economic
factors, which may need to be taken into account when measuring the impacts of
HIV/AIDS-related morbidity and mortality.

As shown in Figure 3.1, the most immediate impact of HIV/AIDS-related illness
and death is on the human capital base, principally in terms of the availability and
allocation of labor (Gohen, 1993; HSRC, 2001a; Rugalema, 1999a; Topouzis and du
Guemy 1999). This may disrupt the flow of labor to farm production and off-farm
activities. Thus, at household level, labor input diminishes as the chronically ill prime-
age adult succumbs to illness, and the labor of other household members and extended
family members is diverted to care. The ultimate death of any productive member of the
household constitutes a permanent loss of one source of labor for agricultural and non-
agricultural activities and potential income, although at the same time resources will no
longer be diverted to caring for the patient (White and Robinson, 2000).

Second, expenditures for medical treatment and funeral costs represent foregone
investment in agricultural production and education. In the event of prime-age adult
mortality, family assets such as livestock and machinery, may be sold off to compensate

for foregone income, thereby delaying the impact of death to a later stage (Barnett and

BI aikie, 1992; Lunderberg, Over, and Munjinga 2000). For food insecure households or

65

those slightly above this threshold, the loss of labor, income and increased expenditures
for medical care can push them further into poverty and food insecurity. AIDS-related
death and illness are likely to progressively de-capitalize highly afflicted rural households
and communities, resulting into loss of savings, livestock assets and drafi equipment, and
other assets. The loss of productive assets that would otherwise be used in production
processes may raise the demand for labor to the extent that labor can substitute for

' capital. However, earlier studies point to the fact that this may not be a problem if
afflicted households can meet the rising demand for labor by attracting family labor from
other sectors.

The way in which households adjust to internal labor supply shocks varies
according to the resources of the households. For example, better off households may be
able to hire additional workers to fill the void left by the deceased. Some afflicted
households are able to attract additional members to at least partially offset the loss of
another member (Y amano and Jayne 2004, Beegle, 2003, Ainsworth, Ghosh, and Semali,
1995, Donovan et a1, 2003, Mather et al., 2005). Households that are initially poor are
faced with very hard challenges to adjust to both human and financial changes. Figure 3.1
summarizes the different coping strategies typically suggested in the literature.

If suitable instruments existed, one would be able to empirically distinguish the
impacts of one factor from the other. Nevertheless, this conceptual framework provides
some insights into the potential pathways by which adult morbidity and mortality may
affect rural households’ livelihoods. From this framework this paper tested the following

hypotheses:

66

First, the death of prime-age adults reduces the households’ supply of labor and
adversely changes its dependency ratio.

Second, to the extent that some households are able to attract new or previously
resident family members from elsewhere after a prime-age adult becomes chronically ill
or dies, these households may incur smaller adverse impacts on production levels, assets,
and other indicators of welfare than households that cannot attract additional adult
members.

Third, a priori, the direction of the changes on household size are ambiguous
since they are likely to be dependent on the gender, age, status of the deceased and the
capacity of the household to attract new members or the extent to which the surviving
household members leave the household to join other families or go and live elsewhere in
search of non-farm opportunities 27

Fourth, if households are experiencing income and labor constraints due to loss of
prime-age adults to illness and death, one might expect to find some evidence that
households are coping by shifting to cultivation of less labor-intensive crops such cassava
and sweet potatoes.

Fifth, if the death of prime-age adults results in a loss of income and shortage of
labor then one would expect to see a decline in the value of crop output and the value of
crop output per hectare. However, the decline in output is expected to be sensitive to the
gender and status of the deceased. In particular: a) cultivation of food crop production
declines more if a female dies than if a male dies because females are more likely to

devote most of their time to food crop production; b) cash crop production (coffee,

 

27 .
(See Ainsworth, and Semali (1995); Janjaroen, (1998); Yamano and Jayne (2004)).

67

$236568 {653. .82 83m use eopam ”32 05 H5% 3&3.

 

 

952 853.555
65:8 2:: £3385 .zozoa EuEEo>om £36 3 $38 .228.“ 35:336.:— 0 m .32 v5 v5. .«o $36.38 3:32 .59: £8.88 ”£88m 259.com.
833% .8553 H 380$ 30.93.. 2m 0 “55%—96v uggmacfi :333 woo—BE -"mocmtouoabno .3395“

82—5. «.58.: E82: .350

 

 

 

# fl

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2:85 i....................
conga—so so ............ . 053266 E a ...... .H u
n 623 es; s a u u m
" Sea 8882. - - - confine we r t
mi" can nob E ._, 83:. Eta 38:23 E A, 380 E25 cog—5:58
. u s . . 0 mm
H t - 308 38:85 W .862: E e m 04
u -32 9 cfiml 9 J.
n J ..... .— 22.5299... 25.... . mars—25¢. 25.2.3 a
u u t ) H m
23:3 “Evan on xomm £333 25% a. *5 N
ha 88 2 3%: mo c0936 finqomumbom SEE 32.: no.“ 28 8 none— uo c2336 . u .
conu— o>uosvoa mo 33 £32 3.526th .«0 $3
) i
u 333 omabEE £9: BB5 28.5w "
.. ....................... .omsoamB8: . . ................... L
”we fifivhmoci
UHEmQZ/wm meafiam

 

 

 

238:0: 55 :23 38%? 528.88 98 bmflfiofi owmbatn mat/Hm 533 an mhmkfimm fizaouom .mm ocswE

68

tobacco, tea, and other export oriented horticultural crops) declines more if a man dies
than if a woman dies because it is likely that cash crop production is the responsibility of
the adult males in the household . c) if (b) is found to be true then it is expected that the
death of male prime age adult members result in a higher decline of gross value of crop
production compared to the death of a female member.

Sixth, if households cope with PA death by liquidating their assets, one would
expect to see a decline in livestock assets, in particular cattle and small animals.
Households may liquidate their assets starting with small animals (goats, sheep), which
have the least impact on long-term productivity. However, in response to severe cash
requirements households sell off larger animals such as cattle.

Seventh, the death of prime-age adults results in a loss of non-farm income, in
particular remittances and wage income. If however, non-farm opportunities are
available, the death of prime-age adults may encourage other household members to
move off the farm to seek better paying jobs or activities if they have the necessary

required skills. Therefore, a priori the direction of the impact of AIDS on non-farm

income and remittances is ambiguous.

3.3.5. Econometric model

To measure the impacts of PA mortality and morbidity on outcome Y,, we
consider the estimation of a panel model that contains a binary variable for PA death as

an eXplanatory variable. The following base model is formulated:

Ya = y,+t*D,5+a,+g, i=1,...,N t=l,...,T [3.1]

69

where Y“ denotes an outcome, such as household composition, area under cultivation,
value of farm output, and non-farm income for household 1' at time t; D, = 1 if a
household experienced prime-age death between 2001 and 2004 and 0 otherwise; the

parameter 7, denotes a time-varying intercept”; or, captures the household-level fixed

effects (assumed constant over time); and 8,, is an error term.

A comparison of the changes in outcomes (Y) over time between the treatment
group (households incurring a prime-age death and/or chronic illness) and the control
group (households not incurring prime-age chronic illness or death) provides an estimate
of the impact of prime-age mortality. Differencing the time 1 and time 0, equation 3.1

yields:

AY, = 7 + Did + A8,. i=1,..., N [3.2]

where AY, is the 2004 - 2001 difference for a given outcome measure for each household
1' , Di is the treatment indicator, 6 is the treatment effect, 7 is a constant, and A8,. is the

difference between errors at time 1 and time 0.

Estimation of equation 3.2 by OLS gives the average treatment (6 ) which is
essentially the impact of prime-age death on outcome Y. Assuming that neither initial
household conditions nor attributes of the deceased person affect 5, and nothing else
Changes between afflicted and non-afflicted households, one could use this simple

difference-in-difference estimator to evaluate the impact of death.

 

28 .
Wooldridge, 2002 page 254.

70

However, rural households are heterogeneous in many variables that change and
evolve differently for different households such as effective dependency ratios, the stock
of education, incomes and assets levels. There is growing evidence that the effects of
prime-age death differ between households depending on their initial conditions in terms
of assets, income and stock of education (see Yamano and Jayne, 2004; Ainsworth and
Dayton, 2000; Beegle, 2003; Yamano and Jayne, 2005). To control for these
heterogeneous factors, a vector of exogenous household initial covariates (X,) are

introduced into equation 3.1 as follows:

Yr: =7,+t*Di6+t*X-0go+ai+si, i=1,..,Nt=l,...,T [3.3]

l

Differencing the time 1 and time 0, equation 3.3 yields:

AY,=7+D,5+X{’¢+A8, i=1,..., N [3.4]
However, in order to analyze the differential impacts of PA mortality, these initial
(pre-death) characteristics were interacted with the treatment (D). The estimated
treatment effect remains 5 but it is now interpretable as a ceteris paribus effect. The

model in equation 3.4 could then be re-expressed as:

AY,=7+D,6+Xi°(p+X,°*Di77+A8, [3.5]

71

3.3.6. Empirical model and estimation strategy

Very little is known about the dynamics of household behavioral response to
premature PA adult mortality in Africa and evidence to date shows great heterogeneity.
Using the conceptual framework presented in section 3.3.4, I adopt and extend the model
and estimation methods of Yamano and Jayne (2004) who chose to use methods that did
not put a lot of restrictions on the data. As an extension of their study, this paper
estimates impacts of mortality on various household outcomes taking into account the
initial (pre-death) household variables as well as testing for the likely endogeneity of
death variables before choosing the estimation method. Using equation 3.5 and adding
provincial dummy variables (P), and interaction terms of deaths between 2001 and 2004

and pre-death household characteristics (X°*D) the following model is estimated:

AYi =7+Di§+Xi0go+Xi0*D,-n+Pg+Aei [3.6]

Outcome variables (AYi) : The changes in outcome household level variables were

grouped into four categories: (a) household composition; (b) agricultural production and
cropping patterns; (c) value of livestock assets; and (d) non-farm income. Household

composition variables included: changes in household size and composition of men,

72

women, boys and girls aged 11 and under”; agricultural production and cropping patterns
variables: changes in total land cultivated, land cultivated by crop category (cereals,
roots and tubers crops, high value crops) and gross value of crop production per hectare;
value of livestock assets included: change in value of small and large animals, and non-
farm income variables included off-farm income.

Death variables (DJ: D, is a vector of deaths occurring in households between
2001 and 2004. Because the impact of prime-age mortality may differ depending on the
gender of the deceased member, D, was specified as two categorical variables, one for
households suffering the death of a prime—age male (DM) and one for households
incurring the death of a prime-age female (DF). I further differentiate the impacts of
mortality according to the position of the deceased in the household by introducing
categorical variables for male household head death (DMH), female heads/spouses death
(DFH), other prime-age male death (DMO) , and other PA females death (DFO). This

specification enables us to test for the possible status-differentiated effects of adult death.

Household pre-death conditions (X19) : X 0 is a vector of initial household

conditions. This vector is comprised of asset poverty status, land holding size, and
households’ effective dependency ratio in 2000.30 Effective dependency ratios,
following de Waal (2003), are defined as the number of children, elderly, and chronically
ill prime-aged adults divided by the number of healthy prime-aged adults. These

variables were interacted with mortality variables to test whether the impacts of PA

 

29 Due to data limitations the number of children could not be disaggregated into boys and girls age 6 to 11
and children under the age of 5. There is evidence in the literature that suggests that households cope with
the loss of prime-age adults by either taking young boys and girls out of school to replenish the pool of
pgysical labor or sending them away to live with relatives to ease the burden on food security.

See Rosenzweig ( 1998) for a discussion of supposition that households’ composition responds to the
economic environment facing households.

73

mortality are sensitive to households’ initial (2000) poverty levels, landholding size, and
effective dependency ratios. Ideally, these initial conditions should be measured prior to
the onset of chronic illness, but due to data limitation the onset of illness cannot be
precisely determined.

The model specified in equation 3.6 is static in the sense that it measures a
snapshot of afflicted households’ changes in welfare outcomes in 2003/04 compared to
pre-mortality conditions in 2000/01. Because households were re-visited only once, the
analysis is not able to capture any dynamics associated with an adult death and outcomes
for the surviving household members. Household responses to a death may not be
constrained to one-time adjustments. Moreover, in the case of HIV/AIDS, the lag
between infection and death (typically one year or less) may result in ex ante adjustments
by the household already having been reflected in the first survey in 2001 in cases where
an adult became chronically ill before this time. We considered the inclusion of a
dummy variable that controls for current prime-age chronic illness in the household but
decided against this because current chronic illness is likely to be endogenous given the
fact that individuals to some extent may select which households to die in. As mentioned
earlier, many individuals return to the homes of their parents to receive terminal care
(Kitange et al., 1996). Moreover, chronic illness is believed to be measured with
considerable error because of its subjective and self-reported nature (see Strauss et. a1,
1993, Beegle, 2003).

Province x time dummies (A) .° Although the difference-in-difference estimator

presented in this chapter controls for unobserved time-invariant household characteristics,

there may be area-specific time-variant effects that might be corrected with both the

74

prime-age death and the outcome. To control for such area-specific time-variant effects,
Provincial x time interaction dummies were added to the estimation models.

It must also be noted that the control group may be tainted by the fact that in areas
where the epidemic is more widespread, no household in the community may remain
unaffected. The increasing deaths and illness due to HIV /AIDS may for example result
in a breakdown of social capital and local institutions that affect the whole community
(afflicted and non-afflicted households alike). With a tainted control group our results
may be biased. We address this concern by including lagged provincial HIV /AIDS
prevalence rates in the model as a way to control for the extent of disease in an area and

the probability of individuals contracting the disease.

3.3.7. Econometric issues

The model discussed above is faced with two econometric issues, namely the
likely endogeneity of death variables and attrition bias. Ignoring these issues may result
in inconsistent and biased results. Therefore, this paper attempted to address these issues

simultaneously as discussed below.

3. 3. 7. 1. Attrition bias

As mentioned earlier, the longitudinal data used in this study suffers from an
attrition rate of approximately 19%. If this attrition occurs randomly, then there is no
reason to worry about selection bias due to attrition, although efficiency will be lost

because of a reduced sample. It is possible that the incidence of prime-age mortality is

75

higher among households that attrited but there is no way to determine this. If attrition
bias is non-random then it is imperative to control for such attrition bias. Comparison of
mean characteristics in 2001 (Table 2.3 in Chapter 2) seem to suggest systematic
differences between attritors and non-attritors. Coupled with high attrition rate, this may
cause concern about inference with this data. Also, if the attrited households suffered a
higher incidence of PA mortality between 2001 and 2004, there would be attrition bias
when estimating the impact of premature adult HIV /AIDS-related mortality.

The literature addressing the correction of selection bias is extensive, and a
complete review of this literature is beyond the scope of this paper.31 Similar to Chapter
2, the inverse probability weighting method (IPW) is used to deal with the potential

attrition bias. The re-interview model is specified as follows

Pr0b(Rit z 1) = f(HIVt-j’Xi.20009Eit) (3-7)

R, is one if a household (i) is re-interviewed at time t, conditional on being
interviewed in the previous survey, and zero otherwise; HIV 1., is the district HIV-
prevalence rate at the nearest surveillance site in 1995; XiZOOO is a set of household
characteristics in the 2001 survey including landholding, productive assets, demographic
characteristics (number of children ages 5 and under, number of prime age males and
females), and ownership of various assets and E" is a set of 59 enumeration teams. All
of the variables in (3.5) are observable even for households that were not re-interviewed

in 2004. Equation (3.7) is estimated with Probit for attrition between the 2001 and 2004

 

3‘ For an overview of sample selection see Fitzgerald, Gottschalk, and Moffit (1998), and Alderman et a1.
(2001).

76

surveys, obtaining predicted probabilities (Przoor). Then, the inverse probability (1/Pr2001)
is computed, and then applied to the models estimated in section 3.4.

The results from the re-interview model (Table 3.3) show that households headed
by older adults were 0.4 percent more likely to be re-interviewed compared to households
headed by younger adults. Households with more adult males, adult females, and
children were more likely to be re-interviewed. Thus, larger households were less likely
to have dissolved and also it was more likely for an enumerator to find a qualifying
respondent in larger households during the second survey.

Households who experienced an adult death between 1996 and 2000 were less
likely to be re-interviewed compared to households experiencing no death during the
same period. Also, landholding size and production assets are positively associated with
reinterview. The lagged HIV prevalence variable is negatively associated with re-
interview and statistically significant at the 10 percent level of significance. This
suggests that AIDS exacerbates attrition in standard household surveys. Households
suffering from adult mortality due to AIDS may have moved away or dissolved, although
the lagged HIV prevalence rate may be picking up the effects of other spatial factors
correlated with district-level attrition rates, such as migration and mobility.

Households located in a district that is on the line of rail were on average 5
percent less likely to be re-interviewed compared to those households not on the line of
rail. Districts along the line of rail also contain most of the main roads linking Zambia to
other southern African countries. Households in these districts may be considered
generally more mobile and integrated with the main thoroughfares by which HIV has

been spread. Mobility is also considered to reduce the likelihood of re—interview. Other

77

Table 3.3. Household-level re-interview model (Probit‘)

 

l=Households contained in 2001 and

 

 

Covariates 2004 Surveys, O=contained only in 2001
dy/dx Z p>z
Demographic characteristics in 2000
Female headed (=1) -0.01989 -1.42 0.154
Age of household head (years) 0.00426 1.92 0.055
Age of household head squared (years) 000003 -1.44 0.15
Mean years of education of head and spouse (years) -0.00257 -1.65 0.099
Number of male adults 0.01643 2.76 0.006
Number of female adults 0.01844 3.03 0.002
Number of children under age 6 years 0.01625 2.61 0.009
Number of children age 6-11 0.00713 2.05 0.041
Prime-age adult mortality and illness in 1996-2000
Death of head/spouse in 1996-2000 (=1) -0.06482 -1.87 0,062
Death of non head/spouse on 1996—2000 (=1) -0_01047 -0, 5 5 0.581
Household assets in 2000
In (Value of assets (ka)) 0.00271 2.54 0.011
In (Landholding size(I-Ia)) 0.02859 4.77 0.000
Community variables
District HIV prevalence rate in 1999 -0.00058 -1.66 0.097
Distance to the nearest tarmac (Km) -0.00027 -l.l3 0.258
Distance to the nearest district town (Km) -0.00006 -0.23 0.817
On line of rail (=1) -0.04973 -2.30 0.021
Enumeration team dummies includedb Yes Yes Yes

 

Joint tests (x2)
Household characteristics
Enumeration Team effects

158.77 [p=0.000]
209.26 [p=0.000]

 

‘ Community variables 15.66[p=0.029]
Predicted probability of re-interviewc 0.795
Number of households 6922

 

Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004

Notes: aEstimated coefficients are marginal changes in probability. Absolute z-scores, calculated using

heteroskedasticity robust standard errors clustered for households.

78

Enumeration teams are included but
not reported in the table. cPredicted probability of re-interview is estimated at model mean values

community characteristics such as distance of the household to the nearest tarmac road or
to the district town appear to reduce the probability of being re-interviewed, although this
effect is statistically insignificant at 10 percent. This may be because enumerators were
less likely to attempt to re-visit households in remote or relatively inaccessible locations.
The enumeration team dummies are jointly significant, suggesting that differences in
enumeration team effort could be a strong predictor of re-interview. Also, the 2000
household characteristics are jointly significant as determinants of re-interview. In any
case, the results in Table 3.3 suggest the importance of controlling for attrition, as is done

in section 3.4.

3.3.7.2. Identification of impact of death

The DID fixed effects estimator of equation 3.6 is confounded by the possibility
that PA death variables are endogenous, hence OLS results may be biased. There is
growing evidence that households afflicted by prime-age mortality are not randomly
distributed, for they tend to display certain features with respect to initial income, asset
levels, education, etc. (see Ainsworth and Semali, 1998; Ainsworth and Dayton, 2000;
Yamano and Jayne, 2004; Beegle, 2003). Unfortunately, these studies did not
empirically test for the likely endogeneity of prime-age mortality.

Beegle (2003) outlines some of the reasons why AIDS-related prime-age death
could be viewed as endogenous. First, AIDS-related mortality is caused by behavioral
choices rather than random events. Contraction of HIV /AIDS is an endogenous

occurrence, resulting from distinct patterns of behavior, particularly with respect to

79

sexual activity, and perhaps influenced by economic and social conditions. Second,
individuals have some scope to choose where to live once they become ill. For example,
seriously ill individuals may move into a household seeking terminal care and would like
to die and be buried in their home area. Since these individuals are selecting which
households to die in, the death variable(s) are not likely to be independent of the
disturbance term in the household outcomes of interest. Over one third (3 6%) of the
people who died between 2001 and 2004 in our sample were persons who moved back
into the household and died before the second survey. So there is a possibility that the
decision to return to a household for terminal care is related to the receiving household’s
or the returning individual’s economic circumstances.

In the early years of the epidemic in Sub-Saharan Africa, evidence suggests that
men and women with higher education and income were more likely to contract HIV than
others because they were more likely to have numerous sexual partners (Ainsworth and
Semali, 1998; Gregson, Waddell, and Chandiwana, 2001).32 Using the same data set,
results in Chapter 2 on the characteristics of individuals who died of disease related
deaths between 2001 and 2004 in Zambia exhibited the following characteristics: (1)
61% of the prime-age deaths observed in the nationally-representative rural sample were
women; (2) single women and men are 2 to 5 times more likely to die of disease-related
causes than women and men who are the heads or spouses of their households; (3)
females are more likely to die at an earlier age than their male counterparts; (4) relatively

wealthy men (defined according to initial household assets and income) are 1.4 to 1.8

 

32 . . . . . . .
As information about HIV transmrssron spreads, however, it rs believed that educated people are more

likely to change their behavior in ways that reduce their vulnerability to the disease compared to less
educated people.

80

times more likely to die than relatively poor men; (5) relatively wealthy and poor women
are equally likely to die of disease-related causes; (6) among relatively poor women,
those having some form of formal or informal business income are 15% less likely to die
of disease-related causes than those without any form of business income; (7) by contrast,
among relatively non-poor women, those with business income were 7% more likely to
die than those without business income; (8) irrespective of income status, prime-aged
men and women experiencing a death in their household in the past 6 years are 23.0 and
18.1 times more likely to die of disease-related causes than men and women in
households with no prime-age deaths in the past 6 years; and (9) men and women living
two or more months away from home per year are 2 to 10 times more likely to die than
men and women living at home throughout the year. These characteristics seem to
buttress the argument against treating premature adult death in the household as a random
event. If prime-age mortality remains correlated with individual and household
characteristics such as social status, education, and wealth — which are also important
determinants of incomes and other welfare indicators — failure to control for these
characteristics may generate biased estimates of the impact of adult mortality on
household welfare.

The DID fixed effects methods employed by Yamano and Jayne (2004) may
result in biased estimates if the death variables remain endogenous even after controlling
for time-invariant individual and household characteristics. Several methods have been
proposed in the literature to deal with endogenous dummy variables when estimating
treatment effects models including: "Heckrnan-type" selection models (Goldberger 1972,

based on Heckman’s [1976] sample selection model) in which a selection equation and

81

an outcome equation are jointly estimated, assuming a bivariate normal error term in the
two equations; and instrumental variables (IV) estimators and nonparametric matching
methods, most prominently propensity score matching (Rosenbaum and Rubin 1983), in
which the probability of each unit selecting treatment is first estimated, and control
observations are chosen by matching this score to the treatment observations. However,
all of these methods are dependent on the availability of instruments to identify the
impact of death.

Using the two-step IV method that exploits the binary nature of the endogenous
explanatory variable(s), endogeneity of PA mortality variables by gender position in the
household is tested for at two levels. First, the two years of data are pooled and tested to
determine whether mortality is endogenous.33 Second, the data are tested again to see if
death variables are still endogenous after controlling for time-invariant unobservable
effects by differencing the outcome variables. Evidence of endogeneity at this stage
would warrant the use of instrumental variable fixed effects; otherwise, OLS on the
difference models will be sufficient.

The challenge is finding instruments with some explanatory power in
distinguishing between afflicted and non-afflicted households, and not directly correlated
with the welfare indicators of interest. Time-invariant variables from ex ante survey data
(e.g. , distance from the household to a main road or distance to health facilities) could

potentially be used, but their usefulness in distinguishing between afflicted and non-

 

33 . . . . . . .
F rr‘st, the method mvolves estimating the probit model P(D=l |x,z)=G(x,z; y) and obtaining fitted

probabilities . Second, equation 5 is estimated by IV using instruments 1, G, and x.. The method has a

unique robustness property because using G.

I as an instrument for D,, the model for P(D=l|x,z), does not

have to be correctly specified and identification is achieved off the non-linearity of P(D=1 Ix). However,
@(yOtx 71) and x are usually highly correlated which may result in imprecise estimators (see Wooldridge,
2002, pages 621-625).

82

afflicted households may be limited. On the other hand, a variable such as educational
attainment of the most highly educated person in the household may have some
explanatory power in distinguishing between afflicted and non-afflicted households.
However, education is also likely to be directly correlated with income, so the direct link
between education and income is likely to distort or bias the effect of adult mortality on
income if education is indeed correlated with income as one might imagine. Moreover, it
is educational attainment prior to the onset of illness that would be appropriate; this
variable might change after the death of an adult if that adult was the most highly
educated person in the household. In other words, education level after the death of a
household member is likely to be endogenous. At worst one could have used the non-
linearity of the first stage regression to identify the impact of death but our results would
be less convincing in the absence of plausible exclusion restrictions.

This study considers the use of rainfall shocks as a proxy for migration in and out
of the community, lagged district HIV prevalence rates, and prior death in household as
likely instrurrrents.34 Below is a discussion about the possible pathways in which these
instruments are linked to PA death.

Logged HIV/Prevalence: An investigation of the correlation between prime-age
mortality rates from our household survey data and district HIV prevalence rates from
antenatal clinics as reported in Zambia’s Demographic Health Survey (CSO, MoH and
Macro International, 2003) show a strong relationship between prime-age mortality and

HIV prevalence rates making HIV prevalence a possible instrument (see Chapter 2,

 

34 . . . . . . .
An instrumental variable must satisfy two requrrements: it must be correlated With the included
endogenous variable(s), and orthogonal to the error process.

83

Figure 2.1).35 The Pearson correlation coefficient of 0.84 suggested that provincial-level
adult mortality rates observed in our survey data are closely associated with HIV -
prevalence rates. However, the use of lagged HIV/AIDS prevalence as an instrument
could be problematic in the sense that prevalence rates based on sentinel site data may be
biased upward, but this problem would less severe to the extent that the upward bias is
uniform across all regions. However, if some differential bias existed, then lagged
HIV/AIDS prevalence would not be a good instrument because the variable may also be
correlated with the outcome variables. For example, women’s use of the clinics where
HIV testing is performed may be correlated with income levels, thus high income pe0ple
are more likely to use these facilities than poor people. Despite the high correlation
between HIV prevalence and prime-age mortality, HIV prevalence failed to pass the
overidentification test, suggesting that the variable is also correlated with the outcome
variables. Therefore, as suggested earlier, HIV prevalence is included as a control in
both the first and second stage models.

Prime-age deaths in 1996-2000: In the 2001 survey, respondents were asked
about prior mortality of household members during the 1996-2000 period. Results in
Chapter 2, showed that individuals in households experiencing deaths between 1996-
2000 were more likely to die of disease-related causes in the 2001-2004 period, making
prior PA death good proxy for future deaths. Thus, the correlation between prior deaths’
and current mortality makes prior death during 1996-2000 period a good instrument for
PA mortality due to illness in 2001-2004. However, it is also possible that mortality in

the 1996-2000 period may be related to the dependent variables in the 2001-2004 period.

 

35 . . .

National estimates of HIV prevalence in sub-Saharan Africa are almost exclusively based upon surveys
of antenatal clinics, the majority of which are located in urban areas. The Zambia Demographic Health
Survey figures are derived from blood sample testing of a randomly selected national sample of PA adults.

84

Therefore, prior PA death is not used as an instrument in pooled models because it is
likely to be correlated with the unobserved household characteristics. In contrast, as a
robustness check of the endogeneity tests in differenced model, prior PA death is
included as an instrument in the first model (model 1) and excluded in the second (model
2). The results in Table 3.5 show that we still reach the same conclusion in both cases.
Rainfall shocks: A long history of variation in rainfall may be correlated with
earlier migration in and out of the community. For example, a drought tends to induce
people to leave the rural areas and their families to seek income in the cities and send
food home. (See Thiam et a1. 2002). Once away from the family, some may rely on
risky livelihood strategies that expose them to HIV infection and on their return to the
community may pass the disease to unsuspecting partners. On the other hand, during
good years, rural areas may attract traders from towns and cities and farmers may travel
to markets and spend some time away from their families and communities. Such
activity results in increased interaction with the outside community increasing their

susceptibility to the spread of the disease.36

Also, some studies have suggested that rural
people may be infected with AIDS because of the interaction of drought and poverty,
thus poor people (especially young girls) with no other survival alternatives may be
forced into transactional sex in order to survive thereby exposing themselves to HIV.

For example, a study by Bryceson et al. (2005) of smallholder farmers in three rural

villages in Malawi's Lilongwe district revealed that hunger was a greater contributing

 

36 .
A study in Senegal found that 27 percent of the men who had previously traveled in other African

countries and 11.3 percent of spouses of men who had migrated were infected with HIV. In neighboring

villages where men had not migrated less than one percent of the people were HIV positive (see Thiam et
al., 2003).

85

 

fit

as

nfigr
ofsu
thep

inter:

With

 

 

 

factor to increasing susceptibility to HIV/AIDS, as these communities were engaging in
risky sexual practices to survive.

Since deaths that occurred between 2001 and 2004 are being dealt with, it is more
likely that people dying of HIV/AIDS-related causes might have contracted the disease
five to eight years ago. Therefore, this study uses annual rainfall shocks in the 1994/95
drought season (crop season rainfall in 1994/95 minus mean rainfall over the 10-year
period from 1990/ 1991 to 1999/2000) as a proxy for migration. This particular year is
used because it was a severe drought year in the country and most likely induced
significant migration that could affect mortality with a 7-10 year time lag, taking into
account the mean period between HIV infection and death. However, not everyone
migrates to other areas in search of other opportunities to help out their families in time
of such hardship. It is likely that gender and age influences who migrates. To improve
the predictive power of the instruments, age group of the deceased and rainfall shock
interaction variables are computed. In particular, eight variables are computed to interact
with deviations in rainfall from the 1994/95 drought season from the 10 year mean with
eight age groups 20-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54 and 55-59. The results
were not sensitive to which drought year was selected, 1991/92 or 1994/95. Due to high
multicollinearity between the deviations in rainfall by year only deviations from the

1994/95 drought year are used.

86

 

1'6

515

3.4

Wit

forr

5130

b

 

3.4. Results

This section begins with a discussion of the results from the first stage regression
models, where the suitability and strength of the instruments is tested followed by a
discussion of the results from the Hausman Wu Chi-square test for endogeneity of prime-
age death variables by gender and position in the household for pooled OLS and DID
regression models. Then, having determined whether the prime—age death variables are
still endogenous or not after purging the time-variant unobservable effects by taking
differences, the remainder of the section presents the results fiom appropriate models

measuring the impacts of PA death on rural household variables.

3.4.1. First-stage regression models

An instrumental variable must satisfy two requirements: it must be correlated
with the included endogenous variable(s) and orthogonal to the error process. The
former condition is tested by F-test of the joint significance of the instruments in the first-
stage regression. The inverse of the F-statistic is proportional to the bias in the second
stage (Duncan and Strauss, 1997). Table 3.4 reports the first-stage F-statistics for the
significance of the identifying instruments. In all cases, the joint F -test for prior death
and the 1994 drought age-group shocks are highly significant. Surprisingly, prior prime-
age death is a significant predictor of other males and females mortality. In spite of these
fairly large F-statistics and high percentage correctly predicted, a good deal of
unexplained heterogeneity remains, as indicated by the low pseudo R2. Also, in one case,

changes in area under hi gh-value crops, the null hypothesis that our instruments are not

87

 

con
inter
pres

are:

3.3.1

prod

 

 

correlated with the error term is rejected. So the test for endogeneity should be
interpreted with these shortcomings in mind. Tables A3.8 to A3.14 in the appendix
present the first- and second-stage regression results from which the results in Table 3.4

are derived.

3.3.2. Is prime-age death endogenous?

As discussed in section 3.2.1 there are reasons to believe that PA mortality from
disease-related causes is likely to be endogenous and OLS estimates are biased in such
instances. Table 3.5 columns A to F, summarizes the results from the Hausman-Wu test
for endogeneity and Sargan N*R-squared test for overidentification of exclusion
restrictions. A complete set of the first and second stage results are presented in tables
A3.6 to A3.14 in the appendix. The results in Table 3.5, column A show that prime-age
death is endogenous when OLS and IV results for the pooled sample are compared.
Thus, in all the cases except when measuring the impact of death on gross value of crop
production and area under roots and tubers, the null hypothesis that all the prime-age
gender and position mortality variables are exogenous is rejected at the 1-5% level of
significance. This finding implies that any attempt to measure impacts of prime-age
death on rural household welfare with pooled cross-sectional data would yield biased
estimates because of the unobserved effects, which are correlated with the error term.
Taking advantage of the availability of panel data, the time-invariant unobserved
household characteristics are differenced out as shown in equation 3.5 and further tested
for endogeneity of prime-age death variables. Any evidence of endogeneity at this stage

would indicate that even after differencing out the time-invariant unobserved

88

 

10.9....Uunn k—uUULLOU c\: 3:1.

 

, . . ii. Jew r: 3:. . - - O p ....~-
NM- CUSUmL .mainu—quivél WZTAkmzhnvtmko DOC-wucwlwmw hoiw Otwcnuv L 0. . um. ~ ..~ V r ~ ~

 

 

 

 

.mofimtg x027. an? 98 95cm owe 28 58v Btu Son .33 £258 aofimuawuc own; “we Sod £32 “worms

.805ng nu 8 Snow SEQ 5055 £258 :23“:on uwsm “we Bot 33%: 584$ .o\o_ “a “anomamwm .1. ”fem Hm ammo—mama _.. ”$3 “a unmowawfi + 5802

voom was SON Sufism ~8noEoEn—sm can oooQaae NAeta—m «magi «mom :mEOUEOmU H850m

 

 

 

 

 

 

 

 

 

 

 

A??? 6.83.8 332.8 :33 ....L.~$ .3 10:3 :mnflm see Ease 550
3.358 GENE 2.0%.; 3.9%“. :33 :62 :23 1.26% sec 8:8 550
8.9%.; 33:8 3.3an :32 :33 v3 :23 a; :8 ese messages; 295m
cesowm 2.3:.mm A? :33 :32 3.98m So :26; int: sec BB 0:2
g Q m: m: R: Q a: g 5332 “@385
r. o o
p N 382 t 382 DN £5: a _ e 2 38%
860nm 862% 95% own
85ng 9.on own £95 .3on own 98 285:6“.
295m =< 23 Maggie SE and 8255c granny.“
038mm voououuwfin Moo—com Edged =£5~M
2362 @858me £95m moo—com
mm 253% 23

BEBE 388.80 .x. 355:me How «worm

 

33695 buooboo c\o 23 NM 23on .858ng grating mo oozeowEwfi 8m oumufimi omfiwemhm .vd 03.8.

89

Ta

lieu
Hr

Fe
Bc
6
Crap
(in
(in

 

Asset

 

 

 

Table 3.5. Summary table of Hausman Wu Chi-square test and Sargan N*R square
test for overidentiflcation for pooled and differenced samples.

 

Sargan N*R square test of

 

 

 

 

 

Hausman Wu Chi-square testa overidentificationb
Pooled First differenced Pooled First differenced
Sample Model 1 Model 2 Sample Model 1 Model 2
(A) (B)c (C)d (D) (E)° (FF

Land cultivated

Total area cultivated 22.26" 0.82 0.38 4.24 2.72 2.85

Area under cereals 18.32“ 2.09 1.89 7.26 3.36 3.97

Area under tubers 6.17 2.64 3.06 5 .21 6.82 7.25

".66986966.68.13381966986 ....... 1.95.1.1: - - - "Hi-fl} .......... 5. :5_l----------§'.‘.5. ........ $831“? ...... 53 3735...

Household demographics

Household size 29.26" 1.80 0.65 5.01 7.57 3.16

Males 22.69" 7.19 2.37 8.91 5.02 7.67

Females 22.87M 6.04 6.88 5.85 8.15 8-72

Boys 29.26" 1.80 0.65 ' 5.01 7.57 3.16

Guls 14 56" 5 85 6.09 4 76 8 69 8 71
Crop production

Gross value of output 5.26 1.32 1'23 7.14 3.71 3'96

Gross value of output/ha 9.46+ 2.05 5'51 4.28 2.14 3’22
Assets and ofl-farm income

Values ofcattle 20.11" 1.49 0.77 5.12 6.72 7.85

Values of small animals 1040* 4.22 2.99 8.90 7.59 8.12

Off.farm income 25.01" 7.53 8.61 6.99 9.01 9.18

 

Source: Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004

Notes: 1Tests of endogeneity of prime—age mortality (male heads, female heads/spouse, other males and
other females): Ho: Regressors are exogenous. Tests of overidentifying restrictions: Ho: All instruments
are exogenous. + significant at 10%; * significant at 5%; ** significant at 1%. cResults from models

without prior death as an instrument. dResults from models with both prior death and age group and
rainfall shock variables.

90

 

 

01
fir

l'fl

The
p08?
than

1881

 

characteristics there still remains time-varying unobserved household characteristics
correlated with the error term, which would require us to consider the use of instrumental
variable DID fixed effects estimation. However, the results in Table 3.5, column B and
C, indicate that differencing of the household time-invariant unobservable characteristics
adequately addresses the endogeneity problem, since the null hypothesis that all prime-
age mortality variables are exogenous is not rejected. These findings offer some support
for the validity of earlier studies using fixed effects, RE or DID (but which did not
explicitly test for endogeneity). Since this is the first study (to my knowledge) that
attempts to test for endogeneity of prime-age mortality when measuring household
outcomes, there is need for further empirical evidence on this issue. Given these
findings, results fiom fixed effects models using differenced data are presented in the

remaining sections.

3.4.3. Impact of prime-age death on household composition

The results in table 3.6 (columns A and B) show that irrespective of gender and/or
position of the deceased person in the household, a prime age death is associated by less
than a one-person decline in household size. For example, the death of any prime-age
male reduces the size of the household by 0.84 members and death of a PA woman
reduces the size of the household by 0.69 members. Compared to non-afflicted
households, the death of male head of the household results in a reduction of household
size by 0.71 person and the death of females who were heads or spouses of their
households reduces household size by 0.81 person. The death of another adult male

reduces the household size by 0.88 persons whilst death of other females reduces the

91

 

 

SU

ma

500

(Hill

 

household size by 0.56. This finding indicates that households respond to mortality by
partially replenishing their household size.

Changes in household size, shown in column A and B, are the sum of changes in
men, women, boys and girls, as shown in columns C to J. As mentioned earlier, due to
data limitations it was not possible to split boys and girls into two age groups, age 5 and
under and age 6 to 11, so I only look at effects of death on boys and girls ages 11 and
under. Looking across the row, it can be seen that the reduction of household size due to
male heads/spouses death is mainly caused by a reduction of 0.55 in the number of adult
males. The change in number of females is positive whilst that for girls and boys are
negative but with the exception of change in number of girls, the estimates are not
statistically significantly different from zero. The drop in adult males is less than one,
suggesting partial replacement of males. In contrast, the death of PA female heads or
spouses and other females reduces the size of the household by 0.81 and 0.56 respectively
and this reduction is due to the changes in the number of females. Similar to the effect of
male heads/spouses death, the reduction in household size is less than unity suggesting
some partial replacement of household members. The results are slightly different for
other non-core male death. The reduction in household size due to other male death is a
result of a decline in the number of adult males by 0.42 as well as a reduction in the
number of girls by 0.34. This additional decline in number of girls explains why the
reduction in household size is close to unity. This finding seems to suggest that
households incurring other male death are coping by sending away small girls to live with

other relatives.

92

 
   

 

5.1-1-: :4 :C.....v...:::c.u 32:98.22: :3 ..:....H..:.:I<LLC .meQZ: 5:8 1.0.» 27:: -

 

 

.886 Ragga “manomeHEBdonn—E 58> 882228
more; can 88:38am 5 83:52 .oocwomiwmm mo 32: $3 + was oofiomawfi mo 323 gm ... .ooawomimmm 80 $32 ex; on." 80qu

voom 28 SON Sufism gage—mnsm v5 ooomama eh35m 53.82 “mom :mEOUEOmU Hooh—om

 

 

 

 

 

8mm mom mom mom 8% 8mm 8% 8mm 8mm 8% 888888 do L88:2
:.o :8 8 .o 8 .o 88 8.8 :.o :.o :8 :.o 8888-8
.83 .83. a; $2 :58 18.8 :32 2.2.8 ta: 18.: £852 8 888
me? my? mu> no? moNr woof mooxo mo> mo? woof mvgv only x noggin
$8 828 2 we and 5.8 88.8 A38 A88 A88 2o. 8
1.2.? 1.2.? 1.8.? 2.8.? :28 :25 :2 .o :2 .o 88 3o 8888
an: 5.: 84.8 :08 A38 A38 8.: §.: 8.: Ge:
8.? a? 8.? 8.? :Ne? .88.? 2 .o S .c in? in? 828:8 888
5.: $8. 8 ANN: cm: 828 5.8 8.8 $08 84.8 5.8
+3.? +3.? 25 2 .o 8.? 8.? 2.8.? ...... a 8.? 8;? 8;? 81.6 382m
A3835 383%
AS. 8 A3.: A28 an: and
2 .? E .? :8? 2 .o 88.? 88888.88 28:8
and A8: 8.8 5.8 8.8
83:? 2.? 8.0 :86? 3.3.? 8:888:88 282
A88 A2 .8 8:8 2 88 $8
3.? 2.? 8.3.? :3 8;? “.8888: 288.8
5.8 :38 so. 8 5 .8 898
8 .? So _ S :3? :3? “888888 222
AR. 8 A88 8 .8 88.8 84.8
_ a? ...: .? 3%.? So 3.8.? 2888
$88 5.8 3.8 A: .8 38.8
*ecmda No.0- V0.0 simmdl envwd- 032
5:385 :33» «St www-mfitm
S 8 £8 68 E A8 88 n8 8 28
2.6 88m 8.888 :32 828: 833. 885.60

 

N we hon—85a E 3330
@8888 05 mo notion 98 83% e3 coEmomEoo 29—028: no 52858 <m mo 83:: BE. .o.m 2an

08m Eonomsom <

93

In contrast to PA mortality which generally is not anticipated very far in advance,
elderly mortality can be anticipated to some extent. Over time, their roles in the
household may progressively be absorbed by other household members. Our results
show that the death of elderly adult male and female members reduces the household size
by 0.81 and 0.57 respectively. Thus, there is greater partial replacement of household
members by males (though not statistically significant) in the event of an elderly female
death compared to elderly male death. The death of elderly males also results in a
decline in number of girls. Although statistically significant at the 20% percent level,
death of elderly males tends to increase the number of boys in the household suggesting
that the tasks performed by elderly males may be assumed by boys absorbed in the
household from the extended family.

Do these findings support the general assumption in most qualitative literature
and popular discussion suggesting that, in general, households incurring prime-age death
face severe agricultural labor constraints based on the notion that death of a prime-age
adult reduces household’s labor supply and increases its dependency ratio?
Unfortunately, results in Table 3.6 alone cannot adequately provide definitive answers to
this important question. But these results do provide some indications. First, there
seems to be evidence suggesting that households are coping by partially replacing lost
household members. Second, a comparison of household size ex ante shows that afflicted
household had initially larger household sizes compared to non-afflicted households (see
Table A3.] and A32 in the appendix). In addition, one has to bear in mind that, in
general, most agricultural households are likely to face labor shortages of some kind at

particular times of the season. Based on these reasons I suggest caution before making

94

any conclusions that assume that the loss of a PA adult will cause the household to face
greater labor shortages than non-afflicted households. To seek more clarity on this issue,
the impact of death on household composition is considered by examining the differential
impacts due to the differences in initial pre-death wealth status, land holding size and
effective dependency ratios.

Also, the sensitivity of the findings is tested by stratifying the sample into two
groups based on the initial pro-death value of household assets in the 2001 survey37 and
then estimate the models in Table 3.6 with gender and position mortality variables. This
is done because the ability of a household to respond to any kind of death shock is subject
to some of these factors and to ignore them in the discussion may lead us to make

incorrect conclusions (Drimie, 2002; Yamano and Jayne, 2004).

Does the impact on household composition difler by initial household conditions?
First, I look at the results fiom models with interactions terms between death by
gender and position and pre-death household poverty status, landholding size and
effective dependency ratio to capture the likely differential impacts of the death shock.
The hypothesis to be tested is that initial poverty exacerbates the impacts of death and
that poor households are less able to cope with the shock.
Wealth status in 2000: The results of the interaction terms between death and wealth
status as shown in Table 3.7 are not statistically significant except for the male
heads/spouses death and wealth status interaction which is statistically significant at the

10% level. The positive coefficient suggests that the reduction in household size is

 

37 . . . .
Value of assets includes farm equipment including scotch-carts, oxcarts and harrows, value of cattle and
small animal stock and non farm assets such as bicycles and motor vehicles.

95

greater in poorer households compared to non-poor households. Non-poor households
are attracting boys and other males (statistically significant at 15%) to replace the labor
lost among households incurring male head/spouse death. Notwithstanding the loss of
knowledge about farm practices and the foregone income of the deceased, older boys and
other males may be filling the labor void left by the deceased male head in non-poor
households. Since all the other interaction terms are statistically insignificant only some
simulations are done to compare the impact on household size due to male head/spouse
death between poor and non-poor households. In order to do this the impact of male
head of household death on household size, change in males and boys is evaluated for
two scenarios: (1) poor households (bottom 50% of value of assets distribution), mean
land size (3.10 hectares) and mean effective dependency ratio (1.33); and, (2) non-poor
households (upper 50% of value of assets distribution), mean land size and mean
dependency ratio. See table A3.3 in the appendix for descriptive statistics on initial (pre-
death) household characteristics. The results indicate that household size declines by -
0.301 in non-poor households compared to -O.838 in poorer households. The
implications of this finding are that poorer households have substantially greater
difficulties in coping with the death of core male household members while non-poor
households are likely to almost fully restore household size to former pre-death levels.
These results seem to support earlier findings in Kenya that poor households
experiencing a male head death are less able to cope with the shock due to limited

household resources to support additional members (Yamano and Jayne, 2004).

96

- -1..¢.u_ru.-:ru :p~..h.—.f.nvnh—._LPVU ~v~Avn~

 

 

  

32:0: :3 ~32:

:C—Z (LLC .-c...»::..

 

 

.c i. F

  

xiv

 

 

 

a : .? 8:? 8.5.? .8? 2 3 8 S :88? :82- .88.? .88.?
328 E88 :38 33: 3:8 :88 $08 :88 a? 8 a? 8 93:588.. 55883:
1.2 md- 1.: md- :mmmd- twwmd- :oood 1mg .o :NSd 13.080 «*mvvd- tomato-
5.8 888 :38 3.8 :88 :88 2 _.8 888 3.8 £88 38 8: 32°85:
:88? £8? :88? .88.? so? So? So? moo? :83? .88.?
$88 $8 88.8 :28 3.8 5.8 888 a 88 :28 8.8 econ :8": was. 8.8:
.we? :8 3. :83. .83.? 23. Rod 83. 2o? .82.? .5? 88.2888 :8 583.2:
5.. 8 3.8 2&8 3.8 5.8 E8 628 $08 688 5.8
:o? 8o? :5? N8? .28.? .MS? :5? :5? Re? 88.? as 8:2?2: Z: 38
898 $88 $8 $8 2: .8 :38 5.8 688 $88 3:8
83.? 85.? $3 Sod .38.? .88.? $3. 838 2...? an? .28.... beam
:28 328 $8 so: 38 $8 888 3.8 :88 :88
o: .? ”3.? +22 +83 so? 85.? :5? top? .5? .83- .28. 802:
5:5on 583%
$8 688 8:8 :88 A38 £38 5.8 $88 388 am:
as? so? 8: .? 88.? 35m? 2.28.? mm; 8 3 .43.? +5.? 0388888: 085:
:88 E8 :38 A88 A28 A88 :88 28.8 5.8 :88
.88? .83? So? go? £3 83 .83.? .88.? :5? .85.? 8888348.. 2.2
828 888 8.0.8 888 A38 :88 A: .8 688 :88 38.8
:2 .? 92.? .88? .58? .88.? :88? :3 83. :33. _- .38.? .8388: 285:
an: at: 8:8 2 n8 :8: 5.8 618 :88 8:8 6:8
2:? 8:? m8? .8? 2 : .o 8 3 :38? 18s? .88.? .88.? 0888.8 2.2
ASS-BE £388 QR» 8M8$Etk
5 8 a8 :8 E8 :8 88 n8 8 :8
2:0 whom 888i 882 885860

 

«o 88:5: 3 owEfiU

08m Eczema—om 4

82858888 588-2: 20:88: 88:: .3 8:80:88 228mg: :0 3:388 <8 .8 83:: BE- .n.m 2an

97

 

 

 

 

Elf
ubnv-uuumsuhcvv Kn R w:

Afi.n.~.

 

 

$98

simmmd

an;
an:
$3
$08
83.
at:
E?
5.8
83.

888
38.?
8.8
83
888
8.3
c :8
So?

8.8
25.?
5.8
8...?
A88
8o?
88
88

:38

3%de

8%

A38

353.0

we?
88
.83
$98
83
5.8

*hde.

5.8
:9?

A88
:58
:88
88.?
80.8
98.0
9.88
.88?

:98
85.?
$18
83
:88
83
$18
.83

:98

.1.va6

we?

88

simmmd

8%

A58
.32 .o
8.8
NS?
5.8
8.3
5.8
Sod

$8
8o?
A38
38.?
$8
:8.
$88
:2.

A28
2:?
688
8.3
828
88.?
5.:
3.?

A88

tmmmd

3%

88
8:.
3%
8.8
.88
308
Sod
388
08.?
688
on;

:98

1,304..

aw:

+Nmod-

: 8.8
98.?
608
to?

Gm:

+mmmd-

$28
38.?
:58
EN?
8.8
83

888
v88
8%

9.88

95mm.“

mo>
A38

SL190

898
N88
:28
8%.?
G88
.88

A28
23.
an:
88.?
A38
Sod
888
$2?

88.8
5.?
5.8
$8?
$88
38?
5.8
+82

688
9.. _ _ m4 88800
8? 8888 088 x 38>on
mam .8388 86m 850
mam .888 :33 $50
mam $888888: £80m
Wan—983: 232

8.8: 858ch8 3 >88ng vi
08m unwrmflgom :38 580
08m cam—$388 :38 550
08m 6878308833: 280..“

08m 8578308883: 352
8.; 33 88 3 588 E

883 €u>o9m£m8£ 550
was... €308.33: .8050
88... .9968 .. ma £85m

33% 3.5.69. .63: £34
38.“. A298& ‘3 583 Vt

 

325:8 5 .38.

98

.ooaaomawa mo ~32 $2 + 98 88.888... mo 3%: c\om .. 608.888., mo 3%: fix; .3. 88 Z

voom was SON $2.8m _S=o8o_aasm m8. ooomRoa 88m “mots: “mom mdeOUEOmU ”880m

 

 

8mm 8mm 8mm 8mm 8mm 8mm 8mm 8mm 8mm 82 258530 Mo H2:52
2.0 :8 a; 8.0 :5 o; .3 .3 25 28 .2388
1.... 8 m :88 2.8.2 $8.2 : SE 8.22: 8.88 58.838
:5. 2L... .8 .88 :8. 3:938 08.0898. 25
.m 8 .N ..m.~ mg m.._ we; 8.58 8.33. €23 .32
3o :3 8: 9.: :8. in... :82 :68 .38 ta? 8388 €85.
*mvN *wm.N am; NBA aimed *uvwd— ELwS aumNdm simfifl *aomfi 3:588: owmuogm
an: E5.

 

335:8 S“ 2.3

99

Sauna-«Emma mo _o>o_ $3 + 85 850.83%.“ 80 352 8m _. .oofiouawu no 388 o\o~ .1. 882

38 2a 88 5:5 3:25.885 2a 8888 85m ES: 888 mam-8032.08 ”8.8m

 

 

 

 

 

8m $8 28 8m 28 8m 8m 8m 33 $8 823530 .8 52.32
28 Ed 2 .o w; 2 .o :8 8 .o 8 .o :8 8.0 8538
83 +8.8 28 and ME .. :4. :88 :52 ME :26 5:558 €285.
+82 83 m3 8: :88 2.“? 2.8% 1.88 L S. 1.8% .8388 own-85
.8\88~ “Sox-
8% 8% 8% 8% 8% 8% 8% 8% 8% 8% 8.8536 2:: x 85>on
GB. 8 3.8 E. 8 628 2.98 8.8 5.8 5.8 85.8 c 8.8
+5.0- 2.82.? 888. :88? .83 :38 2: .o :23 Bed 3 8 .o 2588
58.8 c _.8 a8 58 A88 808 A88 88 5.. 8 A888
33. 23- 83- 888 8.2- :8 no- 83. +33 88- 8.3- 883 €85
8.8 888 A88 828 38 8.8 33.8 A88 8.8 888
8; .o- +23- 8.3 338 E .o 83. .1388 1.38.? @28- a: H 2- 838 332m
A5585 583%
so. 8 A88 A8. 8 838 $8 888 A2 .8 8.8 $88 85.8
5 .o- :88- ~88- :3- tawd- amno- 23. .888 8%.? $3- 0338888: 288
5.8 so. 8 A38 388 8.8 5.8 5.8 $88 58 A38
.883- Ed- 838. 88 .o- 3:. ~88- ...voed- 38. 88.7 +5.8 0838888.. 222
£88 a~8 5.8 8.8 8.8 E88 888 2 M88 :88 68.8
838. mod- 2 S- 83- :88? :23. 5 .o- 8: .o 88- +8.3- 082888; 22.8
5.8 $8 8.8 5.8 5.8 88.8 5.8 3.8 8.8 8.8
Rod 83- +8.8 +82- 88.0- 228 .23- 28:8- 32- :88. H- 0808888 222
ASS-BE #388 «Vt» was-85.2%
Son no Z Sam Sam no Z 808 Room :02 Son “com no Z “com boom :02 Room
£80 8.5 mofifiom 832 8828.60

 

mo 89:56 E owafio

08m 20880: <

 

25:: 3.8.58 .3 2.8.—8:2— E .8588 85 88:8 .3 558888 22.8.5.— E. 3:53.: 43 no «23:: 2:- .w.m 83.;

100

 

 

 

 

 

hi

 

Turning to the models stratifled by asset poverty status, results in Table 3.8 show
that among poor households, male head/spouse death causes a unity decline in household
size whilst among non-poor households the impact is statistically insignificant.
Therefore, the significant negative impact on household size in the full sample is to an
extent influenced by what is going on in households that are in the bottom 50% of the
assets distribution. The reduction in total household size as a result of female
head/spouse death in both poor and non-poor households is mainly explained by the
change in the number of females. Whilst, an additional other male death among non—poor
households results in a more than one person reduction in household size. Looking across
the columns under non-poor, the decline in household size is due to a reduction in males
and young girls consistent with the finding in the full sample models in table 3.7 column
I. .

Eflective dependency ratio (EDR) in 2000: Tables A3.1 and A32 in the appendix
show that both ex ante and ex post EDR ratios of afflicted households are generally equal
to those of non-afflicted households. Households incurring prime-age female death
(heads/spouses and other females) between 2001 and 2004 had a decline in mean
effective dependency ratio in 2001 of 0.16 and 0.35, respectively. Households incurring
PA mortality of male heads/ spouses and other males had a decline in EDR of 0.34 and
0.41. Non-afflicted households had a decline of mean EDR of 0.34. It is likely that most
afflicted households are able to restore their dependency ratios, at least to some extent, by
attracting new PA members or sending away children to other relatives, or that afflicted

households are in general further along in the household lifecycle, such that some

101

 

 

 

children are old enough to have switched from the numerator to the denominator of the
dependency ratio, or have already left to start their own households.

Given this background, the impact of PA mortality on household composition
may depend on the initial household effective dependency ratio. To empirically test this
apparent relationship, death variables and pre-death EDR in 2000 variables are interacted.
The results in Table 3.7 show that not all interaction terms are statistically significant
except for deaths of female members, heads/spouses and other females. Table 3.7,
column H show that there are differences in the change in number of boys by initial EDR
in households experiencing female heads/spouses and other female death. Among
households with higher effective dependency ratios in 2000, other female death results in
a smaller reduction in numbers of boys compared to households that had a lower EDR in
2000. This could be explained by the fact that households with a higher EDR in 2000
that experienced non-core female deaths were able to partially replace the lost female by
attracting other females in the household.

Land holding size in 2000: The land/labor ratio}8 provides a rough measure of the
household’s potential supply of labor per hectare owned. A priori one expects that
households with lower land/labor ratios in 2000 are less likely to need to attract new
members if the household experiences a death, whilst one might expect to see households
with high land/labor ratios attempting to attract new members in the event of a death.
However, descriptive results in tables A3.1 and A32, do not show any significant
differences in land/labor ratio between afflicted and non-afflicted household, either ex

ante or ex post. Since changes in adult equivalency are endogenous, I estimated models

 

8 .
Adult equivalents are used as a proxy of farm labor.

102

 

 

 

of

3.4

 

 

with interaction of PA mortality between 2001 and 2004 and initial land holding size in
2000 in order to determine the differential impact of death on household composition
according to initial land holding size. Despite the loss of wealth due to illness and prime-
age death, non-poor households may be able to restore their land/labor ratio, at least to
some extent, by hiring in extra labor or attracting new PA members compared to their
less wealthy counterparts. The results in Table 3.7 show that there is differential impact
on household size among households experiencing male head death. Among poor
households with small (25th percentile of landholding size, 1.06 hectares) and large land
sizes (75th percentile in 2000, 4.0 hectares), and using the mean effective dependency
ratio of 1.33, the death of a male head of household results in a decline in household size
of 0.92 and 0.74 respectively. Among wealthier households, the results indicate that
household size rises by 0.04 and 0.22 persons respectively. This finding suggests that the
effect of male head death on subsequent household size varies greatly between poor and
non-poor households. Poor households experience a decline in household size after the
male head dies, while non-poor households are able to partially compensate for the loss

of a male head, primarily through attracting more boys into the household.

3.4.4. Impact of PA death on farm and crop production

There are at least four pathways in which farm production can be affected by
prime-age mortality, through impacts on labor, knowledge, capital, and land. First, the
reduction in household size may result in labor shortages, which force households to cut
back on land cultivated or switch to labor saving crops. As mentioned earlier, this

general assumption has led some development agencies to advocate for greater

103

investment and promotion of labor-saving crop technologies. Results from the impact of
mortality on household composition in the preceding section show that there is partial
replacement of members when a prime-age member dies but less so for relatively poor
households experiencing the death of a household head or spouse. Second, the death of
an adult may also entail a loss of agricultural husbandry, management, and marketing
knowledge, requiring a change in crop mix. Even in households coping by attracting new
members, the skills of the new member may not match the skills of the deceased, who we
found to be primarily boys and girls. Third, crop mix and the intensity of input
application may change because of cash constraints imposed on the households afier
incurring the loss of an adult member. Certain cr0ps require greater use of capital (e. g.,
purchasing farm inputs, chemical Sprayers in the case of cotton, rental of animal traction
services). Fourth, and especially in cases where the male household head dies, the widow
and her dependents may have insecure land tenure rights.

The results presented in this section should be interpreted taking into account the
dynamics of household composition from the preceding section. Table 3.9 (column A)
show that, in general, adult male mortality resulted in an 13% decline in total land
cultivated; this effect is significant at the 5% significance level. Female death of any
kind resulted in a 5% decline of cultivated land but the impact is not significant at the
10% level. Surprisingly, there is a 30% decline in land cultivated when elderly (>59
years) men died. This finding suggests that men (aged 60 and above) remain productive
in their old age and probably devote a greater portion of their time to crop cultivation
than the younger male heads (who tend to be the primary earners of off-farm income in

sampled households).

104

 

 

 

d:

CE

be

3.9

 

Table 3.9 also disaggregates impacts according to the gender and position of the
deceased person (Column B) . Male head/spouse mortality is associated with a 21%
reduction in land cultivation. Examining across the row for “male head/spouse death,” it
is shown that most of the decline in cultivated area is for cereal crops (-19%) and hi gh-
valued cash crops (-4%). Column B results also indicate the important impact of elderly
male mortality on crop cultivation, particularly cereals. All the other mortality
categories are negative but not statistically significant. This finding seems consistent
with our earlier findings that households incurring male head/spouse mortality experience
a higher decline in net household size, which may partially explain the significant cut
back in land cultivated for this particular case.

By disaggregating area cultivated by area under cereals, root and tuber crops, and
high-value cash crops, I examine the impact of mortality on particular crops by (columns
C through H in Table 3.9) and test the hypothesis that households experiencing prime-age
mortality switch to less labor-intensive crops Examining the first row of data in Table
3.9, we see that the 13% decline in land cultivated among households incurring a male
death is due to the reduction in area under cereals and, to a lesser extent, high-value
crops. The coefficient on changes in roots and tubers is positive but not statistically
significant. By contrast (examining the second row of data), PA female mortality is
associated with a 5% decline in area under roots and tubers. When distinctions are made
between gender and position of the deceased (in the subsequent rows of Table 3.9), we
see that mortality is associated with a subsequent decline in cereal area of 19% in the
case of male heads/spouses, 8% in the case of female heads/spouses, 11% when the death

was another non-head male, and 20% for men over 59 years of age. These results support

105

 

 

 

 

 

mal

SOll

cm;

decl

spou

 

 

the importance of disaggregating mortality by gender and position in the household of the
deceased as well as by crop mix. Otherwise, one would have concluded that households
incurring male deaths had a statistically significant decline in land cultivated but missed
the fact that mortality of older members has the most important impact on crop area
cultivated.

Also, the results in table 3.9 column H show that among households experiencing
male head/spouse death, area under high value cr0ps declined by 4%, suggesting that, to
some extent, the death of a male core member of the household adversely affects cash
cropping.

Regarding impacts on area under roots and tubers, there is a significant 5%
decline among households experiencing the death of women other than the head or
spouse. No significant impacts on root and tuber cultivation are found in the case of men
in any category or for female head/spouse or elderly women. These findings do not seem
to support the hypothesis that households experiencing prime-age death cope with the
reduction in family size by switching to labor saving crops such as roots and tubers.
However, the small and generally statistically insignificant declines in root and tuber
cultivation are substantially less than the estimated declines in cereal area in response to
adult mortality. Moreover, there might be differential impacts by pre-death household

characteristics in 2000 survey. This is analyzed in the next section.

Does the impact on land cultivated and crop mix differ by initial household conditions?

Turning to results from models with interactions terms between mortality

variables and pre-death household characteristics, the likely differential impacts of the

106

.33 can 83.8w? .9808 :8: .0838 .8888 cospoudsm .588 08205 808 02.3
awn-r. 88808 823 .8688 08203 880 80m; .808 team SEE 850 =a 88 88:38 .8883 .858 .388 88203 £880.. .838 88288 fisnom
8833-898: 55 888328 888.“ 08 885888 E 83:92 dunno—8:86 .«o _o>o_ $3 + 88 condo—mama 80 ~92 o\om _. dunno—mama mo _o>u_ .XL .1. ”8on

EON 85 SON Sugm anaoaasm can coo-«.383 85m 8882 “mom EWQOUEOmU U850m

 

 

 

 

83:882-— momm £589,830 .8 89852
8.0 8.0 3o 26 2.0 28 cm... 88 Bang-m
8: 8.. :3 +82 $3 . :2 .3... .88 £382 .5 .8...
m0? we? 83% we? .06? we? no? mo> mogfi 05w. x “Sugar—m
688 an. 8 5.8 5.8 5.8 $88 23.8 2.8
cod 8... 8.8 8.8 +8.? +8.? 38 38 .5388
23.8 28.8 a... 8 9.....8 28.8 98.8 $8 3.8
3o 88 3o 8... E .o- E .c- 8.8 8.? .225... scam
$8 5.8 :48 8.8 888 5.8 $8 8.8
So- 5o- 8.? 3o- .28. .88 tomo- ..om.o- .22. 3.28
8.3.2.. 8.8.3
9:8 5.8 5.8 :88
No.8- .38. 3.0- nod- omsommEaua-:o—H 298m
328 an 8 82.8 3.8
:3. 88 L 3- 8...- 08.888888: 2.2
8.8 98.8 388 9.88
8.0 8.8 +8.8 8.8 8&88: 2.88
3.8 8.8.8 688 $8
.38- 88- .88- .13- .83888 032
888 8.8.8 8.8 $8
:3. .38- 8.? 8.8 one...
an. 8 888 2 M8 688
+8.8 :3 .28- .88- 2.2
.9322: £388 «Vat www-mfitk
a8 68 8 m8 R8 68 88 28

 

umac-5 eighwi

ammo-G Hoe can 838%

.8880

 

”$5 88:: 88 .3 m2 8233 E ownanu

83988 3285 a

888.60

 

22880: E notion 85 88:8 .3 8:3 883:3 so MES-SE <8 .«o 8888. 2:. .m.m 2an

107

death shock on land cultivated and crop mix by wealth status, effective dependency ratio,
and landholding size are discussed.

The interaction terms between mortality and wealth status are jointly statistically
significant. However, there are differential impacts by wealth status among households
experiencing mortality of male heads/spouses and other males as shown by the
statistically significant interactions in Table 3.10. The negative interaction coefficients
suggest that mortality leads to a greater reduction in land cultivated among non-poor
households compared to poor households. The impact of mortality of male
heads/spouses and other males on land cultivated was evaluated for two scenarios: (1)
poor households (bottom 50% of value of assets distribution), mean land size (3.10
hectares) and mean effective dependency ratio (1 .33); and (2) non-poor households
(upper 50% of value of assets distribution), mean land size andmean dependency ratio.
From these simulations, land cultivated and area under cereals decline by 75% and 61%
among non-poor households experiencing mortality of male core members of the
household compared to 35% and 36.1% decline among households in the same situation
but in the bottom 50% of the 2000 assets distribution. By contrast, the impacts are lower
in both poor and non-poor households experiencing female heads/spouse mortality. For
example, among poor households experiencing female head/spouse of household
mortality, total land cultivated and area under cereals declined by 0.4% and 7.3%,
respectively, compared to declines of 6.4% and 4.7% among non-poor households,
holding all other variables constant. There is a less severe but similar pattern of results

for other (non-head) prime-age males. These results indicate that other PA males are also

108

key to crop production in rural Zambia and that their death adversely affects land
cultivation in both poor and non-poor households alike.

Turning to the models stratified by poverty status, the results in table 3.11 seem to
reinforce the above observations that land cultivation and particularly the area under
cereals declines significantly among non-poor households incurring male mortality, both
among heads and non-core men. Similar to the findings by Yamano and Jayne (2004),
mortality of the male head of the household in poor households is associated with
reductions in area under hi gh-value crops. However, there are no other statistically
significant impacts of mortality among households in the bottom 50% of the assets
distribution except in the case of elderly male mortality. The results reinforce findings
reported earlier that men over aged 59 still exert an important contribution to crop
cultivation, not necessarily entirely through their own labor input but also through their
capital resources for renting animal draft power, hiring labor, and other inputs into the
production process.

The interaction terms between PA mortality variables and EDR as well as
mortality and landholding size in 2000 are not jointly significant, suggesting that there
are no differential impacts of PA death by initial dependency ratio or land holding size.
In all cases only the coefficient on male head/spouse death is statistically significant on
changes of area under cereals and total land cultivated, respectively. The interaction term
between male head/spouse death and EDR in 2000 is negative, implying that among
households with higher EDR in 2000, male head of household death results in a larger
decline in area under cereals compared to similar households that had lower EDR in

2000. In contrast, the interaction term on landholding size in 2000 and male head/spouse

109

 

 

 

:98 38 6.88 3.8 $28 698 5.8 :38
so? So? so? :5? wood mood 33. 83 @3833? 588a% 2,80%
$08 :08 am: 828 $28 $8 $08 A38
83 53 $8.? $8.? .58.? 3.8.? 2&8? :38? 38 2% @2223
$8 6:8 :98 8N8 $8 G98 :48 $8
.1de .23 £85.? :35? 2:3? 302.? is? :02? coomuo c838": mesa €28
QQQN E wobwtfioggo 333.33 Sandal
$08 a3 5.8 828 8:8 3.: an: 6...:
:80? :25? 8o? 25.? i 8.? i 8.? Bo? woo? as 8:255 >5 32
A38 $8 an: 3.: so: so: 38 828
23 23 was :3 92.? 3.? $5.? :5? ”29%.. reg
$8 $8 an: so: $28 5.8 $8 2 we
so? So? 03.? 2o? :2? $2.? 23%.? 253.? 8:5 beam
528on ANSEQ
cm: E: 30: 5.8 2:8 2:8 E8 $08
to? go? So? .58.? 83 25.? NS? «8? asoeesnaoa 220»
5.8 5.8 5.: A8. 8 God God :98 5.8
So? So? $38 :56 as? 83.? «3.? so? asoeasnao: 2“:
A88 2 _.8 888 § .8 $98 838 A38 A2 .8
53 Sod 3o? 39? mod Sod 33 8o? usamesn 29:...»
:98 $8 3.8 A2.: 9:8 and 608 9:8
.35.? .33.? So? 58.? tam? .88.. a; mm? .38.? uaogmaafi 222
358on ~32» «Vat mwudfitm
A8 6 A8 5 A8 5 E E
0308 029752 cacao 88 can 335. .mEoSU

 

”EU BUS. 88 mo wo_ 353: E owqmnu

93928 SEE a

855300

8558330 53?on 238:0: 3:5 93 action .uoccow an 95 @8323 so 3:358 <m .«o 83.5 2; .ofim 05mm.

110

 

58
25.?
an:
2o?
A88
80.?
:6:
was

$08
N8?
A38
Sod
G_.8
8o?
$8
8o?

G~8
So?
82:
.38.?
E8
89?
:88
80?

AS:
~85
2%.:
NS?
§8
23
A28
8o?

Gm:

+306-

:58
89?
A28
«8.?
an:
:3:

A2.:

*Nm_ .0.

68
~23
E8
86.?
6:8
33:

828
:2:
6:8
3::
93.:
23
Gm:
32.?

6.0.8
Sod
§.8
53
GN:
So?
as:
+89:

:98
m8?
5.:
+838.
:08
So?
an:
.52.?

5.8
$3:
A28
23
£28
$3:
5.:
$2.?

$8
2o?
5.:
:85
A8,:
2??
Ga:
*vood

A58
E?
3.:
:2?
92.8
”3o
A8:

sowmd-

are... 55:59: $0380: :25 850
038 55:83: $25: :33 680
2:2 55:83: «momsoamivmoa cinch

0:8 honuonoaovumomaommhumon 2a:
95: Aozmhzmmoh 8% 3 58% Ex

undo. 3230389.: “Eva 850
3% Burma—«E :53 550
ofim cam—*momsonmhcaon Baum

0N6 :5783omfl83a £32
3.: mssg 33 88 .3 58:» E

mafia €u>09m2m8£ :33 880
£56 €269... mug—«E :33 $50
mafia Egofimomsomflmgog Baum

mBSm begofimumsomflmcmoa 032
3% been 83 3 seem E

 

32:5... 2d 2.3.

111

SE 835%? 880m 55 .8239 .uoboo .Hosougm 8:8 00205 380 02? swan .8888 “036 6388 ova—0E 880 83? .380 coo.“ SEE
550 2a 98 85:38 .3053 42:6 .oNHE 62:03 £350“ dunno—mama mo _o>o_ obo— + 93 oogowawmm mo _o>o_ o\ow _. dunno—mama mo :53 o\o~ I ”882

voom 98 Sam Sufism Enuaoaqsm 98 OOONRQE ~A33m “meta: “mom mdmfiOUEOmU 632%

303.3530 we Honfisz

 

 

w; w; 85 93 25 26 So So 33%;”
3o 2: E A :5 5:388 frog baggage 380mm
25 $2 N3 :Sw $588 $3.5?va 83
:3 is? 2.8.5 :Sa 3588 omEEquasm €38
So go NM: 22 .85 .85 :30 £3 £538 :35
z; 2: m3 $3 :23 +22 2.3m EN Euros $385
3.48 3.85

5 .9 5.3 $9 $3 am: 9%.: Gus and
:33 :33 23. mad :3 a; .33 .83 83280
8% we? we? we? we? mo> mo> we? mega—v Us K ”Dagny—m

 

UGq—flmflflOU ¢~.M O—QGH.

112

.32 833%? 850m 55 .0839 .8883 .HoBomasm nozoo ova—05 308 2:9, Awe“. .8883 803m 6338 £535 398 «coda .380 team ESE
85o =m 98 85:98 43:3 .858 dug: Hows—2: £3“an dunno—mama .«o ~33 $3 + 28 ooguaawfi .«o :32 o\om * .ooamomfiwwm no _o>o_ o\o~ 1. ”882

vooN ES SON Sufism gage—masw 98 ooomaaa zogm H8383 “mom namEOUEOmU H355m

 

 

 

 

113

 

as $8 $3 $8 $8 $8 $8 $3 Sofituso mo p52.5.2
85 93 m3 Sac o3 ‘ 8d 03 go @2338
woo 25 :.o m: x: i 3. N3 :26 £558 :55
w: +3.8 L 3. N3 in? Ed :84. 85 £39: 032:5
853.2 3.85
8% "3% me? we? mUV we? we? we? mega—v DEE K cog—freak
5.: 8a.: :9: E8 :98 A3.: $08 av:
83 83 38.0- Sod £3- 23- 8.3. $3 assoc
A38 698 A8 . c 808 E8 and 58 A38
Rod 82.. 08¢ 53 C38 33. v88 N26. 8388 beam
:98 808 A88 5.: an: :9: SN. 8 :38
ago. Bod- 83 .68.? N: .o- 3.3. 9: .o- :23- 8:5 355
$3295 ATwEM
5.: $8 A38 9.98 5.8 a: .8 A3.: $8
83.- 89¢. 18 S- 086 N85- wad god. :3. anamasaao: 283m
5.: §.: 5.: :38 5:8 38 E. 8 5.8
M88 :3 08¢ Sod .82.? £3 $2.? 23 a:&€§-=8 2a:
5.8 908 A38 $08 G08 :88 $8 $08
:3. wood 8.90- 53 83 :58 wood Sod- ogoawesn 22$
am: and A88 5.: :28 E8 83 at:
856. .58.? Sad 89¢. :83- as .o- gang. 2.: .o- ogogeafi 222
A5335 :58 «3Q www-mfitl
goon aoz Hoom uoom aoz goon “com doz goon poem :02 “com
0308 vigéwfi nacho 88 US $33. E850

 

€8~>E=o «835 < $35.60

AS: 89:. 38 80 m2 .833 a 0320

 

353.9. 3.55.— .3 22.8.5: 5 .858.— ES .8..qu .3 6:2 cog—«>55 .8 5:53.: <m «a «23:: «an. .:.m 033—.

death is positive implying that among households with larger landholding sizes, male
head of household death results in a slightly less severe decline in land cultivated
compared to households in similar situations but having smaller landholding sizes in
2000. For example, among households with landholdings in the 75th percentile and
incurring mortality of male head/spouse, total land size declined by 54% compared to
73% among similar households but with land sizes in the 25th percentile. This result is
surprising, because it implies that remaining household members are switching to other
non-farm activities because of the departure of the core person responsible for the farm
enterprise. I empirically check this fact later in the paper when I look at impact of

mortality on off-farm income.

Impacts on Crop Production:

Before incurring a death, afflicted households had higher gross value of output
and gross value of crop output per hectare for the crop season 1999/2000 compared to
non-afflicted households. Generally, households incurring prime-age death between
2001 and 2004 experienced a decline in gross value of crop production and gross value of
crop output per hectare (Table A3.1 in the appendix).39 Due to large standard errors, the
difference-in-difference descriptive statistics were not statistically significant even at the
10% level. Unlike in the Kenyan study where gross value of crop output was observed to
decline significantly due to a switch from high value crops, our findings show a slight

decrease.

 

39 There values are calculated by applying mean crop prices from the 1999/2000 and 2003/04 seasons
uniformly to the quantities of crops produced in each season. This method avoids the possibility that
changes in crop production values are caused by weather-induced price changes. All changes in crop
production values between 1999/2000 and 2003/2004 are caused by changes in physical production.

114

Table 3.12 presents the regression results from models measuring impact of PA
death on gross value of crop output and gross value of crop output per hectare. Although
the coefficients on mortality variables are negative as expected, there is no statistically
significant effect on gross value of output for mortality by gender in general or
disaggregated by gender and position of the deceased.

The results are similar for the models measuring the impact of PA mortality on
gross output per hectare except for the statistically insignificant but positive impact
among households experiencing male head of household death. Although insignificant,
this result is surprising because there is no evidence to suggest that these households
shifted to high value crops and, in fact, land cultivated declined by a larger percentage

among households incurring male head of household death.

Does the impact on crop production difler by initial household conditions?

Potential differential impacts by initial household conditions are tested for by
estimating models with interactions of PA mortality with household wealth status, land
holding size and effective dependency ratios in 2000. Table 3.13 column B shows that
PA mortality impact on gross value of output does not differ by initial poverty status. In
contrast, the impact of male head of household mortality on gross value of output per
hectare seems to be influenced by initial poverty status (column D). The statistically
significant positive interaction term on male head death and poverty status implies that
among wealthier households, gross value of crop production per hectare increases more
than in assets-poor households experiencing the same shock. For example, evaluating the

impact of male head death on gross value of output per hectare at mean land size and

115

.flobo 985% #453—
-SEBLonsm 515 333—3 moans 8a 835539 E moon—852 60:85:»? no 786— $2 + 23 ooaomawmm .«0 EB: o\°m .. dunno—Mimi 80 ~32 o\o~ a... 6802

38 ea 38 55m Ensusaam as 25838 3:5 “macaw mom $383880 ”ooaom

 

 

 

 

 

116

88 8 2 3 m 3 m 28 :8 as me n Baggage? Hopesz
Rd 28 So 8o 25 N3 25 2 .o @0538
R: E .o N28 26 n: 83 R.“ L _ .m Emcee 0»?an 8 so:
3% no? 8.». mo> mo? 8.». wow. mo> 3:556 095 x 33>on

E8 2: . 8 A3.: A2 .8 58 $8 $8 85:
25.8 ~88 «8.? wood- 83. Red. ”8.0- +5.8 unsaoo
58 E8 38 $8 A88 A88 $8 5.8
:3- Sod- :3. mod. 32. M: 8- 85.? 33. case beam
888 and $8 $8 58 3v. 8 A38 $8
23 N8 .0 3 3 2 3 Sad- 33. as .o- m2 .o- 8?: 355
5:5on 3%th
A38 :48 9:8 :98 $8 A8 .8
«.88 Sod- mad- 2: .o- Sod- x: .o- asafisaaa 225m
58 E8 :98 3.: 8&8 A38
88.8 33. $3.. $8.? mod «8.? oaoamaagaoa on:
5.8 :38 g8 58 A38 5.:
8 3. Rod- god- 2: .o- 98.8 2 3- ”Samoa: 2§£
A3.: $8 808 as: an: and
+23 ~38 £3 £3- 53- :9. oaoamasa an:
E8 5.:
03¢. 23- 295a
A28 5.:
wood- 5 .o- on:
REESE :33 «vi» www-mfitm
:8 68 E A8 88 Q 88 5
Son :02 “com ...... 3953 BE ....... Son :02 “com title—95m ==m ttttt 835.30

 

3.5593 «c 33> $895 < 9:330 .«o 05:; mmOcUVB <
Eczema—on 5 H8288 can 395w .3 058: com 89:0 8on 98 33:0 mo 33> 88w no 3:585 <m mo Emma: 23. .9 .m 2an

mean effective dependency ratio, we find that output per hectare rises by 32% in
wealthier households experiencing male head death compared to a 19% decline in assets-
poor households. Since there was a net decline in land cultivated in both cases due to
male head death these results seem to suggest greater intensification in wealthier
households experiencing male head death. Also, earlier it was found that wealthier
households incurring male head of household death attract boys and other males
suggesting labor lost from male head of household is taken up by boys and/or other
males.

Similar results are found when the sample is stratified into two groups by value of
assets in 2000. Table 3.12 column H shows a statistically significant positive (27%)
impact on gross value of crop production per hectare among households with male head
of household death. The death of other males reduces the gross value of crop output by
20% (column D). This decline could be explained in two ways. First, it could be
because of labor shortages due to the death of other male PA adults. Earlier, in section
4.4, it was found that mortality of other males resulted in a reduction in the number of
young girls who might have provided extra farm labor. Second, other male members may
possess extra knowledge and farming skills that are lost when they die. The rest of the
impacts of mortality by gender and position of the deceased on gross value of output and
gross value of output per hectare are not statistically significant.

Turning to the likely impact of PA mortality by initial land holding size, results in
Table 3.13 show some statistically significant differential impacts on the gross value of
output due to male head death and on gross value of crop output per hectare due to either

male and female head or spouse death. The negative coefficient on the interaction of

117

 

 

 

ma]

sin

dc;

 

male head death and landholding size implies that among households with larger
landholdings and death of male head, gross value of output decreases more compared to
similar households but with small initial landholdings. Evaluated at mean effective
dependency ratio and mean wealth status, households’ crop output decreased by 57.2 %
among households with landholding sizes in the 75th percentile and having a male head of
household death compared to 41.3% in similar households but having landholding sizes
in the 25th percentile.

Somewhat different results are obtained on the impact of gross crop output per
hectare. At mean effective dependency ratio and mean wealth status, the gross value of
output per hectare decreased by 6.2% among households with landholding sizes in the
75th percentile and having a male head of household death compared to 34.7% in similar
households but having land holding sizes in the 25th percentile.

In the event of female head/spouse death, table 3.13 column D also shows the
impact on gross value of output per hectare is opposite that of male head death. The
significant positive coefficient on female head/spouse death and landholding size implies
that the impact of female head/spouse mortality is smaller for households with larger
initial landholdings. Evaluated at mean effective dependency ratio and mean wealth
status, gross value of output per hectare increased by 17.1% among households with
landholding sizes in the 75th percentile and having a female head/spouse of household
death compared to a decrease of 3.7% in similar households but having landholding sizes

in the 25th percentile.

118

 

 

 

 

A: .8 no.8 58 A38
89c 82. Bod 83 3982882 58833 ӣ85
3.8 348 G28 G08
89¢. 83. :23- 2.48.? 38 0% 32°83
58 A2. 8 A88 308
48.0 33° :83- :2 3- econno 53:88 mass €38
QQQN 5 MUESLNBGLGQU ENSEMSQ.‘ Sumfinmkl
5.8 3.8 3.8 ans
good. mood- .588 $3.0- as Sasgoa >5 32
588 A28 A28 A38
33. ~88 Rod- Rod- Emma beam
64.8 608 $8 :98
E .o 2 3 ~28 ~28 Es: 352m
A5235 Arena
A38 E .8 §8 g8
£3. :3. 83. «So. oaafisaaon 038$
$8 $8 A88 :68
~38 Sod- Rod- Sod- oaoaesaaoa 2...:
3.8 $8 5.8 Go. 8
85.? ”88 Sod- 2:8 08338: 285
3.8 §8 3.8 888
$86 E; 23. $3- unifies; 032
bactofi :53 «Ft omeloEE
88 68 88 28
@3998 no 02.3 390 39:0 80 02? v.85
mo m2 333: E omEfiU 335850

 

22.02::— 5 538m 85 SEE 3 .258.— ..2— 35.3 8:: 2.5.8 no.5 .3 3:55:— «P— .3 3.3:: 2C. .n—.m 033—.

119

 

hvnvnncmuahnvrv ”in -9- awhnfi-u.h.

 

 

 

A38
~85
5.:
2 3-
8.8
23
A88
2 3

£08
83
A38
83
30.8

:Kod

f8

«*am—d-

5.8
~85
A28
v8.8
58
33.
A38

“iofimd

898
~25
A3:
2 3-
A88
2 3
A28
33.

2 _.8
~88
3.:
£3
an:
33
5.8

+Vmo.o-

6.8
$3.
an:
MKS-
A88
33.
9:8
~23

0:8 55338 $3558 :33 350
0:8 xouuvqomou *moEE :23 380
28. 339893 *momsoamEVmo; 255..—

ouS zunowaomuc*mom:o%\mvmun flu:

0.5: 32365.33 88% 3 33h Sm

cum 9878380.“ :33 860
36 warms—«E 56m 550
SR c5_*mogomm\mv~on 295m

36 camrmomaomflmcmon 222

8.; mssfi E3 83 3 saw E

336 €0>09~2~E£ :33 SEC
£68m ate>o9m£mfi :33 350
333 Egofimumsommugoa 295m

£55m 93093309330: £32

33: Aryan. 8% 3 «3% E

 

835:8 2d ~32.

120

880$:me mo _0>0_ $3 + 98 oogomfiwfi mo _0>2 XE * noogomfiwmm m0 80>“: X: “In 8802

88 85 88 8.3% 8882888 23 8388 8&5 :25: 088 8m800<§0m0 “85%

 

 

 

8mm 8mm 8% 8% 823530 8 H9852
-.o .3 :8 2.8 82838
Nnd 5: bzmtofi <muoufi 50080080“. 3:00am
.2.th figs 3:388 08-088“..on,“ 8:3
*ow.~ :E.~ ©2588 ow~-oEE*mBsm €38
.32: No.0 m fl N." E: 3:888 382m
~m.o 38 +83 of €388 08-888
38 3.88

358 68 $8.8 $8.8
«mod Rod Iovmd imvmd 83800
8% 8% 8% 3% 83:56 083 x 30305

 

8:588 28 «3.;

121

 

 

 

of out
ratio.
Also,

diffel

impz
can i
use <
10w:
limi
sun
the ‘

inc

 

 

There is no evidence of any differential impact of PA death on either gross value
of output or gross value of output per hectare by initial (pre-death) effective dependency
ratio. All interactions between mortality variables and EDR are statistically insignificant.
Also, the null hypothesis that the joint effect of the interaction terms is not statistically

different from zero is not rejected.

3.4.5. Impact of PA mortality on value of livestock

Liquidation of assets is cited as one of the coping mechanisms to mitigate the
impact of mortality and other shocks (see Barnett and Blaikie, 1992). Assets depletion
can increase households’ vulnerability to income shocks, and may decrease household
use of cash inputs and animal traction in crop cultivation, which will tend to result in
lower productivity and overall crop production (Mather et al., 2004). Due to data
limitations, I could not obtain a comparable measure of value of assets between the two
surveys, so the focus is limited to value of livestock, which was collected consistently in
the two surveys. In addition I analyze the impact of mortality on non-farm (off-farm)
income.

The results in table 3.14, columns A and B, show that there is a negative impact
of PA mortality on value of cattle but the impact by both gender and position in the
household of the deceased are not statistically significant except among households
experiencing death of a male head of household. Among households experiencing a male
head death, values of cattle declined by about 29.9% but this impact is only significant at
the 10% level. The results on the impact of mortality on values of cattle are somewhat

similar to those found by Yamano and Jayne (2004) in Kenya. Thus, to some extent,

122

households appear to try to hold on to productive cattle and probably sell only as a last
resort. However, this coping strategy does not seem to strongly apply among households
incurring male head of household death.

Table 3.14, columns C and D, show that PA mortality has significant adverse
effects on the values of small animals. The value of small animals declines by 29.5% and
41.5% after the death of adult men and women, respectively. When the results are
disaggregated by gender and position in household of the deceased, the impact is highest
among households with a male head death with a 65.3% decline, followed by mortality of
other females with a 44.3% decline, and a 37.0% reduction in values of small animals
among households with female head/spouse death. The result on female head or spouse
death of a 37.0% decline in value of small animals is only significant at the 10% level.
Generally, these results seem to suggest that afflicted households cope with mortality of

males and/or females by selling off small animals.

Does the impact on values of livestock differ by initial household conditions?

Results in Table 3.15, column B, show no evidence of differential impacts of
mortality on values of cattle by initial pre-death household characteristics. In contrast,
there seem to be differential impacts on values of small animals among households
experiencing a male head of household death. Table 3.15, column D, shows a negative
interaction between wealth status and male head death, implying a greater decline in the
value of small animals among wealthier households experiencing male head death
compared to poor households experiencing a similar shock. For example, evaluating the

impact on values of small animals at mean land size and mean effective dependency

123

 

 

63:08::— :_ :cuzmoq—u..." Lot-age 05cc... ELSA—ac v.5: treat-w..zwlwmemtlusz CL .2. ~.v.....~l:.dl..l.r~..—. 4; .9. .wtle'.

 

 

.0883 80589 3.58
98 Echo-«E .0883 0mm? :5 >028 0:305 0:80.: Sham-bop .3598 98 $320 £00630 .mma .9005 88% 0:303 20880 :08? .388 c.8950 “Spam
-0:E>w¢0n=m 55 85:28 8:83 08 88:30.8: 3 309552 .0280me6 mo _0>0_ $2 + :5 oogocawfi .«o _0>0_ o\om _.. dunno—mama mo _0>0_ .XL .1. ”8qu

38. Ba 88 .323 Eggsegsm 93 8332 Spam 085m 08m mam-$8308 ”seam

 

 

 

 

8350025 oomm 8283030 :0 89:52
98 8o 30 w _ .o 26 2 .o Bang-m
So :3 inn :8 a: 8d 3:588 so awe-m
8% 8% 8% 8% 8% 8% 8:95.: 08: x 85>on
£8: 33: am: 53 3.: an:
:38- :Nwwd- 3&3 tame $3 985 5388
an: :0: Ed: and: 82: c we
22 82 02:. Sod 33. 83:- mega .2520
3.8 :2: 2...: Gm: 82: 3.:
83- 32- $?- mmmd- ammo- £3- 338 3520
ASS-SE 3.83%
$2: at: and:
92.0- $3.? 3 _ .o- 0388888 295m
32: :3: am. :
Sod- 89o $8- omaameafiaoc 0:2
am: 2 w: ES:
83 +33- wsd- 02388: 295m
A88 82: 5.:
83 immed- iawo- 03883; on:
so: 5.: :2:
:3 :33- m: .0. one;
fine 60.: 5.:
god +32:- +53- 2“:
5:8on ham: «3t 0M90Etl
A": m: E: G: a: 2:

 

a

0:803 Emu-b0

.0335 :95 mo 820>

ac ma: 2230: E 0330

038 mo 820>

889300

 

20:080.— 3 .858.“ ES 00::0w .3 0:80.: Egan-b0 was 800.8 :0 3:022: «5 he 8.3:: 0.5. .v—.m 030,—.

124

ratio, the value of small animals declined by 76.9% in wealthier households experiencing
male head death compared to a 54.6% decline in assets-poor households. This result

might be reasonable because wealthier households have more to spend.

3.4.6. Impact of PA mortality on off-farm income

Ofl-farm income: Previous studies have suggested that off-farm income sources are at
risk among households experiencing PA mortality“), particularly among those that are
asset poor and vulnerable to begin with. In this section I test the hypothesis that a
reduction in PA adults due to prime-age mortality results in a reduction of off-farm
income. Yet the results in table 15, columns D and E, show mixed results that are not
statistically significant for all cases of mortality by gender and position in the household
of the deceased.

Turning to differential impacts by initial household characteristics, Table
3.15, column B, shows that there seem to be differential impacts by initial effective
dependency ratio and by landholding size in households experiencing female head or
spouse death. The statistically significant negative interaction between pre-death EDR in
2000 and female head/spouse death implies that there is a greater decline in off-farm
income among households with higher EDR in 2000 and experienced female head or

spouse death compared to households experiencing a similar shock but having lower

 

40
See Mather et al., 2004, Yamano and Jayne, 2004, Donovan et al., 2003.

125

 

 

8.8 88: 88 am: 2:: 8m:
88 28 :88- :2 8- 2.8- 38- c3888“. 358038 3:88
8.8 88: 88.8 8.8 :8: c m:
888- 28- 88- 88. 38 $8 3: 2% 822850
$8 an: 8.8 88: 88: 8.8
:88- :35? :88 :38 :88 :88 828 0898:": was. €28
:::N 5 83820533 30880: Scum-mam
8.8 88 28: GE: 88: 8.8
R8- 28- :Nmm: :mom._- :88- :ofld- ea 88:55 >5 38
3.: A3. : 5.8 5.8 5.8 8.8
$8 $8 :8 E8 2 8- m: .o- 3.8.8 beam
88: E8 8.: 8m: 5.: 8.:
38- R8- 28- 038- 88- 2 8- 8.9: 3520
REESE 38:85
5.8 3.8 8. : 8.8 5.8 8.8
28- 8 8- 88°- .28- 3.8- 5 .o- 3888388.. 288
8.8 8 .8 8.8 8.8 2 m: c m:
28. 3.8- 88- 88- £8- 88- ”38888.8: 2::
a8: 8.8 8.8 an: 88 A28
~28 38 £8- : 8o. 38 .88 ”3388: 2258
8.8 88: 8.8 A8: 8.: an:
88 28 :88- :208- +58- +88. 03283; 2a:
ASS-BE =53 Qt» mus-05.2%
E m: E: G: Am: 7:

 

a

05005 End-to

.5958 :95. mo 85n>

038 «o 85.3

 

mo w2 85:8 5 0850

8.5.8.60

 

835-88888 586.05 20:08:: 1:35.»: 0:805 Sham-b0 can 308: .8 5:585 «d 80 00895 0.:- .m~.m 03¢.

126

 

8&8
888-
8&8
33
8&8

.Imomd-

8&8
M: 3-

8&8
~88-
8&8
288-
28.8
3% — .o
888
38.8-

8&8
8:8
8&8
~88
2 8.8
N88-
88.8
83.

8&8
N88-
8&8
88.8-
8&8
was
$8
888-

8&8
888
5.8
.88
8 _ .8
288
88.8
88.8

8&8
2 _ .o
8&5
8:8-
8&8
8&8-
8&3

+mm~.o-

8&8
~28
8&8
8&8
28.:
8&8
8&8
:88-

8&8
88.8
8&8
~88-
88.:
E8-
8&8
88.8-

88
888-
88.8
_ :8-
8&8
8:8
8&8
8:8

ocfl 558538 $285.“ :23 850
ova 352.3% $25: :33 680
0:8 zone—Eamon *momsommhugn 288-,—

oufi 3530338308888: 382
83k Aurmhzmmmh gem 3 5qu Vi

cum cgrmuiea =25 880
0N6 287338 :23 550
oflm cam—*mogommxodmoa 2.8th

cum unarmomsoaflmwmon 282
8.;- mssfi E3 88. 8 «8% E

mafia 3.8398380.“ :33 850
3:3 buo>on~§m2mfi :38 550
8388 3.039330%888: £55m

33% 9309833838: 232
33.... 5.63% QQQN ‘3 533 Vt

 

825:8 28 «ES

127

.0883 «.8589 358

98 3588 .0885 own? was Sim 09:05 0883 EFFGOQ 3:58 v5 326 .8813 .mwfi .827. .flmow “6qu 2958 28m“ .388 wauasm “waned
-Sgéopam £5 83328 883 03 88:88am S 33:52 888586 mo 3%: $3 + 98 condo—mama mo $53 gm ... .ooawomawa mo :33 o\o~ I. 8on

33 U5 SON Sufism Enofioaasm wad oooQoma 88m 288$ “mom ngOUEOmU H825m

 

8083080 .«o 895:4

 

 

comm
8o 93 N3 N2 :6 E .o 3:538
who LEN R; £88: 5.88 tegauu 2588
of aid .3; £588 omgacfifim 23
£6 :38 18.2 32388 owaéacazasm €28
m2 N2 33 mg mg 85 $82: 332m
25 as :Ea 12+ 8.“ :4 $82: owfiefi
3.8 3.85

awe awe Gas cod and Ed
~86- ~85- :23 :vmwd .3355 :Ed 2388
8% 8% 8% 8% 8% 8% 88:56 25“ x 33>on

 

60:32.3 mfim Bash

128

888.: 82:83 Echo.“
8:.» 3883 .0883 own? was 8:3. 82:08 0885 Emacs .838 88 86:8 .8830 .88 .808 .flaom 8305 £858 swam“ 988 3858 Based
-833-89=m £5 83328 883 8a 885:83 E 83:52 .ooqaociwmm no _u>o_ $2 + 28 88anme mo 8%: .wi * 688:8me mo 3%: .x: .1. 8on

3% Es 88 Sufism 35585 Ba 8332 8:5 got“: mom 88803308 ”858

 

 

 

129

.--------w£.~ .................. mfi--:-:::-:wmw~ .............. 8 Mom ............. 8.. mm“--------::_.mw.m ..................................... m. fifimmfififiém-
28 as Ed 28 88 one 8338-8
2 .e 2: mg 23 m2 3o 5558 352.".
N3 88 :8; m2 :8 :._ 3:585 08.85

.Sxdwm» $.85
8% 8% 8% 8% 8% 8% 88:58 088 x 8585
£8 :88 A28 608 $8 5.. 8
£88.? :38- :38- 3:3. 83 ago 2588
c ".8 528 A38 A38 E8 88
Sod +om.o MES 33. 38. 8 .o 8358 858
2 W8 88 so. 8 48.8 c as $8
23. 82- 2 no- 22- $8.? wood fig: 885
beaten: bumEm
$28 $8 3.: 938 A88 38
28. 5 .o- :3. 88. M: 3- c: .o- ”888888: 228
2 m8 5.8 E: A88 am: 898
N: .o $3- 88- $8 8.3- $3- ”Seafimfiaoa 2.“:
£08 608 8.: 2 n8 808 A38
32 :3 $3. 83- 83 $3. 0.8882 283
898 SN: 5.8 so: 658 as. 8
:3 wood 10:8- £2- $3. $88- 088888 032
A8 A8 58 Q 88 3 83:3 88m $8 www-mstm
goon :02 Moon “com :02 hoonm Hoom GOZ Hoonm

 

92:85 8.5.80 .3538 :55 .«o 83a> 038 .«o 833/
888.60

ac w£ 3.58: E amass

 

2588 3.88.— %n 30:85:: 5 :33ch 6:: 868m .3 083.: Sufi—nae can 388 .3 5:38.: Sn .8 8:95 2:. 6:” 035,—.

EDR in 2000. Evaluated at mean land holding size and mean wealth status, off-farm
income decreased by 2.06% among households with EDR in the 75th percentile and
having a female head or spouse of household death compared to an increase of 22.5% in
similar households but having EDR in the 25th percentile. In contrast, the interaction
between interaction between landholding size in 2000 and female head/spouse death is
statistically significant and positive implying that there is a smaller decline in off-farm
income among households who had bigger landholding in 2000 and experienced female
head or spouse death compared to households experiencing a similar shock but having
smaller landholdings in 2000.

Evaluated at mean EDR and mean wealth status, off-farm income increased by
21.4% among households with landholdings in the 75th percentile and having a female
male head or spouse of household death compared to a decrease in off-farm income of
23.8% in similar households but having landholdings in the 25th percentile. This result
suggests that the deceased female head was central in sourcing income off the
farmamong households with limited land resources. Their death results in a significant

decline in off-farm income

3.4.7. Does the impact differ if the sample excludes homecoming sick prime-age
household members who died between 2001 and 2004?
The impact of PA mortality on rural welfare may differ between (a) households
experiencing deaths of resident members who were enumerated in the first survey and (b)
households with homecoming sick members who were not residents at the time of the

first survey but rather re-joined the household to receive terminal care between the time

130

of the first and second survey. All other factors constant, one would expect to find more
severe impacts of mortality among households in category (a) than in category (b). For
category (a) households, their asset and income levels, and crop cultivation patterns
reflect the productivity of household laborers in 1999/00, prior to the household incurring
a death, and possible before any members became chronically ill. The effects of having
had at least one adult turn ill and die is measured by the change in household behavior
and outcomes between the 2001 and 2004 surveys. By contrast, among category (b)
households, none of the members enumerated in 2001 became sick and/or died between
2001 and 2004. Hence the changes in household behavior and outcomes as measured by
the differences in the 2001 and 2004 surveys reflect not the effects of mortality of
resident members in 2001, but rather the effects of increased care giving, feeding, and
other kinds of adjustments. While the welfare indictors of households in both categories
(a) and (b) are expected to be adversely affected, the magnitude of the death shock may
differ.

In order to analyze if there are differential impacts between household with PA
mortality of homecoming sick and those with PA mortality of members who where in the
2001 survey, one would need to include two dummy variables: one for households
incurring the death of a resident PA member and one for households taking in a sick
member who was not a resident of the household in 2001 and who subsequently died
prior to 2004. However, these variables would be endogenous (see Beegle, 2004 and
earlier discussion in section 3.3.7.2).

I approach this problem by comparing the results of the restricted sample

(households in category (a) only, with category (b) households excluded from the

131

sample) and results from the full sample as presented in earlier sections. The main
shortcoming of this approach is that we do not have any information about how long the
homecoming sick stayed in the household before death. If they stayed for a longer period
the impact may be similar to those households that had a chronically ill adult in 2000 but
died between 2001 and 2004 since the households had to spend some of their time and
financial resources to take care of them.

Although the sign and statistical significance of the results remain the same for all
models, the magnitude of changes in household size is note-worthy especially for
households experiencing male head death and other males. Therefore, the restricted
sample results examining the impact of PA death on land cultivated, values of cattle and
small animals, crop production and off-farm income are presented in appendix, Tables
A321 to A323. Only the results on impact of PA death on household composition are
discussed here because they are slightly different among households with non-core males
and females’ death.

Table 3.17, Columns A and B shows that the reduction in household size is
greater than unity, for example death of other non-core PA male is 2.02 members and is a
result of a decline in the number of males by 0.89 as well as a reduction in the number of
females, boys and girls of 0.32, 0.36 and 0.55 respectively. Mortality of other non-core
females results in the reduction of household size of 1.46 members as well as a
significant decline in number of boys and girls of 0.31 and 0.34 respectively. Thus, in
addition to the person who died, the net effect on household size was negative, suggesting
that the death of non-core members results in the departure of some members especially

boys and girls. This result is surprising because one would expect boys and girls to

132

.meo 3358..." “waned
$255-59st 55 68.2318 moan: 2m momofiuoaa E muongz .ooaaomawi mo 3%: o\oo~ + was 852.3%? mo _u>o_ gm _.. .oonmomawwm mo _o>o_ XL .2. 5802

voom 98 30m Sufism Egan—gnaw wad ooomRaa >256 «Siam «mom namEOUgOmU bouzom

 

 

 

 

 

o: m cm: 02 m cm: 8 z m on: 8 _ n on: cm: 3 m Seascape co 598:2
3 .o E .o 8 .o Ed 88 85 .2 .o a .o m3 25 853-8
me? my? mo? wow wo> we? mo> mo> mo> we? muggy 083 X noggin
:25- :2 .o- 1.8.0- 3.8.8 .2de :26 :86 :85 2 .o N;
25980
3.: an 8 :38 A38 2 N8 8N8 $8 $8 Gm. : Gm:
NS- «3. 3o- 8.? 12.? its- :5 2 .o $98 .30- 8:88 355
Go: God A2.: A8,: E .8 8N8 93.8 898 and 5.8
+8.8 :3- E .o m; 8.8 8.8 1.2.? 2mg- 23.0- toad- Ema beam
A5385 Krona
5.8 $98 5.8 a: .8 608
£2- 2:2- £2. 85. 13.. a- aaoamesnaon £85m
5.8 2 38 A88 $3 5:8
:28 .98- .63. :38 :2 .N- omaommeafiaoa on:
58 A88 $8 a: .8 6&8
8.8 NS- 13¢- :3 .36- ”games: save
A28 A328 2 :8 A8 .8 2 n8
86. mod 2 .o 23.8 *3? cages; 222
30.8 $8 5 .8 2 m8 8N8
L3- :26- :26- 3o- :8. 7 2§£
85.8 E: A2.: 5 .8 5.8
1.2.8 2.8 :6- 2.2.8 2.3.7 032
A5835 :58 «3t www-mstm
S e :8 68 A8 98 68 g 88 $8
£6 £5 8.85 :53 33: :53. 83530
on? 29.025: 6. .
u no 398:: E owafio

 

2953 683.58% "8332. 2: he .85ch 6:: Saga .3 acEmenEeu 22.8.5: a: 3:522: «d .3 33:: 2:. .26. 033.

133

perform caregiving functions for the sick member. The results for the full sample, as
presented earlier in Table 3.17, show that there is indeed partial replacement of
household members among households with male head and female head/spouse
mortality. However, despite the shortcomings of the data on timing, these findings if
supported by other studies are important for targeting of assistance. These somewhat
point to the need to take into account the different types of mortality in the household.
Since, the impact of the death of a member who was in the household in the first survey
in terms of the direct the contribution to the labor supply and demand and welfare needs
of the household may differ compared to the death of a ‘member’ who comes to seek
short-term care before they die. In order to seek more clarity to this issue, in addition to
collecting information about ill and deceased members, future studies need to consider
collecting more complete information about homecoming ill, for example, when they
joined the household, whether lefi any children, as well as information about remittances
from the deceased. These questions may help disentangle impact from the various

illness-related PA mortality that we are unable to answer with this data.

3.5 Summary of findings

Fully two decades since the HIV/AIDS epidemic in Africa has been characterized
as a major economic development crisis, there remains a dearth of micro-level
information on the impacts of the disease on rural African households and their
responses, although this is fortunately beginning to change. Using comprehensive rural
farm household longitudinal data from Zambia, 1 measured the impacts of prime-age

(PA) adult morbidity and mortality on crop production and cropping patterns, household

134

size, livestock and non-farm income. The paper adopted and extended the counterfactual
(difference-in-difference) approach used by Yamano and Jayne, (2004) by controlling for
initial (pre-death) household conditions that may influence the severity of the impacts of
adult mortality. In particular, I controlled for initial poverty status, landholding size,
effective dependency ratio, and the gender and position of the deceased person.
Moreover, this study extends previous research on this topic by explicitly taking into
account the possibility that prime-age adult death in the household is endogenous.

The study highlights a number of noteworthy findings: first, using prior death and
age group-specific drought shocks as instruments for prime-age deaths’ between 2001
and 2004, the Hausman-Wu chi square test for endogeneity shows that indeed death
variables are endogenous for pooled OLS models. However, after differencing out the
time-invariant unobserved household characteristics, the Hausman-Wu test indicates that
the endogeneity problem is addressed and that OLS estimation using household fixed
effects is appropriate. These findings offer some support for the estimates of earlier
studies using fixed effects, random effects or difference—in-difference (but which didn’t
explicitly test for endogeneity). However, since there are very few, if any, other studies
examining the endogeneity of prime-age mortality when measuring household outcomes,
there is need for further empirical treatment of this issue.

Second, based on the difference models, the study reports the impact of prime-age
mortality on rural households’ welfare. In contrast to the general assumption that HIV-
related mortality is typically associated with household heads/spouses, the survey
findings show that, only 36.6% of households with PA death incurred household

heads/spouse death. While most adults are considered to make important contributions to

135

their families, both materially, in their roles as nurturers and teachers, and in less tangible
ways, it is likely that the most severe economic effects would occur when the death is the
household head or spouse. The fact that less than 36.6% of the prime-age deaths
observed in Zambia’s rural areas involved a household head or spouse suggests that the
potential magnitude of rural PA mortality on rural household agricultural and off-farm
incomes and orphaning rates -- while still very serious -- may be somewhat less severe
than often suggested in the conceptual literature on this topic.

Third, irrespective of gender and/or position in the household of the deceased
person, household size declines by less than one member. This indicates that households
are to some extent successful in replenishing their household sizes, though the net impact
on labor productivity is most likely negative.

Fourth, households’ initial poverty levels affect the relationship between a prime-
age death and changes in household composition. For example, among households
experiencing a male head death, household size changes by -0.30 in non-poor households
compared to -0.84 in poorer households. The implications of this finding are that poorer
households have substantially greater difficulties in coping with the death of core male
household members while non-poor households are likely to almost fully restore
household size to former pre-death levels.

Fifth, both ex ante and ex post the effective dependency ratios of afflicted
households are roughly equal to those of non-afflicted households. Actually, households
incurring prime-age female death (heads/spouses and other females) between 2001 and
2004 experienced a decline in mean effective dependency ratio in 2001 of 0.16 and 0.35,

respectively, suggesting that households are partially able to restore their dependency

136

ratios, by attracting older girls into the household with initial high EDR to assist in caring
for other children, the elderly or the sick when other females die. Despite the fact that
women in households experiencing death may well face increased demands on their time
for domestic tasks and crop production out results indicate that the relative burden of
dependents in relation to healthy adults is not much different for non—afflicted and
afflicted households.

Sixth, the net effect on household size among poor households experiencing male
head or spouse death is negative but not unity, implying that there is some partial
replacement of household members. In contrast, under the same circumstances non-poor
households are able to attract boys and girls into the household.

Seventh, the effects of PA death on farm production are sensitive to the gender
and position in the household of the deceased. For example, the death of a PA male
resulted in a 13% decline in total land cultivated whilst death of a PA female resulted in a
5% decline of cultivated land but the impact is not significant at the 10% level. After
disaggregating mortality by gender and position in the household, I find that the death of
male heads/spouses resulted in a 28% reduction in land cultivated. All the other
mortality categories are negative but not statistically significant. This finding seems to
follow from the earlier reported finding that households experiencing male heads/ spouse
death tend to incur a higher decline in household size. Without full replacement of
household members, land cultivation is cut back to cope with the labor shortage.

The area under cereals and other food crops declines among households
experiencing male or female head/spouse death and other females’ death. However, the

impact is more severe among households with male head/spouse death (a decline of

137

19%), followed by the death of other males and female heads/spouses of household (a
decline of 11% and 8% respectively). Also, the death of male core members is associated
with an additional 4% decline in high-value crops.

One would find these results a bit puzzling given that women devote more labor
hours to agriculture than men. However, in about 33% of the cases among households
experiencing a male head of household death, the widow ended up cultivating less land.
This could be due to loss of land, capital and livestock assets to other relatives after the
death of their husband. We also find that relatively wealthy widow-headed households
are particularly vulnerable, as they have more land and assets that can be claimed by
relatives than afflicted households that are poor to begin with. (see Chapoto and Jayne,
2006-forthcoming). Therefore, the severe reduction in land cultivated that we are seeing
among households experiencing male head of household may be due to the loss of labor
as well as decapitalization of agricultural production among widow headed households.

The implication of this finding implies that the responses to mitigating the social
and economic impacts of HIV /AIDS in Zambia may not be successful if they ignore the
gender inequalities that that exist in terms of land access and other productive assets
important for rural livelihood. Therefore, efforts to safeguard widows’ rights to land
through land tenure innovations involving community authorities may be an important
component of social protection and poverty alleviation strategies.

Eighth, in contrast to the general hypothesis that households experiencing prime-
age death cope with the reduction in family size by switching to labor saving crops such
as roots and tubers, I find little evidence of this. The death of other females in the

household results in a 5% decline in area under roots and tubers. The death of all other

138

person categories has no statistically significant impacts on roots and tuber cultivation.
These findings indicate that afflicted households appear to be no more likely to switch to
these less labor-intensive crops than non-afflicted households.

Ninth, the findings of this study seem to be going against conventional wisdom
that suggests initial poverty exacerbated the impact of mortality on cultivated area. Our
results indicate that the area under cereals declines most among wealthier households
incuning male head-of-household mortality (-75%) compared to -35% among households
in the same situation but in the bottom 50% of the 2000 assets distribution. In contrast,
the impact is almost the same among both poor and non-poor households experiencing
female head/spouse of household death. Similar, to point 7 above, one may find this
finding puzzling considering that women devote most of their time to agriculture than
men. However, as mentioned above it is likely that widow are losing their land and
productive assets rights after their husband death and relatively wealthy widow-headed
households are particularly vulnerable, as they have more land and assets that can be
claimed by relatives than afflicted households that are poor to begin with.

Tenth, in terms of value of crop output and gross output per hectare, the results of
this study do not strongly support the contention that households incurring prime-age
death suffer huge declines in crop output except among poor households experiencing PA
male head death. Gross value of output actually went up by 32% in wealthier households
experiencing male head/spouse death (admittedly a finding that is hard to explain) whilst
among poorer households gross value of crop production declined by 19%. There is
evidence to suggest that wealthier households incurring male head of household death

attract boys and other males to join the household. Also since there was a net decline in

139

land cultivated in both cases due to male head death these results seem to suggest greater
agricultural intensification in wealthier households experiencing male head death.

Eleventh, these results indicate that initial landholding size greatly influences the
change in the value of crop output due to male head death (and on gross value of crop
output per head due to either male and female head or spouse death). Evaluated at mean
effective dependency ratio and mean wealth status, gross value of output decreased by
57% among households with land holding sizes in the 75th percentile and having a male
head of household death compared to 41% in similar households but having land holding
sizes in the 25th percentile.

Twelfth, values of cattle appear to suffer greatly from the death of a PA male head
of household (30% decline) whilst the impacts of death of other household members are
negative but not statistically significant. These results seem to suggest that households
appear to try to hold on to productive cattle and probably sell only as a last resort but the
result does not seem to strongly hold among households incurring male head of
household death. In contrast, there is strong evidence to suggest that afflicted households
liquidate small animals to mitigate the impact of PA death. For example, disaggregated
by gender and position in household of the deceased our results show that death of male
head of household resulted in 65% decline in values of small livestock, whilst death of
other females’ death and female head/spouse resulted in a decline of 44% and 37%.
Also, the value of small animals reduced by 77% in wealthier households experiencing
male head death compared to a 55% decline in assets-poor households.

Thirteenth, contrary to the hypothesis that off-farm income sources are at risk

among households experiencing PA mortality, particularly among those that are asset

140

poor and vulnerable to begin with, results of this study show mixed results that are
statistically insignificant for all cases of mortality by gender and position in the
household of the deceased. However, evidence does point to differential impacts by
initial household effective dependency ratio and by landholding size among households
experiencing female head or spouse death. For example, evaluated at mean land holding
size and mean wealth status, off—farm income decreased by 2% among households with
EDR at the 75th percentile and having a female male head or spouse of household death
compared to an increase of 23% in similar households but having EDR at the 25th
percentile. Off-farm income increases by 21% among households incurring the death of
a female head/spouse with landholdings at the 75th percentile. In contrast, off-farm
income declines by 24% in similar households incurring a female/spouse death but whose
landholdings are at the 25th percentile. This result suggests that the deceased female head
was central as a source of off-farm income especially among households with limited
land resources and their death resulted in a significant decline in off-farm income.

Last but not least, there is some evidence that warrants taking into account the
different impacts on household welfare resulting from households experiencing deaths of
resident members versus households with homecoming sick members who joined or re-
joined the household to receive terminal care. Further study is needed, based on

carefully constructed survey information about ill and deceased members.

141

CHAPTER 4

CONCLUSION AND POLICY RECOMMENDATIONS

Using nationally representative panel data on 18,821 individuals surveyed in May,
2001 and May, 2004 in rural Zambia, Chapter 2 of this paper identified important ex ante
socio-economic conditions of individuals who die of disease related causes between the
ages of 15 and 59 years. The findings of the study can help policy-makers and
development agencies better understand current transmission pathways of HIV /AIDS, as
well as help in the formulation of up-to-date AIDS prevention and mitigation strategies.

Overall, the probability that a prime-aged (i.e., 15-59 year) woman would die of a
disease was roughly 1.0 percent over the 3-year period, while the comparable probability
for men was 0.6 percent. Just over 60% of the prime-age deaths observed in this
nationally-representative rural sample were women, supporting other findings that
women are being disproportionately afflicted by HIV/AIDS.

For both asset-poor and non-poor women, the probability was roughly 1.0 percent
whilst for asset-poor and non-poor men the probabilities were 0.5 and 07/08 percent,
respectively. Consistent with findings in the 19803 and early 19903, we find that men in
the upper half of the assets or income distribution are more likely to die of disease-related
causes than men residing in poor households. In contrast, women in the lower half of
assets or income distribution are equally likely to die of disease-related causes as women
residing in the upper half of assets/income distribution.

An emerging strand of the social science literature on HIV /AIDS in Africa

stresses the relationship between poverty, risky sexual behavior, and subsequent

142

 

contraction of the disease. It has been argued that single women unable to sustain
themselves through wage labor or agriculture are more likely to resort to transactional sex
for survival. If this is an important social pathway contributing to the spread of the
disease in Africa, then we might expect to find a relationship over time between
household- and individual-level indicators of poverty, especially for single women, and
subsequent chronic illness and death. We find that relatively poor women who have some
form of formal/informal business income are less likely to die of disease-related causes
than women with the same characteristics and no formal/informal business activity. This
finding suggests that efforts to provide greater income-earning opportunities for poor
women may make at least a modest contribution to reducing female prime-age mortality.
This relationship does not hold, however, for relatively non-poor females. And 47.2
(45.0) percent of the women dying of disease-related causes over the 3-year survey
period came from households in the top half of the asset (income) distribution. These
findings suggest that the social factors driving the spread of AIDS are considerably more
complex than simply poverty-based explanations, although poverty may certainly
contribute to risky behavior and poor health, which are important pathways by which the
disease is spread.

Single women and men in poor households are twice as likely to die of disease-
related causes as poor women and men who are the heads or spouses of their households.
Single women and men in relatively non-poor households are 3.7 and 4.5 times more
likely to suffer a disease-related death compared to married women and men who are the
heads or spouses of their households. Irrespective of gender, individuals who spend

several months or more away from home are 2 to'10 times more likely to die of disease-

143

r‘Ir-v‘. -Vrlum '-

K - ‘.fl“-T

 

related causes in succeeding years. It is possible that the creation of business
opportunities that involve men and women spending more time away from home for
extended periods may exacerbate the AIDS problem in rural Zambia and negate the
positive effects of greater financial independence for women, unless progress is made in
the use of condoms, other forms of safe sex, and prevention interventions.

Educational attainment was found to be largely unrelated to vulnerability to death

 

Is
for men. For women, the evidence is not robust, but the data tend to show that
educational attainment reduces somewhat women’s vulnerability to disease-related death, ;
especially for non-poor women. This finding suggests that education is an empowerment r
tool for women and helps reduce mortality of females and should be encouraged in rural i,

Zambia. Also, HIV/AIDS education campaigns should still target both literate and
illiterate men because men of any education level have roughly the same risk of
contracting HIV.

Most importantly, the prior death of a prime-aged person in the household
substantially increases the probability of another prime-aged member dying. Irrespective
of income status, prime-aged men and women experiencing a prior death in their
household are 23.0 and 18.1 times more likely to die of disease-related causes than men
and women in households with no prime-age deaths in the past 8 years. The predicted
probability of death was 12.4% and 16.3% for men and women experiencing a prior-
disease-related death in their household in the past 8 years versus 0.54% and 0.90% for
men and women not experiencing a prior prime-aged death. Of the 362 households
experiencing prime-age mortality between 2001 and 2004, 15% of them suffered multiple

prime-age deaths. In this way, AIDS differs from other kinds of diseases (e. g., malaria),

144

which do not appreciably raise the likelihood of subsequent death in the family after one
member contracts the disease. To the extent that the death of two prime-age members
from the same household within a few years of each other causes extreme hardships on
remaining members, especially for children, the implication of this finding is that
programs and strategies to support the care and education of children in AIDS-afflicted
households may need to become a critical component of poverty reduction strategies in
areas hard-hit by AIDS, such as most of eastern and southern Africa. More research is
necessary to understand the longer-term impacts of the disease on household behavior
and welfare, and to develop programs that can mitigate the adverse consequences. At this
point in time, the research community still knows very little about the cost-effectiveness
of alternative ways of mitigating the impacts of AIDS, but a solid understanding of the
socio-economic factors associated with the disease is likely to help considerably in
designing appropriate risk messages and prevention strategies.

If resources for ARV therapy remain highly constrained in the foreseeable future,
ARV therapy may be prioritized for single parents in households where the spouse has
already died. This option may help to reduce the strains on communities caused by
double-orphans and help nuclear families remain intact as best as possible. Given the
long period between infection with HIV and the transition to full-blown AIDS and death,
these findings reflect patterns of infection 4-8 years ago. It is critical to continue
assessing the factors associated with AIDS-related mortality to understand how they may
be shitting over time, in order to develop appropriate risk messages and public campaigns

targeted at high-riSk groups.

145

 

Chapter 3 of this study estimated the impact of prime-age mortality occuning
between May 2001 and May 2004 on household composition, land cultivated, changes in
cropping patterns, value of livestock assets and off-farm income. Based on the difference
models, the analysis yields a number of noteworthy findings important for AIDS policy
and mitigation strategies for rural farm households. First, in contrast to the general
assumption that HIV -related mortality is typically associated with household
heads/spouses, the survey findings show that, only 36.6% of households with PA death .1
incurred household heads/spouse death. While most adults are considered to make

important contributions to their families, both materially, in their roles as nurturers and

 

teachers, and in less tangible ways, it is likely that the most severe economic effects E
would occur when the death is the household head or spouse. The fact that less than
36.6% of the prime-age deaths observed in Zambia’s rural areas involved a household
head or spouse suggests that the potential magnitude of rural PA mortality on rural
household agricultural and off-farm incomes and orphaning rates -- while still very
serious -- may be somewhat less severe for many households than ofien suggested in the
conceptual literature on this topic.

Second, irrespective of gender and/or position in the household of the deceased
person, rural farm households in Zambia attempt to cope with the death of PA adults
through changes in household composition. In all cases household size declined by a
factor less than one member, suggesting that afflicted households are partially successful
in replenishing their family size. However, in response to the death of a male household
head, poorer households have substantially greater difficulties in coping than non-poor

households, which are likely to almost fully restore household size to former pre-death

146

levels. These results imply that the widespread view that death of productive members of
the family results into labor shortages needs to be more carefully nuanced, taking into
account the position of the deceased person and the initial conditions of the household.
Nevertheless, the loss of adult members, especially heads and spouses, may have longer
run impacts not measured in the relatively short three-year period of this analysis, such as

the loss of inter-generational knowledge in terms of farming skills and knowledge,

 

I.

especially in households that are poor to begin with and experience the death of a male i

household head. This may require the targeting of assistance and skill training to

relatively poor households headed by widows. I;
Third, both prior to households’ experiencing adult mortality and after, the E

effective dependency ratios of afflicted households are roughly equal to those of
non-afflicted households. Effective dependency ratio is defined as the number of
children, elderly, and chronically ill persons divided by the number of healthy prime-age
adults (between the ages of 15 and 59). Actually, households incurring prime-age female
death (heads/spouses and other females) between 2001 and 2004 experienced a decline in
mean effective dependency ratio in 2001 of 0.16 and 0.35, respectively, suggesting that
households are partially able to adjust dependency ratios, by attracting older girls into the
household with initial high EDR to assist in caring for other children, the elderly or the
sick when other females die. Despite the fact that women in households experiencing
death may well face increased demands on their time for domestic tasks and crop
production, these results indicate that the relative burden of dependents in relation to

healthy adults is not much different for non-afflicted and afflicted households.

147

Fourth, the effects of PA death on farm production were sensitive to the gender
and position in the household of the deceased. For example, death of a PA male resulted
in a 13% decline in total land cultivated whilst death of a PA female resulted in a 5%
decline of cultivated land and the death of male heads/spouses resulted in a 21%
reduction in land cultivated. All the other mortality categories are negative but not
statistically significant. This finding appears to be consistent with the findings above that
households experiencing male heads/ spouse death tend to incur a higher decline in
household size, and thus may experience greater declines in land cultivation associated
with household labor shocks. Also, the results show that land cultivated and area under
cereals decline more in relatively non-poor households than poor households after the
death of a male household head. These results seem to suggest that despite the fact that
the death of productive members may result into a reduction in the total cultivated land,
the position and gender of the deceased member matters.

Fifth, in contrast to the widespread view that households experiencing prime-age
death cope with the reduction in family size by switching to labor-saving crops such as
roots and tubers, the results show positive but statistically insignificant effects on the
cultivation of these crops except among households experiencing the death of non-

head/spouse females. The death of other adult women in the household results in a 5%
decline in area under roots and tubers. These findings indicate that afflicted households
are not more likely to switch to these less labor-intensive crops than non-afflicted
households. While many have identified HIV /AIDS as the cause of recent shifis in area
cultivated from maize to less labor-intensive root and tuber cultivation in Zambia as well

as other parts of southern Africa, it is important to acknowledge that recent crop and

148

input policy changes in the region associated with structural adjustment and food market
reform have affected the relative output/input price ratios for grain crops relative to roots
and tubers, reducing the profitability in some areas of grains as compared to roots and
tubers. The potential for such national and/or regional cropping trends exemplifies the
importance of investigating cropping patterns of afflicted households in comparison with

the non-afflicted population. These results suggest that for afflicted households as a

 

I.
group, the loss of family labor due to a death in the household may not necessarily mean ‘

t
that agricultural labor becomes the limiting input in agricultural production (any more so I
than capital assets, for example, which are likely to be drawn down due to foregone i
income, medical treatment, and funeral expenses among afflicted households). The E.

macro-level picture emerging from recent demographic population projections, which
include the impact of AIDS-related deaths, demonstrates that although the epidemic will
reduce life expectancy and population growth considerably in the hardest-hit countries,
the epidemic has not caused a decline either in the aggregate labor supply or in the labor-
to-available- land ratios in agriculture. In fact, between 1990 and 2000, the rural
population of Zambia has grown at a considerably faster rate than the overall population
— 43.6% vs. 33.9%. Therefore, prioritization of public sector investment in the
development and dissemination of technologies aimed at mitigating the effects of prime-
age adult mortality ideally requires in-depth evaluation of household constraints and
opportunities, as well as consideration of the need for balance between investments in
long-term rural economic productivity growth and targeted assistance to both afflicted
and non-afflicted households. Assessing which labor-saving technologies to prioritize

should involve investigation of the characteristics of affected households, whose labor

149

time is most constrained, who would benefit from these technologies, who has effective
access to new technologies, and which technologies promote efficiency of allocation of
public resources across sectors.

Sixth, in terms of value of crop output and gross output per hectare, the results do
not strongly support the contention that households incurring prime-age death suffer large
declines in crop output -- except among initially poor households experiencing the death
of a male household head. Among this group of afflicted households, the gross value of
crop production per hectare declined by 19% relative to non-afflicted households. There
is evidence to suggest that wealthier households incurring male head-of-household death
attract boys and other males to join the household, while initially poor households have
greater difficulty in doing so. This finding supports the need for creating or and/or
strengthening community-based networks to assist poorer households experiencing
mortality of household heads and spouses. Government and interested donor agencies
may also assist with agricultural extension programs to reach afflicted poor households in
order to strengthen their capacity to cope with the loss of prime-age core members.

Seventh, the value of cattle assets appear to suffer greatly from the death of a PA
male head of household whilst the impacts of death of other prime-age members are
negative but not statistically significant. Similar to the findings by Yamano and Jayne
(2004), there is strong evidence to suggest that afflicted households liquidate small
animals to mitigate the impact of PA death. The sale or liquidation of livestock as a
means of coping with illness and death of prime-age adults is costly in the short-term and
may also compromise the household’s fiiture livelihood (Stokes, 2003). Another possible

explanation why the results show a significant decline in cattle assets among households

150

experiencing male head-of-household mortality is that property of the deceased man
(including cattle) is often redistributed to surviving relatives. Cattle assets are not only a
stock of wealth but are also an input into agricultural production (through draft power for
land preparation) which in some cases raises the average product of other inputs such as
fertilizer (Xu et al., 2005). Therefore, in order to support the food security and farm
productivity of households afflicted by male head mortality, the Zambian govemment
and development agencies may consider targeting households whose capital base is
affected by AID-related illness and death as well as encourage cultural changes that
empower widows who need not be pushed into poverty further by assets redistribution
after their husband’s death. Also, programs similar to the ‘Heifer project’ may need to be
targeted to poor households and especially households with male head of household
death. 1

Eighth, the study shows mixed findings in terms of the impact of PA death on off-
farm income. Contrary to the hypothesis that off-farm income sources are at risk among
households experiencing PA mortality, particularly among those that are asset poor and
vulnerable to begin with, the results were statistically insignificant for all cases of
mortality by gender and position in the household of the deceased. However, evidence
does point to differential impacts by initial household effective dependency ratio and by
landholding size among households experiencing female head or spouse death.
Households with female head or spouse female death and having higher effective

dependency ratios seem to suffer more compared to those with lower initial dependency

ratio.

151

Ninth, the study points to the potential need to take into account the different
types of mortality in the household, especially the potential differences in households
welfare between households experiencing deaths of resident members versus
households with homecoming sick members who joined or re-joined the household to
receive terminal care. The impact of the death of a member who was always resident in
the household in terms of the direct the contribution to the labor supply and demand and
welfare needs of the household may differ compared to the death of a ‘member’ who re-
joins the household to seek short-term care before they die. In order to seek more clarity
to this issue, future studies need to consider collecting more complete information about
the homecoming ill, for example, when they joined the household, whether they left any
children, as well as information about remittances from the deceased. These questions
may help disentangle impact from the various illness-related PA mortality that we are
unable to answer with this data.

Overall, the results of this study question the usefulness of a homogeneous
conceptualization of “afflicted households,” especially in the context of proposals for
targeted assistance, technology development, and other pro grams/policies. In most cases
the gender and household position of the deceased appear to strongly condition the
effects on the household. The death of a male household head is associated with larger

hegative impacts on household size, farm production and livestock assets than any other
kind of adult death. In addition, the results show that initial asset levels, land cultivated

and initial effective dependency ratios also condition the effects of mortality on
households. In general, the impact of adult mortality appear to be most severe for

households in the bottom half of the distribution of assets in 2000. Overall, these

152

 

findings suggest that poorer households headed by HIV/AIDS widows are in especially
precarious positions. This would imply that AIDS mitigation pro grams should target
their scarce resources particularly toward widow-headed households, especially those that
were relatively poor to begin with.
Also, the results in this chapter show more severe impacts on household land
cultivated among households experiencing male head of household death. One would the
find this result a bit puzzling given that women devote more labor hours to agriculture 1
than men. However, in about 33% of the cases among households experiencing a male ,

head of household death, the widow ended up cultivating less land. This could be due to

 

loss of land, capital and livestock assets to other relatives after the death of their husband. E
We also find that relatively wealthy widow-headed households are particularly
vulnerable, as they have more land and assets that can be claimed by relatives than
afflicted households that are poor to begin with. Therefore, the severe reduction in land
cultivated that we are seeing among households experiencing male head of household
may be due to the loss of labor as well as decapitalization of the agricultural production
among widow headed households.

The implication of this finding implies that the responses to mitigating the social
and economic impacts of HIV/AIDS in Zambia may not be successful if they ignore the
gender inequalities that exist in terms of land access and other productive assets
important for rural livelihood. Therefore, efforts to safeguard widows’ rights to land
through land tenure innovations involving community authorities may be an important

component of social protection and poverty alleviation strategies.

153

Caveats and limitations: It is important to take note that the findings from this
study only measured short run effects of prime-age mortality between April 2001 and
April 2004 on a few aspects of Zambia rural farm households. Future research studies
need to be designed in order to measure full long-run effects of prime-age adult death.
This would entail tracking affected households over a long time frame.

Also, the situation in which a relatively small percentage of households incur a
shock, but the shock is spread across households in a community presents methodological
challenges for estimating the full effects of the shock using household survey data. This
study and most prior household-level panel studies, using difference-in-difference,
household fixed-effects, or random-effects models, have measured the effects of
mortality in afilicted households on differenced household-level outcomes, typically over
a 2—5 year time frame, compared to differenced outcomes on non-afflicted households.
Yet if non-afflicted households are likely to be indirectly affected by the mortality
occurring around them, non-afflicted households may not be a valid control group. In
communities hard-hit by HIV /AIDS, households not directly incurring a death may
nevertheless be affected by taking in orphans, losing access to resources owned by kin-
related “afflicted” households, intra-household resource transfers to afflicted households,
and broader effects of high mortality rates on communities’ economic and social

structures. Future studies may need measure the effects of mortality on rural welfare

other than at the household level.

154

APPENDICES

155

 

Table A2.1. Descriptive Statistics of variables used in the analysis

 

Value of variable at 1r"I percentile

 

 

 

Variable Mean Std Dev Min Max 111 m m
25 50 75
Gender (1=male, 0 otherwise) 0.49 0,50 0 1
Head/spouse (=1 , 0 otherwise) 0, 57 0,50 0 1 - - -
Marital status (=1)
Currently married 0, 56 0, 50 0 1 - - -
Never married 0.36 0.48 0 1
Previously married 0.08 0.28 0 1
Age groups in 2000 (=1)
Age 15-19 0.23 0.42 0 1 '
Age 20-24 0.18 0.38 0 1 - - - =
Age 25-29 0.15 0.35 0 1 - - -
Age 30-34 0.11 0.31 0 1 - - -
Age 35-39 0.09 0.29 o 1 - - -
Age 40-44 - - -
Age 4549 0.08 0.28 0 1 - - -
Age 50-54 0.06 0.24 0 1 - - -
Age 55-59 0.06 0.23 0 - - -
Years of education (=1)

None 0.15 0.35 0 1

1'3 Years 0.12 0.32 0 1 - - -

4-6 Years 0.25 0.43 0 1 - - -

7 years 0.24 0.43 0 1 - — .-

8 years and above 024 0,43 0 1 - - -
Salary wage income in 2000 (=1) 0,09 0,29 0 1 - - -
Formal/Informal business activity (=1) 0_ 14 0,3 5 0 1 - - -
Months spent away from home in 2000

Resided all months at home (=1) 0,37 0,3 3 0 1 - - -

Spent one month away (=1) 0,04 0,19 0 1 - - -

Spent two or more months away (=1) 0,09 0,23 0 1 - - -
Polygamous household in 2000 (=1) 0,13 0,34 0,00 1,00 - - -
Prior death in 1996-2004 (=1)‘ 0,07 0,25 0,00 1,00 - _ -
Land holdings (hectareS) in 2000 3.39 3.30 0.03 20.00 1.13 2.25 4.50
Productive assets ( ‘000 ka) in 2000b 814 2502 000 42000 7 50 314
District HIV prevalence rate in 1999 15.88 3.25 10.8 31.0 13.34 15.90 17.20
District on the line of rail (=1) 0.36 0.48 000 1,000 - _ -
Distance to the nearest tarmac road (km) 23.41 33.23 0.00 188.90 3.60 10.40 27.50

Distance to the diStl’iCt TOWN/BOW (km) 33.83 22.48 0.30 104.40 15.40 28.80 46.50

Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004
Notes: 'Refers to other adults ages 15 to 59 in household who died up to 8 years before the individual under
analysis. bProductive assets are the sum of the value of farm equipment (scotch carts, barrows and ploughs)
and livestock.

156

Figure A2.1. Distribution of illness-related prime-age mortality by gender and age-

groups between 2001 and 2004

 

Number of deaths (2001-2004)

 

801

70

 

60

 

50

40

30

20

10

 

 

 

15-19

20-24

25—29 30-34 35—39 40-44 745-49

Age group of deceased

 

 

—e— Female +Male - - -A- - - Total

 

 

50-54

55—59

 

 

157

tee—53w
I i, .

1 Fa ms. A‘ Isl-I.

 

 

 

 

 

3.3 3.3 3.3 33 3.3 9.3
:83 :33 :33 :23 :283 :R3 332 om<
3.3 3.3 3.3 $03 3.3 3.3
:23 .23 :33 :33 :~83 :33 «3% ow<
3.3 3.3 3.3 3.3 :23 3.3
ammod .1. _ mod aamomd *amw ~ .0 «*Nood fawnod 91m? ow<
3.3 3.3 5.3 5.3 3.3 3.3
:33 :33 :83 :33 :R3 :33 3.2. “.3
:3 3.3 3.3 3.3 84.3 5.3
:33 :33 :33 :33 :83 :33 «.3». um<
63.3 63.3 3.3 A: .3 3.3 $3
:33 :33 :33 :33 :83 .383 3.8 8?.
$3 803 3.3 3.3 $33 3.3
sieved atvod «*mwod 43050.0 aamvod siwmod omfim ow<
3.3 3.: 5. 3 3.: 3.3 3.3
.33 33 33 .23 :_ _3 2L :3 3-8 ow<
:03 gen E 3.8..» and.
:33 :33 833 833 see 3.3
83- S3- 33- 30.0- 83- 83- :13 Bras 8:2
and 34.3 5.: 3.3 3.3 3.3
:w _3- .33- :3- 23- :83- :83- :13 BEE $556
:03 and an: 5.: 3.3 9:3
.383- :23- 83- 83- :_ :3- :23- cnv 88338:
- - - - 3.3 3.3
- - - - :33- :33- 38313 £21326
3 A3 A3 3 a 3
Gorga flotsam near—flu
no.“ 6300500 How vouooboU How “532.50
£95m £ng 323m

 

Amwoobo uoxm-omm__§v 83% 43 38-88 5053 5:888 <m mo «3:58 3308 :35 .N.N< 2an

158

eat.

 

:33 84.3 8. 3 3.3 3.3 3.3
88.8- 83- 83 83 83 83 :13 3an 83 588m
883 2 m3 3.3 8. 3 8.3 3.3
83 83 83 83 83 83 :13 88 as 858 8% 388m
33 3.3 3.3 34.3 8.3 3.3
83- 83- 83 83 83 88.8- :13 .8248 8% 8:88
33 3.3 8.3 $3 3.3 :33
83 83 33- 33- 83- 83- :3 328: 888. .5388
33 A83 3; .3 3.3 5.3 3.3
33- 83- .83 33 83 83 :3 838 583 8282on
3.3 3.3 3.3 8.3 3.3 3.3
83- 83- 83- 83- 83- .83- 888 a 8583 8:. ”3338835
833 833 :33 333 A883 A883
:888 :83 :83 :83 :§.o :83 23.88- 83 a 28088 a 588 <8 8.5
84.3 8.3 3.3 3.3 3.3 3.3
83 83 83- 88.8- 83- 83- 21388 a 22825 388:8
3.3 3.3 3.3 3.3 3.3 8.3
:83- :83- 83- 88.8- :83- :3 8.8- 213 88 a Dec: 83 8% 8% €58 25
5.3 2:3 5.3 8.3 3.3 833
:38- :83- :83- :83- :83- :38- :L 88 a so» as 3885 088 a 828m
:33 3.3 5.3 3.3 833 5.3
83- 83- 83 83 83- 83- 3213 88 a 8%: 883m
:33 A33 2 33 3.3 3.3 3 .3
83- 38- .33 .L _3 83 83 :13 88 a 2:82 883 Sam
3. 3 3.3 3.3 3.3 3.3 3.3
33- 33- 83 83 83- 83- 268 85 as»; w
c 3.3 88.3 3. 3 A83 33 3.3
23- :3- :3 33 83 83 3% 8
3.3 3.3 3.3 8.3 3.3 333
:3- : :3- £3 33 83- 83- 3.38 3
3.3 2 N3 3. 3 3.3 8.3 8.3
83- 83- 33 :3 83 88.3 as; m;

:13 88 a 882.3 8 38>

8:588 8.3 28.8

159

hurl...-

 

cfllcd ADHUfiLw

88388 83::

832335 8: Bowen 88% w 8 a: 306 on? Bone-mac: E on 8 2 8mm “£38 850 8 flow-3r .muurfioa 80885 EEO-«E Ho EEO”: ._o>o_ cacao—mama 83qu

m 38235 _.. :32 832-38me 3853 ~ .1. 823830: 83 3883—0 888 33338 8:38 583833888: wfim: 3823—8 .mouoom-N 8203.3 8802

.38 was SON $3.5m Ecofioaasm 33 033332 >255 $3.83 “mom ngOUxJEOmU u350m

 

 

88 88. 88 88 88 88 808.580 8 85:2
85 8% mu.» m“; mo> mo> 3328: 88:56 qu=>
33 53 2:3 83 A83 A23

83 83 83- 83- 83 83 3.085

63 8.3 3.3 323 A83 83

83 83 88.8- 83 88.8 83 88-22

8.3 A83 A83 83 A8. 3 3.3 38: 283:8
83 88.8 83 83 83 88.8 33-22 8 as? 8 8380

 

8:588 8.? as:

160

 

 

 

 

 

33 ans 93.3 are 3.3 3.3
:33 :23 $983 $333 1.93 2.2-3 3-3 om<
3d 3d $9 3.3 3.3 83V
.83 .23 2383 .383 :33 :33 v2; ow<
§ .3 Sad 3.3 853 3.3 G~$
.33 .33 £83 :33 :33 2383 91.3 ow<
6th 5.0 54¢ 33¢ 6&3 3.3
aim-mod tame-o IQOQO **on.o «*wvod *{Qmod 310v ow<
$3; 3.3 3.3 393 65.3 and
*vaod tun-mod .cummod {*omod 3:36 *§vmo.o amfim ow<
2va 3.3 83V 3.8 c we 3.3
aim-v06 sicmcd .ILNQO *anwfiod Sum-no.0 sin-Nod VMuOM uw<
5.3 3.3 5.3 83v 60.8 3.3
**OM0.0 {*ONod fLMOd ainod ELMO-o Eek-Nod mNnmN ow<
S .3 God 5.: Gm: and 5.8
L :3 :83 S3 B3 .83 $83 3-8 ow<
Cuv 25m 5 3:9.» «wt.
3.8 #08 E9 82: 63v 23V
83- 83- 83- S3- 83- 83- :uv BEE 5:2
23V and 8m: and 92V 32v
:2 3- :83- 83- B3- .83- ...83- Al BEE €850
ans ans at: 5.8 3d 5d
.23- 38? 83- S3- :83- :83- :nv 0309.35
- - - - 3.23 and
- - - - 2.083- 333- Esfiuo .uEEEEuo
6v 5 € 8 S 3
GOEEM Gotta-.3 «HOE-bum
How 6303.50 .5.“ deuce-50 How unwoquU
225m 2“: 295m 38;

 

AWN—00:0 fivN—hlouflTrchs human-0M ha— chufiCGN Hm bm-NHHOE audios—Sham HQ am—QUGE quOHH .M.N.< 0—DNr—L

161

.
l‘. I!

 

A88 88 88 ~88 A2 8W 82%
83 83 .88. 83- 888 83- 885
8.: 8.8 3.8 $8 :98 A88
83 83 83 888 83 83 85-22
8.8 5.8 88 A88 A8: 3.8 988 28088
83 83 83 ~88 N88 83 33-22 8 use, 8 £930
88 A38 38 3.: A88 88
.888- 888- 83- 83- .83- :83- 883 @2988 8% ”caséafivs
:3: 838 a3: 5.8: $38 628
.828 :§.o :58 .2838 :58 2.83.0 2888. 88 a 283:8 a .88 8 BE
9&8 :88 8.8 8. 8 8.8 88.8
83 83 83- 83- 83- 83- 2888 a 28088 88838
A88 A88 A88 A88 5.8 5.8
.888- :23- 83- 83- .888- :83- :nv 88 5 98o; Bow 8% 80% £88 25
8.8 A88 5.8 8.8 A88 A88
.888- ...888- :88- :23- .....8o.o- .888- :uv 88 a a...» ”a 58825 0&8 a 323m
38 2 08 9:8 8.8 8.8 5.8
83- 83- 83 83 83- 83. ..:L 88 a 888 30:55
8.: 3.8 A8. 8 8.: A88 88
83- .83- 83 83 83 888- cuv 88 a 83% 083 D28
:08 8.8 Ed 88 88 A88
83- 83- 83 83 83- 83- 28a 8“ ”3% w
5.8 :9: 5.8 8.: 3.8 E .8
83- 83- L :3 >88 83 83 23% n
8.8 A8. 8 35: an: 8.8 $08
83- 83- 83 83 888- 83- as: 3-
:88 8.8 8.: 5.8 :98 $8
83- 83- :3 888 83 83 3% E

:uv 88 a 833:3. 8 28>

6253—80 .m.N< 035,—-

162

 

IL“

 

 

88358 .593 flaw—>65 05 ago-«on 88% w 8 a: 36 on? 20:88: 8 mm
8 2 wows 838m 650 2 who-Homo .moEZuo-m 80585 Ego-HS Ho 88.8”? 432 832838 Eco-Em m 8:365 *. 80>»: oogomaflm
6858 _ 85868 .2. 8205025: How 35830 88.5 @8838 .858 36533888: wfim: 38.328 888- -N 82883 .88 Z

voom 98 SON Sufism _BaoEo_am=m can ooonaS >o>usm 363mm «mom :mHOUEOmU H33cm

 

 

NE. 2: 8: :8 88 2 85 88388 8 595:2
88> 3% we? 8% no? we? nova—2: woe-=56 30555

a; 8.: A88 38 A8. 8 8.8

83 83 83 83- 888 83 as; «88889 8.8.8 as 9 8:85

A88 58 828 3.8 :88 8.8

88.8 88.8 83 83 83 83 as: 88 0983 :28: as 2 85:5

A88 2 n: 838 A28 A8,: A8,:

83- 83- 83- 88.8- 88- «.88. :L as 8 25 as 8 35am

5.8 5.8 :98 8.8 8.8 em:

83 ...... 83 83 888 83 83 82 a 33 88:58 >8 E55

 

dos—5:8 .m.N< «Eun-

163

 

 

 

 

 

 

#3: Q3: 38.: 3.3 3.3 93% :39 5 .3
”83 >83 ::_.o :2 3 .33 .23 :de :35 8-8 ow<
an: A2.: 3.: an: 8.: 83: an: Gm:
~33 33 :33 :83 >83 83 :33 :83 8-8 om<
: a: a3: a: .3 3.3 3.: 5.: 3.: 3.:
£3 23 :33 :083 :3 m3 :2 3 :2 3 3.2 om<
3.: at: an: an: 3.3 $33 at: 8.:
e3 83 :33 :33 :83 .83 :83 :33 312. ow<
3.: 3.3 :83 3.3 an: an: E3 3.3
:33 :33 :33 :33 :3 83 :38 :33 8-8 8<
3.: 5.: 3.8 an: 8.: 3.3 :83 3.3
.23 83 8d :3 :33 :83 .388 :83 8-8 ow<
5.3 A8: 3.3 5.3 3.3 5.3 93.: $3
.83 :3 .23 :33 :83 :33 .883 :53 8-2 uw<
8.: 3.: 3.: 83: 3.: 83: 3.: 8m:
:3 :3 o8? No3 o8.o s3 23 23 8.8 ow<
:n: 88 5 8:2» 84
$3: 3.: c 3: 83: a3: :3: $3: 3.:
83- :83- :83- 83- :83- W83- 83- 83- :u: 8:58 532
5.3 :0: $3: 5.: A8,: A3.: 6:: Go:
:29? : _ ~3- m8.o- 83- :83- 88.? 83- m8.o- A :H: 8E8: @850
83: 3.: 8.8 33: A83 :83 33: A3.:
83 83 S3- S3- ...33- .23- m8.o- 83- an: 03888:
88 a 88:38.26 .3222:
«3 S G: A: A: a 3 3
2933 2m:
3.6m mob “com-a3 0883 3.6m Eaton: 503885

 

A3308 o .T: :33. mum-08E 8385

 

35.53 be 689330 "A332:

38...? tuna-355:5 was: 388.5 3.52:. an: .815” me 33-38 E 5:33:— <m no 2252 ~39:— .v.~< 2:5-

164

 

 

..
A

 

 

 

5.8 3.8 3.8 3.8 43.3 $8 3.8 $38
:83- :83- so? 83- 83 83 83 33 :L :30 on: as 8 .255
5.8 3.8 3.8 $58 3.8 3.8 838 3.8
:33 :83 83 83 S3- 83- 83- 83- 32 a 28 883% >5 .255
853532.59 33.5530
3.: 3.8 3.8 :9: ANN: 2:8 2 _.8 3.8
83- 83. 83- S3- 83- 83- 83- 83- A8588 0% 3635533
3.2 8 - :38 - 838 - GE 8 -
:2 3 - :33 - :mtd - :8 3 - 088- 88 a 5:6 <3 SE
3.8 3.8 5.8 2: .8 A28 5 .8 3.8 3. 8
83 83 83- 83- S3- S3- S3- 83. CL 223:2 Soaambom
88 a 8.3.8835 223.8:
518 - an: - 3.8 - 898 - 88 a 88: w
:_ :3- - 83- - 83- - .83- - soc 5: 3% £88 25 1
388 - 5.8 - 3.8 - E .8 - 8% a so» as
:33- - :23- - :33- - :23- - 52325 2:2. a 828m
38 - 3.8 - 3.8 - 5.8 -
83 - S3 - 83- - 83 - EL 82 a $68 $05.03
3.8 - 3.8 - 3.8 - G38 -
83- - :83 - 83- - 83 - :8 88 a 2:85 owes Sam
38 5.8 5.8 3.: 808 3.8 an: an:
33- S3- 83 23 83 33 M53 23 08% as 93% w
5.8 5.8 A3.: 3.8 638 E8 and am:
33- 33- E3 :23 33- 33- :3 :83 as» B
858 3.8 :38 2 m8 3.8 3.8 5.8 65.8
83- 33- :23 .23 33.. 83.. S3 33 e8» 3
6:8 38 9:8 and 3.8 3.8 38 9:8
33- 33.. .583 333 83 83 33 83 38» 3
:uv 88 5 558:5 3 as»
dos—5:8 6.2 23H

 

 

Ivvbh’.

.82? 32: :05 B 832:; beam—axe :u mafia .338 @823 8838 Eat wav 80 3:53qu 056 .3328 895 132385 05
80qu 98% w 2 a: 36 on)» 20:83: 5 am 8 2 mow—w £33 360 8 flowed“. .3988“ 823:: 3588 8 RES"? ._o>o~ 33anme “:8qu m 8:868
.. :26— ounmouawmm 53th g 882?: .1. 820530: Ho.“ @2833 82.8 9338 .898 365803880: mam: Bum—=28 .muBUmé nus—03¢.“ 880 Z

.voow wan SON Sufism ~SaoEoEE—m can 25$an NA“scum “meta: “mom mmmEOU<$<OmU H350m

 

 

 

8 :m 8: 3m :2 3m 88 8?. 3.8 38530 8 3832
. . . . . . . . umomsmo 8322-83va
3 c 3 o 88 o 8 o 3 o 3 8 8° 8 8o 8 Soc 83 8 8:335 3238
8.» 8% mo? mu> mo> 3% 3% mo> vows—05 8:882”. 32865
3.8 8.8 3.8 318 §8 38 5.8 a 08
88.8 88.8 83 80.8 83. 88.8- 83 88.8 88 @038 338 ”a 2 833
E8 38 698 5.8 3.8 8.8 8.8 5.8
83 83 o8.o 88.8 83 83- 83 83 38 82 cans “838 ea 2 Susan

 

€255.59 .82 033..

166

 

 

 

 

 

 

167

 

E: 688 A88 83% ~88 A88 A28 8.8
head 08.0 .3;de *ivwod *MNod *MNod siocod 8380.0 amumm ow<
an: 68.: A8: 8.8 58 8.8 5.8 8.8
23 83 88.8 .83 83 8.8 .888 .888 8.88 8<
A8: :3: 82: 3.8 3.8 8.8 :88 3.8
:3 83 *23 .83 .83 .83 .888 .888 812 8<
88.: an: 8.8 A88 63: :2: G3: 8.8
83 83 3.8.8 :88 :88 8.8.8 $88.8 88.8 318 8<
5.8 :88 and 63: A8: A8: 8.8 8.8
.888 :83 88.8 .888 83 :3 .888 :83 8-8 .83
so: an: 8.8 5.: 8:: :88 8.8 8::
23 83 83 88.8 2.88.8 3.88 .888 .888 8-8 8<
8.: 2 N: 2 <8 8.8 5.: 88.8 E8 888
83 83 :83 2.88.8 2L :3 :33 .888 :83 8-8 08.
A8: A8,: 3.8 88 A88 58 5.8 5.8
«8.8 88.8 88.8- 83 88.8- 83- :88.8 :83 . 8-8 8<
3": 25m 5 3:9.» uw<
A88 3.: 8.8 5:: 33: A88 88 3.8
83- 88.8- 83- 88.8- 83- 8.8- 83 83 an: BE? 5:2
8.8 8.8 8.: 8.: 28.8 £3: E8 58
2.88.8- a; :3- 83- .83- 83- 8.8. 83- 88.8- 2": 8E5: 85:5
8.8 88.8 8.8 8.8 8.: 8.: 5.: GE:
83 8.8 83- 83 .83- 83- 88.8- 88.8- Cu: 038885
88 a 8:35:56 .3222:
88 S 8 5 S 5 a: 3
o—gom 2&2 03th 032
$08 8: son-82 $08 888: 88

 

Aomsfiofio o J": 23% own-25.5 commooun

 

gum-Ea .5.— EEESU "Am—occu—
3889 88-9888 .253... @888 8.38 E: 558 8 88-88 a £335 E 8 .882 :8: 8.2 2...;

 

 

 

A3.: am: as: A3.: 5.9 82: E: 80:

mood mood oood 89o 82.. 83.. 258.? 38.0.. c": Queue mace 553m
3.: 5.: am: am: an: an: A2.: as:

83. 83. 33 32: 83.. 83. 82.. 83 @2585 a: mauséfids
E: - 5.: - E: - an: -

~85- - Sod - ...8o.o- - Sod - 088- 82 a 596 E SE
$2: $2: A2.: Go: A8: 3.: 62: A86:

53. ~85. 386 $85 Rood- 83. 83. 83 cu: 22038 38338
an: 3.: am: am: at: A3.: 3:: :2: 83 5 83.39220 2282:
N85. 385 Sod- god.

2...: - awe - an: - A8,: - 88 a 0&2

$86 - Sod- - :23- - 2.80.0- - 88 .33“ 50% £88 25
an: - § .8 - A3.: - 8:: - 88 a 3% ”a
8:6. - coo: - 59o- - coed- - Boswsoht 080: an 323%
so: - an: - $5.: - $3: -

:mwod - :2 S - :Ed - :83 - as": coon a 53:8 3&5

five - as: - 5.8 - E: -
Sod - Sod - Sod- - 258.? - cu: 88 a 2:85 owes E3
3.: ch: an: am: an: an: 5.8 so:

83:- 83. 53 32: 53:- :56. 89¢. 83 38“ 23 3% w
E: 5.: 5.: 5.: 32.: SW: ewe Go:

N85. 83. 83 82: $8.? 83. 82: 83 as» 5
£98 an: am: so: fig: 3.: 63: A8:

Sod- Sod- 38.0 $23 .82.- 83. Bed. 82: 38» 3‘
so: :0: 8a.: 8a.: 8:: A3.: :2: :2:

~86. Sod- 35o ~86 Sod- mood- Sod- Sod- 3% m;

an: 83 5 553:5 3 E2;

4.2.52.8 .m.~< ~35.

168

 

 

 

5:323 Bun: _§E>€E 05

8&3 who?» w 2 a: :06 0:3 20:83; B an 9 2 memo $33 350 9 flowed”. .8536.“ 32:55 39:85 8 .9509. ._u>u_ connocawfi E85: w 8:865

.. :26..— oucaoaawu 388: a 830.63.... .mEonomson 8: @2833 $8.6 23:8,.” #458 bwocmfiooxmfiufin mam: 33:28 .mBoUm-N 820%? ”$82
a

.voom :5 Sam Sufism gage—33m :5 883:3 ~Agram 7.03am “mom amEOU<§OmU H330m

 

 

moom moom Noe Noe $8 32 oo: oo: aoomtaoooo 5952
no? mo> mo> we? mo> we.» 8% mo> 3:205 8:956 ow£=>
2%.: an: coo: ooo.o: am: 3on so: 5.: .
ooo.o- zoo- ooo.o- ooo.o- ooo.o _oo.o ooo.o .83 an: 1:23 :95 scooom
so: ooo.o: coo: so: so: so: so: so:
ooo.o Nooo ooo.o ooo.o Nooo ooo.o .83 .83 2:88 o5 05o 58 225%
Go: Go: 3%.: 5.: 60.: so: ooo.o: so:

ioo.o- :oooo- *Nooo .oooo :mooo ..3o.o ooo.o- ooo.o- 2": scene Boo Scooom

 

.8238 .m.~< 2%...

169

 

 

 

 

 

170

 

 

 

so: do: «on: so: 8o: 5.: at: 5.:
zoo oNoo :aoo .Looo o_o.o ooo.o :mooo :Nooo ova uw<
an: am: so: so: A3.: 5.: so: an:

:Roo .omoo :83 :33 ooo.o _oo.o :Goo :oooo Eon ow<
5.: av: Go: so: 39: mo: so: so:
hood mood {*mwod Schood *VNod *mNod {*hmod tenccd 919a om<
so: an: 5.: so: 5.: A3.: A2.: 8::

:23 Lmoo :33 13.3 *mmoo zoo :Nooo :oooo 3on u?
so: so: no: aw: Go: E. : : _.: so:

tomoo :Soo :mmoo :mmoo ooo.o :oo .omoo *omoo ow? 0?
so: so: am: so: so: E: A3.: so:

:mmoo :Eoo .NNoo :Roo :Soo :omoo :oo .23 omoo ow<
so: so: so: so: A3.: fin: 9:: 5.:

:23 :33 :23 :omoo :oo ooo.o .too too 3-2 ow<
so: 3.: E: ooo.o: E: A2.: Go: Go:
ooo.o ooo.o vooo ooo.o ooo.o ooo.o o_o.o Boo «woo om<
Ammo: so: so: so: so: 6:: fin: 5.: an: 82 5 as...» 0?.
ooo.o _oo.o- ooo.o ooo.o goo- ooo.o- ooo.o- ooo.o- cu: BBQ: 532
so: so: coo: go: so: 5.: Go: go:

$8.? .58.? ooo.o- ooo.o- .oooo- ooo.o- ooo.o- .oooo- :u: 3.5% $85“:
Go. : aw: ooo.o: ooo.o: Go: 5.: 5.: A3.:

ooo.o- :oo- 83 ooo.o ooo.o- ooo.o- ooo.o- ooo.o- an: ”3358:

82 a 833.856 .3232:
A: E G: 6 a: a 8 3
omega 0:2 038$ on:
axon nob oooméoa oomm< axon Boson: Sam oomm<

 

cognac o .3 .33. mmméafi 3380a

 

55.5w he 339:8
.2 "E82: 88% 83352:: 3.5... 5.8: E; has» 3 «ooo-ooou 5 £33.: .3 oo 22...: .32.. 6.? 2.3.

 

L

.83

 

I. ”Pr".

 

8.: an: 688: A88: @888 888: $2? $88:
888- 888- 888- 888- 888- 888. 888- 888- :n: 8:8 25 28 S 6585
Gm: 88: A88: 38: 68.: A8: A88: 2 8:
.. _88 .. 888 888- 888- _88 _88 888- 888- 82 a 22 882228 >E .255
838283.35 big-Sou
3: a 8: an: 68.: A88: 888: A88: 88:
£88- 388- ~88- ~88- :88- 888- 888- 888- @585 2% 8528833
:3: - 82: - :88: - A88: -
:88 - :88 - 2:28 - 2.828 - .88- 82 a 588 <8 SE
88: 58: 888: a; 8: A28: 988: 88: 8.8.:
888 888 888- 888- 888 888 .888- 888- cu: 28388 3888388
GaoN E muflmtuuuahaa—U Eon—095:
A8: - 8.... : - 3.: - 88: - 88 a 2:8
2.888- - 888- - .888- - ~88- - Boa 3.: 80% £88 25
58: - £8: - 28.: - E: - 88~ a 3% 28
:388- - 2.388- - $888- - :388- - 388885 0&8 a 8288
88: - A88: - $88: - A88: -
888- - ~88 - 888- - ~88 - as": 88 a $68 $255
:8: - £88: - an: - A8: -
888- - :88 - 888- - .888 - :n: 88 a 0885 ”was 9888
88: $8.: $8.: 88.: 8.: A2.: A88: A88:
.888- 888- 888 888 888 888 _88 888 26% 88 £8 8
5.: 8.: A88: :88: 88.: 88.: 68.: 88:
888- 888- 888 888 888- 888- 888 L 88 as“; ~
88: 88: 68.: $8.: 688: 688: $88: $88:
.888- .888- :88 :88 ~88- ~88- 88 ~88 £8» 8-8
:8. : :8: $8.: 38.: A88: 888: A88: A88:
:88. 88. 288 288 888 .88 888 ~88 as» 3

Su: 88 a 888:8. .8 28>

83538 88¢. 28..-

171

.828, :88 H85 8 85gb? b08538 =8 wnflom .838 @828 0886 Sod waw .«o b23303 2E6 88328 895 3:239: 85

283 88» w 9 a: 8% on? 205850: 5 an 9 2 .88 £38 850 3 8833 83338 88:83 3585 no 358m; ._o>£ condo—mama Huang n 880:2:
* :32 condo—mama 888m _ 880:8: .1. 820830: How 88820 888 @8988 8:98 585388088: mam: 8823—8 .8888 820mm?“ ”8qu

.38 E“ 88 8:8 388088 23 8388 35m 3.95: mom émaoo<§o8 ”830m

 

 

 

:R :R 2% 2mm :3. gm 8? 82 22358 “6 5952
. . . . . . . . ".838 88228886
as o N; c woo o 8o 0 as o :o o 3o 0 go o 88 we? 80 $535 3285
8% 8% 8% 8% 8% 8% 8% 8% 88205 83:56 38:65
8.8 5.: 5.8 88 $8 5.8 §8 8.8
.886 ooo.o ooo.o 83 ooo.o- ooo.o- ooo.o- ooo.o 858 «885$ 39% ”a 9 Susan
A88 58 5.8 58 A28 A38 $8 5.8
ooo.o ooo.o ooo.o ooo.o ooo.o ooo.o ooo.o- ooo.o- an: 22 3&5 .338 ea 2 83.9.5

 

62.528 .92 2.3

172

 

 

173

 

 

 

3.8 :88 E8 95.8 A888 $8 8.8 88.8
89o Sod :2 3 18:8 885 1.85 2.52.0 3888 $8 of.
58 A28 £8 :88 38 A38 888 688
£85 :86 :83 a; _ S :55. 89¢. :23 :83 8-8 0?
so. 8 5.8 688 2K8 A38 808 3.8 898
wood wood .zvaod 3.5890 singed *wZVd 5,050.0 .INhod 915» ow<
:58 :38 3.8 $8 3.8 at: 62.8 3.8
:23 88.0 3.88.0 t. :56 Bod Bod 3.086 .4 God 318. ow<
$88 5.8 8.8 $8 8.8 :98 98.8 5.8
:88 :33 .1.an 1.de ooo.o 83 :83 :23 8mm». 08
5.8 §8 6&8 388 888 25.8 688 A88
**NN0.0 timed 3139.0 **N.No.o nib—0.6 *{W—od *Nood {hood vmém ow<
A88 8.8 $8 $8 an: :38 5.8 $8
.188 :53 .2..de 1.de 89¢ 83. :83 .3de 3-8 03
888 $8 88 88 §8 88 88 A28
.888 .588 Sod 39o ooo.o ooo.o :83 :33 3-8 ow<
Cuv 25m 5 «8.5..» 84..
$08 $8 8.8 808 388 A38 and :88
83 ooo.o ooo.o mood ooo.o ooo.o- ooo.o- 83. :8 BEE 552
A2.: 808 8.28 828 3.: Ga: 5.. 8 A38
v8.8 89o- Sod- Sod- Sod- N88 ooo.o- 83. 2.1.8 BEE $550
8.8 28.8 898 3.8 3.8 8.8 888 888
83. ooo.o- 8a.? ooo.o- :56. N88 :88 :88 :8 ”33888
88 5 83:82:20 .2222.
g E «8 8 A8 8 8 3
£988 £22 £988 228
$8. mob 58-82 30% 823888

 

AomEofio o Jnv ==v<aaéafim vomaoooQ

 

.8323: .8.— 883.989
Bz ”22.8. 38:0 88.9858 .33: 8.83 25 .525» .3 38-88 5 5:33.: 8 a. .2282 .EEm 68... 2.3.

 

 

 

um.- -- - . 4:...
E .q :88 :88 :88 A88 :88 :88 :88
88.8 888 N88- 88. ~88- m8.o- 88.8- 888.8. 2: scene mace 558m
.88 .88 :88 .88 E8 :88 :08 .88
.88- :88- :88- :88- 88... 88.8- 888 888 .8588 2% mazofiafifi
.288 - A88 - $3.8 - .88 -
:83 - :38 - :83 - :58 - .88- 82 a 8% <8 SE
688 :88 :88 8.8 :88 :88 : m8 3.8
:88 :88 :88- 88- 88.8 8.8 88.8. o8.o- 2: 28°82. 308882.
88 5 8:836:26 22.88..
:88 - A88 - : m8 - : m8 - 88 a use.
:88- - 80.8- - .888- - 88.8 - EB: 5: 32.: £88 25
G. .8 - 8.8 - A88 - 5.8 - 88 a a: 2..
:88...- - :89? - :88...- - :88- - 52.885 2:2. a 828..
:28 - :88 - :88 - $8.8 -
O88 - .88- - .88- - 88.8 - £2: 88 a. :58 $8.8m
:88 - :38 - A88 - 5.8 -
N88- - .88 - ~88- - .888 - 2: 88 a 088:. 8:, Sam
.88 :88 A88 A88 :88 :98 .88 .88
:83- :88- 88 8.8 :88 .88 888.8 8.8 26% as an: m
.88 8.8 :88 .88 6.8 8.8 5.8 5.8
.888- :88- 888 8.8 .88- ~8.9 .88 88.8 ea: 5
:88 :88 .38 :88 :88 :88 A88 :88
:88- :v8.o- 88.8 88.8 88.? N88- 888 888 an: 3.
:2 .8 A88 .88 A88 :88 :88 :...8 .88
:88- :88- 88.8 288 .88- .88. 88.8 .88 ea: 8.

3: 88 5 553...... .c 23>

 

60:52.3 .5: 035-

174

I... 11.

8.030.... .08.... 3.6.38... 0...

0.080.. 0.3» w 0. a: 80:. on? 20:030.. 5 on 0. m. 009. 8.0.... .050 0. 0.0.3.0 8052.00 3050.5 .958... .0 .550"... ._0>0_ 00.30.33... 8.00.08 m 00.00%...

_. 20.6. 00.80.8330 8.00.08 . 3306...: 820.380.. .08 80.0.3.0 m.0..0 ESE-Sm .850. b.0umaw0xmo.0.0a mfim: 80.0.3.8 £0.00.-. 0.303.... 80.02

doom 8.... Sam .>0>.:m 32.060395 2.0 883.33 >0>...m .00....03 .00.. namEOUgOmU H00.80m

 

 

.8. .8. .8. .8. 8.. 8.. a... 8... 88.30308 .0852
m0> 00> 00> 00> m0> 00> 00> 00> 80820... 3.0.8.6 0mg;
.88 .28 .88 .88 .28 .88 8.8 .88

.8...- .8.o- .8... .8... 88...- o8.o- 8o...- o8.o- ..u8 .822. .83 55%.
.28 .28 .88 .88 .88 .88 .88 .88

8.8 88.8- .8... .8... .8... .8... o8... .8... .7888 .05 25.. 58 .0558
.2 .8 .88 .. .8 .88 .88 .. .8 ....8 .8. .8

.8...- .8.o- 8...... 888- ....8... :8... 8...... 8..... ..u8 .83.... 80.. 3.5.8

 

605.5000 .52 0308

175

Figure A3.1. Zambia’s HIV Prevalence Rates, by Province, 2001-2002.

     
  
  
  

 
 

o
Kasama

   
 

  
   
  
 
  
     
  
   
   
 
   

Northern

fi
“"8 8.3%

Luapu a

0
Solwezl

 

 

Northwestern
9.2%

Chlpata

Hanna
0

Western
13.1%

 

Southern
17.6%

Livingstone

176

 

 

vcom v5 Sam SoPSm ~SaoEoEq=m 98 oocmaoa ~A052w 63.8mm «mom mamQOUxngmU H350m

 

 

 

 

 

0.: N? 93 new 482. :8 § 88 5 man; .233
3a 0.8 v: «.3 0mm 9% 2m. ”.8 as. to as, 885
0.32 32: SE: q: : m :2 :3 22: Q. ~ a can 88 2:83 5E mo
«.3: ~62 :2 32 2: ca: 3: mg cam 88 2.35% :55? 8:3
M23: 35 5:: 382 <82 ENE «.32 32 cam 88 098 Mo ”83>
2:02: ~56me «=8 33.?
8.0 3.0 85 and 3c 3o Ed 3o 83 $5383 Hauaaeguao 83
3Q 25 985 3% Si 38 33 5mm 93R 88 B an 3&8 .6 as? $.80
33 moo: 3&2 3&2 Q82 39: «.2: v.32 95R 88 3&8 .0 use, 38o
m; S .o Ed :.o m; :5 26 So aeu 3:; :3:
OS :3 «no mno mg as 98 93 $33
$2 8; ”S m: v2 8; m2 m2 285
O? m2 0: NE :2 as m: 3; 822630 23
accuanoxm EEK
we; of :3 of N3 02 o: a: s5
:2 a; x: 2: N3 2: 8; v2 aom
2: 3m 02 8a m2 :.~ 22 .8; ”icon
-.~ 8; OS 92 22 :2 93 $4 222
:3 m3 3 .0 OS ”.3 ts 3w Sm 0% 2338:
3.0 3: So 12 So 2; 93 m2 2E. @8208“. 280mm
aonmmomfiou 33939:
nag—Q, 502
58c page 5? mm 58¢ 22: as, m: fig 5 55 mm mm u30§< 82
38 8:3 38 83a 38 83a 38 85a

 

AG 60353.5: .8 @8381 ac 3.5% .3 .535? mm 22.3.8: 5 .852...» was .39» u: mus—S, :32 Add «3:.

177

 

voom US“ Son Sufism ~Saoaoaasm can oooNRoS NAguzm “meta: “mom ngOUEOmU H355m

 

 

 

 

SM 3:. N? 3% new 3% 8: Do 82 5 3% 5:33

W? a. G Q: mam ado o.mh mdm ofio _._m w.~o 058E go be 55 EoBoA

03.2 032 3.3. ooz— ~.Nm3 oHoMN floom _ NAN: move o. 3o ABMN ooo.o 0883 ES.“ to

oéom msm: 53— Q? o ammo Wm: flow wfiofi 22; no.2 ABMN ooo.o flmfifim :35 mo 33:;

Tatum 033 m.wmom _.mwo~ $33 mdvm Ndmm Nmmm No3: moon ABMN ooo.o 3:8 no mos—Sr
2:85 5.85be 38 336V

36 mod moo omd 2.6 So mod cod woo mod 038 Ban—mince ==3B8~>c=8 was

odmn v4: moms Est. 5.2 N. Wm; oomw @on mono osmo ARV—N ooo.o an 5m 59:0 mo 2:9, $.80

w.mom_ 2km m ~33 083 ”no.2 3.x? . 3x3 nova ~63; v.32 ABMN ooo.o 5&3 oo egg 380

o~.o 26 Ed Nod o~.o moo oo.o boo m_ .o boo macho 33> aw“:

wmo Go omo mmo Nmo wmo wmo 3o omo 3.6 $33.

fin." N5; 9: am; om.“ 34 mod 3: mm.“ mm; 2330

mo; mmd oo.N nmd mm; EN mm; 34 ms; 3; E55530 was
rotuzhogm 5.8K

Q A 3; 36 $4 No; ow.“ go 32 o“; 9: flow

m: VmA mo; on; So om; 36 mm; mo; Vmg whom

cram ~m.m omd ZN we; oo.~ SN :4 £4 No; 3&th

mm.~ Nod mud NYN God mm; mm; mo; mm.“ $4 22“

mod mus no.0 NmH mod $6 26 no.0 mod 36 can 238:3»

2: S; :3 w? 33 2: £3 a: 36 m2 2:2 @Sfiauc 258mm
zofinonfiou SSEQSI
03? :32
53v 228m 636 58v 080% Snow m:

5:8 55 mm 2g 350 55 mm 3 EB 28E 5? mm 23 29: as, m: 8352 82

 

vooN Sow voo~ Sow vooN “com voom Sou vooN _ooN

 

A8 682555: .8 3883—9 .3 25“: can .51qu m5 335.? m_ 2.53:2— : .352; ES .39» E 83.3 :32 .N.m< «Bah

178

 

 

 

 

3o~ can Sou Sputum anaoansm van oooNRoa >256 503mm “mom ngOU<S<OmU Uooh-Sm

 

 

 

 

 

 

 

2.: 3.2 8.2 52 8.: 8.2 8.2 32 8.: 8.2 on: 8.2 g 32 a was 8:293 >5
- - - v2 .o - - - g .o .. - - £3 :uo £2; a Engage .856
m: N: So 52 w: 2: N2 92 . a: 2: 8.0 m2 c353 88 a one aoauuaaoc 2805
3+ mod N: 3.4m 3m #2 :3 2% 8e SN 2: 2 .m 3o 88 a as 322 ES
- - - 898 - - - Sno - - - 83 coon Sauo 88 3 €38 .§<
- - - 3:3 - - - 29o - - - :3 Al 59% fiasco €05
- - - 23o - - - ~85 - - - «So :uo 59% 8E: asam
- - - NEE - - - wed - - - was A To 59% ”22a “2% E
- - - 33o - - - ”So - - - :85 AI anew 338 28 E
- - - 23¢ - - - Rod - - - Rod :no 58c momsoqmzafi 35:3 5
- - - ~23 - - - Sod - - - :3 :uo 58c 833%23 22 E
w» on mm :82 mo. om mm :32 2. ow mm 522 85223
0:68qu 0:38qu 0350qu
Son :02 80A 29:8 =5m

 

£255 3:95 .3 833...; can: But "mount—3w. 953.839 .m.m.< «Bah

179

I .
l.5E-llnh-Pv'ya

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8-8 nae-38 3 38% 838 8392 $88

cog—d can“ xenovnomoc gave-cm Eva Vmém 9.833 3 Macaw :flES 332$ vmmwm

8a 88 a 8:- wauséfl 22080: 22: ea? 95438 3 ~88 zmfie 3:82 $8

Coomno a30m noun $ 83. E 858m 926% 383 20:88: >093 313 $.on own 3 x85 €38 3:803 3%?

$8128 Sagasamesam 32 >3 8-2 9.on 88 3 #88 £32 332 388

AgonfidafloooN 5 83cm =a mo 88: Bet .326 2:82 bass vmém 953mm 3 #09? 2&58 3:83" vomva

saw 338 98 8 88v 58% 888 €85 5320 8-2 $on 88 B 8.8% :83: 3:82 38....

Al A328 28 8 88v 58% 888 baa-n. €58 3-8 as; B 82: 832 3:82 :88

noumtomoo $93 cabana :53

$0M

3cm 98 SS. 5:5 3582&=m 28 8332 8:3. 85m mom mam-8830mm 828m
m 84 m mod m wed m 36. m 8.0 m vod m 36 m :3 m 86 m mod. N Nod- m 86 m Ed m ood m 8.0- m Nod m “not m
u. ....... M-8..._----.._.~.o----n-.8...o..--m-.8¢---n-.~.q.qEM-ad. ..... m-a.a..--n-a..o---n.-a..o..- 3.8.-o.---m-._.q.%-u..a..q---n-a..q---m-.38---M-wd.¢----m-----m_ma--
W ....... m- ........ m-moum----m.-3”?wwouo--.mz_.§----M-§.m ..... Wow-mim-.~.q.o---m..8..o: .m Wm.m--.w-.~.odim- -mo..o.--.m:a.m-:w.38o---.m--.q..q-:-mr--\.,m&mm
m. ....... m- ........ m .......... m. ........ m-.N.§--.m-._.q.q..---M-.8.d. ..... 5.8.--.m-.~.§--888.--..-888---m-._.q.o---m.-8-.q--.m-.88.---m-.a.o---.m-~q.q----m:-----.>-a-m

M ....... m. ........ m. ......... m. ........ gum---m-a.qm::w-oqm ..... Mme-d:--m-.~.q.q---m-_.o..o..- wa.m---m:a¢im. -3..o.--.m-wm..q:-m..~.ouo---.m-wm.m::m.----xm>->m

........ m .......... m. ........ 2: ..... E8. ..... «am-.883“--- -a..o..-- wad...--m-.a..a---m.-_.o.d..--.m-8.8.8..--m-.§----m-8.8.8----m.--§.9...-m

W ....... ........ m- ......... m ........ M ........ .......... flaw ...... M-3..-o...-w:&.-q.. -mouo..- ad..-:W:_.od--M-No..o.:w:~.q.q:-M..~.ouo---w-w-q.q----m $838.8

m ....... m ........ m .......... m ........ m ........ m .......... m ........... 38.8--- mad-J 9......- m-§.---m-§---m .80.-- m-.fi.o...q-.-m-w.§--- m-a..q----m ---.8..-.a..--m

u. ....... "- ........ n. ......... u. ........ W ........ w .......... M ........... W. ........ w-mo..~.---m.8..o: w-omd.--flwz%dim.8.-o.--w-wad.-:m.ad:.w-o.q..q--:m-:.mmwwm

W ....... m. ........ m. ......... m- ........ . ........ m .......... m ........... m ........ m ......... m .34--. ad.-- 3.8-Q---- gd: m-....q.q---m ad--- m-a..q----m ---.....m..~...-m

m ....... u. ........ m .......... u. ........ m ....... .n .......... m ........... m ........ J” ......... H. ....... ama--w-wad:-qu--wzm.q.q:-m douo---w-mq.qz:m ..... .m mwwm-m

w ....... m. ........ m- ......... m. ....... .- ....... -m .......... M ........... M ........ -m ......... m. ....... m ....... 38...--.-8.-8.--.-8..Q---m .88---m-%.q----m.---.8ww...-m

........ m .......... n. ........ m ........ m .......... M ........... M ......... m ......... m ......... .u ......... m ......... u. -8.-.m--u-w.q.m-:m:e.o..o:-wad--4 ..... m mug-«fl

M ....... m- ....... m. ......... m- ........ W ....... .m .......... M ........... M ........ .m ......... m- ....... .m ....... .- ......... m- ........ .m-o-q._.-:m..ano---m-&..q-Em. ....&...8. -m

....... m .......... m ......... m ......... m ......... m ......... m ......... m ......... m .......... m --------- m -.o.o..m---.m:~.§--:m ----- 8 «Wm-M

M ....... M ........ m- ......... m ........ W ........ w .......... M ........... M ........ M ......... m ....... M ......... W ......... M ........ w ......... m- ........ 8.- ..... .- 8.?
3&3 QE- -m.-»m .8. EE- fiwm may-Em m $33....- m .m%%. m--&.mw8.-m -83..“ wmmwwmwwwwfim $.88. awn-WW... «WW-NA..- m:&&_.um-m ............ .-

 

8332; 333.0..on can: «mam "um-.22: news—ouucu 6.2 «Ea-H

180

‘wfivfl‘u. 0.0m...» «trot ml... 1,. J .t.» “h I. a}!

 

 

38% 2; 8 wows e26 Ease 352m EH22”
35% Ba 8 news 55% 838 3:65 3263

 

 

Coonuo coon com“: ooo.o. 5 mafia Egon $9.3m Eonomzom >093 . anew owméfiflq $380M 550 3:5.“
C3855 8:8 xoaovuomov 9600b” “not 53v omméfifia 338$ .550 3608
as 88 a on: magoaas 289.5% 22: 53¢ 0%-an 833%22 038$ £35
moan“ 08285.5 maxi; 32 >3 53c owméfiun 38: 032 59:9:
#853539 385 noumtomoa 323
23M

38 Ba 88 5:5 financing 23 883% 33% “scam son “EgoufieOmo ”859m

 

 

 

 

 

 

 

 

 

 

 

 

 

of 3d :6 So. So :3 8.0 86 8.0 So. 898
2: 36 86. 3d 8d :3 So 85. 8.0- :5
2: Ed- 8d :3 86 86 So- So- 22:
2: So- So So Nod 86 :8. E
2: 85 cod 8d 8d- .86. 252»
2: 3o 85 So- So- 59:20
2: 85 8d 8.0 268
2: :3. So 2:58
2: 85 528$
84 52:8
5&8 a3 22: >3 22%? 526: 268 269: 528$ evens

 

 

 

 

 

 

 

 

 

 

 

 

 

 

£288 Ban—E— 3352: aces—0530 .m.m< 033;

181

Table A3.6. First stage regression results: Pooled probit’.

 

Dependent Variables: Death of prime-age adults

 

 

 

 

Covariates Heads/gaouses Other males Other females
Male Female
(A) (B) (C) (D)
Excluded variables
Age group of the deceased
age 20 to 24 0.002 0.004 0.049" 0.067“
(0.91) (1.12) (5.05) (6.17)
age 25 to 29 0008* 0.014M 0.056" 0.049"
(2.55) (3.02) (6.15) (5.82)
age 30 to 34 0.018" 0.022M 0.039“ 0.067“
(3.70) (4.46) (4.93) (6.88)
age 35 to 39 0.030M 0.025“ 0.036" 0.063“
(5.07) (4.50) (4.00) (6.34)
age 40 to 44 0009+ 0.003 0.047“ 0.038"
(1.81) (0.95) (4.03) (3.45)
age 45 to 49 0.072“ 0.044" 0.008 0.025“I
(7.04) (5.09) (1.18) (2.81)
age 50 to 54 0.114" 0.074" 0.009 0.016“
(7.48) (6.21) (1.37) (2.16)
age 55 to 59 0.067M 0.044" 0.071" 0.073“
(5.08) (4.77) (4.54) (4.90)
Deviation of 1994 mean rainfall from 0.003 0.007" 0.014" 0.019“
10 year average (1.35) (2.97) (2.62) (3.93)
Age group and rainfall shock interactions
1994/95 rainfall shock by age group 20-24 0.001 -0.005 -0.052** -0.064**
(0.14) (0.47) (4.40) (5.29)
1994/95 rainfall shock by age group 25-29 -0.008 -0.016** -0.040** -0.060**
(1.35) (2.82) (3.85) (6.00)
1994/95 rainfall shock by age group 30-34 -0.006 -0.020** -0.058** -0.043**
(0.74) (3.82) (5.30) (3.89)
1994/95 rainfall shock by age group 35-39 -0.009+ -0.021** -0.045*"' -0.033"”"
(1.65) (4.33) (3.26) (2.95)
1994/95 rainfall shock by age group 40-44 0025" ~0.028** -0.036* -0.047**
(3.09) (3.86) (2.17) (2.97)
1994/95 rainfall shock by age group 45-49 -0.015* -0.026** -0.046* -0.035*
(2.56) (4.36) (2.15) (2.19)
1994/95 rainfall shock by age group 50-54 0019" -0.021** 0074'” 0033+
(2.59) (2.66) (4.23) (1.82)
1994/95 rainfall shock by age group 55-59 -0.026** -0.023"‘* -0.015 -0.022
(2.79) (2.69) (0.65) (0.94)
Included variables
1999 HIV/AIDS prevalence rates -0.000* 0.000 0.000 -0.000
(2.56) (1.40) (1.14) (0.84)
Landholding size (Ha) -0.000 -0.000 0.000 0.000
(1.37) (1.33) (1.59) (0.48)
Asset poverty status (1=non poor, 0=poor) -0.001* -0.000 -0.000 0002*
(2.50) (0.47) (0.21) (2.10)
Effective dependency ratio(number) -0.000 -0.001** 0.000 -0.000
(0.07) Q44) (0.52) (0.08)

 

182

 

 

Table A3.6. continued.

 

 

 

Elderly males death (=1) b b -0.000* 0.000
(2.56) (1.40)
Elderly females death (=1) b b -0.000 -0.000
(1.37) (1.33)
Provincial dummies included Yes Yes Yes Yes
R-squared 0.113 0.155 0.161 0.164
F -test for instruments
All excluded variables 147.38" 237.11" 325.19“I 273.53"
Observations 10682

 

Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey, 2001 and 2004

Notes: aEstimated coefficients are marginal changes in probability. Absolute z-scores, calculated using
heteroskedasticity robust standard errors clustered for households in parentheses. + significant at 10%; "'

significant at 5%; ** significant at 1% bExcluded since there was no variation in the dependent variable.

183

km“

 

 

 

. u.W.d .
._ I. ‘
.. - .0 "15,-ell.“ f . f s. .. My»

 

 

 

 

 

£36m own-BEE mo amon— Hmofimum> “savanna

3.8 3.8 3.8 3.: 83.8 3.8 3.: 3.: ”we”: :2 2
S3 33 23- 23- 83 83 :3- :3- Soc :35 use 3% do 8335
2 3v 2 <8 3.3 3.3 63.3 3.3 82.8 83
**omm.o ELMMd *iammd .3;de {*w_N.o siw_N.o anvad **va.o am 9 mm uwm
fine and £33 84.8 got 3.3 $33 3.8
3:3 :33 E3 .53 :33, 3&3 :83 1333 E 2 cm own
8th 83V 3.8 83V 52¢ a; 3.3 3.3
3.323 :53 23 23 teas :23 :23 :2 3 a. 9 2 own
A28 33: 3.8 $2.8 A2.: 3.: 5.: an:
:33 :Q3 :83 :33 :3 23 +33 +33 3. 9 S. own
2 3V 5.8 at; at; 3.3 64$ 3.3 $13
.Ibmvd *ammvd ..:..mm~.0 3.3.-Nd 552.0 *nvmfid Sumo—d «*mc_.o am 9 mm own
is $30 $8 3.8 3.3 33v 94$ 63;
‘30de siNovd **WMN.O six-0N6 .33: .o ianHd an; — —.o EL a fio Vm 9 0m owm
3.3 3.8 83.3 $8 3.9 35 Si 2 S;
*Lmvd *immvd **o§\o 1.3-v.0 *i—N~.O *nONfiO **vwo.o «*vwod am 8 mm own
3.3 and Gas Gas 3.8 3.: 3.3 3.8
:33 :83 3:3 :33 .23 +83 :23 .3383 3. 2 8 a;
338% 2: mo 38w ow<
WNNQGYU.» EOVBNURW
3 3 E m: 3 6v 3 3
...... 338$ 550-1:- ----mo_mE BfiO-l-l- ----o_anm--WMm--5-omm\muMm-l332:}-..- 835300

 

.2308 Boaocotmu 036 EN 05 S cm: com 6:32 $50.5 owfim “mum .534 293.

184

 

at:
.333
:3:
333
G3:
333-

:3:
83
$3:
33
23:
23
a3:
83
53:
33
5::
:33
:3:
33
83:
N33-
63:
33-

$3.:
.333
$3:
333
$3:
333-

33:
~33
$3:
23
$3:
83
$3:
23
an:
:33
83:
e3
63:
83-
$3:
333-

83:
333-
A3.:
.3 25.6
$3:
333

93.:

gimme-o

an:
+386
83.:
$3-
3.:
333-
E:
23
a3:
83-
E3:
23-
3.:
+333
33:
383

53:
333-
83.:
... flood
5.:
S3

5.:
i Ed
3.:
33-
83.:
3333-
GS:
383
63:
83.
5.3:
23-
83.:
+33
83:
:3

5.3:
333-
:3:
333-
63.:
+ooo.o

:3:
.83-
83:
_ 83-
63:
+33
2 2:
:3-
$3.:
:3-
$3:
23-
5.3:
83-
$3:
83
a3:
33

a3:
83-
$3.:
333-
$3.:
333

83:
23.
33:
N33
3.:
:3-
am:
$3.
$3:
23-
$3:
83-
$3:
33
a3:
«83

A3.:

:89?

$3.:
333-
83.:

.3. Sod-

G3:
333-
83:
23-
G3:
:3
:3:
S3
:3:
33.
23:
33
A2.:
:33
5.:
:3
E:
:33

$3.:

*imood-

$3.:
333-
33.:

5.9—00.0-

33:
~83-
63:
:3
3.3:
83
:3:
N83-
:3:
23
A2.:
33
$3.:
:3
E:

... _ mod

Cooano doom can“: 3383 3.53: Howm<

3: ”Na 352323

388 83232: mQESHI 32
833.29» 333335

2333332 3833 583 emu-08:3 SE
3.2 3:83 own 3 ~85 3352 3332
+33 3:23 own 3 :27. :35: 3332
an? 3:23 own 3 35% 3.35.2 3:32
313+ 38» own 3 35 335$ 3:32
333 3:23 own 3 3% $32 332
33.0. anew 03.... 3 3% 3358 8:32
33m 3:83 own 3 :23 335.8 3332

+23 3:83 own 3 “.8.: 3.353 332
2:52:35 x35 :55: E:— nacuu and:

 

33.538 3.3. as;

185

 

BEES, 302096 05 E sought, on 83 305 8:6 wovaoxmn fix; an “magmas 3.. mo\om “a Emu—Mai ... ”$2 «a Emu—Maw? + .momuficoumm E $35.33
8m 38320 38.8 “56:8,.” E58 bmoummwoxmefloa mam: wean—=28 .888-N 3302?. 525305 E mowgo 33ng 03 356508 wougmma “8on

SEN new SON Sogm gage—33m wan ooomama— >256 505$ “mom AdeOUEOmU H8.30m

 

 

 

3m m maocatomno
:9? :33 :32 :22 832:; 8305 .2
2.st ing :33 imam? 2.63m :MQM :82; 2:2». £85 232% Ba 258% om<
*3 :52 no 85 58c 03-08% 85
anmfignfi gobnfii
Em 8.8 o? 9? Q3 8.8 _.3 2% 3533
mu> we? we? we? moxw mo> we? mo> flown—05 mogfi ~amofi>9~m
2 0: am. 3 and 5.3
23 was $.86 $85 a a a a :L 58.... magma €25
$on 33 $on 82:
Sod 89o mood 83 n. a a a :uv 58c 3?: 355
§ .8 92.8 2.98 ans :3 3th § .8 83V
ooo.o ooo.o ooo.o 53 :83- :83- 83- ooo.o- @3858“. 5882.8 2805

 

62.538 63 2...;

186

Notes for tables A3.8 to A3.]4

Source: CSO/MACO/FSRP Post Harvest Survey 1999/2000 and Supplemental Survey,

2001 and 2004

Notes:

-** 1% level of significance, * 5% level of significance and + 10% level of significance.
-Endogenous variables: Death of prime-age male heads of household, female
heads/spouses, other non core males and females.

-Instrumented by:

IV-I: 1994/195 rainfall and age group of the deceased drought shocks.

IV -II: Prior prime-age death in household between 1996 and 2000, 1994/ 195 rainfall and

age group of the deceased drought shocks.

187

 

 

 

 

 

 

 

 

 

 

 

 

 

8 mm 88 28 888 882 88 88 M: R 882 882 823338 8 2:52
cosmomcaovtgo
8.: 8.2 8.8 8.2 8.2 8.: § as 8-8.2 588
a: 8.~ :82 8.8 8.8 :38 as 8.20 agsfiaaépaa
NE 8; :8 88 MS a: a: :3.»- 38 +o~.~ Euros own-22E 8 30E
48.8 28.8 $8.: @888 838 88.8 68.8 «8.8 5.8 838
+82 +83 +23 :81- :88? :83 :58 :83 :83- :88? “385
we? m0? m0? m0? . mp? mo? m0> mo? mow. mew down—05 mug—u ~fl05>0un~
85.8 5.8 38.8 a: .8 8.: 3.8 5.8 5.8 an: 3.:
88.8 88.8 88.8 88.? 88.? 88.8 ooo.o 88.8 88.? 88.? 385888 85888 2685
3.8 3.8 $8 $38 $.88 $8 5.8 £08 2 $8 8888
:82? :88? :82? :88 2.888 :88? . :8_.? :82? :58 :88 coonuo .88 :8": €38 33....
58 $58 5.8 $38 5.5 6&8 8:8 :88 628 GE 8
:88? :88? :88? 2.8.8 :83 :88? :88? :88? :83 :83 38 8: wagofiafi
A8,: 8.8 A8: 5.8 E8 658 5.: 858 A88 A88
88.? .88.? $8.? :_ 8.? :N8.? 88.? i 8.? 88.? a: 8.? :NS? 8.2 85368 832: 883
$8 $88 888 A88 88 A88 A88 :88 a: .8 3.8
2 S. 5.? 2 3. 88.8 88.8 88.? Re? 88.? 28.8 88.8 :uv som 8mg 538 8.25: €88
ems 3.8 5.8 8.8 $8 828 5.8 88.8 8.8 88.8
.82.? 82.? ...8~.? :5? .82.? :8m? :80? :28? :88? :88 .? 2L 88m 888 £88 £8. :55
A88 A88 88 28.8 88.8 $8 $8 A88 $8 $8
N8? 3.? 88.? 83 ~88- 88.? 88.? «8.? £3 88.? :L £38 ”228 sec <8
88.8 68.8 5.8 88.8 88.8 628 628 88.8 68.8 808
m2? :8? 88.? 88.? 88.8 88.? 88.? 88.? 88.? v8.8 at 588 ES: 550 E
88.8 5.8 628 5.8 88 A88 A88 A88 628 88.8
~28 :2 :88 2 3. 88.8 8:8 888 28.? 3.? 88.8 :uv 588 8388888 285 <8
5.: an: and 5.8 88.: 858 38.: A88 $8 8.8
88.? 88.? £8? 8:? .82.? $8.? +82- :88? v8? :2? 2L same 088888; 29: E
E S E a: 68 m8 :8 g 88 28
:2 3: 58 >8 So 8-2 I: 30 2 So
flofioz Gnomes-wig 035mm “go—Com £0602 vegans-HEA— Bgmm «Va—Com
33qu 595 323 €39,230 9835

 

Emu-.3 .36.:- aoua EB 6355...... can. .3 5:33:— <m .8 3:33.58.— wfifianottog can Emu—896:0 no 38,—. .m.m< 035.

188

 

 

 

 

 

 

 

 

 

 

 

w 2 8 8 2 8 M: 8 882 882 8 2 .8 88 82mm 882 882 3235308 388:2
80882-8852830
8.8 88 28 82 88 _&n .8 :2 3.8.2 .3me
2&8 :8. .882 83 88 :8 as 8.20 895853-5852
88 8.8 +8.2 23 8&2 :2 83 .88 «3 .28 88:2: own-08:8 8 22-8
8&8 8&8 82 .8 8&8 8&8 8&8 8&8 8&8 82.88 48.88
:38 :38 2.88.8 :83- :83- 823 :3 8.3 .888 :8 3 8388
mo> 80> 80> 8% mo> 8% mm; 8% 8% 88% 80822: 828888 383on
8&8 8&8 2&8 8&8 8&8 2&8 28.8 8&8 88.8 88.8
83- 83- 83- :88? :88? 83- 83- 88.? 88.8 83 235882 888588 388m
8&8 . 8&8 8&8 83 8 83 8 8&8 88.8 8&8 8&8 88.8
:23 :33 .23 2888 288.8 :3? 2.3.? $88.? :888 :38 28818 .822 8:...8 €268 23.3
28.8 88.8 8&8 838 838 88.8 8&8 2&8 82.88 8&88
83 83 83 1.88.8 1.88.8 2.28.? :83- :83- :38 9.88.8 38 as 8588852
8&8 8&8 88.8 88.8 82.8 8&8 8&8 8&8 88.8 82.8
:88? :88? :83- :83 :83 88.? 88.? 83- 83 83 8E 383,28 8328 383
88.8 88.8 88.8 8&8 8&8 8&8 8&8 8&8 8&8 838
M88 M88 83 83 33 $3 $3 $3 83- 83- 28 83 888 588 82858 2:88
8&8 8&8 8&8 8&8 8&8 28.8 88.8 5.8 8&8 28.8
88.? 88.? 83- 83- 83- 3.? 3.? 33. 83- 83- 28 8% 888 588 888 :58
$8 8&8 2&8 8&8 8&8 82.. 8 82.8 8&8 8&8 88.8
88.? :3- 88.? 288 :3 2 3- 83- .88.? 88.? .88.? 28 £38 3888 see <8
8&8 8&8 8&8 8&8 8&8 2&8 8&8 88.8 28.8 82 .8
88.? 83- 88.? 83 83 8 3 83 83 83 83- 28 538 885 35o E
88.8 2 &8 8&8 8&8 8&8 8&8 8&8 88.8 88 88.8
23 :3 83 88.8 +3.8 88.? 88.? 83- 23 88.? 28 588 8338888 2988 <8
2&8 88.8 88.8 88.8 2&8 88.8 8&8 88.8 8&8 8&8
M83 33 .88.? 88.8 83- 83 83 3.? 3.? .3? 28 588 ”30833.2 22 <8
38 28 28 28 68 2,8 28 68 g 28
8-2 2-2 2-30 >2 30 8.2 2-2 So >2 30
£380: flooaouo-MEQ 2&8—mm Ufloom £0802 fleeces-adm— o~§mm “yo—Com
A389 .050 58:: 3385 A333 83 808 38:: 3383

 

2.96

.85: 8.5 8.525 8:: 38.. .88.:— auaa :c hum—3.3:. «S no 8533.580.— ufibzaoutgc 8:8 EBaowc—Eo no 88H 53.. 83:-

189

 

 

 

 

 

 

 

 

 

 

oz m om 8 m 5 m 968 $2: MN _ m E n E m 3:: $2: 88338 .6 59:52
nomaauuuaowtgo
NS EN ”3. 3m Kg. v2. 5 as 3-8.2 3%
Em 22 $3.. 3; NE 86 :3 3.30 gamamaaénaa
85 v3 two :5: :38 who :3 a: a: +3; Essa owgufi no 52¢
$8 38 A88 9 28 A38 $8 $8 A38 @38 $28
Bed 83 ES :33 :52 :23 3526 :62 and :vzd 2538
8% my? 8% mo> mo? moxfi no? mo> we? 3% 609505 mogfl WEEK/ohm
928 §8 38 $98 9.8 $8 618 6:8 8%: 628
89¢ :55 83 2:3 83 Bod Bed 83 23. ~88 cognac? 58236 3803
5.8 6.6.8 3.8 2: .8 A88 3.8 $08 $38 338 82.8
.585 #85 .885 :vao :Sto :2 3- . :83- :2 3- :33 :ano eooano .33 :8": thin .32
5.. c 5.: 5.. 8 A28 3.: $98 5.8 3.8 :38 :28
Sod- Sod- Sod- v8.8 23. £88 .38.? :vmod- :2 3 :2 3 38 9% ”avenues
am: am: Ga: 3.: 63 $28 5.8 998 am: E:
wood- wood- wood- :3 .o- .3; .o- :86- :w :5. :85- ooo.o 83 was ”803% mesam 833
5.8 5.8 :28 5.: an: A28 @028 a M8 3.8 5.8
33.. ~86- 83.. $3 «2 .o no.8 ~85- 88.? m8? mad- :uv Gem 83 538 83:8 332m
3v. 8 3.: 5.: G~8 A38 6%: an: A3.: 358 3.:
RS 23 83 m8? :3. $3- $3- $2.? +3.3- N25- :uv Sam 33 55 3.8. 385
§8 A88 A88 828 80: A38 938 :58 3.8 so:
mod- 33- 83.. $2 $3- $3- and- MES- Sod- :3. at as“. Essa sac é
3.8 $08 5.8 c N: an: :88 8:8 38 6:8 an:
83.. 82 a8? ooo.o- 5 .o- 33. me .o 39¢. 33 83- cuv sec 838 sec E
$8 $8 308 3.8 838 25.8 5.8 $8 Gm: $8
«8.0. $3.. wood- .56. $3.0- £2. 33- 33. +23- 835. :uv saw 833%23 2.25.: 5
A8: am: 58 3. 8 A3.: 5.8 $8 $28 95.8 no:
33 3.3 83.0 +2 S 89. 33. 85.8 .23- :45 83- :uv as“. ”80388; 28. E
38 5 3 Q: g E 88 g 88 28
:4: I: 38 B So :2 I: So 2 30
20602 wooaouomwa 295mm Ugoom m—ovoz floocouobwfl 2ng vofioom
33:93 $.on .«o mus—«>85 c3950 3on mo 8:333

 

9:38.— .5.—

EEE. 39.» an: 2.9:: $on .3 82a.» .8 3:388 a:— .3 38838.. ufibuaoutog can 53:33.5 a: 38,—. A2.? «3:.

190

 

 

 

 

 

 

 

 

 

 

 

cm; as m oz m $2: $68 as n mg g m 3:: $2: gag-.30 80 2:52
nonmomccoctgo
2 .w a: 93 a: NS 2 .m 5 :3 38.2 5&5
gm N3 .2: Rd 8... 2; SN :2 3.30 :EBE-aaénaa
$3 $3 a? a: we; 36 m2 :2 x: a; games own-2:3 8 53¢
5.8 38.8 $3 Es: :3: £8 £8 and $58 SW8
SEQ-.0 autumn-o 3.0056 thew-N Iknow-N *vaFd 31.3.6 nimt-d wad owe-o «:83—00
8». we? mo> mo> mo> 8% mo> mo> mo> 8% 39:2: mew-:83 30:3on
3.8 5.8 §8 an: AS. 8 Ga: Ga: Gm: 9:8 8N8
ooo.o- wood 53. Rod : So 83. Rod $3 «88 83 c3583? 388538 268$
83: 638 as: $38 A888 9.28 $28 5.8 $38 $28
:82- :owm._- tam;- :o:._ _.....§._ 333.0. _.....m;.o- :83- :ooa._ :33 coogno coon San: €33 “3.2
3.8 $8 5.8 Ea: $.28 and 858 3.8 838 $.28
3.. u no.0 «*Nhod *{Ohod *uhmod .30de 3.53.0 .3936 fame-o «*oomd as; _N.o Adm-G 0N5 wad—£0553
an: 3.: A3: 5.8 $08 21.8 818 938 $8 5.8
:3. ES- Sod- 308.c- ......o8.o- :Nvod- 3.96.0- :mvod- 358.c- _.....$o.o- 3:: 88:35 832: wafi
828 A38 $28 A28 A88 5.8 3.8 $8 3.8 5.8
83 83 $3 .36 .533 02 .o. 2:8 RS- 3:6 #2 2L 8% 83 69% 3388 :35
am: A2.: 5.: 5.: 5.: A3.: A3.: A2.: so: so:
and- 828 $2. RNo- 83- wmmd- :3- $2. $3 83 :uv 80m 83 588 338 332m
858 5.8 A3.: E8 9:8 :08 828 E8 :58 an:
~38- 28.0- +£2- RS v8.8 89¢ 83 was- Re... 83 at 588 8388 5:8 §
A88 :9: E8 698 5.8 A38 A58 92.: a9: $8
E .o- $3- Ead- weod Sod 8.3- :3- Ed. SE E; :nV age 835 sec 5
A3: $8 an: 2 <8 an: $98 $8 5.8 $8 $8
32. nad- 32. ES. wwmd- Sod zed 08d- 83 83- :n8 saw 8838888 28E 5
58 A28 E8 am: 80: A28 A38 3.: 8&8 sod
38 $3 toad- +82 3.3- 23. Ste. +3m.o- 82- .22- :uv snow 0338888 29: E
38 5 a $8 68 E 88 Q 88 28
=->_ 8-2 8-30 2 30 8-2 72 30 B 30
£352 “BEGUM-«ma 035mm mun—00m m—Uflo—Z wvoaosm-SQ Dam-Gum Mum—Com
ESE—8 :95. mo moi—gr: A093 .3 82865

 

3.255 :25 can 233 .3 82a.» .3 £3.88 <m .«c 3:558.— wfibuuoeflgo can genome—Ea .3 38B A3.-é. «Eu-H

191

 

 

 

 

 

 

 

 

 

 

 

:3 3mm 3% $2: $2: 3% 3mm :2 $08 $2: 203E030 :o 5:52
nemumuwfldofitoxro
w? o? 8.: 2a Rs 8w 5 as 3-8.2 538
6: M? an 85 8: :33 so. 3.20 5&8E-§-§50
m: 83 So 85 $2 :3 :3. :8.» .8.~ :2 £355 ”misc“. 8 22¢
A88 9.98 888 A86: :3: 5.: 63: a2: G3: 83:
Rod- 23:. ES. :33 :a2 :2 2 :32 :22 :3+ :83 “388
we? mo? m0? m0? mo> my? no? no.” mu> mo> wot—:05. muggy 335‘,on
6:: AB: so: 58 E8 83: $3: 33: 6mm: 63.:
25.8 85.0- Sod- 83:. Rod. :de- :58 2.8:? :vmvo :xvd Assessors 582096 2585
a: .8 c _.8 § .8 am : so: 3.: an: 998 an”: :3:
83. ~38 Rod :2 .o :26 ago- :33- :«Sa. :3 2 :3: cooano .83 8f: E22: .83
5...: 5.: an: 3.: Gm: 808 an: 808 $3: $3:
*fimood- {imoodu axe—wod- *ONodn «cmod- .SLaOdu «*Nmodu {imaodu {*oONd **O~N.O Ammv 0N5 wEEOS—Ej
3.: E: A8: 38 E8 3.: $1: an: 6:: A88
£6. was- $8.? :33 :33 $8.? Rod- mac- ~88 mood- was 883% mafia: 8me
AR: ANN: am: 9.2.8 A38 3.: am: 5.: $98 $98
was ~33 RE $3. $3. 2.8- 38- 33- 20.0 3:. an: 8% wow: anew £85: 335
A38 80.: 35.8 a; 8 5.: A2.: 3w: 3.8 5.8 A2.:
22. 3.8- $2- :m§._- 3:57 :23- :38 .33- :88: 2.20:. an: Sam mow: 88c 8:8 355
E: 8...: 3&8 E: 69: 5.8 3.8 3.: 6:: 93.:
+33 33 33- ME: L58 83- :2. 3:3- $2 .53 2:5“: 8.98: “£3 E
an: A38 898 5.: A38 so: G88 A3.: 95.8 E:
22- £2- 88¢ 33. 23- £3- ooo.o- :55- 5.8 :3 2": 58¢ Ea. sac §
5.8 $98 828 3.8 A3: A2.: 5.: 80: A38 so:
8:: 85.8 $3 83 +82: 8.3- $5- .08.? 22.- RS- 2n: 5% 883333 one”: E
A8: A8. : 3.8 398 A8,: 3.8 A38 A8: 8N8 89:
22- $3- 23 23- ~83 82- N28- :36. :3- :83- an: an: 033:2»: 29: E
3: S e a: 6: a: E: G: Am: 2:
:2 I: 28 2 Bo =->_ 72 30 2 Bo
flute: 6025.5me 33mm wogoom m—ovoz Uoofleomfla o—gm “VD—com
$885 =§§o§ c383: 3:. 2238:

 

Ehauéo can on? 22.82:. _c 829» .8 mists:— <m .3 395958.. ”5&35235 :5 565365 .3 38,—. .264 «Sub

2:35

192

 

 

 

 

 

 

 

 

 

 

 

82 :88 88 ~88 ~88 :88 =me :88 N88 ~88 223538 8 888:2
cosmoataovcgo
NZ 2 8 88 $8 88 88 .8 82 83:2 swam
88 88 :88 R8 8: 3.8.8 as 8-38 magma-83-5832
:88 :88 :38 :88 :88 .88 .888 :38 :88 :88 Ease own-85E .5 82-8
8n: 88: 88: 888: 888: 88.8 3.8 5.8 8: : 88: :
:83 :888 :38 :88: :82 :88 888 288 :E: :2 2 8528
we? mo> moxw mo> m0? mo? 3? m0? we? 3% UUUEUE mega—u Emu—Papa
88: 88: 88.8 88.8 88: 88: 8:: 88: 88.8 88.8
:88 :8: 8 :8: 8 :888- :888- :82 8 1.8.8 :82 8 2.888- :888- 38888882 888888 883%
88.8 88.8 88.8 888: 888: 28.8 88.8 88.8 :88: 888:
N88 288 :88 :988 :38 :88- 888- 888- :888 :88 c8818 .888 Sin: €268 ~33
88.: 88.: : _.: 88: : 82 : 8...: 83: 88.: 888: 888:
888- 888- 888- :888 :888 8888- 888- 8888- :888 :888 8:: :8 8522883
E: 88.: 8:: 88.8 88.: 88.8 88.8 88.8 88.: 88:
i 88- 888- :88- 888- 888- 888- 888- 888- 888 888 8E 882928 mesa: 8883
8_ .8 88.8 88: 88: cm: 88.8 3.8 88.8 88: 88:
:828- :88- :88- 388 888 8888 888 888 2.8888 £8888 an: 8% 88: 588 8388 :58
88.8 88.8 88.8 88.: :8: :8: 88.8 88: 88: 88.:
888- :88- 388- +888 +888 :88- :38- :58- +888- we 8- an: 8% 88: 888 3:8 852m
888 28.8 88: 88.: 88: 88.8 88.: 8...: 88.8 88:
888- $.88- :83- +82 :38 88. 888- 828 888 1.8888 2...: 888 £888 358 <8
88.8 88.8 88.8 8 fl .8 c H: 88.8 88.: 88: 88. : A: .8
888- 888- 888- M: _8- :88 888- .888 2.8.88. 88; :888 Cu: 888 83: .288 E
88: :8: 3.: :8: E: 88.8 88.8 88.8 88.: 88:
888.7 :E: :82- 888- :28- 888 ~88 888 £88 .888 an: 888 8888:8888 28.8 E
88.: 88.: A2.: 88.: 88: 8:: 88.: 88: 88.: 88:
+888 :88 828 :8: .888 .888- .8888- :888- £2- :888- :u: 888 83883888 2:8 <8
3: S a a: 40: C: a: 6: E: 2:
8-2 F: Ego >8 30 8-2 I: 30 >: So
£0602 “gouache-«fin— o—gmm Uofioom flofioz ©0255me BnEuwm Una—Com
Conga: meadow A3955: 332

 

338$ ES 838 .3 boa—Es: .3 5:538 <m 3 8:33:83 “5:532:33 can bmonoucuao 3 38,—. .264- 2.3.—-

193

 

 

 

 

 

 

 

 

 

 

 

88 82 :88 8:: 8:: 8% :88 :38 82: 8888 88338 .8 .3882
nocmomtaovto>0
:8 $8 8:. 2 8 R8 88 3.: :2 838.2 583m
88 £8 :88: G8 88.: 2.888 :2 3-30 gmaSam-saénaa
.88 .83 +2.8 3: :88 8.: 8.: R: 88.: :88 8:588 own-oat: 8 32-..:
:88 888 888 A888 :88: 5.8 8:: 88.8 @888 E88
amod- Sod- Sac-o- ibmmd ELNWN tum—m4 .8.;va 3.5mm; _..*mv—.v tyne—é ESQ-50
80> .85 we? we.» 8.» 3.» 85 m“; 3% 8% coca—2: 83:56 1.633 E
:88 A88 88. 8 58 E8 8888 8888 3888 G88: 8888
:88- 888- 888- 888- 888- 2.8-9.8- 3.888- :888- 2.898 :888 3858582 8588808 3:385
G: .8 2 _ .8 a: .8 858 $8.8 88.8 88.8 5.8 8888 2888
888- N88- 888 :28 :88 1.288- .388- :388- :3 2 :82: A825 .88 88": €23 :82
$88 5.8 88.8 $8.8 $8.8 88.8 88.8 88.8 33.8 $888
:888- :888- :888- .888- ...888- :888- :888- :888- :83 :888 38 3: 8529853
3.8.8 A? 8 $8.8 :88 E8 28.: 88.: an: a: .8 :38
888- 888- +888- :388 :888 8.88- 888- 888- 888- 888- was 8:29.28 832: 80883
5.: :2. 8 A2.: 9:8 88.8 888 88.8 38.8 88.8 888
8:8 S8 38 888- 888- 888- 88.8- 888- 288 888 :8 88m 888 £88 8:25.: 352m
:88 88.8 88.8 c; .8 5 .8 $88 $8.8 88.8 $8 a: .8
888.8. 388- RN8- :§:. :88:- ER8- .8888- .LE8- :88:- :~8._- :8 8%. 888 838 335 352m
E: 88.8 88.8 88. 8 88.8 :28 28.8 g. 8 $8.8 88.:
+888 38.: 2:8- 88.: .888- 828- :88- +888- ~38 :83 cuv 538 3:58 358 E
88.8 88.8 38.8 5.8 :88 88.8 38.8 28.8 88.8 5.8
88:- 888:- 888 888:- 888- 88:- 888- :58- E8- 888 :8 £88 8:8. 5:8 E
5.8 88.8 88.8 $8.8 $8.8 88.8 88.8 88.8 88.8 98.:
2:8 888- 828 88.: +3.8 8.8.:- SE- .888- 888- 288- :8 £88 masoavasn 2888 E
88.: A8. 8 5.8 $8.8 88.8 :38 38.8 :88 88.8 88.8
22- 88.. 7 828 88:- 3.8 8: m8- 888- :888- £88- :8888- :L 588 0808888: 28: S:
38 S a 8:8 68 E 88 68 d8 98
:5: :5: :30 2 8.5 :5: _->: 30 2 30
£380: voodouobwa 23mm “we—00m £0602 tonnage-wig o—gm Un—OOA
Conn—=5 mEO €35.58 8.3m

 

.8.-aw ES 93.. «a .395:— no mam—3.5:. «S .3 8:38:88.— wiéunoutog 6:8 585385 m: 88H .364 «Sah-

194

 

852832..” mo 38. $3 + 28 guano—mama mo :32 {on ... .oogomammm .«o _u>o_ ..\o_ 1. 882

38 Ba 88 5:5 355023 Ba 8382 3:5 :25: mom mam-33:58 H858

 

 

 

 

 

 

«o .5956 E owaanu

33 20:83: <

8m 5m 28 .0.ch 8m $3 28 8m 23 MSN ”Santana 80 3:52
3o w ~ .o 2 .o m; 26 t .o m; e .o S .o 26 vegan-m
$8 +2: o: 28 a: :2: 2:3 :85 x: :33 games 355

+2: 2: :3 m: :53 :ME :26 :58 a; :2: 93:2: own-2:5
Lox. 58“ “2.85
8% 8% 8% 8% 8% 8% 8% 8% 8% 8% 8:556 0:5 x 85>on
A8,: 8 .8 am. 8 5.8 and 6+8 2 m. 8 5.8 $8 $8
+83- :m: .o- a: .8 288.8 s: 3o :33 me .o :23 mad- £2. 2988
858 :9: 2A.: 858 an: $28 8.8 89: 5. 8 $8
a _ .o- 23- S 3. me .o +32- +22- 83 +33 $8- ano- §§a 332m
858 3. 8 SM: 828 A58 2 ”.8 A88 A38 A38 A88
3 _ .8 $2.? $3 God a: .0 +33- .Lfid- :89? £2- ,324- 838 852m
RES-BE A1385
$8 808 $8 E8 5.8 A38 88 9:8 8.8 $8
+2.8 «.8.? S S- m: .8 2.5.8 .23- on; .2 no :88 me .8 ”38888-8.“ 295m
5.8 5.: $8 8.: 5.8 308 A88 2 n: 8.8 80.8
.2 2- 83- 33- 33- 3; EE- :_8.o- Nad- +m£._- .36.? 8888888: 2“:
A88 G~8 :08 f: §8 808 A88 888 an: ac:
wood- Sod- me .o- 2: .8 +888 :89? 83. 3-3 ~28- +83- %888: 283
5.8 5.: an: ah; 8+8 65 8 and :88 E8 :88
”So 33- 228 +32- :3. +83 +5.? :33- flno- :88- 038838 2+:
«ENS-BE :33 «Ft www-mfitm
boom :02 “com “com aoz noon Hoom QOZ “com Hook aoz Hoom “com GOZ Hoom
2:0 83m mofifiom 832 839300

55.53 .8..

cote-=8 «e Z 835» 5.825 as I: 3 538a can Scam» an 55.9-38.5 I: .8 3:38.: «E go 8.3:: 2:- .m~.m< «Eu-H

195

 

9:5 was 338%? .3809 55 .8233 .uoboo .EBOGASm £330 ova—2: 305 029,
swig .3983 «023 .33me 06205 30.8 Hood» .380 v08 338 550 =a ES «5538 anon—3 43:8 .055: ”09:05 £3“an .388 @3283 “£59m
6:55-59sz 53 33328 moans 3a 8855qu 3 $3552 .oonuoaawmm mo ~96— o\oo_ + 98 cacao—mafia mo 33: o\om ... .35anme mo 3%: o\°_ .1. ”8on

voom was SON .xoPSm Educ—293m wan oochoe >o>§m $332 “mom :mEOUEOmU 682%

 

 

 

 

 

mEBEmsoE 8mm meccatomno Mo 82:52
mg m _ .o m3 m3 2 .o a .o 88 one team-m
+8; 8: EA +82 $2 Ed #2 .8? Ego: 8 :28
m0? m0? my? m0? m0? my? me? mo? mugfi 08% x noggin
A3. 8 5.. 8 A38 A38 5. s an: $8 5.8
So 85 So- So- 2.? 2.? 3o So 3588
so: A3: 5.: Gm: A88 $58 $8 8&8
So .85 3o 35 N;- NE- 35. 3.? 33:8 355
5.8 2 58 an 8 8%: $8 818 88.8 €88
So- So- 86- 3o- :3. .LS- :28. ...Rd- 838 252m
ASS-SE Arms-m
A2.: 5.8 $8 so:
So. .8.? 3.? Bd- owaamfifiaoa 220m
308 Gm: 68.8 A38
:3. 3o .2 .o- Bd- asoqmesn-aoa as:
E .8 5.8 so. 8 2 m8 .
86 So- +8.0- Nod- 25%88: 038$
888 SN. 8 $08 92.8
.36- 23- ...mg- :53. oaamag: as:
$8 :58 ANN-.8 $8
8.0- .8.? So. So. 22$
5. c 5.8 $8 Gad
+8.? :3 .2 .o- .3 .o- 222
ASS-BE “Nana «3t mMQ-mEN-i
:8 68 3 m: 88 g #8 28
own—08 039-35 £308 Hook can muons-- .m~mEoU meatgov

60528 SEE <

 

AS: 56:: «98 80 m2 1233 E owgo

 

35.5: E 688:8 .cz “22.8.8.— 5 .858; ES .52—cm 3 ES— eSa-E—au .8 5:38.: dd 8° 33:: 2E. 63% 035,—-

196

 

.385 $.58»me 8808 5.5 .0832 £0888 £36895 8038 08305 808 03? awn-mu .8883 «83m @23me 08305 380 83F .380 8008 SEE
5&0 =~ 38 58:38 43:3 43:8 6.5.38 6832: £330“ .ooqaumiwmm no .32 $2 + 85 ougomawmm .«0 _o>o_ o\om .. .ounaociwwm mo ~o>o_ fix; 3.. “3on

vooa 85 58m SEE—m Egan—masm 85 808892 NAo>H=m $053 «mom namEOUAngmU H855m

 

 

 

 

 

 

”38 888: 83 mo w2 583 E owfifiu

82928 8:85 <

$8 :3 E8 :8 $8 :3 «SN :8 maoumtanfio H0.8.32
86 «.2 m3 owe 8.0 85 3o .83 Bang-m
Ed 2 .o 88 $2 35 +22 who a: 3:358 352m
82 $2 :3” 8.8 2.3+ :d :3. 83 ESE: awn-25$
ugh-r8“ 3.85
muxfi mo> mo> no? m0? m0? m0? mgr mogw 095 x flatworm
888 80.8 688 3». 8 $28 828 93.8 $8
~39 358.c- :o.o $8.? :83- m2 .o- :88? +33- 5588
5. 8 388 65 8 58 A38 9: .8 A88 5.8
838 :55 +82.- Sod 2 2.- Rod- Sod- 83 32050 o .E 333 a £83050
A38 A38 $8 $8 2 88 5.8 308 § .8
Sod ooo.o- $3. £3 23. 83- Sod «.2 .o- 33:8 355
3.8 5.8 5.8 5 .8 E: 8 388 SN. 8 5.8
SE. Sod- ooo.o .2 S- SS- +mw~.¢- 33. t; _ no- 888 :55
A5235 Atmfim
328 $88 28.8 a «.8 $28 888 898 A28
was- mood- 188.c- :28 Rod- 83 N30- :58 ”Sawing-:8 258»
21.8 E .8 88.8 828 888 898 5.8 $0.8
v8.9 was «88 98° .83- 33 +5.? ooo.o omaoamasa-Sa 2m:
88 .8 $88 A38 A38 888 A88 A88 688
Sod- 89o m3? Sod was Sod- wood Sod- 0308.88,. 22$
A88 80.8 628 $58 5.8 858 £38 58
N86- 388.c- :o.o $8.? :22- RS- :88? +33- asoaasa 222
ASS-SE “388 «vi» mum-wail
80m 802 “com 80m no Z “com “com :02 Boom “com :02 “com
ammob 029133 n.305 88 van muonnh .8880

$35.50

 

83:53 .8.—

Euuotcu 32 85:3 €9.68 .3 I: 5 .8388 can .515» .3 EE— 88353 .8 .0253:— <m .8 «can: 2; Rafi-<- «Ea-n.

197

 

.8880 888888; $880M
6823-8383 585 888—823 888-8 03 8858088 E 985882 .388me6 mo 3%: $3 + 888 38858me 80 _o>o_ o\om _.. .oogomawmm 80 ~26— o\o_ .1. ”8qu

38 28 88 82:5 38808.85 23 885% team 8258 888 mmmaooSéOmu H850m

 

 

 

 

 

 

$588980 80 0285 $8085 4

08980 80 08—? $8085 <

83 8 2 cu m as m 23 :2 Ms m as macaw-capo 80 585:2
Rd 2.8 $8 :8 mg NS 28 m 8 .o 8.23.9-8
N3 N3 3o «.8 £8 83 .88 .52 games own-08E 8 83-8
mo> mu> we? moxw mo> “Wu-W mo> we? mug—u 08...“ x 005.»on
88.8 828 83.8 as. 8 898 A28 888 $8
838- :3. mod- 53. .88.? :3. 89¢. :28. 8388
82.8 698 3:8 :88 388 3.8 $8 888
85.8 83 :3. 83. ooo.o 83- :ad- 88.? £988 3528
$8 $8 $8 3.8 :98 88.8 5.8 88.8
858 22 as .o 2 8 .o «8.? ~23- $2 .8 +38 .8 888 :35
ASS-BE ”€th
88.8 808 888 3.8 $8 cm. 8
Sod 85.? £3. ES- Sod- 83- 888888888 2858
5.8 93.8 5.8 38.8 38.8 83.8
~88 Sod- v3.8 +82- 83 88.8 asafimfiaoa 0::
888 $88 $8 3.8 888 so. 8
$3. @88- 23- 59o- 83.- 8: .o- 080%838 2858
88.8 5.8 3.8 608 88.8 8.0 8
+83 «.8.? ”So 23. ammd- £28- oaafisn 082
898 am. 8
8.3- 83. DEE-.8
85.8 :8. 8
v8.8 «8 .o- 222
ASS-BE .8388 «Ft 8&8-28.2%
:8 68 E 38 R8 68 88 :8
Son 80 Z 800m --l--o_QEmm =88 ........ Boa 8oz Sam ........ £988 :88 .......... 388.5260

 

885-58 .8.“ 8382.88 82 "22.882—
5 83:38 888 .5880” E. 9:38.— ..on «8&8: $8..» 888 «8988 .3 08.8.» 88..» 88 3:88.85 <m .3 .8895 2: 88.2 838,—.

198

 

 

 

.0883 $0535 358..
can EEO-«E .0883 own? cum hum—am ova—ca 2:85 Sufi-boa .EBQE U5 83.6 £3330 .mmfi dooam .Baow own—03 £958 :35. duo-Co @8233 aged
-o:§>¢un:m 55 35:28 wows?» o3 885559 3 $3852 .ooaaoaawfi mo 3%: $3 + 23 850$?me .«o _o>o_ gm .. £0.80me: mo _o>o_ ex; .2. ”8on

32 Ba 88 Sufism Bagging-W Ba 8383 325m Eta: mom agony-«280 83cm

 

 

 

 

 

32:35 Ema-b0

.mEEEa =«Em mo 83m>

038 mo 32.;

 

mEonomso: 8mm 3232030 Mo 69:52
98 and M: .o w; 25 m3 Essa-m
Rd 2 .o 2.3m :an S." RN £388 8 ma.»
8% mo> 8% mo> 3% an; 8:556 can x 353$
M: fl .o +83- $8.?
$8 2 m8 awed $8 2 n: 5.: 2588
£36. 2.5.? and .33 ~86 355
A2.: a: .3 5.8 Gas 808 82: 83.5 352m
«33 5d 29¢ 33 02 .o- 5 .0-
A38 A38 5.. 3 8:: Ed A2.: 815 352m
was- 82- 82- 32- $3- 82- $235 $55
808 2 NS 8qu
£3- .83- 82- omaoamemfiaoa ”Baum
see :2: § . 3
23. Sad ~39 umsoameaoaaou 2“:
$58 5.: cm:
23 $8.? 3 5o- asamesn 295m
awe and Ge. 3
2 no :83- +omm.o- games: 2“:
A38 fins E . s
_ :3 32¢. § .o- 255»
5.8 3.: Go. 3
M: w .o +82- +33- 2“:
REESE “Nana «Ft mMU-mfitl
Q m: Re 6V EV as

83550

 

we me. 353: E owzfio

EVE-Ea you @8828 «o Z ".35ch 6:: Macaw ha 2:35 Shauna: “in 30mm.“ .5 325.55 «S no Baa—E 2E. .m~.m< «Zah-

199

.0883 “”3355 1258
was BEBE .0885 own? van aha—am ova—03 0885 Sunfibo... .8398 23 £26 .3330 .me £35 .flaow 322: £555 :39 .383 “:8qu “£50m
-o~§-uon:m £3, 33333 333 08 3353an E 33:32 dunno—mama mo _o>o_ $3 + 28 330536 .«o .33 o\om .. dunno—mama no ~32 £2 1. 3on

38 as 88 .525 53535 23 8382 35m it“: mom $303503 H830m

 

 

 

 

 

a

2:02: 83.380

«2955 =«Em .«o 32a>

uo wB 3:58 E 09:30

233 mo 33a>

$8 :3 $8 :3 $3 :8 323530 we Benz
03 33 v3 mg 88 OS 8:333
25 2: as who 24 :8 59.2: 355
Ed :3 3.5m R; 85 m: 382: $355
38‘3““: 3.8%
3% 3% 3% 3% 3% 3% 3:556 85 x 353E
5.8 5.8 58 §8 6&8 80.:
1.00%.? «*chwdu *uo~m.o- 3,206 hfiod mmod “magma—CU
88 am. 8 8N8 $28 $8 $8
89o- +2: .o 336 $3 22.- 83 8358 355
A38 58 A38 A8. 8 2 m8 2 m8
83- 22- M33- $3- 83. 338 fine 352m
A5385 533m
888 8M8 Gm: 358 2 n8 a8
33. ~28 $2. $3. 5 .o- a: .o- aaogwegnaoq 220m
58 808 3.: cm. s so: 938
23 N28 22. 82 $3- $3. usamasnaoa 2“:
$8 808 5.: $8 5.8 § . 8
$3 3:5 +33- $3- 9; .o £3- 03838; 295m
58 2 _ . 8 A38 3. 8 58 58
:35 £2 :8 .o- .28- $3. $2.? 5&38: 032
E m: 88 Q 88 A8 3385 38 «$8 .3-»er
50m 82 Son Son .82 Room 80m aoZ Room

33530

 

.8553 .8.. 339:8

32 8:3: 5.52:— .E .853: can 393m .3 2:35 Banana: 1:: 83.3 :c 3:338 Sm .«c 3.3:: 2E. .cn.m< 93.;

200

.315 and 338%?» .853 Amt“ .8258 .oobOo 33095..” .838 £535 303 03a.»

swan .3238 “33m .3830 09:03 30.5 80%..

.395 698 SEE 350 an wan 8.598 .833 42:8 .oNEE ”0333 “.3030“ .383 ESE—am “waned

-Bgéunsm £5 33323 3:93 03 33533m E 33:32 .ooqaomammm .«o :32 $3 + was 852.“?me mo _o>o_ o\om .. 603.0%»? .3 BR: .x; 8.. ”3on

vocm v3 SON Sufism _S:qu_nE:m ca.» 833%— >255 “33$ “mom namEOUgOmU u3.30m

 

 

 

mEonomsom ca m aeratompo Mo 39.52
m; w; 03 £5 2... Ed NE. N3 3.23m
3% wow. 3% mo? mo> we? mokw mo> modded—w 053 x uofi>em
an: an: 808 A88 60.: so: 828 .8.:
3o 86 S8 88 i _8 i _8 35 2:. 8888
$8 .88 SN. 8 828 A28 A88 A88 A38
Nod 88 38 3° .2 8 38 S 8 S8 .22.... 3.02....
58 A88 AS. 8 85. 8 .28 38 $8 $8
88 88 +88 +88 .88 .88 :28 1:8 .28. 3.05
A3835 bumEM
38 A28 2...: 3.8
88 .88 :8 +m _ 8 8.23888. 283..
A38 58 8.8 am:
88 3o .28 a 8 8.23888: 222
E8 .88 A28 2 _8
8.0 88 88 S8 08338.. 25......
9:8 .88 A28 908
.38 88 .88 :28 08838.. 2.2
.38 as. c 58 am. 8
88 +38 88 88 28......
5.: A88 58 Ed
+88 :3 :38 :88 2.2
5:885 aura «3t www-mfitm
:8 68 Q m: 88 Q 88 28

 

0303 vagfiwi

macho 88 93 £33.

.2330

 

A33 395 33 .3 MB 353: E omafio

832:8 855 q

33330

 

2953 183.53% "22.332. 3 353a 15. .563” .3 US.— uoua>§=o no mum—«22: dd .3 3.3:: 2:. 4&2 039—.

201

 

 

 

.08005 03503 385

53 3.855 .0885 0mm? 53 358 05:05 08005 83.3.50; .8598 53 3050 803050 .0.wa .3050 .flmow 05:05 £953 =8m. .0380 @3933 330M
6555.385 55 “00350—3 358-“ 03 3358058 5 039852 308.0058me no _0>0_ $2 + 53 0035538 mo _0>0_ o\om .. 3032.583 .«0 _0>0~ o\o~ .1. 38 Z

38 05 88 .325 335330 05 08888 3350 .858 0.3.8 33005803 .855

 

202

 

mU—Ofiowzom Ow — w aeratomno mo MODEM—Z.
.23 v2 8 .o 2 .o 35 2 .o 0288-8
3? mo> my? 0.0%. m0? moxw mogfi 083 x 005.?on
88 808 A88 A88 and an:
:58 :83- :2 no :2 no 88.0 80.0 88.80
am. 8 E8 :28 g8 88 88
8:3 033 830 Sod 308 2:8 .228 €003
$08 658 ANN. : A8,: 2 Ev 5.:
~38 38 :8- ~38 v8.8 88 .28. 232..“
5.5885 533%
38 808 58
808 :58 88 0833888.. 283
A38 88 as. c
808 08.0 088 0383888: 222
E8 65: 88
83 +8 m8 v.18 08338; 285
A88 A38 3.:
£20 :33- 338 0333.3 222
38 3.8 $8
208 :58 M58 228
5.8 A28 A28
83 3.38 :2 m8 2.2
A5835 :83 «Ft 0MU-0E.EQ
E Q 80 Q 80 A8

 

08005 8580 .5850 :80 mo 35m>

mo wo_ 388 5 3850

0330 Mo 350>
.. 3:53.60

 

0883
33:33“ "22.02::— 5 5053a can 353» .3 08005 83.3.3 53 3003 :0 3:338 4m .8 3a98_ 0.; .NN.M< 035.

 

20.5 83853 fissoMéHEk/Lonsm 85 808328

moan; 08 885588 E E09882 .8528in .«o 3%: $3 + 8.8 8:85ch we _o>o_ AXE * .ooaaomiwmm .«o 3?: £1 1, $802

32.. can SON $3.5m Euaoaoannm can oooQamm— ~A058m $053 “mom mamEOUgOmU H850w

 

 

 

 

2mm 3% $3 6% £2 3% 8% $3 823530 .6 £832
28 a3 8.0 8.0 m3 «.8 2 .o t .o 8823-8
mo? 8% my? mo> mu> me? my? mu> muggy 05% x ouam>oum
:38 A38 A88 25.8 38.8 $8 828 $8
ooo.o 83. $3. Sod- Rod- :38- 88.8 828- 6585
A88 A88 A88 58 9&8 5.8 828 5.8
Sod- 39o Sod :55. 88.? 3.3- 88.? Rod- 8888 beam
808 A28 am. 8 888 3.8 68.8 5.8 £8
23 mm 8 .o 2: .c 2: .o ooo.o- 8.28. +33- +8 _ .o- 838 352m
b.8835 38.3%
a; .8 a; .8 30.8 5. 8 808 A38
23. Sod- Bod- SN? 83. 83- ”macaweafiaoa 285
$8 $08 E8 308 § .8 8&8
23. £28- 2 S- 82- 33 2.8- asafisnaoa 0::
$8 £8 A88 A38 :08 5.8
088. Rod- $3. 83. Sod- .88.? 088.0588: 22$
38 388 $8 3.8 5.8 an 8
+8.3 Sod- Sod 2.8- 83. 33- 0338328 0:2
A88 A88
Sod- E .o- 285
88.8 5.8
:3. 83- 082
5:335 8388 QR» mm8-mE.2n~
:8 68 E Q 68 Q 88 $8
goon co Z boom ..... 038mm ==m-----:- Boa no Z 80m 2.3!---2888 =37 ........ 38530

 

$559.5 mo 02? 38085 <

959:0 mo 2:? 38085 <

touomhmom “22.02::—

5 .5525 ES .385” 3 c.382— 3; 35.5 36..» can 33.8 no 2:3 39.» .8 @5388 <.— .8 83:: 2: .364 ~33.

203

REFERENCES

Ainsworth, M., and I. Semali (1998). Who is most likely to Die of AIDS ? Socio—
economic correlates of Adult Deaths in Kagera Region, Tanzania, in; Ainsworth,
M, Fransen, L. Over, M (eds) Confronting AIDS: Evidence for Developing
World, European Commission, and Brussels.

Ainsworth, M., and J. Dayton (2000). Is the AIDS epidemic having an impact on the
coping behavior and health status of the elderly? Evidence from Northwestern
Tanzania. World Bank, Washington, DC.

Ainsworth, M., Ghosh, R., Semali, I. 1995. The impact of adult deaths on household
composition. Mimeo. World Bank, Washington, DC.

Alderman, H, Behrman J ., Hans-Peter Kohler H.P., Maluccio J., and Watkins S., 2001.
“Attrition in longitudinal household survey data: Some tests for three developing
country samples”. Demographic Research. 5: 78—124.

Araya T, Reniers G, Schaap A, Kebede D, Kumie A, Nagelkerke N, Coutinho R, Sanders
E. Lay diagnosis of causes of death for monitoring AIDS mortality in Addis
Ababa, Ethiopia. Tropical Medicine of International Health. 2004 Jan;9(1):1 78-
86.

Barnett T., and Blackie P.M., 1992. Aids in Africa: It’s Present and Future Impact.
London, UK, Belhaven Press.

Barnett, T. (1994). The eflects of HI V/AIDS on farming system and rural livelihood in
Uganda, Tanzania and Zambia, FAO, Rome.

Barnett, T. and Whiteside, A. (1999). HIV /AIDS and development: case studies and a
conceptual framework. Health Economics and HIV /AIDS Research Division,
University of Natal, Durban. Online
atzhgttp://www.und.ac.za/und/heard/papers/1999/EJDR.PDF

Barnett, T., J. Tumashabe, G. Bantebya, R. Ssebuliba, J. Ngasongwa, D. Kapunga, M.
Ndelike, M. Drinkwater, G. Nitti, and M. Haslwinimer, (1995). Field report:
The social impact of HIV /AIDS on farming systems and livelihoods in rural
Afiica: Some experiences and lessons from Uganda, Tanzania and Zambia.
Journal of International Development 7, pp 163-1 76.

Barnett, T., and A. Whiteside. 2002. AIDS in the T wenty-F irst Century: Disease and
Globalization. Palgrave Macmillan, Basingstoke, UK.

204

Beegle, K., (2003). “Labor Eflects of Adult Mortality in Tanzanian Households, " World
Bank Working Paper, Washington, DC.

Birbeck, G. 2004. Assistant Professor, Department of Epidemiology and Neurology,
Michigan State University, personal communication, November, 2004.

Booysen, F. 16 R., (2003 ) "HI V/AIDS and Poverty: Evidence from the Free State
Province, ” South African Journal of Economic and Management Sciences, N.S.
vol. 6, no. 2, pp 419-38.

Bryceson, D., Deborah F. and F onseca, J. (2005). Risking death for survival: Household
and village responses to the threat of hunger and HIV/AIDS in Rural Malawi .
Paper presented at the International Conference on HIV /AIDS, Food and
Nutrition Security, South Afiica, April 14-16 . Available at:
http://www.ifpri.org/events/conferences/2OOS/durban/papers/brycesonWP.pdf

Central Statistical Office, Ministry of Health, Macro International (2003). Zambia
Demographic Health Survey (ZDHS), 2001/2002, Lusaka, Zambia.

Chapoto, A. and T. Jayne. 2005. Characteristics of individuals afflicted by AIDS-related
mortality in Zambia. Working Paper, Food Security Research Proj cot/Zambia.
Lusaka, Zambia.

Chin, James. 2003. Understanding the Epidemiology and Transmission Dynamics of the
HIV/AIDS Pandemic. School of Public Health, University of California/Berkeley.
Paper presented at USAID, Washington, DC, October, 2003.

Cuddington, J .T. (1993). Modeling the macroeconomic effects of AIDS, with an
application to Tanzania. World Bank Economic Review, vol. 7, pp 173-189.

Davison, J. (1988). Who owns what? Land registration and tensions in gender relations of
production in Kenya. In Davison, J. (Ed), Agriculture, Women and Land: The
African Experience, Boulder, Westview, Colorado.

de Waal, A. (2003). How Will HIV /AIDS Transform African Governance? African
Affairs, 102, 1-23

De Walque, Damien, 2004. "How does the impact of an HIV/AIDS information
campaign vary with educational attainment? Evidence from rural Uganda,"

Policy Research Working Paper Series 3289, The World Bank.

Doctor V. H. and Weinreb AA. 2003. Estimation of AIDS adult mortality by verbal
autopsy in rural Malawi. AIDS 2003, Vol. I 7 No 1 7 pages 2509-2513.

Donovan, C., L. Bailey, E. Mpisi, and M. Weber (2003). Prime-Age Adult Morbidity
and Mortality in Rural Rwanda: Eflects on household Income, Agricultural

205

 

Production and Food Security Strategies. Research report, Food Security
Research Project, Ministry of Agriculture, Livestock, and Forestry, Kigali.

Dorward, A. (2003). Modeling Poor Farm Household Livelihoods in Malawi: Lessons
for Pro-Poor Policy (DRAFT). Center for Development and Poverty Reduction,
Imperial College, London; Wye Campus. Paper available at:
http://www.wye.ic.ac.uk/AgEcon/adu/research/proiects/ppag/mlhhmdrfpdf

Drimie, S. (2002). HIV/AIDS and Land: Case Studies fiom Kenya, Lesotho, and South
Africa. Report for Southern Africa Regional Office of the Food and Agricultural
Organization (F AO), Human Sciences Research Council, Pretoria, South Africa.

Drinkwater, M. (1993). The eflects of HI V/AIDS on agricultural production systems in
Zambia. An analysis and field reports of case studies out in Mpongwe, Ndola
rural district and T eta, Serenje district. Rome, FAO.

Du Guemy, J. (1999). AIDS and agriculture in Africa: can agriculture policy make a
difierence? Rome, FAO.

Epstein, H. 2002. The Hidden Cause of AIDS. New York Times Review of Books,
49(8), May 9, 2002.

Epstein, H. 2003. AIDS in South Africa: The Invisible Cure. New York Times Review
of Books, 50(12), July 17, 2003.

F A0 (1993). The eflects of H1 V/AIDS on agriculture production system in Zambia:
Analysis and field report of Case Studies carried out in Mpongue, Ndola Rural
District and Serenje District: Adaptive Research Planning Team Report, Ministry
of Agriculture. FAO, Rome.

FAO (1995). The eflects of HT V/AIDS on farming systems in Eastern Africa. FAO,
Rome.

FAO (2003). HIV/AIDS and Agriculture: Impacts and Responses. Case studies from
Namibia, Uganda and Zambia. FAO, Rome. Downloadable at:
http://www.fao.or/sd/2003/PE12013f1en.htm

F A0 (2004). HIV/AIDS, gender inequality and rural livelihoods. The impact of
HIV/AIDS on Rural Livelihoods in Northern Province, Zambia. F AO, Rome.

FASAZ (Farming System Association of Zambia) (2003). Interlinkage Between

HIV/AIDS, Agricultural Production, and Food Security. Integrated Support to
Sustainable Development and Food Security Program. F AO, Rome.

206

Fitzgerld, J ., Gottschalk, P., and Moffrtt, R. (1998). An analysis of sample attrition in
panel data. The Michigan Panel Study on Income Dynamics. Journal of Human
Resources, 33: 251-299.

Garenne M, and O. Fontaine (1990). Assessing probable causes of death using a
standardized questionnaire: A study in rural Senegal. In: Measurement and
Analysis of Mortality. New Approaches (eds J Vallin, S D’Souza and A Palloni),
Clarendon Press, Oxford, pp 123-142.

Gillespie, S. (1989). Potential impact of AIDS on Farming Systems. Land use Policy 4,
pp 301-312

Glyrm R. J ., Carael M., Buve A., Anagonou S., Zekeng L., Kahindo M., Musonda R.,
“Does increased general schooling protect against HIV infection? A study in four
African cities. Tropical Medicine and International Health Volume 9 No 1 pp 4-14
January 2004.

Goldberger, A. (1972). Selection Bias in Evaluating Treatment Efi’ects: Some Formal
Illustrations. Madison, Wisconsin: University of Wisconsin Press.

Gray R.H., Wawer, M.J, Brookmeyer R, Sewankambo, N.K, David Serwadda, D.,
Wabwire-Mangen F., Lutalo T., Li X., VanCott, T., Quinn, T.C., and the Rakai
Project Team Probability of HIV-1 transmission per coital act in Monogamous,
heterosexual, HIV -1 discordant couples in Rakai, Uganda. 2001. Lancet, Vol. 357,
1149-1153, Apri1.14, 2001.

Gregson, S., H. Waddell, and S. Chandiwana. 2001. “School education and HIV control
in Sub-Saharan Africa: from discord to harmony?” Journal of International
Development, vol.13: 467-85.

Hallrnan, K., and Grant M., 2004. “Poverty, educational attainment, and livelihoods:
How well do young people fare in KwaZulu Natal, South Africa?” Horizons
Research Summary. Washington, DC: Population Council.

Hargreaves, J. R., and Glynn, J ., “Educational attainments and HIV-1 infection in
developing countries: a systematic review. Tropical Medicine and International
Health, Volume 7 No. 6 pp 489-498 June 2002.

Harvey, P. (2004). HI V/AIDS and Humanitarian Action. Humanitarian Policy Group
Report 16. London: Overseas Development Institute. Downloadable
at:http://www.odi.org.uk/hDR/papers/hpgreportl 6.pdf

Haslwimmer, M. (1994b). The social and economic impact of HI V/AIDS on Nakambala
sugar Estate. F AO, Rome.

207

 

Hausman, J .A., and WE. Taylor, 1981. Panel data and unobservable individual effects,
Econometrica 49, 1377-1398.

Heckman J .J ., and T. E. MacCurdy (1986). Labor Econometrics. In: Griliches, Z., and
Intrilli gator, M.D. (eds.) Handbook of Econometrics. Elsevier Science
Publishers, Vol 3, pp 1945-51.

Heckman, J. (1976). The Common Structure of Statistical Models of Truncation, Sample
Selection, and Limited Dependent Variables, and a Simple Estimator for Such
Models. Annals of Economic and Social Measurement, 5, 475-492.

Hosegood V, McGrath N, Herbst K & Tirnaeus I.M. 2004. The impact of AIDS mortality
on household dissolution and migration in rural South Africa. AIDS 18(1 l):1585-
1 590.

J anj aroen, W. (1998). The impact of AIDS on Household composition and consumption
in Thailand. In; Ainsworth, M, Fransen, L. Over M (eds) Confronting AIDS:
Evidence for Developing World, European Commission, and Brussels.

 

Jayne, T.S., Villarreal, M., Pingali, P., and Hemrich, G. HIV/AIDS and the Agricultural
Sector: Implications for Policy in Eastern and Southern Afiica. The Electronic
Journal of Agricultural and Development Economics Vol 2, N02 (2005), pp 158-
181. Available online at www.fao.org/es/esa/eJ ADE

Kitange H.M., H. Machibya, and J. Black. (1996). Outlook for survivors in sub-Saharan
Africa: Adult mortality in Tanzania, British Medicine Journal, PubMed Abstract,
312, 216-220.

Kwaramba P. (1997). The Socio-Economic Impact of HIV/AIDS on communal
agricultural systems in Zimbabwe, ZFU, Friedrich Ebert Stifiing Economic
Advisory Project, Working Paper 19, Harare.

Lacey, C. J .N., D.W. Merrick, D. C. Bensley, and I. Fairley (1997). Analysis of the
Socio-demography of Gonorrhea in Leeds, 1989-93. British Medical Journal 314
(I) pp 715-18.

Mather, D., Donovan C.,, Jayne, T.S., Weber, M., Mazhangara, E., Bailey, L., Yoo, K.,
Yamano, T., and Mghenyi, E., 2004. A Cross-Country Analysis of Household
Responses to Adult Mortality in Rural Sub Saharan Africa: Implications for
HI V/AIDS Mitigation and Rural Development Policies. MSU International
Development Working Paper, IDWP 82. East Lansing, Michigan.

Mather, D., H. Marrule, C. Donovan, M. Weber and A. Alage (2004). Household

Responses to Prime Age Adult Mortality in Rural Mozambique: Implications for
HIV/AIDS Mitigation Eflorts and Rural Economic Development Policies.

208

Maputo: MADER Directorate of Economics, Department of Policy Analysis
(MADER/DE/DAP).

Megill, DJ, 2004. Recommendations on Sample Design for Post Harvest Surveys in
Zambia based on the 2000 Census. Working Paper 10, Food Security Research
Project/Zambia. Lusaka, Zambia.

Ministry of Health and Central Board of Health (1999) HIV/AIDS in Zambia:
Background, projections, impacts and interventions, Lusaka, Zambia.

Ministry of Health and Central Board of Health (1999). HI V/AIDS in Zambia:
Background, projections, impacts and interventions. Lusaka, Zambia

Mushati, P., G. M. Mlilo, J. Lewis, and C. Zvidzai (2003). Adult Mortality and
Economic Sustainability of Households in Towns, Estates, and Villages in AIDS-
aflected Eastern Zimbabwe. Paper presented at the Scientific Meeting on
Empirical Evidence for the Demographic and Socio-Economic Impact of AIDS,
Durban, South Afiica, 26-28 March 2003. University of Durban.

Mutangadura, G, and D. Webb (1999). Socio-Economic Impact of Adult Mortality and
Morbidity on Urban Households in Zambia. AIDS Analysis Africa 9(4).

Ngom, Pierre and Samuel Clark. 2003. Adult Mortality in the Era of HIV/AIDS: Sub
Saharan Africa. Workshop on HIV /AIDS and Adult Mortality in Developing
Countries, Population Division, Dept of Economic and Social Affairs, United
Nations Secretariat, New York, 8 13 September 2003.

Pitayanon S., S. Kongsin, W.S. J anjaroen (1997). Economic Impact of H1 V/AIDS
Mortality on Households in Thailand. In: David Bloom and Peter Godwin, eds.
The economics of HIV and AIDS: The case of South and South East ASIA,
Dehli Oxford University Press.

Reardon, T., E. Crawford, and V. Kelly (1995). Promoting Farm Investment for
Sustainable Intensification of Agriculture, International Development paper 18,
Department of Agricultural Economics, Michigan State University, East Lansing.

Rosenbaum, P. and Rubin, D. B. (1983), The Central Role of the Propensity Score in
Observational Studies for Causal Effects," Biometrika, 70, 41—55.

Rosenzweig, M., 1988. Risk, implicit contracts and the family in rural areas of low-
income countries. Economic Journal 98, 1 148-1 170.

Rugalema, G. (1998). It is not only the loss of labor: HIV/AIDS, loss of household assets
and household livelihood in Bukoba District, Tanzania. Paper Presented at the
East and Southern Africa Regional Conference on Responding to HIV/AIDS:
Development Needs of Afiica Smallholder Agriculture, Harare June 8-10.

209

 

Rugalema, G., with S. Weigang and J. Mbwika. (1999). HI V/AIDS and the Commercial
Agricultural Sector of Kenya. Report. Food and Agriculture Organization,
United Nations Development Programme, Rome.

Sachs, J. et a1. (2001). Macroeconomics and Health: Investing in Health for Economic
Development, Report of the Commission of Macro-economics and Health to the
Direct-General of the World Health Organization (20 December 20010, p. 42.

SAFAIDS (1998). Socio—Economic Impact of Adult Mortality and Morbidity on
Households in Kafue District, Zambia, Harare, Zimbabwe.

Shah, M., N. Osborne, T. Mbilizi, and G. Vilili (2002). The impact of HI V/AIDS on
Agricultural Production system and rural livelihoods in the central region of
Malawi. Care International in Malawi, Lilongwe.

Stokes, C, S (2003). Measuring impacts of H1 V/AIDS on rural livelihoods and food
security. Sustainable Development Department (SD) FAO.

Strauss, J., Gertler, P.J., Rahman, O., and Fox, K. (1993). GendeL and Life-Cycle
Differentials in the Patterns and Determinants of Adult Health. The Journal of
Human Resources, Vol. 28, No. 4, Special Issue: Symposium on Investments in
Women's Human Capital and Development. (Autumn, 1993), pp. 791-837.

Thiam Macournba, Perry R., and Piché V. 2003. Migration and HIV in Northern
Senegal. Population Reference Bureau. Downloadable at:
http://www.prb.org/Template.cfrn?Section=PRB&template=/ContentManagement
/ContentDisplay.cfm&ContentID=9699

Thomas, D., and Strauss, J. Health and wages 1997. Evidence on men and women in
urban Brazil. Journal of Econometrics, Mar 1997, 77(1): 159-185.

Tibaijuka, A., K. (1998). AIDS and Economic Welfare in Peasant Agriculture, Case
Studies from Kagabiro Village, Kagera Region, Tanzania,” World Development,
vol. 25, no. 6 (June 1997): 963-975

Topouzis, D. (2000). Measuring the impacts of H1 V/AIDS on the agricultural sector in
Africa. Geneva, UNAIDS Best Practice collection

Topouzis, D., and J. du Guemy (1999). Sustainable Agricultural/Rural Development and
Vulnerability to the AIDS Epidemic. UNAIDS and Food and Agriculture
Organization (FAO), Rome.

UNAIDS, (1999). A review of household and community responses to the HIV/AIDS
epidemic in the rural areas of Sub-Saharan Africa. GENEVA, SWITZERLAND.

UNAIDS, 2003. AIDS Epidemic Update. GENEVA, SWITZERLAND.

210

 

UNAIDS/WHO. 1998. AIDS Epidemic Update (December). UNAIDS/WHO, Geneva.

Urassa, M., Boenna, J ., T., Isingo, R., Ngalula, J ., Ng’weshemi, J ., Mwaluko, G., abd
Zaba, B (2001). The impact of HIV/AIDS on mortality and household mortality
in rural Tanzania. AIDS, 15, 2017—2023.

USAID, 2004. Prosperity, Hope and Better Health for Zambians, Lusaka, Zambia.

White. J ., and E., Robinson, 2003. HIV/AIDS and Rural Livelihoods in Sub-Saharan
Africa. Policy series 6, National Resources Institute. Downloadable at :
http://www.livelihoods.org/info/docs/aidspolicyseries6.pdf

Wooldridge, J. M., (2002). Econometric Analysis of Cross Section and Panel Data. MIT
Press, Boston.

Wooldridge, J .M (2002). Inverse Probability Weighted M-Estimators for Sample
Selection, Attrition and Stratification. Portuguese Economic Journal, 1: 117-139.

World Bank. (1999). Confronting AIDS: Public priorities in a global epidemic. Revised
edition. Oxford University Press, New York.

World Heath Organization. Fact Sheet 1 HIV/AIDS: the infection. http://librlarvunesco-
iicba.org[ English/HlVfiAIDS/cdrom%20materials/WHO/WHOhtm1%204.htm,
retrieved April 28, 2004.

Xu, 2., J. Govereh, R. Black, and T. Jayne. 2005. Maize Yield Response to Fertilizer
and Profitability of Fertilizer Use Among Small-Scale Maize Producers in
Zambia, Staff Paper, Department of Agricultural Economics, Michigan State
University, East Lansing.

Yamano, T., Jayne, T. S. 2004. Measuring the Impacts of Working-Age Adult Mortality
on Small-Scale Farm Households in Kenya, World Development 32 (1): 91-119.

Yamano, T., and TS. Jayne. 2003. Working-Age Adult Mortality and Primary School
Attendance in Rural Kenya. Discussion Paper Series on International Development
Strategies: No. 2003-006. Foundation for Advanced Studies on International
Development, Tokyo, Japan.

211