iéjw.fiw ME“?

0... .. .. v.”

Eufimrs i.
.

{Mr

 

 

; . .m 5."...
ufiuhflduh
44%".

 

r .12...
. I‘vflnw

3.93
a
tail)

r. 9". v...
.353: n.» 25.2... ..

 

 

Cat: -
.dfl: t... v.‘
. %
av!- 93:...
.5

3“. y 4
a . .1 4.4.1.9...3.
. “Lanalfifirlr hfx .33.
.. .. ... u
. .1. . .1. at.“

 

 

.82.. a.
6:. 2... l... !O?. A 7"
315‘»? nu}! $12.13.
.13.! .3 $9.— 51‘ . . .

{116. .H 5!

. .... .: .L. V‘ r... :1 .1316 I. .livAS. . ,
. . . , . . mi 2 h . 1.3! l t l A . v.1 . r x , .I v. 3 . fi ,
, .L. . . a 4.. . .. .. L»... hwyyfi‘n.
. . . :i- .. xii. .2? . . .
. .. . 1...... .2. . .1! 5.... 32:12.: . . . . : . .
1|! '
I Ei|l

 

 

 

    

 

   

 

  

 

 

   

 

 

 

"NM
’1 a;
.| =
:

 

"F‘s ' LIBRARY
'4 o o 9 Michigan Stite
University

This is to certify that the
dissertation entitled

ESSAYS ON DISEASE-RELATED WORKING-AGE ADULT
MORTALITY: EVIDENCE FROM RURAL KENYA

presented by
Lilian Wambui Kirimi

has been accepted towards fulfillment
of the requirements for the

Doctoral degree in Agriculture, Food, and
Resource Economics

 

 

 

 

 

MSU is an Affinnative Action/Equal Opportunity Employer

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SIOB K:IProj/Acc&Pres/ClRC/DateDue.indd

 

 

ESSAYS ON DISEASE-RELATED WORKING-AGE ADULT
MORTALITY: EVIDENCE FROM RURAL KENYA
By

Lilian Wambui Kirimi

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Agriculture, Food, and Resource Economics

2008

ABSTRACT

ESSAYS ON DISEASE-RELATED WORKING-AGE ADULT
MORTALITY: EVIDENCE FROM RURAL KENYA

By
Lilian Wambui Kirimi

The onset of HIV/AIDS has led to a remarkable increase in working-age adult
mortality. Using a nationally representative panel data of rural Kenyan farm households
surveyed between 1997 and 2007, the first essay of this dissertation examines whether the
relationship between socioeconomic characteristics and adult mortality has changed over
this period. Hazard analysis is applied to a 10-year period data of 5,755 working-age
adults. There is a downward trend in the risk of working-age adult mortality across the
observation period, with the decline being faster for men than women. In the early stages
of the HIV/AIDS epidemic, wealthier, more educated individuals, those who were mobile
and spent extended periods of time away from home were more likely to die of
HIV/AIDS-related diseases. A comparison between the findings within the 10-year
period and the early stages reveals that the relationship has changed in a number of ways.
First, the hazard of death varies only slightly by wealth status and there is a weak
significant shift toward an inverse relationship between asset value and mortality in some
periods. Second, the hazard of death is converging among men of all education levels.
Third, the risk and spread of HIV/AIDS is no longer related to mobility and spending

time away from home. Fourth, relative remoteness and isolation from the initial epicenter

of HIV/AIDS is no longer associated with lower HIV infection and hazard of death.
These changes suggest that over time, mortality risks are converging among groups that
could in the past be clearly distinguished with regard to the likelihood HIV infection and
AIDS-related mortality.

Essay Two determines the dynamics of the impacts of adult death on household
composition, cultivated land, value of crop output, household asset base and income. I
use panel data over a 7-year time period and a statistical model borrowed from the
program evaluation literature that exploits a counterfactual strategy, taking into account
the impacts of death across both time and households. In contrast to the Essay 1, in which
the individual is the unit of observation, the unit of observation in Essay 2 is the
household. Findings indicate that: (i) small negative impacts of adult mortality begin to
emerge in the period preceding death, presumably due to pre-death morbidity; (ii) there
are substantial and significant negative impacts observed in the year of death, and in most
cases, these impacts are larger compared to those in other years before and after death;
(iii) impacts are not just a one-time post-death adjustment but rather they persist over
time; (iv) negative effects are larger in the period closest to the death event, signifying
some recovery over time for some outcomes; (v) contrary to the conventional wisdom
that afflicted households will shift to labor-saving crops such as roots and tubers, there is
weak evidence of a such a shift but a modest shift away from capital-intensive high-value
crops; (vi) sale of small animals and participation in off-farm activities are important
coping strategies for afflicted households; and (vii) impacts of adult mortality differ by

the gender and position of the deceased adult.

Copydghtby
Lilian Wambui Kirimi

2008

 

DEDICATION

To my husband, Kirimi Sindi
And our daughters Kendi and Gakii,
who cheered me on along the way.
And to my mom, Bilha Njoki Muigai,
who prayerfully and eagerly waited for

the completion of this work.

ACKNOWLEDGEMENTS

I am greatly indebted to my major professor and dissertation supervisor, Dr.
Thom Jayne, for his supportive role and guidance in this work, particularly during the
moments when I needed a reminder to keep the faith. I would also like to thank my other
committee members: Dr. Robert Myers, Dr. Roy Black, and Dr. Steven Haider for their
useful comments and suggestions.

I would like to extend my deepest appreciation to my husband Kirimi Sindi who
ably and patiently played a supportive role at home and in school. I am grateful for all the
assistance and insights he provided particularly with regard to data analysis. I rejoice to
have our daughters Kendi and Gakii who cheered me on, asked me questions and
provided answers to some of them, prayed for me to get “unstuck” and their prayers were
heard. They have been an inspiration along the way.

I am grateful for the support of fellow graduate students in the department of
Agriculture, Food, and Resource Economics. Special thanks to Elan Satriawan, the late
Lesiba Bopape, Antony Chapoto and Denys Nizalov for their valuable suggestions at
various stages of this work.

I would also like to extend my appreciation to many other individuals who
contributed to our family welfare in very unique ways during our stay at Michigan State
University. Lucy and Gilbert Nyaga were always but just a call away. They worked
through the maze with me, feeling my joy and heartaches, praying and offering words of
advice and encouragement, a true source of refreshment. Marthe Diallo consistently and

faithfirlly braided our hair while Chifumi Takagi, Hikuepi (Epi), Mary Mwiandi, Lucy

vi

—A

'. .I'I

 

 

and Gilbert Nyaga, Patience Mshenga, Mwanarusi Saidi, Mukwiti Mwiinga, Nicole Fisk
and Eric Bailey took time to provide “uncle/auntie” support for our children. I thank the
many friends and neighbors, particularly the families of Danny Campos, Tom
Engelsman, William Ola, Julia Grant, Patrick Ogutu, Qaiser, and Achila, who provided
child care and opportunities for Bible study and fun for our children. I will miss great
friendships such as that of Tomokazu Nagai that have been born and nurtured in East
Lansing. I gratefully acknowledge prayer support from family and friends and
particularly my mother, Bilha Njoki and her friend mama Karuga, Mary and Barnabas
Rotich, Esther Kamau, the Church that meets at Reannah’s house and the Bible study
group that meets at Campos’s house.

I am thankful for the tuition and stipend support I received from the Food Security
Research Project/Kenya and grants from Miriam Kelly Scholarship fund, Graduate
school (ALANA), and Office for International Students and Scholars.

Through it all, God’s grace has been sufficient and to Him be the glory!

vii

 

 

 

TABLE OF CONTENTS
LIST OF TABLES ......................................................................................................... ix
LIST OF FIGURES ..................................................................................................... xiii
Chapter 1 1
Chapter 2 6
CHARACTERISTICS ASSOCIATED WITH WORKING-AGE ................................. 6
ADULT MORTALITY IN RURAL KENYA ................................................................ 6
2.1 . Introduction ...................................................................................................... 6
2.2. Previous research on adult mortality, HIV/AIDS and socioeconomic factors 10
2.3. Data ................................................................................................................. 18
2.4. Methodology ................................................................................................... 24
2.4.2 Empirical analysis ....................................................................................... 24
2.4.4 Correcting for potential attrition bias .......................................................... 31
2.5 Results ..................................................................................................................... 33
2.5.1 Descriptive analysis ......................................................................................... 33
2.5.2 Factors influencing re-interview ................................................................. 47
2.5.3 Results from hazard analysis ...................................................................... 53
2.5.4 Conclusion and policy implications ............................................................ 66
Chapter 3 70
DYNAMICS OF THE IMPACTS OF WORKING-AGE ADULT MORBIDITY AND
MORTALITY ON RURAL FARM HOUSEHOLDS IN KENYA ............................. 70
3.1 Introduction ............................................................................................................. 70
3.3 Limitations of reviewed literature .................................................................. 83
3.4 Data and attrition ............................................................................................ 84
3.5 Disease progression and the economic impacts of AIDS-related mortality 90
3.6 Empirical analysis of the dynamics of mortality impacts ............................... 94
3.6.2 Model variables ........................................................................................... 99
3.6.3 Econometric concern: household attrition ................................................ 100
3.7 Results .......................................................................................................... 105
3.7.1 Determinants of household re-interview ................................................... 105
3.7.2 Impacts of working-age adult mortality on household composition ........ 108
3.7.3 Impacts of working-age adult mortality on farm production .................... 118
3.7.4 Impacts of working-age adult mortality on assets ................................... 132
3.7.5 Impacts of working-age adult mortality on household income ............... 139
3.8 Conclusion and policy recommendations ........................................................ 144
3.8.1 Summary of main findings ....................................................................... 144
3.8.2 Implications for mitigation and poverty reduction programs ................... 146
References ................................................................................................................... l 5 1
viii

 

LIST OF TABLES

Table 2-1 Households and adults interviewed, and working-age adult mortality rates,

1997-2007 ......................................................................................................................... 22

Table 2-2 Individual and household level characteristics of working-age adults by
attrition status .................................................................................................................... 23

Table 2-3 Characteristics of working-age adults in 2000 who remained alive, and those

who died due to illness-related causes between 1997 and 2000 ....................................... 35

Table 2-4 Characteristics of working-age adults in 2002 who remained alive, and those

who died due to illness-related causes between 2000 and 2002 ....................................... 38

Table 2-5 Characteristics of working-age adults in 2004 who remained alive and those

who died due to illness-related causes between 2002 and 2004 ....................................... 40

Table 2-6 Characteristics of working-age adults in 2007 who remained alive and those

who died due to illness-related causes between 2004 and 2007 ....................................... 42

Table 2-7 Probit models for individual-level reinterview by gender and survey periods 49

Table 2-8 Parameter estimates, marginal effects and hazard probabilities for the initial
baseline hazard model for the period 1997-2007 .............................................................. 54

Table 2-9 Results of time-invariant hazard models of working-age adult mortality in rural
Kenya, 1997-2007 ............................................................................................................. 56

Table 2-10 Results of time-varying hazard models of working-age adult mortality in rural
Kenya, 1997-2007 ............................................................................................................. 62

ix

"' 17—1—1). air-f

 

Table 3-1 Relationship between disease-related working-age adult mortality, household
attrition dissolution, and initial household size for the 1997-2000 period ....................... 85

Table 3-2 Relationship between disease—related working-age adult mortality, household
attrition and initial household size for the 2000-2002 period ........................................... 85

Table 3-3 Relationship between disease-related working-age adult mortality, household
attrition, dissolution and initial household size for the 2002-2004 period ....................... 86

Table 3-4 Relationship between disease-related working-age adult mortality, household
attrition, dissolution and initial household size for the 1997-2004 period ....................... 86

Table 3-5 Number of households incurring working-age adult morbidity and mortality by

survey perioda ................................................................................................................... 89
Table 3-6 Mean levels of 1997 household characteristics by attrition status ................. 103
Table 3-7 Probit models for household re-interview for the period 1997-2004 ............. 107

Table 3-8 The impact of working-age adult mortality on household composition, 1997-
2004 (OLS with district dummies) ................................................................................. 110

Table 3-9 Impact of working-age adult mortality on household composition by years
before and after death, 1997-2004 (OLS with district dummies) ................................... 111

Table 3-10 Impact of working-age adult mortality on household composition by gender
and position of the deceased and by years before and after death, 1997-2004 (OLS with
district dummies) ............................................................................................................ 116

Table 3-11 The impact of working-age adult mortality on total cultivated land and area by
crop types in for the period 1997-2004 (OLS with AEZ dummies) ............................... 119

Table 3-12 The impact of working-age adult mortality on total cultivated land and area by
crop types for years before and after death in the period 1997- 2004 (OLS with AEZ

dummies) ......................................................................................................................... 122

Table 3-13 The impact of working-age adult mortality on total cultivated land and area by
crop types, by gender and position of the deceased for years before and after death in the

period 1997-2004 (OLS with AEZ dummies) ................................................................ 124

Table 3-14 The impact of working-age adult mortality on gross value of crop output in
the period 1997-2004 (OLS with AEZ dummies) .......................................................... 127

Table 3-15 The impact of working-age adult mortality on value of crop output for the
years before and after death in the period 1997-2004 (OLS with AEZ dummies) ......... 128

Table 3-16 The impact of working-age adult mortality on gross value of crop output by
gender and position of the deceased, for years before and after death in the period 1997-
2004 (OLS with AEZ dummies) ..................................................................................... 130

Table 3-17 The impact of working-age adult mortality on value of assets for the period
1997-2004 (OLS with AEZ dummies) ........................................................................... 133

Table 3-18 The impact of working-age adult mortality on value of assets for years before
and after death in the period 1997-2004 ......................................................................... 134

Table 3-19 The impact of working-age adult mortality on assets by gender and position of
the deceased, for years before and after death in the period 1997-2004 (OLS with AEZ

dummies) ......................................................................................................................... 13 7

Table 3-20 The impact of working-age adult mortality on income for the period 1997-
2004 (OLS with AEZ dummies) ..................................................................................... 140

xi

Table 3-21 The impact of working-age adult mortality on income for years before and

after death in the period 1997-2004 (OLS with AEZ dummies) .................................... 141

Table 3-22 The impact of working-age adult mortality on income by gender and position
of the deceased for years before and after death in the period 1997-2004 (OLS with AEZ

dummies) ......................................................................................................................... 143

xii

 

LIST OF FIGURES

Figure 2-1 Percentage of disease-related deaths within age groups by gender, for the
period 1997-2007 .............................................................................................................. 44

Figure 2-2 Provincial adult mortality rates from household survey data (1997-2007) ..... 46

Figure 3-1 Modeling the impact of working-age adult mortality on household outcomes

over time ........................................................................................................................... 95

Figure 3-2 Modeling the impact of working-age adult mortality with binary variables for
years before and after death .............................................................................................. 95

xiii

 

in the region became newly infected, while 2.0 million adults and children died of
AIDS(UNAIDS, 2006).

According to the 2006 Report on the Global AIDS Epidemic (UNAIDS, 2006),
1.3 million people in Kenya were living with HIV/AIDS by the end of 2005, 740,000 of
whom were women with ages 15-49. The HIV prevalence rate was 6.1% among adults in
the 15-49 age range. The majority of new infections occur among the youth, especially
for young women aged 15-24 and young men under the age of 30. The number of deaths
due to AIDS was estimated to be 140,000 while there were about 1.1 million children
who lost their mother, father, or both parents to AIDS by the end of 2005.

Until recently, HIV/AIDS was mainly regarded as a health issue. However, the
epidemic is now being considered as one with profound economic, political, and social
impacts. It is also recognized that certain aspects of the disease including stigma, rapid
spread of infection, and heightened working-age adult morbidity and mortality, pose
major problems for governance, development and economic growth (Barnett and
Whiteside, 2002; FAO, 1994). The increased mortality of working-age adults in sub-
Saharan countries due to the AIDS epidemic is associated with potentially severe
consequences for surviving household members and the community at large (Ainsworth
and Dayton, 2000; Ngom and Clark, 2003).

While there are other more prevalent diseases in Africa, the characteristics of
HIV/AIDS suggest that its economic and demographic impact will be profound. In the
absence of the AIDS epidemic, working-age deaths are relatively rare in Africa, about six
deaths per 1,000 persons per year for ages 15-50. In an area with high levels of HIV

infection, working-age mortality rates can double. triple, or quadruple (World Bank, 1993

Chapter 1

INTRODUCTION

Working-age adult deaths historically have been relatively rare in Africa.
However, in eastern and southern sub-Saharan Africa, the working-age adult mortality
rates have more than doubled since 1980 (UNAIDS, 2003). There is overwhelming
evidence that the dramatic rise in disease-related adult mortality is predominantly a result
of HIV/AIDS, especially within the 15-45 year old range. According to the 2006 UNDP’s
Human Development Report (UNDP, 2006), the life expectancy in sub-Saharan Africa is
46 years, which is 32 years less than the average life expectancy in countries of advanced
human development, and 20 years less than it would have been without HIV/AIDS. The
onset of HIV/AIDS has changed the pattern of mortality dramatically. As Topouzis
(1998) points out, it is not just a matter of mortality decline being reversed. In the past,
declines in mortality occurred at the infant and child levels and resulted in a marked
increase in life expectancy. However, in the era of HIV/AIDS, life expectancy is
declining dramatically due to young adult mortality.

HIV/AIDS emerged in the early 19803 and has continued to spread all over the
world with alarmingly high prevalence and incidence rates. Afi'ica and particularly Sub-
Saharan Africa continues to bear the brunt of the scourge. In 2006, an estimated 38.6
million people worldwide were living with HIV at the end of 2005, 4.1 million became
newly infected with HIV and 2.8 million lost their lives to AIDS (UNAIDS, 2006). Sub-
Saharan Africa is home to just over 10% of the world’s population but has more than

60% (about 24.5 million) of all people living with HIV. An estimated 2.7 million people

 

W'TTT‘:

cited in Beegle, 1997). Because HIV in Africa is transmitted primarily through
heterosexual contact, the epidemic is having a dramatic impact on the mortality of men
and women in their working, earning and childbearing years. As a result, HIV/AIDS has
reduced life expectancy by nearly 15 years in some countries. It is because it is poised to
destroy 2-5% of the productive population each year over the next 20 years that
HIV/AIDS is top priority for development agencies (Engel, 2002).

This analysis looks at the characteristics associated with working-age adult
mortality and the economic impacts of adult mortality more generally, realizing that
much of the mortality observed in our data is due to HIV/AIDS. While all working-age
adult deaths cannot be attributed to AIDS, we know that AIDS is a leading cause of
working-age mortality. Efforts by development planners to deal with the heightened
prime-age adult mortality require accurate knowledge of the characteristics of those most
likely to die in their prime age, and the effects of mortality on household behavior and
welfare.

Increases in adult mortality have led to a sustained focus on economic impacts and
mitigation strategies among policy makers and development practitioners. As a result,
there have been a number of studies directed at understanding the characteristics of
individuals dying of HIV/AIDS and the economic impacts of HIV/AIDS on households’
welfare. Although these studies have generated some useful and interesting results, much
of the evidence is based on case studies, cross-sectional surveys, sector-specific research,
or surveys of specific risk groups, including medical and clinical studies (Nyaga et al,
2004). These types of studies cannot adequately measure the effects of mortality and are

not representative of how it affects the population in a wider sense.

Another drawback of these impact studies is that they tend to focus on very short-
term impacts, usually up to two years after death. However, we would expect short-term
effects to be different from long term-effects. This essay uses data over a 7-year time
period to understand the longer-term effects of death, effects which may be large and
different from those experienced during the period immediately following a death. In
addition, I examine the temporal pattern of the effects of death in the years after the
occurrence of death, which accounts for the likelihood that adult death is not associated
with just a one-time or permanent adjustment in household outcomes. As a result, this
study uncovers the dynamics in household welfare and responses as a result of prime-age
death.

In this regard, this study addresses the following questions. First, what are the
characteristics associated with working-age adult mortality in Kenya and is there
evidence of shifts over time of socioeconomic characteristics of persons who are afflicted
by death? In particular, are individuals who are relatively less wealthy, with lower
education, and without a formal job or business activity more likely to die in their
working-age period than their counterparts? Second, how does disease-related adult death
affect rural farm households, how severe and persistent are these effects, and what is the
ability of afflicted households to respond to impacts of mortality? Effects of mortality on
household demographic structure, agricultural production, assets and off-farm income
will be examined. Insights into death effects on rural agriculture are particularly
important since a shock in this sector can affect the Kenyan economy as a whole. This is
because agriculture is vital for domestic production, exports, employment and foreign

currency earnings.

Answers to questions addressed in this study will contribute to the growing
empirical literature and inform responses on mitigation measures. Understanding the
determinants of adult mortality is an essential step towards curtailing the spread of
HIV/AIDS. Knowledge of impacts of adult mortality and household coping strategies is
vital in generating the best responses toward the epidemic. Such information can improve
targeting and effectiveness of prevention activities, and inform the design of programs to
help afflicted and affected households and communities deal with the impacts of the
disease. The study will use nationwide panel data on rural households surveyed between
1997 and 2004. The national coverage of the data set will overcome the non-
representativeness problem that plagues past case-studies, while its panel structure will
make possible the use of methods that control for problems such as unobserved
heterogeneity (which are not possible with cross-section data).

This dissertation is organized as follows. The second chapter addresses the
question of socioeconomic characteristics of adults afflicted by disease-related death, and
whether the relationship between these socioeconomic characteristics and adult mortality
has changed over time as the disease has progressed. Chapter 3 examines the dynamics of
the impacts of adult death on household composition, cultivated land (total and by crop
types), value of crop output, household asset base (cattle, small animals and farm
equipment) and income. Each essay contains a section on conclusions and policy

recommendations.

 

Chapter 2

CHARACTERISTICS ASSOCIATED WITH WORKING-AGE

ADULT MORTALITY IN RURAL KENYA

2.1. Introduction

Prime-age adult mortality rates in eastern and southern sub-Saharan Africa have
more than doubled since 1980 (UNAIDS, 2003). There is overwhelming evidence that
the dramatic rise in disease-related adult mortality is predominantly a result of HIV/AIDS
(Ainsworth and Dayton, 2000; Ngom and Clark, 2003). Using data from the earlier part
of the period covered in this analysis, Yamano and Jayne (2004) applied an ecological
approach to estimate the portion of deaths attributed to AIDS. They found that AIDS
accounted for a large proportion of the recorded deaths for particular age/sex categories,
particularly in the Nyanza area. This essay looks at the characteristics associated with
working-age adult mortality, realizing that much of the mortality observed in our data
especially within the 15-45 year range, is due to HIV/AIDS.

HIV/AIDS has been typically viewed as an urban epidemic. However, there is
concern that over time, prevalence rates have continued to rise in rural areas through
migration, trade and strengthened rural-urban linkages. In the early stages of the
HIV/AIDS epidemic, studies in Eastern and Central Africa showed that a positive
correlation existed between HIV infection and socioeconomic status indicators namely
education, income, wealth and occupation. Infection and mortality rates were higher

among the urban, wealthy and educated (Ainsworth and Semali, 1998; Nyaga et a1,

 

2004). Ainsworth and Semali (1998) present a summary of eleven studies, which show
that HIV infection is positively correlated with indicators of socioeconomic status such as
schooling, income, wealth and occupation even in rural areas. The accepted explanation
has been that men with higher education and income find it easier to attract casual sexual
partners, and that men and women with higher education and income are likely to travel
more. These studies covered the early periods of the epidemic. However, as the disease
has progressed, there are reasons for expecting that this positive relationship between
socioeconomic status and HIV contraction, and consequently adult mortality, may have
changed. This may probably be due to greater availability and access to information
about HIV/AIDS.

In addition, later in the development of the epidemic, 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 (e. g. Gillespie and
Kadiyala, 2005; Poku, 2001; Desmond, 2001). For example, health factors correlated
with poverty reduce individuals’ immune response and raise the probability of
contraction with HIV-positive partners (Stillwaggon, 2005). Some policy makers have
increasingly recognized poverty as an important social pathway contributing to the spread
of the disease and working-age adult mortality, particularly among women. It has been
argued that women unable to sustain themselves through wage labor or agriculture are
more likely to resort to risky sexual behavior for survival (Bryceson and Fonseca, 2006).
According to this perspective, poverty predisposes certain groups to risk of infection. As
a result, a common programmatic response to AIDS has been the provision of income-

earning activities to targeted groups, in the hope that this may reduce risky behaviors that

spread AIDS. In spite of growing perceptions that the spread of HIV (and subsequent
mortality) in eastern and southern Africa is increasingly associated with poverty, we
know of no statistical evidence using nationally representative survey data that would
support this conclusion.

This study seeks to determine the socioeconomic characteristics of individuals who
are subsequently afflicted by working-age adult mortality. Also, it examines whether over
time, the relationship between socioeconomic status and adult mortality has changed as
HIV/AIDS (a leading cause of prime-age mortality) has progressed. The study uses a
nationwide longitudinal data of 5,755 working-age individuals in 1,500 small-scale farm
households in rural Kenya surveyed across a ten—year time period, to determine the
hazard of death and how this relates to socioeconomic characteristics of individuals who
are afflicted by death. Because sampled households were surveyed five times over a ten-
year period, from 1997 to 2007, this study provides a unique opportunity to assess
potential shifts in the socioeconomic correlates of disease-related adult mortality over
time.

The essay first examines the spatial relationship between disease-related prime-
age mortality in our survey data and lagged provincial HIV prevalence rates as reported
by the Kenyan government; the strong correlation found indicates that the disease-related
mortality observed in our survey data largely reflects the growing influence of AIDS in
mortality rates. Next, we construct various indicators of household and individual poverty
status and use these in hazard models of disease-related mortality to assess potential
changes in the relationship over time between socioeconomic characteristics and

subsequent death of working-age adults. The study covers a ten-year period and thus

includes a period in which AIDS-related mortality has acutely heightened. Therefore, it is
now possible to capture more of the dynamics of adjustment in socioeconomic
characteristics over a longer period of time. We use inverse probability weighting

procedures to control for potential attrition bias.

The study contributes to literature and practice in various ways. It adds to our
understanding of the relationship between adult mortality and socioeconomic
characteristics. This is in turn important for understanding behavior and designing
mitigation policy. Working-age adults comprise most members of the society who are
economically productive and responsible for the support of children and elderly
dependents. Hence, there is need to determine socioeconomic characteristics of
individuals afflicted by adult morbidity and mortality in order to avert potential adverse
effects on individual and household welfare. Knowledge of the mechanisms that are
behind mortality trends and variations is essential for the prediction of the course of the
epidemic. The study informs the ongoing debate on the relationship between poverty,
risky sexual behavior, and subsequent contraction HIV/AIDS and adult mortality in

Africa.

While not all working-age adult deaths can be attributed to AIDS, high HIV
infection rates in Sub-Saharan Africa are producing dramatic increases in the mortality of
working-age adults (N gom and Clark, 2003). Campaigns to prevent the spread of
HIV/AIDS require accurate knowledge of the characteristics of those most likely to
contract the disease. Therefore, understanding the determinants of individuals’ risk of
infection and death is an essential step towards curtailing the spread of this disease. Such

information can improve targeting and effectiveness of prevention activities, and inform

the design of programs to help affected households and communities deal with the
impacts of the disease. Also, given that HIV/AIDS epidemics are heterogeneous across
and within countries (in terms of intensity, pace and impact), the study will in particular

provide guidance on AIDS prevention and mitigation strategies in Kenya.

The remainder of this chapter is organized as follows. Section 2.2 presents
findings from previous studies and the contribution of my study. Section 2.3 describes the
data and addresses the issue of attrition bias which is potentially an important problem in
analysis of longitudinal survey data. Section 2.4 presents the methodology. Results are

discussed in section 2.5 and section 2.6 presents conclusions of the chapter and policy

implications.

2.2. Previous research on adult mortality, HIV/AIDS and socioeconomic
factors

A well established pattern in the literature is that people with high education, high
income, and a higher status occupation experience a lower risk of death. One possible
explanation is the “access to health care” argument. If better-educated or richer people
have better access to health care, and if health care has a major effect on morbidity and
mortality, then it follows that education or wealth will be positively correlated with health
status (Deaton, 2002). Further, income provides nutrition, housing, clean water, and
sanitation and thus protects individuals from hunger and infectious diseases. Education
protects health by enhancing a person’s efficiency as a producer of health, conferring
advantage in access to health information or through the establishment of a more patient

and forward looking behavior (Grossman, 1972).

10

An individual’s probability of dying is primarily a function of gender, age, health,
genetic endowment, and the environment, all of which determine the risk of falling
victim to illness or accident. These factors work together and are influenced by a
complex and ever changing set of social and historical determinants (INDEPTH, 2002).
HIV/AIDS has changed the landscape of adult mortality and its socioeconomic correlates.
The impact of HIV/AIDS on an individual’s probability of dying is complex, and
depends on many factors such as gender, mode of infection, number of infections, age at
infection, immune competence, overall health, and access to treatment (N gom and Clark,
2003). The HIV/AIDS epidemic in Sub-Saharan Afiica is primarily one of heterosexual
transmission. Consequently, the epidemic is having a dramatic impact on the mortality of
men and women in their prime childbearing and earning years. Mortality rates of adults
aged 15—50 have increased dramatically in areas affected by the epidemic. In addition,
the number and gender-age composition of the infected sub-population is in turn strongly
influenced by population-specific behavioral factors such as sexual networking
preferences (N gom and Clark, 2003).

Analysis of the socioeconomic characteristics associated with working-age adult
mortality using micro data has been previously uncommon, since mortality has been a
rare event for the middle-aged and for children past the age of five years (Duleep, 1986).
However, the onset of HIV/AIDS in Sub-Saharan Africa in the 19803 has led to a
substantial increase in working-age adult mortality and as a result, has made this a subject
of interest among researchers.

In response to the AIDS epidemic, the initial focus of analysts and policymakers

was to understand the characteristics of those who were infected and died of AIDS.

11

Analysts and health practitioners argued that information on background characteristics
of those most likely to be infected and those who have died could: i) improve targeting
and effectiveness of prevention activities, and ii) affect the economic impact of the AIDS
epidemic on households and the economy, including the distribution of skills in the labor
force and the demand and ability to pay for curative medical care.

Past studies in sub-Saharan Africa that followed this line of inquiry generally
found a positive relationship between HIV -related deaths and socioeconomic status — as
measured by education, income, and occupation (Ainsworth and Semali, 1998; Kapiga et
al, 1994). The explanation for this finding was that the ability to support multiple partners
may rise with income and that men and women with more education and higher incomes
are probably more likely to travel and thus have more opportunities for casual sexual
contacts. However, Ntozi et al (1997) found that in Uganda, those with primary education
are more likely to die of AIDS than those with post-primary education. They argue that
that the more educated have greater access to AIDS prevention facilities such as condoms
and information than those with primary education.

In terms of occupation, people of other occupations were more likely to die of
AIDS than farmers (Ntozi et al, 1997); executives had higher HIV infection rates than
foremen and casual laborers (Over, 1992); the epidemic was disproportionately striking
groups with higher levels of productive skills and human capital (Ndongko, 1996).

Other studies report that women are more vulnerable to AIDS than men in Africa
(Anarfi, 1995; Berkley et al, 1990, Ntozi et a1, 1997 and Ryder et al, 1990). Women's
vulnerability to HIV/AIDS may be due to a complex interplay of biological, social, and

behavioral conditions. Women often have vulnerable employment status, lower incomes,

12

and least entitlements to or ownership of assets and savings, and may resort to
transactional sex for survival. As a result, they may be more economically dependent on
their partners and less empowered to protect themselves against HIV infection (Greig and
Koopman, 2003). It is also argued that there is more efficient transmission of HIV virus
from men to women than the reverse, and sexual behavior favors men having more
sexual partners. In sub-Saharan Africa, the highest rate of HIV infection is among
sexually active young women, less than 35 years old. The 15-24-year-old women are
more likely to be infected than men in the same age group, and men are infected at a later
age (Topouzis, 1994). Thus, although the typical female advantage of lower mortality has
been smaller in developing compared to developed countries, this may have been eroded
more recently in Africa, due to the differential gender impact of HIV/AIDS.

Mixed results emerge with respect to marital status and type of marriage union
but generally people in polygamous marriages are more likely to die of AIDS than those
in monogamous marriages (Ankrah et al, 1992; Allen et al, 1991; Kapiga et a1, 1994;
Nunn, 1989). The institution of marriage is believed to play a beneficial role in mitigating
one’s risk of exposure to HIV. Since the transmission of HIV in the population is mainly
through sexual activity, avoiding infection depends on risk-avoiding behavior.
Consistently, empirical results show that excess mortality is concentrated in not-yet
married adults aged 20-39 among both men and women (Yamauchi, 2007). In most cases,
widows may have no legal rights to land and property (due to customary inheritance laws
or the difficulties of enforcing existing remedial legislation) after their husbands' death.

Their incomes may be too low and thus impoverishment may force them as well as single

13

women to engage in occasional sex for money or earn a living as commercial sex workers
(Topouzis, 1998).

HIV/AIDS was initially primarily prevalent in urban areas. However, it has
spread to rural areas due to factors like migration, trade, refugee movements,
urbanization, strengthened rural-urban linkages, improved transport and relatively high
incomes that enhance mobility, as well as economic and socio-cultural factors that
promote an increasing number of people who maintain strong footholds in rural areas and
commute between urban and rural areas on a regular basis. Restricted access to health
facilities and fewer institutions delivering HIV/AIDS information has also been cited as a
factor (Topouzis, 1998). Consequently, all these factors have been suggested as important
in spreading HIV/AIDS and contributing to working-age adult mortality in rural areas.

Another focus of analysts and policymakers has led to a set of studies on patterns
and determinants of sexual behavior rather than the characteristics associated with HIV
infection and working-age adult mortality. These patterns are seen as one of the primary
determinants of the spread of HIV, and researchers have sought to determine the
socioeconomic correlates of sexual behavior. This was based on the recognition that
socioeconomic status may affect the probability of HIV infection through differences in
sexual behavior, condom use and visits to voluntary counseling and testing centers
(Deheneffe et al, 1998; de Walque, 2002). These studies provide strong empirical
evidence of great heterogeneity in sexual behavior within and across countries, with some
individuals having a great many more sexual partners than others. The probability of
having had a non-regular partner is often unrelated to asset ownership, but when the

relationship is significant it tends to be negative for women and positive for men. The

14

probability of using a condom with a non-regular partner tends to be positively associated.
with education for both men and women. These studies consistently find that condom use
and visits to voluntary counseling and testing centers have been more widespread among
educated individuals. This finding implies that the positive correlation between HIV
infection and education is likely to reverse itself in the longer run.

Indeed some reversal has been shown to occur (de Walque, 2002). In Uganda, de
Walque investigates the hypothesis that the ability to process information is one channel
through which education affects health outcomes. He does this by examining the
effectiveness of an information campaign aimed at preventing the HIV/AIDS epidemic.
He finds that during the early years of the epidemic in 1990, there was no robust relation
between HIV/AIDS and education. However, in 2000, education was found to lower the
risk of being HIV positive among young individuals. He suggests that as soon as
information on existence and modes of transmission of HIV is available, more educated
individuals modify their behavior more strongly. Also, he argues that if there indeed was
an initial positive correlation between income and risk of contracting HIV, then the
subsequent decline in prevalence is more likely to be due to education than to income‘.
This is because more educated people have better access to information about existence,
transmission and prevention modes of HIV/AIDS epidemic. Education enables
individuals to avoid infection by using condoms or a combination of HIV testing and
reciprocal fidelity, and enables them to mitigate consequences of the disease through

better nutrition and health care. In addition, as educated people have invested in their

 

' Unlike most cases of health conditions, malnutrition for example, it is not very expensive- at least in
monetary terms- to avoid being infected by HIV/AIDS.

15

future by going through a longer education process, they may feel that they have more to
lose if their life expectancy is drastically shortened by AIDS.

Additionally, more recent studies have found that wealth and income continue to
be strongly and positively related to HIV prevalence and risky behaviors (Central Bureau
of Statistics (CBS), Kenya, 2004; Shelton _e_t_ a_l, 2005; Annim & Awusabo-Asare, 2006).
Risky behaviors such as unsafe sex, frequent changes of sexual partners or multiple
concurrent partnerships are key factors in driving the AIDS epidemic. Such behaviors
have been found to be positively related to wealth and income for a number of reasons.
First, wealth and income promote such partnerships because they are associated with
mobility, time, and resources necessary to maintain concurrent partnerships. Second,
wealth and income are very closely linked with social interaction, and therefore they
might increase the number of opportunities for such partnerships to develop. Third, this
relationship could arise partly because household wealth and income may relate to urban
residence, and HIV rates are higher in urban areas. Hence, urbanization may mean higher
income and wealth for the household but not necessarily improved health status. Fourth,
perhaps wealth and income accord individuals status in the society and simply enable
them, and especially men to have more sexual partners (Shelton e_t g, 2005). Individuals
with more education and therefore more human capital will have better employment and
higher earnings. Hence, education and occupation may also be strongly positively related
to HIV prevalence just like income and wealth for similar reasons. In general, individuals
that are relatively wealthier, richer, or more educated (especially males) put themselves
or their health at risk through use of their resources in sexual networking. However, as

the disease has progressed, this positive relationship between socioeconomic status and

16

HIV contraction, and consequently adult mortality, may have changed, probably due to
greater availability and access to information on HIV/AIDS.

This study aims to extend existing research on the socioeconomic characteristics
of working-age adult mortality. The study makes four contributions to the literature on
this topic. First, most studies to date investigating the socio-economic attributes of
AIDS-related adult mortality in Africa are based on relatively small samples like in case
studies of areas with high HIV prevalence rates (e. g., Kagera, Tanzania and Rakai,
Uganda) and of high-risk groups such as commercial sex workers and truck drivers.
Thus, the generalizability of these results to the general population may be limited, and
only partial evidence can be obtained from such studies. My study is based on a large-
scale nationwide set of household surveys over a lO-year time period.

Second, the longitudinal nature of the data provides the opportunity to assess
whether the socio-economic correlates of disease-related adult mortality have changed
over time. Most studies report results from cross-sectional analysis and those that use
panel data are limited to the early period of the epidemic. This study will use a 10-year
panel to address the question of whether the positive relationship between working-age
adult mortality/HIV infection and indicators of socioeconomic status reverses in the
longer run, as frequently contended.

Third, past studies that have focused on sexual behavior do not provide adequate
information with regard to the individual characteristics associated with AIDS-related
adult mortality because sexual behavior is just one of the variables through which
socioeconomic indicators influence mortality. The analysis undertaken in this study aims

in part at addressing some limitations of past studies.

17

Lastly, some analysis at the macro level (e.g., Deheneffe et al, 1998, across 12
countries/sites) is too large-scale and presumes homogeneity of institutional or
community features thought to be crucial in imparting variance in sexual behavior. On
the other hand, the purely individual-level survey reveals little about the institutional and
cultural environment surrounding the household. This study improves on past research by
combining measures of socioeconomic characteristics at the individual, household and

community levels.

2.3. Data

The study uses nationwide data set of Kenyan rural households collected in 1997,
2000, 2002, 2004 and 2007. The sampling frame for the surveys was prepared in
consultation with the Central Bureau of Statistics, although its agricultural sample frame
was unavailable. First, 24 districts were purposively chosen to represent the broad range
of agro-ecological zones (AEZs) and agricultural production systems in Kenya. Next, all
non-urban divisions were assigned to one or more AEZs based on secondary data. Third,
proportionally to population across AEZs, divisions were chosen purposively from each
AEZ. Fourth, within each division, villages and households within selected villages were
randomly selected. A total of 1,578 households were chosen from 24 districts. Forty
households from two pastoral districts were excluded since they differed considerably
from households in other zones and had high rates of attrition. Also, thirty-eight
households whose landholding was larger than 20 acres were excluded from the survey,
in order to focus on small-scale households. As a result, this study is based on 1500
households from the 1997 survey. Out of these households, 1,406 households (93.7%)

were re-interviewed in 2000, 1,245 households (83% of the original 1,500 households in

18

the 1997 survey) in 2002, 1,356 households (90.4%) in 2004 and 1308 households
(87.2%) in 2007 (Table 2-1). The household attrition rate between 1997 and 2007 was
12.8 percent. Note that over 100 households that dropped out of the sample in 2002 were
re-interviewed in 2004, indicating that attrition rates are influenced by whether
enumerators were able to find eligible household members. A relatively high rate of
household attrition is found in Nyanza and Rift Valley provinces. Differences in attrition
rates across regions could be due to greater household dissolution and relocation in some
areas (which could be influenced by AIDS) or by differences in the quality of enumerator
and supervisor teams (an issue examined below) or both.

According to household respondents, the individuals who left their households
after the 1997 survey departed for various reasons including: search for employment,
attending school, marriage, death, divorce/separation, moved away to live on their own or
with other relatives, and moving back to the households they lived in prior to 1997.
Working-age individuals lost to follow-up include those that were in households that
were not relocated and reinterviewed after the 1997 survey. Household attrition is
attributed to the following reasons: household moved away or dissolved, no one was at
home or available to be interviewed, or a household refused to participate in the survey.

There is a great deal of fluidity in household composition, with many working-
age individuals moving into and out of their rural households over time. Of the 5,755
working-age individuals (15-59 years) who were in the original 1997 sample, 13.1% had
left the households between 1997 and 2000, 23.4 percent by 2002 survey, 37.0 percent

by 2004 survey, and 53.3 percent by 2007 survey, for reasons other than death.

19

This study focuses on adults 15-59 years old who were resident in the sampled
households at the time of the initial survey in 1997. We continued to track the
whereabouts of individuals who left their households after 1997 and were able, therefore,
to calculate mortality rates based on a consistently defined sample over the 10-year
period. A panel data survey spanning a lO-year period provides 22,654 person-years of
follow-up generated from 5,755 adults. In this study, the main outcome measure is time
from study entry to death. For the calculation of person-time, the starting point was the
beginning of the study period, while the end-points were the time of death or end of
observation, defined as the end of the study period, or the time of the prior survey before
the whereabouts of the individual became unavailable. The outcome in each survey
interval was yes if the person died of disease related causes and no, otherwise.

The data provide multiple years of observation for each individual and a survival
outcome at the end of the survey interval. In a person-period dataset, each person has one
row of data for each discrete time-period, each containing: (i) a variable, which labels the
time-period to which each row of the person-period dataset refers; (ii) another variable,
which records whether an individual died in the particular discrete time-period in
question; (iii) a set of other variables that influence the likelihood of death.

Hazard analysis is performed on this dataset, where some of the individuals died
of disease-related causes during the ten year period. However, the majority of the adults
are considered as “censored” observations, since they were alive at the end of the
observation period and we have no further information about them.

Individuals joining their households after 1997 are excluded from the analysis. It

has been noted in the literature that many people afflicted by HIV/AIDS returned to their

20

rural homes to receive terminal care and to die (Kitange e_t a_l, 1996). Therefore, including
individuals who joined their households after the 1997 survey in the analysis may lead to
an overestimation of working-age adult mortality. Table 2-1 presents mortality rates
within the 15-59 year age group computed two ways, first based on individuals who were
residents at the beginning of each survey, and second, based on the inclusion of newly
arriving members after each survey. Indeed, we find that mortality rates are higher (3.8
vs. 3.5 persons per 1000 person years) when including members joining their homes after
the prior survey.

Table 2-2 presents initial individual- and household-level characteristics (based
on the 1997 survey) of individuals by attrition status. Individuals who “attrited” were
slightly younger and had more years of education. However, there is no considerable
difference between attrited and non-attrited individuals in terms of months spent at home
in 1997, and the age and education of heads of households in which they resided.
Generally, the percentage of men and women attriting each year is very similar, except in
2007 when more men than women attrited. A larger percentage of individuals who
attrited comprises of sons and daughters compared to heads/spouses or other members of
the household. A negligible proportion of female household heads attrited while attrition
rates among individuals from polygamous households are lower than those of
monogamous ones. Households with larger landholdings, more income, and more wealth
showed greater rates of attrition. Therefore, there appears to be differences between
individuals who remained in the survey and those who attrited. These differences will be

investigated further through probit re-interview models.

21

Table 2—1 Households and adults interviewed, and working-age adult mortality rates, 1997-2007

 

 

Original Reinterviewed households and working-age adults who remained in
1997 the households, among those in 1997 survey (number and percent)1
survey
2000 2002 2004 2007
Households 1500 1406 (93.7) 1245 (83.0) 1356 (90.4) 1308 (87.2)
Adults (15-59 5755 5002 (86.9) 4411 (76.6) 3625 (63.0) 2690 (46.7)
years)
Reinterviewed households among those in 1997 survey
(Number and percent)
Province 1997 2000 2002 2004 2007
Coast 91 90 (98.9) 73 (80.2) 88 (96.7) 84 (92.3)
Eastern 242 232 (95.9) 218 (90.0) 226 (93.4) 219 (90.5)
Nyanza 280 258 (92.1) 220 (78.6) 246 (87.8) 240 (85.7)
Western 303 284 (93.7) 258 (85.1) 273 (90.1) 259 (85.5)
Central 181 173 (95.6) 169 (93.3) 169 (93.4) 167 (92.3)
Rift Valley 403 369 (91.6) 307 (76.1) 354 (87.8) 339 (84.1)
Mortality rate for adults (15-59 years)2
Province [997-2000 2000-2002 2002- 2004-2007 Average
2004
Coast 5.2 3.9 2.6 3.0 3.7
Eastern 3.3 1.8 2.4 2.5 2.5
Nyanza
Kisii 3.4 1.2 2.3 1.5 2.1
Kisumu/Siaya 20.0 5.5 9.7 6.4 10.4
Western 4.5 2.7 3.0 3.1 3.3
Central 3.0 1.8 3.1 2.1 2.5
Rift Valley 2.9 1.3 2.3 3.1 2.4
Total 5.2 2.3 3.4 3.2 3.5
Mortality rate for adults (15—59 years)3
Province 1997-2000 2000-2002 2002- 2004-2007 Average
2004
Coast 4.3 3.4 2.7 3.0 3.4
Eastern 2.9 2.0 2.8 2.6 2.6
Nyanza
Kisii 3.5 1.2 2.3 1.5 2.1
Kisumu/Siaya 18.1 5.5 12.9 8.0 11.1
Western 3.9 2.4 5.1 2.7 3.5
Central 3.0 2.6 5.1 2.7 3.3
Rift Valley 2.7 1.3 3.1 3.3 2.6
Total 4.7 2.4 4.7 3.3 3.8

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997, 2000, 2002, 2004, and 2007.
Notes:

I Excludes pastoralist districts

2 Working-age adult mortality is per 1000 working-age person years; this is for adults (15-59 years) who
were listed in the demographic roster at the beginning of each period.

3 This includes adults (15-59 years) who joined the household after the previous survey; some of these died
within the period.

22

Table 2-2 Individual and household level characteristics of working-age adults by attrition status

 

Attrited in Attrited in Attrited in Attrited in Did not

 

2000, 2002, 2004, 2007, attrit,
contained in contained contained contained in remained
1997 in 1997 in 1997, 1997, 2000, in
sample and 2000 2000 and 2002 and household
samples 2002 2004 for all 5
samples samples survey
years
Age (mean of all adults) 29.5 31.3 24.1 24.5 34.5
Age of household head (mean) 48.6 43.0 44.5 43.1 44.6
Months at home (mean of all 10.0 1 1.0 10.0 10.0 11.0
adults)
Education (mean years of all 7.4 7.5 7.8 7.6 6.9
adults)
Education of household head 7.2 7.8 7.2 8.0 7.7
(mean)
Female (%) 48.9 50.1 51.3 43.3 52.0
Male (%) 51.1 49.9 48.7 56.7 48.0
Relationship to household head
Head/spouse 25.2 45.5 7.3 13.8 65.1
Son/daughter 74.8 39.4 69.4 70.0 32.1
Others 0.0 15.1 23 .4 16.3 2.8
Has formal job (%) 13.1 13.0 10.6 8.9 13.7
No formal job (%) 86.9 87.0 89.4 91.1 86.3
Has informal activity (%) 12.5 22.2 10.9 10.2 22.2
No informal activity (%) 87.5 77.8 89.1 89.8 77.7
Female household head (%) 0.1 3.2 0.5 0.6 3.8
Other female (Not heads) (%) 99.9 96.8 99.5 99.4 96.2
Monogamous household (%) 87.8 93.0 95.2 92.6 93.5
Polygarnous household (%) 12.2 7.0 4.8 7.4 6.5
Land holding (mean acres) 8.6 5.7 6.6 6.7 5.9
Household income (‘000 Ksh., 145.0 125.0 132.4 142.6 122.3
mean)
Asset value (‘000 Ksh, mean) 180.1 102.9 1 14.1 134.4 99.3
Income quartile (%)
Lowest 25.1 30.3 26.0 22.6 23.6
Mid-low 25.4 25.8 21.5 24.6 26.7
Mid-high 20.1 22.3 25.2 25.6 26.9
Highest 29.5 21.5 27.3 27.3 22.8
Asset value quartile (%)
Lowest 31.7 28.4 21.3 23.8 24.0
Mid-low 20.7 25.6 23.8 25.2 26.8
Mid-high 17.4 20.9 27.7 28.3 26.3
Highest 30.1 25.1 27.2 22.7 22.9
Number of individuals 753 716 1 142 806 2338

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997, 2000, 2002, 2004, and 2007.

23

2.4. Methodology
2.4.1 Working-age adult mortality model

Based on cited literature, there is mixed evidence about the correlation between
adult mortality and socioeconomic status. On the one hand, morbidity and mortality rates
are inversely related to indicators of socioeconomic status such as income, wealth,
education, or occupation. Conversely, in the era of HIV/AIDS, individuals who are
relatively wealthier, richer, more educated or in more prestigious occupations may have
higher mortality risks. This is because they tend to put their health at risk through use of
their resources in sexual networking. Therefore, the impact of variables like education,
income, occupation and wealth on adult mortality is ambiguous. Even over time, their
influence will depend on the extent to which their potential perverse effects on health
serve to counteract the expansive or protective effects.

The econometric model for the relationship between mortality and socioeconomic
characteristics can be written as:
P(WAM,-) = f(1C,- ,HC;,CC,-) (2.1)
where WAM ,- is a binary variable equal to 1 if individual i died of disease-related

causes; IC, HC and CC represent a set of covariates measured at the individual,

household and community levels, respectively, for individual 1'.

2.4.2 Empirical analysis

In this essay, the probability of death (death hazard) is estimated using hazard
analysis. Hazard models (also referred to as duration or survival models) focus on the

timing of a change in state or a transition from one category to another such as from

24

unemployment to employment, a change in marital status, or as in this essay, a change in
the survival status. The dependent variable is usually the length of time that elapses from
the start of the initial state until its end or until measurement is taken and the observation
is censored. This study focuses on transitions from life to death, and the effects of various
explanatory variables such as individual, household and community characteristics on the
probability of such transitions.

Duration analysis has its origins in what is called survival analysis, where the
duration of interest is survival time of a subject. In this case, the objective is to determine
how various treatments or socioeconomic characteristics affect survival times. Recent
treatments of duration analysis tend to focus on the hazard function, where it is not only
the duration of the event, per se, that is interesting, but also the likelihood that the event
will end in “the next period”, given that it has lasted as long as it has. The hazard function
allows us to approximate the probability of exiting the initial state within a short interval,
conditional on having survived up to the starting time of the interval (or given that the
state has been occupied for some length of time t) (Wooldridge, 2002).

Three equivalent functions are used to describe the distribution of survival time:
the survivor function, the probability density function, and the hazard function. These
functions can be derived from each other mathematically. The survivor fimction (also
called survival function) is defined as:

S (t) = Pr (time before failure is at least t)

= Pr (T > t) (2.2)
where T is the length of time a subject lives and t is a realization of T. The survivor

function depicts the probability (Pr) that failure has not occurred at time t. It is the

25

probability of “surviving” past time t. By definition of the cumulative distribution
function (cdf) of any random variable, the survivor firnctions are related to cdfs such that

S (t) =1 - Pr (T_<_ t)
=1 — F (t) (2.3)
where F (t) denotes the cdf.

The probability density function for survival times is defined in the usual way for
a density of a continuous random variable:

f“) = lim Pr{farlure in period (t, t + At)}

2.4
At—>0 At ( )

 

Hazard functions describe the conditional failure rate or the probability of failure
during a very small interval, given survival up to the beginning of the intermission. The

hazard rate is defined as:

 

 

, Pr(t s T s t+ AtlT 2t)
h(t) = 11m
At——>0 At
h (t) = Alimo Pr{subject with survival to timZtt fails in the period (t, t + At)} (2.5)
t—)

It is the probability of leaving the initial state in the interval (t, t + At) given survival up
until time t. Thus h(t) gives the instantaneous conditional probability of death
immediately after t, given survival until time t. The hazard function can be expressed in

terms density and cdf of survival time:

_ f(t) =fl
h(t)—14:0) sm (2'6)

These conditional probabilities, the h(t)s , are the fundamentals of discrete-time hazard

models.

26

The event of interest in this study is disease-related death. Individuals who do not
die are considered censored by the end of the observation period. The presence of the
censored cases suggests that the time-to-event is probably longer than the period of
observation. In order to obtain an unbiased estimate of the time-to-event, the censored
cases cannot be ignored. They are included in the analysis so that they can contribute
whatever information they contain.

Hazard models explicitly account for time and recognize that an individual’s risk
of dying changes over the course of time. They adjust for changes in individuals’ risk of
death in different time periods. This is necessary since some individuals die within the
first survey period while others do so in subsequent survey periods, and yet others are
still alive at the end of the observation period. Hazard models can incorporate time-
varying covariates and utilize time-series data by including annual observations as time-
varying covariates. These models also make use of much more data since they can be
constructed as binary logit models with each person-period as a separate observation.
Therefore, each individual contributes multiple years of observation. In this case, we
have multi-period logit models that can be interpreted as hazard models, and this is the
approach followed in this study. The dependent variable in a multi-period logit model is
constructed such that it is equal to one only in the survey interval in which an individual
dies.

A hazard probability that describes the “risk of dying” in each time-period can be
estimated. Hazard probability is the (conditional) probability that an individual will die in
a particular time-period, given that he/she has “survived” up until this period. I estimate a

discrete-time hazard model since the death event is observed at discrete times, i.e.,

27

between two surveys or equivalently, within a survey interval. Discrete-time analysis is
an extension of the basic logistic regression analysis. It uses logit analysis to examine the

occurrence and timing of events. Let hit be the hazard (or the probability) that an

individual i dies in survey interval t, given that he’s still alive at the beginning of that
interval.

Since the hazards are probabilities, they can be reparameterized so that they have
a logistic dependence on time periods and the predictors (Cox, 1972):

1
-(a1 Dr it +012 DZit +613 D3it +a4D4it +flX it )

 

hit:
1+e

(2.7)

where D1 i, D21, D3 ,-, and D4,- are a sequence of dummy variables indexing the four

survey periods in the study namely: 1997-2000, 2000-2002, 2002-2004, and 2004-2007.

The survey-period dummies are defined for each person as follows: Dm is equal to one

when t = 1 , and equal to zero when t takes any other value (2 through 4). The model does
not contain an overall intercept, but has multiple intercepts, one per survey-period, as

represented by the parameters, al to 0L4. The variable X i, denotes a person-level vector of

the individual (IC), household (HH) and community characteristics (CC). This is shown
here as a vector of time-varying variables but could also represent time-invariant
variables.

Taking the logistic transformations of both sides of the equation (2.7) gives:

 

h.
“(13:17 j: [alDlit + aZDZit + 0530311 +a4D4itI+ 5 X it (2-3)
I

28

Therefore, the probability that an individual will die is related to predictors by the above
standard logistic regression model.

The parameters or 1 to a4 represent the baseline logit hazard function or the

baseline level of hazard in each time period. It shows the survey-period to survey-period
hazard function for the entire sample, assuming a homogenous population in which
individuals are not distinguished by the values of any of the explanatory variables (Singer
and Willett, 1993). These parameters describe the overall profile of risk of death over
time and represent the main effect of a single predictor, time. The vector of slope

parameters, [3 describes the main effects of the explanatory variables on the baseline

hazard function. These parameters are associated with shifis in the baseline hazard
function.

The above model (2.8) incorporates a proportional odds or proportionality
assumption that the effect of the individual, household and community characteristics on
the log-odds of death is the same in every time period. It postulates that the entire family
of logit-hazard profiles represented by all possible values of the explanatory variables
shares a common shape, and is mutually parallel, and is shifted only vertically for
different values of the explanatory variables (Singer and Willet, 1993). This assumption
does not allow the effect of the variables to vary over time, and so the variables have a
time-invariant effect. This applies even in the case of time-varying variables. The

assumption can be relaxed by estimating the model shown in equation (2.4):

 

h.
ln[ ” JzialDlit +a2D2i, +a3D3it+a4D4it]+,31(Dlizx Xit)+flZ(D2it X Xit) (2.9)

I—hit
+fl3(D3,-, xXz't)+.54(D4it XXit)

29

To examine whether the effect of any variable varies over time, we include the
interaction between indicators of time and the variable in the hazard model. This
interaction implies that the effect of the variable on the hazard profile varies from survey-
period to survey-period. The model enables us to investigate whether and how the effects
of the various characteristics on the hazard of death vary across the 10-year observation

period.

2.4.3 Model variables

Individual characteristics

Individual characteristics include age, years of education, months residing at
home, and whether the adult is a household head/spouse or not. Age is a time-varying
variable and is defined as an individual’s age at the beginning of each survey period. It is
entered in liner and quadratic terms in order to detect potential non-linearities of age on
the hazard of death. Education is entered as a continuous variable after specification tests
rejected non-linearities in years of education. The number of months over the past year
that a person resided at home is included as a continuous variable.
Household characteristics

The vector of household characteristics includes: acres of land owned, value of
assets (productive assets and farm animals), and binary variables for security of land
tenure and whether the household is monogamous. Tenure security is defined as the
household head having a title deed or well-recognized right over a parcel of land. The
relationship between mortality, household assets, and landholding size may be non-linear.
However, specification tests rejected the non-linearity hypotheses and therefore only

linear terms of these variables are used in the reported regressions.

3O

Community characteristics
Community variables include distance to the tarmac road and fertilizer store,
which act as proxies for the extent of interaction among local residents and contact with

people from other areas.

2.4.4 Correcting for potential attrition bias

It is possible that the AIDS epidemic contributes to sample attrition in the data
used, resulting in selection bias. Characteristics of individuals that have attrited and
those remaining in the sample are explored using probit reinterview models. To control
for possible attrition bias, the inverse probability weighting method is used as outlined
below.
Re-interview model

We control for possible attrition bias using the inverse probability weighting
method (IPW). The IPW method assumes that the probability of being re-interviewed
(non-attrition) as a function of observable information is the same as the probability of
being reinterviewed as a function of observables, plus unobservables that are only
observed for non—attrited observations (Wooldridge, 2002). Thus, the IPW method works
well if observations on observed variables are strong predictors of non-attrition and if
observations on unobserved variables are not strong predictors of non-attrition. Following

Yamano and Jayne (2005), the re-interview model can be written as:
P(Riht =1) = f(HC1997, ETiht , PD) (2.10)
where Riht is equal to 1 if an individual i remained in household h which was re-

interviewed at time t, conditional on the individual being in the household in the previous

31

time period and on the household being interviewed in the previous period, and zero
otherwise. The function f is specified as the standard normal cumulative distribution. The

set of explanatory variables includes individual and household characteristics in the 1997

survey (HC1997) , binary variables for enumeration teams (E Tim) , and provinces (PD).

Team dummies account for potential differences across enumeration teams in the amount
of effort put into locating households in follow-up interviews. This procedure identifies
statistical differences in the attributes of re-interviewed vs. attrited individuals. Stratified
by gender, a set of probit models are rim to determine an individual’s probability of
reinterview during each of the follow-up surveys.

Equation (2.10) is estimated with probit for attrition by gender between the 1997

and 2000 surveys, obtaining predicted probability, P2000; between 2000 and 2002 to
obtain the predicted probability, P2002; between 2002 and 2004 to obtain the predicted
probability, P2004 and; between 2004 and 2007 to obtain the predicted probability,
P2007. For observations in the 2000 survey, the inverse probability weight is

R2000 =1/P2000 . But for the observations in the 2002 surveys, the inverse probability
weight is R2000 , a product of 1/ @000 and 1/15002 because these observations survived
attrition twice. It follows then that for observations in 2004, the inverse probability
weight R2004 is a product of 1/ P2000 , 1/15002 and 1/ 15004 , while R2007 is a product

of 1/P2000 , 1/‘2002 , 1/112004 , and l/P2007 . The models on working-age adult
mortality described in the previous section (2.4.2) are estimated using these inverse

probabilities as weights in order to correct for attrition bias.

32

2.5 Results

2.5.1 Descriptive analysis

Basic characteristics of working-age adults who remained alive and those who
died due to disease-related causes are shown in Tables 2-3, 2-4, 2-5 and 2-6, for the four
survey periods: 1997-2000, 2000-2002, 2002-2004 and 2004-2007. Table 2-4 focuses on
the period 2000-2002 and has a smaller set of characteristics since information on some
characteristics is not available. Results are provided separately for deceased individuals
who were in the demographic roster at the beginning of the period and those who came
into the households anytime during the period under consideration.

Results from Table 2-3 indicate that women in the 15-59 year age-range who died
tend to be younger than men. Working-age adults who joined households after 1997 and
died by 2000 are younger than the mean of those in the demographic roster in 1997.
Deceased working-age men have fewer years of education than those who remained
alive, but women who joined households and later died are more educated than those who
remained alive and the other deceased women. Most of the working-age adults in the
sample have primary or secondary education, and as a result mortality is predominant
among adults with these levels of education. There is little difference in the number of
months spent at home among all working-age adults who were in the 1997 survey.
However, deceased men and women who joined the households after 1997 have resided
at home for fewer months. This may be consistent with the suggestion that such

individuals return home to seek terminal care and die.

33

Adult daughters account for more than half of the female deaths. Among
deceased men who were resident household members in 1997, most deaths relate to

household heads while among those who joined, sons account for half of the deaths.

34

Table 2-3 Characteristics of working-age adults in 2000 who remained alive, and those
who died due to illness-related causes between 1997 and 2000

 

Working-age adults in 1997 Deceased working-age adults
survey who remained alive
Were in 1997 Joined household

 

 

survey after 1997 survey
Female Male Fem Male Female Male
Age (mean) in 2000 33 32 37 42 23 35
Months residing at home 1 1.0 10.0 12 11 2.6 6.8
Years of education (mean) 6.8 7.9 5.7 6.8 7 7
Relationship to household
Head/spouse 45.6 31.6 34.3 61.7 0.0 33.3
Son/daughter 45.2 57.8 51.4 27.7 70.0 50.0
Others 9.3 10.6 14.3 10.6 30.0 16.7
Education categories (%)
No formal education 14.9 7.2 28.6 19.1 0.0 33.3
Primary 58.3 55.9 54.3 53.2 80.0 16.7
Secondary 24.2 31.2 17.1 23 .4 20.0 50.0
College/university 2.6 5.8 0.0 4.3 0.0 0.0
Has formal job (%) 5.6 19.0 2.9 19.1 0.0 16.7
No formal job (%) 94.4 81.0 97.1 80.9 100.0 83.3
Has informal activity (%) 15.8 18.2 20.0 25.5 10.0 0.0
No informal activity (%) 84.2 81.8 80.0 74.5 90.0 100.0
Female household head (%) 4.2 N/A 14.3 N/A 0.0 N/A
Other female (Not heads) (%) 95.8 N/A 85.7 N/A 100.0 N/A
Monogamous household (%) 95.1 97.9 85.7 93.6 100.0 83.3
Polygamous household (%) 4.9 2.1 14.3 6.4 0.0 16.7
Land holding (mean acres) 6.6 6.4 6.2 5.7 4.4 3.5
Income quartile (%)
Lowest 27.5 22.3 25.7 23.4 20.0 33.3
Mid-low 24.8 25.0 20.0 27.7 30.0 16.7
Mid-high 24.3 25.8 25.7 27.7 20.0 33.3
Highest 23.4 26.9 28.6 21.3 30.0 16.7
Asset value quartile (%)
Lowest 25.6 24.2 22.9 27.7 40.0 0.0
Mid-low 25.3 24.8 34.3 17.0 20.0 33.3
Mid-high 25.1 25.1 20.0 29.8 30.0 16.7
Highest 24.1 25.9 22.9 25.5 10.0 50.0
Province
Coast (N) 182 232 3 3 0.0 0.0
(%) 6.4 8.2 8.6 6.4 0.0 0.0
Eastern (N) 478 502 3 6 0.0 0.0
(%) 16.8 17.7 8.6 12.8 0.0 0.0
Nyanza
Kisii (N) 165 151 1 2 0.0 1
(%) 5.8 5.3 2.9 4.3 0.0 16.7
Kisumu/Siaya (N) 286 245 12 20 4 3
% 10.1 8.7 34.3 42.6 40.0 50.0
Western (N) 580 535 6 6 3 1
(%) 20.4 18.9 17.1 12.8 30.0 16.7
Central (N) 354 365 3 4 l 0.0
(%) 12.5 12.9 8.6 8.5 10.0 0.0
Rift Valley (N) 798 800 7 6 2 1
(%) 28.1 28.3 20.0 12.0 20.0 16.7
Number of individuals 2843 2830 35 47 10 6

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997 and 2000.

35

The sample contains few adults with formal jobs and informal/business activities,
who are female household heads and from polygamous households; the same pattern in
numbers is reflected in mortality levels for people with these characteristics. None of the
women who died had a formal job and none of those in the 1997 survey who died had
any informal/business activity.

Deceased adults who joined households after 1997 appear to be from households
with less land. Men and women who remained alive and those who died don’t seem to
coalesce into any income quartile. With respect to asset quartiles, deceased working-age
women are from poorer households compared to men. Most female deaths among those
who joined households after 1997 are from the lowest asset quartile while for men, no
deaths are found in this quartile. It appears that mortality rates are higher among poorer
adults, especially for women.

There are apparent regional differences in working-age adult mortality. Kisumu
and Siaya districts in Nyanza province have the highest percentage of both female and
male mortality, followed by Western province while Coast province has the lowest adult
mortality. Although Nyanza province represents about 15 percent of the working-age
adults in the sample, it accounts for almost 40 percent of death cases.

Table 2.4 shows characteristics in 2002 of working-age adults who remained
alive, and those who died due to disease between 2000 and 2002. Thirty-three working-
age adults who were in the 2000 survey passed away. Only 4 adults who joined the
households sometime after 2000 died; we report statistics for this group in Table 2-4 for
completeness, but of course a sample of this size cannot be considered in any way

statistically representative. Adults, who were in 2000 survey and passed away are much

36

older, less educated, and have resided at home for fewer months than the individuals who
remained in the sample. Most of the working-age adults who remained alive have
primary or secondary education while those who died have either primary or no formal
education. Mortality rates are higher among sons and daughters compared to heads and
spouses. The sample contains few adults with formal jobs and informal/business
activities, who are female household heads and from polygamous households. As a result
we find fewer deaths among adults with formal jobs and informal/business activities, who
are female household heads and from polygamous households. None of the women who
died had any informal activity or was a household head. Adults who joined the household
within the period and died did not have a formal job and were not involved in
informal/business activities. There are differences in mortality rates across the provinces,

with Coast province having the least number of deaths while Nyanza and Western

provinces have the highest deaths within this period.

37

Table 2-4 Characteristics of working-age adults in 2002 who remained alive, and those
who died due to illness-related causes between 2000 and 2002

 

Working-age adults in Deceased working-age adults
2000 survey who

remained alive

 

 

 

Were in 2000 Joined household
survey after 2000 survey
Female Male Female Male Female Male
Age (mean) in 2002 29 28 34 35 34 26
Months residing at home 8.1 7.7 3.0 3.5 4.3 0
(mean)
Years of education (mean) 8 9 6 8 8 10
Relationship to household
head (%)
Head/spouse 30.4 19.5 38.9 26.7 0.0 0.0
Son/daughter 51.9 61.4 61.1 46.7 66.7 100.0
Others 17.8 19.1 0.00 26.7 33.3 0
Education categories (%)
No formal education 10.6 5.8 22.2 13.3 33.3 0.0
Primary 56.0 51.8 61.1 60.0 33.3 0.0
Secondary 28.0 33.0 16.7 13.3 0.0 100.0
College/university 5.4 9.4 0.0 13.3 33.3 0.0
Has formal job (%) 5.8 11.4 5.6 13.3 0.0 0.0
No formal job (%) 94.2 88.6 94.4 86.7 100.0 100.0
Has informal activity (%) 13.3 17.0 0.0 6.7 0.0 0.0
No informal activity (%) 86.7 83.0 100.0 93.3 100.0 100.0
Female household head (%) 3.2 N/A 0.0 N/A 0.0 N/A
Other female (Not heads) (%) 96.8 N/A 100.0 N/A 100.0 N/A
Monogamous household (%) 94.0 95.0 94.4 93.3 100.0 100.0
Polygamous household (%) 6.0 5.0 5 .6 6.7 0.0 0.0
Province
Coast (N) 231 229 2 2 0 0
(%) 7.1 7.0 11.1 13.3 0.0 0.0
Eastern (N) 543 564 3 1 l 0
(%) 16.6 17.2 16.7 6.7 33.3 0.0
Nyanza
Kisii (N) 171 185 1 0 0 0
(%) 5.2 5.7 5.6 0.0 0.0 0.0
Kisumu/Siaya (N) 337 293 3 5 0 1
(%) 10.3 8.9 16.7 33.3 0.0 100.0
Western (N) 731 727 6 2 0.0 0
(%) 22.4 22.2 33.3 13.3 0.0 0.0
Central (N) 414 428 0 3 2 0
(%) 12.7 13.1 0.0 20.0 66.7 0.0
Rift Valley (N) 838 848 3 2 O O
(%) 25.7 25.9 16.7 13.3 0.0 0.0
Number of individuals 3265 3274 18 15 3 l

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 2000 and 2002.

38

Characteristics in 2004 of working-age adults who remained alive and those who
died between 2002 and 2004 are shown in Table 2-5. Out of 65 working-age adults who
died between 2002 and 2004, 52 of them were in the 2002 survey. Adults who died were
older while men who came to the households after 2002 and later died, were 11 years
younger than those who were in 2002 survey and passed away. There appears to be no
difference in education between adults who died and those who remained alive, with
most of them having either primary or secondary education. A large proportion of
females who were in the 2002 survey and later died were relatives of the household head
while most cases of male mortality pertained to household heads. Among the men who
joined the households after the 2002 survey and died by 2004, most were sons of the
household head while women were either spouses or daughters. There were no deaths
among female heads of households. Unlike previous periods, we find a few deaths among
those with formal jobs and informal/business activities. However, this is only among
adults who were in 2002 survey. In addition, deceased adults appear to be from
households with less land but men who join households and later die seem to join
households with more land.

For working-age adults who were in 2002 survey, there are fewer deceased
women in the poorest income quartile while a larger percentage of the men who died
belong to this income quartile. For adults who joined the households after 2002 survey
and later died, women appear to belong to the bottom two income and asset quartiles
while men (who are principally sons) are in the top two quartiles. Once more, we find

more deaths in Nyanza province than in other provinces.

39

Table 2-5 Characteristics of working-age adults in 2004 who remained alive and those

who died due to illness-related causes between 2002 and 2004
Working-age adults in

Deceased working-age adults

 

 

 

2002 survey who
remained alive
Were in 2002 Joined household
survey after 2002 survey
Female Male Female Male Female Male
Age (mean) in 2004 30 29 38 44 38 33
Months residing at home 8.1 7.4 1.9 4.0 2.4 3.0
Years of education (mean) 8 9 8 9 7 9
Relationship to household
Head/spouse 34.1 21.3 16.1 47.6 42.9 16.7
Son/daughter 48.3 60.7 35.5 33.3 42.9 66.7
Others 17.6 18.0 48.4 19.0 14.3 16.7
Education categories (%)
No formal education 9.3 3.4 6.5 4.8 14.3 0.0
Primary 55.3 54.6 74.2 52.4 71.4 66.7
Secondary 27.7 32.0 12.9 38.1 14.3 33.3
College/university 7.7 10.0 6.5 4.8 0.0 0.0
Has formal job (%) 9.9 15.1 3.2 9.5 0.0 0.0
No formal job (%) 90.1 84.9 96.8 90.5 100.0 100.0
Has informal activity (%) 12.0 13.8 3.2 4.8 0.0 0.0
No informal activity (%) 88.0 86.2 96.8 95.2 100.0 100.0
Female household head (%) 4.8 N/A 0.0 N/A 0.0 N/A
Other female (%) 95.2 N/A 100.0 N/A 100.0 N/A
Monogamous household (%) 93.7 94.8 93.5 100.0 85.7 66.7
Polygamous household (%) 6.3 5.2 6.5 0.0 14.3 33.3
Land holding (mean acres) 4.80 4.81 2.67 2.65 2.12 4.85
Income quartile (%)
Lowest 26.5 23.4 19.4 33.3 28.6 16.7
Mid-low 24.5 25.7 25.8 23.8 42.9 0.0
Mid-high 24.5 25.4 22.6 28.6 28.6 33.3
Highest 24.5 25.5 32.3 14.3 0.0 50.0
Asset value quartile (%)
Lowest 25.5 24.5 25.8 23.8 14.3 33.3
Mid-low 24.9 25.2 12.9 42.9 42.9 0.0
Mid-high 25.5 24.4 32.3 14.3 14.3 50.0
Highest 24.5 25.9 29.0 19.0 28.6 16.7
Province
Coast (N) 205 204 1 2 0 O
(%) 7.5 7.3 3.2 9.5 0.0 0.0
Eastern (N) 470 472 3 3 O O
(%) 17.1 16.9 9.7 14.3 0.0 0.0
Nyanza
Kisii (N) 153 169 2 0 0 0
(%) 5.6 6.1 6.5 0.0 0.0 0.0
Kisumu/Siaya (N) 304 296 6 10 2 2
(%) 11.1 10.6 19.4 47.6 28.6 33.3
Western (N) 577 585 6 4 4 0
(%) 21.0 20.9 19.4 19.0 57.1 0.0
Central (N) 336 346 4 2 1 1
(%) 12.2 12.4 12.9 9.5 14.3 16.7
Rift Valley (N) 698 721 9 0 0 3
(%) 25.4 25.8 29.0 0.0 0.0 50.0
Number of individuals 2743 2793 31 21 7 6

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring and Policy

Analysis Household Surveys in 2002 and 2004.

40

Table 2-6 shows the characteristics in 2007 of working-age adults who remained
alive and those who died between 2004 and 2007. The results show that working-age
adults who joined households after 2004 and died by 2007 are younger than the deceased
adults who were in the 2004 survey. Among the adults in the 2004 survey, the deceased
ones spent more months at home than those who did not die.2 The former may have been
too sick to participate in any activity outside the home and were receiving care at home.

Heads of households account for most deaths among individuals who were in the
2004 survey. However, among those who joined later, sons and daughters account for a
larger percentage of the deaths.

Most of the working-age adults in the sample have primary or secondary
education. However, we find that a slightly larger number of women deaths among those
with primary or no formal education, while more deceased men have primary or
secondary education.

The sample contains few adults with formal jobs and informal activities, who are
female household heads and from polygamous households. The same pattern is generally
reflected in the number of deaths among people with these characteristics. However, we
note a slightly higher number of deaths among males with a formal job and women from
polygamous households.

Adults who died appear to come from households with slightly smaller
landholdings, and particularly for those who joined the household after 2004 and died by

2007.

 

2 Among deceased working-age adults who joined the households after 2004, information on months at
home, as well as participation in formal employment and informal activities is missing for some of them. A
dash (-) in the last 2 columns of Table 2-6 indicates that information on the variable is missing for some
individuals and so no means or percentages are reported.

41

Table 2-6 Characteristics of working-age adults in 2007 who remained alive and those
who died due to illness-related causes between 2004 and 2007

 

 

 

 

Working-age adults in Deceased working-age adults
2004 survey who
remained alive
Were in 2004 Joined household
survey after 2004 survey
Female Male Femal Male Female Male
Age (mean) in 2007 33 32 39 39 34 31
Months residing at home 7.9 7.3 10 10 - -
Years of education (mean) 8 9 7 8 8 10
Relationship to household
Head/spouse 31.9 20.4 52.5 45.2 0.0 0.0
Son/daughter 45.1 56.5 37.5 25.8 62.5 75.0
Others 23.0 23.2 10.0 29.0 37.5 25.0
Education categories (%)
No formal education 9.0 3.9 20.0 3.2 0.0 0.0
Primary 56.2 56.3 57.5 58.1 75.0 50.0
Secondary 27.7 30.8 17.5 32.3 25.0 25.0
College/university 7.1 9.0 5.0 6.5 0.0 25.0
Has formal job (%) 9.2 14.2 10.0 29.0 - -
No formal _iob (%) 90.8 85.8 90.0 71.0 - -
Has informal activity (%) 11.5 13.4 20.0 12.9 - -
No informal activity (%) 88.5 86.6 80.0 87.1 - -
Female household head (%) 0.0 N/A 0.0 N/A 0.0 N/A
Other female (%) 100.0 N/A 100.0 N/A 100.0 N/A
Monogamous household (%) 93.4 94.6 85.0 100.0 100.0 100.0
Polygamous household (%) 6.6 5.4 15.0 0.0 , 0.0 0.0
Land holding (mean acres) 4.9 4.8 4.3 4.6 3.5 3.1
Income quartile (%)
Lowest 26.5 23.3 27.5 35.5 37.5 0.0
Mid-low 24.4 25.7 30.0 16.1 12.5 50.0
Mid-high 24.7 25.2 22.5 29.0 25.0 25.0
Highest 24.3 25.8 20.0 19.4 25.0 25.0
Asset value quartile (%)
Lowest 26.1 25.8 27.5 29.0 37.5 0.0
Mid-low 25.5 23.8 20.0 25.8 25.0 25.0
Mid-high 25 .1 24.6 25.0 29.0 12.5 50.0
Highest 23.3 25.0 27.5 16.1 25.0 25.0
Province
Coast (N) 288 261 2 3 0 l
(%) 8.4 7.5 5.0 9.7 0.0 25.0
Eastern (N) 548 527 6 2 2 1
(%) 16.1 15.1 15.0 6.5 25.0 25.0
Nyanza
Kisii 229 209 1 1 0 0
6.7 6.0 2.5 3.2 0.0 0.0
Kisumu/Siaya 376 386 8 7 l 0
( 11.0 11.1 20.0 22.6 12.5 0.0
Western (N) 673 696 8 6 3 0
(%) 19.3 19.9 20.0 19.4 37.5 0.0
Central (N) 380 404 4 3 l 0
(%) 11.1 11.6 10.0 9.7 12.5 0.0
Rift Valley (N) 919 1006 11 9 1 2
(%) 26.9 28.8 27.5 29.9 12.5 50.0
Number of individuals 3413 3489 40 31 8 4

 

Source: Tegemeo Institute Michigan State University Agricultural Monitoring and Policy Analysis
Surveys in 2004 and 2007. A dash (-) indicates information is missing for some individuals.

42

A slightly higher percentage of deceased men and women who were in 2004 are
in the bottom two income and asset quartiles. Unlike in previous years, we find more

deaths in Rift Valley province compared to Nyanza province.

Mortality by gender and age

Figure 2-1 presents percentage of deaths within age-group by gender. It reveals
that there are age/cohort and gender differences in mortality. In the earlier stages of the
life-cycle, there are more deaths among women than men but the trend reverses in later
years. Percentage of deaths among women rises and is generally greater for men up to
age group, 30-34 years. From 35-39 years age-group, this percentage declines and is less
than that for men particularly for the age-groups 50-54 and 55-59. The differences in
percentage of deaths between men and women are more pronounced for the age-groups
20-24, 30-34, 35-39 and 50-54, with the largest gap in the 20-24 age group. Therefore,
age is a critical factor in working-age adult mortality. Data in 2004 and 2007 report cause
of death by type of disease but in the other surveys, it is reported as ‘disease’. Based on
these surveys, the five leading causes of death among women are malaria, HIV/AIDS,
TB, meningitis and cancer while among men they are malaria, TB, HIV/AIDS, cancer
and stroke/heart disease.

In general, three broad categories of explanations have been suggested for gender
differences in ill-health: biological, reporting differences and differences in behavior
(Strauss et al, 1993). Biological explanations relate to genes and hormones. With respect
to health reporting behavior, women are likely to report more health problems. It is

commonly argued that women are more complete and detailed than men in any interview,

43

including a clinical setting (Verbrugge, 1985). Strauss et a1 (1993) point out that
behavioral choices have a significant effect on gender differences in health for various
reasons. Child-bearing may affect physical functioning of women through maternal
depletion and particularly where health facilities are poor and fertility rates high. Gender—
based occupational and time allocation decisions may cause women and men to face
different types and levels of stress. This leads to different levels of health between men

and women.

 

 

 

 

 

 

 

m i
e as
g .5 :«r
u- ' w , “I
g j‘: a»: ‘ C Ifemale
g if?“ g E lmale
C [3‘51 -‘ I
0 I-
o m , n
E {ha 5 a
5.: . .
3:3“ 7"” a E3
E“ 3"" 5‘ 5
it; {”1 d t”
—t N N 00 (A) A A U! 0..
5" F’ S" ‘P S“ f j: ‘P cln
F5 R 8 g 8 «h to CA” 3
Age Groups

 

 

 

Figure 2-1 Percentage of disease-related deaths within age groups by gender, for the
period 1997-2007

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997, 2000, 2002, 2004 and 2007.

44

Different activities with respect to smoking, drinking and sexual behavior may
also lead to gender differential in health. It is also possible that women and men have
different access to health care arising from household choices. In this sample of working-
age adults, the pattern of deaths by age-group and gender reflects that which has been
reported for HIV/AIDS, where mortality is higher among relatively younger women and

slightly older men.

Correlation between HI V prevalence rates and mortality rates from survey data

Before reporting the main results, it is important to establish the extent to which
the disease-related mortality among individuals aged 15-59 years in this nationwide
survey is associated with AIDS. To test this, I computed mortality rates at the province-
level, which correspond to available data provided by NASCOP (2005) on HIV
prevalence rates among women tested at Ministry of Health antenatal clinics. Taking into
account the lag time between the median years since HIV sero-conversion in an HIV
positive sample and full-blown AIDS, we match available data on HIV prevalence rates
from the 1990-2001 period with our survey-based mortality rates over the 1997-2007
period. Figure 2-2 shows a strong correlation (Pearson’s coefficient of 0.69), providing
reasonably solid evidence that the disease-related mortality in our data is to a large extent

picking up the effects of HIV-related mortality.

45

 

12.0
' Kisumu o
'é
“.1 10.0
O
01
O)
H
g; 3.0 ,
4.0 /
I! 1/
w
g /'1’
_g 6.0 /.
9 ,/
e C . .,
a Rift Valley _/ 0‘15
2 4.0 \ ,r/ o
I Easterm, \V ,,./ .
— \g “ ./" Western
05 j) ’ .
g . go ’ / Central
.5 2.0 ‘ “’J /" '
o ' Kisii
L-
o.
0.0 ,
i 0.0 5.0 10.0 15.0 20.0
Provincial Adult mortality rates, 1997-2007

 

- m..- ...4". “mm. -.¢- __... ..-... —

 

25.0

Figure 2-2 Provincial adult mortality rates from household survey data (1997-2007)

and HIV prevalence rates (1990-2001)

46

 

2.5.2 Factors influencing re-interview

Results in Table 2-7 are based on equation (2.10) in section 2.4.3. Findings show
that household heads and spouses are more likely to be re—interviewed than unmarried
working-age adults. Age has a non-linear effect on probability of re-interview for all
model specifications. At young age ranges, the probability of reinterview declines with
age, but then increases beyond the ages of 42 and 41 for males and females, respectively
(based on results in columns 9 and 10). Older individuals are more likely to be re-
interviewed compared to younger individuals. In all specifications, the probability of re-
interview decreases with years of education but the results are statistically significant
only in the cases of men in the 2000 and 2002 surveys.

While men with formal jobs are more likely to be reinterviewed (columns 7 and
9), women are less likely to be reinterviewed (column 10). Generally, individuals who
spend more time at home are more likely to be reinterviewed. The results also show that
in general, adults in households with more children are more likely to be reinterviewed.
The effect of household wealth on reinterview is generally negative, often statistically
significant in the case of females. However, the effect is rather small. Holding other
factors constant, women from households at the 25th and 75th percentile of the assets
distribution, are nearly equally likely to be re-interviewed. The probability of re-
interview for the two groups are, respectively, 0.895 vs. 0.894 in 2000, 0.839 vs. 0.835 in
2007, and 0.847 vs. 0.846 in the case of the pooled model. Security of land tenure is
positively related to reinterview (columns 2, 3, 4, 9 and 10) but amount of land owned is
negatively related to reinterview for males in the 2000 survey only (column 1). Adult

women from polygamous households are less likely to be reinterviewed (columns 2, 8

47

and 10), possibly because lack of adequate economic resources in such large households
maybe forcing young women to move out in search of better opportunities or even get
married. However, we find mixed results in the case of men. Men from polygamous
households were less likely to be reinterviewed the 2000 survey, but more likely to be
reinterviewed in 2004.

Generally, there is evidence of joint significance of team dummies as well as
individual and household characteristics, implying that these categories of variables are
strong predictors of reinterview. Community characteristics are, however, jointly
significant in a few of the specifications.

The implication of the findings in Table 2-7 is that attrition bias is a problem in
this data set and there is need to control for attrition while using estimating the hazard

models of working-age adult mortality.

48

 

 

 

 

49

3. c 3.9 33v 5.3 8.8 8.8 SE 2 m. s 3.8 3.3
+83 :33 33 .83 S3 S3- :83 83 83 :23 2:2 a was:
3.8 83V 3.8 8. : Gm. : 3.8 Ge. : 8. z : _.: an s
:3- 83 83- $3 3.3- 83- 83- 83- 23 83 st 33:8 3.585
A8. a 5.9 an. : 63v 63v 3.8 83v 8. : § . 5 3.8
+83. .83 8a.? :S_.o :3- :3- 83- 83 83- w _3 at 8.. ago...
5. c 8.8 33V 8. s 5.: 8.8 3.8 a». c 5.8 893
83- 83- 83- 83- 33- 83- 83. +33- 83- 33 8:828 .8 we;
2 as 3.5 A8. 5 83¢ 83. a and 3. a 6»: 33 63V
:3 2.35 .35 +mo
-2 ._ -08 +3-8; :383 3-3; :32.“ -2 ._ -03 :38; :38; coascm ow<
93¢ 5:3 5. c 3.3 3.: 8.5 383 3.8 8i 5 .3
:83 :33
- - +_ _3. : _ 23. :23- :33- .583- ...83- :29? :33- 8....
$3: 32.: G3: 6.2 s 83: A83: 8.: 8.: 83: 83V
:83 :33 :53 :83 :33 :23 83 83- .323 :8_.c 2808838.;
32 5 88:32:23 .3222:
a: 5 as E g 6 € 5 Q 5
285m 222 295-.— 222 285m 322 e_mEon— 0.22 225.; 222
8.988
.255 38¢ sagas 38 e 88 88 ea 32,
.88 .32 a a as. $383 88 2a 88 a 2 ea .88 nae: 82 a a as
.38 s 8323852 .82 s a as .38 5 a Baas-.252 .88 a 33252?
.2222. 853.532 33%.: 32:3 32%...

 

: _u 033:; 806530

mooted .833 28 Hots-ow .3 322852 $573229: Sm £088 “505 TN 2an

 

 

 

 

 

83: 8.: 33: G3: 3.8 A8,: 83: 83: 28.: 8.:
83 83 83 83 83 83 83 83 .8...- .83- 9288:8685.
6...: 8.: $3: 83: :3: .88: Go: 3.: 8.: E:
*
.83 83 23 83 83- :3 :88 +83 .88... 83 2:880:83
8.: :3: 8.: 83: Comm 8.: 33: A8: 8.: 93:
89
+8 8... .8
-3...- - 8.? 8-8... .. 8.28. 8.08.8- 8.88.. .883 .8-8. .- 8-8. s- 8... .98
.8: 8.: 5.: 5.: $3: 83: 9...: 5.: $3: 8.8
i...
8.3 :3 :3 L8... 83 83 :83 :8... 83- 83 8.5528» 3.38m.
83: 83: .88: an: :3: 3...: 8.: A8,: 9...: 3.:
e858:
83 83 88.8- 83- 83 8.3 :3 :3 .83- 83- 38355.8.88
63: 8.8 .38: ES: :3: an: .8: an: an: .8:
83 83 83 8.3 83. :3- ...8... 83 88.? 88.8- €385.38 2-3.5
8.8 8.: 83: 5.: A8,: :3: a3: 8.8 93: 8.:
Con—5.5
88.8- 83 83 83 +83 83 88.8- 83- 8...? 23 $0355525
53 E 838-.3922? 233.3:
8: A: A: S G: 5 a: 5 A: 3
2.2.5: 222 285.... 252 2.2.8: 282 225-.— 282 2.25... 2.82
8:528
.82: 8.8.. 88 a 88 .88 3.95. 88 Ea $.95.

.33 E m. :5 .83
5 3303.858

.8238.

Son .32 E m. :5
.38 5 3333852

.8238.

coom :5 32 E m. :5
Noon 5 3303.852

t _u oEatg Botswana

.8238.

28.8 58. a. m. as.
.OOON 5 60322358

.8288.

323:8 TN via-_-

50

 

 

 

 

my? mo> mo> mo> mo> mo> mo> mo> mo> mo> moi—:56 00:30.5
mo> mo? mo> mo> mo> mo> mo> mo> mo> mo> moms—=56 Each.
ave A8: A3: cod 8.8 2 <8 Gmd A35 8m: cod
88.8. 88.8 .888- A88- 88.8. 88.8 88.8 .388 _88- _8.o 8% 33:5 2 85:5
8d 3.8 82: 8.8 E: Ewe fine 32: ans 88.8
.. .88- 88.8. 88.8. 88.8 ~88. :88. 88.8- 88.8 .. .88. :88. 82 8.55 9 8555
8:358:35 3:55:80
$38 868 and 808 62: Ga: Gm: an: A33 ASS
tecwod ELF—d
- 288. - _88 no.8. :88 Rod- 38.8. .828- :29? ct 28032. 38338
83 Q as E § 6 E 3 5 a
2.“th 232 22:“; 222 e_mEom 252 25th 222 03th 232
3.928
388 868 8N a. 88 .88. 8.95 88 Ba 8363

.hoe E m_ 98 .boom
5 3303.259.

.8223

88 .32 a 2 Ea

.: _u e_nmtg Eovcomoa

.38 5 3322852

383:5.

ooom can 32 5 m_ can
.88 E 3303::th

3:239:

29:3 32 a 2 Ba
.88 2 8322352
.2285

 

82.88 Z 22;

51

acts—.3325 05 mo 03.: 8 $3 3 8:35sz 83 85880335 3325800 .Som 8

$3.: .8 8:3 voom 8 383852 0380.: 83 038 5.: 8588 .X; 3 838.88%? 23 85:88.35 29385: :88 .x: 3 835%? 23 38888335 3:238:—
.moom 39A 5 3358 .3 383882 3.: £088 $2 3 8803:»? 03 Quote :38. ”88553: E 8:33: .
80882 8: 8:: 2080:. 05 E 38:38 mo_EE:u 883:: :83 Each
33.. 8:388:85 $2 8388: + ”32: 350$:wmm 8m 838E: .. :32 8:85ch 8.: $389: 1.

88553: 8: 83838 N 5:33:08: 5 mowcmzo 3:338 98 8:88:98 wouaezmm

 

 

 

 

3.082
.8888
:5 3cm .88 .88 .32 E 9835 228:0: 88:82 .8on :83 2:83:22 3.5::otw< b39532: 38m 38:83:93:sz 83:8: 8:8on ”330m
8; 88 $8 88 E8 88 88 88 88 :8 828580
38.8 88.8 88.8 E _8 :8 m8 88.8 38.8 28.8 :88 3.8.8
8... 8 ._ .3 8+ 8.8 8.8 8 .m 8.8 8:. 8: 885635 88858
88.8 38.8 88.8 :88 $88 38.8 :88 88.8 88.8 88.8
8.8 8.8 8.8 8.. _ 88 8.8 8.8 8.8. 8.3 8.8 8883838 2288:
88.8 888.8 88.8 A888 A888 A888 888 88.8 88.8 88.8
8.8% 8.88 8.8 m 3.8.8 8.8». 8 .88 8.8 8.8 8.8. 8.8 388333;”. .8832.
888.8 888.8 88.8 88.8 88.8 :88 88.8 38.8 3.88 98.8
8.8 8.8 8.88 8.88 938 NS 8.8 8:. $8 8.8 385 88:.
"1:: .8. 2:3. 0355-30
88 5 8 E 8 5 A8 5 A8 3
03:5”: 03.2 0380“: 03.): 03:5“: 03:2 03:5,.— 032 03th 032
8.983 .
388 8.88 88 a. 88 .88 8.8.5” 88 83 38:5

Sam .32 E m: :83
.voom 5 3323850..
3:238:

.2 _n 23:? 30:20:00

.82 8 m: 83 .38
:3 3383888

.8228:

2853 88 s 8 8a
.88 5 8335.52
.8228:

ooom :83 32 E m: 23
.83 5 80383858
3:239:

 

32:38 Wm 23,—.

52

2.5.3 Results from hazard analysis

During the 10-year study period, 22,654 person-years were generated from 5,755
adults aged 15-59 years. There were 216 deaths and the crude mortality rate was 9.53 per
1000 person-years, 9.19 for males and 9.87 for females. This section presents results

from three different discrete-time hazard models for the observation period 1997 to 2007.

Initial discrete-time hazard model

The results of the initial hazard model are shown in Table 2—8. This model
contains survey-period dummy variables as the only explanatory variables and therefore
describes the overall temporal profile of risk. Columns 1 and 2 show the parameter
estimates while the corresponding marginal effects are shown in columns 3 and 4. The
parameter estimates in columns 1 and 2 describe the shape of the fitted whole-sample
hazard function, for men and women, respectively. If the risk of death were not related to
time, these estimates would be equal and the hazard function would be flat. Results show
that the parameter estimates of the later periods are smaller than those of the earlier
periods for men, implying that men’s overall risk of death declines over time. In the case
of women, the hazard decreases in the second period, rises in the third period and then
declines in the fourth period. Wald tests on the specifications for men (chi2 (3)=11.31, p
value=0.0101) and women (chi2 (3)=8.87, p value=0.0310) indicate that we reject the
null hypothesis that the parameter estimates are equal. Therefore, the overall risk of death
is different in each survey period for both men and women in this sample. In general,
there is a downward trend in the probability of working-age adult mortality across the

observation period. However, mortality risk decreases faster for men than women. This is

53

consistent with decreases in AIDS-related deaths and HIV prevalence rates in Kenya,
which can be attributed to positive changes in sexual behaviour and significant progress
in treatment programs. Demographic health surveys indicate that both age at first sex and
use of condoms are rising, and the percentage of adults with multiple partners is falling
(Cheluget et al, 2006).

The risk of dying in each survey period can be recovered using equation (2.7)

(page 24). For example, in period 1 where D1 =1 and D2 through D4 are equal to zero,

the hazard probability for men can be obtained as:

1
"1 ’ 1 +e—(—4.004462x1) ‘ 0'0179'

 

The fitted hazard value of 0.0179 indicates that among this sample of working-age adult
men, there is a 1.79% risk of dying of disease-related causes between the 1997 and 2000
surveys. Similar calculations lead to the estimates of the discrete per period hazards for
men and women reported in columns 5 and 6, respectively.

Table 2-8 Parameter estimates, marginal effects and hazard probabilities for the initial
baseline hazard model for the period 1997-2007

 

Parameter estimates

Marginal effects

Per period hazard

 

 

 

 

probabilities
Survey period Male Female Male Female Male Female
(1) (2) (3) (4) (5) (6)

1997-2000 4019*" 4209"“ 0025*“ -0.027*** 0.0179 0.0118
(0.194) (0.196) (0.003) (0.003)

2000-2002 -4.696*** -5.209*** -0.029*" ~0.034*** 0.0093 0.0044
(0.348) (0.304) (0.003) (0.003)

2002-2004 4758*" 4386*" -0.029*** -0.028"* 0.0087 0.0100
(0.279) (0.254) (0.004) (0.003)

2004-2007 -5.212*** -4.755*** -0.033*** 0030*" 0.0056 0.0069
(0.344) (0.231) (0.004) (0.003)

Observations l 1311 l 1343 11311 1 1343

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997, 2000, 2002,2004, and 2007.

Notes: Standard errors are in parentheses; *" indicates 1 percent significance level; ** indicates 5 per

cent significance level;"‘ indicates 10 percent significance level

54

Discrete-time hazard model with time-invariant effects

The main effects of the individual, household and community characteristics, in
addition to the main effect of time, are shown in Table 2-9. The reported estimates are
marginal effects. These models are specified such that the effect of each explanatory
variable is the same in all survey-periods and therefore, the variables have a time-
invariant effect. This applies even in the case of time-varying variables. For instance,
although the values of the variables age and age squared vary from survey-period to
survey-period, their effects are constrained to be the same in all periods under the
proportionality assumption.

Time-invariant effects reported in column 1 of Table 2-9 indicate that in every
survey-period, men’s mortality hazard is unrelated to their age, even when age is
specified as five-year age group dummies for the 15-59 age-range (results not reported
here). This may be consistent with an AIDS’ flattening of the normal adult mortality
schedule as suggested by Weinreb (2004). Also, in each period, male heads of
households have greater hazard of death compared to non-head male adults. Perhaps this
may be attributed to the fact that heads are much older than the non-heads, a factor that
may increase the probability that they are infected by the HIV virus, particularly if they
engage in extramarital relationships.

More educated men are less likely to die in each survey period. This is because
more educated people have better access to information about existence, transmission and
prevention modes of HIV/AIDS epidemic. Education enables individuals to avoid

infection by using condoms or a combination of HIV testing and reciprocal fidelity, and

55

Table 2-9 Results of time-invariant hazard models of working-age adult mortality in rural
Kenya, 1997-2007

 

 

 

 

Male Female
Explanatory variables (1) (2)
1997-2000 period (=1) -0.024*** -0.072***
(0.008) (0.017)
2000-2002 period (=1) 0032*" -0.092"*
(0.010) (0.023)
2002-2004 period (=1) -0.028*** -0.077***
(0.009) (0.020)
2004-2007 period (=1) -0.027**"‘ -0.087*"
(0.009) (0.024)
Head/spouse (=1) 0.0080" -0.004l**
(0.0038) (0.0018)
Age 0.0005 0.0015***
(0.0004) (0.0005)
Age squared -4.57e-06 -l .64e-05***
(4.34e-06) (5.74e-06)
Years of education -0.0003* -0.0002
(0.0002) (0.0002)
Months at home -0.0003** 0.0003
(0.0002) (0.0005)
Monogamous household (=1) —0.0005 0.0049"
(0.0026) (0.0020)
Land tenure (=1) -0.0001 0.0042*
(0.0019) (0.0024)
Land owned (acres) 5.84e-05 -1.46e-04*
(6.85e-05) (8.27e-05)
Assets -2.24e-06 -7.78e-07
(2.98e-06) (l .88e-06)
Distance to fertilizer store (km) 0.0004 -0.0003
(0.0005) (0.0008)
Distance to tarmac road (km) -0.0008 -0.0008
(0.0007) (0.001 1)
Observations 1 131 l l 1343

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000, 2002, 2004, and 2007.

Notes: Standard errors are in parentheses; *** indicates 1 percent significance level;
** indicates 5 percent significance level;* indicates 10 percent significance level

56

enables them to mitigate consequences of the disease through better nutrition and health
care (Deheneffe et al, 1998; de Walque, 2002).

The number of months an adult man spends at home is negatively related to
mortality hazard. It has been hypothesized in the literature that spending more time away
from home is associated with higher risks of death due to disruption of family and
spousal or sexual relationships, as well as the existence of more social interactions and
opportunities for risky multiple sexual partnerships.

Landholding size and the value of productive farm assets (including livestock
assets) are important indicators of a household’s wealth. Results show that there is no
significant relationship between amount of land owned, household assets and male
mortality hazard. It has been argued that wealthier men are more likely to be infected by
HIV because their greater resources provide access to more sexual partners (Shelton Qt al,
2005; Annim & Awusabo-Asare, 2006). However, their wealth may confer an advantage
once they are infected. It enables them to alleviate the consequences of the disease
through better nutrition and health care and access to ARV treatment. The sum effect
may lead to a situation where wealth does not influence the probability of death, thus
flattening the wealth-mortality gradient.

The distance variables are not significantly related to men’s risk of death. This
finding may imply that the disease is no longer concentrated along the main highways
and points of contact with travellers but has permeated throughout rural areas.

In column 2, we find that in every survey period, women who are heads or
spouses are less likely to die compared to the non-head adult women. This maybe

because women who are not household heads tend to be younger; and it is the younger

57

women in their 203 who face greater risks of HIV/AIDS. This finding is in contrast with
the case for men and may be linked to the following explanations. First, men tend to
marry younger women, with the older men having higher chances of being infected by
the HIV virus. Thus, the presence of cross-generational sexual relationships implies that
the risk of dying is greater among younger women and relatively older men. Moreover,
this may be due to gender differences in sexual behaviors. For example, having multiple
sexual partners and extramarital liaisons is more dominant among men than women,
possibly because the society tends to somewhat condone such behaviour among men but
not women. Also, while women are likely to acknowledge their risk of infection, men
tend to downplay their vulnerability (Parikh, 2007). This leads to a greater risk of
infection and death among male heads in contrast to women heads/spouses.

Age has a non-linear effect on women’s mortality hazard. The linear term in age
is positive while the quadratic term is negative. Therefore, the risk of death among adult
women first increases with age and later declines after a certain age. This suggests that
mortality is higher among women in the middle of the 15-59 age distribution. This age-
mortality pattern is expected given that majority of new HIV/AIDS infections in Kenya
occur among the youth, especially for young women aged 15-24 (UNAIDS, 2006). It is
also consistent with the UNAIDS (2003) report which indicates that the risk of death
from AIDS is highest in the mid-30$ to early 405.

Adult women residing in monogamous households have higher probability of
death in each time period, a finding that appears surprising and counterintuitive.
Transmission of HIV in Kenya is mainly through sexual activity, and so avoiding

infection depends on risk-avoiding behaviour. Therefore, in theory, marriage could

58

reduce the risk of HIV infection and HIV-related mortality risk by promoting
monogamous relationships. However, research shows that married women’s greatest risk
of HIV infection is their husbands’ extramarital sexual activities (Parikh, 2007). Based on
a study in Uganda, Parikh suggests that social and economic contexts surrounding men’s
extramarital sexuality may increase marital HIV risk. First, she points out that HIV
prevention efforts have worked to change the landscape of social morality in ways that
present new obstacles to HIV prevention. There has been widespread circulation of
social, religious, and public health messages that present infidelity and polygamy as
risky, immoral, backward, and dishonorable. This has heightened men’s personal
motivations and avenues for sexual secrecy (to avoid public scorn and domestic conflict)
and denial that an extramarital relationship increases risk for HIV. Parikh finds that
although all participants in the Ugandan study recognized the health risks of extramarital
relations, they were more concerned about the risk of getting a bad public reputation by
being caught in sexual scandals, rather than the distant, unforeseen effects of contracting
HIV. As a result, men have used strategies such as secrecy to manage sexual relationships
that were considered as socially immoral in order to minimize risks to their reputations.
Secondly, Parikh echoes what other studies (e.g., Brockerhoff and Biddlecom,
1999) have found. Urbanization, education, and changing labor markets have created new
forms of men’s labor-related migration, providing opportunities for extramarital
relationships. It has also been widely recognized that women tend to have much older
husbands, a factor that may increase the probability that their husbands are infected and
weaken their bargaining power within the marriage. In addition, women who are more

economically dependent on their husbands are less empowered to protect themselves

59

against HIV infection (Greig and Koopman, 2003; Parikh, 2007). Thus, the risk posed by
their husbands’ extramarital relationships is exacerbated by the fact that women cannot
negotiate condom use or inquire about their husbands’ extramarital activities. Taking
these arguments into consideration, the finding that women in monogamous marriages in
this Kenyan sample are more likely to die of disease-related causes maybe plausible.
Another surprising result is that women residing in households with secure land
tenure have greater hazards of death in each time period. While the model results cannot
explain the reasons behind this finding, a plausible line of conjecture is as follows. In
Kenya, as in many countries in the region, economic necessity demands that a high
proportion of rural men migrate elsewhere in search of employment. If households have

3, men usually leave the women behind in the countryside with the

secure land tenure
responsibility of tilling the land and caring for homes and children. Generally, the male
migrants remit money to their wives or bring their savings back home to supplement
income from the farm. Husbands returning from the cities (where risk of HIV infection
and opportunities for extramarital relationships are greater) may infect their wives. The
male migrants may take up low-income jobs (e.g. guards) and so what they remit back
home may not be enough to supplement farm income. Parikh, (2007) finds that younger
and poorer married men who migrate tend to leave their wives in rural areas because they
cannot afford to relocate with them. Due to long separation from their husbands, and
insufficient income from the farm and remittances, some women may develop

“residential” sexual relationships in the village (Epstein, 2003; Gillespie and Kadiyala,

2005). Therefore, economic necessity and migration of men create opportunities for

 

3 Tenure security is defined as the household head having a title deed or well-recognized right over a parcel
of land.

60

multiple sexual partnering among women, which is a major risk factor in HIV/AIDS
infection for women. While these social and economic dynamics may explain the
significant relationship between tenure security and the probability of death among
women, these explanations are offered only as conjecture, and may be a useful topic for

further research.

Discrete-time hazard model with time-varying effects

The results shown in Table 2-10 allow the effects of the explanatory variables on
the mortality hazard to vary over time, from one survey-period to another. Hence, the
proportionality assumption is relaxed and the variables permitted to have time-varying
effects.

In the case of men (columnl), education has a significant and negative effect in
the first two time periods (1997-2000 and 2000-2002). Setting the education variable at
the 25th vs. 90th percentile changes the probability of death from 0.0140 to 0.0072 in the
first time period, and from 0.0034 to 0.0020 in the second period. However, in the later
two periods (2002-2004 and 2004-2007), mortality hazard does not vary by education.
Therefore, the effect of education is confined to the earlier periods. Within the 10-year
period, education largely affects early but not late male mortality hazards. This suggests
that education causes mortality hazards among men to be separated initially, and to
converge over time. In the earlier periods, information about HIV/AIDS was less
available. The majority of people had poor, incomplete or inaccurate information about
the probabilities of HIV infection. Education may have been protective by conferring an

advantage in accessing and processing health information including HIV/AIDS

61

Table 2-10 Results of time-varying hazard models of working-age adult mortality in rural

Kenya, 1997-2007

 

 

 

Male Female
Explanatory variables (1) (2)
1997-2000 period (=1) -0.014** -0.044
(0.005) (0.028)
2000-2002 period (=1) -0.021* -0.046
(0.011) (0.031)
2002-2004 period (=1) -0.023** -0.038
(0.010) (0.025)
2004-2007 period (=1) -0.034*"' -0.017
(0.015) (0.010)
Head/spouse (=1) 0.006“ -0.002*
(0.003) (0.001)
Age 0.0003 0.001*
(0.0003) (0.000)
Age squared -2.79e-06 -5.82e-06
(3.29e-06) (3.58e-06)
Education“ 1997-2000 -0.0005** -2.05e—05
(0.0002) (1.7le-04)
Education‘2000-2002 -0.0005* -0.0003
(0.0003) (0.0003)
Education*2002-2004 0.0002 0.0002
(0.0003) (0.0002)
Education*2004-2007 0.0006 -0.0002
(0.0004) (0.0001)
Months at home*l997-2000 —0.0004** 0.0005
(0.0002) (0.0003)
Months at home*2000-2002 -0.0004 0.001
(0.0003) (0.000)
Months at home*2002-2004 4.74e-05 2.93e-05
(2.58e-04) (3. l 8e—04)
Months at home*2004-2007 0.0001 -0.0002
(0.0003) (0.0002)
Monogamous“ 1997-2000 0.0001 0.003
(0.0029) (0.003)
Monogamous*2000-2002 -0.0001 0.005
(0.0050) (0.008)
Monogamous*2002-2004 -0.003 -0.0001
(0.002) (0.0014)
Land tenure“ 1997-2000 -0.0008 0.0033
(0.0018) (0.0055)
Land tenure*2000-2002 0.006 0.003
(0.009) (0.003)
Land tenure*2002-2004 0.002 0.019
(0.004) (0.019)
Land tenure*2004-2007 -0.003 -0.0009
(0.002) (0.0018)
Land owned“ 1997-2000 0.0001 —3.77e-05
(0.0001) (4.57e-05)
Land owned*2000-2002 -0.0002 -0.0002
(0.0003) (0.0002)
Land owned*2002-2004 2.40e-05 -0.0001
(1.65e-04) (0.0001)

 

62

Table 2-10 continued

 

 

Male Female
Explanatory variables
(1) (2)
Land owned*2004-2007 0.0001 8.70e-06
(0.0001) (5.25e-05)
Assets" 1997-2000 -9.52e-06 1.01e-07
(8. l7e-06) (6.03e-07)
Assets*2000-2002 5.82e-07 -3.55e-05**"‘
(2.25e-06) (9.006-06)
Assets*2002-2004 -3.51e-05* -1 .67e-06
(2.00e-05) (3. l 3e-06)
Assets*2004-2007 6.96e-07 -5.62e-06
(1 .906-06) (5 .86e-06)
Distance to fertilizer store” 1997-2000 -0.0005 0.0004
(0.0007) (0.0003)
Distance to fertilizer store*2000-2002 -0.0001 -0.0008
(0.0013) (0.0006)
Distance to fertilizer store*2002-2004 0.0009 0.0002
(0.0006) (0.0004)
Distance to fertilizer store*2004-2007 0.0010* 0.0007*
(0.0006) (0.0004)
Distance to tarmac road“ 1997-2000 0.0000 -0.0007
(0.0007) (0.0009)
Distance to tarmac road‘2000-2002 -0.0035 0.0007
(0.0023) (0.0007)
Distance to tarmac road*2002-2004 -0.0023 -0.0010
(0.0015) (0.0012)
Distance to tarmac road*2004-2007 0.0005 -0.0004
(0.0010) (0.0006)
Observations 1131 l 1 1343

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000, 2002, 2004, and 2007.

Notes: Standard errors are in parentheses; **"‘ indicates 1 percent significance level;
** indicates 5 percent significance level;* indicates 10 percent significance level

campaign. However, as information about the disease has become more available, the
education advantage is eroded and education-mortality gradient is flattened. Therefore,
we see no trend in this dataset that would suggest that mortality is increasingly correlated
with low education as has been found in other studies (de Walque, 2002). This finding
may suggest that the public AIDS awareness campaigns have been effective in delivering

AIDS messages to all regardless of their education levels.

63

The negative effect of months at home on the risk of death among men is limited
to the 1997-2000 period. Although the direction of the effect changes in subsequent
periods, there is no statistically significant relationship. This finding would be consistent
with the notion of the effects of migration on social and sexual networks. Men who
migrate from their rural villages to towns in search of work and so spend extended
periods away from home have opportunities for multiple sexual partnering in areas where
HIV/AIDS prevalence has been higher for a longer time. Therefore, they face a higher
risk of HIV infection and death. Migration of men and associated time spent away from
home may have served as an important initial pathway through which HIV/AIDS was
spread. Patterns of return or circular migration to rural areas facilitated this spread since
rural dwellers may have been socialized by the experiences of the migrants (Brockerhoff
and Biddlecom, 1999). However, in later periods, migration may not be a critical factor
since the disease has now spread throughout the rural areas. Consequently, over time
number of months spent at home has no effect on male mortality, meaning that the hazard
profiles tend to converge.

The effect of household assets on risk of death among men is different over time.
In the earlier two survey-periods, assets have no effect on mortality hazard. However, in
the subsequent period (2002-2004), there is a negative significant effect. Holding other
factors constant, the probability of death for a man at the 25’11 and 75th percentiles of the
assets distribution is 0.0088 and 0.0050, respectively. This negative effect does not
persist for long. It fades away by the last period. Thus, the effect of assets fluctuates over
time, leading to coincident hazard profiles initially, followed by divergent profiles, and

again coincident ones later.

64

The distances of individuals’ homes to tarmac roads and retail input stores are
generally not significantly related to the probability of death except in the 2004-2007
period. These results may imply that the disease is no longer concentrated along the main
highways, trading centres and points of contact with travellers but has permeated
throughout rural areas. Indeed, overtime, the risk of death is found to be higher among
men who reside in relatively remote rural areas.

Over the whole observation period, there is no statistically significant effect of
amount of land owned, land tenure and whether the men reside in monogamous
households or not.

For women, results in column 2 of Table 2-10 show that there is no statistically
significant relationship between the variables and mortality hazard except for household
assets and distance to fertilizer store. The effect of assets on female mortality fluctuates
over time. First, in the 1997-2000 period, assets have no effect on mortality. Then, in the
2000-2002 period, women from relatively wealthier households are less likely to die.
Holding other factors constant, the probability of death for a woman at the 25th and 75"1
percentiles of the assets distribution is 0.0068 and 0.0033, respectively. Again, there is no
observed relationship between assets and risk of death in the last two periods. This
pattern is similar to what we find in the case of men. Men and women in relatively
wealthy households are slightly less likely to die during one of the periods. This
indicates that there is some evidence of a weak but statistically significant shift over time,
suggesting that poverty may be increasingly associated with adult mortality over time.
This pattern is in contrast to earlier findings in the literature that it is wealthier

individuals who were more likely to die. It provides some evidence that the direction of

65

the association between wealth and mortality has shifted slightly over time among men.
Overall, however, the results within the 10-year period do not support the view that
poverty is becoming an increasingly important social pathway contributing to the spread
of HIV/AIDS and working-age adult mortality, particularly among women.

In the earlier periods, the hazard of death among women does not differ by
distance to the fertilizer store. However, over time, women residing in households farther
away from towns where retail input stores are found, are more likely to die. As in the
case of men, being far away from highways and points of contact with travellers/outsiders
is no longer protective. This is because HIV/AIDS has permeated throughout rural areas
and the risk of disease-related deaths in general and AIDS-related deaths in particular

maybe even greater in the remote rural areas.

2.5.4 Conclusion and policy implications

The objectives of the study are to identify the relationships between socio-
economic characteristics and rural adult mortality within the 15-59 year age range, and to
examine whether these relationships have changed over time as the HIV/AIDS disease
has progressed. An accurate evidence-based understanding of these relationships based
on nationwide survey data can usefully guide the design of AIDS response strategies.

Hazard analysis is applied to a lO-year period data of 5,75 5 working-age adults in
rural Kenya. The results show that in general, there is a downward trend in the risk of
working-age adult mortality across the observation period, with the decline being faster
for men than women. This is consistent with decreases in AIDS-related deaths and HIV
prevalence rates in Kenya, which can be attributed to significant progress made in-

HIV/AIDS prevention and treatment programs.

66

A number of findings derive from the time-invariant models. First, while age is
unrelated to male mortality, it has a non-linear effect on women deaths. Second, the
hazard of death is greater for men who are household heads, less educated, and those who
have spent more time away from their homes. Third, the risk of death is greater for
women are not spouses or heads, and those residing in households that are monogamous,
and those with secure land tenure. Fourth, there is no statistically significant relationship
between household assets, distance to tarmac roads and retail input stores and mortality
hazards for both men and women.

Results from the time-varying models indicate that within the rural areas, there
have been changes in the effects of some variables on the hazard of death among men and
women, over time. While more educated men are less likely to die in the first two time-
periods, a lack of relationship between mortality and education in the later periods
implies that the hazard of death is converging among men of all education levels. When
the education variable is set at the 25th vs. 90th percentile, the probability of death
changes from 0.0140 to 0.0072 and from 0.0034 to 0.0020, respectively, for the earlier
two periods. Although this may indicate that the HIV/AIDS information campaigns have
been effective, it is necessary to keep the efforts up. It is particularly important to focus
on prevention information, since new cases of HIV infection are still being reported.

The effect of assets on male and female mortality fluctuates over time, but the
pattern of fluctuations does not indicate a clear trend over the 10-year period. Findings
generally show a lack of relationship between assets and mortality. However, there is a
weak significant shift toward an inverse relationship between asset value and mortality in

some periods. Holding other factors constant, the probability of death at the 25th

67

percentile of the wealth distribution is 88% and 68% for a man and a woman,
respectively, whereas at the 75th percentile of the wealth distribution, the probability is
50% and 33%, respectively. Therefore, the risk of death varies only slightly by household
wealth. This does not support the hypothesis that over time, by predisposing people to
risk of infection, poverty has become an important factor in adult mortality, although it
may certainly be one of the pathways. Hence, there is need to study this issue further and
to reassess the common programmatic response to AIDS which focuses on the provision
of income-earning activities to targeted groups, in the hope that this may reduce risky
behaviors that spread the disease. In particular, expanding the scope of research to take
into account non-poverty risk factors and how they interact with poverty-related factors
may be more instructive for policy and interventions.

Men who reside at home for longer periods have a lower hazard of death in the
1997-2000 period but there is no observed significant relationship in subsequent periods.
This is an indication that the risk of HIV/AIDS is no longer related to mobility and
spending time away from home. Also, there is a shift from a lack of relationship to a
positive relationship between mortality and distance to the fertilizer store for both men
and women. This may be a sign that the risk of death is increasing faster in the relatively
remote localities compared to towns and trading centres within the rural areas. Therefore,
there is need to ensure that the remote rural areas are not overlooked in the provision of
HIV/AIDS prevention and treatment programs.

In the early stages of the HIV/AIDS epidemic wealthier, more educated
individuals, those who were mobile and spent extended periods of time away from home

were more likely to die of HIV/AIDS-related diseases. A comparison between the

68

findings within the 10-year period and the early stages reveals that the relationship has
changed in a number of ways. First, the hazard of death does not vary by wealth status.
Second, the hazard of death is converging among men of all education levels. Third, the
risk and spread of HIV/AIDS is no longer related to mobility and spending time away
from home. Fourth, relative remoteness and isolation from the initial epicenter of
HIV/AIDS is no longer associated with lower HIV infection and hazard of death. All
these changes suggest that over time, mortality risks are converging among groups that
could in the past be clearly distinguished with regard to the likelihood HIV infection and

AIDS-related mortality.

69

Chapter 3

DYNAMICS OF THE IMPACTS OF WORKING-AGE ADULT MORBIDITY AND

MORTALITY ON RURAL FARM HOUSEHOLDS IN KENYA

3.1 Introduction

There has been growing concern about the increase of working-age (WA) adult
morbidity and mortality in rural Africa and its impact on farm households. This rise in
mortality is attributed to the onset and spread of HIV/AIDS. The death of a working-age
adult may affect the social and economic well-being of households through several
mechanisms such as potential loss of labor time, farming experience (knowledge and
skills), and cash income including remittances. As HIV/AIDS has progressed and its
presence become more visible, more research has been undertaken to understand the
impacts of adult death. However, micro-level studies on impacts of adult mortality and
HIV/AIDS on rural households remain limited.

Existing panel data studies examining the impact of disease-related adult
mortality on rural household responses and changes in welfare (e. g. Beegle, 2005;
Yamano and Jayne, 2004; Chapoto and Jayne, 2008) show that the impacts of working-
age mortality depend greatly on the gender and position of the deceased person in the
household; in some cases these impacts are large and significant. However, these studies
generally cover a 2-3 year time frame and are unable to measure the impacts of adult
mortality beyond this relatively short time frame. One exception is a study by Beegle et al
(2006) which uses a 13-year panel of individuals in Kagera, Tanzania to assess how adult

mortality shocks affect both short and long-run consumption growth of surviving

70

household members. This highly interesting study, however, is confined to a particular
district known to be particularly hard-hit by HIV/AIDS, and hence it may not provide a
meaningful indication of the general nationwide impacts of the disease.

In addition, some studies treat afflicted households as a homogenous group with
similar recovery paths after the death shock and therefore the impact measured is for the
average household and is considered to be constant regardless of the timing of the adult
death. However, the full effects of mortality on household behavior and welfare may
occur over many years.

This essay uses data over a 7-year time period to understand the longer-term
effects of death, effects which may be large and different from those experienced during
the period immediately following a death. Therefore, the essay adds to the currently
sparse literature on the long-term effects of adult mortality in several ways. First, I
estimate a model with longitudinal data to control for pre-death differences between
afflicted and non-afflicted households that may bias estimates of the impacts of death.
Second, I examine the temporal pattern of the effects of death in the years before and
after the occurrence of death. This accounts for the likelihood that adult death is not
associated with just a one-time or permanent adjustment in household outcomes. Third, I
explore how the recovery patterns of households vary depending on the gender and
position of the deceased adult. Fourth, this study uses nationwide data and hence provides
estimates that are more indicative of general impacts rather than being confined to areas
known to have particularly high HIV prevalence rates. Therefore, this essay addresses
major knowledge gaps by assessing: the short- and longer-term impacts of adult

mortality; whether household outcomes recover or decline over time; whether the gender

71

and position of the deceased members affect the recovery path; and impacts at the
national level in a country with moderately high HIV prevalence rates by East African
standards. As a result, this study uncovers the dynamics in household welfare and
responses as a result of prime-age death.

The potential persistence of mortality effects beyond the time of death has
important implications for policy makers and development practitioners on current and
future policies and intervention strategies intended to mitigate these effects. The evidence
of large short-term impacts of death from previous studies has led many to argue the
importance of developing appropriate mitigation measures. In hard-hit areas of Africa, if
the impacts of death were found to persist over time, this would motivate even greater
concern for long-term response strategies to assist afflicted households and communities,
and would perhaps have important implications for the design of poverty reduction
strategies.

The next section contains a review of past studies on impacts of HIV/AIDS and
working-age adult mortality on rural households. Section 3.3 outlines the shortcomings of
these studies, which this essay seeks to address. Section 3.4 describes the data while
section 3.5 briefly discusses the progression of HIV/AIDS and the associated economic
costs, and section 3.6 contains the estimation methods. Study results are presented in

section 3.7 and the last section contains the conclusion and policy recommendations.

3.2 Review of literature on impacts of working-age adult death

Research on the economic impact of working-age adult death can be broadly

categorized into two types i.e., macro-economic level studies that estimate impacts on

72

economic growth and development, and microeconomic research that focuses on impacts
at the household level. Most of the early empirical impact studies were at the macro-
level, where simulations were done on GDP or GNP per capita, and their rates of growth
under various levels of projected HIV prevalence rates. These studies depict mixed
conclusions and wide variations in predictions". For example, Cuddington (1993)
employed a single sector, full employment Solow-type growth model to simulate impacts
of HIV/AIDS on the growth of GDP and GDP per capita in Tanzania. The simulation
results show that AIDS reduced the average real GDP growth rate by 0.6-0.9 percent. On
the other hand, Sachs et a1 (2001) estimate that Sub-Saharan Africa faces a 35 percent
loss in GNP due to HIV/AIDS. Most existing estimates of the macroeconomic costs of
AIDS in Africa range between 0.3 and 1.5 percent reduction in the growth rate of GDP
annually. However, as Bell et al (2003) argue, these estimates may greatly underestimate
the impact of HIV/AIDS since they fail to account for the potential impact on human
capital formation and the mechanisms through which knowledge and abilities are
transmitted from one generation to the next. Young (2005) incorporates this detrimental
impact on the human capital accumulation in his model. He tests for the competing
effects of reduced human capital of orphaned children and lowered fertility. He finds that
the fertility effect dominates, with the implication that the AIDS epidemic, could lead to
improved future per capita consumption possibilities of the South Afiican economy.

A growing interest in understanding the household-level impacts of HIV/AIDS
has resulted in microeconomic research that examines how this disease affects the

behavior of rural households and farming systems. Most of the early research involved

 

4 See Cuddington (1993); Cuddington and Hancock (1995); Over (1992); Sachs e_t a_l (2001);

73

qualitative studies and rapid appraisals, and quantitative findings based on small samples.
Evidence from earlier microeconomic studies is at best anecdotal and speculative, where
the impacts of AIDS are hypothesized and conjectured but hardly quantified. However,
studies measuring the impacts of HIV/AIDS on rural households are now emerging, but
evidence is still modest. This section reports some of the existing evidence of the impacts

of adult mortality and HIV/AIDS on various aspects of the rural farm household.

Household composition

Death in a household leads to changes in household size. However, as Beegle
(1997) notes, the death of an adult does not necessarily imply a reduction of household
size by one adult. Instead, such a death introduces changes in household size and
composition or structure, which include in- and out-migration of household members, an
increase in the rate of fostering, and higher rates of remarriage for surviving spouses
(World Bank, 1999; Ntozi et al, 1997). Menon and others (1998) find that the death of a
working-age adult leads to an increase in dependency ratio. The dependency ratio
increased from 1.2 to 1.5 as a result of an adult death from AIDS. Consequently, there
was an increasing number of households where children and the elderly have less
support. Changes in household structure have been shown to be part of households'
coping strategies. For example, fostering spreads the impact of HIV/AIDS over several
households (Bledsoe, 1994). Yamano and Jayne (2004) find that household composition
is affected in different ways depending on the gender and position of the deceased in the
household. Death of a household head or spouse resulted in changes in household size but
the death of other working-age adults is partially compensated for by entry or return of

other members. Chapoto (2006) finds that irrespective of the gender and position of the

74

deceased individual, household size declines by less than one person, implying that
households are somewhat successful in replenishing their household sizes. However,
changes in household size and composition depend on initial household conditions. For
instance, non-poor households are more likely than poor households to restore their size
to pre-death levels, mainly by attracting young boys and girls. Therefore, a household’s
flexibility is a critical component in successfully responding to extreme crises such as
illness and death of an adult. Without such adaptation, either through altered structures or
roles, or through external assistance, families and households may become non-
functioning social and productive units and ultimately dissolve (Hosegood et al, 2004).
Such demographic changes may have profound short- and long-term consequences for
surviving individuals by impacting on a household’s labor availability and allocation.
These impacts begin to emerge when an adult’s labor contribution declines as he falls
sick. The labor of other household members is also diverted to care giving during the
period before death. The death of the sick adult implies permanent loss of one source of
household labor but at the same time frees time and cash that were previously devoted to
care giving. Dynamics in household composition and the attendant changes in intra-
household labor allocation have implications for a farm household’s wealth and asset
base, income sources and crop production, as households change their demand for and
supply of labor. Household composition affects consumption and time allocation or labor
supply decisions, and may or may not affect farm production decisions (Singh et al 1986;
Benjamin, 1992). In addition, composition influences a household’s ability to insure

against risk (Rosenzweig, 1988) and the preferences of its members for consumption and

75

investment (Chiappori, 1988). Edmonds et al (2001) conclude that household
composition is an important component of a household’s response to changes in income.
Wealth and asset base

The loss of human capital leads directly to a loss of financial capital. Households
experience financial loss as a result of medical and funeral expenses and labor diverted
from economically productive activities to care giving. Financial capital effects of adult
death include: liquidation of savings accounts; seeking remittances from family; sale of
stores of value (jewellery, household goods); borrowing from informal sector sources
such as relatives, friends, neighbors, rural cooperatives, rotating and savings club
associations, rural traders and money lenders, pledging of firture crops, and sale of
livestock (Mutangadura et al, 1999; Stokes, 2003; Topouzis, 2000). Most of the
households respond initially by disposing of assets that are reversible (e. g., savings,
remittances and credit) (Topouzis and du Guemy, 1999). However, as indicated by
Stokes (2003), if the effects of mortality are severe enough, households may follow with
the sale of productive assets, a strategy that may be costly in the long-run. Indeed,
evidence from Kenya (Yamano and Jayne, 2004) and Zambia (Chapoto, 2006) shows that
households first attempt to dispose of small animals and other assets with the least impact
on long-term production potential. Households tend to hold onto cattle and productive
farm equipment and possibly sell them only in response to severe cash requirements after
incurring a death in the family but in Zambia, the death of a male head results in a 30%
decline in the value of cattle assets.

There is concern that the above mentioned coping strategies may have even greater

longer-run adverse consequences on households’ future welfare. For instance, it has been

76

noted that farm households rely on remittances and non-farm income to capitalize their
farm operations (Reardon et al, 1995). Unfortunately, as Donovan and others (2003)
indicate, such sources of income are often at risk among AIDS-afflicted households, and
particularly so for those households that were already asset poor and vulnerable to begin
with. Therefore, through loss of remittances and non-farm income, the death of a
household member is likely to tighten cash constraints on agricultural production. This
situation is aggravated by increases in medical and funeral expenses and forgone income
by surviving household members who provide care for the sick. It is also the case that if
these coping strategies are not sustainable or viable in the long-run, we may observe the
following: first, households may be forced to adopt other strategies, implying that the
long-term impacts of death may be different from the short-run effects, and; second, the
capacity of households to recover from adult death shocks even in the long-run, may be
eroded. In general, as fewer households cope or even merely survive, there may be
increased vulnerability to other shocks or households may pursue more damaging
strategies.
Agricultural production

An adult death is associated with at least a temporary loss of labor, labor re-
allocation, potential loss of farming knowledge and skills, and increased medical and
funeral expenses. All of these responses may be hypothesized to affect agriculture
through changes in cropping patterns, land use, area cultivated, input use, farm output
and income. FAO (2003) estimates that seven million agricultural workers have died
from AIDS-related sicknesses since 1985 and that another 16 million of the agricultural

labor force in sub-Saharan Africa could die by 2020. Thus, the epidemic currently has

77

and is expected to have devastating impacts on millions of rural farm households over the
next two decades (UNAIDS, 2004).

Existing research has presented mixed evidence on impacts of mortality on
agricultural households with respect to changes in area cultivated, cropping mix and
patterns, and frequency of weeding. As a result, there is no clear direction on the best
policy responses to adopt in order to mitigate the effects of adult death. For instance,
using data (for 1989-1993 period) from a rapid rural assessment survey for communities
in Zambia and Tanzania, and households in Uganda, Barnett (1994) and Barnett et al
(1995) find mixed evidence. They find little evidence of HIV/AIDS-related impacts on
agricultural production in Tanzania. However, in Uganda they find some discernible
evidence where households, particularly the poorer ones, shift to subsistence crops. In
Zambia, Barnett (1994) reports that farming systems that are most vulnerable to labor
loss are not those greatly affected by AIDS epidemic. Also, other research such as
Barnett and Blaike’s (1992) study in Uganda finds no significant change in agricultural
production. As Barnett et al (1995) argue, the impact of death may vary across
communities and households in accordance with differences in HIV prevalence rates,
population density, existing farming system and the size of the local labor market. Such
factors need to be taken into account in making comparisons on such study findings.

Increasing dependency ratio and lower labor availability are hypothesized to
cause households to shift from labor-intensive crops to less labor-intensive ones (which
may be less nutritious or productive). It has been reported that the shift in area cultivated
from maize to root tubers in most parts of Africa may be an indication of labor shortages

and hence a need to shift to less labor-intensive crop systems (Barnett, 1994; F A0, 1995;

78

FAO, 2004). Other land use changes have been observed as a response to HIV/AIDS. In
Cote d’Ivoire, households facing labor shortages have resulted to either hiring more labor
or allocating some land for crop sharing (FAO, 1997). However, Barnett (1994) finds a
reluctance to rent out land in Uganda and particularly among widows. This may be
attributed to a lack of guarantee that the land owners can maintain the use rights to the
land.

Using panel data from Kagera in Tanzania, Beegle (1997, 2005), finds that
although adult male death causes temporary scaling-back of cash cropping and a fall in
wage income, afflicted households do not shift towards subsistence crops. She argues that
this result is consistent with the high labor-land ratios in Kagera region. Beegle’s study is
based on a short panel (1991-1993) and is therefore unable to assess the longer time
impacts. In addition, this study does not examine changes in crop output and liquidation
of assets. Yamano and Jayne (2004) also use panel data from Kenya and find that
afflicted households experienced a decline in farm output compared to non-afflicted ones.
However, the effect of death on crop production is sensitive to gender and position of the
deceased in the household and the households’ initial asset levels. Results show that
households suffering death of the household head or spouse were largely unable to
replace labor lost through death, whereas households suffering death of another adult
were able to attract new household members. They also find that relatively better off
households experienced a greater decline in family labor, especially among adult
members. However, they find no evidence of significant losses in cultivated land and net

crop output among these households.

79

Chapoto (2006) finds that the effects of prime—age adult mortality on farm
production in Zambia are sensitive to the gender and position in the household of the
deceased member. The death of a male head causes households to cut back the area
cultivated, possibly due to the inability of households to fully replace deceased members.
The area under cereals and other food crops declines as a result of death, with the most
severe effects observed in the case of a male head death and there is no evidence that
afflicted households are switching to labor-saving crops or that initial poverty
exacerbates the impact of mortality on cultivated land. In addition, afflicted households in
Zambia don’t suffer declines in value of crop output except in the case of poor
households experiencing a male head death. Generally, evidence from Tanzania, Rwanda,
Kenya and Zambia suggests that better-off households may be able attract new members
or to hire additional workers to partly compensate for the loss of adult member (Beegle,
1997; Donovan et al, 2003; Yamano and Jayne, 2004; Chapoto, 2006). These results raise
questions about the view that loss of labor poses the major constraint on crop production.
This issue may be addressed in part, by accounting for differences in conditioning factors
(such as labor-land ratios) that vary across production regions. It has also been
hypothesized that households may withdraw from time-consuming marketing activities
and revert to subsistence production. As a result of the inability to add value to produce,
households are likely to experience decreases in crop income flows.

Household diversification strategies

Not all costs of illness are borne by the person who is sick. In a household, ill

health of an individual is likely to evoke resource adjustments by other members of the

household or surviving members in case of death. Adjustments in participation across

80

income earning activities may be expected in response to an illness shock. Morbidity
lowers the sick person’s productivity and often entails care time spent by another
household member who may otherwise be engaged in productive work. In Indonesia
there is evidence for reallocation of time from market activities and school to household
care activities of adult women and older children when infants fall sick (Pitt and
Rosenzweig, 1990). Kochar (1995) finds that market labor supply (wage income) is
responsive to illness of a household member and concludes that farm households may be
more vulnerable to demographic shocks such as sickness and death than to crop income
shocks. The effects of adult mortality on agricultural production and yields will depend
on several conditioning factors such as the agricultural system, labor/land ratios, labor
requirements of the cropping system, farm sizes, and land tenure, which vary across
production regions.

Apart from agricultural production, households pursue other income-generating
activities either all year round or on a seasonal basis. Many farming households
supplement their income with non-farm activities. There is evidence that non-farm
earnings are an important source of household income in Africa (Reardon, 1997; Yamano
and Jayne, 2004). However, these earnings are sometimes seasonal because of
competition for labor from the agricultural sector during the peak seasons. Extra labor
demands (e. g. due to care giving) during critical times of year such as ploughing,
planting, or weeding can have disastrous impacts on farm and non-farm incomes.
Studies have shown that households may be protected from idiosyncratic crop shocks
when there are well functioning credit and labor markets, through compensating increases

in non-farm labor supply (Eswaran and Kotwal, 1989; Morduch, 1990; Kochar, 1995).

81

This raises the possibility that households whose crop production and/or income may be
adversely affected by morbidity and death may use market labor supply as an avenue to
cope with effects of death. The importance of off-farm/labor income for Kenyan rural
farm households increases the likelihood of such a responses. Incomes from both farm-
and off-farm sources may decline as HIV infections and AIDS deaths are
disproportionately concentrated in the most productive age groups (15-59). There is
evidence that households and particularly female-headed ones are responding to
HIV/AIDS by turning to off-farm income generating opportunities (FAO, 1995). The
studies by FAO find that: (i) households that did not engage in off-farm activities did not
do so because they lacked capital and (ii) households that were already engaged in some
off-farm activities were better cushioned from effects of HIV/AIDS. This implies that
diversification over income sources may be important in mitigating the effects of adult
mortality. The study by Yamano and Jayne (2004) reports that death of a male head of
household results in large reductions in off-farm income. Chapoto (2006) reports mixed
but statistically insignificant impacts on off-farm income for all cases of death by gender
and position in the household.

It is also often assumed in some of the theoretical literature that the negative
effects of adult mortality on agricultural production, as well as losses in non-farm or self-
employment income from the deceased, will force many afflicted households into
poverty. However, there are mixed findings. There is evidence that the effects of
working-age adult mortality on agricultural production and incomes tend to be more

severe on households that were already relatively poor (Drimie, 2002). Chapoto (2006)

 

5 Nearly 30 percent of farm household income in 1997 and 2000 was derived from off-farm activities
(Yamano and Jayne, 2004).

82

finds that the area under cereals declines most among wealthier households incurring a
male head death while the impact does not differ between poor and non-poor households
in case of a female head/spouse death. In addition, Lundberg et al (2000) show that
among afflicted households, poorer households are less likely to receive assistance,

financial and otherwise, from social networks.

3.3 Limitations of reviewed literature

The preceding review of literature on impact of adult mortality suggests some
limitations in research on this topic. First, few studies use nationwide representative panel
data. For instance, although case study panel data sets like the Kagera demographic and
health survey have generated estimates of the impact of adult mortality and AIDS in
particularly hard-hit areas, these studies lack representativeness and could produce
overestimates of the impacts of the disease if they are generalized more broadly to the
region or continent. Hence, their results cannot provide a comprehensive understanding
of the national level impacts.

Second, most prior micro studies are unable to estimate the enduring or prolonged
impact of mortality, starting from the onset of illness, death and after death short- and
longer-term effects. The few existing longitudinal studies use data from the early years of
the epidemic and available evidence has typically been from panels separated by only 2-3
years. In addition, some studies focus on impacts on households regardless of the timing
of death. Therefore, only short-run effects are measured. Long run impacts, dynamics of
recovery process, whether recovery occurs and under what conditions, are still unknown.
Put differently, it is likely that less is known about the long-term impacts of adult

mortality and HIV/AIDS epidemic than is generally acknowledged.

83

Some longitudinal research does not account for attrition, which may
underestimate incidence and impact of adult mortality. Bias due to attrition may be
particularly severe where there is considerable mobility arising from migration between

rural and urban areas and/or dissolution of households as a result of death.

3.4 Data and attrition

The study uses a four-year nationwide panel data set of Kenyan rural households
collected in 1997, 2000, 2002 and 2004. The 2002 survey does not contain information
on the household outcomes that are of interest in this essay. However, it contains valuable
information on mortality. Therefore, changes in household outcomes are described for the
1997, 2000 and 2004 surveys while mortality information is derived from all surveys.

As is typical of any longitudinal study, the data used in this study suffer from
attrition, with 6.2 percent of the initial 1,500 households dropping out between 1997 and
2000 while 9.3 percent were not re-interviewed in the 2004 survey. Over the entire 7-
year period the attrition rate is 11.7 percent. Tables 3-1, 3-2, 3-3 and 3-4 present
additional information on attrition by survey wave and also show the relationship

between attrition, dissolution", disease-related mortality and household size.

 

6 . . . . . .
Dissolution means that remaining members dispersed, or in the case of one-person households,
the household no longer existed.

84

Table 3-1 Relationship between disease-related working-age adult mortality, household
attrition dissolution, and initial household size for the 1997-2000 period

 

 

 

 

Household Households in Households Households Households
size in 1997 survey attriting in 2000 attriting due to apparent incurring a
1997 dissolution in 2000 working-age
death
Number Number Number (%)a Number (%)a Number (%)
1 l3 0 (0.00) 0 (0.00) 0 (0.00)
2 43 3 (6.97) 0 (0.00) 1 (2.33)
3 86 7 (8.14) 0 (0.00) 6 (6.98)
4 158 10 (6.33) 3 (1.90) 8 (5.06)
5 197 15 (7.61) 3 (1.52) 17 (8.63)
6 249 14 (5.62) l (0.40) 9 (3.61)
7 234 14 (5.98) 0 (0.00) 17 (7.27)
8 172 11 (6.40) 0 (0.00) 8 (4.65)
9 118 8 (6.78) 1 (0.80) 6 (5.08)
>=10 230 11 (4.78) 1 (0.40) 18 (7.83)
Total 1500 93 (6.2) 9 (0.60) 90 (6.00)

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997 and 2000.

Note: a The percentages of the total number of households sampled in the initial period are in parentheses

Table 3-2 Relationship between disease-related working-age adult mortality, household
attrition and initial household size for the 2000-2002 period

 

 

 

 

Household Households in Households attriting in Households incurring a
size in 2000 2000 survey 2002b (%) working-age death
Number Number Number (%)a Number (%)
l 18 7 (38.89) 0 (0.00)
2 48 12 (25.00) 1 (2.08)
3 80 12 (15.00) 3 (3.75)
4 152 31 (20.39) 8 (5.26)
5 200 30 ( 15.00) 6 (3.00)
6 220 34 (15.45) 5 (2.27)
7 216 36 (16.67) 3 (1.39)
8 151 34 (22.52) 0 (0.00)
9 123 4 (3.25) 2 (1.63)
>=10 238 26 (10.92) 6 (2.52)
Total 1446 226 (15.63) 34 (2.35)

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 2000 and 2002.

a The percentages of the total number of households sampled in the initial period are shown in
arentheses
The percentage of households attriting due to dissolution could not be computed because there
was no information on the reasons for household attrition in the 2002 survey.

Note:

85

Table 3-3 Relationship between disease-related working-age adult mortality, household
attrition, dissolution and initial household size for the 2002-2004 period

 

 

 

 

Household Households in Households Households Households
size in 2002 survey attriting in 2004 attriting due to incurring a
2002 apparent dissolution working-age death
in 2004
Number Number Number (%)‘I Number (%) Number (%)
l 19 3 (15.79) 1 (5.20) 0 (0.00)
2 46 7 (15.22) 1 (2.17) l (2.17)
3 97 27 (27.84) 1 (1.03) 0 (0.00)
4 145 15 (10.34) 0 (0.00) 9 (6.21)
5 208 35 (16.83) 3 (1.44) 5 (2.40)
6 210 35 (16.67) 1 (0.48) 6 (2.86)
7 193 30 (15.54) 0 (0.00) 8 (4.15)
8 I61 21 (13.04) 0(0.00) 5(3.ll)
9 122 11 (9.02) 0 (0.00) 3 (2.46)
>=lO 219 20 (9.13) I (0.46) 12 (5.48)
Total 1420 204 (14.37) 8 (0.56) 49 (3.45)

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 2002 and 2004.

Note: ’ The percentages of the total number of households sampled in the initial period are shown in
parentheses

Table 3-4 Relationship between disease-related working-age adult mortality, household
attrition, dissolution and initial household size for the 1997-2004 period

 

 

 

 

Household Households in Households Households Households
size in 1997 1997 survey attriting in 2004 attriting due to apparent incurring a
dissolution in 2004 worlgg-age death
Number Number Number (%)a Number (%) Number (%)
1 13 l (7.69) 0 (0.00) 0(0.00)
2 43 7(16.28) 0 (0.00) 4(9.30)
3 86 9 (10.46) 3 (3.48) 7(8.14)
4 158 21(13.29) 3 (1.90) 20(12.66)
5 197 l9(9.64) 3 (1.52) 25(12.69)
6 249 28(11.24) 5 (2.01) 22(8.84)
7 234 21(8.97) 4 (1.71) 34(14.53)
8 172 ll(6.40) 0 (0.00) l6(9.30)
9 1 l8 7(5.93) 0 (0.00) 12(10.17)
>=10 230 15(6.52) l (0.43) 3103.48)
Total 1500 139(9.27) 19 (1.27) l7l(l 1.40)

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring and Policy
Analysis Household Surveys in 1997, 2000, 2002 and 2004

Note: a The percentages of the total number of households sampled in the initial period are shown in
parentheses

86

Between 1997 and 2000, 20 percent of the households had 4 or fewer members
(i.e., are small households) and these households accounted for 22% of attrited
households, while between 2000 and 2002, such households accounted for 20% and 27%
of total and attrited households, respectively. Similarly, in the 2002-2004 period, 22% of
the households had 4 or fewer members and they accounted for 25% of attrited
households.

The reasons for attrition include: the household moved away or dissolved, no one
was at home to be interviewed, or household refused to participate in the survey.
Household dissolution is likely to be associated with mortality. For example, Urassa et al
(2001) found that in Kisesa, Tanzania, 42.5 percent of households had dissolved after the
death of a male household head. With respect to mortality, our data show that these
smaller households represent 17%, 35%, and 20% of households experiencing adult
mortality in the periods 1997-2000, 2000-2002, and 2002-2004, respectively. Thus,
larger households are more likely to experience working-age adult mortality. As regards,
dissolution, the smaller households represent 33% of dissolved households between 1997
and 2000, and 38% between 2002 and 2004 surveys. Therefore, dissolution is a more
important reason for attrition among smaller households than larger households. In
general, household attrition may lead to attrition bias, an issue which is examined further
in subsequent sections.

The surveys contain information on household demographics, socio-economic
characteristics of household members, agricultural, livestock and off-farm activities,

household assets as well as child and adult mortality. Possible disease-related mortality of

87

working-age adults (15-59 age-ranges) is tracked between 1997 and 2004, and is related
to various household outcomes to assess the impact of death on households. Table 3-5
shows number of households afflicted by working-age adult morbidity and mortality by
survey period. Between 1997 and 2000 surveys, 93 working-age adults passed away due
to disease-related causes, with 89 households having at least one death. In the 2000-2002
period, there were 37 cases of death and none of the afflicted households had multiple
cases of death. Between 2002 and 2004, there were 65 cases of working-age adult deaths
in 59 households, 6 of which had 2 cases of working-age adult mortality. Additional
information on the number of individuals dying from disease and the adult mortality rates

is presented in Table 2-1 (Chapter 1).

88

.8888? 8:5 .8 838888 88 82:88 8.:
:82: 88880885 :04” .888 888 88» 838. :28 8.: .889 8:3 :8 80:: 88 5888:. :85 88:88: 0
.8388 888 88:. 85 .8 8.88 8 95 >8 5 :88 82.805 88:88: a 8:: 888 88... a
.888 888 85 £58» 88 88— 88 E 88:8 883 85 8:3 88:88: 85 82:3 8:3 88888.: 8:88
85 .8: 8:8 838 :88 .8 was—Ev: 85 8 88:88: 85 E 883 8:3 88888.: 9 888.. 888 25 E 88:88 o:._. a 88 Z

38 Ea 88 .88 .82 5 3:8 2288: 2822
88m 8.8 8:382): _8_8:w< 3883:: 88m 8w_:o_2\c:o:mz.v 888:— 8:8on 89.8w

 

 

 

 

om N. w w ooom
vm S 2 b 33
A: a S 3
89A
:88 888: 88» N :88 8.8.8: 88» _ :88 8.8.8: 88» N :88 88.8: :8» _ 88m
82:: 88: 2.8-8: :8: 88:88: 82:: 88: 8.8.. :23 88:88:

 

 

383.2: .33 888.33

 

 

 

88

a E _E No 8 mm 8 -82

Sam

m mm 2.. Q 2 E .88

meow

o 2.. E N. n v .88

ooom

m on em mm _ _ mm -32
E 8 5 :3 E 5 E

:88 8288

:88» 8:8 888 :88 :88 :88 88888: :88 hotsm
88:88 E a E 8:88 28¢. 88 228.: 8:8 888 8:88 8.28.: 88: 888
:88 m :8, 88:88 :85 88— 8 :23 8:82: 8:82: a 8:82: a waEsoE
88:88: 88:88: 88:88: 88:88: 88888: 88:88: 88:88:

 

3:38.: .88 nag—2.83 88.9.8885

 

«8:88 888 a: 5:888 88 38888 :28 88-8383 wagon: 88:88: .8 8:82 m-m 28H

89

3.5 Disease progression and the economic impacts of AIDS-related
mortality

There are multiple channels by which by disease-related adult mortality may
affect household outcomes such as agricultural production, assets and income. Such
channels have often been studied using the sustainable livelihoods framework (Gillespie,
2003) or its variants that view the household as a system of interactions between its
supply of labor to farm production and off-farm activities and the return flow to the
household of food and cash to sustain their livelihoods (FAO, 1995; Chapoto, 2006). As
mentioned earlier, disease-related adult mortality in Africa has increased significantly
with the advent of HIV/AIDS. In addition, it has been recognized that HIV/AIDS is a
long-wave crisis (Barnett and Whiteside, 2002), suggesting that it has long-term effects.
As a result, attempts to measure the economic impacts of disease-related mortality may
benefit from the recognition that these impacts are expected to be different over time
since there is a progression of stages at which afflicted households are likely to suffer.

First, the onset of AIDS symptoms leads to chronic illness which is characterized
by increasingly frequent bouts of sickness as the disease progresses. This long-term
illness is associated with diminished labor capacity and lost earnings of the sick member
while the families hit by AIDS have to draw down savings and sell assets to meet
treatment expenses. In addition, other household members may have to take time off
work or withdraw from school to provide care. Often, the immediate spending on
treatment can be large and the economic costs of illness may increase over time as the
disease becomes more severe. Overall, this reduces the financial resources of the

household and may impair a household’s future economic prospects. In response to these

90

effects, households adopt ex-post illness (or ex-ante death) strategies that influence the
behavior of their members concerning many decisions such as farm production and
schooling. For example, Evans and Miguel (2007) show that there is a decrease in child
school participation before the death of a parent, presumably due to pre-death morbidity.

Second, following this period of illness is the death event, which is associated
with large funeral expenses which are often a heavy burden on the family budget. Funeral
costs appear to be even higher than medical expenses in some settings (FAO, 1995). The
death of an individual constitutes a permanent loss of labor (for agricultural, off-farm and
home-related activities) and related income, but at the same time frees time and cash that
were previously devoted to care giving (White and Robinson, 2000). In addition, there is
a loss of acquired human capital investments, management and caring skills, as well as
farming experience, knowledge and skills. The loss of labor, human and financial capital
implies a loss in productive resources. Assets and savings may be depleted as households
use them to meet their increased needs.

Third, a household enters the post-death period in which it may either recover
from the impacts of death, remain at a lower constant level of welfare, or decline further
into poverty. Death effects might either compound over time if a households financial
situation does not recover quickly afier the death of the sick member or perhaps diminish
if effective coping mechanisms emerge. In other words, time matters when measuring
the impacts of AIDS-related illness and death.

Therefore, in order to capture the mortality dynamics (ex-ante and ex-post
effects), we need to consider the period ranging from before death to several years after

death. However, most studies that attempt to measure the impacts of death tend to

91

capture the one-time, average and permanent effects rather than the dynamic effects. An
exception is Evans and Miguel (2007) who consider both parental pre-death and post-
death effects on child school participation.

This temporal pattern of the impacts of death is analogous to that observed in
empirical studies on the effects of job displacement on wages and earnings (Jacobson et
al, 1993). Evidence from these studies shows that workers’ earnings begin to decline
significantly in years prior to job loss, drop discontinuously when they actually leave
their jobs, and then recover to varying degrees in the years following job displacement. In
this literature, the time pattern of earnings losses is therefore seen to occur mainly along
three dimensions namely, the rate at which earnings “dip” in the period before job loss,
the size of the “drop” that occurs at the time of the job loss, and the rate of “recovery” in
the period following job displacement (Jacobson et al, 1993). The dip is usually referred
to as Ashenfelter’s dip7. A similar pattern may characterize how smallholder farm
households’ welfare responds to death shocks. To capture the death dynamics over time,
this essay follows the dip-drop-recovery approach in specifying a statistical model. This
type of model, combined with relatively long panel data, provides a more comprehensive
assessment of the temporal pattern of adult death dynamics than has previously been
possible.

Using a nationally representative panel dataset from Kenya, this study examines

the dynamic effects of adult mortality on household composition, cultivated land (total

 

7 In the context of evaluating training programs, Ashenfelter (1978) observed that the mean earnings of
participants in government training programs decline in the period prior to program entry or participation.
This drop in earnings of trainees relative to the comparison group has become known in the economics
literature as the “Ashenfelter’s dip” or the “pie-program dip” and it has subsequently been found to be a
feature of virtually all training and adult education programs. Ashenfelter identified the bias for inferring
long-term impacts of training that can arise when there is a pre-program earnings dip.

92

and by crop types), value of crop output, household asset base (total assets, cattle, small
animals, farm equipment and other assets) and income (i.e.; total, off-farm, crop and
livestock). This is perhaps the first study to have measured the longer-run impacts of
working-age adult mortality using panel data based on national household surveys, which
cover areas of diverse HIV/AIDS prevalence rates.

The preceding discussion provides some insight into the potential pathways by
which adult morbidity and mortality may affect household outcomes over time. Based on
this, I consider four hypotheses. The first hypothesis is that there are significant
morbidity impacts on household outcomes because negative effects, particularly financial
stress, are usually observed in the period of illness preceding death. The second
hypothesis is that there are large, negative and significant mortality impacts during the
year of death as households’ financial resources decline further. Third, a priori, the
direction of the post-death impacts cannot be determined because they depend on how
mortality affects the resources of the household and how effective the coping
mechanisms are. The fourth hypothesis is that working-age adult death may have
differential effects on the dip-drop-recovery pattern depending on the deceased’s gender
and position in the household. For instance, to the extent that the contribution of a male
head to household resources is more important than that of all other working-age adults,

we may expect male head mortality to have large effects on household outcomes.

93

3.6 Empirical analysis of the dynamics of mortality impacts
3.6.1 Econometric model

To measure the impacts of adult mortality, I use an econometric approach similar
to that used by Jacobson et a1 (1993) to examine the earnings losses of displaced workers.
The statistical model used by these authors is borrowed from the program evaluation
literature, such as that used in studies which evaluate the earnings impact of public-sector
training programs. I use panel data on households that are afflicted by death within the
survey period and a comparison group of households that are not afflicted at any time
during this period, thus exploiting a counterfactual strategy that takes into account the
impacts of death across both time and households.

The effects of adult mortality on household outcomes over time can be depicted
as in Figure 3-1, which shows a very simple representation of the dynamics of adult
mortality over time. This representation can be modeled as in equation (3.1).

Yit =a+flEAi+6WDit+Xul+YRty+5it (3.1)
where Yit denotes an outcome, such as value of farm output or off-farm income for
household 1' at year of survey t; EA stands for ever afflicted and is a dummy variable
equal to 1 if a household experienced a prime-age death at any time between 1997 and
2004 (i.e. the “treatment group”), and 0 otherwise; W0 is a single binary variable equal
to 1 if household i experienced a working-age adult death between the previous survey

and the survey year t, and 0 otherwise; X it is a vector of time-invariant and time-varying

household and community characteristics that influence household outcomes but are not

94

Figure 3-1 Modeling the impact of working-age adult mortality on household outcomes
over time

________________ Non-afflicted

 

Household outcome

e Afflicted

 

 

t = 0 Years since death

Figure 3-2 Modeling the impact of working-age adult mortality with binary variables for
years before and after death

__________________________ Non-afflicted

— Afflicted

Household outcome

 

 

 

t = 0 Years since death

95

themselves affected by deaths; YR, is a vector of year dummy variables that take a value

of one in year t for t=2000 or 2004. These variables control for secular economic

conditions that may cause changes in Yit regardless of whether households are afflicted or

not; and an is an error term.
The coefficient ,6 in equation (3.1) is a measure of any pre-death persistent
differences between afflicted and non-afflicted households. In Figures 3-1 and 3-2, ,8 is

shown as negative but it can also take on positive values. The parameteré' captures an
average effect of adult mortality. The approach used in this study to estimate the impacts
of death is cast in the spirit of the difference-in-difference estimator by using a
counterfactual approach to capture impacts of death across both time and households.
The difference-in-difference method first considers the pre-death and post-death
dimension. Then, because not all households are afflicted by working-age mortality, it
accounts for this by considering the “with dea ” (treatment group) and “without death”
(control group) dimension. Therefore, it takes into account differences across households

both before and after death. While the coefficient ,8 accounts for any pre-death persistent

differences between afflicted and non-afflicted households, 6 captures the after death
differences between these two categories of households. Therefore, with longitudinal
data, the approach used here controls for time-invariant household characteristics just as
in the difference-in—difference method.

The 6 parameter imposes a constant effect of adult death for all households and

regardless of when the adult died. However, an adult death is not necessarily associated

 

8 A priori, we cannot think of many household-level variables in X, that are unlikely to change as a result
of prime-age death. However, X, may contain variables like age and education of the head, and community
characteristics that are likely to remain unaffected by mortality in a particular household, such as distance
to the market.

96

with just a one-time or permanent adjustment in household outcomes. This model can be
extended to examine if the estimated mortality impacts are different over time and
whether they tend to persist. This is done by including a series of binary variables that
indicate the number of years between death and the date of survey. In addition, in the
case of higher mortality due to AIDS-related illness, deaths are likely to be preceded by a
potentially severe and lengthy episode of illness. Therefore, we include variables that
account for the fact that household outcomes may begin to deteriorate prior to death as
households deal with illness of their members. Therefore, the impacts of adult mortality

can be modeled as in Figure 3-2 and equation (3.2):

s=7 k=2
Y,-, = a + flEAi + Xuxi + Z Dposti"; 6S + Z Dpreg 6k + YR,y + 8,, (3.2)
s=0 k=l

Figure 3-2 is also a simple representation of the dynamics of adult mortality over
time but unlike Figure 3-1, it accounts for: (i) the ‘dip’ that may be expected before death

as depicted for example, by 6, and (92 for one and two years before death, respectively;
(ii) distinguishes between the ‘drop’ during the year of death (as measured by 60) and the

‘recovery’ that takes place in the post-death period (for example, as measured by 5, for

one year after death). The post-death pattern shown here is for illustrative purposes.
Although we don’t expect the post-death effects to be constant, households may or may
not recover from the death shock. Whether these post-death effects decay or persist is an
empirical question.

The representation in Figure 3-2 can be modeled as in equation (3.2) where
DPostl-St is equal to 1 if in the year of the survey (t), household i experienced a death 5

years earlier (i.e. 5 years have elapsed since a household experienced death) and so there

97

are multiple DPost binary variables for various years after death. The coefficients on
these variables will reflect the persistence of the effects of death over time, for up to
seven years after death. These variables account for the fact that the impacts of death in
the post-death period are not necessarily constant and may not decline or grow linearly
over time during this post-death stage.

The variable Dprei]; is equal to 1 if household i experienced death k years after the

year of the survey for both 1997 and 2000 surveys. This vector consists of two binary

variables since k takes the value of l or 2 for deaths occurring one and two years after the
survey, respectivelyg. This means that Dprel! is equal to 1 if death occurred in 1998 for

the 1997 observations and equal to 1 if death occurred in 2001 for the 2000 data, and zero
otherwise. These variables capture the fact that household outcomes may begin to be
affected during the period preceding death.

The specification in (3.2) investigates the dynamics of the effects before and after
death. It allows us to fully capture the dip-drop-recovery temporal pattern of the

dynamics of death through Dpre and DPost binary variables. Hence, we can estimate the

pre- and post—death impacts, and the extent to which these impacts persist over time.
Previous studies on the impacts of adult mortality show that these impacts depend
on the gender and position of the deceased member (Yamano and Jayne, 2004; Chapoto
and Jayne, 2008). It has also been argued that the impact of death on a household may be
greater if the deceased is a male head since he is an influential member of a household

and is more likely to have access to greater financial resources and social capital. To

 

9 AIDS deaths are typically preceded by 4 to 17 months of illness and so negative effects might be expected
up to two years before an adult death due to AIDS-related morbidity (Evans and Miguel, 2004).

98

examine any differential effects, I interact each of the Dpre and DPost variables with a

binary variable (MH) equal to 1 if the deceased person is a male head and zero,

otherwise. The analysis that looks at possible differential effects is represented as:

5:7 8:7 k=2
Yit = or + BEAi + Xitk+ Zleostist + ZDpost?t * MH 65 + Z Dprei’: 9k +
5:0 5:0 k=l (3.3)

k=2
2 Dpreil: 6k * MH + YRty+eit
k=l

3.6.2 Model variables

The vector of dependent variables consists of household outcomes namely: (i)
household composition which consists of household size, number of adult men, adult
women, boys (6-14 years) and girls (6-14 years); (ii) farm production variables, including
total cultivated land, cultivated land by crop types (cereals, roots and tubers, high-value
crops), gross value of crop production and gross value of crop production per acre; (iii)
total asset values and also represented by various components: farm equipment, small
animals, cattle, and other assets, and; (iv) income (total, crop, livestock and off-farm).
Although household composition is not a welfare measure, results on the changes in the
composition provide a useful understanding of how households respond to working-age
mortality and may aid in the interpretation of the effects of mortality on farm production
(Yamano and Jayne, 2004). In particular, there is a widely held view that the death of an
adult adversely affects supply of labor in a household. However, as Yamano and Jayne
(2004) and Chapoto (2006) find, the effect of an adult death on a household’s labor
supply is likely to depend on the ability of a household to attract new family members or

even send away others.

99

As explained above, the variable Post consists of a vector of time-since-death
binary variables which indicate the number of years that have elapsed since a household
experienced an adult death. These variables are also interacted with a male head dummy
variable in order to examine the effect of gender and position on the persistence of the
impacts of death.

The vector X it consists of locational and market access variables namely district

dummy variables, agro-ecological zone (AEZ) dummy variables and the distances to the
tarmac road and fertilizer store. The district dummy variables capture geographical or
locational differences and even farm characteristics such as soil types and land quality.
The distance variables are used here as measures of the effective price that farm
households receive (Strauss and Thomas, 1998), the remoteness of an area, the quality of
road infrastructure and thus the associated transport costs. Therefore, we expect that the
longer the distance the more expensive inputs costs become and so the less the area
cultivated. Also the longer the distance, the less likely that people will participate in off-
farm activities because the associated high transportation costs reduce the net benefits
from these activities.

Other variables include the gender, education, age and age squared of the

household head, which account for a household’s life cycle stages and age structure.

3.6.3 Econometric concern: household attrition

The analysis of mortality impacts on households may be confounded by attrition
bias which may lead to biased estimates. Our panel data exhibits an attrition rate of 11.7
percent over the entire 7-year period. Attrition raises a problem of selection bias if it is

not random. To examine potential bias, I compare the means of variables measured in

100

1997 for households that were re-interviewed vs. those that attrited. The results presented
in Table 3-6 indicate that there are statistically significant differences in means of
variables between households that attrited and those that were re-interviewed.
Households that were re-interviewed in either 2000 or 2004 exhibit the following
characteristics: they have older and more educated heads, larger household sizes with
more working-age and senior adults but fewer children in the 6-14 age-ranges. The last
four rows of Table 3-6 show that households incurring a death or morbidity in 1997-2000
or 2000-2002 were less likely to attrite out of the 2004 survey.

The systematic differences between households that were re-interviewed and
those that were not may lead to biased results. The inverse probability weighting (IPW)
method is used to correct for attrition bias. The IPW method assumes that the probability
of being re-interviewed (non-attrition) as a function of observable information is the same
as the probability of being re-interviewed as a function of observables, plus
unobservables that are only observed for non-attrited observations (Wooldridge, 2002).
Thus, the IPW method works well if observations on observed variables are strong
predictors of non-attrition and if observations on unobserved variables are not strong
predictors of non-attrition.

Following Yamano and Jayne (2005), the re-interview model can be written as:

P(Rit =1) = f(HC1997, ETitaPD) (3.4)
where Rn is equal to 1 if a household i was re-interviewed at time t, conditional on the

household being interviewed in the previous period, and zero otherwise; HC1997 is a set of

household characteristics in the 1997 survey; ET“ is a set of enumeration team dummies;

and PD is a set of provincial dummies. All of the variables are observable even for

101

households that were not re-interviewed after the 1997 survey. Three to six enumeration

teams conducted the surveys, each headed by a supervisor who was authorized to decide.

102

Scam “.5
.38 .33 doow .32 E mag-5m 22.8.5: mix—£3. .828 E... wctoxzo—z 223.310.... .063th 88m gwmnsgonohazv 883m... 8......on "33.5

 

 

103

8... t... Room-88. b.8385 own-wise were... 8.2.8:... .o .x.
8... SN scam-88. 58.. 88.8.8.8... a 8.58. 8.2.82... 8 as.
8... «2 883-8.. 8.2.88 88-8.82» 352.. 8.2.83. .8 .x.
mm... 2% 8.8-8... 58.. own-mafia a wagon. 8.2.82... 8 as
8... 8... ac... m5. E... o. .. a: an... .5... Boa 85.38. 3 8.885
.2 8.. 8.... 5... mm...- Nm...- m. ... 2.... .8... .82 8:5. 9 8:8...
om... 8.... ”a. 8.... Ed (mm... 8.. 2.. .5.. .88 83828. o. 88......
on..- R..- mm... 8... R... m... on.» RR .88... 8.. 3.8.2.2....
8.... 8.2. 83. . 8...... .2 8mm .38. 28... £8. .5.... .88.. .o 8:5
8...- 8...- 9... m3 2.. on... 9... N3- 283 8.88.8 .82.»... 8.28...
S. .- ...8...- mm... 3... .m... 2 ... a... E... .28.... 8.88.8 .85... 0...:
S... h . ... «N. 8.. «3 <3... 3.. 8.... e882. 2.... 85.888.
8....- ...... E. a... ....m <2... an a... .825... E-.. 8.8 88......
8.... 8.....- E... 9.... 8...- 8.... 3.... 3... .85.... $8... 88......
«a... so... .2... t... .2 +........ .2. : ... cones... $88 82.8 28......
vom- én... an. 3.. .2... +2...- .H. 8.. e882. 8... 8.. :28 22......
R..- 8.... E... on... ”ma-.- ..o....- 2... mm... €2.82. omA own 82:8 0.82
num- ._m.....- 3.. 8.. .w. .- +8...- 8.. 8.. 92.5... 8-2 8.. =28 8.8.).
an..- .8...- m... .3 2.. .- km...- mm... on. €8.55 8.. 22.88..
2....- 8.... 2.... S... 3. .- +R...- ... ... S... .88.. 88%.... 8.883.
8... 9.... 8.8 R... .2.- 1....- a. .m a... .88.. .82. .o 8.88....
mm..- .8... on: 8.2 8...- S...- 3... 3.... .88.... 8.8%... 8...
.3- 8.2.- 3: .2... a..- .2..- 2...... 8.... .28.. .82. 22.88.. 8 8..
“Sm-“ :00: SW02 SW02 “Sm-.. 502 50: e802
858...... 2.8 a. 38 a. 8888... 88 5

v030m>u0~5 $0303.35 00303035. OOON

-0.— »O: -0.— -0._ “o: 5 60303.02:

20:88: e_osomaom 22.82.: -2 32.8.5:

 

£58m courts .3 mocmtogfimso 22.830: 33 .«o 20>...— 582 ©-m 2an

when enumerators stop visiting households to establish contact. Thus, team dummies are
an indicator of the amount of effort teams put in locating households in follow-up
interviews.

Probit regressions based on equation (3.4) provide predicted probabilities of re-
interview. For attrition between the 1997 and 2000 surveys, the predicted

probability, &000 is obtained while P2004 is obtained as the predicted probability

between 2000 and 2004. The inverse probability weight for observations in the 2000

survey is given by R2000 = l/PZOOO . For the observations in the 2004 survey, the inverse
probability weight R2004 is a product of 1/&000 and 1/ I§004 because these observations

survived attrition twice. All models on impacts of working-age adult mortality are

estimated using these inverse probabilities as weights.

104

3.7 Results

In this section, I present results on two aspects of this essay: determinants of
household re-interview and the time dynamics of the impacts of working-age adult
mortality on various household aspects namely household composition, amount of land

cultivated, value of crop output, value of assets and income.

3.7.1 Determinants of household re-interview

The results of the re-interview models are presented in Table 3-7 and they
indicate that households that were re-interviewed differ from those that attrited in terms
of their characteristics in 1997. Household characteristics are jointly significant as
determinants of re-interview. In particular, households with older heads and those with
more male and female adults and children less than 6 years old are more likely to be re-
interviewed. This may be because larger households are less likely to move or dissolve
and it is easier for enumerators to find large households as well as potential respondents
in these households during follow-up. Household assets are also positively associated
with re-interview.

The community characteristics are jointly significant as determinants of re-
interview. Distance to the tarmac road reduces the probability of re-interview in 2004, an
indication that enumerators may have found it more difficult to follow up households in
remote areas. HIV prevalence rate is positively associated with re-interview. This may
suggest that in this sample, AIDS does not necessarily exacerbate attrition. It is possible

that households that have been afflicted by AIDS-related deaths are not more likely than

105

other households to dissolve or move away. Rather, such households may be

predominantly composed of the elderly and young children. Also, HIV prevalence rates

106

Table 3-7 Probit models for household re-interview for the period 1997-2004

 

Dependent variable =1 if:
Re-interviewed in Re-interviewed in
2000, contained in 2004, and is in 1997 Pooled

 

 

1997 sample and 2000 samples Model
Female head (=1) 0.026 -0.032 -0.002
(1.44) (1.57) (0.13)
Age of household head 0.002" 0.001 0.002"
(2.24) (0.71) (2.21)
Age of household head squared -0.000 -0.000 -0.000
(1.45) (0.18) (1.17)
Years of education of head 0.001 -0.002 -0.000
(0.77) (1.38) (0.25)
Number of male adults 0.005 0.007 0006*
(1.20) (1.46) (1.88)
Number of female adults 0.004 0.007 0.007*
(0.83) (1.50) (1.82)
Number of children 6-14 years -0.007* 0.004 -0.002
(1.84) (0.87) (0.79)
Number of children (<6 years) 0.005 0010* 0.007*
(0.86) (1.82) (1.83)
Ln (landholding size) -0.009 0.001 -0.005
(1.22) (0.10) (1.03)
Ln (Asset value) 0.006" 0005* 0006*“
(2.21) (1.92) (2.83)
Distance to tarmac road 0.000 -0.001* -0.000
(0.33) (1.70) (0.80)
Distance to fertilizer store 0.000 -0.000 0.000
(0.29) (0.08) (0.17)
HIV prevalence 0.006" 0.001 0.003*
(2.39) (0.40) (1.83)
Population density -0.0006** -0.000 -0.0003*
(2.31) (0.40) (1.86)
Precipitation/evapotranspiration 0.058 0.094“ 0.073"
ratio
(1.25) (1.89) (2.09)
Team dummies included Yes‘ Yes Yes
Year 2004 (=1) -0.008
(0.40)
Chi-square joint test for a
Household characteristics 38.58 (0.00) 25.45 (0.00) 46.91 (0.00)
Community characteristics 10.47 (0.06) 8.86 (0.11) 11.37 (0.04)
Team effects 15.23 (0.00) 5.38 (0.15) 21.16 (0.00)
Observations 1500 1407 2907

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000 and 2004

Notes: Estimated coefficients are marginal changes in probability; 2 statistics in parentheses, calculated
usrng heteroskedasticrty robust standard errors, clustered for households; *** indicates 1%; ** indicates
5% and;"‘ indicates 10% significance level; ‘ p-values in parentheses

107

may be picking up other regional factors that may be negatively correlated with mobility
and migration.

Households in highly populated districts are less likely to be re—interviewed while
those from villages with a high precipitation/evapotranspiration rate are more likely to re-
interviewed. Potential evapotranspiration is a representation of the environmental demand
for evapotranspiration and the precipitation/evapotranspiration ratio is an aridity index; a
numerical indicator of the degree of dryness (harshness) of the climate at a given
location. The influence of these two variables on the probability of re-interview is as
expected since in Kenya, migration and particularly urban-rural migration occurs
typically in areas with harsh climatical conditions or high population density.

The enumeration team dummies are jointly significant at 1% for the 2000 survey
and in the pooled model, signifying that team effort in contacting households during
follow up greatly influences the probability of re-interview. The implication of these
results from the re-interview probit models is that there is need to control for attrition and
this is done in the regressions on the impacts of adult mortality on household outcomes

by using the inverse probabilities as weights.

3.7.2 Impacts of working-age adult mortality on household composition

Changes in household structure have been shown to be part of households' coping
strategies. This section examines how households adjust their composition as a means to
cope with working-age adult death.

Table 3-8 presents ordinary least squares (OLS) results based on the specification
in equation (3.1). They show that compared to households that have not experienced

adult death, afflicted households tend to be larger and with more male adults. The death

108

of a working-age adult reduces the household size by 0.684 person and the number of
adult females by 0.326 person. This less than one-person decline suggests that partial
replenishment is taking place, indicating that households are unable to fully replenish
their numbers after incurring the death of a working-age adult member. The death of an
elderly female reduces the household size by more than one-person and 0.557 person for
female adults.

Male headed households have larger families and statistically significant larger
numbers of all categories of household members except female adults. Results with
regard to age of the head suggest a household composition that is consistent with a
household’s life cycle stages. As households mature, they have more adults and elder
children but fewer young children and elderly adults. However, later in their lifecycle, the
opposite scenario takes place, except in the case of younger children. This may be
because older sons and daughters may themselves start families. Households with more
educated heads tend to have fewer younger children which is consistent with the theory
that education and fertility are negatively related.

The results in Table 3-9 are based on the specification (3.2) which includes
dummy variables for years before and after death. The objective is to examine in a more
comprehensive manner the likely dip-drop-recovery temporal pattern of the dynamic

effects of death on household composition.

109

.x; “a Hemogewfi 3.2.. ween an «:85ch 1. ”aeo— 3 “:85ch _. swap—€23 5 £95 235% ~953— H882

 

 

 

110

 

38 23 88
doom .32 E 9335 Ecsomzom £82.22. 5:?— was Max:302 _a§_=otw< basis 88m 5w££2ksohazv 8855 8:5on "328m
N3 3 .c 88 3° 2 .o N3 85 8338-8
2:. S :. E... 3:. 5:. 3:. 3:. 823530
2:8 62.8 83.8 :28 $3.8 $8.8 $2.8
:Kbod aaimvfim .Iimed eiamad traced- fibeagu .31.me “c.8300
mo> mo> mo> mo> mo> mo> mo> moi—:56 wombfla
:2 O8 888 83.8 83.8 688 £38 8. _8
unsaved ouwaod mgd hmod unemvmd ouumcmd «asgwd VOON how .5:—:56 30>
33.8 88.8 88.8 88.8 82.8 83.8 :8 .8
*anwwod bmod 1.1.9— .c hmod ESL. _ m6 ***va.o 3.32.”— OOON he bah—5.6 30>
$8.8 $8.8 $8.8 88.8 :88 688 £88
8a.? .2898 :55 83. ......~8.o tags :33 Be; as .8 8:825
88.8 2898 2°88 58.8 88.8 88.8 88.8
Stu—cod :Lwoood *aamocoda **§Nooc.o. a: :56. .1... :56. 2!..NOOd- flog—um two: 05 mo ow<
68.8 $8.8 $8.8 :88 £38 888 98.8
***ch.o. unuwgdn :aNch nan—o6 :nmtd aiamgd «ustNd use: 05.3 ow<
@888 $3.8 $3.8 63.8 88.8 68.8 E; .8
3.2.. _ acd as...” _ Nd :aonmd «nemcflc :avwvd we .o 3.3va CHM—95 tau..— .«0 .5250
5N8 82.8 :38 c 2.8 88.8 :2 N8 :38
£2 82- 8o... 2&8 :3- time. :22- 58.8 225 beam
628 ASN.8 5 m8 888 82.8 63.8 as. 8
52$ 2: .o- 32 $3- 83. ES- ES- 5.8 2.2: €85
$8.8 $8.8 88.8 88.8 a: .8 a: .8 38.8
:3. .23. t 3. 85.8 _8 .o. :82. :83- 58.6 :23 088553
c 8.8 $88 88.8 $8.8 $58 :38 3:8
83. 88.8 ”8.0- ~86 1.8.8 2: .o :83 8.28s .3.»
E 8 6 A8 5 Q 5
mi mi 3%
muoEom To .5.—EEO 3-0 88m mi 3-0 2:0 933 232 833 03808 22833 mozmtg boaacflmxm
H 80 35:52

 

$28846 63va 53> mqov 33-32 doEmanoo 228:0: so .3358 :35 03-3353 .«o 83:: 2:. Wm 038%

 

 

 

 

.8...8 88.8 ...88 88.8 .88.... 88.8 88.8
.12.. :56 Stucco-o ......wocoda 3.2.52.6. .....iNocdu ...: _oodu .....iNOodn “5.535 two..— 05-«0 ow<
.888 88...... ...-8...... .888 ...-......8 .88.... .888
3.30— voda tinged- ......»— ~Nod ......Nw _ o6 ......»%b— .O ......onc— .O ......VMBNd can: van-«o ow<
38.8 .5....8 ......8 ......8 88.8 .88.... 62.8
....immod zine .o ...»...hvmd ......«Nomd ....immvd ...oo— .o ......meA CHEM—.5 95: .«o .5950
... m8 ......8 .58 .5.... ...88 .82.. .... N. ..
8m... 8m..- 8..... ...-m...- ..m...- .8...- 8m. .- 58.. 258. 5......
....N. .8 .388 88.8 .82.. .N. 08 5...... .8... ..
:83... 8...? SN... 5......- 88. .8..- 828 58.. “5.5 38......
......8 .8 . .8 .8. .8 ...... .8 ....N8 .858 ...28
58..
:38...- S. ... 8.....- 8. ...- 2.... .82. 8...... 8...m 88. 883......
G... .8 58.8 .858 .. 8.8 .588 .... m8 588
2......- ...c....... 8 . ...- ...-8...- m8... 8...... ...8.~. .- 58.. 8.... e8. .8“.
:88 .22.. ...-8.8 88.8 ...88 .588 G88
......o. 88... ... .... m. .... :8.....- 8.....- .8... .- 58.. 8.... e8. 85
...... .8 EN... .258 .358 5...... ......8 68.8
5...... e. .... 2...... 2...... ..R8 8.... .8... 58.. 8...... a8. ......
.528 ...... .8 .8. .8 .3. .8 EN... .528 ....n8
..8... 8.... 58...- w. ..8 8.....- .8... 8m... 58.. 8.... .8. 80
.88 ..88 ...-8.8 88.8 .58.... .32.. .858
8.... 88... 8.... : .8... 2.... 2:88... ......R.~ 582.. 8..
...... .8 ....m8 .. ....8 .858 8.....8 ......8 68.8
SN... ...-0....- 8.....- .8... 8m... 8...... 85...- 58.. 258 .8. 25
.8. .8 .858 .....N8 .588 .5.... 88.8 .898
58... ..N. ...- 8.... E... ... .... 8.8...- 8m..- 58.. 25.... a8. 95
.. 8.8 88.8 .888 .888 .888 ...-8.8 .8 . .8
.8... «8...- 5......- ........ :88... R. ... ...-Sm... 885-... 83..
.8 .8 .8 .... .8 .8 ...
828m 5. m-.. 85.50 ...-o m...m. ...-... 2.6 5.5.... 8...: 5...... 285...“. S... 5.2.8.8: 8.5.5... 5885.55
”.5 59:52

 

$388.6 85m...
5.3 m-HOV gem-So. £80.. ..on E... 283 8.8.. .3 50:50:58 20:08.0: ..o .3135... :3... own-matte? mo 85...... o-m 03....-

111

.... .. 2.858.. .1. xx... .. 2.888.. : .8... .. E858... ..

835.58.. E 88.8 ...—«v.88 .853— .832

 

 

 

 

....8 ...a
Sam .88 .32 .... 9.3-Sm 22.3-so: mix—«.2. 5:.... was mavens—2 .5155... .0533: 88m :mw_:o.2\cao-._nzv 83.8... 880on "co-Sow
8.. .. ... 8... ...... .. ... m. ... 8... 88.9.-..
R... 8... 8... 8... 8... 8... 8... 82.38....
...: .... .888 .888 .888 G8... .82.. .82..
......8... :83. :88... :8... ......RN. ......88- $8.. .5388
8... mo> mo> mo> mo> mo> 8; 3.58.... 8:85
.8. ...... .808 ......8 .....o... .828 .89... .... ..8
:88... :8.... 3...... 8.... :88... :18... :8... ....8 .... .55.... 58>
8...... .888 8...... .808 .5.... .888 . . . ..8
:88... 8.... :8... ... 8...... ...:. .m... :88... :88. ....8 .... .55.... 8..
......8 .25.... ......8 ........... 688 SS... 8......
8...? 2.888 8.... 8...... :88... :88... :88... ..8.. .5... 858......
.8. .8 .8 4... .... .N. ...
......8m 2. m-.. 85...... ...-e ...... .. 8 ...... 9...... ...-.... 3...... 2...... 8.. 22.88.... 83...... b.3558...
”mo 595:2

 

3:538 8m 2......-

112

These results show that there are no significant changes in household size and
composition in the periods preceding death, except for the case of female adults. Two
years before death, there is a significant decrease in the number of adult females as
households begin to experience the effects of illness. This result is somewhat surprising
and inconsistent with the expectation that additional female adults will both provide care
to the sick and assist in other tasks related to home and farm production or income
generation, which were previously undertaken by the sick household members or the care
givers. However, the finding suggests that households afflicted by working-age adult
mortality are not just making ex post responses to cope with death. Rather, they are also
making pre-death illness adjustments in household composition.

In the year of death, we find an increase in household size and the number of
female adults, girls and young children. A household’s flexibility is a critical component
in successfully responding to economic hardship. For instance, the presence of additional
members can promote flexibility of households by facilitating alternative work
arrangements among the core household members. New members can take on most of the
domestic work, allowing core members to reallocate more time to income-generating
activities. For afflicted households, inclusion of members may be motivated by
economic considerations, owing to their greater economic hardships as a result of
sickness and death.

There are few significant impacts of mortality in the years following death.
Households experience a more-than-one person decline in household size in the third and
fourth years after death. This is due to the death of the adult male member (column 3) and

a reduction in the number of girls (column 4) and young children (column 6). A decline

113

in the number of girls may indicate two things. First, the support of young girls in
providing help with household chores and care giving is no longer needed four years after
the adult’s death. Second, households may send young girls to live with other extended
family members or to seek for employment opportunities upon the death of a working-
age adult. The out-migration of young children suggests that fostering is one of the
coping strategies employed by afflicted households. Fostering helps by spreading the
impact of death over several households. In the period five to seven years after death,
there is an increase in the number of adult females but a decrease in the number of
seniors. It is worthwhile to note that adult mortality does not appear to affect the number
of boys in a household.

Overall, these findings show that changes in household composition take place
over time, implying that such changes are part of households' coping strategies.
Therefore, a household’s flexibility is a critical component in successfully responding to
extreme crises such as illness and death of an adult. Additionally, households afflicted by
working-age mortality are not just making ex post responses to cope with death. Rather,
they are also making pre-death illness adjustments as a coping strategy. It is also evident
that the impacts of mortality on household composition are not just a one-time permanent

adjustment.

114

Impacts of working-age adult mortality on household composition by the deceased ’s
gender and position in household

The foregoing analysis indicates that on average, afflicted households experience
changes in household composition and size. To determine how the pattern of these
changes varies by gender and position of the deceased in the periods before and after
death, I estimate model (3.3).

The results of this analysis are presented in Table 3-10. In the period 1-2 years
before death, the impact of death does not vary by gender and position of the deceased.
In the year of death, we find that compared to other deaths, a male head death leads to a
decrease in the number of boys and young children. This means that when a male head
dies, boys and young children are more likely to be fostered out since the household loses
the main bread-winner. Also, four years after death, in comparison to other deaths, a
male head death is associated with a more-than-one person decrease in the number of
adult males. Overall, we find that changes in household composition depend to a limited

extent on the gender and position of the deceased member.

115

 

 

 

 

836 82.8 898 $98 83. 8 93.8 82.8
:3. 838. 9.2. 33. $2 «8.? ”an. _- Esau as; 0.2....38 8% as; 95
88.8 S n8 53.8 8&8 32.8 6:..8 $8. 8 .
8.9- $3 «8... 828. 23- 5.8 $2- 38% 82 gafsgc 8:: as» 25
§ N8 88.8 9.38 2 8.8 9S8 $3.8 ASN.8
02.8 .23- .18.? 238. mm. .o :28 an. _- Sea. 22 23.58“. .6 a;
§ _.8 98.8 82.8 9.88 82.8 E «:8 98.8
”:8 8:8 «85 88.8 82 :23 83 age 85... as...» 58,353...
G: .8 83.8 88.8 53.8 98.8 638 35.8
33. ~93. 8:8. 83. was $.28- $3- 58o 8:: as...» sou
:8 .8 :38 82.8 $3.8 $3.8 $88 68. 8
~86 828 :5de and :28. S 3. E3 58c 8% can; 8::
88.8 9.88 $8.8 88.8 $8.8 $2.8 $2.8
m2; 228 838 m: .o 83- 828 82 58c 8% who» 2:
a: .8 338 8.2.8 :28 88.8 25.8 33.8
mm. .c 328 RS 338 2: .o .88 m: ._ 58c 8% a...» 25
$2.8 $28 838 5.08 33.8 83.8 28.8
”8.8 :83 :3... :83 $28 an; ...... _ mom 586 Mo 58>
33.8 8&8 33.8 :38 9.8.8 82 .8 22.8
mm; 83. 3:. 8.3 no? 32 N2 ._ 255 22 22.58c 283 so» 2.0
828 2.8.8 628 88.8 3&8 $58 $38. 8
82 83. 8.3 :2 82 ~35- 83 38% as; e_meissc 2&3 e8» 2;
$8.8 $.08 $3.8 $3.8 82.8 § 08 55. 8
$28 $2. .8828. m: .o. :2 $3.? an. _- 58a 2&3 so» 25
a: .8 82.8 $3.8 $2.8 $3.8 A3328 3. ”.8
.13 ”8.8 $3. 2: .o- 29o «8.8 .2 . _- 53% 283 was» 2:
c 8.8 888 £88 $8.8 $8.8 £38 5 _ .8
N36- :28- 85.8 :38 ...:82 mm. .o :ng BEE“ 35
E 68 5 c8 6 S A 8
PS $.33 ofim
seem so 5.220 :a £8 :6 ...Eo 3.28 2a: 22E 223.5: 8335 bosfiaxm
co 598:2

 

$28888 88.3% 53> mAOV 288-33 .588 coca can 8&3
menu» B 85 83388 05 80 82:88 98 Hot—How 3 82:88:50 20:039.— :o @2358 2:3 owmmEv—coB mo 83:: o Tm 033.

116

.x: E “:35ch ...: ”can an ESE—gm ..i ”$.8— 3 ~565qu . momoficocwq E Echo 8.38.83 330M ”832

 

 

 

 

38
8.8 88 doom .32 E mag-5m 20:83: max—92. 3:0“— 85 mug-88:02 =§£=on< basicD 88m :wwEom—zxsoazv 8355 880on "325m
~28 2.0 23 8.: E .o 26 o~.o Bang-m
3:. 3:. 3:. 3:. 3:. 3:. 3:. 22.3080
2:8 838 83.8 :38 83.8 32.8 :38
3.1.3906 ...:wbmd ......caod ...: _ mad eeuwah.mn alive—d- mnwd “53800
mo> mo> mo> mo> mo> mo> mo> momegfi “ombmmfl
88.8 88.8 23.8 83.8 $3.8 $28 5 _.8
:33 .. 328 $28 328 ...:32 ...: GS ...: _ m; 38 é 55% 50>
88.8 88.8 83.8 88.8 398 9.8.8 2 _ _.8
.....imnod «No-o ......mO-d {mood seem—m6 susoavd 3.....:.N._ OOON .5.“ .3556 30>
£88 $8.8 $8.8 $8.8 68.8 68.8 38.8
88.8. 338.8. 58.8 coed. ......Nmod ...:NNQO .....immcd 83: 05 mo cosmosum
88.8 88.8 88.8 88.8 88.8 88.8 88.8
.21. :56 ...:Good evegodn ...:oood- ...»...Nood. .....i _ood. see-mood- 80.883 Una: 05.3 ow<
$8.8 $8.8 88.8 :88 88.8 88.8 88.8
3.5—36: .....incodn 3:.—Nod aim—o6 {such—.0 ....inb—d scam-Nd use: 05.3 ow<
588 2:58 $3.8 $3.8 A888 3.88 85 .8
:emmaod ...:hva_.o essehvmé {senammd ...:Nvmvd a: _ _d ......ach— GHQ—95 two..— .wo 38:00
8-.8 A-v8 $2.8 32.8 83.8 :38 $-. 8
:.~.o 328- 83 ~83. 3...»? v3.8 33- 596 285. 8.85
5:8 :38 33.8 83.8 33.8 38.8 88. 8
:32 3:.? 3~.o m: .8 2:8 33. c~w.o- save 29: beam.
388 83.8 33.8 338 A358 638 82.. 8
~38- ~8.o- 3:8- 33. :.~_ .c @38- 35. _- 55% 98.. gas-58¢ 8:: as; 5335
32.8 88.8 33.8 33.8 $2.8 $3.8 89.8
838- 328 82 9.3. :o-. _- 3~.o :3. _- 2.5% Bo: safes-6 8:: as...» .8»
:._ m8 388 $3.8 $3.8 33.8 88.8 3?. 8
god 3 _ .o- 8:.- m _ :3. w _ 3. «~38- o-.~- 55% 22 one-58‘s 82m 3% 85
E 68 5 c8 5 A8 :8
mi 33% 338“ «NE
fioEom m-o 56::0 3-8 98m 3-8 2:0 222 225““ 22.3.5: 832:; bo~m§_axm
co :58: Z

 

8:588 8-... use.

117

3.7 .3 Impacts of working-age adult mortality on farm production

In this section, we examine the impacts of working-age adult mortality on farm
production in terms of total cultivated area, area cultivated by crop types, and the value of
crop production. Our findings show that adult mortality affects households’ farm
productions decisions. A reduction in household size due to adult death may lead to labor
shortages and labor re-allocation. Also, an adult death may result in the loss or decline of
farming knowledge and skills, and increased medical and funeral expenses which may
impose considerable cash constraints. As a result, we see effects of mortality on farm
production through changes in area cultivated, cropping patterns and farm output. In
particular, cash constraints and loss of labor and farming knowledge may lead. to a
reduction in cultivated area and shifts in crop mix away from labor-, management- and
capital or input-intensive crops. For example, maize and high-value crops (e.g. tea, some
horticultural crops) are considered more labor- and capital- intensive than root crops like

sweet potatoes, yarns and cassava.

3.7.3.1 Impacts of working-age adult mortality on cultivated land

Table 3-11 presents the results of the impact of adult mortality on cultivated area,
in total and by crop type. These results show that although working-age adult mortality is
associated with a decline in total land cultivated and the area devoted to the different crop
types, none of these impacts are statistically significant. However, results in subsequent
sections, which examine the dynamic effects of death reveal that mortality has significant

effects on cultivated land.

118

Table 3-11 The impact of working-age adult mortality on total cultivated land and area by
crop types in for the period 1997-2004 (OLS with AEZIO dummies)

 

Natural logarithm of acres of cultivated land:

 

 

 

Total Cereals High value Roots and
Explanatory variables crops tubers
(1) (2) (3) (4)
Ever afflicted -0.020 -0.013 -0.033 0.004
(0.036) (0.036) (0.036) (0.026)
Working-age adult death -0.065 -0.019 -0.033 -0.032
(0.062) (0.060) (0.060) (0.044)
Elderly male death -0.244 -0.153 -0.286*"‘ -0.042
(0.151) (0.178) (0.134) (0.123)
Elderly female death 0.067 0.034 0.028 -0.121
(0.243) (0.197) (0.274) (0.173)
Gender of head (male=l) 0075*" 0.017 0085“" 0054*"
(0.028) (0.028) (0.028) (0.020)
Age of the head 0015*" 0.007 0020*" 0.012“"'“'l
(0.005) (0.005) (0.004) (0.003)
Age of the head squared -0.0001* -0.0000 -0.0001*** -0.0001"‘*"‘
(0.0000) (0.0000) (0.0000) (0.0000)
Education of the head 0025"" 0019*" 0017*" 0.002
(0.003) (0.003) (0.003) (0.002)
Distance to tarmac road (km) 0011*" 0012‘“ 0005*" 0.000
(0.001) (0.001) (0.001) (0.001)
Distance to fertilizer store (km) 0.003" 0003*" -0.003*** 0001*"
(0.001) (0.001) (0.001) (0.001)
Year dummy for 2000 0086*“ 0081*" 0417*" 0124*"
(0.023) (0.024) (0.023) (0.017)
Year dummy for 2004 0072*" 0.049“ 0365*" 0105*"
(0.024) (0.024) (0.023) (0.017)
AEZ dummies Yes Yes Yes Yes
Constant 0554*“ 0647’" 0.031 0.219"
(0.136) (0.141) (0.123) (0.088)
Observations 4157 4157 4157 4157
R-squared 0.1 1 0.13 0.16 0.08

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004
Notes: Robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at

1%

 

'0 AEZ refers to agro-ecological zone

119

Other factors significantly affect the area of land that is cultivated. For instance,
male-headed households have more acres of cultivated land in total and also under high-
value crops and roots and tubers. The age of the household has a significant non-linear
effect on total cultivated land and area devoted to high-value crops and roots and tubers.
Observed differences in land area cultivated are consistent with both a resource-
endowment and a demographic (based on a household’s life cycle) explanation. When the
head is young, the household age structure is young (due to the presence of infants and
young children) and the household commands fewer resources. The high dependency
ratio and fewer resources imply that relatively little land is farmed due to limited labor
and capital available for agriculture. As time passes, the household accumulates more
resources and the average age among the children increases. As children grow up and
become more economically active, the dependency ratio decreases. This, coupled with
more resources allows for the expansion of the land area cultivated.

Households with more educated heads have more cultivated land in total and
under cereals and high-value crops. Also, the distance variables are positively related to
land cultivated. The distance variables indicate the remoteness of an area, the quality of
road infrastructure and thus the associated transport costs. Livelihoods of households
residing in remote rural areas depend to a larger extent on farming. Therefore, they have
more land under cultivation and particularly for cereals in order to ensure food security.
However, as expected, the longer the distance to fertilizer store, the more expensive
inputs costs become and so the less the area cultivated. This is true for the high-value
crops which are intensive in the use of purchased inputs but is surprising in the case of

roots and tubers.

120

The results in Table 3-12 show the temporal pattern of the dynamics of death
effects on land cultivated. The coefficients on two years before death indicate that
households have more cultivated total land and area under high-value crops and roots and
tubers. One year before death, only the area under high-value crops appears to be
significantly larger. In the year of death, there is no statistically significant effect on area
cultivated both in total and by crop types. However, one year after death, area under
roots and tubers declines while two years after death, total cultivated land and area under
high-value crops, roots and tubers declines. Also, five to seven years after death, there is
a significant decline in the area devoted to high-value crops. This pattern of effects
reflects the progression of the severity of disease. Two years before death, illness is
relatively mild and so households have enough labor and capital to farm their land. As
the disease progresses and is followed by death, households may begin to face labor
shortages and cash constraints causing them to cut back the area under cultivation.

The decline in area under high-value crops even five to seven years after death may
signify tighter cash constraints, causing households to shift away from high-value crops
which require larger capital investment and possibly more labor than other creps.

To summarize, these results show that afflicted households experience different
impacts of death on cultivated land as time passes since death. Except in the case of
high-value crops, the impact of death does not persist beyond two years after death. This
finding indicates some recovery in cultivated area. There is no statistically significant
impact of morbidity and mortality on the area under cereals, indicating that households
appear to be successful at safeguarding their food security position. However, they are

less able to maintain the area under the capital- and labor-intensive high-value crops.

121

Table 3-12 The impact of working-age adult mortality on total cultivated land and area by
crop types for years before and after death in the period 1997- 2004 (OLS with AEZ

dummies)

Natural logarithm of acres of cultivated land:

 

 

 

Explanatory variables Total Cereals High-value crops Root and
tubers
(1) (2) (3) (4)
Ever afflicted -0.020 -0.01 1 -0.037 -0.001
(0.036) (0.037) (0.035) (0.027)
Two years before death 0.261 "‘ 0.151 0.323“ 0.298"”""l
(0.145) (0.150) (0.143) (0.107)
One year before death 0.229 0.173 0.373“ 0.090
(0.181) (0.188) (0.179) (0.134)
Year of death 0.083 0.137 -0.040 0.080
(0.133) (0.138) (0.131) (0.098)
One year since death -0.096 -0.059 -0. 160 -0.143*
(0.112) (0.116) (0.111) (0.083)
Two years since death -0.286* -0.214 -0.389** -0.253**
(0.167) (0.173) (0.165) (0.123)
Three years since death -0.072 -0.083 -0.097 -0.002
(0.130) (0.134) (0.128) (0.096)
Four years since death -0.034 -0.048 -0.073 -0.085
(0.136) (0.141) (0.134) (0.100)
Five-seven years since death -0.029 0.036 -0.199"‘ -0.109
(0.1 14) (0.118) (0.112) (0.084)
Elderly male death -0.235 -0. 148 -0.262 -0.033
(0.207) (0.215) (0.204) (0.153)
Elderly female death 0.073 0.028 0.008 -0.088
(0.251) (0.260) (0.248) (0.186)
Gender of head (male=l) 0.067" 0.007 0080*" 0.052M
(0.028) (0.029) (0.028) (0.021)
Age ofthe head 0015*" 0.007 0020*" 0012*"
(0.004) (0.004) (0.004) (0.003)
Age of the head squared -0.0001" -0.0000 -0.0001*** -0.0001***
(0.0000) (0.0000) (0.0000) (0.0000)
Education ofthe head 0026*" 0019*" 0018*“ 0.003
(0.003) (0.003) (0.003) (0.002)
Distance to tarmac road (km) 0011*" 0012*" 0005*" -0.000
(0.001) (0.001) (0.001) (0.001)
Distance to fertilizer store 0003*" 0004*" -0.003*** -0.001***
(km)
(0.001) (0.001) (0.001) (0.000)
Year dummy for 2000 0086*“ 0079*" 0417*“ 0124*"
(0.023) (0.024) (0.023) (0.017)
Year dummy for 2004 0074*” 0.050" 0362*" 0104*"
(0.024) (0.025) (0.023) (0.018)
AEZ dummies Yes Yes Yes Yes
Constant 0468*” 0595*" 0.088 0.228"
(0.156) (0.162) (0.154) (0.115)
Observations 4157 4157 4157 4157
R-squared 0.1 l 0.13 0.16 0.08

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004
Notes: Robust standard errors in parentheses; *significant at 10%; "significant at 5%;"*significant at 1%

122

Impacts of working-age adult mortality on cultivated land by the deceased ’s gender and
position in household

In the results shown in Table 3-13, I examine if the effects of adult mortality in
the periods before and after death vary depending on the gender and position of the
deceased. In the period preceding death and up to two years after death, the effect of
morbidity and mortality of male household heads on land cultivated does not differ from
that of all other adult household members. By the third year after death, households
afflicted by male head deaths as opposed to other adult deaths, face a large reduction in
total cultivated land and area devoted to cereals and high-value crops. This suggests that
the loss of a household head imposes greater hardship for households for various reasons.
First, the household loses the main bread winner who has great influence and control of
financial resources and social capital. In addition, the death of male heads may be
associated with land sales and subdivision or widows may face great difficulties in

retaining access to land after the death of their husbands.

123

Table 3-13 The impact of working-age adult mortality on total cultivated land and area
by crop types, by gender and position of the deceased for years before and after death in
the period 1997-2004 (OLS with AEZ dummies)

 

Natural loggithm of acres of cultivated land:

 

 

 

Total Cereals High-value Root and
Explanatory variables crops tubers
<1) (2) <3) (4)
Ever afflicted -0.029 -0.022 -0.042 -0.002
(0.036) (0.037) (0.035) (0.026)
Two years before death 0.102 0.018 0.223 0.279“
(0.169) (0.175) (0.166) (0.125)
One year before death 0.099 0.063 0.213 0.086
(0.224) (0.232) (0.221) (0.166)
Two years before death*ma1e head dummy 0.199 0.131 -0.062 -0.078
(0.386) (0.400) (0.381) (0.286)
One year before death*male head dummy -0.495 -0.521 -0.275 -0.310
(0.513) (0.532) (0.507) (0.380)
Year of death 0.189 0.279 0.080 0.181
(0.172) (0.179) (0.170) (0.128)
One year since death -0.1 19 -0.025 -0.055 -0.l30
(0.138) (0.143) (0.136) (0.102)
Two years since death —0.248 -0.151 -0.206 -0.260"'
(0.190) (0.197) (0.187) (0.141)
Three years since death 0.174 0.193 0.278 0.079
(O. 198) (0.206) (0.196) (0.147)
Four years since death -0.041 -0.072 0.075 -0.010
(0.164) (0.171) (0.162) (0.122)
Five-seven years since death -0.027 0.023 -0. 173 -0. 123
(0.137) (0.142) (0.135) (0.102)
Year of death'male head dummy -0.307 -0.353 -0.108 -0. 165
(0.267) (0.277) (0.264) (0.198)
One year since death*male head dummy -0. 129 -0.252 -0.085 0.070
(0.328) (0.340) (0.323) (0.243)
Two years since death*male head dummy 0.239 0.227 0.077 0.363
(0.431) (0.446) (0.425) (0.319)
Three years since death*male head dummy -0.937** -0.959** -0.923"”" -0.245
(0.365) (0.378) (0.360) (0.270)
Four years since death‘male head dummy -0.185 -0.039 -0.441 -0.318
(0.308) (0.319) (0.304) (0.228)
Five-seven years since death*male head 0.333 0.382 0.303 0.150
dummy
(0.310) (0.322) (0.306) (0.230)
Elderly male death -0.259 -0.l91 -0.309 -0.048
(0.21 1) (0.219) (0.208) (0.156)
Elderly female death 0.014 -0.049 -0.053 -0.111
(0.254) (0.263) (0.251) (0.188)
Gender of head (male=l) 0.067" 0.007 0079*" 0.053"
(0.028) (0.029) (0.028) (0.021)
Age ofthe head 0015*“ 0.007 0020*" 0012*"
(0.004) (0.004) (0.004) (0.003)
Age of the head squared -0.0001** -0.0000 -0.0001"* -0.0001*"
(0.0000) (0.0000) (0.0000) (0.0000)
Education of the head 0026*" 0019*" 0018*" 0.003
(0.003) (0.003) (0.003) (0.002)

 

124

Table 3-13 continued

-....- -. m. *3.

Natural logarithm of acres of cultivated land:

 

 

Total Cereals High-value Root and
Explanatory variables crops tubers
Distance to tarmac road (km) 0.01 1*" 0012*“ 0005*“ -0.000
(0.001) (0.001) (0.001) (0.001)
Distance to fertilizer store (km) 0003*" 0004*“ -0.003*** -0.001***
(0.001) (0.001) (0.001) (0.000)
Year dummy for 2000 0085*" 0078*" 0417*" 0124“"
(0.023) (0.024) (0.023) (0.017)
Year dummy for 2004 0072*" 0.048" 0.361 *** 0.104***
(0.024) (0.025) (0.023) (0.018)
AEZ dummies Yes Yes Yes Yes
Constant 0549*" 0651*" 0.030 0.212”
(0.127) (0.131) (0.125) (0.094)
Observations 4157 4157 4157 4157
R-squared 0.1 l 0.13 0.16 0.08

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

Notes: Robust standard errors in parentheses; *significant at 10%; "significant at 5%;***significant at 1%

3.7.3.2

Impacts of working-age adult mortality on the value of crop output

The impact of working-age adult mortality on value of crop output is shown in

Table 3-14. Results indicate that mortality has no statistically significant impact on either

the total or per acre gross value of crop output. However, the death of elderly male adults

leads to a decline in the total gross value of crop output. This is probably due to two

reasons. First, older men (60 years and above) are likely to be the ones who have title to

land and may control a reasonable amount of productive resources. Second, they are also

likely to be strong and productive even at this age, and compared to the working-age

adults, they may allocate most of their time to crop production.

There are other factors that influence the value of crop output. Male-headed

households have more crop output in total and on per acre basis. Age of the head has a

non-linear effect on total but not per acre gross value of crop output. Also, households

125

with more educated heads tend to have greater crop output and higher land productivity.
The distance variables are positively related to total gross value of crop output but
negatively related to gross value of crop output per acre. This may be because
households residing in remote rural areas derive their livelihood mainly from crop
cultivation. However, their location implies that inputs costs are likely to be high, leading
to poor land productivity.

Table 3-15 shows that the impact of adult mortality on value of crop output does
not vary over time, both for the periods before and after death. This result may reflect the
fact that there are very few significant changes in area cultivated both in total and by crop
types over time (Table 3-12). This may suggest that afflicted households may be well-
cushioned from the potential effects of illness and death arising from loss of income,
labor, farming knowledge and skills. It is possible that informal social safety nets provide

support to afflicted households, particularly in the form of labor and seeds.

126

Table 3-14 The impact of working-age adult mortality on gross value of crop output in
the period 1997-2004 (OLS with AEZ dummies)

 

Natural logarithm of value of:

Gross value of crop output

Gross value of crop

 

 

Explanatory variables output/acre
(1) (2)
Ever afflicted 0.022 0.033
(0.069) (0.056)
Working-age adult death -0.092 0.025
(0.120) (0.097)
Elderly male death -0.493* -0.132
(0.288) (0.321)
Elderly female death 0.128 0.033
(0.460) (0.435)
Gender of head (ma1e= 1) 0338*" 0209*"
(0.065) (0.051)
Age of the head 0026*" 0.006
(0.010) (0.008)
Age of the head squared -0.0002* -0.0001
(0.0001) (0.0001)
Education of the head 0053*" 0018*"
(0.006) (0.005)
Distance to tarmac road (km) 0010*“ -0.004*
(0.003) (0.002)
Distance to fertilizer store 0.003” -0.000
(km)
(0.001) (0.001)
Year dummy for 2000 0.381 **"‘ 0263*"
(0.056) (0.048)
Year dummy for 2004 -0.055 -0.l62***
(0.049) (0.040)
AEZ dummies Yes Yes
Constant 7995*" 8172* **
(0.309) (0.264)
Observations 4157 4157
R-squared 0.24 0.27

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

Notes: Robust standard errors in parentheses; *significant at 10%; ”significant at 5%;"*significant at 1%

127

 

Table 3-15 The impact of working-age adult mortality on value of crop output for the
years before and after death in the period 1997-2004 (OLS with AEZ dummies)

 

Explanatory variables

Natural logarithm of value of:
Gross value of crop output

Gross value of crop output/acre

 

 

(1) (2)
Ever afflicted 0.01 1 0.032
(0069) (0.056)
Two years before death -0.068 -0.346
(0291) (0.265)
One year before death -0.008 -0.289
(0.341) (0.341)
Year of death 0.154 0.172
(0.283) (0.179)
One year since death -0.013 0.185
(0.267) (0.247)
Two years since death -0.008 0.432
(0.356) (0.295)
Three years since death -0.050 0.161
(0.267) (0.247)
Four years since death -0.229 -0.203
(0.254) (0.185)
Five-seven years since death -0.029 -0.004
(0.234) (0.21 l)
Elderly male death -0.513"' -0. 145
(0.301) (0.318)
Elderly female death 0.101 0.027
(0.470) (0.473)
Gender of head (male=1) 0339*" 0210*"
(0.065) (0.051)
Age of the head 0027*" 0.006
(0.010) (0.008)
Age of the head squared -0.0002"‘ -0.0001
(0.0001) (0.0001)
Education of the head 0053*" 0018*"
(0.006) (0.005)
Distance to tarmac road (km) 0010“" -0.004
(0.003) (0.002)
Distance to fertilizer store (km) 0.003" -0.000
(0.001) (0.001)
Year dummy for 2000 0380*" 0265*”
(0.057) (0.048)
Year dummy for 2004 -0.059 -0. 165*"
(0.049) (0.040)
AEZ dummies Yes Yes
Constant 7985*" 8.166***
(0.310) (0.265)
Observations 41 57 4157
R-squared 0.24 0.27

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

Notes: Robust standard errors in parentheses; *significant at 10%; "significant at 5%;***significant at 1%

Impacts of working-age adult mortality on value of crop output by the deceased ’s
gender and position in household

In this section, I examine whether the impact of adult mortality on the value of
crop output depends on the gender and position of the deceased adult. The results in
Table 3-16 show that to a great extent, this impact does not vary by gender and position
of the deceased. Three years after death, households afflicted by the death of male heads
have greater value of crop output per acre, compared to households that experienced
other adult deaths. However, in the period five-seven years since death, the opposite
effect is observed. This may be attributed to combined effects of the loss of farming
skills, human capital, and financial resources. In the longer term, these effects may be
expected to be larger upon the death of the male heads due to their greater influence and
control of these resources and social capital. In addition, the death of male heads may be
associated with land sales and subdivision or widows may face great difficulties in
retaining access to land after the death of their husbands. All these factors diminish the

ability of afflicted households to maintain or improve land productivity over time.

129

Table 3-16 The impact of working-age adult mortality on gross value of crop output by
gender and position of the deceased, for years before and after death in the period 1997-
2004 (OLS with AEZ dummies)

 

Natural logarithm of value of:
Gross value of crop Gross value of crop

 

 

Explanatory variables output output/acre
(1) (2)
Ever afflicted 0.01 1 0.032
(0.069) (0.056)
Two years before death -0. 196 -0.247
(0.306) (0.291)
One year before death 0.010 -0.105
(0.377) (0.388)
Two years before death*ma1e head dummy 0.187 0.014
(0.739) (0.566)
One year before death*male head dummy -0.825 0.190
(0.619) (0.523)
Year of death 0.248 0.080
(0.368) (0.265)
One year since death -0.105 0.071
(0.282) (0.273)
Two years since death -0.040 0.301
(0.3 82) (0.327)
Three years since death 0.015 -0.190
(0.297) (0.302)
Four years since death -0.121 -0.126
(0.278) (0.221)
Five-seven years since death 0.071 0.045
(0.251) (0.236)
Year of death*male head dummy -0. 163 0.231
(0.580) (0.341)
One year since death*male head dummy 0.590 0.704
(0.743) (0.549)
Two years since death*male head dummy 0.759 0.168
(0.831) (0.604)
Three years since death*male head dummy 0.045 0265*"
(0.571) (0.048)
Four years since death*male head dummy -0.485 -0.247
(0.593) (0.404)
Five-seven years since death‘male head -0.344 -0.852**
dummy
(0.524) (0.399)
Elderly male death -0.491* -0.121
(0.290) (0.319)
Elderly female death 0.092 0.072
(0.435) (0.459)
Gender of head (male=l) 0347*" 0217*"
(0.065) (0.051)
Age of the head 0027*“ 0.007
(0.010) (0.008)
Age of the head squared -0.0002* -0.0001
(0.0001) (0.0001)
Education of the head 0053*" 0.018***
(0.006) (0.005)

 

130

Table 3-16 continued
Natural logarithm of value of:
Gross value of crop Gross value of crop

 

 

Explanatory variables output output/acre
Distance to tarmac road (km) 0010*" -0.004*
(0.003) (0.002)
Distance to fertilizer store (km) 0.003M -0.000
(0.001) (0.001)
Year dummy for 2000 0380*“ 0265*"
(0.057) (0.048)
Year dummy for 2004 -0.059 -0.166***
(0.049) (0.040)
AEZ dummies Yes Yes
Constant 7969*" 8154*”
(0.310) (0.266)
Observations 4157 4157
R-squared 0.24 0.27

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring
and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

Notes: Robust standard errors in parentheses; *significant at 10%; "significant at 5%;***significant at 1%

131

3.7.4 Impacts of working-age adult mortality on assets

Households may sell assets in order to deal with many shocks including additional
financial constraints imposed on them by adult illness and mortality (Barnett and Blaike,
1992). In this section, we examine the impacts of adult mortality on the value of
households’ cattle, small animals, farm equipment and other assets.

Results in Table 3-17 indicate that adult mortality has adverse and significant
effects on total assets, small animals and farm equipment but not on cattle and other
assets. Consistent with findings by Yamano and Jayne (2004) and Chapoto and Jayne
(2008), households experiencing adult death appear to rely heavily on the sale of small
animals rather than cattle and farm equipment.

While male elderly death has no significant effect on household assets,
surprisingly the death of an elderly female leads to a significant increase in total assets.
Compared to female-headed households, male-headed households have more assets. The
age of a household head has a non-linear effect on asset levels. This reflects the fact that
the ability of a household to accumulate more assets over its lifecycle is greater at the
earlier stages but diminishes later. Also, households with more educated heads tend to
have more assets. Distance variables are positively related to the amount of assets,
particularly small animals and farm equipment. This is because households that live in
relatively more remote rural areas are mainly engaged in crop and livestock production.

The complete temporal pattern of the impact of working-age adult mortality on
assets is shown in Table 3-18. The results show that the value of small animals, farm

equipment and other assets varies over the period of illness and after death.

132

Table 3-17 The impact of working-age adult mortality on value of assets for the period

1997-2004 (OLS with AEZ dummies)

 

Natural logarithm of value of:

 

 

 

Total assets Cattle Small Farm Other
Explanatory variables animals equipment assets
(1) (2) 43) (4) (5)
Ever afflicted 0.052 0.053 -0.075 -0.041 -0.187
(0.150) (0.278) (0.182) (0.253) (0.229)
Working-age adult death -0.129*** -0.637 0387*" -0.083*** 0.306
(0.022) (0.471) (0.127) (0.030) (0.313)
Elderly male death -0.615 -0.306 0.309 -0.179 0.646
(1.176) (0.246) (0.912) (0.990) (0.974)
Elderly female death 0122*“ 0.290 -0.l66 0.151 0.034
(0.427) (0.233) (1.225) (0.234) (1.018)
Gender of head (male=1) 0468*" 0.336 0662*" 0632*“ 0692*"
(0.143) (0.227) (0.148) (0.203) (0.170)
Age ofthe head 0047*" 0093*" 0038* 0.059" 0.107***
(0.017) (0.030) (0.020) (0.029) (0.025)
Age of the head squared -0.0003* -0.0005* ~0.0002 -0.0001 -0.0006***
(0.0001) (0.0003) (0.0002) (0.0003) (0.0002)
Education ofthe head 0086*“ 0131“" 0.030“ 0257*" 0201*"
(0.011) (0.019) (0.012) (0.018) (0.015)
Distance to tarmac road 0.002 -0.003 -0.001 0026*" 0.004
(km)
(0.006) (0.01 1) (0.006) (0.008) (0.007)
Distance to fertilizer store 0010*" 0.001 0014*" 0.004 0.004
(km)
(0.002) (0.005) (0.003) (0.004) (0.004)
Year dummy for 2000 -0.583*** 0830*" -0.617*** 0830*" 3763*"
(0.107) (0.174) (0.117) (0.167) (0.145)
Year dummy for 2004 -0.296*** -0.534*** -0.372*** 0.418" 3329*"
(0.085) (0.168) (0.1 10) (0.166) (0.142)
AEZ dummies Yes Yes Yes Yes Yes
Constant 6805*" -0.907 5641*” -2.400*** -2.313***
(0.540) (0.936) (0.618) (0.883) (0.763)
Observations 4157 4157 4157 4157 4157
R-squared 0.13 0.17 0.06 0.11 0.30

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004
Notes: Robust standard errors in parentheses; * significant at 10%; *"‘ significant at 5%; *** significant at

1%

133

Natural logarithm of value of:

Table 3-18 The impact of working-age adult mortality on value of assets for years before
and after death in the period 1997-2004

 

 

 

Total assets Cattle Small Farm Other
Explanatory variables animals equipment assets
(1) (2) (3) (4) (5)
Ever afflicted 0.053 0.044 -0.077 -0.033 -0.180
(0.150) (0.278) (0.181) (0.253) (0.229)
Two years before death -0.875 -0.300 -0.020** -0.019 0.210
(0.692) (0.233) (0.010) (0.934) (0.652)
One year before death 0.820 -0.672 -0.274** -0.350 0.185
(0.926) (1.338) (0.1 10) (0.261) (0.739)
Year of death 0.719 0.545 -0.465 -0.006"‘ 0.017
(0.455) (1.031) (0.659) (0.003) (0.012)
One year since death 0.025 0.855 -0.293 -0.321 0.095
(0.018) (0.821) (0.246) (0.722) (0.420)
Two years since death -0.046 -0.230 0.052“ -0.078* -0.093*
(0.731) (1.204) (0.025) (0.045) (0.050)
Three years since death -0.612 0.859 0.016 -0.012 -0.088*
(0.886) (1.078) (0.015) (0.865) (0.049)
Four years since death -0.425 0.411 -0.008 0.356 -0.465
(0.525) (0.977) (0.700) (0.762) (0.641)
Five-seven years since death -0.503 -0.450 -0.293 -0.493 -0.360
(0.563) (0.878) (0.670) (0.693) (0.551)
Elderly male death -0.826 -2.189 -0. 133 -0.123 0.435
(1.199) (1.471) (0.888) (0.923) (0.944)
Elderly female death 0.734 0.166 -0.547 0.273 -0.269
(0.463) (0.234) (1.301) (0.266) (1.018)
Gender of head (male=1) 0484*” 0.345 0677*" 0641*” 0.709***
(0.143) (0.227) (0.148) (0.203) (0.170)
Age ofthe head 0049*" 0095*" 0.039" 0.059“ 0108*“
(0.017) (0.030) (0.020) (0.029) (0.025)
Age of the head squared -0.0003* -0.0005* -0.0002 -0.0001 -0.0007***
(0.0001) (0.0003) (0.0002) (0.0003) (0.0002)
Education of the head 0085*“ 0131*" 0.028" 0258*" 0200*"
(0.011) (0.019) (0.012) (0.018) (0.015)
Distance to tarmac road 0.002 -0.004 -0.001 0026*" 0.005
(km)
(0.006) (0.01 1) (0.006) (0.008) (0.007)
Distance to fertilizer store 0010*" 0.001 0.014*** 0.004 0.004
(km)
(0.002) (0.005) (0.003) (0.004) (0.004)
Year dummy for 2000 -0.573*** -0.818*** -0.610*** 0837*" 3764*"
(0.108) (0.175) (0.1 18) (0.169) (0.146)
Year dummy for 2004 -0.306*** -0.551*** -0.381*** 0.414” 3329*“
AEZ dummies Yes Yes Yes Yes Yes
(0.085) (0.169) (0.1 10) (0.167) (0.143)
Constant 6448*“ -1.262 4983*" -2.239** -1.843**
(0.631) (1.124) (0.789) (1.032) (0.868)
Observations 4157 4157 4157 4157 4157
R-squared 0.13 0.17 0.06 0.12 0.30

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

134

Notes: Robust standard errors in parentheses;*significant at 10%;**significant at 5%;***significant at 1%

Significant declines in the value of small animals begin in'the two years before
death, but there are no significant effects on the value of cattle, farm equipment and other
assets. This suggests that households may sell assets because of increased expenditures
(medical and otherwise) and the financial loss they experience as a result of labor
diverted from economically productive activities to care giving. The decline during one
year before death is greater than that in the period of two years before death. This is
consistent with the increased financial burden as the disease becomes more debilitating.
The value of small animals increases in the second year after death but beyond this time,
there is no clear pattern on the value as time passes by. The rebound in the value of small
animals is a sign of recovery.

Significant reductions in the value of farm equipment emerge during the year of
death and are also observed two years since death. However, in successive years, there is
no clear pattern on how the value of farm equipment changes. It is likely that farm
households rely heavily on remittances and non-farm income to capitalize their farm
operations (Reardon et al, 1995) and maintain farm equipment. Consequently, mortality
of working-age adults who may be the major sources of remittances and non-farm
income diminishes the households’ ability to maintain the number and quality of existing
farm equipment as well as purchase new farm equipment.

The value of other assets (mainly durable assets like radios and TVs) declines in
the second and third years after death. This suggests that households are coping with the
financial capital effects of adult death through the sale of stores of value such as

household goods.

135

There are no significant effects of death on value of cattle during the period
running from two years before death and up to seven years after death. Therefore,
findings from this study suggest that one coping strategy adopted by households facing
illness and subsequent death is the sale of small animals and stores of value. This is
consistent with findings from other studies. For instance, Topouzis and du Guemy
(1999), show that most of the households are responding to illness and death in the short-
term by disposing assets that can be easily replaced. In addition, households seem to be
holding on to their cattle and rely on the sale of small animals (Yamano and Jayne, 2004)

in the face of illness and mortality.

Impacts of working-age adult mortality on assets by the deceased ’s gender and position
in household

In the preceding section, we find that working-age adult mortality has significant
effects on household assets even in the years following death. In this section, I examine
whether these effects depend on the gender and position of the deceased household
member. The results are presented in Table 3-19 and they show that the impacts of adult
mortality on total assets, cattle, farm equipment and other assets does not vary by gender
and position of the deceased person. However, households experiencing a male head
death compared to any other death, face a reduction in the value of small animals, one
year before death. In terms of recovery, a male head death is associated with a smaller

recovery in small animals compared to other deaths in the second year afier death.

136

Table 3-19 The impact of working-age adult mortality on assets by gender and position of
the deceased, for years before and after death in the period 1997-2004 (OLS with AEZ

dummies)

 

Natural logarithm of value of:

 

 

 

Small Farm Other

Explanatory variables Total assets Cattle animals equipment assets

(1) (2) (3) (4) (5)
Ever afflicted 0.050 0.075 -0.068 -0.070 -0.194

(0.149) (0.277) (0.180) (0.252) (0.227)
Two years before death -0.320 -0.269 -0.953 -0.083 -0.108

(0.233) (0.247) (0.922) (1.069) (0.748)
One year before death 0.835 -0.057 0.182 0.212 0.428

(1.131) (1.728) (0.233) (0.481) (1.089)
Two years before 0.291 -0.006 -2.032 0.094 0.390
death‘male head dummy

(1.600) (0.787) (1.572) (0.358) (1.459)
One year before death*male -0.154 -0.197 -0.115* 0.140 0.095
head dummy

(0.723) (0.503) (0.065) (0.486) (0.213)
Year of death -0.410 -0.025 0.388 0.094 0.267

(0.635) (1.504) (0.706) (0.988) (0.599)
One year since death 0.411 0.013 0.747 0.543 0.800

(0.386) (1.122) (0.553) (0.965) (0.493)
Two years since death -0.361 -0.102 0.102* -0.847 -0.119

(0.902) (0.419) (0.058) (1.201) (0.853)
Three years since death -0.747 0.273 0093* 0.322 0.209

(1.119) (1.649) (0.050) (1.304) (1.031)
Four years since death -0.945 -0.l43 0.061 1.071 -0.159

(0.743) (1.168) (0.093) (0.860) (0.779)
F ive-seven years since death -0.016 -0.245 -0.651 -0.776 -0.131

(0.655) (1.085) (0.804) (0.848) (0.632)
Year of death*male head 0.1 17 0.524 -l.127 0.561 -0.625
dummy

(0.781) (2.114) (1.243) (1.632) (1.322)
One year since death*male 0.182 0.140 0.439 -0.128 -0.471
head dummy

(0.810) (0.236) (1.084) (1.255) (1.102)
Two years since death‘male 0.983 0.159 -0.067* -0.187 024]
head dummy

(1.679) (0.469) (0.039) (0.453) (0.270)
Three years since 0.324 -0.237 -0.047 -0.l84 -0.120
death*male head dummy

(1.855) (2.381) (0.090) (2.028) (0.083)
Four years since death"‘male 0.011 1.270 -0.1 17 -0.048 0.052
head dummy

(0.028) (2.010) (0.427) (0.062) (1.493)
Five-seven years since -0.024 0.304 -0.031 0.477 -0.107
death‘male head

(0.016) (1.750) (0.051) (1.561) (1.141)
Elderly male death -0.800 -0.032 0.101 -0.572 0.233

(1.140) (0.059) (0.861) (0.951) (0.939)
Elderly female death 0.664 0.359 -0.516 0.272 -0.386

(0.551) (0.435) (1.362) (0.228) (1.049)

 

137

Table 3-19 continued

Naturallflarithm of value of:

 

 

Small Farm Other
Explanatory variables Total assets Cattle animals equipment assets
Gender of head (male=l) 0.502*** 0.364 0689*“ 0619*" 0694*“
(0. 144) (0.229) (0.149) (0.204) (0.171)
Age of the head 0049*“ 0096*" 0.040" 0.059" 0107*"
(0.017) (0.030) (0.020) (0.029) (0.025)
Age of the head squared -0.0003** -0.0005* -0.0002 -0.0001 -.0006***
(0.0001) (0.0003) (0.0002) (0.0003) (0.0002)
Education of the head 0085*" 0130*" 0.028" 0259*" 0200*"
(0.011) (0.019) (0.012) (0.018) (0.015)
Distance to tarmac road (km) 0.002 -0.003 -0.000 0026*" 0.004
(0.006) (0.01 1) (0.006) (0.008) (0.007)
Distance to fertilizer store 0010"" 0.001 0.014*** 0.004 0.004
(km)
(0.002) (0.005) (0.003) (0.004) (0.004)
Year dummy for 2000 -0.573*** -0.815*** -0.610*** 0836*" 3762*"
(0.108) (0.175) (0.118) (0.169) (0.146)
Year dummy for 2004 -0.307*** -0.545*** -0.380"'** 0.405" 3330*"
(0.085) (0.169) (0.1 10) (0.167) (0.143)
AEZ dummies Yes Yes Yes Yes Yes
Constant 6714*" -l.012 5537*" -2.369*** 2310*"
(0.540) (0.936) (0.618) (0.887) (0.763)
Observations 4157 4157 4157 4157 4157
R-squared 0.13 0.17 0.06 0.12 0.30

 

Source: Tegemeo Institute (N airobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004
Notes: Robust standard errors in parentheses;*significant at 10%;"significant at 5%;***signif1cant at 1%

138

3.7 .5 Impacts of working-age adult mortality on household income

Table 3-20 presents the results on the impacts of working-age adult mortality on
total, crop, livestock and off-farm income. Adult deaths lead to a decrease in livestock
and off-farm income.

Other factors influence the level of household income. Male—headed households
and those with more educated heads have higher incomes. The age of the head has a non-
linear effect on total and livestock income only. Also, the distance to fertilizer store is
negatively related to off-farm income, suggesting that it is harder to find off-farm income
opportunities farther away from towns and trading centers.

The results in Table 3-21 show how death effects on income vary from two years
before death up to seven years after death. Significant reductions in livestock and off-
farm income begin in one year before death. This may be due to the loss of productive
labor time for the sick, potential reduction in hours worked off-farm by the care givers,
and sale of livestock assets.

There is a significant drop in livestock and off—farm income in the year of death,
reflecting the additional burden of lost income experienced by households as the sick
adult dies. In the period one to three years afier death, there is no significant change in
income. However, four years after death, we find a significant increase in livestock and
off-farm income. Also, five to seven years after death, off-farm income increases. This
recovery in livestock income may be due the ability of households to restock their small
animals after death (Table 3-18). The post-death increase in off-farm income may be

consistent with a situation where afflicted households experience a large financial loss as

139

Table 3-20 The impact of working-age adult mortality on income for the period 1997-
2004 (OLS with AEZ dummies)

 

Natural logarithm of income:

 

 

 

Explanatory variables Total Crop Livestock Off-farm
(1) (2) (3) (4)
Ever afflicted -0.061 -0.047 0.131 -0.237
(0.084) (0.137) (0.208) (0.260)
Working-age adult death -0.011 -0.034 -0.657* -0.274**
(0.1 16) (0.204) (0.394) (0.110)
Elderly male death -0.455 -0.330 0.534 0.613
(0.305) (0.327) (1.344) (0.670)
Elderly female death 0.191 0.432 -0.716 -2.926
(0.348) (0.488) (1.898) (2.495)
Gender of head (male=1) 0387*" 0415*" 0569*“ 0.508“
(0.067) (0.105) (0.183) (0.197)
Age of the head 0030*" 0.021 0.097*** 0.045
(0.008) (0.016) (0.026) (0.030)
Age of the head squared -0.0002*** -0.0001 -0.0007*** -0.0004
(0.0001) (0.0002) (0.0002) (0.0003)
Education of the head 0077*" 0058*" 0064*” 0222*"
(0.005) (0.009) (0.016) (0.017)
Distance to tarmac road (km) 0.003 0.006 -0.000 0.002
(0.003) (0.005) (0.007) (0.007)
Distance to fertilizer store -0.003 0.002 -0.006 0012‘"
(km)
(0.002) (0.002) (0.005) (0.004)
Year dummy for 2000 0227*“ 0900*" -0.573*** 0992*"
(0.051) (0.088) (0.139) (0.166)
Year dummy for 2004 0236*" 0498“" -0.891*** 0554“"
(0.050) (0.083) (0.141) (0.169)
AEZ dummies Yes Yes Yes Yes
Constant 10.300"* 7098*" 2967*“ 8383*"
(0.246) (0.484) (0.777) (0.871)
Observations 4157 4157 4157 4157
R-sgared 0.17 0.21 0.09 0.10

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

**"‘ significant at 1%

140

Notes: Robust standard errors in parentheses; "' significant at 10%; ** significant at 5%;

Table 3-21 The impact of working-age adult mortality on income for years before and
after death in the period 1997-2004 (OLS with AEZ dummies)

 

Natural logarithm of income:

 

 

 

Explanatory variables Total Crop Livestock Off-farm
(1) (2) (3) (4)
Ever afflicted -0.060 -0.047 0.126 -0.240
(0.084) (0.137) (0.208) (0.260)
Two years before death -0.165 -0.413 -0.179 -0.105
(0.217) (0.447) (0.132) (0.135)
One year before death 0.248 -0.472 -0.007* -0.033**
(0.265) (0.570) (0.004) (0.016)
Year of death -0.175 -0.155 -0.088* -0.213*
(0.224) (0.347) (0.049) (0.1 18)
One year since death -0.l38 0.228 -0.026 -0.573
(0.173) (0.352) (0.765) (0.699)
Two years since death -0.268 0.581 -0.332 -0.372
(0.250) (0.595) (1.120) (0.326)
Three years since death -0.272 0.225 -0.763 0.121
(0.198) (0.272) (0.799) (0.235)
Four years since death -0.006 -0.230 0.109M 0.005*
(0.224) (0.344) (0.044) (0.003)
Five-seven years since death -0.214 0.034 -0.883 0.095“
(0.177) (0.346) (0.768) (0.03 8)
Elderly male death -0.476* -0.346 0.404 0.521
(0.271) (0.362) (1.159) (0.661)
Elderly female death 0.180 0.450 -0.858 -3.053
(0.325) (0.509) (1.462) (2.417)
Gender of head (male=1) 0390*" 0415*" 0576*" 0523*"
(0.067) (0.105) (0.183) (0.197)
Age of the head 0030*" 0.022 0101*“ 0.047
(0.008) (0.017) (0.026) (0.030)
Age of the head squared -0.0002"'** -0.0001 -0.0007*** -0.0004
(0.0001) (0.0002) (0.0002) (0.0003)
Education of the head 0077*" 0057*" 0065*" 0222*"
(0.005) (0.009) (0.016) (0.017)
Distance to tarmac road (km) 0.003 0.006 0.000 0.003
(0.003) (0.005) (0.007) (0.007)
Distance to fertilizer store (km) -0.003 0.002 -0.006 -0.012***
(0.002) (0.002) (0.005) (0.004)
Year dummy for 2000 0231*" 0902*" 0548*" 1003*"
(0.052) (0.088) (0.140) (0.167)
Year dummy for 2004 -0.240"* 0495*" -0.921*** 0543*“
(0.050) (0.083) (0.142) (0.170)
AEZ dummies Yes Yes Yes Yes
Constant 10.372"* 7018*" 2856*" 8726*”
(0.301) (0.587) (0.956) (1.031)
Observations 4157 4157 4157 4157
R-squared 0.17 0.21 0.10 0.11

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004

*** significant at 1%

141

Notes: Robust standard errors in parentheses; "‘ significant at 10%; ** significant at 5%;

a result of medical and funeral expenses, lost“ income of the deceased member and labor
diverted from economically productive activities to care giving. Consequently, some
surviving members may participate in off-farm activities in order to restore the lost
income streams. Participation in such activities seems a reasonable strategy for afflicted
households since it provides an avenue to generate faster returns and income that is

required to meet the large and urgent financial needs.

Impacts of working-age adult mortality on income by the deceased ’s gender and
position in household

Results in Table 3-22 show that over time, the impacts of adult mortality don’t
generally vary by gender and position of the deceased individual. However, compared to
other deaths, households that experience a male head death show smaller recovery in
livestock income four years after death. In the year of death, the decline in off-farm
income is smaller for households with a male head death, suggesting that non-head adult
members may be the major sources of off-farm income. Also, five-seven years after
death, households that experience death other than that of the male head have better
recovery in off-farm income. Therefore, surviving members of afflicted households turn
to off-farm activities as a means to replace lost income streams due to morbidity and
mortality of working-age adults. As time passes, more non-head members of the family

become economically active and engage in off-farm activities.

142

Table 3-22 The impact of working-age adult mortality on income by gender and position
of the deceased for years before and afier death in the period 1997-2004 (OLS with AEZ

dummies)

 

Natural logarithm of income:

 

 

 

Explanatory variables Total Crop Livestock Off-farm
(I) (2) (3) (4)
Ever afflicted -0.062 -0.071 0.177 -0.201
(0.071) (0.122) (0.216) (0.249)
Two years before death 0.099 -0.423 -0.057 -0.107
(0.334) (0.574) (0.039) (0.297)
One year before death 0.377 -0.430 -0.010 -0.085
(0.444) (0.763) (0.007) (0.927)
Two years before death‘male head dummy -0.178 0.041 -0.246 -0.379
(0.765) (1.313) (2.333) (2.688)
One year before death*male head dummy -1 . 197 -0.311 -0.071 -0.060
(1.018) (1.748) (0.596) (0.108)
Year of death 0.424 0.126 -0.125 -0. 165*
(0.342) (0.587) (0.081) (0.086)
One year since death -0.062 0.007 -0.097 0.032
(0.274) (0.471) (0.836) (0.963)
Two years since death -0.246 0.295 -0.166 -0.123
(0.377) (0.646) (1.148) (0.203)
Three years since death -0.097 0.040 0.135 -0.l88
(0.394) (0.676) (0.169) (0.237)
F our years since death 0.104 -0.287 0.105" -O.340
(0.326) (0.560) (0.046) (1.147)
Five-seven years since death -0.196 0.343 0.061 0.097*
(0.272) (0.467) (0.093) (0.055)
Year of death*male head dummy -0.401 0.011 -0. 187 0.015“
(0.530) (0.911) (0.117) (0.009)
One year since death‘male head dummy 0.273 0.593 -0.036 -1.612
(0.650) (1.1 15) (1.982) (2.283)
Two years since death*male head dummy 0.809 0.962 0.958 0.381
(0.854) (1 .466) (2.605) (0.767)
Three years since death*male head dummy -0.443 0.577 -3.520 0.134
(0.723) (1.242) (2.206) (2.542)
Four years since death‘male head dummy -0.325 -0.082 -0.007* -0.507
(0.61 1) (1.049) (0.004) (2.147)
Five-seven years since death*male head 0.163 -1.013 0.103 -0.043*
dummy
(0.616) (1.057) (0.093) (0.024)
Elderly male death -0.517 -0.314 0.753 0.318
(0.418) (0.718) (1.275) (1.469)
Elderly female death 0.100 0.457 -0.645 -0.023
(0.504) (0.865) (1.538) (0.061)
Gender of head (male=l) 0382*" 0406*" 0539*" 0.489"
(0.056) (0.096) (0.171) (0.197)
Age of the head 0031*" 0.021 0099"" 0.045
(0.008) (0.014) (0.025) (0.029)
Age of the head squared -0.0002*" -0.0001 -0.0007*** -0.0004
(0.0001) (0.0001) (0.0002) (0.0003)
Education of the head 0077*" 0058*" 0062*" 0224*“
(0.005) (0.009) (0.016) (0.018)

 

143

Table 3-22 continued

Natural logarithm of income:

 

 

 

Explanatory variables Total Crop Livestock Off-farm
(1) (2) (3) (4)
Distance to tarmac road (km) 0004* 0.006 0.000 0.005
(0.002) (0.004) (0.007) (0.008)
Distance to fertilizer store (km) -0.003**"‘ 0.002 -0.006 -0.013***
(0.001) (0.002) (0.004) (0.005)
Year dummy for 2000 0225*" 0914*" -0.561 *** 0978*"
(0.046) (0.079) (0.139) (0.161)
Year dummy for 2004 -0.257*** 0497*" -0.945*** 0498*"
(0.047) (0.080) (0.143) (0.165)
AEZ dummies Yes Yes Yes Yes
Constant 10.258*** 7.105*** 2946*" 8354*”
(0.251) (0.431) (0.766) (0.883)
Observations 4157 4157 4157 4157
R-squared 0.18 0.21 0.10 0.11

 

Source: Tegemeo Institute (Nairobi)/Michigan State University Agricultural Monitoring

and Policy Analysis Household Surveys in 1997, 2000, 2002 and 2004
Notes: Robust standard errors in parentheses; "' significant at 10%; ** significant at 5%; *** significant at

1%

3.8 Conclusion and policy recommendations

3.8.1 Summary of main findings

This study has examined the dynamics of the impacts of working-age adult

morbidity and mortality on household composition, total cultivated area, area cultivated

by crop types, value of crop production, assets and income for a period of time extending

from two years before death and up to seven years after death. In addition, it has

examined whether the temporal pattern of the mortality effects differs by the gender and

position of the deceased individual. The main contribution of this work is that it reveals

more information about the dynamics of household responses and the longer-run impacts

of mortality on household outcomes than any prior quantitative study based on household

survey data. Moreover, it provides much greater detail than past studies about how

144

impacts of mortality depend on the gender and position of the deceased over time. Lastly,
the study is based on a nationally representative panel dataset and hence provides an
understanding of the dynamic effects of adult mortality from an area of diverse rates of
HIV/AIDS. Therefore, it complements earlier instructive studies from particular areas
known to suffer from high rates of HIV/AIDS.

The essay highlights six key findings that will better inform intervention policies
and strategies. First, working-age adult death affects various household outcomes. For
instance, it is associated with changes in household structure, reduction in value of small
animals and farm equipment, as well as a decline in livestock and off-farm income.

When we take into account the dynamics of prime—age death, the second finding
is that small negative impacts of working-age adult mortality begin to emerge in the
period preceding death, presumably due to pre-death morbidity. Third, significant
negative impacts are observed in the year of death. Fourth, mortality impacts are not just
a one-time post-death adjustment but rather they persist over time. The fifth conclusion is
that in most cases, negative effects are larger in the period closest to the death event,
suggesting that households are somewhat able to recover over time. However, we note
that afflicted households are less able to recover from the impacts of death on area under
high-value crops.

Sixth, impacts of adult mortality differ by the gender and position of the deceased
adult. Overall, changes in household composition depend to a limited extent on the
gender and position of the deceased member. When a male head dies, boys and young

children are more likely to be fostered out owing to the loss of the main bread winner.

145

Households afflicted by male head deaths as opposed to other adult deaths, face a large
reduction in total cultivated land and area devoted to cereals and high-value crops. Three
years after death, households afflicted by the death of male heads have greater value of
crop output per acre, compared to households that experienced other adult deaths.
However, in the period five-seven years since death, the opposite effect is observed.

One coping strategy adopted by households facing illness and subsequent death is
the sale of small animals. Households with a male head death compared to any other
death face a reduction in the value of small animals one year before death, and have a
smaller recovery in small animals two years after death.

In terms of income, households that experience a male head death show smaller
recovery in livestock income four years after death, smaller decline in off-farm income in
the year of death, and smaller recovery in off-farm income five-seven years after death. It
appears that surviving members of afflicted households turn to off-farm activities as a
means to replace lost income streams due to morbidity and mortality of working-age

adults.

3.8.2 Implications for mitigation and poverty reduction programs

Findings from this study have a number of implications for the appropriateness of
existing programs and interventions meant to mitigate the effects of adult mortality as
well as the design of future interventions. First, for the mitigation efforts to succeed there
is need to recognize that illness preceding death and timing of death matter. In particular,
since mortality impacts are not just a one-time post-death adjustment, blanket
recommendations for interventions should not be made for all afflicted households.

Instead, interventions should be made while taking into account the time that has elapsed

146

since death. In most cases, the impact of death is largest in the period closest to death,
which suggests that policy interventions should be made in the short-term when the
impact is greatest. Although most households will generally recover after some years,
early interventions as well as interventions on a descending scale in the short to medium-
terrn will help afflicted households recover faster and with more case.

Secondly, given that illness before death has negative impacts on household
outcomes, there may be gains from policies that reduce the burden of taking care of the
sick at home and the cost of treatment. The provision of universal health care and life
insurance, free medical care for AIDS-patients and improved community health care
systems, may help afflicted households deal with the burden of illness without selling
their small animals or compromising their agricultural incomes.

Third, the finding that the decline in cultivated area in total and by crop types is
greater for households that experience a male head death compared to all other deaths
raises concern for the welfare of widows, particularly with respect to their food security
position. Therefore, policy makers need to take into account that gender and position of
the deceased matters, a recognition that may lead to the provision of targeted
interventions for the more vulnerable widow-headed households. For example, direct
food aid as well as targeted inputs can be given to widow-headed households to minimize
effects on food security. Also, fostering women farmers’ participation in programs such
as extension activities and farmer field schools could reduce the impact of death on
agricultural output and income for the widows’ households. In addition, this finding
raises the issue of potential gender inequalities that exist in terms of access to and control

over land and other productive resources. Although it is generally observed that women

147

devote more labor time to agriculture than men, households experiencing the death of a
male household head face a large reduction in total cultivated land and area devoted to
cereals and high-value crops. Therefore, efforts to mitigate the impacts of adult mortality
and improve household livelihoods should take into account such gender inequalities.
Also, there is need to encourage cultural changes that empower widows to retain access
to land and assets after the death of their husbands. Introduction of such changes may be
an important component of poverty alleviation strategies. However, because this may
upset the traditional norms of property inheritance, it will arguably require careful and
sustained campaigns to garner support from the local communities.

The fourth implication relates to the lack of evidence of a shift toward labor-
saving crops but a modest tendency to shifi away from capital-intensive high-value crops.
This suggests that cash constraints may be more important than labor constraints for most
afflicted households. Easy access to affordable credit will enable afflicted households to
maintain cultivated area under high-value crops. Therefore, rather than overly prioritizing
labor-saving crop and input technologies, there may be great gains arising from the
removal of policies and distortions that bring about great disparity in access to farm
credit. For instance, support by government, donor agencies and the private sector for
input provision schemes and inter-linked credit arrangements may be more effective in
helping afflicted households cope with death faster and in more robust ways that do not
threaten their future livelihood. Providing afflicted households with short-term subsidized
input credit for fertilizers, seed, and cash for hiring labor will enable them to maintain
cultivated area under hi gh-value crops in the short-term. Further research that addresses

this issue will provide useful information on how access to credit under appropriate

148

institutional structures and arrangements can be helpful in the recovery process by
creating opportunities for financing farm operations.

Fifth, the finding that in- and out-migration of individuals is a potential coping
strategy suggests that more community resources can be mobilized to care for afflicted
households and help them develop other effective coping mechanisms. For instance,
more linkages across households can be developed in terms of labor-support groups, cash
or in-kind remittances as coping mechanisms. However, it has been argued that
increasing deaths and illness due to HIV/AIDS may lead to a breakdown of social capital
and local institutions, particularly in hard-hit communities. Therefore, there is need for
government and donor agencies support to make such community initiatives sustainable.

Sixth, a decline in area cultivated and off-farm income may imply a reduction in
household food supply and the share of food expenditures. If higher food shares translate
into better nutrition and health status, the lack of adequate nutrition will comprise
members’ immunity and labor quantity and/or quality. This may increase the chances of
more disease-related mortality and reduce agricultural productivity. Therefore,
HIV/AIDS mitigation programs must include improvements in food security. Food aid
and community based mutual help groups can provide income and boost nutrition. While
this may provide relief temporarily, there is need to integrate HIV/AIDS prevention and
mitigation into agricultural extension systems, development projects/programs and
microfinance. For instance, increased agricultural productivity will lower the cost of
food and contribute to improved food security and better health.

Seventh, we find that afflicted households suffer a loss of off-farm income one

year before death and during the year of death. This is probably due to the loss of

149

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) Confionting 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.

Allen, Susan, Philippe Van de Perre, A. Srufilira, et al. (1991). Human
immunodeficiency Virus and malaria in representative sample of childbearing
women in Kigali, Rwanda. Journal of Infectious Diseases 164267-71.

Anarfi, J .K. (1995). The condition and care of AIDS victims in Ghana: AIDS sufferers
and their relations. Pp. 253-263, in the Third World AIDS Epidemic, ed. 10.
Orubuloye, J .C. Caldwell, P. Caldwelland S. Jain. Supplement to Health
Transition Review 5. Canberra: Australian National University.

Ankrah, E.M., M. Lubega and S. Nkumbi. (1992). Stress and coping among rural families
in Uganda.Paper presented at Eighth International Conference on AIDS/Third
STD World Congress,Amsterdam, 19-24 July.

Annim, Samuel Kobina and Awusabo-Asare, Kofi. (2006). Wealth status and risky
sexual behaviour in Ghana and Kenya, University of Cape Coast, Ghana.

Ashenfelter, Orley (1978). Estimating the effect of training programs on earnings, Review
of Economics and Statistics: 60:47-57

Barnett, T. (1994). The effects of HIV/AIDS on farming system and rural livelihood in
Uganda, Tanzania and Zambia, FAO, Rome.

Barnett T., and Blaikie, P.M.(1992). Aids in Africa: It’s Present and Future Impact.
London, UK, Belhaven Press.

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
Africa: 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 Twenty-First Century: Disease and
Globalization. Palgrave Macmillan, Basingstoke, UK.

151

Beegle, K. (1997). The Impact of Adult Mortality in Agricultural Households: Evidence
from Rural Tanzania, PhD Dissertation, Department of Economics, Michigan
State University, East Lansing, Michigan.

Beegle, K. (2005). Labor Effects of Adult Mortality in Tanzanian Households, Economic
Development and Cultural Change, Vol. 53 No. 3: 655-683

Beegle, Kathleen, Joachim De Weerdt, and Stefan Dercon (2006). Adult Mortality and
Economic Growth in the Age of HIV/AIDS. World Bank Working Paper Series:

Draft.

Bell, Clive, Shanta Devarajan and Hans Gersbach. (2003). The Long-run Economic Costs
of AIDS: With an Application to South Africa. World Bank Policy Research
Working Paper No.. 3152.

Benjamin, Dwayne. (1992). Household Composition, Labor Markets, and Labor
Demand: Testing for Separation in Agricultural Household Models,”
Econometrica, Vol. 60, no. 2, pp 287-322.

Berkley, S., W. Naamara, S. Okware, R. Downing, J. Konde-Lule, M. Musagara and S.
Musgrave. (1990). AIDS and HIV infection in Uganda - are more women infected
than men? AIDS 4,12:1237-1242.

Bledsoe, Caroline. (1994). Children are like young bamboo trees: Potentiality and
reproduction in Sub-Saharan Africa, in: K. Lindahl-Kiessling and H. Landberg,
(eds).

Brockerhoff M. and Biddlecom A.E., (1999). Migration, sexual behaviour, and risk of
HIV in Kenya. International Migration Review 33: 833 - 856.

Bryceson, D. F ., and J. Fonseca. (2006). Risking death for survival: Peasant responses to
hunger and HIV/AIDS in Malawi. World Development 34(9), pp. 1654-1666.

Central Bureau of Statistics (CBS) [Kenya], (2004). Ministry of Health (MOH) [Kenya],
ORC Macro 2004. Kenya demographic and health survey 2003. Calverton,
Maryland: CBS, MOH, and ORC Macro, 2003:
http://www.measuredhs.com/pubs/pdfioc.cfm?ID=462 (accessed March 10,
2005).

Chapoto, Antony. (2006). The Impact of Aids-Related Prime-Age Mortality on Rural
Farm Households: Panel Survey Evidence From Zambia, PhD Dissertation,
Department of Agricultural Economics, Michigan State University, East Lansing,
Michigan.

152

Chapoto, Antony and TS. Jayne. (2008). Impacts of AIDS-Related Mortality on Farm
Household Welfare in Zambia, Economic Development and Cultural Change,
Volume 56, Number 2.

Chiappori, Pierre Andre. (1988). Rational Household Labor Supply, Econometrica, 56,
63-90.

Cox, DR. (1972). Regression Models and Life-Tables, Journal of the Royal Statistical
Society, 34 (Series B): 187-202.

Cuddington, J .T. (1993). Modeling the macroeconomic effects of AIDS, with an
application to Tanzania. World Bank Economic Review, vol. 7, pp 173-189.

Cuddington, J ., and Hancock, J. (1995). The macro-economic impact of AIDS in Malawi:
a dualistic, labor-surplus economy. Journal of African Economics, 4, 1-28

Deheneffe, Jean-Claude, Michel Caraél, and Amadou Noumbissi (1998). Socioeconomic
determinants of sexual behaviour and condom use in; Ainsworth, M, Fransen, L.
Over, M (eds) Confronting AIDS: Evidence for Developing World, European
Commission, and Brussels.

de Walque, Damien (2002). How does educational attainment affect the risk of being
infected by HIV/AIDS? Evidence from Uganda and Zambia. The University of

Chicago.

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.

Deaton, A. (2002). Policy Implications of the Gradient of Health and Wealth. Health
Affairs 21 (2): 13—30.

Desmond, D., (2001). Joint Epidemics: Poverty and AIDS in Sub-Saharan Africa.
Harvard International Review 23(3): 54-58.

Donovan, C., L. Bailey, E. Mpisi, and M. Weber (2003). Prime-Age Adult Morbidity
and Mortality in Rural Rwanda: Effects on household Income, Agricultural
Production and Food Security Strategies. Research report, Food Security Research
Project, Ministry of Agriculture, Livestock, and Forestry, Kigali.

Drimie, S. (2002). HIV/AIDS and Land: Case Studies from Kenya, Lesotho, and South
Africa. Report for Southern Africa Regional Office of the Food and Agricultural
Organization (FAO), Human Sciences Research Council, Pretoria, South Africa.

Duleep H0. (1986). Measuring the effect of income on adult mortality using longitudinal
administrative record data. The Journal of Human Resources, Vol. 21, No.2.

153

Edmonds, Eric and Douglas Miller. (2001). Cash Constraints and the Migration of the
Elderly in South Africa. Work in progress.

Epstein, H. 2003. AIDS in South Afiica: The Invisible Cure. New York Times Review of
Books, 50(12), July 17, 2003.

Eswaran, Mukesh and Kotwal, Ashok. (1989). Credit as Insurance in Agrarian
Economies. Journal of Development Economics, Vol 31(1), pp. 37-53.

Evans D.K. and Miguel, E. (2007). Orphans and schooling in Africa: A longitudinal
analysis. Demography, Volume 44-Number 1, February 2007: 35—5 7

FAO (1994). What has AIDS to do with agriculture? FAO, Rome.

FAO (1995). The effects of HIV/AIDS on farming systems in Eastern Africa. F AO,
Rome.

F A0 (1997). The rural people of Africa confronted with AIDS: A challenge to
development. Summary of F AO studies on Aids. Rome: FAO

FAO (2003) AIDS-a threat to rural Africa. Accessed at
1m://www.fao.org/focus/e/aids/aids1-e.htm

FAO (2004). HIV/AIDS, gender inequality and rural livelihoods. The impact of
HIV/AIDS on Rural Livelihoods in Northern Province, Zambia. FAO, Rome.

Gillespie, Stuart. (2003). HIV/AIDS and Livelihoods. Paper presented at the Technical
Consultation on Measuring Vulnerability in the Light of the HIV/AIDS
Pandemic, 9-11, September 2003, Johannesburg, South Africa.

Gillespie, SR. and Kadiyala, S. (2005). HIV/AIDS and Food and Nutrition Security:
From Evidence to Action. Food Policy Review No. 7, IFPRI, Washington DC.

Greig, F. and Koopman, C. (2003). Multilevel Analysis of Women’s Empowerment and
HIV Prevention: Quantitative Survey Results from a Preliminary Study in
Botswana. AIDS and Behavior 7(2): 195-208; Jun 2003.

Grossman, M., (1972). On the Concept of Health Capital and the Demand for
Health, Journal of Political Economy, Vol. 80:223-255.

Hosegood V, McGrath N, Herbst K & Timaeus, I.M. (2004). The impact of AIDS

mortality on household dissolution and migration in rural South Africa. AIDS
18(11):1585-1590.

154

INDEPTH. (2002). INDEPTH Mortality Patterns for Africa, in Population, Health, and
Survival at INDEPTH Sites, vol. 1, Population and Health in Developing
Countries, edited by INDEPTH. Ottawa: IDRC Press.

Jacobson, Louis, S; LaLonde, Robert J; Sullivan, Daniel G (1993). Earning Losses of
displaced workers, The American Economic Review: 83,4:685-709

Kapiga S.H. et a1. (1994). Risk factors for HIV infection among women in Dar-es-
Salaam, Tanzania, Journal of Acquired Immune Deficiency Syndromes: 7: 301—
309.

Kitange H.M,. Machibya, and J. Black. 1996. Outlook for survivors in sub-Saharan
Africa: Adult mortality in Tanzania, British Medicine Journal, PubMed Abstract,
312: 216-220.

Kochar, Anjini.(1995). Explaining household vulnerability to idiosyncratic income
shocks, The American Economic Review, Vol.85, No.2.

Lundberg, M., M. Over, and P. Mujinja. (2000). Sources of financial assistance for
households suffering an adult death in Kagera, Tanzania, World Bank,
Washington DC.

Menon, R, M.Wawer, J. K. Konde-Lule, N. Sewankambo, and C. Li (1998). The
Socioeconomic Correlates of HIV Infection among Household Heads in Rakai
District, Uganda. In: Confionting AIDS: Public Priorities in a Global Epidemic.
New York: Oxford University Press.

Morduch, Jonathan. (1990). Risk, Production and Saving: Theory and Evidence from
Indian Households, Mimeo, Harvard University.

Mutangadura, G, and D. Webb (1999). Socio-Economic Impact of Adult Mortality and
Morbidity on Urban Households in Zambia. AIDS Analysis Afiica 9(4).

Ndongko, Theresa M. (1996). A preliminary study of the socio-economic impact of
HIV/AIDS in Africa. Africa Development 11,1:39-55.

National AIDS and ST] Control Programme (N ASCOP)/ Ministry of Health (2005).
AIDS in Kenya: Trends, Intervention and Impact, 7th ed. NASCOP, Nairobi.

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.

Nunn, P. (1989). Acquired Immunodeficiency Syndrome and Human Immunodeficiency
Virus infection in Bujumbura, Burundi. AIDS Technical Bulletin 2:88.

155

Nyaga, R.K.; Kimani, D.N.; Mwabu, G.; Kimenyi, M.S.(2004). HIV/AIDS in Kenya: A
review of research and policy issues, KIPPRA Discussion Paper No. 38

Ntozi, James P. M.; Yovani M. A.; Lubaale and Immaculate M. Nakanaabi. (1997).
AIDS mortality in Uganda: circumstances, factors and impact of death. Health
Transition Review, Supplement to Volume 7, 199: 207-224.

Over, M. (1992). The Macroeconomic Impact of AIDS in Sub-Saharan Africa. AFTPN
Technical Working Paper 3. World Bank, Africa Technical Department,
Population, Health, and Nutrition Division, Washington, DC.

Parikh, A. Shanti. (2007). The Political Economy of Marriage and HIV: The ABC
Approach, “Safe” Infidelity, and Managing Moral Risk in Uganda. American
Journal of Public Health, Vol 97, No. 7

Phillipson, T.J. and RA. Posner (1993). Private Choices and Public Health: The AIDS
Epidemic in an Economic Perspective, Cambridge, MA: Harvard University
Press.

Pitt, M. and Rosenzweig, M. (1986). Agricultural prices, food consumption and the
health and productivity of Indonesian farmers, in Singh, 1., L. Squire and J.
Strauss. Eds (1986). Agricultural Household Models: Extensions, Application
and Policy, Baltimore Johns Hopkins U. Press. USA

Pitt, Mark M., and Mark R. Rosenzweig. (1990). “Estimating the Intrahousehold
Incidence of Illness: Child Health and Gender-Inequality in the Allocation of
Time.” International Economic Review 31, no. 4: 969—89.

Poku, N.K. (2001). Africa’s AIDS crisis in context: ‘how the poor are dying’. Third
World Quarterly 22(2): 191-204.

Reardon, T. (1997). Using evidence of household income diversification to inform study
of rural nonfarm labor market in Afiica. World Development, 25(5):735-47.

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.

Rosenzweig, MR. (1988). Risk, Implicit Contracts and the Family in Rural Areas of
Low-Income Countries, The Economic Journal, Vol. 98, No. 393, pp. 1148-1170.

Ryder, R.W., M. Ndilu. S.E. Hassig e_t a_l. (1990). Heterosexual transmission HIV-1

among employees and their spouses at two large businesses in Zaire. AIDS 4:725-
732.

156

Sachs, J. D. et al. (2001). Macroeconomics and Health: Investing in Health for Economic
Development. Report of the Commission on Macroeconomics and Health, to the
Director-General of the World Health Organization

Shelton, James D., Michael M. Cassell and Jacob Adetunji (2005). Is Poverty or Wealth
at the Root of HIV? The Lancet 366: 1057-1058.

Singer, J. D., & Willett, J. B. (1993). It's about time: Using discrete-time survival
analysis to study duration and the timing of events. Journal of Educational
Statistics.

Singh, 1., L. Squire and J. Strauss. Eds (1986). Agricultural Household Models:
Extensions, Application and Policy. Baltimore Johns Hopkins U. Press. USA.

Stillwaggon, Eileen. (2005). The Ecology of Poverty: Nutrition, Parasites, and
Vulnerability to AIDS, IFPRI conference on HIV/AIDS and Food and Nutrition
Security: From Evidence to Action, April 14-16, 2005, Durban, South Africa.

Strauss, J. and D. Thomas. (1998). Health, Nutrition and Economic Development,
Journal of Economic Literature: 766-817.

Strauss, J .; Gertler, J .P.; Rahman,O.; and Fox, K.(1993). Gender and life cycle
differentials in the patterns and determinants of adult health, The Journal of
Human Resources, V0128, No.4

Stokes, C. S. (2003). Measuring impacts of HIV/AIDS on rural livelihoods and food
security, Sustainable Development Department (SD) F AO.

Topouzis, D. (1994). Uganda: the socio-economic impact of HIV/AIDS on rural families
with an emphasis on youth. Rome: Food and Agriculture Organization.

Topouzis, D. (1998). The implications of HIV/AIDS for Rural Development Policy and
Programming: Focus on Sub-Saharan Africa, Sustainable DeveIOpment
Department, Food and Agriculture Organization, Rome.

Topouzis, D. (2000). Measuring the impacts of HIV/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 (2003, 2004, 2006). AIDS Epidemic Update. GENEVA, SWITZERLAND.

Urassa, M., Boerma, J ., T., Isingo, R., Ngalula, J ., Ng’weshemi, J ., Mwaluko, G., and
Zaba, B. (2001). The impact of HIV/AIDS on mortality and household mortality

157

in rural Tanzania. AIDS, 15, 201 7-2023.

Weinreb, A. Alexander (2004). Convergence of mortality risk across sociodemographic
characteristics in rural Nyanza Province, Kenya, Department of Sociology and
Anthropology Faculty of Social Sciences Hebrew University of Jerusalem,
Working Paper No. 2004-02.

White, Joanna and Robinson, Elizabeth. (2000). HIV/AIDS and rural livelihoods in Sub-
Saharan Africa. Policy series 6. Chatham, UK: Natural Resources Institute.

Wooldridge, J. M., (2002). Econometric Analysis ofCross Section and Panel Data. MIT
Press, Boston.

World Bank. (1999). Confronting AIDS: Public priorities in a global epidemic
A World Bank Policy Research Report. New York: Oxford University Press.

Yamano, T., T. S. Jayne, (2004). Measuring the Impacts of Working-Age Adult Mortality
on Small-Scale Farm Households in Kenya, World Development Vol.32, No.1, pp
91 -1 1 9.

Yamano T. and TS. Jayne, (2005). Working-Age Adult Mortality and Primary School
Attendance in Rural Kenya, Economic Development and Cultural Change,
Vol.53, No.3.

Yamauchi, Futoshi. (2007). Marriage, schooling, and excess mortality in prime-age
adults: Evidence from South Africa, IF PRI Discussion Paper 00691.

Young, Alwyn. (2005). The gift of the Dying: The Tragedy of AIDS and the Welfare of

Future African generations. Quarterly Journal of Econometrics. Vol. 120 No. 2:
423-466

158

 

 

ulglllllgllllljjitilt: