awn IJBRARY
l Michigan State
w" University

This is to certify that the
thesis entitled

A TEST OF THE NEW VARIANT FAMINE HYPOTHESIS:
PANEL SURVEY EVIDENCE FROM ZAMBIA

presented by

NICOLE MARIE MASON

has been accepted towards fulﬁllment
of the requirements for the

 

 

MS. degree in Agricultural Economics

  

 

r / /
erssor’s Signature
ET ArrV'2008

Date

 

MSU is an afﬁrmative-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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 KIProj/Aoc8PrelelRC/DaleDueindd

A TEST OF THE NEW VARIANT F AMINE HYPOTHESIS:
PANEL SURVEY EVIDENCE FROM ZAMBIA

By

Nicole Marie Mason

A THESIS

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

MASTER OF SCIENCE
Department of Agricultural Economics

2008

ABSTRACT

A TEST OF THE NEW VARIANT FAMINE HYPOTHESIS:
PANEL SURVEY EVIDENCE FROM ZAMBIA

By
Nicole Marie Mason

The new variant famine (N VF) hypothesis suggests, inter alia, that I-HV/AIDS is
causing a decline in agrarian livelihoods and that the epidemic is making agrarian
communities more vulnerable and less resilient to drought and other transitory shocks.
The NVF hypothesis has become an important part of the conventional wisdom
surrounding the relationship between HIV/AIDS and food crises in southern Africa;
however, there is a dearth of empirical evidence to support the NVF hypothesis and, to
date, no studies have been speciﬁcally designed to test the NVF hypothesis. This thesis
uses econometric analysis of nationally-representative district-level panel data from
1991/2 to 2002/3 to examine two main questions with the goal of ‘testing’ the NVF
hypothesis in Zambia: (1) Is HIV/AIDS having a negative independent effect on various
indicators of agricultural production? And (2) Is HIV/AIDS indirectly affecting
agricultural production by exacerbating the impacts of drought and other shocks?
Estimation results for the most drought-prone and highly HIV/AIDS-afﬂicted
agroecological region of the country suggest statistically signiﬁcant negative effects of
HIV/AIDS on agricultural production and support the NVF hypothesis that HIV/AIDS
exacerbates the impacts of drought. Results from the other agroecological regions
suggest much weaker impacts of HIV/AIDS and do not provide strong support for the

NVF hypothesis as it relates to agricultural production at the district level.

ACKNOWLEDGEMENTS

I would like to thank my major professor and assistantship advisor, Dr. Thorn
Jayne, for his support, feedback, and helpful ideas and comments on this thesis, and for
the opportunity to get involved with MSU’s research on HIV/AIDS and rural livelihoods.
I also greatly appreciated the suggestions from and learned from the interaction with the
other members of my thesis committee: Drs. Robert Myers, Cynthia Donovan and Jack
Meyer. Dr. Antony Chapoto was instrumental in helping me develop the skills to write
this thesis. Antony took me under his wing during my ﬁrst semester at MSU and has
continued to be an excellent mentor and colleague throughout my graduate studies. Other
MSU faculty, particularly Drs. Scott Swinton, Roy Black, and Dave Tschirley, also
provided helpful feedback at various stages of the thesis development. I am grateful for
the assistance of Dr. Jones Govereh, Stephen Kabwe and Misheck Nyembe of the Zambia
Food Security Research Project in tracking down data in Zambia. I also wish to thank
Margaret Beaver for sharing with me her expertise on the Post-Harvest Survey data and
Hunter Nielson for assistance in developing 618 maps used in the thesis research.

Thanks to my parents, brother and sister, and friends from many chapters of my
life for their unconditional love and support and to my fellow graduate students in the
department for their camaraderie and encouragement. I owe a special debt of gratitude to

Nicole Olynk, on whose support, patience, and “chipperness” I can always rely.

iii

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... v
LIST OF FIGURES ..................................................................................................... viii
LIST OF ACRONYMS .................................................................................................. ix
CHAPTER 1: INTRODUCTION .................................................................................... 1
CHAPTER 2: BACKGROUND ON THE NEW VARIANT F AMINE HYPOTHESIS ...4
2.1 The new variant famine hypothesis and its predictions ........................................... 4
2.2 Evidence for the new variant famine hypothesis ..................................................... 5
CHAPTER 3: METHODS & DATA ............................................................................. 12
3.1 Theoretical Framework ........................................................................................ 12
3.2 Empirical model .................................................................................................. 15
3.3 Estimation ........................................................................................................... 23
3.4 Data ..................................................................................................................... 24
CHAPTER 4: RESULTS ............................................................................................... 28
4.1 Independent effects of HIV/AIDS on agricultural production indicators ............... 28
4.2 How HIV/AIDS affects the impacts of exogenous shocks on agricultural
production indicators ................................................................................................. 36
CHAPTER 5: CONCLUSIONS & POLICY IMPLICATIONS ...................................... 50
APPENDIX ................................................................................................................... 56
REFERENCES .............................................................................................................. 81

iv

LIST OF TABLES

Table 1a. Long-run partial effects of a one-percentage point increase in HIV prevalence
on selected agricultural production indicators ......................................................... 30

Table 1b. Long-run partial effects of a one-percentage point increase in the AIDS-related
mortality rate on selected agricultural production indicators ................................... 30

Table 2. Summary of model results by agrozone according to their estimated long-run
partial effects of HIV/AIDS on selected agricultural production indicators ............. 33

Table 3. Summary of model results by agricultural production indicator according to their
estimated long-run partial effects of HIV/AIDS on selected agricultural production
indicators ............................................................................................................... 34

Table 4a. Partial effects of a one-percentage point increase in the negative rainfall shock
on selected agricultural production indicators (evaluated at mean and high HIV
prevalence) ............................................................................................................ 39

Table 4b. Partial effects of a one-percentage point increase in the negative rainfall shock
on selected agricultural production indicators (evaluated at mean and high AIDS-
related mortality) ................................................................................................... 39

Table 5a. Partial effects of a one-kg/ha (4%) fertilizer subsidy increase on selected
agricultural production indicators (evaluated at mean and high I-HV prevalence)....41

Table 5b. Partial effects of a one-kg/ha (4%) fertilizer subsidy increase on selected
agricultural production indicators (evaluated at mean and high AIDS-related
mortality) ............................................................................................................... 41

Table 6a. Partial effects of a one-percentage point increase in female-headed households
on selected agricultural production indicators (evaluated at mean and high HIV
prevalence) ............................................................................................................ 44

Table 6b. Partial effects of a one-percentage point increase in female-headed households
on selected agricultural production indicators (evaluated at mean and high AIDS-
related mortality) ................................................................................................... 44

Table 7a. Partial effects of a 100,000 ZMK productive asset base increase on selected
agricultural production indicators (evaluated at mean and high HIV prevalence)....47

Table 7b. Partial effects of a 100,000 ZMK productive asset base increase on selected
agricultural production indicators (evaluated at mean and high AIDS-related
mortality) ............................................................................................................... 47

Table A1. Govemment-subsidized and commercial/private sector channels for fertilizer
acquisition by Zambian smallholders in the Post-Harvest Surveys for years 1997/8-
2002/3 ................................................................................................................... 58

Table A2. Preferred lag structure on HIV/AIDS for each dependent variable and
agrozone ................................................................................................................ 59

Table A3. Agrozone 1 - Summary statistics for dependent and explanatory variables ....60
Table A4. Agrozone 2 - Summary statistics for dependent and explanatory variables ....61
Table A5. Agrozone 3 - Summary statistics for dependent and explanatory variables ....61
Table A6. Agrozone 4 - Summary statistics for dependent and explanatory variables ....62
Table A7. Correlation matrix of explanatory variables (data from all agrozones pooled) 62

Table A8. Correlation between HIV prevalence and AIDS-related mortality (all
agrozones) ............................................................................................................. 63

Table A9. Mean and 90th Percentile of HIV prevalence by agrozone and lag ................. 63
Table A10. Mean and 90th Percentile of AIDS-related mortality by agrozone and lag ....63

Table A11. Short-run partial effects of a one-percentage point increase in HIV prevalence
on crop output (‘000 ZMK) .................................................................................... 64

Table A12. Short-run partial effects of a one-percentage point increase in HIV prevalence
on crop output per capita (‘000 ZMK/capita) ......................................................... 65

Table A13. Short-run partial effects of a one-percentage point increase in HIV prevalence
on crop output per hectare (‘000 ZMK/ha) ............................................................. 66

Table A14. Short-run partial effects of a one-percentage point increase in HIV prevalence
on crop output per hectare per capita (‘000 ZMK/ha per capita) ............................. 67

Table A15. Short-run partial effects of a one-percentage point increase in HIV prevalence
on cultivated area (0.1 ha) ...................................................................................... 68

Table A16. Short-run partial effects of a one-percentage point increase in HIV prevalence
on cultivated area per capita (0.1 ha/capita) ............................................................ 69

Table A17. Short-run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on crop output (‘000 ZMK) .................................................. 70

Table A18. Short-run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on crop output per capita (‘000 ZMK/capita) ........................ 71

vi

 

Table A19. Short-run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on crop output per hectare (‘000 ZMK/ha) ............................ 72

Table A20. Short-run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on crop output per hectare per capita (‘OOO ZMK/ha per capita)

.............................................................................................................................. 73
Table A21. Short—run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on cultivated area (0.1 ha) ..................................................... 74
Table A22. Short-run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on cultivated area per capita (0.1 ha/capita) .......................... 75
Table A23. Summary of Almon lag estimates for the short-run partial effects of
HIV/AIDS on select agricultural production outcomes ........................................... 76
Table A24. Regression results from Agrozone 1 models using HIV prevalence and
including interaction terms ..................................................................................... 77

vii

LIST OF FIGURES

Figure A1. Agroecological regions of Zambia ............................................................... 57

viii

AER
AIC
AIDS
CGAIHS

CSO
FAO
FEGLS
FRA

F SP

F SRP
ha

HIV
IFPRI
kg
MACO
MSU
NGO
NVF
PA
PAM
PHS
PPI
SADC
SAHIMS
SCF
SIDA
UNAIDS
WHO
ZMK

LIST OF ACRONYMS

Agroecological region
Akaike Information Criteria
Acquired immune deﬁciency syndrome

Collaborative Group on AIDS Incubation and HIV Survival including the

CASCADE EU Concerted Action
Central Statistical Ofﬁce

Food and Agriculture Organization of the United Nations

Fixed effects (feasible) generalized least squares
Food Reserve Agency

Food Security Pack

Food Security Research Project

hectare

Human immunodeﬁciency virus

International Food Policy Research Institute
kilogram

Ministry of Agriculture and Cooperatives
Michigan State University

Non-govemmental organization

New variant famine

Prime age

Programme Against Malnutrition

Post-harvest survey

Producer price index

South African Development Community
Southern African Humanitarian Information Network
Save the Children Foundation

Swedish International Development Agency
Joint United Nations Programme on HIV/AIDS
World Health Organization

Zambia kwacha

ix

CHAPTER 1: INTRODUCTION

The ‘new variant famine’ (N VF ) hypothesis has become an important part of the
conventional wisdom surrounding the relationship between HIV/AIDS and food crises in
southern Africa, and has begun to shape HIV/AIDS mitigation and food security policies
and programs of governments and development agencies (de Waal and Tumushabe,
2003). The NVF hypothesis suggests, inter alia, that HIV/AIDS is causing a decline in
agrarian livelihoods and that the epidemic is making agrarian communities more
vulnerable and less resilient to drought and other transitory shocks (de Waal and
Whiteside, 2003; de Waal, 2004). Although a growing body of literature suggests a
decline in agricultural productivity and productive assets among HIV/AIDS-afﬂicted
households compared to non-afﬂicted householdsI (reviewed in Gillespie and Kadiyala,
2005; and Barnett and Whiteside, 2002), there remains a dearth of empirical evidence to
support the NVF hypothesis (de Waal, 2004), which emphasizes how HIV/AIDS
compounds the effects of other exogenous shocks on agrarian household and community
livelihoods. To date, no studies have been speciﬁcally designed to test the predictions of
the NVF hypothesis (de Waal, 2007).

This study represents a ﬁrst step towards testing the predictions of N VF . I
estimate the impact of AIDS—related morbidity and mortality on indicators of agricultural
production in Zambia. I focus speciﬁcally on the impact of HIV/AIDS on district-level

crop output, output per hectare, and area cultivated (henceforth referred to as ‘agricultural

 

I This thesis uses the terminology of Barnett and Whiteside (2002): the term “afflicted” is used to refer to
households that have incurred an HIV/AIDS-related death and to households with an HIV positive
household member. Households that are not afﬂicted but have otherwise been indirectly affected by the
impacts of HIV/AIDS on the broader community are referred to as “affected’.

production indicators’). The study is based on econometric analysis of district-level panel
data derived from nationally representative household surveys from 1991 to 2003. The
analysis is designed to: (1) understand the potential lagged effects of AIDS morbidity and
mortality on current agricultural production indicators; (2) measure the extent to which
HIV/AIDS may exacerbate the impacts of other factors affecting agricultural production,
such as drought and agricultural sector policy changes; and (3) determine whether these
trends and impacts are consistent with the predictions of the NVF hypothesis. The study
aims to strengthen the empirical foundation of food security policies and programs
responding to the HIV/AIDS crisis in southern Africa.

In addition to the availability of nationally representative longitudinal district
level data, Zambia is a suitable test case of the NVF hypothesis because agrarian
communities in the country are experiencing the ravages of HIV/AIDS as well as
recurrent droughts. In 2006, UNAIDS estimated that the HIV prevalence rate among
adults (aged 15-49) was 17%, placing Zambia among the six most-highly afﬂicted
countries in the world (UNAIDS, 2006). Drought has also plagued the country, with ﬁve
droughts occurring between 1991 and 2003 (Govereh and Wamulume, 2006; Del Ninno
and Marini, 2005). In addition to being faced with recurrent drought and HIV/AIDS,
smallholder farmers in Zambia have had to adapt to structural adjustment reforms
implemented in the 19905. These reforms included large reductions in government
fertilizer subsidies, the withdrawal of marketing boards infrastructure, and the elimination
of pan-territorial pricing for maize, the main staple food crop in the country (World Bank,

2004). Understanding the impact of HIV/AIDS on agrarian communities requires

controlling for these other exogenous shocks facing the agricultural sector, as well as
accounting for potential interaction effects between these processes.

The remainder of this thesis is organized as follows: Chapter 2 describes the
predictions of the new variant famine hypothesis and reviews the evidence to date in
support of NVF. Chapter 3 presents the methods and data used in the study. Chapter 4
describes the results and Chapter 5 outlines the conclusions and policy implications of the

thesis.

CHAPTER 2: BACKGROUND ON THE NEW VARIANT FAMINE
HYPOTHESIS
2.1 The new variant famine hypothesis and its predictions

The background discussed below is drawn from papers by de Waal and
Whiteside, 2003, de Waal and Tumushabe, 2003, and de Waal, 2003. The NVF
framework predicts two main trends over time due to the HIV/AIDS epidemic in
southern Africa: (1) declining agrarian livelihoods; and (2) increasing sensitivity and
decreasing resilience of agrarian communities to drought and other external shocks. The
rationale for the ﬁrst trend is that the burden of care giving, lost labor, money, and lost
social and emotional support are all eroding agrarian livelihoods. Care is expensive in
terms of time (e. g., caring for the sick and orphans, attending funerals) and money (e. g.,
paying for medical and funeral expenses) for both afﬂicted and non-afﬂicted households
in farming communities. Furthermore, HIV/AIDS-related mortality among prime-age
adults, particularly women, reduces the number of productive adults that provide for
dependents while also increasing the number of dependents. This results in high effective
dependency ratios2 and potential labor shortages.

The rationale for the second trend predicted by the NVF hypothesis is that
longstanding coping strategies used by households to mitigate the impacts of drought and
other shocks may no longer be effective since the onset of AIDS. For example, even in
the absence of other shocks, households may sell assets to pay for AIDS-related medical
or funeral expenses. Such asset depletion may undermine the households’ and

communities’ ability to cope with drought-related shortfalls in crop production. In

 

2 The effective dependency ratio is deﬁned as the number of dependents (children, the elderly, and the ill)
divided by the number of productive adults (de Waal and Whiteside, 2003).

addition, AIDS-related morbidity and mortality can disrupt the intergenerational transfer
of knowledge related to famine coping strategies (e. g., gathering wild foods or income
generating activities) on which households rely to reduce the impacts of drought and
other shocks. According to de Waal and Whiteside (2003), a common coping strategy
historically in times of drought was for adults to reduce their food consumption; however,
in the era of HIV/AIDS, such a strategy is dangerous because poor nutrition accelerates
the progression from HIV to AIDS and also increases the susceptibility of non-afﬂicted
individuals to HIV infection. De Waal and Whiteside (2003) argue that kinship networks
are also being stressed by AIDS-related illness and deaths with the result that kinship
networks may be less able to absorb the impacts of other shocks. Finally, de Waal
suggests that the burden of care described above forces households and communities to

adopt less productive and less resilient farming practices.

2.2 Evidence for the new variant famine hypothesis

The two broad predictions of the NVF hypothesis described above grow out of
three NVF ‘sub-hypotheses’: that HIV/AIDS leads to (1) “new patterns of vulnerability to
destitution and hunger”; (2) “new trajectories of destitution during crisis”; and (3)
changes in the “ecology of nutrition and infection, and thereby the pattern and level of
child mortality” (de Waal, 2007). In his chapter on NVF in the 2007 book, The New
F amines (Devereux, 2007), de Waal outlines the evidence to date in support of each of
these NVF sub-hypotheses. The majority of this evidence is either indirect or
circumstantial. de Waal acknowledges that no studies have been designed speciﬁcally to

test the predictions of NVF (de Waal, 2007) —- this study is the ﬁrst to do so. Few of the

 

papers cited as evidence of NVF control for other factors affecting agrarian livelihoods
nor do they attempt to empirically test the hypothesis that HIV/AIDS is exacerbating the
effects of other shocks on agricultural livelihoods. Furthermore, most of the works cited
as evidence are from limited geographic areas or are based on a ‘snapshot’ of the
epidemic’s impact rather than trends over a number of years. Given the ‘long—wave’
nature of HIV/AIDS, it is important to consider the immediate and delayed impacts of the
epidemic (Gillespie, 2006; Barnett and Whiteside, 2002). Moreover, case studies based
on localities known to be hard-hit by the disease may generate conclusions that do not
accurately reﬂect broader community- or national-level impacts.

In terms of evidence for the NVF sub-hypothesis that HIV/AIDS is creating “new
patterns of vulnerability to destitution and hunger” (de Waal, 2007), de Waal points to
several small-scale studies (Barnett and Blaikie, 1992; Webb and Mutangadura, 1999;
and Baylies, 2002) that have explored the impacts of HIV/AIDS on household and
community food security in rural Aﬁica. While these studies provide valuable insights
into the social and economic impacts of HIV/AIDS on rural households and
communities, all three of the studies focus on relatively limited geographic areas, often
with high HIV prevalence rates, and so are not nationally representative or easily
generalized. The work by Barnett and Blaikie (1992) and Baylies (2002) is largely
qualitative and so does not permit measurement of the magnitude of the HIV/AIDS
impact on households relative to other shocks and factors affecting agrarian livelihoods.
The work by Webb and Mutangadura (1999) has a quantitative component and includes
comparisons between afﬂicted and non-afﬂicted households with respect to various

socio-economic indicators; however, it is not clear if differences in income and other

indicators are statistically signiﬁcant or if there are other differences between the afﬂicted
and non-afﬂicted households that could be responsible for the income differential.
Nonetheless, these and a plethora of other household-level studies (see Gillespie and
Kadiyala, 2005 for a thorough review) suggest that HIV/AIDS is negatively affecting
afﬂicted households’ incomes, asset levels, and agricultural production.

None of the aforementioned papers allows comparison of the impact of drought
on highly HIV-afﬂicted or affected households and communities versus those where the
epidemic has had less of an impact. However, a pilot study in Tanzania which examined
the impacts of the 2002/3 drought on households experiencing adult morbidity and
mortality versus non-afflicted households found, contrary to the predictions of NVF, that
afﬂicted households were actually better off than non-afﬂicted households in the face of
drought (Tumushabe, 2005). One reason cited for this counter-intuitive ﬁnding is that the
households with chronically ill adults and households incurring a prime-age adult death
tended to be better off to begin with, underscoring the importance of controlling for other
factors in order to isolate the impact of HIV/AIDS on rural livelihoods (Jayne et a1.
2005)

While HIV/AIDS has undoubtedly had a devastating impact on many of the
households it has touched, de Waal cites several papers that call attention to the
effectiveness of household coping strategies in many instances. One such coping strategy
is so-called “replacement” in which households experiencing a prime-age death attract
new household members, thereby mitigating the labor force impact of the death (Yamano

and Jayne, 2004; Mather et al., 2003). de Waal suggests that households that use the

 

replacement coping strategy might be more vulnerable to subsequent shocks; however, to
date, no studies have empirically tested this hypothesis (de Waal, 2007).

A weakness in much of the evidence presented by de Waal is that it consists of
facts and ﬁgures on how HIV/AIDS is affecting afﬂicted households but gives little
ground for comparison with non-afflicted households, nor does it directly support the
claim that HIV/AIDS exacerbates the effects of drought and other shocks on agrarian
livelihoods. For example, de Waal cites a study from rural South Africa (Steinberg et al.,
2002) that indicates that half of the AIDS-afﬂicted households in the sample “reported
that their children were going hungry as a result” (de Waal, 2007: 93). But to what extent
were children going hungry in the non-afﬂicted households in the study area? Were there
factors other than HIV/AIDS that were contributing to the child malnutrition problem?

In several other papers on the NVF hypothesis, de Waal lists among the evidence
for NVF increased cassava cultivation in several southern African countries (de Waal and
Whiteside, 2003; de Waal, 2004). While this upward trend is striking, other factors such
as agricultural policy changes, including dismantling of marketing boards infrastructure
and reductions in maize and fertilizer subsidies are likely to be as, if not more, important
than HIV/AIDS in inﬂuencing this shift (Jayne et al., 2005; Chapoto, 2006).

While the evidence presented by de Waal in support of the NVF sub—hypothesis
of “new patterns of vulnerability to destitution and hunger” shows that HIV/AIDS often
has a negative effect at the household level, none of the evidence directly supports the
claim that afﬂicted households are harder hit by drought and other shocks than non-

afﬂicted households. While it is certainly plausible that such households are more

vulnerable to shocks, as their ability to cope is worn down by the impacts of the
epidemic, to date no explicit empirical evidence supports this claim.

In terms of the NVF sub-hypothesis that afﬂicted and affected households and
communities will follow “new trajectories of destitution during crisis”, de Waal states
that “this prediction has yet to be tested” (de Waal, 2007: 95). The two main sources of
circumstantial evidence for this NVF sub-hypothesis are a report by the Southern African
Development Community (SADC) Food and Natural Resource Vulnerability Assessment
Committee (SADC, 2003) and a paper on AIDS, child malnutrition, and drought in
southern Africa by John Mason and colleagues (Mason et al., 2005). Similar limitations
to those outlined above recur for this evidence for the NVF hypothesis. de Waal points to
a ﬁnding in the SADC report that 57% of households with a chronically ill adult (used as
a proxy for AIDS-afflicted households) had not eaten for entire days (de Waal, 2007: 95).
Although this is higher than the percentage of non-afﬂicted households using this coping
strategy (46%), there were no signiﬁcant differences in income between afﬂicted and
non-afﬂicted households and other socio-economic differences were not controlled for
between the two groups (SADC, 2003).

The Mason et a1. (2005) paper is one of the few sources of evidence cited by de
Waal that is based on regression analysis. It is also the only study other than this thesis
of which I am aware that has explicitly tested for an interaction effect between
HIV/AIDS and drought on a food security- or rural livelihoods-related outcome variable.
Using pooled time-series and cross-sectional data from Zambia (1996-2001/2) and
Zimbabwe (1999-2002), Mason et a1. (2005) regressed change in child underweight

prevalence (a measure of malnutrition) on a drought year dummy variable, HIV

prevalence, and an interaction term between drought and HIV prevalence. The
coefﬁcients on the HIV prevalence and drought variables were statistically insigniﬁcant
but the interaction term was statistically signiﬁcant and positive, indicating that the effect
of drought is exacerbated by high HIV prevalence and vice versa. However, the
parameter estimates from this regression analysis may be biased because no other factors
affecting change in child underweight were controlled for in the model (omitted variables
bias).

de Waal also points to anecdotes of increased engagement in transactional sex by
women during food crises as evidence in support of the NVF hypothesis (SCF, 2002;
SAHIMS, 2003; and Semu-Banda, 2003, cited in de Waal, 2007). While such practices
may indeed be occurring, there is very little evidence to indicate whether it has increased
in recent years (as opposed to people simply becoming more aware of it). Furthermore,
even it if is a widespread practice, increased engagement in transactional sex lends
credence to the idea that food crises are exacerbating the spread of HIV/AIDS (Bryceson
and F onseca, 2006), rather than HIV/AIDS worsening the effects of food crises as
postulated by NVF.

de Waal presents no evidence in support of the third and ﬁnal NVF sub-
hypothesis that the ecology of nutrition and infection changes in poor, vulnerable
populations in the presence of a generalized HIV/AIDS epidemic (de Waal, 2007).

Overall, the evidence in support of the various components of the NVF hypothesis
is weak at best. This is not to say that the theory is invalid, only that it has yet to be
tested empirically in any rigorous way. In this study the use of econometric analysis is a

critical contribution, as it allows us (subject to the constraints of the data available) to

10

control for the effects of other processes and identify the ceteris paribus effects of

HIV/AIDS, rainfall and other shocks, and their interactions, on agrarian livelihoods as
proxied by district-level crop output, output per hectare and cultivated area.3
Furthermore, this analysis is based on nationally representative survey data and considers

a 13-year period rather than being based on a limited geographic area or short time

period.

 

Other aspects of agrarian IIVeIIhOOdS that would have been interesting to examine Include consumption or
expenditure, income (farm and off-farm), and nutrition-related outcomes (e.g., anthropometrics); however,
the district-level panel data required for such analyses are not available for Zambia.

ll

 

l mesa—.-

CHAPTER 3: METHODS & DATA

3.1 Theoretical Framework

The goal of this thesis is to empirically test the NVF hypothesis. The NVF
hypothesis predicts a “downward spiral” in the well being of HIV/AIDS-afﬂicted
agrarian communities due to the interactions between the epidemic, drought and other
shocks. This implies dynamic responses and impacts over time, and hence reasonably
long time-series data is required to measure and detect such dynamics.

I test the NVF hypothesis using data on agricultural production from the Zambia
Post-Harvest Surveys (PHS) for agricultural years 1991/ 1992 to 2002/2003. The PHS is
a nationally representative longitudinal survey of smallholder agriculture in 52 districts.4
Approximately 7,000 small and medium scale farming households5 are included in the
PHS each year; however, the speciﬁc households interviewed are not the same from year
to year. The data set can therefore be considered a panel data set over 12 years, in which
the cross-sectional unit of observation is the district (not the household). (For more
details about the PHS survey design and samplings procedures, see Megill (2004).) The
methodology used to go from household- to district-level measures of agricultural

production is discussed in further detail in section 3.2 below.

 

4 Since the 2000 census, the nine provinces of Zambia have been divided into 72 districts; however, at the
time of the 1990 census, the country was divided into 57 districts. PHS data for five (5) of these districts is
not complete for one or more years during the period 1991/2-2002/3 so I use in the analysis the 52 “old”
districts for which'l do have complete data.

5 Small and medium-scale farming households are defined as those households that cultivate fewer than 20
hectares of land and produce crops, raise livestock or poultry, or farm ﬁsh. I refer to these households as
‘smallholder’ households throughout the text.

12

1 base my model on a supply response framework.6 Supply response models (e. g.,
Nerlove models) are used to study the effects of changes in prices and other exogenous
factors of interest on agricultural production at some aggregate level (community, region,

country, etc.). Such models are of the general form:

y=y(pe,px.Z) [1]

where y is a measure of agricultural output; [9" is the expected output price; p, is a vector
of input prices; and Z is a vector of other exogenous factors that affect supply (N erlove,
1958; Askari and Cummings, 1977). Input prices of potential relevance to agrarian
communities in sub-Saharan Africa include the prices of fertilizer, seed, pesticides, labor
(human and draft animal), and credit (i.e., interest rates).

In the supply response literature on crop production, the dependent variable in [l]
is usually crop output, output per hectare or total acreage. Askari and Cummings (1977)
suggest that acreage is preferable to output (tonnage) or output per hectare (yield) as the
dependent variable in supply response models because acreage is under farmers’ control
while output and yield are affected by factors beyond farmers’ control.

In many supply response models, Z includes factors such as rainfall,
infrastructure, and technology. Smallholder agriculture in Zambia is predominately
rainfed so rainfall is expected to have an important effect on agricultural production, and
so is included in the supply response model (RAIN). Beyond these ‘standard’ variables,
researchers commonly add other relevant explanatory variables to [1] depending on the
particular research question (Askari and Cummings, 1977). I add four such explanatory

variables. First, to examine the effects of HIV/AIDS on agrarian communities in Zambia,

 

6 Although I adopt an output supply function approach in this essay, 3 production function approach would
have provided additional insights as to the pathways through which HIV/AIDS is affecting smallholder
agricultural production. Data limitations prevent me from adopting a production function approach here.

13

I add the district-level HIV prevalence rate or AIDS-related mortality rate (AIDS).
Second, I add the community’s asset base (ASSET). The asset base is a potentially
important determinant of farm supply decisions because with imperfect credit markets,
the ability to ﬁnance input purchases and invest in productive assets will be constrained
by available resources. Third, I add the percentage of female-headed households in a
district (FEM). Women are disproportionately affected by the epidemic (Gillespie and
Kadiyala, 2005; UNAIDS, 2006) and widow-headed households often cultivate less area
than non-afﬂicted households or households in which the deceased prime-age adult
member is not the male head of household (Chapoto et al., 2007). Furthermore,
descriptive studies suggest that AIDS-related mortality exacerbates gender inequalities
(Mutangadura, 2005). For these reasons, I want to control for the number of female-
headed households in communities. Finally, I add the quantity of government-subsidized
fertilizer acquired by the community (SUB) to control for the effects of structural
adjustment-related reductions in government fertilizer subsidies on agricultural

production. With the addition of these variables, [1] becomes:
y = y(p§,px,Z*,AlDS) where 2* = (RAIN,ASSET,FEM,SUB) [2]

Another consideration is how to model py", the expected output price. The
prevailing market price for agricultural output at harvest time is not observable at
planting time; however, expected crop output prices are a major inﬂuence on farmers’
planting decisions (Nerlove, 1958). Two reasonable models of price expectations in the
Zambian smallholder sector are naive expectations and adaptive expectations. Rational
expectations is unlikely to be an appropriate model of price expectations in this context

because it requires that smallholders have information about the demand curve they face

14

for their output. Such information is typically not available to Zambian smallholders.

The naive and adaptive models of price expectations assume less information is available
to the decision makers. In the naive expectations framework, expected output price (pyf)
is deﬁned as:

Pyte = Pyt—l [3]

where py ,-, is the observed output price in the previous period. In the adaptive

expectations framework, expected output price is deﬁned as:
e _ e l _ 4
Pyt ‘ apyt—l +( a)p_yt—l I I
where 0 < (x < 1. This assumes that current price expectations adjust partially to the

difference between last period’s expectation and the actual price realization last period.

3.2 Empirical model

Model speciﬁcation

To test the predictions of the NVF hypothesis that HIV/AIDS: (1) negatively
affects agrarian livelihoods (as proxied here by various agricultural production
indicators); and (2) exacerbates the impacts of drought and other shocks on agricultural
production, I consider the estimation of an agricultural supply response model based on

equation [2]. I hypothesize the following linear panel data econometric model:

yi, — a + ypyit + p,“ w + AIDS-,6 + Zita) + T,0 + 1,4 81-, [5]
and z" = (RAINi, ,ASSETi, ,FEMi, ,SUBi,)

where i indexes the district; I indexes the year; y is the agricultural production indicator;

py" is the expected crop output price; pJr is a vector of input prices; AIDS is a measure of

15

the current productivity effect of the HIV/AIDS epidemic and is a function of the
estimated district HIV prevalence or AIDS-related mortality rate (this is discussed further
below); RAIN is a vector of positive and negative rainfall deviations (in percentage terms)
from the 20-year district mean rainfall level (following Hoddinott, 2006); ASSET is an
index of the mean household livestock and productive asset base; SUB is the mean
household acquisition of government-subsidized fertilizer (kg/ha); FEM is the percentage
of female-headed households; T is a vector of year dummies intended to capture the
effects on agricultural production of unobserved factors that change over time, such as

infrastructure or agricultural technology; ’1,- is the time invariant district-level unobserved
effects; and 8,, is the idiosyncratic error term. Assumptions made about the nature of the

relationships among the explanatory variables, A, and 8,, are described in detail in section

3.3.

The agricultural production indicators used as dependent variables (y) in the
analysis are mean household cultivated area, mean household crop output, and mean
household crop output per hectare, in both level and per capita terms. (I henceforth refer
to these six different dependent variables collectively as ‘agricultural production
indicators’.) For models in which the dependent variable is in per capita terms, the ASSET
variable is also speciﬁed in per capita terms. I use mean household averages in each
district rather than district totals for the dependent variables (as well as for ASSET and
SUB) in order to control for changes in the number of households (or number of members

in each household) in each district over time. In the case of cultivated area, the total area

16

cultivated by each household for 17 different crops7 included in the PHS was computed.
The household-level data were then weighted and mean household cultivated area was
calculated for each district and year.
In the case of crop output, a household-level crop output index (CM) was
17 T
computed as C hit = Z pjkcjhi, where pjk = T—1 2 p jkt , cm, is the kilograms of crop j
j=1 t=l
(1': 1 , ..., 17) produced by household h, in district i in year t, and pjk, is the real median
price for crop j in year tin province k (k=1,. . .,9). Thus, the household-level crop output
index is a weighted sum of the physical quantities of 17 crops produced by the household,
where the weights are the average median provincial crop prices over the period 1991/2
to 2002/3.8 To get a measure of household crop output per hectare, the household crop
output index (CM) was divided by the household’s total area planted (ha) in those 17
crops. Mean household crop output and mean household output per hectare were then
computed by weighting the household-level data and taking means for each district and
year.
In this analysis, py=PPI, where PPI is an agricultural producer price index. I use
the PP] as the output price rather than individual crop prices because the dependent

variable in [5] is mean household area planted or crop output (total or per hectare) for 17

crops potentially cultivated by the household. I calculate the PP] from PHS survey data

 

7 These 17 crops are: maize, sorghum, rice, millet, sunﬂower, groundnuts, soybeans, seed cotton, Irish
potatoes, Virginia tobacco, burley tobacco, mixed beans, cowpeas, velvet beans, coffee, sweet potatoes, and
cassava.

8 Other quantity indexes, such as the Divisia, Laspeyres, or Paasche quantity indexes, could also have been
used. The indexing method used in the thesis is similar to the Laspeyres quantity index in that prices are
held at base year levels; however, rather than choosing a single year as the base year, 1 computed average
crop prices over the entire period of analysis and used those prices as the ‘base year prices’. For the crop
output dependent variable, I am interested in changes over time in the physical quantities produced of the
17 different crops. Crop prices are used as weights to enable aggregation of physical quantities of different
crops. By using the average crop prices as weights, I am holding the weights constant over time.

17

Hit
17 ijt * chhit

for each district and year (PPIu) as PPIi, = 2 5,], p jkt where Sift = 17 [2:];
It

Fl 2 ijt * 2 thit
j=1 h=1
cm, and pjk, are deﬁned as above, and 5,], is the share of crop j in the total value of crop
output in district i in year t. Thus, PP!” is a weighted combination of real crop output
prices, where the weights are based on the relative importance of each crop to total

agricultural production in the district in each year.9 In the analysis, I model expected

crop output price using naive expectations (the model became overparameterized when

adaptive expectations were used); therefore, p31 1 = PPli,_l in [5].
. l —

The input prices included in the pJr vector include the real provincial fertilizer
price per kilogram and the lagged real national interest rate. Ideally, I would also have
included other input prices such as prices for seed and pesticides, rental rates for draught
animals, and wage rates (agricultural and non-agricultural); however, such data were not
available. I include input prices in the model to control for other factors that inﬂuence
supply so that I can get good estimates of the effects of HIV/AIDS and other shocks on
agricultural production. It is likely that fertilizer prices and interest rates are correlated
with the prices of inputs for which I do not have data; therefore, including fertilizer prices
and interest rates should adequately control for other input prices that affect farm supply.

The SUB variable (mean household acquisition of government-subsidized
fertilizer (kg/ha» was computed from the PHS data as follows. Prior to the 1997/8 crop

year, the PHS does not identify the channel used by households to acquire fertilizer,

 

9 As discussed above in the context of quantity indexes, alternative price indexes, such as the Divisia,
Laspeyres, Paasche or Fisher’s ldeal price indexes, could also have been used here.

18

reﬂecting the dominant role of the government in fertilizer importation, sales, and
distribution during that period. As described in detail in Jayne et al. (2002), prior to
1997/8 it was either illegal or legal but unproﬁtable in most cases for private ﬁrms to
participate in fertilizer marketing unless they were contracted by the government to do so.
Thus, in the analysis, I assume that all fertilizer acquired by households during crop years
1991/2 to 1996/7 was from government channels. For years 1997/8 through 2002/3, PHS
data on household fertilizer acquisitions are disaggregated by channel and these channels
were identiﬁed as either government-subsidized or commercial/private sector. Table A1
in the Appendix details the speciﬁc government-subsidized and commercial/private
sector channels in each year during the period 1997/8 to 2002/3, and provides additional
information on speciﬁc government fertilizer subsidy programs. Fertilizer acquisitions
through government-subsidized channels were computed for each household (kg/ha);
these data were then weighted and the mean household government-subsidized fertilizer
acquisition was computed for each year and district.

In the empirical speciﬁcation of the model, I add squared terms of RAIN and
AIDS to equation [5] to allow for the possibility of non-linear relationships between these
variables and the agricultural production indicators. 1 also add interaction terms between
AIDS and all elements of the 2‘ vector. This allows the elements of the 2‘ vector to have
differential effects on agricultural supply depending on the levels of HIV/AIDS. It also
enables us to test the key prediction of the NVF hypothesis: that HIV/AIDS exacerbates
the effects of drought and exogenous shocks (i.e., the variables in Z.) on agrarian
livelihoods as proxied here by various indicators of agricultural production.

(Smallholders’ responsiveness to input and output prices might also be sensitive to

19

HIV/AIDS but this is not a prediction of the NVF hypothesis, so such relationships are
not explored in this thesis.) Together these additions to [5] give:
yi, = a + ypv 2 l + p,” v1 + Alps-,5, + 24105562 + RAINitwl + RAIN: (02

. l —

+w3ASSETi, + w4FEM,-, + wSSUBi, + AIDS“ * RAIN ”th + AIDS” * RAIN : (pz [6]
+AlDS,-, * ASSE7},(p3 + AIDS” * FEM ”(p4 + AIDS,-, * SUBWS + T,9 + 11,- + 8;,

Modeling the dynamic relationship between HI V/AIDS and agricultural production

indicators

1 use two different variables to model the severity of the HIV/AIDS epidemic in a
given district and year: 1) the estimated HIV prevalence rate; and 2) the estimated AIDS-
related mortality rate, deﬁned as the number of AIDS-related deaths divided by the total
population. The HIV prevalence rate can be thought of as an advanced indicator of the
AIDS-related mortality rate, as there is typically a lag of 8-10 years between
seroconversion and death in the absence of antiretroviral treatment (Morgan et al., 2002;
CGAIHS, 2000; Zaba, Whiteside and Boerma, 2004).lo The HIV prevalence rate is the
estimated percentage of the population currently living with HIV/AIDS, and I expect it to
reﬂect the effects of HIV/AIDS-related illness and morbidity on agricultural production
indicators. I expect the AIDS-related mortality rate to reﬂect the effects of AIDS-related
deaths on such indicators.

There is often a signiﬁcant time lag between seroconversion and the onset of

symptomatic illness and death, as well as a lag between illness and death and the socio-

 

'° HIV prevalence rates are measured without reference to when people became HIV positive. Based on
epidemiological estimates that the mean period between seroconversion and death is 8-10 years, it is likely
that the HIV prevalence rates are computed based on people that have been HIV positive for the mean of
this period (4-5 years). Therefore, it is likely that HIV prevalence rates are an advance predictor of AIDS
deaths with a lag of roughly 4-5 years.

20

economic impacts of the epidemic. This ‘long-wave’ nature of HIV/AIDS (Gillespie,
2006; Barnett and Whiteside, 2002) implies that both current and past HIV/AIDS-related
morbidity and mortality may be affecting agricultural production. To model both the
immediate and delayed impacts of HIV/AIDS-related illness and morbidity, I estimate
models including the contemporaneous HIV prevalence rate as well as models including
contemporaneous and lagged HIV prevalence. I estimate similar models using the AIDS-
related mortality rate instead of HIV prevalence to test the relationships between AIDS-
related deaths and agricultural production.

Including numerous lags of the HIV/AIDS variables creates potential problems
with multicollinearity and degrees of freedom. To address this issue, I estimate models
using a polynomial distributed (Almon) lag structure for the HIV/AIDS variables in
addition to models with a traditional distributed lag structure. To identify an appropriate
number of years to lag the HIV/AIDS variables (J), I follow the recommendations in
Pindyck and Rubinfeld (1997) and Gujarati (2003) and add additional lags until the
Akaike Information Criteria (AIC) stops declining. In the Almon lag structures, I impose

a second-degree (quadratic) form on the polynomial. That is, I assume in

J
2 ajHM,_j , that a,- can be approximated by or,- = a0 + au' + azjz, where HIV" is the
i=0

HIV prevalence rate or AIDS-related mortality rate in district i in year t, and j is the

length of the lag. Substituting in for a], we obtain:

21

J J J J
,3 . .2
Hlvi, = 2 ame,_j = a0 2 Him—,- +01 2 JHqu— j +02 2 I “1 Vii—j
i=0

j=0 j:0 jzo

*
or HM, = aOAlmonli, + alAlmonzl-t + azAlmon3i,

J I
where Almon”, = Z Hm,_j, Almonzl-t = 2 jHIV,-,_,-.
i=0 i=0

J
-2
Almon” = 2 J HIVit—j
1:0

Such a lag structure allows the impact of the HIV prevalence rate in year t to grow over
time to a peak (as the disease progresses and ultimately results in the death of the HIV-
positive individuals reﬂected in that HIV prevalence rate in year t) and then for the effect
to eventually diminish as the households and communities affected by that wave of
illness and death eventually recover.11 The implications of such a lag structure on the
AIDS-related mortality variable are similar.

Zambia is divided into three agroecological regions (AERs): AER 1 covers the
southern border of the country and receives less than 800 mm of rainfall per year; AER 11
covers western and central Zambia and receives 800 to 1,000 mm of rainfall; and AER III
covers the northern part of the country and receives in excess of 1,000 mm of rainfall per
annum (see Figure A1 in the Appendix for a map of the AERs in Zambia). Some
districts’ borders encompass parts of both AERs II and III so I deﬁne four “agrozone”
categories and assign districts to each of them: districts in AER I are assigned to
agrozone 1 (lowest rainfall); districts entirely in AER II are assigned to agrozone 2 (lower
rainfall); districts with area in both AERs II and III are assigned to agrozone 3 (higher

rainfall); and districts in AER III only are assigned to agrozone 4 (highest rainfall). The

 

'1 Note that this lag structure allows for any quadratic lag weight path, i.e., either concave or convex. The
pattern described above is the intuitively expected result.

22

 

"Tr-11

impacts of HIV/AIDS and other explanatory variables on agricultural production
indicators may vary by agrozone. To determine if it is appropriate to pool agrozones, I
conduct a series of Chow Tests. The results of the Chow Tests suggest that the impacts
of HIV/AIDS and other regressors do indeed vary by agrozone, indicating that I should
run separate regressions for each agrozone.

For each agrozone and dependent variable combination, I compute the AIC value
for various ﬁnite distributed lag and Almon lag structures on HIV/AIDS (HIV prevalence
or AIDS-related mortality rate). The ‘preferred’ lag structure for each model (agrozone-
dependent variable combination) is that with the lowest AIC value; it is these ‘preferred’
models (summarized in Table A2 in the Appendix) that I estimate and analyze throughout

the remainder of the thesis.

3.3 Estimation

I take advantage of the panel-nature of the data and use an estimation technique
that allows us to difference out the time invariant district-level unobserved effects (ll,- ).
Under strict exogeneity of the explanatory variables conditional on 3.,- (assumption

FE. l) and the assumption that rank[E(X,-’X,-)l = K , where T is the number of years, K is

the number of explanatory variables, and X, is the TxK matrix for district i of time-

demeaned data on the explanatory variables in all T years (assumption FE.2), the ﬁxed

effects (FE) estimator is consistent (Wooldridge, 2002: 269). (Under these assumptions,

ll,- may be correlated with the observed explanatory variables and with the idiosyncratic

error term, 8i, .) I ﬁnd evidence of heteroskedasticity and serial correlation in the 8f, . In

23

this case, the ﬁxed-effects estimator is not the most efﬁcient estimator (ibid). Therefore, I
assume that E(8,-8,‘ I x,- ,/l,-) = A , where A is a TxT positive deﬁnite matrix,

8,- E (8,-1,£i2,...,8,-T) and x,- E (in ,xi2,...,x,-T)(assumption FEGLS.3), which relaxes the
assumptions of homoskedasticity and no serial correlation (Wooldridge, 2002: 276).
Under FE.1, FEGLS.3 and an additional assumption, FEGLS.2, the ﬁxed-effects

(feasible) generalized least squares (FEGLS) estimator is consistent and is the most

efﬁcient estimator (Wooldridge, 2002: 278). FEGLS.2 is the assumption that
ranklEO? {9"}? [)1 = K where Q a E(é,é,.’) , E,- is the (T -1 )x1 vector of time-demeaned
errors, and X,- is the (T - I )xK matrix of time-demeaned data on the explanatory variables.

I use the FEGLS estimator in the analysis (Wooldridge, 2002: 278).

3.4 Data

District—level indicators of agricultural production (mean household crop output,
output per hectare, and cultivated area), government fertilizer subsidies, producer price
indexes, and asset base indexes used in the analysis are drawn from PHS data. Median
provincial fertilizer prices are based on data from the Ministry of Agriculture and
Cooperatives (MACO) Agricultural Marketing Information Center. Real interest rates are
from World Development Indicators, 2005 (World Bank, 2005). Rainfall data are from
36 rainfall stations throughout Zambia. Estimates of district HIV prevalence and AIDS-
related mortality rates are drawn from the report, Zambia: HIV/AIDS Epidemiological
Projections, 1985-2010(CSO, 2005) and the Zambian population censuses from 1980,
1990, and 2000 (CSO, 1975; CSO, 1985; CSO, 1994; CSO, 2003). The Epidemiological

Projections report includes estimates of the HIV prevalence rate and of the number of

24

I.“ i!

AIDS-related deaths in each district in Zambia. The HIV prevalence estimates are based
on sentinel surveillance site (ante-natal clinic) data and the projections are computed
using the cohort component method. The AIDS-related deaths ﬁgures in the report are
generated by a mathematical model of the relationship between HIV prevalence and
AIDS-related deaths under assumptions about the life expectancy rate, fertility rate, and
the percentage of pregnant women on treatment to prevent mother-to-child-transmission
of HIV/AIDS (Chewe, 2006). The Epidemiological Projections report lists estimated
HIV prevalence rates and numbers of AIDS-related deaths for each district for the years
1985, 1990, 1995, and 2000-2010; extrapolation was required to estimate the HIV
prevalence and AIDS deaths in the years for which no values were reported. To control
the number of AIDS deaths for population growth, I computed a variable for the “AIDS-
related mortality rate”, deﬁned as the number of AIDS-related deaths divided by the total
population. As with the HIV prevalence and AIDS deaths ﬁgures, some extrapolation
was required to estimate population ﬁgures for the years in the intercensal periods.
Summary statistics, correlation matrices and other information on the variables used in
the analysis are included in the Appendix.

Given the methodology used to generate the district-level estimates of the HIV
prevalence rates, AIDS-related deaths, and AIDS-related mortality rates, there are
multiple sources of potential measurement error in these estimates. First, not all districts
have a sentinel surveillance site; the Zambia Central Statistical Ofﬁce (CSO) generated
estimates of the HIV prevalence rates in districts without sentinel surveillance sites by
assuming that the HIV prevalence rates in districts with similar socio-economic

characteristics are the same (Chewe, 2006). In other words, CSO assumed that the HIV

25

prevalence rate in a district without a sentinel surveillance site was the same as the HIV
prevalence rate in a district with similar socio-economic characteristics that did have a
sentinel surveillance site. There will clearly be measurement error in such estimates.
Second, even for districts with sentinel surveillance sites, estimates of HIV prevalence in
the sexually active adult population of men (aged 15-59) and women (aged 15-49) in
those districts are based on the estimated HIV prevalence among pregnant women
visiting antenatal clinics (sentinel surveillance sites). This can be problematic for two
main reasons: (i) pregnant women who visit antenatal clinics may not be representative of
all pregnant women (those who do and do not visit antenatal clinics); and (ii) HIV
prevalence among pregnant women may not be a good proxy for HIV prevalence among
the sexually-active adult population of men and women (WHO & UNAIDS, 2003).
Third, since the AIDS-related adult death estimates are derived from the HIV prevalence
estimates using a mathematical model, there will be measurement error in these
estimates. Fourth, in order to estimate the AIDS-related mortality rate, population
estimates had to be made for years in the intercensal period using '
extrapolation/interpolation — another source of measurement error. Fifth, there is also
measurement error due to extrapolation/interpolation in the estimates of AIDS-related
deaths and HIV prevalence for the years for which no ﬁgures were published in the
Epidemiological Projections report (CSO, 2005). Finally, the HIV/AIDS estimates in the
Epidemiological Projections report (CSO, 2005) are for the combined rural and urban
populations of each district but the agricultural production data analyzed in this thesis is

predominately for rural households.

26

One important result to note in the data on HIV/AIDS is the high sample

correlation (p = 0.90) between the estimated district level HIV prevalence and the

contemporaneous estimated AIDS-related mortality rate. Because of the lag between
seroconversion and AIDS-related death, the correlation between the AIDS-related
mortality rate and HIV prevalence increases as I lag HIV prevalence (see Table A8 in the

Appendix for a table of these correlations). This correlation is highest (p i: 0.95) between

the AIDS-related mortality rate in year t and the HIV prevalence rate from three to ﬁve
years earlier.

These relationships between HIV prevalence and lagged AIDS-related mortality are
consistent with a priori expectations and the epidemiology of HIV/AIDS. In general, the
time from seroconversion to death is 8-10 years. So, for any given HIV-positive
population, on average, individuals have been living with HIV for 4-5 years. The high
correlation in the Zambia data between HIV prevalence from 3-5 years ago and AIDS-
related deaths today roughly corresponds with the average time period for an HIV-

positive individual to die of AIDS-related causes.

27

CHAPTER 4: RESULTS

4.1 Independent effects of HIV/AIDS on agricultural production indicators

I estimate models of the form

y” = a + ypyit_l + p,” w + Alas-,5l + Alosﬁaz + RAIN,-,wl + RAJNfﬁwZ [7]
+w3ASSETi, + w4FEMi, + w5SUB,-, + T,9 + A,- + 13,-,
using an appropriate AIC-minimizing lag structure on AIDS and AIDS‘ to get an initial
measure of the effects of LUV/AIDS on mean household crop output, output per hectare,
and cultivated area (levels and per capita). Note that in these models I exclude the
interactions of HIV/AIDS with drought and other exogenous factors.

To determine the long-rtm partial effect of a one percentage point increase in the
HIV prevalence rate or AIDS-related mortality rate on the dependent variable of interest,
I take the partial derivative of equation [7] (which may include dynamics) with respect to
AIDS. I evaluate this partial derivative at mean HIV prevalence or AIDS-related
mortality and then at the 90th percentile of these HIV/AIDS measures. The long-run
partial effects of a one percentage point increase in HIV prevalence or the AIDS-related
mortality rate on each dependent variable are reported by agrozone in Tables 1a and lb

below. The short-run partial effects of HIV/AIDS and the partial effects of HIV/AIDS at

each lag are reported in tables A11 to A22 in the Appendix.12 Because of

 

'2 The short-run partial effect estimates show us the ‘shape’ of the Almon lag structure for each outcome
variable and agrozone combination (see Table A23 in the Appendix). A priori, I expected the Almon lag
estimates of the partial effects of HIV/AIDS on agricultural production outcomes to be zero, then
increasingly negative, and eventually zero again as agrarian communities recover from the HIV/AIDS
shock, i.e., a U shape. This was the estimated shape in 10 of 31 models where the preferred lag structure
was an Almon lag; the estimated shape in the other 21 Almon lag models was A contrary to a priori
expectations. One possibility for further research in this area would be to impose the u shape rather than
allowing ‘the data to decide’ the shape (i.e., either U or n) and compare the long-run partial effect
estimates to those reported in this chapter.

28

multicollinearity, these short-run effects are imprecisely measured; however, the long-run
partial effects reported in Tables 1a and 1b should be more reliable. (Due to the large
number of models estimated and space limitations, I report only the parameter estimates
for the key variables of interest, rather than the full regression results. In the Appendix,
Table A24, I include one full set of regression results that are consistent with the
implications of many of the other models estimated in the analysis.)13

Consider ﬁrst the long-run partial effect of an increase in the HIV prevalence rate
on agricultural production indicators in agrozone 1, the lowest rainfall zone. When
evaluated at mean HIV prevalence (17.33%), there is a negative and statistically
signiﬁcant long-run partial effect of HIV prevalence on all six indicators (p<0.10). (Refer
to Tables A9 and A10 in the Appendix for summary statistics of contemporaneous and
lagged HIV prevalence and AIDS-related mortality rates for each agrozone.) For four of
the six agricultural production indicators, the negative impact of HIV/AIDS is more
negative at high levels of HIV/AIDS relative to mean levels. The results are similar for

agrozone 1 when I use the AIDS-related mortality rate instead of HIV prevalence.

 

'3 In general the parameter estimates on the covariates (i.e., those variables that are not interacted with
HIV/AIDS) are as follows: the coefﬁcient on lagged agricultural PPI is either positive and statistically
signiﬁcant (p<0.10), as I would expect, or not statistically different from zero. I expect the coefficients on
the provincial fertilizer price and lagged national interest rate variables to be negative and signiﬁcant;
however, in all models but one (area cultivated in agrozone 2 with AIDS-related mortality as the
HIV/AIDS measure), the estimated coefficient is either positive and signiﬁcant or not statistically different
from zero. These counter-intuitive findings may be because these variables are correlated with other input
price variables that I was unable to include in the model due to data limitations. However, the parameter
estimates on these variables are not the key results of interest in this essay - I only include the lagged PPI,
lagged interest rate and fertilizer price variables in the models as controls.

29

Table la. Long-run partial effects of a one-percentage point increase in HIV
prevalence on selected agricultural production indicators

 

 

 

 

 

 

 

 

 

 

Evaluated Dependent Variable
A ro- at Area
zogne HIV— Output Output . Output/ha Output/ha Area cultivated
prevalence per capita per capita cultivated per capi ta
1 Mean -1 1.3% -7.3% -1.0% -12.4% -3.8% -2.3%
(17.33%) (2.75)" Q85)+ (2.64)" (3.08)M (2.08)* (2.20)*
High -16.5% -7.0% -1.1% -5.5% -4.3% -2.4%
(25.00%) (2.87)" (1.29) (2.24)* (1.10) (1 .71)+ $.36)
2 Mean 4.9% 7.4% -0.2% 12.9% -9.6% 8.8%
(11.77%) 42.40? (1.53) (0.52) (2.22)* (2.62)“ (2.71)M
High 4.3% 6.6% -0.4% 8.8% -3.8% 9.2%
(18.78%) (3.78)“ (1.88)+ (1.02) (2.08)* (1.41) (2.69)"
3 Mean -3.5% -4l.6% -9.8% -105.5% -7.7% -38.9%
48.56%) (0.84) (1.48) (5.30)" (4.02)" (0.21) (0.85)
High 4.7% -24.6% —3.5% -50.9% 0.3% -8.0%
(12.00%) (0.54) (1.47) (3.52)" (3.46)M (0.02) (0.45)
4 Mean -1 . 1% -3.5% -1.0% -10.3% 3.5% 0.8%
(10.43%) (0.53) (1.51) (3.22)“ (2.85)M (1.22) (0.31)
High 5.2% 3.8% 0.5% 6.0% -6.2% -5.1%
(21.03%) (1.82)+ (1.25) (1.99)* (1.97)‘ (2.64)“ (2.12)*

 

Table lb. Long-run partial effects of a one-percentage point increase in the AIDS-
related mortality rate on selected agricultural production indicators

 

 

 

 

 

 

 

 

 

 

Evaluated Dependent Variable
Agro- ill—[TS— Area
zone - Output Output . Output/ha Output/ha Area cultivated
related per capita per capita cultivated .
mortality per capita
1 Mean -11.1% -5.5% -1.3% -4.2% 8.1% 10.8%
(6.87%) (1.99)* (0.95) (2.24)* (0.88) (2.12)* (2.96)"
High -17.8% -15.5% -1.2% -5.3% -l.6% -5.4%
(12.34%) (4.61)" (4.11)" (3.12)“ (1.38) (0.64) (2.10)*
2 Mean 6.1% -1.5% 0.3% -3.2% 18.3% 8.8%
(3.66%) (1.59) (0.34) (0.85) (0.89) (1.92)+ (2.25)*
High 3.8% -0.3% 0.0% -2.1% 21.9% 8.8%
(7.23%) (1.79)+ (0.13) (0.22) (1.1 1) (3.18)" (2.49)*
3 Mean -56.7% -92.0% 9.7% 103.8% 9.7% -5.7%
(2.66%) (1.20) (1.92)+ (3.49)M (2.82)M (0.12) (0.29)
High -56.0% -6S.0% 4.2% 57.9% -1 1.2% 7.9%
(5.07%) (1.32) (1.59) (2.58)M (3.23)" (0.35) (0.7g
4 Mean -4.2% -0.5% -1.4% -13.2% 14.9% 12.4%
(1.63%) (1.37) (0.21) (5.06)" (2.56 ** (4.11)" (3.31)"
High -0.4% 0.0% -1 . 1% 1.9% 4.7% 5.6%
(13.73%) (0.15) (0.00) (5.27)“ (0.54) (1.86)+ (2.14)“

 

Source: Author’s calculations
Notes: Absolute value of 2 statistics in parentheses under partial effect estimates. + signiﬁcant at 10%;

* signiﬁcant at 5%; ** signiﬁcant at 1%. Mean = partial derivative with respect to HIV/AIDS evaluated at
mean HIV prevalence or AIDS-related mortality for the agrozone. High = partial derivative with respect to
HIV/AIDS evaluated at the 90th percentile of HIV prevalence or AIDS-related mortality for the agrozone.

30

In agrozone 2, the next lowest rainfall zone, the picture is dramatically different.
For only one of the dependent variables (area cultivated) is there a negative and
statistically signiﬁcant long-run partial effect of mean HIV prevalence (-9.6%, p<0.01);
however this effect becomes less negative in magnitude and is not statistically signiﬁcant
(p>0.10) when evaluated at high HIV prevalence. I ﬁnd no evidence of signiﬁcant
negative impacts of AIDS-related mortality on any of the six indicators. Compared to
agrozone 1, the mean HIV prevalence rate in agrozone 2 is substantially lower (11.77%),
which might explain why I ﬁnd little evidence of a negative impact of HIV/AIDS on
agricultural production in agrozone 2. The ﬁndings, in some cases, of positive long-run
partial effects of HIV/AIDS in agrozone 2 (and in the other agrozones) are
counterintuitive. I discuss and analyze the cases of positive long-run partial effects in
further detail at the end of this section.

In agrozone 3, the second highest rainfall zone, there is no statistically signiﬁcant
long-run partial effect of HIV prevalence on output (levels and per capita) or area
cultivated (levels and per capita). The results are similar when I use AIDS-related
mortality. For output per hectare (in both levels and per capita terms), the long-run
partial effect of IHV prevalence is negative and highly signiﬁcant (p<0.01); however, the
magnitude of the negative effect is less negative when evaluated at the 90th percentile
than it is when evaluated at the mean. (Also, the ﬁnding of a 105.5% decrease in
output/ha per capita when HIV prevalence is at the 90th percentile, is unreasonable as it
suggests total crop failure.) On the contrary, the estimated long—run partial effects of
AIDS-related mortality on output per hectare (in both levels and per capita terms) are

positive and statistically signiﬁcant (p<0.01), again with an unreasonably large estimated

31

partial effect (103.8%) of mean AIDS-related mortality. Given these contradictions and
the weak evidence of any statistically signiﬁcant effect whatsoever of HIV/AIDS on
output or area cultivated, I conclude that there ”is no robust relationship between
HIV/AIDS and agricultural production indicators in agrozone 3. Of the four agrozones,
zone 3 has the lowest average HIV prevalence and AIDS-related mortality rates (8.56%
and 2.66%, respectively).

In agrozone 4, the highest rainfall zone, the results are also inconclusive. Neither
HIV prevalence nor AIDS-related mortality has a statistically strong long-run partial
effect on output or output per capita (all p—values are greater than 0.10). There is some
evidence of a decline in output/ha in both levels and per capita terms when the partial
effect of HIV/AIDS is evaluated at mean levels. However, the results also suggest,
contrary to a priori expectations, that the long-run partial effect of HIV prevalence is
positive and signiﬁcant when evaluated at the 90th percentile. The opposite is true of the
relationship between HIV prevalence and area cultivated: the long-run partial effect of
mean HIV prevalence is not statistically signiﬁcant (p>0.10) but the long-run partial
effect of HIV prevalence is statistically signiﬁcant (p<0.05) and negative when evaluated
at the 90th percentile. As in agrozone 2, 1 ﬁnd a signiﬁcant positive effect of increased
AIDS-related mortality on area cultivated in agrozone 4.

On balance, only in agrozone 1 do I consistently ﬁnd evidence of a signiﬁcant
negative direct effect of HIV/AIDS on agricultural production. This agrozone is
characterized by the lowest mean annual rainfall levels and the highest mean HIV
prevalence and AIDS-related mortality rates of the four agrozones. This ﬁnding of a

weak relationship between HIV/AIDS and agricultural production at the district level is

32

consistent with other community and aggregate level evidence from Zambia. For
HIV/AIDS impacts on cultivated area, Larson et a1. (2004) ﬁnd that, for the entire sample
of cotton farming households in‘ their study, an increase in prime-age (PA) deaths was
associated with a decline in area cultivated of less than 1%. Similarly, Jayne et a1.
(2006) ﬁnd a decrease of only 6% in area cultivated at the community level when adult
mortality rates in Zambia increase from 0 to 24.4%; furthermore, this effect is short-lived,
as it becomes statistically insigniﬁcant after 3-8 years. These studies also ﬁnd weak or
non-existent impacts of AIDS-related mortality on crop output: increases in PA deaths in
Zambia reduce aggregate crop output by less than 1% (Larson et al., 2004) and there is no
independent effect of increases in the adult mortality rate on gross value of crop output or
output per hectare (Jayne et al., 2006). Likewise, despite incurring 52 AIDS related-
deaths between 1993 and 2005, 35 clusters studied in Mpongwe, Zambia were able to
increase maize production over the period (Drinkwater et al., 2006).

Table 2 summarizes the signs and signiﬁcance levels of the long-run partial
effects of HIV/AIDS on agricultural production indicators in each agrozone.

Table 2. Summary of model results by agrozone according to their estimated long-

run partial effects of HIV/AIDS on selected agricultural production indicators
% of total model speciﬁcations according to their estimated long-run partial effect of
HIV/AIDS on agricultural production indicators

 

 

 

Agrozone Negative, signiﬁcant Not statistically drﬁezent from Positive, signiﬁcant
zero at the 5 A: level
at 5% level . . . at 5% level

Negative Posrtrve

I 54.2% 37.5% 0% 8.3%

2 4.2% 29.2% 29.2% 37.5%

3 33.3% 33.3% 16.7% 16.7%

4 29.2% 20.8% 29.2% 20.8%

Overall 30.2% 30.2% 18.8% 20.8%

 

Source: Author’s calculations (based on Tables 1a and 1b)

Notes: Percentages indicate the percentage of the 24 total models estimated for each agrozone where the
estimated long-run partial effect of HIV/AIDS on agricultural production indicators was negative and
signiﬁcant at the 5% level, positive and signiﬁcant at the 5% level, and not statistically different from zero
at the 5% level.

33

Table 2 indicates that the counterintuitive ﬁndings, in some cases, of positive and
statistically signiﬁcant (p<0.05) long-run partial effects of HIV/AIDS are most prevalent
in agrozone 2. In each of the other agrozones, approximately 80% of the estimated long—
run partial effects of HIV/AIDS on agricultural production indicators area either negative
and statistically signiﬁcant at the 5% level or not statistically different from zero at the
5% level (as expected a priori). Using a 5% signiﬁcance level, we can expect to make
Type 1 errors 5% of the time.

Disaggregating the counterintuitive positive long-run partial effects results by the
agricultural production indicator (rather than by the agrozone) reveals additional insights
(Table 3).

Table 3. Summary of model results by agricultural production indicator according

to their estimated long-run partial effects of HIV/AIDS on selected agricultural

production indicators
% of total model speciﬁcations according to their estimated long-run partial

 

 

 

Agricultural effect of HIV/AIDS on agricultural production indicators
production Negative, Not statistically different from zero at the 5% level Positive,
indicator signiﬁcant , , , signiﬁcant
at 5% level “33"“ ”5"” at 5% level
Output 25.0% 37.5% 25.0% 12.5%
Output per capita 6.3% 68.8% 25.0% 0%
Output/ha 56.3% 12.5% 12.5% 18.8%
Output/ha per capita 31.3% 31.3% 6.3% 31.3%
Cultivated area 18.8% 31.3% 31.3% 18.8%
Cultivated area per capita 18.8% 25.0% 12.5% 43.8%

 

Source: Author’s calculations (based on Tables 1a and 1b)

Notes: Percentages indicate the percentage of the 16 total models estimated for each agricultural production
indicator where the estimated long-run partial effect of HIV/AIDS was negative and signiﬁcant at the 5%
level, positive and signiﬁcant at the 5% level, and not statistically different from zero at the 5% level.

Table 3 indicates that the ﬁndings of positive and statistically signiﬁcant (p<0.05)
long-run partial effects of HIV/AIDS are most prevalent when the agricultural production
indicator used is cultivated area per capita or output/ha per capita. In the case of

cultivated area per capita, if a household member dies and the household continues to

34

cultivate the same total area of land (and does not attract new household members),
cultivated area per capita must increase, ceteris paribus. Thus, there is a plausible
explanation for what initially appears to be a counterintuitive ﬁnding. In the case of
output/ha per capita, a similar explanation applies. If total output/ha is held constant and
the number of household members decreases due to an AIDS-related death, per capita

output/ha must increase, ceteris paribus. A reduction in household size as a result of an

AIDS-related death in the household is consistent with empirical evidence from Zambia.

Chapoto and Jayne (2008) ﬁnd that households incurring the prime-age death of a long-
time resident household member other than the household head or spouse experience a
reduction in household size of 2.21 members in the case of a male adult death and a
reduction of 1.47 household members in the case of a female adult death.

For the other agricultural production indicators, over 80% of the estimated long-
run partial effects are either negative and statistically signiﬁcant at the 5% level or not
statistically different from zero at the 5% level. The ﬁnding of negative and signiﬁcant
long-run partial effects of HIV/AIDS on output/ha at the district level in 56% of the
models estimated is consistent with ﬁndings of household-level studies that suggest that
prime-age male deaths (Yamano & Jayne, 2004) and male head of household deaths
(Chapoto and Jayne, 2008) are associated with a decrease in production of cash (hi gh-
value) crops, and that the (HIV/AIDS-related) death of the head of household or his/her
spouse adversely affects the value of crop production per hectare (Yamano & Jayne,

2004)

35

.. '"—l-'

4.2 How HIV/AIDS affects the impacts of exogenous shocks on agricultural
production indicators

In the previous section, I examined the independent impacts of HIV/AIDS on
selected agricultural production indicators. This section analyzes how the impacts on
agricultural production indicators of drought and other exogenous factors are affected by
HIV/AIDS, namely: policy changes such as those that occurred as part of structural
adjustment reforms, gender inequalities, and shocks to communities’ productive asset
base.

In this section I detail the estimation results from models as speciﬁed in equation
[6] with an appropriate AIC-minimizing lag structure. To determine the long-run partial
effect of a one unit increase in the rainfall shock, fertilizer subsidy, female-headed
household, or productive asset base variables on the dependent variable of interest, I take
the partial derivative of equation [6] (which may include dynamics) with respect to RAIN,
SUB, FEM, or ASSET. I evaluate the partial derivative at mean HIV prevalence or AIDS-

related mortality and then at the 90th percentile of these HIV/AIDS measures.

Evidence that HIV/AIDS exacerbates the effects of drought?

Tables 4a and 4b present the results for the partial effects of a one-percentage
point increase in high (i.e., 90th percentile) negative rainfall shocks (droughts) evaluated
at mean and high levels of HIV/AIDS. As in the case of the long-run partial effects of
HIV/AIDS on agricultural production indicators, of all the agrozones, the results from
agrozone l are most consistent with a priori expectations with respect to the impacts of

negative rainfall shocks. In agrozone 1 models where I use the HIV prevalence as the

36

measure of the severity of the HIV/AIDS epidemic in a district, I ﬁnd negative and
statistically signiﬁcant (p<0. 10) partial effects of negative rainfall shocks (droughts) on
ﬁve of the six indicators when the partial effect is evaluated at mean HIV prevalence, and
on four of the six indicators when evaluated at the 90th percentile of HIV prevalence. For
crop output and output/ha, the negative impact of drought is more negative when HIV
prevalence is high. This ﬁnding is consistent with the prediction of the NVF hypothesis.
However, when I use the AIDS-related mortality rate instead of the HIV prevalence,
there is no robust relationship between drought and the agricultural production indicators
used in the analysis.

In agrozone 2, whether I use the HIV prevalence or AIDS-related mortality rate, I
consistently ﬁnd a statistically signiﬁcant negative partial effect of drought on output and
output per hectare. In HIV prevalence models, the relationship between the partial
effects evaluated at mean and high HIV prevalence is consistent with NVF for three of
the agricultural production indicators. In AIDS models, the magnitude of the negative
impact of drought becomes more negative when evaluated at high AIDS-related
mortality; however, in some cases the partial effects evaluated at the 90th percentile are
imprecisely measured and are not statistically signiﬁcant at the 10% level. In general, in
agrozone 2 I ﬁnd some support for the NVF hypothesis that HIV/AIDS exacerbates the
negative impacts of drought, particularly those effects on crop output and output/ha.

In agrozone 3, output, output per capita, and cultivated area per capita are
negatively impacted by drought (p<0.10) in both HIV prevalence and AIDS models. The
negative impact of drought on output and cultivated area per capita is exacerbated by

high HIV prevalence rates and AIDS-related mortality rates as predicted by the NVF

37

hypothesis; and the negative impact of drought on output per capita is more negative at
high AIDS-related mortality rate levels but not at high HIV prevalence levels. There is
no robust relationship between drought and the other agricultural production indicators in
both the HIV prevalence and AIDS-related mortality models. As in agrozones 1 and 2, in
agrozone 3 1 ﬁnd some evidence to support the NVF hypothesis, but only for a subset of
the agricultural production indicators examined.

In agrozone 4, the highest rainfall zone, there is little statistically signiﬁcant
impact of negative rainfall shocks on agricultural production indicators when I use HIV
prevalence to model the HIV/AIDS epidemic. In models using the AIDS-related
mortality rate instead, I ﬁnd a weak negative impact of drought on crop output and
cultivated area (both in levels and per capita terms) but this impact is small in magnitude
(less than 2%) and only in the case of output per capita is the negative effect of drought
exacerbated by AIDS-related mortality.

These results provide some support for the NVF hypothesis in agrozones 1, 2 and
3, and particularly for the effects of drought on crop output and output per hectare (both
in levels and per capita terms). In many cases, the negative impact of drought is at least
twice as negative when HIV/AIDS is high relative to when HIV/AIDS is held at its mean.
However, NVF-like phenomena are far from universal even within these agrozones, and

the results are sensitive to the HIV/AIDS variable used.

38

Table 4a. Partial effects of a one-percentage point increase in the negative rainfall
shock on selected agricultural production indicators (evaluated at mean and high

 

 

 

 

 

 

 

 

 

 

 

HIV prevalence)
Evaluated Dependent Variable

Agro- at _

zone HIV Output Output . Output/ha Output/ha Area cultivated
prevalence per capita per capita cultivated per capi ta

1 Mean -1.4% -1.0% -0.2% -2.4% -0.9% -78.6%

41 7.33%L (2.22)* (1.52) (3.65)" (3.70)" (2.22)* (1.77)+

High -2.7% -2.1% -0.3% -4.3% 1.8% -97.7%
(25.00%) (2.60)” (2.17)* (3.27)" (4.30)* * (0.46) (1.584

2 Mean -1.6% -2.0% -0.2% -2.7% 0.7% -0.3%
(1 1.77%) (2.36)* (2.53)* (2.67)M (4.09)” (1.49) (0.58)
High -0.4% -3.8% -0.4% -7.2% 4.3% 1.8%
(18.78%) (0.32) (1.90)+ (2.81)" (3.84)” (3.39)" (1.43)

3 Mean -2.1% -8.3% 0.1% 0.5% -2.7% -3.9%

48.56%) (2.25)* (6.1 1)M (0.60) (0.26) (1.60) (2.35)*

High -3.4% -8.0% 0.5% 6.2% -8.7% -18.9%
(12.00%) (2.01)* (1.79)+ (1.48) (1.41) (1.45) (3.11)“

4 Mean -0.5% -0.5% 0.0% 0.0% -0.2% -0.3%
(10.43%) (1.70)+ (1.73)+ (1.07) (0.15) (0.57) (0.924
High 0.4% -0.2% 0.0% -1.0% 1.1% 0.4%
(21.03%) (0.79) (0.26) (0.39) (1.11) (1.36) (0.46)

 

Table 4b. Partial effects of a one-percentage point increase in the negative rainfall
shock on selected agriculturall production indicators (evaluated at mean and high
AIDS-related mortality)

 

 

 

 

 

 

 

 

 

Evaluated Dependent Variable

Agro- {AID—S

zone - Output Output . Output/ha Output/ha Area cultivated
related per capita per capita cultivated .
mortality per capita

1 Mean -0.3% -0.9% -0. 1% -0.2% -0.3% -0.6%
(6.87%) (0.46) (1.66)+ (1.42) (0.26) (0.68) (1.52)
High 1.1% -0.7% -0. 1% 0.1% -0.8% -0.2%
Q2.34°/o) (0.68) (1.23) (0.56) (0.1 1) (0.76) (0.25)

2 Mean -2.6% -2.1% -0.3% -2.7% 0.4% 0.6%

Q.66%) (3.08)" (2.64)" (4.79)" (3.56)" (0.81) (1.23)

High -3.3% -3. l % -0.7% -6.9% 2.2% 2.3%
(7.23%) (1.62) (1.64) (4.17)" (3.79)" (1.01) (1.06)

3 Mean -8.7% -l6.2% 0.2% 2.2% 5.4% -4.4%
(2.66%) (4.60)* * (7.46)" (1.36) (0.90) (0.58) (2.64)M
High -21.0% -41.8% 0.3% 4.4% 3.3% -10. 1%
(5 .07%) (3.63)* "‘ (6.29)“ (0.68) (0.91) (0.16) (1.90)+

4 Mean -0.8% -0.8% 0.0% 1.1% -1.3% -0.9%

44.63%) (2.5 8)" (2.72)" (0.74) (l .88)+ (2.50)* (l .75)+

High -1.4% -1.8% 0.0% 1.7% -1.6% -1.3%
(13.73%) (1.56) (2.28)* (0.34) (1.49) (1.56) Q20)

 

Source: Author’s calculations
Notes: Absolute value of 2 statistics in parentheses under partial effect estimates. + signiﬁcant at 10%;

* signiﬁcant at 5%; ** signiﬁcant at 1%. Mean = partial derivative with respect to high negative rainfall
shocks evaluated at mean HIV prevalence or AIDS—related mortality for the agrozone. High = partial
derivative with respect to high negative rainfall shocks evaluated at the 90‘h percentile of HIV prevalence or
AIDS-related mortality for the agrozone.

39

 

Evidence that HIV/AIDS exacerbates the effects of other shocks? iv

In much of the literature on the NVF hypothesis, de Waal suggests that
HIV/AIDS may be exacerbating the effects of other shocks in addition to drought. To
test this hypothesis, I consider ‘other shocks’ such as changes in fertilizer subsidies and
communities’ asset bases, and gender inequality embodied in the effect of female
household headship on agricultural production indicators.

Tables 5a and 5b present the results of the simulations for the partial effects of a 1
kg/ha (4%) increase in the fertilizer subsidy per household evaluated at mean and high
levels of HIV/AIDS. (Fertilizer subsidies did not have a statistically nor a practically
signiﬁcant effect on cultivated area or area per capita and so the SUB variable was
dropped from these models. For this reason, no partial effects of SUB results are reported

for cultivated area and area per capita.)

4o

Table 5a. Partial effects of a one-kg/ha (4%) fertilizer subsidy increase on selected
__agricu1tural production indicators (evaluated at mean and high HIV prevalence)

 

 

 

 

 

 

 

 

 

Agro- Evaluated Dependent Variable
at Ou ut Ou ut/ha
zone HIV prevalence Output per ctgpita Output/ha pertgapita
1 Mean 0.1% 0.0% 0.0% 0.0%
(17.33%) (0.52) (0.26) (0.60) (0.33)
High 0.2% 0.1% 0.0% 0.1%
(25.00%) (1.49) (0.68) (1.53) (0.40)
2 Mean 0. 1% -0.2% 0.0% 0.2%
41 1.77%) (0.72) (0.84) (2.40)* (0.984
High 0.0% -1.1% 0.1% -0.3%
(18.78%4 (0.13) (2.01)* (1.82)+ (0.77)
3 Mean 0.0% 0.5% 0.0% 0.5%
(8.56%) (0.31) (2.90)" (1.03) (2.15)*
High -0.6% -0.1% 0.0% 0.5%
(12.00%) (3.18)M (0.33) (0.95) (1.254
4 Mean 0.5% 0.4% 0.0% 0.1%
410.43%) (7.53)" (5.47)" (2.00)* (0.88)
High 0.3% 0.2% 0.0% 0.1%
(21.03%) (5.43)" (3.30)" (0.63) (0.324

 

Table 5b. Partial effects of a one-kg/ha (4%) fertilizer subsidy increase on selected

agricultural production indicators (evaluated at mean and high AIDS-related

 

 

 

 

 

 

 

 

 

 

mortality)
Evaluated Dependent Variable

Agro- at _

zone AIDS-related Output OWN} Output/ha Output/ha
mortality per capita per capita

1 Mean 0.7% O. 1% 0.0% 0.0%

46.87%) (0.52) (1.18) (1.40) (0.144

High 0.4% 0.0% 0.0% 0.0%
(12.34%) (1.34) (0.16) (0.87) (0.26)

2 Mean —0. 1% -0.2% 0.0% 0.1%

4469/0) (0.56) (0.91) (1.51) (0.81

High -0.3% -0.6% 0.0% 0.2%
(7.23%) (0.78) (1.33) (1.41) (0.65)

3 Mean 12.1% 16.8% 0.0% 0.1%
(2.66%) (3.734" * (5.26)" (0.13) (0.36)
High -0.7% -0.3% 0.0% -0.2%
(5.07%) (1.34) (0.60) (1.06) (0.51)

4 Mean 0.2% 0.2% 0.0% 0.0%
(4.63%) (1.40) (2.13)* (1.08) (0.10)
High -0.6% -0.4% 0.0% 0.1%
(13.73%) (1.74)+ (1.23) (0.55) (0.17)

 

Source: Author’s calculations

Notes: Absolute value of 2 statistics in parentheses under partial effect estimates. + signiﬁcant at 10%;

"‘ signiﬁcant at 5%; ** signiﬁcant at 1%. Mean = partial derivative with respect to fertilizer subsidy shocks
evaluated at mean HIV prevalence or AIDS-related mortality for the agrozone. High = partial derivative
with respect to fertilizer subsidy shocks evaluated at the 90th percentile of HIV prevalence or AIDS-related

mortality for the agrozone.

 

r

 

In agrozone 1, I ﬁnd no statistically signiﬁcant (p>0.10) partial effect of fertilizer
subsidies on agricultural production indicators after controlling for other exogenous
shocks. This is the case for both the HIV prevalence and AIDS-related mortality models.
Similarly for agrozone 2, there is no robust relationship between fertilizer subsidies and
agricultural production indicators. However, when HIV prevalence is high I do ﬁnd a
weak, negative impact (-1.1%, p<0.05) on output per capita and a practically small,
positive impact on output/ha (+O.l%, p<0.10).

In agrozone 3 for mean HIV prevalence models, the impact of increased fertilizer
subsidies is positive and statistically signiﬁcant (p<0.05) on both output per capita and
output/ha, but these partial effects are practically small (+0.5%). However, when I
evaluate the partial effects at the 90‘h percentile of HIV prevalence, the positive partial
effects of increased fertilizer subsidies are eliminated. This is in line with what I would
expect under the predictions of the NVF hypothesis, but the results are quite weak. I ﬁnd
a similar pattern in agrozone 3 for AIDS-related mortality models. In both HIV
prevalence and AIDS-related mortality models for agrozone 3, I ﬁnd results that are
consistent with the NVF hypothesis for two of the four agricultural production indicators.

In agrozone 4 for mean HIV prevalence models I ﬁnd statistically signiﬁcant
(p<0.01) but practically small increases in output, output per capita, and output/ha
associated with an increase in the fertilizer subsidy. The magnitude of the positive effect
on output and output per capita is smaller when these partial effects are evaluated at the
90th percentile of HIV prevalence; in the case of output/ha, the positive partial effect is no
longer statistically signiﬁcant (p>0.10) when evaluated at high HIV prevalence. For

mean AIDS-related mortality models, the partial effect of fertilizer subsidies is only

.. l

statistically signiﬁcant (p<0.05) on output per capita, and this effect is practically small
(+02%).

Overall, increases in fertilizer (subsidies have a practically small, if any, positive
effect on output and output/ha (in both levels and per capita terms). In all cases but one
where there is a statistically signiﬁcant, positive partial effect of fertilizer subsidies at
mean HIV/AIDS levels, this effect is less positive in magnitude when evaluated at high
HIV/AIDS levels. This is consistent with the predictions of the NVF hypothesis, but
occurs mainly in agrozones 3 and 4, and only for a subset of the agricultural production
indicators analyzed.

Tables 6a and 6b present the results of the simulations for the partial effects
(evaluated at mean and high levels of HIV/AIDS) of a one percentage point increase in

female-headed households in a district.

43

 

Table 6a. Partial effects of a one-percentage point increase in female-headed
households on selected agricultural production indicators (evaluated at mean and

 

 

 

 

 

 

 

 

 

 

41igh HIV prevalence)

Evaluated Dependent Variable

Agro- at _ . Cultivated

zone HIV Output OUtPUt per Output/ha Output/ha Cultivated area per
prevalence capita per capita area capita

1 Mean -1 . 1% -0.4% -0. 1% -0.7% -O.4% -26.9%
(17.33%) (2.65)" (1.02) (2.96)“ (1.58) (1.36) (0.92)
High -0.4% -0. 1% -0. 1% -1.0% -l .0% -13.5%
(25.00%) (0.59) (0.10) (1.48) (1 .79)+ (2.30)* «1.29)

2 Mean -0.9% -0.1% 0.0% 0.9% -0.5% -0.8%
(1 1.77%) (2.44)* (0.25) (0.15) (2.29)* (2.10)* (2.91)"
High 0.2% O. 1% -0.1% -0.9% 0.4% 0.6%
(18.78%) (0.34) (0.06) (1.43) (1.20) (0.83) (1.18)

3 Mean -0.4% 1.5% 0.0% 0.6% -0.3% 1.1%

48.56%) (1.09) (2.75)“ (1.33) (1.32) (0.14) (0.51)

High -1.7% 0.3% 0.1% 1.9% 1.0% 1.5%
(12.00%) (2.33)* (0.37) (1.26) (l .98)* (0.80) (1.234

4 Mean 0.0% 0.5% 0.0% 0.5% -0.7% -0.7%

410.43%) (01.4) (1 .67)+ (0.74) (1.67)+ (1.15) (1.20)

High 0.4% 0.7% 0.1% 1.5% -0.2% 0.2%
(21.03%) (0.75) (0.99) (1.77)+ Q78)" (0.54) (0.55)

 

Table 6b. Partial effects of a one-percentage point increase in female-headed

 

 

 

 

 

 

 

 

 

 

households on selected agricultural production indicators (evaluated at mean and
_hjgh AIDS-related mortality)
Evaluated Dependent Variable
at .

30%;: AIDS- Output Output per Output/ha Output/ha Cultivated 2:21;?“
related caprta per caprta area .
mortality caprta

1 Mean -0.3% -0.6% -0. 1% 0.1 % -0.5% 0.3%
(6.87%) (0.22) 470» Q.34)* (0.24) (1 .79)+ (0.36)
High 0.0% -0.6% 0.0% 0.2% -0.9% -0.6%
(12.34%) (0.03) (0.85) 41.06) (0.34) (1.57) (1.25)

2 Mean -0. 1% 0.3% 0.0% 0.7% -0.6% -0.9%
(3.66%) (0.23) (0.64) (0.52) (1 .85)+ (2.54)* (3.58)"
High 0.8% 1.4% 0.0% 0.2% - l .6% -l .8%
(7.23%) (1.1 1) (1 .76)+ (0.48) (0.34) Q..29)* (2.54)*

3 Mean 12.6% 17.3% 0.0% -0.6% -l.8% 0.6%
(2.66%) (4.10)" (5.44)** (0.744 (1.44) (0.55) (1.23)
High 2.7% 3.8% 0.0% -2.2% 3.8% 0.6%
(5.07%) (2.06)* (3.14)” (9.42) 47wr (0.65) (0.54)

4 Mean 0.0% 0.6% 0.0% 0.5% -0.2% 0.2%
(4.63%) (0.09) (2.02)* (0.58) (2.19)* (1.14) (1.10)
High 0.4% 0.6% 0.1% 0.1% -0.5% 0.0%
(13.73%) (0.59) (0.39) (1.46) (0.13) Q26) (Oﬂ

 

Source: Author’s calculations

Notes: Absolute value of 2 statistics in parentheses under partial effect estimates. + signiﬁcant at 10%; *
signiﬁcant at 5%; ** signiﬁcant at 1%. Mean = partial derivative with respect to female headship shocks
evaluated at mean HIV prevalence or AIDS-related mortality for the agrozone. High = partial derivative
with respect to female headship shocks evaluated at the 90th percentile of HIV prevalence or AIDS-related

mortality for the agrozone.

 

 

In agrozone 1 for HIV prevalence models, the sign of the partial effect of a one- d"
percentage point increase in female-headed households is consistently negative, as
expected a priori. When evaluated at mean HIV prevalence, this negative effect is only
statistically signiﬁcant (p<0.01) on output (-1.1%) and output/ha (-0.1%). For these two
agricultural production indicators, the negative effect of female headship shocks is not
statistically different from zero (p>0.10) when evaluated at the 90th percentile of HIV
prevalence. This ﬁnding is contrary to what I would expect under the NVF hypothesis.
For output/ha per capita and cultivated area, however, while female headship shocks
don’t have a signiﬁcant effect when HIV prevalence is at its mean, the effect is more
negative and signiﬁcant (p<0.10) at high HIV prevalence levels, as predicted by NVF.
When I model HIV/AIDS using the AIDS-related mortality rate, partial effects of female
headship shocks that are negative and signiﬁcant at mean AIDS-related mortality become
statistically insigniﬁcant at high AIDS-related mortality rates, contrary to NVF.

In agrozone 2, I again have conﬂicting results with respect to the hypothesis that
HIV/AIDS exacerbates the impact of female headship shocks on agricultural production
indicators. In HIV prevalence models, none of the ﬁndings support this hypothesis and
for three of the indicators (output, cultivated area and cultivated area per capita), the
results are opposite of what NVF would predict: the negative impacts of female headship
shocks are less negative at high HIV prevalence relative to mean HIV prevalence. For
AIDS-related mortality models, NVF is only supported for the effects of female headship
shocks on cultivated area and cultivated area per capita.

In agrozone 3, the ﬁndings are similarly weak. Although I do not ﬁnd direct

contradictory evidence against NVF, in only two of the twelve models (6 models each for

45

““11

HIV prevalence and AIDS-related mortality) do I ﬁnd evidence to support the NVF
hypothesis. In the case of HIV prevalence, NVF is supported for the negative effects of
female headship shocks on output, and in the case of AIDS-related mortality, NVF is
supported for output/ha per capita.

In agrozone 4 there is no evidence to support the NVF hypothesis as it relates to
female headship shocks and the agricultural production indicators analyzed, nor do I ﬁnd
evidence to unequivocally reject the NVF hypothesis in this agrozone.

Overall, there is little evidence to support the NVF prediction that the negative
impacts of female headship shocks will be exacerbated by HIV/AIDS. Of the 48
simulations done, the results of only 6 of them are consistent with the predictions of the
NVF hypothesis as it relates to female headship shocks.

Tables 7a and 7b present the results of the simulations for the partial effects
(evaluated at mean and high levels of HIV/AIDS) of a 100,000 ZMK increase in the
mean household productive asset base (or mean productive asset base per capita for
models in which the dependent variable is in per capita terms). This 100,000 ZMK
increase corresponds to a 10% increase for models in which the dependent variable is in
levels, and to a 67% increase for models in which the dependent variable is in per capita

terms.

46

Table 7a. Partial effects of a 100,000 ZMK productive asset base increase on
selected agricultural production indicators (evaluated at mean and high HIV

 

 

 

 

 

 

 

 

 

 

prevalence)
Evaluated Dependent Variable

Agro- at _ . Cultivated

zone HIV Output Output per Output/ha Output/ha Cultivated area per
prevalence capita per capita area capi ta

1 Mean 0.4% 0.3% 0.1% 1.2% 0.0% 233.1%
(17.33%) (1.38) (0.14) (2.93)* * (0.65) (0.20) (1.62)
High 1.5% 6.6% 0.1% 0.6% 0.4% 357.0%
(25.00%) (2.74)" (2.02)* (1.41) (0.18) (1.18) (1.80)+

2 Mean 0.7% - l .0% 0.0% -6.0% 0.7% -0. 1%
(1 1.77%) (2.43)* (0.48) (1.21) (2.87)" (3.78)“ (0.04)
High 0.6% -3.1% 0.0% -5.5% 1.0% 0.0%
(18.78%) (1.27) (0.68) (0.19) (1.35) (2.91)“ (0.02)

3 Mean 1.3% 23.0% 0.2% 18.9% -0.2% 17.7%
(8.56%) (1.29) (3.56)“ (2.01 )* (2.70)“ (0.06) (2.13)*
High -0.5% -3.0% -0. 1% -21.4% 1.9% 0.8%
(12.00%) (0.32) (0.29) (0.83) (1.76)+ (0.87) (0.06)

4 Mean 3.5% 14.8% 0.0% -0. 1% 3.2% 13.6%
(10.43%) (4.00)“ (3.23)“ (0.42) (0.02) (4.15)“ (3 .96)* *
High 2.2% 3.5% -0.2% -7.9% 4.1% 9.3%
(21.03%) (1.19) (0.41) (1.30) (0.90) (2.30)* (1.37)

 

Table 7b. Partial effects of a 100,000 ZMK productive asset base increase on
selected agricultural production indicators (evaluated at mean and high AIDS-

related mortality)

 

 

 

 

 

 

 

 

 

 

Evaluated Dependent Variable

A _ at_ . .

zognr: AlDS- Output Output per Output/ha Output/ha Cultivated 3:33;“
related capita per capita area capi ta
mortality

1 Mean 2.0% -0.7% 0.1% 0.8% -0.3% -2.0%
(6.87%) (1.46) (0.37) (3.72 ** (0.36) (1.15) (1.14)
High 1.6% 4.2% 0.1% 0.5% 0.2% 3.9%
(12.34%) (2.07)* Q85) (3 .20)* "' (0.19) (0.43) (0.86)

2 Mean 0.7% -0.4% 0.0% -3.0% 0.7% 0.8%
(3.66%) (2.61)“ (0.18) (0.83) (1.75)+ (4.36)M (0.70)
High 0.6% -4.7% 0.0% -3. 1% 1.2% 2.8%
(7.23%) Q .22) (1.48) (0.97) (0.88) (2.86)" (0.87)

3 Mean 9.3% -30.3% 0.0% 13.7% 2.7% 11.8%
(2.66%) (2.38)* (2.20)* (0.10) (1.62) (0.39) (1.96)*
High -2.5% -5.4% 0.0% 14.0% 6.3% -29.1%
(5.07%) (1.43) (0.46) (0.07) 41.5 8) (0.62) (2.37)*

4 Mean 2.6% 14.3% -0.1% 6.3% 3.8% 14.9%

44.63%) (2.83 ** (2.98)” (1.97)* (1.44) (5.66 ** (4.60)"

High -3.6% 8.0% -l.1% 2.6% 6.9% 27.9%
(13.73%) (1.26) (0.72) (4.81)“ (0.22) (3.62)“ (2.99)“

 

Source: Author’s calculations
Notes: Absolute value of 2 statistics in parentheses below partial effect estimates. + signiﬁcant at 10%; *
signiﬁcant at 5%; ** signiﬁcant at 1%. Mean = partial derivative with respect to productive asset base
shocks evaluated at mean HIV prevalence or AIDS-related mortality for the agrozone. High = partial
derivative with respect to productive asset base shocks evaluated at the 90th percentile of HIV prevalence or
AIDS-related mortality for the agrozone.

47

 

13...... _

In agrozone 1, I ﬁnd little evidence for the NVF prediction that the positive
effects of increases in the mean household productive asset base are reduced by
HIV/AIDS. In HIV prevalence models, this NVF scenario is only observed for output/ha
and I ﬁnd evidence contrary to NVF for output and output per capita. I ﬁnd no support
for NVF in the AIDS-related mortality models and ﬁnd evidence of effects contrary to
the NVF prediction in the case of crop output.

In agrozone 2, the results lend very little support to the NVF hypothesis. Only for
crop output do I ﬁnd that the positive effects of productive asset base increases are
decreased by HIV/AIDS, and these declines are minimal: from +0.7% to +0.6%. And
contrary to the NVF prediction, positive productive asset base partial effects on cultivated
area are more positive when HIV/AIDS is high relative to when it is at its mean.

Findings from agrozone 3 are most consistent with the predictions of the NVF
hypothesis. In HIV prevalence models, for four of the six indicators (output per capita,
output/ha, output/ha per capita, and cultivated area per capita), the statistically signiﬁcant
positive effects of productive asset base increases are less positive (and in some cases
negative and signiﬁcant) when the HIV prevalence is at the 90th percentile, compared to
when it is held at its mean. In AIDS-related mortality models, there is less evidence of
NVF-like phenomena, but I do ﬁnd that high AIDS-related mortality reduces the positive
partial effect of productive asset base increases on output and cultivated area per hectare.

In agrozone 4, results from the HIV prevalence models support the NVF
hypothesis for output, output per capita, and cultivated area per capita, but the results
from the cultivated area models contradict the predictions of NVF. Results from the

AIDS-related mortality models are inconclusive.

48

Overall, only in agrozone 3 do I consistently ﬁnd weak evidence that HIV/AIDS
reduces the positive impacts of productive asset base increases, as would be expected
under the NVF hypothesis. In mean household terms, the productive asset base in
agrozone 3 is lower than the other three agrozones; in per capita terms, the productive
asset base is the second lowest in agrozone 3 after agrozone 4). Perhaps because
communities in agrozone 3 have fewer productive assets to begin with, those few assets
are more important for their agricultural production but are also more vulnerable to being

liquidated as HIV/AIDS puts more stress on the communities.

49

CHAPTER 5: CONCLUSIONS & POLICY IMPLICATIONS

The new variant famine hypothesis posits that HIV/AIDS is eroding rural
livelihoods and exacerbating the impacts of drought and other shocks on agrarian
communities. In southern Africa, where HIV prevalence rates are among the highest in
the world and there are recurrent droughts, the NVF hypothesis has become an important
framework for understanding the relationship between HIV/AIDS and food crises. The
NVF hypothesis has also begun to inﬂuence the HIV/AIDS mitigation and food security
policies and programs of development agencies and governments.

Despite this inﬂuence, the empirical evidence base for the NVF hypothesis is
weak for several reasons. First, many of the studies cited as evidence in support of NVF
are based on case studies or data collected from relatively small (and often highly
afﬂicted) geographic areas or only for a short period of time. Second, few of the studies
cited as evidence of NVF control for factors other than HIV/AIDS that affect rural
livelihoods. And third, none of these studies directly tests the central tenet of NVF: that
HIV/AIDS exacerbates the effects of drought. Indeed, this thesis is the ﬁrst study to test
the NVF hypothesis econometrically using nationally representative panel survey data
from an extended period of time. Given the scarcity of development resources for
HIV/AIDS mitigation and food security initiatives, it is important that such interventions
be based, to the extent possible, on a solid empirical foundation. With this thesis, I aim
to strengthen the evidence base concerning the NVF hypothesis speciﬁcally and the

relationship between HIV/AIDS and smallholder agricultural production more generally.

50

In the thesis, I use nationally-representative district-level panel data from Zambia
from 1991/2 to 2002/3 and econometrically estimate several supply response models to
examine two main questions with the goal of ‘testing’ the NVF hypothesis in Zambia: (1)
Is HIV/AIDS having a negative independent effect on agricultural production (a proxy
for agrarian livelihoods)? And (2) Is HIV/AIDS indirectly affecting agricultural
production by exacerbating the impacts of drought and other shocks?

The analysis generates a number of ﬁndings that may help evaluate the validity of
the new variant famine hypothesis as an analytical framework in the context of agrarian
livelihoods and food security in Zambia. First, only in agrozone 1 do I consistently ﬁnd
evidence of a signiﬁcant negative independent effect of HIV/AIDS on mean household
agricultural production at the district level. This agrozone is characterized by the lowest
mean annual rainfall levels and the highest mean HIV prevalence and AIDS-related
mortality rates of the four agrozones. In agrozone 1, a one-percentage point increase in
the district HIV prevalence rate is estimated to result (in the long-run) in a 1% decrease in
mean household crop output/ha and an 11% to 17% decrease in mean household crop
output. In the other three agrozones, the estimated partial effects of HIV/AIDS on several
agricultural production indicators are negative but imprecisely measured. Crop output/ha
is the agricultural production indicator for which statistically signiﬁcant (at the 5% level)
negative partial effects of HIV/AIDS are most commonly found. This ﬁnding of a weak
relationship between HIV/AIDS and agricultural production at the district level is
consistent with other community and aggregate level evidence from Zambia (e.g., Larson

et al., 2004; Jayne etal., 2006; and Drinkwater et al., 2006).

51

Second, for the key NVF prediction that HIV/AIDS exacerbates the impacts of
drought on agrarian livelihoods, the results of this study lend some support to this
prediction for agrozones 1, 2 and 3, particularly when the outcome variable is crop output
or output per hectare (both in levels and in per capita terms). In many cases, the negative
impact of drought is at least doubled when HIV/AIDS is high relative to when HIV/AIDS
is held at its mean. For example, a one-percentage point increase in a negative rainfall
shock in agrozone 1 results in a 1.4% decrease in crop output when the HIV prevalence
rate is 17.33% (the mean), but a 2.7% decrease in crop output when the HIV prevalence
rate is 25.00% (the 90th percentile). However, NVF-like phenomena are far from
universal even within agrozones 1, 2 and 3, and the results are sensitive to whether the
HIV prevalence rate or the AIDS-related mortality rate is used to measure the severity of
the HIV/AIDS epidemic.

Third, increases in fertilizer subsidies have a practically small, if any, positive
effect on output and output/ha (in both levels and per capita terms). In all cases but one
where there is a statistically signiﬁcant, positive partial effect of fertilizer subsidies at
mean HIV/AIDS levels, this effect is less positive in magnitude when evaluated at high
HIV/AIDS levels. This is consistent with the predictions of the NVF hypothesis, but
occurs mainly in agrozones 3 and 4, and only for a subset of the agricultural production
indicators analyzed.

Fourth, there is little evidence to support the NVF prediction that the negative
impacts of female headship shocks will be exacerbated by HIV/AIDS. Of the 48
simulations done, the results of only six of them are consistent with the predictions of the

NVF hypothesis as it relates to female headship shocks. While I ﬁnd some evidence of

52

negative impacts of female-household headship on agricultural production indicators (a
result that is consistent with household-level studies that ﬁnd a negative impact of PA
male head of household deaths on agricultural production (Yamano and Jayne, 2004;
Chapoto and Jayne, 2005) and widows’ access to land (Chapoto et al., 2007; Mather et
al., 2004)), the results do not suggest a differential impact of female household headship
shocks depending on the severity of the HIV/AIDS epidemic.

And ﬁfth, only in agrozone 3 do I consistently ﬁnd weak evidence that HIV/AIDS
reduces the positive impacts of productive asset base increases as would be expected
under the NVF hypothesis. In mean household terms, the productive asset base in
agrozone 3 is lower than the other three agrozones (although in per capita terms, the
productive asset base is the second lowest in agrozone 3 after agrozone 4). Perhaps
because communities in agrozone 3 have fewer productive assets to begin with, those few
assets are more important for their agricultural production but are also more vulnerable to
being liquidated as HIV/AIDS puts more stress on the communities.

None of these ﬁndings lend unequivocal support to the NVF hypothesis in
Zambia. There is some evidence that in low rainfall areas, HIV/AIDS exacerbates the
effects of drought on crop output and output per hectare — a ﬁnding that is consistent with
the predictions of the NVF hypothesis. The evidence is much weaker that HIV/AIDS
exacerbates the impacts of other shocks on agricultural production (such as reductions in
fertilizer subsidies, a rise in the percentage of households that are female-headed, and a
reduction in productive farm assets). Furthermore, the results vary by agrozone, by the
agricultural production outcome analyzed, and by the HIV/AIDS measure used. Thus, as

is the case with household level analyses, it is important not to lump all highly afflicted

53

districts (or agrozones) into one category and overgeneralize as to the effects of
HIV/AIDS (and its interaction with other shocks) on agrarian communities.

The ﬁndings of this study suggest that efforts to target assistance toward
communities that are drought-prone (have low annual rainfall) or have a weak productive
asset base and are also highly AIDS-afflicted may be an important aspect of food security
and HIV/AIDS mitigation programs and policies. Efforts to improve social protection and
safety nets in communities whose asset bases have been eroded may also be an effective
way to mitigate the impacts of the epidemic. The ﬁnding of no robust negative effect of
HIV/AIDS on district level agricultural production except in the lowest rainfall areas
suggests that agrarian communities may be more resilient in the face of HIV/AIDS than
predicted. Therefore, it will be important for governments, donors and NGOs to continue
to invest in AIDS mitigation and rural development, broadly deﬁned, to bolster resilient
livelihood strategies in HIV/AlDS-afflicted agrarian communities.

Future research to further test the NVF hypothesis as it relates to agricultural
production could explore the impacts of HIV/AIDS, drought, and their interaction on the
median or variance of household agricultural production at the community level. This
thesis focuses on the effects of HIV/AIDS and drought on mean household agricultural
production indicators, but the impacts on the median or variance may tell a different
story. Furthermore, studies to assess the extent to which HIV/AIDS is exacerbating the
effects of drought on aspects of agrarian livelihoods other than agricultural production,
such as non-farm income; health, nutrition, and anthropometrics; and consumption/

expenditure among agrarian households, are also needed to further evaluate whether

54

empirical evidence supports the NVF hypothesis in Zambia and other countries in

southern Africa.

55

APPENDIX

56

Figure A1. Agroecological regions of Zambia

 

Agra-Ecological Zones & Average Annual Rainfall
1: less than 800 mm

:3 Ha: 800 - 1000 mm (clay soils)

' IIb: 800 - 1000 mm (sandy soils)

III: greater than 1000 mm

 

Source: Food Security Research Project, Lusaka, Zambia

57

Table A1. Government-subsidized and commercial/private sector channels for
fertilizer acquisition by Zambian smallholders in the Post-Harvest Surveys for years

 

 

 

 

 

 

 

1997/8-2002/3
Crop Government-subsidized channels Commercial/ Description of selected
year private sector channels government subsidy
prggrams
1997/8 -Co-op union -NGO
-Credit coordinator -Private trader
-Fellow farmer -Out-growers
1998/9 -Co-op union -NGO
~Credit coordinator -Private trader
-Fellow farmer -Out-growers
1999/2000 -Food Reserve Agency (F RA) through -NGO -FRA: subsidized
farmers association -Private trader fertilizer distributed on
-Farmers association -Out-growers credit to smallholders
-Govemment agent - FRA/Omnia -FRA/Omnia: FRA
-Fellow farmers contracts private ﬁrm
Omnia to distribute
fertilizer
2000/1 -F RA loan -Other fertilizer loan -FRA: subsidized
-Free government fertilizer -Direct exchange/barter fertilizer distributed on
-Other sources of free fertilizer -Cash purchases credit to smallholders
2001/2 -F RA loan -Other fertilizer loan -F RA: subsidized
-Govemment Food Security Pack- -Direct exchange/barter fertilizer distributed on
Programme Against Malnutrition -Cash purchases credit to smallholders
(FSP-PAM) -FSP-PAM: free
-Other sources of free fertilizer fertilizer (grant);
targeted at relatively
poor smallholders
2002/3 -Govemment cash and loan programs -Direct exchange/barter -FSP-PAM: free

through co-ops & farmers associations
-Govemment FSP-PAM
-Other sources of free fertilizer

-Commercial cash
purchase

fertilizer (grant);
targeted at relatively

poor smallholders

Sources: Post-Harvest Surveys, 1997/8-2002/3, Central Statistical Ofﬁce, Lusaka; Jayne et al. (2002);
World Bank (n.d.)

58

Table A2. Preferred lag structure on HIV/AIDS for each dependent variable and

 

 

 

 

_agrozone
Dependent Variable
HIV/AIDS Agro- Output Output/ha Area Area
measure zone Output . Output/ha . . cultivated
per capita per capita cultivated .
per capita

HIV 1 At—5 At-5 At—S At—5 At-6 t only

2 t only At-6 At-3 At-6 At-6 At-3

3 t only Dt-2 At-3 At-3 At-7 At-7

4 t only t only At-6 At-6 At-7 At-7
AIDS 1 At-7 At-6 At-6 t only At-6 At-7

2 DH Dt-l t only t only At-7 At-7

3 At-7 At-7 At-5 At-4 t only Dt-l

4 At-3 Dt- 1 At-4 At-7 At-7 At-7

 

Source: Author’s calculations
Notes: HIV = HIV prevalence rate; AIDS = AIDS-related mortality rate; D t-j = ﬁnite distributed lag with
maximum lag length ofj years; A t-j = Almon lag with maximum lag length ofj years; t only =

contemporaneous only.

59

Table A3. Agrozone 1 - Summary statistics for dependent and explanatory variables

 

 

 

 

 

Variable Obs. Mean Std. Dev. Min Max 9001 .
percentile
OUTPUT 156 1186.81 767.04 0 3976.22 21 14.52
OUTPUTPC 144 196.80 1 17.83 0 554.94 363.95
YIELD 156 852.76 442.74 0 2190.32 1429.14
YIELDPC 144 169.62 95.35 0 490.68 285.99
AREA 156 14.69 6.63 1.70 37.25 24.56
AREAPC 144 2.59 0.93 0.39 5.60 3.80
HIV 156 17.33 7.65 3.05 34.51 25.00
AIDS 156 6.87 4.22 0.32 17.69 12.34
POS 156 11.05 16.34 0 70.13 37.75
NEG 156 9.56 13.67 0 51.63 31.95
PPI 144 11.22 6.24 4.30 48.26 19.52
FERT 156 2048.41 816.84 747.43 3708.02 3489.62
IR 156 -1.17 24.08 -48.09 25.12 21.14
SUB 144 33.51 71.89 0 594.68 86.92
FEM 156 22.34 7.32 0 42.67 31.33
ASST 144 21.80 13.17 1.66 67.59 37.97
ASPC 144 3.14 1.85 0.27 10.38 5.35
Source: Based on raw data from PHS surveys, 1991/2-2002/3, Central Statistical Ofﬁce, Lusaka.
Where:
Variable Description Units Level Years
OUTPUT Mean household crop output '000 ZMK District Agric
OUTPUTPC Mean household crop output per capita '000 ZMK District Agric
YIELD Mean household crop output per hectare '000 ZMK/ha District Agric
YIELDPC Mean household crop output per hectare per capita '000 ZMK/ha District Agric
AREA Mean household cultivated area 0.1 ha District Agric
AREAPC Mean household cultivated area per capita 0.1 ha District Agric
HIV Estimated HIV prevalence rate % District Calen
AIDS Estimated AIDS-related mortality rate % District Calen
POS Positive rainfall shock, deviation from 20-year % District Agric
district avg.
NEG Negative rainfall shock, deviation from 20-year % District Agric
district avg.
PPI Real agricultural producer price index in year H ’00 ZMK/kg District Agric
PERT Real price of fertilizer ZMK/kg Provincial Calen
IR Real interest rate in year t-1 % National Calen
SUB Mean household fertilizer subsidy kg/ha District Agric
FEM Percentage of female-headed households % District Agric
ASST Mean household productive asset base '00,000 ZMK District Agric
ASPC Mean household productive asset base per capita '00,000 ZMK District Agric

 

Notes: Agric=Agricultural years (October through September) 1991/2-2002/3; Calen=Calendar years 1991-

2002.

60

Table A4. Agrozone 2 - Summary statistics for dependent and explanatory variables

 

 

01
Variable Obs. Mean Std. Dev. Min Max p622" tile

OUTPUT 130 1686.57 814.93 331.25 3876.07 2720.91
OUTPUTPC 120 305.11 130.43 48.40 634.87 484.89
YIELD 130 1030.35 347.45 150.34 1938.95 1494.18
YIELDPC 120 228.03 75.01 27.41 404.49 335.36
AREA 130 15.98 4.53 6.43 29.40 22.22
AREAPC 120 3.16 0.78 1.39 5.54 4.17
HIV 130 11.77 5.01 3.21 23.10 18.78
AIDS 130 3.66 2.43 0.36 11.76 7.23
POS 130 10.68 15.48 0 79.73 34.68
NEG 130 8.64 11.46 0 45.88 27.03
PPI 120 12.90 5.15 4.27 37.35 20.73
FERT 130 2100.57 809.15 747.43 3708.02 3489.62
IR 130 -l.17 24.10 -48.09 25.12 21.14
SUB 120 23.79 34.87 0.00 167.24 69.53
FEM 130 24.83 6.70 12.35 40.74 34.84
ASST 120 17.15 10.62 4.33 69.41 29.60
ASPC 120 2.66 1.33 0.66 7.43 4.51

 

Source: Based on raw data from PHS surveys, 1991/2-2002/3, Central Statistical Ofﬁce, Lusaka.

Table A5. Agrozone 3 - Summary statistics for dependent and explanatory variables

 

 

Variable Obs. Mean Std. Dev. Min Max 9001 .
percentile

OUTPUT 91 1201.14 589.04 316.90 3574.74 1790.47
OUTPUTPC 84 242.40 108.01 79.85 553.09 355.42
YIELD 91 1064.35 334.66 354.00 1895.76 1428.07
YIELDPC 84 241.39 76.73 80.04 461.84 327.12
AREA 91 11.45 3.44 5.10 22.24 15.29
AREAPC 84 2.41 0.69 1.35 4.87 3.20
HIV 91 8.56 2.81 2.23 13.00 12.00
AIDS 91 2.66 1.72 0.24 7.13 5.07
POS 91 6.89 10.78 0 36.70 24.73
NEG 91 8.24 11.40 0 64.10 25.23
PPI 84 11.32 4.52 6.30 26.28 18.49
F ERT 91 2120.43 804.51 747 .43 3777.15 3489.62
IR 91 -1.17 24.14 —48.09 25.12 21.14
SUB 84 33.55 51.96 0 187.34 129.59
FEM 91 27.65 8.02 12.15 55.08 38.89
ASST 84 4.24 4.81 0 21.01 12.66
ASPC 84 0.70 0.82 0 3.79 1.92

 

Source: Based on raw data from PHS surveys, 1991/2-2002/3, Central Statistical Ofﬁce, Lusaka.

61

Table A6. Agrozone 4 - Summa4ry statistics for dependent and explanatory variables

 

 

Variable Obs. Mean Std. Dev. Min Max 90th .
percentile

OUTPUT 299 1254.1 1 537.33 222.46 3585.52 I 2008.50
OUTPUTPC 276 254.83 108.72 49.35 681.32 399.00
YIELD 299 1139.22 428.98 446.75 4533.39 1618.33
YIELDPC 276 258.96 103.31 86.73 1071.70 368.38
AREA 299 11.88 4.04 2.92 30.38 17.31
AREA PC 276 2.51 0.79 0.57 5.44 3.55
HIV 299 10.43 7.27 1.53 29.53 21.03
AIDS 299 4.63 5.09 0.14 21.05 13.73
POS 299 2.45 5.70 0 30.74 9.84
NEG 299 10.87 10.64 0 42.38 24.71
PPI 276 12.21 4.83 5.16 29.94 19.25
F ERT 299 2166.43 787.14 747.43 3777.15 3489.62
IR 299 -1.17 24.04 -48.09 25.12 21.14
SUB 276 21.03 35.64 0 424.58 59.71
FEM 299 20.77 6.95 0 46.15 29.00
ASST 276 2.69 3.19 0 17.42 6.86
ASPC 276 0.49 0.58 0 3.16 1.35

 

Source: Based on raw data from PHS surveys, 1991/2-2002/3, Central Statistical Ofﬁce, Lusaka.

Table A7. Correlation matrix of explanatory variables (data from all agrozones

 

 

Jrooled)
HIV AIDS POS NEG PPI FERT IR FEM ASST ASPC
HIV 1.00
AIDS 0.90 1.00
POS 0.05 0.02 1.00
NEG 0.05 0.05 -0.46 1.00
PPI -0.18 -0.23 -0.12 0.03 1.00
FERT -0.18 -0.28 -0.06 -0. l 3 0.48 1.00
[R 0.23 0.29 0.13 -O. 18 -0.71 -0.34 1.00
SUB 0.04 -0.05 -0.07 0.05 0.39 0.23 -0.42
FEM -0.22 -0.24 0.17 -0.08 -0.10 0.00 0.03 1.00
ASST 0.16 0.06 0.16 0.06 0.04 0.00 -0.08 0.00 1.00
ASPC 0.13 0.03 0.18 0.02 0.03 0.02 -0.08 0.06 0.96 1.00

 

Source: Author’s calculations

62

Table A8. Correlation between HIV prevalence and AIDS-related mortality (all

 

 

_agrozones)

HIV prevalence Correlation with AIDS- AIDS-related Correlation with
in year related mortality mortality in year HIV prevalence

H 0.923 t-1 0.879

t-2 0.938 t-2 0.854

t-3 0.948 t-3 0.827

t-4 0.952 t-4 0.797

t-5 0.950 t-5 0.765

t-6 0.939 t-6 0.728

t-7 0.920 t-7 0.701

 

Source: Based on raw data from Zambia: HIV/AIDS Epidemiological Projections: [985-2010 (C80, 2005)
and Zambian population census data (C80, 1975; C80, 1985; CSO, 1994; C80, 2003).

Table A9. Mean and 90tll Percentile of HIV prevalence by agrozone and lag

 

 

La Zone 1 Zone 1 Zone 2 Zone 2 Zone 3 Zone 3 Zone 4 Zone 4
g Mean 90th Mean 90th Mean 90th Mean 90th

t 17.33 25.00 11.77 18.78 8.56 12.00 10.43 21 .03
t-l 16.85 25.00 10.96 18.73 7.93 12.00 9.92 21.03
t-2 16.27 24.98 10.08 18.46 7.25 11.73 9.35 20.90
t-3 15.55 24.78 9.13 17.76 6.51 11.28 8.69 20.80
t-4 14.72 24.38 8.11 16.81 5.73 10.37 7.95 20.80
t-5 13.77 23.28 7.08 15.35 4.94 9.86 7.12 19.90
t-6 12.71 21.83 6.06 14.00 4.18 9.10 6.23 18.40
t-7 12.22 21.45 5.47 12.22 3.72 8.35 5.79 17.01

 

Source: Based on raw data from Zambia: HIV/AIDS Epidemiological Projections: [985-2010 (CSO, 2005).

Table A10. Mean and 90"I Percentile of AIDS-related mortality by agrozone and lag_

 

La Zone 1 Zone 1 Zone 2 Zone 2 Zone 3 Zone 3 Zone 4 Zone 4
g Mean 90‘h Mean 90th Mean 90th Mean 90th

t 6.87 12.34 3.66 7.23 2.66 5.07 4.63 13.73
t-l 6.42 12.09 3.21 6.56 2.31 4.72 4.19 12.85
t-2 5.92 11.77 2.78 5.78 1.96 4.34 3.73 11.75
t-3 5.40 11.25 2.36 5.06 1.64 3.64 3.25 10.37
t-4 4.86 10.60 1.97 4.24 1.34 3.19 2.79 8.83
t-5 4.32 9.78 1.62 3.62 1.07 2.74 2.35 7.24
t-6 3.78 8.92 1.31, 3.12 0.84 2.22 1.93 5.74
t-7 3.49 8.52 1.11 2.73 0.70 1.77 1.68 5.51

 

Source: Based on raw data from Zambia: H1 V/AIDS Epidemiological Projections: [985-2010 (C80, 2005)
and Zambian population census data (C80, 1975; C80, 1985; C80, 1994; CSO, 2003).

63

larhl

.bco 385286850 0 3:0 a when.» Co 5mg— wﬂ 825me 5:: we. :oE_< 0 $3128 coco—«>03 >5 0 >5
.e\e_ E “53$ch .1. me\em 6 E8536 .. Me\eo_ 8 “585%; + demon—=28 E mozmuﬁm N do 029 3283». 2:523 53035 .3383 9 Echo.— ocoN 2882

macaw—:28 m.._o£:< SEEM

 

 

 

2%.: ﬁne see is .1353 .846 4.25.3 .4358
3.8 83:. as? 89$- Suﬁ Neva Same- $3.2- Bebe Eager/ESE
rinse
e-.>Ee%
.32—Na 3328.8
Sedan. 8342. 22:96
rand +3.:
Same 43.5, range
**A.VO.MV 0.3.35.8
$2.8 £28 22%}.
*Lvodv **CQ.NV
33mm :22 45596
5.8 95.8
Ram an: games
1%.: see see 3.8 195.9 Load $25.0 tees
gee 33:. 23.? Same- Sam $4.9 Base? 333. 5:58
nwf :32 swam :82 :wi coo—2 :3: :82 .55 >5 E 33395—
bcou Econ 3:0“ 3:0“ bee“ bee“ m¢< m1?
v ocoN v ocoN m ocoN m ocoN m ocoN N ocoN _ocoN _o:oN

 

AMEN 25; 2.9.8 no.8 :e ooze—«>2; >—: E 83.65 «Eon owaunooaoalouo a ..o 38...? 33:3 SEES—5 .:< 039—.

64

.bco mzooceanoEoo
0 3:3 "Eng Co 5w:2 we. E:E_me 5:5 mm. :oE_< u _..u< MES» .Co Swan: ms Ens—:88 5:5 ms :82:me 2:5 0 To ”8.: 85.96.... >_I n 2:
.e\e_ “a 385%? i. ”e\em 2.... “52::ch .. 2x6— 3 33:2»? + .momoﬁcoaa E motmzﬁm N we 33> 328n< .235an econswf .ocououwa 8 2&2 ocoN ...882

28:23.8 mLo5=< ”8.59m.

 

 

 

Gm: so: 5.: as: 13.: an: an: +5.:
Red $3- Show- 832. SEN NBS $5.2- $4.2- 98E :aaeeczmse
22:98
AR: $2:
9.9%. Seem era/ESE
3.5.: 28.: $5.: ism:
E _ .8. Emma. ”we. 3. 2 _. :1 22:96
83: .22.: an: awe
e8. _ «S. S- o3.» 5“. m 32%}:
E: 3.: :80: Lee:
SEN _ 38. 2mg anew range
:3: .28.: .28.: 83: team: iced
”mama- exams Seem 83. 23$ 30% range
:3: :5: ice: :32: .22.: Law:
$28 Snow: ”RAN $5.8 vows meme _-.>Ee%
am: 5.: Lax: :85: God: +3.: .52.: .23.:
Res Saw- 54.5. $362- 83. 33% imam- Sea: 5:98
:wE .802 :wE 2.82 :wE :82 :wi :32 46E >5 8 Bums—gm
bee. been So So e-.< e4... 3... S...
v oaoN v ocoN m ocoN m oaoN N ocoN m oaoN ﬂocoN ﬂocoN

 

Ease—2N

25.: .3an .5.— ..325 no.8 :e 3.3—«>03 >~E 5 8.8.55 .32— owﬁaoononnono a he Scuba 15...:— ==....to__m .N~< 033—.

65

.38» Co 5mg. wa. EzmeaE 5:5 mm. =oE_< u _..u< ”88 coco—~55 >5 n >:._

..x: 8 “:85ch .1. ”exam 8 2.85%?- .. ”foo— :w ESE—$3 + 33553: E 85%: N no 2:? 820:2 23:09.3 econcwf 6232:: 9 2&2 ocoN .332

macaw—:28 m.-_o£:< ”8.33m.

 

 

 

:3: :89: :80: 3.8:: A8.: 82: .23.: $3.:
83 mm:- ngm- 82.2:- 83- E3- Sod- 3S- coo:o Sc-wco: 2:98
22:98
28.8 38.:
2a.- 88.:- 32:98
.53.: so: .22.: tan:
GS- Sw: 3.2- 8::- 22:98
1%. : .22.: 1%. : *2 am:
83. a? E.»- oova 12:98
3.: *5: ...._.§.: 523.: 3&9 so: is: 528.:
$3- 83 $33. EOSN- 83 ca:- mvne 23.: 22:98
and: 1%.: 3.8:: is: $08 an: +3.: 3:3,:
8m.- 53 On _ .3: 232 83 an? 2: .m Sq: N..2:98
*LVwNV +ANm. C VIRGO. mv VIC-m. MV Quodv A2. 3 GNdv 35.8
82 $8- $38 39:: 83- 89mm 8:. om». 22:98
.8%: .28.: :85: :83: A8.: 3:: .28.: 3.28.:
8? on;- omit- 832- 35.2- BEM- $3:- mmodm- 2:98
:wm: :32 .35 :32 FE: :32 :wi :82 .>8W>E E BEES/m
8-2. 8-2 92 92 92 m-E 92 92
v UGON v oGON m QCON m QGON N oGON N ocoN m ocoN ﬂ QGON

 

35.2:

25.: 9:33.— ..on 2.3.5 :95 E. cone—«>9:— >—: 5 3.3.8.: «52— uwauauohonénc a .«c macaw—o 33...:— EthSE .m_< «Ea--

66

 

n
m I. II. ,._ : Pb

.23 33:83:.238 u 2:: a 3:3» Co 5:3. mm. 83:32: 53, mm. 8.5.: u 72 3:: 33.32: 2: n Z:
.3. 3 E3235 1. 33m 3 E3536 ... ”3o. 8 :aoEcwv. + 33535: :_ 3:223 N .8 23> 82034: 23:38: seal—E: 633%: 9 who-8: o:oN .382

32223.3 @853: 8.23%

 

 

 

+3.: 1.3.: 1.8:: 1.3.: +3.: +3.: 2:: 2.2.3.:
3:2 $3: $3:- 533- 2:8 8:2 :3- $3: 98:0 5:38: 2:98
22:98

3.: +3.: 3.8 *3:

3. 3:8- 2:: 23% 22:98

* *AQONV am. C *AwoNv +Ahw. C * *Amva * "28:3
83. 083. 8mm- :3. 3.3. $2:- 22:98

+3.: 3.9 3.: 3.3.: 3:: 2.3.:

8:. 3+ v2.2- $2? 33 3.3 22:98
3.: 3.: +3.: 2.3.: 3.: 2.3.: 2.3.: 23.28.:

N3»- 83 82%. M33- 3::- oﬂdw 2:8 223 22:98
3.: 3.8 +3.: 2.3.: 3.: +3.: +3.: 2.6::

:2- 32 23:. 5.3 :3- 5:: 3.3 5.3 N22:98

3.3.: 3.: +3.: 3:: 2.3.: 2.2.3.: 2.3.: 2::

3.: 83- 3.3+ 333 +3.: +2.: 53 2} 22:98

:3: 8:: *3: £22.: 5.: 3:: +3.: 13.:

3:3 3::- 33? 3.3- 5.8 3% 33.3- 83.3- 2:98
3:: :32 :wE :32 35 :32 :wE :32 Sega: a: 3323mm
8-2 3-2 9.: 3.: 8-2 8-2 2.: E:

v oaoN v ocoN M ocoN m UGON N QGON N oaoN _ ocoN ~ OGON

 

.8: 8.882— 8: 35.8 :98 .8 3.8.98.5 >=—

338 3: 25.2: 25.: 3:8
E 33.8.: 2.8: 33.88.8928 a .8 33:8 15...:— :P.-.8:m .34.. «San-

rFI-Fl

67

 

880» Go 5::0. :8. 82:38:: 5.3 :8. :oE.< n .72 .88 85.85:. >.: u >.:

...\o. :8 5:82-2:28 3.. ”fan :8 55055:? ... 2x6. 8 ::8.....::.m + 80855.8: :. 8.8.38 N ..o 258.» 8283‘ 2.28:2. =.oon;:.: .05-88:8 9 8:52 o:oN .8632

85.8.3.8 8.55.14. ”mu-Sow

 

 

 

£88.: 8:: 8:8 9:8 2:: 3.8:: +::: ...8o.:
8:- 8: 88. 88.- 8:- 82- 9”:- En- 85:0 5:88: 2:98
:88 3.3.8:: +::: 8::
8:. 8:: 3:8- 83: 32:98
Chad “*3va Amo- 3 GM. 3 3.32933 **Am_.mV ".3850. MV “683.. mV
8::- oﬂ. 8o: 8:: 831 88 .v- :8. _- $8. _- 32:98
8:8 .3: 8:8 88.8 8:: 3.8:: +88: 388.:
83.- 8o:- 8: 88.- ME..- 88; 8:.- 88: 22:98
8:8 3.8:: 88.: 88.8 3.8:: 8:: £8: +88:
88.- t m.- §8 88: E: 3: _ 88. 8: 12:98
Aomdv 3328.5 “Rm—NV 8o. 3 “.3200. MV **A.vw.Nv **A:N. mV **AN©.NV
2.8.. 8:- 88.8 88.: 63.. 8:8 :8: R: m2:98
8:: 3.8:: +88: .88.: 3.88.: 38:: 3.88.: 3.8::
:2.- .cm: 88.8 :88 at: 88.: 8:: m8. N..2:98
.88.: 8o. : 8:8 88.8 8: : 2:8 8:8 88
88.- 5. 8%. 88: SN. 88. 88.- :2. 22:98
88.: +3.: 88.8 83: $28.: .88.: 3.28.: +88:
:8.- om: to: 88.. NS: 88.: 88:- :3- 2:98
2:5 :32 ::.I :32 2:5 :32 ::.I :82 $8: >~I Hm 832.8%.
E: 3.: E: E: f: f: 8-2 8-2
v DQON v OCON m OCON m QGON N DEON N OGON ﬂ OGON ﬂ DGON

 

A3. 2: 3:: 53:22.5 .5 3:28.53 >—= E 8:22.. «5:: $559392: 8 a: 33:: .83-Em 5:48;: .m_< 038,—.

 

00
6

 

SEO 303888028 n bco a $30» Co camco— wﬂ 8.5:qu 5:: ma. :oE_< u _..u< ”88 coco—gen >5 u >_:
.o\.._ a 280$ch .1. ”fan a 28$ch .. ”foo. B ESE—Em + .momoﬁcoan 5 853.3 N mo o:_m> 838n< 2:522. soon-FEE 6:35th 8 “co-«8 ocoN .6802

223.328 mLoﬁs< ”8.39m.

 

69

 

 

£26 :2: A39 A38 1.80.8 $5.8 6n: Load
@2- 88. may ”8.- 8m. ”R. 88.- 83.- comma 5:-w8:>E©%
Awwdv **AmN.mV **Am©.NV *Amodv
88. Ba. 83- 83. :ngm
an. s an. 3 AS. : L38
#5.. 28. Eq- 82. 22%}-
:t: :6an 82: $3:
28.. Smo; 98. an. 9.2%-Rm
5.8 :6an *Eé 818
28.- v2.- 93. am. :2%%
A38 *Lﬁé i890 8w: awe +5.:
”MS.- NN..- 82 mow. won- o8. m-.>E©%
AmN. C *VLNQMV "Iago-NV **A.v©.mv Codv "Inge-NV
$8.- 28.- a: N8. 35.. 22- ~-.>E©%
95.8 an a 3.8 so. : ans 1%. 3
$8.- 38. En. an my. go.- 2:58
63: 28.8 5.: 8on 308 ..de 3n: 18.3
$8.. §. 83- SN- 09. 8: 88.- 88.- Siam
:3: :82 :3: :82 nwi :82 E: :8: sea 23 a 88:35
$< E< W2 W2 m-Z mi 38 H Nee H
v OGON v DCON m OGON M OGON N DSON N QSON _ UGON — @GON

 

A3383

2: 333 .39 3.3 3.95.5 .5 3.5.553 >—= E 83.5.: .52— owaeuoouoauono a a: 88:9 .52.:— Ezutcam .e—< «Zah-

 

.38» Co Swan. mm. .25—£98.: 5:5 mm. :oE_< u TE $30» Co sumac. mm. 5.5588 5:: ms 32::me exec u TO ”BE 2585:. Baa—0.7mm; n me<
52; E Ewe-tam; 1. 32% 8 .50c_:w_m .. 322: 8 EnocEwi + $985228 5 85:8,.“ N .3 o:_m> 820mn< 65:082. soon—.3: .05-3%.“ 2 who-*2 ocoN .6292

323.328 9.553.. ”mu-59m

FIT-2.

 

70

 

 

29.9 22.9 an: 899 +239 2%.: 228.9 28.9
m9;- oooam- 33$- Qméc- 9.39 399: 22:. 529- 3-90 5.329829%
322.9 228.9 1239 329.9
3.339 v8.39 Sum-9.- 83mm- 23296
23.: 123.9 28.9 .3239
”a: _ m- 2 ZS- :Sm- 98.3- 28296
22.9 23.9 323.9 22.9
83% 5.9% :38 $39 awe-«9%
...2mm9 323.9 328.9 .3239
3209 3239 09.9% $8.8 33296
2%: .1899 129.9 322.9 128.9 .2289
$3:- 233- $25 83:; 82: N892 28:95
28.9 228.9 239 3239 15.9 323.9
3.3; €1.99 ”3.5 2.39: 3;: @232 28296
22.: 2.3.9 28.: 699 28.9 23.9 32999 25.:
Room“ 828 :32. 83m $95- woodmw- 9.3:- www.8- 7.829%
28.9 .3239 5239 328.9 .239 28.9 .2239 $399
Sad:- 93? 29.82- 833- 823 $38 3.2%. 29%? $929229
:9: :82 a? :82 as; :32 E: :82 2.3% mm?- a B§§m
92 92 92 92 Zn in 92 92
v DGON v DGON m “EON m ocoN N DQON N ocoN _ DGON ~ UGON

 

92a

25.9 39.8 99.5 .5 8a.. mam—8.88 tuna—enumn:< 2: 5 3.3.5.: «£2. «wauaoohon.o=c a .3 38am Emu-:3 5.2-225 .5:- 03:.

.832 Co 582 m2 E3552: 5:: mm. :oE_< n _..u< $302 Co 582 ms 82:98.: 5:5 :9 :82:me SEC u TO 6:2 258.88 BEE-ma?- " me<
52.. E 230$ch i. 32% “a Emoﬁcwﬁ .. 322: a 2.onch + 88553: E 83922.. N 20 029 3:.08< 6:288 ...oonnwE .258th 3 who-22 ocoN .882
8253.8 @8532 ”839m.

 

 

 

~89 :99 an: 28.: 299 :99 22:9 239
:8. $2- ”8.2:- :8.-N- $9. 8:- 923m- oté- 98:0 :E-w=o_:8292©
sigma-Q “.2..qu
$2.9m- $28- 2.829%
:9: 223.9 2299 28:
08.9:- 8929 89%. 25.2- 0.8296
23.: 28: 59 289
£99: 92% ”92 99 28292v
2289 223.9 222.: 2%.:
99.8 99.28 8.8 99.2 2.829%
223.9 225.9 22%.: 2a.:
:33 88% 898 90.3 M.8292:
23.: 2229 2239 25.9
$3: @828 £99- 82 2.829%
239 6%: 2:9 2:: .289 2%.: .289 $9
an: 30.: 53m- 92:: 982:- «9.22- 259- :99 28296
289 2%.: .289 .299 129.9 22: 2289 .2929
9:2- :83- $88- 8929 was: 822 83:- 92%. 82922V
:9: :82 :9: as: :9: :82 :9: 582 88¢ 82 a 83:3:
:8 .8 $2 $2 8 :8 :2 0-2
v DGON v OCON m OGON m OCON N OGON N ocoN ﬂocoN ﬂoGoN

 

23389.2: 25.: £23
.3: 35.5 :95 .8 8a.. 23.3.8:— uoeﬁouﬁaz 2: 5 980.53 «5.5 «wax—3390:: a ac 398.3 23.3.— :E-tc__m .a~< 03:-

71

L

.35 macezﬂanoEoo u 3.5 u Mano» Co Swag mm. EzEcaE .23 wﬂ =oE.< u 3(1an €23.55 Bum—2&9: u we?

.o\o_ 8 28533 1. ”ﬂow 5 E8E:w_m * ”obo— 3 «:85ch + 3855.8 E 83.33 N we 02.; 8203< 23:3th soonswi .oconohwa 3 ﬂow“: ocoN ”.882
228.228 {053‘ ”830%

 

 

 

328.3 .1893 1&3 $23.0 32: $3 3.5.0 5&8
and. 30.2- 23;. 322 mg. 33 5.2:- 3.2- 98% 5:-wcocm9<©%
28:6}.
:6me :38
E3- SEN- gazgm
328.3 ...God 5.: +65:
£3: 55.3: tam $2- 282sz
1.3.8 3229 6:: $2.: :Swd 13.:
5.9.- 92%. 3:. 35% £52 392 182%
1.5.3 .imwd 2%.: ANN: :Gwd 22.3
02.8 22$ 30.3. ”2.8- 89% at: £9<$€
125.8 .503 A8.: +39: 1.3.9 1658
Ed. $13 82m- 33% ~32 S3; N-.m9<$€
3:99 3.80.8 ﬁne *amd Lid $2:
$9: 2.3 5.2- $3. 23- 3.. 7.829%
2303 39%;; $8 5.: Gas Gas 18:3 Land
5.3. 03.2. N3: $23 3. 33 23¢- SEN- .329me
3:36
:3: :82 $5 .302 .3: g: :9: :82 82 II a B§§m
v-2 v-2 9% WE baa H Nae H oi 92
v ocoN v ozoN m ocoN m oaoN N ocoN N ocoN _o=oN ~ocoN

 

925:5 2:5 258..
..on 2.5:: AFB :9 3a.. hem—3.8:. B.a.?..mﬁ—< 05 am 8.33.: «£2— owﬁuogonénc a a: 398.3 15.3.. 52-9.25 $3.. 033.

72

r

.bco msoocﬂomEoEoo u 3:0 a ”28» Co paw—.2 mm. 525qu 5:5 mm. coE_< M 73128 5:358 :Bm_2-mn=< n mn=<
.o\.._ 8 Enoc_:w_m .3. ”Avon an 282.:ch .. ”$2 8 “535%; + $8955.89 E 8523 N :0 o:_m> 8203< 65:3th scone—mm: 6:39? 8 2&2 ocoN .6202
30:23.8 9853‘ ”8.30%

 

 

 

$2: 160: in: 32%.: 3.: am: an: 9on
23 ENE- 8E2 832 $3. 33- Rod- $3. 98% ca-w8:me<©%
:2: :69:
53 23:. zmozmim
15.: 5.:
2m: 83. e-.me<$€
28.: 32%.:
8?: 32m 9.836%
39: ...qumv 1"?on IANONV
$3: Rn; 5.8m 82; £9<$€
$5.: 28.: +86: 3.:
33m- 30% v8.8: 838. £9<®€
328.: AR: Low: .23.:
$08- Sad 35%. 83mm- zmazgm
La: :20: so: is:
32:- :38. 222. 833. 7.829%
:E: :93: +69: 2%.: 5.: am: am: 3on
3:8 23%. $33 83% £3- was- 2?. 5:. .829@
:3: :82 :3: 582 :3: :82 E: :82 23% mm? a Baggm
W2 :2 v-2 «.2 baa H 38 H 3:3 :8 H
v ocoN v ocoN m ocoN m ocoN m ocoN N ocoN _ ocoN _ ocoN

 

ASE-3 ..om «Ev—2N 25.: .3an ..2— 9.392.
..2— 2.9.8 :98 .5 8a.. mum—3.8.: :33??mﬁ—< 2: 5 33.53 .52. $8523.05 a we 38:9 15...:— :=..-t2_m 52 «Bah.

73

.225 mzoocﬁanoHEo u 2.5 H .88» Co Swan. ma. £38.52: 5.3 wa. :oE.< u T< .28 9.8.8:. :on.2-mQ.< u me<
..xH. Hm H§oc_=w_m I .me Ha 2.35—Hm? H. 2.22: 8 HERVE—Hm; + 38558.. 5 8.3.83 N ..o 2:? 3283‘ 6.3522. ...oannwi 6:85.? 8 2&2 ocoN .382

223.228 3.053.. ”8.50%

 

 

 

23.: 22:: 2mm: 22.: 22:: +28: 2%.: .22.:

Non. E. gm..- NH: .53” 23 mm; 3: 98% §-m§:m9<©%
2:: 22%.: 228.: .22.:

3o. 33 S: :2 28.6%
2:: .28.: 323.: 2a.: 28.: 322.:

88.- a. a: 22 +8. 33. 282::
25.: 2a.: .223: 328.: 23.: 2mm:

8:. an- 2:- £3- 2:- 3o. H-.m9<:€
2%.: .22.: 228.: 32;: 2%.: 32%.:

NE; +8.- ”8+- Ram- Rn- $3- 382:»:
2%.: .28.: 32%.: 329: 2:: 1.62.:

mom. Maw; 85.- ”a: £8. 83. awmgim
2%.: 28.: 32%.: +23: 8:: 32%.:

3.. 3H..- 63. £3. 2m. .8..- wweﬁtﬂm
2%.: .23.: 2%.: 22%.: 2.2.: 28.:

N2. 8m. 2:. a3. :8. EN. 7.829%
28.: 2mm: 2mm: 2:: 32%.: 222.: 3:: .23.:

o3- a? SN..- 22 E; 83 2a.- :3 gain.
HE: :82 E: :32 E: :82 E: 502 23% mm? Hdw BH§§m
:2 2-2 2.8 H 2.8 H WH< 2-2 92 0-2.
v ocoN v ocoN m ocoN m ocoN N ocoN N ozoN . oaoN . ocoN

 

3 2:

3.3 6355.3 .3 03.. 29:35:. 625.97mnz 2: a. 8.33:. 2...:— owauuouuuauog a .3 38¢» .33.:— :=...u.::.m .—N< 035.

74

.88» Co 5w=2 wa. 8255:: .23 wﬂ :oE_< u _..u< $802 Co Swan. wﬂ EEExaE 5:5 wa. :83:me BEE n .30 ”BE b:mtoE BEE-m0; n me<
52.. an 282::ch ..i 82% an “505$? .. 322: an “:85ch + .8355th E mozmzﬁm N no o=_m> 8203.2 .0228th eoouswf duo-28w“ 2 "co-:2 ocoN .6202

32223.3 m...o£=< “mu-Som-

 

 

 

22: 22:: 28.: 2%.: .23.: .28.: .22.: 228.:
E. Sm. ca. 8:.- oS. m8. 8:.- SN. 2woomo§-wco:m9<:€
2K: 228.: 228.: 2:: 28.: 225.:
«mm. 82. 82 NR: 82. 32. awe-«:28
28.: 228.: 22s.: .28.: :2: 28.:
R8. 5. 8m. 82 E8.- or. 28296
:2: 22.: .28.: cm: 2mm: 228.:
$8.- 88.- 8:. 8? 88.- 98.. 282:8
28.: 25.: 22%.: .22.: $2: 222.:
:8.- owT 82.- an- S8.- 8? 2.828%
28.: 28.: 22mm: 226.: 2:2: 128.:
88.- 8..- 2? :8.- _m8. m8..- 282:6
23.: .28.: 228.: 228.: 28.: 225.:
88. M: _.- 8m- 5} 88. M: _ .- 282928
28.: 28.: an: 28.: 28.: 28.: 28.: .28.:
$8. 8.8. 82- 88.- $8. $8. 888. 38. 28298
am: 28.: 5.: 28.8 228.: 228.: 28.8 .28.:
88.- :N. a? «a. 82 88. m3.- 8 G. .8298
:wE :82 :wi :82 gwi :82 E: 502 328 mm:- 2 B§§m
2-2 WE in 75 2-2 2-2 E< 2-2
v ocoN v 28M m ocoN m ocoN m osoN N oaoN _ BEN _ 28M

 

Astana}:— 2: .3an .3:

3.5 6355—3 :9 8a.— b=ut¢5 toga—97mm:< 2: E 02353 «E...— owﬁuoaoméuc a .3 30qu 32.2. 5.2.225 .NN< 03am-

 

75

Table A23. Summary of Almon lag estimates for the short-run partial eﬂ'ects of
HIV/AIDS on select agricultural production outcomes

 

 

 

 

 

 

 

 

 

Agrozone l 2 3 4
HIV/AIDS variable HIV AIDS HIV AIDS HIV AIDS HIV AIDS
Crop output 0 n NA NA NA 0 NA 0
Crop output per capita n n U NA NA n NA NA
Crop output/ha m n 0 NA 0 U n 0
Crop output/ha per capita n NA U NA n U n Ambig
Cultivated area m Ambig n U n NA n U
Cultivated area per capita NA Ambig U U n NA U U

 

 

 

 

 

 

 

 

 

 

Source: Summarized from tables A10-A2]

Notes: HIV = HIV prevalence; AIDS = AIDS-related mortality rate; 0 = lagged partial effects are
negative/zero, then positive, then negative/zero again; U = lagged partial effects are positive/zero, then
negative, then positive/zero again; NA = not applicable — preferred lag structure was not an Almon lag for
that combination of outcome variable and agrozone; Ambig = ambiguous ‘shape’, i.e., not clearly n or U.
A priori, I expected the Almon lag estimates to be zero, then increasingly negative, and eventually zero
again as agrarian communities recover from the HIV/AIDS shock, i.e., a U shape. This was the estimated
shape in 10 of 3] models where the preferred lag structure was an Almon lag; the estimated shape in the
other 21 Almon lag models was 0 contrary to a priori expectations.

76

'_'li

u. D _ .l
..m . D. ... ..t U..Fou‘-mﬂ

 

 

.8: E: .8: 2 . 2: 2:8: 28.:
.83 88.: 8...: SM... :88 2.28.8 GEN 8: .7: 8.5. 8.5. .8585. 85.8.82
.8. : .8: 28.: a . .: :8. : :2:
+8.8- 8...: 2.8...- ..._..8¢.o- .88- 8m..- 88:8 .585 8.8.: 52. .5838 8 .82.” .855 2.882
28.: 28.: 28.: 28.: .8. : .8. :
.2: m8...- 228... 2.83 .8... .88 =85 8.8.: 58. .5838 8 .82.... 8.55 2.882
28.: .8. : 28.: .8. : : ...: 28.:
m8...- +8...o 8.8- 8.8- 8°... :8... 88:8 .585 8.8.: 5o... .5838 8 .82.a .855 08.8..
28.: 28.: .8. : :8. : 28...: 28.:
:n. + .88- 23 +N8.N 8.... M8,... =85 8.8.: 52.. 88.5.. 8 .82; .855 28.8..
:2: 28.: 22.: :2: .8. :
88.8- .. . 8... . S...- 8 . .8. Re. .- 8.8.8 .2552
28.: .8: :2: 28.: .8. :
8.8 .88. .- 8.8 23 888 88.8 .2852
28.: :N: 28.: 82: .8. :
8.....- 2.8.. .88- ~88- .88- 88.8 .5052
82: 28.: :2: 28.: .8. :
8.8- 1.08.8- :8m.:- .802- +888- 2552
:3: 28.: .8: 8m: .8. :
.88 1.82.8. :3 n8 .. . 8.8 88...:- N852
3...: 28.: :m: E. : .2. :
$3. 28887 88.8. 8.2: 88......- .852
:2:
w. ..o- 3 383.8 02.0.80... >=..
28.:
8.... 8 88.56:. >.:
3.5 : 3-2: .22: 3-2: .92: 3.2:
.5. ..o. .5. ..o. .mémzw 8?. .825 .8% 8.... .88
8.98208 :08 8.93 .0: coo: 8.98 coo:
“00830.30 v0.8.2.5 8.28930 «£8920 ..0: 89:0 89:0 333:; 2585.93.

 

228.5... 8.23.: a. 2.85; 8:
.85..

030.88 80.82.00

 

@828 88008.58. $58.08. 88 00:0_u>0.:ﬁ>—=wm_.m= £0.88 ~ 0ccnca< 89¢ 8:50.. 88.80w0m .vnd- 033—.

 

77

 

.[F

 

78

 

 

5......5> 823.: 5. 5.52.... w...
7.5:.

535...... 25:59.0

.2... .85. 8...... $2. .8...
8...... :85...- :N~....- .58...- ... ....- 5..5...:...-..._.:8 .5... :55... 35.5..
...m... .8... .53.. .53. :v. ..
8.... :RN... :3 .... .8... En... 5552.853... .5... :55... 55......
.8... .55.: .8... an: .3...
8...... : .2...- :8...- ..........- NS..- _5.52....5._.5..m .55 :55... 26.5.
.2... .2... .53.. ..N. .. .8...
8.....- :8... :85... 5...... mm... 5552.55... :55... 22.5..
:N... ...3. :3. .3. .. .8...
m S... :55? :8...- ch- .2...- N555.4555... :55... 2,...8.
.2... .25. .55.. :5. .. .8...
o_o.ou 33:06 **w©v.m vow.N MGNd EOE—<3??? awn—:2 03.30;
.8...
coo..- 8.8.55... 2:53.535 .62.. =m-EE. 03......—
.... ...
~No.o- coco—«>95 >=.__..v_oor_m zﬂcme o>£mom
...,... .8. .. .8... .8. .. :5... .2...
. .N. .- 5.....- 35... N3... 3 m...- Nmz- .. ...EN 5.5.5.. 5.5.5 m5.53.5... 25825.. :83.
.52: .2... :5. .. Gm. .. .3... .8...
5.... .- .. 3.... +33 2. .. .5... Sm... .0... 5.5.... 5 8.2552. 882.5555.
.8... :5... :5... :5...
cm...- ......- m . N... 5......- ....25.=o§£w._. .258 55...... 55.2
.8. .. 5.: .2... .8... .85. .2...
3 ..m :53... :23. :83 :8... :52... . .-.. ......n......~. .5. .555. .8:
.8. .. .8... ...3. .8... .8... .85.
+82. :8... 358.8 :83. 5.5.8 .5252 .5532 5.... 8.... 55...... 5.5.5... 5.8...
3.5 : ..-2. .92. ..-2.
. . .22. 3-2.
5 ... 5 2. 5 :22... .
SLMEBM _ F. no.” H a. 8.98 .0.“ TED—EN ﬁx... EZNMQL EZN 8....
8.53.30 8.53.30 2.52.50 2...: ..O 5.. 3.9.5 5.9.5 83...; beggaﬁ

 

8:558 ...2 m...,...-

 

 

 

79

 

 

no.3 ems and ~33
t. 33 :85 :83 :23 geiksﬁa SEEe so:
25.3 :«d 3.8 and
:82- 3:5- :Qvo- .13.? aoe_<_..§33 Earn; 522
23.3 so: cod And
:9? +2 ..o $8.0 3.: 50523233 .355 :82
00:02;th >_I*%Em03m Size—om Gav—2
25.8 8 ..e 5:: 5.8 55
83.- ~85 22. 33 *8; ”852.3333 v.85 .552 2332
:2: Gas 3.8 82: : ..NV
83 m _ 3.. Rod- 08.? :85. aoéfuoaag v.85 .552 2332
A38 82: 93: ﬁne A3. :
53. :3 :3 23. .32 _85::Ea...um v.85 .525 02302
868 25.8 83: 32: she
83 :2 $3 82 33. £323.85 359 02332
32: age 95.8 €on 8::
v8.9- :3- EN..- 5.. _- 03.2 8082385 .558 2:532
GE: 62: 80.8 €98 5:8
V SS. 82 83 E: 23- ...erLooﬁ .358 2332
God
~Bod 85355 22.38:?“ :85 =££E o>tmwoz
so. :
mNméu oo=o_m>o.a >_I*v_oo£m =£EE o>tmwoz
3-2;
2.5 a 6-3; 3-5 3.35
.2 ..a .2 .2 HSEN EDEN .25 83 a .3
SEmoBEa 83 coo; coop 8&8 EZN coo;
382230 Bags—:0 «:98 Ba «£8930 5a 59:0 59:0 8 b a
. . «£395 32:; 922 E
@325, 832: 5 232.5. was

323
93253 E02250

 

8:528 £4 2...;

.bco 303888028 u b5 H 3.30» .Co ﬁmcﬂ ms 525me :83 mm—
coE_< n E < 55:8 8&8 Sn 5 2 ﬂoat? Benson—ow 05 5:3 2:8 838 Sn 5 0.8 $38 252605 22.8.5: :38 n + 89.8352: 05 .3 N.m .5303 E
cop—Cow mm 08 2582 EB .NcoE_< .5082 ..x: 3 ESE—Em 3.. ”ﬂow a ESE—hm? .., xxx: 3 “:85ch + .835:qu 5 833.5 N 80 33> 8285‘ ”882

223328 m1052< ”8.50m.

 

 

N. N. N_ N_ N. N_ 925.8% 525.2
N8 N8 N2 N2 N2 N: 2232030
:08 :38 SN. 8 ANNNV 5.8
258 :Nono- +8 No. .2 _ Nam- .23.- 82:2 L338 3538a 223:2 :82
828 8N8 as. 8 A88 8va
:38- in 3Nm $88 :NNNNN ...SNNN 2552 1388 3:883 228:2. :82
$08 3.8 A8. 8 sod and
:38 :SNdN- $38- :SNé- ...2 E _- 52:2 Leona 3:885 22352 :82
Go. 8
w :3 85—305 >5138mm 028.605 20:83: :82
3.. 8 5.8 $08 85.8 $8
838 Sod 88.8 3.83 NNod 2852 $22.38; 883-0850”.
3. 8 :08 828 g. 8 : N.8
55.8 23- Nde +883- Em. _- 82:2 322.88; 882.230".
Gm. 8 8 _ .8 3o. 8 ANN. 8 $8
A 8 :58 Rod- and- 88.. v3. _- .852 $22.88; 882-230“.
N28
356 coco—«>95 >—:*m20£0m:or— nutmeg—no—anm
SE 8 3.28 3.28 3.28
. . $-28 $-28
8 _ 8 a _ 8 a £6 88 2 8 .
«NE—Nuke; H z 8.8 _ N: 838 SN“ FéMZN who; 3. Nghuv EZN ﬂog
839230 889230 «£5930 SS: :O .2. 39:0 3 =0 8323> boumcmixm

 

@325 mesa: 5.. 232:8 mm:
was.

033:; Eovcoaoo

328:8 ...N< 2...;

80

REFERENCES

81

REFERENCES

Askari, H., and J.T. Cummings. 1977. Estimating Agricultural Supply Response with the
Nerlove Model: A Survey. International Economic Review 18.2: 257.

Barnett, T., and P. Blaikie. 1992. AIDS in A frica: Its Present and Future Impact. London:
Belhaven Press.

Barnett, T., and A. Whiteside. 2002. AIDS in the Twenty First Century: Disease and
Globalization. New York: Palgrave Macmillan.

Baylies, C. 2002. The Impact of AIDS on Households in Rural Africa: A Shock Like Any
Other? Development and Change 33.4: 611.

Bryceson, D., and J. F onseca. 2006. A Dying Peasantry? Interactive Impact of Famine
and HIV/AIDS in Rural Malawi. In AIDS, Poverty and Hunger: Challenges and
Responses, ed. S. Gillespie. Washington, DC: IFPRI.

CGAIHS. 2000. Time from HIV-1 Seroconversion to AIDS and Death before
Widespread Use of Highly-active Antiretroviral Therapy: A Collaborative Re-analysis.
Lancet 355.9: 210.

Chapoto, A. 2006. The Impact of AIDS-related Prime-age Mortality on Rural Farm
Households: Panel Survey Evidence from Zambia. Ph.D. dissertation, Michigan State
University.

Chapoto, A., and TS. Jayne. 2005. Impact of HIV/AIDS-related Deaths on Rural Farm
Households’ Welfare in Zambia: Implications for Poverty Reduction Strategies. Food
Security Research Project Working Paper No. 15. Lusaka: FSRP.

Chapoto, A., and TS. Jayne. 2008. Impact of AIDS-Related Mortality on Farm
Household Welfare in Zambia. Economic Development and Cultural Change 56: 327.

Chapoto, A., T.S. Jayne, and N. Mason. 2007. Security of Widows’ Access to Land in the
Era of HIV/AIDS: Panel Survey Evidence from Zambia. Food Security Research Project
Working Paper No. 25. Lusaka: F SRP.

Chewe, J. 2006. Personal communication. Statistician, Central Statistical Ofﬁce, Zambia.
June 7.

C80. 1975. Population Projections for Zambia, 1969-1999. Lusaka: C80.
C80. 1985. 1980 Census of Population and Housing. Lusaka: C80.

C80. 1994. 1990 Census of Population, Housing and Agriculture. Lusaka: C80.

82

 

tn"

 

CSO. 2003. 2000 Census of Population and Housing. Lusaka: CSO.
C SO. 2005. Zambia HI V/AIDS Epidemiological Projections: [985-2010. Lusaka: CSO.

del Ninno, C., and A. Marini. 2005. Household’s Vulnerability to Shocks in Zambia.
World Bank Social Protection Discussion Paper No. 0536. Washington, DC: World
Bank.

Devereux, 8., ed. 2007. The New F amines: Why Famines Persist in an Era of
Globalization. London: Routledge.

de Waal, A. 2003. New Variant Famine’: Hypothesis, Evidence and Implications.
Humanitarian Exchange 23: 20.

de Waal, A. 2004. Evidence for the ‘New Variant F amine’ Hypothesis in Africa. Justice
Africa GAIN Brief No. 2. London. Available at: http://www.justiceafrica.org/wp—
content/uploads/2006/07/DeWaal_NewVariantFamine.pdf

de Waal, A. 2007. AIDS, Hunger, and Destitution: Theory and Evidence for the ‘New
Variant F amines’ Hypothesis in Africa. In The New F amines: Why F amines Persist in an
Era of Globalization, ed. S. Devereux. London: Routledge.

de Waal, A., and J. Tumushabe. 2003. HIV/AIDS and Food Security in Africa: A Report
for DFID. Available at: http://www.sarpn.org.za/documents/d0000235/P227_AIDS_Food
_Security.pdf

de Waal, A., and A. Whiteside. 2003. ‘New Variant Famine’: AIDS and Food Crisis in
Southern Africa. Lancet 362: 1234.

Drinkwater, M., M. McEwan, and F. Samuels. 2006. The Effects of HIV/AIDS on
Agricultural Production Systems in Zambia: A Restudy 1993-2005: Analytical Report
IFPRI/Renewal Project, CARE, International HIV/AIDS Alliance, FAO, SIDA.

Gillespie, 8., ed. 2006. AIDS, Poverty and Hunger: Challenges and Responses.
Washington, DC: IF PRI.

Gillespie, S., and S. Kadiyala. 2005. HIV/AIDS and Food and Nutrition Security: From
Evidence to Action. Food Policy Review 7. Washington, DC: IFPRI.

Govereh, J ., and M. Wamulume. 2006. Strategic Review and Explanation of Trends in
Agriculture, Poverty and Major Related Issues in Zambia. Lusaka: FSRP. Draft report.

Gujarati, D.N. 2003. Basic Econometrics. New York: McGraw-Hill.

. I

Hoddinott, J. 2006. Shocks and Their Consequences Across and Within Households in
Rural Zimbabwe. Journal of Development Studies 42.2: 301.

Jayne, T.S., J. Govereh, A. Mwanaumo, J .K. Nyoro, and A. Chapoto. 2002. False
Promise or False Premise? The Experience of Food and Input Market Reform in Eastern
and Southern Africa. World Development 30.11: 1967.

Jayne, T.S., M. Villarreal, P. Pingali, and G. Hemrich. 2005. HIV/AIDS and the
Agricultural Sector: Implications for Policy in Eastern and Southern Africa. Electronic
Journal of Agricultural Development Economics 2.2: 158.

Jayne. T.S., A. Chapoto, E. Byron, M. Ndiyoi, P. Hamazakaza, S. Kadiyala, and S.
Gillespie. 2006. Community-level Impacts of AIDS-Related Mortality: Panel Survey
Evidence from Zambia. Review ongricultural Economics 28.3: 440.

Larson, B., P. Hamazakaza, C. Kapunda, C. Hamusimbi, and S. Rosen. 2004. Morbidity,
Mortality, and Crop Production: An Empirical Study of Smallholder Cotton Growing
Households in the Central Province of Zambia. Boston: Center for International Health
and Development, Boston University of Public Health.

Mason, J., A. Bailes, K. Mason, 0. Yambi, U. Jonsson, C. Hudspeth, P. Hailey, A.
Kendle, D. Brunet, and P. Martel. 2005. AIDS, Drought and Child Malnutrition in
Southern Africa. Public Health Nutrition 8.6: 551.

Mather, D., C. Donovan, M. Weber, H. Marrule, and A. Alage. 2003. Household
Responses to Prime Age Adult Mortality in Rural Mozambique: Implications for
HIV/AIDS Mitigation Eﬂorts and Rural Economic Development Policies. Maputo:
Government of Mozambique Ministry of Agriculture and Rural Development and
Michigan State University.

Mather, D., C. Donovan. M. Weber, H. Marrule, and A. Alage. 2004. Prime Age Adult
Mortality and Household Livelihood in Rural Mozambique: Preliminary Results and
Implications for HIV/AIDS Mitigation Efforts. Reprint of paper prepared for the Annual
Conference of the Centre for the Study of African Economics, Oxford, UK.

Megill, DJ. 2004. Recommendations on Sample Design for Post Harvest Surveys in
Zambia Based on the 2000 Census. Food Security Research Project Working Paper No.
10. Lusaka: F SRP.

Morgan, D., C. Mahe, B. Mayanja, J .M. Okongo, R. Lubega, and IA. Whitworth. 2002.
HIV-1 Infection in Rural Africa: Is There a Difference in Median Time to AIDS and
Survival Compared with that in Industrialized Countries? AIDS 16.4: 597.

Mutangadura, G. B. 2005. Gender, HIV/AIDS and Rural Livelihoods in Southern Africa:

Addressing the Challenges. JENDA: A Journal of Culture and African Women Studies 7.
Available at: http://www.jendajournal.com/issue7/mutangadura.html

84

 

Nerlove, M. 1958. The Dynamics of Supply: Estimation of Farmers ’ Response to Price.
Baltimore: Johns Hopkins Press.

Pindyck, R., and D. Rubinfeld. 1997. Econometric Models and Economic Forecasts. New
York: McGraw-Hill.

SADC. 2003. Towards Identifying Impacts of HIV/AIDS on Food Insecurity in Southern
Africa and Implications for Response: Findings from Malawi, Zambia and Zimbabwe.
Vulnerability Assessment Committee, Harare, Zimbabwe.

SAHIMS. 2003. Girls Exchanged for Food. SAHIMS editorial. Available at:
http://www.sahims.net/archive/Briefcases/Z003/08/review_ma1_06_08_03new.htm

SCF. 2002. The Livelihoods of Commercial Sex Workers in Binga. Save the Children
Foundation Report. Available at: http://www.sarpn.org.za/documents/d0000080/P85_
SCF_Binga.pdf

Semu-Banda, P. 2003. Daughters for Loan under Fire. The Nation Online. Available at:
http://www.sardc.net/Widsaa/Gender&Media/daughters.htm

Steinberg, M., S. Johnson, G. Schierhout, and D. Ndegwa. 2002. Hitting Home: How
Households Cope with the Impact of the HIV/AIDS Epidemic: A Survey of Households
Affected by HIV/AIDS in South Africa. Report compiled for the Henry J. Kaiser Family
Foundation and the Health Systems Trust. Available at: http://www.kfforg/southafrica/
20021 125a-index.cfm

Tumushabe, J. 2005. HIV/AIDS and Changing Vulnerability to Crisis in Tanzania:
Implications for Food Security and Poverty Reduction. Paper presented at the
International Conference on HIV/AIDS, Food and Nutrition Security, Durban, South

Africa. Available at: http://www.ifpri.org/events/conferences/2OOS/durban/papers/
tumushabeWP.pdf

UNAIDS. 2006. Report on the Global AIDS Epidemic. Geneva: UNAIDS.

Webb, D., and G. Mutangadura. 1999. The Socio-economic Impact of Adult Morbidity
and Mortality in Kafue District, Zambia. Lusaka: SafAIDS.

WHO and UNAIDS. 2003. Reconciling Ante-natal Clinic-Based Surveillance and
Population-Based Survey Estimates of HIV Prevalence in Sub-Saharan Africa. Geneva:
WHO and UNAIDS. Available at: http://data.unaids.org/UNA-docs/ANC-
Population_surveys_report_en.pdf

Wooldridge, J .M. 2002. Econometric Analysis of Cross Section and Panel Data.
Cambridge, Massachusetts: MIT Press.

85

World Bank. n.d. Zambia Government Programs to Promote Fertilizer. Available at:
http://www.worldbank.org/afr/fertilizer_tk/bpractices/ZambiaGovProg.htm

World Bank. 2004. Zambia Country Assistance Strategy. Available at:
http://www.wds.worldbank.org/servlet/WD S_IBank_Servlet?pcont=details&eid=0001 60
016_20040423185130

World Bank. 2005. World Development Indicators, 2005. Electronic database. Available
at: http://web.worldbank.org/WBSITE/EXTERNAL/DATASTATISTICS/O,,content
MDK220398986~pagePKz64133150~piPKz64133175~theSitePK:239419,00.htm1

Yamano, T., and TS. Jayne. 2004. Measuring the Impacts of Working-Age Adult
Mortality on Small-Scale Farm Households in Kenya. World Deve10pment 32.1: 91.

Zaba, B., A. Whiteside, and J .T. Boerrna. 2004. Demographic and Socioeconomic Impact
of AIDS: Taking Stock of the Empirical Evidence. AIDS 18.2: 51.

86

   

IIlilillllllllliliI