2007

 

This is to certify that the
thesis entitled

IDENTIFYING THE DETERMINANTS AND
CHARACTERISTICS OF MOTHERS' INCOME IN RURAL
MALI

presented by

MEGAN ELIZABETH MCGLINCHY

has been accepted towards fulﬁllment
of the requirements for the

MS. degree in AGRICULTURAL
ECONOMICS

 

Vaﬂ g/é‘asX

 

1/" Major Professor’s denature

/ 1/ ?/ ﬂail:

Date

MSU is an Afﬁrmative Action/Equal Opportunity Institution

 
    
   

  

UBRARy

Michigan State
University

I
I
i
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
‘i
I
I
i
I
.
I
I
I
.
I

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/C|RC/DaIeDue.indd-p.1

IDENTIFYING THE DETERMINANTS AND CHARACTERISTICS OF MOTHERS’
INCOME IN RURAL MALI

By

Megan Elizabeth McGlinchy

A THESIS

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

MASTER OF SCIENCE
Department of Agricultural Economics

2006

ABSTRACT

IDENTIFYING THE DETERMINANTS AND CHARACTERISTICS OF MOTHERS’
INCOME IN RURAL MALI

By

Megan E. McGlinchy

This study investigates the determinants and characteristics of women’s income in Mali.
Malian men and women do not entirely pool their incomes within the household, and
women’s income is particularly important in inﬂuencing child health and nutritional
outcomes. The study estimates two different models: an income determinants model and
a model that describes different categories of women based on their income-generating
activities. Results from model 1 show that women in the irrigated rice zone have the
highest incomes, ceteris paribus, followed by women in the coarse grains zone. The
cotton zone produces the lowest women’s incomes, despite having the highest men’s and
household income levels. Other signiﬁcant determinants include the women’s age, being
married to the head of the household, the composition of the household and asset levels.
In the second model, different income-generating activities have different impacts on
incomes, depending on the region in which the women live. Women in the cotton and
coarse grains zones receive the highest marginal beneﬁt from participating in the sale of
wild products, such as shea butter, but not in the irrigated area, where more proﬁtable
agricultural activities exist for women. The results imply. that agricultural grth and
higher household incomes do not automatically lead to gender equity or better welfare for
women and children. In the cotton zone, particularly, complementary interventions are

needed to counteract the negative impact of cotton production on women’s incomes.

Copyright by
MEGAN ELIZABETH MCGLINCHY
2006

To my parents, for everything

iv

ACKNOWLEDGEMENTS

I would like to thank my committee, composed of Dr. John Staatz, Dr. Valerie Kelly and
Dr. Steve Esquith, for their constant support and encouragement. This project beneﬁted
greatly from their in-depth knowledge of Mali; I have beneﬁted greatly from their

mentorship. Dr. Nango Dembele also provided valuable comments on the paper.

I am extremely grateful to USAID through the Food Security 111 Cooperative Agreement,
with funding from Economic Growth, Agriculture and Trade (EGAT), USAID/Mali and
the Africa Bureau, Sahel Regional Bureau, for funding the research that generated the
data. I would also like to thank our local partners in Mali who participated in the
research, including the Institut du Sahel (INSAH), the Planning and Monitoring Unit of
the Ministry of Health (CPS/MS), the National Institute fpr Public Health Research
(INRSP), the Institute of Rural Economy (IER) and the National Direction of Statistics

and Computing (DNSI).

I am indebted to my family, friends, and classmates and faculty at Michigan State
University for their support and encouragement. Special thanks go to Joleen Hadrich for
submitting and printing my thesis after I left the country, and to Crystal Jones for

assistance in preparation of my defense.

TABLE OF CONTENTS

LIST OF TABLES ........................................................................................................... viii
LIST OF FIGURES ........................................................................................................... ix
1. INTRODUCTION ......................................................................................................... l
1.1 Problem Statement .............................................................................................. 1
1.2 Objectives of the Study ...................................................................................... 4

2. SETTING AND DATA ................................................................................................. 6
2.1 Description of Survey Zones .............................................................................. 6
2.2 Sampling Methods and Data ............................................................................... 8

3. CONCEPTUAL MODEL AND HYPOTHESES ........................................................ 12
3.1 Human Capital ................................................................................................. 12
3.2 Status and Social Capital ................................................................................. 14
3.3 Physical Capital and Financial Resources ....................................................... 16
3.4 Labor ................................................................................................................ 17
3.5 Enabling Environment ..................................................................................... 18

4. EMPRICIAL ESTIMATION AND RESULTS ........................................................... 20
4.1 Model 1: Determinants of Mothers’ Income .................................................... 20
4.1.1 Dependent and Explanatory Variables Employed .................................... 20
4.1.2 Estimation Strategy ................................................................................... 24
4.1.3 Empirical Results ...................................................................................... 31

4.2 Model 2: Proﬁles of women earning income .................................................... 42
4.2.1 Additional Explanatory Variables Employed ........................................... 43
4.2.2 Estimation Strategy ................................................................................... 46
4.2.3 Empirical Results ...................................................................................... 47

5. DISCUSSION AND CONCLUSIONS ....................................................................... 55
APPENDIX A ................................................................................................................... 65
APPENDIX B ................................................................................................................... 66
APPENDIX C ................................................................................................................... 67
APPENDIX D .................................................................... ‘ ............................................... 68

vi

BIBLIOGRAPHY ............................................................................................................. 70

vii

LIST OF TABLES

Table 1: Characteristics of Survey Zones ........................................................................... 9
Table 2: Number of wives per husband by region ............................................................ 33
Table 3: Summary statistics for key characteristics by region ......................................... 34
Table 4: Determinants of mothers’ income in Rural Mali: ZSLS regression results ........ 36

Table 5: Number of women participating in income-generating activities by region ...... 44

Table 6: Number of activities per mother by region ......................................................... 46
Table 7: Determinants of mothers’ income in Rural Mali: ZSLS regression results ........ 48
Table 8: Summary statistics for regression variables ....................................................... 65
Table 9: Weights for the calculation of the Agricultural Equipment Indexa’b .................. 66
Table 10: Weights for the calculation of the Vehicle Indexa ............................................ 66
Table 11: F irst-stage regression for husbands’ expenditures ............................................ 67

Table 11: Determinants of mothers’ income in Rural Mali: Comparison of ZSLS models

with and without the vehicle index as an explanatory variable ................................ 68

viii

LIST OF FIGURES

Figure 1: Map of study area ......................................................................... 7
Figure 2: Average annual expenditures by production system ................................ 32
Figure 3: Mothers’ average expenditures by degree of income diversiﬁcation ............. 54

ix

1. INTRODUCTION

1.1 Problem Statement

Despite great strides in reducing global child undemutrition rates over the past few
decades, child malnutrition in Africa remains problematic. Sub-Saharan Africa has
experienced the smallest percentage decrease in malnutrition prevalencel (3.9%) of any
region between 1970 and 1995. The number of malnourished children increased by 70
percent over the same period (Smith and Haddad, 2000). Rural Mali is no exception. A
study of the linkages between child nutrition and agricultural productivity growth in Mali
(LICNAG) revealed that by 18 months, 25% of children are wasted and 40% are stunted

(Tefft, Kelly and Staatz, 2003).

Abundant studies have attempted to identify and measure the determinants of child
malnutrition. It is generally accepted that higher income, and speciﬁcally mother’s
income, leads to increased food expenditures and better child nutrition. One study by the
International Food Policy Research Institute found the effect of women’s income to be
positive on household calorie level, food expenditures, child weight for height and child
survival. In every case, the effect of women’s income was more signiﬁcant than that of
men’s income (Quisumbing et al., 1995). Thomas (1997) reported that a greater share of
household budget under the control of women is more likely to be spent on household
services, health and education. In Mali, the LICNAG study points to a higher (and
signiﬁcant) correlation between mother’s expenditures and height-for-age (HAZ) scores

relative to father’s expenditures or total household expenditures (Tefft, Kelly and Staatz,

 

’ Share of children who are malnourished.

2003). However, despite this empirical evidence, there is a lack of key information on
the sources and determinants of women’s income in Mali. This paper attempts to address

this knowledge gap.

Household Economic Models

The differing effect of mothers’ and fathers’ income on child nutritional status challenges
traditional household economic models that depict households as utility-maximizing units
motivated by stable, uniform preferences. In such models, the household is assumed to
act in the same way as an individual; all resources are pooled and then reallocated to
achieve common objectives (Thomas 1992). Under this assumption, any income that
enters the household - whether it is under the control of the mother, father or head of

household — will have identical impacts on the nutritional status of household children.

However, as the above studies demonstrate, who controls income does matter when
explaining child malnutrition. This suggests that individual members can, and often do,
have separate, different, and sometimes conﬂicting preference functions. Given these
different preferences, researchers have developed new models to describe household
behavior. One such model, the bargaining model, posits that individual members pursue
their own interests, given their relative bargaining position in the household. Bargaining
power is determined, in part, by an individual’s income-earning power and “threat point”
— the point at which the individual believes he or she will be better off outside the

household unit than within it (Zeitlan et al. 1995).

Another model, termed the “implicit contracts” model by Folbre, “sees the family as

governed by culturally determined expectations about the entitlements and obligations of

individuals in different positions with the household unit (Zeitlan et al. 1995). In this
relationship, each spouse contributes to the well-being of the household, although they do
so in a fairly restricted, well-defmed and socially recognized manner. The separate
priorities of husbands and wives are often reﬂected through gender-speciﬁc expenditures
(Fapohunda 1989). These traditional and cultural norms serve both to decrease the time
and energy expended in bargaining over alternative allocations of resources, and to deﬁne
the upper and lower bounds on acceptable behavior within which bargaining can occur

(Zeitlan et al. 1995).

In rural Mali, there is a strong implicit contract between husbands and wives regarding
the responsibilities of each to the household. Lilja et al. (1996) also found evidence of a
bargaining relationship that resulted in higher wages for work on communal ﬁelds paid to
women who were able to exert their bargaining power. Incomes earned by husbands and
wives are not pooled (Wooten 2003; Ward et al. 2004); each, therefore, needs to have his

or her own source of resources to fulﬁll their obligations.

In general, men in Mali are responsible for providing cereals and, in the wealthier
households, meat. Women, in turn, provide the condiments and sauce ingredients for the
meals and may supplement food supplies with purchases of fruits, vegetables and meat or
ﬁsh. In addition to these primary obligations, men pay for luxury items, such as tea,
sugar (for tea) and cigarettes, taxes, children’s school fees, clothes for themselves (and
sometimes for their wives and children) and some ceremonial costs. Women buy clothing
for themselves and their children and pay for expenses connected to weddings and

circumcisions (Wooten 2003; Ward et al. 2004). Both women and men contribute to the

costs of health care, although women tend to pay for traditional health remedies, while
men cover the higher costs of modern health care (Tefft and Kelly 2004). Gifts are also
paid for and given by both men and women, although they are more important for
women, dependent males and poorer households than male heads of household and
members of wealthier families (Wooten 2003). In this sense, the practice of gift-giving is
a form of insurance; by investing in extended family relationships and social networks,
women are improving their bargaining position within their marital relationship and
decreasing the risk of asset loss under traditional practices of inheritance and/or

polygamy (Fapohunda 1989).

While these gender-speciﬁc expenditures are widely accepted, there exist ample
examples where these norms are not respected. Some men provide their wives with
money to purchase sauce ingredients, while women increasingly have to cover household
expenses (e. g., cereals) traditionally charged to men due to economic hardship (Ward et
al. 2004). Ward et al. also observe that an individual’s obligations change with a change
in their relative income. An increase in women’s income may elevate her ﬁnancial

responsibility to the household.

1.2 Objectives of the Study

Accepting the generally separate economies of men and women in Mali and the
important role that mothers’ income appears to play in alleviating child malnutrition, this
study aims to understand the characteristics of women’s income and identify the factors
that determine different income levels attained by mothers. Speciﬁcally looking at the

characteristics and determinants of mothers’ income in Mali offers several advantages.

Describing income patterns and identifying the determinants of income and their relative
importance provides useful information to policy makers designing programs to reduce
poverty and malnutrition and improve gender equity. It is also a useful ﬁrst step in
efforts to model the determinants of malnutrition, as the estimated values of mothers’
income can be used in lieu of the income/expenditure data, which are usually endogenous
in such models. If the relationship between mothers’ income and child nutrition holds up
to multivariate analysis, knowledge of the determinants could serve as leverage points in
the design of interventions to improve child nutrition through increases in mothers’
incomes. The beneﬁts of increasing mothers’ income far exceed its direct effect on child
nutrition. Women with income are likely to be better off, have more bargaining power to
affect household decisions, and are better able to provide health care for children and

assist with clothing and school expenditures.

2. SETTING AND DATA

2.1 Description of Survey Zones

Analysis in this study is designed to add value to an existing study of the linkages
between child nutrition and agricultural productivity grth in Mali (LICNAG)2. The
LICNAG survey data cover three agricultural production systems experiencing different
levels of productivity and economic growth: the irrigated rice growing area in the Segou
region, the cotton-growing area of the Sikasso region, and the millet/sorghum zone in the
Mopti region (see Figure 1 for map of Mali and survey zones). The Segou and Sikasso
regions have been characterized by signiﬁcant increases in agricultural production
through intensive and extensive agriculture practices, respectively, in the late 19905.
Conversely, Mopti has experienced stagnant cereals productivity growth and slow overall

production growth (due largely to area expansion) during the past decade.

While these three production systems deﬁne their respective regions, agricultural
activities are diverse in each. In addition to cotton production, Sikasso is one of Mali’s
most important rain-fed cereals zone. Since the mid-19805, the region has accounted for
approximately 30% of national sorghum and maize production (Tefft and Kelly 2004).
Rice farmers in the Segou region take advantage of irrigation schemes in the Oﬂice du
Niger to grow vegetables (mainly tomatoes and onions) in the dry season, and also
cultivate rainfed cereals in upland ﬁelds surrounding the irrigated areas. Cultivation of
traditional grains in the Mopti region is supplemented by dry-season onion production

and a heavier dependence on livestock.

 

2 Funded through the Food Security 11 Cooperative Agreement between USAID and MSU. Information on
the study can be found at http://www.aec.msu.edu/agecon/st/mali_nut/index.htm

Figure 1: Map of Study Areas

 

     
  

MALI

TOMBOUCTOU

 
   

Legend

'__' _: sruov AREAS

1:] REGIONS

mp Dosim a Productim: Marian Mitchell.
Geography Department. Michigan State University

100 0 100 200 lﬁlometors
E KOLCNDIEBA Soumuzooo

 

 

 

Additionally, these three regions differ with respect to the involvement of state-sponsored
agencies in rural and agricultural development. In the cotton zone, the Compagnie
Malienne pour le Développement des Textiles (CMDT) has been involved in research and
extension, input supply and credit, output marketing, and processing activities for the
cotton sector. It has also, at times, provided support to cereal production, livestock,
forestry and rural infrastructure and services (Kelly and Staatz 2006). In Segou, the
Office du Niger (ON) has provided similar services for the rice sector. Recent
institutional reform of the ON has narrowed the responsibilities of the agency to

infrastructure and water management, transferring most of its other functions to farmers.

The Mopti region has depended largely on non-govemmental organizations (N G05) and

poorly-funded extension services for development assistance.

Each of the three production zones is further divided into administrative zones, or
cercles, two of which were chosen in each zone for their divergent levels of infrastructure
and agricultural opportunities. The key characteristics of the three regions and six zones

are summarized in Table 1.

2.2 Sampling Methods and Data

The data include household-level information for 750 households3 with children under
ﬁve years of age. These households represent a subset of the Demographic and Health
Survey (DHS III) undertaken by the Malian government. The DHS rural households were
selected using a two-stage cluster sample technique. In the ﬁrst stage, 38 “Sections
d’Enumeration” (SE5) used by the Malian Census Bureau were systematically selected
from each of Mali’s administrative regions. In the second stage, households were selected
from a complete list of households for each SE. The resulting DHS sample is
representative nationally and for each of the regions, but is not representative at the level

of cercle (Tefft and Kelly 2004).

For the LICNAG survey, sampling was purposive at the regional and cercle levels to
examine the effects of (1) different agricultural production systems (i.e., cotton, irrigated

rice and millet/sorghum) and (2) different levels of infrastructure on child health and

 

3 For the purpose of this study, a household is deﬁned as an agricultural production unit or “work-eat”
group. It consists of members engaged in common production and consumption, and may include more
than one “nuclear” family (husband, spouse(s) and unmarried children).

Table 1: Characteristics of Survey Zones

 

 

 

 

 

 

 

 

Production Zone-level
Systems Regional Characteristics Zones Characteristics
High rainfall (1000-1500 mm .
Cotton System annually); in addition to cotton, Kolendieba New cotton zone; farmers tend to
(S ikasso) region is largest producer of rain- be'less experienced-and have less
fed cereal crops (maize, sorghum); animal traction equipment than
high livestock ownership farmers in Koutiala
Intervention OfCMDT for cotton Koutiala Old cotton zone; farmers well
and cereal inputs, mar kettng, established and infrastructure
"3d“ and rural infrastructure better developed than in the newer
, . zone; little diversiﬁcation outside
Large extended families under of cotton and cereal crops
strong patriarch; dominated by
Bambara and Mianka ethnic
grows.
Rainfall equals 500-750 mm
Irrigated Rice annually; ﬂoodplains along the Niono Located in part of irrigation
System/ Niger River delta suitable for scheme with-full. water control and
(Segou) irrigated rice and recession upgraded urrgatron Infrastructure;
agriculture; dwersrﬁcatton Into primary occupation of households
horticultural crops; livestock and is rice production; dry-season
ﬁshing also important horticultural production common;
better access to transport than
Institutional reform of Ofﬁce du Macina
Niger in recent years, improved
irrigation infrastructure and Macina Water control less developed than
technological innovations in Niono; for some households,
. production of rainfed millet and
Smaller, nuclear family structure; sorghum crops is more important
Silver“? ethnic population, , than irrigated rice; sample includes
including Bambara, Bozo, Mtanka, important share of households
Sarakole and Fulani whose primary occupation is
ﬁshing.
Rainfall equals 500-750 mm
Millet/Sorghum annually; millet/sorghum produced Bandiagara Some households have access to
System in rainy season, onions are water retention structures that
(Mopti) produced in dry season; livestock permit onion cultivation during the
important part of production dry season; road and market access
system easier than in Koro; migration to
provide harvest labor in Office du
Rural and agricultural Niger irrigated perimeters
development mainly through common.
NGOs and extension service
Koro Few dry season agricultural

 

Extended family structure;
dominated by Dogon ethnic group

 

 

opportunities, as wells tend to dry
up; difﬁcult access to/from major
urban centers.

 

Source: Kelly et al., 2004

 

nutritional status. Within each of the cercles, sampling was random at the household

level.’

The data collected in the LICNAG survey are comprised of monthly anthropometric
measurements for the children, and information on household demographics, assets,
individuals’ expenditures and agricultural production and non-farm activities. Data were
collected for the period between May 2001 and April 2002 using bi-monthly interviews
with a variety of household members. In addition to the bi-weekly interviews, a
Knowledge, Attitudes and Practices (KAP) survey was administered to the parents of 457
children <25 years of age. Focus discussion groups with men and women were held, as

well as health center interviews, mayor interviews and surveys of village infrastructure.

Of the 1615 mothers of children under the age of ﬁve years who were weighed for the
study, 984 were retained for this analysis. Five hundred forty-ﬁve were dropped because
either the woman or her husband did not provide information on their respective
expenditures. Thirty-eight women were dropped because they did not report involvement
in any income-earning activities, despite having positive expenditures. Another 46 were

dropped due to insufﬁcient data on the woman and/or her household.

Forty-one percent of the cases dropped came from the Mopti region, compared to 38%
from Sikasso and 21% from Segou. There are some signiﬁcant differences between the
women retained in the study and those who were dropped. On average, the women

retained in the sample were 3.2 years older than their excluded counterparts. Forty-four

percent of the women retained in the sample were married to the head of the household,

 

’ See Tefﬁ and Kelly (2004) for a detailed description of sampling methods.

10

compared to 33% in the overall sample. The woman’s average level of schooling and her
position within the family (i.e., wife orders) were not signiﬁcantly different. The average
size of the household was signiﬁcantly larger in the dropped cases, which is likely due to
the disproportionate number of dropped cases from the Sikasso and Mopti regions
relative to those from Segou. It is also extremely difﬁcult to track individuals’ income
and expenditures accurately for households with more than 50 people. Total household
income was not signiﬁcantly different between the two populations for the Sikasso and
Segou regions. Households retained for the analysis in Mopti had a signiﬁcantly higher

average household income than those that were dropped.

 

5 The majority of ethnic groups (and total population) in Mali are Muslim and practice polygamy. Each
man can have a maximum of four wives. The term “wife order” in this paper refers to whether a woman is
the ﬁrst, second, third or fourth wife. A negative relationship between wife order and income implies that

the ﬁrst wife has a higher income than the second wife, who in turn has a higher income than the third wife,
etc.

11

3. CONCEPTUAL MODEL AND HYPOTHESES

The goal of this research is to describe key characteristics and identify the determinants
of mothers’ income in three agricultural production zones in rural Mali. The study

focuses on testing hypotheses about factors thought to be correlated with incomes for a
sample of mothers with children under the age of ﬁve years who participated in a study

6 These factors include various individual—,

on the determinants of child malnutrition.
household- and community-level assets and labor availability along with community
characteristics that serve as an “enabling environment” for women to earn income. The
conceptual model can be presented as follows:

Y=f(a, b,c,d,e)
Where Y is the income level of the mother, a is a vector of human capital, b is a vector of

social capital, c is a vector of physical capital and cash ﬂow, (I is a vector of labor input

and e is a vector of community characteristics.

3.1 Human Capital

Components of human capital that are hypothesized to contribute to increased women’s
income include her own level of schooling and experience gained through informal
education, the level of schooling of her husband and the woman’s nutritional and health
status. One would expect the income-earning capacity of a woman to increase with her
level of schooling. Hill and King (1994) attribute both market and non-market beneﬁts to
women’s education, including higher family income, increased own-earnings, increased

labor productivity, improved health and nutrition and greater fertility control. However,

 

6 See Tefﬁ and Kelly (2004) for the preliminary results of this study.

12

other empirical studies have produced indeterminate results on this issue. Studies of
female non-farm income in Ghana and Uganda showed negative and positive inﬂuences

of education, respectively (Canagarajah et al., 2001).

Similarly, it is hypothesized that being married to an educated husband will have a
positive effect on a woman’s income. This may occur through several paths. First, in its
State of World Population, the UNF PA (2002) reports that households with educated
members have more secure employment, higher incomes and better access to economic
assets and credit. Access to these assets may improve a woman’s capacity to earn income
even if she, herself, has not directly beneﬁted from educational opportunities. Second, a
study investigating human capital and labor allocation in rural Pakistan (F afchamps and
Quisumbing, 1998) found that households with better-educated males diverted labor
resources away from low-productivity activities toward higher-return nonfarm work.7 It
is possible, then, that an educated husband is more likely to encourage his wives to
pursue more lucrative activities outside the home. For a review of the returns to

investment in education, see Psacharopoulos and Patrinos (2004).

It is expected that the health and nutritional status of the woman will have a positive
effect on her income, in that healthier women have more time available for agricultural
production and nonfarm activities and work more productively. The positive relationships
between nutritional status and labor productivity, health and wages, and health and
incomes are supported by Haddad and Bouis (1991), Schultz (2005) and Bloom and

Canning (2000), respectively. However, in a review of studies investigating

 

7 In the case of Pakistan, where purdah prevents many women from engaging in activities outside the
home, women did not experience the same returns to nonfarm income from education as their male
counterparts.

13

nutrition/productivity links, the authors ﬁnd the linkages to be less consistent for women

than for men (Kennedy and Garcia, 1994).

3.2 Status and Social Capital

Robison and Stiles deﬁne social capital as “a person's or group's sympathy that provides
another person or group potential beneﬁts, advantages, support and access beyond that
which might be expected between strangers in an exchange relationship” (2000, 3).
Several factors contribute to the sympathy a woman receives; each has implications for
her ability to earn income. At the individual level, these include her age, whether she is in '
a polygamous marriage, her status within the household (ﬁrst, second, third or fourth
wife) and ethnicity. It is hypothesized that a woman’s capacity to earn income is
positively correlated with her age. As the woman becomes older, she can pass household
chores to younger girls, allowing more time for other work. A woman’s age may also be
an indicator of experience and knowledge gained from sources other than formal
education. Whether or not a woman is in a polygamous marriage can have contrasting
effects. Having co-wives may reduce the amount of housework required by each woman;
however, limited resources (e. g., land) may have to be shared amongst the wives. For
wives in a polygamous relationship, we expect to see a negative relationship between the
wife order and income earned. As a husband acquires more wives, the ﬁrst wife is
relieved from many household duties. Ethnicity can be signiﬁcant if cultural norms
dictate whether or not women are educated and can work outside the home, and if these
norms differ signiﬁcantly amongst ethnic group. Poulton and ag Youssouf (1998) argue
that the latter does not hold in Mali. They claim that — due to migration and inter-

marriages — ethnicity has become a very ﬂuid concept, giving it less inﬂuence in Mali

14

than it may have in other places.8 Still, certain ethnic groups do tend to engage in certain
economic activities (e. g., the B020 are associated with ﬁshing, while Fulani depend

heavily on livestock and milk).

At the household-level, aspects of a woman’s social capital that are hypothesized to
inﬂuence her income include the position of her husband within the household and the
demographic make-up of the household’s members. The husband’s position in the
household is critical. In the Malian context, a “household” may be comprised of several
“nuclear families” within the extended family. The household is deﬁned by those
members who are engaged in common production and consumption. As such, one
patriarch controls most of the income, assets and expenditures, giving other adult males
very little decision-making power and control over ﬁnancial resources. A woman who is
married to the head of household may therefore have greater access to household
resources, such as using the household’s animal-drawn equipment to plow her personal

ﬁelds or using household cash to invest in nonfarm activities.

At the community level, it is hypothesized that the number of village-level associations
will have a positive impact on incomes. Village associations are an important source of
credit for women who often do not have the collateral to qualify for formal bank loans.
Associations organized around income-generating activities also provide labor- and

resource-sharing opportunities for women.

 

8 Conversely, in their intra-household analysis on access to resources in Southern Mali, DeGroote and
Coulibaly (1998) found ethnicity to be an inﬂuential variable.

15

3.3 Physical Capital and Financial Resources

A women’s ability to generate income is hindered by the competing demands on her
time, including agricultural production, domestic production (including food preparation
and the collection of fuel and water) and reproduction. Assets and technologies which
increase women’s productivity in both agriculture and home production are expected to
have a positive effect on women’s income by allowing for more time to engage in other
activities, without sacriﬁcing additional time, their children’s welfare or their own health
and nutritional status (Quisumbing et al. 1995). Other household assets could have a
more direct impact by providing income-generating opportunities to women (e. g., land

and livestock to milk).

The impact of agricultural technology on women’s incomes and welfare is more
complicated. As opposed to assets which reduce the time required to perform certain
activities (such as energy-efﬁcient stoves), agricultural technology packages often require
increased labor input. This higher demand for labor is often met by women, pulling them
away from their own-income activities to work on communal ﬁelds (Kumar 1987).
Conversely, Lilja et al. (1996) found the introduction of new agricultural technologies in
southern Mali to have a positive effect on wages paid to women for work on communal
lands. They also found that technology, proxied by land under cotton cultivation, reduces
the female-male wage differential by generating new income streams for women. These
effects are magniﬁed when woman have the ability to bargain and/or refuse to work on
communal ﬁelds. This last point is critical to the discussion on technology. It is often
not the technology itself that affects women’s income, but rather, the rules governing

who gets the income streams generated from the new technologies. If a woman controls

16

the income streams herself, presumably she would not adopt the technology if it made her
worse off. Where women do not have a voice in how their labor is utilized, they may see

their welfare decrease as a result of the technology.

The income level of a woman’s husband can have contrasting effects on her own income.
Under certain production systems and household structures, non- or low—remunerated
work by women on their husbands’ ﬁelds may simultaneously increase the men’s
incomes while suppressing those of the women. If married couples pool their income or
allocate resources to their best use, then a husband’s income could positively inﬂuence
his wife’s earning capacity by providing the capital necessary to start certain income-
generating activities. Even where funds are not given directly, a women may beneﬁt from
her husband’s income if be, for example, provides better nutrition to the family, hires
labor that frees up the wife’s time or provides access to income-generating assets. As
with physical assets and technoldgy, informal institutions and the woman’s social capital
will determine if, and to what degree, she can access her husband’s earnings to increase

her own.

3.4 Labor

A woman’s income level will likely be inﬂuenced by the intensity of time and effort she
dedicates to income-generating activities. The time a woman has to participate in extra-
household activities, in turn, will depend on both the size of the household and the
composition of its members. It is plausible that households with more members can
diversify their activities, enabling them to engage in more non-farm activities

(Canagarajah et al. 2001), depending on the amount of land they cultivate. Speciﬁcally, it

17

is hypothesized that having more women who share common household responsibilities
(such as cooking and collection of fuelwood and water) will reduce the time that each
dedicates to these chores, allowing more time to explore other work. If few women and
girls are responsible for the childcare and cooking of a large family, time to participate in

other work will be limited or nonexistent.

3.5 Enabling Environment

An enabling environment is a set of interrelated and interdependent conditions such as
policies, laws, institutional mechanisms (both formal and informal), infrastructure and
resources which facilitate the ability of women to earn income.9 These conditions are
deﬁned at the levels of community and/or region; however, a women’s ability to take

advantage of the environment may differ at the individual- and household-levels.

At the community level, it is hypothesized that levels of infrastructure, such as water
resources and schools, should have positive effects on women’s income. Nearby wells or
public water faucets will reduce the time women spend gathering water, allowing them to
devote more time to activities outside the home. The presence of schools could be
indicative of private and public investment in the community and reduce the hours per
day that mothers need to care for school-aged children. Schools also provide an
opportunity for women to generate income by selling snacks and drinks to students.
Proximity to markets could reﬂect an overall healthier economy and opportunities for

commerce. Evidence from MSU work in Africa shows that proximity to markets leads to

 

9 This deﬁnition is adapted from “Enhancing Participation of Women in Development through an Enabling
Environment for Achieving Gender Equality and the Advancement of Women” (DESA, 2005).

I8

higher household income, which could be expected to carry over to higher women’s

incomes (Dioné 1989).

Location may play an important role in determining mothers’ income. However,
interpreting location variables is difﬁcult because they can capture a variety of climatic,
economic and social factors. For example, one might expect higher incomes in the rice-
producing region of the Office du Niger relative to the cotton and course grains regions,
due to the existence of irrigation mechanisms for gardens and more non-farm activities.
However, a dummy variable for this zone might also be picking up income differences
due to the formal and informal institutions that exist in the regions, such as the different
historical approaches to agricultural and rural development taken by the Office du Niger
and CMDT, and household structure. Households in the rice-growing region tend to be
smaller and more nuclear, while the cotton and millet/sorghum zones are more likely to
have many nuclear families living in the same household. In the extended families,
productive assets (equipment) and income from household ﬁelds (cotton, cereals) are
controlled by the patriarch of the extended family rather than by the heads of the
individual nuclear families. The mean numbers of members per household are 24.27 for

Sikasso, 18.97 for Segou and 22.69 for Mopti.

l9

4. EMPIRICAL ESTIMATION AND RESULTS

This study attempts to measure the effect and relative inﬂuence of the determinants of
mother’s income in rural Mali and provide information on the activities in which women
participate. This will be done with two different empirical models. The ﬁrst model
focuses on the determinants of mothers’ income, based on the conceptual model
discussed above. The second model will build on the ﬁrst model by adding dummy
variables to represent participation in the different activities by which women earn
income. Since the causality of participation in these activities and income levels is not
clear, these variables were excluded from the ﬁrst model. However, they do provide a
better proﬁle of what women who earn different levels of income look like, as
participation in different activities are more speciﬁc indicators of the opportunities and

constraints faced by women than the determinants alone.

4.1 Model 1: Determinants of Mothers’ Income

4.1.1 Dependent and Explanatory Variables Employed

The explanatory variables in the model of the determinants of the mothers’ incomes
represent the different kinds of capital and “enabling environment” discussed in Chapter
3. The choice of explanatory variables, described below, was guided by the hypothesized
factors thought to be positively or negatively associated with mothers’ income,

information from the literature and the available data.

The natural logs of mothers’ expenditure data were used to represent income, the

dependent variable. The natural log was chosen to minimize the effect of extreme values.

20

Only women who had positive expenditures were considered for this study, since it is
unclear whether the 23 percent of women (373 cases) with no reported expenditures are a
result of missing data or truly have no income. The latter is unlikely in the Malian
context.10 Expenditures were utilized in lieu of income because of the difﬁculty in
collecting information on earned income and for the ability of expenditures to reﬂect
long-run income potential (Haddad et al. 2002). Careﬁil measures were taken during the
interviews to assure that only expenditures made with money earned by the interviewee
were recorded. Speciﬁcally, there was a question on the survey regarding the “source of
money” for the expenditure that indicated whether the money spent was earned by the
respondent, someone else or whether it represented general household income managed
by the head of the household. The responses to this question were used to assign the
expenditures to the person who actually paid for them. For all respondents (male and
female), expenditures on cereals were excluded from the analysis, since males —
particularly the heads of household —- may be responsible for provisioning the children of
several wives and other dependents. Inclusion of cereal expenditures would increase the

incomes of males relative to women.

Human capital variables were represented by the number of years that the mother and her
husband attended formal school. There is nothing in the data set to represent the women’s
health or nutritional status, so these are necessarily excluded from the model.ll Several

variables represent the mothers’ status and social capital. Mother’s age was represented

 

'0 If we were conﬁdent that all zero expenditures were true zeros, a Tobit regression would be preferable to
least squares. However, it is reasonable to believe that the number of women with no income is much lower
than 23 percent, and that “zeros” in many cases represented missing responses, not zero expenditures. In
this case, including the cases with “zero expenditures” in the analysis would introduce a bigger bias due to
measurement error than taking them out.

” Mother’s height was recorded for women participating in the KAP survey. It was not found to be
signiﬁcant when the model was run on the smaller sub-sample of women.

21

by a continuous variable. Her position within the polygamous marriage (I , 2, 3 or 4) was
found to be insigniﬁcant when represented by both continuous and dummy variables, and
was therefore dropped. A dummy variable representing whether the woman was involved
in a monogamous or polygamous marriage was also dropped because its coefﬁcient was
not signiﬁcantly different than zero. Ethnicity was dropped from the model because it
was found to be insigniﬁcant for all ethnic groups and including it added a signiﬁcant
number of parameters to the model. In addition, members of the same ethnic group tend
to live in the same area, and thus ethnic group dummy variables are highly correlated

with the location variables.

A dummy variable indicated whether or not the husband was the head of the household.
If he was the head, then his expenditures included not only his personal expenditures, but
also major expenditures for the entire household which fell under his control, with the
exception of cereals and agriculture equipment. All else equal, we would expect the
expenditures for a husband who is “head” to be much higher than for a husband who is
“dependent” within the extended family. We also assume that a woman married to the
head (versus a dependent male) will have better access to common household resources

for pursuing her own income generating activities and fewer household tasks.

A continuous variable was used to represent the number of village-level associations.
This variable considers all the associations reported to exist in the village. There is no
indication of whether the women included in the survey were active members or

beneﬁciaries of the associations.

22

Several variables were used to represent household assets. Time-saving assets that were
considered include energy efﬁcient stoves (which require less wood, and hence, less time
spent gathering wood) and whether or not there was a well in the compound. Both of
these assets were found to be insigniﬁcant and ultimately dropped from the model.
Measures of assets that represent household wealth and potential access to resources by
the women include the value of livestock holdings per adult equivalent, total household
and individual land holdings, an index measuring agricultural equipment, and an index of
vehicles per adult equivalent. The livestock variable includes non-traction cattle, goats,
sheep and pigs. The average value of each animal per zone at the time of the survey was
used to calculate total value. Total household land holdings were measured in hectares
and divided by the number of adult equivalents. The estimated area (in ha) of women’s
ﬁelds was reported by the heads of households, and not explicitly measured. The
agricultural equipment index is calculated based on the “completeness” of the full animal
traction chain and its perceived contribution to productivity. '2 The vehicle variable is an
index based on the relative cost of motorcycles, mopeds, bicycles and animal-drawn
carts, divided by the number of adult equivalents in the household. There were no cars or
trucks owned by households in the sample. These assets were found to be signiﬁcant in
models determining per capita household income from the same data set. The natural
logs of husbands’ expenditures were used to represent husbands’ income, following the
same assumptions cited above for mothers’ expenditures. These values include cash

expenditures only and do not include the value of production for home consumption or

 

'2 A full animal traction chain includes two oxen and one plow. This combination of assets is given a value
of 1, and is weighted higher than having one ox and two plows or three oxen and no plow. The total value
of the index is a summation of the weights of full and partial traction chains. The full list of weights can be
found in the appendix.

23

expenditures on major assets. When cereal production and expenditures are factored into

total expenditures, total “incomes” are much higher across all zones.

A variable to represent the intensity of effort and time with which the women labored in
income-earning activities could not be found in the data. As a proxy for the potential
time available to her for these activities, a variable was created to indicate the ratio of
females aged 10 and 45 to other members of the household. Women within this age range
often share the responsibility of household chores, including cooking, childcare and
laundry. The greater the number of people for which each woman performs these duties,
the less time she would have available for other activities. The numbers used in this
calculation were adjusted to account for the amount of time (in months) that each

member was present in the household.

Dummy variables were used to indicate if the mother lived in the Kolendieba, Macina,
Niono, Bandiagara or Koro zones. Koutiala, which experiences the lowest mean income
levels for women, was omitted; the impact of this zone is reﬂected in the constant term.
The distance between the village and the nearest market is measured in kilometers. A

dummy variable represents whether there is a school in the village.

4. 1. 2 Estimation Strategy
The dependent variable, mother’s expenditures (ME), was hypothesized to be determined
by vectors of explanatory variables, denoted by V, W, X, Y and Z and indexed by j, k, l, m

and n, respectively. The basic model took the following form:

J K L M N
Mfr =0 + Zﬂthj,i + 2.3qu +Zﬂ3tX1,i + Z ﬂ4mYm,t + Zﬂsﬂmt +5,- (1)
j=l k=1 1:] m=l n=l

24

where i denotes the observations for each mother included inthe analysis
a is the constant term
V is a vector of human capital variables
W is a vector of social capital variables
X is a vector of physical capital variables
Y is a vector of labor variables
Z is a vector of community characteristics (enabling factors)
[31 is a J x 1 vector of parameters associated with the human capital vector, V
[32 is a K X 1 vector of parameters associated with the social capital vector, W
[33 is an L X 1 vector of parameters associated with the physical capital vector, X
B4 is an M x 1 vector of parameters associated with the labor vector, Y
[35 is an N X 1 vector of parameters associated with the enabling factor vector, Z,
and e, is the error term.

To determine whether equation (1) can be estimated using ordinary least squares (OLS),
the endogeneity of certain explanatory variables was tested. Endogenous variables are
inﬂuenced by some of the same forces that inﬂuence the dependent variable being
studied, resulting in correlation between the concerned variable and the error term. When
explanatory variables are endogenous, OLS gives biased and inconsistent estimates of
their causal effect on the dependent variable (Bound et al. 1995). By contrast,
instrumental-variables estimation, such as two-stage least. squares (ZSLS), will yield

parameter estimates that are both consistent and efﬁcient.

Endogeneity of variables was determined using the Hausman-Wu instrumental variables
(IV) test (following Smith and Haddad 2000). The test is performed in two steps. First,
the potentially endogenous variable(s) is (are) regressed on the remaining variables
(assumed to be exogenous) and the predictors, or “instruments,” of the endogenous
variable(s). These instruments should be correlated with the endogenous explanatory

variable but have no direct association with the dependent variable (Bound et al. 1995).

25

To test the endogeneity of one of the physical capital variables, X1, in equation (1), given

instruments H l .,. . ., H1 3 , the following estimating equation is used:

.1 K L M N G
XL: =w+ ZalejJ +Z5ZkaJ + Z531X1J+ Z54mYm,t + 25$:sz + Z’lgHg +Wi-
j=l k=l 1:2 m=l n=l g=l

(7-)

In the second step of the test, the dependent variable, ME, is regressed on all explanatory

variables plus the predicted residuals from equation (2), #7::

J K L M
Mr =7+Z‘PIjVjJ+Z¢2kaJ+Z¢31XLi+ Z ¢4mYm,r+Z¢5nZn,1+wI/?r+ﬂz- (3)
j=1 k=1 [=1 "7:1

The null hypothesis is that the explanatory variable is not endogenous. If the coefﬁcient
on the predicted residuals, w , is statistically signiﬁcant, the null hypothesis is rejected

and OLS will not yield unbiased and consistent estimates.

However, before the Hausman-Wu test is performed, the credibility of the instruments
must be tested (i.e., whether the instruments are proper substitutes for the potentially
endogenous variables). This is accomplished using two tests: a “relevance test” and an
“overidentiﬁcation test.”[3 The relevance test is an F -test on the joint signiﬁcance of the
instruments. If the F-statistic is signiﬁcant, the instruments are considered to be relevant
to the variable they attempt to explain. The overidentiﬁcation test examines whether the
instruments directly affect the dependent variable other than through the potentially

endogenous variable they replace. It tests whether there is a correlation between the

 

'3 See Smith and Haddad (2000) for a thorough explanation of these tests.

26

instruments and the error term from the IV estimation. The test is performed in several
steps. In the ﬁrst step, the predicted residuals from the ZSLS regression of mothers’
expenditures are calculated, as in equation (2). These predicted residuals are then
regressed on the exogenous variables and the instruments:

M
Z

N G
¢I4mYm,i+ Z ¢SnZn,i+ Z ¢5gHg+4i

J K
wr=t+ Z ¢IjVj,i+ Z
: = m=1 ”=1 g=l

L
¢2ka,1+ Z ¢31XI,1+
1 k 1:2

1

Next, the statistic N X R2, where N is equal to the number of observations, is calculated. N
x R2 has a chi-squared ([7) distribution, with degrees of freedom equal to the number of
instruments minus the number of potentially endogenous variables. Under the null
hypothesis, the instruments are uncorrelated with the error term and the model is

correctly speciﬁed. If the null is rejected, the model is overidentiﬁed by the instruments.

If a suitable instrument fails the Hausman-Wu test (i.e., the variable is endogenous to
mothers’ expenditures), ZSLS will be used to estimate the model in two stages. In the
ﬁrst stage, each of the explanatory endogenous variables is regressed on the entire set of
exogenous variables (Pindyck and Rubinfeld 1998), as in equation (2). In the second
stage, the dependent variable is regressed on its remaining exogenous variables and the

predicted values of X 1:

J K L M N
MEi =a+ Z91jVj,i +292kaJ + Z931X/,r + Z 94mYm,i + 29%sz +9X1,i +01. (5)
j=I k=l [=2 m=l n=1

The variables tested for potential endogeneity to mothers’ expenditures include the

husbands’ expenditures, the equipment and vehicle indices and the value of household-

27

level livestock holdings. Husbands’ expenditures are likely to be endogenous if some of
the same household assets or production systems, for example, determine the
expenditures of both the husbands and wives simultaneously. Since the household-level
assets are controlled by men, there is less theoretical evidence to indicate endogeneity of

these variables.

After selection of the estimation model, the model was tested for heteroskedasticity.
Heteroskedasticity is found, most often with cross-sectional data, when the variance of
the error terms is not equal across realizations. Heteroskedasticity is caused if (I) the
variance of the dependent variable increases (or decreases) with an increase in the level
of the dependent variable, (2) the variance of the dependent variable increases (or
decreases) with a change in the levels of the explanatory variables, (3) there are outliers
in the data, or (4) there is speciﬁcation bias (e. g., missing variables or incorrect functional
form). Pindyck and Rubinfeld (1998) expect to ﬁnd heteroskedasticity in models that use
expenditures as the dependent variable. They propose that low-income individuals are
likely to spend at a rather steady rate, while the spending patterns of high-income
variables are relatively more volatile. As a result, the variances associated with high-

income families are likely to be greater than their low-income counterparts.

For heteroskedastic models estimated by OLS, parameter estimates remain unbiased and
consistent, although not efﬁcient. However, the variance estimates of OLS estimators are
both biased and inconsistent, yielding biased t- and F-statistics, which cannot be used for
hypothesis testing. To test for heteroskedasticity using the Breusch-Pagan test, we ﬁrst

estimate the model with least squares, assuming no heteroskedasticity. For simplicity, the

28

model in equation (5) will be presented with all the explanatory variables under one
vector, X:
K

ME; =a+Z§ka,,-+ Vi (6)
k=l

We then compute the squared residuals, t7} , and regress these on the explanatory

variables thought to be responsible for causing the heteroskedasticity:

K

.2

V,- =7+2kak,i+ vi (7)
k=1

The Breusch-Pagan test statistic, BP , is constructed from the results of this auxiliary

model:

BE = Nx R2 ~ my}; (8)

where the degrees of freedom, k, is equal to the number of restrictions in the auxiliary

model.

The null hypothesis of constant variance in the error terms, or homoskedasticity, is

rejected if the p-value associated with B13 is less than 0.05 (5% signiﬁcance level). If
heteroskedasticity is accepted, the standard errors will be corrected using White’s
heteroskedastic-robust covariance matrix estimator. White’s method obtains the

corrected standard errors of each coefﬁcient in equation (6) as follows:

29

. 2:23»?

var (a,)=_.__. (9)

(mi

where I3,- are the residuals obtained from the original regression (6) and iii. are the

residuals obtained from the auxiliary regression (7) (Gujarati 2003).

Results of endogeneity and heteroskedastic tests

The Hausman-Wu test revealed that husbands’ expenditures are endogenous to those of
their wives at a signiﬁcance level of 1%. Since the credibility tests performed on the
instruments show them to be both relevant (F = 27.41) and not overidentiﬁed ([7 = 3.45),
two-stage least squares (ZSLS), as discussed above and presented in equation (5), was

chosen to estimate the model.14 '5

The age and education level of the head of the household were used as instruments to test
for the endogeneity of the agricultural equipment index, the value of the household-level
livestock holdings and the vehicle index. The instruments passed the relevance and
overidentiﬁcation tests for all three of the potentially exogenous variables. However, the
explanatory power of the instruments was very weak (R2 S 0.01). The data do not contain
anything else that is an appropriate instrument. The Hausman-Wu statistics for the
equipment index and the value of household-level livestock holdings were not signiﬁcant
at the 10% level. These variables are therefore assumed to be exogenous (the latter was

found to be insigniﬁcant in determining mothers’ expenditures and ultimately dropped

 

" The instruments used to predict husbands’ expenditures can be found in the appendix.
'5 Variables that were exogenous to mothers’ expenditures but endogenous to husbands’ expenditures were
excluded from the ﬁrst-stage of the ZSLS estimation.

30

from the model). The vehicle index was found to be endogenous, but was sensitive to the
choice of instruments. Since it is unlikely that a woman’s income inﬂuences the purchase
of these assets, and the relationship between the instruments and the vehicle index is

weak (R-squared < 0.01), the non-instrumented estimates are presented in the analysis. ’6

An alternate ZSLS estimation with the vehicle index excluded can be found in the

appendix.

The Breusch-Pagan test statistic was signiﬁcant at the 1% level. We therefore reject the
null hypothesis that standard errors are constant across the sample. Heteroskedasticity

was corrected using White’s heteroskedastic-robust covariance matrix estimator.

4.1.3 Empirical Results

Descriptive Analysis of Mothers ’ and their Husbands ' Expenditures

Figure 2 shows average expenditures for mothers and their husbands in the three
agricultural production regions included in the study. Mothers in the irrigated rice-
growing region of the Oﬂice du Niger have the highest average expenditures, followed by
mothers in the millet zone and, ﬁnally, the cotton-growing zone. Within the rice-growing
region, Niono, with better infrastructure and off-season production, has higher average
expenditures than Macina. Mothers in the Bandiagara zone of the millet/sorghum region
have higher average expenditures than those in Koro, which is to be expected due to
Bandiagara’s better access to markets and water retention structures for gardening. In the
cotton region, Kolendieba exhibits higher average expenditures, despite having less

experienced producers and less animal traction equipment than Koutiala.

 

'6 According to Bound et al. (1995), the use of instruments that jointly explain little of the variation in the
endogenous variable can do more harm than good. If this weak relationship does exist, then even a small
correlation between the potential instruments and the error can seriously bias estimates.

31

Figure 2: Mothers' and their husbands' average annual expenditure by
production system

 

100,000-i

80,000—
Q
in
U
t:-
-§ 60.000-
5 I Mothers
E I Husbands
a 40,000—
a.
31
I-‘r-l
20.000—

 

 

 

Cotton Irrigated Rice Coarse Grains

Production Systems

Source: Calculated from the data

a Only positive expenditures are calculated in averages

b l US$=560 FCFA
Amongst husbands, the cotton-growing region exhibits the highest average expenditures,
followed by the irrigated rice zone and those producing mainly coarse grains (Figure 2).
As expected, fathers’ average expenditures exceed those of the mothers in every region.
Koutiala, which exhibits the lowest average expenditures'of all the zones for the mothers,
has the highest average expenditures for the fathers, reﬂecting perhaps a very hierarchical
household structure where men control most income and expenditure decisions and
women have few opportunities to earn their own money. This large difference in
expenditures is also driven, in part, by the large number of husbands in the Sikasso region

who have multiple wives. Husbands with several wives are expected to cover the

32

expenditures for which he is responsible (e.g., school fees, modern health care, taxes,
clothes for holidays) for all his wives and their children. Conversely, a mother is only
responsible for expenditures relating to herself and her own children. Table 2 gives the

percentage of husbands in each region who are in polygamous relationships.

Analysis of the expenditures did reveal differences in the items purchased by men and
women. However, the frequency and relative value of expenditures for different items do
not appear to differ substantially between regions for either sex. This implies that the
levels of expenditures for men and women in the Sikasso region are not simply a result of

greater ﬁnancial responsibilities for men in Sikasso relative to men in the other regions.

An analysis of the summary statistics reveals other differences between regions (see
Table 3). Both women and men in the Segou region have signiﬁcantly more education
than their counterparts in Sikasso and Mopti, although the average level of formal
education is less than one year for both genders across all regions. '7 Signiﬁcantly fewer

women are married to the head of the household in Mopti than in Sikasso or Segou (39%

Table 2: Number of wives per husband by region

 

 

 

Number of wives per husband
% %

Region 1 2 3 4 monog. polyL Total
Sikasso 152 130 15 8 49.6% 50.2% 305
Segou 202 73 1o 2 70.4% 29.6% 287
Mopti 126 70 16 0 59.6% 40.4% 126

Total 460 273 41 10 59.7% 34% 804

 

 

 

 

 

 

 

 

 

Source: Calculated from data.

 

'7 A t-test for equality of means is used to test for signiﬁcance.

33

 

Table 3: Summary statistics for key characteristics by region

 

 

 

All Sikasso Segou Mopti
Mean Mean Mean Mean
Variables (Std. dev) (Std. dev) (Std. dev) (Std. dev)
Mother’s expenditures (in F CFA) 40,771 25,265 67,528 29,924
(47,532) (23,324) (59,747) (42,257)
Human Capital
Years of formal education of mother 0.23 0.13 0.44 0.10
(1.07) (0.67) (1.58) (0.61)
Years of formal education of husband 0.66 0.61 0.89 0.44
(1.86) (1.84) (2.10) (1.52)
Social Capital
Age 29.6 28.7 29.8 30.9
(7.7) (7.8) (7.3) (7.8)
Husband is head of household 0.44 0.46 0.47 0.37
(0.50) (0.50) (0.50) (0.48)
Number of village-level associations 3.0 2.3 2.9 4.0
(2.1) (1.4) (1.0) (3.1)
Physical Capital and Financial Resources
Agricultural Equipment index 0.18 0.22 0.19 0.1 l
(0.13) (0.14) (0.12) (0.09)
Vehicle Index 0.22 0.26 0.22 0.17
(0.21) (.18) (0.26) (0.17)
Husband’s expenditures (in FCFA) 79,094 97,787 88,299 39,276
(103,153) (121,041) (105,682) (42,315)
Labor
Ratio of females aged 10-45 to other 0.39 0.41 .38 0.38
household members (.152) (0.15) (0.17) (0.13)
Enabling Environment
Distance to market in kilometers 8.74 7.23 8.42 11.40
(8.27) (3.97) (12.07) (6.22)
School = 1 0.66 0.82 0.74 0.31
(0.47) (0.38) (0.44) (0.47)

 

Source :Calculated from survey data

compared to 46% and 47%, respectively). 18 The average number of village-level

associations is highest in Mopti, while the average agricultural and vehicle indices are

highest in Sikasso.

 

'8 The numbers of women married to the head of the household in the full sample (n= 1615) for Sikasso,
Segou and Mopti are 35%, 40% and 24%, respectively.

34

Multivariate Analysis

Figure 2 shows that women in the irrigated rice-growing region of the Ofﬁce du Niger
have the highest average expenditures, followed by mothers in the millet/sorghum system
and cotton region, respectively. However, these averages do not reveal information
concerning the variables believed to inﬂuence incomes. A ZSLS regression (with fathers’
expenditures instrumented) was thus carried out to isolate the independent effect of each

variable, holding all others constant. Table 4 presents the results.

Eleven of the variables thought to be signiﬁcant in determining mothers’ income in Mali
were indeed signiﬁcant at the 1, 5 or 10 percent level. Variables that were removed from
the model because they were insigniﬁcant include position of the mother within the
family (i.e., wife order), whether her marriage was polygamous, ethnicity dummy
variables, the value (in FCFA) of household livestock, total per-capita land holdings, area
of mothers’ ﬁelds and the presence of wells and energy-efﬁcient stoves in the household
compound. Seven variables with signiﬁcance levels greater than 10% were retained in the
model, including mothers’ and their husbands’ level of formal education, the number of
village associations, husbands’ expenditures, “distance to market” and the location
variables for Kolendieba and Bandiagara. The location variables were retained to show
how each zone compares to a single constant, the Koutiala zone. Number of associations
was retained because it just surpassed the 10% signiﬁcance level and was signiﬁcant in
other versions of the model. The other four variables were retained because there is a
strong underlying theoretical reason to keep them there, despite their signiﬁcance level.
Distance to market becomes signiﬁcant in the next model, and is therefore also kept in

this model for comparison reasons.

35

Table 4: Determinants of mothers’ income in Rural Mali: ZSLS regression results
Dependent variable = (natural log of mothers’ expenditures)

 

 

EQIanatory variables Coef. t-test Si L
Human Capital
Years of formal education of mother .018 0.43
Years of formal education of husband .019 1.03
Social Capital
Age .026 5.74 "*
Husband is head of household .191 1.78 "'
Number of village-level associations .036 1.60
Physical Capital and Financial Resources
Agricultural equipment index -1.008 -2.31 **
Ag. Equipment index "' Segou Region .943 1.71 *
Vehicle index per adult equivalent .780 4.45 ***
Natural log of husband’s expenditures (instrumented) .178 1.34
Labor
Ratio of females aged 10-45 to other household members .352 1.76 *
Enabling Environment
Distance to market in kilometers -.006 -I .27
School = l .434 3.96 **"‘
Kolendieba = 1 .199 1.32
Macina= 1 .881 4.10 ***
Niono = 1 1.097 5.35 *"'*
Bandiagara = 1 .374 1.42
Koro = 1 .483 1.96 **
Constant Term 6.239 4.37 ***
Prob > F 0.00
R-squared .301 0

 

Source: Estimated from survey data
Notes: The number of observations is 984.
“' Signiﬁcant at the 10 percent level
" Signiﬁcant at the 5 percent level
"“" Signiﬁcant at the 1 percent level

The education level of both the mothers and their husbands were found to be insigniﬁcant

in the model. However, this may be explained by the small variation across observations,

and the failure of a formal education variable to capture knowledge obtained through

informal education and experience. Over 94% of the women included in the study have

no formal education, compared to 84% of their husbands. In Mali today, only 32% of

girls complete primary education (MCC 2006), and the majority of these are situated in

36

the capital and other urban centers. Of the women and men who have some formal
education, 53% and 45% of them, respectively, are found in the Segou region, where

mothers’ average expenditures are the highest.

There is a positive and signiﬁcant relationship between mothers’ age and expenditures.
As a woman gets older, household duties are passed to younger wives and/or other
female members of the household. As a result, the woman has more time to devote to
income-generating activities. However, this effect may be minimized somewhat by the

fact that all the women in the sample have children under. the age of ﬁve years.

Women appear to beneﬁt from being married to the head of the household, holding all
else constant. Women married to heads may have better access to household-level assets,
such as land for private ﬁelds, equipment to plow these ﬁelds or cows to milk.
Additionally, the status afforded to the head’s wife may limit the time she spends on

household chores relative to the wives of dependent males.

There is a positive but marginally insigniﬁcant relationship between the number of
associations in the village and women’s expenditures. While other studies have shown
women to beneﬁt from participation in a cooperative (see Saito, 1994), it is likely that the
variable did not capture this relationship adequately. Our data do not specify whether the

women in our sample directly participated in these associations or beneﬁted indirectly.

Both time-saving asset variables — having a well and/or an energy inefﬁcient stove in the
compound — were found to be insigniﬁcant to mothers’ income. The insigniﬁcance of

these variables could be explained if, as Wooten (2003) notes, unmarried girls are

37

primarily responsible for household tasks, such as gathering water and ﬁrewood and child
care. In the poorer households, the cheapest labor-saving device for a mother may be a
young girl. Thus, in proposing labor-saving devices, their cost needs to be evaluated
relative to the (low) ﬁnancial cost of the alternative technology - using someone else’s

labor.

Assets found to have a signiﬁcant inﬂuence on mothers’ expenditures include a higher
vehicle index and the agricultural equipment index. Having transportation options in the
household is positively correlated, and could be used to drive to markets and towns to
buy inputs or sell products. 19 The effect of agricultural equipment is not uniform across
regions. A higher agricultural equipment index exhibits a negative relationship to
mothers’ expenditures for the Sikasso region and a positive relationship for women in
Segou. The coefﬁcient for Mopti, which has the lowest levels of equipment, is not
signiﬁcant. In Sikasso, the possession of more equipment could indicate that more
hectares of cotton are planted, resulting in higher demand on women’s time for unpaid
labor to assist with the harvest and weeding. This result contradicts the ﬁndings of Lilja
et a1 (1996), which state that the introduction of agricultural technologies in Sikasso
increased women’s wages for agricultural labor on communal lands and reduced
inequality between men and women. Conversely, the effect of agricultural equipment on
women’s income is positive in the Segou region. Here, access to this equipment may
provide women with better ﬁeld maintenance and water control on lands planted to

horticultural crops in the dry season. Additionally, it is plausible to expect that the

 

'9 As noted earlier, the exogeneity of this variable is questionable, so causality should be interpreted with
caution.

38

equipment index variable is correlated with better access to controlled irrigation. If so,

this variable may be picking up some of the effects of better water control.20

The value of livestock holdings per adult equivalent was found to be insigniﬁcant in
determining mothers’ expenditures, despite being signiﬁcant to total household
income/capita. This is telling of women’s access to these assets, controlled by the men in
the household, and speciﬁcally, the head of the household. A more precise instrument
would be to isolate female possession of small ruminants and chickens, which are more
ﬂuid than cattle and oxen in their ownership, and may fall under the control of women.
Both total household land holdings (per capita) and the estimated areas of women’s
personal ﬁelds (in ha) were insigniﬁcant in the analysis. In Lilja (1996), the latter was
found to be signiﬁcant and negative in determining the daily female wage rate on
communal ﬁelds, as the granting of these personal ﬁelds by the head of household is
considered part of the compensation for communal work.21 In the same study, women’s
earnings on private ﬁelds were sixteen times higher than mean incomes from the
communal ﬁelds, and represented the single greatest source of income for ﬁve of the six

regions under study.

The level of the husband’s expenditures has an insigniﬁcant effect on the expenditures of
his wives. This lends support to the hypothesis that women and men in Mali have
separate economies. As demonstrated in Figure 1, the expenditures of women in the

cotton zone are relatively low, despite high levels of expenditures on the part of their

 

2° The data do not allow for us to control for the effect of irrigation directly.

2' Women are granted temporary use right to these ﬁelds from the head of the household. As stated by
Lilja (1996), “the implicit rent for the private ﬁeld is paid in terms of the ‘obligatory’ production of
vegetables for the household consumption.” This obligation reduces the women’s ability to produce more
proﬁtable crops for the market.

39

husbands. This would indicate that programs that aim only to improve men’s incomes
(such as credit for the purchase of agricultural equipment) do not have a complimentary
effect on women’s incomes. This is not to say, however, that women do not beneﬁt at all
from being married to a high earner: women may still beneﬁt from their husband’s
expenditures on food, medicine, etc. (although the high levels of child malnutrition in the
cotton zone indicate that higher men’s earnings do not increase the welfare of other

household members).

As hypothesized, the number of females aged 10 to 45 relative to other members of the
household has a positive and signiﬁcant effect on the women’s income. Women in this
age range tend to share the responsibility for household chores, such as cooking, cleaning
and childcare. A higher female to “other” ratio would thus indicate less work per female
and more time available to devote to activities outside the home. This variable is likely
associated with age, since the older women would be relieved from chores before

younger girls.

The effect of having a school in the community is positive and signiﬁcant. This does not
reﬂect better education for the mothers, since this was controlled for in the model.
Rather, the presence of the school could indicate a greater amount of investment in the
community by its members, the government, and/or non-governmental agencies. A
school will also reduce the amount of time that women are taking care of young children
during the school year, although this effect is likely dampened for the women in our
sample who all have children under the age of ﬁve years. Finally, schools may present a

market for women selling snacks and drinks to students.

40

Distance to market is negatively correlated as we would expect, although the signiﬁcance
falls just above the 10% level. This may indicate that women’s income-earning activities
are contained within the village in which she lives, such as petty commerce in
agricultural products. By comparison, distance to the market is signiﬁcant for both the
women’s husbands and total household income/capita, suggesting that men rely more on
the larger market system for their income. It can also be interpreted to mean that
remoteness alone is not adequate to discourage agricultural and non-farm activities and
that overall regional infrastructure, discussed above, is more important (Canagarajah et

al. 2001).

The zone dummies account for regional differences not controlled for elsewhere in the
model. The coefﬁcients represent the marginal beneﬁt of living in each of the zones
relative to living in Koutiala, which is situated in the cotton-growing region and
experiences the lowest average women’s expenditures. As expected, living in the
irrigated rice-growing region has the highest marginal effect on income relative to the
base, holding other variables constant. Within this region, Niono had a higher coefﬁcient
compared to Macina. Surprisingly, these zones were followed by Kolendieba in the
cotton-growing region and Koro in the coarse grains region. Bandiagara was not
signiﬁcantly different than the base. This indicates that the factors driving expenditures in
the Mopti region above those in the Sikasso region were captured in the model. Other
factors captured in the location dummies could be overall regional infrastructure,
ethnicity, cultural norms that dictate women’s place in the work arena, distance to the

capital city and production capacity of the land.

41

As discussed at the end of Chapter 2, the sample of women retained for our analysis was
more likely to be married to the head of the household than the women who were
dropped. Since women married to the household head have higher expenditures than
those married to dependent males, holding all else equal, it is likely that the average
expenditures depicted in Figure 2 are higher than those found in the true population.
Similarly, the average expenditures of the husbands are likely inﬂated due to the over-

representation of heads of households.

4.2 Model 2: Profiles of women earning income

To exploit available information on the income—eaming activities of women in our
sample, a second model was estimated which added activity dummy variables to the 17
variables used in the previous model. Inclusion of these variables helps to identify
further the characteristics of women with varying levels of expenditures. As participation
in different activities is often conditional on the women’s income and access to resources,
we cannot establish a causal path from activity to income (i.e., participation in the
activities are not “determinants” of income). However, women participating in activities
associated with lower incomes are likely facing resource constraints and barriers to entry
of higher-retum activities. It is expected that these outcomes will be region-speciﬁc. The
data do not allow us to perform capital budgets to determine the average return on

investment associated with each activity.

42

4. 2.1 Additional Explanatory Variables Employed

Dummy variables were used to indicate the mothers’ participation in various income-
generating endeavors. Eight different activities were identiﬁed from the data: non-farm,
agricultural labor and gardening, sales of cereals, rice or “other” agricultural products,
animal sales, animal product sales and collection and sale of wild products. Two sources
of information, drawn from two different questionnaires, were used to determine the
woman’s participation in each activity.22 In the ﬁrst, the woman was asked to report any
income that she earned since the last interview date (bi-monthly), and to name the
activity by which she earned it. In the second, the woman was asked to give the source of
money for each expenditure made during the interview period. For each of the income
activities listed above, the woman was given a value of one if she identiﬁed the activity
in either of the questionnaires. Unfortunately, the data do not allow one to capture the
intensity with which she was involved in the activity or the amount of resources which
were invested. The percentage of women participating in each activity by region is

presented in Table 5.

The non-farm variable is the broadest in scope, since the expenditure questionnaire does
not allow for further disaggregation. The large majority of non-farm activities identiﬁed
in the income questionnaire involve some form of petty commerce — including the
purchase and re-sale of agricultural products — followed by crafts. Very few women were

involved in high-input or skilled trades.

 

22 Since the interviews were carried out periodically throughout the year, seasonality in expenditures is
captured.

43

.83 Eat Basso—«O 68:5
53:8 some 5 newsman 5&2 2: E 5an ho owaucooba 2: E0858 8855.3 5 mo:_e>

 

 

 

 

 

 

 

 

 

 

 

 

a! E. o8 3m 8 SN «2 a: 08 .28
can...“ : ashes 3.3 s 3.39 3.3 a 4.50. 3 3.3% 6.3
9. 3a 8 a: 3. B. P we EN :8:
3.6.9 cs3: §39 Ashes cant can»: $.33 Asoae
2 SN 3: a: 8 E E. 2m 8m 835
3.5.3 $.35 :38 eﬁag can”. $.33 Ashes 3... .3
8 m8 8? 3a 9 2. 5 2” man 359.5
mum—aw moﬁw mmﬁw QUE mgmw moﬁw 69¢ mgmw boom; 9:81:02 coaom
6:85 9:26 328 .653 .955 252.? 5:63
2.2, .28
493.2

 

 

 

:3»?— E 8523: mamaaaoaowioﬁeoﬁ E ”Egan—Eta.— ..oEoB no Lona—=2 ”m 93:.

44

 

Agricultural activities, such as transplanting and weeding rice ﬁelds, are distinctly
identiﬁed from the sale of own-production agricultural goods. The latter include the sale
of rainfed cereals, rice and “other” agricultural products. Rainfed cereal sales include
millet, sorghum, maize and fonio and exclude rice. “Other” agricultural products include
high-value horticultural crops such as tomatoes, cucumbers, onions and melons, as well
as pulses (including groundnuts), sesame, leaves and legume hay. Horticultural products
are most prevalent in the irrigated-rice region of Segou, while pulses dominate in Mopti

and Sikasso.

Wild products include the sale of all products that are collected rather than produced or
purchased. Examples include karité (shea nut butter), soumbala (a spice made from the
African locust tree and used in sauces) charcoal, fuelwood, baobab leaves and honey
(Sundberg, 1989). The collection of most of these products is from common pool
resources: permission to harvest is given by the chief of the village. Shea nuts are the
exception. For those trees found in the ﬁelds and fallow lands of particular families,
permission to harvest is given by the head of the household (Becker 2000). Most wild
products include some form of processing to add value. Shea nut butter is the most

transformation-intensive and the most proﬁtable.

To examine the different income effects of each activity by region, interaction variables
were created by multiplying the regional and activity dummy variables. Those found to
be signiﬁcant were retained in the model. It is expected that the diversity (i.e., number)
of activities in which the mother participates will be reﬂected in her earnings. While in

some cases, participation in only one activity may signal specialization and a high level

45

of intensity, it is more likely that these women lack the time and/or opportunity to engage
in multiple activities. On the other extreme, women who spread themselves too thin
cannot devote signiﬁcant time or resources to any one of the activities. It is therefore
hypothesized that women who engage in a middle-range number of activities will have
the highest incomes. Categorical variables were created to represent the number of
activities. Participation in one activity is omitted from the model; its effects will be
picked up by the constant. Women participating in two to four and ﬁve to eight different
activities were each represented by a dummy, respectively.23 Table 6 shows the number

of activities in which women participated by region.

Table 6: Number of activities per mother by rgion

 

 

 

 

 

 

Number of different activities per mother
Region 1 2 3 4 5 6 7 8 Total
Sikasso 41 123 69 83 62 1o 4 o 392
Ségou 23 77 125 65 35 3 3 2 333
Mopti 37 61 89 45 24 4 1 o 261
Total 101 261 283 193 121 17 8 2 986

 

 

 

 

 

 

 

 

 

 

Source: Calculated from data.

4. 2. 2 Estimation Strategy

Model 2 is estimated with two-stage least squares (ZSLS), in similar fashion to the ﬁrst
model, but with an additional vector of variables representing participation in income-
generating activities. In the ﬁrst stage, the natural log of fathers’ expenditures is
predicted using a set of instruments, denoted by H l 1,. . ..H 1 g, and the exogenous variables

determining mothers’ expenditures:

 

23 The number of activities could not simply be represented by a continuous variable because of the issue of
multicollinearity between the number of activities and the dummy variables for the individual activities.

46

J K L M N
XL; =§+ lejVj,i+Z'12ka,i+ 2'13IXIJ + Z l4mYm,i + Eisnzm +

j=l k=1 [:2 m=l ":1
F G
Zia/VJ + 217gHgJ+Ei (9)
f:] gzl

where A is a vector of income—generating activities, and 7% is an F X 1 vector of
parameters associated with these activities. In the second stage, the estimated value of

the dependent variable from equation (9), 191,1: is added as a regressor in the model for

mothers’ expenditures:

J K L M N
MEI =r7+ 2101;ng +ZP2kaJ + 2P31X1J + Z P4mYm,i + ZPSnZnJ +
j=l k=l [=2 m=l n=l
F A
ZP6fAfJ+ Muwz‘ (10)
f=l

As with the previous model, this model was found to be heteroskedastic with a p-val <
0.0005, and was corrected using White’s heteroskedastic-robust covariance matrix

estimator.

4. 2.3 Empirical Results

The signiﬁcance of a few of the variables in the ﬁrst model changed with the addition of
the activity dummy variables (Table 7). The signiﬁcant negative relationship between
mothers’ expenditures and distance to market indicates that for women involved in the
different activities, those closer to the market tend to have higher expenditure levels.
Being married to the head of the household loses its signiﬁcance. This may indicate that

being married to the household head provides a woman the means to participate in

47

Table 7: Determinants of mothers’ income in Rural Mali: ZSLS regression results

 

 

Explanatoq variables Coef. t-test Sig.
Human Capital

Years of formal education of mother .024 0.68

Years of formal education of husband .007 0.48
Social Capital

Age .017 4.25 ***

Husband is head of household .123 1.38

Number of village-level associations .031 1.50
Physical Capital and Financial Resources

Agricultural equipment index -.716 - l .99 **

Ag. Equipment index * Segou Region .809 1.67 *

Vehicle index per adult equivalent .624 3.86 ***

Natural log of husband’s expenditures (instrumented) .227 2.12 **
Labor

Ratio of females aged 10-45 to other household members .452 2.60 ***
Enabling Environment

Distance to market in kilometers -.009 -1.70 *

School = 1 .408 4.18 ***

Kolendieba = 1 .534 3.65 ***

Macina = 1 1.266 4.82 *"‘*

Niono = l .982 3.88 ***

Bandiagara = 1 .325 1.44

Koro = 1 .675 3.23 ***

Income-generating activities
Participation in:

Non-farm activities .196 1.78
Non-farm activities * Ségou Region .286 1.73
Agricultural labor -.203 -2.44 **
Animal sales ' .468 6.75 ***
Animal product sales .373 2.43 ”
Cereal sales (excl. rice) .151 1.50
Cereal sales * Segou Region -.614 -4.20 "‘**
Rice sales .357 3.47 *"‘*
Rice sales * Sikasso Region -.599 -4.30 "*
Other agricultural sales (incl. horticultural crops and pulses) .310 3.17 ***
Sale of wild products .620 5.83 *"‘*
Sale of wild products * Ségou Region -.605 -3.31 ***
Participation in two to four distinct activities .668 4.01 ***
Participation in four to eight distinct activities .930 4.46 ***
Constant Term 4.508 3.88 ***
Prob > F 0.00
R-squared .5 166

 

Source: Estimated from survey data
Notes: The dependent variable is natural log of mother’s expenditures. The number of observations is 984.
*, ", ‘" indicate signiﬁcance at the 10, 5 and 1 percent level, respectively.

48

different activities (e. g., access to household livestock). After these activities are
accounted for, being married to the head does not add to a woman’s expenditures. The
natural log of the husband’s expenditures gains signiﬁcance in the second model. This is
harder to explain, and could be due to problems with the instrumental variable. Finally,
the Kolendieba zone becomes signiﬁcantly different from the constant, Koutiala.
Kolendieba and Koutiala are both found in the cotton region. The insigniﬁcant difference
between the two zones in the previous model is likely a result of the women having
access to similar income-generating activities. The signiﬁcant difference in this model
indicates that there is something else going on between the two zones that is not
accounted for. Once activities are controlled for, Bandiagara is not signiﬁcantly different

than the constant.

The activity dummy variables provide information on the relationship between
participating in a certain activity and the mothers’ income levels. Seven of the eight
activities had a signiﬁcant relationship with income across the three regions. Cereal sales

were insigniﬁcant for Sikasso and Mopti, but signiﬁcant for the Segou region.

Of the signiﬁcant activities, earning income through agricultural labor was associated
with lower incomes. This was surprising, particularly for women in the rice-growing
region of Segou, who report signiﬁcant earnings from transplanting rice (See Box 1).
Since many women engaged in this particular activity are paid in-kind with rice, which
they in turn sell, it is possible that the beneﬁts are captured in the “sale of rice” variable
(discussed below). The negative coefﬁcient for the Sikasso region could indicate that

women who resort to working on others’ ﬁelds do not have opportunities to produce their

49

own crops or engage in other income-generating activities. Only one woman in the Mopti

region participated in this activity, so it is not possible to gauge its impact on incomes.

Of the agricultural products produced by, or given to, the women, the sale of cereals —
including sorghum, millet, maize and fonio - had the smallest marginal impact on
incomes. The coefﬁcient was positive but marginally insigniﬁcant for women in the
Sikasso and Mopti regions, where more proﬁtable opportunities are limited. Conversely,
the coefﬁcient is negative and signiﬁcant for women in Segou. This implies that in this
region, there is an opportunity cost of selling traditional grains instead of engaging in
more proﬁtable activities, such as rice and horticultural crop production. Women who sell
cereals, therefore, likely lack access to these other opportunities (e.g. land, irrigation,

inputs).

The effect of the sale of rice is positive and greater than that for cereals in Segou and
Mopti, and negative for women in Sikasso. The large difference in coefﬁcients can be
due to several factors. First, the technologies for growing rice differ greatly between the
regions. Segou is the main rice-growing region in the country and is dependent on
irrigation schemes. Yields of greater than six tons per hectare are common in this region,
compared to an average yield of 1.54 tons per ha for the country (FAOSTAT, 2004). In
Sikasso, rice is predominantly rain-fed and grown by women. Unlike Segou, where rice is
grown for the market, women in Sikasso do not typically sell their crop (Kelly). Women

who are selling in this region may do so because they are in an economic or food crisis.

The sale of “other” agricultural crops is positive for women in all regions, with a

coefﬁcient of 0.310. However, the data do not allow for further disaggregation into

50

 

 

Box 1. Women’s Role in Agricultural Production in the Ofﬁce du Niger

Women play an important role in agricultural production in the Ofﬁce du Niger.
Women’s participation in rice production is typically limited to the transplanting and
harvesting of rice in ﬁelds controlled by men, although examples do exist of female-
headed rice parcels (both individual and cooperatives). Women have been credited for the
widespread adoption of transplanting techniques in the ON, which were responsible for
signiﬁcant gains in yields in the late eighties and early nineties. Women are preferred
more often than men for this task because they reportedly produce work of higher quality
(Bélieres et al., 2003). Since 1992, the formation of women’s “working groups”
speciﬁcally for transplanting and harvesting have become common. In 2000, these groups
were responsible for completion of these tasks on 69% of the ﬁelds on rehabilitated land
(Bélieres et al., 2003). Women are compensated with cash and/or payment in kind. While
women are usually paid for work completed on their familial lands, it is reported that
women are working less frequently on lands controlled by their head of household (Osté,

2001).

It is common for women to invest the money earned during the rainy season rice
campaign in horticultural production (Kelly, personal comm.) Typically, women produce
horticultural crops in the off-season on land controlled by male members of their
families. It is estimated that 10 to 20% of the land is used in the dry season for gardening
(Béliéres et al., 2003; Aw and Diemer, 2004). Use of this land is provided freely,
although women are required to pay the water fee out of their proﬁts.

Osté (2001) cites several examples of group gardens cultivated by women’s associations.
The ﬁrst market gardening perimeter speciﬁcally for women was installed in Macina in
1984. Group gardens can now sporadically be found in districts throughout the Ofﬁce.
According to Osté (2002), women’s gardening plots produced an annual gross revenue of
420 000 FCFA (640 euros). The level of revenues realized by women depends on the
availability of land, the differing levels of investments, and degree of processing. The
cultivation of onions (shallots) is often the most practical due to their low perishability. In
the ON, the cultivation of onions was found to be more than twice as proﬁtable as the
cultivation of rice (Osté, 2001). Total revenues from gardening in the Ofﬁce du Niger in
1999 are estimated at 18.3 million euros (Chohin-Kuper et al., 2002).

 

different categories of crops. It is likely, therefore, that the coefﬁcient masks signiﬁcant
differences between high-value crops, such as horticultural crops and groundnuts, and
lower-value crops, such as leaves and hay. With irrigation and stronger market linkages,
Segou produces more horticultural crops than the other regions. It is possible that the

differences in crop production are being captured by the location dummies.

51

 

Eighty percent of the women in the sample earned income through non-farm sources. For
women in Sikasso and Mopti, the activity yielded a coefﬁcient of .196 — which lies
within the range of coefﬁcients associated with the sale of agricultural crops. For women
in Segou, however, the coefﬁcient increases to .482. Unfortunately, the variable does not
distinguish between different types of non-farm activities. Therefore, it is unclear
whether this difference is driven by a different set of activities in the regions or results
from a more active and dynamic market system in Segou. The higher average incomes in
Segou relative to Mopti and Sikasso could allow women there to invest more resources in
higher-return activities. In addition, the more dynamic economy in the Ofﬁce du Niger
may increase the demand for non-agricultural products, making those activities more

proﬁtable for women.

Women who sold animal products, including milk, eggs, butter and skins, saw a positive
marginal beneﬁt of participating in the activity. Over half of these women reside in the
Mopti region. The coefﬁcient on this variable (.373) was greater than those for the
agricultural and non-farm activities. However, the relatively few women who reported
income from this source (3.3% for Sikasso, 7.8% for Segou and 15.7% for Mopti)
indicate that access to animal products is limited. This observation is corroborated by the
insigniﬁcance of total household livestock assets on women’s income. Conversely,
women who sell animals (poultry and small ruminants) under their own control see the
log of their expenditures increase by .468, holding all other variables constant. Twenty-
ﬁve percent of women across the sample sold at least one animal during the survey

period, with the highest percentage of women coming from the Mopti region (41%).

52

Finally, there is the large marginal beneﬁt of participating in the collection and sale of
wild products for women in Sikasso and Mopti. For women involved in shea butter
production in Sikasso, this outcome is consistent with other studies that place shea among
the most proﬁtable activities for rural women in Mali (Sundberg 1989; DeGroote and
Coulibaly 1998). However, for women in Mopti, where shea nut trees are not found, this
result is more surprising. Here, women engaging in this activity are largely restricted to
low-input, low-return products, such as fuelwood and baobab leaves. Collection and sale
of these products may represent one of the few income-opportunities for these women.
Conversely, women from Segou - with more attractive activities - do not signiﬁcantly
gain from the sale of wild products. Shea is also less present in this region than in the

south.

It was hypothesized that women who practice some diversiﬁcation of income-generating
activities would have a higher marginal beneﬁt than women who participate in either one
or many activities. However, our data show that more is better (Figure 3). Women who
participated in ﬁve to eight different types of activities had higher average expenditures
than women engaged in one to four types of activities. This result holds up to multivariate
analysis (Table 8). However, it should be noted that only 15% of women fall into the
latter category, and looking at the averages masks the variation within the categories.
Amongst women in the top decile of expenditures, only two participate in six or more

types of activities.

53

Figure 3. Mothers' average expenditures by degree of income diversiﬁcation

 

80,000-

60,000‘

40,0004

20,000#

Mothers' Average Expenditures (FCFA)

 

 

 

-—l
—

I l l l l
1 2 3 4 5 6 7 8

Number of Activities

Source: Calculated from data
1 USS = 560 FCFA

54

5. DISCUSSION AND CONCLUSIONS

Several of the results presented above merit further discussion. First is the large (and
signiﬁcant) discrepancy between female and male income, particularly in the cotton zone
(Figure 1). Here, average husbands’ expenditures are nearly four times the value of their
wives’. This gap decreases when the heads of households are removed, although
dependent males still earn, on average, more than three times their wives do. This
indicates that agricultural growth and higher household incomes do not automatically
lead to gender equity or better welfare for women and children.24 Conversely, production

systems that depend on cheap family labor may lead to greater intra-household inequities.

Women in the cotton zone also fare worse than their female counterparts in the irrigated
rice and millet/sorghum regions. This raises questions about family structure, access to
resources and opportunities to work off of communal lands in the different regions. A
fundamental issue is whether the production of cotton is dependent on a large family
structure and high levels of non- (or low-) compensated family labor. Indeed, a study of
“Cotton and Gender Politics in Cote d’Ivoire” revealed that “male household heads now
ﬁnd it necessary to contest women’s [activities]. . ...in their attempt to control their wives’
labor” (Bassett 2002). Social mechanisms employed by the heads of households to
restrict women’s own-activities included limiting the number of days per week that
women could work on their own enterprises and giving women the right to consume food

grown in the household ﬁelds only on days that they worked in them.25 In the cotton-

 

2’ This is based on studies, including one performed with these data, that indicate a positive and signiﬁcant
correlation between mother’s income/expenditures and the nutritional status of children under 5 years. See
Tefft and Kelly (2004).

5 These mechanisms were not uniform across ethnic groups.

55

growing region of Mali, there is pressure for institutional change in family structure even
among dependent males, who want to break the traditional control over productive assets

and labor by the head of the household (Lilja 1996).

The government of Mali continues to promote cotton expansion as a means to reduce
poverty. Clearly, if this strategy is to continue, complementary interventions are needed

to counteract the negative impact on women’s incomes.

Lilja (1996) found that a woman’s perception of her right to refuse work on the
communal ﬁelds had a signiﬁcant and positive impact on her daily wage rate on those
ﬁelds. However, only 15 percent on the women interviewed perceived having this right.
Conversely, women in the Office du Niger region of Segou, where average women’s
expenditures are highest, are working less frequently on lands controlled by their head of

household, due to higher compensation in others’ ﬁelds and/or off-farm (Oste 2001).

In lieu of working on household lands in the rice zone, many women have formed
“working groups” for transplanting and harvesting the rice. In 2000, these groups were
responsible for completion of these tasks on 69% of the ﬁelds on rehabilitated land
(Be’lieres et al. 2003). Using the terminology of Bassett (2002) and the notion of “scalar
politics,” participation in these working groups allows women to “jump” the scale of the
household to secure their own income. “Jumping scales,” as deﬁned by Bassett, “is a
sociospatial process in which a group seeks to transcend the routinized constraints of a
speciﬁc scale (e. g., the household) by engaging in activities at a different scale (e.g., the
village) to realize its objectives” (p. 353). Women who work in these groups may be less

dependent on assets controlled by the household head, who may have motives to

56

undermine women’s independent initiatives in order to secure their labor on communal

ﬁelds.

Anthropological and/or sociological research is needed to investigate the links between
production systems, household structure and women’s autonomy. Econometric analysis
does not shed light on why women in the rice zone have been able to beneﬁt from

increases in production while women in the cotton zone have not.

Human capital indicators, namely the education levels of women and their husbands, did
not play a signiﬁcant role in determining the women’s income levels. However, this does
not suggest a reason to stop investing in women’s education. Data from many other
countries (see, for example, Smith and Haddad, 2000) show the important link between
education, women’s incomes and child health. In rural Mali, there are simply not enough
educated women for this to be reﬂected in the data. Also, there may be payoffs to non-
forrnal education that the model failed to capture. Of the women in our sample, more than
half of them who had at least some formal education, and 90 percent with ﬁve or more
years of schooling, live in Segou, the region with the highest observed incomes for

mothers.

Variables representing a women’s social capital were signiﬁcant and positive. First, the
age of the woman is highly signiﬁcant. In rural Mali, status comes with age. Senior
women (and men) do not have obligations to the household economy. Once retired from
these domestic roles, they can spend most of their time tending private plots or engaging
in other activities (Wooten 2003). This also indicates that the youngest women, who are

likely to have the youngest children, are the least likely to be earning the income needed

57

to care for them. When designing income-generating programs for women, one needs to
be proactive in identifying income-earning activities in which younger women are likely
(and able) to participate. Alternatively, the provision of support services that increase

women’s productivity and/or relax their time constraints will be beneﬁcial.

All else equal, being married to the head of the household is signiﬁcant for women’s
income levels, while being in a polygamous marriage and wife order were not. This
would suggest that being a fourth wife to the head of household is generally more
beneﬁcial than being the ﬁrst wife of a dependent male. This relates back to the issue of
family structure discussed above. The low percentage of .women married to the
household head in the Mopti region (37%, compared to 46% and 47% in Sikasso and
Segou, respectively) could contribute to the low average expenditures found in that

region.

The discrepancy between husbands’ and wives’ incomes indicates that men beneﬁt
disproportionately from access to productive assets and inputs. This claim is supported
throughout the literature.26 In our study, the insigniﬁcance attached to household land and
livestock holdings suggest that access to these assets is limited for women. The effect of
agricultural equipment in the Sikasso region is signiﬁcanl and negative on women’s
incomes. As growth in cotton production is largely extensive, more equipment is an

indicator of larger ﬁelds and increased demand for women’s labor that is rarely

remunerated.

 

2" Aldermen et a1. (1995) use econometric evidence to suggest that total household productivity could be
increased by 10 to 20% if existing resources were shifted from male- to female-controlled plots; Saito et al.
(1994) conclude that if women had access to the same inputs as men, productivity of identical crops would
be approximately 15% higher in women’s ﬁelds than in men’s; Udry (1996) concluded that approximately
6% of output is lost because of inefﬁcient factor allocation within the household.

58

There was a lot of controversy in Mali after the devaluation of the currency (FCF A) in
1994. Banks gave loans to men to purchase motorbikes, a move that was criticized as a
frivolous waste of resources that could lead to over-indebtedness and a neglect of
household responsibilities. However, our results show that women also appear to beneﬁt
from the household having invested more money in vehicles, which likely improve their
ability to obtain inputs and market their output. It would be interesting to investigate this
issue further to determine the speciﬁc effect of motorbikes vis a vis other vehicles, and to
see if different regions and levels of infrastructure beneﬁt disproportionately from these

investments.

The signiﬁcance of the zone dummies shows that there are zone-level characteristics that
we have not captured in the model that are inﬂuencing women’s ability to earn income.
Certain zones may have stronger afﬁliations with non-govemmental organizations or
better access to government resources. There may simply be more income-generating
opportunities in some zones relative to others. Isolating the prevalent activities found in
each zone and better enumerating infrastructure, agro-climatic conditions and the state of

the local economy could shed more insight on this phenomenon.

The second model offers some intriguing insights into the activities performed by women
in the three regions under study. While we cannot conclude that participation in these
activities leads to higher or lower incomes (causality is not determined), we can make
some inferences about the women who participate in each activity. This information

could help the targeting and design of projects by those interested in increasing women’s

59

incomes. Future work should examine the costs of entry into certain activities and their

respective net returns.

There are low barriers to entry into the non-farm sector. However, the wide level of
participation across regions (ranging from 76% in Mopti to 82% in Segou) and relatively
low marginal beneﬁt mask a great diversity of activities included in this sector. Women
with low access to resources can sell their labor for menial tasks such as pounding millet
or doing laundry. Other activities, such as tailoring, require larger start-up costs, asset
accumulation and skills. There is a signiﬁcant and higher coefﬁcient for women engaged
in non-farm activities in Segou. This could signal more and/or better non-farm activities.
Due to their higher average incomes, women in this region may be able to invest in
higher-return activities, and the more dynamic local economy may generate higher

demand for non-farm activities.

The negative coefﬁcient associated with agricultural labor suggests that women who
participate in this activity do not have resources to produce and market their own crops,
which yield higher returns. This could be due to land constraints or inability to access
other inputs, such as seeds and fertilizers. In the market-oriented cotton and rice zones,
opportunities exist for women to work on others’ ﬁelds. This contrasts with Mopti, where
only one woman reported participating in this activity. This may have implications for
resource-strapped and/or landless women, who are faced with few other opportunities to

earn income.

Roughly 62 percent of women in both Sikasso and Mopti‘participate in the sale of rain-

fed cereals. These women have incomes slightly higher than those who do not, all else

60

equal. The low return of this activity could signal that women are selling their crop right
after the harvest when prices are low. It could also mean that women are selling cereals,
which they were given for food preparation, in very small amounts to purchase
condiments. Women who sell these crops in Segou have incomes signiﬁcantly lower than
women who do not, ceteris paribus. This indicates an opportunity cost to participating in
this activity. Women who sell rain—fed cereals in this region do so at the expense of
engaging in higher-return activities, such as selling rice. The constraints and bottlenecks
to entering, or increasing their participation in, more lucrative activities need to be

identiﬁed.

Over half of the women in Segou sell rice. Their marginal beneﬁt is higher than for sale
of any of the other agricultural crops. Conversely, women who sell rice in Sikasso have
signiﬁcantly lower incomes. As rice in Sikasso is primarily grown for home

consumption, its sale may indicate some form of distress and/or lack of opportunities to

earn income through other activities.

Almost 80% of women participated in the sale of crops other than cereals, rice or cotton.
Across the zones, women who participated in this activity had higher incomes than
women who did not, holding all else constant. However, the data do not distinguish
between horticultural crops and peanuts, which tend to be high-value, and other crops,
such as bean leaves and tubers. A subset of 372 women participating in an additional
Knowledge, Attitudes and Practices (KAP) survey allowed us to investigate the incomes

of women participating in “market gardening” (14.5% of respondents, of which 63%

61

reside in Segou). 27 On average, incomes of those who did participate were 70% higher
than their counterparts who did not. This difference in incomes was only signiﬁcant for
women in Segou, pointing to the lack of gardening infrastructure and access to markets in
the other zones. Studies by Wooten (2003), Becker (2000) and Ward et al. (2004)
highlight the potential of market gardens for income generation and food security in

Mali.28

Across regions, women who sold livestock and animal products had signiﬁcantly higher
incomes than women who did not, all else equal. The marginal beneﬁt of participating in
these activities is greater than for any other agricultural endeavor. In both categories,
participation was highest in Mopti, where agricultural opportunities are limited and
animal production is more widespread. Access to animals for sale is greater than access
to animal products, which may be an indicator of the high cost of cows for the latter, and
their control by the head of the household. Conversely, women typically sell poultry and
small ruminants, which they buy with income earned through other activities (DeGroote

and Coulibaly 1998).

Finally, participating in the sale of wild products was associated with the largest marginal
beneﬁt of any activity for women in Sikasso and Mopti (women in Segou beneﬁted
signiﬁcantly less). This may be due to several factors. First, as common pool resources
(with the exception of shea trees), most women have access to at least some wild
products, regardless of personal or household assets. For poor women, this may be one of

the few income-generating opportunities available to them. As such, these women would

 

27 For the purpose of this analysis, a woman is considered to participate in market gardening if she reported
that she sells the majority of produce that she is responsible for growing.
2’ All three studies focus on communities within 100 km of the capital city, Bamako.

62

experience extremely low incomes in the absence of this activity. Second, certain wild
products, such as shea nut butter, can be highly proﬁtable ventures. Women in southern
Mali have reported in interviews that the production of shea nut butter is much more

proﬁtable than the sale of agricultural produce (Sundberg 1989).

The importance of wild products to women’s income has serious implications with
regards to tenure and sustainability. As a common pool resource, permission to harvest is
given by the chief of the village. An exception is for shea'nut trees, which belong to the
family on whose land the tree is located. Permission to collect is thus at the discretion of
the head of household. Women in households without such land ask and regularly
receive permission to collect on the land of other households (Becker 2000). Since the
activity is recognized as belonging almost exclusively to women, conﬂicts over their
access are rare. However, as agricultural production (particularly cotton), expands into
previously uncultivated areas, or as land in peri-urban areas is converted to non-
agricultural uses, women have no formal protection over these productive assets. Policies
which transfer legal rights to usufruct users should be considered. Secure land tenure for

women may also encourage sustainable use of the resources.

It is obvious from this study that any analysis of women’s incomes and activities need to
be disaggregated by region. Activities that are proﬁtable to women in one region —- with a
speciﬁc set of natural resources and assets — may lower incomes for women in other
regions through foregone opportunities to participate in more appropriate activities. One
set of national policy recommendations is unlikely to be successful given the huge intra-

country diversity.

63

APPENDIX

64

.mcozatomno c: no woman 83:86. =< ”862
8% .333 :8: 38.3.60” 850m

 

 

_ o wmm. .2. _n 83—
_ o Em. 3 _. _ u Smwﬂvgm
_ o 23m. 3:. _ M 6:22
_ o oom. mi. _ n «Eon—2
_ o amv. com. _ n 2.2—6:23—
. o mg. Noe. . . _ n Begum
a o cmo.m amad 323688 _o>o_-owm_=> .«o .38: Z
co 9 ”cm.” SH.» 336522 E woo—BE 9 62835
wetntmauuxgb §§EEQD
Em; N3. mﬂ. cam. £382: 20:96.20: .28 2 31: vowm 83:5; we 6:3—
amvg o EN. NNN. E67353 :25 Ba mo_oEo>
m3.— c m _ _. 3o. 3me ... x265 25:533— .23_:utw<
no _ ._ c an _. mm _. x265 E66933“ .m._==:o_uw<
_ o cav. nmv. _ u 228:6: .«o «626 m_ .653
N. o Now; So. cosmos? .258 he memo» $2333
92.: Nmod 03:6. 392 ABEoEEHmEV mung—2.56668 92:33: go wo— .6582
Suntaoeaexb Sexowzeﬁ
a o 36.. mwm. cosmos—co 3E5. mo 33>
mm 2 «mos emodm ow<
85126539 333935
682 52 .>oD .3m :82 833:3

 

83mg; 868qu 8m 8593 waE:m ”w 6331

< X—GZm—mm<

65

APPENDIX B

Table 9: Weights for the calculation of the Agricultural Equipment Indexa‘b

 

Full oxen chain (2 oxen and plow) 1.0

Additional full chain .75
Cart .50
Extra plow .35
Extra pair of oxen .35
Extra animal .25
Cultivator .25
Seeder .25
Sprayer .25

 

a Weights are based on farmer perception of the value of each animal/piece of equipment to production.
b Values are divided by the number of adult equivalents in the household.

Table 10: Weights for the calculation of the Vehicle Indexa

 

Bicycle 1
Moped 3
Motorcycle 5
Car 8
Truck 10

 

a Values are divided by the number of adult equivalents in the household.

66

APPENDIX C

Table 11: First-stage regression for husbands’ expenditures

 

 

Explanatory variables Coef. Std. errora Sig.

Exogenous variables in ME regression
Years of formal education of mother .005 0.22
Years of formal education of husband .067 4.29 ***
Mother’s age .010 1.62
Husband is head of household .539 7.54 ***
Number of village-level associations .061 2.51 **
Ratio of females aged 10-45 to other household members -. 148 -0.72
Distance to market in kilometers -.184 -3.70 ***
School = l -.210 -2.10 **
Kolendieba = 1 -.381 -3.00 ***
Macina -.712 -5.72 ***
Niono .191 1.49
Bandiagara -1 .302 -9. 10 * * *
Koro -1 . 109 -6.66 ** *

Instruments
Husband’s age .052 2.59 ***
Husband’s age squares -.001 -2.88 ***
% time husband spent in village 1.502 2.23 **
Per capita land holdings (hectares) .101 1.29
Area of husband’s ﬁelds (hectares) .034 5.08 ***

Constant Term

Prob > F 8.167

R-squared 0.00

Adjusted R-squared .3282

 

Source: Estimated from survey data

Notes: The dependent variable is natural log of husband’s expenditures (lnfaexp). The number of

observations is 984.
* Signiﬁcant at the 10 percent level
“ Signiﬁcant at the 5 percent level
*** Signiﬁcant at the 1 percent level

67

APPENDIX D

Table 11: Determinants of mothers’ income in Rural Mali: Comparison of ZSLS models
with and without the vehicle index as an explanatory variable

 

with without
vehicle ind. vehicle ind.

 

 

Explanatorxvariables Coef. Sig. Coef. Sig.
Human Capital

Years of formal education of mother .018 .013

Years of formal education of husband .019 .022
Social Capital

Age .026 *** .024 ***

Husband is head of household .191 * .148

Number of village-level associations .036 .034
Physical Capital

Agricultural equipment index - l .008 ** -4.32

Ag. Equipment index * Segou Region .943 * .984 *

Vehicle index per adult equivalent .780 *** -- --

Natural log of husband’s expenditures .178 .219 *
Labor

Ratio of females aged 10-45 to others in hh .3 52 * .369 *
Enabling Environment

Distance to market in kilometers -.006 -.008 *

School = l .434 *** .453 ***

Kolendieba = 1 .199 .227

Macina= l .881 *** .870 ***

Niono =1 1.097 *** 1.146 ***

Bandiagara = l .3 74 .437 *

Koro = 1 .483 ** .567 **
Constant Term 6.239 *** 5.904 ***
Prob > F 0.00 0.00
R-squared .3010 .2867

 

 

 

 

Source: Estimated from survey data

Notes: The dependent variable is natural log of mother’s expenditures. The number of observations is 984.

"' Signiﬁcant at the 10 percent level
" Signiﬁcant at the 5 percent level
**" Signiﬁcant at the 1 percent level

68

BIBLIOGRAPHY

69

BIBLIOGRAPHY

Abdulai, A. and A. CroleRees. 2001. Determinants of income diversiﬁcation amongst
rural households in Southern Mali. Food Policy 26: 437-452.

Alderman, H., J. Hoddinott, L. Haddad and C. Udry. 1995. Gender Differentials in Farm
Productivity: Implications for Household Efficiency and Agricultural Policy.
F CND Discussion Paper No. 6. Washington, DC: IFPRI.

Aw, D. and G. Diemer. 2004. Making a Large Irrigation Scheme Work: A Case Study
from Mali. Washington, DC: The World Bank.

... .I-f 'L;'

Barrett, C.B., T. Reardon, and P. Webb. 2001. Nonfarm income diversiﬁcation and
household livelihood strategies in rural Africa: concepts, dynamics, and policy
implications. Food Policy 26: 315-331.

 

 

Bassett, T.J. 2002. Women’s Cotton and the Spaces of Gender Politics in Northern Cote
d’Ivoire. Gender, Place and Culture 9(4): 351-370.

Becker, Laurence. 2000. Garden Money Buys Grain: Food Procurement Patterns in a
Malian Village. Human Ecology 28(2): 219-250.

Bélieres, J .-F., Y. Coulibaly, A. Keita and MK. Sanogo. 2003. Caracte'risation des
exploitations agricoles de la zone a'e l’Oﬂice du Niger en 2000 : Résultats d ’une
enquéte des exploitations agricoles dans un e'chantillon de villages. Mali, Ofﬁce du
Niger, DADR-URDOC.

Bloom, D. and D. Canning. 2000. The Health and Wealth of Nations. Science 287(5456):
1207-09.

Bonneval, P., M. Kuper, and J .-P. Tonneau, eds. 2002. L ’Oﬂice du Niger, grenier a riz du
Mali: Succe‘s économiques, transitions culturelles et politiques de développement.
Paris : CIRAD/Karthala.

Bound, J. D.A. Jaeger, and RM. Baker. 1995. Problems with instrumental variables
estimation when the correlation between the instruments and the endogenous

explanatory variable is weak. Journal of the American Statistical Association
90(430): 443-450.

Canagarajah, S., C. Newman and R. Bhattamishra. 2001. Nonfarm income, gender, and
inequality: evidence from rural Ghana and Uganda. Food Policy 26 : 405-420.

70

Chohin-Kuper, V. Kelly, R. Ducrot and Y.Coulibaly. 2002. La diversiﬁcation des
cultures: Le maraichage comme complement de la riziculture. In Bonneval, P., M.
Kuper, and J .-P. Tonneau (eds.), L ’Ofﬁce du Niger, grenier a riz du Mali: Succés

économiques, transitions culturelles et politiques de développement. Paris :
CIRAD/Karthala.

DeGroote, H. and N. Coulibaly. 1998. Gender and generation: an intra-household
analysis on access to resources in Southern Mali. African Crop Science Journal

6(1): 79-95.

Josué Dione. 1989. Informing food security policy in Mali: Interactions between
technology, institutions and market reforms. PhD dissertation. E. Lansing, MI:
Department of Agricultural Economics, Michigan State University.

Fafchamps, M. and A. Quisumbing. 1998. Human capital, productivity, and labor
allocation in rural Pakistan. F CND Discussion Paper 48, IFPRI.

Fapohunda, E. 1989. The Nonpooling Household: A Challenge to Theory. In Dwyer, D.
and J. Bruce (eds.) A Home Divided: Women and Income in the Third World.
Stanford, California: Stanford University Press.

Gujarati, D. 2003. Basic Econometrics: Fourth Edition. Boston, MA: McGraw-Hill
Higher Education.

Haddad, L., H. Alderman, S. Appleton, L. Song and Y. Yohannes. 2002. Reducing Child
Undemutrition: How Far Does Income Growth Take Us? IFPRI Food
Consumption and Nutrition Division Discussion Paper No. 137.

Haddad, L. and H. Bouis. 1991. Impact of Nutritional Status on Agricultural
Productivity: Wage Evidence from the Phillippines. Oxford Bulletin of Economics
and Statistics 53(1): 45-68.

Hill, M. and E. King. 1995. Women’s education and well-being. Feminist Economics
1(2):21-46.

Kelly, V. and J. Staatz. 2006. Mali Country Economic Memorandum (CEM): Agriculture
and Rural Development. Draft submitted to the World Bank, June 26.

Kelly, V., J. Tefﬁ, J. Oehmke and J. Staatz. 2004. Identifying Policy Relevant Variables

for Reducing Childhood Malnutrition in Rural Mali. Agricultural Economics Staff
Paper, 04-28.

71

Kennedy, E. and M. Garcia. 1994. Body Mass Index and Economic Productivity. In
James, W. and A. Ralph (eds.), The Functional Signiﬁcance of Low Body Mass
Index (BMI): Proceedings of an [DEC G Workshop held at FAO Headquarters,
Rome, Italy on 4-6 November, 1992. European Journal of Clinical Nutrition
Volume 48, Supplement 3.

Kumar, SK. 1987. Women’s Role and Agricultural Technology. In Mellor, J ., C.
Delgado and M. Blackie (eds.), Accelerating Food Production in Sub-Saharan
Africa, chapter 10. Baltimore, MD: Johns Hopkins University Press.

Lilja, N.K. 1996. Technological Change in Agriculture and the Welfare of Women in
Southern Mali. PhD dissertation, Purdue University.

Lilja, N., J. Sanders, C. Durham, H. DeGroote, and I. Dembele. 1996. Factors inﬂuencing
the payments to women in Malian agriculture. American Journal of Agricultural
Economics 78(5): 1350-1345.

Millennium Challenge Corporation. 2006. FY06 Country Indicator Rankings: Mali.
Available at: http://www.mccgov/countries/rankings/FY06/indexshtml

Robison, L. and M. Stiles. 2000. Social Capital: Sympathy, Socio-Emotional Goods and
Institutions. AEC Staff Paper No. 2000-45. E. Lansing, MI: Department of
Agricultural Economics, Michigan State University.

Osté, R. 2002. Les exploitations agricoles de l’Ofﬁce du Niger sont-elles rentable grace
aux femmes? In Bonneval, P., M. Kuper, and J .-P. Tonneau (eds.), L ’Office du
Niger, grenier a riz du Mali: Succés économiques, transitions culturelles et
politiques de développement. Paris : CIRAD/Karthala.

Osté, R. 2001. Diagnostic global de genre et développement a 1 ’Office du Niger. Rapport
ﬁnal. Mali, Ofﬁce du Niger, ARPON 4.

Pindyck, R. and D. Rubinfeld. 1998. Econometric Models and Economic Forecasts:
Fourth Edition. Boston, MA: Irwin/ McGraw-Hill.

Poulton, R-E. and I. ag Youssouf. 1998. A Peace of Timbuktu: Democratic
Governance, Development and African Peacemaking. New York and Geneva:
United Nations Institute for Disarmament Research, UNIDIR/98/2.

Psacharopoulos, G. and H. Patrinos. 2004. Returns to investments in education: a further
update. Education Economics 12(2): 113-134.

Quisumbing, A., L. Brown, H. Sims Feldstein, L. Haddad and C. Pena. 1995. Women:

the Key to Food Security. Washington, DC: International Food Policy Research
Institute.

72

Saito, K., H. Mokonnen and D. Spurling. 1994. Raising the Productivity of Women
Farmers in Sub-Saharan Africa. Africa Technical Department Series 230.
Washington, DC: The World Bank.

Schultz, TR 2005. Productive Beneﬁts of Health: Evidence from Low-Income
Countries. 1&1 Discussion Papers 1482, Institute for the Study of Labor (IZA).

Smith, L., and L. Haddad. 2000. Explaining Child Malnutrition In Developing Countries:
A Cross-Country Analysis. Washington, DC: International Food Policy Research
Institute.

Sundberg, Shelly. 1989. OHV Food Consumption and Expenditure Survey: Preliminary
Results on Income Sources in the OHV. Unpublished Resource Notes. Food
Security in Africa Cooperative Agreement, Bamako.

Tefft, J ., V. Kelly and J. Staatz. 2003. Linkages between Child Nutrition and Agricultural
Growth: Insights from Mali. Powerpoint slides from presentation to
USAID/Washington, December.

Tefft, J. and V. Kelly. 2004. Understanding and Reducing Child Malnutrition in Mali:
Interim Research Findings for the Project on Linkages between Child Nutrition
and Agricultural Growth. (LICNAG). AEC Staff Paper No. 2004-27. E. Lansing,
MI: Department of Agricultural Economics, Michigan State University.

Thomas, D. 1992. Intra-Household Resource Allocation: An Inferential Approach. The
Journal of Human Resources 25(4): 63 5-664.

Thomas, D. 1997. Incomes, Expenditures and Health Outcomes: Evidence on
Intrahousehold Resource Allocation. In Haddad, L., J. Hoddinott and H.
Alderman (eds.), Intrahousehold Resource Allocation in Developing Countries.
Baltimore, MD: Johns Hopkins University Press.

Udry, C. 1996. Gender, Agricultural Production, and the Theory of the Household. The
Journal of Political Economy 104(5): 1010-1046.

UNFPA. 2002. Chapter 7: Poverty and Education. State of World Population 2002.
Available at: http://www.unfpa.org/SWp/2002/english/ch7/page4.htm.

Ward, C., B. Ballif-Spanvill, A. F uhriman, Y. Solomon, and K. Widdison-Jones. 2004.
Human Ecology 32(4): 509-521.

Wooten, S. 2003. Women, Men, and Market Gardens: Gender Relations and Income
Generation in Rural Mali. Human Organization 62(2): 166-177.

73

Zeitlan, M., R. Megawangi, E. Kramer, N. Colletta, E. Babatunde and D. Garman. 1995.
Strengthening the Family: Implications for International Development. Tokyo:
United Nations University Press.

74

 

 

P

 

 

 

 

 

uuuuuuuuuuuuuuuuuuuuuu
”ll

   

1293 02845

  

. “- _ ‘
. w
_._._‘

ﬁll/llWill/llIll/ll’llll/l

IIIIIIIII

Will/ll!

5933