A STUDY OF HOUSEHOLD INCOME DETERMINANTS AND INCOME INEQUALITY IN
THE TOMINIAN AND KOUTIALA ZONES OF MALI
By
Brenda Nicole Lazarus

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food and Resource Economics - Master of Science
2013

ABSTRACT
A STUDY OF HOUSEHOLD INCOME DETERMINANTS AND INCOME INEQUALITY IN
THE TOMINIAN AND KOUTIALA ZONES OF MALI
By
Brenda Nicole Lazarus
According to the UN's Millennium Development Indicators, 57.6% of the rural
population in Mali was living below the national poverty line in 2006. To improve on this
statistic, it is important to understand the following about communities in rural Mali: 1) the
makeup of household incomes, 2) factors associated with higher income levels, and 3) the levels
of income inequality in these communities. This thesis used panel household data from the
Cercle of Tominian and the Cercle of Koutiala to examine these issues. More specifically, a
descriptive statistics analysis of reported household incomes was performed, comparing incomes
across zones, years, and income quartiles. This showed that households in both zones were poor
with only 8-16% of all households earning more than $1/day per capita. It also showed that
households in Koutiala earned considerably more income than households in Tominian and that
food crops are the most important income source for households in both zones.
A Heckman two-step model was also estimated to better understand the determinants of
income for cropping, livestock, and nonfarm activities. This analysis showed that having a larger
household size and living in Tominian zone were associated with lower probabilities of activity
participation and/or lower incomes, while wealth and durable goods indicators, easy road access,
and having a household head with at least a primary school education had the opposite effect.
Finally, to determine whether certain income activities increase or decrease income
inequality levels, regional Gini coefficients were calculated and decomposed. This analysis
showed low levels of income inequality with Gini coefficients ranging from 0.37 to 0.42.

ACKNOWLEDGMENTS

This thesis would not have been possible without the guidance and support of numerous
people. First and foremost, I would like to thank my major advisor, Dr. Valerie Kelly, for all of
her advice, assistance, and knowledge about Mali that she has provided from the day that I
started this thesis. I would also like to thank Dr. Songqing Jin for all of his guidance as I
developed my econometric models, and Dr. Jim Bingen for his willingness to serve on my
committee and provide input on this thesis.
I would also like to thank several organizations for funding this research project. First, I
would like to express my gratitude towards the Bill and Melinda Gates Foundation and the
United States Agency for International Development (USAID) - Mali for funding the project that
collected the data used in this thesis. In addition, I am grateful for the assistantship funding that I
received from the Food Security III Cooperative Agreement between MSU and USAID, through
the Bureau for Food Security, Office of Agriculture, Research, and Technology.
Beyond my thesis committee and funding sources, I would also like to thank several
people who have provided immense support during my graduate career. First, I would like to
thank my friends (both on campus and back home) who have helped remind me to take the time
to enjoy life outside of graduate school. Second, I would like to thank my two brothers for all of
their support. In particular, I would like to thank Greg for putting up with my endless chatter
about this thesis, and Nathan for pushing me to finish up my revisions after I joined him in
Washington DC. Finally, I would like to express my immense gratitude towards my parents for
all of their love and support. Mom and Dad - This thesis is for you.

iii

TABLE OF CONTENTS

LIST OF TABLES ......................................................................................................................... vi
LIST OF FIGURES ..................................................................................................................... viii
1. BACKGROUND INFORMATION ON MALIAN HOUSEHOLD INCOMES AND
MOTIVATION FOR THIS STUDY .............................................................................................. 1
1.1 Purpose of Study and Research Questions............................................................................ 4
1.2 Previous Studies on Household Incomes in Mali ................................................................. 5
1.3 Limitations of the Previous Studies on Malian Household Incomes ................................... 7
1.4 Structure of Thesis ............................................................................................................. 10
2. LITERATURE REVIEW ......................................................................................................... 11
2.1 Household Income Portfolios and Livelihood Diversification .......................................... 11
2.1.1 Common Income Sources ........................................................................................... 11
2.1.1.1 Livestock Income ................................................................................................. 11
2.1.1.2 Nonfarm Income .................................................................................................. 12
2.1.1.3 Migration Remittance Income ............................................................................. 13
2.1.2 Relationship between Farm and Nonfarm Activities ................................................... 13
2.1.3 Reasons for Livelihood Diversification ...................................................................... 16
2.1.3.1 Risk Reduction ..................................................................................................... 16
2.1.3.2 Coping after a Shock ............................................................................................. 17
2.1.3.3 Seasonality ........................................................................................................... 17
2.1.3.4 Credit Market Failures ......................................................................................... 17
2.1.3.5 Asset Strategies .................................................................................................... 18
2.1.3.6 Returns from Income Activities ........................................................................... 18
2.1.4 Determinants of Household Income ........................................................................... 19
2.2 Income Inequality .............................................................................................................. 21
2.2.1 National-Level Income Inequality .............................................................................. 21
2.2.2 Community-Level Income Inequality in Rural Areas ................................................ 23
2.2.3 The Relationship between Income Sources and Community-Level Inequality.......... 23
2.2.3.1 Crop Income......................................................................................................... 23
2.2.3.2 Nonfarm Income .................................................................................................. 25
2.2.3.3 Livestock Income ................................................................................................. 26
2.2.3.4 Migration Remittance Income ............................................................................. 27
3. DATA AND INCOME DEFINITIONS ................................................................................... 28
3.1 Definition of Household .................................................................................................... 37
3.2 Definitions of Income and Income Categories .................................................................. 37
3.3 Basic Characteristics of the Surveyed Households ............................................................ 39
3.4 External Events in the Koutiala and Tominian Zones that may have Impacted Household
Income Portfolios ...................................................................................................................... 42
4. HOUSEHOLD INCOME PROFILES FOR TOMINIAN AND KOUTIALA........................ 43
iv

4.1 Methodology ...................................................................................................................... 43
4.2 Average Total per Capita Income Levels .......................................................................... 44
4.3 Household Income by Source ............................................................................................. 46
4.4 Distribution of Income Across Household Types .............................................................. 53
4.4.1 Distribution of Income Across Income Quartiles ........................................................ 53
4.4.2 Distribution of Income Across Landholding Quartiles ................................................ 56
4.4.3 Distribution of Income between Households Above and Below the $1/Day/Capita
Poverty Line .......................................................................................................................... 59
4.5 Discussion of the Descriptive Statistics Analysis of Household Income Portfolios in the
Tominian and Koutiala Zones................................................................................................... 61
5. DETERMINANTS OF HOUSEHOLD INCOME .................................................................. 63
5.1 Methodology ...................................................................................................................... 63
5.2 Factors Correlated with a Higher Probability of Participation and Higher Income Levels
Earned from Livestock, Nonfarm, and Cropping Income Activities ........................................ 72
5.2.1 Livestock Income ......................................................................................................... 73
5.2.2 Nonfarm Income .......................................................................................................... 77
5.2.3 Crop Income................................................................................................................. 81
5.3 Discussion of the Results from the Econometric Analysis of Household Income
Determinants in the Tominian and Koutiala Zones .................................................................. 83
6. COMMUNITY INCOME INEQUALITY AND INCOME SOURCES ................................. 86
6.1 Methodology ...................................................................................................................... 86
6.2 Results ................................................................................................................................ 90
6.3 Discussion of the Gini Decomposition Results ................................................................. 96
7. CONCLUSIONS...................................................................................................................... 98
7.1 Limitations of this Study.................................................................................................. 101
7.2 Future Research ................................................................................................................ 102
APPENDICES ............................................................................................................................ 103
Appendix A: Method for Calculating Households Living Above and Below the $1/day Poverty
Line ......................................................................................................................................... 104
Appendix B: Real per Capita Income from Various Sources, Including F-Tests to Determine
Statistical Significance of the Income Differences found between Survey Years (in 2010 Franc
CFA) ....................................................................................................................................... 106
Appendix C: Real per Capita Income from Various Sources, Including T-Tests to Determine
Statistical Significance of the Income Differences found between Zones (in 2010 Franc CFA)
................................................................................................................................................. 107
Appendix D: Average Share of Household Income from Various Sources, Including F-Tests to
Determine Statistical Significance of the Income Differences Found Between Survey Years
................................................................................................................................................. 108
Appendix E: Average Share of Household Income from Various Sources, Including T- Tests
to Determine Statistical Significance of the Income Differences found between Zones ....... 109
BIBLIOGRAPHY ....................................................................................................................... 110

v

LIST OF TABLES

Table 1: Number of Villages in Each Village Selection Criteria Category for the Koutiala Zone
....................................................................................................................................................... 33
Table 2: Income Category Definitions .......................................................................................... 39
Table 3: Descriptive Statistics on Household Demographical Information ................................. 41
Table 4: Total per Capita Household Income by Zone and Year ................................................. 45
Table 5: Average per Capita Income Levels by Income Source among all Surveyed Households
....................................................................................................................................................... 47
Table 6: Average per Capita Income Levels for only Households that Participated in a Given
Activity ......................................................................................................................................... 50
Table 7: Activity Participation Rates ............................................................................................ 51
Table 8: Average Share of Household Income by Source ............................................................ 53
Table 9: Average Share of Household Income by Source and Income Quartile (2006/07
Cropping Season) .......................................................................................................................... 54
Table 10: Average Share of Household Income by Source and Income Quartile (2008/09
Cropping Season) .......................................................................................................................... 55
Table 11: Average Share of Household Income by Source and Income Quartile (2009/10
Cropping Season) .......................................................................................................................... 56
Table 12: Average Share of Household Income by Source and Land Quartile (2006/07 Cropping
Season) .......................................................................................................................................... 57
Table 13: Average Share of Household Income by Source and Land Quartile (2008/09 Cropping
Season) .......................................................................................................................................... 58
Table 14: Average Share of Household Income by Source and Landholding Quartile (2009/10
Cropping Season) .......................................................................................................................... 59
Table 15: Average Share of 2006/07 Income Earned by Source and by Household Poverty Status
(2010 International Dollars) .......................................................................................................... 60
Table 16: Average Share of 2008/09 Income Earned by Source and by Household Poverty
Status (2010 International Dollars) ............................................................................................... 60

vi

Table 17: Average Share of 2009/10 Income Earned by Source and by Household Poverty Status
(2010 International Dollars) .......................................................................................................... 61
Table 18: Abbreviations for Explanatory Variables Included in the Determinants of Household
Income Model ............................................................................................................................... 68
Table 19: Descriptive Statistics of Independent Variables Included In The Determinants of
Household Income Model ............................................................................................................. 69
Table 20: Determinants of Participation in Livestock Activities.................................................. 74
Table 21: Determinants of per Capita Livestock Income Levels ................................................. 76
Table 22: Determinants of Participation in Nonfarm Activities ................................................... 78
Table 23: Determinants of per Capita Nonfarm Income (Excluding Transfers) .......................... 80
Table 24: Determinants of per Capita Cropping Income Levels .................................................. 82
Table 25: Gini Coefficients for Koutiala and Tominian (2006/07) .............................................. 91
Table 26: Gini Coefficients for Koutiala and Tominian (2008/09) .............................................. 91
Table 27: Gini Coefficients for Koutiala and Tominian (2009/10) .............................................. 91
Table 28: Gini Decomposition of Household Incomes in Koutiala (2006/07) ............................. 93
Table 29: Gini Decomposition of Household Incomes in Koutiala (2008/09) ............................. 93
Table 30: Gini Decomposition of Household Incomes in Koutiala (2009/10) ............................. 94
Table 31: Gini Decomposition of Household Incomes in Tominian (2006/07) ........................... 94
Table 32: Gini Decomposition of Household Incomes in Tominian (2008/09) ........................... 95
Table 33: Gini Decomposition of Household Incomes in Tominian (2009/10) ........................... 95
Table 34: Real per Capita Income from Various Sources, Including F-Tests to Determine
Statistical Significance of the Income Differences found between Survey Years (in 2010 Franc
CFA) ........................................................................................................................................... 106
Table 35: Real per Capita Income from Various Sources, Including T-Tests to Determine
Statistical Significance of the Income Differences found between Zones (in 2010 Franc CFA) 107
Table 36: Average Share of Household Income from Various Sources, Including F-Tests to
Determine Statistical Significance of the Income Differences Found Between Survey Years .. 108
Table 37: Average Share of Household Income from Various Sources, Including T-Tests to
Determine Statistical Significance of the Income Differences found between Zones................ 109
vii

LIST OF FIGURES

Figure 1: Poverty Headcount Ratio in Mali at the National, Rural, and Urban Poverty Lines
(2006) .............................................................................................................................................. 2
Figure 2: Cotton Production Levels in Mali ................................................................................... 8
Figure 3: National Gini Coefficients by Country – 2006 ............................................................. 22
Figure 4: Map of the Cercle of Tominian in the Ségou Région and Cercle of Koutiala in the
Sikasso Région .............................................................................................................................. 29
Figure 5: Map of Surveyed Villages in Tominian and Koutiala ................................................... 34
Figure 6: The Gini Coefficient ...................................................................................................... 87

viii

1. BACKGROUND INFORMATION ON MALIAN HOUSEHOLD INCOMES AND
MOTIVATION FOR THIS STUDY
Poverty reduction is often a key goal of economic development programming pursued by
international development agencies, as well as national governments. This focus on poverty can
been seen through international initiatives, such as the United Nation’s Millennium Development
Goals which aim to halve the proportion of the world’s population suffering from extreme
poverty (defined as earning less than $1/day) between the years 1990 and 2015. While the world
as a whole is on track to meet this goal, much of this success is due to drastic reductions in
poverty levels in East Asia. Meanwhile, other regions of the world have seen only modest
improvements. For example, during the 1990-2005 time period, sub-Saharan Africa has only
seen poverty levels drop from 58% to 51% of the population (United Nations, 2010).
Due to limited successes in sub-Saharan Africa, researchers and policy makers need to
consider what types of household livelihood strategies and income activities have the greatest
potential to serve as motors of economic growth, reducing poverty while improving income
distribution in this region. A better understanding of these parameters should contribute to
improved economic development policies and programming in Africa. In particular, information
on how livelihood strategies differ between poor and non-poor households, as well as
information on whether certain income activities increase or decrease community inequality
levels, can be useful to policy makers and international development agencies as they develop
new poverty reduction initiatives.
This thesis will examine household livelihood strategies and income inequality, as it
relates to poverty, for two rural zones in Mali. Mali is an example of a country that has
experienced only modest improvements in poverty levels in recent years. As of 2006, 47% of

1

Mali’s population was living below the national poverty line (The World Bank, 2012c). In
addition, Mali is far from meeting the UN Millennium Development Goal of halving poverty by
2015 (The World Bank, 2012b). The United Nations Development Program (2011) currently
th

ranks Mali 175 out of 187 countries on its Human Development Index.
Poverty issues are particularly serious in rural areas of Mali where 67% of the country’s
population was located as of 2010 (United Nations Population Division, 2012). As of 2006, 57%
of Mali’s rural population was living below the rural poverty line. In comparison, only 26% of
Mali’s urban population was living in poverty during the same time period (See Figure 1) (The
World Bank, 2012c). In addition, it was estimated in 2010 that the incomes of 80% of Mali’s
rural population were too low to provide even a basic 2,450 kcal/day diet during the entire year,
suggesting that low incomes contribute to food insecurity and malnutrition (Boughton, Staatz, &
Dembélé, 2010). Given these statistics, programs and policies that raise rural incomes are
urgently needed in Mali.

Percentage of
population in poverty

Figure 1: Poverty Headcount Ratio in Mali at the National, Rural, and Urban Poverty Lines
(2006)
60
50
40
30
20
10
0
National Population

Rural Population

Urban Population

Source: Developed by the author using data from World Bank (2012c)
As policy makers and development agencies in Mali work to reduce rural poverty, they
need to practice caution and consider all the implications of promoting certain income activities
2

(agricultural or non-agricultural) as the pathway to poverty reduction. In particular, consideration
of which groups will likely reap the majority of benefits from the development of a given sector
or income activity is needed before the implementation of policies or programs. For example, a
study on poverty in Mozambique found that recent large investments in mining, manufacturing,
and utilities megaprojects contributed to overall economic growth for the country but did not
significantly reduced poverty levels (Cungara, Fagilde, Garrett, Uaiene, & Headey, 2011). This
example shows that while certain economic development projects may generate income in a rural
community, the project will only have a poverty-reducing effect if the beneficiaries of the
additional income were previously impoverished. If, on the other hand, only wealthier
subsections of the population gain additional income, poverty levels will likely remain
unchanged and community inequality levels will increase.
Policy makers can also use information on household income sources to identify potential
vulnerabilities that may cause a household to fall into poverty or become food insecure. For
example, households that earn income solely from rainfed crop agriculture may be more
vulnerable to droughts than households with a more diversified income portfolio that includes
both farm and nonfarm activities. On the other hand, households that earn a high percentage of
their income from nonfarm activities and that generally purchase cereals on the market would be
more vulnerable to changes in consumer food prices. As these examples show, understanding
household income sources can play an important role in predicting how shocks to a community
will likely impact poverty levels and food security.
For these reasons, understanding what types of income activities are practiced by the poor
and non-poor subsections of a country’s rural population, as well as how certain income

3

activities either increase or decrease community inequality levels, can help policy makers
improve policy and program design.

1.1 Purpose of Study and Research Questions
The purpose of this study is to explore the issues described above as they relate to two
rural, agricultural zones in southern Mali. In particular, panel household survey data from the
Tominian and Koutiala zones of Mali was analyzed to answer the following questions:
1)

What were the income levels and sources reported by households in Koutiala and

Tominian during the 2006/07, 2008/09, and 2009/10 cropping years? To what extent did
household income portfolios differ across zones, years, and income quartiles?
2)

What household and community characteristics were associated with higher levels of

income earned by households participating in the following three income categories: crops,
livestock, and nonfarm activities?
3)

What were the income inequality levels in Koutiala and Tominian during the three

cropping seasons? Does income from selected categories (nonfarm activities, transfers, cash
crops, food crops, livestock, and other agricultural activities) increase or decrease income
inequality in these communities?
To answer the first research question, a descriptive statistics analysis of reported
household incomes was performed, comparing incomes across zones, years, and income quartiles.
Standard statistical tests were performed. To examine the second research question, a Heckman
two-step model was estimated to better understand the determinants of income for cropping,
livestock, and nonfarm income. Finally, to answer the third research question on whether or not
certain income activities increase or decrease income inequality levels, regional Gini coefficients

4

were calculated and decomposed using a Gini decomposition method proposed by Lerman and
Yitzhaki (1985).

1.2 Previous Studies on Household Incomes in Mali
Two previous studies used household survey data from 1987-88 and 1994-96 from rural
Mali to examine household incomes and livelihood patterns (Debrah & Sissoko, 1990; Abdulai
& CroleRees, 2001). A third, more recent study interviewed national and local experts to
construct livelihood zones for all regions of Mali (Famine Early Warning Systems Network
2010). None of these studies used the Malian household survey data to examined issues relating
to income sources and inequality levels.
Debrah and Sissoko (1990) collected data on cash incomes (i.e., no valuation of
production produced and consumed at home) by surveying 15 households during the 1987/88
crop season in the Banamba zone (a coarse grain zone not far outside of Mali’s capital city,
Bamako). They determined that the surveyed households were generally not subsistence farmers
but rather were active participants in the local markets. In addition, households in this study were
found to have earned, on average, 20% of cash income from crops, 39% from cattle, 32% from
small stock, and 9% from nonfarm activities. Debrah and Sissoko also found that cash incomes
were used for a variety of purposes. Surveyed households reported that they sold crops to the
local market in order to raise cash to purchase grains and animal feed that were not produced by
the household. In addition, households reported selling livestock during the “hungry season”
(when food stocks are low) to purchase grains and livestock feed, as well as to pay for farm
machinery repairs, hired agricultural labor, and to purchase additional livestock. Finally, Debrah

5

and Sissoko found that incomes from nonfarm activities were often reinvested into the farm to
support various livestock and cropping activities.
The study by Abdulai and CroleRees (2001) examined the incomes of 120 households in
the Malian cotton basin (the Sikasso and Koutiala cercles) during the 1994/95 and 1995/96
growing seasons. This study found that, on average, 70% of total income came from crop
production (44% from cotton) while about 30% came from non-cropping activities. In addition,
the study found that as total household income rose, the share of income earned from food crops
declined, while the share of income earned from non-cropping, cotton, and livestock activities
rose. In addition, Abdulai and CroleRees used a conditional fixed effects logit model to identify
factors found to affect the probability of a household’s participation in cotton, livestock, and
nonfarm activities. They determined that variables representing the number of male adults in the
household, household landholdings, an interaction term of household size multiplied with
landholdings, and value of agricultural equipment were all positively correlated with a higher
probability of participation in these three activities. The landholding variable was found to have
the largest effect on participation with calculated coefficients ranging from 0.79-0.97, depending
on the income source. Finally, they concluded the study by arguing that poorer households in the
Malian cotton basin are less diversified than wealthier households because they face barriers to
entry into higher-return activities.
Finally, the authors of the Famine Early Warning Systems Network (FEWS NET) study
created livelihood profiles for all regions of Mali through a two-step process which included 1)
interviewing local, regional and national experts, and 2) organizing a national livelihood
profiling conference attended by technical staff from FEWS NET, the United Nations, and local
and international nongovernmental organizations (2010). This was the only study identified that

6

examined household incomes and livelihood strategies in both the Koutiala and Tominian areas
of Mali. The FEWS NET study classified the Tominian zone as part of the West and Central
Rainfed Millet/Sorghum livelihood zone where households generally participate in rainfed
agriculture, as well as sedentary livestock rearing. The FEWS NET study also found that poor
households in this zone generally have fewer household members, poorer access to land,
agricultural equipment, and formal credit, and generally produce smaller quantities of cash crops
than wealthier households.
The FEWS NET study placed the Koutiala zone within the sorghum, millet, and cotton
livelihood zone. In this zone, households generally have good market access, and agricultural
credit and inputs are provided by the Banque Nationale de Developpement Agricole (BNDA).
Similar to the Tominian zone, the authors of this study argued that poor households in this region
generally have smaller families and landholdings.

1.3 Limitations of the Previous Studies on Malian Household Incomes
While these three studies revealed notable patterns relating to household incomes in Mali,
all three studies have their limitations. First, the data from the two studies using household
income data was over 15 years old, and Mali has experienced considerable changes since these
surveys were conducted. For example, the Abdulai and CroleRees study was conducted the year
after Mali’s currency, the Franc CFA, was devalued. The devaluation caused exported cotton to
become significantly more profitable, resulting in several subsequent booms in the Malian cotton
sector. As shown in Figure 2, the devaluation was followed by more variable cotton production
levels from year to year, exposing farmers to additional risks not seen before the Franc CFA
devaluation. One might expect household income portfolios to change as a result of the currency

7

devaluation, although such changes might occur over several years as it takes time for
households to adjust their fixed assets. Since Abdulai and CroleRees’ study was performed in the
two years immediately after the devaluation, their study’s results may not have fully reflected
any income portfolio changes that occurred after this policy change. In addition, recent issues
relating to the Malian cotton sector, such as low producer prices, concerns about the privatization
of the national cotton company, and institutional inefficiencies within the sector, may cause
households to change their income portfolios. As a result, repeating the analysis with more recent
data may show different household livelihood strategies than those reported in earlier studies.

Figure 2: Cotton Production Levels in Mali
700,000

Devaluation of Franc CFA

600,000

Metric Tons

500,000
400,000
300,000
200,000
100,000

1961
1963
1965
1967
1969
1971
1973
1975
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009

0

Source: Developed by author using data from FAOSTAT, a database created and managed by
the Food and Agriculture Organization of the United Nation’s Statistical Division (2012)
8

In addition, Abdulai and CroleRees’ study used a fixed effects logit model. Fixed effects
models cannot show the direct effects of time-invariant factors, such as distance from a market or
ethnic group of the household (Wooldridge, 2009). Given this constraint, a re-evaluation of this
problem using an econometric model that can take into account time-invariant variables may
reveal new information.
A limitation of the Debrah and Sissoko study (1990) is that it only surveyed 15
households and only examined cash incomes. The small sample size makes it difficult to draw
conclusions about the community and limits the usefulness of the study for policy purposes. In
addition, most economists agree that household income should include both cash and in-kind,
non-monetary income (Ellis, 2000a). In developing country situations, examining only cash
income has been found to be problematic because a considerable share of household production
is intended for home consumption and never enters the market. For this reason, studying only
cash income may miss an important component of a given household’s livelihood strategy.
Finally, the FEWS NET livelihood profiles study is considerably more recent but a
weakness of this study is that it is based primarily on interviews with national and local experts
and no household survey data was used. As a result, the quality of this study is highly dependent
on how well the experts interviewed understood the local dynamics of each region profiled.
Given the limitations of past studies, further research on household incomes in Mali using
more recent data is needed. In particular, understanding household income levels and sources can
help policy makers identify potential vulnerabilities that households in a given community face
and can help them determine how shocks (such as a drought, changes in commodity prices, etc.)
may impact household incomes and food security. In addition, an awareness of household and
community characteristics that differentiate higher income-earning households from lower
9

income-earning households can help policy makers focus their attention on particular issues and
policies that could raise incomes for the poor and reduce poverty rates. For example, if it is
shown that high income-earners generally have more education or better access to transportation
than poorer households, policy makers would perhaps want to focus their attention on improving
access to education and transportation for the poor. Finally, understanding whether inequality
levels increase or decrease as a result of certain income activities will help policy makers
understand if promoting a given income source may actual help the poor and reduce poverty
levels, or just increase the incomes of the wealthy sub-sections of the population.

1.4 Structure of Thesis
The issues discussed above will be explored in more depth in the next six chapters of this
thesis. Chapter 2 is a literature review on the topic of household livelihood strategies, household
income determinants, and income inequality. Chapter 3 presents the data used in this study, as
well as income definitions applicable to this analysis. Chapter 4 describes household income
levels and the relative importance of different income sources in these two zones of Mali. In
Chapter 5, a Heckman two-step econometric model is used to examine the determinants of per
capita household income and the extent to which they differ by income type (e.g., crop, livestock,
and nonfarm activities). Chapter 6 then uses a Gini decomposition method to better understand
the relationship between income sources and income inequality. Finally, Chapter 7 discusses the
implications of this study’s results, policy recommendations, and suggestions for future research.

10

2. LITERATURE REVIEW
In order to better understand the context of this study on household incomes and
livelihood strategies in Mali, a literature review is presented. The first section of this literature
review discusses income activities commonly found in West Africa, as well as related theories
proposed by researchers on household livelihood diversification in developing countries. This
section also includes a discussion of factors found to be correlated with agricultural and nonagricultural participation and income levels in various countries throughout the world. The
second section of this literature review discusses the relationship between income sources and
income inequality.

2.1 Household Income Portfolios and Livelihood Diversification
2.1.1 Common Income Sources
Traditionally it was thought that rural households in developing countries only
participated in agriculture, with a focus on cropping activities. However, research from various
developing countries has shown that rural households actually participate in a variety of income
activities — both on and off the farm.
2.1.1.1 Livestock Income
In addition to crops, one income source common in West Africa is livestock. In much of
rural Africa, where there are few financial and banking alternatives, livestock serves as a
relatively liquid asset and is often used as a savings mechanism (Dercon, 1998). Livestock can
also be culturally significant and can play an important role in local customs, such as in the case
of bride payments. A common hypothesis is that households use livestock as a consumption
smoothing mechanism, selling off livestock to ensure that consumption levels remain relatively

11

constant throughout the year when faced with an income shock. However, this idea has been
questioned by Fafchamps, Udry, and Czukas (1998) in a study of rural households in Burkina
Faso. This study found that households that faced the most significant income losses during a
drought reported that their primary motivation for selling livestock was to meet household
consumption needs. However, statistical analysis of these households’ reported income showed
that, at most, livestock sales only made up 30% of the income lost as a result of the drought.
Fafchamps et al. argue that although livestock income can help make up a sizable share of lost
income due to a drought, it does not make up for all of it, and therefore is unlikely to be the only
consumption smoothing mechanism employed by households in Burkina Faso.
A study of household incomes in several countries neighboring Mali, including Burkina
Faso and Senegal, used data collected between 1981-1990 to find that households in the
Sudanian agroclimatic zone (which would be similar to the climate of Mali’s Tominian and
Koutiala zones) earned, on average, 5-10 percent of total income from livestock activities. This
study also found that the share of income earned from livestock was fairly constant across
income quartiles (Reardon et al., 1993).
2.1.1.2 Nonfarm Income
Another common income source is nonfarm activities. Nonfarm income includes income
earned from non-agricultural rural wage employment, self-employment activities, land rentals,
and domestic and international migration remittances. Nonfarm income has been found to be a
significant part of household income and has been estimated to account for approximately 51%
of rural income in Asia, 34% in Africa, and 47% in Latin America (Haggblade, Hazell, &
Reardon, 2009). In the Sudanian agroclimatic zone of Burkina Faso and Senegal, local nonfarm
activities made up on average 20-27% of total income during the 1980s (Reardon et al., 1993).

12

2.1.1.3 Migration Remittance Income
Within the nonfarm income category, income from migration remittances merits further
discussion. In Africa, it has been argued that most household income diversification is not only
nonfarm in nature but also non-rural, suggesting that people are moving to urban areas to search
for income opportunities (Ellis, 2000a). In addition, there has been evidence to suggest that
migration can play an important role in household risk reduction and consumption smoothing.
For example, a study of households in rural India found that households commonly send their
daughters to other villages or regions to marry. In these agricultural communities, where income
risks are often correlated with location, the study argued that this migration spatially diversifies
Indian families’ risks. During periods of low income, remittance income could be sent between
the two locations, smoothing household consumption patterns (Rosenzweig & Stark, 1989).
Several studies have examined the role of migration remittances on household incomes in
West Africa. The Reardon et al. study (1993) of the Sudanian zone of Senegal, Niger, and
Burkina Faso found that remittance income made up, on average, a small percentage of total
income during the 1980s, ranging from 2-3%. Another study of households in northern Mali,
where poor climatic conditions cause higher agricultural risk, found that migration is extremely
common, and that the remittances received by households in this region often corresponded with
agroclimatic shocks and the death of household members (Perakis, 2011). Finally, Gubert,
Lassourd, and Mesplé-Somps (2010) estimated that international remittance income reduced
national-level poverty rates in Mali by 5-11%.
2.1.2 Relationship between Farm and Nonfarm Activities
The relationship between agricultural and non-agricultural activities/sectors in developing
countries has been discussed in previous literature (see, for example, Reardon et al., 1993). In

13

this literature, three questions are frequently explored: 1) does a vibrant agricultural sector aid in
the development of a strong non-agricultural sector or vice versa? 2) Do nonfarm income
activities lead to agricultural labor constraints within rural communities? And 3) do households
usually reinvest nonfarm income into farming activities?
The rural growth linkage approach is one model created to answer these questions. This
approach states that a community’s agricultural sector creates forwards and backwards
production linkages and expenditure/consumption linkages with the community’s nonfarm sector
(Ellis, 2000a). For example, if the cocoa industry in Nigeria experiences a boom, there will likely
be more demand for inputs, such as fertilizers, pesticides, and herbicides (backward production
linkages). In addition, employment opportunities for cocoa traders, transporters, and processors
will also increase (forward production linkages). Finally, as cocoa farmers become wealthier,
they will likely increase consumption, which will create a multiplier effect throughout the
community (expenditure/consumption linkages) (Ellis, 2000a; Delgado, Hopkins, & Kelly, 1998).
Attempts have been made to estimate this multiplier effect for various developing countries. For
each additional dollar earned on farm, researchers have estimated that an additional $0.30-1.90 is
created in the community, depending on the country and research study (Ellis, 2000a; Delgado et
al, 1998; Haggblade & Hazell, 1988). These results suggest that improvements in the agricultural
sector will lead to additional off-farm opportunities in the community.
One issue examined by Haggblade and Hazell (1988) is whether production or
consumption linkages produce greater multiplier effects. They argue that in rural Africa,
consumption linkages tend to dominate while production linkages have a much smaller effect.
They also argue that improvements in agricultural income should be focused on poorer
households, rather than larger landholders, because higher income in the hands of the poor will

14

have a greater multiplier effect in the community. Larger households, meanwhile, tend to spend
more of their money outside of the community and on high priced goods that are not generally
produced or sold by poorer subsections of the population.
The rural growth model was explored in the study mentioned earlier by Reardon et al.
(1993) of rural incomes in Burkina Faso, Niger, and Senegal using data from the 1980s. This
study found that almost all income earned by the surveyed households was either directly related
to agriculture or linked via production-side linkages. This study also examined the issue by
agroclimatic zones, and found that as one moved farther north into areas with riskier agriculture,
the linkages between agricultural and non-agricultural activities became less strong. In particular,
for the Sudanian zones of Burkina Faso and Senegal, 85-98% of household incomes on average
were from agriculture or directly-linked to agricultural activities.
Past studies have also tried to understand whether or not participation in nonfarm income
activities reduces liquidity constraints and leads to more agricultural investments and input
expenditures. Studies using household surveys from Vietnam, Bulgaria, and Nigeria found
evidence which indicates that the presence of nonfarm income increases household expenditures
on agricultural inputs (Hertz, 2009; Oseni & Winters, 2009; Stampini & Davis, 2009). In
addition, a study of a Dogon community in Mali found that migration remittances were often
used to purchase farm machinery, livestock, and to pay for hired agricultural labor (David, 1995).
Finally, another issue commonly discussed in the literature is whether or not participation
in nonfarm activities creates farm labor constraints, leading to a reduction of agricultural
productivity. A case study of a Dogon community in Mali found evidence to support this idea
(David, 1995). This case study found that the interviewed households stated that migration,

15

particularly by young adults, to urban areas created a labor gap, making the cultivation of family
fields in rural Mali more difficult.
2.1.3 Reasons for Livelihood Diversification
Frank Ellis defines rural livelihood diversification as “the process by which rural
households construct an increasingly diverse portfolio of activities and assets in order to survive
and to improve their standard of living” (Ellis, 2000a, p. 15) The following six reasons are often
given for why households might diversify their incomes:
2.1.3.1 Risk Reduction
It has been argued that risk averse households prefer lower incomes with lower risk
to higher incomes with higher risk. One way in which a risk averse household might reduce
risk is through income diversification into activities that are not positively correlated with
each other (Ellis, 2000a; Ellis, 2000b; Reardon et al., 2000; Reardon et al., 1992). For
example, a Malian household that is currently only growing coarse grains might have the
opportunity to diversify into cotton (a cash crop) or send a few household members to
Mali’s capital, Bamako, to seek employment. Coarse grain production and cotton production
levels can be highly correlated. In other words, an agroclimatic shock, such as a drought,
would reduce both crops’ production levels at the same time. Nonfarm income earned in
Bamako, on the other hand, will likely be uncorrelated with coarse grain production. As a
result, income diversification through migration may reduce risk more than diversification
into cash crops. Finally, researchers have stressed that diversification for risk reduction
purposes is considered ex-ante and happens before an income shock, such as a drought,
occurs (Ellis, 2000a; Ellis, 2000b; Reardon, Taylor, Stamoulis, Lanjouw, & Balisacan,

16

2000). Risk reduction strategies differ from coping strategies in response to income shocks,
which will be discussed in the next section.
2.1.3.2 Coping after a Shock
Diversification can also be ex-post and occur after an income shock. In this situation,
households are forced into other activities for survival purposes because income from one
activity is not sufficient to live on (Ellis, 2000a; Ellis, 2000b; Reardon et al. 2006). For
example, a study of households in the Lacustre zone in Mali identified migration, livestock
sales, and receiving gifts from friends and relatives as coping strategies commonly used in
this area (Harrower & Hoddinott, 2005).
2.1.3.3 Seasonality
In most areas of West Africa, agriculture can only occur during a limited period of
the year. Consumption, on the other hand, occurs all year long. As a result, households
might attempt to smooth consumption by participating in other types of income activities
during the months when they are not busy with agriculture (Ellis, 2000a; Ellis, 2000b). For
example, Wooten’s ethnographic study of the Niamakoroni village, located near Bamako in
Mali, found that during the short rainy season, household members were generally too busy
with agricultural activities to be active in nonfarm activities. However during the dry season,
household members frequently participated in various, local nonfarm activities, such as the
collection of forestry items and the production of charcoal and crafts (Wooten, 2003).
2.1.3.4 Credit Market Failures
In many rural areas of developing countries, credit is unavailable. As a result,
farmers may struggle to accumulate enough cash to purchase inputs and equipment needed
for agricultural activities. One solution to this liquidity constraint is to diversify into other

17

cash-generating activities (Ellis, 2000a; Ellis, 2000b). As mentioned earlier, research on
rural households in Vietnam, Bulgaria, and Nigeria reported that the presence of nonfarm
incomes decreases liquidity constraints facing farmers and increases household expenditures
on agricultural inputs (Hertz, 2009; Oseni & Winters, 2009; Stampini & Davis, 2009).
2.1.3.5 Asset Strategies
Households may also choose to diversify their income activities to enable them to
make investments today that will enable higher incomes tomorrow (Ellis, 2000a; Ellis,
2000b). For example, a study in Senegal found that mothers’ participation in off-farm
horticulture employment increased their children’s primary school enrollment levels
(Maertens & Verhofstadt, 2011). This finding suggests that rural household diversification
in Senegal has led to educational investments which will likely lead to higher incomes in the
future.
2.1.3.6 Returns from Income Activities
A household may diversify its income sources if the returns from other activities are
higher than the returns from the household’s current activities (Reardon et al., 2000). This
reason for diversification is often explored using the household economic model, which
compares the returns from the household’s current activities with the returns from other
potential income sources (Ellis, 2000a).
The reasons for household livelihood diversification described above are often combined
into two larger categories labeled “survival” and “choice”. If one diversifies for “survival”
reasons, the diversification is usually involuntary and may be in response to an income shock,
such as a drought, death in the family, etc. This differs from diversification for “choice” reasons
which are voluntary (Ellis, 2000a). An example of “choice” driven diversification could be

18

household diversification into a new sector in order to take advantage of high local demand for a
certain product or service.
2.1.4 Determinants of Household Income
Many studies have used household survey data from various regions of the world to
examine determinants of household income. A review of these studies reveals several factors
commonly found to be correlated with higher income levels and a higher probability of
participation in certain income sources:
Education: The importance of education is frequently discussed in past studies. For
example, studies from Mexico, Ghana, and Nicaragua found that higher levels of education
were associated with an increased probability of participation in nonfarm activities
(Yunez-Naude & Taylor, 2001; Abdulai & Delgado, 1999; Corral & Reardon, 2001). In
the study of households in Mexico, low levels of education (1-3 years) was associated with
a higher probability of participation in staple crop activities (Yunez-Naude & Taylor, 2001)
while the Nicaragua study found that primary and secondary school education decreased
the probability of farm wage employment (Corral & Reardon, 2001). This suggests that
households with no education or very low levels of education are more likely to participate
in farm activities because they do not have the skills to participate in other income
activities. However, with higher education levels, households generally diversify into
nonfarm activities because they have the necessary education to overcome nonfarm entry
barriers. One interesting finding from the Mexico study (Yunez-Naude & Taylor, 2001) is
that the education level of the household head was found to have no impact on income
levels or participation rates for any activity. Education was also found to increase nonfarm

19

income levels in studies from Peru and Burkina Faso (Escobal, 2001; Wouterse & Taylor,
2008).
Agricultural Assets: Several studies (Wouterse & Taylor, 2008; Yunez-Naude &
Taylor, 2001), including one study from Mali (Abdulai & CroleRees, 2001), identified
agricultural assets as important factors associated with a higher probability of participation
in agricultural-related income sources. Types of agricultural assets found to be positive and
statistically significant in these papers include farm size (ha), access to irrigated land, and
the value of a household’s agricultural equipment.
Local Infrastructure and Market Access: Market access, road access, and the state of
local infrastructure were found to be positively correlated with the probability of
participating in nonfarm activities in several studies of households living in Burkina Faso,
Ghana, and Nicaragua (Wouterse & Taylor, 2008; Corral & Reardon, 2001; Abdulai &
Delgado, 1999). However, the Nicaraguan study found the opposite relationship for farm
income to be true (i.e. road access was negatively correlated with income from farm
activities) (Corral & Reardon, 2001).
Other Household Characteristics: Two studies found that a household member’s age
was positively correlated with income from non-agricultural activities, at least up until a
certain age (Corral & Reardon, 2001; Olale & Hensen, 2012). In addition, the relationship
between family size and agricultural income was examined by several studies although the
direction of this relationship remains unclear. One study of households in Mexico found
that larger family sizes decreased farm income (Yunez-Naude & Taylor, 2001). However,
a different study of households in Burkina Faso found that larger family sizes were
associated with higher levels of staple crop income (Wouterse & Taylor, 2008). Finally,

20

access to credit was identified as an important factor relating to both farm and off-farm
self-employment income activities (Escobal, 2001).

2.2 Income Inequality
In addition to examining household livelihood strategies in Mali, this thesis will also
examine income inequality issues. Income inequality can be examined at either the national or
local level, and in this next section, literature on inequality at both levels was reviewed. In
addition, there is a discussion of the literature relating to how income from certain sources (cash
crops, livestock, nonfarm, remittances, etc.) either increases or decreases community income
inequality levels.
2.2.1 National-Level Income Inequality
One measure commonly used to determine income inequality levels is the Gini
coefficient. The Gini coefficient is based on the Lorenz curve and is normally a value between
zero and one. A value of zero represents perfect income equality and a value of one represents
perfect inequality (see Chapter 6 for additional information). The World Bank estimated Mali’s
national Gini coefficient to be 0.39 in 2006. As shown by Figure 3, this value suggests that
Mali’s income inequality level is similar to the levels found in most countries in Western,
Eastern, and Central Africa. It is higher than those found in many countries in East Europe, and
lower than those found in many countries in Latin America and Southern Africa (The World
Bank, 2012a).

21

Country

Figure 3: National Gini Coefficients by Country – 2006
South Africa
Colombia
Honduras
Brazil
Bolivia
Panama
Paraguay
Zambia
Ecuador
Dominican Republic
Chile
Peru
Costa Rica
Argentina
Uruguay
El Salvador
Venezuela, RB
Philippines
Macedonia, FYR
Ghana
Uganda
Russian Federation
Georgia
Turkey
Mali
Kyrgyz Republic
Moldova
Vietnam
Togo
Poland
Armenia
Pakistan
Romania
Kazakhstan
Serbia
Montenegro
Belarus
Slovak Republic
0

10

20

30
40
50
Gini Coefficient

Source: Graph created by author using data from the World Bank (2012a)
22

60

70

80

2.2.2 Community-Level Income Inequality in Rural Areas
Rural community income inequality levels are most relevant to this study on household
livelihood strategies and income inequality in Mali. Some have argued that in Africa, rural
communities are generally “poor but equal” (Haggblade & Hazell, 1988). In addition, evidence
from certain household surveys supports this statement. For example, a study of several rural
communities in Burkina Faso calculated Gini coefficients ranging from 0.25-0.34, which
suggests relative equality within these rural communities (Reardon et al., 1992). Using the
2006/07 round of income data from the same data set analyzed for this thesis, Samake et al.
(2008) reported Gini coefficients of 0.36 and 0.30 for Tominian and Koutiala, respectively.
2.2.3 The Relationship between Income Sources and Community-Level Inequality
Many studies have examined the relationship between income from certain activities
(agriculture, livestock, nonfarm, and migration remittances) and community-level income
inequality levels. In this next section, literature relating to this topic is reviewed.
2.2.3.1 Crop Income
By using the rural growth linkage approach discussed earlier, some have argued that if
agricultural income improvements are focused on smaller/poorer farmers, agricultural activities
can be inequality decreasing for the community (see, for example, Haggblade & Hazell, 1988).
The assumption behind this argument is that smaller farmers are more likely to spend their
money locally, creating positive production and consumption linkages in the community. These
local linkages will create additional nonfarm opportunities for the rural poor and will reduce
income inequality levels. The argument also states that if increases in agricultural income are
focused on larger farms, community-level inequality will increase. This is due to the fact that

23

larger farmers will likely demand higher-priced inputs and consumer products that are not
produced locally in the community. As a result, the local income multiplier effect will be smaller.
Other researchers who have examined the effect of agricultural incomes on income
inequality levels have focused on cash crops. Most of these researchers argued that cash crops
are income inequality increasing. For example, a study of Pakistani households found that
income from sugar cane, a major cash crop in that country, contributed to income inequality
(Adams & He, 1995). In addition, Maxwell and Fernando (1989) argued that cash crops are
generally income inequality increasing for several reasons. First, early adopters tend to be from
favored groups (e.g. larger farmers, men), and as a result of being early adopters, these groups
tend to financially benefit more in the short-run compared to late-adopters. Second, they argue
that government policies promoting cash crops tend to benefit larger farmers, augmenting
inequality. Finally, Maxwell and Fernando state that a focus on cash crops may lead to land
tenure issues within a country. For example, if a household loses access to some or all of its land
as a result of a national policy to increase cash crop production, this can be inequality increasing.
In general, studies have found that land access is an important factor when it comes to
whether agricultural activities are inequality increasing or decreasing. For example, studies from
both Pakistan and Egypt, where land ownership is highly unequal, found that cropping income
contributed a sizable share of overall income inequality (35-45%) in both countries (Adams &
He, 1995; Adams, 2002). On the other hand, a study of households in three zones of Burkina
Faso, where land distribution is relatively equal, found that crop income was inequality
decreasing (Reardon & Taylor, 1996). Since land distribution in Mali is likely to be similar to
that found in Burkina Faso, I hypothesize that this thesis will find crop income to be inequality
decreasing.

24

2.2.3.2 Nonfarm Income
The relationship between the share of total household income earned from nonfarm
activities and total wealth has been found to vary depending on the region of the world. In Latin
America and Asia, this relationship is often found to be negative and linear, or U-shaped. In
these areas, high labor-capital ratio jobs with low entry barriers are widely available providing
the poor with easy access to nonfarm activities. In addition, these areas often have a high number
of poor, landless households due to highly unequal land distribution. Since agriculture
opportunities are limited for these households, they move into nonfarm activities while wealthier
households with land access remain in agriculture. Other characteristics of these regions are that
they generally have a strong agricultural sector, high population density, easy market access, and
good infrastructure (Reardon et al., 2000).
Unlike Latin America and some parts of Asia, the relationship between the share of
household income earned from nonfarm activities and total wealth has been found to be positive
and linear in much of Africa. Relatively equal land distribution is common in much of Africa so
there are few landless households. In addition, these areas usually only have limited nonfarm
employment opportunities with low entry barriers in which poor households can participate. As a
result, the poor generally remain in agriculture while the wealthy overcome nonfarm entry
barriers and diversify into high-return, nonfarm activities (Reardon et al., 2000).
In their literature review of research from Africa, Latin America, and Asia, Reardon et al.
(2000a) argue that there is an important relationship between land access and nonfarm income
activities. In particular, they state that “inequality in access to scarce land translates into
inequality in non-farm employment opportunities because agricultural cash incomes, use of land
as collateral for credit, and the confounding of land wealth and political pull are all determinants

25

of non-farm business starts” (p. 282). However, land can only be a useful source of collateral if
active land markets exist, social norms allow for foreclosure of land, and formal lending sources
are available; in Africa, this is often not the case (Atwood, 1990). This is confirmed by evidence
from Kenya, where land registration programs have been implemented, yet land is not generally
used for collateral (Cotula, Toulmin, & Hess, 2004). No literature on Mali was identified that
found land being used as collateral for credit.
Other authors who have tried to determine whether nonfarm income has an inequality
increasing or decreasing effect have broken down the nonfarm income categories into more
specific sub-categories. For example, Adams and He (1995) found that income from government
employment was inequality increasing while income from unskilled labor employment was
inequality decreasing. They also concluded that the relationship between self-employment
activities and income inequality was unclear. Adams and He state that the higher costs involved
with entering government employment cause this activity category to be income inequality
increasing while the other categories are not.
2.2.3.3 Livestock Income
Another source of income that might influence community inequality levels is livestock
income, although the results from previous studies have been mixed. In Adams and He’s study of
Pakistani households (1995) and Adams’ study of Egyptian households (2002), livestock income
was found to have a small income inequality decreasing effect. Reardon and Taylor (1996)
found in Burkina Faso that livestock income increased inequality slightly in the Sahelian zone
and decreased it slightly in the Guinean zone. A common finding across all of these studies is
that livestock income has a very minimal effect on income inequality when compared to other

26

income sources because 1) livestock income generally makes up a very small share of total
income and 2) the correlation between livestock income and total income is small.
2.2.3.4 Migration Remittance Income
The relationship between migration remittances and income inequality levels has been
examined in several previous studies. The Adams and He study (1995) found that internal
migration remittances were important to poor households and had an income inequality
decreasing effect. However, this study also found that international migration remittances had an
income inequality increasing effect. This is due to the fact that international travel, as well as
visa applications, is very expensive in Pakistan and therefore, poorer households were less able
to find the resources to migrate internationally. Results on the inequality effects of migration
remittances in West Africa have been mixed. Reardon and Taylor (1996) reported that
remittances had a slight inequality increasing effect on communities. On the other hand, a study
in Mali found that international remittances were inequality decreasing (Gubert, Lassourd, &
Mesple-Somps, 2010).

27

3. DATA AND INCOME DEFINITIONS
This analysis used household survey data that was collected in two rainfed agricultural
1

zones of Mali (Cercle of Tominian in the Ségou région and Cercle of Koutiala in the Sikasso
région) during the 2006/07, 2008/09, and 2009/10 growing seasons. The first year of the survey
was funded by the World Bank under its RuralStruc program and was conducted by a consortium
formed by IER (Institut d’Economie Rurale du Mali), CIRAD (Centre de Coopération
Internationale en Recherche Agronomique pour le Développement), and Michigan State
University. The purpose of the World Bank’s RuralStruc program was to study how seven
countries throughout the world (Kenya, Madagascar, Mali, Morocco, Mexico, Nicaragua, and
Senegal) responded to recent economic integration and liberalization policies. After the
RuralStruc survey was completed in 2006/07, Michigan State University and IER received
funding through USAID (US Agency of International Development) and the Bill and Melinda
Gates Foundation to follow-up with the same households during the 2008/09 and 2009/10
growing seasons to create a three-year panel data set. The household survey covered numerous
topics including household demographics and assets, crop production and sales, livestock income
and changes in stock levels, cereal consumption, perceptions of well-being and food security,
and nonfarm income.

1

Cercle is a local administrative unit in Mali that is smaller than region. The country of Mali is
broken down into 8 régions and 49 cercles.
28

Figure 4: Map of the Cercle of Tominian in the Ségou Région and Cercle of Koutiala in the Sikasso Région

TOMINIAN
KOUTIALA
Design: Steve Longabaugh, Food Security Group, MSU
Spatial Files: FAO, cloudmadecom, CarteAdminRoutesMali
29

While both zones selected for this study are rural and are based on rainfed agricultural
systems, the Koutiala and Tominian zones differ geographically and economically in several
ways. Tominian is located within a traditional coarse grain production zone that has received
very limited public investment. The area has faced several severe droughts in the past which is
believed to have encouraged some income diversification (such as internal migration) although
incomes are still primarily focused on subsistence agriculture (Samake et al., 2008). Important
2

crops produced in the zone include millet, sorghum, peanuts, cowpeas, and fonio . Rainfall
averages in the zone are approximately 600-900 mm/year (Murekezi, 2012), placing the zone on
the border of the Sudanian and Sahelo-Sudanian agroclimatic zones.
In contrast, the zone of Koutiala is located within Mali’s traditional cotton basin and has
benefited from extensive public investments (e.g., road infrastructure, agricultural research and
extension, farmer literacy training, and cotton ginning capacity). These investments have
benefited both cotton and coarse grain production in the Koutiala zone. In addition, the stateowned cotton company (Compagnie malienne pour le développment des textiles (CMDT)) has
provided cotton farmers with input credit and a guaranteed market. Although this zone has
historically done well, low cotton prices and management problems within the CMDT have
caused the zone to struggle since 2005 (Samake et al., 2008). Rainfall in the zone is slightly
higher than in the Tominian zone (averaging 750-1000 mm/year), placing it in the Sudanian
agroclimatic zone (Murekezi, 2012).
The differences in the infrastructure and public investment levels found in these two
zones can be partially explained by historic differences in agricultural policies for the cotton and
coarse grain industries in Mali. In Koutiala, policies relating to the Malian cotton industry have
2

Fonio is an annual herbaceous plant that is traditionally grown in West Africa as a cereal
(CIRAD, 2012).
30

greatly impacted the zone. The beginning of many policies relating to the Malian cotton industry
can be traced back to the French colonial era. In 1949, France created a cotton monopoly in Mali
called the Compaigne Francaise pour le Developpement des Textiles (CFDT). The purpose of
this monopoly was to serve as the single buyer of cotton in Mali and ensure that France had easy
access to cotton. In 1974 during Mali's post independence era, the CFDT was transformed into
the public CMDT (Compagnie malienne pour le developpement des textiles) which still exists in
Mali today. In more recent years, the CMDT has served as the primary provider of inputs for
cotton farmers and the sole purchaser of their cotton production. The CMDT also assisted in the
creation of local cotton farmer associations to assist in the distribution of inputs and credit, as
well as serving as a vehicle for cotton farmers to organize and defend their rights. Finally, the
CMDT has been very active in the economic development of cotton producing areas, through the
construction of roads, wells, health centers, schools and through programs such as adult literacy
programs (Smale, Diakité, & Keita, 2011).
In contrast, the historical policies for coarse grains, which would greatly impact
households in the Tominian zone, have followed a considerably different path from that of the
cotton industry despite similar beginnings. In 1964, the government created an official, public,
grain marketing agency in Mali called the Office Malien des Produits Agricoles (OPAM).
Similar to the CMDT for cotton, OPAM was developed to act as a monopoly for grains in Mali.
However unlike the CMDT, only 15% of Mali's marketed coarse grains were sold through
OPAM and only 3-6 percent of Mali's national coarse grain production ever went through
OPAM (Dembélé & Staatz, 2002). In addition to the OPAM, the government created a state rural
development program called the Opérations de développement rural to assist farmers with
improving crop production and marketing. This institution also served as assembly agents

31

collecting and purchasing coarse grains for OPAM. Due to several difficulties within the industry,
the Malian government agreed to start the liberalization process of OPAM Mali in 1981 in
exchange for food aid from several international donors. OPAM's monopoly was eliminated and
the sector was privatized. Since privatization, course-grain farmers have struggled with reduced
access to inputs and new technologies (Staatz, Dioné, & Dembélé, 1989). For example, adoption
of new seed varieties has been low and farmers have used only limited amounts of mineral
fertilizer (Smale, Diakité, & Keita, 2011).
Another Malian government program, Operation Riz-Segou, which promotes rice
production using controlled flooding techniques has also operated in the Tominian zone for
many years (Steedman et al., 1976) although there is no indication that the villages included in
this study ever produced any significant quantities of rice or were impacted by this program.
Given the different cropping systems and related government policies affecting the
Tominian and Koutiala zones, it is hypothesized that the income portfolios of households in each
zone will be different. In particular, one might expect that household incomes will be more
diversified in the Tominian zone than in Koutiala, with a greater percentage of total income
coming from non-cropping activities. In addition, it is hypothesized that total income levels will
be higher in Koutiala than in Tominian given the higher levels of public investments, importance
of the cotton sector, and better access to agricultural inputs in the Koutiala zone.
Within the Koutiala and Tominian zones, the villages included in the household survey
were selected based on certain characteristics (Figure 5). In the Tominian zone, it was thought
that households with better market access may be able to cope with the zone’s difficult climate
and low levels of public investment better than other households. To test this theory, three
villages with easy market access and three villages with poor market access were selected to be

32

included in the survey. Easy market access was defined as having either 1) access to a good road,
2) a weekly market in the village, or 3) easy access to a neighboring village’s market. In the
Koutiala zone, the selection of villages was based on market access and land access. Land access
was taken into account because it was believed to be a growing constraint for some parts of the
zone. To test the latter hypothesis, six villages were selected on a combination of market and
land access criteria as follows:
Table 1: Number of Villages in Each Village Selection Criteria Category for the Koutiala Zone

Market
Access

Easy
Difficult
Source: Samake et al., 2008

Land Access
Average
Difficult
1
2
2
1

33

Figure 5: Map of Surveyed Villages in Tominian and Koutiala

Design: Steve Longabaugh, Food Security Group, MSU
Spatial Files: FAO

34

Limited land availability was defined as having no additional land that could be cleared
for agriculture and limited quantities of land in fallow. However, an analysis of the 2006/07
survey data showed that land access did not differ significantly across the villages selected, so
only the market access criterion was used in the present study (Samake et al., 2008).
Once the villages were selected, researchers constructed a list of family farms in each
village with assistance from each village’s head, taking care to include family farms from all
sections – geographical and social – found in the village. Once this list was created, 25-30 family
farmers were randomly drawn from each village’s list to be surveyed (Samake et al., 2008).
The survey was conducted slightly differently during the three years of the study. During
the first year (2006/07), the household head was the primary person interviewed. The survey also
required at least one woman to respond to questions about women’s incomes for a separate
questionnaire. Finally, married men who headed smaller nuclear households within the larger
farm unit were also interviewed. During the 2008/09 survey year, the data collection changed
slightly and only the household head was interviewed. Given the large size of families and the
yearlong recall of income, there is evidence that perhaps only interviewing the household head
led to the under-reporting of household incomes. In addition, there was concern about respondent
fatigue due to the long nature of the survey. As a result, the third year of the survey (2009/10)
was changed in several major ways. First, the survey was split into two separate interviews with
the first interview covering primarily information relating to agricultural production during the
rainy season, and the second interview covering livestock income, nonfarm income, and
agricultural production during the dry season. In addition, during the third round, efforts were
made to interview all household members who earned nonfarm income and all respondents were

35

asked to recall incomes earned during both the 2009/10 and 2008/09 years to compensate for the
under-reporting of income during the second round of the survey.
The strengths of this household survey are (1) that a relatively large number of
households in several different zones were interviewed, and (2) that the survey questionnaires
covered a wide range of topics (household demographics and assets, crop production and sales,
livestock income and changes in stock levels, cereal consumption, and nonfarm income), which
allows for the development of econometric models that cannot be constructed with smaller data
sets. A weakness of the survey data is that the one-year recall of earned income and the
interviewing of only certain household members each year may have led to the under-reporting
of income. A second weakness is that much of the income data was collected at the household
level and therefore total individual income levels could not be determined. This prevented any
major analysis of the gender component of total income within households. For a brief analysis
of certain types of women's incomes using data from this household survey, please refer to
"Nonfarm Income in the Tominian, Macina, and Koutiala Zones of Mali" by Lazarus and Kelly
(2012). Finally, a third weakness of the survey is that it does not include significant information
on household expenditures.
It has been argued that given the difficult nature of collecting income data in developing
countries and since large variations are typically from year to year with income data, expenditure
data can more effectively represent a household’s long-term income potential than a direct
income measure can (McGlinchy, 2006). However, using expenditure data as a proxy for income
has its own limitations. First, expenditure data can only be used to measure total income levels. It
cannot be used for detailed analyses of individual income sources. In addition, it requires
intensive data collection, such as weekly or monthly interviews with households, which can be

36

expensive. Since an analysis of individual income sources is an important component of this
paper, the lack of expenditure data is not a major problem.
During the 2006/07 cropping year, 151 family farms (or extended household units) were
surveyed in Tominian and 153 family farms were surveyed in Koutiala. Attrition rates for the
following years were low. During the 2008/09 year, 149 farms were interviewed in Tominian
and 150 were interviewed in Koutiala. Finally, during the 2009/10 year, 151 farms were
surveyed in Tominian and 150 farms were surveyed in Koutiala.

3.1 Definition of Household
One should note that unlike in the United States, the extended family farm unit in Mali is
generally made up of several smaller, nuclear household units that share in the responsibilities of
the family farm under the direction of the extended household head. The multiple, nuclear
households join together and work as one unit in cropping activities and share the same
residency, food produced on collective fields, and some of the farm’s income and assets.
However, these smaller households are also semi-autonomous in that the members of the smaller
household units can also pursue their own income-earning activities and consumption strategies
separate from the larger farm (Samake et al., 2008). When the term “household” is used in this
thesis, it is in reference to the family farm, unless it is otherwise specified.

3.2 Definitions of Income and Income Categories
As noted in Chapter 1, most economists believe that in the context of developing
countries, household income should include both cash and in-kind, non-monetary income. Nonmonetary income should be included because in many cases, household production is intended

37

for home consumption and never enters the market (Ellis, 2000a). In this paper, all efforts were
made to take into account both cash and non-cash forms of income—the latter including crop,
livestock, and non-agricultural production that was consumed by the household, as well as
changes in animal stocks. Net crop income was calculated as the sum of total crop sales plus the
value of any unsold production (valued at the average sale price received by the household for
sold production or if the household did not sell the crop, the average sale price at the village level)
minus the costs of production (including both cash and in-kind inputs but excluding family labor
costs). For livestock income, net household income was calculated as the sum of total livestock
and livestock product sales plus the value of changes in stocks (resulting from livestock
consumption, births, deaths, gifts, etc. and valued used village or zone livestock prices). Finally
nonfarm income was calculated by taking gross nonfarm income and subtracting any input costs.
Given the limitations of the survey and issues relating to missing sale, revenue, and input cost
data, the net incomes used in this report are only rough approximations in some cases.
Once the household survey data was collected, incomes were categorized into six
household income categories: cash crops, food crops, livestock, other agricultural activities,
other nonfarm activities, and transfers. Detailed definitions of these categories can be found in
Table 2. Data on income from land and capital rentals was not collected for all rounds of the
survey, and therefore was excluded from this analysis. Since previous studies have found that the
income from land rentals made up a small and unimportant share of total income in the zones of
Tominian and Koutiala (Abdulai & CroleRees, 2001; Samake et al., 2008), it was assumed that
the results of this analysis are unlikely to be significantly different if information on these
income sources had been included.

38

Table 2: Income Category Definitions
Income Category
Cash Crops
Food Crops
Livestock

Agricultural Wages

Non-Agricultural Wages
Other Nonfarm Activities

Transfers

Definition
Net income from cotton, peanuts, and fonio
Net income from all crops except cotton, peanuts, fonio.
3
Income from tree crops was included.
Net income (including the valuation of changes in stocks
and home consumption) from the purchase and sale of
animals and animal products, including eggs, milk, hides,
and leather.
Income earned from agricultural wage work on other
farms, as well as income from agricultural machinery
rentals and custom hire.
Wage income earned from non-agricultural activities
Net income from the transformation of agricultural
products, forestry and other primary sector activities,
artisan and commerce activities, and self-employment.
Income from shea nuts and butter was included in this
category.
Income from public and private transfers including food
aid, gifts, and short/long term migration remittances

3.3 Basic Characteristics of the Surveyed Households
Basic descriptive statistics of household demographics provide information about the
typical household interviewed for this study. As Table 3 shows, the average farm size ranged
from 6.71 ha in Tominian to 13.8 ha in Koutiala. Farm size per capita was also larger in the
Koutiala zone than in the Tominian zone (0.99 ha/capita versus 0.69 ha/capita, respectively).
Household size also varied across zones with an average of 12 persons per household in
Tominian and 15 in Koutiala. Ninety-nine percent of household heads were male in both zones.
The average age of a household head was 54 in Koutiala and 56 in Tominian. Household heads
3

In regions of Mali with irrigation agriculture, such as the Office du Niger, onions are an
important cash crop. However, very few households reported income from onions in the
Tominian and Koutiala zones, and the reported incomes from onion production by households
who grew this crop were very low. As a result, this study assumes that onions are grown
primarily as a food crop in Tominian and Koutiala zones. Therefore, this income source has been
placed in the food crop category.
39

reported very limited education levels: only 24% of household heads in Tominian and 15% of
heads in Koutiala had attended at least one year of school. Finally, the majority of households in
Koutiala were net sellers of cereals (61%) while only 10% were net sellers in Tominian. The null
hypothesis that the two zones’ means were the same was tested for all variables that were not
dummy variables, and in all cases, t-tests indicate that the null hypothesis could be rejected at the
5% significance level or better.

40

Table 3: Descriptive Statistics on Household Demographical Information
Tominian
Mean

Stand.
Dev.

-

-

Total Number of Present Household
Members

11.73

Age of Household Head

Koutiala
Mean

Stand.
Dev.

24

-

-

15

-

6.70

-

15.42

7.60

-

0.000

56.04

15.19

-

54.17

13.38

-

0.050

-

-

10

-

-

61

-

Farm Land Owned (Ha)

6.71

5.09

-

13.80

9.27

-

0.000

Farm Land Owned Per Capita (Ha)

0.69

0.58

0.99

0.70

-

-

1

-

-

1

1.02

0.62

-

1.19

0.65

-

0.000

2.31

1.21

-

2.15

1.07

-

0.034

Household Head Attended at Least One
1
Year of Formal Schooling

Household Was a Net Seller of
1
Cereals

1

Household Head is Female
Dependency Ratio

3

Number of Smaller, Nuclear
Households Living on the Farm

% of
Total

2

Significance of T-Test

% of
Total

0.000

Source: Survey data
1
Represents a dummy variable
2

Null hypothesis for T-test: Tominian and Koutiala’s means are the same

3

The dependency ratio is defined as the ratio of the number of present inactive household members (such as children or the elderly) to
the number of present active household members.

41

3.4 External Events in the Koutiala and Tominian Zones that may have Impacted Household
Income Portfolios
To better understand household livelihood practices in Koutiala and Tominian during the
2006/07, 2008/09, and 2009/10 cropping years, one must understand external events having
occurred during these three cropping years that may have impacted household incomes in each
zone. In the Tominian zone, rainfall levels were low during the 2009/10 cropping year (452 mm
of precipitation in 2009/10 compared to a range of 557-732 mm of annual precipitation during
the 2006 to 2009 years. In addition, the global food crisis that began in 2007 and continued into
2008 led to higher cereal prices in the zone. Since most households in this zone are net cereal
purchasers, these high cereal prices likely put pressure on households to acquire more cash
income to cover consumption needs. Finally, farmers in this zone generally did not receive
support from government programs to purchase agricultural inputs during these three years
(Lazarus & Kelly, 2012).
In the Koutiala zone, several issues relating to the cotton sector may have impacted
household incomes during the survey period. First, there were concerns that the national cotton
company, CDMT, might be privatized although no official efforts towards privatization were
made. In addition, farmers received payment for their 2007 cotton production several months late
(November 2008), which was well after the 2008 cotton planting season had begun in June 2008.
This angered farmers and lead to a boycott of cotton production in the zone (Lazarus & Kelly,
2012). This boycott can be seen in the data set used for this paper. Between the 2006/07 and
2008/09 cropping years, the number of cotton producing households dropped 44% (Murekezi,
2012). Finally, subsidized fertilizer was made available to cotton and maize farmers in the
Koutiala zone during the 2009/10 cropping year (Lazarus & Kelly, 2012).

42

4. HOUSEHOLD INCOME PROFILES FOR TOMINIAN AND KOUTIALA
4.1 Methodology
In order to better understand the income levels and sources reported by households in
Koutiala and Tominian during the 2006/07, 2008/09, and 2009/10 cropping seasons, a
descriptive analysis was performed. For this analysis, average per capita income levels were
compared across the two zones as well as across the three years of data. T-tests and F-tests were
used to determine whether incomes are similar across the zones and years. To take into account
inflation, the nominal 2006/07 and 2008/09 incomes were adjusted to real income (based on
2010 Franc CFA) using GDP deflators for Mali, as reported by the World Bank.
Income levels, participation levels, and the share of income earned from various income
categories (cash crops, food crops, livestock, agricultural wages, non-agricultural wages,
transfers, and other nonfarm activities) were also examined for all three cropping seasons. For
the 2006/07 season, the costs of production were not disaggregated by crop in the survey
questionnaire so net incomes from cash and food crops could not be separated. Therefore, the
analysis was performed on total crop income for all three years. To better understand how
income activities differ between the poor and non-poor households, the share of total income
earned from each income category was also analyzed across income and landholding quartiles, as
well as between households living below and above the $1/day poverty line. The determination
of households living above and below the $1/day/capita poverty line was made by converting
household incomes (reported in Franc CFA) into international dollars using a purchasing price
parity approach. A detailed explanation of this methodology can be found in Appendix A.

43

4.2 Average Total per Capita Income Levels
During the 2006/07, 2008/09, and 2009/10 cropping seasons, average annual real per
capita incomes ranged from 35,932 to 52,379 Franc CFA in Tominian and from 77,023 to 84,881
Franc CFA in Koutiala. As shown in Table 4 by the 5%, 25%, 75%, and 95% percentile columns,
the difference in income levels between high income earners and low income earners was
relatively large. In addition, the median income levels were consistently lower than the mean
values, suggesting that the inclusion of a few households with very high incomes increased the
mean income values.
Both mean and median incomes declined across the three periods of data for Tominian.
This finding is confirmed by an F-test that rejected the null hypothesis that incomes in Tominian
were stable across all three years of survey data. In Koutiala, the median income levels also
decline over the three survey rounds although mean incomes levels do not show this pattern. An
F-test on this zone could not reject the null hypothesis that income levels were stable across the
three survey years.

44

Table 4: Total per Capita Household Income by Zone and Year
Zone

Year

Mean

th

5 percentile

th

25 percentile

Median

th

75 percentile

th

95 percentile

Signif.*

(real FCFA/capita using 2010 as the base year)
‘06/07

25,778

42,476

62,171

114,078

‘08/09

51,922

9,655

24,290

38,026

67,551

133,316

35,932

7,742

15,586

24,625

43,269

97,453

‘06/07

84,881

20,816

51,878

74,135

110,711

169,103

‘08/09

77,023

21,327

45,200

68,207

94,826

160,158

‘09/10

Koutiala

9,993

‘09/10

Tominian

52,379

81,267

21,817

45,573

59,970

85,873

149,942

0.005

0.771

Source: Survey data
*Probability that null hypothesis could be rejected (significance column) was determined by running an F-test through SPSS’ one-way
Anova command to compare annual mean values in each zone.

45

4.3 Household Income by Source
To better understand household income portfolios and livelihood strategies for the
2006/07, 2008/09 and 2009/10 cropping seasons, individual income sources (cash crops, food
crops, cash and food crops combined, livestock, agricultural wages, non-agricultural wages,
4

transfers, and other nonfarm activities) were also examined. Table 5 shows average real per
capita income by source in Tominian and Koutiala amongst all surveyed households (including
households that reported zero income for a particular category). By studying average income
amongst all surveyed households, we gain an understanding of the importance of various income
sources at the community/zone level. The analysis of all surveyed households showed that across
the two zones, households in Koutiala consistently earned on average higher incomes than
households in Tominian in almost all income categories. These regional differences were found
to be statistically significant at the 5% level during one or more cropping seasons for cash crops,
food crops, livestock income, agricultural wages, and other nonfarm activities. The only income
category where Tominian households consistently earned more than households in Koutiala was
non-agricultural wages. This finding was statistically significant at the 5% level for the 2008/09
cropping season (Appendix C).

4

2006/07 household income from cropping activities could not be disaggregated into income
from cash and food crops. As a result, crop income for this survey year is presented in the
category "cash and food crops combined."
46

Table 5: Average per Capita Income Levels by Income Source among all Surveyed Households
Tominian
06/07

08/09

Koutiala
09/10

06/07

08/09

09/10

(FCFA/capita)*
Cash Crops

7,740

5,154

12,798

12,539

Food Crops

20,985

16,637

42,924

37,714

30,856

28,725

21,792

60,485

55,723

50,254

Livestock

-386

3,306

2,210

9,471

6,103

4,864

Ag Wages

213

1,753

156

575

572

789

Non-Ag Wages

1,065

4,288

2,496

660

1,625

1,035

Transfers

9,123

6,162

4,194

2,052

3,538

14,507

Other Nonfarm

11,512

7,688

5,084

11,645

9,462

10,257

Cash and Food Combined

Source: Survey data
*Reported in real values using 2010 as the base year.

47

An analysis of the 2008/09 and 2009/10 cropping years showed that food crops were the
most important source of income in both zones, with average incomes from this source ranging
from 16,637-20,985 Franc CFA in Tominian to 37,714-42,924 Franc CFA in Koutiala. In
addition, the large difference in total per capita income mentioned earlier between Koutiala and
Tominian appeared to be primarily due to differences in income earned from food crops.
Average per capita income earned from cash crops was considerably lower than income
earned from food crops and ranged from 7,740 Franc CFA in 2008/09 to 5,154 Franc CFA in
2009/10 for Tominian and 12,798 in 2008/09 to 12,539 Franc CFA in 2009/10 for Koutiala. In
addition, the difference in income levels between the two zones was found to be less large for
cash crops compared to food crops (ranging from 5,058-7,385 for cash crops versus 21,07721,939 for food crops, depending on the year). This suggests that, for the survey years, cotton
income in Koutiala does not, on its own, account for the large income differences between
Tominian and Koutiala.
F-tests and t-tests were also performed to determine whether per capita income levels,
including all surveyed households, were relatively stable or unstable across the three survey
years (Appendix B). For the two years where cropping income could be separated into cash and
food crop incomes (2008/09 and 2009/10), t-tests indicated that the null hypothesis that incomes
during the two years were the same could be rejected for cash crop income in Tominian.
However, the null hypothesis could not be rejected for food crops in both zones and cash crop
income in Koutiala. When cash and food crop incomes were combined and data from all three
crop seasons were included, f-tests showed that the null hypothesis that the cropping incomes
were stable could be rejected at the 5% level for both zones. Finally, F-tests showed that this null
hypothesis could be rejected for livestock income, non-agricultural wage income, and other

48

nonfarm income for the Tominian zone. In the Koutiala zone, F-tests failed to reject this null
hypothesis for all non-cropping income sources.
The income level data was also analyzed including only households that participated in
each income activity (Table 6). This analysis showed the importance of each income source for
those households that participate in the activity. As one would expect, when households
reporting zero income from a given income source were removed, the mean income levels
increase for almost all income activities, cropping seasons, and zones. The only exception is for
food crops were average incomes remain unchanged because all surveyed households reported
income from this activity.
After removing the households reporting zero income, non-agricultural wage income is
still found to be higher in the Tominian zone compared to the Koutiala zone while income from
cash crops, food crops, and livestock were found to be consistently higher in the Koutiala zone.
Results for agricultural wage, other nonfarm, and transfer income do not show that incomes from
these three sources were consistently higher in one zone during all three years cropping seasons.

49

Table 6: Average per Capita Income Levels for only Households that Participated in a Given Activity
Tominian
06/07

08/09

Koutiala
09/10

06/07

08/09

09/10

(FCFA/capita)*
Cash Crops

9,399

6,380

16,693

14,386

Food Crops

21,123

16,637

42,924

37,714

30,856

28,914

21,792

60,485

55,723

50,254

Livestock

-503

3,747

2,567

9,925

6,270

5,216

Ag Wages

5,473

12,191

840

2,254

1,996

2,247

Non-Ag Wages

32,803

29,823

26,922

16,838

10,599

8,506

Transfers

16,528

8,729

5,864

4,617

5,769

20,448

Other Nonfarm

13,637

12,126

9,842

12,459

11,634

12,341

Cash and Food Combined

Source: Survey data
*Reported in real values using 2010 as the base year

50

Table 7 presents household participation levels for each income source. Since the
household survey was focused on rural agricultural households, it is not surprising to find that
almost all surveyed households in both zones grew food crops. In addition, the majority of
households earned incomes from cash crops, livestock, transfers, and other nonfarm activities
during most or all years of the survey. Agricultural and non-agricultural wage work had the
lowest participation rates. Given the low average per capita income levels reported by
households participating in agricultural wage activities, it is not surprising that participation rates
for this activity were found to be low. However, those households that participated in nonagricultural wage work generally reported high income levels so the low participation rates
suggest that entry barriers are preventing some households from participating in this lucrative
income source.
Table 7: Activity Participation Rates
Tominian
06/07
08/09
09/10
Cash Crops
Food Crops
Cash and Food Combined
Livestock
Ag Wages
Non-Ag Wages
Transfers
Other Nonfarm
Source: Survey data

100%
77%
4%
3%
55%
84%

82%
99%
100%
88%
14%
14%
71%
63%

81%
100%
100%
86%
19%
9%
72%
52%

06/07

Koutiala
08/09

09/10

100%
95%
25%
4%
44%
93%

77%
100%
100%
97%
29%
15%
61%
81%

87%
100%
100%
93%
35%
12%
71%
83%

Table 8 shows the average share of total household income earned from various income
activities during the three cropping seasons. This analysis showed that food crop income made
up the majority of household income for both the Tominian and Koutiala zones, confirming the
findings discussed earlier. Other smaller but notable income sources include cash crops, transfers,
51

and other nonfarm activities. F-tests showed that the null hypothesis that the share of total
income earned from each income activity was the same across the three cropping seasons could
not be rejected at the 5% level for any income source or zone except for transfer income for
Koutiala and other nonfarm income in Tominian (Appendix D). However, t-tests comparing
income levels across zones showed that Tominian households earned a statistically significant
higher share of their total income from transfers all three years and higher income from other
nonfarm activities in 2006/07. Households in Koutiala, on the other hand, earned statistically
significantly higher share of their income from food crops in 2008/09 (Appendix E).
Given the focus on cotton in the Koutiala zone, one might have expected to find cash
crops representing a larger share of total income than is shown by this analysis. In fact for
2008/09 and 2009/10, if all off-farm income sources (agricultural wages, non-agricultural wages,
transfers, and other nonfarm) were combined into one larger off-farm income category, this
analysis showed that the average share of total income earned from off-farm activities was
actually higher than the share of total income earned from cash crops in both zones. However,
this finding may be due to the 2008 cotton boycott in Koutiala and may not hold true during a
typical production year. Since the crop income from 2006/07 could not be disaggregated into
separate cash and food crop income categories, it is difficult to say what percentage of household
income came from cash crops during this year prior to the cotton boycott.

52

Table 8: Average Share of Household Income by Source

-

Tominian
2008/09
15%
49%

2009/10
14%
53%

62%

64%

0%
0%
1%
16%
21%

4%
1%
4%
12%
15%

2006/07
Cash Crops
Food Crops
Cash and Food
Crops Combined
Livestock
Ag Wages
Non-Ag Wages
Transfers
Other Nonfarm
Source: Survey Data

-

Koutiala
2008/09
14%
59%

2009/10
16%
57%

67%

73%

73%

73%

2%
1%
3%
13%
13%

8%
1%
1%
2%
15%

6%
1%
2%
5%
13%

5%
1%
1%
6%
14%

2006/07

4.4 Distribution of Income Across Household Types
4.4.1 Distribution of Income across Income Quartiles
Tables 9, 10, and 11 show that in Tominian, households in higher income quartiles
generally earned a lower share of their total income from food crops and a higher share of their
total income from non-agricultural wage activities. These tables also show that in the Koutiala
zone, higher income earners generally received a higher share of total income from cash crop
(most likely cotton) and livestock activities, and a lower share from agricultural wages. These
results were all confirmed to be statistically significant at the 10% level or better using F-tests for
at least two out of the three years of survey data.

53

Table 9: Average Share of Household Income by Source and Income Quartile (2006/07
Cropping Season)
Tominian

Koutiala

1st

2nd

3rd

4th

Signif
*

1st

2nd

3rd

4th

Signif
*

70%

63%

63%

52%

0.785

77%

73%

75%

68%

0.738

Livestock

a

-5%

-2%

-1%

6%

0.950

-3%

9%

10%

15%

0.114

Ag Wages

0%

1%

1%

0%

0.328

2%

1%

0%

0%

0.045

Non-Ag
Wages

0%

1%

0%

4%

0.091

0%

2%

1%

0%

0.327

Transfers

16%

15%

18%

13%

0.881

2%

2%

3%

3%

0.935

Other
Nonfarm

19%

22%

19%

24%

0.769

21%

12%

11%

14%

0.077

Quartile
Crops

Source: Survey data
*Probability that null hypothesis could be rejected (indicated by the significance column) was
determined by running an F-test through SPSS’ one-way Anova command.
a
In some cases, the average share of household income earned from livestock activities was
negative, which signifies that households were making livestock investments.

54

Table 10: Average Share of Household Income by Source and Income Quartile (2008/09
Cropping Season)
Tominian

Koutiala

1st

2nd

3rd

4th

Signif
*

Cash
Crops

14%

14%

20%

12%

0.134

Food
Crops

60%

53%

43%

38%

Livestock

a

-3%

3%

9%

Ag Wages

1%

0%

0

Transfers
Other
Nonfarm

Quartile

Non-Ag
Wages

2nd

3rd

4th

Signif
*

8%

14%

15%

21%

0.059

0.008

67%

59%

55%

55%

0.164

8%

0.280

0%

6%

10%

7%

0.088

0%

2%

0.455

3%

1%

1%

0%

0.026

1%

2%

14%

0.000

2%

2%

2%

2%

0.999

14%

13%

10%

10%

0.669

7%

3%

3%

4%

0.195

15%

15%

16%

16%

0.985

14%

15%

13%

10%

0.645

1st

Source: Survey data
*Probability that null hypothesis could be rejected (indicated by the significance column) was
determined by running an F-test through SPSS’ one-way Anova command.
a
In some cases, the average share of household income earned from livestock activities was
negative, which signifies that households were making livestock investments.

55

Table 11: Average Share of Household Income by Source and Income Quartile (2009/10
Cropping Season)
Tominian

Koutiala

1st

2nd

3rd

4th

Signif
*

1st

2nd

3rd

4th

Signif
*

Cash
Crops

11%

15%

13%

17%

0.359

10%

16%

17%

22%

0.014

Food
Crops

69%

57%

49%

39%

0.000

63%

62%

60%

42%

0.000

Livestock

a

-10%

5%

4%

10%

0.004

2%

5%

3%

11%

0.003

Ag Wages

1%

1%

1%

0%

0.243

3%

1%

1%

0%

0.008

Non-Ag
Wages

0%

2%

2%

9%

0.011

1%

1%

0%

3%

0.123

Transfers

17%

12%

18%

6%

0.033

8%

3%

6%

5%

0.383

Other
Nonfarm

12%

8%

13%

20%

0.064

14%

12%

12%

16%

0.558

Quartile

Source: Survey data
*Probability that null hypothesis could be rejected (indicated by the significance column) was
determined by running an F-test through SPSS’ one-way Anova command.
a
In some cases, the average share of household income earned from livestock activities was
negative, which signifies that households were making livestock investments.

4.4.2 Distribution of Income Across Landholding Quartiles
An analysis of the average share of household income earned from various income
activities broken down by landholding quartiles (Tables 12, 13, and 14) showed that the share of
income earned from agricultural wages declined as landholdings increased (statistically
significant at the 10% level in Tominian for the 2006/07 and 2009/10 cropping years and in
Koutiala for the 2008/09 cropping year). This may suggest that households with higher
landholdings require more labor to farm the additional land and therefore, cannot spare
household members to go work on other farms in the community. Another possible explanation
of this result is that households with higher landholdings are able to produce more food on-farm
56

and therefore, do not need household members to work on other farms in order to earn money to
cover food production shortfalls. For the 2009/10 cropping year, the share of income from
transfers also declined as landholdings increased (statistically significant in Tominian at 0.002
and in Koutiala at 0.009) for likely the same reasons as described above. The share of total
income from transfers, which was mostly made up of short-term and long-term remittance
income, was found to be higher across all landholding quartiles in Tominian as compared to
Koutiala. This may be due partially to stark differences in average farm size between the two
zones. For example in 2009/10, farms in Tominian averaged 5.73 ha while farms in Koutiala
averaged 14.12 ha.

Table 12: Average Share of Household Income by Source and Land Quartile (2006/07 Cropping
Season)
Tominian

Koutiala

1st

2nd

3rd

4th

Signif
*

1st

2nd

3rd

4th

Signif
*

62%

69%

58%

60%

0.926

66%

85%

72%

70%

0.146

Livestock

a

-6%

-9%

3%

9%

0.787

11%

-2%

12%

11%

0.200

Ag Wages

2%

0%

0%

0%

0.044

2%

2%

0%

1%

0.175

Non-Ag
Wages

0%

0%

3%

2%

0.266

1%

0%

1%

0%

0.729

Transfers

17%

18%

15%

12%

0.760

3%

3%

2%

2%

0.671

Other
Nonfarm

25%

22%

20%

17%

0.536

17%

12%

13%

17%

0.556

Quartile
Crops

Source: Survey data
*Probability that null hypothesis could be rejected (indicated by the significance column) was
determined by running an F-test through SPSS’ one-way Anova command.
a
In some cases, the average share of household income earned from livestock activities was
negative, which signifies that households were making livestock investments.

57

Table 13: Average Share of Household Income by Source and Land Quartile (2008/09 Cropping
Season)
Tominian

Koutiala

1st

2nd

3rd

4th

Signif
*

1st

2nd

3rd

4th

Signif
*

Cash
Crops

12%

16%

17%

16%

0.502

10%

10%

19%

18%

0.120

Food
Crops

46%

50%

48%

50%

0.928

64%

59%

59%

54%

0.530

Livestock

a

3%

2%

7%

5%

0.882

1%

9%

6%

6%

0.338

Ag Wages

0%

2%

0%

1%

0.518

2%

1%

1%

1%

0.097

Non-Ag
Wages

2%

7%

2%

6%

0.217

0%

4%

2%

3%

0.216

Transfers

16%

12%

12%

8%

0.327

4%

6%

5%

3%

0.360

Other
Nonfarm

21%

13%

14%

14%

0.188

18%

11%

8%

15%

0.049

Quartile

Source: Survey data
*Probability that null hypothesis could be rejected (indicated by the significance column) was
determined by running an F-test through SPSS’ one-way Anova command.
a
In some cases, the average share of household income earned from livestock activities was
negative, which signifies that households were making livestock investments.

58

Table 14: Average Share of Household Income by Source and Landholding Quartile1 (2009/10
Cropping Season)
Tominian
Quartile

1st

Koutiala

2nd

3rd

4th

Signif
*

1st

2nd

3rd

4th

Signif
*

Cash
Crops

6%

11%

16%

21%

0.000

11%

16%

18%

19%

0.207

Food
Crops

51%

53%

54%

55%

0.951

55%

61%

57%

54%

0.536

Livestock

a

3%

6%

1%

-1%

0.645

4%

7%

3%

7%

0.264

Ag Wages

2%

0%

1%

0%

0.002

2%

2%

1%

0%

0.175

Non-Ag
Wages

3%

1%

4%

5%

0.413

2%

0%

1%

2%

0.462

Transfers

25%

15%

9%

8%

0.002

10%

2%

4%

6%

0.009

Other
Nonfarm

9%

14%

16%

12%

0.516

15%

12%

16%

11%

0.352

Source: Survey data
*Probability that null hypothesis could be rejected (indicated by the significance column) was
determined by running an F-test through SPSS’ one-way Anova command.
a
In some cases, the average share of household income earned from livestock activities was
negative, which signifies that households were making livestock investments.
4.4.3 Distribution of Income Between Households Above and Below the $1/Day/Capita Poverty
Line
Tables 15, 16, and 17 show the average share of total income earned from various income
sources with households divided by whether they earned over a dollar a day per capita,
calculated using a purchasing price parity approach described in Appendix A. While t-tests
suggested that for certain income sources during certain years, the percentages were statistically
different when comparing households earning above and below the $1 poverty line, there was
little consistency in the findings when compared across all three rounds of survey data. Therefore,
no major conclusions were drawn from this analysis.

59

Table 15: Average Share of 2006/07 Income Earned by Source and by Household Poverty Status
(2010 International Dollars)
Less
$1/Day
Crops
68%
Livestock
2%
Ag Wages
1%
Non-Ag Wages
1%
Other Nonfarm
18%
Transfers
10%
Total # HH
257
Source: Survey data

Over
Significance
$1/Day
(two tailed)
64%
0.618
13%
0.293
0%
0.238
1%
0.800
18%
0.970
5%
0.080
50

Table 16: Average Share of 2008/09 Income Earned by Source and by Household Poverty Status
(2010 International Dollars)

Cash Crops
Food Crops
Livestock
Ag Wages
Non-Ag Wages
Other Nonfarm
Transfers

Less
$1/Day
14%
55%
5%
1%
3%
14%
8%

Total # HH
Source: Survey data

260

Over
Significance
$1/Day
(two tailed)
17%
0.426
47%
0.101
7%
0.534
2%
0.290
7%
0.020
12%
0.482
8%
0.941
43

60

Table 17: Average Share of 2009/10 Income Earned by Source and by Household Poverty Status
(2010 International Dollars)

Cash Crops
Food Crops
Livestock
Ag Wages
Non-Ag Wages
Other Nonfarm
Transfers

Less
$1/Day
14%
56%
3%
1%
2%
10%
13%

Total # HH
Source: Survey data

Over
Significance
$1/Day
(two tailed)
20%
0.086
38%
0.001
12%
0.036
1%
0.508
2%
0.946
9%
0.880
18%
0.150

276

23

4.5 Discussion of the Descriptive Statistics Analysis of Household Income Portfolios in the
Tominian and Koutiala Zones
In conclusion, the descriptive statistics presented above relating to the three cropping
years studied in this thesis showed that per capita income levels in both the Tominian and
Koutiala zones were low and were highly dependent on crop income (particularly food crop
income). Despite a strong focus on the cotton sector in Koutiala, income from cash crops was
found to be considerably smaller than income from food crops for households in this zone for
both survey rounds where cash and food crop incomes could be separated (2008/09 and 2009/10).
In addition, cash crop income was only a slightly larger share of total household income than
nonfarm income for these two years. This result may be due to the 2008 cotton boycott so further
research using data from a typical cropping year is needed to confirm or refute this finding.
Another finding from this analysis is that per capita incomes in Koutiala were about 1.5-2
times larger than per capita incomes in Tominian. In addition, this difference appeared to be
primarily due to differences in the levels of food crop income between the two zones. Larger
61

farm sizes and access to agricultural credit and inputs in Koutiala may partially explain this
finding.
These descriptive statistics also showed that wealthier households in Tominian and
Koutiala generally had different income portfolios than poorer households. As households in
both zones become wealthier, the share of income from food crops generally declined,
suggesting that wealthier households were moving away from subsistence agriculture to other
income activities. In Koutiala, wealthier households often earned a higher share of their income
from cash crops and livestock activities as compared to poorer households. In Tominian,
wealthier households tended to earn a larger share of income from non-agricultural wage
activities. Despite high income levels for non-agricultural wage activities, participation rates
were found to be low for this activity. This suggests that entry barriers were preventing poorer
households from participating in this income source. These issues are explored in more detail in
the next chapter of this paper.

62

5

DETERMINANTS OF HOUSEHOLD INCOME

5.1 Methodology
To better understand factors associated with higher levels of household income, an
econometric model of household income determinants was developed. For the econometric
model section of this thesis, data on household income per capita was combined into three larger
categories: cropping income (includes income from both food and cash crop activities), livestock
income, and nonfarm income (includes income from agricultural wages, non-agricultural wages,
and other nonfarm activities). For this analysis, nonfarm income did not include income from
transfers, such as migration remittance income, because the focus of this analysis was on locallyearned income. Attempts were made to model transfer income separately but the model
presented later in this chapter did not fit the transfer income data well. It is possible that transfer
income is less impacted by the characteristics of the transfer-receiving household and is more
influenced by factors not covered by this household survey, such as individual characteristics of
the migrant or the household’s domestic and international social networks. Additional research
on transfer income is needed to confirm this hypothesis and is beyond the scoop of this study.
The determinants of household income model used in this analysis needed to address two
issues relating to the Malian household survey data set: 1) the panel or longitudinal nature of the
data, and 2) potential selectivity bias.
Panel Data
Data sets that have multiple observations for the same individual, household, country, etc.
over a given period of time are considered panel data set (Wooldridge, 2010). An issue with
panel data is that the observations are not necessarily independently distributed. Factors that
impact a given individual/household/country/etc. in one year will likely impact observations in

63

other years as well. These factors are shown by the variable c in Wooldridge’s basic panel data
set model (2010):

More precisely, the variable c represents omitted explanatory factors that impact Yt but
are constant overtime. It is also referred to as the unobserved effect (Wooldridge, 2010). For
example, in an econometric model of household income determinants, c might represent
household members’ intelligence, ability, or drive to succeed. While these factors may be
important to the model, it is difficult to collect quantitative data to measure these factors and thus,
these factors are usually omitted. In cases where the unobserved effect (c) is correlated with any
of the explanatory variables, such as when unobserved ability and intelligence are correlated with
an observed variable like education level, a pooled OLS model will result in inconsistent and
biased results (Wooldridge, 2010).
One solution to this problem is to eliminate the time constant, unobserved effect using a
differencing econometric model such as a fixed effects model (Wooldridge, 2010). A fixed
effects model was used in Abdulai and CroleRees’ study of factors affecting the probability of
participation in cotton, livestock activities, and nonfarm activities (2001). One constraint of a
fixed effects model is that while it solves the unobserved effect problem, it also drops out all
time-invariant explanatory variables in the model. In the case of modeling household income
determinants, it seems probable that time-invariant variables, such as location or road access,
could be important factors in household income. For this reason, a fixed effects model was not
used in this thesis.
Another econometric model commonly used by researchers working with panel data sets
is called the random effects model. Unlike the fixed effects model, this model can include time64

invariant explanatory variables. However, a random effects model requires the assumption that
there is zero correlation between any explanatory variable and the unobserved effect, c
(Wooldridge, 2010). In the case of household income determinants, it seems difficult to assume
that there exists zero correlation between these variables. For example, an explanatory variable
like education level is likely correlated with unobserved variables, such as ability or intelligence.
For this reason, a random effects model was not ideal for this analysis.
Given the constraints of both random effects and fixed effects models, this econometric
analysis used a correlated random effects (CRE) model, also referred to as the MundlakChamberlain device. Proposed by Mundlak (1978) and Chamberlain (1984), this is a random
effects model that includes an additional vector

. The vector

represents the average values of

each time-variant explanatory variable across the time period covered by the panel data set. A
5

correlated random effects model can therefore be written as follows :

where:
,
The benefit of a correlated random effects model is that it combines several positive
aspects of both the fixed effects and random effects models. It allows for time-invariant factors,
such as location and road access, to be included in the analysis like a random effects model. In
addition, it also solves the unobserved heterogeneity issues associated with the unobserved effect,
c, like a fixed-effect model (Ricker-Gilbert, 2011).

5

The generic presentation of the correlated random effects model presented here is a slight
variation of the model presented by Muyanga, Jayne, & Burke (2010).
65

Potential Selection Bias
In addition to issues relating to panel data, the econometric model needed to address any
potential selection bias present in the data. As the last chapter showed, not all households
participated in livestock or nonfarm income activities, and the decision to participate in these
income activities was not likely random. This raised concerns that the subset of the population
that earned income from one of these two activities may have been different from the general
population in some way which allowed them to earn higher returns or have a comparative
advantage in the activity. If this type of selectivity bias was present in a data set and the
econometric model was not corrected to take selectivity bias into account, the model’s
coefficients could be biased (Lanjouw, 2001; Yunez-Naude & Taylor, 2001).
To resolve this problem, several authors have used Heckman two-step models (also
known as a Heckit model) to evaluate the determinants of nonfarm income (see, for example,
Berdegue, Ramirez, & Reardon, 2001; Lanjouw, 2001). A Heckman model has two stages. First,
a probit regression is performed to determine the factors affecting the yes/no decision to
participate in a given activity. From the probit model, an adjustment factor correcting for
selectivity bias, called an inverse mills ratio, is calculated. The inverse mills ratio is then
included in the second stage of the analysis where typically an OLS regression of the factors
affecting income levels is performed. Mathematically, the Heckman two-step model can be
written as:

where Yi is only observed if Zi=1 in the following probit regression:

66

In this model, Yi represents the dependent variable (household income in our case), Xi and Vi
represent vectors of explanatory variables, and Wi represents the inverse mills ratio. If the OLS
regression finds the inverse mills ratio to be significant, then selectivity bias is present in the
model.
The econometric model presented in this thesis combined a Heckman two-step model and
a correlated random effects model. To do this, a probit analysis of factors impacting activity
participation was first performed for each year and from this information, each year’s inverse
mills ratio (IMR) was calculated. Then, all of the inverse mills ratios were included in a
correlated random effects model for the second stage of the model. The vector

representing

the average values of the time-variant explanatory variables were only included in the second
stage of the model. Tables 18 and 19 present the explanatory variables, as well as abbreviations
and descriptive statistics for each variable used in this econometric model.

67

Table 18: Abbreviations for Explanatory Variables Included in the Determinants of Household
Income Model
Abbreviation

Definition

DEPEND

Dependency Ratio

AGE

Age of Household Head

POP
FEMALE

Number of People in Household
Female Household Head Dummy

CE_EDU_1_6
CE_EDU_7+
EDU_MAX_1_6

Household Head Has 1-6 Years of Education
Household Head Has 7+ Years of Education
Most Educated Person (14 yrs. old +) has 1-6 Years of Education

EDU_MAX_7+

Most Educated Person (14 yrs. Old +) has 7+ Years of Education

FARM_SIZE

Land Owned (Ha)

EASY_ACCESS
TOMINIAN

Easy Road Access Dummy
Tominian Regional Dummy

ROOMS

Number of Rooms In Household Head’s Home (proxy for wealth)
in 2006/07

MOTO
CART

Household owns a Motorcycle or Motorbike Dummy
Household owns an Animal Cart Dummy

MOVED_LATE

Household Moved To Village After the Village’s Creation (proxy
for poor access to land)

DIVIDED

Parent’s Farm Was Divided (proxy for land access constraints)

6

6

The dependency ratio is defined as the ratio of the number of present inactive household
members (such as children and the elderly) to the number of present active household members.
68

Table 19: Descriptive Statistics of Independent Variables Included In The Determinants of Household Income Model

Factors
DEPEND
AGE
POP
FEMALE
CE_EDU_1_6
CE_EDU_7+
EDU_MAX_1_6
EDU_MAX_7+
FARM_SIZE
EASY_ACCESS
TOMINIAN
ROOMS
MOTO
CART
MOVED_LATE
DIVIDED

ZONES COMBINED
Stand.
Mean
Valid %
Dev.
1.10
0.64
55.11
14.34
13.56
7.39
0.8 %
16 %
4%
29 %
37 %
10.23
8.26
47 %
50 %
1.75
0.90
44 %
74 %
10 %
41 %

TOMINIAN
Stand.
Mean
Valid %
Dev.
1.02
0.62
56.04
15.19
11.73
6.70
0.9 %
19 %
6%
25 %
44 %
6.71
5.09
44 %
100 %
1.52
0.72
36 %
67 %
0%
36 %

Source: Survey data

69

KOUTIALA
Stand.
Mean
Valid %
Dev.
1.19
0.65
54.17
13.38
15.42
7.60
0.7 %
13 %
3%
33 %
30 %
13.80
9.27
49 %
0%
2.00
1.00
52 %
82 %
20 %
46 %

It is possible that some of the explanatory variables, particularly those relating to durable
goods and wealth, such as the animal cart and motorcycle ownership variables could be
endogenous in one or more of the econometric models. This is particularly true for the models
representing income sources that made up a greater share of total income, such as cropping and
nonfarm income. Endogeneity may occur if, for example, rather than ownership of a motorcycle
or animal cart leading to higher income levels, higher income levels actually cause households to
purchase animal carts or motorcycles.. Any explanatory variable representing household wealth
could also be a source of endogeneity. In this analysis, wealth was proxied by the number of
rooms reported to be in the household head’s home in 2007 as it was thought to be less
vulnerable to endogeneity issues compared to other wealth indicators. Since access to
transportation and household wealth seem like factors that could be correlated with higher
income levels, the motorcycle, animal cart, and number of rooms variables were left in the model
7

and no efforts were made to correct for endogeneity . However, if endogeneity existed in the
models, the predicted coefficients may be biased and inconsistent.
One benefit of a Heckman two-step model is that the set of variables used to explain the
yes/no decision to participate in each income activity does not need to be the same as the set of
variables that are used to explain differences in income levels among households that participate
in the activity. With Heckman two-step models, there should generally be one or more
explanatory variables in the probit analysis that are not included in the second stage of the model
to reduce the model's standard errors. The selection of these excluded variables should be based
7

Lagging these variables was tried in an effort to reduce endogeneity within the models but did
not yield better or considerably different results (i.e. the signs on coefficients and significance
levels were very similar). Given that this did not significantly change the results, the econometric
models presented in this thesis do not include lagged variables nor are any other attempts made
to correct for endogeneity.
70

on the following: (1) the variables need to impact a household’s participation decision, and (2)
the variables should not affect income levels once a household decides to participate in the
activity. If variables are not excluded from the second stage and the explanatory variables in both
stages are the same, the inverse mills ratio will be highly correlated with the explanatory
variables in the second stage. In this case, the model would still be a best linear unbiased
estimator (BLUE) but the model’s standard errors would be high. An examination of other
studies that have used Heckman two-step models to study determinants of household incomes
shows that the explanatory variables chosen to be excluded from the second stage have varied
greatly, suggesting that there is not a consensus among researchers on the best explanatory
variables to exclude. This lack of consensus is likely due to the fact that it is difficult to find
variables within household survey data sets that only impact the yes/no decision to participate in
an income activity without impacting income levels. For this analysis, a dummy variable
indicating whether or not the farm of the household head's parents was divided up among their
children was selected to be omitted from the second stage. The reason that the division of the
parents’ farm might be important for the yes/no activity participation decision is that if the farm
is divided among many children, access to farm land becomes more constraining. Therefore, the
children may have an incentive to move into other income activities. Based on the econometric
models presented in previous studies, the exclusion of other explanatory variables was also tried
but did not yield better or considerably different results (i.e. the signs on coefficients and
significance levels were very similar).
Three dummy variables representing female-headed households, households with a head
who had more than 7 years of education, and households that had moved to their village after the
village’s creation were omitted from the first step of the Heckman model (the probit analysis)

71

and were only included in the second stage. These variables were excluded from the probit
analysis because they were found to perfectly predict activity participation due to the small
number of households where the explanatory variable was equal to one. The omission of these
variables is acceptable in a Heckman model as the set of variables used to explain the yes/no
decision to participate in each income activity does not need to be the same as the set of variables
that are used to explain differences in income levels among households that participate in the
activity
As noted in Chapter 4, income levels were significantly higher in the zone of Koutiala
than in the zone of Tominian. Despite these differences, incomes from both zones were pooled
together for this econometric analysis in order to increase the total number of household
observations. To take into account differences in income levels between the two zones, a dummy
variable indicating one for households living in Tominian and zero otherwise was included in the
model. Interacting regional dummy variables with other explanatory variables was also tried but
did not yield significantly different results. As a result, no regional interaction terms were
included in the model presented in this thesis.

5.2 Factors Correlated with a Higher Probability of Participation and Higher Income Levels
Earned from Livestock, Nonfarm, and Cropping Income Activities
Factors correlated with a higher probability of participation, as well as factors correlated
with higher levels of per capita income, were examined for three common income sources:
livestock income, nonfarm income, and crop income. In all of the models, the inverse mills ratios
were found to be statistically insignificant at the 5% level (one inverse mills ratio was significant
at the 10% level for nonfarm income). This suggests limited to no selectivity bias present in the

72

models meaning that a regular correlated random effects model that does not correct for
selectivity bias could also be used. For comparison purposes, results from both a model that
corrected for selectivity bias and one that does not are presented in this paper.
2

2

The R and pseudo R values reported below were found to be relatively low and varied
from 0.073-0.393, depending on the specific model. That said, tests on the null hypothesis that
all coefficients were equal to zero (ex. the likelihood ratio chi-square test and the wald chi-square
statistic) could be rejected at the 0.001 significance level or better for all of the econometric
models.
5.2.1 Livestock Income
The results from a probit regression of factors impacting participation in livestock income
activities showed that larger farm sizes (measured in hectares) and owning an animal cart or
motorcycle were correlated with a higher probability of participating in livestock income
activities (Table 20). For these three variables, owning a animal cart or motorcycle had the
greatest effect (increasing the probability of participation by 0.09% and 0.026%, respectively)
while an additional hectare of farmland has a considerably smaller effect (only increasing the
probability by 0.005%). This model also showed that living in the Tominian zone and older
household heads were correlated with a lower likelihood of participating in livestock income,
with living in Tominian decreasing the probability by 0.034% and older household heads
decreasing the probability by 0.001%. The model also showed that the cropping year 2008/09
was correlated with a greater likelihood of participation in livestock activities.

73

Table 20: Determinants of Participation in Livestock Activities
Factors

dF/dx

a

Standard
error

c

Z-value

All Three Cropping Seasons
b
Pooled
DEPEND

-0.010

0.094

-1.49

AGE
POP
CE_EDU_1_6

-0.001
0.002
-0.008

0.005
0.014
0.191

-3.40
1.90
-0.51

EDU_MAX_1_6

-0.005

0.174

-0.37

EDU_MAX_7+

0.002

0.179

0.18

FARM_SIZE
EASY_ACCESS
TOMINIAN

0.005
-0.003
-0.034

0.020
0.155
0.182

3.22
-0.27
-2.56

ROOMS
MOTO

-0.003
0.026

0.100
0.181

-0.45
2.03

CART
DIVIDED
2009

0.090
-0.016
0.034

0.149
0.147
0.175

5.59
-1.44
3.10

0.016

0.163

1.41

2010
# of Observations
2

Chi Likelihood
Prob > chi

175.44

2

Log Likelihood
Pseudo R

901

0.000
-211.499

2

0.293

Source: Survey data
a
dF/dx represents the change in the probability of participation due to a one unit increase in the
explanatory variable
b
Calculated using dprobit in STATA
c

Z-value indicates statistical significance: >1.645 = 10% significance level, >1.96 = 5%
significance level, >2.576 = 1% significance level

74

The second stage of the Heckman two-step model showed factors associated with higher
levels of livestock income. Table 21 showed that a higher dependency ratio (a higher ratio of
inactive, dependant household members to every active household member), owning at least one
motorcycle or motorbike, and moving to the village sometime after the village’s creation were
correlated with higher levels of livestock income. Meanwhile, living in the Tominian zone and
having a larger household size were correlated with lower income levels.
The magnitude of the effect of each of these variables on livestock income varied. A one
unit increase in the dependency ratio, or owning a motorcycle or moving to the village after the
village's creation all increased per capita livestock incomes by somewhat similar levels, ranging
from 3,599 F CFA to 4,160 F CFA. Living in the Tominian zone also had a relatively large effect,
decreasing income levels by 2,906 F CFA. Meanwhile, the household size variable had the
smallest, statistically significant effect, with each additional person only decreasing per capita
income by 731 F CFA.

75

Table 21: Determinants of per Capita Livestock Income Levels
Dependent
Variable:
Livestock
income/capita
(FCFA)
Factors
DEPEND
AGE
POP
FEMALE
CE_EDU_1_6
CE_EDU_7+
EDU_MAX_1_6
EDU_MAX_7+
FARM_SIZE
EASY_ACCESS
TOMINIAN
ROOMS
MOTO
CART
MOVED_LATE
2009
2010
CONSTANT

CRE results corrected for
selectivity bias

Coef.

b

Z-value

a

-4,618

a

1,246

6,060

IMR (2009/10)

a

139

5,408

# Observations

808

808

60.57

59.53

0.001

2,042
159
284
18,955
10,187
18,237
2,131
2,882
126
1,204
1,360
611
2,223
3,050
1,882
1,264
1,338
3,421

0.00

-0.95

2

Prob > chi

2

2

0.075
Overall R
Source: Survey data
a
The abbreviation IMR stands for the inverse mills ratio
b

0.073

Z-value indicates statistical significance: >1.645 = 10% significance level, >1.96 = 5%
significance level, >2.576 = 1% significance level
76

b

Z-value

0.03

Wald-Chi
Statistic

3,867
242
-714
2,362
-3,977
8,620
1,513
4,254
79
-1,889
-3,143
-396
4,247
-2,842
3,591
633
98
7,200

Standard
Error

0.21

IMR (2008/09)

1.95
1.58
-2.55
0.13
-0.45
0.39
0.73
1.54
0.58
-1.58
-2.05
-0.70
1.85
-1.00
1.91
-0.19
-0.45
2.23

Coef.

2,066
159
287
19,007
10,211
18,331
2,133
2,896
128
1,213
1,416
614
2,245
3,134
1,886
1,524
1,646
3,571
4,875

IMR (2006/07)

4,021
251
-731
2,501
-4,601
7,094
1,556
4,451
74
-1,919
-2,906
-427
4,160
-3,141
3,599
-288
-734
7,970

Standard
Errors

CRE results with no selectivity
bias correction

1.89
1.53
-2.51
0.12
-0.39
0.47
0.71
1.48
0.63
-1.57
-2.31
-0.65
1.91
-0.93
1.91
0.50
0.07
2.10

5.2.2 Nonfarm Income
Living in the Tominian zone, having an older head of household, and the 2008/09 and
2009/10 cropping season variables were found to be negatively correlated with the probability of
participating in nonfarm activities (Table 22). Meanwhile easy market access and having a head
of household with 1-6 years of formal education was found to be correlated with a higher
probability of participating in nonfarm income activities.
As for the magnitude of the effects of these variables on the probability of activity
participation, living in a village with easy market access had a relatively large positive effect
(increasing the probability by 0.067%), while the dummy variables representing living in
Tominian zone, 2008/09 cropping year, and the 2009/10 cropping year all had relatively large
negative effects, ranging from -0.141% to 0.173%.

77

Table 22: Determinants of Participation in Nonfarm Activities
Dependent
Variable:
Participation in
Nonfarm
Activities
(Yes =1, No=0)

dF/dx

a

Standard
Error

c

Z-value

All Three Cropping Seasons
b
Pooled

DEPEND

-0.002

0.081

-0.10

AGE
POP

-0.002
-0.002

0.004
0.009

-2.18
-0.97

CE_EDU_1_6

0.062

0.159

1.72

EDU_MAX_1_6
EDU_MAX_7+

0.008
0.052

0.131
0.136

0.25
1.59

FARM_SIZE
EASY_ACCESS

0.003
0.067

0.010
0.114

1.43
2.39

-0.141
0.017
-0.019

0.128
0.070
0.121

-4.46
0.96
-0.63

0.026
0.042

0.129
0.112

0.78
1.52

2009

-0.166

0.133

-4.59

2010

-0.173

0.134

-4.73

TOMINIAN
ROOMS
MOTO
CART
DIVIDED

# of Observations
Chi2 Likelihood
Prob > chi

107.78

2

Log Likelihood
Pseudo R

901
0.000
-392.501

2

0.121

Source: Survey data
a
dF/dx represents the change in the probability of participation due to a one unit increase in the
explanatory variable
b
Calculated using dprobit in STATA
c

Z-value indicates statistical significance: >1.645 = 10% significance level, >1.96 = 5%
significance level, >2.576 = 1% significance level

78

For the second stage of the Heckman two-step model, a larger number of people within
the household and moving into a community after the village was created were correlated with
lower per capita income levels (Table 23). On the other hand, easy road access, the 2008/09 and
2009/10 cropping seasons, living in Tominian and more rooms in the household head’s home (a
proxy for wealth) were found to be correlated with higher income levels.
When one considers the magnitude of the effects of these statistically significant
variables, household size had the smallest effect with one additional person decreasing income
by 929 F CFA. Meanwhile, the 2008/09 and 2009/10 dummy variables had the greatest effect,
increasing incomes by 6,088 to 6,480 F CFA.
Note that the inverse mills ratios were only found to be significant at the 10% level or
better for the 2009/10 cropping year. For both other years, the inverse mills ratios were
insignificant, suggesting no selectivity bias.

79

Table 23: Determinants of per Capita Nonfarm Income (Excluding Transfers)
Dependent
CRE results corrected for
CRE results with no selectivity
Variable:
selectivity bias
bias correction
Nonfarm
Standard
Standard
b
b
income/capita
Coef.
Coef.
Z-value
Z-value
Errors
Errors
(FCFA)
DEPEND
-3,301
2,234
-1.48
-3,461
2,220
-1.56
AGE
-136
196
-0.70
-129
193
-0.67
POP
-929
342
-2.71
-939
342
-2.75
FEMALE
-739
21,791
-0.03
-561
21,830
-0.03
CE_EDU_1_6
6,264
10,556
0.59
8,257
10,502
0.79
CE_EDU_7+
1,866
21,171
0.09
-32
21,04
-0.00
EDU_MAX_1_6
-713
2,625
-0.27
-957
2,613
-0.37
EDU_MAX_7+
-5,772
3,629
-1.59
-5,267
3,625
-1.45
FARM_SIZE
-12
154
-0.08
26
151
0.17
EASY_ACCESS
3,912
1,708
2.29
4,521
1,632
2.77
TOMINIAN
3,913
2,207
1.77
1,990
1,842
1.08
ROOMS
1,725
819
2.11
1,774
810
2.19
MOTO
4,265
2,904
1.47
3,786
2,890
1.31
CART
-563
3,600
-0.16
-477
3,602
-0.13
MOVED_LATE
-5,204
2,696
-1.93
-5,281
2,689
-1.96
2009
6,480
2,805
2.31
4,031
1,523
2.65
2010
6,088
2,870
2.12
2,438
1,584
1.54
CONSTANT
5,613
4,690
1.20
6,583
4,543
1.45
a
10,017
-1,452
-0.14
IMR (2006/07)
a

-7,246

6,437

-1.13

a

-10,294

5,967

-1.73

IMR (2008/09)
IMR (2009/10)
# Observations

724

724

2

122.83

119.41

2

0.000

0.000

Wald-Chi Statistic
Prob > chi

2

0.167
Overall R
Source: Survey data
a
The abbreviation IMR stands for the inverse mills ratio
b

0.163

Z-value indicates statistical significance: >1.645 = 10% significance level, >1.96 = 5%
significance level, >2.576 = 1% significance level

80

5.2.3 Crop Income
For cropping income, a Heckman two-step model was not used because nearly 100% of
households reported participating in cropping activities during all three years. As a result,
selectivity bias could not be present and a correlated random effects model could be used without
any adjustments.
This correlated random effects model showed that household heads who had 1-6 years of
education and larger farm size were correlated with higher per capita income levels (Table 24).
The household head education variable, in particular, had a relatively large effect on income
levels, increasing per capita income by 22,568 F CFA. Meanwhile, living in the Tominian zone,
having a higher number of people in the household, and 2008/09 cropping year were correlated
with lower income levels. Of these variable, living in Tominian had the largest negative effect,
decreasing per capita incomes by 20,120 F CFA

81

Table 24: Determinants of per Capita Cropping Income Levels
Dependent Variable:
Cropping
income/capita (FCFA)
DEPEND
AGE

CRE results with no selectivity bias
correction
Coef.

Standard
Errors

b

Z-value

2,968
-240

2,382
192

1.25
-1.25

-1,635
-17,011
22,568

393
27,706
12,801

-4.16
-0.61
1.76

CE_EDU_7+
EDU_MAX_1_6

1,423
-2,665

25,768
2,887

0.06
-0.92

EDU_MAX_7+
FARM_SIZE

-4,948
910

3,903
178

-1.27
5.12

EASY_ACCESS
TOMINIAN
ROOMS

-1,276
-20,120
2,033

2,396
2,711
1,242

-0.53
-7.42
1.64

2,462
-600

3,067
4,049

0.80
-0.15

-569

3,865

-0.15

4,396
961

1,713
1,778

2.57
0.54

43,997

6,661

6.60

POP
FEMALE
CE_EDU_1_6

MOTO
CART
MOVED_LATE
2009
2010
CONSTANT
# Observations

901

2

393.21

2

0.000

2

0.393

Wald-Chi Statistic
Prob > chi
Overall R

Source: Survey data
a
The abbreviation IMR stands for the inverse mills ratio
b

Z-value indicates statistical significance: >1.645 = 10% significance level, >1.96 = 5%
significance level, >2.576 = 1% significance level

82

5.3 Discussion of the Results from the Econometric Analysis of Household Income Determinants
in the Tominian and Koutiala Zones
The results of these econometric models of household income determinants reveal several
important findings. First, having a larger household size was correlated with lower levels of per
capita livestock, nonfarm, and crop income. While the magnitude of this effect was generally
relatively small, this suggests that households in Tominian and Koutiala were facing diminishing
returns to additional labor. For cropping income, this finding might also be due to increasing
land constraints as households become larger. Given that Mali’s annual population growth rate
and fertility rate are high (3.0% and 6.3 total births per woman in 2010, respectively) (World
Bank, 2012c), the results of this analysis suggest that policies and public health initiatives that
reduce the size of families could increase income and reduce rural poverty levels.
Living in the Tominian zone was also found to be correlated with a lower probability of
participating in livestock and nonfarm activities as well as considerably lower income levels
among those who participated in livestock and crop incomes. However among households
participating in nonfarm income activities, living in Tominian was weakly correlated with higher
nonfarm incomes. One possible explanation for this finding may be that households in Tominian
faced entry barriers that were restricting participation in non-cropping activities. Entry barriers to
nonfarm income activities would vary depending on the activity but might include lack of access
to credit, capital, technical skills, or social networks needed for a nonfarm activity. Efforts to
reduce entry barriers into these activities could increase their participation rates. In addition,
access to agricultural inputs, research, and extension is more constrained in Tominian than in

83

Koutiala because the CMDT provides these services in Koutiala. A lack of these types of
assistance in Tominian could partially explain the lower crop income levels.
In general, wealth and the durable goods indicators were found to be positively correlated
with livestock participation and income levels, as well as nonfarm income levels, although they
were not found to be significant in crop income. This result suggests that as households become
wealthier, they may be better able to diversify their income portfolios by moving into livestock
and nonfarm activities. However, as noted earlier in this paper, the direction of the relationship
between the durable goods and income levels is not clear. As a result, it is possible that
endogeneity existed in the model and that the models' coefficients were biased and inconsistent.
Interestingly, easy road access was found to be significant only in the nonfarm income
model, suggesting that it is positively correlated with higher nonfarm income levels and higher
nonfarm income participation rates. However, easy road access was not found to be correlated
with livestock or crop income levels. This finding suggests that building roads and bridges to
connect rural communities to local and regional markets, as well as improving the condition of
road infrastructure (ex. repairing washed out roads after the rainy season) in rural areas could
increase nonfarm incomes and reduce poverty levels.
The only education variable that was found to be significant was the household head’s
primary school attendance, which had a positive impact on cropping income levels and nonfarm
income participation levels. The results were particularly large for cropping income, where
primary school attendance increased per capita income by 22,568 F CFA. One possible
explanation for this result is that some exposure to primary school education may improve the
household head’s access to information on both farm and nonfarm practices and techniques,
credit, and market information, which can increase agricultural production levels, improve

84

marketing skills, and help households overcome barriers to nonfarm activities. None of the
variables relating to the education levels of other household members were significant. This may
be due to the fact that family structures in Mali are hierarchical and as a result, the household
head may make most decisions on the household’s choice of income activities.
The results from this analysis are somewhat similar to those reported in previous studies.
For example, the result that easy access to roads is important for nonfarm activities was also
reported by a study from Nicaragua (Corral & Reardon, 2001). In addition, larger family sizes
were found to be negatively correlated with farm income in a study from Mexico (Yunez-Naude
& Taylor, 2001) and increased levels of rural poverty in Nigeria (Anyanwu, 2005). The Mexican
study reported that low levels of education were positively correlated with cropping activities,
which is similar to the results reported in this thesis. However that study did not find that the
education level of the household head mattered, which differs from the results found for
Tominian and Koutiala in this analysis. This variation could be explained by differences in the
family structure between rural households in Mali and those in Mexico.

85

6. COMMUNITY INCOME INEQUALITY AND INCOME SOURCES
6.1 Methodology
This thesis has so far examined household income levels found in the Tominian and
Koutiala zones of Mali, as well as factors associated with higher participation rates and income
levels for crop, livestock, and nonfarm activities. However, attention should also be paid to who
benefits the most from the income earned from each income source. For example, if nonfarm
incomes rise but only the wealthy are participating in nonfarm activities, community inequality
levels will increase and poverty levels will not decline. For this reason, this chapter will examine
how income from various income sources impacts community inequality levels.
To study inequality, one needs to decide on an appropriate income inequality index. Past
research has proposed various indices for measuring income inequality. According to the
literature, an ideal income inequality measure will have the following features (Adams & He,
1995):
1)

Pigou-Dalton transfer sensitivity: If income is transferred from a poorer individual to a

wealthier individual, the income inequality measure should increase.
2)

Symmetry: If two individuals switch incomes with each other, the income measure should

remain unchanged.
3)

Mean Independence: If all people in the community have their income increased by the

same proportional level, the income inequality measure should remain unchanged.
4)

Population homogeneity: If the population size increases but the distribution of income

does not change within the population, the income inequality measure should remain unchanged.
5)

Decomposability: The income inequality measure should have the ability of being broken

down so one can determine what types of income increase or decrease inequality.

86

The Gini coefficient, based on the Lorenz curve, is one type of income inequality index
that meets all five of these requirements (Adams & He, 1995) and for this reason, it is the income
inequality measurement used in this thesis. According to the Food and Agriculture Organization
of the United Nations (FAO), the Lorenz curve “relates the cumulative proportion of income to
the cumulative proportion of individuals” (Bellù, 2005, p. 2). If there is zero inequality in the
community, then the Lorenz curve will be identical to a 45 degree line and the Gini coefficient
will be zero. As income inequality increases, the Lorenz curve stretches farther and farther away
from the 45 degree line, and the Gini coefficient becomes larger. Mathematically, the Gini
coefficient is the ratio of the area between the line of equality and the Lorenz curve divided by
the total area under the 45 degree line. In Figure 6, the Gini coefficient is equal to A/(A+B) (The
World Bank, 2011).

Figure 6: The Gini Coefficient

Line of Equality
Gini Coefficient = A/(A+B)

Lorenz Curve
A
45°

B

Source: Adapted by author from a figure in World Bank (2011)

87

If all household incomes included in an analysis are positive, the Gini coefficient will
take a value between 0 and 1. However, Gini coefficients can also handle negative incomes
(Wodon & Yitzhaki, 2001). If negative values exist, the Gini coefficient may be greater than one.
While this does not indicate perfect inequality, it is an acceptable finding that was commonly
reported in previous studies (see, for example, Pyatt, Chen, & Fei, 1980; Lopéz-Feldman, Mora,
& Taylor, 2007; Mishra, El-Osta, & Gillespie, 2009; Lerman & Yitzhaki, 1985).
The three rounds of household survey data from Mali contain one case where total
income was reported to be below zero. As a result, the regional Gini coefficients were calculated
twice, with and without this one household included. To determine whether differences in the
Gini coefficients between the two zones are statistically significant, standard errors and 95%
confidence intervals were calculated through bootstrapping using the STATA command ineqerr.
Bootstrapping is a resampling method that is frequently used to calculate standard errors
of an estimate, . To bootstrap, a random drawing of n observations from the overall population
is performed p times with replacement. For each drawing, the estimate

is calculated where
8

a=1, 2, 3, … , p and then the sample standard errors are estimated using the following equation :

In addition, this study decomposed the Gini coefficients to determine how incomes from
various income categories (cash crops, food crops, livestock, agricultural wages, non-agricultural
wages, transfers, and other nonfarm activities) contributed to overall income inequality in the
two zones. To decompose the Gini coefficients, the method proposed by Lerman and Yitzhaki
8

The general equations and description of basic bootstrapping is based on information from
Wooldridge, 2009.
88

(1985) was used. Lerman and Yitzhaki state that the Gini coefficient can be written as the
following:

In this equation, Sk represents the share of total income earned from a given activity, Gk
represents the Gini coefficient of income inequality within each income category, and Rk
represents the Gini correlation between income from a given activity and total income. Rk can be
calculated through the following equation:

In this equation, Yk represents income from a given activity and F(Y) and F(Yk)
represent the cumulative distributions of total income and income from the given income
category, respectively. Once these two equations are calculated, the percentage change in the
Gini coefficient associated with a 1% increase in income from a given activity can be calculated
with the following equation:
% Change

=

-

The Gini coefficient decomposition was performed using the descogini command in the
STATA statistical software. Confidence intervals and standard errors were calculated by
bootstrapping for the percentage change in the Gini coefficient associated with a 1% increase in
income earned from a given activity.
When interpreting the Gini decomposition results, several issues should be noted. First,
this Gini decomposition method can suggest income sources that are income inequality
89

increasing or decreasing, but it does not show how income portfolios may have been different in
the absence of a given income source. For example, a Gini decomposition analysis might suggest
that a given income source (Income Source A) is income inequality increasing. However if in the
absence of Income Source A, poor households participate in a different income activity with
lower earnings (Income Source B), the households are worse off and community inequality may
be higher. Therefore, despite the findings from the Gini decomposition analysis that Income
Source A is inequality increasing, it may be a better income source for the community than its
alternatives. A second limitation of this Gini decomposition method is that it cannot identify
factors that cause an income source to be inequality increasing or decreasing (Reardon et al.,
2000). Thus, the analysis presented during the next several pages of this thesis may propose
hypotheses on why a given income source is inequality increasing or decreasing although further
research using more complex methods would be needed to confirm these hypotheses.

6.2 Results
For the three survey years included in this analysis (2006/07, 2008/09, and 2009/10), the
regional Gini coefficients of household incomes ranged from 0.409-0.422 for Tominian and
0.373-0.4 for Koutiala (Table 25, Table 26 and Table 27). These values are slightly larger than
the regional Gini coefficients calculated by Reardon et al. (1992) for areas in Burkina Faso with
agroclimatic zones similar to Tominian and Koutiala using data from the early 1980s. By
excluding the one household with negative total income, standard errors and 95% confidence
intervals could be calculated for the two coefficients using bootstrapping. The confidence
intervals were found to overlap, suggesting that the Gini coefficients in Tominian and Koutiala
may not be statistically different from each other.

90

Table 25: Gini Coefficients for Koutiala and Tominian (2006/07)
Zone

Gini Coefficients

Tominian

0.410

Koutiala

0.373

Gini Coefficients
0.392
(0.020)
95% CI: 0.359-0.435
0.366
(0.018)
95% CI: 0.333-0.404

Table 26: Gini Coefficients for Koutiala and Tominian (2008/09)
Zone

Gini Coefficients

Tominian

0.409

Koutiala

0.391

Gini Coefficients
0.409
(0.021)
95% CI: 0.370-0.455
0.390
(0.024)
95% CI: 0.350-0.442

Table 27: Gini Coefficients for Koutiala and Tominian (2009/10)
Zone

Gini Coefficients

Tominian

0.422

Koutiala

0.400

Gini Coefficients
0.401
(0.016)
95% CI: 0.368-0.432
0.400
(0.022)
95% CI: 0.359-0.444

Source: Survey data
() reports standard error, calculated through bootstrapping using STATA command ineqerr,
1,000 repetitions performed
95% CI: stands for 95% Confidence Intervals, also calculated through STATA’s ineqerr
command
The Gini coefficients for both zones were also decomposed by income source (Tables 2833). In both zones, this showed that food crop income was generally income inequality
decreasing while livestock income was inequality increasing. Given that food crops made up a
significant share of total income (as shown by Sk) and income inequality within this income
91

source was low (Gk), the magnitude of the impact that food crop income had on income
inequality was relatively great. However, the inequality increasing effect of livestock income
was smaller because 1) livestock incomes made up a relatively small share of total income (Sk)
and 2) the large number of negative livestock incomes (signifying livestock investments) resulted
in a Gini coefficient that is greater than one. As a result, the income inequality effects of
livestock income showed by the Gini decomposition analysis may have been overstated and thus
appear to be larger than they may have actually been in reality. Another notable finding from the
Gini decomposition analysis is that in Tominian, non-agricultural wages were found to be
inequality increasing in both 2008/09 and 2009/10. In Koutiala, agricultural wages were found to
be income inequality decreasing in all three years and cash crops were inequality increasing for
the last two cropping years. All of these results were found to be statistically significant at the 1%
level through bootstrapping.

92

Table 28: Gini Decomposition of Household Incomes in Koutiala (2006/07)
Income
1
2
3
Share
% Change
SK
GK
RK
Source
Crops (Food
0.705
0.379
0.909
0.651
-0.054*
and Cash)
Livestock
0.112
0.908
0.621
0.170
0.057***
Ag Wages
0.006
0.886
-0.162
-0.002
-0.008***
Non-Ag
0.008
0.969
0.197
0.004
-0.004
Wages
Transfers
0.023
0.825
0.309
0.016
-0.007
Other
0.147
0.638
0.648
0.163
0.016
Nonfarm
Source: Survey data
*, **, *** signify 10%, 5%, and 1% statistical significance, respectively; significance
determined using bootstrapping with 1,000 repetitions
1
Sk represents the share of income from a given activity in total income
2

Gk represents the Gini coefficient of income inequality within each income category

3

Rk represents the Gini correlation between income from a given activity and total income

Table 29: Gini Decomposition of Household Incomes in Koutiala (2008/09)
Income
1
3
2
Share
% Change
SK
RK
GK
Source
Cash Crops
0.183
0.707
0.741
0.245
0.063***
Food Crops
0.537
0.399
0.841
0.461
-0.076***
Livestock
0.088
1.024
0.551
0.127
0.039
Ag Wages
0.007
0.874
-0.016
-0.000
-0.007***
Non-Ag
0.022
0.922
0.391
0.020
-0.002
Wages
Transfers
0.035
0.718
0.233
0.015
-0.020***
Other
0.129
0.662
0.600
0.131
0.002
Nonfarm
Source: Survey data
*, **, *** signify 10%, 5%, and 1% statistical significance, respectively; significance
determined using bootstrapping with 1,000 repetitions
1
Sk represents the share of income from a given activity in total income
2

Gk represents the Gini coefficient of income inequality within each income category

3

Rk represents the Gini correlation between income from a given activity and total income
93

Table 30: Gini Decomposition of Household Incomes in Koutiala (2009/10)
Income
1
2
3
Share
% Change
SK
GK
RK
Source
Cash Crops
0.186
0.650
0.751
0.226
0.041*
Food Crops
0.498
0.346
0.854
0.367
-0.131***
Livestock
0.081
1.044
0.613
0.130
0.049**
Ag Wages
0.009
0.854
0.116
0.002
-0.007*
Non-Ag
0.016
0.945
0.543
0.020
0.004
Wages
Transfers
0.068
0.817
0.697
0.097
0.029
Other
0.143
0.677
0.656
0.158
0.015
Nonfarm
Source: Survey data
*, **, *** signify 10%, 5%, and 1% statistical significance, respectively; significance
determined using bootstrapping with 1,000 repetitions
1
Sk represents the share of income from a given activity in total income
2

Gk represents the Gini coefficient of income inequality within each income category

3

Rk represents the Gini correlation between income from a given activity and total income

Table 31: Gini Decomposition of Household Incomes in Tominian (2006/07)
Income
1
2
3
Share
% Change
SK
GK
RK
Source
Crops (Food
0.593
0.392
0.793
0.450
-0.144***
and Cash)
Livestock
-0.003
-30.914
0.604
0.155
0.159***
Ag Wages
0.003
0.973
0.043
0.000
-0.003
Non-Ag
0.020
0.977
0.674
0.033
0.012
Wages
Transfers
0.159
0.747
0.492
0.142
-0.017
Other
0.228
0.619
0.639
0.220
-0.008
Nonfarm
Source: Survey data
*, **, *** signify 10%, 5%, and 1% statistical significance, respectively; significance
determined using bootstrapping with 1,000 repetitions
1
Sk represents the share of income from a given activity in total income
2

Gk represents the Gini coefficient of income inequality within each income category

3

Rk represents the Gini correlation between income from a given activity and total income
94

Table 32: Gini Decomposition of Household Incomes in Tominian (2008/09)
Income
1
2
3
Share
% Change
SK
GK
RK
Source
Cash Crops
0.152
0.593
0.654
0.144
-0.008
Food Crops
0.406
0.422
0.693
0.290
-0.116***
Livestock
0.086
1.142
0.608
0.146
0.060***
Ag Wages
0.016
0.979
0.822
0.031
0.015
Non-Ag
0.071
0.920
0.739
0.118
0.047***
Wages
Transfers
0.120
0.769
0.555
0.125
0.005
Other
0.149
0.682
0.588
0.146
-0.003
Nonfarm
Source: Survey data
*, **, *** signify 10%, 5%, and 1% statistical significance, respectively; significance
determined using bootstrapping with 1,000 repetitions
1
Sk represents the share of income from a given activity in total income
2

Gk represents the Gini coefficient of income inequality within each income category

3

Rk represents the Gini correlation between income from a given activity and total income

Table 33: Gini Decomposition of Household Incomes in Tominian (2009/10)
Income
1
2
3
Share
% Change
SK
GK
RK
Source
Cash Crops
0.144
0.603
0.630
0.130
-0.015
Food Crops
0.462
0.406
0.711
0.315
-0.147***
Livestock
0.051
2.242
0.584
0.159
0.108***
Ag Wages
0.005
0.916
0.204
0.002
-0.003
Non-Ag
0.060
0.957
0.804
0.110
0.049***
Wages
Transfers
0.130
0.774
0.490
0.116
-0.013
Other
0.148
0.768
0.626
0.168
0.020
Nonfarm
Source: Survey data
*, **, *** signify 10%, 5%, and 1% statistical significance, respectively; significance
determined using bootstrapping with 1,000 repetitions
1
Sk represents the share of income from a given activity in total income
2

Gk represents the Gini coefficient of income inequality within each income category

3

Rk represents the Gini correlation between income from a given activity and total income
95

6.3 Discussion of the Gini Decomposition Results
This analysis of Gini coefficients shows that income inequality rates in the Tominian and
Koutiala zones of Mali were relatively low (0.373-0.422) and comparable to Mali’s national
2006 Gini coefficient of 0.39 reported by the World Bank (World Bank, 2012a). The Gini
decomposition analysis also shows some interesting findings that policy makers should consider
when making decisions on economic development projects in these two zones. First, food crop
income was inequality decreasing. Given that almost all surveyed households reported income
from food crops (suggesting low entry barriers) and that this income source has an income
inequality decreasing effect, policies and programs that increase food crop income levels may
decrease poverty and reduce community inequality levels. This could include projects that
increase food crop production levels, such as improving farmers’ access to inputs, credit,
agricultural extension programs, or new technology. In addition, programs that help farmers
make higher profits or better market their food crop production would also have a positive effect
on inequality levels. This might include improving market information systems or supporting
rural farmer cooperatives, which might help farmers strengthen their bargaining power with
traders, make more informed market decisions, reduce input and transaction costs, and gain
access to credit.
Since the cotton sector receives considerable attention by policy makers, the finding that
cash crop income in the Koutiala zone was inequality increasing should be noted. This suggests
that while projects focused on improving the zone’s cotton sector may increase income levels,
those who primarily benefit from these projects may be some of the wealthier subsections of the
population.

96

Finally in the Tominian zone, non-agricultural wages were found to be income inequality
increasing. This information, plus the finding from Chapter 4 that participation in this income
source was limited despite high incomes in this sector, suggests that poorer households had
difficulty gaining access to this income source. As a result, if policies are put into place that
increase non-agricultural wage incomes in rural Mali, it seems likely that only the wealthier
subsections of the population will benefit unless the issues of access and entry barriers are
addressed.

97

7. CONCLUSIONS
The purpose of this study was to understand household income sources and levels, as
well as how certain income activities either increased or decreased community inequality levels
in two zones of rural Mali. The first zone, Tominian, is located within Mali’s traditional course
grain production zone, has received little public investment, and has been struggling
economically in recent years. The other zone, Koutiala, is located in Mali’s cotton basin, and has
traditionally received considerable public investment, particularly relating to the cotton sector.
The results of this study show that households in both zones were very poor with only 816% of all surveyed households earning more than $1/day per capita during each of the three
cropping years. Households in the Koutiala zone were found to have earned about 1.5-2 times
the income of households in Tominian during all survey rounds.
Overall, this study shows that food crops were the most important income source for all
households in the Tominian and Koutiala zones. In addition, this study showed that participation
rates for food cropping activity were found to be high (nearly 100%) and that this income source
was inequality decreasing. Given the importance of this income source and the low entry barriers
for this activity, policy makers should focus policies and programs on increasing production,
marketing, and sales of food crops. Examples of such programs might include programs that
improve farmers' access to agricultural inputs, agricultural extension programs, better storage
techniques, or market price information systems.
The econometric model of determinants of crop income presented in this thesis showed
that household heads with some primary school education earned higher levels of crop income
than household heads with no primary school education or more advanced levels of education at
the secondary school level or beyond. This suggests that policies that encourage greater primary

98

school attendance could be effective in raising rural incomes. The econometric model showed
that larger households generally earned less crop income, which is consistent with the findings of
other similar studies on incomes and poverty (see, for example, Yunez-Naude & Taylor, 2001 or
Anyanwu, 2005). This suggests that households are facing diminishing returns to additional labor.
As a result, programs that help promote family planning in rural areas of Mali might be
beneficial.
Nonfarm income, particularly income from self-employment, commerce, artisan, primary
sector activities, and transformation of agricultural products, was found to be a smaller but still
sizable share of total income. Although the income inequality effects of nonfarm income
activities was inconclusive for all nonfarm income sources except non-agricultural wages, it
seems that promoting nonfarm income activities through policy initiatives, such as training
programs, improved access to non-agricultural credit, or assistance for the development of
artisan cooperatives, could reduce poverty. Based on the results of the econometric model,
improving road access in rural areas (i.e. building, repairing, and maintaining roads and bridges
to connect rural communities to local and regional markets) and reducing family sizes may
positively impact nonfarm income levels.
That said, it should be noted that not all nonfarm income activities are the same. In
Tominian, non-agricultural wage activities were found to be income inequality increasing. Given
the high levels of income earned from these activities but low activity participation rates, entry
barriers may be an obstacle preventing many poor households from engaging in non-agricultural
wage activities. Entry barriers to nonfarm income activities in general would vary depending on
the activity but might include lack of access to credit, capital, technical skills, or social networks
needed for a nonfarm activity. Research to identify these barriers and projects aimed at helping

99

the poor overcome these constraints could be beneficial. However, without such programs, the
promotion of non-agricultural wage activities will likely just increase the incomes of the wealthy
subsections of the population, increasing inequality and have little direct impact on poverty
levels.
Livestock income was found to make up only a small share of total income for the poor
and middle-class subsections of the surveyed households, despite a relatively high participation
rate. In addition, this income source was found to be slightly income inequality increasing. This
suggests that while many households participate in livestock activities, only the wealthy are able
to amass enough animals to earn significant income from this source. Focusing policies on other
income areas will likely have a larger impact on reducing poverty.
Finally, it should be noted that despite great attention given to the cotton sector,, cash
crop income in Koutiala was found to be a considerably smaller share of total income than food
crops and was only a slightly larger share than the share of total income earned from nonfarm
activities. In addition, evidence shows that cash crops in Koutiala were inequality increasing and
that wealthier households earned a larger share of income from cash crops than poorer
households. However, before moving policies away from the cotton sector, additional research is
needed. First, the cropping years included in this study may not be particularly representative of
a normal cotton production year in Koutiala. In particular, due to the late 2007 cotton production
payments, many farmers in the zone boycotted cotton in 2008 (Lazarus & Kelly, 2012). As a
result, the inequality increasing effect of cotton found in this analysis may mean that wealthier
farmers were less likely to abandon cotton production during difficult times for the sector. Also,
one should consider that cotton may have a positive, indirect impact on food crop production that
is not picked up by this analysis. For example, extra fertilizer provided by the CMDT for cotton

100

production is often used on cereal crops in the Malian cotton basin (Staatz, Dioné, & Dembélé,
1989). Therefore, further research on the effects of cotton income on poverty levels and
inequality in Mali should be performed

7.1 Limitations of this Study
Despite the information learned from this analysis of household incomes, there are
several limitations to this study. First, the income data was collected through annual field
interviews and required household members to recall all income earned during the previous year.
Given that incomes in Mali are rarely salaried and low levels of literacy limit financial record
keeping, one year recall data can be inaccurate. Survey collection methods that interview
households weekly or monthly about their income levels may result in more accurate income
data.
Second, this household survey only included households active in agriculture. As a result,
it is possible that some households in the community that are solely involved in nonfarm
activities were omitted from the study.
Finally according to the results of the study very few households earned more than a
$1/day per capita. While this is possible, this statistic also suggests that there was perhaps underreporting of household incomes. The under-estimating of income could have several causes. First,
not all income-earning household members were interviewed during data collection. As a result,
interviewed household members were asked to estimate incomes for household members not
present during the interview. Given the large family sizes in Mali, interviewed household
members may not have known the exact income levels of other members of their household,
leading to under-estimation of household income. In addition, income levels can be sensitive

101

information that some households may have been hesitant to accurately reveal to survey
enumerators.

7.2 Future Research
In addition to the study limitations mentioned above that could be further researched, the
results of this study suggest several research areas that could be explored with additional
household survey data from Mali. First, additional research on the effect of cotton income on
income inequality using data from a typical cotton production year is needed to confirm or refute
the finding that cotton is income inequality increasing. Second, the Gini decomposition method
used in this thesis cannot identify factors that may contribute to an income source being
inequality increasing or decreasing. While this study offers hypotheses on why a given income
source may be inequality increasing or decreasing, further research using more complex methods
would be needed to confirm these hypotheses. Finally, further research on transfer and
remittance incomes could help researchers and policy makers gain a better understanding of the
income dynamics in the Tominian zone, where transfer income makes up a sizable share of total
income among poorer households.

102

APPENDICES

103

Appendix A: Method for Calculating Households Living Above and Below the $1/day Poverty
Line
A common poverty line used by international development organizations is based on
whether or not household income earnings exceed a $1/day. In order to determine what
percentage of the surveyed households fell above or below this threshold, total household
incomes reported for Tominian and Koutiala were converted into international dollars using a
purchasing price parity approach.
Converting currencies using a purchasing price parity approach is not the same as
converting currencies using official exchange rates. In the case of Mali in 2010, the average
official exchange rate was 495.28 Franc CFA/1 US dollar (World Bank, 2012d). This compares
to the private consumption purchasing price parity conversion factor which was 301.4 Franc
CFA per international dollar (equivalent to the purchasing power of one US dollar) in 2010
(World Bank, 2012e). The purchasing price parity approach converts one country’s currency into
another currency at the rate in which one could buy the same amount of goods in each country.
For example, a very simplified purchasing price parity index used by The Economist magazine,
called the “Big Mac” index, compares the prices of hamburgers sold at MacDonald’s in various
countries throughout the world. More complex purchasing price party indices, such as those
presented by the World Bank, usually compare prices from various countries using a basket of
consumer goods. For most purposes, the purchasing price parity approach is believed to be a
better indicator of overall well-being than official exchange rates (Callen, 2007). In addition, the
United Nations uses this approach in its calculations of households living above and below the
$1/day/capita poverty line (United Nations Statistical Division, 2012b). For these reasons, a
purchasing price parity approach was used in this thesis.

104

This appendix briefly covers the three steps used to determine a household’s location
relative to the $1/day poverty line using a purchasing price parity approach. These steps are:
1) To match the methodology used by the United Nations, the World Bank's purchasing
price conversion factor for private consumption from 2010 was used (301.4 Franc CFA
per international dollar).
2) In order to convert the reported household incomes into international dollars, the incomes
had to be divided by this conversion factor. For example, if a household of 10 people
reported earning 1,155,115 Franc CFA in 2010, then the household’s annual earnings in
international dollars was:

3) To determine the household’s per capita earnings per day, the international dollars were
divided by the number of people present in the household and the number of days in one
year. For example, using the household described in Step 2:

This shows that our example household was living above the $1/day poverty line in 2010.

105

Appendix B / Table 34: Real per Capita Income from Various Sources, Including F-Tests to Determine Statistical Significance of the
Income Differences found between Survey Years (in 2010 Franc CFA)
Tominian
2006/07

2008/09

Koutiala

2009/10

Signif.
(twotailed)

2006/07

2008/09

2009/10

Signif.
(twotailed)

1

-

7,740

5,154

0.009

-

12,798

12,539

0.903

1

-

20,985

16,637

0.055

-

42,924

37,714

0.084

30,856

28,725

21,792

0.005

60,485

55,723

50,254

0.036

Livestock

-386

3,306

2,210

0.010

9,471

6,103

4,864

0.071

Ag Wages

213

1,753

156

0.351

5,751

572

789

0.437

Non-Ag
Wages

1,065

4,288

2,496

0.041

660

1,625

1,035

0.225

Transfers

9,123

6,162

4,194

0.063

2,052

3,538

14,507

0.343

1,152

7,688

5,084

0.000

11,645

9,462

10,257

0.459

Cash Crops

Food Crops
Food and
Cash Crops
Combined

Other
Nonfarm
Source: Survey data
1
Represents a t-test

106

Appendix C / Table 35: Real per Capita Income from Various Sources, Including T-Tests to Determine Statistical Significance of the
Income Differences found between Zones (in 2010 Franc CFA)
2006/07 Cropping Season
Tominian

Koutiala

2008/09 Cropping Season

Signif.
(2tailed)

Tominian

2009/10 Cropping Season

Koutiala

Signif.
(2tailed)

Tominian

Koutiala

Signif.
(2tailed)

Cash Crops

-

-

-

7,740

12,798

0.090

5,154

12,539

0.000

Food Crops

-

-

-

20,985

42,924

0.000

16,637

37,714

0.000

30,856

60,485

0.000

-

-

-

-

-

-

Livestock

-386

9,471

0.000

3,306

6,103

0.051

2,210

4,864

0.042

Ag Wages

213

575

0.042

1,753

572

0.445

156

789

0.000

Non-Ag
Wages

1,065

660

0.527

4,288

1,625

0.034

2,496

1,035

0.140

Transfers

9,123

2,052

0.001

6,162

3,538

0.095

4,194

14,507

0.366

Other
Nonfarm

11,512

11,645

0.940

7,688

9,462

0.200

5,084

10,257

0.001

Crops (Food
and Cash
Combined)

Source: Survey data

107

Appendix D / Table 36: Average Share of Household Income from Various Sources, Including F-Tests to Determine Statistical
Significance of the Income Differences Found Between Survey Years
Tominian
2006/07

2008/09

Koutiala

2009/10

Signif.
(twotailed)

2006/07

2008/09

2009/10

Signif.
(twotailed)

Cash
1
Crops

-

15%

14%

0.452

-

14%

16%

0.457

Food
1
Crops

-

49%

53%

0.168

-

59%

57%

0.376

Crops
(Food and
Cash)

62%

64%

67%

0.666

73%

73%

73%

0.965

Livestock

0%

4%

2%

0.766

8%

6%

5%

0.670

Ag Wages

0%

1%

1%

0.707

1%

1%

0.605

Non-Ag
Wages

1%

4%

3%

0.055

1%

2%

1%

0.196

Transfers

16%

12%

13%

0.278

2%

5%

6%

0.003

Other
Nonfarm

21%

15%

13%

0.004

15%

13%

14%

0.618

Source: Survey data
1
Represents a t-test

108

1%

Appendix E / Table 37: Average Share of Household Income from Various Sources, Including T- Tests to Determine Statistical
Significance of the Income Differences found between Zones
2006/07
Tominian

Koutiala

2008/09
Signif.
(twotailed)

Tominian

Koutiala

2009/10
Signif.
(twotailed)

Tominian

Koutiala

Signif.
(twotailed)

Cash
Crops

15%

14%

0.744

14%

16%

0.209

Food
Crops

49%

59%

0.002

53%

57%

0.277

Crops
Combined

62%

73%

0.099

Livestock

0%

8%

0.295

4%

6%

0.576

2%

5%

0.224

Ag Wages

0%

1%

0.133

1%

1%

0.311

1%

1%

0.057

Non-Ag
Wages

1%

1%

0.608

4%

2%

0.071

3%

1%

0.078

Transfers

16%

2%

0.000

12%

5%

0.000

13%

6%

0.000

Other
Nonfarm

21%

15%

0.015

15%

13%

0.211

13%

14%

0.812

Source: Survey data

109

BIBLIOGRAPHY

110

BIBLIOGRAPHY

Abdulai, A., & CroleRees, A. (2001). Determinants of income diversification amongst rural
households in southern Mali. Food Policy, 26(4), 437-452.
Abdulai, A., & Delgado, C. L. (1999). Determinants of nonfarm earnings of farm-based
husbands and wives in northern Ghana. American Journal of Agricultural Economics,
81(1), 117-130.
Adams, R.H. (2002). Nonfarm income, inequality, and land in rural Egypt. Economic
Development and Cultural Change, 50(2), 339-363.
Adams, R. H., & He, J. J. (1995). Sources of income inequality and poverty in rural Pakistan
(Research Report No. 102). Washington, D.C.: International Food Policy Research
Institute.
Anyanwu, J. C. (2005). Rural poverty in Nigeria: Profile, determinants and exit paths. African
Development Review, 17(3), 435-460.
Atwood, D.A. (1990). Land registration in Africa: The impact on agricultural production. World
Development, 18(5), 659-671.
Bee, A. (2012). Household income inequality within U.S. counties: 2006–2010 (American
Community Survey Briefs # 10-18). Retrieved from
http://www.census.gov/prod/2012pubs/acsbr10-18.pdf
Bellù, L. G. (2005). Charting income inequality: The Lorenz curve. Retrieved from
http://www.fao.org/docs/up/easypol/302/charting_income_inequality_000en.pdf
Berdegue, J., Ramirez, E., & Reardon, T. (2001). Rural nonfarm employment and incomes in
Chile. World Development, 411-425.
Boughton, D., Staatz, J., & Dembélé, N. (2010). Mali’s food security challenges: An overview.
East Lansing: Michigan State University.
Callen, T. (2007). PPP versus the market: Which weight matters? Finance and Development, 44
(1). Downloaded from http://www.imf.org/external/pubs/ft/fandd/2007/03/basics.htm.
Chamberlain, G. 1984. “Panel data”. In Grilliches, Z., Intriligator, M.D., eds. Handbook of
Econometrics, Vol. 2 North Holland, Amsterdam. 1247-1318.
CIRAD (Centre de Coopération Internationale en Recherche Agronomique pour le
Développement). (2012). Fonio. Retrieved from http://fonio.cirad.fr/index.php/fonio_en
Corral, L., & Reardon, T. (2001). Rural nonfarm incomes in Nicaragua. World Development,
29(3), 427-442.
111

Cotula, L., Toulmin, C., & Hess, Ced. (2004). Land tenure and administration in Africa: Lessons
of experience and emerging issues. London: International Institute for Environment and
Development.
Cunguara, B., G. Fagilde, J. Garrett, R. Uaiene, and Derek Heady. (2011). Growth without
change: The elusiveness of agricultural and economic transformation in Mozambique. A
study commissioned by IFPRI as part of the CAADP initiative in Mozambique.
David, R. (1995) Changing places?: Women, resource management and migration in the Sahel:
Case studies from Senegal, Burkina Faso, Mali and Sudan. London, UK: SOS Sahel
International.
Debrah, S., & Sissoko, K. (1990). Sources and transfers of cash income in the rural economy:
The case of smallholder mixed farmers in the semi-arid zone of Mali (African Policy
Analysis Network Paper No. 25). Addis Ababa, Ethiopia: International Livestock Centre
for Africa.
Delgado, C.L., Hopkins, J., & Kelly, V.A. (1998). Agricultural growth linkages in Sub-Saharan
Africa (IFPRI Research Report 107). Washington, DC: International Food Policy
Research Institute.
Dembélé, N. N., & Staatz, J. (2002). The impact of market reform on agricultural transformation
in Mali. In T.S. Jayne, I.J. Minde, & G. Argwings-Kodhek (Eds.) Perspectives on
Agricultural Transformation: A view from Africa (163-189). East Lansing: Michigan
State University.
Dercon, S., (1998). Wealth, risk and activity choice: Cattle in Western Tanzania. Journal of
Development Economics, 55(1), 1-42.
Ellis, F. (2000a). Rural livelihoods and diversity in developing countries. Oxford, NY: Oxford
University Press.
Ellis, F. (2000b). The determinants of rural livelihood diversification in developing countries.
Journal of Agricultural Economics, 51(2), 289–302.
Escobal, J. (2001). The determinants of nonfarm income diversification in rural Peru. World
Development, 29(3), 497-508.
Fafchamps, M., Udry, C., & Czukas, K. (1998). Drought and saving in West Africa: Are
livestock a buffer stock? Journal of Development Economics, 55(2), 273-305.
Famine Early Warning Systems Network (2010). Livelihood zoning and profiling report: Mali.
Washington, D.C.
Food and Agriculture Organization of the United Nation’s Statistical Division. (2012).
FAOSTAT [Data file]. Retrieved from http://faostat.fao.org/

112

Gubert, F., Lassourd, T., & Mesple-Somps, S. (2010). Do remittances affect poverty and
inequality? Evidence from Mali (Working Paper DT/2010-08). Paris, France: Institut de
recherché pour le développement, Dauphine Université Paris
Haggblade, S., & Hazell, P. (1988). Prospects for equitable growth in rural sub-Saharan Africa
(AGRAP Economic Discussion Paper No. 3). Washington, D.C.: The World Bank.
Haggblade, S., Hazell, P., & Reardon, T. (2009, February). Transforming the rural nonfarm
economy: Opportunities and threats in the developing world. Retrieved from
http://www.ifpri.org/sites/default/files/publications/ib58.pdf
Harrower, S., & Hoddinott, J. (2005) Consumption smoothing in the zone Lacustre, Mali.
Journal of African Economies, 14(4), 489-519.
Hertz, T. (2009). The effect of nonfarm income on investment in Bulgarian family farming.
Agricultural Economics, 40(2), 166-177.
Lanjouw, P. (2001). Nonfarm employment and poverty in rural El Salvador. World Development,
529-547.
Lazarus, B. & Kelly, K. (2012). Nonfarm income in the Tominian, Macina, and Koutiala zones
of Mali. Manuscript in progress.
Lerman, R., & Yitzhaki, S. (1985). Income inequality effects by income source: A new approach
and applications to the United States. The Review of Economics and Statistics, 151-156.
Lopéz-Feldman, A., Mora, J., & Taylor, J. E. (2007) Does natural resource extraction mitigate
poverty and inequality? Evidence from rural Mexico and a Lacondona rainforest
community. Environment and Development Economics, 12, 251-269.
Maertens, M., & Verhofstadt, E. (2011). Maternal off-farm wage employment and primary
school enrolment: Evidence from a natural quasi-experiment in Senegal. Unpublished
paper presented at The European Association of Agricultural Economists 2011 Congress
Change and Uncertainty, Zurich, Switzerland.
Maxwell, S., & Fernando, A. (1989). Cash crops in developing countries: The issues, the facts,
the policies. World Development, 17(11), 1677-1708.
McGlinchy, M. E. (2006). Identifying the determinants and characteristics of mothers’ income in
rural Mali. (Unpublished master’s thesis). Michigan State University, East Lansing.
Mishra, A., El-Osta, H., Gillespie, J.M. (2009). Effect of agricultural policy on regional income
inequality among farm households. Journal of Policy Modeling, 31, 325-340.
Mundlak, Y. 1978. “On the pooling of time series and cross section data.” Econometrica 46: 6985.

113

Murekezi, A. (2012). Household cereal production and marketing patterns in Mali: Implications
for additional policy analysis. Unpublished manuscript.
Muyanga, M., Jayne, T.S., & Burke, B. (2010). Pathways into and out of poverty: A study of
determinants of rural household wealth dynamics in Kenya (Working paper). Retrieved
from MSU Food Security Groups website:
http://www.aec.msu.edu/fs2/income/index.htm
Olale, E., & Hensen, S. (2012). Determinants of income diversification among fishing
communities in Western Kenya. Fisheries Research, 125-126, 235-242.
Oseni, G., & Winters, P. (2009). Rural nonfarm activities and agricultural crop production in
Nigeria. Agricultural Economics, 40(2), 189-201.
Perakis, S. M. (2011). The short and long run effects of migration and remittances: Some
evidence from Northern Mali. Unpublished paper presented at The Agricultural and
Applied Economics Association’s 2011 AAEA and NAREA Joint Annual Meeting,
Pittsburgh, PA.
Pyatt, G., Chen, C., & Fei, J. (1980). The distribution of income by factor components. Quarterly
Journal of Economics, 95(3), 451-473.
Reardon, T., Berdegue, J., Barrett, C. B., Stamoulis, K. (2006). Household income
diversification into rural nonfarm activities. In S. Haggblade, P. Hazell, & T. Reardon, T.
(Eds.), Transforming the Rural Nonfarm Economy. Baltimore , MD: Johns Hopkins
University Press.
Reardon, T., Delgado, C., & Matlon, P. (1992). Determinants and effects of income
diversification amongst farm households in Burkina Faso. Journal of Development
Studies, 28(2), 264-296.
Reardon, T., Fall, A., Kelly, V., Delgado, C., Matlon, P., Hopkins, J., & Badiane, O. (1993).
Agriculture-led income diversification in the west African semi-arid tropics: Nature,
distribution, and importance of production-linked activities. Paper presented at the
International Conference on African Economic Issues, Abidjan, Cote d’Ivoire.
Reardon, T. & Taylor, J.E. (1996). Agroclimatic shock, income inequality, and poverty:
Evidence from Burkina Faso. World Development, 24(5), 901-914.
Reardon, T., Taylor, J. E., Stamoulis, K., Lanjouw, P. & Balisacan, A. (2000). Effects of nonfarm employment on rural income inequality in developing countries: An investment
perspective. Journal of Agricultural Economics, 51(2), 266–288.
Ricker-Gilbert, J. (2011). Household-level impacts of fertilizer subsidies in Malawi. (Doctoral
dissertation). Michigan State University, East Lansing.
Rosenzweig, M. R., Stark, O. (1989). Consumption smoothing, migration and marriage:
Evidence from rural India. Journal of Political Economy, 97(4), 905-926.
114

Samake, A., Belieres, J.-F., Corniaux, C., Dembele, N., Kelly, V., Marzin, J., Sanogo, O., &
Staatz, J. (2008). Changements structurels des economies rurales dans la mondialisation:
Program RuralStruc Mali – phase II. East Lansing: Michigan State University.
Sheahan, M.B. (2011). Analysis of fertilizer profitability and use in Kenya. (Master’s thesis).
Michigan State University, East Lansing.
Smale, M., Diakité, L. & Keita, N. (2011). Location, vocation, and price shocks: cotton, rice, and
sorghum-millet farmers in Mali. Development in Practice, 21(4-5).
Staatz, J.M., Dioné, J., & Dembélé, N.N. (1989). Cereals market liberalization in Mali. World
Development, 17(5), 703-718.
Stampini, M., & Davis, B. (2009). Does nonagricultural labor relax farmers’ credit constraints?
Evidence from longitudinal data for Vietnam. Agricultural Economics, 40(2), 177-188.
Steedman, C., Daves, T., Johnson, M., & Sutter, J. (1976). Mali agricultural sector assessment.
Ann Arbor: The University of Michigan.
United Nations, The. (2010). GOAL 1: Eradicate extreme poverty and hunger [Fact sheet].
Retrieved from http://www.un.org/millenniumgoals/pdf/MDG_FS_1_EN.pdf
United Nations Development Program. (2011). International human development indicators.
Retrieved from http://hdr.undp.org/en/statistics/
United Nations Population Division. (2012). World urbanization prospects:
The 2007 revision population database [Data file]. Retrieved from
http://esa.un.org/unup/p2k0data.asp
United Nations Statistics Division. (2012a). Mali. Retrieved February 15, 2012, from
Millennium Development Goals Indicators: http://unstats.un.org/unsd/mdg/Data.aspx
United Nations Statistical Division. (2012b). Population below $1 (PPP) per day, percentage.
Retrieved from
http://mdgs.un.org/unsd/mdg/Metadata.aspx?IndicatorId=0&SeriesId=580
Wodon, Q., & Yitzhaki, S. (2001). Inequality and social welfare. In J. Klugman (Ed.), Poverty
Reduction Strategies Source Book. (75-431). Washington, D.C.: World Bank.
Wooldridge, J. (2009). Introductory econometrics: A modern approach. Mason, OH: South
Western, Cengage Learning.
Wooldridge, J. (2010). Econometric analysis of cross section and panel data. Cambridge, MA:
The MIT Press.
Wooten, S. (2003). Women, men, and market gardens: Gender relations and income generation
in rural Mali. Human Organization, 62(2), 166-177.

115

World Bank, The. (2011). Measuring Inequality. Retrieved from
http://go.worldbank.org/3SLYUTVY00.
World Bank, The. (2012a). GINI index [Data File]. Retrieved from
http://data.worldbank.org/indicator/SI.POV.GINI.
World Bank, The. (2012b). Goal 1: Eradicate extreme poverty and hunger. Retrieved from
http://go.worldbank.org/T1S8XUXT40
World Bank, The. (2012c). Mali [Data File]. Retrieved from
http://data.worldbank.org/country/mali#cp_wdi.
World Bank, The. (2012d). Official exchange rate (LCU per US$, period average) [Data File].
Retrieved from http://data.worldbank.org/indicator/PA.NUS.FCRF.
World Bank, The. (2012e). PPP conversion factor, private consumption (LCU per international $)
[Data File]. Retrieved from http://data.worldbank.org/indicator/PA.NUS.PRVT.PP.
Wouterse, F., & Taylor, J.E. (2008). Migration and income diversification: Evidence from
Burkina Faso. World Development, 36(4), 625-640.
Yunez-Naude, A., & Taylor, J. E. (2001). The determinants of nonfarm activities and income of
rural households in Mexico, with emphasis on education. World Development, 29(3),
561-572.

116