THE EFFECTS OF PARTICIPATION IN FARMER
ORGANIZATIONS ON PRODUCTION EFFICIENCY:
EVIDENCE FROM COARSE GRAIN FARMERS IN MALI
By
Manda Dite Mariam Sissoko

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
2017

ABSTRACT
THE EFFECTS OF PARTICIPATION IN FARMER ORGANIZATIONS ON PRODUCTION
EFFICIENCY: EVIDENCE FROM COARSE GRAIN FARMERS IN MALI
By
Manda Dite Mariam Sissoko
Increasing agricultural productivity is at the core of the Government of Mali’s strategies to reduce
poverty and achieve food security. Dry cereals, such as maize, sorghum, and millet, support the
livelihoods of more than 60 percent of the Malian population. Public interventions to boost dry
cereal productivity have mostly focused on investments to develop and promote farm
intensification strategies. Historical data shows that yields have not responded proportionally. This
study looks at an alternative approach on how to address the problem of low production
performance. It emphasizes efficiency gains in the farm resource allocations, as a sustainable way
to achieve productivity gain. Following a parametric approach, stochastic frontier Cobb-Douglas
production functions for sorghum and maize are estimated. The levels of technical and allocative
inefficiency in the production of these cereals are derived. Furthermore, the role of farmer
organizations, as potential drivers to enhancing production efficiency, is investigated. Using a
Propensity Score Matching approach, the marginal effects of membership in farmer organizations
(FOs) and under different participation regimes are estimated. The results show that overall,
membership in farmer organizations improves technical efficiency and reduces the costs of
technical and allocative inefficiencies. Moreover, the results corroborate the fact that farmers who
participate more actively in their FOs (e.g., group purchase of inputs, group sale or both) are more
efficient. These findings suggest that addressing market imperfections and covering the marketrelated risks could incentivize farmers to adopt efficiency-increasing production decisions.

Copyright by
MANDA DITE MARIAM SISSOKO
2017

I dedicate this work to my parents Mamadou Sissoko and M’Madama Sylla for
the education I received and for being a model of courage and integrity. I am
grateful for your trust, blessings, and love. To my sister Tounko and brothers
Moussa and Fousseyni, I am humbled by your respect and affection. To my friend
and confidant, Mahmud Dicko, I thank you for standing with me in all
circumstances.
In your love, I took ground and found the strength to persevere on the path that
has led me here.

iv

ACKNOWLEDGMENTS
I praise the Almighty “Allah” for the health, the strength and the opportunity I have been granted
with to accomplish this work. Every obstacle overcame, has been possible by your “Grace.”
Indeed, your Grace is sufficient….
I would like to express my profound gratitude to my major professor, Veronique Theriault, for her
patience, the constructive feedbacks and her overall academic supports throughout my Master’s
program. I thank my committee members, Saweda Liverpool Tasie and Judith Whipple for their
invaluable guidance and supervision. Special thanks to Pr. Robert Myers for the insightful advice
and the constructive discussions which have helped me gain clarity on the subject. I also would
like to thank Amidou Assima for his assistance with the data and all the PROMISAM staffs in
Mali. Mr. Jelili Adebiyi (a.k.a Mr. Gana), receive the expression of my sincere gratitude for your
support from the time I applied to MSU. I am profoundly grateful to Ange Kakpo, Awa Sanou,
Aissatou Ouedraogo for helping me settle in and acclimate to the life in East Lansing. To Edward
Opoku, Hamza, Ryan, Tian, Sarah, Ari, David and Christine, I feel privileged to being part of this
learning journey with you!

v

TABLE OF CONTENTS
LIST OF TABLES ..................................................................................................................................... viii
LIST OF FIGURES ..................................................................................................................................... ix
KEY TO ABBREVIATIONS ....................................................................................................................... x
1.

INTRODUCTION ................................................................................................................................ 1
1.1. Motivation ..................................................................................................................................... 1
1.2. Objectives of the study.................................................................................................................. 7
1.3. Contributions from the study ........................................................................................................ 9

2.

LITERATURE REVIEW ................................................................................................................... 11
2.1. Farmer organizations and farm performance .............................................................................. 11
2.2. Cooperative movements in the Malian context: history, opportunities, and challenges ............. 17

3.

CONCEPTUAL FRAMEWORK ....................................................................................................... 24
3.1. Enhanced knowledge of farm management practices ................................................................. 26
3.2. Improved access to inputs ........................................................................................................... 27
3.3. Reduced transaction costs ........................................................................................................... 28

4.

MEASUREMENT OF FARM EFFICIENCY .................................................................................... 34
4.1. Theoretical approach ................................................................................................................... 34
4.2. Empirical approach ..................................................................................................................... 38
4.2.1
Estimating technical and allocative efficiency .................................................................... 40
4.2.2
Estimating the costs of allocative and technical inefficiency ............................................. 43
4.2.3
Modeling the impact of participating in farmer organizations on efficiency ...................... 44

5.

STUDY AREA AND DATA .............................................................................................................. 51
5.1. Study area: .................................................................................................................................. 51
5.2. The Data ...................................................................................................................................... 53
5.3. Descriptive statistics ................................................................................................................... 55

6.

ECONOMETRIC MODELS .............................................................................................................. 59
6.1.
Group membership decision model: ................................................................................... 59
6.2.
Participation decision model: .............................................................................................. 60
6.3.
The Cobb-Douglas frontier production model .................................................................... 61

7.

RESULTS ........................................................................................................................................... 64
7.1. Estimates of the stochastic frontier functions and Technical Efficiency scores ......................... 64
a)
Maize: ................................................................................................................................. 64
b)
Sorghum: ............................................................................................................................. 67
7.2. Allocative inefficiency scores: .................................................................................................... 69
a)
Maize................................................................................................................................... 69
b)
Sorghum .............................................................................................................................. 70
7.3. Marginal effects of Farmer Organizations on the production efficiency: ................................... 72
a)
Maize: ................................................................................................................................. 72
b)
Sorghum: ............................................................................................................................. 75

vi

8.

CONCLUSION ................................................................................................................................... 78

APPENDIX ................................................................................................................................................. 83
BIBLIOGRAPHY ....................................................................................................................................... 95

vii

LIST OF TABLES
Table 1: The decision space defining the participation regimes ................................................................. 49
Table 2: Total cereal production per region, 2015/2016 ............................................................................. 51
Table 3: Summary statistics of farm and plot characteristics (N= 964) ...................................................... 56
Table 4: T-Test statistics comparing farm and plot characteristics of members and non-members of the
FOs .............................................................................................................................................................. 56
Table 5: Summary statistics of production and input use at the plot level.................................................. 57
Table 6: T-test statistics comparing the production and inputs used by members and non-members of the
FO ............................................................................................................................................................... 58
Table 7: Maximum Likelihood Estimates of a Cobb-Douglas Production function with heteroscedastic
inefficiency (half-normal distribution of δu) for Maize .............................................................................. 65
Table 8: Summary statistics of technical efficiency and cost of technical inefficiency ............................. 66
Table 9: Maximum Likelihood Estimates of a Cobb-Douglas Production function with heteroscedastic
inefficiency (half-normal distribution of δu) for sorghum .......................................................................... 68
Table 10: Resources allocation inefficiency in maize production .............................................................. 69
Table 11: T-test for differences in technical efficiency and costs of inefficiency across participation
regimes in maize production ....................................................................................................................... 70
Table 12: Resources allocation inefficiency in Sorghum production ......................................................... 71
Table 13: t-test for differences in allocative inefficiency cost across participation regimes in Sorghum
production ................................................................................................................................................... 71
Table 14: Covariates balance summary for maize ...................................................................................... 73
Table 15: Marginal effects of membership and the participation regimes on maize production efficiency
using the PSM ............................................................................................................................................. 75
Table 16: Covariates balance summary for sorghum .................................................................................. 76
Table 17: Marginal effects of membership and the participation regimes on sorghum production
efficiency using the PSM ............................................................................................................................ 77

viii

LIST OF FIGURES
Figure 1: Production of coarse grains in Mali, from 1985 to 2015 ............................................................... 2
Figure 2: Areas cultivated of coarse grains, from 1985 to 2015 ................................................................... 3
Figure 3: Yields of coarse grains in Mali, from 1985 to 2015 ...................................................................... 3
Figure 4: Framework of farm household-level decision making ................................................................ 25
Figure 5: Frontier production function and technical inefficiency ............................................................. 35
Figure 6: The efficient isoquant and allocative inefficiency ....................................................................... 37
Figure 7: The study area ............................................................................................................................. 52

ix

KEY TO ABBREVIATIONS
GOM

Government of Mali

FO

Farmer Organization

GRPSM

Rural Associations of Producers and Mutual Supports

OPAM

Office of Malian Agricultural Products

SOMIEX

Malian Import and Export Company

CMDT

Malian Company for Textile Development

NGO

Non-Governmental Organization

ON

Office of Niger

OPA

Professional Agricultural Organizations

PAPAM

Agricultural Productivity Growth Project of Mali

PNISA

National Plan of Investment in the Agricultural Sector

ULPC

local union of cereal producers

x

1. INTRODUCTION
1.1. Motivation
Agriculture is the backbone of Mali’s economy, accounting for 40% of its GDP. Agricultural
exports represent more than half of the Total Malian exports (World Bank 2016). Over two-thirds
of the country’s labor force is involved in agricultural activities (INSTAT 2017). About 60% of the
17 million Malians live in rural areas, where agriculture is the primary source of income and
employment (Lazarus 2013). Yet, poverty is still high among rural people. For these reasons, the
agricultural sector remains at the heart of the national strategies aimed at reducing poverty and
promoting economic growth in Mali.
Cereals are the most important agricultural commodities in Mali, both regarding food
security and cropland use. About 90% of farmers produce cereals for subsistence purposes
(Chauvin, Mulangu and Porto 2012). Cereals, such as rice, maize, millet, and sorghum, covered
77% of the total cultivated area (CountrySTAT 2008). These crops provide approximately 65% of
the total food supply (kcal/capita/day) in Mali compared to an average of approximately 45% in
West Africa (FAOSTAT 2013).
Coarse grains, which include maize, millet, and sorghum, are the main staple crops grown
by rural farmers. Millet and sorghum are particularly well-adapted to the harsh agro-ecological
conditions of Mali. They are less demanding of water and other inputs than many other crops,
making them accessible to poor smallholder farmers (CGIAR 2014). Coarse grains account for
about 85% of the total land devoted to cereal crops (CPS/SDR 2016). Current per capita
consumption of coarse grains in Mali is about 168 kg/year compared to approximately 83kg/year
and 14kg/year for rice and wheat, respectively (FAO/GIEWS 2016).

1

Over the last 20 years, maize, millet, and sorghum production have been increasing (Figure
1). During the same period, the cultivated areas have notably increased while yields have remained
stagnant1 (Figures 2 and 3). A Pearson correlation test between the total production of coarse grains
and areas cultivated gives an average correlation coefficient of 0.87, while the average correlation
coefficient between production and yields is 0.59. Thus, the increase of the production of coarse
grains over the last two decades has been a result of land extensification. Overall, the coarse grain
sector has poorly performed regarding productivity gains.

Quantity harvested (tons)

Figure 1: Production of coarse grains in Mali, from 1985 to 2015
2000000
1800000
1600000
1400000
1200000
1000000
800000
600000
400000
200000
0
1980

1985

1990

1995

2000

2005

2010

2015

2020

Year
Millet

Sorghum

Maize

The spike observed in maize yields in 2011 can partially be explained by the cotton sector crisis between 2005 and
2010, which led to a drastic decline in cotton prices and areas cultivated (MAFAP 2013). Maize is cultivated in the
main cotton production zones. It is the second most important cash crop after cotton. Anecdotal stories report that the
drop in cotton prices has led some cotton growers to deviate subsidized inputs destined to cotton to their maize plots.
1

2

Total cultivated areas (ha)

Figure 2: Areas cultivated of coarse grains, from 1985 to 2015
2500000
2000000
1500000
1000000
500000
0
1980

1985

1990

1995

2000

2005

2010

2015

2020

Year
Millet

Sorghum

Maize

Figure 3: Yields of coarse grains in Mali, from 1985 to 2015

Quantity harvested per hectare

4000
3500
3000
2500
2000
1500
1000
500
0
1980

1985

1990

1995

2000

2005

2010

2015

2020

Year
Millet

Sorghum

Maize

Source: (INSTAT 2014)
Mali is self-sufficient in producing coarse grains. However, in the long run, the land-intensive
production of those crops cannot be sustained, first, because “extensification” can be a driver of
natural resource degradation and deforestation. Second, the Malian population is expected to more
than double by 2050. Over half of the population will be urban and per capita income is also
3

expected to grow (Hollinger and Staatz 2015). With population growth and expansion of cities, the
agricultural sector would be in direct competition with other sectors of the economy to access
whatever land remains available. Note that most of the fertile land has already been exploited.
Within the agricultural sector, the limited fertile land available would have to be competitively
allocated across agricultural activities. On the one hand, rice, wheat, meat and milk products are
highly demanded, mainly by urban consumers, as evidenced by their high-income elasticity
(Hollinger and Staatz 2015). On the other hand, coarse grains are pivotal to rural food security and
often seen as the food of the poor. It is thus imperative to increase agricultural productivity,
especially for coarse grains, to ensure that the supply of the staple food meets the demand for the
low-income segment of the population.
As highlighted by Christiaensen, Punam and Aly (2013), poverty alleviation in Africa depends
on increasing productivity of staple crops. They claim that 1% growth in agriculture, induced by
cereals and tubers productivity gains, reduces national poverty by more than 1 % and generates a
higher poverty reduction than if export crops drove the growth. Their claim is consistent with the
fact that most poor smallholder farmers rely on staple crops for their livelihoods.
Increasing agricultural productivity is at the core of the Government of Mali’s (GOM)
strategies to reduce poverty and achieve food security. The PAPAM2 and PNISA3 are two national
programs that translate the political will and commitment to support agricultural productivity
growth in Mali (GOM 2015, GOM 2010). These programs allocate considerable funding to
develop and promote agricultural intensification, including modernization strategies. They support
investments in large irrigation infrastructures, inputs subsidies as well as research and development

2
3

PAPAM: Agricultural productivity growth project in Mali initiated in 2010 and funded for a period of 7 years
PNISA: National Plan for Priority Investment in the Agricultural Sector

4

of new crop technologies (i.e., improved seeds). For instance, through the PNISA, about USD 119
million is budgeted to encourage intensification and modernization of the coarse grain production.
However, given their high costs, these policy interventions are unlikely to be sustained over more
extended periods of time.
Moreover, adoption of agricultural intensification technologies by smallholder farmers in
developing countries continues to be highly constrained. According to Kaminski, Elbehri and
Samake (2013), the persistent low sorghum and millet yields in Mali can be attributed to the lack
of input use as smallholder farmers face difficulties in accessing inputs due to liquidity and credit
constraints. Despite the existence of improved seed varieties and management techniques,
constraints linked to technology adoption, market failures and lack of public and private support
to extension services are significant drivers of the poor performance in the coarse grain sector
(Kaminski, Elbehri and Samake 2013, Staatz, et al. 2011).
Efficiency gains could be a cost-effective way to achieve productivity gains in the Malian
agriculture. Following Farrell (1957), efficiency is defined as the success of a firm in producing
the maximum possible outputs given a set of inputs (this refers to technical efficiency) or as the
success in choosing the optimal cost-minimizing set of inputs given a level of outputs (this refers
to allocative efficiency). In Fried, Schmidt and Lovell (1993), Grosskoft defines factor
productivity growth as the net change in output due to changes in efficiency and technical change.
While technical change captures output growth resulting from the shift in production technology
between two periods of time, efficiency reflects output gains due to the changes in inputs
utilization. Latruffe (2010) identifies three sources of productivity growth: 1) increase in
efficiency, 2) economies of scale, and 3) technological progress. In the context of this study,
smallholder farmers use their inputs in an efficient way when they can produce more outputs with
5

the same level of inputs. Efficiency, thus, implies the possibility of achieving higher results with
the resources already available. It is an indicator of production performance.
Provision of technical assistance and support services, such as credit, can be crucial for
efficiency gains since they facilitate farmers’access to inputs as well as information on better farm
management practices. Bokusheva and Kumbhakar (2008) argue that non-optimal use of
production inputs may result from financial constraints that limit their purchase. Abdallah (2016)
investigated the link between access to agricultural credit and technical efficiency in Ghana and
found that access to credit increases farmers’efficiency by 4%. Furthermore, exposure to extension
services has been found to improve efficiency. Nordin and SĂśren (2017) found that extension visits
contributed to more efficient uses of fertilizer and, thereby, to increased production in Sweden.
Following the liberalization of cereals markets in the 1990s, the GOM has withdrawn from
providing support services to smallholder farmers and has transferred much of its responsibility to
farmer organizations (Staatz, et al. 2011). Farmer organizations (FOs) have played a key role in
providing smallholder farmers with information, technical assistance, and support services. Here,
an FO refers to an “autonomous association of men or women who unit voluntarily to meet their
common economic, social and cultural needs and aspirations through a jointly owned and
democratically controlled enterprise “ (FAO 2012).
Farmer organizations provide their members with a package of services that can induce
productivity and efficiency gains. For instance, dissemination and exchange of knowledge is a core
component of the activities undertaken by FOs and, as discussed above, are efficiency-enhancing
factors. Through their membership to FOs, smallholder farmers can share their farming experience
and learn about new management techniques and technologies from their peers as well as from

6

external experts. Moreover, farmer organizations can facilitate access to input credits and input
subsidies, which relax the financial constraints faced by smallholder farmers and in turn, encourage
input use. They can also provide farmers with input and output price and market information,
which contributes to reducing transaction costs (Verhofstadt and Maertens 2014).
In Mali, farmer organizations have served as an interface with buyers, by grouping production
and negotiating the terms of exchange (Vroegindewey, Theriault and Staatz Forthcoming). Some
FOs with stronger organizational capacity have even been able to negotiate input loans for their
members, make a bulk purchase of inputs, and acquire subsidized fertilizer. Others have provided
farm equipment loans, rental services, crop insurance, and literacy training to their members
(Vroegindewey, Theriault and Staatz Forthcoming). The provision of such supports services
confers to the organizations the capacity to enhance production efficiency of their members.
1.2. Objectives of the study
The broad objective of this study is to investigate whether participation in farmer organizations
(including cooperatives and farmer associations) improves technical and allocative efficiency on
maize and sorghum cultivated plots in Mali. The study aims to identify the effect of membership,
as well as the effect of different types of membership, referred to as “participation regimes.” Four
participation regimes are defined: 1) members who use the FO channel for input provision service
only, 2) members who use the FO channel for group marketing service only; 3) members who use
the FO as a channel for both input and marketing services and, 4) members who do not use any
input or marketing services.

7

To achieve this objective, the study addresses the following research questions:
▪

Q1: What is the average technical efficiency among sorghum and maize smallholder
farmers?

▪

Q2: How are inputs of these crops (i.e., labor, fertilizer, herbicides, and seeds) allocated
vis a vis the optimal use?

▪

Q3: What are the costs associated with technical and allocative inefficiency?

▪

Q4: Does membership in farmer organizations influence technical and allocative efficiency
as well as their related costs?

▪

Q5: Does efficiency differ across different participation regimes?

The hypotheses underlying those research questions are:
▪

H1: The Malian maize and sorghum farmers are operating at a low level of efficiency.

▪

H2: Inputs are non-optimally allocated

▪

H3: There are high costs associated with low efficiency

▪

H4: Being a member of an FO improves both technical and allocative efficiency and
reduces the costs of productive inefficiency.

▪

H5: Farmers who participate in input provision and/or output marketing services are, on
average, more efficient than non-participants.

Answering those questions will shed light on the types of services offered by FOs that sorghum
and maize smallholder farmers most value. It will also inform on the channels through which
membership affects efficiency. This information is valuable to formulate policy interventions
tailored to the needs of the organizations and their members. The ultimate goal is to find ways to
improve the productivity of coarse grains to reduce poverty and achieve food security in Mali.
8

1.3. Contributions from the study
This study makes several contributions to the literature. First, studies on farmers’ decisions to
patronage their organizations (referring to the participation regimes) have not been examined from
an efficiency ‘s point of view. So far, previous studies have explained the motivating reasons
behind the choice of the participation regimes. However, these studies have overlooked the
relationship between the participation regimes and technical and allocative efficiency (Fischer and
Qaim 2014, Mujawamariya, D’Haese and Speelman 2013, Cechin, et al. 2013, Pascucci,
Gardebroek and Dries 2012, Bhuyan 2007).
Second, studies that estimated agricultural production efficiency show only evidence of
membership effect on efficiency scores (Abate, Gian Nicola and Kindie 2014, Addai, Victor and
Gideon 2014, Boubacar, et al. 2016, Debebe, et al. 2015). To our knowledge, none of them have
analyzed the individual effects of different participation regimes on technical and allocative
efficiency.
Third, this study describes in detail the different pathways through which participation in
farmer organizations can influence farm productivity and efficiency. It does so by drawing on
institutional economics theories.
The rest of the thesis is organized as follows: Chapter 2 reviews the relevant literature and
outlines the gap that this thesis aims to fill. Chapter 3 presents the conceptual framework describing
how farmer organizations affect production efficiency. The theoretical and empirical frameworks
for measuring production inefficiency and its costs, as well as the effects of FO membership, are
discussed in Chapter 4. Chapter 5 describes the contextual background, including the historical
evolution of cooperative movements in Mali. Chapter 6 presents the dataset. The empirical

9

estimation models are specified in Chapter 7. Results of the analysis are discussed in Chapter 8.
Main conclusions and key recommendations are drawn in Chapter 9.

10

2. LITERATURE REVIEW
2.1. Farmer organizations and farm performance
The literature widely covers the role of farmer organizations in improving farm performance. One
strand of the literature has analyzed factors that confer those organizations the capacity to mitigate
high transactions costs and market imperfections (Verhofstadt and Maertens 2014, Fischer and
Qaim 2012a, Shiferaw, Hellin and Muricho 2011, B. Shiferaw, G. Obare and G. Muricho, et al.
2009, Markelova, et al. 2009, Bernard and Spielman 2009, Vroegindewey, Theriault and Staatz
Forthcoming). For instance, Verhofstadt and Maertens (2014) mentioned that participation in
cooperatives with joint input purchase and collective marketing could improve small farmers’
bargaining power vis-a-vis input suppliers and output buyers and can reduce transaction costs on
input and output markets. This has been supported by Shiferaw, Hellin and Muricho (2011) who
pointed out that through the range of services provided by the FOs (such as, marketing, financial,
education, advocacy, resources pooling) members are in a better position to reduce their transaction
costs. Moreover, members can better access information and new technologies and compete with
larger farms and agribusinesses.
Moreover, collective marketing can reduce buyers’ transaction costs considerably, by
aggregating products of homogenous quality, which in turn allows FOs to negotiate better prices
for their members. Vroegindewey, Theriault and Staatz (Forthcoming) developed a framework that
built upon transaction cost economics to analyze market coordination choice among cereal
producers in Mali. Their findings show that transaction characteristics, which determine the
presence of high transaction costs, play a significant role in predicting the choice of a horizontal
coordination structure (e.g., marketing cooperatives, bargaining associations or service provider
cooperatives) over other types of market coordination structure. Shiferaw et al. (2009) reported
11

that farmer organizations have the potential to mitigate the effects of imperfect markets by enabling
contractual linkages to input and output markets and by promoting economic coordination in
liberalized markets. According to Bernard and Spielman (2009), group inclusiveness is a major
determinant explaining FOs’ success to providing services to their members effectively. Group
inclusiveness can be determined by the degree of openness to all interested individual farmers, the
extent to which participatory decision-making is conducted and to which all individuals in the
locality benefit from the activities of the organizations.
Another strand of the literature has assessed the effect of FOs on farm productivity.
Findings from these studies indicated that the portfolio of services offered by the organizations to
their members is a significant driver of productivity gain. By offering financial services, FOs can
induce productivity-enhancing investments (Verhofstadt and Maertens 2014). Moreover,
knowledge dissemination through training and extension activities may have a significant and
positive impact on technology adoption and farm management practices, which ultimately result
in higher productivity.
The impact of FOs on production efficiency has also been investigated. For instance,
cooperative membership and exposure to extension services have been found to affect technical
efficiency gains among smallholder farmers positively (Boubacar, et al. 2016, Kelemu and
Workneh 2016, Debebe, et al. 2015, Hailu, Alfons and Bart 2015, Addai, Victor and Gideon 2014,
Abate, Gian Nicola and Kindie 2014, Theriault and Serra 2014, Jaime and CĂŠsar A. 2011). In their
study on rice farms in South-Western Niger, Boubacar et al. (2016) used a data envelopment
analysis approach and a Tobit regression to estimate technical efficiency and identify its
determinants. Their results showed an average technical inefficiency of 0.52, meaning that rice

12

producers could decrease their inputs use by 52% while keeping their outputs at the same level.
Their results also showed evidence of the mitigating effect of a cooperative on inefficiency.
Debebe et al. (2015) used a parametric approach to estimate technical and allocative
efficiency among maize farmers in Ethiopia. Assuming a Cobb-Douglas production function, their
stochastic frontier and dual cost frontier estimates indicated an average technical and allocative
inefficiency of 38% and 43%, respectively. The results from the Tobit regression model, in the
second stage of their estimation, also suggest that cooperative membership significantly affects
both technical and allocative efficiency. In line with these studies, Addai, Victor and Gideon (2014)
and Abate et al. (2014) used a propensity score matching approach, to draw similar conclusions
regarding the effects of FOs on production efficiency. Likewise, Theriault and Serra (2013)
conducted a parametric estimation of technical efficiency and its determinants using a CobbDouglas frontier production function for Malian cotton growers. They estimated an average
technical efficiency score of 72% and found that services provided by cotton cooperatives, such
as access to credit, extension services, and timely payment, are significant determinants of
efficiency. They concluded that policies aiming at reducing farmers’ financial stress and improving
access to input-credit markets should be encouraged to improve technical efficiency in West
Africa.
The differential effects of different types of organizations on the farms’ performance have
also been addressed in the literature. Verhofstadt and Maertens, (2014) identified a broad typology
of cooperatives. These include 1) marketing cooperatives, in which farm outputs are sold
collectively; 2) production cooperatives, in which production resources are pooled and; 3) supply
cooperatives, in which production inputs are collectively acquired. The diversity of the
organizational forms and thereby, the activities were cooperatively undertaken, may lead to
13

different effects of FOs on farm performance. As highlighted by the authors, the effects of different
organizational forms on the farms’ performance have not been sufficiently covered in the literature.
The study carried out by Vroegindewey, Theriault, and Staatz (Forthcoming) in the context of the
Malian cereal sector also pointed out the existence of different types of FOs regarding their roles
and activities undertaken at the village-level. For instance, the authors distinguished between
bargaining associations, marketing cooperatives, and service provider cooperatives. While the first
two types intervene as an interface between producers and potential buyers to facilitate the linkages
to downstream markets, the latter primarily assume the provision of training, education and credit
services. Francesconi and Heerink (2011) examined the differential impact of Ethiopian
cooperatives on the marketed share of outputs across different types of organizations and found
that members of marketing cooperatives exhibited significantly higher marketed share as
compared to non-members.
However, when farmers belong to livelihood organizations (or service provider
organizations), the membership effect on the marketed share was insignificant, or even negative,
in some instances. In meso-America, Hellin, Lundy and Meijer (2009) analyzed whether forming
an organization is beneficial to producers of undifferentiated crops, such as maize, compared to
high-value crops, such as vegetables. The authors argued that because of low transaction costs in
the maize sector, FOs have a greater impact in facilitating access to inputs (e.g., improved seeds)
than access to output market. In contrast to maize, tomato production is more subject to volume
and price volatility due to its high level of perishability, which contributes to higher transaction
costs. As such, only medium to large producers with high capital endowment can operate in the
sector. For these reasons, the authors considered farmer organizations to be more beneficial to the
vegetable sector than the staple crop sector regarding output market access.

14

One of the main threats that weaken farmer organizations and jeopardize their long-term
sustainability is the low level of active participation in the organization’s activities, which we refer
to as “patronage decision.” This leads to another strand of literature, which looks at factors that
affect the decision of the members to patronize the services provided by their organizations (Wollni
and Fischer 2015, Mujawamariya, D’Haese and Speelman 2013, Cechin, et al. 2013, Pascucci,
Gardebroek and Dries 2012, Bhuyan 2007, Fulton and Giannakas 2001, Klein, Richards and
Walburger 1997, Fulton and Adamowicz 1993). The word “patronage” is used in this literature to
refer to members’ participation in the activities of the organization or the actual use of the services
offered by the organization. Bhuyan (2007) analyzed the determinants of the patronage behavior
and found that members’ socioeconomic characteristics (such as farm size, off-farm income, farm
income, duration of membership) and their attitude toward the cooperative management system
have a significant effect. On the other hand, Klein, Richards and Walburger (1997) found that older
Albertan farmers with more landholding are more likely to patronage their cooperatives. Higher
prices for farm outputs and refunds (rebates) have also been found to positively influence farmers’
decisions to patronize their cooperatives (Fulton and Adamowicz 1993, Fulton and Giannakas
2001). Recently, Wollni and Fisher (2015) have argued that liquidity constraint, and the discounted
value of late payments from collective marketed output are strong determinants of the patronage
decision.
While previous studies on the patronage decision have focused on analyzing its determinants
both as a discrete and continuous decision-making, none of them has investigated its impact on
farm performance indicators, such as technical and allocative efficiency. Moreover, previous
studies on its determinants have not disaggregated between different levels of patronage decisions
(a.k.a participation regimes). First, some members may find it more advantageous to use the input
15

provision services offered by the FOs only while relying on other marketing channels for their
outputs. For instance, for farmers with diversified crops, collective marketing may result in higher
opportunity costs if their FOs focus on specific crops or if better marketing alternatives are
accessible to them (Fischer and Qaim 2014). Second, some members may be interested in
collective marketing services only because they face high transaction costs in selling their outputs.
Third, some farmers may face market failures in both input and output markets and thus,
patronaging the FO for both input provision, and collective marketing may be an optimal decision.
Fourth, some farmers may only value the information and education services offered by the FOs
and therefore, do not patronage either for input provision or access to output markets. Perhaps
those members are wealthier farmers with a stronger social network, less liquidity-constrained,
more entrepreneurial and thereby, the reasons for their membership to FOs are to enhance their
knowledge. Alternatively, they may keep their membership status for non-economic reasons, such
as conformity to social norms or loyalty to the organization. How these decisions translate into
economic performance, such as technical and allocative efficiency, have been overlooked in the
literature.
One exception is Klein, Richards and Walburger (1997) who analyzed discrete and continuous
decisions to patronage cooperatives for both inputs (e.g., seed, fertilizer, fuel) and marketing (e.g.,
grain sales) services. They analyzed factors that determine the decision to use the cooperative
channel for input procurement and output sale as well as factors that explain the amount of inputs
and outputs transacted through the cooperative channel. Their results show that older farmers with
large farms tend to patronage more the cooperatives both intensively and extensively. Farmers
‘perception about the cooperatives’ performance and competitive prices that they offered have also
been found to be major determinants of the intensity of the patronage behavior. However, their
16

analysis was limited to only identifying farmers’ socioeconomic characteristics and attitudes that
predict the different levels of the patronage decision. The effects of such behavior on production
performance were not explored.
2.2. Cooperative movements in the Malian context: history, opportunities, and challenges
Current rural organizations in Mali are the legacy of post-colonial cooperative movements initiated
by the government to structure the rural economy. The first forms of cooperatives emerged under
the socialist government regime of the first Republic (1960-1968) with the creation of the GRPSM
(“Groupements Ruraux de Producteurs et de Secours Mutuels”) at the villages level. The GRPSM
reflected the government ideology of a modernized agriculture following the collectivist socialism
ideals. Participation in the GRPSM was mandatory, and each village possessed a “collective field”
managed by the members of the GRPSM (Tag 1994). Surplus generated from the “collective
fields” were invested in public services (e.g., roads, schools and hospitals).
Another function of the GRPSM was the provision of consumption goods to both urban
and rural population in coordination with parastatal agencies such as OPAM4 (Office of Malian
Agricultural Products), SOMIEX5 (Malian Import and Export Company), CMDT6 (Malian
Company for Textile Development) (Traore 1993). Before the liberalization of markets, OPAM
was the parastatal agency controlling the cereal markets and possessed a legal monopoly over
cereals trade. The CMDT was the equivalent of OPAM operating in the cotton sector. SOMIEX
was a state company in charge of the imports and exports of agricultural and manufacturing goods.
The undemocratic nature of the GRPSM management and the strict control over prices by

4

OPAM: Office des Produits Agricoles du Mali
SOMIEX: Societe Malienne Import Export
6
CMDT: Compagnie Malienne du Development des Textiles
5

17

parastatal agencies led to the disengagement of rural producers, causing poor production
performance on fields managed by GRPSM compared to collective and individual fields managed
by the household’s head and household members.
In 1982, the government of the second Republic instituted the “Tons Villageois,” a village
level association, to represent the primary legal, organizational unit of rural communities (Traore
1993). They were instituted as a new model of “cooperative development” with a more democratic
character and intended to correct the weaknesses of the previous GRPSM models. The “Tons
Villageois” were then used as an institutional tool to promote and manage economic, social and
cultural development in rural areas. The main objectives assigned to these organizations were to
1) increase the productivity of rural farms by facilitating individual or group access to production
inputs; 2) the organization of the bulking, storage, and commercialization of farm products; 3) the
provision to members of consumption goods, equipment, and services; 4) the organization of
collective savings to enable access to credits supporting production and consumption. The
activities of the “Tons Villageois” were kept inscribed in the political agenda and closely related
to the operations of parastatal agencies that had a legal monopoly on production and trade of the
main agricultural products (such as cereals and cotton). The parastatal agencies were at the same
time the only buyers of agricultural products and the main suppliers of services to the “Tons
Villageois”- private operations were restricted. To accommodate the “Tons Villageois” to the
democratic principles, the government allowed the creation of voluntary rural organizations in the
form of “Village Associations” as transitory institutions to the establishment of the Ton Villageois.
The “Village Associations” have existed since the 1970’s. In contrast to “Tons Villageois,”
they have been created on a voluntary basis. They are informal associations of people with no legal
status. The initial goal of the creation of the “Village Associations” was to let them mature and
18

acquire enough credentials in cooperative practices to upgrade to the status of a “Ton Villageois”
(Bélières, et al. 2008). Thus, the Village Associations” before 1990, were pre-cooperative
organizations with a transitory status. The first forms of these types of organizations were
promoted by CMDT in the mid-1970s in the cotton production area (including the Southern Mali
zone). Their functions were centered on the commercialization of cotton, especially the grading,
weighing and transportation of cottonseed (Theriault and Sterns 2012). They also stood as the
primary channel for farmers to access inputs and technical assistance provided by CMDT
(BÊlières, et al. 2008, Theriault and Sterns 2012). At that time, very few of them adopted the legal
status of “Ton Villageois,” mainly to avoid the heavy administrative and legal requirements
expected from such cooperative movements during the State interventionism period.
The early 1990’s experienced the reform and restructuration of state-interventionism
policies with the liberalization of cereal markets. These reforms translated into the retreat of the
State from production and marketing activities in the Malian cereal sector and the transfer of the
market functions to the private sector. The roles of OPAM became limited to managing the national
security stock and providing market information services to the private sector. The implications of
these market reforms became the necessity to empower the rural organizations to take over the
functions previously fulfilled by the State Agencies (Mercoiret 2006). These contextual changes
led to the emergence of multiple forms of organizational movements and the engagement of NGOs
(Non-Governmental Organizations) to support and defend the interests of rural producers. The
Village Associations, primarily established to support cotton production, evolved into several
specialized organizations (including village-level cooperatives and regional and national unions
and federations) around different agricultural value chains (including coarse grain, rice, livestock,
and cotton). Given their strategic and economic importance for the country, as main export crops,
19

cotton and rice farmer organizations continue to benefit from direct government supports through
inputs subsidies, credit and technical assistance provided by the CMDT and ON7. For cereal farmer
organizations, such supports are mainly obtained from NGOs or research and development projects
with the resulting downside of a dependence of some FOs lifespan on the duration of the projects
supporting their activities.
In 2001, the government reformed the status of cooperative movements and established a
legal framework that promotes autonomy for producers in the realization of their collective
economic and social needs. The term “OPA”(Professional Agricultural Organizations) is the actual
terminology used to refer to any groups of people with an agricultural vocation, who voluntarily
unite to advocate their interests before public authorities and ensure the provision of goods and
services to the members (CRA n.d.). The “OPAs” can assume different functions: production (e.g.,
inputs services), marketing (e.g., storage, group sale, and market information), credit and financial
assistance, representation and advocacy of members ‘interests. From the article 27 of the Malian
agricultural policy (LOA 2006), the “OPA” denomination includes cooperatives, village
associations, unions of cooperatives, federations, confederations, and syndicates. Village
associations, cooperatives, and their unions are the types of “OPA” concerned in this study. The
difference between these types of organizations depends on the jurisdiction that defines their legal
status and the scope of their activities. Unlike village associations, cooperatives have a legal status,
a more sophisticated governance structure and are financially autonomous (Vroegindewey,
Theriault and Staatz Forthcoming, Theriault and Sterns 2012).

7

The ON (Office du Niger) is a government agency in Mali that manages a large irrigation scheme supporting rice
production in Segou region

20

The rapid changes occurring in the agri-food system of developing countries, including the
consolidation in the downstream sector into supermarket chains and the emergence of many
modern mini retail stores and processing factories create new challenges to the rural food
production systems. For instance, farmers must meet the quality and quantity requirements of a
downstream sector characterized by an increasing market power, which can be costly (Reardon, et
al. 2000). While some of these costs come as threshold investment costs, others are transaction
costs. In this context, it is essential to identify institutions and modes of rural organizations that
should govern the food production activities while mitigating these challenges (MĂŠnard and
Valceschini 2005). Since the retreat of the government from cereal markets, farmer organizations
have been the main (if not the only) institutional support available to coarse grain growers, who
continue to face issues of high transaction costs and inefficiency in production.
These organizations have considerable potential to scale up production in the coarse grain
value chain while minimizing transaction costs (Kaminski, Elbehri and Samake 2013). Take for
instance the three cases of the local union of cereal producers (ULPC), Faso Jigi, and the Cereal
Banks in the Malian cereal production zone. The local union of cereal producers (ULPC) in the
Dioila district, offers inputs (e.g., fertilizers, pesticides, seeds) to its members at the beginning of
the cropping season, bought from its own funds. It sets a reference price, which is an average
output price from three villages plus a premium (Kaminski, Elbehri and Samake 2013). After
harvest, farmers deliver part of their production to the union to repay their credits. The union
borrows money from banks or micro-credit institutions to purchase cereals from members and
organizes grouped sales with higher prices provided to the supplying farmers. Beyond the input
credit repayment scheme, members can sell their grains to the union, which organize the collective
marketing.
21

Faso Jigi in Koutiala is another producer organization providing input credits to farmers in
the form of payment advances on the quantity they committed to deliver, based on a fixed base
price at the beginning of the season. The payment advances allow farmers to make investments to
procure inputs for the cropping season. The remaining payment is made after harvest with
deductions of interests. Collective outputs are sold to local retailers and processors, sometimes
under marketing contracts with premium prices. This type of grouped marketing can be a source
of income improvement and stability for farmers (Kaminski, Elbehri and Samake 2013).
The “Cereal Banks” are local development instruments used by rural organizations to
address the issue of food insecurity collectively. The “Cereal Banks” act as a sort of village-level
cereal warehouses, which is managed by the village associations. After the harvest period, the
banks purchase grains from producers (members and non-members of village associations), collect
and store them in the warehouses. The grains kept until the food shortage period are then sold to
the local population at non-speculative prices (lower than market price) (AfriqueVerte n.d.).
Priority is given to members of the associations managing the banks for the purchase of the grain.
While the primary objective of the “Cereals Banks” is food security, a crucial implicit role of this
type of collective action is to smoothen consumption risks. For most smallholder farmers in
developing countries who are both buyers and sellers of staple crops, the needs to smoothen
consumption can be a considerable obstacle to investing in the production activities. Therefore,
participation in cereal banks can decrease farmers’ risk averseness toward consumption and
increase their willingness to make productivity and efficiency increasing investments.
The diversity in rural organizations as well as in their functions (input provisions, group sale,
food security, resources pooling) raises the question of how these organizations address the needs
of smallholder farmers in this “market economy” environment. Addressing this question requires
22

an analysis of the effects of farmers’ participation in FOs on their farming activities. Since
efficiency gains can be an important determinant of performance in a market economy, the focus
of this study is to examine, to which extent participation in FOs can induce technical and allocative
efficiency gains and reduce the costs of inefficiencies.

23

3. CONCEPTUAL FRAMEWORK
Technical efficiency refers to the capacity of the farm household to achieve the maximum possible
output (the frontier output) with a minimum use of inputs. Alternatively, a technically efficient
firm maximizes output with a given level of inputs. Allocative efficiency refers to the ability to
choose the optimal input quantities that minimize costs, given the level of output and the relative
factor prices. A firm’s ability to achieve higher efficiency is conditioned on the decision on its
input utilization.
Here, the framework of farm household-level decision making developed by FAO (1995)
is adapted to illustrate the potential links between input allocation decisions and participation in
farmer organizations. As seen in Figure 2, the economic activity of a farm household involves two
main types of management decisions: 1) investment and marketing decisions and 2) production
and conservation decisions. The first type of decisions focuses on farm product marketing
decisions. The second type of decisions is related to the adoption of farm management practices,
such as soil conservation and crop management practices as well as resource allocation decisions,
such as labor, land, capital, and input acquisition. These two types of management decisions are
influenced by both internal (e.g., farmers’ socioeconomic characteristics, biophysical
characteristics of the farm) and external factors to the farm household (e.g., markets, policies,
support services, technical information).

24

Figure 4: Framework of farm household-level decision making

Markets
conditions
(local, national,
international)

Support services
(infrastructure,
credits, supply of
inputs)

Socioeconomic
characteristics

MARKETING DECISIONS

PRODUCTION DECISIONS
•
•
•
•
•

Crops to grow
Land, labor allocation
Capital
Types/ quantity of
Inputs
Inputs sources

•
•
•

FARM HOUSEHOLD

Biophysical
factors

crops to sell
Quantity to sell
market channels

Technical
information
(indigenous and
introduced)

Policies, rules,
regulations

Adapted from FAO (1995)
Decisions taken by the household
Internal factors
Influence of external factors on household

External factors

Following this framework, FOs can affect production efficiency through their ability to
create an enabling environment for optimal decision- making regarding both farm inputs
investment and utilization. For instance, Falkowski and Ciaian (2016) explained that when farmer
organizations succeed in insuring against the risks of contract hold up faced by farmers, they can
indirectly contribute to moving farm investments closer to the optimum. The FOs can affect farm
productivity or efficiency by reducing the investment risks on productive resources. When farmers
expect that their investments will leave them vulnerable, they are unlikely to make these
investments in an efficient way (Falkowski and Ciaian 2016). As farmer organizations reinforce
the bargaining power of individual members, they mitigate the market risks faced by farmers and,
25

hence, promote optimal investments and allocation of farm resources. Besides the influence of
FOs on market-related factors, the provision of support services and the dissemination of technical
information by FOs constitute other factors that can lead to efficiency gains.
As mentioned previously, this study hypothesizes that farmers’ participation in farmer
organizations, among other factors, has a positive effect on their productive efficiency by reducing
output losses due to both technical and allocative inefficiencies. Drawing from the literature, there
are three pathways through which the effect of FOS on efficiency can be seen: 1) knowledge and
information dissemination; 2) increased access to farm inputs and; 3) reduction of transaction
costs. The next sections elaborate on those pathways and how they can increase technical and
allocative efficiency.
3.1. Enhanced knowledge of farm management practices
Membership in farmer organizations provides a setting for learning. As such, it influences
members’ productive efficiency by facilitating their access to technical and market information.
Indeed, improving farmers’ access to technical training and advisory services is a function that
FOs aim to perform. Farmer organizations have made key contributions to the provision and
enhancement of extension services by facilitating linkages between extension providers and
smallholder farmers (FAO 2010). These advisory or extension services might originate from
external providers (such as government extension officers, NGOs or development projects) who
are more willing to work with groups of farmers than individuals (Shiferaw, Hellin and Muricho
2011).
Farmer organizations have bridged the gap between local agricultural knowledge and
technical research knowledge. The flows of information received through either extension services
or from other members, enhance learning about optimal input combinations that maximize the
26

farm outputs. Farmers may learn about the optimal timing for applying inputs, which all contribute
to better production outcomes. Exposure to the extension services may also make them aware of
and more willing to adopt, for instance, new cropping practices, including soil fertility
management practices, that improve land productivity (Junge, et al. 2009, Somda, et al. 2002).
Previous studies have identified two sources of uncertainty that constraint farmers’
willingness to adopt new farm technologies (e.g., soil conservation practices and improved
seeds):1) their knowledge about the optimal input mix needed to realize the maximum output level
and; 2) their expectations about the profitability of the new technologies. Learning can be a crucial
determinant of adoption (Foster and Rosenzweig 1995, Conley and Udry 2010, Bandiera and Rasul
2006, Besley and Case 1993, Ma and Shi 2015, Munshi 2014). Members of farmer organizations
can learn from the experiences of their peers (social learning) on the optimal ways to utilize a given
farm technology. By reducing the risks perceived in the adoption process of a new farming practice
and by improving farmers’ own experience, the learning dynamic within FOs can be an important
contributor to members’ production performance. As a platform for social learning, information
sharing, and capacity building, farmer organizations can be a source of improvement in technical
and allocative efficiency.
3.2. Improved access to inputs
Production efficiency is built upon input decisions. Thus, improved access to inputs, both
physically and economically, is critical to achieving higher allocative efficiency. Indeed, ceteris
paribus, farmers with better access to inputs are expected to have higher allocative efficiency than
farmers with more limited access. Everything else equal, farmers with limited access to farm inputs
are more likely to underinvest in those resources than peers whose access are less constrained.
Two factors may constrain farmers’ access to inputs: 1) liquidity constraint and; 2) transaction
27

costs. While the effect of the former on access to inputs is obvious and straightforward, the effect
of the latter will be explained more in details in the next section. Participation in farmer
organizations can mitigate both constraints. However, in contrast to the learning benefits procured
by the membership status, here members must decide to use the FO as a channel for input
procurement and/or output sale to overcome liquidity constraint and transaction costs.
Most FOs distribute inputs, such as chemical fertilizers, pesticides or seeds to their
members under different procurement systems. Input credit is a type of input procurement systems
that some FOs make available to their members. Another procurement system offered by FOs is
the provision of subsidized inputs. The FOs usually have facilitated access to input subsidies
through their collaboration with NGOs, government extension services, and local development
projects. However, while enhancing members’ access to inputs, subsidies may lead to undesirable
effects on productive efficiency. Subsidies create distortions to the free-market economy and as
such, affect the relative factor prices that inform optimal input decisions.
Therefore, farmers with access to input subsidies may report higher allocative inefficiency
than others, assuming the subsidies affect relative prices. Nevertheless, input credits and subsidies
are mechanisms through which FOs may relax the liquidity constraint faced by members.
Consequently, organizations can help improve members’ access to productivity-increasing inputs.
Considering the criterion of improved access to inputs, membership to FOs can be a source of
improvement in productive efficiency, particularly the allocative efficiency.
3.3. Reduced transaction costs
Transaction costs incurred by farmers on both output and input markets can be another potential
source of inefficiency for similar reasons (accessibility criterion) discussed above. A good

28

understanding of the effects of transaction cost on inefficiency and of the way participation in FOs
can mitigate those effects depend on knowing first what transaction costs are.
The theory of transaction costs is anchored in the seminal work of Coase, ''The Nature of
the Firm''. Coase (1937) was the first to recognize that market exchange is not costless and that the
rationale of the existence of a firm is to internalize some of its activities (through vertical
integration), which would otherwise generate higher costs if transacted through the open market
system. Firms seek to minimize their costs. Following the New Institutional Economics (NIE),
transaction costs are the costs associated with the use of market institutions for the transfer of
property rights between parties (Garfamy 2012, MĂŠnard and Valceschini 2005). Since market
transactions are costly, the firm arises as an alternative institution, which internalizes the
organization of transfers.
Any exchange of goods and services under the open market is subject to costs shared by
both parties involved in the exchange. Here, the definition of transaction costs adopted by Coase
(1937) and later summarized in Hobbs (1996) and Hobbs (1997) is used. Coase defined transaction
costs as the costs of discovering what prices should be, the costs of negotiating individual contracts
for each transaction and the costs of accurately specifying the details of a transaction in a longterm contract (contract enforcement). Hobbs (1996, 1997) provided three main classifications of
these costs:
•

Information costs: costs of searching information about products, prices, inputs, buyers,
and sellers. The price discovery costs depend on the extent to which there are easily
accessible sources of market price information.

•

Negotiation costs: costs incurred from the physical act of the transaction. These costs
include transportation costs, both the monetary value and the opportunity costs of the
29

producer time and effort of organizing the transportation of the goods to the market (or
buyers). Other coordination costs include assembling the output and storage costs.
•

Monitoring and enforcement costs: arise after the exchange has been negotiated and
may include the monitoring of the quality of goods from a supplier and the monitoring
of the buyer behavior to ensure that all the pre-agreed terms of the contract are met.

Although the transaction cost literature has focused mostly on the output market side, all three
types of transaction costs (as defined above) can be observed in input markets.
Williamson (1981) defined three transaction attributes that influence costs: i) the frequency
to which transactions occur, ii) the uncertainty surrounding the transactions, and iii) the degree to
which transaction-specific investments are required to realize least costs supply (a.k.a. asset
specificity). Menard (2007) and Garfamy (2012) explained that relative costs of using markets for
exchange depend on these attributes of the transaction. Asset specificity can arise in three different
ways: site specificity (characterized by the degree of immobility of resources invested), physical
asset specificity (technological advantage) and human asset specificity (arises from learning by
doing, the know-how advantage). In general asset’s specificity is reinforced when the resources
invested in a specific production sector generate low alternative values when invested in a different
sector.
Coase (1937) attributed the existence of transaction costs to the limitations of the
neoclassical theory, which assumes perfect competition in a frictionless economy. Hobbs (1996)
grouped these limitations under what he called the key concepts supporting transaction cost
analysis: bounded rationality, opportunism, asset specificity (as defined by Williamson), and
information asymmetry. Bounded rationality implies that, although individuals intend to take
rational decisions, their capacity or ability to explore all possible decision options is limited.
30

Opportunism refers to the risk that some individuals or firms involved in transactions will seek to
exploit the situation to their advantages (for instance, when few alternative suppliers are available
to buyers or vice versa). Finally, the asymmetrical distribution of information among parties of an
exchange arises when all parties to the transaction do not possess the same level of information.
Parties might all have access to public information, but access to private information is limited to
a selected group of people. Information asymmetry creates a situation of uncertainty which may
lead to opportunistic behavior. Garfamy (2012) extended the list of factors explaining the costs of
using market for exchanges by adding the following: low bargaining power (or the degree to which
a firm has alternative suppliers or buyers to meet its needs) and loss of resource control (when a
firm outsources a product it should naturally own or produce). Following the author, transaction
costs can be conceptualized as a function of the cost-determining attributes of the transaction
(frequency, uncertainty, asset specificity) and factors creating transactional difficulties (bounded
rationality, opportunism, information asymmetry, bargaining power, and resource control).
With this theoretical review in mind, we now turn to how transaction costs may affect the
allocative efficiency and how membership to FOs can mitigate their effects. The optimal condition
defining allocative efficiency in production is an input quantity choice, such that the marginal
value product equals the marginal cost of the inputs (pf1= w1; pf2 = w2), alternatively the ratio of
the marginal products equals the ratio of the prices (f1/f2 = w1/w2). Transaction costs may affect
allocative efficiency through their distortion effects on output prices, p, or on the relative factor
prices, leading to market failures. In rural economies of developing countries, transaction costs are
the principal drivers of market failures (Alene 2008, Key, Sadoulet and De Janvry 2000). When a
farm household participates in markets to sell and buy a given food crop or service, the difference
between the sale price and buying price of the commodity determines a ''price band'' within which
31

it can be advantageous for the household to participate. If within the price band, the household’s
shadow price (its valuation of the good or service) is less than the purchase price and higher than
the sale price of the good on markets, the household will not participate. Market failures arise when
the shadow price of a commodity or service falls within the price band (De Janvry, Fafchamps and
Sadoulet 1991). Transaction costs can lead to market failures by enlarging the width of this price
band in such a way that farmer’s supply price of a commodity is higher than the prevailing market
price, and his/her demand price is lower than the market price. Transaction costs thus reduce the
price of a commodity or service sold by the farmer and increase the purchase price of the same
goods relative to the market price. Consequently, exchange becomes disadvantageous for the
farmer in either case.
Market failures can apply to both output and input goods. For instance, most of the farm
households in developing countries purchase on markets (especially during the period of food
shortage) the same food crops that they grow on their farms. In this case, high transaction costs
will affect output prices. On the other hand, households may participate in the labor market by
supplying their labor force or by hiring external labor force. For this latter case, the differences
between earned wages from supplying the labor force and the costs of hiring external labor force
determine the transaction costs that may create failures in this input market. Thus, through these
distortion effects on prices and given that costs are household- specific, high unobserved
transaction costs incurred by the farm households can be responsible for the observed allocative
inefficiency.
Due to unobserved transaction costs, farmers can look inefficient in their input decisions, even
though their decisions might reflect a rational choice. The mode of operation of FOs can reduce
the distortion effects on input decisions due to the presence of high unobserved transaction costs.
32

The reason is that; FOs are more capable than individual farmers to address some of the issues
regarding the transaction attributes inherent to market exchanges (i.e., the seven transaction
attributes beforehand mentioned) that lead to higher costs. Membership and participation in
different activities of the organizations (e.g., input procurement and group marketing) may reduce
uncertainty and opportunism behavior surrounding transactions. By bulking outputs, pooling
resources, organizing group sales, offering guaranteed output prices, and ensuring access to inputs,
participation can lower uncertainty and risk for members.
Such collective activities can also contribute to addressing the bargaining power and the issues
of control over the resources (Verhofstadt and Maertens 2014). The learning dynamic that takes
place within the organizations, as well as the education services, can address the information
asymmetry and bounded rationality problems when it comes to looking for low costs or highquality inputs and better market opportunities for outputs (Shiferaw, Hellin and Muricho 2011).
The effect of membership and participation in FO’s activities on asset specificity and
frequency of transactions can be nuanced. For instance, when production is highly diversified for
a given farmer, collective marketing may result in high opportunity costs, especially if the
organization concentrates on some specific crops only or better alternatives are available to farmers
(Fischer and Qaim 2014). In this case, dealing with FOs can increase members’ asset specificity
and inflate the transaction costs incurred. On the other hand, when members are isolated or located
in remote villages, participating in group activities can reduce the costs related to the location
specificity since FOs can help to bulk and coordinate the transport of members’ outputs, generally
a village level warehouse. In this case, the location specificity effect on transaction costs is
attenuated.

33

4. MEASUREMENT OF FARM EFFICIENCY
4.1. Theoretical approach
The standard producer theory provides the basic framework for production efficiency measures.
Farrell (1957) has been the first to introduce the concept of productive efficiency. He distinguished
two components of productive efficiency: technical efficiency (TE) and allocative efficiency
(AE). In the attempt to provide a theoretical representation of these two concepts, let us assume
that the farm household produces a single output, Y, using a bundle of inputs X (x 1, x2). A
theoretical representation of the farm production function is given by its production possibility
frontier, F, which delineates the set of all possible outputs the farm can realize with the production
technology available for a different level of the input bundle X (Schmidt 1985).
In Figure 5, each level of X represents a combination of inputs x1 and x2 used to produce
output Y. The production frontier, F, determines the output of a fully technically efficient farm
household, meaning the set of maximum possible output levels for each level of input bundle X.
The set of points represented by, v, reflect any random shocks to the production; any deviation
from the frontier output caused by factors outside of farmer’s control (e.g., weather and economic
shocks). If at input bundle level, Xa, the farmer produces output level, Y0, his/her technical
inefficiency, u, is represented by the deviation from the production frontier function while taking
into account random shocks (v) to the production. At the input level, Xa, the fully efficient output
level would have been to produce the amount Y1. The term, u, measures the Output-Oriented (OO)
technical inefficiency of the production given the technology available to the farmer. The OO
technical inefficiency is interpreted as the proportion by which outputs can be increased, holding
the level of inputs constant or the losses of outputs due to inefficiency.

34

Alternatively, the level of inputs can be reduced from Xa to Xb while holding the output at
the same level Y0 and moving to the frontier function, F. The proportion by which the level of
inputs can be reduced holding outputs constant at its efficient level represents the Input-Oriented
(IO) technical inefficiency. In Figure 5, the IO technical inefficiency is represented by, Ρ, and
expresses the overuse of inputs due to inefficiency or the increase in input costs resulting from
technical inefficiency.
Figure 5: Frontier production function and technical inefficiency

V (random shocks to production)
y1
F

u= Technical inefficiency
y0

Ρ
X
Xb

Xa

There is no significant distinction between OO and IO technical inefficiency measures,
except that they represent alternative perspectives of technical inefficiency. As pointed out by
Kumbhakar, Wang and Horncastle (2015), the choice of examining IO or OO technical
inefficiency depends on whether outputs or inputs are considered exogenous. For instance, in a
sector where the output level that each firm can produce is regulated (e.g., production quota), the
firm optimization decision-making will be reduced to choosing the level of inputs that maximize
outputs or reduce the costs of production. In this case, estimating IO-technical inefficiency
35

becomes more suited. In the framework of this study, the level of outputs that each farm can
produce is not restricted or regulated, therefore estimating the OO technical inefficiency makes
more sense. However, since the estimation of the costs of inefficiency is one of the objectives of
this study, the IO technical inefficiency will also be estimated as it offers a more straightforward
interpretation of the costs induced by technical inefficiency.
One can also assume either a heteroscedastic or a homoscedastic technical inefficiency. A
homoscedastic technical inefficiency implies that the variance of the firm’s inefficiency (δu) is
constant. In contrast, a heteroscedastic inefficiency assumption implies that the variance of u varies
in function of some variables. For this study, the output-oriented heteroscedastic technical
inefficiency is assumed. Following Kumbhakar, Wang and Horncastle (2015), a production
function with OO technical inefficiency can be mathematically written as:
Y = F(X).ev-u, u ≥ 0
Where u measures the OO technical inefficiency, e-u is the measure of OO technical efficiency
(TE) and 0 ≤ (e-u) ≤ 1. A fully efficient farmer has a value of e-u equals to 1.
Regarding the allocative inefficiency, let us consider Figure 6, where the curve y0 represents
the efficient isoquant of the firm or the isoquant of its production frontier. On the curve y0, each
combination of the bundle of inputs X(x1, x2) yields the output level, y0, which is the maximum
feasible outputs given the technology available. Note that, while the efficient isoquant can be
conceptualized in theory, in practice, it is not observed and can only be estimated.
Now, let us assume that the slope of the line, P, represents the input price ratio (w1 /w2), so that
line P gives us the isocost curve of the production. Following the producer theory, the optimal
quantities of inputs to use are given at point C, at the tangency of the isoquant and isocost lines.
36

Figure 6: The efficient isoquant and allocative inefficiency
X2

Allocative Inefficiency= Îť
B

A

C
y0

P
X1
Xb

Xc

Xa

At point C, the farmer allocates the inputs in a way that minimizes its costs, while
producing a technically efficient output. Therefore, the farmer is both technically and allocatively
efficient at point C. The ratio of the marginal product of the two inputs is equal to the ratio of the
prices of the inputs, f1/f2 = w1/w2 (optimality condition). At point B, this equality is not satisfied,
and the departure from the optimality condition (cost-wise) is represented by the allocative
inefficiency Îť (Kumbhakar and Wang 2006). The difference between the slope of line P and the
slope of the line tangent to y0 at point B (dotted line) measures the firm allocative inefficiency.
Following Kumbhakar and Wang (2006), the input quantities at point B, are optimal with respect
to the price ratio given by (w1/w2)eÎť. The term eÎť captures the distortion effect induced by the
allocation inefficiency that creates a deviation from the optimal condition. Notice that at point B,
the farmer is technical efficient since it produces on his/her efficient isoquant.
If at the point A the combination of the input bundle X (x1, x2) yields the output level y0
then the farmer will be both technically and allocatively inefficient. Indeed, if the technology used
was technically efficient, the combination of inputs at point A should yield an output level y1
37

higher than y0. Allocative inefficiency is the possible reduction (point A to C) or increase (point
B to C) in input costs to move to the optimal input use condition.
The estimation of production efficiency requires a few assumptions regarding the
characteristics of the underlying efficient isoquant. The isoquant must be convex at the origin and
must display a non-positive slope at every point. The convexity assumption ensures the most
conservative estimates of the efficient isoquant and the non-positive slope ensures that outputs do
not decrease from increasing inputs level (M. J. Farrell 1957). Details on the estimation procedure
of u (OO technical inefficiency) and Îť (allocative efficiency) are given in section 4.2.1 below. For
each input J, Îťj can take both positive and negative values. A positive value means that the input
pair (J, 1) is being used at a level above the cost minimizing and a negative value means that the
input pair is being used at a level below the cost-minimizing (Kumbhakar, Wang and Horncastle
2015). For instance, at point B, the bundle of inputs (x1, x2) is above the least cost curve.
Therefore, Îť should take a non-zero positive value.
4.2. Empirical approach
When it comes to specifying an empirical strategy for estimating productive efficiency, two
distinctions can be made. The first distinction is between the deterministic and stochastic frontier
models. Deterministic models are nonparametric; they attribute any deviations from the frontier
output to inefficiency. In other words, deterministic models force the error term of the frontier
production estimation to represent inefficiency only; there is no white (or random) noise. In
addition to the restriction imposed on the error term, another limitation of the deterministic models
is their high sensitivity to extreme observations or outliers in the data. Deterministic models,
however, exhibit more flexibility in the estimation since they do not impose a specific functional
form to the data (Bravo-Ureta and Rieger 1991).
38

In contrast, the stochastic frontier model better handles extreme observations and
decompose the disturbance term into two components: a symmetric normal component, v,
referring to random shocks to production and a one-sided distributed error term, u, referring to the
OO inefficiency term (Aigner, Lovell and Schmidt (1977) and Meeusen and van Den Broeck
(1977)). Another key advantage of stochastic frontier models is that they allow each firm to be
inefficient relative to their own frontier rather than to a sample norm (Schmidt and Lovell 1979).
Stochastic frontier models also have a weakness worth noting. Being heavily parameterized,
estimates from such models may be biased and inconsistent if the models are misspecified and/or
the distributional assumption about the error term breaks down (Bravo-Ureta and Pinheiro 1993).
A second distinction is made between models that assume exact cost minimization
(measuring technical inefficiency only) and models with inexact cost minimization. Inexact cost
minimization models disaggregate the measure of firm’s total efficiency (or inefficiency) into its
technical and allocative components. Early stochastic frontier studies such as Aigner, Lovell and
Schmidt (1977), Meeusen and van Den Broeck (1977) are based on exact cost minimization.
Schmidt and Lovell (1979) extended the model built by Aigner, Lovell and Schmidt (1977) to
allow for the estimation of allocative inefficiency considering the duality between stochastic
frontier production and cost functions. Other examples of inexact cost minimization studies, which
have decomposed total efficiency measure into its different components are: Greene (1980); Kopp
and Diewert (1982) ; Jondrow, et al. (1982); Bravo-Ureta and Rieger (1991); S. C. Kumbhakar
(1987); Kumbhakar and Wang (2006); Kumbhakar and Tsionas (2005); Kumbhakar, Wang and
Horncastle (2015).
The empirical estimation model in this study capitalizes on the historical evolution of the
state of knowledge on modeling technical and allocative efficiency. Following the inexact cost
39

minimization approach, the duality between the stochastic frontier production function and the
cost function is used to estimate farm technical inefficiency, allocative inefficiency and the cost
associated with each inefficiency measure. A propensity score matching approach will be used to
estimate the effects of the participation regimes on inefficiency measures and the induced costs.
Acknowledging the sensitivity of the stochastic frontier model to the misspecification problem,
robustness check will be conducted using an exponential functional form of the inefficiency term
u (Table A15 in the appendix).
4.2.1 Estimating technical and allocative efficiency
It is assumed that the farmers aim to minimize their costs subject to a stochastic production
frontier. Given the duality between the production frontier and cost functions, this assumption can
alternatively be stated as the maximization of output level subject to given input levels. Technical
inefficiency occurs when the farm operates below its stochastic frontier production function. The
firm is allocatively inefficient if it operates off the least cost curve (Schmidt and Lovell 1979).
Following Kumbakhar and Wang (2006) and Kumbhakar,Wang and Horncastle (2015), a
derivation of the technical inefficiency and allocative inefficiency starts with a system of equations
including a standard production function that relates output to inputs and the FOCs (first-order
conditions) of cost minimization:
Yi = F (xik , βk, z1, z2)e(v -u)

k = 1, 2,3, 4…. Lth input
i= 1, 2, ….

FOCs:

Ə𝑦
Ə𝑋𝐽
Ə𝑦
Ə𝑋1

=

𝑓𝐽
𝑓1

=

𝑤𝑗
𝑤1

eÎťj

(1)

m th plot

J = 2,3, 4…. Lth input

(2)

Where F(.) represents the technology used by plot manager (i), which transforms the vector
of variable inputs and quasi-fixed factors x, into a quantity of outputs Y. Quasi-fixed factors are
40

variables such as land or capital. The other variables included in x are fertilizer, labor, seed, and
herbicide. The vector z1 captures differences in soil quality and input quality (e.g., local seed vs.
improved seed). The vector z2 represents plot manager, and farm characteristics assumed to
influence the variance of the technical inefficiency. These characteristic variables include farm
size, household size, education and gender of the plot manager, assets owned and distances from
the household’s house to markets. 𝛽𝑘 is the vector of parameters to be estimated.
A Cobb-Douglas functional form is assumed for the production technology F(.), because it
allows the analytical derivation of the costs associated with technical and allocative inefficiencies.
Also, the Cobb-Douglas offers a dual property which is more suited to the models that account for
both technical and allocative inefficiency. Moreover, this functional form has been used in many
studies from developing countries, facilitating comparisons.
ui is the measure of technical inefficiency for the plot (i), which is explained by the vector of
factors z2.
vi, captures the random shocks to production (or the white noise).
The ratio fj/f1 is the relative factor productivity, with factor X1 taken as a numeraire.
The ratio (wj /w1) is the relative factor prices, where Îťj is the measure of allocative inefficiency for
the input pair (j, 1).
Taking the input elasticity of output, equation (2) can be transformed to obtain:
Ə𝑦
Ə𝑋𝐽
Ə𝑦
Ə𝑋1
Ə𝑙𝑛𝑦
Ə𝑙𝑛𝑥𝑗
Ə𝑙𝑛𝑦
Ə𝑙𝑛𝑥1

=

Ə𝑙𝑛𝑦 𝑦
∗
Ə𝑙𝑛𝑥𝑗 𝑋𝐽
Ə𝑙𝑛𝑦 𝑦
∗
Ə𝑙𝑛𝑥1 𝑋1

𝑆

𝑤𝑗𝑥𝑗

=

⇔ 𝑆1𝑗 = 𝑤1𝑥1eλ

𝑤𝑗 λ

e

(3)

J = k+1…… L

(4)

𝑤1

41

Where:

Ə𝑙𝑛𝑦

Sj = Ə𝑙𝑛𝑥 = βj ;
𝑗

Ə𝑙𝑛𝑦

S1 = Ə𝑙𝑛𝑥 = β1
1

Assuming a Cobb-Douglas production technology, the estimation form of the model given by the
system of equations (1) and (4) is:
𝑙𝑛𝑦𝑖 = 𝛽0 + ∑𝐿𝑘=1 𝛽𝑘 ∗ 𝑙𝑛𝑥𝑖𝑘 + 𝑣 − 𝑢
𝛽𝑗

(5)

𝑤𝑗

ln(𝛽1) − ln(𝑤1) − 𝑙𝑛𝑥𝑗 + 𝑙𝑛𝑥1 = 𝜆𝑗 ,

j =2,3…..L

(6)

u ~ N+(0, δu(z2))
v ~ N (0, δv)
Νj ~ MVN (0, Ί)
The parameters of the production frontier model are estimated following the maximum
likelihood estimation method. The estimation of the model requires some distributional
assumption on the error terms v, u and Îť. The error term v is assumed to follow a normal
distribution with mean zero and a constant variance δv. The inefficiency term u is assumed to
follow a half-normal distribution with mean zero and a heteroscedastic variance δu(z2). The
variance of inefficiency is expected to vary with farm characteristics z2. The allocative inefficiency
term Ν, is distributed with no systematic error (mean zero) and variance Ί.

After estimating the Cobb-Douglas production function, equations (6) is followed to
compute the allocative inefficiency term Îťj of each input pair (j,1). A positive value of Îťj indicates
that input j is being underutilized relative to the numeraire input 1 and a negative value indicates
that input j is being overused relative to the numeraire input 1 (Kumbhakar, Wang and Horncastle
2015).

42

The notion of the underutilization or overutilization of inputs does not tell much about the
extent of the farm household’s allocative inefficiency. A measure of the extent of allocative
inefficiency can be the cost incurred from deviating from the optimal input choice condition. The
next section provides the framework to estimate the costs of allocative and technical inefficiency.
4.2.2 Estimating the costs of allocative and technical inefficiency
Inefficiency is costly to producers; the presence of technical inefficiency implies an excess cost
resulting from the overuse of inputs. This excess cost is directly captured by the IO technical
inefficiency term, 𝜂𝑖 , discussed earlier. Indeed 𝜂𝑖 represents the proportion by which inputs can be
reduced while keeping outputs constant; it represents the cost index of being technically
inefficient. A relationship can be established between IO inefficiency and OO inefficiency
applying the formula: 𝑢𝑖 = (∑ 𝛽𝑘 )*ηi) (Kumbhakar, Wang and Horncastle 2015). The production
frontier model specified in equation (5) will directly estimate 𝑢𝑖 and the relationship above can be
used to recover Ρi.
Allocative inefficiency also creates an excess cost due to resource overuse or underuse.
The quantification of the excess costs related to allocative inefficiency requires a preliminary
estimation of a frontier cost function corresponding to the minimum level of input costs to produce
the efficient level of outputs. Input quantity decisions deviate from the frontier cost due to technical
and allocative inefficiency.
Given the self-dual property of the Cobb-Douglas production function, a dual cost frontier
function can be derived analytically by using the parameters estimated from equation (6) only.
Following Schmidt and Lovell (1979) and Kumbhakar, Wang and Horncastle (2015), the dual cost
function takes the form:

43

𝐿𝑛𝐶 = 𝐾 +

1
𝑟

𝑙𝑛𝑦 + ∑𝐿𝑘=1

𝛽𝑘
𝑟

∗ 𝑙𝑛𝑤𝑘 −

1
𝑟

(𝑣 − 𝑢) + (𝐸 − 𝑙𝑛𝑟)

(7)

Where:
𝑟 = ∑𝐿𝑘=1 𝛽𝑘 ;
𝐸 = ∑𝐿𝑗=2

𝛽𝑗
𝑟

∗ 𝜆𝑗 + 𝑙𝑛[𝛽1 + ∑𝐿𝑗=2 𝛽𝑗 ∗ 𝑒 −𝜆𝑗 ];
𝐿

1
1
𝐾 = 𝑙𝑛𝑟 − ∗ 𝛽0 − ∗ 𝑙𝑛 [∏ 𝛽𝑘 𝛽𝑘 ]
𝑟
𝑟
𝑘=1

The minimum or stochastic cost frontier component of the cost function is
𝐿

1
𝛽𝑘
1
𝐿𝑛𝐶 = 𝐾 + 𝑙𝑛𝑦 + ∑ ∗ 𝑙𝑛𝑤𝑘 − (𝑣)
𝑟
𝑟
𝑟
𝑘=1

Observed cost exceeds the frontier cost by two factors: 1) excess cost due to technical
inefficiency captured by the term,

1
𝑟

(𝑢) and 2) excess cost due to allocative inefficiency captured
1

by the term (E-lnr). Note that the term, 𝑟 (𝑢), corresponds to the IO technical inefficiency, η,
beforehand defined. Once equation (6) is estimated the costs associated to technical and allocative
inefficiency are then computed following equation (7).
4.2.3 Modeling the impact of participating in farmer organizations on efficiency
4.2.3.1. Modeling group membership’s effect on efficiency outcomes
To estimate the effects of membership in farmer organizations on technical and allocative
efficiency, a propensity score matching (PSM) approach is used.
The PSM is a non-experimental method used to address the issue of counterfactual in
impact evaluation studies. The PSM approach consists of constructing a comparison group using
44

non-members, who can approximate the characteristics of members. Non-members of FOs
represent the control group, and members of FOs form the treatment group. The method, first,
looks for a common support region, where the control and treatment group are matched based on
their propensity scores. Then the average treatment effect (ATE) is derived by taking the difference
of the outcomes between the two groups. The propensity scores represent the probability
distribution of being a member of an FO. This probability distribution is constructed based on a
set of observed characteristics that predict the membership.
Under the Conditional Independence Assumption (CIA), there exists a vector of
observable covariates, X, such that after controlling for these covariates, the potential outcomes
are independent of treatment status. The CIA is an important assumption for the identification of
the treatment effect (Caliendo and Kopeinig 2008).
Coming back to the framework of this study, we are interested in estimating the effects of
membership in farmer organizations on technical efficiency (TE= e-u), costs of technical
inefficiency (Ρ) and costs of allocative inefficiency (E-lnr). A look at the history of cooperative
movements in Mali indicates that current organizations are legacies of government-led initiatives
to further develop the rural economy. Note that NGOs have also played a crucial role in
encouraging the establishment of FOs and supporting their activities. Given that most FOs were
established decades ago, membership is assumed to be independent of the outcomes analyzed in
this study.
To identify the treatment effect, time-invariant covariates are included in the propensity
score matching of non-members (i.e., control group) and members (i.e., treatment group) based on
the probability distribution of membership decision. Such covariates are specified in equation (10)
below, and include farmer socio-economic characteristics (e.g., age, and gender) and household
45

characteristics (e.g., farm size and household size) among other things. A logit model of
membership decision is first estimated (as specified below).
The Nearest-Neighbor matching strategy is used to produce statistically comparable
groups. To ensure that a match is found for each treated subject, the matching procedure with
replacement will be adopted. It is worth noting that the PSM approach rules out selection bias on
observable characteristics only. There may still be unobservable characteristics that influence the
membership decision (unobserved heterogeneity) and may confound the relationship between
membership of FOs and observed efficiency scores. The sensitivity of the results to potential
unobserved heterogeneity bias is evaluated using the Rosenbaum bounds analysis (Table A8-A11
in the appendix).
Using matched observations, the ATE of the membership decision is calculated by taking
the difference between the treatment group’s outcomes (members ‘outcomes) and the control
group’s outcomes (non-members’ outcomes).
If Πc (d) is the vector of outcome variables such that:
▪

c = refers to technical efficiency (TE), cost of technical inefficiency (Ρ), or cost of
allocative inefficiency (E-lnr),

▪

d= denotes the membership status (yes=member; no= non-member)

Then the PSM estimator of the ATE on the outcome c is identified as follows:
ATEcPSM = {[E (Πc(d)| d= yes, P(d=yes)] - [E (Πc(d)| d = no, P(d=no)]},
Where P(d=yes) is the propensity score of the members and P (d= no) is the propensity score of
non-members. E represents the expectation operator.

46

The logit model of group membership decision
To identify the effect of FO’s membership on farm inefficiency measures, factors that predict the
membership decision need to be controlled upfront to avoid selection bias. As hypothesized,
enhanced learning, reduction in transaction costs and improved access to inputs are the main
channels through which FO’s membership mitigates production inefficiencies. The choice of
variables used to proxy these channels will be emphasized.
An individual farmer i aims to maximize his/her utility Ui of participating in an FO, subject
to his/her resource constraints. The observable utility function is expressed as a function of
observable characteristics, Xi, and parameters, Îą, to be estimated, such that:
Ui = F(ÎąXi) + Îľi.

(8)

The probability (Prob) of a farmer i to be a member of an FO is given by:
Prob (grpi =1) = Prob(Îľi< ÎąXi) = ÎąXi + Îľi.

(9)

The unobservable part of the utility is given by Îźi which is assumed to be independently and
identically distributed with mean zero. If Um is the utility derived from membership and U-m the
utility derived from non-membership in an FO, then the membership decision represented by grpi
is such that:
grpi = 1 if Um > U-m;
grpi = 0 if U-m > Um.
The observable characteristics, Xi, include proxied variables for transaction costs, access
to inputs, and knowledge of farming practices along with other plot manager and household
characteristics that are expected to influence farmer’s likelihood to participate in FOs.
Prob (grpi =1) = F (Xi, Îą) = F (L, TC, A, H, Îą)

(10)

Îą: is the vector of parameters to be estimated
47

TC: is a vector of variables used to proxy transaction costs. TC includes distance of the household’
house to a paved road, to markets and travel costs to the nearest markets
L: is a vector of variables used to capture the likelihood for a given farm household to benefit from
the learning dynamic that takes place within the organizations. Components of vector L are
communication assets (such as TV, radio, cell phones), transport assets (motorcycle, bicycle, car)
and the education of plot manager.
A: is a vector of variables used to capture both the physical and economical access to inputs. Vector
A includes distance to input sources and off-farm incomes.
H: is a vector of household and plot manager characteristics and includes plot manger age and
gender as well as farm size, household size, and the number of adults in the household.
4.2.3.2. Modelling the effects of the participation regimes
Four participation regimes are identified: 1) farmers who do not use any services; 2) farmers who
use the input provision service only; 3) farmers who use the marketing services only; and 4)
farmers who use both input and marketing services. The first regime is used as the comparison
group to the remaining three treatment groups. To isolate the effect of each of these regimes, a
series of logit models for the participation decisions are estimated (equation 12), and common
support regions between participants and non-participants are found. The average treatment effect
of the participation regimes on efficiency outcomes is then estimated following the procedure
described in the previous section.
Logit model of the participation regime decision
A farmer’s decision to deliver outputs to, and/or buy inputs from FOs (referred to as the
participation regimes) is modeled as in section (4.2.3.1). Let’s Pin denotes the patronage decision

48

of the farmer i for services n, where n= 1 if no service is used, n= 2 if uses input service only, n=
3 if uses marketing services only, and n= 4 if uses both input and marketing services.
The farmer uses the FO channel for a given service n if the utility Uin associated with this decision
is greater than the utility associated with the alternative participation regimes Ui-n. If Uin is the
utility function derived from choosing service n, such that:
Uin = F(θYin) + Ξi

(11)

The probability of a farmer i to patronize the service n is:
Prob (Pin= 1) = Prob(Ξin< θYin) = θYin + Ξin

(12)

Pin = 1 if Uin > Ui-n
Pin = 0 if Uin < Ui-n
Table 1 illustrates the decision space of farmer’s patronage behavior, where four
participation regimes in FOs’ activities are defined.

Table 1: The decision space defining the participation regimes

Does the farmer obtained
inputs from the FO

Does the farmer sell outputs through the FO
channel
YES

NO

YES

Ui4: (Pi4=1| θYi4)

Ui2: (Pi2=1| θYi2)

NO

Ui3: (Pi3=1| θYi3)

Ui1: (Pi1=1| θYi1)

Adapted from Pascucci, Gardebroek and Dries (2012)
From the data available, the vector of observable characteristics Yin is defined (in equation
13 below) to predict the four participation regimes. It includes variables capturing farmers’
49

characteristics, transaction costs, knowledge on farm management practices, access to inputs and
other potential predictors of the patronage behavior such as members’ commitment to their FOs
and the extent of involvement in cotton cultivation.
Bhuyan (2007) and Cechin et al. (2013) proxied commitment to FOs with variables such
as frequency of meeting, attendance, participation in training sessions, and frequency of exchange
with the management committee of the organization. The higher the values of those variables are,
the higher the commitment and probability of participating in group’s activities. In our analysis,
commitment to FOs is proxied by the ownership of transport assets (considered in the vector L). It
is assumed that farmers who have limited access to transport are less likely to participate in
meetings and training sessions than others.
Cereal farmers who also grow cotton have long benefited from a highly institutionalized
and cooperative-led cotton sector. Therefore, through their involvement in cotton cultivation, they
are more inclined to acquire their farm inputs, for both cotton and cereal crops, through a cotton
cooperative (more patronage behavior). The involvement in cotton cultivation is captured by the
number of hectares devoted to cotton at the household level.
Prob (Pin= 1) = F (Yin, θ) = F (L, TC, H, A, C, θ,)

(13)

L, TC, A, and H are defined following equation (10). Variable C measures the extent of cotton
cultivation (cotton areas cultivated), and θ represents the vector of parameters to be estimated.

50

5. STUDY AREA AND DATA
5.1. Study area:
Sorghum and maize are mostly produced within the cereal belt production zone in southern Mali.
The zone also represents a major production zone of cotton. The Koulikoro and Sikasso regions
within the southern Mali are the main production zones of those crops. Those two regions account
for approximately 60% and 90% of the national sorghum and maize production (Table 2) and half
of the national agricultural population (CPS/SDR 2016).
Table 2: Total cereal production per region, 2015/2016
Millet

Sorghum

Rice

Maize

Wheat

Fonio

Total

Unit:
(ton)

This study covers farm households selected in the districts of Kati and Dioila in the
Koulikoro region and the district of Koutiala in the Sikasso region. Based on FewsNet’s report on
livelihood zoning and profiling in Mali, the study area is in livelihood Zones 10 and 11
respectively, where sorghum, millet, cotton, maize, and fruits are the primary sources of
subsistence to farm households (Figure 7).
The Koutiala district is part of zone 10. It is characterized by rain-fed agriculture and
sedentary livestock rearing. It is the most populous district and the second industrial city in the
country. The population is estimated at 575, 235 with a density of 66 inhabitant/sq km. The zone
is relatively productive due in part to its more favorable agroecological conditions, with annual
51

precipitation ranging from 700 to 1300 mm. The vegetation is the type Sudano-Sahelian and soils
are classified as sandy clay, sandy loam (FEWSNET 2010). Soils are degraded and deficient in
phosphorus, especially in the higher reaches of the topo sequence (Smale, Assima and Weltzien,
et al. 2014)). Soils in the lower reaches are more likely to be fertilized since it is where cotton and
maize are cultivated (Smale, Assima and Weltzien, et al. 2014). Given that they are well-adapted
to harsh conditions, farmers grow sorghum and millet on the more degraded and marginal soils,
which receive tiny fertilizer amendment.
Figure 7: The study area

Source: Adapted from FewsNet ( (2010) (FEWSNET 2010))
52

Historically, cotton has been the major cash crop in the zone. Cotton growers are organized
in cooperatives and receive inputs on credit, farm equipment, and technical assistance from the
CMDT- the public company supporting cotton production and textile development in Mali.
Sorghum and maize are often cultivated in rotation with cotton. Farmers do not directly receive
government assistance for growing these cereals. Maize has benefited from the fertilizer residuals
on fields previously grown with cotton (FEWSNET 2010). Due to the recent crisis in the cotton
sector (e.g., falling prices, late payments) and recent talks about the privatization of the CMDT,
farmers have been diversifying their production out of cotton and more toward cereal-based
production systems (FEWSNET 2010).
The districts of Kati and Dioila are in the livelihood zone 11 of the country, where the
average annual rainfall ranges from 1000 to 1300 mm, allowing a rainfed cultivation of maize,
sorghum, millet, and cotton. Soils are clay, loam, lateritic and the vegetation is Sudano-Sahelian
with a flat topography. Cotton and maize are the main cash crops within the zone with similar
production characteristics as in zone 10. Other cash crops include fruits, groundnut, cowpeas,
sesame and vegetables (FEWSNET 2010). Kati is home to 948,128 people (56 inhabitant/sq km)
and Dioila, the least populated district, counts 249 403 people (19 inhabitant/sq km).
With the decline of cotton production, the cereal-based production system so-called the
“traditional sector” is confronted to the constraints of low access to inputs, farm equipment and
technical supports (Staatz, et al. 2011, Kaminski, Elbehri and Samake 2013).
5.2. The Data
The data for this study come from a cross-sectional farm household survey conducted under the
Guiding Investments in Sustainable Agricultural Intensification in Africa (GISAIA) project by the

53

Food Security team of MSU. More information about the data, including the strategy of sampling,
is available in the Survey Report (2015).
The villages are from the Koutiala district in the Sikasso region and Kati and Dioila districts
in the Koulikoro region. The objective of the survey was to address research questions related to
sustainable intensification of sorghum cropping in the Malian agriculture. The sampling frame was
drawn from a village census previously undertaken, that enumerated all sorghum and maize
growers in each village. The village census was conducted to evaluate the adoption rates of
improved seed varieties of sorghum. The data that include information at the household and plot
levels were collected in four rounds. At the household level, a simple random sampling was used
to determine the sample size needed based on the village census frame. The formula used to
calculate the sample size is as follow:
N= t2 (1-p) *p / e2
Where N is the total sample size, p (=22%) is the proportion of households that adopt the
improved sorghum seed from the census frame, t (= 1.96) is the statistic representing 95%
confidence interval and e (= 3.5%) is the marginal error tolerated.
A sample size of 538 households is derived from this formula. To account for nonresponse rate, 5% additional observations were considered for a rounded sample size of 580
households. Then, 48 hybrid-variety growers were added to generate a total sample size of 628
households. In each 58 villages, ten sorghum-cropping households were randomly selected, and
48 hybrid producers in the three districts were also surveyed. All cultivated plots by these farm
households were inventoried during the first round of data collection. The plots level information
collected during the first round was used as the sampling frame for the rounds 2, 3, and 4.
Considering only the plots where maize and sorghum were grown, collective and individual plots

54

were randomly selected and used for the survey in the other rounds. A total number of 1382 plots
were randomly selected from the frame elaborated during the first round of data collection.
The sampling design did not allow detailed information to be collected on all the crops (plots)
grown by the households. Detailed data were collected for sampled maize and sorghum plots only
during the second round. Since the sample did not include all maize and sorghum plots grown by
each household, a household level analysis is not feasible. The analysis is therefore done at the
plot level considering maize and sorghum plots.
5.3. Descriptive statistics
The analysis of the productive efficiency is done at the plot level and includes observations from
964 maize and sorghum plots after deleting for non-matched plots across survey rounds. The plots
belong to 575 farm households. Summary statistics of farm and plot manager characteristics are
presented in Table 3. Plot managers in the sample are almost exclusively male (94%), and most
belong to an organization (78%), whether a cooperative or a village association. Only 19% of them
attended a formal school, but 48% are literate. They average 49 years old. Plot size averages 2
hectares. The plot managers travel, on average, about 3 kilometers to access farm inputs. The mean
of household size is 16 people, who get their livelihood from an average farm size of 11 ha.
The households supplement their farming activities with external income sources as well
as from remittance or off-farm employment. The farm households surveyed receive, on average,
an annual amount of 52,500FCFA (about 100$) in remittances and 32,600FCFA (about 50$) from
off-farm economic activities. Farm and plot characteristics disaggregated between maize and
sorghum crops are presented in columns 2 and 3.

55

Table 3: Summary statistics of farm and plot characteristics (N= 964)
Variable

All Plots

Maize Plots

Sorghum Plots

Mean

SD

Mean

SD

Mean

SD

Age (years)

49.3

13.8

49.7

13.6

17.0

85.0

Household size

15.6

8.1

15.8

8.1

3.0

57.0

Farm size (hectare)

11.2

6.5

11.1

6.4

2.0

35.0

Asset value (FCFA)

1,347,204.0

1,085,593.0

1,383,857.0

1,096,505.0

58,000.0

7,890,000.0

Herd size (TLU)

11.3

15.8

11.7

16.0

0

119.1

Oxen (TLU)

2.6

2.2

2.6

2.2

0

13.8

Plot area (hectare)

1.9

1.7

1.7

1.6

0.1

10.0

Off-farm workers(Number)

0.4

0.8

0.5

0.9

0

6.0

Off-farm income (FCFA)

32,663.9

116,282.7

36,347.5

125,433.0

0

1,500,000.0

Remittance (FCFA)

52,541.8

133,137.6

57,893.1

137,806.0

0

1,015,000.0

Distance to input source (km)

3.2

2.9

3.4

3.0

0

13.8

Distance to market (km)

12.1

10.3

11.7

9.9

0

60.0

Distance to paved road (km)

27.1

26.1

25.0

25.9

0

93.0

Village level wage (FCFA)

1,334.5

405.5

1,382.6

423.9

750.0

2,500.0

Number of Bikes

2.4

1.4

2.4

1.4

0

6.0

Number of Motorbikes

1.1

1.1

1.1

1.1

0

5.0

Number of Tv

0.4

0.6

0.4

0.6

0

2.0

Number of Phones

3.1

2.1

3.2

2.1

0

9.0

Key characteristics of members and non-members of FOs are compared in Table 4, using
t-test statistics. Members of FOs live in households with more landholding (farm size), assets,
oxen, and bicycles. They are significantly closer to input sources than non-members and reside in
villages whith lower average wage rate of a male labor.
Table 4: T-Test statistics comparing farm and plot characteristics of members and non-members
of the FOs
Variables

Member=1 Non-Member = 0
diff =Mean (0) - Mean (1)

Age (years)
Household size
Number of adults
Farm size (hectare)
Asset value (FCFA)
Herd size (TLU)
Oxen (TLU)

-0.533
-0.662
-0.522
-1.185*
-173923*
-0.787
-0.471**

56

SD
(-0.49)
(-1.04)
(-1.36)
(-2.32)
(-2.05)
(-0.64)
(-2.79)

Table 4 (cont’d)
Plot area (hectare)
Off-farm workers
Off-farm income (FCFA)
Remittance (FCFA)
Distance to input source (km)
Distance to market (km)
Distance to paved road (km)
Village level wage (FCFA)
Number of Bikes
Number of Motorbikes
Number of Tv
Number of Phones

-0.236
0.059
36.220
5158
0.525*
-0.740
-2.799
160***
-0.241*
0.046
0.053
-0.159

(-1.80)
(-0.9)
0
(-0.49)
(-2.34)
(-0.92)
(-1.37)
-5.12
(-2.19)
(-0.54)
(-1.15)
(-0.99)

The summary statistics on maize and sorghum production and input use are reported in
Table 5. The average total quantity of fertilizer applied on maize plots (274 kg) is by far higher
than on sorghum plots (36 kg). On average, sorghum plots are more labor intensive (about 342
hours of labor force used), which may compensate for the low level of fertilizer use. Land size is,
overall, comparable; 1.8 ha for sorghum compared to 1.5 ha for maize. Maize yields average three
tons per hectare whereas sorghum yields average about 2 tons per hectare. Although the use of
agricultural equipment is higher on sorghum plots, the difference in hourly cost of farm machinery
used suggests that more sophisticated equipment might be allocated to maize plots. The unit cost
(price) of each input is obtained by dividing total expenditures by total quantity used.
Table 5: Summary statistics of production and input use at the plot level
SORGHUM (N=423)
Variables
Production(kg)

Mean

Std. Dev.

Min

Max

1,874.6

3,309.6

40.0

60,000.0

17.4

18.4

1.0

161.0

2.9

20.7

0

375.0

Fertilizer (kg)

34.9

108.3

0

1,750.0

Labor (hours)

341.7

236.1

32.0

1,579.0

1.8

1.5

0.1

8.9

200.7

215.7

6.0

2,000.0

Herbicide price (FCFA/Kg)

4,565.8

2,009.2

7.5

16,250.0

Fertilizer price (FCFA/Kg)

204.5

52.4

73.3

944.2

Seed (kg)
Herbicide (kg)

Land (hectares)
Seed price (FCFA/Kg)

57

Table 5 (cont’d)
Hourly labor cost (FCFA/hour)

1,347.1

1,218.8

78.1

8,083.3

MAIZE (N=541)
Variables
Production(kg)

Mean

Std. Dev.

Min

Max

3013.8

5338.2

7.0

100000.0

22.0

39.4

0.5

751.0

Herbicide (kg)

2.8

5.2

0

100.0

Fertilizer (kg)

274.2

309.0

0

2200.0

Labor (hours)

267.6

193.0

10.0

1512.0

1.5

1.3

0.03

8.8

202.5

245.2

5.0

2666.7

Herbicide price (FCFA/Kg)

5164.3

3559.5

40.0

50750.0

Fertilizer price (FCFA/kg)

211.8

182.1

26.5

4333.3

1517.9

3868.5

20.8

85575.0

Seed (kg)

Land (hectares)
Seed price (FCFA/Kg)

Hourly labor cost (FCFA/hour)

Table 6 compares members to non-members regarding their production characteristics. The
results of the t-test statistics show that members of FOs produce significantly more of sorghum
and maize than non-members. They use a higher amount of fertilizer and face a significantly lower
price of fertilizer compared to non-members. The members also allocate more land to sorghum
and maize than non-members.
Table 6: T-test statistics comparing the production and inputs used by members and nonmembers of the FO
Variables

Member=1 Non-Member = 0
diff =Mean (0) - Mean (1)

Production(kg)
Seed (kg)
Herbicide (kg)
Machinery (hours)
Fertilizer (kg)
Labor (hours)
Land (hectares)
Seed price (FCFA)
Fertilizer price (FCFA)
Hourly labor cost (FCFA)

-772*
-7.047**
1.173
-12.760
-68.860**
7.331
-0.392***
-22.120
2.493
-106

58

SD
(-2.15)
(-2.82)
(-1.06)
(-0.86)
(-3.28)
(-0.43)
(-3.58)
(-1.22)
(-0.23)
(-0.45)

6. ECONOMETRIC MODELS
6.1. Group membership decision model:
The first stage in the propensity score matching method consists of matching non-members to
members of FOs based on relevant covariates that predict membership decision. The probability
distribution of membership decision is modeled following the specification below (equation 14)
where Cm is a vector of explanatory variables. The CIA assumption imposes that Cm includes all
variables that can influence the probability of treatment, such that after controlling for them, the
assignment into treatment is random. The squared term of age and farm size are also included in
the model to account for a potential U-inverted shape distribution of the effects of these variables
on membership. The logistic regression method is used to estimate the equation.
Prob (grpi=1) = 𝛼0 + ∑19
𝑚=1 𝛼𝑚 ∗ 𝐶𝑚 + εi

(14)

C1 = age of plot manager (years)
C2 = if plot manager is male =1; Otherwise=0
C3 = if plot manager attended formal school=1; Otherwise=0
C4 = if plot manager is literate =1; Otherwise=0
C5 = farm size (hectare)
C6= household size (persons)
C7 = average distance to inputs sources (km)
C8 = number of mobile phones used in the household (number phones)
C9 = number of TVs used in the household (number of TVs)

59

C10 = number of motorcycle used in the household (number of motorcycles)
C11 = number of bicycles used in the household (number of bicycles)
C12 = wage level in the village (FCFA/hour)
C13 = distance to the paved road (km)
C14 = age squared
C15 = farm size squared
6.2. Participation decision model:
Similarly, to identify the effects of the participation regimes, the users of the regimes 2, 3, 4
(defined beforehand) are first paired with non-users (regime 1) based on relevant covariates that
predict the participation decision. The probability distribution of these decisions is modeled
following the specification below. The logistic regression method is used to estimate the equation.
Prob (Pin=1) = 𝜃0 + ∑19
𝑚=1 𝜃𝑚 ∗ 𝐷𝑚 + 𝜉𝑖

(15)

n= 1, 2, 3, 4 corresponds to the four regimes
D1 = age of plot manager (years)
D2 = if plot manager is male =1; Otherwise=0
D3 = if plot manager attended formal school=1; Otherwise=0
D4 = if plot manager is literate =1; Otherwise=0
D5 = farm size (hectare)
D6= household size (persons)

60

D7 = number of mobile phones used in the household (number of phones)
D8 = number of TVs used in the household (number of TVs)
D9 = number of motorcycle used in the household (number of motorcycles)
D10 = number of bicycles used in the household (number of bicycles)
D11 = wage level in the village (FCFA/hour)
D12 = distance to the paved road (km)
D13= areas of cotton cultivated (hectare)
D14 = age squared
D15 = farm size squared
6.3. The Cobb-Douglas frontier production model
The estimation form of the Cobb-Douglas frontier production function is specified below. A
frontier production function is estimated for maize and sorghum separately. In addition to the
standard inputs of production, dummy variables are also included in the models to control for zero
values of some inputs, type of seeds, and whether anti-erosion practices are used. First, we observe
zero values for inputs variables such as fertilizer and herbicide. Not accounting for these zero
values may result in biased estimates of the production parameters (Battese 1997).
To address the zero input values, the method suggested by Battese (1997) is followed,
where a dummy variable DXk of the input Xk is included in the model to address the issue. DXk
is equal to 1 if the observed quantity of the input Xk is zero and equals 0 otherwise. Then, the value
of the corresponding input variable is replaced by the maximum of (Xk, DXk). While some authors
address the “zero observation” problem by excluding the zero values from the analysis or replacing
61

them by small arbitrary values, Battese (1997) points out to the fact that these approaches may
result in biased estimates. As robustness checks, results using the method of replacement by a
small value (0.1) are reported in the appendix.
Second, since the plot managers in the sample are not using the same seed varieties of crops
especially for sorghum, a dummy variable of the seed varieties is included. Dummy variables are
also included to account for the presence of anti-erosion practices.
𝑙𝑛𝑦𝑖=𝛽0 + 𝛽1𝑙𝑛𝑥1 + 𝛽2𝑙𝑛𝑥2 + 𝛽3𝑙𝑛𝑥3 + 𝛽4𝑙𝑛𝑥4 + 𝛽5𝑙𝑛𝑥5 + 𝛽6𝑙𝑛𝑥6+ 𝛽7𝐷1+𝛽8𝐷2+
𝛽9𝐷X1+𝛽10𝐷X2+ 𝑣𝑖− u(𝑍2)
𝛽𝑗

(16)

𝑤𝑗

ln(𝛽1) − ln(𝑤1) − 𝑙𝑛𝑥𝑗 + 𝑙𝑛𝑥1 = 𝜆𝑗

(17)

y = quantity of cereals produced (kg)
x1 = quantity of fertilizer applied (max (X1, DX1)) (kg)
x2 =quantity of seeds applied (kg)
x3 = labor force used on each plot (hours)
X4 = plot area allocated to the crop (ha)
X5 = quantity of herbicide applied (max (X6, DX6)) (kg)
D1= dummy variable=1 if anti-erosion practices are adopted, Otherwise= 0
D2 = dummy variable =1 if hybrid seeds are used; Otherwise=0
DX1 = dummy variable = 1 for zero value of fertilizer; Otherwise =0
DX2 = dummy variable =1 for zero value of herbicide; Otherwise =0

62

Z2= Factors affecting inefficiency, including membership to FO, age, gender, farm size, education,
assets owned.

63

7. RESULTS
7.1. Estimates of the stochastic frontier functions and Technical Efficiency scores
a) Maize
Table 7 presents the parameter estimates of the stochastic production frontier for maize. Column
(1) displays the estimates of the production function, and Column (2) presents the determinants of
technical inefficiency. All the inputs considered are significant determinants of maize production
(Column (1)). Non-application of fertilizer on maize reduces the quantity of maize produced as
suggested by the negative coefficient of the fertilizer application dummy variable. The use of antierosion practices appeared to be negatively associated with the production. The positive coefficient
of the herbicide application dummy variable indicates that non-utilization of herbicide is
associated with higher maize outputs. These results may reflect the fact that anti-erosion measures
and herbicide are intended to mitigate losses of production due to weed infestation or poor soil
quality. The primary purpose of their utilization is not to increase the production per se but to
minimize the losses observed firsthand. Therefore, the sign of the estimated coefficients is not
unusual.
The result corroborates the hypothesis of the study – membership is highly significant and
negative, meaning that it reduces technical inefficiency in maize production. This result may be
sensitive to potential bias from unobserved heterogeneity. The presence of those biases is tested
applying Rosenbaum bounds test and are discussed in subsequent sections. Technical inefficiency
in maize production is reduced with larger households owning telecommunication devices such as
TV.

64

Table 7: Maximum Likelihood Estimates of a Cobb-Douglas Production function with
heteroscedastic inefficiency (half-normal distribution of δu) for Maize
Maize (N=541)
(1)
Determinants
of production
Fertilizer (kg)
Seed (kg)
Herbicide (kg)
Labour (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

Mean
(SE)
0.185***
(0.0631)
0.120**
(0.0501)
0.142**
(0.0631)
0.105*
(0.0597)
0.371***
(0.0608)
0.00957
(0.0939)
-0.907***
(0.321)
-0.172**
(0.0864)
7.427***
(0.396)

(2)
Determinants of inefficiency
Group membership
Age (years)
Farm size (hectare)
Dummy education
Dummy literacy
Tv
Phone
Motorcycle
Bicycle
Dist.input source (km)
Household size
Constant

Mean
(SE)
-0.993***
(0.219)
-0.000316
(0.00675)
-0.0188
(0.0176)
-0.380
(0.261)
0.265
(0.201)
-0.515***
(0.175)
0.0315
(0.0553)
-0.121
(0.108)
-0.0343
(0.0828)
-0.00223
(0.0336)
-0.0458***
(0.0152)
1.906***
(0.457)

Variance of random error
constant -1.040**
(0.128)
Note: Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1
The inputs are expressed in logarithmic values

The likelihood ratio test (sigma_u=0: chibar2(01) = 38.41. Prob >= chibar2 = 0.000) rejects
the null hypothesis of a fully efficient maize production asserting that farmers operate with some
level of technical inefficiency (Table A1 in the appendix). The average technical efficiency (TE)
score in maize production is 51%, which is lower than the efficiency score estimated by Theriault
and Serra (2014) from cotton farmers in Mali. A preliminary report by Mansur, Gaskell and
Gautam (2017) on coarse grain productivity across different agro-ecological zones in Mali
estimated an average technical efficiency ranging from 53% to 57%. Thus, the TE score found is
65

closely in line with findings based on a nationally representative dataset (i.e., LSMS). About 75%
of the sample have a technical efficiency score below 65%, and none of them is fully efficient. The
highest efficiency score is about 86%. Indeed, the TE score ranges from 0.6% to 86%. The cost of
technical inefficiency averages 68%.
Table 8: Summary statistics of technical efficiency and cost of technical inefficiency
MAIZE (N=541)

1%
5%
10%

Percentiles
0.020
0.189
0.252

TE
Smallest
0.006
0.012
0.013

25%

0.369

0.015

50%

0.533

75%
90%

0.651
0.738

Largest
0.837
0.845

95%
99%

0.776
0.835

0.857
0.857

Percentiles
0.195
0.275
0.330

Obs
Sum of
Wgt.

541

1%
5%
10%

541

25%

0.465

Mean
Std. Dev.

0.507
0.186

50%

0.683

Variance

0.035
0.376
2.477

Skewness
Kurtosis

Ρ_i
Smallest
0.168
0.168
0.182
0.193

75%
90%

1.082
1.498

Largest
4.531
4.754

95%
99%

1.807
4.237

4.809
5.579

Obs
Sum of
Wgt.

541

Mean
Std. Dev.

0.858
0.633

Variance

0.401

Skewness
Kurtosis

3.146
18.506

541

The robustness of these results is evaluated against an alternative approach to dealing with
the zero values observed in herbicide and fertilizer inputs. Table A2 and A3 in Appendix report
the Maximum Likelihood estimates of the Cobb-Douglas production function for maize when zero
input values are replaced by an arbitrary value close to zero (a value of 0.1 is used). No significant
differences in the frontier estimates are found between the two approaches. The average technical
efficiency and the cost of technical inefficiency are also consistent between the two approaches.
A T-Test statistic for differences in the mean scores of technical efficiencies and its cost is
estimated (Table 11). Results show that, on average, members of FOs achieve higher technical
efficiency than non-members and lower costs induced by inefficiency for maize production. The
mean differences in the efficiency scores between members and non-members are strongly
66

significant. These results reflect the findings of previous studies relative to FO’s membership
effect on farm’s technical efficiency (Abate, Gian Nicola and Kindie 2014, Addai, Victor and
Gideon 2014, Boubacar, et al. 2016, Debebe, et al. 2015). Participation in the group’s activitiesinputs acquisition, group marketing or both- is associated with higher average technical efficiency
compared to non-participation.
While these results give a descriptive overview of the effects of the collective action on farm
technical efficiency, the statistical estimation of the marginal effects is obtained from the results
of the PSM analysis (Table 15).
b) Sorghum
Estimates of the frontier production function for sorghum are presented in Table 9. Except for
fertilizer, the coefficients of all input variables are significant. The non-significant coefficient of
fertilizer may reflect the meager amount used as sorghum is primarily produced for household’s
consumption needs and is agronomically less responsive to fertilizer than maize. The coefficient
of herbicide is negative as often expected since this input is used to minimize the losses of output
and not necessarily to increase productivity.
The likelihood ratio test (sigma_u=0: chibar2(01) = 0.000. Prob >= chibar2 = 1.000) fails
to reject the null hypothesis of a fully efficient sorghum production indicating the absence of
technical inefficiency in sorghum cultivation (Table A4, appendix). The stochastic frontier model
can thus be reduced to a standard OLS model with all errors forced into the white noise. For these
reasons, only results related to allocative inefficiency are reported for sorghum in the next sections.
Note that these results remain consistent even when zero inputs values are replaced by a
small arbitrary value (0.1), as reported in the appendix (Tables A5, A6). These results may reflect
the limitation of our model specification, in adequately representing the underlying production
67

function of sorghum. Sorghum is a staple crop like maize, but in contrast with maize, it is primarily
produced for domestic consumption and is rarely marketed. Therefore, the use of inputs such as
fertilizer, for instance, is marginal given the low value to costs ratio and this type of input may be
excluded from the model.
Table 9: Maximum Likelihood Estimates of a Cobb-Douglas Production function with
heteroscedastic inefficiency (half-normal distribution of δu) for sorghum
Sorghum (N=423)
(1)
Determinants
of production
Fertilizer (kg)

Mean
(SE)
0.137
(0.0842)
0.154***
(0.0531)
-0.145*
(0.0801)
0.265***
(0.0660)
0.171***
(0.0546)
0.164
(0.122)
-0.563
(0.379)
0.126
(0.104)
5.616***
(0.376)

Seed (kg)
Herbicide (kg)
Labour (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

(3)
Determinants of inefficiency
Group membership
Age (years)
Farm size (hectare)
Dummy education
Dummy literacy
Tv
Phone
Motorcycle
Bicycle
Dist. input source (km)
Household size
Gender
Constant

Variance of random error
Constant

-0.661***
(0.118)
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
The inputs are expressed in logarithmic values

68

Mean
(SE)
-0.0513
(0.374)
-0.00278
(0.0112)
-0.138***
(0.0532)
-0.617
(0.442)
0.471
(0.364)
-0.587
(0.387)
0.268**
(0.110)
-0.367
(0.243)
-0.0611
(0.154)
0.128**
(0.0561)
-0.0344
(0.0289)
0.233
(0.715)
0.188
(0.952)

7.2. Allocative inefficiency scores:
a) Maize
The computed inefficiency in resources allocation is summarized in Table 10. The positive nonzero values, first, indicate the presence of allocative inefficiency and second, that the different
input pairs are, on average, allocated above their cost minimizing level. The positive values also
indicate that labor is overutilized on average compared to the other inputs used in production,
(Kumbhakar, Wang and Horncastle 2015). Labor is more easily accessible than modern inputs.
As these results do not indicate the extent of allocative inefficiency, the proportional
increase/decrease in input costs imputable to allocative inefficiency (E-lnr) is estimated. As seen
in Table 10, the average cost of resource allocation inefficiency for maize is about (-17%).
Meaning, the actual maize production cost is below the minimum cost by 17%. More quantity of
maize can be produced by using more inputs while moving to the efficient frontier function.
Table 10: Resources allocation inefficiency in maize production
Variables

Mean

SD

Min

Max

Seed/Labor

3.48

1.16

0.27

9.04

Herbicide/labor

2.83

1.85

-0.73

8.77

Fertilizer/labor

4.74

1.34

0.36

9.73

E-lnr

-0.17

0.40

-0.75

1.94

Following Table 11, members of FOs display a significantly lower cost of allocative
inefficiency, on average, compared to non-members. Also, participation in group activities for the
inputs provision service and/or marketing service reduces the cost of allocative inefficiency, on
average, for maize. The difference in the allocative inefficiency costs between participants and
non-participants is highly significant considering the participation regime 2 (inputs provision

69

only). However, no significant differences are found in the allocative inefficiency costs between
participants and non-participants of the regimes 3 (marketing service).
Table 11: T-test for differences in technical efficiency and costs of inefficiency across
participation regimes in maize production
Participation regimes

diff = Mean (no) - Mean (yes)

SD

Frequency

TE

-0.0910***

(-4.67)

No =110

Ρ_i

0.357***

(5.23)

Yes = 431

E_Lnr

0.143***

(3.36)

Total= 541

TE

-0.0527**

(-3.09)

No =235

Ρ_i

0.179**

(3.15)

Yes= 223

E_Lnr

0.131**

(3.55)

Total= 458

TE

-0.0823*

(-2.31)

No =235

Ρ_i

0.183

(1.42)

Yes = 34

E_Lnr

0.136

(1.70)

Total= 269

TE

-0.106***

(-3.59)

No =235

Ρ_i

0.245*

(2.22)

Yes = 49

Membership in FOs

Use inputs service only

Use marketing service only

Use both input & marketing services

E_Lnr

0.151*
(2.17)
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

Total= 284

b) Sorghum
Allocative inefficiency in sorghum production has been computed considering three variable
inputs: seed, fertilizer, and labor. The computation of inefficiency is based upon the assumption
of a well-behaved production function with positive elasticities of production (such as the CobbDouglas). As this assumption is violated in the case of herbicide, the model could not derive a
value for allocative inefficiency associated with herbicide.
Considering only the three variable inputs - with positive elasticities, the average increase
in inputs costs due to allocative inefficiency is 96% (Table 12). This finding indicates that given
the current level of output, the observed input costs are almost double the value of the minimum
70

optimal cost. Sorghum production cost can be reduced by 96% while keeping the current quantity
produced as constant. As in maize production, seed and fertilizer are underutilized compared to
labor.
Table 12: Resources allocation inefficiency in Sorghum production
Variable

Mean

Std. Dev.

Min

Max

Seed/Labor

4.47

1.17

1.74

7.70

Fertilizer/labor

5.21

2.33

-2.00

9.52

E-lnr

0.96

0.50

-0.10

2.26

The cost of allocative inefficiency is significantly lower for members of FOs and those
who participate in the inputs provision service (Table 13). Farmers who use the inputs provision
service only or both input and marketing services of the organization have lower costs imputable
to allocative inefficiency. However, farmers who use the marketing service only have higher costs
of allocative inefficiency than farmers who do not participate in this regime, although the
difference is not significant.
The marginal effects of the participation regimes on the costs associated with allocative
inefficiency are analyzed in section 7.3.
Table 13: t-test for differences in allocative inefficiency cost across participation regimes in
Sorghum production
Participation regimes

diff = Mean (no)- Mean yes)

SD

0.168**

(2.85)

Frequency

Membership of an FO
E_Lnr

No= 91
Yes =332
Total = 423

Use Inputs service only
E_Lnr

0.202***

(3.95)

No= 187
Yes = 175
Total = 362

71

Table 13 (cont’d)
Use marketing service only
-0.140

E_Lnr

(-1.49)

No= 187
Yes= 27
Total= 214

Use both input & marketing services
E_Lnr

0.006

(0.07)

No= 187
Yes = 34
Total= 221

t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

7.3. Marginal effects of Farmer Organizations on the production efficiency:
a) Maize
The propensity score matching approach is used to determine the marginal effects of group
membership and four participation regimes on three efficiency outcomes: technical efficiency
score (TE), cost of technical inefficiency (Ρ_i) and cost of allocative inefficiency (E-lnr). The
average treatment effects (ATE) is estimated using the 1:1 Nearest Neighbor (NN) matching
approach. To improve the matching quality and covariates balance, a caliper of 0.15 is imposed on
the data indicating the maximum range within which the propensity scores of the treatment and
control group observations can be dissimilar (Caliendo and Kopeinig 2008). For the membership
effect, the logistic regression model is employed (default STATA command) to predict the
propensity scores using the covariates defined in equation (7). For each of the participation regime,
the propensity scores are determined using covariates in equation (9). Results of the logistic model
displaying the determinants of FO’s membership are reported in Appendix Table A7.
Under the 1:1 NN matching approach of the PSM model, covariate balance is achieved for
the variables that appear in Table 14. As can be seen in Table 14, the standardized differences in
mean between members and non-members for the covariates considered are minimized with the

72

matching in contrast to the raw data. Also, the variance ratio is closer to 1 for the matched
observations compared to the raw data, together with the mean differences, this attests a reasonable
level of covariates balance and a large region of common support. A formal covariates balance test
undertaken with the Inverse Probability Weight (IPW) model fails to reject the null hypothesis that
covariates are balanced. Thus, the matching approach followed is reliable.
Table 14: Covariates balance summary for maize
Standardized differences
Raw

Matched

Variance ratio
Raw

Matched

Age (years)

0.094607

-0.10939

0.798272

0.937595

Age squared (years)

0.062126

-0.11521

0.887756

0.952738

Dummy education (0/1)

0.225906

0.004612

1.460073

1.006928

Dummy literacy (0/1)

0.180758

-0.03694

1.013452

1.000684

Farm size (hectares)

0.242197

0.068245

1.123334

0.943766

Farm size squared (hectares)

0.189481

0.033721

1.212132

0.814895

Household size (persons)

0.068192

0.148161

1.488582

1.359076

Distance to inputs source (km)

-0.25486

0.046475

1.276053

1.736365

Number phone (units)

0.012886

0.086191

0.966145

1.044769

Number TV (units)

-0.10281

0.089022

1.20218

1.476307

Number motorcycle (units)

-0.06644

-0.05292

0.842534

0.939508

0.1196

0.073391

1.274507

1.359348

-0.4988

0.069145

0.704034

0.91589

Distance to paved road (km)
0.292528
-0.00835
1.424689
Overidentification test*
H0: Covariates are balanced: chi2(15) = 7.865 Prob > chi2 =0.9291

1.188297

Number bicycle (units)
Village Wage level (FCFA)

The PSM results for maize confirm the hypothesis of the study that membership in farmer
organizations reduces productive inefficiencies (Table 15). The average technical efficiency of
members is 9% higher than non-members. Membership also significantly reduces the costs related
to technical and allocative inefficiency by 32% and 16%, respectively. To check for the sensitivity
of these results to potential biases from unobserved heterogeneity, a Rosenbaum bounds analysis
is performed (Appendix, TableA8). Selection into FO’s membership is robust to unobserved

73

variables for two of three outcomes (TE and n_i). For technical efficiency and cost of technical
inefficiency outcomes, the lowest value of Gamma (Г) generating a 95% confidence interval that
includes zero is 2.2. Moreover, considering the Hodges-Lehmann point estimates (the upper and
lower bounds), the lowest value of Gamma for which the upper and lower bounds interval
encompasses zero is 2.8. Since these minimum values of Gamma are higher than 2, the treatment
effect estimates are robust to hidden biases for these two outcomes (Mirrlees, et al. 2011,
Duvendack and Palmer-Jones 2012, Rosenbaum 2002). However, the cost of allocative
inefficiency outcome (E-lnr) is sensitive to potential biases.
The results also support the hypothesis that engaging in FO’s activities is beneficial to
farmers in general. Members who patronage their FO for either input and/or marketing services
have significantly higher technical efficiency and lower cost of technical and allocative
inefficiency than non-participants. Overall, the PSM’s findings corroborate the positive effects of
membership and the patronage behavior on the efficiency scores.
Specifically, when allocative inefficiency is concerned, participation in the input provision
service is the most beneficial as it strongly reduces the cost of allocative inefficiency. This result
aligned with the idea carried that improved access to inputs can address the underutilization. The
effect of using both the input and marketing services on the costs of allocative inefficiency, albeit
positive, is not as strong compared to the other regimes.
When technical efficiency is concerned, participation in group marketing service appears
to be the most beneficial as it yields the highest positive significant effect compared to the other
regimes. Using the input provision service has no significant impact on technical efficiency. This
finding is of great interest as it suggests that farmers tend to get the most of their productive
resources when they are ensured of the availability of a market for their outputs. In other words,
74

reduced market risks and unobserved transaction costs enabled by the FOs induce more efficient
utilization of farm inputs. To promote agricultural productivity gains, research and public supports
have traditionally focused on getting the upstream segment of farm business right, through
investments in input subsidy programs and research programs on crop seed variety improvement.
This result calls for a broader perspective, as it shows that getting the downstream segment of the
farm business right can also play a significant role in stimulating production behaviors that
promote productivity gain.
The treatment effects of the participation regimes must be interpreted with some reserve as
the Rosenbaum sensitivity analysis fails to reject the presence of hidden biases that confound
outcomes and the treatment assignment (Appendices Table A9-A11)
Table 15: Marginal effects of membership and the participation regimes on maize production
efficiency using the PSM
TE
Mean
(SE)

Ρ_i
Mean
(SE)

E_Lnr
Mean
(SE)

Group membership

0.089***
(0.0244)

-0.319***
(0.0920)

-0.157***
(0.055)

Input service only

0.031
(0.0206)

-0.108*
(0.0586)

-0.170***
(0.0605)

Marketing service only

0.190***
(0.0198)

-0.444***
(0.0617)

-0.150***
(0.0262)

Input & marketing

0.080***
(0.0200)

-0.215***
(0.0724)

-0.136*
(0.0742)

Participation regimes

N

No =110
Yes = 431
Total= 541
No =235
Yes= 223
Total= 458
No =235
Yes = 34
Total= 269
No =235
Yes = 49
Total= 284

Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
TE = technical efficiency; Ρ_i= cost of technical inefficiency; E_Lnr = cost of allocative inefficiency

b) Sorghum
Regarding the sorghum production, as indicated earlier, the LR test fails to reject the null
hypothesis of the absence of technical inefficiency. Therefore, only results on allocative
inefficiency are interpreted. For the treatment models, covariates balance is achieved for variables
75

presented in Table 16. A formal covariates balance test fails to reject the null hypothesis that
covariates are balanced. Thus, the matching approach followed is reliable.
Table 16: Covariates balance summary for sorghum
Standardized differences
Raw

Matched

Variance ratio
Raw

Matched

Age

0.067918

0.072961

0.731757

1.029672

Age squared

0.021632

0.075108

0.800651

1.136755

Dummy education

0.213167

-0.04253

1.486828

0.93341

Dummy literacy

0.302419

-0.14204

1.087757

1.003376

Farm size (hectare)

0.069623

0.09404

0.668408

1.000596

-0.0371

0.063809

0.56159

0.97395

0.312418

-0.0774

0.545574

1.223691

-0.0122

-0.07308

0.947964

1.200375

Number Phone

0.177204

-0.01669

0.931938

1.016611

Number TV

0.016195

0.136809

1.335417

1.534663

-0.0498

0.075268

0.823092

1.038816

Number bicycle

0.182422

0.014953

1.033362

1.238146

Wage (FCFA)

-0.20102

-0.07424

0.675379

0.760481

Farm size squared (hectare)
Dummy gender
Distance to inputs source (km)

Number motorcycle

Distance to paved road (km)
-0.12351
-0.01001
1.031185
Overidentification test
H0: Covariates are balanced: chi2(15) = 7.85449 Prob > chi2 =0.9295

1.160453

As seen in Table 17, the findings on the cost of allocative inefficiency for sorghum are
aligned with those of maize, except for two of the participation regimes. As found in the case of
maize, group membership and participation in the input provision service significantly reduce the
cost incurred due to the presence of allocative inefficiency. In contrast, participation in the
marketing service only has an adverse effect on the costs of allocative inefficiency in sorghum
production, but this result is weakly significant. Unlike maize, the effect of using both the input
and marketing services, albeit positive, is statistically non-significant on sorghum production.

76

Table 17: Marginal effects of membership and the participation regimes on sorghum production
efficiency using the PSM
E_Lnr
Participation regimes
Mean
N
(SE)
-0.205***
No= 91
Group membership
(0.0605)
Yes =332
Total = 423
-0.189***
No= 187
Input service only
(0.056)
Yes = 175
Total = 362
0.196*
No= 187
Marketing service only
(0.119)
Yes= 27
Total= 214
-0.094
No= 187
Input & marketing
(0.0575)
Yes = 34
Total= 221
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

77

8. CONCLUSION
Cereal production is the main source of livelihoods for a large segment of the Malian rural
population and remains an important component of the Malians’ diet in general. A parametric
estimation approach is used to examine the productive efficiency of maize and sorghum farmers
in Mali, and particularly the role played by farmer organizations. A Cobb-Douglas frontier
production function with inefficiency is estimated using a 2015 cross-sectional farm household
data. Both technical and allocative efficiency estimates are derived. The study aimed to address
five research questions: Q1. What is the average technical efficiency among sorghum and maize
smallholder farmers? Q2: How are inputs of these crops (i.e., labor, capital, fertilizer, herbicides,
and seeds) allocated with respect to the optimal use? Q3: What are the costs associated with
technical and allocative inefficiency? Q4: Does membership of farmer organizations influence
technical and allocative efficiency as well as their related costs? Q5: Does efficiency differ across
different participation regimes?
An average technical efficiency of 51% is found on maize, which suggests that there is
considerable room to achieve higher productivity gain. Our model specification could not identify
the presence of technical inefficiency on sorghum as the Likelihood Ratio test from the frontier
production estimation fails to reject the null hypothesis of the absence of inefficiency.
Inefficient allocation of farm resources is a driver of low productivity. Indeed, the results
show that input costs can be reduced by 96% from their current level to move to the efficient output
level in the case of sorghum and increased by 17% in the case of maize. Consequently, the levels
of inputs used on these crops are non-optimal. There is a relatively high technical and allocative
inefficiency in the production of maize and sorghum.

78

While the GOM’s concerns about agricultural productivity have primary been translated
into substantial investment funds to support input subsidy programs or crop seed improvement,
historical data show that dry cereal production has not responded proportionally regarding yield
improvement. The results of this study emphasize the necessity to couple these interventions with
assistance services on farm management practices to maximize production performance and
induce the optimal use of resources. Attention must be paid to some socio-economic characteristics
of farm households and plot managers, given their influence on the technical inefficiency. For
instance, membership in farmer organizations, farm size, education, gender, assets owned,
especially communication assets, are potential avenues for policy interventions as they
significantly affect technical inefficiency.
To deeper insights on the determinants of production inefficiency, the role of farmer
organizations is further investigated. A treatment effect model, namely a propensity score
matching technique (PSM), is used to, first, estimate the marginal effect of FO’s membership on
efficiency/inefficiency scores. Then, the membership effect is disentangled from the effects of
different participation levels into FO’s activities. This is a key contribution of this study to the
literature. The PSM model finds that FO’s membership improves farmers’ technical efficiency
while reducing the incremental costs induced by technical and allocative inefficiency. Members
achieve a 9% technical efficiency gain over non-members and members incurred on average 24%
lower costs of productive inefficiencies than non-members in maize production. For sorghum
production, members of farmer organizations are 20.5% more efficient in allocating their farm
inputs than non-members.
Similarly, farmers who participate in FO’s activities do better when compared to nonparticipants regarding the optimal allocation of their resources. Participation in the input provision
79

service of FOs significantly reduces allocative inefficiency in both sorghum and maize production.
When it comes to technical efficiency, participation in group sale has the most significant positive
impact. This result shows that, when market-related risks faced by farmers are attenuated (as it is
the case with the group sale), they tend to adopt a production behavior that promotes efficiency
and productivity gain.
While research efforts and agricultural development supports have traditionally focused on
the upstream farm segment, this study argues that it also matters to correct for output market
imperfections to improve production performance among smallholder farmers. Farmer
organizations can provide a compelling institutional setting for facilitating linkages to markets.
Thus, their organizational capacity must be strengthened in such a way that their activities are
sustained beyond the lifespan of the NGOs or development projects which often provide them with
supports. There are many pathways to productivity enhancement, but increased productive
efficiency can be a costless and sustainable way to achieve it.
In addition to promoting access to productive resources, policy interventions in the
agricultural sector of developing countries should facilitate farmers’ access to training and
advisory services for a more efficient utilization of resources. Investments in transport and
communication infrastructures should be promoted along with marketing strategies, such as
contract farming and/or organized group sale, since they all contribute to mitigating transaction
costs and market-related risks faced by farmers. Insuring against market-related risks could lead
to efficiency-increasing decision-making among farmers.
Some limitations of this study are worth mentioning. The analysis involves cross-sectional
data, which cannot control for some potential bias due to unobserved heterogeneity in the treatment
model. Such an issue can be effectively corrected with panel data. This constitutes a limitation
80

with cross-sectional data analysis. The study attempts to account for the potential biases induced
by running a Rosenbaum bounds analysis. While some treatment effects estimates are found to be
robust to hidden biases, such as the effects of group membership on technical efficiency and the
cost of technical inefficiency, other results need to be interpreted with reserve. Another limitation
is that of the use of a Cobb-Douglas production function, which is not the most flexible functional
form, but its compelling duality property makes it appropriate for this analysis. Other functional
forms such as the translog may be more flexible, but they are costly regarding the degrees of
freedom, especially for a small dataset.
An interesting follow-up research topic would be to investigate the effects of different
contractual links to output or input markets on farmers’ production efficiency. Examples of these
contractual links are the “warrantage” system and contract farming. The analysis did not
disaggregate between different types of FOs (e.g., cooperatives versus village associations). A
comparative analysis of the effects of these different types of organization is another research
avenue that would be interesting to explore in future works.

81

APPENDIX

82

APPENDIX
Table A1
Likelihood ratio test of the presence of technical inefficiency in maize production
(1)
Determinants of
production
Fertilizer (kg)
Seed (kg)
Herbicide (kg)
Labour (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

Mean
(SE)
0.178***
(0.0652)
0.184***
(0.0520)
0.156**
(0.0671)
0.110*
(0.0622)
0.396***
(0.0633)
0.0261
(0.0974)
-0.831**
(0.329)
-0.196**
(0.0931)
7.288***
(0.414)

Maize (N=541)
(2)
Variance of
Mean
Random Error
(SE)
Constant
-0.987***
(0.138)

(3)
Technical
inefficiency
Constant

LR test of sigma_u=0: chibar2(01) = 38.41. Prob >= chibar2 = 0.000
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

83

Mean
(SE)
0.168
(0.148)

Table A2
Maximum likelihood estimates of a Cobb-Douglas production function for maize with zero input
values replaced by 0.1
Maize (N=541)
(1)
Determinants
of production
Fertilizer (kg)
Seed (kg)
Herbicide (kg)
Labor (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

Mean
(SE)
0.185***
(0.0631)
0.120**
(0.0501)
0.142**
(0.0631)
0.105*
(0.0597)
0.371***
(0.0608)
-0.317
(0.209)
-1.333***
(0.459)
0.172**
(0.0864)
7.837***
(0.431)

(2)
Determinants of inefficiency
Group membership
Age (years)
Farm size (hectare)
Dummy education
Dummy literacy
Tv
Phone
Motorcycle
Bicycle
Dist. input source (km)
Household size
Constant

Variance of random error
Constant

-1.040***
(0.128)

Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

84

Mean
(SE)
-0.993***
(0.219)
-0.000316
(0.00675)
-0.0188
(0.0176)
-0.380
(0.261)
0.265
(0.201)
-0.515***
(0.175)
0.0315
(0.0553)
-0.121
(0.108)
-0.0343
(0.0828)
-0.00223
(0.0336)
-0.0458***
(0.0152)
1.906***
(0.457)

Table A3
Summary statistics of technical efficiency and cost of technical inefficiency for maize with zero
input values replaced by 0.1
MAIZE (N=541)

541

1%
5%
10%

Percentiles
0.197
0.276
0.320

Ρ_i
Smallest
0.170
0.171
0.176

Obs

541

Sum of Wgt.

541

25%

0.465

0.193

Sum of Wgt.

541

Mean
Std. Dev.

0.508
0.186

50%

0.675

Mean
Std. Dev.

0.854
0.632

Variance
Skewness
Kurtosis

0.034
-0.390
2.493

Variance
Skewness
Kurtosis

0.399
3.187
18.991

1%
5%
10%

Percentiles
0.020
0.194
0.254

TE
Smallest
0.005
0.012
0.012

Obs

25%

0.370

0.016

50%

0.537

75%
90%
95%
99%

0.651
0.744
0.775
0.834

Largest
0.837
0.851
0.854
0.855

75%
90%
95%
99%

85

1.078
1.487
1.778
4.237

Largest
4.463964
4.779667
4.782262
5.705907

Table A4
Likelihood ratio test of the presence of technical inefficiency in sorghum production
(1)
Determinants of
production
Fertilizer (kg)
Seed (kg)
Herbicide (kg)
Labour (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

Mean
(SE)
0.159*
(0.0914)
0.216***
(0.0524)
-0.175**
(0.0790)
0.270***
(0.0694)
0.217***
(0.0563)
0.171
(0.120)
-0.622
(0.406)
0.0758
(0.106)
4.944***
(0.630)

Sorghum (N=423)
(2)
Variance of
Mean
Random Error
(SE)
Constant
-0.290***
(0.0697)

(3)
Technical
inefficiency
Constant

LR test of sigma_u=0: chibar2(01) = 0.000. Prob >= chibar2 = 1.000
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

86

Mean
(SE)
-7.991
(68.59)

Table A5
Maximum likelihood estimates of a Cobb-Douglas production function for sorghum with zero
input values replaced by 0.1
Sorghum (N=423)
(1)
Determinants
of production
Fertilizer (kg)
Seed (kg)
Herbicide (kg)
Labour (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

Mean
(SE)
0.137
(0.0842)
0.154***
(0.0531)
-0.145*
(0.0801)
0.265***
(0.0660)
0.171***
(0.0546)
0.497*
(0.283)
-0.878
(0.569)
0.126
(0.104)
5.598***
(0.476)

(2)
Determinants of inefficiency
Group membership
Age (years)
Farm size (hectare)
Dummy education
Dummy literacy
Tv
Phone
Motorcycle
Bicycle
Dist.input source (km)
Household size
Gender
Constant

Variance of random error
Constant

-0.661***
(0.118)
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Note: The inputs are expressed in logarithmic values

87

Mean
(SE)
-0.0512
(0.375)
-0.00279
(0.0112)
-0.138***
(0.0533)
-0.617
(0.442)
0.472
(0.364)
-0.587
(0.387)
0.269**
(0.110)
-0.367
(0.243)
-0.0611
(0.154)
0.129**
(0.0561)
-0.0345
(0.0289)
0.233
(0.716)
0.187
(0.953)

Table A6
Likelihood ratio test of the presence of technical inefficiency in sorghum production when zero
input values are replaced by 0.1

(1)
Determinants of
production
Fertilizer (kg)
Seed (kg)
Herbicide (kg)
Labour (hours)
Land (hectare)
Dummy herbicide
Dummy fertilizer
Dummy soil fertility
Constant

Mean
(SE)
0.159*
(0.0914)
0.216***
(0.0524)
-0.175**
(0.0790)
0.270***
(0.0694)
0.217***
(0.0563)
0.574**
(0.278)
-0.989
(0.613)
0.0758
(0.106)
4.908***
(0.702)

Sorghum (N=423)
(2)
Variance of
Mean
Random Error
(SE)
Constant
-0.290***
(0.0697)

(3)
Technical
inefficiency
Constant

LR test of sigma_u=0: chibar2(01) = 0.000. Prob >= chibar2 = 1.000
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

88

Mean
(SE)
-7.991
(68.24)

Table A7
Determinants of FO membership
Member =1 Non-member =0
Variables
Age (years)
Age square (years)
Gender (0/1)
Education (dummy)
Literacy (dummy
Farm size (hectare)
Farm size square
Household size
Remittances
Number of off-farm workers
Asset value(FCFA)
Distance to input source (km)
Phone (number)
Tv (number)
Motorcycle (number)
Bicycle (number)
Herd size (TLU)
Village wage level (FCFA)
Distance to paved road (km)
Observations

Mean

SD

0.0132**
-0.000136***
-0.0916*
0.0940**
0.0222
0.0244***
-0.000653***
-0.00191
-0.0342
-0.00221
5.49e-08***
-0.00684
0.00606
-0.0450*
-0.0327*
0.00778
-0.000266
-0.000145***
-0.000623
964

(0.00518)
(5.06e-05)
(0.0528)
(0.0419)
(0.0304)
(0.00732)
(0.000211)
(0.00198)
(0.0284)
(0.0162)
(1.94e-08)
(0.00526)
(0.00816)
(0.0255)
(0.0169)
(0.0123)
(0.00108)
(3.28e-05)
(0.000639)

Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

89

Table A8
Rosenbaum bounds analysis for group membership effect(Maize)
Hodges-Lehmann point estimates

95% confidence interval

Outcomes
Upper bound

Lower bound

Upper bound

Lower bound

Г
TE
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

0.107237
0.087624
0.071145
0.057247
0.044939
0.034031
0.023787
0.01476
0.006489
-0.00105
-0.00786

1.07E-01
0.127066
0.14329
0.15763
0.169881
0.181579
0.191776
0.20116
0.209907
0.217779
0.224801

0.083314
0.063683
0.047081
0.032699
0.019483
0.008259
-0.00226
-0.0115
-0.02031
-0.02828
-0.03589

0.131467
0.150943
0.167682
0.183005
0.196281
0.208174
0.218906
0.228679
0.237547
0.24566
0.253451

1
1.2
1.4
1.8
2
2.2
2.4
2.6
2.8
3

-0.25601
-0.30718
-0.35151
-0.38966
-0.42516
-0.45694
-0.48722
-0.51521
-0.54276
-0.56774
-0.5933

-0.25601
-0.20705
-0.16614
-0.13057
-0.09853
-0.07054
-0.0462
-0.02254
-0.00262
0.016293
0.033961

-0.31838
-0.37229
-0.4192
-0.46105
-0.50004
-0.53768
-0.57177
-0.60626
-0.638
-0.66908
-0.70027

-0.19611
-0.14664
-0.10426
-0.0675
-0.0348
-0.0064
0.019415
0.043441
0.065765
0.086596
0.106521

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

-0.12642
-0.17835
-0.22169
-0.25881
-0.29138
-0.32011
-0.34622
-0.37003
-0.39168
-0.41066
-0.42943

-0.12642
-0.07454
-0.03142
0.007365
0.040342
0.070539
0.096112
0.121047
0.142711
0.162586
0.182259

-0.18978
-0.24197
-0.28555
-0.32414
-0.3574
-0.38821
-0.41364
-0.43936
-0.46172
-0.48262
-0.50298

-0.06329
-0.01
0.034871
0.074164
0.107797
0.138924
0.165849
0.191935
0.21499
0.235814
0.256456

Ρ_i

1.6

E_Lnr

90

Table A9
Rosenbaum bounds analysis for the effect of participation in inputs provision service (Maize)
Hodges-Lehmann point estimates

Outcomes

Upper bound

Lower bound

95% confidence interval
Upper bound

Lower bound

Г
TE
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

0.06415
0.041319
0.023312
0.006824
-0.00726
-0.01984
-0.03124
-0.04267
-0.0522
-0.06079
-0.06933

0.06415
0.086286
0.103545
0.119158
0.132055
0.143466
0.153933
0.163868
0.172154
0.179855
0.186665

0.026487
0.003677
-0.01589
-0.03284
-0.04843
-0.06262
-0.07438
-0.08613
-0.09705
-0.10661
-0.11538

0.100665
0.122137
0.139843
0.155164
0.169468
0.180783
0.191334
0.201228
0.20981
0.217889
0.224868

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

-0.16538
-0.22189
-0.26651
-0.303
-0.33763
-0.36942
-0.39706
-0.42176
-0.44251
-0.46305
-0.48196

-0.16538
-0.11115
-0.06597
-0.02387
0.010144
0.042056
0.06798
0.094485
0.118623
0.14184
0.164347

-0.2586
-0.31011
-0.35967
-0.40107
-0.43568
-0.46692
-0.49563
-0.52121
-0.54627
-0.56729
-0.58647

-0.07508
-0.01614
0.030696
0.072042
0.110373
0.145081
0.17932
0.207376
0.234081
0.260065
0.284056

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

-0.17611
-0.22846
-2.70E-01
-3.10E-01
-3.44E-01
-0.37376
-0.40286
-0.43067
-0.45408
-0.47689
-0.49777

-0.17611
-1.26E-01
-8.79E-02
-5.21E-02
-2.35E-02
1.22E-03
0.025577
0.046658
0.066767
0.08329
0.100436

-0.26179
-0.31772
-0.36445
-4.07E-01
-4.45E-01
-4.80E-01
-0.51149
-0.54171
-0.56973
-0.5939
-0.61537

-0.09538
-0.04609
-0.00603
0.029008
0.06001
0.086042
0.111469
0.131519
0.151832
0.170195
0.189518

Ρ_i

E_Lnr

91

Table A10
Rosenbaum bounds analysis for the effect of participation in marketing service (Maize)
Hodges-Lehmann point estimates

Outcomes

Upper bound

Lower bound

95% confidence interval
Upper bound

Lower bound

Г
TE
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

0.084063
0.060263
3.86E-02
2.08E-02
1.07E-02
0.004166
-0.00803
-0.0118
-0.02108
-0.02395
-0.03027

0.084063
1.03E-01
1.16E-01
1.24E-01
1.35E-01
1.44E-01
0.152984
0.162816
0.174893
0.187846
0.194247

-0.01005
-0.02675
-0.04786
-6.68E-02
-8.76E-02
-9.96E-02
-0.1142
-0.12425
-0.13118
-0.13693
-0.14999

0.156053
0.189231
0.215938
0.232954
0.244301
0.254142
0.260303
0.277533
0.289507
0.313153
0.328943

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

-0.22774
-0.25182
-0.28716
-0.30584
-0.343
-0.3919
-0.41878
-0.44224
-0.45442
-0.48741
-0.49723

-0.22774
-0.17723
-0.14044
-0.07204
-0.03882
-0.02131
-0.00397
0.015682
0.033001
0.052609
0.060018

-0.43073
-0.49529
-0.52512
-0.54726
-0.57276
-0.65201
-0.68481
-0.72987
-0.75829
-0.7809
-0.79158

0.012347
0.054679
0.098193
0.121209
0.169421
0.206379
0.233981
0.25715
0.273255
0.299961
0.310789

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

-0.00872
-0.07019
-0.11969
-0.16429
-0.22656
-0.25401
-0.27364
-0.29566
-0.31226
-0.33332
-0.36591

-0.00872
0.026451
0.052464
0.08738
0.122804
0.153592
0.171686
0.198508
0.218867
0.249735
0.26751

-0.28682
-0.34328
-0.41188
-0.43129
-0.44462
-0.47438
-0.54083
-0.55496
-0.57396
-0.62069
-0.67428

0.185835
0.254378
0.313656
0.389
0.431459
0.462043
0.51228
0.537215
0.558976
0.592366
0.633279

Ρ_i

E_Lnr

92

Table 11
Rosenbaum bounds analysis for the effect of participation in both inputs and marketing services
(Maize)
Hodges-Lehmann point estimates

Outcomes

Upper bound

Lower bound

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

0.06014
0.041936
0.024171
0.012793
0.000647
-0.00932
-0.01652
-0.02668
-0.03299
-0.03969
-0.04456

1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3
1
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
2.8
3

95% confidence interval
Upper bound

Lower bound

0.06014
0.077559
0.094189
0.105973
0.11853
0.128483
0.136072
0.143281
0.151122
0.157178
0.160897

-0.00983
-0.03119
-0.04473
-0.061
-0.0766
-0.08727
-0.10291
-0.1108
-0.1266
-0.13283
-0.13832

0.130145
0.148553
0.161251
0.177992
0.190338
0.20758
0.216655
0.230682
0.239684
0.249567
0.263624

-0.1295
-0.17137
-0.20685
-0.23501
-0.26659
-0.2864
-0.31368
-0.32733
-0.34788
-0.3646
-0.38206

-0.1295
-0.08249
-0.04712
-0.01553
0.009556
0.035938
0.049589
0.066558
0.080702
0.096283
0.112758

-0.28969
-0.33739
-0.38404
-0.424
-0.46568
-0.50178
-0.52581
-0.55321
-0.57309
-0.60906
-0.63316

0.036904
0.07514
0.114116
0.155849
0.201589
0.244059
0.283958
0.316765
0.367852
0.460832
0.530614

-0.10672
-0.16183
-0.2096
-0.23874
-0.27064
-0.30447
-0.32918
-0.36412
-0.38404
-0.40988
-0.43287

-0.10672
-0.06192
0.000128
0.04054
0.067986
0.095088
0.12053
0.157856
0.18254
0.20696
0.233623

-0.30673
-0.37578
-0.43369
-0.48063
-0.51448
-0.54034
-0.56932
-0.58966
-0.6179
-0.64922
-0.66877

0.099324
0.170867
0.237005
0.295365
0.331686
0.369726
0.401492
0.439669
0.486197
0.538071
0.591898

Г
TE

Ρ_i

E_Lnr

93

BIBLIOGRAPHY

94

BIBLIOGRAPHY

Abate, Gashaw Tadesse, Francesconi Gian Nicola, and Getnet Kindie. 2014. "Impact of
Agricultural Cooperatives on Smallholders’Technical Efficiency: Empirical Evidence
from Ethiopia." Annals of Public and Cooperative Economics 85 (2): 257-286.
Abdallah, Abdul-Hanan. 2016. "Agricultural credit and technical efficiency in Ghana: is there a
nexus?" Agricultural Finance Review 76 (2): 309-324.
Addai, Kwabena Nyarko, Owusu Victor, and Danso-Abbeam Gideon. 2014. "Effects of farmer–
Based-organization on the technical efficiency of maize farmers across Various AgroEcological Zones of Ghana." Journal of Economics and Development Studies 2 (1): 141161.
AfriqueVerte. n.d. "Les Banques de cĂŠrĂŠales." Fiche Documentaire.
Aigner, Dennis, CA Knox Lovell, and Peter Schmidt. 1977. "Formulation and estimation of
stochastic frontier production function models." Journal of Econometrics 6 (1): 21-37.
Akeredolu, Mercy O. 2008. "Private Sector Involvement In Agricultural Extension Service
Delivery in Mali: Views and Constraints from the Pilot Phase."
Alene, Arega D., V. M. Manyong, G. Omanya, H. D. Mignouna, M. Bokanga, and G. Odhiambo.
2008. "Smallholder market participation under transactions costs: Maize supply and
fertilizer demand in Kenya." Food Policy 33 (4): 318-328.
Bandiera, Oriana, and Imran Rasul. 2006. "Social networks and technology adoption in northern
Mozambique." The Economic Journal 116 (514): 869-902.
Battese, George E. 1997. "A note on the estimation of Cobb‐Douglas production functions when
some explanatory variables have zero values." Journal of Agricultural Economics 48 (13): 250-252.
BÊlières, Jean-François, Michel Benoit C., Luc Barret, Hamady Djouara, and Demba KÊbÊ.
2008. Les organisations de producteurs en zone cotonnière au Mali: Conditions
d’émergence et perspectives. Electronic Ressource, Économie rurale.
doi:10.4000/economierurale.498.
Bernard, Tanguy, and David J. Spielman. 2009. "Reaching the rural poor through rural producer
organizations? A study of agricultural marketing cooperatives in Ethiopia." Food Policy
34 (1): 60-69.
Besley, Timothy, and Anne Case. 1993. "Modeling technology adoption in developing
countries." The American Economic Review 83 (2): 396-402.
95

Bhuyan, Sanjib. 2007. "The “people” factor in cooperatives: an analysis of members' attitudes
and behavior." Canadian Journal of Agricultural Economics 55 (3): 275-298.
Bokusheva, Raushan, and Subal C. Kumbhakar. 2008. "MODELLING FARMS'PRODUCTION
DECISIONS UNDER EXPENDITURE CONSTRAINTS." Seville, Spain: In European
Association of Agricultural Economists, 107th Seminar.
Boubacar, Oumarou, Hui-Qiu Zhou, Abdullah Rana Muhammad, and Ghazanfar Sidra. 2016.
"Analysis of Technical Efficiency of Rice Farms and Its Influencing Factors in Southwestern of Niger." Journal of Northeast Agricultural University (English Edition) 23 (4):
67-77.
Boughton, Duncan, and Thomas Reardon. 1997. "Will promotion of coarse grain processing turn
the tide for traditional cereals in the Sahel? Recent empirical evidence from Mali." Food
Policy 22 (4): 307-316.
Bravo-Ureta, Boris E., and AntĂłnio E. Pinheiro. 1993. "Efficiency analysis of developing
country agriculture: a review of the frontier function literature." Agricultural and
resource economics Review 22 (1): 88-101.
Bravo-Ureta, Boris E., and Laszlo Rieger. 1991. "Dairy farm efficiency measurement using
stochastic frontiers and neoclassical duality." American Journal of Agricultural
Economics 73 (2): 421-428.
Caliendo, Marco, and Sabine Kopeinig. 2008. "Some practical guidance for the implementation
of propensity score matching." Journal of economic surveys 22 (1): 31-72.
Cechin, Andrei, Jos Bijman, Stefano Pascucci, and Onno Omta. 2013. "Decomposing the
member relationship in agricultural cooperatives: Implications for commitment."
Agribusiness 29 (1): 39-61.
CGIAR. 2014. CGIAR developement Dialogues. Accessed November 20, 2016.
http://dialogues.cgiar.org/blog/millets-sorghum-climate-smart-grains-warmer-world/.
Chauvin, Nicolas Depetris, Francis Mulangu, and Guido Porto. 2012. Food Production and
Consumption Trends in Sub-Saharan Africa: Prospects for the Transformation of the
Agricultural Sector. Working Papers, Regional Bureau for Africa: United Nations
Development Program.
Christiaensen, Luc, Chuhan-Pole Punam, and Sanoh Aly. 2013. Africa’s Growth, Poverty and
Inequality Nexus-Fostering Shared Prosperity. World Bank.
Coase, Ronald H. 1937. "The nature of the firm." Economica 4 (16): 386-405.

96

Conley, Timothy G., and Christopher R. Udry. 2010. "Learning about a new technology:
Pineapple in Ghana." The American Economic Review 100 (1): 35-69.
CountrySTAT. 2008. RĂŠpartition de la superficie ensemencĂŠe by AnnĂŠe, Produit (Ha). Comp.
CPS/SDR. Mali. http://countrystat.org/home.aspx?c=MLI&ta=133CPD016&tr=7.
CPS/SDR. 2016. Annuaire Statistique 2015 du Secteur du DĂŠveloppement Rural . Technical
Report, Mali: MINISTERE DE L’AGRICULTURE / REPUBILQUE DU MALI.
CRA. n.d. Les Organisations Professionnelles Agricoles : dĂŠfinition, fonctions et formes
juridiques. Comp. Chambre Regionale d'Agriculture de Segou. Mali.
De Janvry, Alain, Marcel Fafchamps, and Elisabeth Sadoulet. 1991. "Peasant household
behavior with missing markets: some paradoxes explained." The Economic Journal 101
(409): 1400-1417.
Debebe, Sisay, Haji Jema, Goshu Degye, and Edriss Abdi-Khalil. 2015. "Technical, allocative,
and economic efficiency among smallholder maize farmers in Southwestern Ethiopia:
Parametric approach." Journal of Development and Agricultural Economics 7 (8): 282291.
Falconnier, Gatien NoĂŤl. 2016. Trajectories of agricultural change in southern Mali.
Wageningen University.
Falkowski, Jan, and Pavel Ciaian. 2016. Factors Supporting the Development of Producer
Organizations and their Impacts in the Light of Ongoing Changes in Food Supply
Chains: A Literature Review. JRC Technical Reports, Institute for Prospective
Technological Studies. doi:10.2791/21346.
FAO. 2012. Agricultural Cooperatives: Paving the Way for Food Security and Rural
Development. Fact sheet, Rome: Food and Agricultural Organization.
FAO. 1995. Framework for farm household-level decision making. Rome.
FAO. 2010. Producer Organizations: Reclaiming Opportunities for Development. Policy Brief,
FAO Regional Office for Africa.
FAO/GIEWS. 2016. Cereal supply and demand balance for Sub-Saharan African countries.
Country Report, ROME: FAO.
FAOSTAT. 2013. Food Supply - Crops Primary Equivalent.
http://www.fao.org/faostat/en/#data/CC.
Farrell, M.J. 1957. "The measurement of productive efficiency." Journal of the Royal Statistical
Society 253-290.
97

Farrell, Michael James. 1957. "The measurement of productive efficiency." Journal of the Royal
Statistical Society 120 (3): 253-290.
FEWSNET. 2010. ZONES ET PROFILS DE MOYENS D’EXISTENCE AU MALI. USAID.
Fischer, Elisabeth, and Matin Qaim. 2012a. "Linking smallholders to markets: determinants and
impacts of farmer collective action in Kenya." World Development 40 (6): 1255-1268.
Fischer, Elisabeth, and Matin Qaim. 2014. "Smallholder Farmers and Collective Action: What
Determines the Intensity of Participation?" Journal of Agricultural Economics 65 (3):
683-702.
Foster, Andrew D., and Mark R. Rosenzweig. 1995. "Learning by doing and learning from
others: Human capital and technical change in agriculture." Journal of Political Economy
103 (6): 1176-1209.
Francesconi, Gian Nicola, and Nico Heerink. 2011. "Ethiopian agricultural cooperatives in an era
of global commodity exchange: Does organizational form matter?" Journal of African
Economies 153-177.
Fried, Harold O., Shelton S. Schmidt, and CA Knox, eds Lovell. 1993. The measurement of
productive efficiency: techniques and applications. Oxford university press.
Fulton, Joan R., and Wiktor L. Adamowicz. 1993. "Factors that influence the commitment of
members to their cooperative organization." Journal of Agricultural Cooperation 8.
Fulton, Murray, and Konstantinos Giannakas. 2001. "Organizational commitment in a mixed
oligopoly: Agricultural cooperatives and investor-owned firms." American journal of
agricultural economics 83 (5): 1258-1265.
Garfamy, Reza Mohammady. 2012. "Supply management: A transaction cost economics
framework." South East European Journal of Economics and Business 7 (2): 139-147.
GOM. 2010. PLAN NATIONAL D’INVESTISSEMENT PRIORITAIRE DANS LE SECTEUR
AGRICOLE AU MALI- 2011-2015 . CELLULE NATIONALE CEDEAO .
GOM. 2015. Projet d’accroissement de la productivité agricole (PAPAM). IFAD.
Greene, William H. 1980. "On the estimation of a flexible frontier production model." Journal of
Econometrics 13 (1): 101-115.
Hailu, Getu, Weersink Alfons, and Minten Bart. 2015. "Rural organizations, agricultural
technologies and production efficiency of teff in Ethiopia." Milan: In Conference of the
International Association of Agricultural Economists. 9-14.

98

Hellin, Jon, Mark Lundy, and Madelon Meijer. 2009. "Farmer organization, collective action and
market access in Meso-America." Food Policy 34 (1): 16-22.
Hobbs, Jill E. 1996. "A transaction cost approach to supply chain management." Supply Chain
Management: An International Journal 1 (2): 15-27.
Hobbs, Jill E. 1997. "Measuring the importance of transaction costs in cattle marketing."
American Journal of Agricultural Economics 79 (4): 1083-1095.
Hollinger, Frank, and John M Staatz. 2015. Agricultural Growth in West Africa: Market and
policy drivers. Rome: FAO and AfDB.
IFDC. 2015. Mali Fertilizer Assessment. Muscle Shoals, Alabama.
INSTAT. 2015. "Consommation, Pauvrete, Bien etre des Menages 2014-2015."
INSTAT. 2014. Enquete Agricole de Conjoncture. Bamako.
INSTAT. 2017. ENQUETE MODULAIRE ET PERMANENTE AUPRES DES MENAGES
(EMOP) . Survey Report, Bamako: Institut National de la Statistique du Mali.
Jaime, MĂłnica M., and Salazar CĂŠsar A. 2011. "Participation in organizations, technical
efficiency and territorial differences: A study of small wheat farmers in Chile." Chilean
Journal of Agricultural Research 71 (1): 104.
Jondrow, James, CA Knox Lovell, Ivan S. Materov, and Peter Schmidt. 1982. "On the estimation
of technical inefficiency in the stochastic frontier production function model." Journal of
Econometrics 19 (2-3): 233-238.
Junge, B., Deji O., Abaidoo R., Chikoye D., and Stahr K. 2009. "Farmers' adoption of soil
conservation technologies: A case study from Osun state, Nigeria." Journal of
Agricultural Education and Extension 15 (3): 257-274.
Kaminski, Jonathan, Aziz Elbehri, and Michel Samake. 2013. An Assessment of Sorghum and
Millet in Mali and Implications for Competitive and Inclusive Value Chains. Rome: FAO.
Karamba, R.Wendy, and Winters Paul C. 2015. "Gender and agricultural productivity:
implications of the Farm Input Subsidy Program in Malawi." Agricultural Economics 46
(3): 357-374.
Kelemu, Kaleb, and Negatu Workneh. 2016. "Analysis of levels and determinants of technical
efficiency of wheat producing farmers in Ethiopia." African Journal of Agricultural
Research 11 (36): 3391-3403.

99

Key, Nigel, Elisabeth Sadoulet, and Alain De Janvry. 2000. "Transactions costs and agricultural
household supply response." American journal of agricultural economics 82 (2): 245259.
Klein, Kurt K., Timothy J. Richards, and Allan Walburger. 1997. "Determinants of Co‐operative
Patronage in Alberta." Canadian Journal of Agricultural Economics 45 (2): 93-110.
Kopp, Raymond J., and W. Erwin Diewert. 1982. "The decomposition of frontier cost function
deviations into measures of technical and allocative efficiency." Journal of Econometrics
19 (2-3): 319-331.
Kumbhakar, Subal C. 1987. "The specification of technical and allocative inefficiency in
stochastic production and profit frontiers." Journal of Econometrics 34 (3): 335-348.
Kumbhakar, Subal C., and Efthymios G. Tsionas. 2005. "Measuring technical and allocative
inefficiency in the translog cost system: a Bayesian approach." Journal of Econometrics
126 (2): 355-384.
Kumbhakar, Subal C., and Hung-Jen Wang. 2006. "Estimation of technical and allocative
inefficiency: A primal system approach." Journal of Econometrics 419–440.
Kumbhakar, Subal C., Hung-Jen Wang, and Alan P. Horncastle. 2015. A Practitioner’s Guide to
Stochastic Frontier Analysis Using STATA. New York: Cambridge University Press.
Latruffe, Laure. 2010. Competitiveness, productivity, and efficiency in the agricultural and agrifood sectors. OECD Papers, France: OECD Food, Agriculture and Fisheries.
Lazarus, Brenda N. 2013. A Study Of Household Income Determinants And Income Inequality In
The Tominian And Koutiala Zones Of Mali. Master Thesis, East Lansing: Michigan State
University.
LOA. 2006. LOI N°06-045/ DU 5 SEPTEMBRE 2006 PORTANT LOI D'ORIENTATION
AGRICOLE. JOURNAL OFFICIEL DE LA REPUBLIQUE DU MALI.
Ma, Xingliang, and Guanming Shi. 2015. "A dynamic adoption model with Bayesian learning:
an application to US soybean farmers." Agricultural Economics 46 (1): 25-38.
MAFAP. 2013. Review of Food and Agricultural Policies in Mali 2005-2011. Country-Report,
Food and Agricultural Organization of the United Nations.
MAFAP. 2013. Review of Food and Agricultural Policies in Mali 2005-2011. MAFAP Country
Report Series, Rome: FAO.
Mansur, Ahmed, Joanne Gaskell, and Madhur Gautam. 2017. Is There Potential for Dryland
Agriculture? Some evidence from Mali. Unpublished, World Bank.
100

Markelova, Helen, Ruth Meinzen-Dick, Jon Hellin, and Stephan Dohrn. 2009. "Collective action
for smallholder market access." Food Policy 34 (1): 1-7.
Meeusen, Wim, and Julien van Den Broeck. 1977. "Efficiency estimation from Cobb-Douglas
production functions with composed error." International economic review 435-444.
MĂŠnard, Claude. 2007. Cooperatives: hierarchies or hybrids?." In Vertical markets and
cooperative hierarchies". Springer Netherlands.
MĂŠnard, Claude, and Egizio Valceschini. 2005. "New institutions for governing the agri-food
industry." European Review of Agricultural Economics 32 (3): 421-440.
MĂŠnard, Claude, and Egizio Valceschini. 2005. "New institutions for governing the agri-food
industry." European Review of Agricultural Economics 32 (3): 421-440.
Mercoiret, Marie-Rose. 2006. Les organisations paysannes et les politiques agricoles. Afrique
contemporaine , 135-157.
MRD. 2014. "Plan de Campagne Agricole 2014-2015."
Mujawamariya, Gaudiose, Marijke D’Haese, and Stijn Speelman. 2013. "Exploring double sideselling in cooperatives, a case study of four coffee cooperatives in Rwanda." Food Policy
39: 72-83.
Munshi, Kaivan. 2014. "Social learning in a heterogeneous population: technology diffusion in
the Indian Green Revolution." Journal of Development Economics 73 (1): 185-213.
Nordin, Martin, and HĂśjgĂĽrd SĂśren. 2017. "An evaluation of extension services in Sweden."
Agricultural Economics 48 (1): 51-60.
Pascucci, Stefano, Cornelis Gardebroek, and Liesbeth Dries. 2012. "Some like to join, others to
deliver: an econometric analysis of farmers' relationships with agricultural cooperatives." European Review of Agricultural Economics 39 (1): 51-74.
Reardon, Thomas, Jean-Marie Codron, Lawrence Busch, James Bingen, and Craig Harris. 2000.
"Global change in agrifood grades and standards: agribusiness strategic responses in
developing countries." The International Food and Agribusiness Management Review 2
(3): 421-435.
Schmidt, Peter. 1985. "Frontier production functions." Econometric Reviews 289-328.
Schmidt, Peter, and C.A. Knox Lovell. 1979. "Estimating Technical and Allocative Inefficiency
Relative to Stochastic Production and Cost Frontiers." Journal of Econometrics 343 366.

101

Shiferaw, Bekele, Gideon Obare, and Geoffrey. Muricho. 2008. "Rural market imperfections and
the role of institutions in collective action to improve markets for the poor." In Natural
Resources Forum (Blackwell Publishing Ltd) 32 (1): 25-38.
Shiferaw, Bekele, Gideon Obare, Geoffrey Muricho, and Said Silim. 2009. "Leveraging
institutions for collective action to improve markets for smallholder producers in lessfavored areas." African Journal of Agricultural and Resource Economics 3 (1): 1-18.
Shiferaw, Bekele, Jon Hellin, and Geoffrey Muricho. 2011. "Improving Market Access and
Agricultural Productivity Growth in Africa: What Role for Producer Organization and
Collective Action Institutions." Food Security 3 (4): 475-489.
Smale, Melinda, Amidou Assima, Alpha Kergna, Assitan TraorĂŠ, and Naman Keita. 2015.
"Survey Research Report: Diagnostic Survey of Sorghum Production in the Sudanian
Savanna, 2014." FSP Innovation Lab Working Paper.
Smale, Melinda, Amidou Assima, Eva Weltzien, and Fred Rattunde. 2014. An Overview and
Economic Assessment of Sorghum Improvement in Mali. Michigan State University,
Department of Agricultural, Food, and Resource Economics.
Somda, Jacques, A. Joseph Nianogo, Suleymane Nassa, and Seydou Sanou. 2002. "Soil fertility
management and socio-economic factors in crop-livestock systems in Burkina Faso: a
case study of composting technology." Ecological Economics 43: 175-183.
Staatz, John, Kelly Valerie, Boughton Duncan, and DembĂŠlĂŠ Niama Nango. 2011. MALI
AGRICULTURAL SECTOR ASSESSMENT. Bamako: USAID/Mali.
Tag, Sylvia. 1994. Paysans, Etat et dĂŠmocratisation au Mali: enquĂŞte en milieu rural. Vol. 4.
GIGA-Hamburg.
Theriault, Veronique, and James A. Sterns. 2012. "The Evolution of Institutions in the Malian
Cotton Sector: An Application of John R. Commons's Ideas." Journal of Economic Issues
46 (4): 941-966.
Theriault, Veronique, and Serra Renata. 2014. "Institutional environment and technical
efficiency: A Stochastic frontier analysis of cotton producers in West Africa." Journal of
Agricultural Economics 65 (2): 383-405.
Traore, Walo. 1993. Organisation et fonctionnement des organisations paysannes (AV/TV) Ă 
l’Office du Niger: leur évolution, origines des dificultés rencontrées et leurs
conséquences, perspectives d’avenir. Thesis dissertation, IPR.
Verhofstadt, Ellen, and Miet Maertens. 2014. "Smallholder cooperatives and agricultural
performance in Rwanda: do organizational differences matter?" Agricultural Economics
45 (S1): 39-52.
102

Vroegindewey, Ryan, Veronique Theriault, and John Staatz. Forthcoming. "Coordinating Cereal
Farmers and Buyers: Evidence from Mali." Journal of Agribusiness in Developing and
Emerging Economies.
Williamson, Oliver E. 1981. "The economics of organization: The transaction cost approach."
American journal of Sociology 87 (3): 548-577.
Wollni, Meike, and Elisabeth Fischer. 2015. "Member deliveries in collective marketing
relationships: evidence from coffee cooperatives in Costa Rica." European Review of
Agricultural Economics 42 (2): 287-314.
World Bank. 2016. World Development Indicators. http://data.worldbank.org/datacatalog/world-development-indicators.

103