ESSAYS ON MARKET ACCESS THROUGH COLLECTIVE ACTION:
ANALYSES OF FACTORS AFFECTING THE FORMATION, SUCCESS, AND
IMPACT OF FARMER MARKETING GROUPS IN KENYA

By
Julius Kirimi Sindi

A DISSERTATION
Submitted to
Michigan State University
In partial fulfillment of the requirement
For the degree of
Agricultural, Food and Resource Economics-Doctor of Philosophy
2013

ABSTRACT
ESSAYS ON MARKET ACCESS THROUGH COLLECTIVE ACTION:
ANALYSES OF FACTORS AFFECTING THE FORMATION, SUCCESS, AND IMPACT OF
FARMER MARKETING GROUPS IN KENYA

By
Julius Kirimi Sindi
Many smallholder farmers in developing countries suffer from poor access to vital financial
services, improved inputs, and product markets. Collective action (CA), through the fostering of
organizations such as farmer marketing groups, is frequently advocated as a solution to these
problems. This study looks at the factors that contribute to the ability of such groups in semi-arid
areas of rural Kenya to emerge, survive, and grow; the willingness of farmers to join and
patronize them; and the impact of these groups on farmers’ incomes and price risk.
This work is based on transaction cost economic theory, which explains why economic
agents choose different institutional arrangements (“governance structures”) to mediate different
types of economic transactions as the structure of transaction costs associated with the exchange
vary (Williamson, 1979). We also use the work of Mancur Olson (1965) and Elinor Ostrom on
CA to help frame our analysis.
Using household panel and producer-level data collected from Kenya in 2003, 2005, and
2007, the dissertation is organized in five chapters. Chapter 1 reviews briefly the literature on
CA and the study area background. Chapter 2 explores in depth the conditions necessary for
effective and sustainable producer marketing groups (PMGs) to emerge. The study utilizes both

qualitative approaches (case studies) and quantitative approaches (fractional logit and quantile
regression) in the analysis. We develop a Group’s Success Index and a Group Analysis
Framework to identify intervention areas to improve groups’ success rates. Chapter 3 explores
the correlates of participation in PMGs (whether or not to join the organizations), patronage (the
share of sales through the PMG once one has joined) using a Heckman selection model; and the
number of CA efforts a farmer joins, estimated using Poisson regression. Chapter 4 examines the
impact of PMGs on members’ crop incomes and the price risk they face for their crops. We
apply a difference in differences (DiD) model for this impact assessment. Chapter 5 discusses
important results throughout the study as well as policy implications of the results and limitations
in the study.
The results from chapter 2 show that the presence of an altruistic leader or a “core
group”, good governance structures, diverse activities and regular financial subscription to the
group improves success rate. Using a blend of quantitative and qualitative (methods
triangulation) analysis, the results provide a broader and a deeper perspective of producer
marketing group dynamics and sources of their success. In chapter 3, we find that households
that had received assistance from a development agent were more likely to join a group and
increase patronage. Access to mobile phones reduced farmer CA participation. Factors such as a
democratic group, diverse membership, reciprocity, and risk-mitigating strategies increased CA
participation. In chapter 4, the study finds that PMGs do not reduce members’ price risk;
however, participation did improve crop incomes for PMG members compared to non-members.

Copyright by
Julius Kirimi Sindi
2013

DEDICATION

To my dear wife, Lilian W. Kirimi
And our lovely daughters Kendi and Gakii,
Who cheered me and endured the many days
I spent away from their lives to finish this work.
And to my loving mom who sacrificed so much
for me and prayed for the completion of this work.

v

ACKNOWLEDGMENTS

The completion of this dissertation and the entire graduate program would not have been
possible without the support and encouragement from many people and institutions. I am deeply
indebted to Dr. John Staatz, my major professor, who patiently mentored me through this
program and supervised this dissertation. I consider it a privilege to work with a man like him
and to you John, I express my appreciation. He read patiently numerous drafts and offered
invaluable suggestions to get to this level. I am also indebted to my guidance committee
members: Dr. Jeffrey Wooldridge for helping me greatly in my econometric analysis, Dr. Scott
Swinton, Dr. Robert Meyers and Dr. Christopher Peterson for their diligence in reading my work
and offering positive feedback.
I should also mention the valuable support of Dr. Shiferraw Bekele, Dr. Said Silim and
the staff at ICRISAT Nairobi, who facilitated my one-year fellowship for data collection. I would
have not done much without your assistance and financial support. Special gratitude goes to
CAPRi for giving me a grant that allowed me to stay in Kenya for data collection. Thank you
Ruth Meinzen-Dick and Helen Markelova for the facilitation at CAPRi; you made my life easy. I
am also grateful to the department of Agricultural, Food, and Resource Economics and the Food
Security Team for their financial assistance at various stages of my program. I also want to thank
Michigan State Graduate School for granting to me a Dissertation Completion Fellowship that
helped me concentrate on dissertation writing and to the MSU-USAID Food Security III
Cooperative Agreement for graduate assistantship support. Thank you again to the Department
of Agricultural, Food, and Resource Economics for granting me the summer bridge fellowship

vi

that helped me work through summer of 2009 on my dissertation. Thank you, Dr. Crawford, for
your encouragements from the beginning of my program and your support up to the last day.
Thank you, Dr. Larry Hamm and Dr. Dave Weatherspoon for believing in me.
The support of my fellow students at the department made it possible to move on. I
treasure the support from my wife Lilian who over the years read my work and listened to me
patiently. Your support and advice Vandana Yadav, Honglin Wang, Julius Okello, Meeta
Punjabi, Benjamin Gramig, Wei Zhang, the late Lesiba Bopape, Elan Satriawan, Antony
Chapoto, Mary Mathenge, and Zhiying Xu at various stages of my work and program is
appreciated.
There are many others who made my stay in the department very memorable: Chifumi
Takagi, Tomokazu Nagai, Marthe Diallo, Hikuepi (Epi), Ramziath Adjao, Milton Ayieko, Milu
Muyanga, Keneilwe (Kay), Subu, Abdoul Murekezi, Shan Ma, Fang Xie, Alyse, Athur Mabiso,
Nicole Mason, Wolfgang Penjuan, Sonja, Alexandra, Alda Tomo, Sara, Shaufique, Angula, Eric
Bailey, Thasanee, Aleksan, Nathalie, Mgenyi, Andrew Kizito, Song Feng, Malika Chaudhry,
Laura Donnet, Lourdes Martinez, Sarma Aralas, Denys Nzalov, Monika Tothova, Ricardo
Labata, Lili Gao, Joshua Ariga, Jenny Cairns, Megan, Danilo and all others who made my stay in
the department enjoyable, thank you.
Many others have made my stay at Michigan State University enjoyable. I appreciate
Lucy and Gilbert Nyaga— you were my family and just a call away. Thank you, Muthoni
Imungi, Patrick Ogutu, Emma Ogutu, Betty Okwako, Esther Mueni, Betty Kiplagat, Racheal
Ayieko, Amanda James, Achila David, Esther Gachango, Esese, Maureen, Phoebe Fay, Wez,
Jenny Kinoti, Amrita Sen, Zarini, Patience Mshenga, Mwanarusi Saidi, Mukwiti Mwiinga,

vii

Nicole Fisk, Danny and Annalie Campos, Drs Evangeline and Rex Alochilja. My bible study
group Dr. Alvin Smucker, Mike Doktar, Eric, Potter, Doug. Thank you, Emmy and Jason Block,
for adopting me into your family. Thank you, Mary Mollel and Dr Ngonyani, for coming to my
aid when I needed housing.

viii

TABLE OF CONTENTS
LIST OF TABLES .……………………………………………………………………………………xiii
LIST OF FIGURES .. ..………………………………………………………………………………….xiv
Chapter 1 : Essays on Market Access through Collective Action................................................................. 1
1.1 Introduction ......................................................................................................................................... 1
1.2 Background on farmer collective action in Kenya .............................................................................. 4
1.3 Organization of the study .................................................................................................................... 8
1.4 Brief description of the project that led to the study ........................................................................... 9
1.5 Brief description of the study areas................................................................................................... 12
1.6 Data description ................................................................................................................................ 15
Chapter 2 : Pre-conditions for emergence of successful collective action for improving farmer access to
markets: Evidence from semi-arid areas in Kenya ..................................................................................... 19
2.1 Introduction ....................................................................................................................................... 19
2.1.1 Research questions ..................................................................................................................... 20
2.1.2 Hypotheses ................................................................................................................................. 20
2.1.3 Methods...................................................................................................................................... 21
2.1.4 Data and study areas .................................................................................................................. 22
2.1.5 Organization of the chapter ........................................................................................................ 23
2.2 Literature review and theoretical framework .................................................................................... 23
2.2.1 Literature review ........................................................................................................................ 23
2.2.2 Theoretical Framework .............................................................................................................. 29
2.3 Why PMGs did not emerge in semi-arid areas in Kenya .................................................................. 37
2.4 Methods............................................................................................................................................. 37
2.4.1 Group performance .................................................................................................................... 37
2.4.2 Case study .................................................................................................................................. 43
2.4.3 Producer marketing group’s classification/stratification ........................................................... 43
2.4.4 Design of a group analysis framework....................................................................................... 46
2.5 Results and discussion ...................................................................................................................... 48
2.5.1 Geographical location of the groups according to their categorization ..................................... 48
2.5.2 Regression results ...................................................................................................................... 49
2.5.3 Case study I (Commercial groups)............................................................................................. 56
2.5.4 Case study II (Food crop based groups) ..................................................................................... 64
2.5.5 Case study III (Reconstituted groups) ........................................................................................ 69
2.6 Testing of the group analysis framework .......................................................................................... 72
2.6.1Descriptive Analysis of the Group Analysis Framework ........................................................... 73
2.6.2 Regression results of the group analysis framework factors ...................................................... 78
2.6.3 Implementation of the group analysis framework ..................................................................... 84
2.7 Conclusions ....................................................................................................................................... 88

ix

Chapter 3 : Determinants of farmer’s participation in collective action, patronage, and intensity: the case
of farmers in semi-arid areas in Kenya ....................................................................................................... 91
3.1 Background ....................................................................................................................................... 91
3.2 Research questions ............................................................................................................................ 92
3.3 Literature review ............................................................................................................................... 92
3.4 Theoretical Framework ..................................................................................................................... 98
3.4.1 Collective action participation and patronage model ................................................................. 98
3.4.2 Reduced-form market choice ................................................................................................... 100
3.4.3 Hypotheses ............................................................................................................................... 101
3.5 Methods........................................................................................................................................... 104
3.5.1 Decision to participate in the PMGs and patronage ................................................................. 104
3.5.2 Indexes ..................................................................................................................................... 108
3.5.3 Collective action intensity model ............................................................................................. 108
3.5.4 Data .......................................................................................................................................... 112
3.6 Results ............................................................................................................................................. 113
3.6.1 Project placement impact on results......................................................................................... 113
3.6.2 Descriptive statistics ................................................................................................................ 113
3.6.1 Participation model .................................................................................................................. 118
3.6.2 Group patronage....................................................................................................................... 124
3.6.3 Collective action intensity ........................................................................................................ 130
3.7 Conclusions ..................................................................................................................................... 135
Chapter 4 : Can membership and participation in Producer Marketing Groups contribute to improved
smallholder producers’ welfare? Evidence from semi-arid areas in Kenya.............................................. 138
4.0 Background ..................................................................................................................................... 138
4.1 Research questions .......................................................................................................................... 139
4.2 Literature review ............................................................................................................................. 140
4.3 Theoretical framework .................................................................................................................... 144
4.4 Estimation strategy.......................................................................................................................... 151
4.4.1 The Difference-in-Differences Estimator ................................................................................ 153
4.4.2 Assumptions for an unbiased estimator ................................................................................... 156
4.5 Data ................................................................................................................................................. 161
4.6 Results ............................................................................................................................................. 161
4.6.1 Descriptive statistics and initial discussion .............................................................................. 161
4.6.2 Livestock value ........................................................................................................................ 166
4.6.3 Welfare measures trend ............................................................................................................ 166
4.6.4 Price volatility .......................................................................................................................... 167
4.6.5 Random Effects or Fixed Effects model .................................................................................. 175
4.6.6 Correlates of crop production value ......................................................................................... 175
4.6.7 Price volatility regression results ............................................................................................. 182
4.7 Conclusions ..................................................................................................................................... 186
x

Chapter 5 : Conclusions ............................................................................................................................ 188
5.1 Policy implications.......................................................................................................................... 191
5.2 Study limitations and recommendations for further studies ........................................................... 191
REFERENCES …………………………………………………………………………………………192

xi

LIST OF TABLES
Table 1-1 Distribution of households and groups surveyed in the panel ...................................... 17
Table 2-1: Regional location of groups according to their categorization.................................... 48
Table 2-2: Test of the success score index against self-assessed success rating .......................... 49
Table 2-3: Factors determining group success.............................................................................. 52
Table 2-4: Cross-tabulation of group performance and important factors ................................... 73
Table 2-5 : Model to test the group analysis framework .............................................................. 78
Table 2-6: Testing of the factors that used in the group analysis framework ............................... 82
Table 2-7 Summary of the group analysis framework.................................................................. 85
Table 3-1: Social network impact on collective action ............................................................... 114
Table 3-2: Asset ownership effect on collective action .............................................................. 116
Table 3-3: Relation of transaction and other costs relation with collective action ..................... 117
Table 3-4: Sources of information and its influence on collective action .................................. 118
Table 3-5: Factors that affect collective action participation...................................................... 119
Table 3-6: Factors that influence members’ level of group patronage ....................................... 124
Table 3-7: Factors that affect the number of collective action groups joined ............................ 131
Table 4-1: Variations in various measures of household welfare measure from 2003 to 2007.. 162
Table 4-2: Value of livestock ownership across PMG members and non-members .................. 166
Table 4-3: Welfare measures by Gender .................................................................................... 167
Table 4-4 Determinants of net crop income ............................................................................... 177
Table 4-5 Determinants of crop price volatility for Eastern and Western Kenya ...................... 183

xii

LIST OF FIGURES

Figure 1-1: Map of the areas where the survey was conducted in Kenya .................................... 14
Figure 2-1: Group success histogram ........................................................................................... 41
Figure 2-2: Group Success Rate Analysis Framework ................................................................. 47
Figure 2-3 : Polynomial plot of group-analysis framework test ................................................... 80
Figure 4-1: Rainfall patterns in the study areas from 1995 to 2005 ........................................... 164
Figure 4-2: Movement of incomes and net crop production value over the study period .......... 165
Figure 4-3: Different crop price means and standard deviations for 2006 ................................. 168
Figure 4-4: Graph of Groundnuts crop price for the three-survey .............................................. 171
Figure 4-5: Graphs of Groundnuts crop price for the three-survey period ................................. 171
Figure 4-6: Graphs of Pigeon peas and Groundnuts crop price for the three-survey period ...... 172
Figure 4-7: Distribution of price standard deviation of groundnut for Western Kenya (2007) .. 173
Figure 4-8: Distribution of price standard deviation of pigeon peas for Eastern Kenya (20032007) ........................................................................................................................................... 174

xiii

Chapter 1 : Essays on Market Access through Collective Action
1.1 Introduction
In the last few years, we have witnessed many changes in the agricultural sector in both
developed and developing countries. In developed countries, there is a push for consolidation
and creating bigger farms, whereas production in developing countries continues to rely on
smaller and smaller farms (Dimitri, et al., 2005, Gebremedhin and Christy, 1996). The smallscale farms are getting smaller over time due to land fragmentation (Niroula and Thapa, 2006).
And access to product markets has been and continues to be a challenge to these small farmers.
This is because they produce small quantities per household and have small marketable surpluses
that are distributed over huge geographical areas. In the past, in many developing countries,
government attempted to control marketing of most crops through marketing boards and
cooperatives and built chains of grain collecting centers in the rural areas accessible to the
farmers. But following structural adjustment programs in the 1990s, input supply and marketing
services shifted from greater government control through parastatals and marketing boards to
more private-sector participation (Barrett and Mutambatsere, 2008). Although in some cases
1

producer prices improved after the phasing out of government control through boards , some
smallholders who sell through spot market with little competition did not benefit (Fafchamps,
2004). There was lack of an institutional framework to support the small-scale farmers to gain
from liberalization benefits. In general, farmers who market through the spot markets and
middlemen usually get lower prices than if they were able to sell to the bigger market in the
cities (Shiferaw, et al., 2008). Therefore, even with better commodity prices, smallholder farmers
1

For instance from 2007 to 2009, the world has witnessed commodity price hikes and some
producers have benefited from these high prices (FAO. 2008. "Crop Prospects and Food
Situation." Global Information and Early Warning System on Food and Agriculture (2, April), --. 2008. "Food Outlook " Global Market Analysis June
1

may not be in a position to benefit from these new opportunities. This is because without
competitive markets, prices are not passed through the marketing chain efficiently. Also without
good access to distant markets that can absorb excess local supply, even the adoption of more
productive agricultural technologies will lead to a drop in farm-gate product prices (Barrett and
Mutambatsere, 2008). One solution is collective action, where farmers can bulk their products
and market as an organization rather than individually.
In Kenya, there are two types of collective action employed by farmers, namely
cooperatives and farmers’ organizations. In contrast to cooperatives, which are governed by an
2

act of parliament and have a government ministry dedicated to their promotion, farmers’
organizations are operated with minimal government regulation or support. This dissertation is
geared at exploring the second group. This is because, unlike cooperatives, farmers’ groups are
found mainly in areas of low agricultural potential. They are characterized by few capital
resources and are often led by members with little management training. However, when these
organizations are successful, farmers are able to derive similar benefits to those accessible
through cooperatives. One type of farmers’ group is producer marketing organizations (PMGs),
which have the potential to be a tool for improvement of economic welfare of smallholder
farmers by providing solutions to market failures that often characterize rural agricultural
markets. In the presence of market failure, a PMG may be efficient in any of three ways or a
combination of the three: a) marketing the produce at a lower marketing and transaction cost
than the alternative channel, which is usually the spot market, b) if the buyer exerts market

2 In Kenya, cooperatives societies are governed by an act of parliament (ACT NO. 12 of 1997 Co-operative Societies Act amended in 2004) relating to their constitution, registration and
regulation by the central government. The commissioner of cooperatives can order a probe, audit
or even change of management of any cooperative society she feels does not meet the laid down
requirements.
2

power and pays lower prices per unit of a commodity than would prevail under perfect
competition, the PMG may provide countervailing power; and c) the PMG may allow farmers to
3

access factors of production at a lower cost (Sexton and Iskow, 1988) .
Previous studies in Kenya have found that collective marketing allows small-scale farmers to
pool their produce and spread the costs of marketing over a large amount of marketable surplus.
This enables them to reap economies of scale and scope. Collective marketing also enables
smallholders to enhance their ability to negotiate collectively for better prices, improving their
market power (Shiferaw, et al., 2005, Shiferaw, et al., 2008). The basic functions of farmers’
organizations include input supply, storage, processing, bulking or aggregating and selling
products provided by their members, credit sourcing, training, member education, and political
action (Von-Pischke and Rouse, 2004). Agricultural producer groups have been promoted both
in developing and developed countries as a solution to the coordination and marketing failures
farmers face. Justification for farmer organizations includes their potential to play a critical role
in both the delivery and coordination of services to smallholder producers. It also includes their
ability to reduce transaction costs (Dorward, et al., 2005) and deal with problems of barriers to
market access and economies of scale (Gonzalez, 2008).
In general farmers’ organizations (FOs) and other collective action (CA) can be used to
access economies of scale and scope, improve bargaining power, provide inputs, and manage
risk in many rural communities in developing countries (Barrett and Mutambatsere, 2008). These
actions are in the form of cooperatives or PMGs that can be set up for defensive or proactive
purposes. Defensive strategies focus on protecting small farmers from unfair competition, while
3

Producer marketing groups also play many other roles such as members’ education, access to
factors or production, and social interactions.
3

proactive strategies respond to new business opportunities. Either strategy enables farmers to
improve their economic condition by working together in ways that are more productive than
working individually (Von-Pischke and Rouse, 2004). Understanding these FOs is therefore an
important step in helping smallholder farmers’ access factors of production and marketing
services. This dissertation is an addition to work done by other researchers and adds to the body
of knowledge studies that seek understanding of farmers’ organizations. A lot of work has been
done on analyzing the impact of cooperatives in developed countries, but little has been done to
understand FOs that mainly deal with smallholder farmers growing food crops for local markets
in developing countries. Even the little work done in developing countries is largely focused on
cooperatives, usually dealing with export crops. In Kenya and many other Sub-Saharan
countries, these cash crops are only grown in small parts of the country, whereas most of the
country grows cereals, pulses, and legumes for local use, including in areas that grow export
crops. Understanding factors affecting smallholder FOs in terms of organization dynamics,
operations, and patronage and their impact on farmers’ welfare may help policy makers to target
this part of society.
1.2 Background on farmer collective action in Kenya
Kenya has a long history of formal cooperatives involving smallholder farmers in high
4

potential areas (Heyer, 1976), which were initiated during the colonial era. Formal or informal
collective action groups perform various functions, such as income generation, asset building,
commodity marketing and social functions (Place, et al., 2004). However, most of these formal
groups were formed in the Kenyan highlands, where cash crops for export are grown. Few
formal farmers’ organizations emerged in the semi-arid areas of Kenya that mainly grow cereals
4

This study refers to groups that have not gone through formal registration as informal.
4

and pulses. This dissertation is focused on smallholder farmers’ organization in semi-arid areas
5

in Kenya . The semi-arid areas are characterized by poor rainfall and lack of irrigation systems.
Therefore, farmers mainly grow food crops and market only small surpluses. Despite the need
for some form of collective marketing services, few formal organizations were formed in these
areas. However, in the recent past, some farmers have started income diversification by growing
horticultural crops, and some have formed self-initiated farmers’ organizations (Okello and
Swinton, 2006). Formation of formal PMGs could enable producers to add value, differentiate
their products, and bulk their produce for the market, thus reducing per unit costs. However, it is
safe to say that most of the agricultural produce in these areas is marketed through periodic open
markets. These markets operate with few formal regulations and lack public goods such as
physical infrastructure, contract law, public market price information systems, and codified
product grades and standards (Fafchamps, 2001).
6

In the study areas (the Eastern and Western parts of Kenya ), many of the smallholders
grow pulses (pigeon peas, groundnuts, cowpeas, beans, and many others) and cereals (such as
sorghum, millet, finger millets and maize) under conditions characterized by unreliable rainfall
and sandy soils. Farming in these areas is characterized by low inorganic fertilizer and manure
use. Because of increased soil erosion due to reduced soil cover and continuous cropping, soils
have become nutrient deficient (particularly in regard to nitrogen and phosphorus). Poor soils
and low use of improved seeds lead to low crop production per unit area (Gachimbi, et al., 2002).
These semi-arid areas are often far removed from markets in major urban areas, and the road
5

The dissertation study areas are described later under the data and study area section of this
chapter.
6
In Eastern region, the study was conducted in Mbeere and greater Makueni districts, whereas in
the Western region, the survey was in Homabay, Siaya and Teso districts (see the map in a later
section of this chapter).
5

conditions are poor, often becoming impassable during the wet season. To improve food security
7

in these areas, development agencies such as ICRISAT have been promoting improved varieties
of some cereals and pulses. But as has been the case with many new agricultural technologies in
developing countries, these efforts have encountered low adoption rates (Moser and Barrett,
2006). One way to encourage uptake of new technologies in the semi-arid areas is by linking
producers to the market (Feder, et al., 1985, Zeller, et al., 1997). This is the “pull strategy”,
which is likely to work only if the technology is profitable (Doss, 2003). Many farmers market
their surplus products in local markets or at farm-gate to intermediaries, usually at low prices
(Shiferaw, et al., 2005). That is the reason why PMGs are being promoted to address this
marketing problem.
However, there is no consensus among researchers that PMGs are the best solution for
the farmers’ marketing woes. The debate about the ability of PMGs to address smallholdermarketing problems emanates from their past experiences with farmer cooperatives in
developing countries. Cooperatives have had mixed success in developing countries in the past.
In Kenya, there were many cooperatives that were used for providing marketing and other
services to farmers. However, due to mismanagement, many of them failed. The reasons for their
failure are explored further in Chapter 2. Although there are potential problems with any
collective action, some researchers have argued that for small-scale farmers to survive there are
few alternatives to farmers’ organizations of some sort. This is reflected in what Chirwa et al.
(2005) refer to as failure of a critical part of the Washington consensus. Reasons for failure of
liberalization have been attributed to halfhearted liberalization and weak or non-existence of
7

International Crop Research Institute for Semi-Arid Tropic (ICRISAT), which belongs to the
alliance supported by the Consultative Group on International Agricultural Research (CGIAR)
and works on crops grown by the poor in semi-arid areas of developing countries.
6

critical institutions (Jayne, et al., 2002). Therefore, some researchers have argued that
smallholder farmers will benefit from increased liberalized output and input markets if a greater
role is played by farmers’ organizations (Bingen, et al., 2003, World-Bank, 2002). Even
researchers opposed to the views of Washington consensus also argue that farmers’ organizations
are important for access to services (Peacock, et al., 2004) or see them as a new form of
economic organization that replaces government agencies phased out by structural adjustment
programs (Collion and Rondot, 2001). In their study, Lele and Christiansen (1986) conclude that
8

although farmers’ organizations cannot fill the gap left by marketing boards , with enough
external support from Government and NGOs to deal with the complex organizational,
technological, and financial requirements, FOs can be beneficial to producers in rural areas.
These new producer organizations are promoted in developing countries as they move away
from welfare transfers toward empowerment for income generation and social development. We
argue that liberalization has brought challenges and opportunities, and an individual smallholder
farmer often lacks the skill and resources to cope with the resulting uncertainty as well as take
advantage of the emerging opportunities. Thus, collective action can be an important vehicle for
smallholders to access better markets for their products and other supporting services (Dorward,
et al., 2005, Gonzalez, 2008, Omamo, 2005, Shiferaw, et al., 2005). However, the main
challenge is getting FOs to succeed, and often they have not been successful, implying that they
may not always be a solution to the pesky market access problem.
8 Before liberalization, governments created marketing boards as a marketing monopoly for
different crops. The aim was to improve a particular crop’s production and improve farmers’
income. However, most boards failed to deliver on their promise. Usually they paid low prices,
which made it unprofitable to grow the crops, and delayed payments were common. This led to a
decline in the production of most of the crops they handled. The decline was due to farmers
neglecting the affected crops and some outright moving out of the crop production. This is still a
thorny issue in Kenya because crops such as coffee and tea are protected by law and cannot be
uprooted once planted without government approval.
7

In Eastern Kenya, three years after the formation of PMGs, there was evidence that the
PMGs paid a premium of an average of 6 Kshs/kg (22-24%) above the price paid by
9

intermediaries per kg of pigeon peas (Obare, et al., 2006, Shiferaw, et al., 2008). However, the
big concern is that many of the groups in the early part of their lifespan will not have enough
volume to move to the flat portion of the L-shaped per-unit cost curve. Such a scenario can lead
to their marketing cost being sometimes larger than that of the competition. A solution to this
dilemma can be the formation of a bargaining organization rather than a PMG that directly
handles the product. Hence, group success will hinge on the organization’s ability to control
enough products to force a price concession from buyers. Producers must then be committed to
the collective bargaining effort and not cheat on the resulting agreement by selling for less than
the bargained-for price. This has been achieved by some horticultural PMGs that do not handle
any produce but instead are essentially a bargaining entity and virtually control all the produce in
a certain geographical area.
1.3 Organization of the study
From the preceding brief discussion of the background on farmers’ organizations,
especially in Kenya, we find that collective action efforts have the potential to improve the
welfare of rural small-scale farmers. It is against this background that this dissertation seeks to
extend the work done by other researchers by adding to the body of knowledge more insights
into the inner working of FOs in the context of semi-arid agriculture in East Africa. No work we
know of has thoroughly looked at operations of FOs at the level of PMGs. Most of the work
done has looked at cooperatives, and more so in developed countries and in emerging economies.
A major challenge faced in empirically analyzing the performance of farmers’ groups in
9

Intermediaries are traders and brokers operating the area who sell to the wholesalers.
8

developing countries is lack of record keeping, which makes it very difficult to analyze them
effectively. However, it is important to understand the process of group formation, group
dynamics, and management. There is also a need to understand what makes some groups succeed
while others fail. These questions are dealt with in chapter two of the dissertation. Members also
do not always patronize the groups effectively. This dissertation also seeks to investigate the
factors that contribute to group patronage. In doing so, we investigate the factors that are
important in the decisions to join a farmer group and to use its services. These questions are
investigated in chapter three. There is also the ongoing debate alluded to earlier on the role that
farmers’ organizations can play in improving their members’ welfare. We will investigate in
chapter four whether being a member of a group improves a household’s revenue from crop
income and reduces annual price volatility. From all this work, we hope to contribute to a better
understanding of FOs in developing countries and give policy makers information on
opportunities to make these organizations more effective and targeted where they are needed.
Chapter five concludes the dissertation by giving the overall findings and recommendations.
1.4 Brief description of the project that led to the study
For smallholder farmers to have food security and market some produce for income, a
steady supply of high-quality seed is central for sustainable production. Legumes in particular
are an important source of food, feed, oil, and cash for many smallholder farmers in the semiarid tropics. They fix atmospheric nitrogen biologically into mineralized form accessible for crop
nutrition and hence sustain productivity in N-limiting systems. Legumes in Kenya are
traditionally grown as subsistence crops, and seed supply is primarily through the informal
sector. The yields are low, and subsequently inadequate volumes of marketable quantities make
legumes’ commercialization difficult. There are several production and marketing constraints

9

that have contributed to the low legume production among smallholder farmers. These are low
quality seed, poor crop management practices, and poor market linkages. A USAID-funded pilot
project, Technology Application for Rural Growth and Economic Transformation (TARGET)
that ran between 2002 and 2004 in Kenya identified compromised seed quality, poor-quality
produce, and price fluctuations as some of the problems farmers face as the consequences of
using poor-quality seed (ICRISAT, 2006). Therefore, capacity building of producers, traders, and
processors and increased availability of high-yielding, disease-tolerant varieties with both farmer
and market-acceptable traits are pivotal for sustained production. A new project to address these
shortcomings, called the Lucrative Legumes Project (LLP), was started in 2005.
The objective of LLP was to address identified constraints such as poor seed supply, lack
of proper crop production techniques, and inadequate access to markets. LLP was also to develop
a seed supply system capable of sustaining innovation beyond the life of the project. The project,
which ran from 2005-2007, was funded by the United States Department of Agriculture (USDA).
It was implemented by TechnoServe, in partnership with Catholic Relief Services (CRS) and
ICRISAT with collaboration with many private and public institutions. The LLP project was
implemented in 10 districts in Kenya—six in Western Kenya (Bomet, Homa Bay, Kisumu,
Siaya, Suba, and Teso), and four in Eastern Kenya (Machakos, Makueni, Mbeere, and Kitui)
(ICRISAT, 2006).
The problem of poor seed stems from the fact that most legumes are self-pollinated, with
no clear distinction between seed and grain, so farmers save harvested grain to use as seed in the
next season. While this might reduce input costs for smallholder farmers, it has discouraged the
development of a commercial legume seed sector, and most seed companies shun the legume
business. ICRISAT and its partners have been working to address the legume seed supply
10

constraint by educating farmers on the benefits of high-quality improved seed and of the
potential increased production through modest investments in fresh improved quality seed. The
LLP project relied on ICRISAT’s breeding technology and extension support and worked in
partnership with Kenya Agricultural Research Institute (KARI) for seed multiplication, CRS for
extension services to the farmers and TechnoServe to link farmers to the markets. ICRISAT
supplied high-quality legumes seeds and trained farmers and local extension staff on seed
multiplication and bulking techniques.
To address the problem of access to the markets for the smallholder farmers, LLP
encouraged the formation of PMGs through a farmer education campaign. The rationale for
encouragement of PMGs was to reduce marketing costs for individual farmers and help them to
establish links with reliable product markets. In Western Kenya, TechnoServe also helped
establish an umbrella farmers’ trading company, named the Kenya Smallholder Farmer
Investment Company (KESFIC). Launched in July 2006, the company was to be wholly owned
by smallholder farmers and was expected to become the vehicle for facilitating input supplies
and collecting and bulking grain for collective marketing from the smaller PMGs. It was
envisioned that KESFIC was to provide farmers with extra services after the life of the LLP
project, such as training, sourcing of credit, and transport services. However, by the time of the
survey that generated the data for this study, in 2007, the trading company was still struggling to
stand on its feet and provide services to farmers. In Eastern Kenya, no such trading company was
formed, and each individual PMG had to market its members’ produce on its own.
This study was conducted in sample of the districts in which the LLP project operated
from 2005 to 2007, and drew upon earlier surveys that ICRISAT carried out with farmers in
these areas.
11

1.5 Brief description of the study areas
The study is based in Kenya, with all the districts surveyed being in semi-arid lands.
They have low density of paved roads and limited access to major marketing centers. Semi-arid
areas in Kenya are characterized by low and highly variable bimodal rainfall that varies on
average between 500 and 800 mm per year. Soils vary in depth depending on the parent rock
material and slope. They are generally low in organic matter and deficient in nitrogen and
phosphorus, but with adequate levels of potassium. Low infiltration rates and susceptibility to
sealing making the prevailing soils prone to erosion, as heavy rains fall mainly at the beginning
of the growing seasons when the land is bare (Gachimbi, et al., 2002). Farming systems in these
areas are based on rain-fed crop production integrated with varying levels of livestock rearing,
and where water supplies permit, there is limited furrow irrigation. The main rain-fed crops are
maize, beans, pigeon peas and groundnuts. Depending on the district, these are grown in
monoculture or as mixed crops complemented by smaller areas of chickpeas, cowpeas, sorghum,
and millet. These semi-arid lands have experienced increasingly frequent droughts in the last two
decades. Households in these areas have not had sufficient time to recover from previous
droughts before the next one hit. Consequently, farmers produce limited marketable surplus,
increasing their vulnerability to food insecurity and poverty.
This dissertation uses both household- and producer-level datasets. Three sets of
household data sets were obtained in 2003, 2005, and 2007. The producer-organization level
dataset used was collected in 2007. The author participated in collecting all the 2007 data both at
the household and group levels through a grant by Collective Action and Property Rights
institute (CAPRi) in conjunction with ICRISAT and conducted most of the PMG-level
interviews. We also drew information from the previous rounds of survey conducted by
ICRISAT in order to have multi-year comparisons.
12

The data was collected by the International Crops Research Institute for the Semi-Arid
Tropics (ICRISAT) in Mbeere, Makueni, Homabay, Siaya, and Teso

10

districts in Eastern,

Nyanza, and Western Provinces of Kenya (see figure 1-1). ICRISAT targeted the areas because it
believed that dry-land legumes like groundnuts, pigeon peas, and chickpeas could be exploited in
these areas to reduce poverty and food insecurity.

10

Teso district was a part of greater Busia district. It was curved out of Busia district recently
and the actual map has not been input into the GIS map software used to create fig 1-1 below.
13

Busia
Siaya
Homabay

Mbeere

Makueni
Figure 1-1: Map of the areas where the survey was conducted in Kenya
(For interpretation of the references to color in this and all other figures, the reader is

14

1.6 Data description
In 2003, a total of 400 households were surveyed in Mbeere (240) and Makueni (160) to
form the baseline dataset. This was before the PMGs were established as part of a research
project that aimed to pilot alternative institutional innovations for improving market access for
smallholders. Farmers were sensitized to form PMGs through voluntary participation.
Initially, ten PMGs, mostly based on existing local social networks of different types,
were established in Eastern Kenya. These groups were formally registered as welfare societies
(self-help groups) under the Ministry of Culture and Social Services.
From the initial sample of 400 households, some households joined the PMGs, but others
did not. This necessitated the re-sampling of the study sample to get an even distribution of
members and non-members in subsequent surveys. From the original sample, a total of 250
households remained and were re-interviewed in 2005 and 2007.
An additional 150 households were added during the 2005 survey period in Eastern
Kenya. In 2005, data was collected from 400 households (210 from Mbeere and 190 from
Makueni districts) in the ten PMG villages, comprising of 250 members and 150 nonmembers.
In 2007, households interviewed in 2005 were re-interviewed, in addition to added
households in Western Region. As shown in table 1-1 (below), 208, 189, 162, 152, 149
households were interviewed in Mbeere, Makueni, Homabay, Siaya, and Teso districts,
respectively. Two households in Mbeere and one in Makueni dropped out of the survey because
the households either were dissolved or could not be traced. Also 23, 21, 16, 15, and 15 PMGs
were surveyed in Mbeere, Makueni, Homabay, Siaya, and Teso districts respectively in 2007.
The districts were purposely chosen to get locations where the project was to start in
2003 or had already been started for the additional areas in 2007. However, care was taken to

15

include districts that cover a wide variety of zones representing semi-arid areas. A stratified
sampling procedure was followed at the district level. First, the divisions

11

were randomly

sampled, then the locations, then groups in the chosen locations. Two lists of all households were
compiled within the group service area, one with group members and the other with nonmembers, and from those lists equal numbers of households were chosen at random. If for any
reason a particular person chosen could not be located, then another name was chosen as a
replacement at random from the list.

For the repeat interviews, care was taken to get all the households previously
interviewed, and only 3 households dropped out of the survey between 2005 and 2007. A total of
850 households were interviewed in the 2007 survey. These included the 250 that were followed
from the baseline survey in Mbeere and Makueni districts; hence, the data has a short panel from
the three rounds.

11

In Kenya, the local government administration is divided starting from a province at the
highest level, followed in descending order by the district, the division, and the locations, with
the smallest units of administration being villages.
16

Table 1-1 Distribution of households and groups surveyed in the panel
Districts
Year

Mbeere

Makueni

Homabay

Siaya

Teso

152

149

15

15

Households surveyed
2003

240

160

2005

210

190

208

189

162

2007

Groups
23

21

16

Information obtained at the household level included data on socioeconomic
characteristics, assets, credit and savings, production, buying and selling, and participation in
other groups or social networks and in PMG collective marketing. In the 2007, data was
collected regarding the previous history of households’ involvement in earlier collective action
efforts, since this data was not collected in earlier surveys. Because we hypothesized that
involvement with other development agencies affects the households’ decision regarding
whether to participate in collective action, we also collected information on the history of the
household’s involvement with development agencies operating in the area. The previous surveys
did not ask any question about social networks of the households prior to the introduction of
PMGs and before the households’ decision to join or not join the groups. In the 2007 survey,
households were asked a series of social network questions regarding current and pre-existing
networks (retrospectively before the PMGs were introduced). Households were asked if they
belonged to another collective action effort before 2003. They were also asked whether they had
friends who participated in collective action before introduction of the PMG, followed by a
17

question about whether their friends, and how many, had joined the PMG. There were also
general questions to measure the trust in the community as well as communal risk management
strategies.
At the PMG level, all ten PMGs formed after the 2003 sensitization in Eastern Kenya
were interviewed, plus an additional 80 PMGs covering the entire study area. About five to
seven respondents selected from the PMG management and ordinary members served as key
informants. A focus-group approach was employed, guided by a semi-structured questionnaire.
The data obtained included objectives and aspirations of the groups when they were formed;
general group characteristics (e.g., size and composition, frequency of meetings, capital, etc.);
asset ownership (e.g., store, weighing scales, operating capital, etc.); credit access; bulking and
marketing; governance; and perceptions about major constraints limiting the PMGs’ success. The
initial questions were general in order to allow respondents to talk about their PMG, such as,
“Tell us the history of the PMG. How successful do you think the PMG is?” These were
followed up with the “why” questions that led to more follow-up questions. The approach
elicited information to construct the histories of the PMGs. A structured question followed to
record specific information from all the groups.

18

Chapter 2 : Pre-conditions for emergence of successful collective action for improving
farmer access to markets: Evidence from semi-arid areas in Kenya
2.1 Introduction
Structural adjustment programs and increased globalization that took off in the 1990s
increased hardships for many smallholder farmers in developing and transitioning countries
(FAO, 2002, Motiram and Vakulabhranam, 2007). Specific reasons for this hardship vary from
one place to another, but adverse trends in global markets, poorly organized credit markets, and
weak supporting marketing institutions have been cited as major causes (Baffes, et al., 2003,
Jayne, et al., 2002, Rao and Storm, 2003). Collective action has been touted as a way to improve
smallholder farmers’ access to necessary services (Omamo, 2005). However, the history of
collective action through cooperatives in East Africa is characterized by failures, in spite of a few
isolated successes. Yet policy makers and development agencies still look at farmer
organizations (FOs)

12

as a vehicle to use for small-scale farmers to access markets and other

supporting services. Although there is little debate about the ability of successful FOs to help
farmers in the presence of market failures, the key questions are when should they be used and
how to make them successful? If FOs are to be promoted as an alternative to government direct
action for assisting small-scale farmers, then it is important to increase understanding of their
formation, operation and sources of success.
To date, most studies geared to understanding of farmers organizations have been
focused on cooperatives. Though understanding cooperatives is important, they have been most
effectively employed in developed countries. Yet most smallholder farmers in rural areas in poor

12

In this study, we use farmers’ organizations (FOs) as a generic term for all types of farmers’
collective action effort.
19

countries use smaller informal FOs and not cooperatives for services. Usually FOs are initiated
and organized differently from many cooperatives. Unless there is better understanding of these
smaller FOs, then efforts by policy makers and development agencies geared to helping rural
smallholder farmers might not deliver the intended benefits. And the cycle of group formation
and failure will continue.
2.1.1 Research questions
This study contributes to literature on collective action by identifying conditions
necessary for PMGs

13

to emerge, which could be similar to the findings regarding the conditions

leading to the emergence of cooperatives, but not necessarily identical. In particular, the study
focuses on identifying what ingredients were lacking in the semi-arid areas in Kenya that
prevented PMGs from emerging before TARGET and the Lucrative Legume Program launched
by ICRISAT and its partners in 2003-2004. The study attempts to determine the success and
sustainability of smallholder collective action efforts in these areas. In this regard, the study
focuses on addressing the following research questions for these areas:
1.

What are the conditions necessary for smallholder farmers’ organizations to
spring up?

2.

What are the key ingredients of success for smallholder farmers’ organizations?

3.

Is it possible to develop a series of questions that helps to analyze the success
rate of a farmer organization and identify intervention areas?

2.1.2 Hypotheses
We test the following propositions based on the theories and literature reviewed:
13

Producer marketing groups (PMGs) are the forms of FOs mainly used by the smallholder
farmers to access various services such as marketing, inputs, new technology, and information.
20

1.

Collective action will emerge where there is high asset specificity and, in
particular, commodity perishability.

2.

Producer marketing groups will be more successful if they operate in a small
geographic area for ease of coordination.

3.

PMGs will be more successful where they have an informed leadership that is
able to articulate the members’ aspirations.

4.

Strong social networks are important for group success.

2.1.3 Methods
Different approaches have emerged over the years that contribute to the understanding
and modeling of the internal organization of collective action. This essay follows mainly the
transaction costs economics framework (Williamson, 1985, 1975). We also use game theory to
explain some phenomena. The advantages of applying a transaction cost approach to PMGs or
business firms in general is the understanding it offers about organizational strategy and of why
the market is organized as we see it.

In this chapter, we extend the work done by Shiferaw, et al. (2008) by developing an
index for measuring the PMGs’ success. Then we evaluate how well each group has performed
in meeting its members stated objectives and expectations. The study further develops a new
group analysis framework that we use to predict group success as well as identify intervention
areas to improve the success rate, based on the identified success factors. This is the first time to
our knowledge that such a framework has been developed.
Most studies on collective action analyze their data using qualitative approaches. In this
study, we apply both quantitative and qualitative methods. First, using the developed index of
21

success, we propose a model that analyzes PMGs’ success rating. Then we estimate a fractional
logit regression using the index as the dependent variable. We also disaggregate the index further
into its constituent components and run fractional logits on each. A case study approach is next
used to peer deeply into selected groups to find out if identified factors in the model are
corroborated by individual cases. A case-study approach adds more insights that are not easily
captured by an econometric model. The last section is a brief quantitative testing of the group
analysis framework followed by a qualitative analysis of two archetype groups.
2.1.4 Data and study areas
The study was conducted in Western and Eastern regions of Kenya. In Western Kenya,
PMGs were promoted, and a new overarching trading body was created to market produce
bulked by the PMGs. PMGs as well as individual farmers marketed their produce through this
new trading company. In contrast, in Eastern Kenya only PMGs were promoted without a similar
trading company. The study is therefore focused in these two areas and in particular actual areas
where PMGs were formed.
This study uses data from both household and group surveys collected in 2007 by
ICRISAT. In the randomized survey, a total of 850 households were interviewed in Eastern and
Western Kenya. And a total of 90 group interviews were conducted using a focus group
approach in the same areas. The focus groups were comprised of seven individuals drawn from
management and ordinary members of the PMGs, and care was taken to include both genders as
much as possible in the interviews. Most of the analysis in this essay utilizes data gathered from
the discussions conducted in these meetings, which involved both structured and unstructured
questions. For more information about how both surveys were conducted, see chapter one of this
dissertation.
22

2.1.5 Organization of the chapter
This chapter proceeds with the following sections. First, we provide a brief review of the
literature on collective action, and in particular, studies that target factors influencing the success
of cooperatives and farmers’ organizations. Second, the chapter discusses the methods used in
the chapter. Third, results of both quantitative and qualitative analysis are presented and
discussed. Finally, a brief conclusion of the findings is presented.
2.2 Literature review and theoretical framework
2.2.1 Literature review
Farmer groups, cooperatives, and other farmers’ collective efforts under different names
are all a form of collective action used by farmers to access various services or address a local
problem. We define collective action as voluntary act taken by a group of individuals or entities
to achieve common interests (Meinzen-Dick and Di-Gregorio, 2004). As mentioned earlier,
studies on farmers’ collective effort have focused on cooperatives. A cooperative is an
association or corporation established for the purpose of providing services on a nonprofit basis
to its shareholders or members who own and control it. A key characteristic is that residual
claims are distributed in accordance with patronage, not capital investment.
In Kenya, cooperatives are governed by an act of parliament that provides close
monitoring of the operations by the relevant government agencies. Cooperatives go through
stringent government scrutiny before registration. There is a yearly auditing by governmentapproved auditors to ensure good governance for all cooperatives, and strict following of the
cooperative constitution is usually enforced. This paper deals mainly with farmers’ organizations
formed in rural areas of developing countries and referred to as producer marketing groups
(PMGs) or community-based organizations. Registration of PMGs in Kenya is required by the

23

government. However, there are few requirements for registration of PMGs, and the major
obligations are to provide minutes of an agreement by a group of individuals agreeing to work
together, open a bank account to safeguard any members’ finances and provide a copy of the
group’s constitution. Once these conditions are fulfilled, the only possible source of monitoring
is a requirement to file a yearly report to the relevant government agency—a requirement that is
usually not enforced. Based on these differences, we claim that there are fundamental
distinguishing features between cooperatives and PMGs in the manner of formation, membership
composition, and operation. To meet the government requirements, cooperatives are usually
formed by leaders who are more educated, have better management capacity, and are probably
wealthier members of society. Usually cooperatives are also managed by professionally hired
staff, whereas PMGs are operated by members on a volunteer basis. Therefore, factors that are
necessary for a cooperative to emerge and succeed might be different from the factors that are
necessary for a PMG. As a PMG grows, it might evolve into a cooperative, which enables it to
benefit from favorable government incentives and professional management.
As in many developing countries, the history of cooperative performance in Kenya has
had mixed results. Some cooperatives, like the Kenya Tea Development Authority (KTDA), have
been generally successful, while others, like coffee cooperatives, have witnessed spectacular
failures. The dairy industry was plagued with failure of the big cooperatives such as the Kenya
Cooperative Creameries

14

(KCC) but has been successful with small cooperatives like the

Githunguri and Meru dairy cooperatives. Nyoro et al. (2005) found some general success and
failure factors in dairy and coffee cooperatives. The success factors identified were vertical

14

The current government revived KCC in 2003, which had been inactive for many years, and
now it is operating as a successful business (Atieno and Kanyinga, 2008).
24

integration, volume of membership, proper record keeping, level of manufacturing technology,
skills of management committee and staff, timely dissemination of appropriate information,
15

diversification into profitable enterprises, adoption of a strategic plan, and credit availability .
Conversely, the failure factors identified were credit burden, conflicts, external forces,
investment in non-income generating activities, non-skilled board members, poor or lack of
communication between board members and farmers, unfair competition, dishonesty by staff and
representatives, and deceitful businesspersons. Another contributor to failure was the extraction
of rent by corrupt officials (Mude, 2006). Rent seeking by officials led to lower prices than
would prevail in the spot market. Others have attributed poor performance to technological
problems and poor management (Lele and Christiansen, 1989, Wolf, 1986). These broad findings
are attributed to cooperatives but apply to any business concern.
A few studies have argued that farmers’ organizations among the poor get into problems
when external agencies start supporting them. These studies, generally found in the sociology
literature, have found some evidence that support the view that once there is external support,
elite members of the community join the groups, become leaders, appropriate the benefits to
themselves, and marginalize the poor. This is the so-called “Rockefeller Effect” (Garforth, 1994,
Gugerty and Kremer, 2004, Howes, 1997, Stiles, 2002).
Collective action outcomes could be a function of the initiation history. In Europe, the
United States and Canada, cooperatives and other rural people's self-help organizations were
established in the 19th and 20th centuries as farmer-initiated and financed self-help actions.

15

However, it is not very clear from Nyoro’s work if some of the factors mentioned are the
results of collective action or are necessary precursors for success. We hope to get more evidence
of the causality in the analysis later in this chapter.
25

Governments usually provided enabling legislation and technical assistance. By contrast, in
developing countries, this largely was not the case. Colonial and independent governments were
the main actors in initiation and driving the agenda. Cooperatives were therefore viewed more as
government-driven entities than as independent, farmer-owned, and financed self-help
16

organizations. They were used to push for government policies and objectives,

and

government officers often controlled them (Birchall, 1997, Laidlaw, 1978). In the past,
cooperatives were privileged with monopoly and or monopsony powers on agricultural inputs
and produce in the local areas. And governments used cooperatives as mechanisms for taxing
rural producers to finance national development projects (Munkner, 1992). This prevented
farmers’ ownership of the cooperatives, and politicians used them to advance their political
ambitions. Consequently, many failed when government withdrew support (Simmons and
Birchall, 2008). The bad experience left many producers suspicious of any form of collective
marketing effort (Anderson and Henehan, 2003). PMGs therefore have an uphill task educating
members on the benefits of Collective Action (CA) while changing their management styles to
erase the past negative image and heritage. However, as in the western world’s experience,
governments still have a role to play in the success of farmer organizations through provision of
a favorable enabling environment (Soedjono, 2002).
The need to exploit economies of scale, improve bargaining power, and play a
coordination role leads to some of the arguments in the literature that collective marketing needs
a critical mass of members for success (Chamberlin, 1974, Esteban and Ray, 2001, Marxwell and
Oliver, 1993, Oliver and Marxwell, 1988, Sandler, 1992). However, developing the initial

16

Most government wanted to keep the food prices low while using agriculture to fund the
growth of other sectors.
26

coalition is a major problem because it may require some selfless producers to be the initial
entrepreneurs. Reaching agreement between large numbers of producers at the initial stages can
be a challenge. This is because of differences in bonding and social networks that keep
individuals together or apart. In a situation where there is a mass mobilization, the
entrepreneurial costs are spread among many individuals (Sexton and Iskow, 1988), making it
easier to initiate group action. About 50% of the groups in this study were formed after mass
mobilization by government agents and non-governmental development agencies. To overcome
initial inertia problems, some groups start with few individuals who have common interests.
Then they proceed to accept new members in order to achieve economies of scale, but only after
establishing some foundations for success. This is referred to as ‘concentrated social ties’, which
is the extent to which social ties are concentrated in a few individuals rather than being spread
more evenly across the whole group (Marwell, et al., 1988). Olson argued that success was more
likely in groups with members in a small geographical area, having similar economic and
demographic characteristics, and in frequent contact with each other (Olson, 1965). A strong
social network in the early stages of PMG formation is important because the process is costly.

17

18

Some farmers will play the game of ‘wait-and-see. ’ However, many successful groups
interviewed instituted measures like higher entry fees for late entry and long vetting procedures

17

Joining a PMG is costly in terms of opportunity cost of time spent attending meetings,
traveling costs to meeting venues, membership fees, and regular contributions.
18
We refer this as a free-rider problem. It emerges when property rights cannot be traded, or are
insecure, or unassigned (Cook, 1995). Royer (1999) referred to it as “a type of common property
problem that emerges when property rights are not tradable or are not sufficiently well defined
and enforced to ensure that individuals bear the full cost of their actions or receive the full
benefits they create.” Both internal and external free-rider problems are often associated with
conventional cooperatives.
27

to discourage these types of members joining at a later date. This was a largely successful
strategy.
When a group shows signs of success, more producers are typically willing to join. With
higher volumes, a group is able to exploit economies of scale, lowering it’s per unit marketing
and transaction costs. Simultaneously, these same costs may increase for the non-members,
inducing a cascading effect leading to more producers joining the effort. There are even
instances where some farmers who join due to the cascading effect might have a higher cost of
marketing with the group than they faced in the initial period as non-group members. However,
they cannot revert to that initial cost structure because the mere presence of a collective
marketing effort raises the cost of non-membership. In such case, producers who refuse to join
the farmers group could be hurt by the PMGs if their cost of independence increases (Leathers,
2006).
A study in Québec showed that 60% co-operatives survive more than five years and 40%
survive more than 10 years (Bond, et al., 2000). Banaszak (2007) conducted a very similar study
on cooperatives in Hungary applying transaction costs and game theoretical approaches. In his
study area, 20% of the cooperatives had failed, and of the surviving ones, about 80% were
performing their core function of produce marketing. In a recent study of a sample of 54
registered cooperatives in Limpopo province in South Africa, Van der Walt (2005) found that
65% of these were not operational. Reasons provided include poor management, lack of training,
conflict among members, lack of funds, and operations never started after registration (Ortmann
and King, 2007). This suggests that for PMGs, the first few years are likely to be marked by
uncertainty and struggle for survival since they are a business concern amidst other functions

28

they perform. The majority of the PMGs in our study areas had existed for fewer than 5 years at
the time of the study. About 67% of the groups interviewed were not more than 6 years of age.
Another constraint is the lack of managerial capabilities among the potential members.
Due to lack of formal job opportunities in the rural areas, many young educated members of the
communities migrate to the cities to look for better opportunities. The only migration to the rural
areas of educated members of the society is through retirements and those in the education sector
who have to work in the rural areas. This robs the rural communities of dynamic leadership that
could steer farmer groups to success. Therefore, the few educated people in the communities are
often involved with multiple group leadership, leading to conflicts of interests, spreading their
time too thinly between these responsibilities and their own personal commitments.
Based on the foregoing literature reviewed, we came up with questions on location,
membership characteristics, group history, operations, members’ perceptions, and organizational
aspect of the PMGs. We concentrated on these because such a focus is consistent with previous
studies on collective action (Garforth, 1994, Gugerty and Kremer, 2004, Howes, 1997, Stiles,
2002). Researchers such as Elinor Ostrom (1990) and Olson (1982) have looked at more general
conditions for the emergence and sustainability of collective action by farmers. We used these
factors as questions posed in the focus group meetings.
2.2.2 Theoretical Framework
2.2.2.1 Transaction cost economics theory

Transaction cost economics (TCE) theory seeks to explain the nature of governance that
trading partners choose from a set of possible institutional alternatives. The theory posits that
efficient organization necessitates matching transactions that require higher levels of
29

coordination with organizational governance forms that provide the necessary levels of
coordination in a cost-effective manner. For a more detailed treatment of this topic, see Boerner
and Marcher (2003) and Joskow (1985), who have done extensive reviews of TCE theory and its
applications. The choice of the governance form depends on the nature of the transaction in
terms of degree of transaction-specific assets involved, level of environmental and behavioral
uncertainty, complexity of the transaction arrangement and frequency of the exchange
(Williamson, 1985).
Cooperatives were formed in the Kenyan highlands as a way to safeguard farmers’
specific assets. The perennial bushes producing cash crops like coffee or tea, once established,
are very specific to the transaction, with limited alternative use. In contrast, semi-arid areas grow
mainly annual crops, which could easily be stored as the market required, and the cost of shifting
to another enterprise is low since the crops are annuals. However, the extent of that flexibility is
constrained by the climate in these areas and the capital constraint of individual producers. In
addition, limited capacity or scale of operations of most traders hinders their ability to handle
risks due to price fluctuations and ability to negotiate a contract. Therefore, at harvest time, cashconstrained traders are not able to buy large quantities of produce to reduce supply significantly,
leading to product glut in the rural markets. Farmers usually sell most of their produce to meet
their family financial requirements since they are usually cash-constrained. As long as cashconstrained farmers rely on rain-fed agriculture, they have little flexibility in terms of timing of
when crops are ready for the market to limit oversupply. Without strong collective action, it is
not possible for smallholder producers to regulate a commodity supply in the market to influence
prices through supply control.

30

From TCE theory, we postulate that asset specificity and spatial location of farmers may
favor some other form of governance other than a spot market. Farmers may form PMGs to act
as a competitive yardstick in the market with few buyers (Staatz, 1984). Based on transaction
cost theory, we conclude that PMGs will emerge in the presence of high transaction costs.
However, for farmers to vertically integrate through a PMG, the overall gross benefit (both
monetary and social) they derive from collective action must be higher or at least equal to
benefits they would derive in the alternative marketing channel in the long run. We can express
this as:

VB pmg,i TC pm,i  MC pmg,i  C pmg  VB sm,i  TC sm,i  MC sm,i

(2.1)

Where: VBpmg,i is the total value of gross benefits a farmer i gets from joining a PMG; TCpmg,i
is the transaction costs farmer i faces while operating within a PMG; MCpmg,i is the total
marketing costs the farmer i incurs when marketing through a PMG (with the exception of
transaction costs); Cpmg,i are all the other contributions (both opportunity cost of time and
monetary contributions) to the PMGs; Spmg is any subsidy the group gets from other sources;
VBsm,i is total value of gross benefits farmer i gets from participating in the alternative marketing
channel (for example, the spot market); TCsm,i is the transaction costs an agent i faces in the spot
market; MCsm,i is the total marketing cost the farmer i faces in the spot marketing channel
(except for transaction costs).

The first scenario is where the inequality is true even when there is no subsidy (Spmg=0).
31

This is akin to what is referred to as the Haystack Model that has been used to study evolutionary
dynamics. In our case, this would be a situation in which farmers recognize cooperators and are
ready to partner with them and all prosper (Bergstrom, 2002, Frank, 1988, Maynard Smith, 1964,
Schmid, 2004). In this case, cooperating will produce more benefits for the cooperators when
working together in the long run than would non-cooperation. There is higher probability of
cooperators working together in this model if matching is assortative

19

rather than random. With

cooperators having a higher probability of playing with another cooperator, there is therefore
increasing prospect of each cooperator reaping a higher payoff.

The second scenario is where inequality 2.1 is only true in the long run and not in the
short run due to the initial cost of organizing a PMG and the lack of enough volume moving
through the PMG without external subsidy. In the short-run, some PMGs will emerge and
survive only in the presence of an external agent’s promotion or selfless individuals in the
community subsidizing (S pmg  0) the initial costs of collective effort. In the field, it was
common to see this situation in some organizations. These groups had at least one individual
who seemed to take lots of interest in the success of the group and spent his or her resources on
collective activities. In such instances, groups might emerge and succeed through such influence
even if they face a reversed benefits regime without the subsidy.
The third scenario is where there is a differential in members’ benefit stream flow. Game
theory analysis would predict that producer groups often would not emerge, since free riders’
choices would lead to a low-level Nash Equilibrium of non-cooperation as the dominant strategy,

19

Sewall Wright (1921) defined the assortativeness of mating with respect to a trait as “the
coefficient of correlation m between two mates with respect to their possession of the trait.”
32

20

referred to as social trap (Platt, 1973).

However, in the presence of at least one altruistic

21

individual in the group who will accept less short-run benefit from collective action compared to
their proportional contribution, groups might still be formed and survive. These altruistic
members of the society forego opportunities for gain in the short-run or long- run out of sense of
moral obligation, care for others or even a desire to be approved or liked by other members of
the society (Schmid, 2004). They must contribute a significant amount of resources to make the
pay-off of other members in equation 2.1 positive. Therefore, different individuals face different
benefit flows. In this third scenario, farmer i in the group subsidizes the PMG operations to the
point of significantly decreasing her own returns for the benefit of the success of the group. In
the short run, this agent might get a reversed inequality in expression 2.1 because to him or her
Spmg=0. The monetary net returns might be reversed but the net returns in terms of utility will be
positive. We refer to this as an asymmetric benefit model in monetary terms. In this case, the
farmers with small interests will exploit the large who can afford to contribute and has much to
lose if the group is not formed or fails (Olson, 1965). These types of group might succeed in the
short-run if the altruistic individual has long-term goals of positive monetary returns. Therefore,
this individual is not fully altruistic in the sense of the definition provided earlier. Hence she will
endure short-term monetary loses in the hope that the group will succeed and she will get

20

Social traps are negative situations where people, organizations, or societies get caught in a
direction or relationship that later proves to be unpleasant and they see no easy way to back out
of or avoid.
21
Altruism is selfless concern for the welfare of others. It is a traditional virtue in many
cultures, and central to many religious traditions. It is the opposite of selfishness. Altruism is
different from a feeling of loyalty and duty. An altruistic individual focuses on a motivation to
help others or a want to do good without reward, while duty focuses on a moral obligation
towards a specific individual a specific organization or community, or an abstract concept (for
example, patriotism). Some individuals may feel both altruism and duty, while others may not.
Pure altruism is giving without regard to reward or the benefits of recognition.
33

eventual positive returns in the long run. However, there is another category of truly altruistic
individuals who gain utility just by helping others, even at a material cost to themselves. In this
situation, the individuals will continue supporting the group in the long run even if their
monetary benefits are not positive.

2.2.2.2 Moral hazard and adverse selection

Producer organizations usually elect an executive committee on a pro bono basis that
directs the organization’s operations. This is because they have little operational capital to
employ paid staff. It is very difficult to attract and retain good managers providing voluntary
services since their opportunity cost is high, and some managers have an incentive to convert
their PMGs into personal firms. For some of the groups interviewed, there were few options in
terms of human capital to elect as management. Due to low level of education, they had to elect
the only educated members, even when those individuals had low management skills. We
postulate that lack of management skills within the rural farming society coupled with moral
hazard makes it very difficult for PMGs to emerge and succeed. In cases where there is
significant remuneration for managers, abundant supply of human resources and repeated games,
we can design a contract that can take care of the problem. However, in PMGs, these conditions
are rarely present, and it becomes very difficult to solve leadership problems. Also due to the
lack of capital market discipline, a clear profit motive, and the transitive nature of ownership,
PMGs may have a greater difficulty of designing incentive schemes for managers that will align
their personal objectives with those of the farmer group (Ortmann and King, 2007). For example,
the survey revealed cases of leaders inappropriately using group resources. In a pilot program on
the use of ICT technology for marketing services, some groups had cell phones. In two of the

34

groups provided with the free phones, the phones turned out to be a major problem, as the groups
could not agree on who had the right to operate and keep the phones in their possession. This
issue, among other problems, led to the demise of one of the groups interviewed.
However, a farmers’ group formation is a dynamic process that takes place within a
multiplayer repeated games framework. If groups could restrict entry and players’ types were
observable, cooperators would not admit defectors to their groups, the two types would be
strictly segregated, and only cooperators would be allowed to join an organization. But in a reallife situation, type detection is imperfect. After group formation, subsequent failures will be
observed, usually resulting from the non-cooperators’ free riding. In the field, the dynamic group
formation phenomenon was observed. We refer this to dynamic formation of groups as
“reinforcement learning.” Individuals initially meet at random and play a game. The payoffs
determine which interactions are reinforced, and a social network emerges. In case of failure
(extinction of a group), different farmers will have revealed their dominant type. Out of this
learning, new groups are formed by fission of existing groups. The new groups have better
coordination of strategies and are composed mainly of cooperators, which can support
cooperative outcomes.

2.2.2.3 Social capital
Putnam (1993) defines social capital as those “features of social organization, such as
trust, norms, and networks, that can improve the efficiency of society by facilitating coordinated
actions.” From this definition social capital has two components: it is, first, a resource that is
connected with group membership and social networks. Hence, the volume of social capital
possessed by a given agent depends on the size of the network of connections that she can
35

effectively mobilize (Bourdieu, 1986). Bourdieu states that it is a quality produced by the totality
of the relationships between actors, rather than merely a common "quality" of the group.
Membership in groups and involvement in the social networks developing within these and in the
social relations arising from the membership can be utilized in efforts to improve the social
position of the actors in a variety of different fields. It is the differences in the control of social
capital that may explain why the same amount of economic and cultural capital can yield
different degrees of returns and different powers of influence to different actors (Siisiäinen,
2000). Other studies have shown that group memberships creating social capital have a
"multiplier effect" on the influence of other forms of capital such as capital goods, land or even
human resources that affect the quantity or quality of production (Bourdieu, 1986, Cohen, 1999,
Joppke, 1986).

Development of economic organizations does not occur in a social vacuum. The local
social institutions, farmers’ personal relations, and the structure of the network of relations affect
PMGs’ economic activities and their probability of success. Social capital literature states that
certain features of a social group, such as networks and norms and trust, can improve the
society’s efficiency by facilitating the coordination (Putnam, 1993, Putnam, 2001). According to
Granovetter (1985), the economy is socially embedded, and the farmers’ personal relations, the
structure of the social networks and local institutions will affect group economic activities. The
success of the groups will also depend on their capability to link with formal institutions. This
type of social capital could have a net effect of reducing transaction costs (Woolcock, 1998).
Interaction among economizing agents creates trust, and trust facilitates better future interaction.
In dynamic repeated games, where framers’ actions can be observed and remembered by others,
almost any pattern of individual behavior, including behavior that maximizes group payoff, can
36

be sustained by social norms that include obligations to punish norm violations by others.
However, we note that while reciprocity and the presence of norms can support a great deal of
cooperation, much of the individual farmer’s activity motivation is impossible for others to
observe and, hence, lies beyond the reach of punishment or reward (Bergstrom, 2002). Hence,
even the groups with greatest number of cooperators will still have issues as they move on the
path to success due to imperfect assortativeness.

2.3 Why PMGs did not emerge in semi-arid areas in Kenya
From the above arguments discussed under the literature review on transaction costs
theory, moral hazard, adverse selection, and social networks, we conclude that it was very
difficult for collective action in the form of a PMG to emerge and succeed in semi-arid areas in
Kenya on its own. This is due to the fact that the main goods produced (grains and legumes)
were not highly specific or perishable and that the villages exhibited low social capital, lack of
appropriate human and capital resources, and the presence of moral hazard. In spite of the
difficulties identified that discourage PMG groups from emerging, some farmers in the study
areas were able to initiate some farmers’ groups without external agent support. In the results
section, the study highlights a few archetypical groups, focusing on their history of formation
and their prospects of success for both self- and externally initiated farmers’ organizations.

2.4 Methods
2.4.1 Group performance
Some researchers have evaluated performance of cooperatives based on financial ratios
(Gentzoglanis, 1997, Hind, 1994, Lerman and Parliament, 1990). For example, based on
financial data for cooperatives operating in 36 states between 1994 and 2003, Boyd et al.
determined variables that are determinants of profitability in local farm supply and grain
37

handling cooperatives. They tested for statistical significance of variables affected by director
and manager decisions including liquidity, asset size, risk, the ratio of assets to equity, net profit
margin, asset turnover, the times interest earned ratio, total assets, and lagged average return on
equity. They found that business size, as measured by assets, was not a determinant of
profitability (Boyd, et al., 2007). This study is very similar to a recent one conducted on North
Dakota farm supply and grain handling cooperatives between 2002 and 2006 (McKee, 2008).
Financial statements from 120 cooperatives were used in this study. Various financial variables
were tested as determinants of profitability. Financial ratio analysis was used to observe trends in
liquidity, solvency, and efficiency. Comparisons in ratio trends were made based on relative
profitability. No statistical relationship was found between business size and profitability. The
approach of using of financial ratios to evaluate the performance of cooperatives is not consistent
with cooperatives’ operation model (Sexton and Iskow, 1993). Since cooperatives represent a
vertical integration between the farmer and the cooperative, using financial ratios of only part of
an entity fails to account for all of the financial effects of management decisions on the joint
entity. We also argue that accounting performance, i.e., returns on investment (ROI) or other
measures, is hard to measure and interpret in the case of cooperatives or PMGs. This is because
their aim is to pay the members the highest price for their produce and charge the lowest price
for their services and goods, while still remaining financially viable. Even when possible to use
financial measures, it might be only applicable to sophisticated cooperatives and not PMGs in
developing countries.
Other authors follow some overall evaluation process for cooperatives as proposed by
Deshpande et al. (1993) based on marketing and management science that focuses on customer
satisfaction. In our study, most groups did not have good financial records that could be used for
38

analysis. Also, it was difficult to collect data on customer satisfaction due to time and financial
constraints. Therefore, this study cannot adopt this approach. Many had also operated for a short
time and were only starting to put into place proper record keeping measures.
Bruynis et al. (1997) define success in terms of longevity, business growth, profitability,
and members’ satisfaction. Evaluating group performance may require knowledge of members’
well-being before the group was formed and their well-being after joining the group. That type of
analysis is beyond the scope of this chapter, where our unit of observation is the group.
However, the measure of success defined below takes into account the purpose of group
formation and if those objectives have been accomplished (Berne, 1963, Zander, 1994) . This is
in line with other recent studies in group performance (Banaszak, 2007, Sexton and Iskow,
1988).
We define success of any group as dependent on the objectives of the group and its selfassessment. Some groups interviewed were not set up as a marketing concern but have added this
aspect as part of a diversification of their activities. Hence using volume of sales as a success
benchmark alone may not be appropriate for these groups. However, we use the volume bulked
as one of the measures of success. Extending the work done by Shiferaw (2008) on a subset of
the groups interviewed in this study, we use five indicators of collective action to measure group
success. These included:


The number of elections conducted since formation divided by the number of years
the group has been in existence. This measure is justified by the fact that bylaws
require elections to be held annually.



Per capita cash contributions to the group in 2006
39



Per capita group assets owned



Per capita value of agricultural produce bulked, and



The percentage of the members who adhere to the group’s by-laws

By using per capita measures, we take into account the size of a group to avoid a bias that might
tilt the index to favor bigger groups in terms of membership. To aggregate these measures into
an overall index between zero and one, we converted the per capita values to an index

22

that

gives the highest value to 1. We then summed up all the indexes and divided the sum by 5 to
standardize it to have a value between zero and one. A score of zero is defined as complete
failure and one is defined as complete success. Formally:
5

  contrindex  bulkindex  assetindex  memcomitindex  electindex 

success  i 1

5

(2.2)
The mean value of the success index for the groups studied is 0.54, with a standard
deviation of 0.15 (see figure2-1 below). We classified further the groups into four quartiles for
better disaggregation between the more successful groups and least successful. This
classification is used in the cross-tabulation part of analysis and in testing the group analysis
framework.

22

The success index components are equally weighted since there is no way to justify one
component as having more weight than the others have in measuring the success of a group.
40

Mean= 0.54
Std Dev=0.15
N=90

Figure 2-1: Group success histogram
However, for the regression analysis we will use all the variations computed in the
success index. The success index (y) is a fractional response, that is, 0 ≤ y ≤ 1, that is
fundamentally continuous. We start by fitting a fractional model:

Yi=f(Hi, Ai, Si, Fj, TCi, Mi, Lj)

(2.3)
41

Where:


Hi is the level of human resources available in a particular group. We use the mean
education of the focus group that was interviewed. The rationale is that the group usually
sent the most educated of their members to a meeting, particularly if it involves meeting
with potential donors;



Ai is a vector of the size of a group and spatial dispersion. The number of villages the
group members come from is used as an indication of spatial dispersion, and a variable
that measures the number of members registered is also included in the vector;



Si is a vector that measures the level of group diversification in terms of activities it
undertakes. An actual count of the group’s activities is used. Then a dummy variable that
indicates if the group buys produce from non-members for resale later is also included in
this vector;



Fj is a variable that captures the way the group was initiated--that is, if it was self- or
externally initiated, with external initiation taking the value of 1;



TCi is a vector of transaction costs a group faces. Two variables are used to estimate TC.
These are the group categorization according to the nature of the good, as explained later,
and distance to the village market;



Mi is a vector of group internal management. This is measured by two dummies: one
dummy that captures whether the group has measurable performance targets, one that
indicates whether the group has agreed-upon conventions, and the third that captures the

42

management quality of the leadership by having self-assessed good management being
one.


Lj is a location dummy. We use the district where the group is found as a dummy. This
variable captures the unobserved site-specific factors not controlled for in the model.
To estimate the success model, we use General Linear Model (GLM) to estimate a

fractional logit regression. The latter model estimates E(y|x) as a logistic function (Wooldridge,
2002):
E Y | X   exp  X   1  exp  X  



(2.4)

The model ensures that predicted values of success (y) are in (0,1) and that the effect of any xj on
E  y X  diminishes as X    .

2.4.2 Case study
The study also applies a case study approach to further analyze the 90 groups. By
combining multiple data sources, we can help overcome the weakness or intrinsic biases and
problems that come from single data source (Yin, 1994). The use of multiple sources of evidence
in this study allows us to address a broader range of historical, attitudinal, and behavioral issues
that are not easily modeled in a quantitative econometric analysis. It enables the development of
a converging line of inquiry through triangulation. Thus, the findings or conclusions in the study
are likely to be more convincing and accurate when based on several different approaches.
2.4.3 Producer marketing group’s classification/stratification
According to the nature of the goods they handle, we classified PMGs into two kinds:
food-crop-based PMGs and commercial-crop-based PMGs. The basis for this classification
derives from transaction cost concepts of asset specificity in agriculture arising from the
43

perishable nature of the products and immobility of the farm assets associated with the
transactions (Staatz, 1984). The number of times the farmer and buyer meet to transact business
affects transaction cost. The more frequent the transaction, the higher the transaction cost.
Transactions that are more frequent will encourage greater PMG patronage to safeguard the
transactions. The nature of groups’ governance structures decreases or increases the transaction
cost. The implication is that a group will succeed or fail based on whether the PMG reduces or
increases transaction costs in the long run.
Farmers who are members of food-crop-based PMGs grow and market mainly cereal and
pulses. These crops are easy to store in the absence of storage pests, and farmers sell the produce
slowly as the need arises. Farmers also use the crops they grow as a major portion of their diet.
Therefore, these crops are for food security first, and farmers sell their surplus stock on asneeded basis. In particular, pigeon peas are sometimes the only crops harvested in an area like
Makueni in cases of a drought. Most households will market the produce to meet certain
household cash-flow needs, while the rest is stored. Hence, we can conclude that for these crops,
the transaction costs specific to the transaction in the sale of the crops are low since they are not
very perishable. Then PMGs are not likely to emerge in these areas without external
interventions.
The second category consists of the commercial crop producers. In the study area, we
encountered groups dealing with horticultural crops, ranging from fresh vegetables to tree crops
like mangoes and citrus fruits. Some PMGs organized themselves around these crops.
Horticultural produce has high transaction costs due to the perishable nature of the product.
Common perishable products grown in the study areas are mangoes, citrus, pineapples, and
vegetables. Usually, buyers are few. For instance, mangoes grown in Eastern Kenya are mainly
44

for export, and two traders based in Nairobi dominate the market. They have more information
about the market prices than individual producers do. Traders set the quality parameters based on
their market needs as well as the availability of the products. When mangoes are plentiful, the
traders are very finicky about quality considerations, but in times of shortage, they relax the
quality considerations. All these variable rules and quality requirements can be very bewildering
to a smallholder producer. Once the crop is ready to harvest, an individual farmer has a very
small window of time to find a buyer, negotiate price and pick the fruits. In many of the areas
interviewed, only a single exporter operates. The risk of being stuck with produce with no
alternative market is always a threat to a farmer. Therefore, Individual farmers producing such
crops are more vulnerable to opportunistic action by their trading partners, who may have
substantial market power. In Eastern Kenya, the buyers insist on going to pick from the tree the
produce from an individual farmer and will pick only what they consider to be good quality. The
traders also have the power to prevent the producer from selling the unpicked fruits to an
23

alternative market by threatening not to buy their produce if they deal with another local buyer.
Hence, producers with a high proportion of sunk-assets are vulnerable to unfair trading terms
from powerful trading partners. Patronizing the PMGs may reduce the transaction costs, giving
farmers better control of their market process that increase their perceived value of collective
effort.
We disaggregated the groups further into self-initiated and externally initiated groups.

This was important to test the proposition posited by some researchers that chances of survival of
collective action depends on the mode of initiation.
23

Usually there is only one export buyer operating in an area, and a threat for exclusion from
selling to the trader is credible and can lead to heavy losses. In the study areas, the main crop
under this arrangement is mangoes, usually for export.
45

2.4.4 Design of a group analysis framework
To set the stage for analyzing any group and predicting its success potential or identifying
intervention areas, we developed a group analysis matrix. Guided by the literature reviewed,
focus group interviews, and theory, we came up with a group analysis framework (see figure 2-2
below). Using this framework, producers, policy makers, and development agencies can predict
success potential of a group and pinpoint necessary intervention areas. The matrix is similar to
the decision-making framework for changing vertical coordination strategy, as developed in
Peterson, Wysocki and Harsh (2001) and in Wysocki, Peterson and Harsh (2003), hereafter
referred as to PWH framework.

The first thing potential members do is to analyze their situation and come up with a
decision regarding whether to adopt a form of vertical coordination. The framework analyzes
groups composed of individual farmers who choose to vertically coordinate through PMGs. It is
structured as a series of questions any researcher or a group of individuals intending to start a
group can follow. A basic rule is that when the answer to any question is yes, one proceeds to the
next question.

24

However, when the answer is no, it raises a red flag regarding the viability of

the group, and indicates need for attention. The more red flags are raised, the higher the
likelihood of encountering difficulties and hence a subsequent need for commensurate
intervention to increase the probability of success.
This framework’s assumption is that members have gone through the PWH framework
unknowingly and come up with the decision to shift from spot market to another form. The
rationale for the questions included in the framework and cross-tabulation with success score
24

A yes implies that in the area in question the group is on sound footing and improves the
chances of overall success.
46

will be discussed later in the results section 2.6.2.

Is the good characterized by high asset

No

specificity?

Yes
Is the group self-initiated?

Red Flag

No
Red Flag

Yes

No

Does the group have well articulated by-laws?

Red Flag

Yes
No
Are the by-laws enforceable?

Red Flag

Yes
No
Red Flag

Does the group have well defined objectives?
Yes
No

Does the group have measurable targets?
Yes

Red Flag

No
Red Flag

Does the group have regular
contributions/subscriptions?
Yes

No
Red Flag

Is the group involved with more than one activity?
Yes

Does the group cover a small geographical
area?

No

Yes

Red Flag

No

Does the group have good leadership?

Red Flag

Yes
Higher probability of success

Figure 2-2: Group Success Rate Analysis Framework

47

2.5 Results and discussion

2.5.1 Geographical location of the groups according to their categorization
Before we discuss the results of regression analysis and the case studies, we briefly look
at the geographical distribution of groups as categorized by the nature of good they market.

Table 2-1: Regional location of groups according to their categorization

Quartile group of success†
1
2
3
4
Total
(%) (%) (%) (%) (%)
Location
Western Kenya

Categorization
Food crop
Commercial crop

Eastern Kenya

Food crop group
Commercial crop

40.7
10.5
Total 28.3
38.5
12.9
Total 20.5

25.9
31.6
28.3
15.4
25.8
22.7

14.8
31.6
21.7
38.5
25.8
29.5

18.5
26.3
21.7
7.7
35.5
27.3

100.0
100.0
100.0
100.0
100.0
100.0

Somers’D
(Directional
Asymptotic
significance)

0.327**
(2.120)
0.347**
(2.009)

*Significant at 10%; ** Significant at 5%; ***Significant at 1%
t statistics in the parentheses
† a 1=lowest scores on success index; 4=highest score
In general, irrespective of the region a group is located in, food marketing groups tend to
be clustered in the lower two quartiles of the success index rating (see table 2-1 above). In
Western Kenya, 19% of food crop groups are in the top quartile and 41% in the lowest quartile.
In Eastern Kenya, the food groups have only 8% in the top quartile and 39% in the bottom
quartile. Commercial crop groups are skewed more to the upper quartiles, with 26% and 36% in
the fourth quartile for Western and Eastern Kenya, respectively. The bottom quartile has 11%
and 13% for Western and Eastern Kenya, respectively.

48

2.5.2 Regression results
2.5.2.1 Test of the success index score
We tested the success index score against the focus group self-assessment of their group
success rating. The focus group members were asked if they feel that their group has achieved its
objectives and the answer was either a ‘Yes’ or ‘No’. A ‘No’ was coded as 0 and a ‘Yes’ as 1.
Therefore, we ran a probit regression with group self-assessment being the dependent variable
and the success score being the independent variable. Table 2-2 below gives the results of the
analysis.
Table 2-2: Test of the success score index against self-assessed success rating
Group self-assessment of success is the dependent variable
Success index test (Marginal effects)
success
0.001*
(1.89)
LR chi2(1)
10.48
Prob > chi2
0.0012
Pseudo R2
0.5462
Observations
90
Positive predictive value
98.88%
Negative predictive value 100%
Correctly classified
98.89%
Absolute value of z statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
From the test of the success score index against the self-assessment, we find that the
score correctly identifies a group that the focus group members classified as successful 99% of
the times, and correctly identifies 100% of the times the group classified as not successful.
Therefore, this gives us confidence in moving on and using the score to classify the groups’
success standing in the study.
To test the model, we estimated several quasi-maximum likelihood estimation (QMLE)
fractional logit regressions. The first regression utilizes the composite success score index
49

developed. Then to identify the contribution of each component to the success rating, we ran
disaggregated fractional logit regressions with each of the success score index elements being the
dependent variable but retaining the same independent variables. Later, in section 2.6, we will
present a model focusing on the factors of interest to the group analysis framework alone. In
section 2.5, we use quintile regressions that assist us investigate the tails of the success score to
help us understand how each covariate affects the less successful and more successful groups
separately.
Table 2-3 shows the results fractional logit models estimated. In table 2-4, discussed
later, we present results of cross tabulation and univariate regression of the important factors
applied in the analysis framework. This section presents result of multivariate analysis that
controls for other factors.
In this section, the variable of interest, success score (y), is a proportion, being defined
and observed only on the standard unit interval, i.e., 0 ≤ y ≤ 1. The bounded nature of the indexes
and the possibility of nontrivial probability of mass accumulating at one or both boundaries raise
some interesting estimation and inference issues. Therefore, the standard practice of using linear
models to examine how a set of explanatory variables influence a given proportional or fractional
response variable is not appropriate in general, since it does not guarantee that the predicted
values of the dependent variable are restricted to the unit interval. Papke and Wooldridge (2008)
point out that even in cases where the variable is strictly inside the unit interval using traditional
log-odds transformation, it may not be possible to recover the expected value of the fractional
response from a linear model for the log-odds ratio without making strong independence
assumptions. Also, a linear functional form for the conditional mean might miss potentially
important non-linearities. Therefore, we apply the fractional regression estimation procedure that
50

takes into account the specific characteristics of fractional response variables (Papke and
Wooldridge, 1996, Papke and Wooldridge, 2008). Applying quasi-maximum likelihood
estimation (QMLE) we are able to obtain robust estimators of the conditional mean parameters.
This method has been applied in several studies after its application in the seminal paper by
Papke and Wooldridge (1996). Hausman and Leonard (1997) applied fractional logit to panel
data on television ratings of National Basketball Association games to estimate the effects of
superstars on telecast ratings. Wagner (2003) analyzes a large panel data set of firms to explain
the export-sales ratio as a function of firm size. And the recent paper by Papke and Wooldridge
(2008) extends their earlier work and shows how to specify and estimate fractional response
models for panel data of fourth grade math pass rates with a large cross-sectional dimension and
relatively few time periods.
The results below (Table 2-3) are for fractional logits, and we present only the marginal
effects. However, we only discuss the significant variables in the estimated models. The number
of villages the group serves has a negative sign and is significant at the 1% level in the success
model. As posited earlier, operating in a large geographical area decreases the group’s success
index. It is difficult for members to be committed to a collective action ideal unless they know
one another well, which has a negative relation with spatial location of individual membership.
The ratio of men to women results show that as men increase compared to women
members in a group, there is decrease in their ability to follow their election calendar, contribute
less to the group finances and there is less commitment to the group bylaws by the membership.
However, the amount bulking in the group per capita increases. Men control majority of the
productive access and are usually commercially oriented. Therefore, it is expected that the
bulking amount per member increase as the ratio of men to women rise.
51

Table 2-3: Factors determining group success
(1)
(2)
(3)
(4)
Success
Commitment Election Contribution
Education
0.006
0.004
-0.001
0.001
(0.89)
(0.30)
(0.06)
(0.09)
Men to women -0.002
-0.033***
-0.021** -0.024***
(0.77)
(3.00)
(2.03)
(2.98)
# villages
-0.014*** -0.005
-0.021* -0.026*
(2.63)
(0.59)
(1.84)
(1.91)
# activities
0.028*** -0.002
0.030
0.047**
(3.63)
(0.14)
(1.61)
(2.23)
Membership
-0.0004** -0.001
-0.000
-0.001*
(2.05)
(1.23)
(0.75)
(1.65)
Commercial
0.043
0.113*
-0.062
0.093
(1.59)
(1.92)
(0.89)
(1.29)
Conventions
0.035
0.059
0.016
-0.038
(1.07)
(0.87)
(0.20)
(0.51)
Targets
0.012
0.057
-0.037
-0.041
(0.34)
(0.95)
(0.50)
(0.61)
Altruistic
0.066**
0.083
0.047
0.141*
(2.55)
(1.57)
(0.70)
(1.89)
Buy produce
0.093*** -0.057
0.092
-0.005
(2.65)
(0.92)
(1.10)
(0.07)
Distance
0.000
0.001
-0.000
0.000
(0.30)
(1.10)
(0.15)
(0.31)
Leadership
0.031
0.073
-0.036
0.124*
(0.92)
(1.32)
(0.41)
(1.75)
External
0.008
-0.062
0.098
-0.020
(0.32)
(1.15)
(1.41)
(0.31)
Siaya
-0.024
-0.067
0.067
-0.045
(0.45)
(0.69)
(0.59)
(0.43)
Teso
0.008
0.005
0.119
0.118
(0.14)
(0.04)
(0.84)
(0.90)
Mbeere
0.020
-0.160*
0.166*
-0.051
(0.46)
(1.66)
(1.76)
(0.65)
Makueni
0.036
-0.144
0.135
0.049
(0.58)
(1.31)
(1.22)
(0.40)
Observations
90
90
90
90
Absolute value of t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

(5)
Bulking
0.021
(1.52)
0.098***
(2.79)
-0.024**
(2.16)
0.021
(1.26)
-0.000
(0.60)
0.134**
(2.44)
0.036
(0.53)
0.052
(0.83)
-0.035
(0.58)
0.309***
(6.24)
0.000
(0.95)
-0.013
(0.19)
-0.011
(0.19)
-0.312***
(4.20)
-0.334***
(3.90)
-0.191**
(2.54)
0.045
(0.40)
90

(6)
Asset
0.001
(0.09)
0.023
(1.45)
-0.003
(0.26)
0.060***
(3.27)
-0.000
(0.69)
-0.028
(0.50)
0.102
(1.32)
0.031
(0.46)
0.104*
(1.72)
0.143**
(2.19)
-0.001*
(1.76)
0.024
(0.30)
0.038
(0.58)
0.203**
(2.17)
0.130
(1.17)
0.316***
(4.08)
0.110
(1.07)
90

With a large geographical area, effective coordination involves walking long distances to
attend group activities. Therefore, it is likely that more members will have trouble fulfilling their

52

commitments, hence the negative relationship in with success score. The same negative
relationship is shown on election, contribution, and bulking indexes. This is a reflection of
coordination difficulties a group faces if it covers a large spatial area. This is likely to change
with increased use of cellphones.
As the number of group activities increases, the group has a higher chance of success.
The coefficient on model one is positive and significant at the 1% level. More activities add
value to membership because members derive greater benefits from the collective action effort.
With increased value, members will be more committed to the group affairs and work towards
their success. More activities also improve social capital, which decreases need for stricter
monitoring to avoid free riding by non-cooperators. The coefficient is also positive and
significant on it effect to contribution index at the 5% level and asset index at the 1% level. As
the group has more activities, it earns more incomes from deductions, improving the contribution
index. It then utilizes the contributions to accumulate more assets.
There is a negative relationship between success index and group membership size. The
coefficient on membership is negative and significant at the 5% level in the success model. Big
groups might have individuals with diverse needs and interests. Hence, getting members to agree
on common activities and operations requires skilled management that is usually lacking in the
rural areas given that leadership is on a pro bono basis. More members also increase the
opportunity for free riding and make it difficult to use social pressure to discourage such
behavior. This result was as expected. The finding might be contrary to the theories that posit
25

that collective action needs more members to increase economies of scale . However, the
25

We tested for non-linearity of the variables number of villages, members, and activities and
there was no evidence of nonlinearity.
53

results are consistent with Olson’s (1965) an analysis of collective action and the group size
26

paradox

(Esteban and Ray, 2001).The above two results are an indication of the high

transaction costs in getting collective action to be effective with a large membership and spatial
location in terms of service. Similar results are observed in management of natural resources
through collective action (Antony and Steven, 2005, Shiferaw, et al., 2008). Large membership
is also negatively associated with contribution. The coefficient is negative and significant at the
10% level. This could be an evidence of the free-rider problem. With a large membership
individuals may feel that their contributions to the group have negligible effect. They will
therefore be more likely to default on fees and charges, which leads to lower the contribution
index.
Commercial group members, particularly the horticultural crop growers, will be more
committed to their group and patronize them due to asset specificity. There is positive
relationship between commercial groups and higher bulking and commitment indexes at the 5%
and 10% levels respectively.
As posited earlier in the theoretical framework, the presence of an altruistic leader
increases the success index by 0.07 and is significant at 5% level. An altruistic individual takes
collective action seriously and is able to rally commitment of members and both internal and
external resources to contribute to group success. An altruistic leader also attracts other likeminded people and rallies their efforts to the group enterprise. By demonstrating the sincerity of
one's good intentions, an altruistic leader might encourage others to engage in reciprocal altruism

26

Olson (1965) and earlier work by Pareto (1927) argue that the free-rider problem makes small
groups more effective. This is because with larger groups, the smaller is the perceived effect of
an individual defection and the smaller is the size of individual private prize.
54

(Smith and Bliege Bird, 2000, Sosis, 2000). Hence, the altruistic leader is powerful or skilled
enough to sanction free riding, solving a big concern in collective action effort. In return, the
society accords such leaders some amount of respect as a social tax, increasing their utility
(Hawkes, et al., 2001, Henrich and Gil-White, 2001, Michael, 2003). The presence of altruistic
leader increases the contribution and asset index. The results suggest that these types of leaders
are able to galvanize the support of other members of the society to patronize their groups and
build an asset base. That could be the how they influence groups’ success.
Groups that buy products from non-members increase their success rate. The coefficient
is positive and significant at the 5% level. The variable captures the commercial orientation of a
group. Such groups tend to have well thought-out marketing strategies, as trading involves
incurring costs before selling the product for a profit. In these groups, one tends to find a good
marketing representative with a good social network to gather price information from the market
and contact potential buyers. As expected, a group that buys produce increases the bulking index
as well as the asset index.
Groups that are found in remote areas, as captured by the distance-to-market variable,
have fewer assets, decreasing the asset index score. The coefficient of distance to market is
negative and significant at the 10% level in the assets model. The implication is that these groups
are generally poor and have fewer resources. Good leadership also has a positive relationship
with contribution, but the coefficient is not significant in the composite success score index.
Many of the variables expected to be significant on the composite success score were not.
However, using the method of further disaggregating the score to individual indexes, we are able
to get more insight into the interaction between the overall index and the individual items. From

55

the disaggregated models, we can see that the contribution, bulking, and asset indexes have a
greater influence on the score direction compared to the commitment and election indexes.
2.5.3 Case study I (Commercial groups)
This section gives a brief history and characteristics of some selected cases out of the 90
groups interviewed. We explore first the archetypes of commercial groups in the sample.

Group A in Siaya district has a success score of 0.74 and is in the top quartile of the
success index. It was formed as a religious group with the main aim of improving the welfare of
its members. With time, it has branched out into growing commercial groundnuts and
amaranthus for seed and grain. This is as a response to the felt need to serve the women and
orphans affected by HIV-AIDS. The group has acquired assets like party tents, seats and other
equipment to hire out to the local community. A unique aspect of this group is that it also
operates as a charity. Each year, it chooses a needy family in the area that is not a member of the
group and builds a new house for them. This is sort of a local version of the American TV show
“House Makeover”. The group has good leadership, composed of retired civil servants with wide
outside exposure. It has even branched out into processing of Amaranthus (pigweed) grain into
flour, marketed as a nutritious drink for HIV-AIDS victims as an immune system booster,

27

as

well as making Uji (porridge) and several forms of local bread. The key to this group’s success is
the realization that there is need to add value to ordinary food crops through processing. Their
social bond is also very strong due to their religious affinity and the many activities they perform
as a group. To increase their social capital, this group has a monthly lunch in a member’s home.
27

The flour is a very good source of vitamins, including vitamin A, vitamin B6, vitamin C,
riboflavin, and folate, and minerals including calcium, iron, magnesium, phosphorus, potassium,
zinc, copper, and manganese (http://en.wikipedia.org/wiki/Amaranthus).
56

Such a strong social network decreases the need for higher levels of monitoring to ensure
cooperation from members, avoiding cases of free riding or shirking. The other key ingredients
are self-initiation, visionary leadership by an altruistic leader, good management, proper record
keeping, sense of social duty, and diversification.
Another group in Siaya in the Yala area that is worth mentioning started as a women’s
group. Let us call it group B; it has an index score of 0.92. Over time, the group accepted more
membership to include men. The group started in response to a typhoid problem in the area.
Women felt a collective need to clean up their water sources. They started a program of water
catchment conservation. Once the benefit of working together was realized, the success attracted
external help and encouraged the women to branch into other activities. With the help of their
coordinator, they started multiplying an open pollinated variety of maize for the local market that
has superior traits derived from varieties released from the Kenya Agricultural Research Institute
(KARI) and CIMMYT.

28

Individual members also grow potatoes, for which the group

coordinates in the sourcing of inputs and the marketing of the products. They also grow other
horticultural produce for marketing in nearby Kisumu city. The ability of the group to coordinate
small farmers has attracted attention from the private-sector firms that are forming joint ventures
in seed production. A case in point is Leldet Ltd., a local seed company.

29

Leldet contracted with

group B in 2007 during the long rains season to produce groundnut seed. The formal groundnut
seed system in Kenya is in its infancy and for a private firm to have a contract with a group
28

Kakamega Synthetic-I was released by the KARI research station in Kakamega, Kenya. Its
pedigree traces back to the work of CIMMYT and many partners in southern and eastern
Africa—national maize research programs, private companies, and non-government
organizations—to develop stress-tolerant maize for the region’s small farmers.
29
Leldet Ltd. is a small seed company that produces, promotes, and distributes improved seed of
orphan crops for use by small-scale farmers in arid and semi-arid areas of Kenya.
57

composed mainly of smallholders is a significant step. The group started a food-catering branch
that caters for local social events and provides all services, from food preparation and services to
provision of tents and chairs. This is a group that is highly diversified in terms of its activities.
Group B’s sources of success include self-initiation due to a felt need, strong social capital,
diversification into diverse commercial activities, visionary leadership with an altruistic leader,
and the ability to network with various private, non-profit and government agencies. This high
level of investment in the group increases the sense of ownership by its members. It is unlikely
that individual members will chose to free-ride and risk being excluded from the benefits
accruing from being a member. So the threat of being sanctioned by the group is credible and the
loss incurred would be non-trivial. This takes care of the free-rider problem often witnessed in
collective action efforts.

Moving from Western to Eastern Kenya, the study highlights a few groups in the same
category. Commercial groups in Eastern Kenya are involved in mainly horticultural crop
production, tree nursery establishments, and commercialization of the local cereals and pulses.
The main horticultural crops grown were export-quality mangoes, citrus fruits, and vegetables
(cabbage, tomato) and in cereals, the main crop is maize, with pigeon peas dominating the
pulses.

Group C horticultural group is a case in point, with a success score of 0.65. A retired
navy officer initiated it with the purpose of growing and marketing export mangoes and citrus
fruits for the urban market. The group has a target every year of the number of citrus and
mangoes trees each member should have on his or her farm. For continuation of membership, an
annual audit is conducted to ensure that each member meets his or her targeted seedlings planted.

58

At the time of the survey, the group’s main crop was mangoes geared for the export market. The
group identifies a buyer and negotiates the terms of trade. The buyer picks the fruits from the
members’ farms using his own labor and prepares them for export. Group C is exploring avenues
of providing the extra services of harvesting, packing, and transport to the airport/market. Such a
strategy is likely to increase value to the commodity and encourage other competitors who do not
have enough staff to pick the fruits to do business with them. More buyers competing for the
same product might bid up the price, increasing producers’ returns. The group is also
investigating means of setting up a joint disease and pest management system to reduce costs and
guarantee produce quality. In the export business, guaranteed quality is the key to success, and a
group with a higher quality is more likely to get clients compared to other groups with average
quality produce. The group displays a high level of professionalism in running of its affairs. It
has a permanent staff employed to keep account books and maintain proper records. A hired
consultant had finished conducting a strategic management analysis of the group. Following the
analyst’s recommendations, the group has started a savings and loan department. This is another
way of increasing membership value. Savings and loan services are intended to tap into local
saving capacity of the community as well as get financial grants from the government that are
used to encourage rural development. If this scheme is successful, it will relax the credit
constraint within the membership and improve members’ welfare. The overall management of
the group is very good. The sources of success are visionary leadership, professional
management, enforcement of by-laws, diversification, and presence of a selfless leader.

Another group that has very similar characteristics is Group D (Wote location, Makueni
district) with a success score of 0.79. It was initiated by a local school head. He had been
involved with a similar self-development group in another region. Group D grows export-quality
59

mangoes, citrus fruits, and pigeon peas (seed and grain) for the local market and dairy farming.
Its stated goal is to alleviate poverty in the area. Therefore, diversification of activities is a
strategy to improve incomes for its membership. The group is very innovative. To guarantee
quality mangoes to its export clients, it started a collectively managed disease and pest
management system. This ensures a nearly homogeneous product to the customer. The group
enforces strict collective marketing of all its produce. The group also had been contracted by
ICRISAT as one of the seed propagators to be distributed to other groups in the region. It also
has a dairy animal buying scheme that helps members’ entry into the dairy sector for income
diversification. It has a regular monthly contribution requirement for membership continuation.
The group employs a closed membership policy and admits members from the siblings of the
current members. This in effect preserves the already established culture, and family social
pressure reduces incidences of non-cooperation. Sources of their success are self-initiation,
closed membership policy, and well developed by-laws and culture that are strictly enforced,
diversified enterprises, visionary leadership, value addition, and presence of a selfless leader.

Group E has a success score of 0.85. It was self-initiated long before the LLP project. In
the initial stages, it bulked pigeon peas, maize, and beans. The group is located in Makueni, an
area with unreliable rainfall, and wanted to have a bulking store as well as a seed bank. The
current marketing representative attended a women’s development conference and felt the need
to practice what she learnt at home, and the group was born. Its main aim was to bulk the
produce after the harvest, sell the product during the low supply season,

30

and sell seeds to the

local farmers at planting time. The group also started selling locally made baskets from its
30

In many rural areas, food supply is usually low after the beginning of the rainy season before
the next harvest. During this period, prices of most commodities are very high since few farmers
are able to store enough to last them through this period.
60

members. They transport the baskets to the city (Nairobi) for final finishing before they are sold
for export. The group, in partnership with the Kenya Red Cross, stocks and sells mosquito bednets to the local community to prevent the spread of malaria. It has employed a full-time clerk
who runs a store and keeps proper records. With income from the commercial operations, the
group acquired land and put up a building with the aid of a government grant. They intend to use
the building as an agro-veterinary shop. The marketing representative is resourceful in fundraising activities, and she is the driving force of the group. It has diversified into both food and
commercial crops. Its strategy is to have branches of small groups involved with other activities.
These “baby groups” deal with the local poultry business and export mangoes. All these groups
operate from the same office. The strategy has been to have different names and management of
the baby enterprises to avoid same individuals being overwhelmed by management activities.
The sources of success are self-initiation, innovative management style, a diversified portfolio,
and the presence of an altruistic leader.

In contrast to the success of these commercial groups, there were cases of failure. One
such group is PMG F, with a success score of 0.48, initiated by the Catholic Diocese

31

in

Ndhiwa division of Homabay district in Nyanza province. Members of the group are drawn from
three administrative locations, namely Ndhiwa, Nyarongi, and Koriwa. It is organized as a
Community Based Organization (CBO), and its objective is adding value to farmers’ produce.
The group started by growing and processing sweet potatoes. The aim was to process potatoes
into flour used to make various products such as bread, chips, porridge, and nutritious drinks.
Unfortunately, group management did not do any initial feasibility study on the viability of the

31

The diocese runs a development arm of the Catholic Church in conjunction with Catholic
Relief Services (CRS), an international non-governmental organization.
61

investments. They hoped that the diocese would buy all the products for the church-operated
children’s homes. Through grants, they invested in a mill, a drier, and a store. When the church
did not buy the produce, they had nowhere to take their finished product. Unlike a group like
Group E, where the marketing representative actively seeks markets for their products, PMG F
still expected the external benefactor to play this role. The result is that farmers produced huge
quantities of sweet potatoes with no established market. The glut bred animosity within the
leadership and pitted members against the leaders. Even with substantial assets in terms of
equipment and a subsidy from the diocese, the group had almost collapsed by the time of the
survey. It had poor management, with internal leadership wrangles. The large membership of the
group, dispersed over a large geographical area, could also have made communication and unity
of purpose a challenge. Group F showed no innovation or flexibility to change according to
shifting market conditions. This underscores the role of leadership and initial vision of the group
in achieving any success. Sweet potatoes in the processed form are not very perishable, which
overcomes the asset specificity constraint. The group has good objectives but no targets for
implementation. It has no regular financial contributions from members to maintain its
operations. The results are that the group was not sustainable. This group did not understand that
to add value to a product required either processing a product to the point where it had an
established market or to go out and create one. There was a clear lack of altruistic individuals
with a vision to galvanize this group to a common purpose and exploit its potential. However, by
the time of the survey, a small core group of members was planning to dissolve the main group.
In its place, they intended to start another leaner group composed of cooperators.

Also in Ishiara town in Mbeere district, there is a cooperative society we will refer to as
Group H, which has a success index score of 0.46. This is the oldest group that was interviewed
62

in the study and is unique because it was started as a cooperative society in 1980. The aim of the
group was to bulk and market all types of farmers’ products for its members. In the initial ten
years, it was fairly successful because it was supported by the government. This was the period
when cooperatives were deemed to be very important to the development of the farming sector
and enjoyed government support in many ways. The cooperative had acquired some land and
developed a commercial property in the town that was rented out at the time of the survey.
However, in the 1990s there was a change of the Kenyan government policies due to structural
adjustment programs. The support this cooperative enjoyed was withdrawn and the group was on
its own. Despite having considerable assets, the cooperative declined in both membership and its
activities. Out of the original 100 members, there were only 42 members in the register, and few
were active. The only activity the group was involved with at the time of the survey was bulking
honey, refining and selling it at the local market. The group had plans to expand its activities by
setting up a modern commercial honey processing plant at its premises. Members had started
growing pigeon peas and green grams and hoped to use the group for bulking and marketing.
Despite group H’s long history and having considerable amount of assets in comparison
to the new groups in the study, it was lowly rated by the success rate index. Having been started
as a cooperative, it had a good constitution to guide its operations. However, the constitution was
not followed or enforced effectively, which indicated it had poor leadership. In the group’s
twenty-year history, it had conducted only three elections despite the constitution’s requirement
of an annual election. Group H members are not required to have any contribution to the group
finances. Lack of any form of subscription fees coupled with few activities made it difficult for
the group to command loyalty and monitoring by the members. The percentage of its members
that adhered to the bylaws was only 5%. There were no well-articulated goals and targets that
63

could inspire the members. In contrast to many groups in the survey, this was a group that had
well educated leaders, with some having gone to college. However, there was little evidence of
an altruistic leader who could inspire the group or give it the needed change of direction to
improve its success rating.

The poor rating of Group H was mainly due to lack of visionary leadership despite the
leaders being well educated compared to other groups. Group H’s experience also demonstrates
that access to assets does not guarantee a group success in the absence of good leadership and a
vision. Group H has huge potential for growth, but the lack of a visionary altruistic leader made
it not possible to exploit this potential. The group also demonstrates that a group effort supported
by external agents without internal good leadership will not succeed if that support is withdrawn.

From the above sampled groups we can identify some common characteristics of
commercial groups’ success. These success factors are good leadership, presence of an altruistic
leader, diversified activities, innovativeness in value addition, and self-initiation.

2.5.4 Case study II (Food crop based groups)
There were 40 groups in this category in our survey. Their main activities are bulking and
marketing of cereals and pulses. Households grow cereals for home consumption and only
market surplus when they need cash. A characteristic of these products is that they are not very
perishable and can be stored in the farmers’ stores for some time in the absence of a bad outbreak
of a storage pest. This sector usually has many seasonal intermediaries operating at harvest time.
Farmers store the produce and sell small quantities as cash is required in the household. Such a
product does not encourage producers to undertake collective action for marketing. However,
some groups were self-initiated and performed well. This category has less diversity in terms of
64

groups’ activities. Hence, we will take two groups that are the archetypes of success and failure
in this category and discuss them.

Group G (situated in Mbeere district) has a success index score of 0.67. It was formed
after a period of soil conservation efforts in the area, which were organized as a food-for-work
project following a major drought in 2001. CRS, together with the Catholic Diocese of Embu,
32

had instituted a food-for-work project to help the area’s residents affected by a major drought .
The food-for-work project consisted of formation of village-level groups, with members jointly
working on each other’s farms on soil conservation measures. These efforts had a net result of
increasing crop harvests in the region when rainfall was good. To market the surplus grains,
several village-level groups were encouraged to merge into marketing PMGs. Group G falls in
the top quartile of the success index score. It is a typical representation of a successful food crop
group resulting from these mergers. We therefore use it to illustrate the factors of success in this
category. Like the other successful groups, the leadership has been crucial in directing the vision
and growth of the group. Members were able to get some external funding that enabled the group
to start a commercial grain store. Local farmers, both members and non-members, sell their
produce to the group, which bulks and sells it to wholesalers from the city and other bigger
towns in the area. The group has acquired metal silos, through a grant, to store the grain without
damage by pests. The driving force of the group is the ingenuity of the marketing representative,
just like Group E in Makueni. The group has also started working on value adding of its produce

32

It was part of a USAID-funded project named Technology Application for Rural Growth and
Economic Transformation (TARGET), which ran between 2002 and 2004.This project sought to
identify obstacles to legume crop development in semi-arid areas in Kenya. Low production,
poor quality, and price fluctuation due to intermediaries controlling the market were some of the
problems farmers faced. The LLP project was a follow-up project to address some of the
problems identified by the TARGET project.
65

by packing it in shelf-ready quantities supplied to nearby supermarkets

33

. The ingredients for

success in this group are well-articulated objectives, targets, enforceable by-laws, good
management, value addition, considerable capital investment, and the presence of an altruistic
marketing representative.
Common factors that underpin the success of other food groups are good and transparent
leadership, well laid down objectives, well thought out by-laws that are enforced, regular
meetings, diversified activities, and regular subscription fees to the group. Diverse activities in a
group allow membership to derive greater in-kind and monetary benefits from the association. A
common ingredient for success in many groups across the entire sample was regular financial
contributions by the members to the group. Contributions ensure that active members have a
financial stake in the group that encourages interest in the group affairs and helps ensure good
management.
At this point, we describe Group I located in Mbeere district that has a success score of
0.37. It was formed under similar circumstances as Group G described earlier. Group I was
formed after the soil conservation effort of 2001, and by 2007 had virtually collapsed. It is
chosen as a contrast to group G because it is also found in the same geographical area and hence
is affected by similar unobserved factors, yet it did not succeed like Group G. After group
formation, members elected a local prominent resident as their leader. He was chosen because he
was one of the few educated persons in the group. Unfortunately, his attention was divided
between staying in the area and running a business in another town about 15 miles away. Hence,
he was rarely in the area where the group had a bulking store. He did not also work in
33

Four other groups in Mbeere district have started packing self-ready products to market to
supermarkets.
66

collaboration with the other committee members. He controlled every aspect of the group
operations including assets, group books of accounts and records. Although this group had
established by-laws, the leader broke them by not following established procedures of division of
labor. For instance, the group was given a cell phone to access marketing information through
the LLP project. And instead of having the marketing representative perform her function, the
leader kept the phone and did not give any information to the group members, leading to
animosity within the committee. When the group was given inputs by the partner organizations,
the leader would divert them to a local trader who would in turn sell them for their own personal
profit. Group I also served a very large geographical area, and members had no time to bond and
act collectively. It took five years for a section of the membership to get together to enforce their
by-laws and oust the wayward leader through an election. This group had a huge potential for
success since it served a very productive area and could bulk huge amount of grains with proper
leadership.
Group I was formed under the auspices of a local religious agricultural development
organization. Some members on many occasions sought intervention from this organization to
oust the leader, but without a meeting convened by the members at an annual general meeting, it
was difficult to replace the leader. The group did not have any required subscription that would
build solidarity within the membership. There were no other activities the group performed with
exception of bulking grain for sale and receiving of inputs from donors. Eventually it remained
as a group by name only, and all its operations ground to a halt.
However, by the time of the survey, a section of the membership had called an annual
general meeting and had elected a new leader. The members insisted on getting a woman as their
new chairperson. Their argument was that a woman leader was less likely to misappropriate
67

group assets and women compared to men were more negatively affected by group failure.
Therefore, a female leader would be more committed to the group cause.
In retrospect, we find that several factors led to the poor performance of this group. First
was the manner in which it was formed. This group was formed from loosely connected soil
conservation small groups whose members had never worked together and could not easily
identify free riders and troublemakers before they joined. Therefore, there was no cohesion
among the potential members. Second, the group served a large spatial area that made it difficult
for coordination of the group activities among its members. Third was lack of adherence to the
group’s by-laws. Despite obvious misuse of group assets, members could not come together to
organize new elections. This was a classic case of ‘The tragedy of the Commons’(Hardin, 1968).
Fourth, if the group had had a by-law requirement for regular monetary subscription, it would
have ensured self-selection of cooperators who were likely to take action against a wayward
leader. Fifth, the partnering development agents also contributed to this group’s troubles.
Provision of inputs and other assets like cell phones should have been done under a strict
condition that the group adhere to some basic laid down mode of operations like good
governance. Six, there was an obvious lack of an altruistic leader in this group who could have
provided the necessary leadership to solve their problems with their chairperson early enough.
The lack of infrastructure and the remoteness of the group location made it difficult for
development agencies to visit the group to provide necessary group dynamics education to the
members. The combination of all these factors led to the poor performance of Group I, and it
provides a good contrast to a group with the necessary success factors.

68

2.5.5 Case study III (Reconstituted groups)
Five groups out of all the groups interviewed had sprung up from the ashes of the
collapsed groups that were formed through encouragement of external agents. At the beginning
of the Technology Application for Rural Growth and Economic Transformation (TARGET) and
LLP projects, there was the sensitization process about the need to form groups. Through these
projects, many groups were formed, most of which we interviewed. Producers from a wide
geographic region formed these groups. Unfortunately, these producers were frequently strangers
to one another and in some cases did not have a common goal. Hence, when elections were
conducted, a group of individuals was elected from different village constituent groups as
officials, but not for their leadership abilities. Some groups eventually failed but remained active
only as a conduit for receiving project inputs in the form of seeds and chemicals.
Fortunately, some members of the failed groups were able to perceive the benefits of
being in a collective effort and the reasons for the failure of their old groups. Restructured groups
have eventually sprung up on their own. An example of this phenomenon is Group J, with a
success index score of 0.75. The group is located at Kambi ya mawe in Makueni district. It is
categorized as a commercial group because the members are focused on diversified products.
They grow pulses, cereals and horticultural crops. Like some of the groups in the study, this
group was formed after the 2001 TARGET project. Members came together after sensitization
by development agents. However, individual members were drawn from a large geographical
area. When elections were conducted, they allocated positions based on localities to ensure
equity. Therefore, leaders were not chosen on the basis of their leadership capabilities. There was
little to keep this group together other than the need to remain as a group and benefit from inputs
provided. Initially, the group bulked grain and in particular pigeon peas seed under contract by

69

ICRISAT as part of a seed multiplication program. The group had little bonding among the
members, so when the seed multiplication project ended, they could not agree to diversify its
activities to remain relevant. Eventually, the group ground to a halt with no activities. This
continued until a few members came together and decided to start a new group in place of the old
one. The new members are drawn from a smaller geographical area than its predecessor. It has a
core membership of individuals who seem to value the ideal of collective effort. Since they were
members of a previous collective action effort that failed, the “core group” screens thoroughly
any potential new members before accepting them. Due to membership being drawn from a
small geographical area, they are able to identify easily cooperators and free riders and weed out
potential future trouble makers.
As a prototype of these reconstituded groups, Group J gives us a sneak preview of others
with similar characteristics. The groups are composed of farmers who know one another well,
have identified common interests, and are well aware of the undoing of the previous efforts.
These groups tend to have very good leadership, have a vision, well-designed and usually
enforced by-laws, diversified activities, regular contributions and a core group of altruistic
individuals. These reconstituted groups provide an insight into the dynamic process of group
formation after sensitization and group education. They show that the initial group formation
process could have been improperly executed but the education process of pariticipating in
collective action initially given was worthwhile. In these groups, we see that even the failure of a
group after sensitization and the group education is not per se a bad thing. There was a dynamic
process where groups that were formed for the wrong reason or had bad leadership failed.
Nevertheless, since the members appreciated the benefits of collective effort, they reorganized
their groups on their own to have stronger collective action.
70

In this process, we witness in the field a process of a repeated game alluded to earlier in
the theoretical section. Members of a community come together to play a game of collective
marketing. Unfortunately, they do not know each other’s type. Some are cooperators and others
are free-riders. As they work together, some reveal their type, although not completely. As
predicted in theory, the game reaches a low-level Nash equilibrium with the failure of the group.
If the group is dissolved, a subset of the former group play the second-period game. However,
this time cooperators are able to identify each other and form the “core group” that drives the
reconstitution of the next group. It is likely that in the second period, more cooperators will come
together to form the new group and increase the probability of success. There are more likey to
institute a more costly-to-fake signal for commitment to discourage potential free-riders and lead
to self selection of cooperators.
The results in this case study section mirror the work of Ostrom (1990). In her work,
Ostrom emphasized the connection between ‘The Tragedy of the Commons’(Hardin, 1968)
prediction and the Prisoner Dilemma (PD) game in the provision of a common pool resource.
She shows that in the PD game, prisonners are trapped by individual incentives to defect, leading
to failure of cooperation. However, although a PD game structure is useful for empirical analysis
of CA cases and predicts results of many phenomenon in the real life, it is not a reality in all
circumstances. And just as identified in this study, Ostrom showed that it is possible for
individuals to develop effective structures that solve the ‘common set of problems’ that plague
many cooperation situations. The problems are indentified as free-riding, commitment, supply of
new institutions, individual monitoring and effective sanctions. The study finds some common
condtions that overcome these problems, similar to those identified by Ostrom (1990):

1. Group boundaries should be clearly defined.
71

2. Members participate in creating and modifying the rules.
3. Rules governing the use of collective goods should be appropriate to local needs and
conditions.
4. Members particpate in devising their own governance structures, and this should be
respected by external agents.
5. A system for monitoring member's behavior should be developed by the members
themselves.
6. Presence of a system of effective sanctions that is enforced.

In this study, we also show that many CA situations appear to have the structure of a
prisoners’ dilemma at a given point in time, but often represent not a one-shot game but a
repeated game structure where individuals have the opportunity of updating their beliefs and
modifying their strategies. The players are able to learn from the decisions and consequences of
their actions and those of their trading partners, update their beliefs, and evolve structures that
foster subsequent cooperation. This leads to common poor resources being supplied.
2.6 Testing of the group analysis framework
In the methods section, this study proposed a new group analysis framework. This section
presents descriptive analysis of the Group Analysis Framework and results of the regression
estimation of success rating of groups based on quartile categorization of the success index score
as an empirical testing of the framework. The second part of this section employs a case study
approach to test the analysis framework using two selected groups from the sample.

72

2.6.1Descriptive Analysis of the Group Analysis Framework
We briefly go through the Group Analysis Framework by describing the questions
included and by examining a cross-tabulation of the results from the Framework with the success
score rating index.
Table 2-4: Cross-tabulation of group performance and important factors

Quartile of group of success‡ Somers’ D
1
2
3
4

34

Group issue
(%)
(%)
(%)
Categorization (by
Food
40.0 22.5 22.5
crops)
Commercial 12.0 28.0 28.0
Self
25.5 21.3 23.4
Initiation
External
23.3 30.2 27.9
No
20.0 60.0 20.0
Bylaws
Yes
21.0 20.0 22.0
No
29.4 29.4 19.6
Measurable targets
Yes
17.9 20.5 33.3
No
35.0 35.0 25.0
Subscription
Yes
21.4 22.9 25.7
No
70.0 10.0 20.0
More than 3 activities
Yes
18.8 27.5 26.2
Small geographical
No
24.4 34.1 24.4
area
Yes
24.5 18.4 26.5
No
42.1 21.1 28.9
Altruistic leader
Yes
11.5 28.8 23.1
No
30.3 30.3 27.3
Good leadership
Yes
21.1 22.8 24.6
*Significant at 10%; ** Significant at 5%; ***Significant at 1%
t statistics in the parentheses
‡ Quartile 1 is the lowest with Quartile 4 being the highest

34

(%)
15.0
32.0
29.8
18.6
0
22.0
21.6
28.2
5.0
30.0
0
27.5
17.1
30.6
7.9
36.5
12.1
31.6

0.340***
(3.050)
-0.079
(-0.669)
0.268
(1.392)
0.195*
(1.674)
0.324**
(2.533)
0.566***
(2.814)
0.147
(1.263)
0.412***
(3.878)
0.229**
(1.973)

Somers' D is an asymmetric measure of association between two variables. Given predictor
variable X and outcome variable Y, we may estimate D(YX) as a measure of the effect of X on
Y, or we may estimate D(XY) as a performance indicator of X as a predictor of Y.
73

The first questions in figure 2-2 (shown earlier) seek to know the nature of the good the
group deals with. A good characterized by high asset specificity is more amenable to
establishment of a producer marketing group, as discussed earlier. From table 2-4, we find that
the nature of the produce a group is involved in is important to its success. The quartiles numbers
are arranged in ascending order, with four being the top group.
For the food crop groups, only 15% fall in the top quartile, whereas 32% of commercial
groups fall into that group. Of all the food crop groups, 40% fall in the bottom group, compared
to only 12% of commercial groups. Based on Somers’ D measure of association and direction,
these results are significant at the 1% level.

The second question seeks to find out how the groups were initiated. A value of one
means a group is externally initiated, and that is the reason in table 2-3 the association direction
to success is negative. Though the nature of group initiation is not significant in the crosstab,
30% and 19% of self- and externally initiated groups respectively fall into the top quartile.

The next four questions in figure 2-2 explore the competency of the group management
in setting up institutions that can ensure its success. The questions probe the internal planning
and overall operations of the collective effort. The third question asks if the group has wellarticulated by-laws. A cross tabulation of group success against having well-articulated by-laws
reviews show that among the groups with no bylaws, 20% and 60% of the groups are found in
the lowest and second lowest quartiles, respectively, with no group in the top quartile. In
contrast, groups with bylaws are evenly distributed across all the quartile groups. The fourth
question deals with the ability of the group to enforce its bylaws. Unless there is a credible threat
that contravening group bylaws will result in some form of censure, then bylaws will have few
74

desired effects. However, it was not possible to collect this information and hence this question
is not tested in the cross tabulation. The fifth question also deals with the ability of a group to set
well-articulated objectives and goals. Although we asked this question in the survey, we did not
classify explicitly if the objectives and goals were well defined or not. Hence, this question is not
formally tested.

The sixth question seeks to find out if the group has set up well-defined targets that are
measurable. Groups that had measurable targets had a higher success rate, with 28% falling into
the top quartile compared to 22% for the groups with no measurable targets. And 33% of groups
with measurable targets fell in the third quartile, compared to 20% of group with no measurable
targets. The measure of association is significant at the 10% level.

The seventh question in figure 2-2 explores the commitment of the members to group
welfare by looking at the financial subscription and contributions. Subscription is the regular
financial contribution each member is supposed to give to the group. A group with regular
monetary subscription is able to weed out the non-committed members by eliminating the ones
who do not contribute. Also, regular subscriptions will increase the level of investment in the
group effort and the monitoring each member will put into the group’s affairs to make sure that
the group succeeds in safeguarding his or her investment. Contribution is usually a one-off
monetary contribution to the group and is mainly charged as an entry fee. Groups use
contributions as a barrier to entry and as a way to self-select members who are likely to be
cooperators. A contribution also indicates that the members have invested their resources into the
collective effort. The higher the risk of losing a big investment, the higher the commitment level
each member employs. If the group requires frequent member financial contributions, has many

75

activities, and covers a relatively small geographical area, then the increased interactions will
increase bonding within the group, and members will have more time to know each. Once
members get to be comfortable with one another, they will be more likely to agree on issues
pertaining the running of their group, contributing to success. Table 2-4 shows that about 30% of
the groups with subscription and 5% of those with no subscription fall into the top quartile. This
result is significant at the 5% level. Though it is retained in the crosstab, it can be argued to be
endogenous because it is used in the construction of the success index. However, we retain here
in the explanation because it is described by the groups has a very important tool for signaling
commitment by members.
The eighth question looks at the diversity of the groups’ activities. And 70% of the
groups with no more than 3 activities fall in the lowest quartile, with no group in the top quartile.
In contrast, groups with more than three activities are evenly distributed across all the quartile
groups, with the smallest group being in the lowest quartile (19%). The results and direction of
association are significant at the 1% level.

The most common and cheapest mode of communication in rural areas is verbal;
therefore, spatial and geographical location of the collective effort becomes a crucial factor. It is
difficult to pass information about meetings and other crucial marketing information to the group
members when they are dispersed over a large geographical area. Transport means are also poor,
and often members walk to the meeting location. Producers are very busy, having many
commitments, and will be more averse to walking long distances to attend meetings. The ninth
question captures this effect, with groups operating in small areas having 31% of the groups in
the top quartile groups and in contrast to the other group operating a bigger area having 17% in

76

the same quartile groups.

The last question in the framework focuses on self-assessment of group leadership
quality. At the outset, a group may not admit inadequacies. After going through the previous
questions with members in a focus group setting, it was easy to finally reach the level where
members would honestly gauge the leadership qualities of their leaders. Table 2-4 has two ways
of capturing group leadership. The first is the presence of an altruistic leader and the second is
the self-assessment of overall leadership. Groups with an altruistic leader had 37% of the groups
in the top quartile, while the group lacking such a leader had only 12% in the same quartile. In
contrast, 42% of the groups with no altruistic leader fell into the bottom quartile, whereas group
with an altruistic leader have only 12% in this quartile. This association is significant at the 1%
level. The last question asked the focus group members to assess the overall quality of their
group leadership. Groups with a perceived good leader had 32% in the top quartile, while the
ones with bad leadership had only 12% in the same quartile. The association is significant at the
5% level.

In summary, questions in the analysis framework can be used for a quick assessment of a
group success rating. The framework can be utilized to pinpoint areas of intervention in an
already operating farmer group. We find that the results from this initial analysis conform to the
hypotheses derived from theory and the literature review. Based on the data, the factors the
framework finds key to success are group categorization (based on the asset specificity of a
good), presence of measurable targets, a group having regular monetary subscription, the group
working in a small spatial area, presence of an altruistic leader and having good leadership.
These initial group interview results thus validate the ability of the analysis framework to give

77

quick results for analyzing collective action groups. While the framework focuses only on some
key aspects of a group, section 2.5 gave the results of both econometric and case study analysis
approach of the factors of success. The results of an in-depth testing of the framework are
presented in the next section (2.6.2), followed by two cases that illustrates how the framework
can be implemented.

2.6.2 Regression results of the group analysis framework factors
The group analysis framework was developed to pick up the most important factors that
are necessary for a group to succeed. It is necessary to test whether the identified factors have a
strong relationship to the success of the groups interviewed. Therefore, to use the Framework we
needed to be confident that it has a decent prediction of the ‘success score index’. To test the
efficancy of the framework, we ran a number of regressions and one graphic representation of
the predicted mean success. The first exercise was to go through the 90 groups in the sample and
test the framework’s ability to predict success. Running each group on the framework, we
counted the number of ‘Yes’ and ‘Red Flags’ each group had and put that into the data.

Table 2-5 : Model to test the group analysis framework
Success score index is the dependent variable
(1)
(2)
(3)
‘Yes’
‘Yes’2
‘Yes’3
yes
0.042***
0.096**
0.038
(4.84)
(2.46)
(0.33)
yes2
-0.005
0.007
(1.39)
(0.33)
yes3
-0.001
(0.58)
Observations
90
90
90
Absolute value of z statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

78

Then we run a fractional logit regression on the number of ‘Yes’ each group had and calculated
the marginal effect. The results are in table 2-5 above model 1 column two.
Model one shows that as the number of ‘Yes’ increase, the success score index rises.
Therefore, we estimated model (2) to test for a local maximum. The ‘Yes’ squared coefficient is
not significant at the 10% level, and hence, we find no evidence of a local maximum. This is a
logical conclusion since a ‘Yes’ shows that a group is healthy in that aspect. However, a function
can have no maximum but still have an inflection point. It is at this point that there is critical
number of ‘Yes’, to be sure that a group will be successful. We therefore estimated model (3)
testing for local inflection point. Again we find no evidence of a local inflection point. A graphic
representation of the polynomial model (3) is shown below in figure 2-3. The polynomial graph
shows graphically that there is no inflection point. Howerever, as the number of ‘Yes’ reaches 9
the graph tends to show some leveling off. However, in the statistical analysis of model 2 there
was no evidence of a local maximum. We can conclude that though there is no maximum
number of ‘Yes’, by the 9th ‘Yes’, the 10th ‘Yes’ will not improve the Framework’s ability to
predict higher a ‘success score index’. We may then conclude that if a group has nine ‘Yes’ out
of ‘10’ possible ‘Yes’, then it has a very high probability of succeeding. Figure 2-6 shows that
below four ‘Yes’ the success score index will be below 0.5, which tending to the low success
region.

79

Success score index

1

.8

.6

.4

.2

0

2

4
6
No. of yes in group analysis framework
Success

8

10

Predicted mean success

Figure 2-3 : Polynomial plot of group-analysis framework test
2.6.2.1 Quantile regressions of the ‘Group Analysis Framework’

Two types of regression, a linear OLS and quantile regressions, are estimated. In the
earlier regression analysis, we used fractional logit since the success score is an index bound
between 0 and 1. Employing the same method or least squares regression, the resulting estimates
of various factors on conditional mean success score may not be indicative of the size and nature
of the effect on the lower or upper tail of the score distribution. In an effort to focus attention on
distribution tails directly without censoring, the data we use quantile regression. Quantile
regression gives a complete picture of the covariates’ effects at the tails by estimating a family of
conditional quantile functions (Koenker and Hallock, 2001). In our case, we choose to estimate
quartile regression so that we can gain insight into the effect of the Group Analysis Framework

80

factors on the least and most successful groups as well as the median group(s). To get the
required quartile regression estimation, the following conditional quantile function is solved as
shown in Koenker and Hallock (2001).

∑

[

(

)]

We also estimate the mean effect of the covariates to the success score by OLS for comparison.
Results in table 2-6 below show OLS, lower quartile, median and upper quartile regressions.
Group categorization is positive and significant at the 5% level in the lower quartile
estimation. It is very important for poor performers to be more commercially oriented if they are
to perform better. This factor is not that important as the group rises in the success index score.
This result is as expected and hypothesized earlier in the analysis framework discussion.

The negative sign on the coefficient for external initiation of groups hints that external
initiation decreases a group’s probability of success; however, the coefficient is not statistically
significant. From our sample, we can therefore conclude that there is no statistical evidence that
once groups are involved with external agents, they are more likely to fail, as proposed by some
sociology literature mentioned earlier. But we can equally conclude that there is no evidence
from our data that external initiation increases the chances of success.
The next question in the framework is about bylaws. However, the survey did not rate groups’
bylaws for their articulation and enforceability. Only 5% of the groups interviewed did not have
bylaws at all. Therefore, this variable does not have much variation in this sample. To capture
the groups’ ability to make good rules, the analysis uses the “convention” variable that captures
whether the groups had some accepted conventions apart from the written bylaws.
81

Table 2-6: Testing of the factors that used in the group analysis framework

VARIABLES

Commercial
External
Convention
Target
Buy product
Subscription
# of village
# of activities
Altruistic
Constant

(1)
OLS

(2)
.25 Quartile

(3)
LAD
(Median)

(4)
.75 Quartile

0.453**
(0.225)
-0.0835
(0.221)
0.0930
(0.223)
0.388*
(0.211)
0.561**
(0.258)
8.70e-05
(0.000132)
-0.0815**
(0.0394)
0.184***
(0.0557)
0.360*
(0.214)
1.129***
(0.386)
90
0.343

0.619**
(0.287)
-0.0307
(0.256)
0.0842
(0.235)
0.161
(0.253)
0.726**
(0.300)
0.000156*
(8.70e-05)
-0.0275
(0.0361)
0.217***
(0.0697)
0.139
(0.271)
-0.0587
(0.496)
90

0.392
(0.307)
-0.0955
(0.298)
-0.0365
(0.290)
0.736**
(0.288)
0.365
(0.375)
4.87e-05
(0.000117)
-0.0945*
(0.0522)
0.244***
(0.0877)
0.412
(0.308)
0.765
(0.569)
90

0.434
(0.268)
-0.242
(0.297)
0.0953
(0.298)
0.276
(0.280)
0.427
(0.382)
-0.000187
(0.000116)
-0.102**
(0.0462)
0.160**
(0.0651)
0.708**
(0.284)
2.091***
(0.501)
90

0.2799

0.2540

Observations
Adjusted R2
Pseudo R2
0.2799
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

Having conventions shows that a group is flexible enough to notice the inadequacy of its
bylaws to guide the operation and institute other laws that may not yet be filed with the registrar
of groups. The coefficient on convention variable has the expected positive sign, although it is
not significant.

The variable for the presence of well-articulated targets captures the management
abilities of the group in general. Its coefficient is significant in the OLS and LAD regressions but

82

not in the lower and the upper quartiles. This can be interpreted as the overall need for targets in
general. The expectation was that it would be significant for the lower quartile, but this was not
the case.
Buying of non-members’ product to trade shows the planning and marketing abilities of
the group. The variable was used to capture the groups’ ability to have well thought-out
objectives, as this was not well captured in the survey. It is significant in the OLS and lowest
quartile regressions. Good management abilities and the ability to plan effectively to have
enough capital to buy non-members’ produce and then market it at a profit is essential for a poor
performing group to succeed. It shows that groups that trade with non-members’ product
increase their success index.

The subscription variable is a dummy showing if the individual ever paid any
subscription to the group with 1 being Yes and 0 being No. The variable is significant for the
lowest quartile. The presence of many non-cooperators could lead a group to be in the lower
quartile. Introducing regular contribution to a group weeds out the free riders as well as
increasing commitments of the members due to increased investment. However, this variable
could be endogenous since it is correlated with the success index because we used the per-capita
contribution index in creating the overall success index. This will be investigated in future
studies.

Operating in large geographical areas is captured by the variable indicating the number of
villages. The results show that there is negative relationship between success rating and the
number of villages. This confirms the hypothesis that coordination is a challenge for groups
operating in a large geographical area. The variable is significant for all the estimations with the
83

exception of lowest quartile.

Diversification of group operations is captured by the number of activities a group
performs. The variable shows a positive relationship with success rating. It confirms our earlier
argument that groups need to add membership value by adding more activities. The coefficient is
significant for all estimations at the 1% level.

Having an altruistic leader in a group increases the success index score. The coefficient is
significant only in the top quartile and OLS regression. This variable captures good leadership
quality of a group rather than using the self-assessed leadership ranking which could have been
biased since leaders were present when this question was posed. We expected that the variable
would be significant for all estimations, but it is only the case for the upper quartile.

2.6.3 Implementation of the group analysis framework
After showing that the framework has predictive power on the group success score, we
use two cases to test the group analysis framework further in a case study approach. The two
groups used have a similar background in terms of starting period, agro-ecological zones,
external support services, and main activities. They are Group E and Group K. Group E started
in 1999. The marketing representative attended a world summit in Italy on social work and felt
the need to start a group in her home area after returning. She sensitized the community, and
eight groups merged together to form the current Group E. The current membership is 200 and is
composed of 50 male and 150 female members. It is located in Kathonzweni market some 10
kilometres from Wote, the main district town.

84

Table 2-7 Summary of the group analysis framework
Questions
Is this a commercial
group?
Is the group selfinitiated?
Does the group have
well-articulated bylaws?
Are the by-laws
enforceable
(enforced)?
Does the group have
well defined
objectives?
Does the group have
measurable targets?

Group E
Yes

Comments
Continue
(good sign)
Continue
(good sign)
Continue
(good sign)

Group K
No

Yes

Continue
(good sign)

No

Red Flag raised
(potential problems)

Yes

Continue
(good sign)

No

Red Flag raised
(potential problems)

No

Red Flag
raised
(potential
problems)
Continue
(good sign)

No

Red Flag raised
(potential problems)

Does the group have
regular subscription
fees?
Is the group
involved in more
than one activity?
Does the group
cover a small
geographical area?

Yes

No

Red Flag raised
(potential problems)

Yes

Continue
(good sign)

No

Red Flag raised
(potential problems)

No

Red Flag
raised
(potential
problems)
Continue
(good sign)
Success

No

Red Flag raised
(potential problems)

Does the group have
good leadership?
Conclusion

Yes

No

Red Flag raised
(potential problems)
Collapsed and had to be
divided into several
smaller units

Yes
Yes

The group has 8
yes and 2 no,
which implies
that it is likely to
succeed

No
Yes

The group has 9
no and 1 yes,
which predicts
that the group is
unlikely to
succeed

Comments
Red Flag raised
(potential problems)
Red Flag raised
(potential problems)
Continue (good sign)

Group H is located about 5 kilometres away from Emali town of Kibwezi district. It was
started in 2003 through encouragement by ICRISAT/KARI (through the TARGET project) and
85

the local church development committee of Catholic Diocese of Machakos. It was composed of
336 members--120 males and 216 females, all farmers. Both groups were initially started to bulk
and market members’ surplus cereals and pulses.
We use the framework as shown earlier (figure 2-2) to analyze the two groups. The
process is summarized in table 2-7 below. We also go through step-by-step implementation of
the whole process tree, showing the decision taken at each step for each group. The analysis
framework implementation clearly predicted that Group E, with a success score of 0.85, would
be a successful group and that Group H, with a success score of 0.36, was unlikely to succeed.
Group E is described earlier in the commercial group section. Therefore we highlight a
few contrasting characteristics to show the analysis by the framework. Group E was self-initiated
as a food group and later evolved into a commercial group by diversifying its activities. It has
well-framed by-laws that are read to the members regularly and strictly enforced. The group also
had very well defined objectives, although they lacked measurable targets. Members contribute
Kshs 40 per month (US $ 0.52), which increases their commitment to the group.
The group not only bulks and sells seed and grain cereals and pulses but also has other
activities like selling handmade baskets and other locally made items by the women. It also
stocks and sells mosquito nets. However, the group serves a large geographical area that makes it
difficult to communicate effectively. It regularly holds elections, and in the last 8 years have had
8 elections. It has well-informed and self-assessed good leadership. The group formation and
marketing activities have benefited from an altruistic and visionary marketing representative.
With only two out of ten areas checked raising red flags in the analysis framework, it is safe to
predict a high probability of success (See table 2-7 above).

86

External agents initiated Group K as a food crop-marketing group. The group served a
large geographical area, which made it very difficult for members to communicate. Members did
not know each other well and did not have common interests. Through the effort of the Diocese
and ICRISAT, the group was able to draw reasonably good by-laws. However, the management
committee did not enforce them due to poor leadership. Like Group E, this group had a shop for
bulking and marketing its produce. It is located five kilometers away from the main MombasaNairobi highway. This enables individual traders in the area to buy products and transport them
to the two big cities due to proximity to the main highway.

35

In comparison, Group E is about 60

Kilometers away from the same highway with very bad road conditions. Group K’s by-laws
called for a regular contribution, but this was not enforced. From 2002 until early 2007, Group K
had conducted a single election, which was against the bylaws; the bylaws stipulated that
elections were to be conducted every year. Lack of regular elections brought up many leadership
problems such as the lack of committee cohesion. The group also did not develop workable
objectives with measurable targets. It did not diversify its operations, and its sole activity was to
bulk and sell pigeon peas collectively.
The framework correctly predicted that this group was unlikely to survive (see table 2-7
above). In 2007, members subdivided the group into three small groups, each serving a smaller
spatial location. The new groups have learnt from previous failure and might have a higher
probability of survival in their second phase of their lives.

From the empirical and group framework analysis, the factors selected in the series of
questions give a good view of each group. Hence, the framework can be utilized to investigate
35

At the time of the study, one of the group leaders was buying large quantities of green pigeon
peas (as a private individual) and taking them to Mombasa for sale.
87

the potential of a group and pinpoint critical areas of interventions.

2.7 Conclusions
The study objectives were to identify conditions necessary for smallholder groups to
emerge and succeed. From the analysis and literature reviewed, the conclusion is that the nature
of a good (crop, activities) will decrease or increase transaction costs an individual farmer faces
in marketing her products. The nature of transaction costs will either foster or discourage
collective action. An individual with a perishable product with frequent transaction faces higher
transaction costs than someone producing a less perishable product that can be easily stored.
Therefore, areas dominated by a low transaction specific good will see few producer marketing
groups formed. However, for a successful farmer organization to emerge, regardless of the nature
of the good, it is easier if there is an altruistic leader or a small “core group” of individuals
spearheading the effort than a mass mobilization. Such an individual or core group is likely to
invest more effort and resources in putting into place structures that enhance the probability of
success. They will have moral authority to sanction a free rider and discourage individual
opportunistic behavior that, left unchecked, leads to a low Nash-equilibrium of group failure.
The results suggest that development organizations or government agencies working with
groups should try to work with already existing groups. Such groups would have developed an
internal mechanism that reinforces collective behavior for the good of its members. If an external
agent were to initiate a new collective action effort, it should evaluate if a group is warranted
given the prevailing conditions. And if a group is needed, then the external agent should be
willing to invest more time and resources into education of all members in group dynamics,
management trainings, and strict monitoring. The lesson learnt is that coming up with successful
new groups is tricky. The process of electing good leadership is difficult without members’
88

initial bonding and unity of purpose. Therefore, it is very difficult for the individuals to work
together. In addition, groups formed because of direct intervention by an external agent usually
have very high initial expectations. They expect lots of monetary and in-kind support from the
supporting agency. There is a general lack of ownership of the collective effort. Lack of
fulfillment of expectations within a specified period leaves members disillusioned, making it
harder for a group to succeed.
In the initial stages of group formation, membership should be drawn from a small
geographical area. This will ensure that individuals have interacted with each other in past.
Repeated games in previous interactions help potential members to identify cooperators easily.
Smaller groups in terms of membership make it easier for individuals to build greater social
capital that helps to reduce the resources used to monitor each other. A diversified portfolio of
activities increases the gross benefit a member derives from collective effort and thereby
increases commitment that leads to greater success probability for a group.
Groups’ internal organization and governance contribute to their success. This is a
function of the leadership qualities of the elected officials and the trust the members have in
them. Stronger social networks reduce negative behavior in a group. From the analysis, we can
conclude that requirements to adhere to by-laws improve members’ commitment to a group. If
such measures are enforced, only committed farmers will join, and the non-committed ones will
withdraw from the groups before there are problems.
The study has also demonstrated that the proposed group-analysis framework does a
decent job in identifying intervention areas in operating groups and issues needed to be
addressed in new group’s constitution. Using econometric analysis with all the 90 cases we have

89

also demonstrated that the Group Analysis Framework actually predict the success of a group. If
a group has four or less ‘Yes’ it will be below 0.5 on the ‘success score index’ scale. However,
more studies need to be conducted using this framework to make sure that it can be generalized
to a variety of conditions.

90

Chapter 3 : Determinants of farmer’s participation in collective action, patronage, and
intensity: the case of farmers in semi-arid areas in Kenya
3.1 Background
Before 1989, most of grains and other agricultural produce in Kenya and in many
developing countries were marketed through government-controlled parastatals or marketing
boards. Pre-liberalization, there was the hope that with liberalization, market forces would get
into play and efficient marketing systems would take the place of the often inefficient and
corrupt government agencies. However, since 1990, when most of agricultural marketing
services were liberalized in Kenya, there have been mixed results in the development of
agricultural input, output, and financial markets. The expectations were that new institutions
would spring up and offer attractively priced, timely, and reliable services that are critical for
more intensive cropping, and particularly for intensification of cereal and pulse production in
poorer rural areas (Dorward, 2001, Fafchamps, 2004, Jayne, 1997). One form of a marketing
channel expected to spring up was some form of collective action effort—in particular, Producer
Marketing Groups (PMGs)— to compete side by side with private traders or as a form of
bargaining agent. However, these new institutional arrangements in many cases did not spring up
automatically to cover the gap left by exiting parastatals, and those that did form often did not
deliver the desired results. Members of producer-marketing groups often do not patronize them
effectively, thereby not reaping desired benefits. The 2005 ICRISAT (2006) survey in Kenya
showed that despite the sensitization about the benefits of forming PMGs, not all villagers
participated, and even those who participated sold only a small portion of their produce through
the groups (Shiferaw, et al., 2005).

91

3.2 Research questions
It is important to determine the impediments that cause farmers not to join PMGs and
patronize them effectively. This chapter investigates:
i)

What are the determinants of participation in PMGs?

ii)

What are the factors influencing patronage of producer marketing groups?

iii)

What factors influence the intensity of collective action?

The chapter sheds light on the factors that affect collective action in a low-incomecountry setting and in particular, in low-potential, semi-arid areas such as the study areas in
Kenya. This is very important and timely, as the development community is seeking the best
methods of employing PMGs to improve small-scale farmers’ access to markets. Although some
work has been done in this area, most has been in developed countries. There is also a marked
lack of literature that has applied econometric methods to estimate the magnitude of different
factors affecting collective action. In part, this lack of quantitative analysis has been due to the
difficulties of measurement of factors like transaction costs and social networks.
3.3 Literature review
In much of rural Africa, in order to market farm product surplus, farmers have to grade
the produce according to grades that are not well defined, understood, and accepted by all the
players in the market; search for a buyer; gather market information; transport the produce to the
market; and negotiate for the price. All these tasks have a high transportation, opportunity, and
transaction costs (Barrett, 1997). Since marketing and transactions costs involved are high,
farmers opt either not to participate in the market or, if they do, to sell at the farm-gate to an
intermediary. Often, they have little or no choice of a buyer due to low competition in the
market. Lack of competition gives private traders the ability to exert greater market power
92

(Bellemare and Barrett, 2006, De Janvry, 1991, Kindness and Gordon, 2001). Hence, poor
smallholders capture only a small percentage of the value of the products they sell, while
middlemen and retailers higher up the commodity chain get a much greater share. For example,
an analysis of the charcoal commodity chain in Senegal found that the net profit per sack of
charcoal of a typical woodcutter at the base of the chain is about 21% percent of the total profit
compared to a merchant profit of 52% (Ribot, 1998). In the absence of a direct market for
tomato producers in Turkey, middlemen offered low prices (Esengun, et al., 2005). Hence, there
are incentives for farmers to obtain a higher return for their effort by controlling the initial stages
of the supply chain through a PMG.
Formation of PMGs gives producers another choice of a marketing channel apart from
middlemen and spot markets. Farmers can continue marketing their produce as individuals or
through producer marketing groups. Often producers will employ multiple channels. From
economic theory, we predict that each household will choose a combination of channels that
deliver the highest utility. This chapter explores the individual household’s incentives to join
producer-marketing groups. Underlying these incentives is the perceived distribution of benefits
and costs. These are influenced by factors related to the nature of transactions costs faced by the
individual producers. They are a function of the nature of the products they produce, the
characteristics of the community, the social networks within the community, the resource and
external environment such as the role of external organizations (government and NGOs), and
accessibility to markets by individuals.
Literature on collective action is vast, and we just cover it briefly here. More details are
available in Olson (1965), Reisman (1990), and Sandler (1992). Collective action requires
involvement of a group of people with shared interests in some common action directed towards
93

achieving some shared interests (Marshall, 1998, Meinzen-Dick, et al., 2001, Sandler, 1992).
Usually, economic theories of collective action are concerned with the provision of a high
exclusion cost good (HEC)

36

(Ostrom, 1990, Schmid, 1989, Schmid, 2004) and other collective

consumption goods through the collaboration of two or more individuals.

37

The theory explores

situations where there is presence of market failures, and individual consumer rationality coupled
with firms' profit-seeking behavior do not lead to efficient provision of the good. Features such
as the group’s purpose and existence of a common problem within a society have been identified
as fostering participation in collective action. Collective action theory, therefore, gives us the
starting point in exploring factors that influence producers to join producer-marketing group.
Economic theory posits that individuals are more likely to participate if they share similar
economic goals and uncertainties as well as socially accepted norms.
Since individual behavior towards collective action is influenced by benefits and costs of
participating, we can utilize the theory of utility or profit maximization to analyze participation
(Meinzen-Dick, et al., 2002). There is a broad consensus in the new institutional economics
literature that participation in collective action depends on four classes of determinants: the
characteristics of the collective action problem, the attributes of the group (members and nonmembers), the attributes of the institutional arrangements, and external influences (Sekher,
2001). Differences in perceived benefits of collective action by the households depend on their
36

In most literature, we find the term “public good.” The term ‘public good’ can be misleading:
First, it refers to a private good with unique characteristics. Second, as Schmid (1989) indicates,
the term itself implies that such good should be provided in the public sector. In this essay, we
use high exclusion cost good (HEC). Schmid (1989; 2004) defines HEC as a certain type of good
that the owner of rights cannot exclude non-paying consumers easily from access to the good and
any benefits accruing from it.
37
Mancur Olson’s book “The Logic of Collective Action: Public Goods and the Theory of
Groups” lays out the economic foundation for collective action.
94

characteristics such as age, education, gender, occupation, values, beliefs, ideas, and economic
status (Hertberg, 2001).
Most studies in developing countries have concentrated on the question of market
participation for the smallholder producers under various conditions. They explore the factors
that influence smallholder farmer’s participation in agricultural product markets. Goetz (1992)
studied the participation of Senegalese agricultural households in grain markets, using a probit
model of households’ discrete decision to participate in the market, followed by a second-stage
switching regression model of the supply decision. He finds that better information and access to
new technology play important roles in the participation and supply decision. Holloway et al.
(2005) used a Bayesian double-hurdle model to study the participation of Ethiopian dairy
farmers in the milk market when non-negligible fixed costs lead to nonzero censoring. That work
is similar to studies by Key, Sadoulet, and de Janvry (2000), but Holloway et al. distinguish
between the discrete participation decision and the continuous volume marketed decision, as in
Goetz. They find that intellectual capital stock is a good complement to physical assets in the
decision to participate in the market. Key et al. (2000) developed a structural model to estimate
structural supply functions and production thresholds for Mexican farmers’ participation in the
maize market, based on a censoring model with an unobserved censoring threshold. Their model
differentiates between the effects of fixed and proportional (i.e., variable) transactions costs. The
results show that lowering of transaction costs and promotion of producer marketing
organizations increases both farmers’ output and market participation. Bellemare and Barrett
(2006) model market participation of livestock producers in Kenya and Ethiopia using a Tobit
model to explore whether the producers make simultaneous or sequential market participation
decisions. They find that fixed costs deter market participation of the livestock herders in Eastern
95

Africa. Vakis et al. (2003) explores the Peruvian potato farmers’ choice of the marketing channel
and propose a method to incorporate transaction costs explicitly. The results again underscore the
importance of market information in reducing transaction costs and increasing market
participation. Hernandez et al. (2007 ) study how tomato farmers in Guatemala make a choice
between traditional markets and the supermarket channel using a sequential model. Their
findings show that wealthier producers are able to market through the supermarkets due to lower
transaction costs even though their profit rates are similar to traditional channels. Okello and
Swinton (2006) use case study approach to analyze the strategies smallholder traders use in
complying with strict enforcement of pesticide use on vegetables grown for export in Kenya.
They find that some individual farmers form producer groups to coordinate the enforcement of
the rules to individual farmers to remain compliant. All these studies have looked at determinants
of producer choice of whether to participate in the market or the choice of different channels in
the market. None looks at the choice between the traditional spot marketing channel and a
producer-marketing group. Only the study by Okello and Swinton (2006) looks at the choice
between individual farmers’ direct contracting with an exporter or use of a farmer group to sign a
marketing contract. However, the study uses a case study approach whereas our study uses a
quantitative approach. It is this gap we wish to fill in this chapter. This is very important since
collective action in many forms has been promoted as an alternative or complement to traditional
marketing model.
Many of the studies reviewed underscore the role played by access to marketing
information. A recent development in many developing countries is the widespread use of new
information communication technology. Mobile phone use is found even in the remote parts of
Kenya now, and this redefines the traditional means of access to market information and
96

transaction arrangements, with a likely impact on individual marketing channel choices. Mobile
phones make it easier for a rural farmer to have a social network wider than her locality. An
important source of transaction costs is information asymmetry between actors. Increased use of
mobile phones in the rural areas is affecting the way actors transact business and the distribution
of transaction costs. In 1998, Kenya had 15,000 cell phones; in five years, the number had grown
by 139 times to 2.1 million by 2003. Then in three years, the number of subscribers increased
again by 3 times to 6.5 million users by 2006 (CCK-Kenya, 2006). By 2008, mobile subscribers
had increased to more than 12 million, and since one line is shared by at least two individuals,
the national coverage is over 25 million people in the country (CCK-Kenya, 2008). With new
technologies that enable producers to gather marketing information easily, we expect that the
marketing structure will change and the role the PMG plays in gathering information will also
change. Prior to introduction of mobile phones in the rural areas, traders from the major cities
had to drive to rural areas to search for the produce. This increased the search cost, and traders
had to factor the additional cost in the price offered to farmers. Currently, one finds these phones
in almost every village and indeed in many homes in Kenya. The fact that a non-cell-phone
owner will also access messages from the neighbor who owns one increases the penetration rate
of this form of communication. Therefore, a trader can call in advance to negotiate for a product,
decreasing the search costs. On the other hand, a producer will also call other traders or friends
to find the prevailing market prices. There are other service providers that provide market
information through short message services (SMS). A farmer or trader can find up-to-date prices
of a commodity in most major cities by accessing a server using a SMS for a fee of about US $
0.23 per access. The improved information access will affect the balance of bargaining power
between individual farmers and groups, on the one hand, and traders on the other hand.

97

Communication among group members is also enhanced or altered. Chowdhury (2006) finds that
reduction in information costs through access to a fixed telephone changes the functioning of
markets and households’ participation in factor markets.
A recent study in Niger assesses the impact of cell phones on grain market performance
(Aker, 2008). Aker finds that introduction of cell phones is associated with a reduction of 20
percent in grain price differences across markets, with a larger impact on markets that are farther
apart and those that have poor road network. Akers study looks at the effect of ICT on marketing
structure at the trader and consumer level, whereas the Chowdhury study investigates the effect
of fixed phones on factor markets. We extend Chowdhury’s and Akers’ work by investigating
the effect of cell phones at households’ decisions to participate in CA, and their choice of the
level of patronage. Therefore, our study will continue to shed more light on the way ICT is
affecting various sections of the agricultural sector that has not been studied. No other study
reviewed looks at use of new communication technology in the rural areas and its impact on the
rural marketing channel choices. The paper therefore adds to the literature by looking at
marketing choices producers make between collective action and traditional spot market and the
impact of structural change in information access.
3.4 Theoretical Framework
3.4.1 Collective action participation and patronage model
We hypothesize that households will make decisions on the channel(s) through which to
market their produce in order to maximize their expected utility. We then assume that the utility
a household derives from the choice of channels depends on the set of available channels C. In
the case of many rural areas in Kenya and more so in the areas we conducted the survey, there
were mainly two choices from which to choose. A household had to sell through a combination
98

of the spot market and a producer marketing group or the spot market alone. The spot market
here includes going to the open market to sell the produce or selling at the farm gate to a broker
or a trader. For ease of modeling we include both these channels in our definition of the spot
market.
Let us assume that the household has a utility function of the form:





ij  V Zij , X i , Ei  i

(3.1)

For any given household i a given level of utility will be associated with any choice of the
alternative marketing channel j. The utility derived from the channel choices depends on the
attributes (Z) of the channels, the socio-economic characteristics of the household (X), and the
social capital, location and the environmental factors (E) (Akin, et al., 1995, Alaba and Alaba,
2002, Whittington, et al., 1990). The utility of choice comprises the systematic component (the
first term) on the right hand side of the equation (3.1) and the error component εi. The error
component is independent of the systematic component and follows a predetermined distribution
(Ben-Akiva and Lerman, 1995, Manski, 1973).
A household choice between the alternative channels is a function of the probability that
the utility associated with a particular option (j) is higher than all other alternatives. Assuming
that the relationship between utility and characteristics is linear in parameters and the variables in
the function, and that the error term is identically and independently distributed, then we can
have a reduced-form equation.

99

3.4.2 Reduced-form market choice
The household maximizes its expected utility by choosing the marketing channel that can be
modeled as:
From equation (3.1), we can specify a reduced form model as follows:

E     X ij   Wi i  Z  i   i

(3.2)

Where  is the utility that is associated channel j, where j= [1, 2], which for our purpose
signifies some a combination of the PMG and the spot market or the spot market only. X ij is a
vector of household characteristics that vary across the two alternative marketing channels
considered, Wi are attributes invariant across the market alternatives such as location
characteristics, Z are market characteristics, δ are a vector of social networks,  , i ,  and 

38

are a vector of parameters to be estimated, and  i is idiosyncratic error term.

Let Cij denote the market choice channel combination that maximizes utility for
household i.
If we assume that  ij are normally distributed, and since our interest is to determine if the
household will choose to market through the PMG, we can model this decision using a binary
model:





P Cij  1| X ,W  G   X ,W     P  X ,W 

38

(3.3)

 captures the effect of marketing-channel-specific parameters such as distance to the market,
while i are parameters that show the impact of individual household characteristics on the
market choice.
100

Let (X,W)=λ , where λ is 1×K and β is K×1. Since we assumed the error terms are normally
distributed, we can estimate the choice as a probit model:

G       

  v dv
 

(3.4)



Where     is the standard normal density,

1
 2
 exp   
      2  2
 2 


(3.5)



3.4.3 Hypotheses
In the empirical analysis, we seek to test the following hypotheses:
Hypothesis 1: Previous assistance from an external agent will increase the probability of a
household being involved in producer marketing groups.
External organizations may facilitate collective action efforts by providing technical
support (in terms of training) and other inputs, provided that these interventions are
complementary to local collective action. External agents may also improve participation
because they usually tie access to their services to being a member of these organizations.
However, external organizations may hinder collective action participation if their role
substitutes for local efforts (such as replacing local effort or interfering in management
decisions) or outright undermining collective effort (Berhanu, et al., 2001, Pender and Scherr,
1999). We control for the effect of external organizations by including a dummy variable that
captures whether a household received services from a development agency. Prior experience
with local organizations could favor collective action due to possible learning effects and the
effect of social capital on the costs or ability to enforce collective action (Baland and Platteau,

101

1999, Pender and Scherr, 1999, Rasmussen and Meinzen-Dick, 1995) . However, a bad earlier
experience may hinder subsequent participation in collective action efforts.
Hypothesis 2: The more isolated a household is from the market (distance and conditions of the
roads), the more likely it is to join a PMG.
Better market access may lead to an increased level of competition by traders, arising
from reduced marketing and search costs. Consequently, better market access may decrease
participation in collective action by reducing the incentive of members to abide by collective
action rules by providing more ‘exit’ options and making enforcement of rules more difficult
(Baland and Platteau, 1999, Berhanu, et al., 2001, Pender and Scherr, 1999, Rasmussen and
Meinzen-Dick, 1995). On the other hand, development agencies may target more remote areas
because they work mostly in less developed area. If this is true, then the association between
remoteness may depend more on the decisions of the external agencies pushing the creation of
PMGs than on individual choice by the household about whether to join a PMG because of the
transaction costs it faces in using the spot market. Though we may not be sure which reason is
true in all situations, we will test the association between remoteness of a location and the local
farmers joining of a PMG to access the markets and other services. In this study, market access is
measured by the walking time taken to reach the nearest market town from the village or their
point of produce sales.
Hypothesis 3: Individuals with a higher social network index will participate more in the PMGs.
Individuals share information with the other members of their group in the social network
and reach a common understanding and perception when faced with a new technology (Valente,
1995). A person’s connectedness in a social system will affect his/her adoption of new
102

technology (Rice, 1993). Therefore, a head of household with a big social network will tend also
to participate in collective action. Social network in this chapter is estimated through several
indexes that are discussed later. Having many friends in collective action will over time
overcome the free-riding mentality where individuals wait and see if there is a benefit from the
group before joining. Hence, due to network effects, these producers will eventually join more
groups. This is referred to as “behavioral confirmation” that stems from actors' desire to follow
prevalent behavioral patterns of relevant others (Lindenberg, 1986) . Individuals have additional
motivations to join the PMG if their friends do so. On the other hand, they will not participate in
the PMG if their friends do not participate (Oberschall, 1994). Intensive interpersonal ties
(friendship) are an important avenue of the spread of social influence that facilitates mobilization
in collective action (Janky and Takács, 2009).
Hypothesis 4: Access to mobile phones will decrease individual producers’ participation in the
PMGs.
Access to good communication services in rural areas by households will affect choice of
the marketing channel. We consider access to telecommunication as access to information and
potential reduction in transaction costs through reduction of search costs and solving of the
coordination failure between traders, producers, groups, and within a group. Previous studies
have found that better marketing information plays an important role in market participation, and
we expect also that the new information technology will affect the choice of marketing channels.
Normally, producer marketing groups give an individual producer low-cost access to marketing
information through the collective effort. Access to a mobile phone may provide a farmer an
alternative faster source of information. This decreases the perceived benefit of a PMG, which in

103

turn decreases participation and might necessitate added benefits to induce farmers to join the
collective effort.
3.5 Methods
3.5.1 Decision to participate in the PMGs and patronage
This section addresses the two questions of participation in collective PMGs as well as
39

the level of patronage (i.e., the amount of produce sold through the group) . These are
important questions because if there are producer-marketing groups, then we expect that
members will market a proportion of their marketable produce through them. Answering these
questions will provide development agencies and government policy makers with the crucial
areas of interventions to have maximum impact with such collective action programs.
Previous research among the same respondents in the study area has shown that even the
participating members were not selling all their marketable produce through the producer groups.
If quantities sold through the groups are large enough to have a sizeable share of the market, then
members can benefit from the potential benefits of economies of scale, scope, and greater
bargaining power. Otherwise, being a member of a group will not be important, and the groups
may eventually fail out of irrelevance. So we examine the factors that determine the household’s
decision to use more of the producer groups’ marketing services compared to the spot market.
Producers will use more of PMG’s marketing services than the spot market if equation (2.1)
holds true or there is a strict inequality.
However, we can estimate a supply function for sales through collective action, as
follows:
39

The level of patronage is chosen by the household’s choice of the level of consumption, and
the rest is then marketed through the two channels.
104

q(t )C  f  X , A, E, SN , MC, I , t , L 
j

(3.6)

The factors used to explain differences in participating in PMG marketing include
individual household characteristics (X). These are in line with incentive and constraints that a
household may face that affect its adoption decision. Household wealth (A) variables are also
used since wealthier individuals may have more choices in marketing channels. We include
involvement of external development agencies (E); experience with other organizations and
social networks (SN); distance to market measured in time spent walking and transport costs
(MC); and sources of information (I); a time trend, which is captured by year variable (t), and a
spatial location variable (L).
Past studies on market supply decisions have treated these as a sequence of two steps, a
market participation decision followed by a supply volume decision (Goetz, 1992, Key, et al.,
2000). However, most did not include information sources explicitly. We use information
sources as dummy variables to estimate the effects of access to marketing and collective action
information. We assume a two-step decision-making process. Let i = [1, 2,..., N] denote the
households in question. Each household compares the level of expected utility derived from
participating in C* against its reservation utility attainable without participation, C* . Here, we
j
Nj
use the superscript “*” to denote the fact that both levels of utility are latent (unobservable)
random variables. Following the literature, we hypothesize that the difference between the utility
levels is determined by a vector of covariates as shown below.



CA j  f X i , Ai , L, PE j , N , D j , SNi , I j

105



(3.7)

where CAi

40

is a discrete variable equal to 1 if a household participates and 0 otherwise,

CAi 0,1 , Xi are observable household-head-specific and household characteristics (age,
gender, years of education, and number of members of the household), Ai is a vector of asset
endowment of the household (this includes the number of acres of land owned and value of
productive assets), L is a spatial location dummy, PEj is membership in an earlier collective
action effort (this is a dummy variable with a value of 1 for prior experience), N is a dummy
variable where 1 indicates that the household received extension services from one or more
development agencies working in these areas, Dj is the distance to the market (this is measured
in time taken to reach the market), SNi are the social network variables, and Ij represents a vector
of sources of marketing and collective action information.
We model the decisions as two step processes. The two methods used in the literature to
model the two step process are the double-hurdle model (Dow and Norton, 2003, Duan, et al.,
1984, Goetz, 1992, Jones, 1989, Leung and Yu, 1996, Madden, 2006) and the Heckman sample
selection model (Amemiya, 1985, Heckman, 1976, Wooldridge, 2002). The estimation procedure
used here is the Heckman self-selection model. In chapter 2 we have argued and shown that the
more successful groups have members who have self-selected into the groups. Therefore, the
error term of the selection model is correlated to the patronage model, and the Heckman
procedure takes into account of this correlation to avoid biased results. The observed sales
through a group are conditional on the decision to participate in the group marketing effort. We

40

Notice we have changed the notation from the general choice of channel in the earlier
equations of Cj to CAj since we have two choices that we are evaluating in this chapter.
106

assume that a vector of variables X affects participating in collective action effort, and Z
variables determine the intensity of sales. This implies that X could be the same as Z.
First, a probit model for participating in collective action is estimated as follows.
*

Suppose that the latent variable CAi follows:
CA*  X i   i
j

(3.8)

where  i is independent of Xi , which is a 1 by K vector of factors affecting the decision to
participate for all households (i), β is a K by 1 vector of parameters, and  i ~Normal(0,1).
Instead of observing CA* , we observe only a binary variable indicating the sign of CA*
j
j

 1 if CA*  0 

j 
CA j  
*  0
 0 if CA j 



(3.9)

The second decision on how much to sell through a PMG is modeled by a truncated
model to take account of those who have zero sales through the group.

*
Zi  X i   i



Zi  



, i ~ N (0,  2 )

*
Zi if Zi  0 and yi 1
0 Otherwise.

(3.10)

(3.11)

where Zi is the level of sales through the organizations, which depends on latent variable zi
being greater than zero and conditional to the decision to participate, CAj.
107

*

The first stage estimation examines the determinants of participation, while the second stage
provides the determinants of intensification or degree of patronage in collective action.
3.5.2 Indexes
The indexes used for various measures of social capital in this study were calculated from
various questions posed in the questionnaires. Each of the indexes was a result of various
measures used that were normalized to a percentage for ease of interpretation. Thus:
 N
  snv j * gwt j
 j 1
Socialnetwork indexi  
K

 gwt * k

k 1






 

 *100




(3.12)

Where snvi refers to the various dimensions of the particular social network index i measured.
Since each index is calculated from the responses of several questions, it is weighted by the
 K

maximum possible total   gwt*k  . And gwt is the weight for each question or part of the


 k 1


question that constitutes the index and K denotes the number of questions or parts of a question.
Each of the indexes used will be described later in the results section.
3.5.3 Collective action intensity model
In collective action literature many studies have looked at the factors that influence
individuals to get together to produce a common good. Others like Olson (1965) and many other
subsequent studies have looked at critical mass theory investigating the ideal group size (Esteban
and Ray, 2001, Oliver and Marxwell, 1988). However, when studies are conducted in the field,
one finds that any households join more than one group that give different services and
sometimes join multiple groups that deliver similar services. We therefore felt that there was
108

need for further investigations on the factors that drive households to invest their scarce capital
resources in several collective action efforts. To the best of our knowledge, this is the first study
that formally investigates this phenomenon through quantitative analysis. Resources have
opportunity costs, and joining a group takes away some resources from other household
productive services. Our investigation therefore, is a step forward towards the understanding this
phenomenon.
To investigate the reasons households choose to belong to more than one social network
(social group), we apply a basic utility model developed by Estaban and Ray (2001) who initially
used it to investigate the collective action and group size paradox first stated by Olson. Others
have used variation of the model to look at coalition formation and lobby group membership
(Anesi, 2007) and even in the case of an environmental treaty (Murdoch, et al., 2003). In our
case we use similar notation but instead of looking at a single group and the number of agents
joining the effort, we investigate an agent making decision as to the number of groups she
decides to join.
Let a denote the level of effort the agent contributes to the collective efforts. This effort
can be in form of money or time contributed to the various groups.
We assume that individual preferences are represented by the (additive separable) utility
function

  w, a   w   a 

(3.13)

Where ν is an increasing, smooth, convex function with ν’(0)=0, and w is the per capita benefits
from various collective efforts, a is the level of effort the agent contributes to the collective

109

action and hence v is the disutility of expending resources a (effort, money) to participate in the
collective action. As in Estaban and Ray (2001), the method is equivalent to measuring expected
utility in units of the collective good. From the sum of the benefits we subtract the cost of the
efforts contributed, translated into the equivalent units of the collective good.
If function ν is linear in a then efforts the agent exerts in participating in PMG activities
can directly be subtracted from the benefit, as in (Olson, 1965), and is equivalent to thinking the
agent will borrow extra capital from a frictionless credit market such that the rate of interest I is
insensitive to the amount borrowed. In such a case then ν(a) is just (1+r)a. The implication is
that as long as there is a capital market and there is some benefit from an extra group and the
benefits exceed the cost, the agent will increase the number of groups joined indefinitely.
Although such a scenario theoretically can occur, it is unrealistic because from the survey we
find that there is a limit to the number of groups a household joins. In this study we assume it is
more appropriate to assume that additional units of efforts are increasingly costly. Therefore, the
marginal rate of substitution between reward and effort, that is the amount of extra benefit that
will just compensate an individual for contributing an extra unit to another effort, increases as
total effort increases. This rate of increase, which is the elasticity of the marginal rate of
substitution with respect to effort, is what determines the number of groups (total efforts) an
agent will join.
This is a more realistic case that guarantees that one cannot join an infinite number of groups
even if they offer extra benefits since the cost of capital increases at an increasing rate (Demir,
2007). We then model the number of groups an individual joins as:



SNi  f X i , Ai , Li , Ti , SIi , GD j , PEi , N , TC j , D j , Ii , t
110



(3.14)

Where SNi is the dependent variable and is a count variable of the number of social groups a
household is involved in as a member, Xi are observable household-head-specific and household
characteristics (age, gender, years of education, and number of members of the household), Ai is
the asset endowment of the household (a vector that includes the number of acres of land owned
and value of household assets), L is a spatial location dummy, Ti is the trust index that measures
how much the household trusts the other members of the village (defined below), SIi is an index
measuring how much social insurance a household has within the community (defined below),
GDj is the group democratic decision making process index (defined below), PEj is membership
in an earlier collective action effort (the variable will be a dummy variable with a value of 1 for
prior experience), N is a dummy variable, where 1 indicates that households received extension
services from one or more development agencies working in these areas, TCj are marketing and
transaction costs variables, Dj is the distance to the market (this is measured in terms of time
taken to reach the market), Ij represents the sources of marketing and collective action
information, and t is a dummy variable of year that captures time trend.
We apply the Poisson regression to estimate the model for the following reason. We have
the distribution D  SN | X  when SN is an unbounded count variable. Let   X   E  SN | X 
where SN 0,1, 2,... and  .  0 . Then the distribution is a Poisson if the density is
f  SN | X   exp   X     X  SN / SN ! , where SN!=1.2…(SN-1).Y and 0!=1. We have a




count-dependent variable that ranges from 0 to 9. And the most popular method for estimating
count data is quasi-maximum likelihood estimation using the poisson quasilikelihood function.
111

Poisson is used because it is computationally simple, it also gives satisfactory results, has
compelling robustness and is relatively efficient (Wooldridge, 2002).
3.5.4 Data
Data used in all the models in this essay come from the household survey data previously
described in chapter one. We use pooled regression of the three surveys from 2003 to 2007.

112

3.6 Results
In this section, we present brief descriptive statistics of the main variables in the models
considered in this study. Then we move on to results of the PMG participation and patronage
models, and finally to the intensity model, which investigates the factors that affect households’
choices on the number of PMG efforts to join.
3.6.1 Project placement impact on results
As we discuss the results in this section, it is important to point out that participation in
the PMGs and group formation can be affected directly by program placement decisions made by
external agents working in the area. Apart from the LLP program mentioned in chapter one,
there were other development partners working in the study areas. And many of the development
groups encourage farmers in the rural areas to form groups to access their services. Since each
development group would like to open its operations in new areas, there will be situations where
they will operate in areas where farmers have had no previous experiences with external agents.
Such a situation will affect the impact and direction of some of the social-network and access-toinformation factors tested in this study.
3.6.2 Descriptive statistics
To explore the relationship between social networks and collective action, we calculated
several indexes as measures of the strength of farmers’ social networks and cross-tabulated them
with membership in PMGs. Table 3-1 (below) shows that households that have a higher
solidarity index are more likely to be members of a group.

113

Table 3-1: Social network impact on PMG membership

Solidarity index
Trust index of the
community and group
members
Diversity index
Number of friends in
PMG
Got NGO help*
Previous PMG
experience*

PMG member
No
Yes
21.24
50.14

Yes
No
Yes

510.6

Asymp sign
2 sided
0.000

60.63

58.88

28.6

0.158

28.77

73.81

642.3

0.000

7

Mean
values

Chi2

9

510.6

0.000

56.9%
42.5%
57.5%

43.1%
57.5%
42.5%

23.3

0.000

506.7
No
45.5% 54.5%
Note (*): NGO help and Previous PMG experience are row percentages

0.000

Solidarity in this study refers to extent to which members of the same village or social
network are willing to help one another in times of needs. In other words, we can refer it as the
degree of collective insurance in the society. In the survey instrument, we gave the respondent a
list with groups of items (seed, grain, other food items, cloths, and cash) a respondent would give
(receive) to (from) a fellow group member in times of a famine. The number of items mentioned
are summed up and then divided by the maximum possible sum and turned into a percentage to
create the index. The results indicate that societies with better collective insurance will have
more members in collective action. To create the trust index, the household was asked if a
majority of the village members could be trusted. Then the respondents were given choices of
various groups of people

41

and asked if they trusted them. These are the variables that were

41

The groups given were family members, church leaders, group members, fellow farmers,
political leaders, and friends. Trusting all the groups would yield a value of 100%, whereas
trusting no one gets a value of 0%.
114

summed up to create the index. The trust index was 59 for PMG members and 61 for nonmembers.
From the cross tabulations, it appears that within the sample in this study, more diverse
groups are more likely to be members of the PMG. Diversity in this study stands for diverse
religious affiliation, ethnicity, age group, and wealth endowment. We had seven categories of
diversity. To have the least diversity, a group’s members would have to be drawn from the
narrowest category. Hence, such a group would have a value of 1/7, whereas groups with
greatest diversity have an index value of 1. This value was then converted to a percentage.
The expectation was that the more friends a household had in PMGs before joining a
PMG, the higher the likelihood that the household would join a PMG. On average, members had
nine friends involved in PMGs while nonmembers had seven. From the table, we find that
having previous PMG experience does have a negative impact in being a member of a group.
However, this could be affected by the decisions on program placement made by external agents
promoting group formation. If the external agent decided to promote group formation in a new
area where it had no previous operations, then this decision would lead to a negative correlation
between previous NGO experience and joining a group. Since we do not have enough
information on exact decisions by NGOs working in the study areas on program placements, we
cannot discount this scenario.
The average land ownership for the members and non-members is very similar, with a
difference of about 0.16 acres more for non-members (see table 3-2). However, the total cropped
area is higher for the members; with 65% more cropped land than for non-members. The
differences can be explained by the group members being more dedicated to farming activities

115

and hence putting more land into cropping either through hiring or utilizing more of their land
into production. However, since the PMGs were organized around crops, it is conceivable that
farmers who had a higher percentage of land under pasture would be less likely to participate.

Table 3-2: Asset ownership effect on PMG membership
PMG member
No
(Mean)
7.37
7.38
206,755.29

Yes
(Mean)
7.21
12.17
98,879.77

Total owned in acres
Total acres cropped
Total asset value (Kshs)
Total household farm income
86,404.48 109,575.39
(Kshs)
69,270.40 229,170.61
Transfer payments (Kshs)
* Significant at 10%; ** significant at 5%; *** significant at 1%

Anova (F
test)

0.144
74.903***
25.775***
4.887**
5.024**

%
Δ
-2
65
-109
27
231

The mean level of assets is higher for non-members than for members, with a decrease of
about 109% in the value of assets owned by members compared to non-members. Development
agencies in the rural areas work with the poor members of the society. So once these groups are
formed, the poor are encouraged to join as way of accessing services. Group members have a
higher mean value of transfer payments from other sources than do the non-members, with a
difference of about 231%.
We also tested the differences between the two groups in terms of the reported
transaction and marketing cost variables as well as some other variables that would indicate the
ease of access to markets (Table 3-3).

116

Table 3-3: Relation of transaction and other costs in relation to PMG membership
PMG member
No
Yes
(Mean)
(Mean)
10.50
13.01
Distance to point of sale (time)
18.52
21.38
Transport cost to the market
1.45
1.56
Sales tax
20.00
17.73
Distance to the point of sale (km)
* Significant at 10%; ** significant at 5%; *** significant at 1%

Chi2
179.978***
251.042***
265.291***
743.746***

From table 3-3 above, it can be seen that households that had a higher measure of
transaction cost or higher cost of marketing were more likely to be involved in PMGs for all
measures, with the exception of the distance to the market in kilometers. As expected from
transaction cost theory, the households with higher transaction costs were more likely to choose
an organization form that minimized those costs. The reversal in the result concerning distance to
market as measured in minutes and kilometers could be due to the fact that it is very difficult for
individuals in the rural areas to accurately estimate distances in length measures, whereas they
can measure in terms of time spent traveling more accurately. An alternative explanation could
be that members live in areas with worse roads than nonmembers do, hence the time to travel a
given distance is greater than in areas with good roads.
Farmers’ sources of information for collective action and price information are expected
to have a significant impact on households’ decisions about whether to join groups. If a
household gets its collective action information from the media (newspaper, TV and radio) or
from local traders, then it is less likely to participate in a PMG (see table 3-4). Households who
get their price and CA information from all other sources of information seem to be more likely
to be found involved in some form of collective action. We discuss this further in relation to
mobile phones as a source of information in the participation model.
117

Table 3-4: Sources of information and its influence on collective action
PMG member
No
Yes
(%)
(%)
51.5
48.5

Media is source of CA information

Chi2
0.232

99.5

0.5

214.176***

36.1

63.9

42.864***

32.2

67.8

53.305***

49.8

50.2

6.056**

44.9

55.1

10.164**

12.5
87.5
Mobile phone is the source of price
information
35.2
64.8
NGOs are the source of price
information
27.2
72.8
Groups are the source of price
information
* Significant at 10%; ** significant at 5%; *** significant at 1%

5.532**

Traders are the source of CA
information
Groups are the sources of CA
information
NGOs are the sources of CA
information
Extension service is the source of CA
information
Media is the source of price information

12.976***
51.478***

3.6.1 Participation model
The first model estimated was a PMG participation model, where membership in a
producer-marketing group was the dependent variable. The aim of this model was to answer the
first research question regarding the factors that influence a household’s decision to join a PMG.
The probit model results show most of the variables have the expected signs (see table 3-5).
Among households’ characteristics, age has a positive sign but age squared has negative
sign and a very small magnitude. Both coefficients are significant at the 1% level. Therefore,
there is evidence for an inverted-U relationship between age and participation in the PMGs; the
turning point is at 67 years.

118

Table 3-5: Factors that affect the decision to join a PMG
Membership in a PMG is the dependent
variable
VARIABLES
Probit (marginal
effect)
Individual characteristics
Age

0.04***
(4.14)

Age2

-0.003
(-0.07)
0.01
(1.08)
0.004
(0.51)
-0.08
(-1.63)

Education
No. HH members
Formal job
Location characteristics
Village population
Log rainfall
Household wealth
Livestock value

Asset value
Land cultivated
Other income
Social networks
Previous CA experience
NGO help before CA
No. of friend CA
Diversity index

Probit
(marginal
effect)

Marketing and transaction cost
Transport cost per 90 kg 0.0003
bag
(0.40)
Distant to market
0.001
(minutes)
(1.43)
ICT access
Access to mobile phone -0.12***
(-2.62)
Sources of price information
Farmers
-0.03
(-0.45)
Government agents
0.05
(0.43)
Traders
-0.01
(-0.15)
Mass media
-0.08
(-1.34)
NGOs
0.02
(0.21)
Access to CA
information
Traders
-0.26**
(-2.03)
Government agents
0.06
(1.23)
NGOs
0.15**
(2.21)
Time trend
year
1.20***
(10.24)
Spatial Location
Siaya
0.30***
(4.15)
Teso
0.13*
(1.90)
Mbeere
0.81***

-0.0003***
(-3.95)

Male

VARIABLES

-0.002***
(-3.82)
0.33***
(7.24)
1.66e-06**
(2.26)
0.00
(0.83)
-0.00
(-1.05)
-0.00
(-0.27)
0.03
(0.58)
0.10*
(1.68)
-0.00
(-1.46)
0.01***
(15.24)
119

Table 3-5 cont’d
Solidarity index

0.01***
(7.19)
Trust index
0.00
(0.27)
Observations
1658
Pseudo R2
0.7495
Wald chi2(34)
511.56
Prob > chi2
0.0000
(Std. Err. adjusted for 1013 clusters)
Robust t statistics in parentheses
*** p<0.01, ** p<0.05, * p<0.1

Makueni

(6.61)
0.79***
(8.47)

The implication is that up until the age of 67, as the individuals get older, they are more
attracted to the PMGs. Older people have more free time to get involved in the group activities.
With time, older members of the society are able to appreciate the wisdom of working together
and have less energy to ferry products to the market and wait for buyers. However, after 67
years, they may have less energy to travel to groups meetings and other activities and will reduce
their participation. This is similar to finding of farmer’s participation in dairy cooperatives in
Ethiopia (Francesconi and Ruben, 2007).
On location characteristics, an increase in the population of the village decreases the
probability of a household joining a PMG. The coefficient on village population is negative and
significant at the 1% level. A village with a higher population decreases the transaction cost for
both the trader and farmer in accessing a trading partner. Therefore, there will be greater
competition in such a village, decreasing the need for an organization for marketing their goods.
Hence, the coefficient for village population has the expected negative sign.
The estimated model also controlled for the rainfall amounts and time trend. The two
variables are positive and significant at the 1% level. These variables capture other factors we

120

could not be controlled for in this model that affect the individual farmer’s decision to be a part
of a group. It could also be interpreted that PMG membership has been growing over time for
reasons not captured entirely by the other variables in the model. Farmers are also more likely to
join PMG in areas with more rainfall because they have more produce to sell and require group
services.
The livestock value coefficient has a positive sign and is significant at the 5% level,
although the actual magnitude is small. To increase crop production, a farmer requires capital
resources to access inputs and usually needs animal power for land preparations and weeding.
Therefore, this result is as expected. Increased livestock value might imply that a farmer will
produce more crops and therefore joins a PMG to help in marketing the products.
Previous help by a development agency increases the probability of a household joining a
group. The coefficient is positive but barely significant at the 10% level. Many of the
development organizations encourage individual producers to join some form of collective
action. Being involved in collective action makes it easier for these agencies to improve their
own effectiveness. It is more cost-effective visiting a group than individual clients. Also, some of
these groups were formed as a result of encouragements by agencies working in the regions. The
linkages between farmers and external groups may be essential in the initial stages on group
formation and the group’s survival (Gregorio, et al., 2008, Scherr, et al., 2001).
Farmers in areas with a higher diversity in terms of ethnicity, religious, and other aspects
had a higher probability to be in a PMG. The coefficient of the diversity index is positive and
significant at the 1% level. As discussed earlier, in the literature, diversity’s impact is
ambiguous, but in this case, a more diverse farming community in the PMG operating areas

121

increased participation. Usually, social heterogeneity is hypothesized to have a negative effect on
cooperation because different social norms may make creating and enforcing decisions more
costly. However, the data from this sample supports those studies that have found that diversity
fosters collective action (Bandura and Wood, 1989, Goldenberg and Mazursky, 2002, Lowndes
and Skelcher, 1998).
The solidarity index was created to assess a household’s stance on social responsibility
towards its immediate neighbors in the village. The index was created from two questions. One
sought to know if households would provide assistance to their fellow villagers in the event of a
drought or scarcity with basic needs like seeds or food. The second question reversed the
previous question by asking if they would receive the same basic needs under similar
circumstances. The coefficient of the solidarity index was positive and significant at the 1%
level. Hence, the greater the solidarity a household feels towards its neighbors, the higher the
probability of joining a PMG. In the rural areas, a farmer faces a variety of risks that are not
formally mitigated through the market. The alternative is to have reciprocity that hinges on moral
obligations to meet the basic needs of the groups. Households that are already familiar with
communal reciprocity therefore have a higher probability of joining collective action efforts.
Information communication technology aids the flow of information, especially in remote
rural areas in developing countries through cell phones. In the last few years, Kenya has
witnessed an explosion of access to mobile phone use even among the rural farmers in spite of
lack of electricity. This variable captures access to a cell phone by ownership or access through
friends or relatives. Farmers are able to communicate with relatives, and buyers are able to get
valuable market information and negotiate produce sales through cell phones. Hence, it is not
surprising that access to a mobile phone had a negative effect on membership in a PMG. The
122

coefficient for this variable had a negative sign and was significant at the 1% level. Individual
members of a PMG are able to access market information, other agricultural information, and
negotiation with traders. The mobile phone gives the individual farmer a wider network of
information sources, thereby limiting the range of benefits they enjoy from group membership.
PMGs provide an avenue for individual farmers to interact, socialize, and exchange ideas. A cell
phone provides a similar service without joining any group.
Sources of information about prices were expected to affect farmers’ decisions on joining
a PMG. However, none was significant in the estimated model. However, access to CA
information from NGOs is positive and significant at the 5% level. As expected, CA information
from traders has a negative impact. The coefficient has a negative sign and significant at the 5%
level. CA information from NGOs would encourage individual farmers to join groups because
they normally promote the idea of working in groups. Traders, on the other hand, give negative
information about the groups and discourage their customers from joining them.
We also included location dummies that control for unobserved spatial heterogeneity in
location factors. Homabay district was excluded. It was excluded because there were fewer
promotion activities in this district compared to all others and we expected that to be reflected in
the analysis. All the dummies are positive and significant compared to Homabay. With the
exception of Teso, all other coefficient of the dummies are significant at 1% level. They could
also be capturing the amount of promotion activities the various development partners had in the
different districts. We know from the field survey that there were fewer promotion activities to
encourage group participation in Teso, and this is captured by the coefficient Teso being positive
but is significant at only the 10% level. The differences in magnitude of the dummies across the

123

different zones suggest that that even among these zones there were differences in factors
influencing the choice of participation that the model may not have captured.
3.6.2 Group patronage
From the field interviews, it was apparent that not all group members sold their produce
through the organizations. And even those who sold through the group often made only a small
fraction of their total sales through them.
Hence, it was important to isolate those factors that have the biggest effects on patronage.
The dependent variable in this model was the percentage of sales through the group out of the
total crop sales.
Table 3-6: Factors that influence members’ level of group patronage
(1)
(2)
Dependent variable
PMG member
% to group sales†
VARIABLES
(Selection equation) Heckman
Individual characteristics
Age
0.101***
-1.254*
(3.572)
(-1.956)
age2
-0.001***
0.012*
(-3.459)
(1.928)
gender
0.084
3.879
(0.438)
(1.186)
educate
0.021
0.581
(1.177)
(1.366)
No. of HH members
0.016
0.022
(0.733)
(0.046)
Formal job
-0.228
-1.635
(-1.516)
(-0.516)
Married
0.038
(0.701)
Location characteristics
Village population
-0.006***
(-3.981)
Log rainfall
0.976***
(7.359)

124

Table 3-6 cont’d
Household wealth
Asset value
Land cultivated
Other income
Social networks
Previous CA experience
Had NGO before PMG
No. of friends in PMG
Diversity index
Solidarity index
Trust index

0.000
(0.759)
-0.004
(-0.562)
-0.000
(-0.083)

-0.000
(-0.585)
0.295*
(1.800)

0.082
(0.559)
0.241
(1.452)
-0.005
(-1.078)
0.034***
(15.612)
0.019***
(7.079)
0.001
(0.209)

0.833
(0.273)
10.759***
(3.285)
-0.091
(-1.118)
-0.314***
(-3.878)
-0.076
(-1.266)
-0.006
(-0.115)

Marketing and transaction cost
Transport cost per 90 kg bag
0.001
(0.540)
Market distance (minutes)
0.003
(0.879)
ICT Access
Access to a mobile phone
-0.313**
(-2.346)
Access to price information
Farmers
-0.090
(-0.456)
Government agents
0.133
(0.484)
Traders
-0.033
(-0.175)
Mass media
-0.261
(-1.294)
NGOs
0.075
(0.278)
Access to CA information
Traders
-0.975*
(-1.845)
Government agents
0.190
(1.383)

0.052*
(1.809)
-0.044
(-0.614)
3.184
(1.051)
0.976
(0.230)
-16.924***
(-2.853)
-5.041
(-1.307)
-2.617
(-0.631)
-7.862
(-1.402)

125

Table 3-6 cont’d
NGOs
Time trend
year
Location
Siaya
Teso
Mbeere
Makueni

0.428**
(2.377)
3.422***
(11.252)

12.909***
(3.670)

0.784***
(2.936)
0.370
(1.237)
2.723***
(6.697)
2.543***
(6.893)

-41.529***
(-6.545)
-43.938***
(-6.730)
-41.093***
(-6.091)
-43.793***
(-6.444)
-12.792**
(-2.409)
93.258***
(3.919)
1660
894
766
218.17
0.0000

IMR (lambda)
Constant

-17.144***
(-10.185)
1660

Observations
Censored observations
Uncensored observations
Wald chi2(27)
Prob > chi2
t statistics in parentheses
*** p<0.01, ** p<0.05, * p<0.1
†Percentage sales through the PMG are the dependent variable in model 2.
We estimated a Heckman sample selection model that takes into account the fact that
individuals self-select to join PMGs. The results of the estimation are shown in table 3-6. The
first model is the selection model and the second model is the intensity or patronage model. All
discussions in this section refer to model 2, which is the second stage of the Heckman two-part
self-selection model accounting for self-selection in the PMGs.
The lambda coefficient in the model is negative and significant at the 10% level.
Therefore, we reject the null hypothesis that there no evidence for self-selection. Hence, the
Heckman self-selection model was the right estimating procedure for group patronage.

126

The results show that most of the individual characteristics are not important in
determining patronage intensity. Age in the selection model shows that that as household heads
increase in age they are likely to join PMG at a decreasing rate as expected. The age and age
squared coefficient in the intensity model are significant. The age coefficient is negative and
significant at the 10% level. And the age-squared coefficient is positive and significant at 10%
level. Hence, there is a U relationship between age and level of sales through the group.
Households with older household heads have fewer sales through the group. This can be
explained by the past failures of marketing collective efforts. The middle aged household’s may
be joining the PMG for social reasons. Age in the intensity model has a turning point at age 53
years. This could be the point where the individuals discount the risk associated with group
marketing for the convenience it provides of shorter distances to the market and less waiting time
to sell the product.
As expected from the participation model, the location characteristics are significant and
have same sign as we get in the selection model 1 in the estimation. The reasons are the same as
mentioned above in table 3.5.
Moving to household wealth indicators, the coefficient on land cultivated is not
significant in the selection model while positive and significant at the 10% level in the group
patronage model. Group patronage increases by 0.3% per every increase of one acre of land
owned by a household. This reflects the fact that more cultivation in likely to increase crop
production. The reason more production might increase percentage sales through the PMG is not
clear. But this phenomenon might be driven by the fact that as produce volume to sell increases,
it becomes worthwhile to sustain the transaction costs of being in the PMG in order to get
slightly more per kg sold. If one only had a few kilograms to sell, the transaction costs might
127

outweigh the potential gain. The magnitude of increase of 0.3% is very small. In this model, we
omitted the livestock variable because of high collinearity of that variable with the other
variables in the model.
Receiving NGO help by development agents to individual households did not encourage
self-selection to join but once they joined the households sold a higher percentage of their
produce through the group. The coefficient on NGO help is positive and significant at the 1%
level. If a household had any help from a development agent, it increased its sales through the
group by 11%, which is a very sizable amount. The implication is that development agents build
trust with farmers. Many development agencies tend to work better in a group setting, as
discussed earlier. Hence, it is plausible that households that work with them trust and value
group efforts due to the services they receive. However, the individual households could have
already been members of the PMG and that is what motivated the NGO assistance in the first
place. This finding goes contrary to sociology literature that finds that once a development agent
is involved in a group, the group tends to disintegrate. It also shows that the development work
by the organizations in study areas is viewed positively by the farmers.
Whereas diversity of individuals in an area increases the probability of a household
joining a PMG, it decreases the amount of produce sold through the group. The diversity
coefficient is negative and significant at the 1% level in the patronage model. Diversity therefore
seems to hinder the actual operation of the already existing group. Maybe it makes it difficult for
group members to impose sanctions on free riding, decreasing trust in the ability of a group to
market its produce (Miguel and Gugerty, 2004). Therefore, as in other studies discussed earlier,
this study concludes that group heterogeneity has ambiguous effects on the PMGs. In a

128

somewhat similar manner, an increase in the solidarity index increases the self-selection into
groups, but it has no effect on the patronage intensity in the group.
Higher transport cost of produce would be expected to encourage households to join
groups so that they can have a bargaining power or undertake joint marketing. The coefficient on
transport cost is not significant in the selection model but positive and significant at the 10%
level in the intensity model. An increase in one shilling of the transport cost of a 90 kg bag of
produce to the market increases the amount of sales through the PMG by 0.05%. Though this
amount is small in magnitude, the results are as expected, since we posited that high marketing
cost would increase PMG patronage. The variable also captures the difficulties of access to the
market for individual farmers located in remote areas. The implications are that farmers far
removed from the market physically will use the PMGs more.
Access to price information through all the channels explored does not have an impact on
self-selection, but access through government agents has a negative association with the
percentage of products sold through the group. We would expect that if government extension
agents promote group participation and patronage, the coefficient would have a positive
correlation with patronage. The results show that access to more land for cultivation increases
patronage intensity. There is a likelihood that the poorer members of the groups have smaller
percentage to market anyway and since government agents work with poorer members of the
society, this is being captured in the variable “access to information through government agents.”
However, this needs to be investigated further.
The time trend coefficient is positive and significant at the 1% level. It shows that with
passage of time, group members increase their patronage. This result is important, since at the

129

beginning of the life of a PMG, there might be fewer sales through the group, but as the group
builds trust, the patronage increases. Therefore, it is important to support these groups in their
formative years.
3.6.3 Collective action intensity
The sampled rural households belonged to an average of 2 social groups, and the
household belonging to the most groups was a member of 9 groups. Different groups serving the
same area usually provide different services. Therefore, for a household to access these services,
it is forced to join a variety of CA organizations. During the survey, we found that groups were
performing specialized narrow functions. Different development agencies encourage formation
of groups that are in line with the agencies’ interests, leading farmers to belong to multiple
groups. Table 3-7 shows the results of the estimated Poisson model that examined the factors
influencing multiple memberships. The second column shows the coefficients or probability of
joining multiple PMGs and the third column shows the marginal effects or the magnitude of the
probability of joining more PMGs.
Therefore for the second column we are interested in the signs or the direction of the
coefficient whereas the magnitude of each of the coefficient is shown in the third column.
The results in table 3-7 show that households with a male head of the family join more
groups than do female-headed households. The coefficient on the gender variable is positive and
significant at the 1% level. Our conjecture here is that men in rural areas have more time
compared to women to dedicate to activities outside of household labor requirements. Women
usually bear a disproportionately bigger household labor burden. They are involved in farming
activities as well as house chores, while men are involved in farming activities only.

130

Table 3-7: Factors that affect the number of collective action groups joined
Dependent variable is the number of groups to which the household head belongs
Variables
(1)
(2)
Poisson
Poisson marginal effects
Individual Characteristics
Age
0.011
0.015
(1.20)
(1.19)
Age2
-0.000
-0.000
(1.11)
(1.10)
HH head is male
0.259***
0.333***
(4.98)
(4.65)
Education
0.021**
0.028**
(1.97)
(1.96)
# of HH members
-0.001
-0.001
(0.09)
(0.09)
Formal job
-0.053
-0.071
(1.10)
(1.08)
Household wealth
Total asset value
-0.000
-0.000
(1.53)
(1.49)
Land farmed
-0.004***
-0.006***
(2.64)
(2.61)
Social networks
Had prior CA experience
0.037
0.049
(0.69)
(0.69)
NGO help before CA
0.031
0.042
(0.49)
(0.49)
No. of friends in CA
0.001
0.002
(1.08)
(1.08)
Diversity index
0.002***
0.003***
(2.79)
(2.73)
HH community trust index
0.002**
0.003**
(2.30)
(2.28)
Solidarity index
-0.001
-0.002
(1.39)
(1.38)
Price volatility
Price St.Dev
0.007**
0.009**
(2.20)
(2.25)
Marketing and Transaction cost
Selling time (min)
-0.002***
-0.003***
(2.94)
(2.82)
Transport cost
0.001*
0.001*
(1.78)
(1.76)
Market distant (Min)
0.000
0.000
(0.03)
(0.03)

131

Table 3-7 cont’d
Access to credit
Has credit access
Access to ICT
Mobile phone
Sources of CA information
Traders
Government agents
NGOs
Time trend
year of the interview
Spatial location
Siaya
Teso
Mbeere
Makueni
Constant
Observations
Wald chi2(29) Log Prob >
chi2

0.030
(0.74)

0.041
(0.74)

-0.087*
(1.70)

-0.119*
(1.65)

-2.563***
(4.40)
0.110***
(2.81)
0.272***
(5.19)

-1.742***
(16.45)
0.150***
(2.66)
0.404***
(4.40)

0.603***
(19.83)

0.810***
(9.32)

0.364***
(5.51)
0.276***
(3.98)
0.952***
(10.48)
1.460***
(17.74)
-1.916***
(6.83)
1660
2698.80
0.0000

0.570***
(4.48)
0.416***
(3.48)
1.463***
(6.70)
2.842***
(8.76)

1660

pseudolikelihood
-2584.0778
(Std. Err. adjusted for 863 clusters)
Robust z statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
Also, households with a male head are likely to have both spouses in the home, whereas
households with a female head are likely to have only the woman as the sole family head, with
no male spouse for various reasons. This gives women less time to divide among many groups
and core farm-labor requirements.
More education in the household gives the members of the household an ability to
recognize the benefits of group membership and join many efforts. Educated members of the
132

society in developing countries tend to migrate to the urban areas in search of better job
opportunities. Hence, there are few well-educated people in the rural areas. Therefore, educated
members of the community are usually in leadership positions of multiple collective action
efforts. The education coefficient is positive and significant at the 5% level.
A household with more land under cultivation requires more labor devoted to farm and
will devote less time resources in CA efforts. The coefficient on land farmed is negative and
significant at the 5% level. Members of the society, who have more land cultivated, are usually
the well-to-do and participate in fewer groups. These household heads are usually timeconstrained and their opportunity cost of time is higher than that of less well-to-do households.
They will therefore participate in groups that give them higher marginal utility. Another possible
explanation is that larger farms may be able to get many of the scale economies individually that
smaller farms can only capture acting collectively.
As in the model of group participation, diverse societies encourage more cooperation and
households join more groups. The coefficient on diversity index is positive and significant at the
1% level. Similarly, a household that exhibits higher trust of the community surrounding it will
increase the number of groups it joins. The coefficient is positive and significant at the 5% level.
Therefore, we find evidence that strong social networks improve group participation intensity.
Few other studies have included such variables in their analysis, although social networks are
considered important in general in the CA literature.
An increase in price volatility increases the number of groups a household joins. This
variable could be capturing the uncertainty in the benefit flow of the existing marketing
channels, increasing the need to diversify a household portfolio of groups. Ideally, we should

133

have added the risks associated with the new CA portfolio to the model. However, we did not
have that information. The results shows that an increased annual price risk increases the number
of CA a household invest into to help manage the risk. The coefficient is positive and significant
at the 5% level.
We also tested in the impact of marketing costs and transaction costs to group
participation intensity. The model shows that the transaction cost captured by the selling time
coefficient has a negative correlation with the number of group a household joins. The
coefficient is negative and significant at the 1% level. A household that takes longer selling its
product in the market may not trust the group’s marketing abilities and is ready to invest more
time selling its product. On the other hand, it could also be capturing the presence of higher
transaction costs in the marketing process.
A higher transport cost for the product increases the household participation in more
groups. The coefficient on transport cost is positive and significant at the 10% level. Households
in more remote areas may have fewer services than the ones in more accessible areas. Therefore,
a household in these areas will find the various services provided by groups attractive and join
more groups.
Access to a mobile phone reduces the number of groups a household joins. The
coefficient is negative and significant at the 10% level. Just like the case of joining a group, it
seems that better communication with other individuals outside the community reduces the
benefit an individual gets from cooperating with the community in a village.
The coefficients of variables capturing sources of CA information are all significant.
Households that get information from traders join fewer groups. Farmers’ reliance on traders
134

could indicate their mistrust of CA efforts and hence, they join fewer groups. An alternative
explanation could be that traders are in direct competition with CA efforts and thus do not pass
favorable information to farmers, thereby discouraging potential members. The coefficients on
government and NGOs are both positive and significant at the 1% levels. This indicates that
government and other development agents have cooperated in encouraging farmers to join CA
efforts. This underscores the role that information plays in the decision on collective action
intensity.
The time trend coefficient captures the changes in the perception of participation in a
group benefits over time. The variable is positive and significant at the 1% level. It shows that
with time groups in these rural areas may be providing important service that households value.
Therefore, households increase their intensity of participating.
3.7 Conclusions
Social networks have a big impact on the decision by producers to join a producer
marketing group, patronize it, and on the number of collective action groups a farmer joins. The
implication is that before collective effort is advocated in any area, it is important to make sure
that the social dynamic component is well addressed through community mobilization and
education. For an effort to succeed, greater effort should be put into training the community as a
whole on group dynamics and social responsibility of the members. Therefore, it is necessary to
build trust to avoid future opportunistic behavior of others, which often leads to failure.
Previous assistance by a development agent before joining a PMG improves the
probability of a household joining a group and also increases the level of patronage. Earlier
research had shown that once an external agent is involved in collective action, groups are likely
to fail. Our findings are not consistent with these previous results. However, we did not ask the
135

households if they were involved with self-started groups or those that were encouraged by
external agents. The question asked sought to know if the household had ever received assistance
from a development organization before deciding to join a PMG. All we can say is that previous
help from an external agent encourages an individual to join a group.
As expected, an area with high population is likely to have many traders and the market
to be more competitive than a sparsely populated region. Hence, when encouraging the
formation of groups for the purpose of produce marketing, it is important to take into
consideration market conditions. And if a market is deemed to be competitive, a PMG can only
play the role of a competitive yardstick.
From transaction cost theory, it was expected that TC and marketing costs would play an
important role in the decision to join a PMG and any other CA efforts. We find that that transport
costs have no impact on the participation decision, but they play a part in patronage. Once
farmers join a PMG, they are at a better position to choose between the two channels available.
One explanation is that PMG generates slightly higher prices per kg that make it worthwhile to
patronize the PMG if one has sufficient volume. For those with small volumes, the transaction
costs of using or joining the PMG may exceed the gain.
Information sources play a vital role in a producer’s decision process. Therefore, a
strengthened information delivery system through existing groups, external agents and
government agencies could lead to more households to join and patronize PMGs in rural areas.
Access to a cell phone gives a farmer access to communication with the outside world
beyond the normal village interaction. This increased coordination with others outside the village
may be the reason why access to a mobile phone decreases the probability of joining a group.
136

The study by Aker (2008), discussed earlier, found evidence that mobile phones improve traders’
market coordination, decreasing their search costs. The fact that farmers can coordinate with
traders directly might reduce the need for a PMG as a coordinating body between the farmers
and traders. Though we may not deduce too much from this study since the phenomenon is new,
more research needs to be done on this topic to understand this phenomenon. As new
complementary services are added to ICT technology that specifically target local communities,
it might improve local interactions.

137

Chapter 4 : Can membership and participation in Producer Marketing Groups
contribute to improved smallholder producers’ welfare? Evidence from semi-arid areas in
Kenya
4.0 Background
The vast majority of rural inhabitants in developing countries still depend, directly or
indirectly, on smallholder agricultural production for sustenance and income generation (WorldBank, 2008). Many rural smallholder farmers lack effective organizations, which deprives them
of the power to influence local decisions affecting their lives, to negotiate better terms of trade,
to interact on equal terms with powerful market intermediaries, and to make governmental and
nongovernmental organizations (NGOs) accountable to them (Gulati, et al., 2007, Miehlbradt
and McVay, 2005). In policy research and development-driven agendas, there has been a
renewed interest in agriculture and its ability to stimulate pro-poor growth but under a new
institutional framework (Hellin, et al., 2009, World-Bank, 2008). An institutional approach to
improve farmers’ welfare needs to have as key objectives increasing access to assets and modern
technology in terms of inputs, information, financial services, and markets (Gonzalez, 2008). To
achieve these goals, collective action has been promoted in semi-arid areas in Kenya with the
purpose of improving producers’ countervailing power, reducing transaction costs, and reaping
the benefit of economies of scale and scope. Small farmers’ organizations are some of the
institutional platforms for collective action that can mitigate these inherent problems (Shiferaw,
et al., 2008). When successful, producer groups are expected to increase their members’ farm
productivity and income through access to better inputs and higher prices or low marketing costs,
thereby improving their overall returns to production (Kydd and Dorward, 2003, Kydd, et al.,
2004).

138

However, organizing smallholder producers into a viable and beneficial business
organization is not a small task. There are high transaction costs involved that sometimes
outweigh the potential benefits (Stockbridge, et al., 2003, Stringfellow, et al., 1997). Often due to
lack of essential preconditions for successful collective action, such as altruistic leadership, basic
education, capable management, entrepreneurial skills, social cohesion, and financial capacity,
many organizations fail before they are able to achieve their goals (Berdegue, 2002, Hulme and
Shepherd, 2003, Key and Runsten, 1999, Masakure and Henson, 2005, Pingali, et al., 2005,
Stringfellow, et al., 1997). Even with the potential pitfall of failure, others have argued that
farmers’ organizations should be promoted because it is the best avenue for assisting smallholder
farmers in accessing necessary services (Collion and Rondot, 2001, Ephraim Chirwa, et al.,
2005, Hellin, et al., 2009, Omamo, 2005, Peacock, et al., 2004, World-Bank, 2008). Therefore,
there is still an ongoing debate on whether farmers’ organizations provide enough tangible
benefits to the producers to warrant their support and promotion by government and
development agencies.
4.1 Research questions
This chapter seeks to investigate if the farmers in semi-arid regions of Kenya who
participate in producer marketing groups (PMGs) derive enhanced benefits compared to nonparticipants in the same area. The research questions we hope to answer are:
i) Do members of a PMG face a lower price risk?
ii) Does being a member of a PMG improve a producer’s crop income?
The core economic incentive for joining a producer marketing group is enhanced income,
crop production and/ or reduced variability in output prices.

139

The stated objectives of most of the groups in the study area are to enhance members’
welfare, obtain better prices for their produce, improve access to inputs, develop business skills,
share knowledge, and transform PMGs into business entities with members able to eventually
own shares (Shiferaw, et al., 2005). Farmers will join collective action if they believe that their
incomes will be enhanced or their prices will become more stable through collective bargaining
(Ruben, 1997). Collective action in the form of an active PMG can protect farmers from
temporary price fluctuations by storing the produce during the period of low prices and selling
when prices are higher and/or negotiating a less variable price with traders.
However, a farmer initially might derive little additional benefit from selling her produce
through the PMG, but she may value the access to the other, non-core assets or services provided
by the organization, which improve her overall welfare through other investments (Berdegue,
2001). Successful groups do provide an opportunity for a farmer to access improved seeds and
other inputs (Shiferaw, et al., 2008).
4.2 Literature review
Looking at the literature on farmers’ membership in producer organizations and their
welfare impact, we find the effect on various measures of farmers’ welfare to be ambiguous. The
results depend on the product of interest and sometimes-overarching market conditions.
Some studies show that the net effect of farmer membership in a farmer cooperative on
total household income tends to be neutral (Berdegue, et al., 2001). In his thesis, Berdegue
(2001), using a Heckman self-selection model, found that farmers’ participation in a cooperative
only has a significantly positive impact on members’ farm profit margins when the cooperative
operates in markets with high transaction costs, such as the dairy sector. This is consistent with
what Staal (1997) and Nyoro et al. (2005) found in Kenya. Berdegue (2001) argues that a
140

producer organization may not benefit its members with increased returns if it operates in
markets with low transaction costs for undifferentiated commodities like wheat or potatoes,
which could indicate a more competitive market. Knack and Keefer (1997) find no significant
relationship between group membership and economic performance in their cross-country study.
In their study of collective action and social networks in the Philippines, Godquin and
Quisumbing (2006), using a 2SLS method, find no evidence of positive returns to group
membership in terms of increased per capita expenditures. They explain that the lack of impact
could have been due to the choice of instruments used.
Another strand of literature has found positive association between measures of group
membership and household welfare (typically measured by household expenditures) (Grootaert,
1999, Grootaert, et al., 1999, Haddad and Maluccio, 2002, Maluccio, et al., 2000, Narayan and
Pritchett, 1999). In their case studies of dairy cooperatives

42

in Ethiopia and Kenya, Holloway

and Stall found that cooperatives enhance rural-urban linkages and increase market participation
by the rural producers (Holloway, 2000, Staal, 1997), which could lead to improved welfare.
Chowdhury et al. (2005) found that new market governance like cooperatives reduce transfer
costs and consequently enhance rural-urban linkages, enhancing farmers’ income. A case study
of pig producers in Vietnam illustrates that collective action approaches such as cooperatives or
farmer groups can facilitate the reduction of transaction costs in production and marketing to
improve both productivity and improved incomes (Lapar, et al., 2006). A study in Nicaragua that
investigated why farmers continued to stay in cooperatives after the change of government
system from socialism to a more market-oriented economy found that farmers who were

42

The dairy industry is more likely to be characterized by specific assets than the grain sector we
are analyzing. So some differences in the results are likely.
141

members of cooperatives benefited from access to services and utilized their social capital to
access credit that in return improved their welfare (Ruben and Lerman, 2005). A study of
collective action efforts in the Kenyan central highlands found that the majority of the
households in the survey were involved in some form of collective action and believed that
group effort improved their welfare (Kariuki and Place, 2005). However, this study used
descriptive statistics without any statistical tests to measure welfare gain and individual
perceptions. The study gives a good starting point, but one cannot assess how statistically
significant its findings were. Recent work from the World Bank group on tobacco clubs in
Malawi (Negri and Porto, 2008) show that membership in these clubs increase farm production
by 40-74% per acre. There is also some recent work done in Kenya using a nationwide panel
data collected in 2000 and 2004 by Tegemeo Institute of Egerton University. This study did not
target a particular crop but measured the total rural households’ welfare. Correlating the means
of members and non-members on several performance dimensions, the study finds that
households that had joined farmers’ organizations performed better in terms of agricultural
production, accumulation of assets and poverty alleviation (Ngugi and Kariuki, 2009). This study
used Least Significance Difference (LSD) to test the mean differences across the groups.
However, this method has the drawback of not taking care of potential self-selection problems,
which could bias the results. Another study uses the price received by the pigeon-pea farmers in
the Eastern Kenya, based on the second round of data used in this study. This study found that
PMG members received higher prices and had better access to improved seeds (Shiferaw, et al.,
2008). Another recent study in South East Nigeria also finds that membership in a cooperative
increased a household’s expenditure (Babatunde, et al., 2008). However, this study also used

142

cross-section data and hence could not deal with potential problems due to self-selection and
other confounding factors.
Other studies disaggregate groups by their function when examining the relationship of
group membership with economic outcomes (Grootaert, 1999, Grootaert, et al., 1999, Knack and
Keefer, 1997, Maluccio, et al., 2000). Christian Grootaert finds that membership in producer and
other social groups has a positive association with household incomes, while membership in
religious groups and government/national groups has none (Grootaert, 1999). Maluccio, Haddad,
and May (2000) find that membership in financially oriented groups like Rotating Savings and
Credit Associations (ROSCAs) has a positive association with household per capita expenditure.
From the literature, we can conclude that the impact of participating in collective action
among rural poor has mixed results in terms of welfare improvement. This chapter uses panel
data that was collected in three rounds. The first round was the survey conducted before PMGs
were formed; then there are two rounds of surveys conducted after the PMGs’ formation.
Therefore, unlike most of the studies reviewed that use cross-section data, we can explicitly
compare the effect of treatment with a counterfactual. We therefore hope to add to the
understanding of the impact of collective action on households’ welfare at a time when different
forms of non-market based initiatives are being promoted in developing countries to alleviate
poverty. Also unlike previous studies that were focused on the impact on farmers’ welfare of
membership in cooperatives, this paper focuses on PMGs, which are different from cooperatives.
As argued in the chapters 1 and 2, PMGs are a smaller form of collective action from
cooperatives (see chapter 1). This study is focused on semi-arid areas of Kenya that are usually
food-deficit regions due to cyclical drought. Therefore, this study will be among few studies that
have done empirical work on farmers who usually grow crops for their own consumption, only
143

market small surplus quantities, and are mainly members of PMGs rather than cooperatives.
Previous studies that have looked at the same question have focused on areas with cooperatives
and used methods that do not address issues of self-selection and other confounding factors.
4.3 Theoretical framework
This section briefly discusses transaction cost economics (TCE), which is used as the
basis of developing the behavioral model we use as the conceptual framework. Transaction costs
economics builds on Coase’s article (1937) , “The Nature of the Firm”, which postulates that
economic activity does not occur in a frictionless environment. The main source of friction in the
economy, according to Coase, is the costs of carrying out the exchange (Benham and Benham,
1998). Coase recognized the role of transaction costs in the organization of firms and other
contracts. Transaction costs include the costs of information, negotiation, monitoring, coordination, and enforcement of contracts. He explains that firms emerge to economize on the
transaction costs of market exchange and that the boundary of a firm or the extent of vertical
integration will depend on the magnitude of the transaction costs. Williamson’s articles on the
economics of organization and contracts are an extension of Coase’s line of thinking
(Williamson, 1996, 1979). He combines the concepts of bounded rationality and opportunistic
behavior to explain contractual choice and the ownership structure of firms. Opportunistic
behavior manifests itself as adverse selection, moral hazard, and other forms of strategic
behavior. In Williamson’s framework, a trade-off has to be made between the costs of
coordination and hierarchy within an organization and the costs of transacting and forming
contracts in the market (Drugger, 1983). This trade-off will depend on the magnitude of the
transaction costs. Eggertson (1990) defines transaction costs as “the costs that arise when
individuals exchange ownership rights for economic assets and enforce their exclusive rights. ”

144

In terms of the context of this study, only the transaction costs arising for individual
agents or for basic economic units such as households are considered. This type of transaction
cost includes expenses and opportunity costs, both fixed and variable, arising from the exchange
of property rights. Transaction costs originate typically from the following activities (see
Eggertson, 1990, Jaffee and Morton, 1995): the search for information about potential
contracting parties and establishing the quality of the produce (this includes personal time, travel
expenses and communication costs); screening costs that arise from the uncertainty about the
reliability of potential suppliers or buyers and the uncertainty about the actual quality of the
goods; the uncertainty of the future state of the world (Schmid, 2004); the bargaining that is
needed to find the true position of contracting parties, in particular when prices are not
determined exogenously; making of (formal or informal) contracts—i.e., defining the obligations
of each party; monitoring of contracts; and the enforcement of the contract and collection of
damages when partners fail to observe their contractual obligations.
Transaction cost theory posits that organizational forms or governance structures that
minimize the production and transaction costs for a given activity will have a competitive
advantage. Therefore, even if we assume that producers have bounded rationality, they will
choose the governance structures that they believe give them highest utility (Williamson, 1986).
If TCE theory is correct, we expect that producers will only chose to participate in PMGs if these
organizations have a comparative advantage over dealing directly with traders and provide
higher welfare benefits to participants compared to non-participants. These benefits can arise
through access to affordable production technology, lower marketing and transaction costs,
reduced marketing risks and uncertainty, and selling at higher prices than the alternative market.

145

We next present an investment model that illustrates mechanisms that affect producer
income when a farmer joins a producer marketing group. Our argument is that joining a PMG
involves some up-front costs that the farmer is unlikely to recoup initially, but that the
investment of time and money she or he makes in the PMG is done in hopes that there will be
returns over time. For simplicity, we portray this process through use of a two-period investment
model. We use the household as the unit of analysis, ignoring intra-household allocation. A
household with specific characteristics X , labor L and a capital endowment K produces one good
(this good is a vector of all the on-farm and off-farm production),

43

with its capital and labor

using the production technology f(K, L;X) in two time periods. The household gains utility from
consumption in each period according to its utility function U(C1,C2), where Ci is consumption
in period i=1,2. The household is assumed to consume the same amount as it produces in value
terms, so consumption is equivalent to income. For simplicity, we assume that marketing of its
produce is the only way the household can increase its consumption in period two. The
household can continue marketing its produce through the spot market or through producer
groups by joining one of the PMGs. Assume that everyone can get an average net return Psm for
that portion of production sold on the spot market and that marketed through PMG will yield an
44

average net return Pca. Revenue in period 1, therefore, is Psm*qsp + Pca*qca =R

.This

equation states that the net revenue R is as a result of the quantity sold in the spot market qsp ,
multiplied by the spot market average net return Psm, plus the net return received at the PMG Pca

43

The majority of the farmers’ off-farm income among the sample is in form of transfer
payments from relatives in the urban areas.
44
Note that the total returns (R) are net returns.
146

multiplied by the quantity sold through PMG qca. If the household chooses not to participate in
the PMG, then qca=0, and the revenue will be Psm*Q=R. Revenue R is a function of Psm, and all
farmers receive this type of return from the spot market. Pca(CA) gives the net return R that is
conditional on the farmer joining the group. The differential income

45

a household gets for

participating in collective action can be represented as R=g(CA), where R is a function(g) of CA,
which is a vector of all the effort and contribution involved in PMG activities and CA=F(Z, D),
where D is the differential income earned through using the PMG, Z represents other household
characteristics that affect market choice behavior. (This assumes that inputs into P will be a
function of expected monetary and non-monetary returns).
Previous studies have indicated that marketing of produce through collective action produces
positive returns to the household through higher prices (Shiferaw, et al., 2005, Shiferaw, et al.,
2008) and/or access to new technology and lower input prices. A household has always the
opportunity to sell in the spot market. Therefore, our analysis is concerned with this potential
differential income and other service benefits the household gets if it is a member of a PMG and
sells some produce through the group. And it is this potential differential income and services the
46

household invests in the farm to get differential income in period 2 .

45

The differential income here includes extra value of production from access to improved
technology.
46
In our analysis, we assume that producers will only join collective action if there are a
potential positive benefit though higher prices, lower costs, or access to other services that
increase production within the farm.
147

In period one, the household can choose either to immediately consume any differential
income from participating in the PMG or to invest a portion of it,  : 0    1 in capital goods47
that can be used in period 2. In the second period, therefore, has a capital stock of K   R if the
household did market some of its extra production, and K if it did not. The relative output price
48

in period 2 is expected to be E(P2)

. We assume that both functions f   

and g    are

concave, continuous, and twice differentiable. Therefore, the household attempts to:
Max U (C1, C2 )
C1,C2

(4.1)

Subject to:

(1) Consumption

49

in period 1 is equal to the amount produced in that period less the amount

saved to invest in production in period 2:

C1  f  K , L  CA; X   1    g  CA; Z 

(4.2)

(2) Consumption in period 2 is the amount produced in that period:

C2  P2 f  K   g  CA; Z  , L; X 

47

(4.3)

We assume that the producer will invest in capital goods in the farm in period 1 that will be
used for production activity in period 2. Since we have set up a two-period model, we do not take
account of interest that differential income can earn if invested in period two.
48
The price E(P2) in period 2 can be any price. Therefore, at this point, in order to keep the
model general, we do not want to assign the price to be the PMG price or the spot-market price.
49
In our model construction, we assume that the only way to change consumption from what it
would if they relied only in the spot market is by joining a PMG since all other avenues are open
to all including consuming less in period one.
148

Participating in a producer group is costly due to membership fees and opportunity cost of time
due to meetings members are required to attend. The loss in monetary terms and labor in a
50

household may lead to some income drop .
Households that invest a portion of their differential returns in period 1 in productive
investments will experience an increase in income in period 2 due to the increase in capital.
Therefore, the household’s problem is to maximize its utility by choosing the extent of its
participation in the PMG, and the portion of returns,  that it will invest.





Max U f  K , L  CA; X   1    g  CA; Z  , P2 f  K   g  CA; Z  , L; X 


CA,
s.t.
0  1

(4.4)

To ensure optimality, first-order conditions with respect to collective action (CA) and proportion
of extra income invested in additional capital (  ) must be met as shown below:
Taking the first order conditions with respect to CA, we get equation (4.5), which shows that the
household chooses the collective action until

 

UC   f L  1    gCA   UC P2  f K gCA  0

1
2



(4.5)

where subscripts denote partial derivatives and arguments of the function have been suppressed.
This shows that joining collective action might decrease factor productivity in period 1 and
increase consumption in period 1 through differential income from higher prices and investment
in period 2 through investment of a portion of the extra returns.
50

However, this loss can be compensated for possibly entirely and over and above the earlier
income through differential income in depending on their choice of patronage. This is what the
chapter will test.
149

Taking the first-order condition with respect to  , we get equation (4.6), which shows that
the household equates, in marginal utility terms, the cost of participating in collective action in
period 1 and 2 with the overall gain from CA in periods 1 and 2. The solution implies that the
household maximizes utility with respect to the fraction of differential returns invested in period
2.









UC g CA; Z  UC P2 f K g CA; Z  0
1
2

(4.6)

Equation (4.6) equates a loss of consumption in period 1 with the gain in period 2. Given
expected prices in period 2, the household equates the marginal utility of consuming the
differential returns from CA in period 1 with the marginal utility of those differential returns in
period 2, in terms of income or consumption from its investment in period 2. If a household does
not invest any of its returns (i.e.,  =0), then the household is constrained to have a higher
relative marginal utility of consumption in period 1 than in period 2.
Rearranging equation (4.5) as a function of the marginal product of labor, we get:

 



UC

2 P * f  g
f L   1  * 
2 K  CA
UC



1


(4.7)

Equation (4.7) suggests that the household equates the marginal product of labor within the
household with some function of marginal loss labor as result of attending group meetings and
other labor inputs it has to make in the PMG.
Rearranging equation (4.6) we get:

fK 

UC
1 1
UC
P
2
150

(4.8)

Equation (4.8) states that the household attempts to set the marginal product of capital fK in

period 2 equal to the product of the inter-temporal marginal rate of substitution

UC
1 and
UC
2

relative prices in period 1, 1/P.
From equation 4.5 we find that joining collective action can have two effects. There
could be an initial decrease in production due to the commitments and resources put into
collective effort. But it also shows that there is differential consumption in period two. Our
theoretical model therefore leads to the hypothesis we intend to test in this essay of a long-term
improvement of production that may increase welfare. Households that participate in PMGs may
get some differential returns and invest in the household’s stock of assets, and if the optimal
portion of differential returns  * is positive, participation in PMGs should have a positive effect
on the overall welfare.
4.4 Estimation strategy
This chapter uses several methods for investigating the welfare issues. Here we
describe the DiD estimation strategy that looking at the differences in crop income and then use
other method to investigate the intra-annual price volatility. We use responses to survey
questions conducted in three rounds to evaluate quantitatively the impact of PMG membership.
A potential pitfall is that units of observation are not randomly assigned to participate; rather,
they might have self-selected to participate in producer groups. This becomes a problem in
estimation and inference. The decision to constitute some of the producer groups was promoted
by various development agencies, and the areas they are formed in are not necessary random;
hence, the program placement could be correlated with the outcome measure. In the
econometrics literature, statistical techniques used to analyze such data are often referred to as
151

"treatment effects" models (Goldberg, 1972, Maddala, 1983, Wooldridge, 2002), where the
51

program of interest is the "treatment,” which is joining a PMG in this analysis.

In the treatment effects literature, several approaches to modeling have been suggested.
First are the "Heckman-type" selection models (Goldberg, 1972, Heckman, 1976), in which a
selection equation and an outcome equation are jointly estimated, assuming a bivariate normal
error term in the two equations. Second are the instrumental variable estimators (Brundy and
Jorgenson, 1971). In this method, there is need for a variable to be measured that is related to
selection into treatment but not to the outcome measure, and this "instrument" is used to make
unbiased inference. Third are nonparametric matching methods, most prominently propensity
score matching (Rosenbaum and Rubin, 1983), in which the probability of each unit being
selected for treatment is first estimated and matched with control observations based on the
propensity score to the treatment (Buckley and Shang, 2003). Matching involves pairing
treatment and comparison units that are similar in terms of their observable characteristics. If the
relevant differences between two units are captured in the observable (pretreatment) covariates,
which occurs when outcomes are independent of assignment to treatment conditional on
pretreatment covariates, matching methods yield an unbiased estimate of the treatment effect
(Dehejia and Wahba, 2002).
In this study, we do not use the Heckman selection model because the Heckman method
does not compare two groups explicitly. Rather, the outcome equation only estimates the treated
group alone. We could have applied the instrumental variable approach, but the method has the
drawback that it is very difficult to get a good instrument that is related to the selection and not
51

Any future reference to “treatment” in this chapter will refer to joining a PMG, and the
“treatment effects” are the consequences for the household of joining the group.
152

the outcome. An attempt was made to instrument for participating in a PMG, but unfortunately
no appropriate instrument was identified. We use one of the simplest techniques for estimating
treatment effects with observational data, the "difference-in-differences" (DiD) estimator. The
DiD method is appropriate since we have panel data with baseline survey data and a subsequent
survey data after treatment. Using this method, we avoid the need for a quasi-experimental
survey method, which requires having an explicit counterfactual to the treated population, since
in the DiD method, the baseline becomes the control data. We now briefly describe the DiD
model.
4.4.1 The Difference-in-Differences Estimator
The DiD estimator model estimates the difference between outcomes measured at two
time points for both the treated observations and the controls (those not participating in the
program) and then compares the difference between the groups—hence the difference-indifferences (Ashenfelter, 1978, Ashenfelter and Card, 1985). This strategy ensures that any
variables that remain constant over time (but are unobserved) that are correlated with the
selection decision and the outcome variable will not bias the estimated effect.
This method thus requires repeated observations of the units. Either the technique can be
applied to a true panel, where data is gathered on the same units at both times, or repeated crosssections, such as two national random survey samples. The DiD model is effective in detecting
small treatment effects. We chose this estimation strategy since we have a very short panel and
producers are still slowly adopting the new marketing channel. However, when we use the fixed
effects estimation, we lose all the cases where we have only a single observation. In our study,
Western region had data collected only in 2007. This region will not be used for analysis in this
DiD method, since it requires a baseline survey or repeated surveys. Therefore, we use 400 cases
153

instead of the total 850 cases interviewed in the 2007 survey. The key assumption of the DiD
model is that the average change in the outcome is presumed to be the same for both the nonparticipants and, counterfactually, for participants if they had not participated. We assume that
changes in economic conditions or other policy initiatives affect both the participants and the
non-participants in similar ways. Dee and Fu (2004) provide an excellent discussion of this
assumption in an education research context and how to minimize the possibility of its violation
through the careful selection of independent variables (Ravallion, 2006). If this condition of the
average change in the outcome being the same for both the non-participants and,
counterfactually, the participants had they not participated is violated, then the estimates may be
biased, and one solution is to use a propensity score to get a comparable group as a control
(Abadie and Imbens, 2006).
We evaluate the impact of collective action or treatment on an outcome (household crop
income) Y on the population in the study area i. Suppose that there are two groups indexed by
treatment status T = 0, 1 where 0 indicates households that do not receive treatment, i.e., the
control group, and 1 indicates households that do receive treatment, i.e., the treatment group. We
observe households in two time periods, t = 0, 1, where 0 indicates a time period before the
treatment (before joining marketing groups) and 1 indicates a time period after the group
receives treatment, i.e., post-treatment. Every observation is indexed by the letter i = 1, ..., N ;
households will typically have two observations each, one pre-treatment and one post-treatment.
T

T

For the sake of notation, let Y 0 and Y 1 be the sample averages of the outcome for the treatment
C

C

group before and after treatment, respectively, and let Y 0 and Y 1 be the corresponding sample

154

averages of the outcome for the control group. Subscripts correspond to time period and
superscripts to the treatment status.

The outcome, Yi, is modeled by the following equation:

Yi     Ti   ti   Ti  ti   i

(4.9)

where the parameters α, β, γ, δ , are all unknown and εi is a random, unobserved error term that
contains all determinants of Yi that our model omits for simplicity. The coefficients have the
following interpretation:
α = constant term
β = treatment-group-specific effect (to account for average permanent differences between
treatment and control)
γ = time trend common to control and treatment groups that picks up any differences in the mean
of the latent individual effects between the treatment and comparison units, such as would arise
from initial purposive selection bias into the program
δ = gives the mean impact estimator

52

53

52

This is equivalent to a fixed-effects estimator in which the error term includes a latent
individual effect that is potentially correlated with treatment status.
53

As shown in the next section, the impact of interaction between time and treatment is referred
to as the impact estimator in the DiD literature.
155

4.4.2 Assumptions for an unbiased estimator
A reasonable criterion for a good estimator is that it be unbiased, that is, the expected

ˆ
value of the estimator, E     . The assumptions we need for the difference-in-difference
 
estimator to be unbiased are given by the following:
1. The model in equation (Outcome) is correctly specified.
2. E [εi] = 0.
3. The error term is uncorrelated with the other variables in the equation:
cov (εi, Ti) = 0, cov (εi, ti) = 0, cov (εi, Ti · ti) = 0
where the last of these assumptions, also known as the parallel-trend assumption, is the most
critical. Under these assumptions, we can use equation (4.9) to determine that expected values of
the average outcomes are given by
T
E Y0     
 
E Y1T         
 
C
E Y0   
 
E Y1C     
 

(4.10)

The difference-in-difference estimator is defined as the difference in average outcome in
the treatment group before and after treatment minus the difference in average outcome in the
control group before and after treatment.
T
T
C
C
ˆ
 DiD   Y 1  Y 0    Y 1  Y 0 

 




 

156



(4.11)

The DiD estimator assumes that the selection bias (the unobserved difference in mean
counterfactual outcomes between treated and untreated units) is time invariant, in which case the
outcome changes for non-participants reveal the counterfactual outcome changes,
c
c
E (Y c  Y0 | T  1)  E (Y c  Y0 | T  0)
1
1
1

(4.12)

Equation (4.12) is called the weak exogeneity condition, which holds only in differencein-difference estimation (Ravallion, 2006).
Taking the expectation of equation (4.11) we get



ˆ
ˆT
ˆDiD  E Y1T   E Y0   E Y 1

 
 


C



C
 E  Y 0 







                    

(4.13)

      

Hence, the estimation in equation (4.9) is shown to be unbiased.
The DiD estimator can be readily generalized to multiple time periods. Therefore, we
estimate the treatment effect by running a regression of Yit on the (individual and date-specific)
participation dummy variable T and other independent variables of interest with individual time
fixed effects (Ravallion, 2006).
For this study we use ICRISAT data of 400 households collected in 2003 (baseline) and
in a subsequent follow-up interview of the same households 2005 and a third round of survey in

157

2007 to have a total of three rounds of surveys. We will pool the data over the three time periods
54

and across treatment status and run one OLS Fixed Effect regression on
Yijt    A j   X ijt   t   Ti1   Ti1  t   it

:
(4.14)

Yijt= Welfare measure for household i in village j by time t. For our estimation, we use total
value of crop income

55

for a household in every year.

α = the intercept term

Aj = district fixed effect

Xijt = A vector of household and individual characteristics

t = the time dummy

Ti1 = the dummy that indicates whether the household joined a producer marketing group. It is
included as a separate regressor to pick up any differences in the mean of the latent individual
effects between those who joined PMGs and comparison units, such as would arise from initial
purposive selection bias into the program.

54

We ran a pooled OLS with random effect accounting for the clusters. This takes care of any
serial correlation of the error term within the individual households and any heteroscedasticity.
55
A discussion is provided in the results section for the rationale of using the value of crop
income as the measure of treatment effect in our model.
158

For Ti1.t, coefficient δ captures the interaction of the treatment dummy and the time dummy
variable, giving the mean impact, and is the main (treatment effect) variable of interest in the
model.

εijt = the error term.
For the individual household characteristics used in this model, we include age, gender,
education of the head of the family, and a dummy variable indicating whether the household has
someone in formal employment. The expectations are that the age of the household head is
correlated with farming experience the household has acquired over time and the amount of
resources it has. Younger households will have less experience and fewer productive resources to
employ in their farms. Therefore, it is expected that the older the household head, the more the
household is able to produce. However, since the productivity of the household head declines
over time, it is expected that the effect of age on productivity will level off and decline by a
certain age. Hence, we include the squared age of the household head as well in order to capture
the non-linearity. Also, women in rural areas have less control over productive resources than do
men due to culture and customs. They tend to do lots of house chores that keep them longer
around the homestead instead of being involved in farming activities that increase farm
production. The areas of study also happen to suffer from water scarcity during the dry periods.
Women have to travel long distances to get clean water, and that takes time away from their
farming activities. Therefore, we expect that having a female head of household will reduce the
value of crop production. Having more education is expected to improve the ability to acquire
better farming technologies and improve crop production. Having a household member in a
formal job should have an ambiguous relationship with crop production. It is possible that
159

individuals with a formal job have more resources to acquire better farm inputs and also hire
labor for crop production, thereby increasing production. On the other hand, having a member of
the household in formal job reduces the amount of family labor employed in farming. However,
farming in conditions with unreliable rainfall is unattractive for an individual with alternative
sources of income. Such a household could then cut back on its farming activities and rely more
on non-farm income.
We also included the independent variable “total land owned” in the estimated model.
This is a proxy for the amount of land available for farming and assets the household has to put
into farming. We expect that with more land, the higher the production. To account for
transaction costs, we had a variable for the length of time it took to sell the produce in the
market. This variable, together with the cost of transport to the market, attempts to capture how
far away the household is from the market. Remote areas have fewer alternatives in terms of
income-generation activities; hence, we expect households in these areas to concentrate more on
farming activities, increasing the value of production. We controlled for credit by including a
dummy variable indicating whether the household had received any form of credit. This variable
captures whether credit constraints are an impediment to crop production. Since most of the
households in these areas use their own seed and little in the form of purchased inputs, we expect
that credit access will have little impact on production.
Sources of price information were also taken into account. The expectation is that the
households that receive most of the information from the PMG would have a higher value of
production. The assumption is that these households will be active group members that get not
only price information but also other technology and improved inputs that enhance crop
production from the trainings through the groups. The amount of rain in the production year was
160

added to account for the variation in weather conditions, which have a profound impact on crop
production. A district dummy was added to account for location heterogeneity that is difficult to
observe and account for in the model.
4.5 Data
The data used is from the household surveys described in chapter one.
4.6 Results
4.6.1 Descriptive statistics and initial discussion
This chapter seeks to estimate the welfare impact of the introduction of the producer
marketing groups on the semi-arid farming communities in Kenya. The study uses two indicators
to assess impact. The first is the value of crop production value (taken as a measure of income),
using the panel data from Eastern Kenya only, and the second is the within-year price standard
deviation (taken as a measure of household price risk) of the main crop sold in the area, using all
the data from Eastern and Western Kenya.
Ideally, we would have desired to use a survey-based measure that approaches as closely
as possible one of the household welfare measure of economic theory. A common measure used
is the real income or real expenditure, although consumption is favored in developing countries.
Unfortunately, in our surveys, expenditure data was not collected. The next best measure is
household income. However, as Deaton (1997) argues, agricultural incomes can be extremely
variable month to month and year to year based on the cropping patterns and climate conditions.
Hence, household or individual incomes in semi-arid areas can be a poor indictor of living
standards when assessing impacts of PMG membership. Examining table 4.1 below shows that
other income, which consists of off-farm formal jobs salaries, sale of assets, charcoal, trees, and

161

other income that is not directly from farming activities, is a significant part of the farmers’
income.
Table 4-1: Variations in various measures of household welfare measure from 2003 to 2007

PMG membership
No

Yes

Year of the interview

Year of the interview

2003

2005

2007

Mean (Kshs)
Total asset value
Total other income

2003

2005

2007

Mean (Kshs)

360,232
56

Total crop net-income

111,499

69,830

177,169

60,949

49,901

99,889

46,791

97,836

63,726

16,453

6,530

18,224

1,417

30,565

Source: ICRISAT household survey data 2003-2007
Therefore, using total farmers’ income would be capturing this portion of income, which
may not be related to PMGs’ activities and hence overestimate their impact. In our study, we
argue that household income is not an appropriate measure because of coping strategies
households tend to employ. In times of drought, households rely heavily on off-farm income,
mainly from transfer payments from relatives and members of the households working in the
urban areas and from sale of assets. From 2003, when the baseline survey was conducted, the
rainfall pattern shows a steep decline in annual rainfall (See fig 4-1). The year 2005 recorded a

56

Total other income includes sale of assets, remittances, and total of salaries of members of the
families with formal employment.
162

major drought, which had significant impact on crop production and the household’s assets. The
households disposed some of their assets for consumption smoothing in such weather conditions.
Therefore, in this study it was difficult to use either the household income or assets as a
measure of welfare changes that can be attributed to the presence of PMGs. The time period is
also short for the organizations to have developed enough produce marketing share in the study
area to attract higher prices from the traders. It takes time for a PMG to develop to a level where
it can have enough quantities and financial clout to take the produce to the bigger markets or to
demand a significant price differential from the spot market. Table 4.1 above shows that mainly
between 2005 and 2007 there was a general decline in assets held. In the same period, mean total
crop net income increased for all groups.

Eastern Kenya rainfall per year
Kitui

Makueni

Mwingi

5000

Rainfall per year (mm)

4500
4000
3500
3000
2500
2000
1500
1000
500
0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Year
163

57

Figure 4-1: Rainfall patterns in the study areas

from 1995 to 2005

58

Producer marketing groups were used by the development agents to introduce better
seeds of selected crops in the study areas. In the Eastern region, short-, medium-, and longduration maturing pigeon peas were introduced. Also, more drought-resistant chickpeas and
green grams were also introduced. In Western Kenya, several varieties of disease-resistant
groundnuts were introduced. The net effect was that members were able to harvest from these
new varieties even in times of insufficient rainfall or with a severe disease outbreak, although
production was usually less than the full potential. Figure 4-2 below shows that the net mean
crop production value for members increased significantly from a negligible amount in 2005 to a
value even more than that of non-members in 2007.

57

Note that the rainfall graph has Kitui and Mwingi districts, which are not in the study areas,
and Mbeere, which is in the study area, is missing. There is no meteorological station in Mbeere,
but it shares very similar climate conditions with Kitui and Mwingi. Therefore, the rainfall
amounts for the two districts are used as proxies for Mbeere rainfall.
58
The graph also shows that the rainfall in 2005 tending to zero. There was a major drought in
2005 in Kenya with very little precipitation in the study areas with Kitui, Mwingi, and Makueni
having only 37, 32, and 153 mm of rainfall respectively.
164

300000

250000
Non-members HH
income

Value (Kshs)

200000

Members HH income
Non-members net crop
pdn value

150000

Members net crop pdn
value

100000

Non-members other
income
Members other income

50000

0
2003

2005

2007

Year

Figure 4-2: Movement of incomes and net crop production value over the study period

Since PMGs members would be the first to get the new varieties seeds, the hypothesis is
that before the technology is dispersed to the non-members we should be able to measure shortterm shifts in net crop production value, particularly in times of drought conditions, among PMG
members. Therefore, the use of net crop production value would be a better and appropriate
measure of the short-term impact of the presence of a PMG in a location. However, we point out
that crop production from where we derive the net crop value can be very variable due to
unreliable rainfall, particularly in Eastern Kenya. This is controlled for by having rainfall
variable in the econometric model. Figure 4-2 shows that household incomes display a more
165

volatile pattern than other incomes and crop production value. This finding gives more weight to
use net crop production value to measure the impact on PMGs on the study area.
4.6.2 Livestock value
Table 4-2: Value of livestock ownership across PMG members and non-members
PMG membership
No

Anova

Yes

Mean square

F

Sig

21,080

0.261

1.161

0.022

Mean
Livestock Value

15,923

(Kshs)

From table 4.2 above, there is a significant difference in livestock ownership between the
two groups. Livestock value is higher for PMG group members than non-members. This is
significant because livestock assets can quickly be liquidated to provide income to buy inputs for
crop production. Also, livestock, and especially oxen, are used as animal traction for land tillage
and weeding. In that case, this might bias the result of net crop production value towards the
PMG members. However, this is controlled for in the model by introducing a livestock value
variable.
4.6.3 Welfare measures trend
In general, households with higher mean total assets had less tendency to join a PMG
across the genders (see table 4-3 below). This is not surprising because some groups were

166

formed from the soil conservation village groups established as a food-for-work droughtmitigating effort by external development agents after a major famine in the 2001/2002 farming
season. Therefore, there is some expectation that poorer households could be overly represented
in the PMGs. From the table, non-members have higher mean asset values. However, female
members and non-members had lower mean total incomes. The scenario is reversed for the netcrop production value, where female-headed households had a higher value in comparison to the
traditional male-headed households. This could be an artifact of data driven by the fact that only
32% of the households in the whole survey were female-headed.
Table 4-3: Welfare measures by Gender
Welfare measure

Gender

Total assets

Female
Male
Female
Male
Female
Male

Total Income
Net crop
production value

4.6.4 Price volatility

PMG Membership
No
(Mean in Kshs)
247,539
182,777
77,601
91,580
16,581
14,833

Yes
(Mean in Kshs)
47,360
116,716
75,842
121,254
22,570
20,560

59

The study was conducted in marginal lands in Kenya at risk of droughts, floods, and
pests. Smallholder production in these areas is vulnerable to poor access to markets and frequent
weather shocks. Therefore, farmers’ income based on on-farm agricultural production is
vulnerable to volatile product prices (see figure 4-3 below). Figure 4-3 shows high volatility in
different crop prices for the 2006 production year. Intra-annual price volatility is an important

59

The section on price volatility uses all the data collected in the study (Eastern and Western
Kenya).
167

component of household net returns variability and general farmer welfare, and therefore it is
very important to quantify price volatility of agricultural products. The hardest-hit farmers are
net food buyers who are usually in the market later in the production season to buy stable food
for home consumption. The differences between the volatility in the prices among commodities
is also important for farmers’ private investment decisions in farm enterprise mix and product
marketing (Heifner and Kinoshita, 1994).
Maize

Soya beans

Field beans

Chickpea

Green gram

Millet

Sorghum

Cowpeas

Mangoes

Pigeon peas Citrus

70

Crop mean prices and St. Dev

60
50
40
30
20
10

0
Jan-March

April-June
July-Sept
Months of the year

Oct-Dec

Figure 4-3: Different crop price means and standard deviations for 2006
(Source: 2007 PMG level data set)
Note: The line bars in the graph are standard deviation from the mean

168

Farmers’ organizations are supposed to play a role in dampening the price volatility by
both bulking and storing the product to sell at better prices or negotiating stable prices in with
traders.
Different methods have been used to measure this price volatility, including the standard
deviation of prices, the coefficient of variation, and the Black-Scholes-Merton model (Jordaan, et
al., 2007). Other methods evaluated by Offutt and Blandford (1986) include the percentage
range, the average percentage change, the moving average, and the Coppock index.
In this chapter, we use the standard deviation of prices from the mean group price as our
measure of volatility. Since each farmer was associated with a group (either PMG members or
non-members), a mean price for every year was calculated for both members and non-members
separately. Using members’ annual mean price received for the main crop

60

sold in the region, a

standard deviation was calculated at the group level, and the same was done for the nonmembers associated with each group but using the non-members’ mean price. The annual
standard deviation calculated is then used as the dependent variable in the econometric model
later on for each year. The drawback of using this measure is that the standard deviation and the
coefficient of variation assume that past realizations of price and volatility have no influence on
current or future realizations. Hence, a better method would have been to use Autoregressive
Conditional Heteroscedasticity (ARCH) or Generalised Autoregressive Conditional
Heteroscedasticity (GARCH) approach (Just and Pope, 2002, Moledina, et al., 2003). However,
this requires time series data over a long period of time, which we do not have. Therefore, this
study uses standard deviation as a measure of price volatility.

60

For Eastern Kenya, the crop used was pigeon peas and in Western Kenya, prices for
groundnuts were used.
169

Figure 4-4 shows the price distribution of pigeon peas in Eastern Kenya and groundnuts
in Western Kenya. Both crop prices show evidence of fat tails, although most fall inside the
normal curve. Note that the graph for the pigeon peas has more data points because households
growing this crop in Eastern Kenya were interviewed in the three survey periods, whereas the
graphs for groundnuts have fewer data points because the first interviews in Western Kenya took
place in 2007. The price distribution for the Western region follows the normal curve. This could
be the impact the trading company had on stabilizing prices. The Eastern price distribution is
more skewed. These heavy tails indicate that higher price variability could be mitigated if
farmers had sale price contracts that would stabilize farmers’ prices.

170

Mean=56.85

60

Std. Dev=18.406
N=263
50

40

20

10

0

20

40

60

80

100

120

Figure 4-5: Graphs of Groundnuts crop price for the three-survey period

Figure 4-4: Graph of Groundnuts crop price for the three-survey

171

140

Mean=35.77

Std.Dev=22.421
125
N=638

100

50

25

0
0
0

20

40

60

80

100

120

140

Figure 4-6: Graphs of Pigeon peas and Groundnuts crop price for the three-survey
61
period
One of our expectations in this study was that the price variations would be lower for the
PMG members compared to the non-members. This is true for members’ standard deviations
groundnuts prices in Western Kenya, whose distribution is tightly packed closer to zero than that
of non-members (see figure 4-5 below). However, members’ pigeon peas price standard

61

The frequency is the number of members who sold at a particular price and the price is in
Kshs/kg
172

deviation is more dispersed from zero than that of non-members in Eastern Kenya (see figure 46).

40
30

20
10
0

30
20

10
0
0

10

20

30

40

Figure 4-7: Distribution of price standard deviation of groundnut for Western Kenya
(2007)

173

200
150

100
50

0

150

100

50
0
0

10

20

30

40

50

60

Figure 4-8: Distribution of price standard deviation of pigeon peas for Eastern Kenya
(2003-2007)
However, this could be caused by the inclusion of 2003 data where there were no groups
and hence everyone was a non-member. In the subsequent econometric analysis, this effect
would be picked up by the year dummy.
Figure 4-5 shows that being a member of a PMG is correlated with lower price volatility. The
members’ price standard deviation is tightly close to zero and bound within a narrow band of
zero to ten, whereas for non-members’ standard deviation of prices ranges from zero to twenty.
This is reversed for pigeon peas in Eastern Kenya, with non-members’ price standard deviation
174

being tightly bound from zero to 10, whereas members’ standard deviation is widely dispersed.
Again this could be affected by the inclusion of 2003 data before the groups were formed.
4.6.5 Random Effects or Fixed Effects model
4.6.5.1 Crop production value
In the analysis, we estimated both fixed effects and random effects models, which show
some differences in terms of values, significance levels, and signs in some coefficients.
Therefore, the logical question is which one do we use? If we assume that the unobserved
heterogeneity is uncorrelated with the other independent variables, then a random effects model
is most appropriate. In the model estimated, the results of the fixed effects model has a Corr(ui,xb)=-0.2838. This is a low level of correction, and we assume that the random effects model is
more appropriate. However, to confirm this observation, a Hausman test was implemented,
where the result chi2(16) was 10.98 with Prob>chi2=0.8105. We therefore fail to reject the null
hypothesis that the two methods yield identical coefficients.

4.6.6 Correlates of crop production value

62

In this section, we present both random and fixed effect models (table 4-4) to estimate the
impact of PMG membership on the level of net crop income. We estimate treatment effects
accounting for serial correlation and potential hetoroscadasticity of the error term within a
cluster.
The variable of interest is the interaction of PMG membership and year 2007
(pmgmem2007), which gives the average treatment effect.63 Though we have a three-period data
62

Crop price was deflated to remove the effect of inflation in the models.

63

This is because, as discussed in the DiD model earlier in the methods section, the variable
membership picks up any differences in the mean of the latent individual effects between those
175

set, we can only estimate 2007-treatment-effect because introduction of the 2005 effect will drop
one period due to the identification problem. Choosing the 2007 treatment effect captures the
long term effect. Results from the model show that the treatment effect is positive and equal to
64

Kshs 12,972 (US $ 172) . The inference is that compared to the producers who did not join a
PMG group, members had an increase of about US $ 171 in net crop production value in
2006/2007 production year. An increase of almost $ 171 worth of produce in the rural semi-arid
areas in Kenya is not a trivial figure in a country where 56% of the population lives on less than
a dollar a day (UNDP, 2006).
The dummy variables yr2005 and yr2007 capture the two subsequent rounds of the
survey after the baseline survey in 2003. These time dummies capture the effects on crop
production values that are due to other factors that are common to both treated and untreated
groups. The coefficient of the 2007 variable is positive and significant at the 1% level. This
finding suggests that there could have been a general improvement in drought- and diseaseresistant technology adoption in the study areas that accounts for this general improvement in
production. In other words, it is possible that the seeds made available to the PMG members had
already multiplied and the non-members were able to acquire them in enough quantities to show
this widespread crop production impact in a span of five years.

who joined PMGs and comparison units, such as would arise from initial purposive selection
bias into the program.
64
All conversions of the Kshs to a dollar are at a rate of US $ 1=75.10 Kshs
176

Table 4-4 Determinants of net crop income
Impact models (difference-in-difference) net crop income in the dependent
variable
(1)
(2)
Variables
RE model
FE model
PMG member
-7,473
-12,129
(1.46)
(1.58)
yr2005
-1,789
13,753
(0.29)
(1.43)
yr2007
50,936***
70,353***
(6.10)
(5.07)
pmgmemyr2007
12,972**
17,551**
(2.08)
(2.04)
HH characteristics
Age
409
1,459
(0.77)
(1.04)
Age2
-4
-14
(0.84)
(1.13)
Male
-2,126
-7,588
(0.77)
(1.55)
Education
1,143***
1,543
(3.23)
(1.52)
Formal job
1,413
-3,955
(0.47)
(0.82)
Assets
Land owned
732***
2,196***
(4.66)
(5.66)
Livestock value
0.193***
0.219***
(4.24)
(2.74)
Marketing and Transaction cost
Sell time (min)
-123*
3,348*
(1.71)
(1.73)
Transport cost
-94***
-246
(2.62)
(0.84)
Credit
Credit
1,864
1,536
(0.73)
(0.36)
Source of price information
Traders
7,368
21,778***
(1.44)
(2.64)
Government agents
-3,107
-6,343
(1.04)
(1.21)
NGOs
-155
3,282
(0.04)
(0.51)
Rainfall
177

Table 4-4 cont’d
RE model
11,956***
(5.51)

Log rainfall

FE model
15,329***
(5.73)

Location dummies
(2.81)
16,339**
(2.41)
-111,470***
(4.47)
1656

Makueni
Constant

Observations
R-squared
Number of clusters
1013
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%

-218,054***
(4.16)
1656
0.23
1013

The increase in net crop production value due to passage of time by an average of US
$678 for 2007 exceeds the marginal effects of any of the other variables in the model. All the
PMG groups distribute improved seeds to their members at a lower price than the non-members
can access them. However, informal seed distribution is very important for the open-pollinated
crops, and this could drive the increase in crop production over time. This finding is consistent
with earlier findings that there was widespread adoption of improved seeds in the study areas
(Shiferaw, et al., 2008). However, the time variable could also be picking up weather effects that
were not fully captured by the rainfall variable, which just measured total rainfall but not its
timing in critical periods of the crop cycle. Thus, all these effects of time may not just be due to
better technology. Also, global food prices went up in 2007 in real terms; to the extent that such
global price increases were transmitted to Kenya, the value of production could have increased
even if the physical volume of production did not.
Education in our model specification could be a proxy for the ability to acquire new
farming knowledge and utilize it appropriately that cannot be observed and measured. Education

178

is positive and significant at the 1% level. There is an increase in crop production value of Kshs
1,143 (US $16) for every extra year of schooling for the head of the household. This result is
consistent with our expectations and consistent with what many studies in farm technology
adoption have found— that more education increases technology adoption and hence more
production (Alene and Manyong, 2007, Feder, et al., 1985, Lin, 1991, Liu and Zhuang, 2000).
We therefore find evidence of positive returns to schooling in our model in terms of crop net
income.
The coefficient on acres of land owned was positive and significant at the 1% level. An
extra acre of land owned increased net crop production by Kshs 732 (US $10). This is expected,
with the exception of cases where there is a severe labor constraint. From the group survey
interviews conducted, it was clear that labor was not a constraint in many of the areas in the
study. And there was a well-functioning labor market; hence, households with more land
produced more. Again, this is consistent with other studies in literature (Feder, et al., 1985,
Idiong, 2007, Latruffe, et al., 2002). The magnitude of the increase in this study (the marginal
product of an acre of land) is low. This might reflect that larger farms have lower quality land on
average. However, this needs to be investigated further.
The longer it took to sell the product in the market, the lower the net crop income. The
coefficient of selling time is negative and significant at the 10% level. There is reduction of US
$1.70 in crop production value for every extra minute it took to market the product. The variable
captured the length of time it took to sell the product in the market in minutes. Hence, it captures
the effect of transaction costs on crop production value. As expected, higher transaction costs
reduce household mean crop production value. This variable could also be picking up the effect
of farmers who produce lower quality products relative to their peers receiving lower prices and
179

having to take longer times to sell their products. Or it could be that those who live in more
remote areas receive lower prices and have to take a longer time to find buyers because of their
remoteness.
The transport cost variable captures part of the marketing costs. As expected, the
coefficient is negative and significant at the 1% level. There is a reduction of US $ 1.30 in crop
65

net income for every increase in transport cost of US $ 0.01 per trip . That shows that farmers
who were further away from the market or who lived in areas with poor road infrastructure had
lower net crop production. Hence a PMG with good storage facilities and marketing abilities
could play an important role in increasing farmers’ crop production and thereby income.
Individuals who received most of their market information from government extension
services had reduced net crop income by Kshs 6,651 (US $ 89). The coefficient on government
agencies is negative and significant at the 1% level. The variable could be capturing the fact that
individuals who are more market oriented and therefore with higher production will not wait for
government extension services to get their marketing information. They will use other avenues to
get price and marketing information to get better returns for their products. On the other hand,
poor farmers who produce less crop value are still reliant on government information services.
An alternative explanation could be that government marketing information is poor and does not
help farmers. Government extension services could be targeting its activities more to the poor
and those in remote areas where prices are lower for greater impact.

65

The transport cost was collected per trip and not per kilogram of the produce. In general,
transport cost for produce in Kenya is not per weight but per trip for a certain amount and in the
data collection, it was difficult to collect data per weight. Therefore, it is difficult to assess the
marginal cost per kg and judge the arbitrage opportunity.
180

As expected, a 1% increase in rainfall in the study areas increases net crop production
value by a significant amount—Kshs 11,956 (US $ 160). The implication is that the regions
studied were in general moisture deficit condition and any technology that can conserve
whatever little rainfall occurs in the year could have big impact in crop production and therefore
food security. Simple technologies like zero or double tillage, soil conservation and early
planting may greatly improve crop production. Though this study was not meant to look at the
types of crops produced in the area, with such low rainfall, a crop mix that emphasizes droughtresistant crops could also be helpful in improving food security and producers’ welfare.
Controlling for livestock value captures the effect of owning livestock on net crop
production. An increase in livestock value increases the crop net income. The coefficient on
livestock is positive and significant at the 1% level. However, the actual magnitude is negligible.
The variable could be capturing the fact that livestock is used for animal traction in tillage or
even relaxing credit constraints, since some livestock is easily liquidated. However, it could also
be capturing the wealth effect since wealthier households are likely to have more livestock. The
small magnitude of the effect could reflect that those who had livestock generally had lands less
suited for crop production (mainly grazing areas).
Households who get their information from traders have a higher crop net income by
21,778 KSh (US $ 290). The coefficient is positive and significant at the 1% level. Wealthier
farmers could be more likely to market their produce through traders and are able to negotiate
higher prices due the quantity they produce. However, there is need for further investigation to
find out reasons for this phenomenon.

181

4.6.7 Price volatility regression results
In this section, we present three models on price volatility, of which one is for the
dominant crop in Western Kenya because we had data for only one year in that area. The other
are two for Eastern Kenya, estimated for 2005 and 2007. Because we are looking at intra-annual
price volatility, we are not using the DiD model for this analysis but we estimate the intra-annual
price volatility a household faces each of the periods the data was collected. The dependent
variable is the crop price standard deviation from the mean price received by PMG members (for
those who were members of the marketing groups) or by non-members (for those who did not
belong to PMGS) for each region. An argument can be raised that those facing greater price
volatility are more likely to join the PMGs, bringing the possibilities of endogeneity problems
into the model. Therefore, this was tested using Davidson-MacKinnon test for exogeneity. The
results (exogeneity: 1.7555 F(1,611) P-value = 0.1857) show that the estimated coefficient of
the residual was not significantly different from zero, and thus the test failed to reject the null
hypothesis that the decision to join the PMG was exogenous.
The coefficient for the variable group membership in 2007 models is negative but not
significant (see table 4-5 below) and in 2005 it is positive but again not significant at the 10%
level. Therefore, there is no significant difference in price volatility between the PMG members
and non-members. From theory, we expected that being a member would allow the group to
negotiate a better and stable price for its produce. However, as explained earlier, it takes time for
these producer groups to gain enough clout in the marketing of the produce to have a significant
impact in the area of marketing.

182

4.6.6.2 Impact of PMG membership on price volatility
Table 4-5 Determinants of crop price volatility for Eastern and Western Kenya
Eastern Kenya†
2005
Individual characteristics
PMG member
Age
Male
Education
Location characteristics
Village population

2007

Western Kenya‡
2007

1.04
(1.51)
-0.00
(0.10)
0.92*
(1.72)
0.00
(0.04)

-1.06
(0.94)
0.03
(1.00)
-0.33
(0.45)
0.03
(0.32)

-0.70
(0.45)
0.05
(1.39)
-1.03
(0.98)
-0.02
(0.16)

0.00
(0.03)

-0.00
(0.51)
-0.27
(0.06)

0.002
(0.09)

0.05
(1.23)

0.09**
(2.00)

0.30**
(2.02)

0.04***
(2.82)

-0.01*
(1.91)

-0.01
(0.60)

0.00
(0.44)

0.00
(0.10)

-0.02
(0.95)

-0.99
(1.11)
0.72
(0.92)
-0.15
(0.22)

1.84
(1.63)
1.75*
(1.74)
0.16
(0.19)
2.78**

3.06**
(2.22)
3.61***
(2.66)
1.38
(1.16)

Log rainfall
Household wealth
Land cultivated
Marketing cost
Transport cost
Group characteristic
Decision index
Sources of Price information
Government agents
Traders
Mass media
Mobile phone
Spatial location
Siaya

Makueni

-1.63
(1.45)
-1.54
(0.98)
0.49
(0.87)

-5.60
(1.47)

183

Table 4-5 cont’d
Constant

-0.26
(0.15)
49
0.31

8.84
(0.33)
380
0.14

Observations
R-squared
Robust t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
† Price St. Dev of pigeon peas is the dependent variable in Eastern Kenya
‡ Price St. Dev of groundnuts is the dependent variable in Western Kenya
The whole process is very slow and goes in stages. The first stage is setting up a PMG by
a few devoted members. Then they have to market the idea to other producers. Due to past
failures in such efforts, this is a tough task. As the membership grows, they have to convince
other members to market a large proportion of their produce together. Eventually, if the group is
successful in demonstrating its usefulness, more members join, and with enough volume of
products successful collective marketing is achieved. If a certain volume threshold is achieved, a
PMG can have a big enough market power to negotiate a higher and/or more stable price. With
higher and stable price, there will be a cascading effect where everyone will want to join a group.
Being outside of the group will be undesirable and increase the price variability.
Producers who cultivated bigger parcels of land had higher price volatility in 2007. The
coefficient on land owned is positive and significant at the 5% level in both regions. In the net
crop production model in table 4-4, cultivating more land increases net crop production value;
hence, we expect these farmers to have a bigger bargaining power. Bigger producers, however,
face higher price risk than do smaller farmers, and if the PMGs manage to lower the risk, then it
possible to attract these famers to be members. An alternative explanation for the greater price
variance faced by large farmers could be that big producers consistently produce more products,
and hence sell throughout the year in significant quantities, while the small farmers only produce
small quantities in the year. The small producers’ sales would all be clustered around the same
184

time, while the large producers would be selling throughout the year and thus facing more price
volatility. Similarly, if the large producers have more liquidity, they may not be forced to sell
everything right after harvest to meet cash flow needs. Thus, they could be selling throughout the
season, and hence face more seasonal price changes than the small farmers, forced to sell right
after harvest.
Higher transport cost to the market per trip, as shown by the coefficient of transport cost,
has ambiguous correlation with intra-annual price volatility. It is positive and significant at the
1% level in the Eastern region for 2005. It is also negative and significant at 10% level in the
same region in 2007. The variable is a proxy for the remoteness in terms of location of the
producer in relation to the market. The logical explanation for the negative correlation is that
these producers are far away from the market, few middlemen go to their farms to buy their
produce at farm gate. So the few traders who venture in these areas then offer similar prices to
the farmers with little variation due to lack of alternatives markets. Hence, although in our case
the price variation is measure of risk, less risk in this case might not necessary be a good thing.
They could be getting less variable but lower prices than they would get if there were greater
competition among the buyers. However, it is difficult to account for the reversal of the sign
between 2005 and 2007. Therefore, more research needs to be done in these areas to account for
the ambiguity.
Producers who get their market price information from a trader experienced a higher
annual price variation. The coefficient is positive and significant at the 1% level for groundnuts
in Western Kenya and at the 10% level for pigeon peas in Eastern Kenya in 2007. Two things
could be going on here. One is that since farmers still sell their produce to the local buyers, the
variable could be capturing the inherent price risk in these areas. On the other hand, it could also
185

be capturing the fact that the buyers generally give varying prices to their customers and the
variable is capturing this fact. However, the coefficient for receiving information from a
government extension agent is also positive and significant at the 5% level for the groundnuts in
Western Kenya in 2007. This is against the expectations. Those who received price information
from government extension agents also had lower crop production values. Therefore, more
research needs to be conducted on government extension services operations and their quality in
the study areas.
Farmers in Eastern Kenya (growing pigeon peas) in 2007 who accessed price information
through a cell phone received prices that were more volatile. The coefficient on cell phone
information is positive and significant at the 5% level. A cell phone allows a farmer to search for
market in more markets than is usually possible without the facility. Therefore, it is likely that
such a farmer will sell in several markets or deal with many different traders offering varied
prices. Hence, this variable could be capturing this variation in different market participation.
4.7 Conclusions
In this chapter, we sought to investigate the impact of being a member of a PMG on
farmers’ crop revenue and on price risk.
The first objective was to look for evidence of the impact of producer groups on the crop
net income. The models find that there is indeed evidence that membership in a PMG leads to an
increase in the net crop income among the members through increased production or through
price effects. We also find that overall in the same areas there was an increase in net crop income
to the community as whole despite adverse weather conditions in the area of study. This could
point to the fact that new seed technology was widely accepted and adopted by the farmers in
general.
186

Another important finding is the impact of transaction and marketing costs on net crop
production value. Few studies have captured TC directly, and this study contributes to the
literature by showing the impact these costs have on net crop production values and thereby
farmers’ welfare. We show that TC and marketing costs directly reduce net crop income of
individual households.
On the second objective, we find no evidence of any difference between the intra-annual
price volatility faced by PMG members and non-members. Maybe in the long run, a group might
gain enough market clout and stabilize the price. Therefore, policy makers should take a longterm view on the price stabilization ability of these institutional arrangements. Promotion of the
groups should have a long-term plan on how to work with the groups until they get enough
market power to stabilize prices through contracts or a better bargaining power. These groups
should not be looked at as a quick fix for the small-scale farmers’ price risks problems.
We find that PMGs can be utilized as a tool to alleviate poor food security situations in
marginal areas. This study gives credence to those who have argued in the past that farmer
groups could be used to alleviate poverty among the poor rural farmers in developing countries.

187

Chapter 5 : Conclusions
The three analytical chapters in this study explore how collective action, in the form of
producer marketing groups, is applied in a smallholder farmers’ context in semi-arid areas of
Kenya, and the impacts it has on farmers who embrace it. Whereas agriculture can be used to
alleviate poverty and food security problems among the rural poor in developing countries
(World-Bank, 2008) and in particular Sub-Saharan Africa, its record in SSA has had mixed
results. However, there is the renewed hope that the rapid growth in domestic and global
markets, innovations in communication technologies, and the revolution in biotechnology may
provide the opportunities these farmers have missed in the past. But realizing this promise
requires that farmers in developing countries and especially the smallholder farmers have access
to input and output markets and other supporting services. Producer marketing groups could be
one of the tools that can be used to exploit this potential.
Although smallholder farmers are usually efficient producers, they need supporting
organizational structures that understand the marketplace and the unique characteristics of rainfed smallholder farming. Without an enabling environment or institutions that support the
interface between the traders and farmers, low and volatile farm incomes have been the result in
many instances. However, PMGs can be applied to bridge the gap between the smallholder
farmers and the private sector, government agencies as well as other service providers. Yet the
main constraint has been the understanding of how to apply collective action appropriately to
these farmers.
The gap in the literature has been the understanding of the dynamics of collective action
within the context of smallholder farming. Therefore, the second chapter of this study explored
in depth the conditions necessary for effective and sustainable PMGs. The third chapter explored
188

factors that determine participation in PMGs, patronage and the number of CA efforts a farmer
joins. The fourth chapter contributes to the debate on the impact of PMG membership on crop
income and price risk.
Research on CA is very wide, but this study concentrates on a subset of collective action
called farmers’ organizations. Among a wide array of CA referred to as farmers’ organizations,
the study focused on producer marketing groups, mainly formed by smallholder farmers. The
study finds that if farmers produce a product characterized by high asset specificity, in the
presence of an altruistic leader it is more likely for PMGs to emerge and be sustainable. An
alternative to a single altruistic individual was the presence of a small core group that was
dedicated to the collective action effort who form the nucleus around which a group is formed
and operated.
From the qualitative and quantitative analyses in chapter two, we found a common set of
factors that foster success of PMGs. In addition to the presence of an altruistic leader or core
group that forms as a nucleus around which other members work, good governance structures
were also found to be important for group success. These structures include bylaws well
understood by members, enforceable and enforced. For a group also to remain relevant and
attract patronage, it needs to have diverse activities that increase members’ benefits from the
PMG and increase the cost of being excluded in case one is not cooperative. Regular monetary
subscription requirements by a group increases active members’ commitment to the effort. It can
also be used to signal members’ cooperative characteristics. Successful groups were also found
to serve a smaller geographical location, which eases group coordination.

189

To improve development agents’ and government service providers’ understanding of
PMGs, a Group Analysis Framework was developed. This framework can be utilized to pinpoint
areas of interventions to improve a variety of CA effort success rate. The framework is also
tested in a case-study approach as well as using quantitative analysis. This is an important
addition to the literature of CA among smallholder farmers, since development and various
government agencies promote group formation but lack a yardstick to identify areas of
interventions specific to each group. These agencies are usually important in providing education
to groups in the early life of a group growth cycle. They also play an important role initially in
connecting groups with various market actors and other service providers. Therefore, the
agencies’ understanding of each group they work with could prove to be very important for a
group’s success probability.
Chapter three investigates the determinants of household participation in PMGs,
patronage, and the number of groups joined. The study finds some factors that are common
drivers in at least two areas. An increase in age of the household head and an increase in the
diversity of the individuals in an area had ambiguous effects, both increasing the probability of
joining a group but decreasing the percentage of sales through it. The presence of a more
democratic group increased the probability of a farmer joining a group as well as increasing the
number of groups joined. The study also tested the impact of previous assistance by a
development agent on collective action. We find that farmers who had received assistance from a
development agent were more likely to join a group and increase patronage. This is contrary to
the findings in sociology literature as discussed earlier.
Chapter four investigates the impact PMGs had on crop production and the price risk
members face compared to the non-members. The study does not find any evidence that groups’
190

members face different price risk compared to the non-members. This could be explained by the
fact that in the early stages of a PMG group and its role in product marketing is limited.
However, the study finds that PMGs improved net crop income for its members compared to the
non-members.
5.1 Policy implications
From the study, we find that there is evidence that participating in a PMG group
improves net crop income. Therefore, PMGs can be used to improve rural farmers’ food security
and increase their incomes. However, for groups to emerge and be successful, good leadership
plays a crucial role. Therefore, great emphasis should be put in providing farmers with education
on group dynamics and sensitizing them on the importance of PMGs and good management
practices. Formation of a new group should be encouraged only in the presence of a core group
of individuals or an altruistic leader to spearhead the efforts. Emerging groups ideally should
draw their memberships from a small geographical area and have few members. This fosters
bonding to increase social capital that is needed to mitigate the free-rider problem and draw
proper governance structures. The group also should have diverse activities to increase its value
to the members and raise the cost of being uncooperative.
5.2 Study limitations and recommendations for further studies
The study used a panel of five years. Hence, more research in this area is needed to verify
these results with a longer panel data and in particular on impact assessment with better
measurement indicators utilized. More testing is also necessary for the group analysis framework
to ensure that it can be applied to a wide form of collective efforts.

191

REFERENCES

192

References

Abadie, A., and G. Imbens. 2006. "Large Sample Properties of Matching Estimators for Average
Treatment Effects." Econometrica 74(1):2006.
Aker, J., C. . 2008. "Does Digital Divide or Provide? The impact of Cell Phones on Grain
Markets in Niger." Center for Global Development Working Paper No. 154:61.
Akin, S.J., D.K. Guikey, and E.H. Denton. 1995. "Quality for Services and Demand for Health
Care in Nigeria: A multinomial Probit Model Estimation." Social Science Medicine
40(1527-37).
Alaba, O.B., and O.A. Alaba (2002) Determinants of Demand for Infrastructure in Rural and
Urban Nigeria. Oxford UK.
Alene, A.D., and V.M. Manyong. 2007. "The effects of education on agricultural productivity
under traditional and improved technology in northern Nigeria: an endogenous switching
regression analysis." Empirical Economics 32:141-159.
Amemiya, T. 1985. Advanced Econometrics. Cambridge, MA: Havard University Press.
Anderson, B., L., and B. Henehan, M. 2003. "What Gives Cooperatives a Bad Name?" Paper
presented at The NCR 194 meeting. Kansas City, Missouri, October 29, 2003.
Anesi, V. 2007. "Moral Hazard and Free Riding in Collective Action." Centre for Decision
Research and Experimental Economics 2007-04.
Angus, D. 1997. The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy. Washington,DC: World Bank Publications.
Antony, S.C., and E.D. Steven. 2005. "Getting to “We”: Examining the Relationship between
Geographic Scale and Ingroup Emergence in Collaborative Watershed Planning." Human
Ecology Review 12(1):14.
Ashenfelter, O. 1978. "Estimating the Effect of Training Programs on Earnings." Review of
Economics and Statistics 60(1):47-57.
Ashenfelter, O., and D. Card. 1985. "Using the Longitudinal Structure of Earnings to Estimate
the Effect of Training Programs." Review of Economics and Statistics 67(4):648-660.
Babatunde, R.O., E.O. Olorunsanya, and A.D. Adejola. 2008. "Assessment of Rural Household
Poverty: Evidence from South-western Nigeria." American-Eurasian J. Agric. &
Environ. Sci. 3(6):900-905.

193

Baffes, J., P. Varangis, T. Akiyama, and D.F. Larson. 2003. "Commodity Market Reform in
Africa: Some Recent Experience." World Bank Working paper.
Baland, J.-M., and J.-P. Platteau. 1999. "The Ambiguous Impact of Inequality on Local Resource
Management." World Development 27(5):773-788.
Banaszak, I. 2007. "Testing Theories of Cooperative Arrangements in Agricultural markets:
Results from Producer Groups Poland." Paper presented at Joint IAAE-104th EAAE
Seminar, Agricultural Economics and Transition: 'What was expected, what we observed,
the lessons learned.'. Corninus University of Bundapest (CUB) Bundapest, Hungary,
September 6-8, 2007.
Bandura, A., and R. Wood. 1989. "Effect of Perceived Controllability and Performance
Standards on Self-Regulation of Complex Decision-Making." Journal of Personality and
Social Psychology 56(5):805-814.
Barrett, C.B. 1997. "Food Marketing Liberalization and Trader Entry: Evidence From
Madagascar." World Development 25:763-777.
Barrett, C.B., and E. Mutambatsere (2008) Marketing Boards, ed. S.N. Durlauf, and L.E. Blume,
Palgrave Macmillan.
Bellemare, M., F., and C.B. Barrett. 2006. "An Ordered Tobit Model of Market Participation:
Evidence from Kenya and Ethiopia." American Journal of Agricultural Economics
88(2):324-337.
Ben-Akiva, M., and S. Lerman. 1995. Discrete Choice Analysis Theory and Application to
Travel Demand. Cambridge, MA: MIT Press.
Benham, A., and L. Benham. 1998. "Measuring the Costs of Exchange." Paper Prerared for the
Second Annual Meeting of the International Society for New Institutional Economics,
September 17-19.
Berdegue, J.A. 2001. "Cooperative to Compete: Associative Peasant Business Firms in Chile,
(PhD Thesis).", Wageningen University.
--- (2002) Learning to Beat Cochrane’s Treadmill. Public Policy, Markets and Social Learning in
Chile’s Small-Scale Agriculture, ed. C. Leeuwis, and R. Pyburn. Assen, The Netherlands,
Koninklijke Van Gorcum.
Berdegue, J.A., E. Ramirez, T. Reardon, and G. Escobar. 2001. "Rural Non-Farm Employment
and Income in Chile." World Development 29(3):411-425.
Bergstrom, T.C. 2002. "Evolution of Social Behavior: Individual and Group Selection." Journal
of Economic Perspectives 16(2):67-88.

194

Berhanu, G., J. Pender, and T. Girmay. 2001. "Community Natural Resource Management: The
case of woodlots in Northern Ethiopia." Environment and Development Economics.
Berne, E. 1963. The Structure and Dynamics of Organizations and Groups. Philadephia: J.B.
Lippincott Company.
Bingen, J., A. Serrano, and J. Howard. 2003. "Linking Farmers to Markets: Different
Approaches to Human Capital Development." Food Policy (28):405-419.
Birchall, J. 1997. The International Cooperative Movement. Manchester: Manchester University
Press.
Boerner, C.S., and J.T. Macher. 2003. "Transaction Costs Economics: An Assessment of
Empirical Research in the Social Sciences." Working Paper Georgetown University.
Bond, L., M. Clement, M. Cournoyer, and G. Dupont. 2000. Survival Rates of Co-operatives in
Québec Government of Canada (Gouvernement du Canada) Co-operatives Secretariat
Secrétariat aux.
Bourdieu, P. 1986. The forms of capital. In: John G. Richardson (ed.): Handbook of Theory and
Research for the Sociology of Education. New York: Greenwood Press.
Boyd, S., M. Boland, K. Dhuyvetter, and D. Barton. 2007. "Determinants of Return on Equity in
U.S. Local Farm Supply and Grain Marketing Cooperatives." Journal of Agricultural and
Applied Economics 39.
Brundy, J., and D. Jorgenson. 1971. "Consistent and Efficient Estimation of Systems and
Simultaneous Equations by Means of Instrumental Variables." Review of Economics and
Statistics 53:207.
Buckley, J., and Y. Shang. 2003. "Estimating Policy and Program Effects with Observational
Data: The "Differences-in-Differences" Estimator." Practical Assessment, Research &
Evaluation 8(24).
CCK-Kenya (2008) COMMUNICATIONS STATISTICS REPORT 2008. Nairobi, Kenya.
--- (2006) General Telecommunications Indicators and Market Information, vol. 2006. Nairobi,
Kenya, Communication Commission of Kenya,
http://www.cck.go.ke/market_information-telecommunications/, pp. Market information.
Chamberlin, J., R. 1974. "Provision of Collective Goods as a Function of Group Size." American
Political Science Review 68(June):707-716.
Chirwa, E., A. Dorward, R. Kachule, Ian Kumwenda, J. Kydd, N. Poole, C. Poulton, and M.
Stockbridge. 2005. "Walking Tightropes: Supporting Farmer Organizations for Market
Access." Natural Resource Perspectives (99).

195

Chowdhury, S., K. 2006. "Access to a Telephone and Factor Market Participation of Rural
Households in Bangladesh." Journal of Agricultural Economics 57(3):563-576.
Chowdhury, S., A. Negessa, and M. Torero (2005) Market Institutions: Enhancing the Value of
Rural-Urban Links. Washington D.C., International Food Policy Research Institute.
Coase, R.H. 1937. "The Nature of The Firm." Economica 4:386-405.
Cohen, J. (1999) Trust, Voluntary Associations and Workable Democracy: The Contemporary
American Discourse of Civil Society ed. M.E. Warren, 1 Edition. Cambridge, Cambridge
University Press, pp. 208-248.
Collion, M., and P. Rondot. 2001. "Investing in Rural producer Organisations for Sustainable
Agriculture." World Bank, SASKI.
De Janvry, A., Fafchamps, M., Sadoulet, E. . 1991. "Peasant Household Behavior With Missing
Markets: Some Paradoxes Explained." The Economic Journal 101:1400-1417.
Dee, T.S., and H. Fu. 2004. "Do Charter Schools Skim Students or Drain Resources? ."
Economics of Education Review 23:259-271.
Dehejia, R.H., and S. Wahba. 2002. "Propensity Score-Matching Methods for Nonexperimental
Causal Studies." The Review of Economics and Statistics 84(1):151-161.
Demir, F. 2007. "Private Investment, Portfolio Choice and Financialization of Real Sectors in
Emerging Markets." Munich Personal RePEc Archive, (MPRA Paper No. 3835):1-41.
Deshpande, R., J.U. Farley, and F.E. Webster Jr. 1993. "Corporate Culture, Customer
Orientation, and Innovativeness in Japanese Firms: A Quadrad Analysis." Journal of
Marketing 57(1):23-37.
Dimitri, C., A. Effland, and N. Conklin. 2005. "The 20th Century Transformation of U.S.
Agriculture and Farm Policy." Economics Research Services 3.
Dorward, A. 2001. "The Effects of Transaction Costs, Power and Risk on Contractual
Arrangements: A Conceptual Framework for Quantitative Analysis " Journal of
Agricultural Economics 52(2):59-73.
Dorward, A., J. Kydd, J. Morrison, and C. Poulton. 2005. "Institutions, Markets and Economic
Co-ordination: Linking Development Policy to Theory and Praxis." Development and
Change 36(1):1-25.
Doss, C.R. 2003. "Understanding Farm Level Technology Adoption: Lessons Learned from
CIMMYT’s Micro Surveys in Eastern Africa." CIMMYT Economics Working Paper 0307. Mexico, D.F.:CIMMYT.

196

Dow, W., and E. Norton. 2003. "Choosing Between and Interpreting the Heckit and Two-Part
Models for Corner Solutions." Health Services and Outcomes Research Methodology
4(1):5-18.
Drugger, W.M. 1983. "The Transaction Cost Analysis of Oliver E. Williamson: A New
Synthesis." Journal of Economic Issues XVII(1):95-114.
Duan, N., W. Manning, C. Morris, and J. Newhouse. 1984. "Choosing Between the Sample
Selection Model and the Multi-Part Model." Journal of Business and Economic Statistics
2:283-289.
Eggertson, T. 1990. "Economic Behavior and Institutions." Cambridge Survey of Economic
Literature.
Ephraim Chirwa, Andrew Dorward, Richard Kachule, Ian Kumwenda, Jonathan Kydd, Nigel
Poole, Colin Poulton, and M. Stockbridge. 2005. "Walking Tightropes: Supporting
Farmer Organizations for Market Access." Natural Resource Perspectives 99.
Esengun, K., M. Sayili, and H. Akca. 2005. "Tomatoes Production, Marketing Structure and
Solution Recommendations for problems of Farmers: A Case Study." Journal of applied
Sciences 5(10):1854-1857.
Esteban, J., and D. Ray. 2001. "Collective Action and Group Size Paradox." American Political
Science Review 95:663-672.
Esteban, J., and D. Ray. 2001. "Collective Action and the Group Size Paradox." American
Political Science Review 95(3).
Fafchamps, M. 2004. Market institutions in sub-Saharan Africa: Theory and evidence:
Comparative Institutional Analysis (CIA) Series, vol. 3. Cambridge and London: MIT
Press.
Fafchamps, M., Minten, B. 2001. "Property Rights in a Land Market Economy." Economic
Development and Cultural Change 49(2):229-268.
FAO. 2002. "Comprehensive Africa Agricultural Development Programme." Food and
Agricultural Organization.
---. 2008. "Crop Prospects and Food Situation." Global Information and Early Warning System
on Food and Agriculture (2, April).
---. 2008. "Food Outlook " Global Market Analysis June.
Feder, G., R.E. Just, and D. Zilberman. 1985. "Adoption of Agricultural Innovations and
Developing Countries: A survey " Economic Development and Cultural Change
33(2):255-298.
197

Francesconi, G.N., and R. Ruben. 2007. "Impacts of Collective Action on Smallholders’
Commercialisation: Evidence from Dairy in Ethiopia." Paper presented at The I
Mediterranean Conference of Agro-Food Social Scientists. 103rd EAAE Seminar
‘Adding Value to the Agro-Food Supply Chain in the Future Euromediterranean Space’.
Barcelona, Spain, April 23rd - 25th, 2007.
Frank, R.H. 1988. Passion within Reason: The strategic Role of Emotions. New York, USA:
Norton.
Gachimbi, L.N., A.d. Jager, H.v. Keulen, E.G. Thuranira, and S.M. Nandwa. 2002.
"Participatory Diagnosis of Soil Nutrient Depletion in semi-arid areas of Kenya."
Managing Africa’s Soils (2):1-22.
Garforth, C. 1994. "Rural People's Organizations and Agricultural Extension in the Upper North
of Thailand: Who Benefits." Journal of International Development 6(6):707-720.
Gebremedhin, T.G., and R.D. Christy. 1996. "Structural Changes in US Agriculture:
Implications fro Small Farms." Journal of Agricultural and Applied Economics 28(1):5766.
Gentzoglanis, A. (1997) Economic and Financial Performance of Cooperatives and Investor
Owned Firms: An Empirical Study, ed. J. Nilsson, and G. Van Dijk. Assen, The
Netherlands, van Gorcun.
Godquin, M., and A. Quisumbing, R. 2006. "Groups, Networks, and Social Capital in the
Philippines Communities." Paper presented at International Research Workshop on
'Gender and Collective Action'. Chiang, Mai, Thailand.
Goetz, S.J. 1992. "A Selectivity Model of Household Food Marketing Behavior in Sub-Saharan
Africa." American Journal of Agricultural Economics 74(May):444-452.
Goldberg, A. 1972. Selection Bias in Evaluating Treatment effects: Some Formal Illustrations.
Madison, Wisconsin: University of Wisconsin Press.
Goldenberg, J., and D. Mazursky. 2002. Creativity in Product Innovation. Cambridge:
Cambridge University Press.
Gonzalez, E.T. "Reducing Poverty and Hunger in Asia Developing Rural Institutions That
Strengthen the Voice of the Rural Poor." International Food Policy Research Institute.
Granovetter, M., S. 1985. "Economic Action and Social Structure: The Problem of
Embeddedness." The American Journal of Sociology 91(3):481-510.
Gregorio, M.D., K. Hagedorn, M. Kirk, B. Korf, N. McCarthy, R. Meinzen-Dick, and B.
Swallow. 2008. "Property Rights, Collective Action, and Poverty: The Role of
Institutions for Poverty Reduction." CAPRi WORKING PAPER (81):1-52.
198

Grootaert, C. 1999. "Social Capital, Household Welfare, and Poverty in Indonesia " Local Level
Institutions Working paper No. 6. World Bank.
Grootaert, C., G. Oh, and S. A. 1999. "Social Capital and Outcomes in Burkina Faso." Local
Level Institutions Working Paper No. 7. World Bank.
Gugerty, M.K., and M. Kremer. 2004. "The Rockefeller Effect." Evans School of Public Affairs,
University of Washington.
Gulati, A., N. Minot, C. Delgado, and S. Bora (2007) Growth in High-Value Agriculture in Asia
and the Emergence of Vertical Links with Farmers, ed. J.F.M. Swinnen. Wallingford,
UK, CABI, pp. 91-108.
Haddad, L., and J. Maluccio, A. 2002. "Trust, Membership in Group, and Household Welfare:
Evidence from Kwazulu-Natal, South Africa." International Food Policy Research
Institute: Food Consumption and Nutrition Division Discussion Paper No. 135.
Hardin, G. 1968. "The Tragedy of the Commons." Science 162:1243-1248.
Hausman, J.A., and G.K. Leonard. 1997. "Superstars in the National Basketball Association:
Economic Value and Policy." Journal of Labor Economics 15:586-624.
Hawkes, K., J.F. O'Connell, and N.G. Blurton-Jones. 2001. "Hadza Meat Sharing." Evolution
and Human Behavior 22(113-142).
Heckman, J. 1976. "The Common Structure of Statistical Models of Truncation, Sample
Selection, and Limited Dependent Variables, and a Sample Estimator for Such Models."
Annals of Economic and Social Measurement 5:475-492.
Heckman, J.J. 1976. "The common Structure of Statistical Models of Truncation, Sample
Selection and Limited Dependent Variables and Simple Estimator for Such Models."
Annals of Ecomomic and Social Measurement 5:475-492.
Heifner, R., and R. Kinoshita. 1994. "Differences Among Commodities in Real Price Variability
and Drift." Journal of Agricultural Economics Research 45(3):10-20.
Hellin, J., M. Lundy, and M. Meijer. 2009. "Farmer organization, collective action and market
access in Meso-America." Food Policy 34:16-22.
Henrich, J., and F. Gil-White. 2001. "The Evolution of Prestige: Freely Conferred Deference as a
Mechanism for Enhancing the Benefits of Cultural Transmission." Evolution and Human
Behavior 22:165-196.
Hernandez, R., T. Reardon, and J. Berdegué. 2007 "Supermarkets, Wholesalers, and Tomato
Growers in Guatemala." Agricultural Economics 36(3):281-290.

199

Hertberg, R. 2001. "Determinants and Impacts of Local Institutions for Common Resource
Management." Environment and Development Economics 6:183-208.
Heyer, J. (1976) Achievements, Problems, and Prospects in the Agricultural Sector: In
Agricultural Development in Kenya, an Economics Assessment, ed. J. Heyer,J.K. Maitha,
and W.M. Senga. Nairobi, Kenya, Oxford University Press, pp. 1-32.
Hind, A.M. 1994. "Cooperatives-Underformers by Nature? An Exploratory Analysis of
Cooperative and Non-cooperative Companies in the Agribusiness Sector." Journal of
Agricultural Economics 45(2):213-219.
Holloway, G., C.B. Barrett, and S. Ehui. 2005. "The Double-Hurdle Model in the Presence of
Fixed Costs." Journal of International Agricultural Trade and Development 1:17-28.
Holloway, G., Nicholson, C., Delgado, C., Staal, S., Ehui, S. . 2000. "Agro-industrialization
Through Institutional Innovation Transaction Costs, Cooperatives and Milk-Market
Development in The East-African Highlands." Agricultural Economics 23:279-288.
Howes, M. 1997. "NGOs and the Institutional Development of Membership Organizations: A
Kenyan Case." The Journal of Development Studies 33(6):820-848.
Hulme, D., and A. Shepherd. 2003. "Conceptualizing Chronic Poverty." World Development
3(3):403-423.
ICRISAT-Website. (2006) International Crops Research Institute for the Semi-Arid Tropics,
http://www.icrisat.org/index.htm.
ICRISAT. "2005 Highlights." ICRISAT (International Crops Research Institute for the SemiArid Tropics) Eastern and Southern Africa Region.
Idiong, I.C. 2007. "Estimation of Farm Level Technical Efficiency in Smallscale Swamp Rice
Production in Cross River State of Nigeria: A Stochastic Frontier Approach." World
Journal of Agricultural Sciences 3(5):653-658.
Jaffee, S., and J. Morton. 1995. Private Sector High-Value Food Processing and marketing: A
synthesis of African Experience. Dubuque: Kendal/ Hunt Publishing Company.
Janky, B., and K. Takács. 2009. "Social Control, Participation in Collective Action and Network
Stability." Budapest University of Technology and Economics:34.
Jayne, T.S., J. Govereh, A. Mwanaumo, J.K. Nyoro, and A. Chapoto. 2002. "False Promise or
False Premise? The Experience of Food and Input Market Reform in Eastern and
Southern Africa." World Development 30(11):1967-1985.
Jayne, T.S., Jones, S. 1997. "Food Marketing and Pricing Policy and Southern Africa: a Survey."
World Development 25(9):1505-1527.
200

Jones, A.M. 1989. "A Double-Hurdle Model of Cigarette Consumption." Journal of Applied
Econometrics 4:23-39.
Joppke, C. 1986. "The Cultural Dimensions of Class Formation And Class Struggle: On The
Social Theory of Pierre Bourdieu." Berkeley Journal of Sociology 31:53-78.
Jordaan, H., B. Grové, A. Jooste, and Z.G. Alemu. 2007. "Measuring the Price Volatility of
Certain Field Crops in South Africa using the ARCH/GARCH Approach." Agrekon
46(3):1-17.
Joskow, P. 1985. "Asset Specificity and the Structure of Vertical Relationships: Empirical
Evidence." Journal Of Law, Economics, and Organization 4(1):95-117.
Just, R.E., and R.D. Pope. 2002. A Comprehensive Assessment of the Role of Risk in US
Agriculture. Boston: Kluwer Academic Publishers.
Kariuki, G., and F. Place. 2005. "Initiatives for Rural Development Through Collective Action:
The Case of Household Participation in Group Activities in the Highlands of Central
Kenya." CAPRi Working Paper No. 43; Washignton, D.C.
Key, N., and D. Runsten. 1999. "Contract Farming, Smallholders, and Rural Development in
Latin America: The Organization of Agroprocessing Firms and the Scale of Outgrower
Production." World Development 27(2):381-401.
Key, N., E. Sadoulet, and A. de Janvry. 2000. "Transactions Costs and Agricultural Household
Supply Response." American Journal of Agricultural Economics 82(May):245-259.
Kindness, H., and A. Gordon. 2001. "Agricultural Marketing in Developing Countries: The Role
of NGOs and CBOs." Social and Economic Development Department, Natural Resources
Institute, University of Greenwich (13).
Knack, S., and P. Keefer. 1997. "Does Social Capital have an Economics Payoff? A CrossCountry Investigation." Quarterly Journal of Economics CXII(4):1251-1288.
Koenker, R., and K.F. Hallock. 2001. "Quantile Regression." Journal of Economic Perspectives
15(4):143-156.
Kydd, J., and A. Dorward (2003) Implications of Market and Coordination Failures for Rural
Development in Least Developed Countries. Strathclyde University, Glasgow.
Kydd, J., A. Dorward, J. Farrington, and P. Deshingkar. 2004. "Making Agricultural Markets
Work for the Poor." Renewable Natural Resources and Agriculture , DFID (Policy
Division).

201

Laidlaw, A. (1978) Cooperatives and the poor: a review from within the cooperative movement,
ed. Report-of-an-Experts’-Consultation. London, International Cooperative Alliance.
Lapar, M., Lucila, A., T. Binh, Vu, N. Son, Tuan, M. Tiongco, Jabbar, and S. Staal. 2006. "The
Role of Collective Action in Overcoming Barriers to Market Access by Smallholder
Producers: Some Empirical Evidende from Northern Vietnam." Paper presented at
Research Workshop on Collective Action and Market Access for Smallholders. Research
Workshop on Collective Action and Market Access for Smallholders, Cali, Colombia, 25 October 2006.
Latruffe, L., K. Balcombe, S. Davidova, and K. Zawalinska. 2002. "Determinants of technical
efficiency of crop and livestock farms in Poland." Institut National de la recherche
Agronomique.
Leathers, H.D. 2006. "Are Cooperatives Efficient When Membership is Voluntary?" Journal of
Agricultural and Resource Economics 31(3):667-676.
Lele, U., and R.E. Christiansen. 1989. "Markets, Marketing Boards and Cooperatives in Africa,
Issues in Adjustment Policy." Paper presented at MADIA Discussion Paper. Washington
D.C.
Lerman, Z., and C. Parliament. 1990. "Comparative Performance of Cooperatives and Investorowned Firms in US Food Industries." Agribusiness 6.
Leung, S.F., and S. Yu. 1996. "On the Choice Between Sample Selection and Two-Part Models."
Journal of Econometrics 72:197-229.
Lin, J.Y. 1991. "Education and Innovation Adoption in Agriculture: Evidence from Hybrid Rice
in China." American Journal of Agricultural Economics 73(3):713-723.
Lindenberg, S. (1986) The Paradox of Privatization in Consumption, ed. A. Diekmann, and P.
Mitter. Heidelberg, Physica, pp. 341.
Liu, Z., and J. Zhuang. 2000. "Determinants of Technical Efficiency in Post-Collective Chinese
Agriculture: Evidence from Farm-Level Data." Journal of Comparative Economics
28:545-564.
Lowndes, V., and C. Skelcher. 1998. "The Dynamics of Multi-Organizational Partnerships: An
Analysis of Changing Modes of Governance." Public Administration 76:313-333.
Maddala, G.S. 1983. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge,
United Kingdom: Cambridge University Press.
Madden, D. 2006. "Sample Selection Versus Two-Part Models Revisited: The Case of Female
Smoking and Drinking." Health Econometrics and Data Group, Working Paper 06/12.
The University of York.
202

Maluccio, J.A., L. Haddad, and J. May. 2000. "Social Capital and Household Welfare in South
Africa 1993-1998." Journal of Development Studies 36(6):54-81.
Manski, C. 1973. "The Analysis of Quantitative Choice." Massachusetts Institute of Technology.
Marshall, G. 1998. A Dictionary of Sociology. New York: Oxford University Press.
Marwell, G., P. Oliver, and R. Prahl. 1988. "Social Networks and Collective Action: A Theory of
the Critical Mass. III." The American Journal of Sociology 94(3):502-534.
Marxwell, G., and P. Oliver. 1993. The Critical Mass in Collective Action: A Micro-Social
Theory. Cambridge: Cambridge University Press.
Masakure, O., and S. Henson. 2005. "Why do Small-Scale Producers Choose to Produce Under
Contract? Lessons From Nontraditional Vegetable Exports From Zimbabwe." World
Development 33(10):1721-1733.
Maynard Smith, J. 1964. "Group Selection and Kin Selection." Nature (201):1145-1147.
McKee, G. 2008. "The Financial Performance of North Dakota Grain Marketing and Farm
Supply Cooperatives." Journal of Cooperatives 21:15-34.
Meinzen-Dick, R., A. Knox, and M. Di Gregorio. 2001. "Collective Rights, Property Rights and
Devolution of Natural Resource Management: Exchange of Knowledge and Implications
for Policy." Paper presented at A Workshop Summary Paper. International Food Policy
Research Institute, Washington D.C.
Meinzen-Dick, R., A. Knox, and B. Swallow. 2002. "Innovations in Natural Resource
Management: The Role of Property Right and Collective Action in Developing
Countries." International Food Policy Research Institute.
Meinzen-Dick, R.S., and M. Di-Gregorio. 2004. "Collective Action and Property Rights for
Sustainable Development." International Food Policy Research Institute, 2020 Focus
II(2).
Michael, E.P. 2003. "Pro-Community Altruism and Social Status in a Shuar Village." Human
Nature 14(2).
Miehlbradt, A.O., and M. McVay. 2005. From BDS to Making Markets Work for the Poor.
Geneva, Switzerland: International Labour Organization.
Miguel, E., and M.K. Gugerty. 2004. "Ethnic Diversity, Social Sanctions and Public Goods in
Kenya." Mimeo.

203

Moledina, A.A., T.L. Roe, and M. Shane. 2003. "Measurement of Commodity Price Volatility
and the Welfare Consequences of Eliminating Volatility." Working Paper at the
Economic Development Centre, University of Minnesota.
Moser, C., M, and C. Barrett, B. 2006. "The Complex Dynamics of Smallholder Technology
Adoption: The Case of SRI in Madagascar." Agricultural Economics 35(3):373-388.
Motiram, S., and V. Vakulabhranam. 2007. "Corporate and Cooperative Solutions for the
Agrarian Crisis in Developing Countries." Review of Radical Political Economics
39(3):360-367.
Mude, A., G. 2006. "Imperfection in Membership Based Organizations for the Poor: An
Explanation for The Dismal Performance of Kenya's Coffee Cooperatives (draft)."
Cornell University: Strategies and Analysis for Growth and Access, Policy Brief.
Munkner, H. 1992. African Cooperatives and The State in the 1990s: In Yearbook of
Cooperative Enterprise 1992. Oxford, UK: Plunkett Foundation.
Murdoch, J.C., T. Sandler, and W.P.M. Vijverberg. 2003. "The Participation Decision Versus the
Level of Participation in an Environmental Treaty: a Spatial Probit Analysis." Journal of
Public Economics 87:337-362.
Narayan, D., and L. Pritchett. 1999. "Cents and Sociability: Household Income and Social
Capital in Rural Tanzania." Economic Development and Cultural Change 47(4):871-897.
Negri, M., and G.G. Porto. 2008. "Burley Tobacco Clubs in Malawi: Nonmarket Institutions For
Exports." The World Bank Development Research Group - Trade Team: Policy Research
Working Paper 4561 (4561).
Ngugi, I.K., and D.K. Kariuki. 2009. "Correlates of Group-Membership by Small-Scale Farm
Entrepreneurs in Kenya." African Journal of Business Management 3(1):1-8.
Niroula, G.S., and G.B. Thapa. 2006. "Impacts of Land Fragmentation on Input Use, Crop Yield,
and Production Efficiency in the Mountains of Nepal." Land Degradation &
Development 18:237-248.
Nyoro, J.K., and I. Komo. 2005. "An Analysis of Success, Failure and Demand Factors of
Agricultural Cooperatives in Kenya." Strategies and Analysis fro Growth and Access,
Policy Brief, Cornell University.
Obare, G., B. Shiferaw, and G. Muricho. 2006. "Leveraging Rural Institutions for Collective
Action to Improve Markets for the Poor: Lessons and Policy Options." International
Crops Research Institute for the Semi-Arid Tropics: Strategic Assessments and
Development pathways for agriculture in the Semi-Arid Tropics, Policy Brief No. 8, .

204

Oberschall, A. 1994. "Rational Choice in Collective Protests." Rationality and Society (6):79100.
Offutt, S.E., and D. Blandford. 1986. "Commodity Market Instability : Empirical Techniques for
Analysis." Resource Policy 12(1):62-72.
Okello, J., J. , and S. Swinton, M. . 2006. "Compliance with International Food Safety Standards
in Kenya’s Green Bean Industry: Comparison of a Smalland a Large-scale Farm
Producing for Export." Review of Agricultural Economics 29(2):269-285.
Oliver, P., and G. Marxwell. 1988. "The Paradox of Group Size in Collective Action: A Theory
of Critical Mass, II." American Sociological Review 53(February):1-8.
Olson, M. 1965. The Logic of Collective Action: Public Goods and the Theory of Groups.
Cambridge, Mass.: Harvard University Press.
---. 1982. The Logic of Collective Action: Public Goods and the Theory of Groups. Cambridge,
MA: Harvard University Press.
Omamo, S.W. (2005) The Future of Smallholder Agriculture in Eastern Africa-The Roles of
States, Markets and Civil Society. Imperial Resort Beach Hotel, Entebbe, Uganda, IFPRI.
Ortmann, G.F., and R.P. King. 2007. "Agricultural Cooperatives I: History, Theory and
Problems." Agrekon 46(1).
Ostrom, E. 1990. Governing the commons: The evolution of institutions for collective action:
Political Economy of Institutions and Decisions series Cambridge; New York and
Melbourne: Cambridge University Press.
Papke, L.E., and J.M. Wooldridge. 1996. "Econometric methods for fractional response variables
with an application to 401(k) plan participation rates." Journal of Applied Econometrics
11(6):619-632.
---. 2008. "Panel data methods for fractional response variables with an application to test pass
rates." Journal of Econometrics 145(1-2):121-133.
Peacock, C., A.Jowett, A.Dorward, C.Poulton, and I.Urey. "Reaching the Poor, A Call to Action:
Investment in Smallholder Agriculture in Sub Saharan Africa." FARMAfrica, Harvest
Help and Centre for Development and Poverty Reduction, Imperial College, London
(2004).
Pender, J., and S. Scherr. 1999. "Organizational Development and Natural Resource
Management: Evidence from Central Honduras." International Food Policy Research
Institute (IFPRI): Environmental and Production Technology Division Discussion Paper
11. :59.

205

Peterson, C., H., A. Wysocki, F., and S. Harsh, B. 2001. "Strategic Choice Along The Vertical
Coordination Continuum." International Food and Agribusiness management Review
4:149-166.
Pingali, P., Y. Khwaja, and M. Meijer. 2005. Commercializing Small Farms: Reducing
Transaction Costs. Rome, Italy: Agricultural and Development Economics Division,
FAO.
Place, F., A. Kristjanson, A. Makauki, G. Kariuki, and J. Wangila. 2004. "Assessing The Factors
Underlying Differences in Achievements of Farmer Groups: Methodological Issues and
Empirical Findings From the Highlands of Central Kenya." Agricultural Systems
82(3):257-272.
Platt, J. 1973. "Social Traps." American Psychologist August:641-651.
Putnam, R., D. 1993. Making Democracy Work. Civic traditions in modern Italy. Princeton:
Princeton University Press.
---. 2001. "Social Capital: Measurement and Consequences." Canadian Journal of Policy
Research 2(1):41-51.
Rao, J.M., and S. Storm (2003) Agricultural globalization in developing countries: Rules,
rationale and results, ed. J. Ghosh, and C.P. Chandrasekhar. New Delhi, India, Tulika
Publishers.
Rasmussen, L.N., and R. Meinzen-Dick. 1995. "Local Organizations for Natural Resource
Management: Lessons form Theoretical and Empirical Literature." International Food
Policy Research Institute (IFPRI): Environment and Production Technology Division
Discussion Paper 11. :37.
Ravallion, M. (2006) Evaluating Anti-Poverty Programs. Forthcoming in Handbook of
Development Economics, ed. R. Evenson, E, and P. Schultz, T, vol. 4. Amsterdam,
North-Holland.
Reisman, D. 1990. Theories of Collective Action. New York: St. Martin's Press.
Ribot, J. 1998. "Theorizing Access: Forest Profits Along Senegal's Charcoal Commodity Chain."
Development and Change 29:307-341.
Rice, R.E. (1993) Using Network Concept to Clarify Sources and Mechanism of Social
Influence. (In the book)Progress in Communication Sciences, ed. W.D. Richards, and
G.A. Barnett, vol. XII. Norwood, NJ, Ablex, pp. 44-62.
Rosenbaum, P., and D. Rubin. 1983. "The Central Role of the Propensity Score in Observational
Studies for Causal Effects." Biometrika 70:41-55.

206

Ruben, R. 1997. "Making cooperatives work - Contract choice and resource management within
land reform cooperatives in Honduras." Wageningen University.
Ruben, R., and Z. Lerman. 2005. "Why Nicaraguan Peasants Stay in Agricultural Production
Cooperatives." Revista Europea de Estudios Latinoamericanos y del Caribe 78:17.
Sandler, T. (1992) Collective Action: Theory and Applications. Ann Arbor, Michigan University
Press.
Scherr, S.J., J. Amornsanguasin, M.E. Chiong-Javier, D. Garrity, S. Sunito, and I. Saharuddin.
2001. "Local Organizations in Natural Resource Management in The Uplands of
Southeast Asia: Policy Context and Institutional Landscape." Paper presented at
SANREM Conference on “Sustaining Upland Development in Southeast Asia: Issues,
Tools and Institutions for Local Natural Resource Management”. Makati, Metro Manila,
Philippines, May 28-30, 2001.
Schmid, A.A. 1989. Benefit-Cost Analysis: A Political Approach. Boulder: Westview Press.
---. 2004. Conflict and Cooperation : Institutional and Behavioral Economics. Malden, MA:
Blackwell Publishing.
Sekher, M. 2001. "Organized Participatory Resource Management: Insights From Community
Forestry Practices in India." Forestry Policy and Economics 3(4):255-279.
Sexton, R.J., and J. Iskow. 1988. "Factors Critical to the Success or Failure of Emerging
Agricultural Cooperatives." Giannini Foundation of Agricultural Economics, University
of California, Davis ( 88-3).
Sexton, R.J., and J. Iskow. 1993. "What Do We Know About the Economic Efficiency of
Cooperatives: An Evaluative Survey." Journal of Agricultural Cooperation 8:15-27.
Shiferaw, B., T. Kebede, and V.R. Ratna. 2008. "Community Watershed Management in SemiArid India." CAPRi WORKING PAPER (85):42.
Shiferaw, B., G. Obare, and G. Muricho. 2005. "Rural Institutions and Producer Organizations in
Imperfect Markets: Experiences from Producer Marketing Groups in Semi-Arid Eastern
Kenya." International Crops Research Institute for the Semi-Arid Tropics.
Shiferaw, B., G. Obare, and G. Muricho. 2008. "Rural Market Imperfections and The Role of
Institutions in Collective Action to Improve Markets for The Poor." Natural Resources
Forum 32:25-38.
Siisiäinen, M. 2000. "Two Concepts of Social Capital: Bourdieu vs. Putnam." Paper presented at
ISTR Fourth International Conference "The Third Sector: For What and for Whom?".
Trinity College, Dublin, Ireland, July 5-8, 2000.

207

Simmons, R., and J. Birchall. 2008. "The role of co-operatives in poverty reduction: Network
perspectives." The Journal of Socio-Economics 37:2131-2140.
Smith, E.A., and R.L. Bliege Bird. 2000. "Turtle Hunting and Tombstone Opening: Public
Generosity as Costly Signaling." Evolution and Human Behavior 21:245–261.
Soedjono, I. 2002. "Features and Characteristics of a Supportive Environment for Cooperatives."
Paper presented at Expert Group Meeting on “Supportive Environment for Cooperatives:
A Stakeholder Dialogue on Definitions, Prerequisites and Process of Creation”.
Ulaanbaatar, Mongolia, 15-17 May 2002.
Sosis, R. 2000. "Costly Signaling and Torch Fishing on Ifaluk Atoll." Evolution and Human
Behavior 21:223-244.
Staal, S., Delgado, C., Nicholson, C. 1997. "Smallholder Dairy under Transactions Costs in East
Africa." World Development 25(5):779-794.
Staatz, J.M. 1984. "A Theoretical Persepective on the Behaviour of Farmers' Cooperatives."
Dissertation Michigan State University.
Stiles, K. 2002. "International Support for NGOs in Bangladesh: Some Unintended
Consequences." World Development 30(5):835-846.
Stockbridge, M., A. Dorward, and J. Kydd. 2003. Farmer Organizations for Market Access: A
Briefing Paper. London, UK: Wye College, University of London.
Stringfellow, R., J. Coulter, T. Lucey, C. McKone, and A. Hussain. 1997. Improving the Access
of Smallholders to Agricultural Services in Sub-Saharan Africa: Farmer Cooperation
and the Role of the Donor Community. London, UK: Overseas Development Institute.
UNDP. "Kenya National Human Development Report 2006 Human Security and Human
Development: A Deliberate Choice." United Nation Human Development Program.
Vakis, R., E. Sadoulet, and A. De Janvry. 2003. "Measuring Transactions Costs from Observed
Behavior: Market Choice in Peru." University of California Berkeley: Department of
Agricultural & Resource Economics, Working paper (962).
Valente, T., W. 1995. Network Models of the Diffusion of Innovations. Cresskill, New Jersey:
Hampton Press, INC.
Van-der-Walt, L. 2005. "The resuscitation of the cooperative sector in South Africa." Paper
presented at International Co-operative Alliance XXI International Cooperative Research
Conference. Cork, Ireland, August 11-14, 2005.
Von-Pischke, J.D., and J. Rouse, G. 2004. "New Strategies for Mobilizing Capital in Agricultural
Cooperatives." Food and Agriculture Organization of the United Nations.
208

Wagner, J. 2003. "Unobserved Firm Heterogeneity and the Size-Exports Nexus: Evidence From
German Panel Data." Review of World Economics 139:161-172
Whittington, D., X. MU, and R. Roche. 1990. "Calculating the Value of Time Spent Collecting
Water: Some Estimates for Ukunda, Kenya." World Development 18:269-280.
Williamson, O.E. 1985. The Economic Institutions of Capitalism. New York: Free Press.
Williamson, O.E. 1986. Economic Organization: Firms, Markets, and Policy Control. New
York: New York University Press.
Williamson, O.E. 1975. Markets and Hierarchies: Analysis and Antitrust Implications. New
York: NY: Free Press.
---. 1996. The Mechanisms of Governance. New York: Oxford University Press.
---. 1979. "The Transaction Costs Economics: The Governance of Contractual Relations."
Journal of Law and Economics 22:233-261.
Wolf, T. 1986. "States Intervention at the Cabbage-Roots: A Case Study from Kenya." IDS
Bulletin 17(1):47-50.
Woolcock, M. 1998. "“Social Capital and Economic Development: Toward a Theoretical
Synthesis and Policy Framework." Theory and Society 72(2):151-208.
Wooldridge, J., M. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge,
Massachusetts: The MIT Press.
World-Bank. 2002. "From Action to Impact: The Africa Region’s Rural Strategy." The World
Bank, Rural Development Operations, the Africa Region, Washington, DC.
---. "World Development Report 2008." The World Bank (2008).
Wysocki, A., F., C. Peterson, H., and S. Harsh, B. 2003. "Quantifying Strategic Choice Along
the Vertical Coordination Continuum." International Food and Agribusiness
management Review 6(3):2003.
Yin, R., K. 1994. Case Study Research: Design and Methods. Thousand Oaks, California: Sage
Publications, Inc.
Zander, A. 1994. Making Groups Effective. San Francisco: Josey-Bass Publishers.
Zeller, M., A. Diagne, and C. Mataya. 1997. "Market Access by Smallholder Farmers in Malawi:
Implications for Technology Adoption, Agricultural Productivity, and Crop Income."
International Food Policy Research Institute 35.
209