Received: 3 May 2018 Revised: 2 July 2019 Accepted: 5 July 2019

DOI: 10.1111/agec.12509

ORIGINAL ARTICLE

Big is efficient: Evidence from agricultural cooperatives

in Ethiopia
Tafesse W. Gezahegn1,2 Steven Van Passel3 Tekeste Berhanu2 Marijke D’Haese4
Miet Maertens1

1 Departmentof Earth and Environmental

Abstract
Sciences, KU Leuven, Belgium
2 College
of Business and Economics,
In Ethiopia, there is a renewed interest in agricultural cooperatives as an institutional
Mekelle University, Ethiopia tool to improve the welfare of smallholder farmers. One of the pathways through
3 Department of Engineering Management, which cooperatives benefit their members is scale economies. However, the estab-
University of Antwerp, Belgium
lishment of cooperatives in Ethiopia seems to pay little attention to the size of the
4 Departmentof Agricultural Economics,
organizations. This article aims at investigating the effect of size on cost efficiency
Ghent University, Coupure Links, Belgium
of agricultural cooperatives. More specifically, the purpose is to examine whether
Correspondence a single cooperative can serve a given number of farmers at a lower cost than two
Tafesse W. Gezahegn, College of Business and
Economics, Mekelle University, Ethiopia.
or more smaller cooperatives could. We employ the concept of cost subadditivity to
Email: [email protected] compare the cost efficiency of large versus small cooperatives, and by extension uni-
lateral actions. We estimate a flexible production technology using cross-sectional
Funding information
VLIR-UOS-ZEIN2015PR406 (13V95615T)
cooperative-level data. Findings show that costs would drop by 78% to 181% if farm-
PROGRAMME (TEAM), Belgium ers join hands in relatively large rather than small cooperatives.

KEYWORDS
Agricultural cooperatives, cost efficiency, cost subadditivity, Ethiopia, scale economies

J E L C L A S S I F I CAT I O N
J54, O13, Q13

1 I N T RO D U C T I O N contract farming (Verhofstadt & Maertens, 2015). They

enable small farmers to aggregate production, capture the ben-
Cooperatives are an organizational form for social and eco- efits of downstream economies of scale, and reduce trans-
nomic development. Their role is increasing in response to action costs (Holloway, Nicholson, Delgado, Staal, & Ehui,
market imperfections (Getnet & Anullo, 2012). In develop- 2000; Ito, Bao, & Su, 2012; Reardon, Barrett, Berdegué,
ing countries, especially in areas with poor infrastructure, & Swinnen, 2009; Saitone, Sexton, & Malan, 2018; Soboh,
family farms are disadvantaged by various forms of market Lansink, Giesen, & Van Dijk, 2009). Essentially, farmers form
imperfections. Smallholders face high transaction costs that cooperatives to generate larger profits or incomes by obtain-
significantly reduce their incentives for market participation ing inputs and services at lower costs than they could obtain
(Poulton, Dorward, & Kydd, 2010). There is a renewed inter- elsewhere, and by marketing their products at better prices or
est in cooperatives as an institutional tool to improve market in markets that are not accessible for individual farmers (Ort-
participation of smallholder farmers, increase farm incomes, mann & King, 2007). A farmer is inclined to join a cooperative
and reduce rural poverty (Bernard & Taffesse, 2012; Fis- if the expected return is higher than when producing alone.
cher & Qaim, 2012; Verhofstadt & Maertens, 2015). Coop- A large body of literature provides evidence on the rel-
eratives are often thought to be more inclusive and poverty ative advantage cooperative members have over nonmem-
reducing than other types of institutional innovations, such as bers (e.g., Abebaw & Haile, 2013; Chagwiza, Muradian, &
Agricultural Economics. 2019;50:555–566. wileyonlinelibrary.com/journal/agec © 2019 International Association of Agricultural Economists 555
556 GEZAHEGN ET AL.

Ruben, 2016; Fischer & Qaim, 2012; Ito et al., 2012; Ma & but no study has applied it to the agricultural sector. In this
Abdulai, 2016; Ma & Abdulai, 2018; Ma, Abdulai, & Goetz, study, we use this approach to examine economies of scale in
2017; Mojo, Fischer, & Degefa, 2017; Vandeplas, Minten, the joint provision of services to farmers; that is, whether the
& Swinnen, 2013; Verhofstadt & Maertens, 2014; Wossen cost of service provision is subadditive. Second, while previ-
et al., 2017). These studies mainly highlight a positive impact ous studies keep only output constant in testing for cost sub-
of cooperatives on different performance indicators, such as additivity, this study adapts the approach to cooperatives by
farm income, farm profits, producer prices, poverty, mar- keeping both membership size and output (service) constant
ket participation, and technology adoption. These observed to capture the cost implications thereof. Another contribu-
effects are attributable to various possible pathways through tion of the article is the use of a flexible model that allows
which cooperatives can benefit their members. One such for the estimation of cooperative type–specific technologies.
channel is exploiting economies of scale and reducing the cost Policy wise, findings provide insights on whether agricul-
of operation (Wossen et al., 2017). Several conceptual studies tural cooperatives should grow in size—via merger and/or
highlight the cost advantage of size in agricultural coopera- open membership—or in number—via split and/or closed
tives (e.g., Barton, Schroeder, & Featherstone, 1993; Fulton membership—to harness underutilized opportunities for agri-
& King, 1993; Trechter, 1996). Yet, only a few studies (Arcas, cultural growth and development in Ethiopia. Results serve as
García, & Guzmán, 2011; Lerman & Parliament, 1991) an ex ante empirical investigation of size–performance rela-
empirically investigate the effect of size on performance. tionships before steps to establish, merge, or split cooperatives
In this article, we focus on size and empirically examine are taken, and pave the way for economic efficiency (and not
whether it is cost efficient for farmers to organize in rela- politics) to guide cooperative policies.
tively small or large agricultural cooperatives, or act unilat-
erally. While cost efficiency may not capture overall eco-
nomic efficiency, it is an important objective of cooperatives 2 SA M P L I NG A N D DATA
to minimize cost of service delivery to their members. We
focus on determining whether a single cooperative can pro- For this study, the six zones in Tigray (excluding Mekelle
vide a given level of service at a lower cost than two or more zone) were categorized into four zones. A multistage ran-
smaller cooperatives could, and, by extension, than unilat- dom sampling technique was used to select 511 coopera-
eral actions. Our focus is complementary to studies on the tives from the four zones. In the first stage, we randomly
impact of cooperative membership as it allows to understand selected 12 districts, three from each zone (Figure A1 in the
how efficient cooperatives are in creating member benefits. online appendix). In the second stage, 223 tabias (the smallest
In addition, this focus is particularly relevant in our study administrative unit) were randomly selected from each district
area where cooperatives are very widespread—with 98% of proportionately, accounting for the type and number of agri-
villages having at least one cooperative (Bernard, Abate, & cultural cooperatives. In the third stage, using the probability
Lemma, 2013). In this context, comparing the performance proportion to size technique, we apportioned the 511 coop-
of cooperative members and nonmembers is less relevant (and eratives among the 223 tabias, based on the number and the
difficult as a good control group is lacking). Policy decisions type of cooperatives in each tabia. Finally, using the list of
regarding the size of agricultural cooperatives, whether by tabias and the number of each type of cooperative, we ran-
governments, nongovernmental organizations (NGOs), com- domly selected 511 cooperatives.
munities, or cooperatives themselves, require information on A semistructured questionnaire was used for the survey.
size–cost relationships and the advantages of size (Arcas et al., We conducted the data collection using Qualtrics survey soft-
2011). In addition, the cooperative landscape in Tigray is very ware. In addition to the questionnaire, we also extensively
heterogeneous with the remains of the pre-1991 government- reviewed cooperative bylaws, audit reports, and periodic
controlled cooperative system and new (post-1991) bottom- activity reports (including financial statements and strate-
up collective action initiatives (Bernard et al., 2013). This gic plan documents). Prior to conducting the actual sur-
implies that the efficiency of cooperatives—and how it relates vey in April to August 2017, a pretest was carried out
to size—is important in this region. with 65 randomly selected cooperatives. In the final sam-
This study contributes some methodological innovations. ple, multipurpose cooperatives (MPC) account for 35.23%,
First, it introduces the Evans and Heckman (1984) test for cost cattle fattening cooperatives for 4.7%, beekeeping coopera-
subadditivity to analyze the presence of incentives for coop- tives (BKC) for 24.66%, sheep and goat fattening cooperatives
eration in agriculture in the context of a developing country. for 5.87%, dairy cooperatives for 4.11%, irrigation coopera-
Several studies (e.g., Adhikari & Guru-Gharana, 2014; Ivaldi tives for 21.72%, and forest and grass cooperatives for 3.72%.
& McCullough, 2008; Sánchez, 2000) apply this approach Geographically, 26% of the cooperatives are from the East-
to electric, railway, and banking industries (to analyze cost ern zone, 23.48% from the Central zone, 25.79% from the
savings from splitting a company into two or more entities), South and Southeastern zone, and 24.74% from the West and
GEZAHEGN ET AL. 557

Northwestern zone, indicating that sampled cooperatives are determines whether an output can be produced at a lower
almost uniformly distributed across zones. cost by a single larger firm (cooperative) than by any group
A detailed overview of the surveyed cooperatives and their of smaller ones (Ivaldi & McCullough, 2008). Evans and
characteristics is provided in the online appendix. For the Heckman (1984) develop a local test (that does not require
purpose of this article, we classify the cooperatives in four global information on cost functions) for cost subadditivity,
broader categories: (1) MPC; (2) livestock and dairy cooper- which avoids the need to extrapolate the estimated cost
atives (LDC), including cattle fattening, sheep & goat fatten- function outside the available data in the sample. A cost
ing, and dairy cooperatives; (3) BKC; and (4) natural resource function 𝐶(𝑞) is subadditive at the output level 𝑞 if:
cooperatives (NRC), including irrigation cooperatives and 𝑛
∑
forest cooperatives. 𝐶(𝑞) < 𝐶(𝑞𝑖 ), (1)
𝑖=1
𝑛
∑
𝑞𝑖 = 𝑞, (2)
3 METHODOLOGY 𝑖=1

3.1 Conceptual framework where 𝑞𝑖 ≥ 0, with at least two vectors nonzero for all 𝑞𝑖 satis-
fying (2), and 𝑛 = the number of firms (cooperatives). Measur-
The most important reason for cooperative formation is ing cost subadditivity entails comparing the actual cost struc-
economies of scale, which farmers cannot realize individu- ture of the existing firm with the cost structure that would
ally (Cazzuffi & Moradi, 2012). Organizational economics apply to an n-firm (hypothetical) configuration of the indus-
explains agricultural cooperatives through their ability to try (Gordon, Gunsch, & Pawluk, 2003). To accomplish this,
economize on transaction costs and develop countervailing Evans and Heckman (1984) determine an admissible output
power (Bonus, 1986; Hansmann, 1988; Staatz, 1987; Valenti- region of the new configuration by imposing the restriction
nov, 2007). Increasing returns to scale (IRS) imply that larger that the n-firm output vectors be within the range of output
firms are more competitive. However, large firms (groups) vectors actually observed in the data. This implies that no
may face coordination problems that lead to inefficient use firm in the alternative configuration of the industry is per-
of resources. Smaller groups may be better able to overcome mitted to produce less (more) output than the lowest (high-
this problem, as better information and social sanctions help est) observed output level in the sample. To meet this, they
to ensure cooperation and offset negative effects from profit impose the constraint that output levels (of the hypothetical
sharing and free riding. Larger groups may find this more n-firm scenario) be at least twice the minimum and at most
difficult as monitoring members’ commitment is more costly the maximum observed output in the sample. The Evans and
and social sanctions less effective. At a certain point, group Heckman (1984) approach has been applied to the electric,
interaction problems may outweigh gains from economies railway, and banking industries to test for cost subadditiv-
of scale, resulting in an inverted-U shaped relation between ity by Adhikari and Gur-Gharana (2014), Ivaldi and McCul-
membership size and efficiency (Cazzuffi & Moradi, 2012). lough (2008), Sanchez (2000), and Hunter, Timme, and Yang
This theoretical size–performance relationship motivates (1990). Applying insights from these studies, we adapt the
our question: is the current average size of agricultural approach to the concept of cooperation in agriculture. In the
cooperatives in the study area smaller than what would be context of farmer cooperatives, subadditivity of a cost func-
efficient in terms of cost? tion means that it is cheaper to provide a given level of service
In a joint action where the objective is cost minimization, to a group of members as a whole than to smaller subgroups or
the main condition imposed on the cost function is subad- individuals. It is subadditivity of the cost function that makes
ditivity (Zakharov & Shirokikh, 2018), a property in which joint provision of a service to a group more economical than
the total cost of any coalition is not larger than the sum of providing the same level of service to subunits of the group
the costs of any disjoint partitions (Kimms & Kozeletskyi, (Staatz, 1985).
2016a) or singletons. Subadditivity is a necessary condition
for cost minimization (Zakharov & Shirokikh, 2018), and is
3.2 Application to cooperatives in Ethiopia
important to ensure that the grand coalition (a larger cooper-
ative) can provide a better outcome for its members (Kimms In this study, we model agricultural cooperatives as service1
& Kozeletskyi, 2016b) than two or more smaller ones. If providers that seek to minimize costs subject to the prices of
the cost function is subadditive, there is an incentive to form labor, capital, and material inputs, the prevailing production
larger cooperatives (Chinchuluun, Pardalos, Migdalas, &
Pitsoulis, 2008). Subadditivity of a cost function can be inter- 1 Servicerefers to supply of tangible products (farm inputs, farm tools, and
preted as economies of scale or scope (Drechsel, 2010). But, consumer goods) and does not include provision of intangible services (e.g.,
it is subadditivity—and not scale or scope—which finally market information).
558 GEZAHEGN ET AL.

technology, and the level of service they provide. Following argues. Implicitly, the cost function (of the N-coop) is repre-
the value added criterion that regards outputs as items that sented as:
entail operating costs (Berger & Humphrey, 1991; Park &
Meyer, 1994), output is proxied by the level of sales. A nec-
𝐶 = 𝐶(𝑞, 𝑝, 𝑁; 𝛽), (5)
essary and sufficient condition for cost subadditivity is that
cooperatives used in the cost estimation have at least twice the
minimum and at most the maximum observed membership where 𝑞 = sales; 𝑝 = a vector of prices of labor (𝑝𝑙 ), capi-
size in the sample. Since our sampled cooperatives differ sub- tal (𝑝𝑘 ), and materials (𝑝𝑚 ); N = number of members; 𝛽 = a
stantially in membership size, we treat each cooperative type vector of parameters to be estimated. To estimate the costs
separately when imposing the minimum and maximum mem- of the M-coop and the R-coop, N will be replace by M and
bership restrictions. We split up the number of members of R, respectively in the predicted model. Empirical estimation
each cooperative into two parts, thereby creating three cooper- of equation (5) entails the choice of an appropriate functional
atives of different sizes: (1) larger cooperative (N-coop) with form and decision on a model of the underlying production
N number of members, where N equals at least twice the min- technology. In this study, the translog functional form, which
imum membership size observed in the sample; (2) smaller has the advantage of flexibility with few restrictions on its
cooperative (M-coop) with M number of members, where M input data, is used. The translog specification has been exten-
equals the minimum membership size observed in the sample; sively used as it has proven to be the most flexible form in
(3) a hypothetical complementary2 cooperative (R-coop) with bridging the gap between theoretical and empirical research.
R number of members, where R equals the residual member- It relaxes, for example, the restrictions that inputs be only sub-
ship size (i.e., R = N − M). After estimating the parameters of stitutes (Cobb-Douglas) or substitutes and/or complements
the cost function using the observed data of the N-coop, the (CES) on a constant basis, and allows the degree of comple-
cost of the N-coop is compared against the sum of the costs of mentarities and substitution to be different between different
the M-coop and the R-coop. Since cooperative service provi- sets of inputs (Malmsten & Lekkas, 2010). The translog form
sion has cost implications, not only membership size but also provides a second-order approximation to the true underlying
the level of output is kept constant in the test. Our modified (but unknown) technology; it does not impose any technolog-
definition of cost subadditivity is as follows: ical restrictions, and allows economies of scale and size to
vary with output (Christensen, Jorgenson, & Lau, 1973; Hailu,
𝐶𝑖 (𝑁; 𝑞𝑖 ) < 𝐶𝑖𝑀 (𝑀; 𝑞𝑖𝑀 ) + 𝐶𝑖𝑅 (𝑅; 𝑞𝑖𝑅 ), (3) Goddard, & Jeffrey, 2005).
Our sample spans a range of cooperative types whose pro-
where 𝐶𝑖 (𝑁; 𝑞𝑖 ) is cost of the ith N-coop providing 𝑞𝑖 level duction technologies may differ with the nature of activity
of service to its N members; 𝐶𝑖𝑀 (𝑀; 𝑞𝑖𝑀 ) is cost of the they are engaged in. So, it may be an undue restriction to
ith M-coop providing 𝑞𝑖𝑀 level of service to its M mem- impose a common underlying production technology for all
bers; 𝐶𝑖𝑅 (𝑅; 𝑞𝑖𝑅 ) is cost of the ith R-coop providing 𝑞𝑖𝑅 level the cooperative types. The assumption of a common technol-
of service to its R members; and 𝑁 = 𝑀 + 𝑅; 𝑀 ∩ 𝑅 = ∅; ogy when heterogeneous technologies are present will poten-
𝑞𝑖 = 𝑞𝑖𝑀 + 𝑞𝑖𝑅 . Since N and 𝑞 are positively correlated in the tially lead to biased estimates of costs. On the other hand, a
sample (r = 0.37; p = 000), 𝑞𝑖𝑀 and 𝑞𝑖𝑅 are determined based separate regression approach assumes the existence of differ-
on the proportion of M and R in N. Thus, using equation (3), ent technologies without allowing the possibility of hypoth-
the degree of cost subadditivity (dCS) can be measured as: esis testing with regard to whether this assumption is valid
∑ (Triebs, Saal, Arocena, & Kumbhakar, 2016). Therefore, fol-
𝐶𝑖 (𝑁; 𝑞𝑖 ) − 𝑗 𝐶𝑖𝑗 (𝑗; 𝑞𝑖𝑗 )
𝑑𝐶𝑆 𝑖 = , (4) lowing Triebs et al. (2016), we use a translog cost model
𝐶𝑖 (𝑁; 𝑞𝑖 )
where technology can be fully flexible across cooperative
where 𝑑𝐶𝑆 𝑖 = degree of cost subadditivity of the ith cooper- types. The approach is a generalization of separate regressions
ative; 𝑗 = 𝑀, 𝑅. If 𝑑𝐶𝑆 < 0, the data suggest subadditivity allowing for the restriction and testing of a common technol-
in the cost structure. To practically test for subadditivity of the ogy assumption. Using the same functional form for each of
cost function of the cooperatives under study, we estimate a the technologies with parameters of their own, the model is
single-output and three-input (labor = l, capital = k, materi- given with the use of cooperative type dummies as:
als = m) cost function. Our subadditivity test is based on the
assumption that the hypothetical cooperatives of a kind can ln 𝐶(𝑞, 𝑝, 𝑁; 𝛽) = 𝑚 × ln 𝐶 𝑚 (𝑞 𝑚 , 𝑝, 𝑁 𝑚 ; 𝛽 𝑚 )
access the same technology, so that their cost properties can
be represented by a single cost function, as Sueyoshi (1996) + 𝓁 × ln 𝐶 𝓁 (𝑞 𝓁 , 𝑝, 𝑁 𝓁 ; 𝛽 𝓁 )

+ 𝑏 × ln 𝐶 𝑏 (𝑞 𝑏 , 𝑝, 𝑁 𝑏 ; 𝛽 𝑏 )
2 Complementary refers to the R-coop containing all the members of the
N-coop that are not in the M-coop. + 𝑟 × ln 𝐶 𝑟 (𝑞 𝑟 , 𝑝, 𝑁 𝑟 ; 𝛽 𝑟 ), (6)
GEZAHEGN ET AL. 559

∑ ∑ ∑
where ln 𝐶 = natural logarithm of total cost; 𝑚, 𝓁, 𝑏, and 𝑟 𝛽𝑖 = 1; 𝛽𝑖𝑞 = 0; 𝛽𝑖𝑗 = 0 for all 𝑗;
are dummies for MPC, LDC, BKC, and NRC, respectively; 𝑖 𝑖 𝑖
superscripts refer to the cost, output, number of members, ∑ ∑∑
𝛽𝑖𝑗 = 0 for all 𝑖; 𝛽𝑖𝑗 = 0 for all 𝑚, 𝓁, 𝑏, 𝑟
and parameters of each cooperative type; others as specified 𝑗 𝑖 𝑗
above. This model represents a cooperative-type flexible pro-
duction technology and allows the parameters to vary among (9)
the four cooperative types. It also allows for testing the com-
mon technology assumption through imposition of appropri- Using Shephard’s lemma and the symmetry constraint (8),
ate parameter restrictions. Applying a translog form to equa- we obtain cost share equation (10) for each input 𝑖 = 𝑙, 𝑘, 𝑚:
tion (6), we estimate the following model: [ ]
∑
𝑚 𝑚 𝑚 𝑚
𝑆 𝑖 = 𝑚 × 𝛽𝑖 + 𝛽𝑖𝑗 ln 𝑝𝑗 + 𝛽𝑖𝑞 ln 𝑞
[ 𝑗

ln 𝐶 = 𝑚 × 𝛽0𝑚 + 𝛽𝑞𝑚 ln 𝑞 𝑚 𝑚
+ 1∕2𝛽𝑞𝑞 (ln 𝑞 𝑚 )2 [ ]
∑
+𝓁 × 𝛽𝑖𝓁 + 𝛽𝑖𝑗𝓁 ln 𝑝𝑗 + 𝓁
𝛽𝑖𝑞 ln 𝑞 𝓁
∑ ∑∑ 𝑗
+ 𝛽𝑖𝑚 ln 𝑝𝑖 + 1∕2 𝛽𝑖𝑗𝑚 ln 𝑝𝑖 ln 𝑝𝑗 [ ]
𝑖 𝑖 𝑗 ∑
] +𝑏 × 𝛽𝑖𝑏 + 𝛽𝑖𝑗𝑏 ln 𝑝𝑗 + 𝑏
𝛽𝑖𝑞 ln 𝑞 𝑏
∑ 𝑗
𝑚 𝑚
+ 𝛽𝑖𝑞 ln 𝑝𝑖 ln 𝑞 + 𝛽𝑛𝑚 𝑁 [ ]
𝑖 ∑
[ +𝑟 × 𝛽𝑖𝑟 + 𝛽𝑖𝑗𝑟 𝑟
ln 𝑝𝑗 + 𝛽𝑖𝑞 ln 𝑞 𝑟
+ 𝑣𝑖 , (10)
𝑗
+𝓁 × 𝛽0𝓁 + 𝛽𝑞𝓁 ln 𝑞 𝓁 𝓁
+ 1∕2𝛽𝑞𝑞 (ln 𝑞 𝓁 )2
where 𝑣𝑖 = error term. To get more efficient estimates, we
∑ ∑∑ estimate the cost function (equation (7)) and share equa-
+ 𝛽𝑖𝓁 ln 𝑝𝑖 + 1∕2 𝛽𝑖𝑗𝓁 ln 𝑝𝑖 ln 𝑝𝑗
𝑖 𝑖 𝑗 tions (equation (10), except the materials cost share to avoid
] singularity) as a system using the iterated seemingly unre-
∑ lated regression (SUR) technique (Zellner, 1962). The sym-
𝓁 𝓁
+ 𝛽𝑖𝑞 ln 𝑝𝑖 ln 𝑞 + 𝛽𝑛𝓁 𝑁
𝑖
metry (equation (8)) and linear homogeneity (equation (9))
[ constraints are imposed during the estimation. Our model
regresses cost on sales (which is a function of quantity sold
+ 𝑏 × 𝛽0𝑏 + 𝛽𝑞𝑏 ln 𝑞 𝑏 + 1∕2𝛽𝑞𝑞
𝑏
(ln 𝑞 𝑏 )2
and output price) and input prices. Price variables are likely
∑ ∑∑ exogenous in the study area: input prices are either competi-
+ 𝛽𝑖𝑏 ln 𝑝𝑖 + 1∕2 𝛽𝑖𝑗𝑏 ln 𝑝𝑖 ln 𝑝𝑗 tive or preset by a union; and output prices are determined as
𝑖 𝑖 𝑗 the sum of the input prices and a preset margin. Yet, a simul-
] taneity problem may arise in relation to output quantities. As
∑
𝑏 𝑏
+ 𝛽𝑖𝑞 ln 𝑝𝑖 ln 𝑞 + 𝛽𝑛𝑏 𝑁 we find it difficult to control for this, we interpret our results
𝑖 as correlations rather than causations.
[
+ 𝑟 × 𝛽0𝑟 + 𝛽𝑞𝑟 ln 𝑞 𝑟 + 1∕2𝛽𝑞𝑞
𝑟
(ln 𝑞 𝑟 )2 4 RESULTS AND DISCUSSION
∑ ∑∑
+ 𝛽𝑖𝑟 ln 𝑝𝑖 + 1∕2 𝛽𝑖𝑗𝑟 ln 𝑝𝑖 ln 𝑝𝑗 4.1 Descriptive statistics
𝑖 𝑖 𝑗
] Our econometric model uses data on total cost (C), sales,
∑ input prices, and input cost shares. The article considers costs
𝑟
+ 𝛽𝑖𝑞 ln 𝑝𝑖 ln 𝑞 𝑟 + 𝛽𝑛𝑟 𝑁 + 𝜀, (7) and sales revenues related to the purchase and resale of sup-
𝑖
plies, including farm inputs, farm tools, consumer goods, and
livestock to members and nonmembers. Yearly cooperative-
where 𝑖, 𝑗 = 𝑙 (labor), 𝑘 (capital), 𝑚 (materials); 𝜀 = error
level data on total cost (ETB)3 and sales (ETB) were col-
term; others as defined above. The cost function is required
lected for the year 2016. While the values of total cost and
to satisfy the following symmetry (8) and linear homogeneity
sales were taken from cooperative documents, input prices
in input prices (9) constraints:
and input cost shares (proportion of total cost accounted for

𝛽𝑖𝑗 = 𝛽𝑗𝑖 for 𝑚, 𝓁, 𝑏, and 𝑟 (8) 3 ETB is the Ethiopian currency: 1 ETB ≈ 0.04 USD at the time of the study.
560 GEZAHEGN ET AL.

TABLE 1 Descriptive statistics of variables

Variable Definition (unit) Mean (SD)
Dependent variables
𝐶 Yearly total cost of service provision of the N-coopa (1,000 ETB) 1,121.97 (4,964.34)
lshare Labor cost share: Yearly proportion of labor cost in total cost 0.08 (0.13)
kshare Capital cost share: Yearly proportion of capital cost in total cost 0.45 (0.31)
mshare Materials cost share: Yearly proportion of materials cost in total cost 0.47 (4.53)
Independent variables
𝑞 Yearly total sales of the N-coop (1,000 ETB) 721.78 (1,613.83)
𝑝𝑙 Price of labor (ETB/hour) 2.17 (1.18)
𝑝𝑘 Price of capital (ETB /100-ETB-worth capital/year) 10.29 (2.50)
𝑝𝑚 Price of materials (ETB/unit) 1,169.62 (3,606.33)
𝑁 Number of members of the N-coop 413.05 (564.67)
𝑀 Number of members of M-coop 10.65 (7.19)
𝑞𝑚 Yearly total sales of the M-coop (1000 ETB) 2.34 (3.43)
𝑅=𝑁 −𝑀 Number of members of hypothetical R-coop 373.69 (548.28)
𝑞𝑟 = 𝑞 − 𝑞𝑚 Yearly total sales of the R-coop (1000 ETB) 680.98 (1569.75)
Source: Field survey and own calculations.
a
N-coop is the larger cooperative with at least twice the minimum membership size (N) observed in the sample; M-coop is the smaller cooperative with the minimum
number of members (M) observed in the sample; R-coop is the hypothetical complementary cooperative with the residual (R = N − M) number of members.

by a given input) are the result of own calculation. Total cost average. Theoretically, the price of capital is determined as the
includes labor cost, capital cost, and materials cost. Capital sum of the depreciation rate and the interest rate (Pindyck &
cost includes the monetary value (at purchase price) of fixed Rubinfeld, 2001). Due to lack of data on depreciation costs,
assets such as buildings, office furniture, motor and treadle price of capital is proxied by the interest rate, which varies
pumps, modern and traditional beehives, and dairy equip- among cooperatives based on the source and purpose of the
ment. Capital cost also includes interest paid on borrowings. loan. Labor (lcost), capital (kcost), and materials (mcost) cost
Materials cost includes the yearly expenditure on farm tools, shares are calculated as the proportion of the cost of each input
farm inputs (pesticides, fertilizer, seeds), consumer goods in total cost.
(sugar, coffee, salt, flour, oil. milk), livestock and feed, and Table 1 presents descriptive statistics of the variables used
transport in 2016. Sales refers to the revenue from the resale in the cost estimation. The average cooperative was found to
of the above mentioned farm inputs, farm tools, consumer spend close to 1.1 million ETB per year to supply farm inputs,
goods, merchandise, livestock and livestock products, and nat- farm tools, consumer goods, etc. to members and nonmem-
ural resource products, such as grass, seedlings, and trees in bers; especially MPC are not allowed to discriminate between
2016. members and nonmembers. The very high standard deviation
For MPC, the price of labor was calculated as the ratio (SD = 4,964) of total cost indicates a considerable scale dif-
of total monthly payroll to total monthly number of hours ference among cooperatives: cooperative size ranges from 4
worked. In cooperatives other than MPC, work is done to 2,550 members. Average sales are 0.72 million ETB per
by members themselves. As there were no paid employees year; average prices of labor, capital, and materials are 2.17
(except guards in some cases), no salaries and wages were ETB per hour, 10.29 ETB per 100 ETB of capital per year,
paid by LDC, BKC, and NRC. To avoid potential bias due and 1169 ETB per unit, respectively. Capital costs take about
to neglecting labor, we imputed labor (opportunity) cost for 45% of the total cost while labor and materials account for 8%
the cooperatives with missing payroll based on the number and 47%, respectively.
of their active members and the labor price calculated for The variance of the price of materials is very high, which
MPC. The price of materials was calculated as the average probably relates to the fact that the data are derived from very
of the ratio of total yearly cost to total yearly quantity of each diverse cooperatives in a rather large area and that heteroge-
item in the materials category. For example, we calculate the neous items are included under materials. Table A1 (online
unit cost of fertilizer by dividing the yearly amount of money appendix) highlights the link between membership size and
spent on fertilizer by the yearly number of quintals of fertil- different cost items for each cooperative type, pointing to
izer purchased. This procedure was repeated for each item in a rather mixed relation between size and cost. In the cost
the materials category, and the average of the unit costs of all analysis, membership size is fixed, such that costs per mem-
the items was calculated to get the unit price of materials, on ber rather than total costs give better insights on size-cost
GEZAHEGN ET AL. 561

relations. Except for two cases in MPC and one in LDC and ple are minimizing production costs, and that inputs can be
NRC, per-unit costs (C/N, lcost/N, kcost/N, and mcost/N) tend adjusted flexibly. If some cooperatives are not able to mini-
to decrease with cooperative size. Moreover, the per-unit costs mize production costs, imposing concavity conditions on the
indicate that keeping the number of members constant at two, cost function yields inconsistent estimates of the parameters
the sum of the costs of the two smaller cooperatives is larger (Ogawa, 2011).
than twice the cost of the large cooperative,4 indicating the Some coefficients of the prices of labor and capital are neg-
cost efficiency of relatively large cooperatives. ative, which is not consistent with the property of a cost func-
tion, and might relate to multicollinearity problems (Filippini,
4.2 Econometric results 1996). Variance inflation factors (VIFs) suggest that there are
multicollinearity problems for the price of labor in Model 1
Table 2 presents the results of three models: Model 1 reports for LDC (VIF = 51.17) and BKC (VIF = 61.55), and in Model
the estimates of a joint regression of equations (7) to (10), 3 for BKC (VIF = 50.84); and for the price of capital in Model
allowing for flexible technology; Model 2 reports the esti- 1 for NRC (VIF = 16.23). A possible cause relates to the dif-
mates for a common-technology model; and Model 3 reports ficulty in calculating the labor and capital variables and their
the estimates of separate regressions for each cooperative prices. Consistent with the rejection of the common technol-
type, allowing for type-specific technologies. To get more ogy assumption, parameters in Model 1 are similar to parame-
efficient parameters, we estimate the cost functions and cost ters in Model 3, except for MPC.5 This could relate to the large
share equations (except mcost) as a system using the iterated number of MPC in the sample (n = 179) and in the pooled
SUR technique (Zellner, 1962). All variables are standardized Model 2. We find that a 1% increase in output increases cost
such that the first-order coefficients of the translog function by 0.51%, 0.65%, 0.19%, and 0.23% for MPC, LDC, BKC,
can be interpreted as elasticities at the sample mean. We per- and NRC, respectively, implying that cost is most elastic with
form the standard likelihood ratio test to test hypotheses of respect to output for LDC, and least for BKC. Returns to scale
common technology across all pairs of cooperative types, by at the sample mean for each technology are given in Table 4.
imposing equality restrictions across corresponding parame- All models point to evidence of IRS. Except for LDC, the
ters of the cooperative-specific cost functions (Table 3). The degree of scale economies under Models 1 and 3 are closer
null hypothesis of a common technology is rejected for any to each other than those under Models 1 and 2.
pair of cooperative types at the 1% significance level. More- The larger-than-unity values of returns to scale imply that
over, the goodness of fit of Model 1 (R-sq = 0.72) is better the typical cooperative operates under IRS. In a competitive
than that of the other models. So, we base our subsequent dis- market, IRS lead to economies of scale (Gelles & Mitchell,
cussion and conclusions on this model after testing it for linear 1996). However, if a cooperative is so large that its input
homogeneity, monotonicity in output and input prices, posi- purchase drives unit prices up, it could have diseconomies
tivity, and concavity. of scale even under IRS. Conversely, if the cooperative is
The linear homogeneity and symmetry constraints were able to get bulk discounts on its input purchase, it could have
already imposed during estimation. Monotonicity in input economies of scale even under decreasing returns in produc-
prices is satisfied: the fitted input cost shares are positive at tion. Since IRS are not a necessary condition for economies
each data point. Positivity is satisfied at each data point. Con- of scale (Bell, 1988), the effect of returns to scale on cost effi-
cerning monotonicity in output, there are only about 5% viola- ciency of cooperatives depends on the nature of the market.
tions. Concavity is not satisfied for only about one-third of the
observations, with the highest violation rate in capital inputs
4.3 Estimation of dCS
followed by labor and materials. A possible explanation for
this violation is that a cooperative does not necessarily choose Using the parameters of the cost functions, we estimate the
the optimal combination of inputs if factor prices change, for fitted costs of the N-coop, M-coop, and R-coop and calculate
example, due to borrowing constraints (Ogawa, 2011). More- dCS for each cooperative based on equation (4). A negative
over, concavity will be violated if an input is quasi-fixed and dCS suggests that a larger cooperative is more cost efficient
a cooperative incurs additional adjustment costs in changing than two smaller cooperatives. Table 5 reports the estimated
input levels. This may explain the highest rate of violation for dCS for each cooperative type derived from alternative
the capital input. Imposing concavity either globally or locally models. Findings show that there are cost savings from a joint
may not solve the problem, because such an approach is based action in larger cooperatives. All observations and all cooper-
on the implicit assumption that all cooperatives in the sam- ative types show single cooperative costs to be lower than the

4 For example, for NRC the two smaller cooperatives can serve two members 5 For LDC about 67% and for BKC and NRC about 60% of Model 1 param-
at a total cost of about ETB 29 while the larger cooperative can serve the same eters are more similar to those of Model 3. For MPC only close to 27% of
two members at a total cost of about ETB 11 (=2×5.49). Model 1 parameters are more similar to Model 3 parameters.
562 GEZAHEGN ET AL.

TABLE 2 Parameter estimates of the translog cost functions

Variable Model 1 Model 2 Model 3
MPC LDC BKC NRC All MPC LDC BKC NRC
𝑚/𝓁/𝑏/𝑟 −0.316*** −0.725*** −1.034** −1.232***
(0.118) (0.271) (0.521) (0.224)
lnq 0.514*** 0.653*** 0.188 0.232** 0.409*** 0.552*** 0.467** 0.306 0.075
(0.053) (0.160) (0.218) (0.099) (0.058) (0.093) (0.237) (0.251) (0.148)
lnpl 0.274** −0.230 −0.171 1.017*** −0.192*** −1.010*** 0.043 −0.516** 0.330*
(0.128) (0.294) (0.211) (0.193) (0.056) (0.069) (0.191) (0.202) (0.177)
lnpk 0.684*** 0.888*** 1.018*** −0.099 0.415*** −0.146* 0.701*** 0.842*** 0.267
(0.158) (0.303) (0.203) (0.198) (0.055) (0.079) (0.187) (0.200) (0.180)
lnpm 0.042 0.342*** 0.153 0.082 0.776*** 2.156*** 0.256* 0.674*** 0.403***
(0.115) (0.094) (0.183) (0.087) (0.071) (0.116) (0.155) (0.236) (0.127)
lnq(lnq) 0.066*** 0.070** 0.010 0.000 0.029** 0.015 0.080 0.007 −0.007
(0.014) (0.033) (0.028) (0.018) (0.014) (0.033) (0.056) (0.033) (0.030)
lnq(lnpl ) −0.200*** −0.127*** −0.106*** −0.104*** −0.088*** −0.199*** −0.016 −0.153*** −0.032
(0.020) (0.026) (0.031) (0.028) (0.019) (0.046) (0.074) (0.051) (0.056)
lnq(lnpk ) −0.116*** −0.019 0.145*** −0.009 −0.003 −0.125** 0.092 0.090* 0.017
(0.031) (0.025) (0.025) (0.029) (0.020) (0.056) (0.074) (0.047) (0.058)
lnq(lnpm ) 0.316*** 0.146*** −0.038 0.114*** 0.091*** 0.324*** −0.076 0.063 0.015
(0.038) (0.028) (0.040) (0.022) (0.024) (0.084) (0.064) (0.060) (0.039)
lnpl (lnpl ) 0.098*** 0.149*** 0.135*** −0.097** 0.108*** 0.046 0.154 0.157*** 0.109
(0.024) (0.032) (0.028) (0.041) (0.029) (0.046) (0.095) (0.046) (0.073)
lnpl (lnpk ) −0.101* −0.291*** −0.183*** 0.065 −0.118** −0.193** −0.223 −0.213** −0.182
(0.054) (0.064) (0.050) (0.078) (0.056) (0.094) (0.192) (0.086) (0.136)
lnpl (lnpm ) −0.096** −0.006 −0.086** 0.129*** −0.098*** 0.100 −0.086 −0.102* −0.036
(0.044) (0.022) (0.036) (0.027) (0.027) (0.092) (0.061) (0.057) (0.054)
lnpk (lnpk ) 0.136*** 0.200*** 0.113*** 0.001 0.062* −0.130* 0.162 0.049 0.118
(0.043) (0.034) (0.028) (0.042) (0.032) (0.067) (0.100) (0.048) (0.074)
lnpk (lnpm ) −0.171*** −0.109*** −0.042 −0.068*** −0.006 0.452*** −0.102 0.114** −0.053
(0.063) (0.021) (0.031) (0.026) (0.028) (0.106) (0.063) (0.054) (0.054)
lnpm (lnpm ) 0.133*** 0.058*** 0.064*** −0.030*** 0.052*** −0.276*** 0.094*** −0.006 0.045***
(0.035) (0.014) (0.023) (0.010) (0.013) (0.072) (0.035) (0.033) (0.017)
N 0.083** 0.095 −0.018 0.991** 0.169*** 0.181** 0.104 −0.061 1.689***
(0.037) (0.217) (0.159) (0.407) (0.051) (0.087) (0.390) (0.184) (0.648)
Constant −0.723*** 0.302 −1.484*** −0.404 −1.956***
(0.128) (0.240) (0.390) (0.592) (0.315)
Number of coops 420 420 179 43 108 90
R2 0.72 0.36 0.29 0.30 0.03 0.24
Note: Standard errors are reported in parentheses. ln stands for natural logarithm. Models 1, 2, and 3 represent flexible technology, common technology, and cooperative
type–specific regression models, respectively. MPC = multipurpose cooperatives; LDC = livestock and dairy cooperatives; BKC = beekeeping cooperatives; NRC = natural
resources cooperatives. Significance levels are reported as *** P < 0.01, ** P < 0.05, * P < 0.1.

sum of two cooperative costs, given the same membership base of cooperatives being limited to a specific district or
size and level of service (see Table A2, online appendix, for tabia.
a detailed cost comparison). A single larger cooperative is Estimates of Model 1 show that a given number of farmers
more cost efficient in providing a given level of service to would realize a cost advantage of 78% to 181% if organized
farmers than two or more smaller cooperatives. This implies in a single larger rather than in two smaller cooperatives.
that cooperatives in the study area are too small to exhaust Mainly due to regulation and/or limited demand, the sampled
scale economies. A possible reason relates to the membership cooperatives have no market power. The evidence for IRS
GEZAHEGN ET AL. 563

TABLE 3 Test on common technology across cooperative types TABLE 4 Returns to scale at sample means by technology and
MPC vs. LDC vs. BKC vs. cooperative type
All LDC BKC NRC BKC NRC NRC Cooperative category
Chi2 193.89 50.49 88.48 91.47 56.76 65.77 62.06 MPC LDC BKC NRC
df 34 11 12 11 12 11 12 Model 1: Flexible technology 1.95 1.53 5.32 4.31
P 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Model 2: Common technology 2.44 2.44 2.44 2.44
Note: MPC = multipurpose cooperatives; LDC = livestock and dairy cooperatives; Model 3: Separate regressions 1.81 2.14 3.27 13.16
BKC = beekeeping cooperatives; NRC = natural resources cooperatives. Note: MPC = multipurpose cooperatives; LDC = livestock and diary cooperatives;
BKC = beekeeping cooperatives; NRC = natural resources cooperatives.

suggests that economies of scale likely underlie the cost effi-

TABLE 5 Estimated degree of cost subadditivity (dCS) by
ciency of larger cooperatives. Moreover, larger cooperatives
technology and cooperative type
can exercise bargaining power in accessing farm inputs, tools,
Cooperative category
and credit; limit transaction costs; and spread fixed costs,
All MPC LDC BKC NRC
such as management costs over more items. The presence
of scale economies implies that smaller cooperatives are not Model 1: Flexible −1.17 −0.78 −1.37 −1.18 −1.81
technology (0.86) (0.97) (1.04) (0.13) (0.57)
competitive and would be driven out of the market if govern-
ment support and regulation would be removed and market Model 2: Common −0.92 −0.68 −1.13 −1.08 −1.10
competition would prevail (Wheelock & Wilson, 2011). technology (0.33) (0.28) (0.40) (0.13) (0.24)
Competitive pressure among cooperatives and other types of Model 3: Separate −0.95 −0.68 −1.31 −1.16 −1.06
enterprises is likely to encourage further growth in coopera- regressions (0.73) (1.02) (0.50) (0.15) (0.09)
tive size (Wheelock & Wilson, 2011). This implies there are Note: Standard deviations are reported in parentheses. MPC = multipurpose coop-
economic incentives for farmers to organize in relatively large eratives; LDC = livestock and dairy cooperatives; BKC = beekeeping coopera-
rather than small cooperatives, and in small cooperatives tives; NRC = natural resources cooperatives.

rather than taking unilateral actions. Yet, economies of scale

have limits and policies regarding cooperative size should be outweigh the negative ones. It is more cost efficient to increase
cautious of not exceeding the optimal size. the size of existing cooperatives than to establish new coop-
On average, cost drops by 117% (Model 1) in the single- eratives or split up existing ones.
cooperative as compared to the two-cooperative scenario. Another insight from the findings is that cost savings
That is, a level of service that a single larger cooperative can decrease with cooperative size. Table 6 presents the correla-
provide at a cost of 1.00 ETB would be provided by two tion between cooperative size and the absolute value of per-
smaller cooperatives at a cost of 2.17 ETB. Findings focus centage cost savings within and across cooperative types. The
on the cost advantages at cooperative level. But we argue that negative correlation coefficients imply that percentage cost
members of larger cooperatives are better off than members savings decrease with cooperative size. However, the absolute
of smaller cooperatives because providing services at a mini- magnitude of the cost savings of smaller cooperatives may be
mum cost is consistent with members’ welfare maximization lower or higher than that of larger cooperatives, depending on
(Park & Meyer, 1994). Economies of scale likely underlie the the actual cost of service provision. The relationship between
subadditivity of the cost function of the cooperatives. How- cost savings and size can be extended to size differentials
ever, insights from Joskow (2007) suggest that it may be less across cooperative types. Though not statistically strong, the
costly for an output to be provided by a single cooperative negative correlation coefficients across cooperative types also
rather than multiple cooperatives even if the output of the sin- suggest a negative relationship between cost savings and size.
gle cooperative has expanded beyond the area of economies Generally, there is a negative correlation between size and
of scale. A cost function can be subadditive beyond the point potential efficiency gains from scale across all cooperative
where economies of scale are exhausted, until demand is large types. For example, MPCs have the largest membership size
enough to add a second cooperative. This may be because and the lowest percentage cost savings, which suggests that
the market demand is not large enough to support efficient these cooperatives are closest to the efficient scale and that
service provision by two cooperatives. Though coordination there is not much gain from increasing membership size. This
costs increase with cooperative size, the cost efficiency of is an indication that larger cooperatives are operating at a more
larger cooperatives implies that the typical cooperative has efficient scale than smaller ones. Therefore, mergers and/or
not reached the size beyond which an increase in coordina- open membership policies would be cost reducing. This can
tion costs counterbalances the gains from economies of scale. be extended to argue in favor of cooperation rather than unilat-
The conclusion is that in the study area, with the current size eral action. Farmers are better off working together than alone
of cooperatives, positive effects of membership expansion because unilateral actions are likely to be less cost efficient
564 GEZAHEGN ET AL.

TABLE 6 Correlation between dCS and cooperative size N affiliation because they do not precisely know the benefits
(within and across cooperative types) from alternative options, which forces the bargaining process
Cooperative type Model 1 Model 2 Model 3 to take place in an atmosphere of uncertainty. As Staatz (1983)
Across All −0.58*** −0.75*** −0.75*** argues, uncertainty about what is in one’s best interest may
Within MPC −0.44*** −0.65*** −0.47*** reduce defection from the cooperative. In the study area, infor-
LDC −0.19 −0.35** −0.36** mation on costs and benefits of external opportunities is not
BKC −0.69*** −0.31*** −0.67*** well established. Cooperatives are often not demand driven
but established by donors or the government to combat youth
NRC −0.66*** −0.70*** −0.83***
unemployment or distribute farm inputs, without taking into
Note: Models 1, 2, and 3 represent flexible technology, common technology, and
cooperative-type–specific regression models, respectively. MPC = multipurpose
account the interests and preferences of members. This leads
cooperatives; LDC = livestock and dairy cooperatives; BKC = beekeeping coop- to lack of commitment and problems of defection.
eratives; NRC = natural resources cooperatives. Significance levels are reported
as *** P < 0.01, ** P < 0.05, * P < 0.1.

5 CONC LU SI ON
than joint actions under any cooperative size. This explains
the presence of economic incentives for collective action via This study empirically investigates the cost structure of agri-
cooperatives. When cost is subadditive, cooperative formation cultural cooperatives in Ethiopia. Specifically, it examines
and membership expansion are beneficial. Our conclusions whether the sector provides incentives for cooperation by
can be related to studies that use the same approach in differ- testing for the condition of cost subadditivity. The sampled
ent industries. For example, Evans and Heckman (1984) reject cooperatives are found to exhibit cost subadditivity in service
the hypothesis that the U.S. bell system’s cost function is sub- delivery to members. Given the scale benefits reflected by
additive at the output levels produced, and support its decen- the subadditive cost function, larger cooperatives are likely
tralization (split) into smaller specialized systems. Sanchez to avail services to their members at a lower cost than would
(2000) rejects cost subadditivity for larger railway companies, smaller cooperatives or unilateral actions do. Findings suggest
suggesting that separate supply of freight and passenger trans- that splitting up cooperatives results in cost disadvantages.
portation by two independent companies is more efficient Collective action in larger cooperatives can help farmers to
than supply of both by a single firm. For smaller companies, reap more economies of scale in the sector. Therefore, policies
results support the hypothesis of subadditive cost, implying favoring fewer but larger agricultural cooperatives would be
that freight and passenger transport by a single railway com- beneficial in Tigray, while the current one-cooperative-per-
pany is more cost efficient. Our results contradict some earlier neighborhood kind of approach is cost inefficient.
results that smaller agricultural cooperatives perform better. Some caution is needed finally. First, the test for the con-
The divergent results can be explained by heterogeneity of dition of cost subadditivity is a local test done by splitting
regions, cooperatives, and samples, and different measures of the observed output of each cooperative into two hypotheti-
size and performance (Arcas et al., 2011; Guzmán & Arcas, cal components. Further tests of the condition using different
2008; Kyriakopoulos, Meulenberg, & Nilsson, 2004). hypothetical outputs could result in more conclusive evidence.
Second, though findings support the assertion that in agricul-
ture, collective action (via larger cooperatives) is beneficial
4.4 Implications for cooperative stability
in terms of cost savings, policies toward cooperative mergers
Cost subadditivity is a necessary but not a sufficient condi- should pay attention to managerial issues as well. The scale
tion for a cooperative to be stable, and does not guarantee that benefits to be reaped from larger cooperatives are conditional
members will continue to cooperate. Each individual mem- on technical and human skills to be at least as good as what the
ber’s share of the joint gain must be larger than the gain they currently existing cooperatives have. Otherwise, managerial
could achieve by operating in a smaller cooperative or inde- inefficiencies in larger cooperatives might nullify the benefits
pendently. The cost advantage of a larger cooperative must be of potential scale efficiencies.
allocated in such a way that all members have an incentive to
remain in the cooperative. There may be different ways to do
ACKNOW LEDGMENTS
this but failure to choose an allocation that gives members a
benefit that is larger than what they can get somewhere else The authors acknowledge funding from the VLIR-UOS-
puts the stability of the cooperative at risk. A stable alloca- ZEIN2015PR406 (13V95615T) Programme (Team), Bel-
tion that gives everyone an incentive to stay in the cooperative gium. We also thank participants at the 30th International
might be more difficult if members are more heterogeneous. Conference of Agricultural Economists held in Vancouver,
On the other hand, some factors may deter membership Canada, for their useful comments on an earlier version of the
defection. Members may simply continue their cooperative article.
GEZAHEGN ET AL. 565

REFERENCES Getnet, K., & Anullo, T. (2012). Agricultural cooperatives and rural
livelihoods: Evidence from Ethiopia. Annals of Public and Cooper-
Abebaw, D., & Haile, M. G. (2013). The impact of cooperatives on ative Economics, 83, 181–198.
agricultural technology adoption: Empirical evidence from Ethiopia. Gordon, D., Gunsch, K., & Pawluk, C. (2003). A natural monopoly in
Food Policy, 38, 82–91. natural gas transmission. Energy Economics, 25, 473–485.
Adhikari, D. R., & Guru-Gharana, K. K. (2014). Econometric test of cost Guzmán, I., & Arcas, N. (2008). The usefulness of accounting informa-
subadditivity in US electric industry. Journal of Economics and Eco- tion in the measurement of technical efficiency in agricultural coop-
nomic Education Research, 15, 33–44. eratives. Annals of Public and Cooperative Economics, 79, 107–131.
Arcas, N., García, D., & Guzmán, I. (2011). Effect of size on perfor- Hailu, G., Goddard, E. W., & Jeffrey, S. R. (2005). Measuring efficiency
mance of Spanish agricultural cooperatives. Outlook on Agriculture, in fruit and vegetable marketing co-operatives with heterogeneous
40, 201–206. technologies in Canada. American Agricultural Economics Associa-
Barton, D. G., Schroeder, T. C., & Featherstone, A. M. (1993). Evaluat- tion Annual Meeting, Providence, Rhode Island.
ing the feasibility of local cooperative consolidations: A case study. Hansmann, H. (1988). Ownership of the firm. Journal of Law Eco-
Agribusiness, 9, 281–294. nomics and Organization, 4, 267–304.
Bell, C. R. (1988). Economies of, versus returns to, scale: A clarification.
Holloway, G., Nicholson, C., Delgado, C., Staal, S., & Ehui, S. (2000).
Journal of Economic Education, 19, 331–335.
Agroindustrialization through institutional innovation: Transaction
Berger, A. N., & Humphrey, D. B. (1991). The dominance of inefficien-
costs, cooperatives and milk-market development in the east-African
cies over scale and product mix economies in banking. Journal of
highlands. Agricultural Economics, 23, 279–288.
Monetary Economics, 28, 117–148.
Hunter, W. C., Timme, S. G., & Yang, W. K. (1990). An examination
Bernard, T., Abate, G. T., & Lemma, S. (2013). Agricultural coopera-
of cost subadditivity and multiproduct production in large US banks.
tives in Ethiopia: Results of the 2012 ATA baseline survey. Research
Journal of Money, Credit, and Banking, 22, 504–525.
for Ethiopia’s Agricultural Policy, Report. International Food Policy
Ito, J., Bao, Z., & Su, Q. (2012). Distributional effects of agricultural
Research Institute, Washington DC.
cooperatives in China: Exclusion of smallholders and potential gains
Bernard, T., & Taffesse, A. S. (2012). Returns to scope? Smallholders’
on participation. Food Policy, 37, 700–709.
commercialisation through multipurpose cooperatives in Ethiopia.
Ivaldi, M., & Mccullough, G. (2008). Subadditivity tests for network sep-
Journal of African Economies, 21, 440–464.
aration with an application to US railroads. Review of Network Eco-
Bonus, H. (1986). The cooperative association as a business enterprise:
nomics, 7, 159–171.
A study in the economics of transactions. Journal of Institutional and
Joskow, P. L. (2007). Regulation of natural monopoly. Handbook of Law
Theoretical Economics, 142, 310–339.
and Economics, 2, 1227–1348.
Cazzuffi, C., & Moradi, A. (2012). Membership size and cooperative
Kimms, A., & Kozeletskyi, I. (2016a). Core-based cost allocation in the
performance: Evidence from Ghanaian cocoa producers’ societies,
cooperative traveling salesman problem. European Journal of Oper-
1930–36. Economic History of Developing Regions, 27, 67–92.
Chagwiza, C., Muradian, R., & Ruben, R. (2016). Cooperative member- ational Research, 248, 910–916.
ship and dairy performance among smallholders in Ethiopia. Food Kimms, A., & Kozeletskyi, I. (2016b). Shapley value-based cost allo-
Policy, 59, 165–173. cation in the cooperative traveling salesman problem under rolling
Chinchuluun, A., Pardalos, P. M., Migdalas, A., & Pitsoulis, L. (2008). horizon planning. EURO Journal on Transportation and Logistics,
Pareto optimality, game theory and equilibria. Springer Optimiza- 5, 371–392.
tion and Its Applications (Vol. 17). Springer, https://doi.org/10.1007/ Kyriakopoulos, K., Meulenberg, M., & Nilsson, J. (2004). The impact
978-0-387-77247-9. of cooperative structure and firm culture on market orientation and
Christensen, L. R., Jorgenson, D. W., & Lau, L. J. (1973). Transcendental performance. Agribusiness: An International Journal, 20, 379–396.
logarithmic production frontiers. Review of Economics and Statistics, Lerman, Z., & Parliament, C. (1991). Size and industry effects in the
55, 28–45. performance of agricultural cooperatives. Agricultural Economics, 6,
Drechsel, J. (2010). An economic lot sizing game. In Cooperative lot siz- 15–29.
ing games in supply chains. Lecture Notes in Economics and Mathe- Ma, W., & Abdulai, A. (2016). Does cooperative membership improve
matical Systems (Vol. 644). Berlin, Heidelberg: Springer. household welfare? Evidence from apple farmers in China. Food Pol-
Evans, D. S., & Heckman, J. J. (1984). A test for subadditivity of the cost icy, 58, 94–102.
function with an application to the bell system. American Economic Ma, W., & Abdulai, A. (2018). IPM adoption, cooperative member-
Review, 74, 615–623. ship and farm economic performance: Insight from apple farmers in
Filippini, M. (1996). Economies of scale and utilization in the Swiss elec- China. China Agricultural Economic Review, 11, 218–236.
tric power distribution industry. Journal of Applied Economics, 28, Ma, W., Abdulai, A., & Goetz, R. (2017). Agricultural cooperatives
543–550. and investment in organic soil amendments and chemical fertilizer
Fischer, E., & Qaim, M. (2012). Linking smallholders to markets: Deter- in China. American Journal of Agricultural Economics, 100, 502–
minants and impacts of farmer collective action in Kenya. World 520.
Development, 40, 1255–1268. Malmsten, M., & Lekkas, D. F. (2010). Cost analysis of urban water
Fulton, J. R., & King, R. P. (1993). Relationships among information supply and waste water treatment processes to support decisions and
expenditure, economic performance, and size in local grain market- policy making: Application to a number of Swedish communities.
ing cooperatives in the upper midwest. Agribusiness, 9, 143–157. Desalination and Water Treatment, 18, 327–340.
Gelles, G. M., & Mitchell, D. W. (1996). Returns to scale and economies Mojo, D., Fischer, C., & Degefa, T. (2017). The determinants and eco-
of scale: Further observations. The Journal of Economic Education, nomic impacts of membership in coffee farmer cooperatives: Recent
27, 259–261. evidence from rural Ethiopia. Journal of Rural Studies, 50, 84–94.
566 GEZAHEGN ET AL.

Ogawa, K. (2011). Why are concavity conditions not satisfied in the cost Valentinov, V. (2007). Why are cooperatives important in agriculture?
function? The case of Japanese manufacturing firms during the bub- An organizational economics perspective. Journal of Institute of Eco-
ble period. Oxford Bulletin of Economics and Statistics, 73, 556–580. nomics, 3, 55–69.
Ortmann, G. F., & King, R. P. (2007). Agricultural cooperatives I: His- Vandeplas, A., Minten, B., & Swinnen, J. (2013). Multinationals vs.
tory, theory and problems. Agrekon, 46, 18–46. cooperatives: The income and efficiency effects of supply chain
Park, S. J., & Meyer, R. L. (1994). Scale economies and cost subad- governance in India. Journal of Agricultural Economics, 64, 217–
ditivity of primary agricultural cooperative banking in Korea. Jour- 244.
nal of Rural Development, 17, 73–97. Verhofstadt, E., & Maertens, M. (2014). Smallholder cooperatives and
Pindyck, R. S., & Rubinfeld, D. L. (2001). Microeconomics (5th ed.). agricultural performance in Rwanda: Do organizational differences
Upper Saddle River, NJ: Prentice Hall. matter? Agricultural Economics, 45, 39–52.
Poulton, C., Dorward, A., & Kydd, J. (2010). The future of small farms: Verhofstadt, E., & Maertens, M. (2015). Can agricultural coopera-
New directions for services, institutions, and intermediation. World tives reduce Poverty? Heterogeneous impact of cooperative mem-
Development, 38, 1413–1428. bership on farmers’ welfare in Rwanda. Applied Economics Perspec-
Reardon, T., Barrett, C. B., Berdegué, J. A., & Swinnen, J. F. (2009). tives and Policy, 37, 86–106.
Agrifood industry transformation and small farmers in developing Wheelock, D. C., & Wilson, P. W. (2011). Are credit unions
countries. World Development, 37, 1717–1727. too small? The Review of Economics and Statistics, 93, 1343–
Saitone, T. L., Sexton, R. J., & Malan, B. (2018). Price premiums, 1359.
payment delays, and default risk: Understanding developing coun- Wossen, T., Abdoulaye, T., Alene, A., Haile, M. G., Feleke, S., Olan-
try farmers’ decisions to market through a cooperative or a private rewaju, A., & Manyong, V. (2017). Impacts of extension access and
trader. Agricultural Economics, 49, 363–380. cooperative membership on technology adoption and household wel-
Sánchez, P. C. (2000). A subadditivity test for the cost function of the fare. Journal of Rural Studies, 54, 223–233.
principal European railways. Transport Reviews, 20, 275–290. Zakharov, V. V., & Shirokikh, V. A. (2018). Heuristic evaluation of the
Soboh, R. A., Lansink, A. O., Giesen, G., & Van Dijk, G. (2009). Perfor- characteristic function in the Cooperative Inventory Routing Game.
mance measurement of the agricultural marketing cooperatives: The Journal on Vehicle Routing Algorithms, 1, 19–32.
gap between theory and practice. Review of Agricultural Economics, Zellner, A. (1962). An efficient method of estimating seemingly unre-
31, 446–469. lated regressions and tests for aggregation bias. Journal of the Amer-
Staatz, J. M. (1983). The cooperative as a coalition: A game-theoretic ican Statistical Association, 57, 348–368.
approach. American Journal of Agricultural Economics, 65, 1084–
1089.
Staatz, J. M. (1985). Game-theoretic analysis of decision making in
S U P P O RT I NG IN FO R M AT I O N
farmer cooperatives. Department of Agricultural Economics, Michi-
gan State University. Additional supporting information may be found online in the
Staatz, J. M. (1987). The structural characteristics of farmer coopera- Supporting Information section at the end of the article.
tives and their behavioral consequences. Cooperative Theory: New
Approaches, 18, 33–60.
Sueyoshi, T. (1996). Divestiture of Nippon telegraph and telephone.
Management Sciences, 42, 1326–1351. How to cite this article: Gezahegn TW, Van Pas-
Trechter, D. D. (1996). Impact of diversification on agricultural coop-
sel S, Berhanu T, D’Haese M, Maertens M. Big is
eratives in Wisconsin. Agribusiness: An International Journal, 12,
efficient: Evidence from agricultural cooperatives in
385–394.
Triebs, T. P., Saal, D. S., Arocena, P., & Kumbhakar, S. C. (2016). Esti- Ethiopia. Agricultural Economics. 2019;50:555–566.
mating economies of scale and scope with flexible technology. Jour- https://doi.org/10.1111/agec.12509
nal of Productivity Analysis, 45, 173–186.