Do farmers’ organizations enhance the welfare of smallholders? Findings from the Mozambican national agricultural survey

Do farmers’ organizations enhance the welfare of smallholders? Findings from the Mozambican national agricultural survey

Food Policy xxx (xxxx) xxxx Contents lists available at ScienceDirect Food Policy journal homepage: www.elsevier.com/locate/foodpol Do farmers’ org...

871KB Sizes 2 Downloads 23 Views

Food Policy xxx (xxxx) xxxx

Contents lists available at ScienceDirect

Food Policy journal homepage: www.elsevier.com/locate/foodpol

Do farmers’ organizations enhance the welfare of smallholders? Findings from the Mozambican national agricultural survey Maren Elise Bachke School of Economics and Business, Norwegian University of Life Sciences, P.O. Box 5003, N-1432 Aas, Norway

ARTICLE INFO

ABSTRACT

Keywords: Collective action Economic welfare Propensity score matching estimator Mozambique

Farmers’ organizations have been used as a tool to improve the living conditions of farmers in many countries by improving market access, access to information and capacity to increase production. I employ panel data from Mozambique to investigate how membership in farmers’ organizations impacts smallholders’ welfare. Using difference-in-difference estimators that control for unobservable selection bias, I find a positive impact of membership on the marketed surplus (25%), the value of agricultural production (18%) and on total income (15%, and more than 20% for those whose main source of cash income is the agricultural sector).

1. Introduction The majority of the world’s poor are rural inhabitants who depend on agriculture for their livelihoods. Raising the income of the smallholders is therefore crucial to reduce poverty. It is widely recognized that increased commercialization among smallholders’ leads to higher production, specialization and higher incomes (Barrett, 2008). One policy to increase specialization and commercialization has been to form and support farmers’ organizations in developing countries (Bernard and Spielman, 2009; Lele, 1981; Rondot and Collion 2001). This paper examines whether membership in farmers’ organizations increase farmers’ welfare. Specifically, I investigate the impact of membership on the income, the value of agricultural production and the marketed surplus of the members of farmers’ organizations in Mozambique. Farmers’ organizations can play an important role in making agricultural development both broad based and pro-poor (World Bank, 2007). In particular, farmers organizations’ can improve smallholders livelihood by: (1) reducing transaction costs in output and input markets (Barrett et al., 2012; Kelly et al.,2003; Markelova et al. 2009; Nilsson, 2001; Poulton et al., 2010), (2) strengthening the bargaining power of the farmers in relation to buyers (Glover, 1987; Sivramkrishna and Jyotishi, 2008), (3) providing information about and access to technology and thereby increase productivity (Caviglia and Kahn, 2001; Devaux et al., 2009; Ma and Abdulai, 2016) and (4) being their voice in the political landscape (Jayne, Mather, and Mghenyi, 2010; Poulton et al., 2010). For farmers’ organizations to be pro-poor they must be inclusive as well as income enhancing, meaning that they are open for participation of the poorer part of the population and that all

the members experience an increase in income (Verhofstadt and Maertens, 2014; Fischer and Qaim, 2012). Furthermore, farmers’ organizations can be a good way for government and non-governmental organizations (NGOs) to reach the rural poor (Bernard and Spielman, 2009; Nyyssölä et al. 2014). There is an increasing number of empirical studies evaluating the impact of membership in farmers’ organizations using primary data (Chagwiza et al., 2016; Fischer and Qaim, 2012; Ma and Abdulai, 2016; Tolno et al., 2015; Verhofstadt and Maertens, 2014). These studies find a positive income effect of membership in agricultural cooperatives focused on apple production in China (Ma and Abdulai, 2016); dairy production in Selale, Ethiopia (Chagwiza et al, 2016); smallholder banana production in Kenya (Fischer and Qaim, 2012), potato production in Middle Guinea (Tolno et al., 2015) and maize and horticultural production in Rwanda (Verhofstadt and Maertens, 2014). Earlier empirical economic studies focus on the impact of a particular activity organized by a farmers’ organization, but not the membership as such. Typical market activities studied are organic farming, fair trade, export products and products sold in supermarkets (Bacon, 2005; Becchetti and Costantino, 2008; Carletto et al., 2011; Moustier et al., 2010). The main finding is that participation in these activities is related to enhanced economic welfare (Bacon, 2005; Becchetti and Costantino, 2008; Carletto et al., 2011; Moustier et al., 2010). However, there are few, if any studies, that evaluate the impact of membership at the national level based on national representative survey data. One contribution of this paper is that it evaluates the overall and general welfare impact of membership in farmers’ organizations. Thus, the results show the overall impact of farmers’ organization on economic welfare among smallholders without linking the

E-mail address: [email protected]. https://doi.org/10.1016/j.foodpol.2019.101792 Received 14 February 2018; Received in revised form 30 October 2019; Accepted 4 November 2019 0306-9192/ © 2019 Elsevier Ltd. All rights reserved.

Please cite this article as: Maren Elise Bachke, Food Policy, https://doi.org/10.1016/j.foodpol.2019.101792

Food Policy xxx (xxxx) xxxx

M.E. Bachke

results to a particular farmers’ organization, activity, crop or region in the country. I investigate the impact of farmers’ organization membership on a household’s marketed surplus, agricultural production and total income. An obvious challenge is the selection of farmers with certain valuable characteristics into farmers’ organizations. To solve this issue, I use the panel structure of the Mozambican agricultural household survey (Ministry of Agriculture, 2002a and 2005). First, following farmers in and out of membership using a difference-in-differences estimator eliminates the effect of all unobserved farmer characteristics. To further eliminate potential selection biases, I also employ a matching difference-in-differences estimator where initially comparable farmers are followed along different membership paths. I find a significant and positive impact of membership in farmers’ organizations on the marketed surplus of 25% and the value of production of 18% in the full sample. The effect on total income seems to be around 15%, which is a significant increase in income for a smallholder in Mozambique. For those who mainly depend upon agriculture for their livelihoods, the effect is even larger and the coefficients are respectively 40%, 28% and 20%. Thus, farmers’ organizations seem to reduce transaction costs, and increase market integration and agricultural production for smallholders in Mozambique. Despite this positive welfare impact, I find a surprisingly erratic membership pattern among the smallholders. This paper is organized as follows. Section 2 gives a short overview of the literature on farmers’ organizations. An overview of Mozambique and its farmers’ organizations is presented in Section 3, and Section 4 presents the theoretical framework and the empirical strategy. Section 5 introduces the data. The results are presented in Section 6. Possible mechanisms behind the results are discussed in Section 7, while Section 8 discusses the erratic membership pattern in farmers’ organizations. Section 9 concludes the paper.

Recently there has been a surge of interest in the economic effects of famers’ organization on its members (Chagwiza et al., 2016; Fischer and Qaim, 2012; Ma and Abdulai, 2016; Verhofstadt and Maertens, 2014; Tolno et al., 2015; Sinyolo and Mudhara, 2018). The studies focus on farmers’ organizations in specific regional areas and working on particular crops, and they are based on primary data. In addition to income effects of membership, they study to what degree these organizations are inclusive for both poor and rich participants. Ma and Abdulai (2016) studying apple farmers in China find that membership increases apple yields, farm net return and household income and that the effects of membership are larger for smallholders than medium and large farms. Fischer and Qaim (2012) using propensity score matching methods find that cooperative membership among smallholder banana producers in Kenya enhance income. However, the price effects are small. Information flows and technology adaption are probably as important channel for the increased income as prices. Despite inclusive cooperatives, they find positive selection of richer households into membership. Verhofstadt and Maertens (2014) find similar results in their study of an agricultural cooperative focusing on maize and horticultural products in Rwanda. In addition to positive mean income effects, they find heterogeneous treatments effects of membership where the least likely to participate gain the most. Chagwiza et al. (2016) studying income and productivity effects of membership in cooperative for dairy producers in Ethiopia find that cooperatives are particularly efficient in enhancing productivity and commercialization, while they to a lesser extent manage to positively affect prices. The results of Tolno et al. (2015) from studying potato farmers in Middle Guinea confirm the positive income effects, but also indicate clear selection effects of membership. Sinyolo and Mudhara (2018) find that group membership increase overall income among the participating farming household in KwaZulu-Natal, South Africa. However, the effects are heterogeneous. Members with higher education and maleheaded households gain more than the less educated and female headed households. From these studies, we see that farmers’ organization seem to enhance income among participants and that these organizations seem to be inclusive, but that the membership effects can be heterogeneous. Even though these studies point in the same direction, we cannot know that the results holds for all farmers’ organizations in these countries. Despite these positive findings, research focused on the ability of farmers’ organizations to link smallholders to markets in the long run show mixed results (Jayne et al., 2010; Kelly et al., 2003; Kirsten and Sartorius, 2002; Markelova et al., 2009; Poulton et al., 2010). There is evidence that the functionality of the farmers’ organization depends upon the product choice, group size, and heterogeneity among the farmers (Markelova et al., 2009) as well as the commitment level of participants and the institutional set up of the organization (Vorlaufer et al., 2012). There might be a tradeoff between the inclusiveness of the organization and the economic performance of the farmers’ organizations (Bernard and Spielman, 2009, Lutz and Tadesse 2017). Sustainability of the farmers’ organization might also be challenged by elite capture of the organization where the leadership promote their own interest over the interest of the organization (Penrose Buckley, 2007). Problems related to financing the farmers’ organisations is often a challenge and include a range of issues from membership fee payment problems among the most resource poor to money belonging to the organisation might disappear. Farmers’ organizations can both be formed independently by the farmers or supported by either governments or NGOs. However, it is rare for farmers’ organizations to self-organize, particularly to form formal organizations (Best et al., 2006). Usually there is a need for support to provide information and technical assistance and facilitate the collective action (Best et al., 2006). Support to formation or formalization of farmers groups or producer organizations, including targets on membership, are included in several developing countries agricultural policies, such as Ethiopia (Eriksen et al., 2018, Lutz and

2. Farmers’ organizations Farmers’ organizations are by definition a member owned business (Nilsson, 2001) and they usually focus on issues such as marketing, production or credit (Lele, 1981). Historically, they have a good track record of strengthening the position of the farmer in the developed world, and recently they have received renewed interest as a tool to increase market participation and welfare among smallholders in developing countries (Bernard and Spielman, 2009; Fischer and Qaim, 2012; Verhofstadt and Maertens, 2014). A standard theoretical justification for cooperatives is that they reduce high transactions costs for the economic agent (Nilsson, 2001). Generally, transaction costs are high in agricultural markets in developing countries, and smallholders face even higher transaction costs than large farmers in relation to access to skilled labor, markets and technical knowledge, and input, capital and output markets (Poulton et al., 2010). Farmers’ organizations can reduce farm-level transaction costs in output, input and credit markets (Kelly et al., 2003; Markelova et al., 2009) and costs related to access to technical, marketing and management knowledge (Jayne et al., 2010). Another justification is that farmers’ organizations can rectify the unequal balance of power between the contractor and the smallholder, and be a tool to avoid monopsonistic exploitation of smallholders in contract schemes (Glover, 1987; Sivramkrishna and Jyotishi, 2008). To strengthen smallholders power in negotiations with buyers through farmers’ organizations is even more pertinent in a time of increasing vertical integration in agribusiness (Reardon and Weatherspoon, 2003; Sykuta and Cook, 2001). And particularly, when contract farming is seen as a good way to integrate the smallholders into international markets (Kirsten and Sartorius, 2002; Fischer and Qaim, 2012). This development requires product traceability, quality and adhesion to standards, all of which create higher transaction costs for smallholders than for other farmers (Poulton et al., 2010). 2

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Tadesse (2017), Mozambique (Republic of Mozambique, 2001; 2006) and South Africa (Department of Agriculture, Forestry and Fisheries, 2012). Financial and technical support to initiate farmers’ organizations, though necessary, can create organizational long-term sustainability problems and dependency on the external actors (Bingen et al., 2003; Markelova et al., 2009, Fischer and Qaim, 2012). One challenges is that public policy goals might not be in line with the farmers’ organizations objectives and therefore create tensions within the organizations (Shiferaw et al., 2011). Examples of such policies are social objectives, which do not align itself well to a commercial objective of a famers’ organization (Shiferaw et al., 2011). There is also some evidence that farmers’ organizations activities fade way when the project initiating them ends (Ochieng et al., 2017). Thus, there is a fine balance between support and creating dependency (World Bank, 2007). Despite these challenges, farmers’ organizations can play an important role for smallholders in reaching the market by improving coordination, but public investment such as roads, contract enforcement facilities and literacy programs need to be in place for them to be successful (Kelly et al., 2003; Markelova et al., 2009).

central Mozambique are not formally registered, but associated with UNAC, indicating that farmers’ organizations are even more numerous in Mozambique. The use of representative data offers good options for identifying general income effects, but at the same time, I lose details on the exact activities of each farmers’ organizations in the sample. Unfortunately, there is no overall qualitative information on the different farmers’ organizations in Mozambique and it is therefore not possible to completely group the different farmers’ organisations according to objective, performance or sustainability issues (Gêmo and Babu, 2019). However, based on literature, reports and anecdotal information from field visits in 2007 and 2009,4 it is possible to shed light on different pertinent issues regarding the functioning of farmers’ organizations in Mozambique during the time of the study. The NGOs working with the farmers’ organizations often provide capacity-building activities such as literacy programs and institutional training, technical assistance, access to extensions services and assistance regarding credits, microcredits and marketing issues including warehouses and milling equipment. Technical agricultural assistance includes promotion of high value commercial products, provision of improved seeds and animals for reproduction, and teaching of different agricultural practices such as conservation agriculture and seed multiplication (Kaarhus and Woodhouse, 2012; Kelly et al., 2003). Uaiene et al. (2009) show that membership in a farmers’ organization increases the probability of using new technology in Mozambique. Boughton et al. (2007), in their study on market participation among smallholders in Mozambique, find a positive, albeit not significant, correlation between market participation and membership in farmers’ organizations. According to their study, farmers’ organizations is a source for market and production information for their members. Due to capital and literacy constraint among potential members, they consider farmers’ organizations more as a public good in Mozambique than the club-good that it would normally be regarded as. Thus, the performance of the farmers’ organizations in Mozambique might be affected by public good characteristics such as free-riding and non-excludability. Human-capacity building is important for farmers’ organizations to function and to be sustainable (Bingen et al., 2003; Kelly et al., 2003) as is the formalization5 of the farmers’ organizations. Formalization has been a focus for many of the NGOs as this will enable the farmers’ organizations to access credit and finance (Gêmo and Bahu, 2019). The aid program of the Cooperative League of the USA (CLUSA) in Mozambique from 1995 is a good example of how organizations can work with human-capacity building in farmers’ organizations to turn them into formal and sustainable organizations (Bingen et al., 2003; Kelly et al., 2003). In their project CLUSA managed to organize 26 000 smallholders, improve the farmers’ relations with the traders which resulted in higher farm gate prices and a 85% reported increase in average annual farm revenues (Pearce and Reinsch, 2005). The improved relationship between the smallholder and the trades resulted in better access to credit and finance allowing for better access to inputs (Pearce and Reinsch, 2005). These positive effects were the results of a long process of supporting the farmers’ organizations with the aim of making them self-managed, also financially. Farm credits, credit management and credit monitoring was an objective so the farmers’ organization could manage group loans. They also focused on training on pertinent issues such as marketing, fiscal, organizational and agricultural skills, market knowledge, literacy and community lobbing power (Pearce and Reinsch, 2005). Almost all farmers’ organizations in Mozambique seem to be connected to funding entities such as international NGOs (Kaarhus and

3. Poverty, agriculture and farmers’ organizations in mozambique Agriculture is the main economic activity for the majority of the Mozambican population and poverty rates are higher in rural areas (Arndt et al., 2012). Increased agricultural production was one of the contributing factors for the 15 percentage points reduction in poverty rates in the six-year period prior to 2002/2003 (Arndt, James, and Simler, 2006). In the subsequent six-year period, poverty rates did not fall further. Low growth in agricultural productivity, weather shocks and high international food prices were among the factors hampering poverty reductions in this period (Arndt et al., 2012). The agricultural sector in Mozambique is still characterized by low production, low productivity, low use of inputs, slow adaption of new technology and a high marketing wedge that excludes many subsistence farmers from the market (Arndt et al., 2012; Heltberg and Tarp, 2002; Uaiene et al., 2009). Agricultural markets were not well developed in Mozambique at the time of the study. Market information on staples was scarce and most produce were sold in local markets with the exception of certain cash crops. Development in the agricultural sector is therefore important for poverty reduction in Mozambique. Promoting farmers’ organizations is a priority in Mozambique’s poverty reduction strategies and agricultural sector programs in the new millennium (Republic of Mozambique, 2001; 2006). Policies and activities promoting farmers’ organizations also preceded the poverty reduction strategies and have been ongoing since the mid-1990s with the support of international NGOs and donors (Dorsey and Muchanga, 1999; Pearch and Reinsch, 2005). The facilitation and formation of farmers’ groups have been a priority of both private and public efforts and supported by international NGOs and donors (Gêmo and Babu, 2019). These efforts1 continued during the early 2000s, both through national NGOs such as UNAC2, international NGOs such as CARE, World Vision, Oxfam and CLUSA and through government policies. Consequently, there were around 4600 formally registered3 local organizations in Mozambique focusing on improving rural livelihoods in 2006 (Francisco and Matter, 2007). In addition to the formally registered farmers’ organizations, many organizations in northern and 1 . This session is based on interviews conducted in Maputo, Mozambique, in May 2009. 2 . UNAC – União Nacional de Camponeses. It is also a member of the international movement Via Campensina. 3 . Most of these farmers’ organizations were registered as NGOs under the Association Law (Lei 8/1991) and not under the Cooperative Law (Lei7/79) (Kaarhus and Woodhouse, 2012).

4 The author also worked in the Ministry of Trade and Industry in the period 2002–2004 on among other issues agricultural price information systems. 5 At the time, it was a legal requirement to have at least 10 members to be legally recognized.

3

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Woodhouse, 2012) and farmers’ organizations are also often a vehicle for distributing different types of support, from both NGOs and the local government implementing the national policies (Nyyssölä et al., 2014, Gêmo and Babu, 2019). An example of this we find in the study by Nyyssölä et al. from 2014, where they look at the impacts on farmers participating in a NGO run program in the Gaza province in Mozambique. The objective of the NGO program was to improve agricultural techniques (new seeds, fertilizer) and food security among smallholders through a based-group approach to technology adoption. They find that the program was successful the first year for all outcomes, including a steep increase in the use of fertilizers among the participating farmers. However, this effect disappeared the second year, probably due to a delay in the aid provided (fertilizer) by the NGO combined with a worsening drought. The new farming techniques did not therefore not seem profitable to the farmers and the farmers did not adopt the new farming techniques. Farmer groups were used to distribute the aid. UNAC is another example of a farmers’ organization in Mozambique. This organization is seen as a more political organization with less focus on development activities. Other farmers’ organizations in areas such as Marracuene in Maputo province hold titles to irrigated land. This has a historical reason as certain areas was defined as “machambas de povo” which were created during the war to feed the army. In Boane, members in a farmers’ organization produced bananas in the same irrigated field, but they sold the produce individually. The result was that the buyer could put one seller up against another seller in the same field (field visit 2009). Irrigation, and not marketing, was the focus of this organization. Despite the efforts since the early 90-ties, the organization of farmers was, and still is, at early stages of development both at local and national levels.

farmers’ organization might also provide technical assistance and technology, so that the production function satisfies Q(n;z,m = 1) > Q (n;z,m = 0) for all n, z. Thus, one could also expect to see higher yield among farmers that are members of farmers’ organizations. Higher overall production might also lead to more produce being sold, i.e., Qs(n;z,m = 1) > Qs(n;z,m = 0). My hypothesis is that a member household of a farmer’s organization have higher total income, value of agricultural production and marketed surplus than a non-member household. (b) Empirical method The objective of this paper is to analyze the impact of membership in a farmers’ organization on the three outcome variables. The main challenge with impact assessments is that it is impossible to observe a household that is both a member and a non-member at the same time, and therefore there is a need to construct a counterfactual. There is a large literature on impact assessments, how to construct counterfactuals and estimating the treatment effect. It is not the objective of this paper to review this literature; for more information see, for example, Heckman, Ichimura, and Todd (1997), Heckman, Smith, and Clements (1997), Heckman, Ichimura, and Todd (1998), Blundell and Costa Dias (2000), Heckman and Navarro-Lozano (2004), Smith and Todd (2005) and Ravallion (2007). Formally, let m indicate membership in farmers’ organizations, and Yitkm be household i’s income of type k in period t conditioned on the membership status. If the household is a member (m = 1) of a farmers’ organization, the outcome is Yitk1, and if not (m = 0) the income is Yitk0. As before, k consists of three types of outcome variables (MS, VA and TI). To enhance readability, the subscript k will be dropped in the equations. The impact of being a member of a farmers’ organization is then described by Yit1 Yit0 . The panel structure of the data allows for difference-in-difference estimators that eliminate biases due to temporally invariant omitted variables such as farmers’ ability. Such unobservable variables are difficult, if not impossible, to measure, and are important in determining the outcome variables in this case. I apply both a conventional difference-in-difference estimator, and a propensity score matching difference-in-difference estimator to show that the results are stable across different estimators.6

4. Analytical framework and empirical method (a) Analytic framework An analytical framework is developed to clarify and discuss possible impact of membership in farmers’ organizations on the outcome variables. This framework is based upon the literature review and knowledge of how farmers’ organizations work in Mozambique. In the framework, a farmer with a vector of characteristics z and membership status m {0, 1} obtains income:

Y ( m ) = p ( m ) Q (n ; z , m )

n'r (m )

Fee .

(i) The difference-in-difference estimator The main assumptions for this estimator are: i) a common trend between the members and non-members, and ii) no changes in group composition within each group (Blundell and Costa Dias, 2000). The common trend assumption implies that macro shocks affect both members and non-members in the same manner and therefore do not increase or decrease the impact of the treatment. If the two groups change in composition this might also affect the outcome. The estimated equation is:

(1)

Here, production Q depends on agricultural inputs used (n), as well as characteristics (z) and membership status (m). The latter affects the production through improved production technologies and farmer ability and knowledge. The product price p(m) and the input prices r (m) also depend on membership status as membership can affect the prices the farmers pay and receive. Fee is the financial cost of being a member. This cost is normalized to zero as most members pay little or nothing for membership because most farmers’ organizations in Mozambique are supported by different types of NGOs (Boughton et al., 2007; Kaarhuse and Wouldhouse, 2012). Membership in farmers’ organizations would also normally require investment in the form of time and work, a cost that is not included in the analytical framework. Possible implications of this is discussed later. I hypothesize that membership (m) in a famers’ organization might influence the farmers income through the following channels. First, by securing the farmer a better price for her produce than the farmer otherwise would get, i.e., p(m = 1) > p(m = 0). This results from an improved negotiation position for farmers’ organizations relative to single farmers and reduced transaction costs as larger quantities are sold. Second, membership can provide lower input prices, i.e., r (m = 1) < r(m = 0) as the farmers’ organization buys relatively larger quantities compared with the individual smallholders. Third, the

Yit =

i

+ mit + 'Xit +

t year

+ µit ,

(2)

where Yit is the outcome variable of household (i) at time t, t ∈ {2002,2005}, δ is the impact of membership (mit), αi are unobservable time-invariant factors such as farmers’ ability, which will be differenced away, and β are the parameters for the control variables (Xit), where of the time invariant variables, gender of head of household, ownership of land and the provinces will also be canceled out. The set of control variables (Xit) consists of all the other household characteristics7 6

. Other methods such as instrument variables could have been used. However, good instruments for membership in farmers’ organizations are rare, and I have not been able to identify such a variable in this case. 7 . Except for “Agriculture as the main cash income” because the sample is divided into subsamples using that variable. 4

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Table 1 Descriptive statistics. Comparison of mean values between members and non-members in the full sample. 2002

Head of household characteristics Age (years) Share, male head household Schooling (year) Self-employment (dummy) Salary work (dummy) Ag. main activity (dummy) Household characteristics Household size (nu.) Pensioner (dummy) Ag. main cash income (dummy) Serious disease (dummy) Assets Ag. land (hectares) Ownership of land (dummy) Good roof (dummy) Good walls (dummy) Agricultural practices Family labor (nu.) Animals (nu.) Animal traction (dummy) Irrigation (dummy) Fertilizers (dummy) Pesticides (dummy) Manure (dummy) Number of obs.

2005

Full sample

Members

Non-members

Difference

Full sample

Members

Non-members

Difference

43.02 0.75 2.55 0.32 0.15 0.83

43.18 0.75 3.09 0.36 0.16 0.85

43.01 0.75 2.53 0.32 0.15 0.83

0.16 0.00 0.57*** 0.04 0.01 0.02

45.4 0.71 2.90 0.42 0.25 0.82

44.6 0.75 3.72 0.49 0.28 0.81

45.4 0.71 2.84 0.41 0.25 0.82

–0.83 0.04 0.88*** 0.08*** 0.03 –0.02

5.23 0.03 0.57 0.07

5.96 0.04 0.59 0.08

5.20 0.03 0.57 0.07

0.76*** 0.01 0.01 0.01

6.55 0.03 0.57 0.21

7.53 0.05 0.55 0.19

6.46 0.02 0.58 0.22

1.07*** 0.03*** –0.03 –0.03

1.71 0.99 0.16 0.34

1.91 0.99 0.29 0.43

1.70 0.99 0.15 0.34

0.21 0.00 0.14*** 0.10**

2.01 0.98 0.19 0.41

2.46 0.99 0.28 0.45

1.98 0.98 0.18 0.41

0.48*** 0.01 0.10*** 0.04

2.82 3.40 0.14 0.14 0.04 0.06 0.07 3480

3.22 5.37 0.21 0.27 0.16 0.17 0.15 152

2.80 3.13 0.13 0.13 0.03 0.06 0.07 3328

0.42*** 2.24*** 0.08*** 0.14*** 0.13*** 0.11*** 0.08***

3.78 3.10 0.11 0.06 0.04 0.06 0.04 3480

4.34 4.72 0.15 0.17 0.12 0.11 0.07 265

3.74 2.96 0.11 0.05 0.03 0.05 0.03 3125

0.60*** 1.76*** 0.05** 0.11*** 0.09*** 0.06*** 0.03***

Notes: Significance level of t-test: nu. indicates a number. a Number of people in the household above 15 years of age. *p < 0.10. ** p < 0.05. *** p < 0.01.

presented in Table 1 and the 10 provinces in Mozambique. Finally, ηt is the trend effect for the period between 2002 and 2005 and uit is the disturbance term.

members of a farmers’ organization. The propensity score for membership in farmers’ organizations is obtained using the following probit estimator:

(ii) Difference-in-difference propensity score matching estimator

Pr ( mi=1|Xi ) =

(3)

( 'Xi ),

where Pr is the probit estimator, mi is the change in membership of farmers’ organizations between 2002 and 2005, and β are the estimated parameters for the corresponding vector of covariates (Xi) explaining whether or not the household is a member of a farmers’ organization in 2002. Φ is the cdf of the standard normal. In this analysis, I first study the impact of becoming a member ( mi = 1) between 2002 and 2005, and I compare the farmers that becomes members with farmers who were neither members in 2002 nor in 2005. Then, I look at the impact of leaving a farmers’ organization (Δmi = 1) between 2002 and 2005, and I compare the farmers that leave farmers’ organization with farmers who were a member in both years. The reason I only compare two outcomes is that my estimator only handles binary treatment variables. The second step is to estimate the average treatment effect on the treated (ATT). The average treatment effect on the treated, estimated with the difference-in-difference matching estimator, based on Smith and Todd (2005) is:

The assumption of a common trend, crucial for the validity of standard difference-in-difference estimators, might not be a realistic description of the situation in Mozambique. There are differences in rainfall, productivity and economic growth patterns across the country, as well as differences in governmental and NGO activities related to farmers’ organizations. In this case, the difference-in-difference propensity score matching estimator will be more robust, as it matches similar households and thereby controls for both observed heterogeneity and indirectly unobserved heterogeneity (Heckman, Ichimura, & Todd, 1997). This estimator is particularly well suited in my case as the data fulfills the three common criteria for the estimator to perform well (Blundell and Costa Dias, 2000; Heckman, Ichimura, & Todd, 1997, 1998). First, only one data source is used. Second, the data contain relevant information for both the outcome variables and the membership decisions for the farmers. Finally, all the farmers, both members and non-members, act in local agricultural market exhibiting common characteristics such as a high marketing wedge and similar market channels. As members and non-member are found throughout the country, there is little reason to believe that there are other systematic difference in their market conditions beside the membership aspect. The estimation method used is based on Heckman, Ichimura, and Todd (1997, 1998) and Smith and Todd (2005)’s two step estimator. The first step is to construct a control group by matching member of farmers’ organizations households to similar households that are not

ATTDDM =

1 n1

1

{ (Yti i I1

Yt0'i )

v (i, j )(Ytj0

Yt0'j ) } ,

(4)

where t = 2005 and t′=2002. I1 is the set of those who become members in the area of common support, which is where propensity scores estimated in Eq. (3) are overlapping between member households and non-member households, and I0 are the set of non-members that are in the area of common support. v(i,j) represents the weighting regime in my matching estimator. I use a kernel estimator to match the members 5

Food Policy xxx (xxxx) xxxx

M.E. Bachke

to non-members on observables8 as it has been shown that imposing kernel matching and common support improves the estimates compared with other matching algorithms (Heckman et al., 1998). The estimator require a binary treatment variable. This estimator rests on two assumptions. The first is that E [Yit Yit ' | Xi , m = 1] = E [Yit Yit ' | Xi , m = 0], in words conditional mean independence (Smith and Todd, 2005). This means that based on the matching, you will find non-member households (m2002 = 0) that stay non-members (Δm = 0, m2005 = 0) with outcome variables equal to households that become members (Δm = 1; m2002 = 0, m2005 = 1) based on the observables and independent of membership status. This assumes that selection into the program occurs on the observables and not the unobservables (Blundell and Costa Dias, 2000). The second assumption requires that one cannot use individuals with characteristics for which the variables perfectly predict the change in membership status; 0 < Pr (( m = 1|Xi ) < 1. As I am only interested in homogenous treatment effects, the latter criteria can be relaxed to hold only in the area of common support. Rosenbaum and Rubin (1983) show that one could use propensity score matching instead of matching on each specific covariate, and thereby solve the dimensional problem of matching.

produce sold by the household and is defined as follows:

YiMS = pd ·qis ,

where the vector qis represents the amount sold by the household of each agricultural product and pd is the vector of median district prices10 for all agricultural products of the household (i). The value of agricultural production (VA) is defined as follows:

YiVA = pd ·qi ,

YiTI = Y VA

The data come from the official agricultural household survey produced by the Ministry of Agriculture in Mozambique with the assistance of Michigan State University. I use panel data collected in 2002 (Ministry of Agriculture, 2002a) and 2005 (Ministry of Agriculture, 2005), which represents the only panel in the data. This is the latest panel available from Mozambique. The sampling is based on the Agricultural and Livestock Census from 2000 and use the National Statistics Institute’s standards to make the sample representative at the provincial and national level. The household survey collected detailed information on household characteristics, welfare indicators, landholdings, employment types and remittances, as well as detailed information regarding farming practices, crops produced, harvested and sold, and livestock. Additional information on prices, marketing and certain infrastructure measures was collected in a community level survey. The original sample consists of both small-scale, medium and large farmers, however, I have chosen to only work on the small-scale farmers.9 I use the balanced panel consisting of 3480 households included in both years. These households are later referred to as the full sample. I also use a subsample, later referred to as the restricted sample, consisting of the 1998 households that have agriculture as their main cash income source. I include the restricted sample because membership in farmers’ organizations should be particularly interesting for households where agricultural is the main cash income source and I also would expect to see larger effects on total income for this sample.

The main independent variable in this analysis is a dummy variable indicating membership in a farmers’ organization. The question in the survey is: “Is the person responsible for this farm or any other household member a member of an agricultural association?” 13(Ministry of Agriculture, 2002a, p. 4). The answer is either yes or no with no indication of what type of organization. In the questionnaire guide presented to the enumerator, the following definition of a farmers’ organization is given “an agricultural association is an association of farmers or agricultural producers or livestock producers oriented towards fulfilling common interests, with regards to production, processing or marketing of agricultural products. The organization might or might not be formally

I study the following three outcome variables: the marketed surplus (MS), the value of agricultural production (VA) and total income (TI) per household. The marketed surplus (MS) measures the overall value of the

. The weights aredefined as v (i , j ) =

I0 G

10 . If district prices are not available, then provincial prices are used, and national prices are used if provincial prices are not available. 11 . Both related and not related to the agricultural sector. 12 . The cost of casual labor is not included because of data limitations in 2005. 13 . In Mozambique, for historical reasons, farmers’ organizations have been called associations; however, I will use the word organizations throughout the paper as this is more widely used in the literature.

pscorei a

pscore

pscorei

(7)

(c) Descriptive statistics

(b) The outcome variables

8

ci + other incomes,

where YiVA is the value of the agricultural production (as defined above), ci is the cost of inputs and other incomes consists of salary work, own-business,11 remittances and pensions. The inclusion of these other income items is in line with other studies using the same data (Cunguara and Darnhofer, 2011). The inputs12 (ci) included are seeds, fertilizers and pesticides and costs of buying animals. Family labor and cost of land are not included, which is consistent with the literature. The median price at the district level is chosen because we have farm gate prices for only a few products, and because the district, not the national, price is the best proxy price because of the strong market segmentation in Mozambique (Heltberg and Tarp, 2002). Using the district price means that we lose important price information from choice of marketing channel and product quality. However, product differences and combinations are included in the study as the survey covers all agricultural produce. There is also possible that the local prices are lower than the district prices, however, as members and nonmember are scattered throughout the country, there is little reason to believe that there are any systematic difference between members and non-members. The major drawback with using district prices is that the data do not permit us to study an important potential benefit of being a member in a farmers’ organization, namely, price differences on produce between members and non-members. If anything, using median district prices to calculate the impact variables for all farmers would potentially underestimate the economic benefit of being a member.

(a) Data source

pscorej

(6)

where the vector qi represents the overall agricultural production of the household and pd is defined as above. The estimated value of the agricultural production of the household is a lower bound because we have only sales values and not the overall value of the goods produced for horticulture, fruits and animal products. The share consumed at home is therefore not included. The total income (TI) of the household is defined as follows:

5. Data and descriptive statistics

j I0 G

(5)

where a is the

a

bandwidth of the kernel and κrepresents the number in the kernel group. The kernel used is Gaussian with a bandwidth of 0.06. 9 In the Mozambican National Agricultural Survey a smallholder is defined as a farmer that has less than 10 ha of cultivated land, less than 10 cattle, less than 50 small animals such as goats, sheep and pigs, and less than 5000 poultry. 6

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Table 2 Descriptive statistics. Comparison of mean income values between members and non-members in farmers’ organizations in both samples. Full sample 2002

2005

Outcome variables

Full sample

Members

Non-members

Difference

Full sample

Members

Non-members

Difference

Marketed surplus (MTNa) Value of ag. prod. (MTNa) Total income (MTNa) Number of obs.

1011 4145 8187 3480

2563 6852 11,973 152

940 4021 8015 3328

1623*** 2831*** 3959***

1399 4915 10,248 3480

3440 7796 16,702 265

1231 4677 9716 3125

2210*** 3118*** 6986***

Restricted sample

Members

Non-members

Difference

2028 5739 8386 1998

5424 10,535 16,304 146

1760 5360 7761 1852

3664*** 5175*** 8542***

Restricted sample 2002 Outcome variables

Restricted sample a

Marketed surplus (MTN ) Value of ag. prod.(MTNa) Total income (MTNa) Number of obs.

1408 4657 6162 1998

2005 Members 3257 7048 9328 89

Non-members

Difference ***

1322 4545 6015 1909

1935 2503*** 3314***

Notes: Significance level of t-test: *p < 0.10. ** p < 0.05. *** p < 0.01. a Income is measured in MTN = 2005 Meticais. 1 US$ was about 24 Meticais. 2002 numbers are PPI adjusted.

registered.”14 (Ministry of Agriculture, 2002b, p. 18). Thus, there is no information in the database regarding the specific name, form or functions of the farmers’ organizations, and therefore this study measures the impact of membership in any type of farmers’ organization. Even if this information was included in the survey, it would not necessarily provided good qualitative information on the farmers’ organization, as there is no complete register with qualitative information of famers’ organizations in Mozambique (Gêmo and Bahu, 2019). Despite the efforts to organize smallholders in Mozambique, only about 7.6% of the farmers in the full sample from 2005 belonged to a farmers’ organization, up from 4.4% in 2002 (Ministry of Agriculture, 2002a and 2005). There are certain regional differences with regard to membership rates; the highest is in the most southern and northern provinces. Moreover, these provinces also had the highest membership increase, while other provinces did not have any increase in membership. Finally, only 47 farmers in the full sample stayed members in both years, 218 became members from 2002 to 2005 and 105 left the farmers’ organizations. The comparable numbers for the restricted sample are; 21 members in both years, 125 new members between 2002 and 2005, and 68 members left the organizations over the same period. Thus, the overwhelming majority of the households in both samples were members in neither year. Table 1 shows descriptive statistics as well as the differences in the mean values of the salient characteristics of the households of members and non-members in the full sample for 2002 and 2005. As seen in Table 1, members in farmers’ organizations have better houses and own more land (only significant in 2005) than non-members and apply better agricultural technologies than non-members. The head of member households have more years of schooling, are to a greater degree self-employed (only significant in 2005) and their households have more members and a higher share of pensioners. Table 2 shows the mean values of the three outcome variable measures for both samples, divided into members and non-members. We see from the descriptive statistics that members have significantly more marketed surplus, value of production and income than non-

members in both samples and both years. This difference in the income variables might be explained by the difference in the size of the land holdings between members and non-members. As I control for the size of the land holdings this should not affect my results. The Kolmorogov–Smirnov test of equality of distributions confirms that the income distributions of members are different from the income distributions of non-members at the 1% level in both samples. 6. Results (a) Difference-in-difference estimator Table 3 presents the results of the difference-in-difference estimators presented in equation15 (2) for the full and the restricted sample. I have estimated five different specifications of the model (A–E) based on inclusion of different controls. From Table 3, we see that membership has a strong and significant impact on almost all of the outcome variables. The only exception is for total income in the two specifications, which includes agricultural practices, in the full sample. As expected the coefficients are higher in the restricted sample where agriculture represents the main cash income source. The impact is highest for the marketed surplus (25–34% in the full sample and 40–49% in the restricted sample) compared to the other outcome variables. The impact on the value of agricultural production and the total income is just below 20% in the full sample and just above 20% in the restricted sample. The impact coefficients are higher in the specifications where fewer controls are included, and including agricultural controls reduces the estimated impact the most. This is in line with theory, where lower coefficients are expected with more controls, particularly controls that are arguably important for the outcome such as agricultural practices in this case. (b) The difference-in-difference propensity score matching estimator

15 . The regressions have also been clustered at the household level, the district level and the provincial level, and the results are similar. These results are available from the author upon request.

14

. Registration would be under either the Association Law (Lei 8/1991 – Lei das Associacões) or the Cooperative Law (Lei 7/79). 7

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Table 3 Summary results. Diff-in-diff estimator results for both samples. Full sample A Marketed surplus Value of agricultural production Total income Control variables Head of household characteristics Household characteristics Assets Agricultural practices

Restricted sample B

***

0.335 (2.92) 4834 0.195** (2.58) 6831 0.190** (2.12) 6873

C ***

D **

E ***

A **

0.336 (2.95) 4829 0.183** (2.41) 6822 0.166* (1.93) 6863

0.260 (2.41) 4809 0.152** (2.09) 6787 0.133 (1.55) 6778

0.323 (2.87) 4820 0.188** (2.44) 6804 0.145* (1.68) 6808

0.254 (2.38) 4809 0.164** (2.23) 6787 0.137 (1.59) 6778

x x

x x

x x x

x x x x

x

B ***

0.499 (4.12) 3843 0.260*** (3.03) 3970 0.254** (2.47) 3966

C ***

D ***

E

0.496 (4.15) 3480 0.248*** (2.89) 3969 0.217** (2.29) 3963

0.401 (3.56) 3474 0.197** (2.36) 3959 0.167* (1.76) 3949

0.483 (4.10) 3477 0.252*** (2.93) 3963 0.198** (2.10) 3954

0.401*** (3.64) 3474 0.210** (2.54) 3959 0.175* (1.85) 3949

x x

x x

x x x

x x x x

x

***

Notes: t-statistics in parentheses. Number of observations in italics. The outcome variable is in natural logarithms. These estimations are clustered at village level sites. Details about head of household characteristics, household characteristics assets and agricultural practices controls are specified in Table 1. * p < 0.10. ** p < 0.05. *** p < 0.01.

balancing property16 (Gilligan and Hoddinott, 2007; Morgan and Winship, 2007; Ravailion, 2007). This ensures that there are no conditional differences between the member and non-member households. Furthermore, these specifications predict the difference between the members and the non-members, and thus have explanatory power. Table 5 below shows the balancing table, which presents the results of running the propensity score matching estimator on each of the farmers characteristics described in Table 1 for those included in the treatment and the comparison group. From Table 5, we see that there are few significant differences between those who become members (the treated) and those who stay non-member (untreated). These results indicate that potential biases are handled well in the estimation. In particular for the second specification (Model E), which includes agricultural practices, are there few significant differences between those who become members and those who stay non-members. This is not surprising as this model also secure balancing on agricultural practices, which is not the case for the first specification (Model D). The matching seem to be even stronger in the restricted sample where there is only one significant difference, the good roof variable. In the full sample, members have larger households, more family labor, larger landholdings, more animals and are more likely to use fertilizers and manure. The significant differences in the agricultural practices fall in both size and significance level when agricultural practices are included in the propensity score matching (specification 2 - Model E). In Table 6, the results from the difference-in-difference propensity score matching estimator (Eq. (4)) are presented. We see that all outcome variables except the total income are significant in both specifications. The difference-in-difference propensity score estimator is very robust, but less efficient than the conventional difference-in-difference estimator. Therefore, it is not surprising that the degree of significance for the coefficients would be lower with the difference-in-difference propensity score estimator than with the conventional estimator.

The membership regression is over parameterized in line with the literature, and I have included covariates that are both related to the membership decision as well as to the agricultural output variables (Heckman, Ichimura, & Todd, 1997; Heckman and Navarro-Lozano, 2004). In particular, I included agricultural practices and land in the membership equation as it explains membership in farmers’ organizations quite well. However, agricultural practices might be endogenous to the membership decision; thus, I use two specifications where one includes agricultural practices in the participation regression, and one does not. Table 4 presents the propensity scores estimated by Eq. (3) for whether the household became a member in a farmers’ organization between 2002 and 2005 for both specifications. Table 4 also includes the propensity score estimates for leaving an organization, which I will discuss later. The younger the head of household is, the more likely it is that the household will join a farmers’ organization. Furthermore, it appears that households with the more members, with a member that receives a pension or that have more land are more likely to join farmers’ organizations than households without these characteristics. This indicates that richer households tend to join. There is a slight contradiction as the variable quality of roof indicates the opposite. The use of more modern agricultural practices such as fertilizers also correlates with membership. Finally, the results show membership decision vary geographically suggesting that availability of organizations to join across the country varies. Compared with Maputo Province, living in all other provinces means a household is less likely to be a member except for Gaza and Niassa Provinces in both samples and also Nampula and Sofala in the restricted sample. Gaza Province is the neighboring province to Maputo province, and probably has the same capital effect as Maputo Province. Niassa Province is the most northern province, and far from the ocean. These are also the provinces with the largest increase in membership. The results for Niassa Province are probably due to the considerable amount of NGO activity in this province. In Table 4, ownership of land was dropped in both specifications in the restricted sample as well as gender in the first specification and Gaza province in the second specification. This is to satisfy the

16 . The Stata algorithm is developed by Becker and Ichino (2002). The balancing property is the test that ensures that each parameter is balanced between members and non-members in the groups that the sample is divided into on the basis of the propensity score. The test used is a t-test.

8

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Table 4 First stage propensity score results for the diff-in-diff matching estimator. Becoming a member

Leaving

Full sample

Head of household characteristics Age (years) Gender (dummy) Schooling (years) Self-employed (dummy) Salary work (dummy) Agriculture primary activity (dummy) Household characteristics Household size (nu.) Pensioner (dummy) Serious disease (dummy) Assets Agricultural land (hectares) Ownership of land (dummy) Good roof (dummy) Good walls (dummy) Agricultural practices Family labor (nu.)

Model E

Model D

Model E

Model D

Model E

–0.00574** (–2.16) 0.0285 (0.31) 0.0108 (1.11) 0.00556 (0.07) 0.0445 (0.39) –0.0344 (–0.29)

–0.00577** (–2.14) 0.00581 (0.06) 0.00942 (0.96) 0.0108 (0.13) 0.0561 (0.48) –0.0405 (–0.34)

–0.00982*** (–2.76) 0.0101 (0.72) 0.0606 (0.56) 0.141 (0.83) –0.0976 (–0.52)

–0.00998*** (–2.75) –0.0577 (–0.48) 0.00977 (0.68) 0.0671 (0.61) 0.131 (0.75) –0.116 (–0.60)

–0.00317 (–0.32) 0.196 (0.61) –0.00307 (–0.09) –0.0434 (–0.17) –0.397 (–0.97) 0.136 (0.33)

–0.00713 (–0.65) 0.378 (1.10) –0.0209 (–0.54) –0.220 (–0.79) –0.402 (–0.94) 0.248 (0.57)

0.0264** (2.02) 0.347* (1.79) 0.0924 (0.68)

0.00425 (0.22) 0.377* (1.93) 0.0773 (0.56)

0.0193 (1.03) –0.325 (–0.65) –0.188 (–0.91)

0.00209 (0.08) –0.251 (–0.50) –0.245 (–1.15)

–0.0701 (–1.48) 0.755 (1.00) –0.994** (–2.01)

–0.132* (–1.78) 0.455 (0.53) –1.212** (–2.24)

0.0810*** (3.88) 0.326 (0.69) –0.252** (–2.05) 0.0445 (0.57)

0.0669*** (3.26) 0.272 (0.59) –0.289** (–2.32) 0.0352 (0.44)

0.0628*** (2.67)

0.0587** (2.48)

–0.215* (–1.96)

–0.264** (–2.08)

0.165 (0.91) 0.152 (1.49)

0.113 (0.62) 0.145 (1.41)

0.0621 (0.17) –0.385 (–1.40)

0.0925 (0.23) –0.391 (–1.29)

0.0455 (1.27) 0.00962* (1.83) –0.00243 (–0.02) 0.509*** (3.13) 0.0167 (0.12) 0.262** (2.03)

Animal traction (dummy) Fertilizers (dummy) Pesticides (dummy) Manure (dummy)

Cabo Delgado Nampula Zambezia Tete Manica Sofala Inhambane Gaza Constant

Full sample

Model D

Animals (nu.)

Provinces Niassa

Restricted sample

–0.238 (–1.27) –0.712*** (–3.78) –0.492*** (–2.76) –0.819*** (–4.57) –0.644*** (–3.47) –0.916*** (–4.55) –0.987*** (–4.61) –0.757*** (–4.05) –0.118 (–0.75) –1.353*** (–2.68)

0.0449 (0.88) 0.00154 (0.20) 0.0522 (0.31) 0.434** (2.25) –0.0371 (–0.23) 0.278 (1.59)

–0.206 (–1.07) –0.673*** (–3.49) –0.418** (–2.29) –0.727*** (–3.98) –0.715*** (–3.62) –0.928*** (–4.52) –0.937*** (–4.35) –0.787*** (–3.99) –0.103 (–0.61) –1.369*** (–2.73)

–0.182 (–0.58) –0.577* (–1.85) –0.412 (–1.37) –0.892*** (–2.90) –0.638** (–2.06) –0.711** (–2.24) –0.602* (–1.83) –0.979*** (–2.76) –0.145 (–0.49) –0.816** (–2.15)

–0.0467 (–0.20) –0.392* (–1.70) –0.205 (–0.94) –0.684*** (–3.02) –0.644*** (–2.79) –0.578** (–2.50) –0.404 (–1.63) –0.922*** (–3.20) –0.977*** (–3.08)

0.0740 (0.60) 0.0363* (1.66) –0.940** (–2.37) –0.993** (–2.34) –0.00540 (–0.01) 0.0875 (0.20)

0.969* (1.70) 0.923* (1.66) 0.920* (1.73) 1.699*** (2.58) 1.489*** (2.70) 1.532** (2.01) 0.0664 (0.13) 1.483** (2.50) 0.674 (0.78)

0.619 (0.98) 0.682 (1.11) 0.649 (1.10) 3.033*** (3.51) 1.668*** (2.65) 1.203 (1.39) 0.220 (0.36) 1.950*** (2.95) 1.181 (1.27)

(continued on next page)

9

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Table 4 (continued) Becoming a member

Leaving

Full sample

Pseudo R2 Number of obs.

Restricted sample

Full sample

Model D

Model E

Model D

Model E

Model D

Model E

0.0666 3319

0.0802 3319

0.0759 1902

0.0866 1902

0.164 145

0.246 145

Notes: t statistics in parentheses. Model D excludes agricultural practices variables whereas model E does not. The blank spaces indicate covariates that have been taken out in order to satisfy the balancing property. The provinces are compared with Maputo province. nu. indicates a number. * p < 0.10. ** p < 0.05. *** p < 0.01.

Looking at the coefficients from both estimations, we see that they are roughly the same for the marketed surplus, the value of agricultural production and the total income, indicating that the results are quite robust.

Table 5 Balancing table after matching based on the propensity score matching estimator. Full sample

Head of household characteristics Age (years) Gender (dummy) Schooling (years) Self-employed (dummy) Salary work (dummy) Agriculture primary activity (dummy) Household characteristics Household size (nu.) Pensioner (dummy) Serious disease (dummy) Assets Agricultural land (hectares) Ownership of land (dummy) Good roof (dummy) Good walls (dummy) Ag. practices Family labor (nu.) Animals (nu.) Animal traction (dummy) Fertilizers (dummy) Pesticides (dummy) Manure (dummy) Number of obs.

Restricted sample

Model D

Model E

Model D

Model E

−0.0776 (1.0947) 0.0034 (0.0300) 0.3921 (0.2762) −0.0112 (0.0328) 0.0247 (0.0283) −0.0332

−0.2852 (1.0434) 0.0074 (0.0288) 0.3709 (0.2875) −0.0004 (0.0353) 0.0168 (0.0271) −0.0227

−2.2510 (1.3879) −0.0055 (0.0400) 0.4424 (0.3219) 0.0130 (0.0438) 0.0285 (0.0296) −0.0319

−1.4240 (1.3517) −0.0126 (0.0403) 0.3706 (0.3256) 0.0220 (0.0444) 0.0245 (0.0308) −0.0278

(0.0278)

(0.0280)

(0.0271)

(0.0299)

0.3860* (0.2155) 0.0161 (0.0148) 0.0017 (0.0200)

0.3829* (0.2244) 0.0162 (0.0144) 0.0069 (0.0204)

0.3138 (0.2502) −0.0052 (0.0084) −0.0342 (0.0226)

0.1988 (0.2634) −0.0003 (0.0081) −0.0020 (0.0202)

0.4898** (0.2491) 0.0003 (0.0049) 0.0462 (0.0285) 0.0091 (0.0346)

0.5519* (0.3026) 0.0025 (0.0049) 0.0382 (0.0268) 0.0080 (0.0361)

0.6253 (0.4500) 0.0000 (0.0000) 0.0664** (0.0320) 0.0350 (0.0466)

0.7069 (0.5075) 0.0000 (0.0000) 0.0487* (0.0295) 0.0400 (0.0475)

0.3323** (0.1298) 1.1720** (0.5295) 0.0290 (0.0263) 0.0582*** (0.0207) 0.0205 (0.0208) 0.0571** (0.0226) 3139

0.2601** (0.1310) 0.2294 (0.6768) 0.0256 (0.0259) 0.0434** (0.0188) 0.0172 (0.0210) 0.0408* (0.0221) 3261

0.2020 (0.1455) 0.7247 (0.7228) 0.0366 (0.0340) 0.0651** (0.0300) 0.0122 (0.0326) 0.0629** (0.0320) 1834

0.1348 (0.1462) 0.5002 (0.6708) 0.0315 (0.0350) 0.0310 (0.0316) 0.0031 (0.0332) 0.0437 (0.0311) 1774

7. Possible mechanisms explaining the results From the analytic framework we know that membership in a farmers’ organization can affect both product and input prices, market access and sales volumes, production volume, yield levels and production technology of the members. These factors affect my outcome variables; the marketed surplus, the value of agricultural production and the overall income of the household; directly, but differently. I use these differences, my data and the overview of farmers’ organizations in Mozambique to discuss possible mechanisms explaining my results. (i) Production and production technology Higher production volume is the only possible explanation for why members have a higher value of production than non-members do, since the prices I use mainly are based on median district prices17 and therefore do not dependent upon membership. One possible explanation of the higher production volumes among members is higher yields among members than non-members; however, the data do not permit me to test yields specifically. However, as I control for landholdings, my findings indicate that members use better farming technologies compared with non-members and therefore have higher yields. This is not surprising as training of farmers and improving production methods are important goals for farmers’ organizations in Mozambique as well as for their partners. This result is supported by empirical evidence from Mozambique. Uaiene et al. (2007) show that members in farmers’ organizations have a higher probability of adopting new technology, Nyyssölä et al. (2014) find that the technology adoption occurs, but does not last partly due to late delivery of inputs. (ii) Price and market access effects Since the prices I use are mainly based on median district prices18 and not farm-gate prices, my results cannot be explained by price effects on sales from membership. This indicates that the higher marketed surplus by members compared to non-members is driven by higher total sales volumes by members compared to non-members. There are two possible explanations for higher sales volume; more surplus production to sell or better market access. From the discussion above we know that

Notes: t statistics in parentheses. Model D excludes agricultural practices variables whereas model E does not. *** p < 0.01. * p < 0.10 ** p < 0.05,

17 18

10

. Some are also overall sales values. . Some are also overall sales values.

Food Policy xxx (xxxx) xxxx

M.E. Bachke

Table 6 Summary results. Second stage. Diff-in-diff propensity score matching for becoming a member for both samples. Full sample

Restricted sample

Model D Marketed surplus Value of agricultural production Total income Treated/controls Number of obs. Covariates included in membership estimations Head of household and household characteristics, assets and provinces Agricultural practices

Model E

Model D

Model E

**

0.258* (1.772) 0.181* (1.929) 0.155 (1.374) 217/2922 3139

0.274* (1.867) 0.213* (2.228) 0.175 (1.549) 217/3044 3261

0.401 (2.390) 0.282** (2.441) 0.230* (1.739) 124/1710 1834

0.425** (2.556) 0.323*** (2.738) 0.243* (1.848) 124/1650 1774

X

X X

X

X X

Notes: The numbers in parentheses are t-values. The standard errors are bootstrapped with 500 replications. Models D and E are defined by the covariates included in the propensity score matching estimation. Model D excludes agricultural practices whereas model E includes these variables. The matching estimator used is a kernel estimator. The outcome variables are in natural logarithms. Details about head of household characteristics, household characteristics assets and agricultural practices controls are specified in Table 1. * p < 0.1. ** p < 0.05. *** p < 0.01.

members have a larger production, and therefore, probably more surplus production to sell. Better market access and more commercialization due to membership in farmers’ organizations might also contribute to my results. Membership in farmers’ organization can provide the farmers with more market information. This is supported by the findings of Boughton et al. from 2007 where they find a positive, but not significant correlation between market participation and membership in farmers’ organizations. From theory, we would expect to find a positive price effect of membership in famers’ organizations and not only better market access or more produce to sell, however, my data used does not allow for studying these effects. Thus, this study cannot capture the substantial price effect identified by Pearce and Reinsch (2005) in the project supported by CLUSA in Mozambique. Therefore, the result in this analysis is a lower bound of the impact of membership in farmers’ organizations in Mozambique.

Table 7 Second stage, Summary results for diff-in-diff propensity score matching for leaving a farmers’ organization.

(iii) Access to inputs

Notes: The numbers in parentheses are t-values. The standard errors are bootstrapped with 500 replications. Models D and E are defined by the covariates included in the propensity score matching estimation. Model D excludes agricultural practices whereas model E does not. The matching estimator used is a kernel estimator. The outcome variables are in natural logarithms. General controls, assets and agricultural practices are as defined in the descriptive statistics in Table 1. ** p < 0.05. *** p < 0.01. * p < 0.1.

Full sample

Marketed surplus Value of agricultural production Total income Treated/controls Number of obs. Covariates included in membership estimations Head of household and household characteristics, assets and provinces Agricultural practices

From the descriptive statistics, we know that members use more inputs compared with non-members. Using a dummy variable consisting of the use of irrigation, fertilizer, pesticides, manure and animal traction, and applying the propensity score matching difference-indifference estimator, this finding is confirmed.19 This estimation establishes a causal link between membership and the use of more inputs, indicating that members have better access to and use more inputs than non-members. Thus, it seems my positive results partly come from a higher use of inputs and probably better production techniques in general probably combined with better market access among members than non-members. More use of inputs probably lead to higher production volumes, via higher yields, allowing the members to sell more in the market than non-members. Thus, the organization of farmers in Mozambique seems to reduce market failure in the input market and probably facilitates market access for smallholders. Several studies from Mozambique (Gêmo and Babu, 2019; Nyyssölä et al., 2014) as well as anecdotal evidence from Mozambique also indicate that farmers’ organizations are used as a vehicle to distribute free inputs. My results

Model D

Model E

–0.555* (–1.881) –0.376* (–1.771) –0.469* (–1.792) 98/45 143

–0.416 (–1.283) –0.111 (–0.511) –0.282 (–0.821) 98/45 143

X

X X

indicate that this approach is working. 8. Erratic membership pattern in farmers’ organizations My results show that membership in farmers’ organizations increase the overall income of members, particularly for smallholders whose main cash income is from agriculture. Given the positive impact on all the outcome variables of being a member in a farmers’ organization, it is surprising to see that as many as 64% of the members in 2002 in the full sample left their farmers’ organization before 2005, and as many as 76% in the restricted sample. Two possible explanations are that the

19 . The result is significant in the full sample, whereas it is not significant in the restricted sample; however, the coefficients are similar. These results are available from the author upon request.

11

Food Policy xxx (xxxx) xxxx

M.E. Bachke

farmers leave the organization, as they do not see that there is any benefit from continuing as members or alternatively the farmers’ organization ceases to function in the area where the farmers live. Assuming that the farmers are rational, I expect them to leave a functioning farmers’ organization voluntarily only if the enhanced income stream would continue after terminating their membership or that they feel that the required time they need to invest is too high compared to the benefits. I do not have data to test the time use issue and can therefore not draw strong conclusions on this issue. However, impressions from Mozambique indicate that weekly meetings are the most common time requirement to participate in farmers’ organizations, in addition to participating in trainings and applying the new technologies in the field. To investigate whether the famers continue to enjoy the higher income stream after they leave the farmers’ organization, I use the difference-in-difference propensity score matching estimator to study the impact of leaving a farmers’ organization on the outcome variables. The propensity score matching for these estimations were presented in Table 4, and the impact results are presented in Table7. As we can see from Table 7, all the impact of leaving a farmers’ organization is negative and rather large on all the three outcome variables in both specifications, however, only significant in the first specification. The estimated effect is particularly large for the marketed surplus. Thus, when a household leaves an organization they also lose the positive effect of being a member. This can indicate that the members do not leave the organization voluntarily, but rather that the farmers’ organization disappears or stops functioning. From Table 4, column 5 and 6, we see that the probability to leave a farmers’ organization is smaller if the household has more members, if they have larger landholdings and if there is serious disease in the household. There is no obvious explanation to the latter characteristic, and this might be due to either less mobility among these household or targeting by for example NGOs. If we include agricultural practices in the estimation, we see that if the household use fertilizers or animal traction, they tend to stay while households with more animals leaves the farmers’ organizations. Thus, the continued access to fertilizers seems to be critical for membership in farmers’ organizations in Mozambique. I cannot use my data to test the hypothesis that a farmers’ organizations disappears; however, research from Mozambique indicates that the formation of farmers’ groups or organizations and their success in implementing, for example new technologies, requires the presence of either national or international NGOs (Boughton et al., 2007; Kaarhus and Woodhouse, 2012; Nyyssölä et al., 2014; Gêmo and Bahu, 2019). The lack of sustainability of farmers’ organizations can be explained by their dependency on international NGOs (Bingen et al., 2003; Markelova et al., 2009; Gêmo and Bahu, 2019) and organizational issues regarding member-owned organizations (Bernard and Spielman, 2009; Cook and Ilipoulos, 2000; Poulton et al., 2010, Ochieng et al. 2017). These explanations fit quite well in the Mozambican case, and with my framework and results. First, many of the farmers’ organizations in Mozambique have been formed with support of and most probably as a result of support by NGOs and public extension services, which might have led to technical and financial dependency on these NGOs (Kaarhus and Woodhouse, 2012; Nyyssölä et al., 2014; Gêmo and Bahu, 2019). In fact, in the analysis I assume that the financial cost of membership is zero, which is in line with the view of Boughton et al. (2007) that farmers’ organizations function more as public goods than club goods in Mozambique. Furthermore, the organizations might have been initiated more to serve as recipients of a project rather than to create a farmers’ organization as such. The latter effect is strengthened by the government’s policy of providing free inputs to farmers groups (Nyyssölä et al., 2014; Gêmo and Bahu, 2019). My results indicate that as long as the farmers’ receive fertilizers, they stay members in the farmers’ organizations. Thus, if this is the farmers’ motivation is to reap the benefits, such as

cheap inputs, and not to stay part of an organization in the long-term, the organization would not be sustainable when it no longer receives inputs, probably at a reduced price. If the organizations in fact are vehicles to receive inputs, one would not expect the farmers to put too much effort or time into building sustainable and functioning organizations. This also indicates a strong external dependency, and as noted by Ochieng et al. (2017) the farmers’ group might fade way when the project ends. The fact that illiteracy among members is common, open membership is quite common, and the financial viability of the organization probably is weak due to low membership fees adds to the organizational challenges farmers’ organizations in Mozambique. The performance of the farmers’ organizations can be further challenged by sticky leadership where the first elected leadership might regard the farmers’ organization as their organization (Kaarhus and Woodhouse, 2012). Due to lack of statues, and therefore few elections, the farmers’ organization can in reality be reduced to the board of the organization (Kaarhus and Woodhouse, 2012). These factors might explain why farmers’ organizations stop functioning and the low membership rates in farmers’ organizations, despite the increase in income I find in this study. Similar results are reported by Kelly et al. (2003) from the work done in Mozambique by CLUSA. 9. Conclusion My results show that there is a positive causal impact of membership in a farmers’ organization on the marketed surplus and the value of agricultural production. The results also hold for the total income of the household, though the significance of this result varies. The impact on the value of agricultural production and total income is around 18% and 15%, respectively, which is a significant increase in production and income for poor smallholders in Mozambique. The impact on marketed surplus (25%) is even higher, indicating that the members commercialize more. For the smallholders, whose agricultural income is their main source of cash, the impact is even higher. These results clearly show that farmers’ organizations enhance smallholders’ welfare. The main driver of these results is more use of inputs, which increases production allowing the farmer to market more, and maybe through a better marketing channel with reduced transaction costs. Thus, it seems that farmers’ organizations help reduce market failures at least in the input market and probably also in the output market. As market failures are prominent in Africa in general, and in Mozambique in particular, correcting market failures is important in reducing poverty in rural areas, and farmers’ organizations is an institution that seem to achieve this goal. These positive results show that farmers’ organizations should be a priority both in agricultural policies in developing countries and in the development policies. However, my data indicate that farmers’ organizations face serious challenges as more than 60% of the farmers that were members in 2002 left their organization before 2005. One possible explanation might involve the way the farmers’ organizations are initiated and organized. The organizations are not a result of a bottom-up approach from the farmers, but rather a result of donor and governmental policies and NGO support. Because of this, the organizations depend on continued support, rendering them less sustainable. These challenges have to be solved for the farmers’ organizations to become sustainable in the long run. Given the positive income effects of these organizations, an important policy recommendation is to increase funding, activities and research on farmers’ organizations, their institutions and determinants to solve these sustainability problems. The main contribution of this paper is that it causally shows that membership in a farmers’ organization enhance smallholders’ welfare. This result is not organizational specific, activity specific nor location specific within a country and therefore shows in an general way that farmers’ organizations contributes positively to enhance welfare. One important insight and policy recommendation from this study is to 12

Food Policy xxx (xxxx) xxxx

M.E. Bachke

strengthen work on farmers’ organization in agricultural policy since these organizations mitigate market failure, enhances incomes and thereby contributes to reducing poverty.

Glover, D.J., 1987. Increasing the benefits to smallholders from contract farming: Problems for farmers’ organizations and policy makers. World Dev. 15 (4), 441–448. Heckman, J.J., Ichimura, H., Todd, P., 1997a. Matching as an econometric evaluation estimator: evidence from evaluating a job training programme. Rev. Econ. Stud. 64, 605–654. Heckman, J.J., Ichimura, H., Todd, P., 1998. Matching as an econometric evaluation estimator. Rev. Econ. Stud. 65, 261–294. Heckman, J.J., Navarro-Lozano, S., 2004. Using matching, instrumental variables and control functions to estimate economic choice models. Rev. Econ. Stat. 86, 30–57. Heckman, J.J., Smith, J., Clements, N., 1997b. Making the most out of programme evaluations and social experiment: Accounting for heterogeneity in program impacts. Rev. Econ. Stud. 64, 487–535. Heltberg, R., Tarp, F., 2002. Agricultural supply response and poverty in Mozambique. Food Policy 27 (2), 103–123. Jayne, T.S., Mather, D., Mghenyi, E., 2010. Principal challenges confronting smallholder agriculture in sub-Saharan Africa. World Dev. 38 (10), 1384–1398. Kaarhus, R., Woodhouse, P., 2012. Development of national producer organizations and specialized business units in Mozambique: A study for the Norwegian Society for Development to prepare a new phase of programme collaboration. Noragric Report No. 63. Ås: Department of International Environment and Development Studies (Noragric), Norwegian University of Life Sciences. Kelly, V., Adesina, A.A., Gordon, A., 2003. Expanding access to agricultural inputs in Africa: a review of recent market development experience. Food Policy 28, 379–404. Kirsten, J., Sartorius, K., 2002. Linking agribusiness and small-scare farmers in developing countries: Is there a new role for contract farming? Dev. S. Afr. 19 (4), 503–529. Lele, U., 1981. Cooperatives and the poor: A comparative perspective. World Dev. 9, 55–72. Lutz, C., Tadesse, G., 2017. African farmers’ market organizations and global value chains: competitiveness versus inclusiveness. Rev. Social Economy 75 (3), 318–338. Ma, W., Abdulai, A., 2016. Does cooperative membership improve household welfare? Evidence from apple farmers in China. Food Policy 58, 94–102. Markelova, H., Meinzzen-Dick, R., Hellin, J., Dohrn, S., 2009. Collective action for smallholder market access. Food Policy 34, 1–7. Ministry of Agriculture, 2002. Trabalho de inquérito agrícola 2002 [National Agricultural Survey 2002]. Ministry of Agriculture, Republic of Mozambique, Maputo. Ministry of Agriculture, 2002. Trabalho de inquérito agrícola 2002. Manual do inquiridor 2002 [National Agricultural Survey 2002. Interview guide 2002]. Maputo: Ministry of Agriculture, Republic of Mozambique. Ministry of Agriculture, 2005. Trabalho de inquérito agrícola 2005 [National Agricultural Survey 2005]. Ministry of Agriculture, Republic of Mozambique, Maputo. Morgan, S.L., Winship, C., 2007. Counterfactuals and Causal Inference. Methods and Principles for Social Research. Cambridge University Press, New York. Moustier, P., Tam, P.T.G., Anh, D.T., Binh, V.T., Loc, N.T.T., 2010. The role of farmer organizations in supplying supermarkets with quality food in Vietnam. Food Policy 35, 69–78. Nilsson, J., 2001. Organizational principles for co-operative firms. Scand. J. Manag. 17, 329–356. Nyyssölä, M., Pirttilä, J., Sandström, S., 2014. Technology adoption and food security in subsistence agriculture – evidence from a group-based aid project in Mozambique. Finnish Econ. Pap. 27 (1), 33. Ochieng, D.O., Veettil, P.C., Qaim, M., 2017. Farmers’ preferences for supermarket contracts in Kenya. Food Policy 68, 100–111. Pearce, D., Reinsch, M., 2005. Small farmers in Mozambique access credit and markets by forming associations with assistance from the Cooperative League of the USA (CLUSA) (English). CGAP agricultural microfinance case study series ; no. 5. Washington, DC: World Bank. Penrose Buckley, C., 2007. Producer organisations: A guide to developing collective rural enterprises. Oxfam GB, Oxford. Poulton, C., Dorward, A., Kydd, J., 2010. The future of small farms: New directions for services, institutions, and intermediation. World Dev. 30 (10), 1413–1428. Ravallion, M., 2007. Evaluating anti-poverty programs. In: In: Schultz, T., Strauss, J. (Eds.), Handbook of Development Economics, vol. 4. North-Holland, Amsterdam, pp. 3787–3846. Reardon, T., Weatherspoon, D.D., 2003. The rise of supermarkets in Africa: Implications for agrifood systems and the rural poor. Dev. Policy Rev. 21 (3), 333–355. Republic of Mozambique, 2001. Plano de acção para redução de pobreza absoluta – PARPA I (2001-2005) [Action plan for the reduction of absolute poverty 2001-2005]. Maputo: Republic of Mozambique. Republic of Mozambique, 2006. Plano de acção para redução de pobreza absoluta – PARPA II (2006-2009) [Action plan for the reduction of absolute poverty 20062009]. Maputo: Republic of Mozambique. Rondot, P., Collion, M.H., 2001. Agricultural Producer Organizations: Their Contribution to Rural Capacity Building and Poverty Reduction-Report of a Workshop, Washington, D.C., June 28-30, 1999. RDV, World Bank, Washington. Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70 (1), 41–55. Shiferaw, B., Hellin, J., Muricho, G., 2011. Improving market access and agricultural productivity growth in Africa: what role for producer organizations and collective action institutions? Food Security 3 (4), 475–489. Sinyolo, S., Mudhara, M., 2018. Collective action and rural poverty reduction: Empirical evidence from KwaZulu-Natal, South Africa. Agrekon 57 (1), 78–90. Sivramkrishna, S., Jyotishi, A., 2008. Monopsonistic exploitation in contract farming: Articulating a strategy for grower cooperation. J. Int. Dev. 20, 280–296. Smith, J.A., Todd, P.E., 2005. Does matching overcome LaLonde’s critique of nonexperimental estimators?. J. Econometrics 125, 305–353.

Acknowledgements I would like to thank the editor and two referees for useful comments. In addition, I am grateful for comments and encouragement from a number of people including Frode Alfnes, Arild Angelsen, Cristopher B. Barrett, Therese Dokken, Caroline Wang Gierløff, Jo Thori Lind, Carl-Erik Schulz and Sofie Skjeflo. Furthermore, I am grateful to the Mozambican Ministry of Agriculture, which has granted me access to the data. Any mistakes that reminds are my own. References Arndt, C., Hussain, M.A., Jones, E.S., Nhate, V., Tarp, F., Thurlow, J., 2012. Explaining the evolution of poverty: the case of Mozambique. Am. J. Agric. Econ. 94 (4), 854–872. Arndt, C., James, R.C., Simler, K.R., 2006. Has economic growth in Mozambique been pro-poor? J. Afr. Economies 15 (4), 571–602. Bacon, C., 2005. Confronting the coffee crisis: Can fair trade, organic, and specialty coffees reduce small-scale farmer vulnerability in Northern Nicaragua? World Dev. 33 (3), 497–511. Barrett, C.B., 2008. Smallholder market participation: Concepts and evidence from eastern and southern Africa. Food Policy 33, 299–317. Barrett, C.B., Bachke, M.E., Bellemare, M.C., Michelson, H.C., Narayana, S., Walker, T.F., 2012. Smallholder participation in contract farming: Comparative evidence from five countries. World Dev. 40 (4), 715–730. Becchetti, L., Costantino, M., 2008. The effects of fair trade on affiliated producers: An impact analysis on Kenyan farmers. World Dev. 36 (5), 823–843. Becker, S.O., Ichino, A., 2002. Estimation of average treatment effects based on propensity scores. The Stata J. 2 (4), 1–19. Bernard, T., Spielman, D.J., 2009. Reaching the rural poor through rural producer organizations? A study of agricultural marketing cooperatives in Ethiopia. Food Policy 34, 60–69. Best, R., Westby, A., Ospina, B., 2006. Linking small-scale cassava and sweetpotato farmers to growth markets: experiences, lessons and challenges. Acta Horticulturae 703, 39–46. Bingen, J., Serrano, A., Howard, J., 2003. Linking farmers to markets: different approaches to human capital development. Food Policy 28, 405–419. Blundell, R., Costa Dias, M., 2000. Evaluation methods for non-experimental data. Fiscal Studies 21 (4), 427–468. Boughton, D., Mather, D., Barrett, C.B., Benfica, R., Abdula, D., Tschirley, D., Cunguara, B., 2007. Market participation by rural households in a low-income country: An assetbased approach applied to Mozambique. Faith Econ. 50, 64–101. Carletto, C., Kilic, T., Kirk, A., 2011. Nontraditional crops, traditional constraints: The long-term welfare impacts of export crop adoption among Guatemalan smallholders. Agric. Econ. 42, 61–75. Caviglia, J.L., Kahn, J.R., 2001. Diffusion of sustainable agriculture in the Brazilian tropical rain forest: A discrete choice analysis. Econ. Dev. Cult. Change 49 (2), 311–333. Cook, M.L., Ilipoulos, C., 2000. Ill-defined property rights in collective action: The case of US agricultural cooperatives. In: Ménard, C. (Ed.), Institutions, Contracts and Organizations: Perspectives from New Institutional Economics. Edward Elgar, Cheltenham, pp. 335–348. Chagwiza, C., Muradian, R., Ruben, R., 2016. Cooperative membership and dairy performance among smallholders in Ethiopia. Food Policy, 59, 165–173.Cunguara, B., & Darnhofer, I. (2011). Assessing the impact of improved agricultural technologies on household income in rural Mozambique. Food Policy 36, 378–390. Cunguara, B., Darnhofer, I., 2011. Assessing the impact of improved agricultural technologies on household income in rural Mozambique. Food Policy 36, 378–390. Department of Agriculture, Forestry and Fisheries, 2012. A framework for the development of smallholder farmers through cooperatives development. Pretoria: Department of Agriculture, Forestry and Fisheries. South Africa. Devaux, A., Horton, D., Velasco, C., Thiele, G., Lopez, G., Bernet, T., Reinoso, I., Ordinola, M., 2009. Collective action for market chain innovation in the Andes. Food Policy 34, 31–38. Dorsey, J., Muchanga, S.P., 1999. Best practices in farmers’ association development in Mozambique. ARD-RAISE Consortium, Arlington. Eriksen, S., Lutz, C., Tadesse, G., 2018. Social desirability, opportunism and actual support for farmers’ market organisations in Ethiopia. J. Dev. Stud. 54 (2), 343–358. Fischer, E., Qaim, M., 2012. Linking smallholders to markets: determinants and impacts of farmer collective action in Kenya. World Dev. 40 (6), 1255–1268. Francisco, da Silva, A.A., Matter, K., 2007. Poverty observatory in Mozambique. Final report. Zürich: Gerster Consulting. Gêmo, H.R., Babu, S.C., 2019. Empowering smallholder farmers organizations through non-public extension service providers: A case study and lessons from Mozambique. IFPRI discussion papers 1807, International Food Policy Research Institute (IFPRI). Gilligan, D.O., Hoddinott, J., 2007. Is there persistence in the impact of emergency food aid? Evidence on consumption, food security, and assets in rural Ethiopia. Am. J. Agric. Econ. 89 (2), 225–242.

13

Food Policy xxx (xxxx) xxxx

M.E. Bachke Sykuta, M.E., Cook, M.L., 2001. A new institutional economics approach to contracts and cooperatives. Am. J. Agric. Econ. 83 (5), 1273–1279. Tolno, E., Kobayashi, H., Ichizen, M., Esham, M., Balde, B.S., 2015. Economic analysis of the role of farmer organizations in enhancing smallholder potato farmers’ income in middle Guinea. J. Agric. Sci. 7 (3), 123. Uaiene, R.N., Arndt, C., Masters, W.A., 2009. Determinants of agricultural technology adoption in Mozambique. Discussion Paper No. 67E. Maputo: Ministry of Planning and Development, Republic of Mozambique. Verhofstadt, E., Maertens, M., 2014. Can agricultural cooperatives reduce Poverty?

Heterogeneous impact of cooperative membership on farmers' welfare in Rwanda. Appl. Econ. Perspect. Policy 37 (1), 86–106. Vorlaufer, M., Wollni, M., Mithofer, D., (2012, August). Determinants of collective marketing performance: Evidence from Kenya’s coffee cooperatives. In: Selected Paper prepared for presentation at the IAAE Triennial Conference, Foz do Iguacu, Brazil, pp. 18–24. World Bank, 2007. World development report 2008. Agriculture for development. Washington DC: The World Bank.

14