Journal of Development Economics 125 (2017) 1–20 Contents lists available at ScienceDirect Journal of Development Economics journal homepage: www.elsevier.com/locate/jdeveco Seeing is believing? Evidence from an extension network experiment☆ a,⁎ b c crossmark Florence Kondylis , Valerie Mueller , Jessica Zhu a Development Research Group, The World Bank, United States b Development Strategy and Governance Division, International Food Policy Research Institute, United States c Agricultural and Applied Economics, University of Wisconsin, Madison, United States A R T I C L E I N F O A BS T RAC T JEL Classification: Extension is designed to enable lab-to-farm technology diffusion. Decentralized models assume that informa- O1 tion flows from researchers to extension workers, and from extension agents to contact farmers (CFs). CFs O3 should then train other farmers in their communities. Such a modality may fail to address informational Q1 inefficiencies and accountability issues. We run a field experiment to measure the impact of augmenting the CF D8 model with a direct CF training on the diffusion of a new technology. All villages have CFs and access the same Keywords: extension network. In treatment villages, CFs additionally receive a three-day, central training on the new Information failure technology. We track information transmission through two nodes of the extension network: from extension Technology diffusion agents to CFs, and from CFs to other farmers. Directly training CFs leads to a large, statistically significant Agriculture increase in adoption among CFs. However, higher levels of CF adoption have limited impact on the behavior of Africa other farmers. 1. Introduction (EAs) and other farmers, are ubiquitously used as messengers of information in developing countries. Efficacy of the CF modality rests Agricultural innovation is necessary to accelerate growth and on two key assumptions. First, EAs will effectively train CFs to adopt achieve food security in Africa (Hazell, 2013). Despite availability of and demonstrate new technologies to peers. Frequent EA visits are yield-enhancing technologies, adoption rates in Sub-Saharan agricul- supposed to elicit a process of experiential learning among CFs. ture remain low (Gollin et al., 2005). A growing literature identifies Second, other farmers' exposure to CFs will encourage wider adoption information failures as an impediment to the technological diffusion in the community, through a peer learning process. Despite some process in agriculture (Bandiera and Rasul, 2006; Conley and Udry, evidence of implementation and accountability constraints, and per- 2010; Munshi, 2004). Less documented are the modalities through haps for lack of a viable policy alternative, the CF model persists across which information can best diffuse and boost adoption of productive Africa (Gautam, 2000). Formally documenting returns to additional farming practices. low-cost, scalable interventions to help leverage these large invest- Agricultural extension services are designed to facilitate the diffu- ments in agricultural extension services could significantly affect the sion of innovations from lab to farm. In developing countries, they path of technology diffusion. account for large shares of government expenditures on agriculture We exploit a large-scale, government-run randomized controlled (Akroyd and Smith, 2007). These substantive investments are seldom trial (RCT) to measure the impact of augmenting the CF model with a supported by causal evidence regarding their effectiveness as a whole, direct training on the diffusion of a new technology in central or of a particular modality (Anderson and Feder, 2004). Contact Mozambique. Our treatment consists of adding a direct CF training farmers (CFs), who serve as points of contact between extension agents to an existing CF model, holding everything else constant. In practice, ☆ This research was funded by the International Initiative for Impact Evaluation, Inc. (3ie) through the Global Development Network (GDN); the Mozambique office of the United States Agency for International Development; the Trust Fund for Environmentally and Socially Sustainable Development; the Belgian Poverty Reduction Partnership and the Gender Action Plan; and the CGIAR Research Program on Policies, Institutions, and Markets (PIM) led by the International Food Policy Research Institute (IFPRI) and financed by the CGIAR Fund donors. The authors benefited from comments provided by Jenny Aker, Luc Behagle, Madhur Gautam, Markus Goldstein, Maria Jones, Rashid Lajaaj, Mark Lundell, Mushfiq Mobarak, Tewodaj Mogues, Glenn Sheriff, David Spielman; and seminar participants at the CSAE (Oxford), the Mid-Western Economic Development Conference, the NEUDC at Boston University, The Ohio State University, the Paris School of Economics, the University of Georgia, the World Bank, and IFPRI. The views expressed in this article do not reflect those of the World Bank, 3ie, or their members. The authors would like to thank Pedro Arlindo, Jose Caravela, Destino Chiar, Isabel Cossa, Beatriz Massuanganhe, and Patrick Verissimo for their collaboration and support throughout the project. John Bunge, Ricardo da Costa, and Cheney Wells provided excellent field coordination; Siobhan Murray and João Rodrigues impressive research assistance. Usual disclaimers apply. ⁎ Corresponding author. E-mail addresses: fkondylis@worldbank.org (F. Kondylis), v.mueller@cgiar.org (V. Mueller). http://dx.doi.org/10.1016/j.jdeveco.2016.10.004 Received 23 February 2015; Received in revised form 24 October 2016; Accepted 27 October 2016 Available online 03 November 2016 0304-3878/ © 2016 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by/4.0/). F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 CFs in treatment villages receive two three-day training (one in 2010, when CFs have access to a direct training in addition to the status quo one in 2012) on a yield-enhancing technology at district headquarters, extension modality. Private returns in the form of labor savings and by the same experts and using the same curriculum as provided to EAs. yield benefits in dry years accompany increases in adoption. However, All EAs were trained on sustainable land management (SLM) and were CFs' knowledge scores on the SLM curriculum are unaffected by the expected to train their local CFs to demonstrate the technology to other direct training. farmers in 200 villages. All CFs were provided demonstration kits to Increasing demonstration of SLM practices could reinforce the encourage adoption and diffusion of information to other farmers. We perceived benefits of SLM among peers by increasing knowledge and augmented the CF model by centrally training CFs on SLM in 150 reducing the uncertainty of SLM benefits (Foster and Rosenzweig, randomly selected (treatment) communities. The training format was 1995). Yet, boosting CF demonstration and adoption through a direct part lectures, part hands-on, with similar content and breadth as the training does not affect other farmers' practices within the community EA training. The central training is the only difference between in our context. Patterns of CF-farmer interactions suggest that the treatment and control, and all 200 villages adhere to the status quo direct training did not additionally stimulate CFs to fulfill their role as CF model. We use two rounds of follow-up survey data on 200 CFs and village messengers. Interestingly, variations in treatment effects by CF a random sample of over 5000 other farmers to examine the impact of characteristics and similarity in cropping patterns indicate that rele- adding a central training on knowledge and adoption of the technology, vance of expected cost–benefit margins affect the diffusion process. For as well as agricultural production. example, pit planting adoption rates increase when a CF's crop The CF model enables a process of experiential learning among CFs portfolio matches the farmers'. Hence, our results corroborate the idea through the use of demonstration activities and regular on-site feed- that the “proximity” of the source of information may be the primary back from EAs. This practice is similar to on-the-job, learning-by-doing constraint on changes in farmer behavior (Munshi, 2004; Feder and processes in other labor markets. Neoclassical growth theories suggest Savastano, 2006; Bandiera and Rasul, 2006; Conley and Udry, 2010) learning-by-doing may be of equal importance to formal training in and, therefore, that the process of CF selection may affect the pace of explaining human capital formation as a production input (Lucas, diffusion (Beaman et al., 2014; BenYishay and Mobarak, 2014). 1988). While learning-by-doing theories are supported in the context of Overall, our findings suggest augmenting decentralized extension firm or plant-level studies (Levitt et al., 2012; Thompson, 2010), programs with a direct training modality can improve their effective- empirical evidence of the significance of learning through extension ness in getting CFs to demonstrate new technologies. Our cost–benefit programs on agricultural growth is mixed (Bindlish and Evenson, analysis shows net private returns of up to USD 76 per CF. Yet, a direct 1997; Purcell and Anderson, 1997; Gautam, 2000; Anderson and CF training leads to modest diffusion to others in the community. Feder, 2007; Benin et al., 2007; Davis et al., 2012; Waddington Taken together, these results imply that adding a direct training et al., 2014). We contribute to this research agenda by formally modality on its own may not be enough to reform the speed of documenting the impact of augmenting an existing, decentralized technology diffusion. Further study is needed to build up the evidence extension model with a relatively low-cost centralized training mod- base, using larger samples, improved measurement techniques, and ality. testing complementary policy actions to make extension services work Adding a direct training may affect technology adoption among CFs for farmers. through three broad categories of mechanisms: increased quantity of In what follows, we detail the Mozambique extension policy and information, enhanced learning experience, and channels other than network at baseline (Section 2). We then describe the evaluation design knowledge. First, the curriculum in a direct training may increase the and empirical strategies used to identify the impact of adding a direct quantity of information transferred (e.g., number of techniques CF training on technology diffusion (Section 3). Section 4 presents taught). Central trainings are offered in an enclosed setting under the estimates of impact on CF knowledge, adoption, and productivity, supervision of project staff, and extension agents present the material other farmers' adoption and knowledge, as well as measures of cost- from the course manual. This plausibly increases the chance that the effectiveness. Section 5 discusses implications of this study for policy intended curriculum is covered. and future research. Second, the centralized format of the training may enhance the learning experience. For instance, the formal setting may add cred- 2. Agricultural extension constraints in Mozambique ibility to the information. While the use of course materials hardly affects learning indicators in other settings (Tan et al., 1999; Glewwe 2.1. National extension coverage et al., 2004, 2009), use of computing technology as a complementary input to a standard curriculum has been shown to have a positive effect Mozambique's agricultural extension network was created in 1987 on learning (Banerjee et al., 2007; Linden, 2008). For these reasons, and began to operate in 1992 after the peace agreement. During the the information set shared during a training held at district head- past two decades, the Ministry of Agriculture (MINAG) has promoted quarters may be (perceived to be) of higher quality than what is given and expanded extension networks (Eicher, 2002; Gemo et al., 2005). during field visits from the EAs. Peer learning will also likely be more EAs are employed by the District Services for Economic Activities pronounced during a centralized training, as CFs with similar char- (Serviços Distritais de Actividades Económicas) and operate at the acteristics get to share information and jointly interact with the subdistrict level to disseminate information and new techniques. The material. system assumes that information flows linearly: agricultural innova- Third, a direct training could increase CFs' adoption through other tions are created by researchers, then distributed by extension workers, channels than knowledge. For instance, the training may improve EA- and finally adopted by producers (Pamuk et al., 2014). Countrywide, to-CF accountability. Directly trained CFs may demand more informa- coverage is as low as 1.3 EAs per 10,000 rural people (Coughlin, 2006). tion from EAs (Björkman and Svensson, 2009; Banerjee et al., 2010). Given this shortage, EAs are inclined to visit the same set of villages Being formally trained could also build empowerment, reinforcing the every year based on their achievements and potentials (Coughlin, identity of the CF as community messenger and their propensity to lead 2006). Only 15 percent of farmers report receiving extension services village-level demonstration activities. Similarly, attending a training in (Cunguara and Moder, 2011). the district town for a few days may make CFs feel “special” relative to At the time our study was designed, the present National Plan for control CFs. Alternatively, a centralized training could create a Agricultural Extension and Extension Master Plan aimed to develop momentum among peers to adopt the new practices, akin to herd the decentralization of services at the district level; increase participa- behavior (Banerjee, 1992; Karlan et al., 2014). tion of targeted groups (women and marginal farmers); and enhance We find a statistically significant increase in CF adoption of SLM partnerships with other actors, such as the private sector and non- 2 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 governmental organizations (Gallina and Chidiamassamba, 2010). household-level follow-ups, respectively, 15 and 27 months after the Given the importance the government places on decentralized exten- first SLM demonstration season. By baseline, we refer to September sion services and the lack of rigorous evidence to date, formally 2010 and earlier. The initial demonstration season in our study was documenting the impact of this policy action seems warranted 2011. Our surveys captured the 2012 and 2013 adoption seasons. This (Gautam, 2000).1 In what follows, we describe the details of the status section details the experiment and data sources. quo extension model operating at baseline in our study area. 3.1. Sustainable land management 2.2. Study area Sustainable land management (SLM, or conservation farming) is a We worked in five districts of central Mozambique: Mutarara (Tete yield-enhancing farming technology that consists of a bundle of Province), Maríngue and Chemba (Sofala Province), and Mopeia and techniques adapted to local crops and agro-ecological conditions Morrumbala (Zambézia Province; Fig. 1). This area receives financing (Haggblade and Tembo, 2003; Thierfelder et al., 2015).5 In the from a large World Bank–Government of Mozambique investment to Zambezi valley, the recommended SLM technology package encom- support the development of the extension network (Smallholder passes seven SLM techniques: Mulching, Crop Rotation, Strip Tillage, project). The project provides three levels of agricultural technical Pit Planting, Contour Farming, Row Planting, and Improved assistance: each district has a facilitator, an environmental specialist, Fallowing.6 Mulching covers the soil with organic residues to maintain and eight EAs. A district is subdivided into four administrative posts soil humidity, suppress weeds, reduce erosion, and enrich the quality of (posto administrativo) that include about 8–10 communities (aldeia). the soil cover. Crop rotation rotates crops on a given plot to improve EAs periodically receive training from the district specialists.2 Each soil fertility and reduce the proliferation of plagues. Strip-tillage community has a designated contact farmer (CF) who receives direct prevents opening the soil, such as through plowing, harrowing, or assistance from the two EAs placed in his administrative post.3,4 CFs digging on land surrounding the seed row. Pit planting consists of receive visits from EAs monthly. They were instated to respond to other constructing permanent holes 15 cm deep around the base of a plant, farmers' demands for technical assistance and provide advice through such as maize, to aid water and nutrient accumulation. Contour demonstration activities. farming is the use of crop rows along contour lines fortified by stones A CF model of extension may not foster learning and adoption (or vegetation) to reduce water loss and erosion on sloped land. Row among CFs. EAs are typically challenged to reach the communities they planting improves productivity by improving access to sunlight and serve. Designating CFs may therefore not adequately address the facilitating weeding and other cultivation practices (for instance, supply-side constraints of extension services. Another concern is that mulching and intercropping) by providing space between rows. information may get “diluted” from the central level to CFs. For Improved fallowing reduces temporary productivity losses from fallow- instance, EAs may not cover all techniques, sufficiently train the CFs, ing through targeted planting of species that recharge the soil in a nor adhere to the expected format. Since CFs do not know what shorter time frame. curriculum their EA should follow, accountability may be low. Finally, There are important complementarities across these techniques, periodic visits from the EA may not be sufficient in getting CFs which are expected to generate savings in labor time during the main motivated to demonstrate to others in their community. season. For instance, combining strip-tillage with pit planting will The underlying assumption of the CF model is that, through peer ensure that pits do not need to be excavated every year. This should learning, a change in CF demonstration effort should affect the process save labor at the seeding stage over the traditional methods of tillage of diffusion to other farmers in the community. By exogenously and planting in ridges. Similarly, strip-tillage and contour farming affecting CFs' adoption of a new technology, our experiment directly combined will save time, since the terraces will not have to be tests whether the CF model is suited to promoting technology adoption prepared every year for seeding. Combining mulching and pit planting on a large scale. Allowing the ITT estimates to vary by CF character- can also help maximize the nutrient retention of the soil around the istics provides qualitative evidence of existing barriers to knowledge maize crop and minimize the need for weeding. transfer. We asked CFs to recall their familiarity with these techniques at baseline (Table A1). SLM exposure varied widely across techniques and farmer types. Twenty-one percent of CFs had heard of improved 3. Experiment and data fallowing relative to 10 percent of other farmers. In contrast, 76 percent of CFs knew of mulching, compared to 34 percent of other We run a large field experiment to test for effective knowledge farmers. This suggests that some, if not all, SLM technologies taught in diffusion under the CF model of extension, and isolate the additional the CF training and disseminated by EAs pose as reasonable instru- impact of directly training CFs. A new technology, SLM, was dissemi- ments to track knowledge diffusion in the Zambezi valley.7 nated through the extension network for the first time in 2010. Our study started in October 2010 and ended after the main 2013 cropping 3.2. Training season, thus spanning three main agricultural seasons (Fig. 2). We collected three rounds of data: a rapid CF baseline and two CF and We now describe the trainings delivered to EAs and CFs in the context of our study. First, all EAs serving administrative posts within 1 Recent work has employed a quasi-experimental design to evaluate the impact of our study area received two three-day training courses on SLM extension and found a positive impact of extension on farm income in Mozambique (Cunguara and Moder, 2011) 2 5 In October 2010 and November 2012, these trainings were dedicated to SLM. A direct implication is that, while positive yield effects of SLM are relatively well 3 The ratio of EAs per administrative post in our study area is on par with the 2013 documented for Southern Africa (Haggblade and Tembo, 2003; Thierfelder et al., 2015), national average of 1.89 (Gemo and Chilonda, 2013). This ratio is calculated using the there is little evidence on the returns of individual SLM techniques. 2010 figures from the Direçåo Nacional de Extenså Agraria (DNEA), available at the 6 Intercropping was included in the curriculum, but is excluded from the analysis as it following URL: http://www.worldwide-extension.org/africa/mozambique/s- was already widely adopted at the time of the intervention by CFs (98 percent) and other mozambique. farmers (76 and 81 percent of women and men, respectively). Including the technique 4 EAs can choose which CFs to work with, and do not necessarily split responsibilities. bears little consequence on our point estimates (not reported). 7 Hence, a given CF may interact with both EAs in his administrative post. CFs are typically The project had started to disseminate mulching, strip tillage, row planting, and crop chosen by the community. In 2010, CFs had been in their position for three years on rotation as early as 2008. However, the formal practice was sparse at the time of the average, with a standard deviation of 3. This indicates the majority of CFs were already intervention and most EAs and CFs had not received a formal training on SLM commissioned by the project prior to our intervention. techniques, or been instructed to transfer their knowledge to their peers. 3 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Fig. 1. Study area and spatial distribution of sampled households. Fig. 2. Timeline of training and survey. 4 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 techniques in October 2010 and November 2012 (prior to the main This design allows us to isolate the additional effect of a direct planting season). Technical staff from the Ministry of Agriculture training, implicitly testing for effective knowledge diffusion under the (MINAG) developed the educational agenda on SLM practices, and CF model. For this purpose, we held constant all other extension the training was delivered by MINAG's district technical staff with interventions across treatment and control communities. Specifically, support from one staff from the central project team. Half of the in line with the status quo modality, all CFs in the study area receive training sessions were devoted to in-class lectures, and the other half assistance from their EAs and a tool kit to set up and maintain a consisted of hands-on demonstrations. The syllabus included a thor- demonstration plot within the community. These demonstration plots ough review of the advantages of each SLM technique over less- are used by (1) EAs to teach and assist CFs in implementing at least one environmentally desirable ones.8 The EA training also highlighted of the agricultural practices of the CF's choice, and (2) CFs to good practices in fostering interactions between EAs and CFs. demonstrate the new techniques to other farmers in their community. The centralized CF trainings were held a few weeks after those of In practice, the CF-level random assignment was implemented as the EAs. The content of the CF training was similar to that received by follows. Each EA team at the administrative post level was in charge of EAs, and was delivered by the same district-level and central MINAG inviting treatment CFs to the central SLM trainings. During the EA staff.9,10 The cost of a direct training per CF per year was 74 USD. Over trainings on SLM, district staff explained the physical impossibility of three agricultural seasons, this represents a modest 12.8 percent training all CFs at once and that a lottery had been used to select the increase in the total salary and training cost of the extension network.11 participating CFs. EAs were then given the list of randomly chosen After these trainings were completed, all EAs worked with their CFs treatment CFs. An attendance sheet was taken at CF training by the to disseminate the SLM techniques most pertinent to their local area on district staff. In October 2010, only four treatment CFs did not attend their (own or communal) demonstration plots, regardless of their CFs' the training (all in the Mopeia district), and there was no contamina- treatment status. All CFs received a new toolkit (a bicycle, tools to plow tion to control CFs.14 Since district staff may have an incentive to the land, and smaller articles).12 A second toolkit with similar items misreport attendance, we performed independent audits. First, we (including a bicycle) was provided to all CFs again in July 2012. The verified that the attendance list reflected the (randomly assigned) only difference between treatment and control CFs is that treated CFs eligibility, and found no contamination of the control group. Second, received an additional direct training on SLM. we showed up unannounced at the trainings in all five districts and verified attendance verbally. Finally, attendance lists were back- 3.3. Experimental design checked: a random set of listed participants were visited in November and December of 2010 and asked whether they attended At baseline, CFs and EAs in our five study districts operated under the SLM training. Our results from these audits indicate that atten- the CF model of extension in all communities. From these districts, we dance was genuine. randomly selected 200 communities (with 200 CFs) in 16 adminis- Similar checks were performed on the 2012 training. While the trative posts, to which 30 EAs were assigned. All EAs received SLM attendance list was equally validated, participation was not universal training. We randomly assigned CFs in 150 treatment communities to and contamination was quite substantial. Of the treated communities, the augmented version of the CF model (treatment), stratifying the 63 percent had at least one CF attend the training, and 16 percent of assignment at the district level. Control (50) and treatment (150) CFs control communities had a CF attend.15 These figures signal statisti- received SLM training during visits from their EAs—the status quo CF cally significant exposure of control CFs to the treatment in 2012. extension modality. Treatment CFs additionally received the direct CF While this may hamper our ability to statistically differentiate the two training described above.13 training models on CF behavior in the 2013 (second follow-up) survey round, our results on other farmers at endline are arguably robust to this contamination. Increased demonstration by control CFs in the 8 The main charts from the class can be found here: intrerrefhttp://siteresources.- 2013 growing season is unlikely to have affected farmers' adoption in worldbank.org/INTDEVIMPEVAINI/Resources/ that same season. Flipchart_deAC_anonymized.pdfhttp://siteresources.worldbank.org/ INTDEVIMPEVAINI/Resources/Flipchart_deAC_anonymized.pdf. The general curricu- There are two important limitations to our identification strategy: lum used by the MINAGRI staff is provided on this site: http://siteresources.worldbank. one concerns the intensive margin of EA support to treatment CFs org/INTDEVIMPEVAINI/Resources/Manual_AC_FINAL.pdf. The hands-on component relative to control CFs, while the other relates to the extensive margin of the training was not recorded but followed closely the techniques discussed in class. of EA attention. First, direct training and EA support are likely 9 In some districts, district staff relied on their EAs to help during the hands-on sessions. This could contaminate our results by lowering the amount of on-farm attention complementary inputs. Therefore our estimates capture the overall treatment CFs subsequently received from their EAs. This may lead us to underestimate effect of augmenting the CF model with a direct training. We cannot information flow in the central training arm, and overestimate it in pure CF model. disentangle the impact of learning during the central training from that Reassuringly, as mentioned above, we do not find that EAs devoted more time to visiting of improved learning during regular EA-to-CF tutorials as a result of CFs in treatment communities, relative to control. 10 the direct training. Given the low literacy of farmers, a film covering all techniques substituted the initial lecture format in the second training of the CFs in 2012. Second, our design implies that each EA will work with both 11 The monthly EA salary costs were at USD 210 per EA from data provided by the treatment and control CFs in his administrative post.16 A threat to DNPDR and the Smallholders project team, and EA training costs ran at USD 370 per training. Each EA supervised on average 5 treated CFs over the course of 36 months. There are obviously other, non-wage costs to running an extension network. To the (footnote continued) extent that we do not account for these additional costs, we overestimate the relative cost have any statistically significant effect on our outcomes of interest (not reported). of adding a direct CF training. Nonetheless, we control for this third treatment arm in the regression analysis. 12 14 The toolkit distribution was planned, independently of our intervention, by the These CFs were trained by the EA on an individual basis, and the follow-up training project staff. The previous distribution had been done in 2007 and, by 2010, the items was verified. 15 were deemed too old to function. The contamination likely was caused by a combination of self-selection and EA 13 The full design consists of multiple treatment arms. A second treatment arm was oversight. CFs in the control group could have easily learned about the trainings from overlaid on our central training that randomly assigned 75 of the 150 treated peers in other communities. Since EAs and district staff were involved in organizing the communities to have an additional trained female. This second treatment is the subject training, it is easy to see how a well-connected CF might have been invited in. 16 of a separate study. In the present study, we pool the two treatments together, to examine A limitation of working with an existing extension network is that we could not the impact of having at least one CF in the community trained on SLM on farmers' withhold information from a random group of CFs by shutting down their interactions outcomes. A third randomized treatment arm was overlaid on the first two that with their assigned EAs. Given the small number of extension workers (30), reasonable attempted to provide different performance-based incentives for the CFs to reach levels of statistical power cannot be reached by assigning the intervention at the EA level. farmers in both villages that were assigned to the direct training and control commu- We do verify that extension agent characteristics are balanced across treatment and nities. These incentives were never announced to the CFs, and we show that they did not control communities at midline (Table A2). 5 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 our identification stems from the fact that CFs may request different Table 1 levels of attention from their EAs across treatment assignments, Contact farmers' characteristics by treatment status. displacing EA time away from the other treatment status—the extensive Variable Treated Control Difference margin of EA attention across treatments. For instance, treatment CFs may request more follow-up visits from their EAs, cutting into the time Mean SD N Mean SD N in mean EAs devote to control CFs. Reassuringly, we find that control and Baseline survey treatment CFs received equal amounts and types of attention from Age 38.858 9.348 148 40.160 10.559 50 −1.302 their EAs in the year after the training (Table A3). Formally 0.350 0.479 140 0.447 0.503 47 −0.097 trained Years since 2.157 2.239 51 3.409 3.202 22 −1.252* 3.4. Data formal training We conducted two follow-up surveys after the first training and Years of 2.243 2.401 144 2.653 2.570 49 −0.410 demonstration season (October 2010 to April 2011). A 2012 (midline) experience round and a 2013 (endline) round form a panel of households and CFs as CF Number of 18.034 16.095 147 19.100 14.333 50 −1.066 in the study area.17 We randomly sampled 18 non-CF households in farmers each community from a full listing performed by our enumerators assisted in ahead of the survey. At both midline and endline, households were last 7 days visited twice: pre- and postharvest. This allows us to observe SLM Number of 10.871 9.659 147 10.860 9.064 50 0.011 male practices when they are most visible, just after planting (preharvest, farmers from February to April), and to record production data after harvest assisted in (from mid-May on). Hence, our fieldwork ran from February to April last 7 days and May to August in 2012 and 2013. Number of 37.060 28.320 133 38.370 26.441 46 −1.309 Midline and endline surveys gathered longitudinal CF and house- farmers assisted in hold information on the two main agricultural seasons that followed last 30 days the first demonstration season. Our fieldwork included five survey Number of 22.480 15.145 148 22.240 17.203 50 0.240 instruments: a household questionnaire, a household agricultural male production questionnaire, a CF questionnaire, an EA questionnaire, farmers assisted in and a community questionnaire. The household survey was also last 30 days administered to all 200 CF households, in addition to the specific CF Hours worked 14.813 12.726 144 12.340 11.573 50 2.473 survey. These surveys provide household demographics, SLM knowl- as CF in last edge for the two main agricultural producers in the household, 7 days individual and plot-level SLM adoption, and production information Hours normally 16.322 12.498 143 12.960 12.034 50 3.362 worked as for approximately 3600 non-CF households in 200 communities. Since CF per week the plot roster identifies the adult in charge of making agricultural Total hectares 1.289 0.655 144 1.242 0.624 50 0.047 decisions for each cultivated plot, we obtain individual measures of of cultivated knowledge and adoption for a sample of 5884 and 5071 individuals at land Number of 284.421 267.037 126 244.548 265.410 42 39.873 midline and endline, respectively. households A rapid baseline survey was administered to all CFs in August 2010, in the before the district-level randomization. This provided data to perform community balance tests on the success of the randomization, using the pre- Number of 459.269 430.130 108 436.063 426.578 32 23.206 intervention characteristics of CFs by treatment status. Fig. 2 illustrates plots in the community the timing of the surveys and CF trainings over the course of the four- year study. Midline survey: Recalled Number of 2.839 2.362 137 3.286 2.255 42 −0.446 3.5. Descriptive statistics SLM techniques We briefly describe the average characteristics of farming indivi- learned duals in our sample (Table A4). More than half of the individuals are before 2010 Number of 1.409 1.210 137 1.167 0.935 42 0.242 women, and the prevalence of female headship is consistently high SLM (approximately 30 percent) for the region (TIA, 2008). The average techniques farmer is 38 years old with two years of schooling. Most plot owners are adopted married with three children, and live in a single-room house made of before 2010 mud and sticks with a palm or bamboo roof (not reported). Farmers Sources: Contact farmer baseline survey, 2010; Household survey, 2012. possess 2.2 hectares of land on average, with a standard deviation of Notes: ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. 2.1. CF, contact farmer. CFs are more knowledgeable (Tables 1 and 2), more educated, and wealthier (Table 3) than other farmers. While CFs are positively We also note that usage of demonstration plots was quite high and selected in attributes, they are also well known in their communities: not statistically different across treatment and control communities 84 percent of farmers in the control group declared knowing them (Table A3). Of the CFs in treatment and control communities, 85 personally. However, only 72 percent of the same group of farmers percent maintained a demonstration plot.18 Thus, changes in patterns reported knowing that these individuals assumed a role as CF in their community. 18 There was no instruction, however, as to what type of plot should be used for 17 Operational constraints precluded us from conducting a household survey at demonstration activities. CFs could choose to use their own, private plot or communal baseline. land. Hence, we present the demonstration results for any plot (own or not). 6 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table 2 Table 3 Other farmers' characteristics by treatment status. Socioeconomic and farming characteristics of contact farmers and other farmers. Variable Treated Control Difference Means Difference in mean Mean SD Mean SD in mean CFs Other farmers Midline survey: 2012 Is the head of household 0.585 0.493 0.588 0.493 −0.003 Household Characteristics: In the current year Male 0.420 0.493 0.414 0.493 0.005 Is the head of household 0.994 0.590 0.405*** Age 37.764 19.980 37.843 20.093 −0.079 Age 42.364 38.243 4.121*** Years of schooling completed 2.057 4.866 1.844 4.905 0.213 Years of schooling completed 5.481 2.054 3.427*** Single 0.063 0.504 0.058 0.509 0.005 Single 0.011 0.056 −0.044* Married 0.844 0.546 0.855 0.550 −0.011 Married 0.974 0.849 0.126*** Divorced, separated, or 0.091 0.366 0.085 0.368 0.006 Divorced, separated, or widowed 0.057 0.095 −0.038 widowed Number of children (ages < 3.744 2.830 0.913*** Number of children (ages < 2.756 3.406 2.843 3.432 −0.087 15 years) 15 years) Total hectares of owned land 3.439 2.171 1.269*** Total hectares of owned land 2.004 3.995 1.880 4.033 0.124 Number of rooms in the house 1.763 1.423 0.340** Number of rooms in the house 1.427 2.116 1.444 2.138 −0.017 Housing walls made of bricki 0.168 0.099 0.068 Housing walls made of brick 0.100 0.777 0.096 0.785 0.004 Housing roof made of tinplatei 0.207 0.079 0.128** Housing roof made of tinplate 0.079 0.718 0.079 0.725 0.000 Production: In the current rainy season Midline survey: Recalled Grew maize 0.725 0.637 0.088 Number of SLM techniques 1.236 4.514 1.303 4.563 −0.066 Grew sorghum 0.139 0.255 −0.116 learned before 2010 Grew cotton 0.133 0.076 0.058 Number of SLM techniques 0.509 2.024 0.554 2.045 −0.045 Grew sesame 0.270 0.156 0.113* adopted before 2010 Grew cassava 0.058 0.157 −0.099 Number of observations 4,385 1,499 5,884 Grew cow pea 0.278 0.347 −0.069 Grew pigeon pea 0.191 0.200 −0.009 Source: Household survey, 2012. Notes: t-test inferences are based on standard errors clustered at the community level. Farm characteristics: In the current rainy season ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. Plot size (hectares) 1.215 1.047 0.167 SLM, sustainable land management. Plot was flat 0.727 0.616 0.111* Plot was burnt 0.060 0.244 −0.184*** Used herbicides/pesticides/ 0.133 0.042 0.091*** fungicides of CF-to-other-farmers information diffusion across the two modalities Used natural fertilizer 0.484 0.351 0.133 Used chemical fertilizer 0.099 0.006 0.092*** can be interpreted as resulting from variations at the intensive margin Number of observations 351 10,960 11,311 of CFs' activities (e.g., number of techniques demonstrated, quality of the demonstration). Sources: Household survey, 2012, 2013. Notes: t-test inferences are based on standard errors clustered at the community level. CF, contact farmer. 3.6. Balance ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. i This variable is only available in Midline. We use data from the baseline CF survey as well as time-invariant and retrospective information collected in the 2012 household survey to check plot.21 The knowledge score is a continuous measure based on the for balance across treatments. Table 1 indicates minor differences between number of correct responses provided in the 23-question exam, CFs in the treatment and control communities. Treatment CFs spent almost covering all SLM techniques. For CFs, the majority of the analysis four more hours a week working as a CF (pre-intervention) and had slightly rests on their self-reported adoption of techniques on any plot more recent training when we condition on being formally trained. Control (demonstration or not).22 CFs were exposed to a greater number of techniques prior to the Since the CFs were encouraged to choose the techniques most intervention.19 In spite of these differences, (recalled) pre-intervention relevant to their local conditions, our main results focus on unweighted adoption rates among CFs in control and treated communities were similar, aggregate measures of knowledge and adoption. However, we create a as were other farmers' (recalled) baseline SLM learning and adoption rates second set of weighted knowledge and adoption outcomes as a (Table 2).20 21 3.7. Measuring information diffusion and behavioral change Our decision to focus on the knowledge score and self-reported adoption outcomes is motivated by the conclusions in Kondylis et al. (2015). Using the midline survey data, they find learning outcomes based on knowledge exams provide more precision than Central to identifying variations in information diffusion is know-by-name questions, inasmuch as they reveal the true knowledge of those measuring changes in learning and agricultural practices. Our study individuals less familiar with the name of the technique yet more familiar with its rests on the reliability of our markers of individual SLM knowledge purpose and usage. Objective adoption measures were also collected for two plots per household and largely corroborate the self-reported outcomes. In our triangulation of the and adoption. We focus on three outcomes: a knowledge score, the self-reported versus observed adoption, we find that false reporting is negligible. Since number of techniques the respondent identified by name, and the objective measures of adoption are collected for only a subset of plots (one per number of techniques the respondent reported having adopted on any respondent) at midline and a subset of the sample at endline, we instead focus on a more inclusive measure of adoption provided by self-reports of interviewed men and women. 19 22 Given that CFs in treatment villages spend more hours a week working as a CF at There are slight differences between adoption measures which include and exclude baseline, we will include the variable as a control in the regression analysis. the demonstration plot. This is due to the fact that some CFs demonstrate on communal 20 Balance tests for the CFs' and other farmers' knowledge and adoption of individual land (29%). We verify the results are not driven by communal propriety of the SLM techniques at baseline are reported in Tables A1 and A5. Because these values are demonstration plot (not reported). Since 71% of demonstration activities are carried based on recalled data, the tests should be interpreted with caution. Even though mean out on CFs' own plots, we choose to use pooled adoption on and off demonstration plots comparisons indicate there are no statistically significant differences, recall bias may be as our main marker of adoption. This improves our precision but does not affect our present. We therefore do not exploit the recalled information beyond balance checks. conclusions. 7 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 robustness check. Prior to aggregation, we multiply the technique by a 4. Results weight based on its relative importance to maize revenue. This is done as follows. First, we compute a vector of weights, based on a regression 4.1. CF learning and adoption of maize revenue on adoption of the seven individual practices. Second, we compute adoption and knowledge indices of practices weighted by We first examine the ITT estimates of a direct training on these correlations between adoption and maize production. unweighted aggregate measures of CFs' knowledge and adoption We additionally explore patterns of knowledge and adoption (Panel A, Table 4). While control CFs adopted on average 3.74 specific to individual techniques. We use responses from the same techniques, we detect that CFs adopt on average a 0.73 additional knowledge exam to quantify farmer knowledge of individual SLM technique in response to the training (statistically significant at the 5 techniques, categorizing questions by technique. The knowledge of a percent level). The associated effect size is large at 0.39 standard specific technique is a [0,1] continuous variable that depicts the share deviations in the control group, or a 19.6% increase relative to the of questions pertaining to the practice that the respondent has mean in the control. Next, we run similar specifications with the answered accurately. The adoption of a technique is captured by a weighted versions of the outcome.29 Control CFs adopt on average binary variable that indicates whether the farmer adopted the techni- 60.1% of the practices, and directly trained CFs increased adoption by que on at least one of his plots. about 10.6 percentage points (statistically significant at the 5% level; Knowledge, adoption, and perception of the SLM techniques were Panel B, Table 4). This effect is similar in magnitude to that obtained collected at the individual level from the household questionnaire. on our unweighted index, with a similar effect size of 0.40 standard Two respondents were interviewed: typically, the household head and deviation in the control group, or a 17.6% increase relative to the mean the head's partner or spouse.23 Our sample of CFs and other farmers in the control. Weighting our knowledge index confirms that directly consists of those who reported their personal information, partici- training CFs did not affect their knowledge scores. Overall, unweighted pated in an agricultural knowledge exam with questions related to and weighted results suggest that a direct training was effective in each specific SLM practice, and self-reported their SLM adoption raising CFs' adoption of SLM practices, with little effect on knowledge rates. Our final regressions samples consist of 347 CF-year observa- scores. tions and 10,955 person-year observations.24 Selective sample attri- To shed light on changes in the technique mix, we disaggregate the tion is of concern, and we address it in the next section. ITT estimates of adoption by technique (Table 5). Despite positive point estimates for all practices, statistical significance is achieved for 3.8. Empirical strategy only strip-tillage, pit planting and contour farming (statistically significant at the 10%, 1%, and 10% levels, respectively).30 The We causally estimate the intent-to-treat (ITT) effects of a commu- magnitude of these effects is substantial, ranging from 28.3% to 65% nity being assigned to a direct CF training (relative to a status quo CF increases relative to the control mean. To account for multiple modality) on the SLM knowledge and adoption of CFs25 and other hypothesis testing, we adjust our inferences for familywise error rates farmers in the community, Y, using a simple reduced-form specifica- (Šidák, p-value=0.015; Bonferroni, p-value=0.014), following Abdi tion: (2007). The effect on pit planting adoption is robust to multiple hypotheses testing (Fig. 4), with a 28 percent increase in adoption Yihjt = β0 + β1Tj + β2 Xi, h, j + νt + ϵi, h, j . (1) T takes the value 1 for each community j with a trained CF. Individual i, household h, and community characteristics are included in the vector (footnote continued) X to improve the precision of the estimated coefficients. An indicator appear consistent with those of other studies in the same region (De Brauw, 2014). A for the second follow-up survey,νt, is also included to capture the effect probit regression reveals that the greater the percentage of household members away in 2012 and the incidence of being single produces a greater probability of the household of time-specific events on behavior.26 We estimate all main regression moving out of the sample (Table A11). Age, the number of children of the household models on the pooled sample, controlling for survey-year fixed effects. head, and exposure to a precipitation shock reduce the probability of moving out of the We also use the Huber-White heteroskedasticity-robust estimator to sample. 28 calculate the standard errors when using the sample of CFs. For the We perform two additional robustness checks to examine the sensitivity of our results to attrition (not reported). The first diagnostic estimates (1) using the balanced other farmer regressions, we cluster the standard errors at the panel. We show that the inclusion of individuals present in both rounds affects the community level to allow for arbitrary correlation of treatment effects precision of our point estimates rather than their magnitude and sign. The second check within the community.27,28 bounds the treatment effect for selective attrition using a method proposed by Lee (2009). This check confirms that selective attrition is unlikely to affect our conclusions. 29 The linear model of maize revenue controls for adoption of each SLM technique, as 23 In the case of polygamous households, the main spouse was interviewed. Only 2.7 well as household demographics, as in Table 4, and production inputs. The regression percent of our sampled households are polygamous. estimates are presented in Table A12. The vector of weights consists of the estimated 24 The number of villages that were administered the CF survey were 179 and 172 in regression parameter for each technique divided by the sum of the parameters over all 2012 and 2013, respectively. The number of farmers interviewed in 2012 were 6252, and seven techniques. Since some regression coefficients are negative, we use improved 5290 in 2013. Sample sizes vary in descriptive statistic tables and some regression tables, fallowing as the reference weight. Thus, mulching takes value 0.227, strip tillage, 0.177, due to the addition of variables excluded from the main analysis. pit planting, 0.157, contour farming, 0.201, crop rotation, 0.193, and row planting, 25 CF-level regressions control for community-level CF outcomes and characteristics. 0.045. 30 In those communities where we (randomly) assigned an additional woman farmer to be We note that adoption trended downward in both treatment and control villages trained, we measure increased village-level exposure by regressing the maximum value of (not reported). However, these changes are fully attributable to a fall in demonstration of CF outcomes within the village on the maximum (mean) value of binary (continuous) SLM from midline to endline, while adoption on non-demonstration plots actually covariates. Switching to mean of outcomes, and controlling for mean and max of all increases from midline to endline (not reported). We additionally use rainfall data to covariates does not affect our conclusions. provide partial evidence that this trend cannot be explained by climatic conditions 26 We include variables that reflect CF (or other farmer) demographic characteristics: (NASA 1/2x 1/21981/2013 precipitation data available at http://power.larc.nasa.gov/ age, primary school completion, whether the individual is single (and a separate widow cgi-bin/cgiwrap/solar/hirestimeser.cgi?email=daily@larc.nasa.gov). Indeed, a dry shock dummy for the other farmer sample), number of children, total landholdings, the number during the rainy season prior to our midline survey (2011) and endline (2012) surveys of rooms in the house, the number of hours worked by the CF at baseline, an indicator for could affect adoption. Fig. 3 displays yearly standardized cumulative rainfall in the rainy a missing response for the baseline CF variable, district indicators, and indicators for season over the 1981/2012 period, and their 95% confidence intervals. We observe that treatment arms not analyzed in the present study. Our results are robust to specifications rainfall in the study years (2010/2012) are within normal range. Nonetheless, we test that omit the demographic characteristics (Tables A6 and A7) or replace district with whether adoption decisions vary by exposure to a dry spell over the rainy season (where administrative post fixed effects (Tables A8 and A9). dry is defined as whether the cumulative rainfall during the growing season was below 27 Attrition rates at the household and CF level are not statistically different (Table the 31-year 25th percentile). We find that rainfall anomalies do not explain variations in A10) nor correlated across treatment groups (Table A11). Household attrition rates adoption (Table A13). 8 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table 4 Effect of a direct SLM training on contact farmers' adoption and knowledge. Ctrl. Mean ITT N R2 [SD] Panel A: CFs' Knowledge and Adoption, unweighted Knowledge score 0.633 0.052 347 0.102 [0.173] [0.055] Number of techniques known by name 4.131 0.706 347 0.098 [1.626] [0.546] Number of techniques adopted on own plot 1.786 0.673** 347 0.224 [1.309] [0.225] Number of techniques adopted on any plot 3.738 0.733** 347 0.241 [1.889] [0.250] Panel B: CFs' Knowledge and Adoption, weighted Knowledge score 0.646 0.059 347 0.138 [0.189] [0.056] Number of techniques known by name 0.687 0.079 347 0.071 [0.236] [0.076] Number of techniques adopted on own plot 0.314 0.113** 347 0.248 [0.220] [0.031] Number of techniques adopted on any plot 0.601 0.106** 347 0.226 [0.265] [0.034] Table 5 Effect of a direct SLM training on contact farmers' adoption of individual SLM techniques. Adoption on Ctrl. Mean ITT N R2 any plot Mulching 0.929 0.026 347 0.040 [0.046] Strip-tillage 0.548 0.159* 347 0.125 [0.072] Pit planting 0.560 0.159*** 347 0.093 [0.023] Contour farming 0.226 0.147* 347 0.241 [0.067] Crop rotation 0.726 0.066 347 0.166 [0.051] Row planting 0.440 0.096 347 0.081 [0.058] Fig. 3. Precipitation anomalies over time, Notes: Standardized cumulative rainfall in the Improved fallowing 0.310 0.080 347 0.169 200 study communities over the rainy season period (October–February). The rainy [0.070] season starting in October 2012 is labeled as 2012. Standardized values are computed as a ratio of the distance between cumulative rainfall and average historical cumulative Sources: Contact farmer survey, 2010, 2012, 2013; Household survey 2012,2013. rainfall to the historical standard deviation across all 200 communities. 95% confidence Notes: Regressions include the same explanatory variables as models in Table 4. intervals are presented around each yearly value. Our study period includes 2010 ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. (demonstration season), and 2011 and 2012 (adoption seasons). Midline and endline ITT=intent-to-treat; SLM=sustainable land management. surveys correspond to 2011 and 2012, respectively. relative to the control mean.31 the observed increase in adoption among CFs, beyond their willingness Finally, we examine changes in technique-specific knowledge as a to comply with the training. In practice, we modify (1) to estimate the result of the direct training. In line with our aggregate measures of ITT effects on maize yields and revenue, input use, and on-farm labor knowledge, Table 6 indicates that adding a direct training to the CF allocation. model did little to increase CFs' knowledge scores. Table 7 (Panel A) presents maize yields (revenue per hectare) and total revenue accrued from a direct SLM training. Given our low level of 4.2. Private returns on SLM statistical power, we present results on the full sample and winsorizing yields at 1% to account for outliers.32 Results show positive though Farmers will adopt a technique only if it demonstrates (public or imprecise point estimates, indicating effect sizes on the order of 0.13– private) positive returns. Recent observational and experimental 0.24 standard deviations. evidence documents positive maize yield effects of SLM techniques in Since most SLM practices disseminated have water-conserving southern Africa, as well as substantial labor savings (Beaman et al., properties (Liniger et al., 2011), we further account for the possibility 2014; BenYishay and Mobarak, 2014; Haggblade and Tembo, 2003; that rainfall patterns in the main growing season may mediate the Thierfelder et al., 2015). We examine private returns to SLM to explain impacts of the intervention. In practice, we add a control variable to distinguish effects by whether the community experienced a dry spell 31 We also follow Anderson (2008) and address multiple inference in two additional ways (not reported): (1) using a free step-down resampling method to our p-values for 32 familywise error rate, and (2) employing the false discovery rate control methodology Winsorizing yields at 1% does not affect the control group mean as the entire top 1% proposed by Benjamini and Hochberg (1995). All yield the same results. of the distribution is in the treatment group. 9 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Fig. 4. Effect of SLM training intervention on contact farmers, controlling for familywise error rate. hours spent seeding over the week preceding the interview and the total Table 6 Effect of a direct SLM training on contact farmers' knowledge of individual SLM weeks spent farming at endline (Table 9). In particular, CFs spent 6.6 techniques. fewer hours seeding in the last week (a relative effect size of 0.63 standard deviations) and 7.1 fewer weeks farming over the last year (a Knowledge Ctrl. Mean ITT N R2 relative effect size of 0.37 standard deviations).34 While large, these score point estimates are in line with the magnitude of effects mentioned in Mulching 0.893 0.043* 347 0.135 the literature (30 days per year reported in Haggblade and Tembo, [0.017] 2003). Since use of herbicides remained constant across treatment Strip-tillage 0.512 0.089 347 0.095 arms (Table 8), labor savings at endline are plausibly attributable to [0.057] increased SLM adoption and complementary usage of the tools Pit planting 0.798 0.050 347 0.075 [0.088] provided in the kit to minimize tillage operations. Contour farming 0.520 0.127 347 0.098 [0.116] Crop rotation 0.567 0.008 347 0.072 4.3. Others farmers' knowledge and adoption [0.046] Row planting 0.310 −0.026 347 0.149 We now turn to CFs' ability to spread knowledge and adoption [0.113] among other farmers in the community. We exploit the exogenous, Improved fallowing 0.690 0.007 347 0.046 [0.040] positive shock in CFs' demonstration of SLM induced by the direct training to measure the extent of CF-to-others knowledge transmission. Sources: Contact farmer survey, 2010, 2012, 2013; Household survey 2012, 2013. Since we cannot exclude the possibility that our treatment affected Notes: Regressions include the same explanatory variables as models in Table 4. other farmers' adoption of SLM through channels other than demon- ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. stration, we adapt (1) and estimate the ITT of directly training CFs on ITT=intent-to-treat; SLM=sustainable land management. adoption and knowledge on a random sample of other farmers in the community. during the survey round: Dry Year, which takes value one if the Table 10 reports the ITT estimates of directly training CFs on CF– cumulative precipitation in a given location fell below the 30-year first farmer interactions and other farmers' aggregate knowledge and quartile, zero otherwise.33 In line with recent experimental evidence on adoption of SLM. First, we note that the direct training did not increase pit planting (Beaman et al., 2014), we find that training CFs on SLM farmers' access to CFs. Second, other farmers' SLM knowledge and has positive and large, if noisy, effects on maize yields and revenues adoption are unaffected by their exposure to a directly trained CF, during drier spells (Table 7, Panel B). The magnitude of the effects, on despite the margin of gains in SLM awareness being larger than for the order of 0.35 standard deviations, or 37% increase in both yield and CFs. These zero effects are robust to balancing the panel at midline, total revenue, are in line with findings from the literature that claim accounting for selective attrition at endline (not reported), and cannot increases of 50 to 100% (Haggblade and Tembo, 2003). The measured be explained by anomalous precipitations over the study period (Fig. 3; yield effects from receiving a SLM training corroborate the notion that (FEWS, 2012, 2013)). Looking at ITT effects by technique confirms this farmer adoption of SLM technologies is motivated by short-term yield general pattern (Table A14) . advantages. In the absence of any statistically significant differences in Recall the direct training led to a 15.9 percentage point increase in input use as a result of the direct training (Table 8), these yield effects CF adoption of pit planting. Other farmers in treated communities were in dry conditions are credibly attributable to SLM adoption. more likely to adopt pit planting by 2.8 percentage points (albeit a non- An additional economic benefit of applying SLM is in the form of significant effect). Placing a 95% confidence interval around this point large labor savings that follow from refining tillage operations and estimate allows us to rule out adoption rates higher than herbicide applications (Mazvimavi et al., 2011). These gains are (0.028 + 1.96 × 0.018 = )6.3% for pit planting among other farmers. A expected to materialize from the second adoption season onward, back-of-the-napkin calculation on this (weak) pit planting result rules since the first year requires at least equal amount of land preparation as out a propagation rate higher than (6.3/15.9 = )39.6%. Adoption of pit traditional practices (Haggblade and Tembo, 2003). Building on this planting by a CF would, at best, inspire less than half of a farmer in the literature and recent large-scale experimental evidence (BenYishay and community to adopt pit planting. This implies low and slow CF-to- Mobarak, 2014), our findings support a delayed contribution of SLM to other-farmers technology diffusion in the augmented CF model: labor savings. We witness a substantial reduction in the number of increasing demonstration of SLM has little effect on other farmers' behavior. 33 For these computations, we use NASA 1/2x 1/21981/2013 Precipitation data Learning about other farmers' perceptions of labor savings asso- available at http://power.larc.nasa.gov/cgi-bin/cgiwrap/solar/hirestimeser.cgi? email=daily@larc.nasa.gov. This measure of weather event is used by others in the 34 literature, see for instance Jayachandran (2006). These effects are robust to 1% top and bottom winsorizing. 10 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table 7 Effect of a direct SLM training on contact farmers' maize production. (1) (2) (3) (4) (5) (6) (7) Control mean T Dry year T× N R2 T+T× [SD] Dry year Dry year Panel A: Without controlling for precipitation Revenue per Ha Original data 1441.540 326.399 347 0.089 [MZN/Ha] [1417.638] [319.157] Winsorize at 1% 1441.540 182.322 347 0.147 [1417.638] [121.077] Total revenue Original data 4408.667 1137.593 347 0.121 [MZN] [4687.363] [721.392] Winsorize at 1% 4408.667 722.089 347 0.152 [4687.363] [438.057] Panel B: Controlling for precipitation Revenue per Ha Original data 1441.540 41.837 178.182 866.412* 347 0.097 908.249 [MZN/Ha] [1417.638] [399.865] [451.867] [354.536] [431.334] Winsorize at 1% 1441.540 9.163 133.582 528.280 347 0.157 537.443* [1417.638] [152.561] [440.298] [257.942] [237.038] Total revenue Original data 4408.667 744.990 19.641 1185.839** 347 0.124 1930.829 [MZN] [4687.363] [632.045] [1014.573] [359.641] [977.493] Winsorize at 1% 4408.667 264.976 −213.874 1370.718** 347 0.156 1635.694* [4687.363] [310.816] [996.793] [453.943] [664.713] Sources: Contact farmer survey, 2010; Household survey 2012, 2013; NASA 1/2x 1/21981/2013 Precipitation data. Notes: Dry year is a dummy indicating cumulative precipitation in the rainy season is in the first quartile of the 1981–2013 historical average. All models include the same explanatory variables as models in Table 4. T+T×dry year (col 7) presents the total effect of the treatment T and its interaction with dry year on maize yield and revenue. The associated standard errors are in brackets. Significance on the additive effect is determined by a Wald test. ×=multiplied by; SLM=sustainable land management; MZN=Mozambican Metacais; Ha=hectare. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. reported). These measures of communication and perceptions indicate Table 8 that a direct training did not contribute to increasing CF-other farmers Effect of direct SLM training intervention on contact farmers' input use. interactions, and that CFs' increased demonstration and use of SLM had little impact on other farmers' perceptions of these techniques. Ctrl.Mean ITT Na R2 [SD] Lastly, we explore whether CFs' characteristics provoke heteroge- neous responses among farmers. We focus on four CF indicators: above Burnt farm plot 0.167 −0.015 347 0.037 median educational attainment, above median age, above median [0.060] landholdings, and production of the same two crops as the farmer.35 Used natural fertilizer 0.524 0.129 343 0.166 [0.067] Each regression separately adds an interaction of the treatment Used chemical fertilizer 0.071 0.065 343 0.037 variable with one of the four indicators and the interacted indicator [0.042] on its own. Working with an existing network of CFs, we could not Amount of chemical fertilizer used (kg) 2.681 35.227 341 0.027 exogenously vary their education, age, wealth, or cropping patterns. [21.992] [19.514] Thus, the results that follow cannot be interpreted causally, but as Amount of chemical fertilizer used (l) 3.855 1.282 341 0.033 [27.973] [5.836] descriptive evidence. In addition, CFs are, on average, of higher status Use herbicides/pesticides/fungicides 0.250 0.022 343 0.082 than other farmers, which reduce the number of variations we have [0.032] access to in establishing a counterfactual. Finally, to yield interpretable Amount of herbicides/pesticides/fungicides 0.060 0.374 341 0.014 results, we need this exercise to focus on a single technique rather than used (kg) [0.361] [0.223] on an aggregate measure of adoption. We focus on the adoption of pit Amount of herbicides/pesticides/fungicides 4.367 −0.928 341 0.048 planting among other farmers as the outcome, since it is the only single used (l) practice which was statistically significantly adopted by CFs, when [11.714] [1.582] adjusting our inference for multiple hypothesis testing. Table 11 displays the results from interacted regression models Sources: Contact farmer survey, 2010; Household survey, 2012, 2013. accounting for farmer heterogeneity in the treatment effects, with the Notes: All models include the same explanatory variables as models in Table 4. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. additive effect of the treatment and its interaction with the CF ITT=intent-to-treat; SLM=sustainable land management. characteristic reported in the last row. Overall, we find CFs with above a The main sample has 347 observations. Sample size varies across models due to median total landholdings were 4.4 percentage points more likely to missing values in the dependent variables. convince other farmers to adopt pit planting, a 64.7% increase relative to the control (significant at the 10% level). Credibility in the source of ciated with each SLM practice may shed light on the mechanisms information appears to influence all farmers, CFs with larger farms underlying adoption—or, in our context, a lack thereof. We asked perhaps commanding more trust and respect within the community. farmers whether they perceived each technique to require more labor effort, equivalent labor effort, or less labor effort than traditional 35 cultivation practices. Farmers in the control group perceived all Our specification implies that having similar primary crops as the CF is an techniques to be labor intensive, with a range of less than 1 percent exogenous decision. Specifically, we assume cropping decisions are made before adoption decisions, and cropping decisions are independent of the treatment. While we cannot to 16 percent of farmers declaring the techniques to decrease the verify the order of the respective planting decisions, we find that other farmers' amount of labor required (not reported). Exposure to a trained CF does propensity to grow the same primary two crops as the CF is not affected by the treatment not favorably affect farmers' perceptions of adoption costs (not (not reported). 11 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table 9 Effect of a direct SLM training intervention on contact farmers' labor allocation. Pooled Sample Endline Control ITT N R2 Control ITT N R2 mean [SD] mean [SD] Hours spent on 6.095 −2.822 346 0.032 6.429 −1.272 168 0.073 preparation of land [14.550] [3.356] [15.353] [2.998] Hours spent on seeding 8.214 −3.558** 346 0.021 10.357 −6.567* 168 0.065 [15.996] [1.069] [17.862] [2.943] Hours spent on 2.607 −1.408 346 0.062 1.738 −0.917 168 0.087 transplantation [7.973] [0.957] [6.666] [0.736] Hours spent on irrigation 0.000 −0.038 346 0.028 [0.000] [0.044] Hours spent on sacha 10.583 0.454 346 0.198 5.833 −0.690 168 0.072 [15.576] [1.098] [14.252] [1.514] Hours spent on protection 0.000 0.969 346 0.042 0.000 0.513 168 0.122 [0.000] [0.922] [0.000] [0.499] Hours spent on harvesting 11.012 −2.288 346 0.122 15.810 −2.234 168 0.114 [18.260] [1.417] [19.573] [1.164] Total weeks spent on 28.262 −2.986 346 0.029 30.381 −7.084** 168 0.069 farming in last year [18.386] [2.210] [19.196] [2.508] Sources: Contact farmer survey, 2010; Household survey, 2012, 2013. Notes: All models include the same explanatory variables as models in Table 4. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. ITT=intent-to-treat effect; SLM=sustainable land management. compare the average annual costs of directly training each CF to the Table 10 average annual benefits realized by the CF. We consider three scenarios Effect of a direct SLM training intervention on other farmers' access to contact farmers, adoption, and knowledge. where the private returns to a direct CF training are in the form of maize revenue, labor earnings, and both.36 To compute the value of Ctrl. Mean ITT N R2 labor savings, we multiply the estimate of the intervention's impact on [SD] farm labor savings by the shadow value of labor.37 Benefits to other Access to CF farmers are excluded from all three scenarios given the non-significant Has access to any contact farmer in the 0.170 0.032 10,955 0.046 (negative) ITT point estimates on adoption. last half year Calculations for the three scenarios are presented in Table 12. Positive [0.027] net benefits over one adoption season exist when we account for the labor benefits only, and when we account for both labor and yield benefits. The Other Farmer Knowledge and Adoption, unweighted costs of training represent 12.8 percent of the total annual costs of running Knowledge score 0.341 −0.004 10,955 0.055 the extension system,38 with benefits ranging from USD -55 to USD 76 per [0.200] [0.012] CF per season. Number of techniques known by name 1.654 0.000 10,955 0.022 Although we fail to measure diffusion to other farmers in our [1.538] [0.120] Number of techniques adopted 0.845 −0.034 10,955 0.060 context, the predicted returns from CF's labor savings justify scaling a [0.891] [0.071] program of this nature. This motivates further research in extension modalities to improve the delivery of information to other farmers to Sources: Contact farmer survey, 2010; Household survey, 2012, 2013. augment the pool of program beneficiaries. For instance, providing Notes: Regressions include the following variables: a constant, age, a completed primary performance-based incentives to CFs and tempering the selection of school dummy, a dummy for male, a single dummy, a widow dummy, number of CFs does appear to achieve greater rates of technological adoption children, total landholdings, the number of rooms in the dwelling, baseline CF's number of years since formal training, a dummy for missing the baseline CF variable, district within communities (Beaman et al., 2014; BenYishay and Mobarak, indicators, an incentive treatment dummy, and an endline dummy. 2014; BenYishay et al., 2015). ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. CF=contact farmer; ITT=intent-to-treat effect; SLM=sustainable land management. 5. Discussion Decentralized extension modalities continue to garner support in Africa More interestingly, similarities in crop portfolios between CFs and despite criticism, often anecdotal, of their inefficacy in reaching most other farmers appear to influence adoption rates. Other farmers' farmers and providing relevant information. We designed an experiment in adoption grew an additional 5.2 percentage points when they had Mozambique to examine whether adding an in-depth centralized training access to a directly trained CF who grew similar crops to theirs. This is on a new technology improves the knowledge and adoption of innovative consistent with the idea that homogeneous farming conditions are conducive to social learning (Munshi, 2004). Delays in adoption may 36 stem from differences in production technologies and an inability to This approach ignores the social benefits produced by the technologies which cannot be quantified over a short-term horizon, such as soil and water quality. extrapolate demonstrated activities to their own plot. 37 We price the shadow value of labor at the minimum agricultural wage offered in Mozambique. Minimum wage rates are provided by the U.S. State Department: http:// www.state.gov/e/eb/rls/othr/ics/2013/204700.htm. Agriculture is the lowest wage rate 4.4. Cost–benefit analysis at 74 USD per month. 38 We focus on the costs specific to the SLM intervention, which include the annual per To provide perspective on the cost-effectiveness of the program, we community cost of an extension agent and per community cost of the SLM training. 12 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table 11 Table 12 Effect of a direct SLM training intervention on other farmers' adoption of pit planting, by Cost–benefit analysis of a direct SLM training intervention. CFs' characteristics. Yield Labor Yield and CF characteristics Educ > Age ≥ Land ≥ Same crop benefits benefits labor Median Median Median benefits T 0.029 0.035* 0.018 0.025 Number of Beneficiaries per [0.021] [0.020] [0.019] [0.019] community CF characteristics 0.011 −0.010 −0.014 0.003 CFs 1 1 1 [0.020] [0.016] [0.016] [0.023] Average Costs per T×CF characteristics 0.000 −0.011 0.025 0.026 Beneficiary [0.025] [0.023] [0.022] [0.027] Total cost of trainings 22,262 22,262 22,262 N 9,836 9,836 9,836 9,968 Annual cost of training 11,131 11,131 11,131 R2 0.010 0.011 0.010 0.010 Annual training cost per farmer 74 74 74 Control mean 0.068 0.068 0.068 0.069 Annual cost of extension agent 505 505 505 T+T×CF 0.029 0.023 0.044* 0.052* per farmer characteristics Average Benefits per [0.025] [0.023] [0.024] [0.030] Beneficiary Annual maize revenue 19 0 19 Sources: Contact farmer survey, 2010; Household Survey, 2012, 2013. Weekly agricultural wage rate 19 19 19 Notes: Regressions include the same explanatory variables as models in Table 10. Number of weeks in labor savings 0 7 7 The cutoff values for CF characteristics correspond to the median values of education (7 Annual labor earnings 0 131 131 years), age (41 years at ML, 43 years at EL), and landholdings (2.75 ha at ML, 3.5 at EL) Net Average Benefits per in the sample of CFs. Beneficiary T+T×CF characteristics (bottom row) presents the total effect of the treatment T and its Total net benefits per CF −55 57 76 interaction with the CF characteristic. The associated standard errors are in brackets. Significance on the additive effect is determined by a Wald test. Notes: CF=Contact farmer. Figures presented in terms of 2012 USD, assuming exchange ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. rate of 38 Metacais per 1 USD. ×=multiplied by; CF=contact farmer; SLM=sustainable land management. ML=midline; Annual cost per extension agent is based on the monthly salary of the extension agent EL=endline. (211 USD) and assumes one extension agent services five communities. Annual benefits in maize revenue obtained from estimates of the ITT in the winsorized specification in Table 7. Annual benefits in endline labor savings taken from estimates of the ITT on the number agricultural practices. We show that adding a direct training to an existing of weeks worked in Table 9. CF model increases adoption of SLM among CFs. Net private returns come mostly in the form of labor savings associated with the new practices. Despite these gains in adoption, adding a direct training to the CF modality had little impact on CF knowledge scores. This could of course be geneity in farming conditions (Conley and Udry, 2010; Munshi, 2004), and the result of poor quality testing and measurement error (Laajaj and social distance between messengers and peers (Feder and Savastano, 2006; Macours, 2015). Alternatively, relative to the status quo extension modality, Beaman et al., 2014; BenYishay and Mobarak, 2014). the direct training may not have changed adoption by increasing CFs' An alternative explanation is that farmers may learn more from their knowledge. Knowledge is not a necessary condition for adoption of a new own experience than from their peers (Foster and Rosenzweig, 1995; Bryan technology, as manifestations of herd behavior indicate (Banerjee, 1992; et al., 2014; Dupas, 2014). Failing to notice a gap between knowledge and Karlan et al., 2014). Instead, the direct training intervention may have actual practice, and not the information set itself, may also pose a key gotten more CFs to adopt SLM by strengthening their sense of identity as barrier to learning. Hanna et al. (2014) find that seaweed farmers in communicators in their community. The absence of differences in the use of Indonesia acted on the information received only when it included a demonstration plot, interactions with other farmers, and subjective descriptions of the relationship between yield and pod size from their happiness (not reported) across treatment arms however suggests CFs' own plot. If the main constraint to adoption of a profitable practice such as dedication and esteem were unaffected. Another possibility is that adding a SLM is not a lack of exposure or knowledge, but a failure to notice its centralized CF training may have heightened the quality and credibility of benefits, then augmenting the CF model will have little effect on the pace of the information, beyond the scope of our knowledge test. The participatory diffusion within the community. nature of the training may have helped CFs convert the information into While we cannot reject that adding a direct training to a decen- productive behavior, and fostered higher peer learning. tralized extension model is a cost effective intervention, more work is Although adding a direct CF training successfully encouraged adoption needed to understand the potential of community-level demonstration of a new technology at the village level relative to the status quo model of activities on technology diffusion. The profile of the “seed adopters” extension, it failed to encourage higher diffusion to other farmers in the influences whether farmers act on the information they receive. When community. There are a number of reasons why this may be the case. First, focusing on farmers with similar cropping patterns as their CF, we increased demonstration may not effectively address other barriers to observe modest (yet statistically significant) technology diffusion. adoption. Having access to a demonstration plot may need to be Complementary interventions, such as assigning different types of seed complemented by other learning inputs, such as CF time or other farmers' adopters (Beaman et al., 2014; BenYishay and Mobarak, 2014; time. Adding a direct CF training does not address the fact that CFs' BenYishay et al., 2015) or encouraging experiential learning in the opportunity costs of time may limit interactions with peers. For instance, community (Jones et al., 2015), may increase the pace of technology adding a performance-based incentive payment for contact farmers is diffusion in the context of decentralized extension services. shown to positively affect their impact in Malawi (BenYishay and Mobarak, 2014). Similarly, increasing demonstration of a yield-enhancing practice may not address other demand-side inefficiencies, such as the tendency to Appendix A. Additional Tables delay adoption until profitable (Foster and Rosenzweig, 1995), hetero- See Tables A1-A14. 13 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table A1 Pre-intervention See Tables A1 to A14. SLM training across treatment status (recalled). Variables Treated mean Control mean Difference in mean Contact Farmers: before 2010 Learned mulching 0.620 0.762 −0.141* Learned strip-tillage 0.321 0.429 −0.107 Learned pit planting 0.504 0.524 −0.020 Learned contour farming 0.307 0.381 −0.074 Learned crop rotation 0.591 0.690 −0.099 Learned row planting 0.285 0.238 0.047 Learned improved 0.212 0.262 −0.050 fallowing Number of observations 137 42 179 a Other Farmers : before 2010 Learned mulching 0.306 0.337 −0.031 Learned strip-tillage 0.182 0.227 −0.045 Learned pit planting 0.145 0.113 0.032 Learned contour farming 0.039 0.048 −0.009 Learned crop rotation 0.360 0.360 0.000 Learned row planting 0.104 0.114 −0.010 Learned improved 0.101 0.104 −0.003 fallowing Number of observations 4,385 1,499 5,884 Source: Household survey, 2012. Notes: ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. SLM=sustainable land management. a t-test inferences are based on standard errors clustered at the community level. * Significance at the 10 percent critical level for t-statistics. Table A2 Extension agents' characteristics by treatment status. Variables Treated Control Difference Mean SD Mean SD in mean EA age 35.415 4.646 34.925 4.962 0.489 EA years of schooling 7.192 0.534 7.263 0.601 −0.071 completed Number of years worked as EA 6.388 5.919 5.355 4.329 1.033 Number of years worked in 4.451 2.893 4.412 2.994 0.038 agricultural section, before became an EA Number of training received 9.624 5.265 9.645 5.563 −0.021 over the past 5 years Received training from the 0.344 0.477 0.289 0.460 0.055 Ministry of Agriculture Received training from 0.752 0.434 0.816 0.393 −0.064 Smallholders' project Number of weeks in training 1.244 0.601 1.276 0.601 −0.032 during the last 12 months One of the main topics covered 0.944 0.231 0.974 0.162 −0.030 in the trainings was conservation agriculture Number of villages 125 38 163 Table A3 Effect of a direct SLM training intervention on contact farmers' use of demonstration plots and access to extension agents. Ctrl. Mean ITT N R2 Used demonstration plot during the last 0.845 −0.034 347 0.043 year [0.066] EA visited CF at least once/month 0.512 0.001 347 0.033 [0.073] EA visited CF at least once/half year 0.631 0.055 347 0.047 (continued on next page) 14 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table A3 (continued) Ctrl. Mean ITT N R2 [0.123] EA visited CF at least once/year 0.667 0.143 347 0.048 [0.123] Sources: Contact farmer survey, 2010, 2012, 2013; Household survey, 2012, 2013. Notes: Regressions include the following variables: a constant, age, a completed primary school dummy, a single dummy, number of children, total landholdings, the number of rooms in the dwelling, baseline CF's number of years since formal training, a dummy for missing the baseline CF variable, district indicators, an incentive treatment dummy, and an endline dummy. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. CF=contact farmer; EA=extension agent; ITT=intent-to-treat effect. Table A4 Other farmers' characteristics. Variables Mean SD Is the head of household 0.590 0.492 Age 38.243 14.430 Years of schooling completed 2.054 2.798 Single 0.056 0.229 Married 0.849 0.359 Divorced, separated, or widowed 0.095 0.293 Number of children [ages < 15 years] 2.830 2.041 Total hectares of owned land 2.171 2.064 Number of rooms in the house 1.423 0.724 Number of observations 10,960 Sources: Household survey, 2012, 2013. Table A5 Pre-intervention SLM adoption by treatment status (recalled). Variables Treated mean Control mean Difference in mean Contact farmers: before 2010 Adopted mulching 0.489 0.405 0.084 Adopted strip-tillage 0.248 0.214 0.034 Adopted pit planting 0.190 0.167 0.023 Adopted contour farming 0.007 0.000 0.007 Adopted crop rotation 0.314 0.262 0.052 Adopted row planting 0.124 0.095 0.029 Adopted improved 0.036 0.024 0.013 fallowing Number of observations 137 42 179 a Other Farmers: before 2010 Adopted mulching 0.181 0.203 −0.022 Adopted strip-tillage 0.087 0.118 −0.031 Adopted pit planting 0.059 0.036 0.023 Adopted contour farming 0.002 0.000 0.002 Adopted crop rotation 0.121 0.132 −0.011 Adopted row planting 0.055 0.059 −0.005 Adopted improved 0.005 0.005 0.000 fallowing Number of observations 4,385 1,499 5,884 Source: Household survey, 2012. Notes: ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. SLM=sustainable land management. a t-test inferences are based on standard errors clustered at the community level. 15 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table A6 Effect of a direct SLM training intervention on contact farmers' knowledge and adoption, basic specification. Ctrl. Mean ITT N R2 [SD] CFs' Knowledge and Adoption, unweighted Knowledge score 0.633 0.046 347 0.019 [0.173] [0.049] Number of techniques known by name 4.131 0.646 347 0.018 [1.626] [0.571] Number of techniques adopted on own 1.786 0.594* 347 0.023 plot [1.309] [0.232] Number of techniques adopted on any 3.738 0.752** 347 0.020 plot [1.889] [0.244] Sources: Contact farmer survey, 2010, 2012, 2013; Household survey, 2012, 2013. Notes: Regressions include the following variables: a constant, treatment variables, an endline dummy, and district indicators. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. CF=contact farmer; ITT=intent-to-treat effect; SLM=sustainable land management. Table A7 Effect of a direct SLM training intervention on other farmers' knowledge and adoption, basic specification. Ctrl.Mean ITT N R2 [SD] Other Farmers' Knowledge and Adoption, unweighted Knowledge score 0.341 −0.003 10,955 0.049 [0.200] [0.012] Number of techniques known by name 1.654 0.013 10,955 0.012 [1.538] [0.119] Number of techniques adopted 0.845 −0.020 10,955 0.049 [0.891] [0.071] Sources: Contact farmer survey, 2010; Household survey, 2012, 2013. Notes: Regressions include the following variables: a constant, treatment variables, a male dummy, an endline dummy, and district indicators. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. ITT=intent-to-treat effect; SLM=sustainable land management. Table A8 Effect of a direct SLM training intervention on contact farmers' adoption and knowledge, includes administrative post fixed effects. Ctrl. mean ITT N R2 [SD] CFs' Knowledge and Adoption, unweighted Knowledge score 0.633 0.046 347 0.103 [0.173] [0.036] Number of techniques known by name 4.131 0.859** 347 0.105 [1.626] [0.367] Number of techniques adopted on 1.786 0.726*** 347 0.256 own plot [1.309] [0.216] Number of techniques adopted on any 3.738 0.658** 347 0.244 plot [1.889] [0.258] Sources: Contact farmer survey, 2010, 2012, 2013; Household survey, 2012, 2013. Notes: Regressions include the same explanatory variables as models in Table 4, except replacing district indicators with administrative post indicators. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. CF=contact farmer; ITT=intent-to-treat; SLM=sustainable land management. 16 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table A9 Effect of a direct SLM training intervention on other farmers' access to contact farmers, adoption, and knowledge, includes administrative post fixed effects. Ctrl. Mean ITT N R2 [SD] Other Farmers' Knowledge and Adoption, unweighted Knowledge score 0.341 −0.007 10,955 0.057 [0.200] [0.012] Number of techniques known by 1.654 −0.001 10,955 0.020 name [1.538] [0.116] Number of techniques adopted 0.845 0.003 10,955 0.053 [0.891] [0.066] Sources: Contact farmer survey, 2010; Household survey, 2012, 2013. Notes: Regressions include the same explanatory variables as models in Table 10, except replacing district indicators with administrative post indicators. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. CF=contact farmer; ITT=intent-to-treat; SLM=sustainable land management. Table A10 Attrition of contact farmers and other farmers. Variables Treated Control Difference Mean SD Mean SD in mean CFs attrited from Midline 0.109 0.313 0.048 0.216 0.062 Number of Observations 137 42 179 Household attrited from Midlinea 0.090 0.372 0.087 0.374 0.003 Number of Observations 2750 935 3685 Sources: Household survey, 2012, 2013; Contact farmer survey, 2012, 2013. Notes: CF=contact farmer. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. a t-test inferences are based on standard errors clustered at the community level. Table A11 Determinants of attrition (contact farmers and other farmers). CFs Other farming HH Treatment 1 0.053 Treatment 1 0.014 [0.066] [0.013] Treatment 3 0.002 Treatment 3 −0.013 [0.055] [0.012] Age −0.006** Age −0.001** [0.003] [0.000] Completed at least 0.056 HH head completed at −0.004 primary school [0.060] least primary school [0.012] Single −0.091 HH head Single 0.023 [0.190] [0.020] HH head divorced, 0.045*** widow, or separated [0.017] Total number −0.011 Total number −0.007** of children [0.013] of children [0.003] Total landholding 0.005 Total landholding −0.005 [hectares] [0.011] [hectares] [0.003] Total number of rooms −0.037 Total number of rooms −0.003 [0.032] [0.008] Number of years −0.034* Number of years 0.004 since formal training [0.020] since formal training [0.003] Missing above −0.145* Missing above 0.001 variable [0.076] variable [0.015] Household head −0.035 Household head −0.002 was female [0.088] was female [0.013] % of household −0.436 % of household 0.171*** members was away [0.350] members was away [0.065] HH has non-own 0.023 HH has non-own −0.013 farming work [0.078] farming work [0.012] HH has outside −0.015 HH has outside 0.003 (continued on next page) 17 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table A11 (continued) CFs Other farming HH employment [0.073] employment [0.017] 2012 precipitation −0.001 2012 precipitation −0.001* shock [0.002] shock [0.000] Constant 0.352 Constant 0.043 [0.308] [0.058] N 178 N 3662 R2 0.099 R2 0.014 Sources: Contact farmer survey, 2010, 2012, 2013; Household survey, 2012, 2013. Notes: Regressions include district fixed effect. CF=contact farmer; HH=household. Household attrition measured by whether a household surveyed in 2012 could not be interviewed in 2013. The CF attrition outcome reflects whether the village had at least one CF interviewed in 2012 but not in 2013. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. Table A12 Effects of contact farmers' adoption of SLM practices on maize revenue. Total Maize Revenue Adopted mulching on own plot 1458.547 [1032.677] Adopted strip tillage on own plot 723.044 [572.914] Adopted pit planting on own plot 436.835 [1502.123] Adopted contour farming on own plot 1078.387 [3145.182] Adopted crop rotation on own plot 957.972 [960.541] Adopted row planting on own plot −1206.388 [1472.065] Adopted improved fallowing on own plot −1871.335 [968.155] N 342 R2 0.187 Sources: Contact farmer survey, 2010, 2012, 2013; Household survey, 2012, 2013. Notes: Regressions include the same explanatory variables as models in Table 4. Additional controls include dummies for usage of all inputs displayed in Table 8, as well as labor allocated to maize production as displayed in Table 9. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. Table A13 Effect of a direct SLM training intervention on contact farmers' adoption of individual SLM techniques, controlling for lagged rainfall. Adoption (1) (2) (3) (4) (5) (6) (7) any plot Control T Dry Year T× N R2 T+T× Mean Dry Year Dry Year Mulching 0.929 0.045 0.034 −0.058 347 0.042 −0.013 [0.059] [0.077] [0.060] [0.035] Strip-tillage 0.548 0.174 0.101 −0.050 347 0.127 0.124 [0.099] [0.129] [0.119] [0.072] Pit planting 0.560 0.161*** 0.125 −0.017 347 0.097 0.144** [0.026] [0.104] [0.059] [0.046] Contour 0.226 0.146 0.052 −0.002 347 0.242 0.145** farming [0.097] [0.086] [0.099] [0.043] Crop 0.726 0.059 −0.054 0.025 347 0.166 0.084 rotation [0.047] [0.164] [0.098] [0.100] Row 0.440 0.040 0.198 0.137 347 0.117 0.178** planting [0.089] [0.111] [0.132] [0.050] (continued on next page) 18 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 Table A13 (continued) Adoption (1) (2) (3) (4) (5) (6) (7) any plot Control T Dry Year T× N R2 T+T× Improved 0.310 0.101 0.166*** −0.074 347 0.175 0.027 fallowing [0.065] [0.035] [0.038] [0.078] Sources: Contact farmer survey, 2010, 2012, 2013; Household survey, 2012, 2013. Notes: Dry year indicates that the precipitation amount falls in the first quartile of the historical distribution of cumulative rainfall during the rainy season (1981-2013). Regression model includes the same controls as in Table 4. T+T×Dry Year (col 7) presents the total effect of the treatment T and its interaction with Dry Year on adoption. The associated standard errors are in brackets. Significance on the additive effect is determined by a Wald test. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. Table A14 Effect of a direct SLM training intervention on other farmers' SLM adoption. Adoption Ctrl. Mean ITT N R2 Mulching 0.336 −0.022 10,955 0.056 [0.033] Strip-tillage 0.181 −0.035 10,955 0.031 [0.028] Pit planting 0.071 0.028 10,955 0.016 [0.018] Contour farming 0.005 −0.003 10,955 0.006 [0.003] Crop rotation 0.153 0.017 10,955 0.005 [0.018] Row planting 0.079 −0.012 10,955 0.018 [0.015] Improved fallowing 0.020 −0.007 10,955 0.003 [0.006] Sources: Contact farmer survey, 2010; Household survey, 2012, 2013. Notes: Regressions include the same explanatory variables as models in Table 10. ***, **, and * significance at the 1, 5, and 10 percent levels, respectively. ITT=intent-to-treat; SLM=sustainable land management. References 1671–1748. Conley, T.G., Udry, C.R., 2010. Learning about a new technology: pineapple in Ghana. Am. Econ. Rev., 35–69. Abdi, H., 2007. The bonferonni and šidák corrections for multiple comparisons. Coughlin, P.E., 2006. Agricultural intensification in Mozambique infrastructure, policy Encycloped. Meas. Stat. 3, 103–107. and institutional framework-when do problems signal opportunities. Maputo, Akroyd, S., Smith, L., 2007. Review of public spending to agriculture. A Joint Study by EconPolicy Research Group, sponsored by Sida. the Department for International Development and the World Bank, 5. Cunguara, B., Moder, K., 2011. Is agricultural extension helping the poor? Evidence from Anderson, J.R., Feder, G., 2004. Agricultural extension: Good intentions and hard rural Mozambique. J. Afr. Econ., (ejr015). realities. World Bank Res. Obs. 19 (1), pp. 41–60. Davis, K., Nkonya, E., Kato, E., Mekonnen, D.A., Odendo, M., Miiro, R., Nkuba, J., 2012. Anderson, J.R., Feder, G., 2007. Agricultural extension. Handbook of Agricultural Impact of farmer field schools on agricultural productivity and poverty in East Africa. Economics, vol. 3, pp. 2343–2378. World Dev. 40 (2), 402–413. Anderson, M., 2008. Multiple inference and gender differences in the effects of early De Brauw, A., 2014. Gender, control, and crop choice in northern Mozambique. intervention: a reevaluation of the abecedarian, perry preschool, and early training Technical Report, International Food Policy Research Institute (IFPRI). projects. J. Am. Stat. Assoc. 103 (484), 1481–1495. Dupas, P., 2014. Short-run subsidies and long-run adoption of new health products: Bandiera, O., Rasul, I., 2006. Social networks and technology adoption in northern evidence from a field experiment. Econometrica 82 (1), 197–228. Mozambique*. Econ. J. 116 (514), 869–902. Eicher, C.K., 2002. Building african models of agricultural extension: A case study of Banerjee, A., 1992. A simple model of herd behavior. Q. J. Econ. 107, 797–817. Mozambique. In: Workshop Extension and Rural Development: A Convergence of Banerjee, A., Cole, S., Duflo, E., Linden, L., 2007. Remedying education: evidence from Views on International Approaches, pp. 12–15. two randomized experiments in india. Q. J. Econ. 122 (3), 1235–1264. Feder, G., Savastano, S., 2006. The role of opinion leaders in the diffusion of new Banerjee, A.V., Banerji, R., Duflo, E., Glennerster, R., Khemani, S., 2010. Pitfalls of knowledge: the case of integrated pest management. World Dev. 34 (7), 1287–1300. participatory programs: Evidence from a randomized evaluation in education in FEWS, January 2012. Agromet update: 2011/2012 agricultural season. India. Am. Econ. J. Econ. Policy, 1–30. FEWS, March 2013. Agromet update 2012/2013 agricultural season. Beaman, L., BenYishay, A., Magruder, J., Mobarak, A. M., 2014. Can network theory Foster, A., Rosenzweig, M., 1995. Learning by doing and learning from others: human based targeting increase technology adoption. capital and technical change in agriculture. J. Polit. Econ., 1176–1209. Benin, S., Nkonya, E., Okecho, G., Pender, J., Nahdy, S., Mugarura, S., Kayobyo, G., Gallina, A., Chidiamassamba, C., 2010. Gender aware approaches in agricultural 2007. Assessing the impact of the national agricultural advisory services (naads) in programmes—Mozambique country report: A special study of the national the uganda rural livelihoods. IFPRI Discussion Paper. agricultural development programme (proagri ii). Technical Report, Swedish BenYishay, A., Jones, M., Kondylis, F., Mobarak, A.M., 2015. Are gender differences in International Development Cooperation Agency. performance innate or socially mediated? Gautam, M., 2000. Agricultural extension: The Kenya experience: An impact evaluation. BenYishay, A., Mobarak, A., 2014. Social learning and communication. NBER Working World Bank Publications. Paper 20139. Gemo, H.R., Chilonda, Pius, 2013. Why did Mozambique’s public extension halt the Bindlish, V., Evenson, R.E., 1997. The impact of t & v extension in africa: The experience implementation of the National Agrarian Extension Program (PRONEA)?. of Kenya and Burkina Faso. The World Bank Research Observer, pp. 183–201. International Food Policy Research Institute (IFPRI), Washington, DC, USA. Björkman, M., Svensson, J., 2009. Power to the people: evidence from a randomized field Gemo, H., Eicher, C.K., Teclemariam, S., 2005. Mozambique's Experience in Building a experiment of a community-based monitoring project in Uganda. Q. J. Econ. 124 (2), National Extension System. Michigan State University Press. 735–769. Glewwe, P., Kremer, M., Moulin, S., 2009. Many children left behind? Textbooks and test Bryan, G., Chowdhury, S., Mobarak, A., 2014. Under-inversment in a profitable scores in Kenya. Am. Econ. J. Appl. Econ. 1 (1), 112–135. technology: the case of seasonal migration in Bangladesh. Econometrica 82 (5), Glewwe, P., Kremer, M., Moulin, S., Zitzewitz, E., 2004. Retrospective vs. prospective 19 F. Kondylis et al. Journal of Development Economics 125 (2017) 1–20 analyses of school inputs: the case of flip charts in Kenya. J. Dev. Econ. 74 (1), Overview of Conservation Approaches and Technologies ((WOCAT)) and Food and 251–268. Agriculture Organization of the United Nations ((FAO)). Gollin, D., Morris, M., Byerlee, D., 2005. Technology adoption in intensive post-green Lucas, R., 1988. On the mechanics of economic development. J. Monet. Econ. 22 (1), revolution systems. Am. J. Agric. Economics 87 (5), 1310–1316. 3–42. Haggblade, S., Tembo, G., 2003. Conservation farming in Zambia. International Food Mazvimavi, K., Myathi, P., Murendo, C., 2011. Conservation agriculture practices and Policy Research Institute EPTD Discussion Paper #108. challenges in zimbabwe. In: Proceedings of the 5th World Congress of Conservation Hanna, R., Mullainathan, S., Schwartzstein, J., 2014. Learning through noticing: theory Agriculture Incorporating 3rd Farming Systems Design Conference, September 2011 and evidence from a field experiment. Q. J. Econ. 129 (3), 1311–1353. Brisbane, Australia. Hazell, P., 2013. What makes african agriculture grow?. 2012 global food policy report. Munshi, K., 2004. Social learning in a heterogeneous population: technology diffusion in Washington, DC: International Food Policy Research Institute (IFPRI). the indian green revolution. J. Dev. Econ. 73 (1), 185–213. Jayachandran, S., 2006. Selling labor low: wage rresponse to productivity shocks in Pamuk, H., Bulte, E., Adekunle, A.A., 2014. Do decentralized innovation systems developing countries. J. Polit. Econ. 114 (3), 538–575. promote agricultural technology adoption? Experimental evidence from africa. Food Jones, M., Kondylis, F., Mobarak, A.M., Stein, D., 2015. Evaluating the integrated Policy 44, 227–236. agriculture productivity project in Bangladesh. Purcell, D., Anderson, J.R., 1997. Agricultural extension and research: Achievements and Karlan, D., Ratan, A.L., Zinman, J., 2014. Savings by and for the poor: a research review problems in national systems. World Bank Publications. and agenda. Rev. Income Wealth 60 (1), 36–78. Tan, J.-P., Lane, J., Lassibille, G., 1999. Student outcomes in Philippine elementary Kondylis, F., Mueller, V., Zhu, S.J., 2015. Measuring agricultural knowledge and schools: An evaluation of four experiments. World Bank Econ. Rev. 13 (3), pp. 493– adoption. Agric. Econ. 46 (3). 508. Laajaj, R., Macours, K., January 2015. The reliability and validity of skills measurement Thierfelder, C., Matemba-Mutasa, R., Rusinamhodzi, L., 2015. Yield response of maize in rural household surveys. (zea mays l.) to conservation agriculture cropping system in southern africa. Soil Lee, D.S., 2009. Training, wages, and sample selection: estimating sharp bounds on Tillage Res. 146, 230–242. treatment effects. Rev. Econ. Stud. 76 (3), 1071–1102. Thompson, P., 2010. Learning by doing. Handbook of the Economics of Innovation 1, pp. Levitt, S., List, J., Syverson, C., 2012. Toward an understanding of learning by doing: 429–476. evidence from an automobile assembly plant. J. Polit. Econ. 121 (4), 643–681. TIA, 2008. National agricultural household survey. Michigan State University and The Linden, L.L., 2008. Complement Or Substitute?: The Effect of Technology on Student Ministry of Agriculture of Mozambique (MINAG). URL 〈http://fsg.afre.msu.edu/ Achievement in India. InfoDev. Mozambique/survey/index.htm〉. Liniger, H., Studer, R.M., Hauert, C., Gurtner, M., 2011. Sustainable land management Waddington, H., White, H., Anderson, J., 2014. Farmer field schools: from agricultural in practice-guidelines and best practices for sub-saharan africa. TerrAfrica, World extension to adult education. Syst. Rev.Summary 1. 20