None of these studies consider whether engaging the supply side of the market can increase adoption by farmers. Survey data in agriculture suggest that input suppliers act as the second most popular source of information for Indian farmers. Informing private input suppliers about technological benefits is one way to take advantage of their intrinsic motivation to spread information. Yet, it has not been looked at as source of potential information agents with well-aligned incentives in promoting the adoption of new technologies. Our paper is the first to implement and test this as a new approach to doing agricultural extension.The rest of this paper is organized as follows. Section 2 gives more background information on the setting and outlines the experiment. Section 3 describes the data collection. Section 4 presents the main results on how targeting agricultural extension to input dealers increases technology adoption by farmers, particularly those with the highest potential benefit. Section 5 turns to analyzing a potential explanation for this result. Particularly, it focuses on whether dealers spread information to their customers and what motivates them to do so. Section 6 concludes. This section starts by providing background information on the standard methods used in agricultural extension. It also gives a description of how the public sector delivers information to farmers in our particular study area. We then outline the design of our main experiment to compare these standard methods with the more business-oriented approach of using agrodealers as information agents. Governments all over the world support agricultural extension services as a mode of information delivery. Ministries of agriculture typically have entire departments dedicated to providing these services. These departments oversee local administrative offices that hire front line extension agents whose role is to diffuse information about new agricultural technologies and practices to farmers.
The specific techniques used by agents vary across contexts, but the basic methods are largely consistent,indoor garden especially in poor countries. Agents usually work with selected “contact farmers” who are keen on trying new approaches and are ideally able to transmit knowledge to others in their social networks. They also organize farmer field days with cluster demonstration plots, where new seeds are grown by multiple farmers, to boost the diffusion of information. The public sector provides agricultural extension services for at least two reasons. First, markets do not exist for many new innovations. For instance, a new planting method may only be promoted by government agents because there is no scope for profiting from its sale in private markets. This contrasts with new seed varieties or material inputs that are produced and sometimes marketed by private firms. Second, many agricultural innovations are not developed, and hence not marketed, by private firms. Public entities such as national agricultural research systems or international research organizations frequently develop new seeds, inputs, and agricultural management practices. The public extension service then transmits information about these developments to farmers. In the context of our experiment, agricultural extension workers use many of these standard techniques. Each of the 10 districts in the sample is organized into blocks, where a block has an average of 135 villages.The BAO employs Assistant Agricultural Officers and Village Agricultural Workers who work in the field with farmers. Our sample consists of 72 blocks in 10 flood-prone districts of Odisha.We selected these areas because the technology being promoted — a flood-tolerant rice variety called Swarna-Sub1 — is most suitable for flood-affected areas.
The blocks in the sample represent around 20 percent of the blocks in the state. We randomly assigned 36 of these blocks to the treatment group where agrodealers were targeted to receive seeds and information. This randomization was stratified by district. The remaining 36 blocks serve as a comparison group where we supported the government extension service to carry out normal extension activities. Figure 1 displays the timeline of these interventions. Starting in May 2016 — about 6-8 weeks before planting time — we partnered with the government’s extension service to introduce Swarna-Sub1 into control blocks. We did this in a way that mirrors three common practices in agricultural extension. First, field staff provided 10 seed minikits of 5 kilograms each to the BAO, who then helped identify contact farmers to use the kits. The BAO chose 2 villages and 5 farmers in each village. Each kit contained only seeds for testing and some basic information about Swarna-Sub1. Our field staff then delivered the kits to the recommended farmers. Second, we provided another 150 kg of seeds to the BAO so that he could set up a cluster demonstration where the seeds would be used by several farmers on a contiguous set of plots. Based on seeding rates in the region, 150 kg allows for cultivation of 5-10 acres. The BAO chose where to do the demonstration and which farmers to target. Official government guidelines for organizing these clusters suggest that they be carried out in sites that are easily accessible to be viewed by many farmers. Moreover, sites should be representative of average conditions in the area. Third, we helped the BAO carry out a farmer field day in November — at the time right before harvest. The BAO selected the location of the field day and whom to invite. The purpose of the field day was for extension staff to train farmers about Swarna-Sub1, share information from the demonstrations, and hope that information will spread throughout the block. The objective of such an active control group is twofold. First, it ensures that each block is endowed with the same quantity of seeds. Therefore, the dealer-based treatment only differs on who received the new seeds and information. Second, the demonstrations and partnerships with contact farmers may not have taken place without our involvement. Forcing these activities to happen makes the treatment-control comparison more meaningful. Most importantly, it sets a higher bar for the dealer-based treatment by eliminating any possibility that the new technology would not be promoted by the government extension service.
Turning to the 36 treatment blocks, we obtained a list of 2,087 seed suppliers from the state Department of Agriculture. These include suppliers of two types: private seed dealers and Primary Agricultural Cooperative Societies . PACS are farmer groups that handle credit, seed supply, and procurement of output for farmers. We did not include them in the intervention because their incentives are not the same as those of private dealers. Seed sales are usually handled by a member that is not the residual claimant on any profits from the sale. Despite being fewer in number relative to PACS, private dealers account for almost 60 percent of the seeds sold to farmers. The sample consists of 666 private dealers, 327 of which were located in the treatment blocks. Armed with this list, our field staff entered each treatment block and located five dealers interested in receiving seed minikits and an informational pamphlet about SwarnaSub1. In some blocks fewer than 5 dealers were available. We provided additional seed to each dealer in these cases to guarantee that a full 200 kilograms were introduced. The list provided by the Department of Agriculture did not have enough locatable dealers in some cases. In these circumstances, our field staff provided the seeds to other local agrodealers.Overall, seeds and information were provided to 151 dealers across the 36 treatment blocks.119 of these were from the original list. Once provided with seeds and information, the dealers were left alone to decide how to use them. We asked dealers about their intended uses. They overwhelmingly stated that they would use the seeds for testing on their own farms and would provide them to good customers for testing.Our intervention did not include any additional assistance to dealers. This differs from standard methods in agricultural extension where agents continually revisit their contact farmers. We allowed dealers to learn on their own because in theory they should be motivated to learn about a new product that could enhance their business. The goal of our treatment is to measure whether this motivation causes information to flow to farmers and ultimately increases adoption. Not intervening further ensures that our treatment effect is driven by any real-world incentives dealers have to learn, rather than heavy monitoring by our partners. Dealers in our sample are small business entrepreneurs. Some operate out of their homes, while others maintain shops in rural towns. 44 percent of dealers sell only seeds,hydroponic farming with fertilizers and pesticides being the most common inputs sold by the other dealers. Dealers are highly local. The median dealer sells enough rice seed to cover roughly 400 acres, which amounts to the rice area cultivated by 150 farmers.Importantly, dealers tend to serve the same customers from year to year. Another important feature of our context is that 84% of the seeds sold by dealers in our sample are produced by the state-run Odisha State Seed Corporation . Government subsidies explain this. Seeds produced by the state are subsidized at a rate of approximately 40 percent. No subsidies exist for seeds produced by private companies. As licensed agents, dealers receive a fixed commission that amounts to about 8 percent of the pre-subsidy price. All varieties have the same final price for farmers. Thus, the margin for dealers is the same across all types of varieties. Hence dealers have no direct financial incentive to sell one variety over another. Turning to the second season , we ran an SMS messaging experiment to compare our intervention with this “lighter touch” information treatment.
The random delivery of SMS messages allows us to test whether our dealer treatment substitutes basic knowledge that can be easily transmitted via ICT technology. Furthermore, it allows us to compare the direct effects of the two approaches. The messaging was simple. It informed farmers that Swarna-Sub1 is a new variety that is suitable for medium-low land in terms of elevation, matures in 145 days, and can tolerate up to two weeks of flooding. The message also stated that it was being produced by OSSC and could be available at local dealers. As a sampling frame, we obtained mobile numbers for 75,616 farmers that had registered for the state government’s Direct Benefit Transfer scheme to obtain seed subsidies.These farmers are located across the 261 gram panchayats that cover our main estimation sample, as outlined below. The SMS treatment was randomized at the gram panchayat level, resulting in messages being delivered to 37,783 of the names on the list.We anticipated that dealers and contact farmers would use the demonstration minikits for learning in 2016 and any possible treatment effects could first be detected during year two . Our main followup survey therefore took place in August-September 2017 — around 15 months after the interventions. Its purpose was to measure adoption of seed varieties by rice farmers. To minimize measurement error, we timed the survey to be right after planting. Our sample consists of 7,200 farmers. These farmers were drawn from a random sample of 261 gram panchayats — an administrative unit usually consisting of around 8 villages.Before drawing this sample, we excluded gram panchayats that had any village within 1.5 kilometers of the block boundary.We removed these areas to reduce any interference caused by farmers possibly obtaining seeds from other blocks. The 261 sample gram panchayats had 75,616 farmers registered for the DBT program for seed subsidies. Using this database as a sampling frame, we randomly drew 100 farmers from each block . These farmers are spread across 1,333 villages. Figure A1 shows their geographic dispersion across the 10 districts in the experiment. Survey teams succeeded in locating and surveying 6,653 of the farmers. Of these, 93 percent were currently cultivating rice. Table A1 shows no significant differences in the probabilities of being surveyed or growing rice between treatment and control groups. The survey focused on which seed varieties were currently being used for rice cultivation. Surveyors went through a list of 30 varieties and asked farmers which ones they were currently using and the amount of land being grown.In addition to these adoption data, we obtained information on contacts with agricultural extension agents during the last year, topics discussed during these conversations, whether the farmer had seen any seed demonstrations, and whether they had recently learned about Swarna-Sub1.