Finance & PSD Impact SEPTEMBER 2020 The Lessons from DECFP Impact Evaluations ISSUE 56 Our latest note tests how different ways of providing best agricultural practices to maize farmers in Mexico affects adoption during and after the intervention. Empowering Farmers to Adopt Agricultural Recommendations Carolina Corral, Xavier Giné, Aprajit Mahajan and Enrique Seira A quick glance at agricultural input use data package of agricultural extension services from developing countries reveals a large designed to help them implement the dispersion in take up of improved inputs and recommendations. In addition, three of the four practices across farms. One explanation is that treatment arms were offered an in-kind grant of this is a problem resulting from limited 2,000 pesos (roughly 150US$ at the time of the information, credit constraints, risk, poor input intervention): quality, and/or behavioral biases. An alternative • T1 received individualized or plot level view is that farmers are in fact making optimal soil analyses with input adoption decisions, and differences among recommendations and an inflexible in- farmers instead reflect heterogeneity in a fixed kind grant. factor such as soil quality. • T2 was the same as T1 except that the We test whether rainfed farmers in Tlaxcala, soil analysis and recommendations were Mexico adopt tailored recommendations based averaged in the cluster. Comparing T1 on soil analyses and whether productivity with T2 allows us to estimate the effect improves as a result. We vary the level of of varying the level of the specificity of information specificity (whether the soil analysis and recommendations. recommendations are based on the farmer's own • T3 differed from T2 in that the in-kind plot or on a larger geographical area) because grant was flexible. A comparison individually tailored information may be more between T2 and T3 thus measures the effective but is also more expensive. We also effect of autonomy, i.e. the ability to vary whether farmers can choose what inputs to choose the inputs purchased. purchase by offering an inflexible grant that • T4 provides the same recommendations subsidizes only the recommended inputs or a as T2 and T3, except that no grant was flexible one that gives farmers the choice of provided. Comparisons between T4 and which inputs to purchase. T2 (or T3) measure the effect of providing the in-kind grant. 1. Experiment • A control group C did not receive any The experiment consists of 678 farmers divided interventions during the experiment, but into a control group and four treatment arms that instead received soil analyses and combine soil analysis and recommendations recommendations (without extension (either at the plot or cluster level) with flexible services) the year after the intervention or inflexible in-kind grants. Farmers in all ended. Comparisons between C and T4 treatment arms were offered a soil analysis estimate the effect of providing localized report, a set of input recommendations and a soil analyses, recommendations and approximately 12-17%. The increase in profits extension services. is more muted and not statistically significant at conventional levels. Since neither specificity nor 2. Results autonomy affected yields or profits, during the We document substantial heterogeneity in soil intervention there appears to be no significant quality, mostly within (rather than between) downside to providing farmers with autonomy. clusters. This heterogeneity implies a Finally, we examine the persistence of the corresponding variation in the optimal mix of recommended practices in 2017, two seasons fertilizers. after the intervention ended. Farmers in T4 We then examine adoption using a standardized adopted 0.39 s.d. more practices in 2017 relative index of new agricultural practices introduced to control farmers. Since control farmers by the intervention. By this metric, farmers in T4 received recommendations in 2016, this adopted 0.33 s.d. more practices relative to suggests complementarities between extension control farmers. Farmers that received the in- services and recommendations, consistent with kind grant adopted considerably more work in other contexts. practices (ranging from 1.68 to 1.96 s.d. More interestingly, farmers who had received depending on the arm) underscoring the the flexible grant, were substantially more likely importance of the in-kind grant. Since the value to persist with the new practices in 2017 relative of the grant was designed to be roughly equal to to farmers with the inflexible grant (an increase the amount spent on fertilizer by control of 0.55 s.d.) and to farmers in the control group farmers, we view the grant as primarily (an increase of 1.08 s.d.). The fact that some encouraging experimentation rather than farmers continued to use the new practices two relaxing liquidity constraints. Surprisingly, T3 years after the intervention suggests that they farmers, who could ignore the recommendations were perceived as valuable. at no cost, adopted these new practices at the same rate as T2 farmers. Finally, we find no 3. Policy Recommendations evidence for specificity since T1 farmers did not These results have the following relevant increase adoption of new practices relative to implications: farmers in T2. 1) Cluster recommendations are as effective as We also examine the impact of the treatments on the more expensive plot-level ones, even in productivity and find results similar to those on a context with substantial heterogeneity. adoption. Average yields and profits for farmers 2) Recommendations should be paired with in T4 are not statistically different from those in extension services to help implement them. the control group but despite a drought, yields 3) Program beneficiaries should be given for farmers that received a grant were 0.2-0.4 autonomy, particularly if the program is top- tons/hectare higher relative to those for control down and involves expert advice. farmers, corresponding to an increase of For further reading see: Corral, C. X. Giné, A. Mahajan and E. Seira “Autonomy and Specificity in Agricultural Technology Adoption�, World Bank Policy Research Working Paper 9367, August 2020. Recent impact notes are available on our website: https://www.worldbank.org/en/research/brief/finance-and- private-sector-impact-evaluation-policy-notes