The role of mechanistic explanations in technology adoption across contexts: Evidence from Uganda

We pursue a lab-in-the-field RCT design with our partner organization, Agriworks, who provide an on-demand irrigation service in Eastern Uganda. We invite tomato farmers irrigating with Agriworks, and who grow in different soils, to a fertilizer demonstration predetermined soil context. In collaboration with extension officers, who will present the information accompanying the demonstration, we randomize the provision of mechanistic explanations to causally identify whether, and how much, mechanistic explanations help farmers translate the demonstration information to their own context.

2023-12-13

2024-04-30

Fertilizer knowledge and practices for each soil type, Farmer beliefs about returns to fertilizer on soil types, WTP for fertilizer information, WTP for fertilizer application on their plot

We will use both incentivized and unincentivized elicitations of farmers' beliefs about agricultural production, in particular their mental models of fertilizers and agriculture. The incentivized measures will be obtained from giving farmers a small endowment, which they can use to pay to reveal advice from agronomist-simulated vignettes. In addition to structured survey responses, we will also collect unstructured qualitative data, for example free responses transcribing farmers' mechanistic descriptions of why a particular (simulated) agronomic result occurred.

Heterogeneity by soil type, demographic characteristics (education, wealth), comprehension of demonstration

We will recruit 200 tomato farmers from the smallholder region of Mbale in Eastern Uganda, evenly distributed between farmers who grow in two soil contexts: ``black soils" and "grey soils". These soil contexts are archetypes familiar to farmers, and correspond to a particular ensemble of textural, mineral, organic matter, and drainage properties that impacts fertilizer results. We collaborated with agricultural scientists at the International Institute for Tropical Agriculture (IITA) to identify and develop a vocabulary around the most common soil archetypes in the region. The eligible pool of farmers are those who are registered with Agriworks and have hired their services at least once in the prior year. Agriworks possesses approximate knowledge of registered farmers plot locations and soil types. We will randomly invite farmers from the eligible pool stratified by soil type. We will confirm soil types by phone when we invite farmers to attend demonstrations to achieve a balanced distribution of soil contexts.

Farmers will be randomly invited to attend two treatment arms: 1. Mechanistic demonstration, 2. Non-mechanistic demonstration. One-half of eligible farmers will be invited to a demonstration that includes mechanistic explanations (arm 1), and one-half will be invited to a demonstration that includes a training on fertilizers that is void of mechanistic explanations and only provides a recipe for fertilizer application without information on *why* and *how* fertilizers work (arm 2).

By stratifying on soil type, half of farmers will attend a demo where tomatoes are grown in a soil context that matches the soil type of their own plot (“matched farmers”). The other half will attend a demo where soil type does not match their context (“mismatched farmers”). The goal is to set up a situation where mismatched farmers will have to translate the demonstration information to their context in order to use it. Our logic is that matched farmers can more easily copy the fertilizer ‘recipe’ shared during demos. By contrast, the mismatched farmers will have to learn from the mechanistic explanations to translate the fertilizer recipe catered to soil type A to one that fits their soil type B.

Of 200 farmers invited to demonstrations, we expect that 75% will come to the first demonstration. We will then administer the baseline, the first of three surveys that we formulate and administer with the help of the experienced local research firm Matrice360. This baseline elicits basic demographic information, social networks, farming experience, current and past practices, plot characteristics, and risk aversion. The endline survey will take immediately after the demonstration and we are planning a short phone survey to ask which practices they implemented during the prior season.

Randomization in office by a computer

Individual

No

18 villages

126 farmers

63 farmers in arm 1, 63 farmers in arm 2