Can Network Theory-based Targeting Increase Technology Adoption?

How to cite

Beaman, Lori; Mobarak, Ahmed; BenYishay, Ariel; Magruder, Jeremy. 2018. "Can Network Theory-based Targeting Increase Technology Adoption?." AEA RCT Registry. AEARCTR-0002017. https://doi.org/10.1257/rct.2017-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Lori Beaman* , Ahmed Mobarak , Ariel BenYishay , Jeremy Magruder

* Primary investigator

Registry ID
AEARCTR-0002017
DOI
https://doi.org/10.1257/rct.2017-1.0
First registered
Last updated
Status
Completed
Trial dates
2010-12-01 – 2014-11-01
Countries
Malawi (Machinga, Mwanza, and Nkhotakota districts)
JEL codes
O16, O13
Registry record
http://www.socialscienceregistry.org/trials/2017

Abstract

In order to induce farmers to adopt a productive new agricultural technology, we apply simple and complex contagion diffusion models on rich social network data from 200 villages in Malawi to identify seed farmers to target and train on the new technology. A randomized controlled trial compares these theory-driven network targeting approaches to simpler strategies that either rely on a government extension worker or an easily measurable proxy for the social network (geographic distance between households) to identify seed farmers. Both reduced form and structural estimates suggest a learning environment in which most farmers need to learn about the technology from multiple people before they adopt themselves.

Intervention

The PIs conducted a randomized field experiment to evaluate the use of network-based diffusion theory to select optimal farmers to train in the use of pit planting, a new agricultural technology to the area, so that knowledge and adoption of it are best spread through communities. The Malawian Ministry of Agriculture's current method of introducing and spreading new technologies was for an extension agent to select a "seed" farmer to train in the technology, and later encouraging the farmer to discuss it with neighbors. The PIs used the status quo method as a benchmark to test three different methods of selecting seed farmers using network-based diffusion theory.

The PIs chose to test threshold diffusion models, in which individuals adopt a new technology if they are connected to a tipping-point number of other adopters. Different formulations of the model yield different predictions of the best seed farmer in a given social network. The PIs focus on two versions of the threshold model. The first, a "simple contagion" model, only requires that an individual be connected to one other adopter to be induced to adopt also. In a simple contagion learning environment, seed farmers with the fewest redundant connections would be the most effective. The second, a "complex contagion" model, requires connections to multiple adopters to induce an individual to adopt as well. In a complex contagion learning environment, some level of redundant connections is desirable so that a threshold number of connected adopters is met.

The field experiment included four treatment arms. Two hundred Malawian villages were randomly assigned to have two seed farmers selected according to a simple contagion model, a complex contagion model, a second complex contagion model where geographic proximity is used as a proxy for connectedness (the "geo" model) , and the benchmark method of relying on extension agents to choose seed farmers. Optimal seeds in the threshold model villages were chosen using a social network census completed prior to the intervention for every sample village. The seed farmers were then trained to use pit planting to cultivate their fields.

The PIs found that the theory-driven selection of seed farmers led to greater diffusion of pit planting than the benchmark method, especially on the extensive margin (whether anyone adopts at all). This indicates that simply changing 'who' is trained based on network theory can increase adoption of new technologies. In addition, while the 'geo' treatment produces some gains relative to the benchmark, geographic proximity is a poor proxy for social network connectedness. Lastly, estimates suggest that a majority of people require more than one connection to an adopter to themselves adopt, indicating a complex contagion learning environment.

Experimental design

The PIs used a linear threshold model as the theoretical basis for the seed farmer selection and technology dissemination strategy. In this model, an individual will adopt a new technology once they are connected to a tipping-point number of adopters. In order to identify the optimal seed farmers under each model, the researchers collected social network data during a household census of all the sample villages. Using this data, they created network adjacency matrices for the households in every village. Next, they conducted hundreds of simulations of the intervention to identify which pair of individuals would be optimal seed farmers under simple and complex contagion environments. To identify the optimal seeds in a simple contagion environment, the PIs conducted simulations with the threshold number of connected adopters set at one. To identify optimal seeds in a complex contagion environment, the PIs specified the threshold number of connections as two. The simulations were run for four periods. Afterwards, the PIs used the average village-level adoption rate for each pair of possible seed farmers to determine the optimal seeds for each village for both contagion models.

Villages were then randomly assigned to one of the four treatment arms. Seed farmers for villages in the simple contagion treatment arm were selected by their effectiveness in simulations assuming a simple contagion environment. Likewise, seed farmers for villages in the complex contagion and geographic treatment arms were selected by their effectiveness in simulations assuming a complex environment. Training was the same for all seed farmers; the experimental variation only altered the theoretical basis for choosing who to train. After training the seed farmers, the PIs followed up with three rounds of households surveys over the next three years. They collected data on farming techniques, input use, yields, assets, and other characteristics from a sample of 5,600 households.

Design details

Randomization method
Randomization performed in office by computer
Randomization unit
Village
Clusters
200 villages
Observations
5600 households
Arms
50 villages in simple contagion diffusion treatment group, 50 villages in complex contagion diffusion treatment group, 50 villages in geographic diffusion treatment group, 50 villages in benchmark group
Primary outcomes
Aware of Pit Planting, Conversations with other farmers about Pit Planting, Pit Planting Adoption Rate, Pit Planting Number of Adopters

Papers from this trial

  1. L. Beaman, A. BenYishay, J. Magruder and A.M. Mobarak, "Can Network Theory based Targeting Increase Technology Adoption?" Mimeo. http://faculty.som.yale.edu/mushfiqmobarak/papers/MNW.pdf

Cited by

No citation data available for this registration yet.

Keywords