Female Social Networks and Farmer Training: Can Randomized Information Exchange Improve Outcomes?
American Journal of Agricultural Economics, 95(2):376-383
As Good as the Networks They Keep? Social Networks and Information Exchange in Rural Uganda
With Ken Leonard
Conditionally Accepted, Economic Development and Cultural Change
We examine the role of social networks in technology adoption for female farmers in rural Uganda growing a cash crop with which they have little previous experience. We devised a social networking intervention (SNI), randomized at the village level which introduced new agricultural information via randomly assigned network links. Difference in difference estimates of the intervention show that the new links and information significantly increased the productivity of all farmers in the treatment villages except those who were already in the highest quartile of productivity (Vasilaky, AJAE 2013). The SNI intervention has its strongest impact on females’ production, but also spilled over to males’ yields. We find that these effects are comparable to the effects of a concurrently-run conventional agricultural training program. Examining the mechanisms by which the program was successful, we focus on the importance of adding weak links to women’s networks: links between people who do not know each other well but which bring new information to existing networks. [This work was funded by Markus Goldstein, at the World Bank.]
Coverage on this research is in:
- World Bank Policy Brief, Gender and Innovation Lab, Brief, and Report.
- Social Networks in Developing Countries, Annual Review of Resource Economics, Vol. 7: 451-472
- World Bank’s World Development Report 2015: Mind, Society, and Behavior.
- Blog post on this work at the World Bank Development Impact
- World Bank’s coverage of 2015 Oxford CSAE conference
Measuring the Impact of Educational Insurance Games on Weather Insurance Demand, With Rahel Diro, Michael Norton, Geoff McCarney, and Daniel Osgood.
There is increasing concern that voluntary index insurance will fail to scale with sustainable impacts unless it can transition beyond the premium subsidy and liquidity programs that would be prohibitively expensive to donors as projects grow. Because farmer’s index insurance demand choices are typically studied in the non-market environment of large subsidies, mandatory purchase requirements or research oriented, but non-commercially viable insurance interventions, there is little known about developing farmer behavior in the market-priced, more commercially focused context necessary to understand for commercial scaling. We therefore study the demand for index insurance in the context of an unsubsidized, non-loan linked commercial insurance program in Ethiopia that has scaled to tens of thousands of farms through a mix of labor based liquidity programs and cash only sales. Using a randomly administered educational game, we estimate that the total impact of offerring financial education increased the likelihood of purchase by 11% across our study sample, and increase the amount purchased by 34%, on average. [This work was supported by NSF-SES0345840, NSF-SES0957516, with the Center for Research on Environmental Decisions (CRED) and United States Agency for International Development (USAID) AID-0AA-A-1-00011.] Revise and Resubmit Journal of Development Studies.
Learning versus transmission: Competitive or Team Incentives
With Asif Islam
This study explores the behavioral aspect behind learning by quantifying the amount of information learned by smallholder female farmers in Uganda under different incentive schemes (Vasilaky, AJAE 2013). The paper shows how competitive versus team incentives compare in motivating Ugandan farmers to learn and share information relevant to adopting a new agricultural technology. The study finds that tournament-based incentives provide greater outcomes in terms of information learned overall than threshold-based team incentives. Furthermore, order matters. More information is learned when a round of tournament incentives is followed by a round of team incentives than when a round of team incentives is followed by a round of tournament incentives. In addition, new information introduced every round was learned by more individuals under team-based incentives than tournament-based incentives. The study provides direct practical policy recommendations for training farmers in the context of Uganda.[This work was funded by Markus Goldstein, at the World Bank.] Under Review.
Mothers, Daughters and Sisters: Is Competitiveness Socialized?
With Ken Leonard, Magda Tsaneva, and Jeff Flory
Using experiments adopted from Niederle and Vesterlund (2007) we look at the turning point of competitive differentials between boys and girls across and matri- and patri- linear societies. We look at the possible cultural determinants along a female’s lifecycle that may affect the differences in competitiveness that we observe between the two types of societies, where females are significantly less competitive in patri- societies than males on average, but change of their life-course, while we observe no gender gaps in competition in matri- societies. [This work was supported by the NSF Grant, SES 0922460.] Field video. Under Review.
Formal Insurance and Real Ties: Randomizing Social Distance in Group Insurance
With Sofia Martinez, Radost Stanimirova and Daniel Osgood.
We look at how existing informal social networks interact with the introduction of formal index insurance, as well as the periods (seasons of the year) that groups versus individuals choose to insure. Our design elicits preferences for purchasing formal index insurance as a group versus as an individual, where groups in our game are formed based on existing network ties between participants, and participants are randomly selected for the group insurance versus individual insurance offering. This setup enables us to test whether offering weather index insurance at the group level versus the individual level increases take-up on the extensive and intensive margins. We also test whether take-up under group insurance varies based on the existing connections (social distance) between group members. To identify social distance effects, we randomly vary the distance between individuals in assigned groups by using participants’ privately recorded knowledge of other participants’ assets. By allowing participants to choose among the periods (seasons of the year) they’d like to insure, we can also observe how demand for insurance varies for individuals and group purchasers. Under Review.
Followup research in Honduras can be seen in this video documentary of our work.
An Iterative Approach to Inverse Problems Using Python’s Numpy
Inverse problems commonly occur in statistical inference, and any field where one wants to compute some interior properties using exterior measurements. Linear regression would be one such problem which solves the least squares problem (where the solution to the least squares problem requires the inversion of a matrix X in order to solve for the optimal beta parameter estimates, b). In many such problems (Ax=b) the data or matrix X is ill-conditioned, meaning the matrix is is close to singular (e.g. a Hilbert matrix that has several close to dependent columns),and the solution, is therefore is noisy. A small perturbation in b will lead to wild adjustments in the solution, (x=A/b), as is well-known.
One way to deal with such instability is to regularize the problem and solve a “nearby” problem using Lasso methods or the more general Tikhanov reguarlization problem, which add a penalty, lambda, to the optimization problem. However, currently, no efficient method exists for searching for lambda. In this method, I have derived an iterative approach to solving the general Tikhanov regularization problem, which converges to the noiseless solution, does not depend strongly on the choice of lambda, and yet still avoids the inversion problem.
My interest in the inverse problem was motivated by the fact that least squares solution does not give a reasonable result when the data matrix is singular or ill-conditioned. For example, the pseudo inverse solution to the least squares problem produces a worthless result when the data matrix is singular or ill-conditioned, due to the fact that small perturbations of the data produce large deviations in the solutions. Yet, researchers often accept the estimated coefficients that the pseudo inverse returns from linear regression, without checking for the conditionality of their data. Inverse problem are becoming more relevant in economics due to the fact that many non-parametric methods are gaining ground, which often result in an inverse problem [1,2].
Test cases show that the approach is either better or significantly better than existing methods. I have posted the algorithms written in Python, but not the derivations since it is still a work in progress. This alogirithm is written in a few lines of code using Python’s numpy package, and primarily relies on the SVD composition. Package at pypi or git.
Paper upon request. Pygotham Slides and video. Pydata talk and slides.
Traversing the Landscape of Experimental Power
With Michelle Brock
This research gives an overview of the use of power calculations in experimental economics; the reasons for using and reporting power calculations, the methods of computing them, the parameters to be considered, and a comparison to other fields.
Hierarchal Bayesian Estimation of Weather Index Insurance Model
With Kenny Shirley, Dan Osgood, Helen Gretrex.
This paper models two sources of variation that enter into the weather indices and ultimately pricing of weather index insurance: both the site specific data source and the tool use to measure the rainfall (satellite or otherwise). Using a hierarchical Bayesian model with one level for locations, and another level for multiple data sources within a location we capture each of these two sources of variation. Bayes Draft. and Poster.
Evaluating the Timeliness to Disaster Response in the Philippines, Co-Principal Investigator, With Aurelie Harou, Chia-Ying Lee and Daniel Osgood
With a grant from the World Bank-GFDRR Disaster Risk Financing and Insurance (DRFI) Program under the umbrella of DFID’s Sovereign Disaster Risk Financing and Insurance Impact Appraisal Project we are studying the impacts of timely disaster response in the Philippines. Using several datasets we document results of the the effects of budgetary spending on economic outcomes in the Philippines post disasters. We use an instrumental variables approach to identify the effects of government spending on labor outcomes, with a vector of extreme weather rainfall and typhoon indices to identify the effects caused by exogenous variance in government spending.
Video Based Learning, Self Efficacy and Perceptions in Training Farmers via Digital Green, Co-Principal Investigator
With Dean Karlan and Kentaro Toyama and JPAL SA
We’re working with Digital Green (DG), a non-profit headquartered in Delhi, which specializes in a specific video-based extension methodology, recently won Google’s Global Impact Award. DG involves local production of videos for agricultural technologies and techniques; and group-based instruction that uses the videos as a base for mediated instruction. Think Kahn Academy for rural areas, but with personal mediation, where the mediator and protagonist belong and mirror their students. A small-scale controlled trial of DG suggested that it is 10 times more cost-effective than classical “training and visit” style agriculture extension.
We are studying both the effectiveness of this teaching methodology with a controlled study, but aslo various other questions: whether self efficacy plays a role in self learning (do I believe I can learn this on my own?), whether external risk affects take-up of the information and new technology, and whether social distance between the video protagonist and viewer affects take-up and learning. It’s an exciting study with a great NGO that studies self-learning beyond the agricultural setting. For more information see Social Registry Trial 313.
Adopting Water Conservation, Principal Investigator
With Katherine Alfredo, Aurélie Harou, Manu Lall
In conjunction Columbia University’s Water Center, I am leading a study on the economic and behavioral incentives behind water conservation in high value crops in India. Using a 2 armed RCT, we are looking at how various factors impact water: electricity availability, irrigation technology, or group incentives (a la Opower). This is the first study, to our knowledge, that is also metering actual pump use over the season, sensoring water levels, and collecting water samples. Irrigation messages as well as group incentives are being sent via SMS using Access Mobile’s SMS platform. The baseline took place May-June 2015. Field Notes on Water Conservation in Kurukshetra can be viewed here. [This research is supported by Columbia University’s Cross Cutting Initiative Grant.] See this blog post at State of the Planet.
Mobile SMS for Weather Insurance, Principal Investigator/Software Developer
With Sara Jayne Terp (Ushahidi) and IRI. We are developing the software for a SMS based platform that collects data on farmers’ weather outcomes as well as allows for remote insurance registration. The former database will assist in cross referencing satellite data with on-the-ground weather observations, helping to miniize farmers’ exposure to basis risk in insurance contracts. The pilot will also test the degree to which small groups of individuals can jointly purchase insurance on one cell phone. This would enable farmers who do not own a mobile to still register for micro insurance. A small beta test will help us understand how tractable this is. Group purcahsing has would improve scalability and reduce costs. We are building this product with Flask and Twilio in Python with some support from Twilio.org. [This work is funded by Columbia University’s Earth Clinic Seed Funding.]
Applications for interning on this project can be found here.