In the 1990s, pineapple for export began to replace cassava and maize as a principal crop in parts of Ghana. Farmers did not initially know the optimal fertilizer/labor combination. How do they learn this? Conley and Udry collect data on the information networks of agents (by directly asking “who do you ask for advice about your farm?”) in three villages. Over a period of two years, they note when new pineapple crops are planted and how much fertilizer is used. Positive shocks in growth of friends who use high amounts of fertilizer is associated with increases of fertilizer use in future periods of roughly the magnitude expected by positive own-farm shocks in a Bayesian learning-by-doing model. This model controls for spatial-temporal location, since otherwise “learning effects” may simply reflect nearby farmers reacting to the same shock to their local soil. The pineapple data also show that more experienced farmers change their fertilizer use less after shocks than less experienced farmers. By applying the same analysis to growers of cassava-maize, the principal crop since the 1930s in the area, people rarely update their fertilizer use to shocks among friends, which is consistent with the idea that people learn in a Bayesian fashion from their friends. Also, without the spatial-temporal controls, it *does* appear that people learn from their friends about cassava-maize, which again is simply a spurious inference resulting from nearby farmers reaction to geographic shocks, not to learning.
This paper is interesting in and of itself, but also interesting in showing how little we know about the spread of information, the critical issue of global growth. Surely some agent – a pineapple export company, perhaps – knows precisely the optimal amount of fertilizer for each farm to use. Why is he not listened to? How does trust play into learning? Particularly with rare events, can it be that learning will not converge to the optimal innovation in some regions but will in others? These issues seem critical to me, and are wide open to future study.
http://www.econ.yale.edu/~udry/pdf/july2005a.pdf (link to WP – final version in AER 2010.1. Also, this is a great example of the ridiculousness of economics publishing. Working papers, essentially identical to the final version, have been around for 10 years. This paper has been cited hundreds of times. Yet it is only appearing in a peer-reviewed journal now. No other scientific field delays publication to such a ridiculous extent.)