Testing Models of Social Learning on Networks: Evidence from a Lab Experiment in the Field

Arun Chandrasekhar
Horacio Larreguy
Juan Pablo Xandri
Publication Type: 
Working Papers
Journal Name: 
MIT Department of Economics Working Paper
Publication Year: 
2011

Studying the diffusion of information is particularly important in developing nations. Due to the lack of formal institutions, many markets, and information aggregating mechanisms, agents in developing economies often rely on social networks for information. Models of social learning provide the environment in which various economic processes take place. Before employing these models and drawing policy recommendations from them, we must first understand which models best describe features of empirical social learning.

We consider the two leading models of social learning studied in the network literature: Bayesian learning and rules of thumb (DeGroot). Bayesian and DeGroot models often yield differing behavior. For instance, individuals employing rules of thumb often double-count information and are more prone to make poorer decisions in the long run. We conduct a unique lab experiment in 22 villages in rural Karnataka to test which models best describe the social learning process. We study games in which six individuals are placed into a social network, each having knowledge of its structure, and attempt to learn the underlying state of the world which is either 0 or 1. Individuals receive independent, identically distributed signals about the state of the world in the first period only; thereafter, they are privy to the historical guesses of their network neighbors and based on this, update their guess about the state of the world. That is, individuals make guesses about the underlying state of the world and these guesses are transmitted to their neighbors at the beginning of the following round. Having considered various environments including complete information and incomplete information Bayesian models, we provide both structural and reduced form evidence showing that individuals are best described by particular rules of thumb wherein they either take simple majority of opinions in their neighborhood or put more weight on more popular friends. Moreover, agents double-count information that they have already had access to, and they overweight information if they receive it from multiple channels.

JEL Codes: 
D82, D83, D85, C92, C93
Region: 
South and Central Asia
Country: 
India
Topic: 
Experiments
Topic: 
Networks