Information and Diffusion in Networks

Paper Session

Friday, Jan. 6, 2017 10:15 AM – 12:15 PM

Hyatt Regency Chicago, McCormick
Hosted By: Econometric Society
  • Chair: Nicole Immorlica, Microsoft Research

How Heterogeneous Timing of Participation Across People Improves Diffusion

Mohammad Akbarpour
University of Chicago
Matthew O. Jackson
Stanford University


Recent studies have shown that non-Poisson (``bursty'') behaviors in human interactions can impede the diffusion of information or infectious diseases in social networks. Those studies generally consider models in which nodes are independently active according to the same random timing process, and vary that timing. In reality, people differ widely in the patterns of their activity. In this paper, we develop a simple model of diffusion on networks in which agents can differ in the autocorrelation of their activity patterns. We show that bursty behavior does not always hurt the diffusion, and depending on the features of the environment, having some (but not all) of the population being bursty significantly helps diffusion. Moreover, we prove that in a variety of settings maximizing diffusion requires heterogeneous activity patterns across agents and does not involve any Poisson behavior.

Diffusion Games

Evan Sadler
Harvard University


I study how the structure of a social network affects the diffusion of a new product or technology. The model deals explicitly with the network's discrete structure, in contrast with most extant work that uses a mean-field approximation. My findings highlight important qualitative differences in predicted diffusion patterns: long-run outcomes are stochastic, individuals can remain isolated, and the likelihood of a large cascade is sensitive to early adoption patterns. The analysis requires technical advances that leverage recent mathematical work on random graphs. A key contribution is a set of structural results for a large class of random graph models that can exhibit observed features of real networks---features like homophily and clustering. These results allow us to characterize the extent and rate of diffusion as a function of network structure.

Reaching Consensus via non-Bayesian Asynchronous Learning in Social Networks

Nicole Immorlica
Microsoft Research


Information is the quintessential example of a replicable good: it can be simultaneously ``consumed'' and sold to others. We study a decentralized market where sellers and prospective buyers of information can negotiate over its price, and the buyers of information may resell it. We study how the potential for resale influences the pricing of information, and the incentives to acquire information when trading frictions are small. We prove that in a no-delay equilibrium, all prices converge to 0, even if the initial seller is an informational monopolist. The seller-optimal equilibrium features delay: the seller is able to sell information at a strictly positive price to a single buyer, but once two players possess information, prices converge to 0. The inability to capture much of the social surplus from selling information results in sellers underinvesting in their technology to acquire information. By contrast, a ``patent policy'' that permits an informed seller to be the sole seller of information leads to overinvestment in information acquisition. Socially efficient information acquisition emerges with patents that have a limited duration.

Learning in Local Networks

Wei Li
University of British Columbia
Xu Tan
University of Washington


Agents in a social network learn about the true state of the world over time from their own signals and reports from immediate neighbors. Each agent only knows her local network, consisting of her neighbors and any connections among them. In each period, every agent updates her own estimates about the state distribution based on her perceived new information. She also forms estimates about each neighbor’s estimates given the new information she thinks the neighbor has received. Whenever a neighbor’s report differs from the agent’s estimates of his estimates, the agent attributes the difference to new information. The agents form the correct Bayesian posterior beliefs in any network if their information structures are partitional. They can also do so for more general information structures if the network is a social quilt, a tree-like union of completely connected subgroups. Under this procedure, the agents make fewer mistakes than under myopic learning; and they learn correctly if the network is common knowledge.
JEL Classifications
  • A1 - General Economics