Working Paper: CEPR ID: DP18093
Authors: Chihsheng Hsieh; Lachlan Deer; Michael Koenig; Fernando Vegaredondo
Abstract: We study a dynamic model of collective action in which agents interact and learn through a co-evolving social network. Our approach highlights the importance of communication in this problem and conceives the social network – which is continuously evolving – as the structure through which agents not only interact but also communicate. We consider two alternative scenarios that differ only on how agents form their expectations: while in the “benchmark” context agents are completely informed, in the alternative one their expectations are formed through a combination of local observation and social learning à la DeGroot. We completely characterize the long-run behavior of the system in both cases and show that only in the latter scenario (arguably the most realistic) there is a significant long-run probability that agents eventually achieve collective action within a meaningful time scale. This, we argue, sheds light on the puzzle of how large populations can coordinate on globally desired outcomes. Finally, we illustrate the empirical potential of the model by showing that it can be efficiently estimated for the so-called Egyptian Arab Spring using large-scale cross-sectional data from Twitter.
Keywords: collective action; networks; riots; protests; Degroot; social learning
JEL Codes: D74; D72; D71; D83; C72
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| local information (R53) | successful collective action (D70) |
| local observation and social learning (Z13) | successful collective action (D70) |
| reduced linking costs (F12) | fraction of rioting agents (C69) |
| belief manipulation (D83) | rioting agents (D74) |
| local information (R53) | clusters of collective action (D70) |
| clusters of collective action (D70) | successful collective action (D70) |