1: \begin{abstract}

2: 

3:   We consider a Bayesian game of pure informational externalities, in

4:   which a group of agents learn a binary state of the world from

5:   conditionally independent private signals, by repeatedly observing

6:   the actions of their neighbors in a social network.

7: 

8:   We show that the question of whether or not the agents learn the

9:   state of the world depends on the topology of the social network. In

10:   particular, we identify a geometric ``egalitarianism'' condition on

11:   the social network graph that guarantees learning in infinite

12:   networks, or learning with high probability in large finite

13:   networks, in any equilibrium of the game. We give examples of

14:   non-egalitarian networks with equilibria in which learning fails.

15:   

16:   \noindent{\bf Keywords:} Bayesian learning, rational expectations,

17:   informational externalities, social networks, aggregation of

18:   information.

19: 

20: \end{abstract}

21: