1: \begin{abstract}

2: %Disagreement among individuals in a society, even on central

3: %questions that have been debated for centuries, is the rule;

4: %agreement is the rare exception. How can disagreement of this sort

5: %persist for so long? Existing models of communication and learning,

6: %based on Bayesian or non-Bayesian updating mechanisms, typically

7: %lead to consensus provided that communication takes place over a

8: %strongly connected network.

9: %

10: %We analyze a stochastic model of communication combined

11: %with the assumption that there are some ``stubborn" agents in the

12: %economy who never change their opinions. We show that the presence

13: %of these stubborn agents leads to persistent disagreements among the

14: %rest of the society---because different individuals are within the

15: %``sphere of influence" of distinct stubborn agents and are

16: %influenced to varying degrees. Under general conditions, there is no

17: %convergence to a consensus. Instead, the expected cross-sectional

18: %distribution of beliefs in society converges (in distribution), and

19: %generally, the opinion of a single individual, and in fact that of

20: %the whole society, potentially fluctuates forever.

21: %Thus this model provides a new approach to understanding persistent opinion fluctuations and disagreements,

22: %and in the process, introduces new tools for the analysis of opinion formation and consensus models.

23: %

24: %We also show that in ``highly fluid'' social networks, where the product between  the fraction of edges incoming in the stubborn agent set times the mixing time of the associated random walk is small, even though there is no consensus, the expected stationary opinions and their variances concentrate, in the limit of large population size, around a certain value, so that one can think of a form of ``homogeneous influence'' of the stubborn agents on the rest of the society.

25: %\end{abstract}

26: