Dynamic Choice of Information Sources

Abstract

I characterize the optimal learning strategy when there are two information sources and learning is modeled as a search process. There are four possible states of the world, indicating whether search in each direction could be successful or not. No matter what the set of actions are (the agents has to choose from these actions when he stops learning) and no matter what the payoffs from these actions are in different states, an optimal strategy always consists of two phases. During the first phase, one chooses the source with the highest index. Source i index is equal to P(theta_i=1)+P(theta_1=theta_2=1)u_{11}(a_i), where P(theta_i=1) is the probability that the search in direction i could be successful, P(theta_1=theta_2=1) is the probability that the search in both directions could be successful, u_{11}(.) is the payoff in the state when the search in both directions could be successful, and a_i is the action taken after the successful search in direction i. During the second phase, one never switches between the sources. Since the index does not depend on the payoff in the state theta_1=theta_2=0, I show (1) how the availability of information sources resolves disagreement between groups with opposed interests, and (2) why social welfare consequences of publication bias might not be as severe as one might think.

Keywords: optimal learning, sequential optimal experimental design, Poisson process, limited attention, disagreement, publication bias

JEL Classification: D83