Credibility Learning, Social Network, and Political Belief Formation

• In Event: Trust in News: Comparative Perspectives

Abstract

How does a citizen evaluate the credibility of information sources? How does credibility assessment affect her choice of information outlets? How does the outlet choice influence citizens' political belief formation? This study attempts to answer these three questions with a computational model of social learning. Specifically, my model combines existing political learning models with the reinforcement learning approach from machine learning literature. Reinforcement learning illustrates how an agent finds the optimal decision throughout repeated experiences. This mechanism can be applied to a scenario where people simultaneously gather political information and update their political beliefs. In this study, it is assumed that agents evaluate an information source's credibility based on its messages' bias and (or) the consistency of messages. Also, agents repeat sampling messages to learn both source credibility and the state of the world. I begin with endogenizing the sampling rule reflecting an agent's credibility learning result: The number of samples drawn from a source depends on how much the agent trusts the source over the others. Eventually, this paper delineates, depending on the choice of sampling rules, 1) how the information consumption diet evolves and 2) how the distribution of agents' beliefs about the world changes. To illustrate this mechanism, I employ agent-based modeling. The agent-based simulations were implemented in environments where 1) there is only dyadic communication between an information source and an agent and 2) agents can interact with both information sources and their neighboring agents in a social network.

Author

• Minsu Jang, University of Milano-Bicocca