Sharing Models to Interpret Data

By: Joshua Schwartzstein and Adi Sunderam

Format:Print
| Language:English
| Pages:61

Abstract

To understand new data, we share models or interpretations with others. This paper studies such exchanges of models in a community. The key assumption is that people adopt the interpretation in their community that best explains the data, given their prior beliefs. An implication is that interpretations evolve within communities to better fit prior knowledge, potentially making final reactions less accurate than initial reactions. When people entertain a rich set of possible interpretations, social learning often mutes reactions to data: the exchange of models leaves beliefs closer to priors than they were before, untethering beliefs from data. Our results shed light on why disagreements persist as new information arrives, why ideological bubbles need not be hermetically sealed, and the structure of meetings in organizations.