Skip to main content
eScholarship
Open Access Publications from the University of California

UC Irvine

UC Irvine Electronic Theses and Dissertations bannerUC Irvine

Heterogeneity in Learning

  • Author(s): Williams, Cole Randall
  • Advisor(s): Carvalho, Jean-Paul
  • et al.
Abstract

This dissertation contributes to the understanding of heterogeneity in rational learning.

The first chapter proceeds from the observation that, in an environment of social learning under unobserved heterogeneity, those with opinions closest to one's own may in fact be the most informative for oneself. For example, a similar opinion of a restaurant may suggest similar tastes and similar political views may suggest a similarity in values or interpretations of evidence. In this environment, individuals will display rational forms of confirmation bias and other ostensibly anomalous patterns of behavior.

The second chapter is a research collaboration with Aydin Mohseni studying learning by agents in social networks. Each agent has a preference for both accuracy (choosing the ``correct" action) and conformity (selecting the action taken by the majority of her neighbors) with the relative weight placed on these two concerns being heterogeneous between agents. Related literature finds the star network to possess optimality properties. In contrast, our analysis finds that agents in highly centralized networks, such as star networks, take longer to settle on the optimal action than agents in other standard networks.

To combat the replication crisis in science, a group of prominent scholars has proposed {\it redefining statistical significance} by reducing the p-value significance threshold from 0.05 to 0.005. The third chapter shows that, if researchers can exercise their ``degrees of freedom" to obtain significance and if they are heterogeneous, then this proposal may exacerbate the problems with reproducibility. I provide an example demonstrating that even a small amount of researcher bias can produce this effect and give a general characterization of the conditions when it will occur.

Main Content
Current View