Computational Perspectives on Individual and Collective Decision-Making

Abstract

Individual decisions determine the success of companies (Pepsi or Coke?) and the structure of our social networks (Alice or Bob?), while collective decisions determine the composition of our governments and the outcomes of criminal trials, among countless other facets of our lives. As such, understanding the factors that contribute to these decisions is crucial, both for predicting future decisions and for designing interventions. In this dissertation, we use computational techniques to address two core questions towards this end. First, can we learn about how people make choices from individual decision-making data? Second, how do we aggregate group preferences in collective decision-making and what are the consequences of different aggregation mechanisms? After a brief introduction in Part I, Part II describes several methods to learn the effects of social and contextual factors on preferences in individual discrete choice settings, synthesizing tools from interpretable machine learning, causal inference, and graph learning. In Part III, we turn to collective decisions, focusing on theoretically characterizing the behavior of two commonly used voting systems, plurality and instant runoff voting (IRV). In particular, we explore what happens under IRV when voters are forced to submit top-truncated preferences, prove that IRV favors moderate candidates in a way plurality does not, and examine the dynamics of candidate policies under a boundedly-rational imitative model. We conclude in Part IV with closing thoughts and directions for future work.