Evaluating strategic decision-making with iterative voting
Loading...
Authors
Kavner, Joshua
Issue Date
2024-05
Type
Electronic thesis
Thesis
Thesis
Language
en_US
Keywords
Computer science
Alternative Title
Abstract
Social choice theory is prolific with paradoxes and impossibilities that prevent cogent justifications for decisions made by groups of people. For example, Gibbard and Satterthwaite proved that no reasonable voting procedure exists that is non-dictatorial and immune to agents misrepresenting their preferences. Significant work has sought to overcome this impossibility by either restricting the domain of agents' preferences or dissuading this form of strategic behavior through computational hardness. The recent approach of iterative voting (IV), rather, aims to characterize the complex interactions ensuing from agents reporting their preferences strategically. In particular, agents may update their votes, given information about other agents' reports, prior to finalizing the group decision. Prior work has documented properties about IV equilibrium and conditions for convergence according to various social choice rules, information agents have access to, and agents' behavioral schemes. Still, only preliminary work has studied the effect IV has on social welfare of equilibrium outcomes relative to the truthful vote. This thesis advances our understanding of strategic behavior in social choice, via IV, on two fronts. First, we study the effect iterative plurality has on the social welfare of the chosen outcome with respect to the worst-case preference profile and as agents have arbitrary rank-based utility. To overcome a poor worst-case result, we study expected performance when agents' preferences are independent and identically distributed according to the impartial culture. Our finding surprises us in that IV helps agents choose higher quality alternatives on average, regardless of the order of their strategic manipulations. We go on to characterize certain classes of preference distributions for which IV improves or degrades social welfare, thus helping to explain why prior experiments attained varying results. Second, we generalize iterative plurality to multiple issues while agents have uncertainty about each alternative's score. In this setting, we identify sufficient conditions for convergence, including O-legal preferences and a novel model about what information agents have access to. Our study through both fronts characterizes agents' behavior given the opportunity to deliberate their votes. Our results provide insight for mechanism designers choosing whether or not to encourage such deliberation, and call for further study of non-incentive compatible mechanisms.
Description
May 2024
School of Science
School of Science
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY