Statistics and Strategic Behavior in Algorithmically Mediated Systems

ABSTRACT

Algorithmic systems affect many aspects of our lives, influencing how we allocate time and money, how resources are distributed, and how information spreads. Understanding how these systems shape social and economic processes requires looking beyond the algorithms themselves to the context surrounding them and the interactions they mediate. In this thesis, we contribute to developing an understanding of these systems through two lenses: statistics and analyses of strategic behavior. On the statistical side, the thesis develops foundations and methods for settings in which algorithmic systems violate classical assumptions or create new design constraints: It shows how such systems can break standard assumptions in causal inference and develops methods for valid and efficient treatment effect estimation; it develops approaches for machine-learning models used in housing, employment, and credit decisions to satisfy legal and regulatory requirements; and it adapts statistical procedures for ranking generative models from pairwise preferences to make them robust to (near-)duplicates. On the strategic behavior side, the thesis analyzes how individuals and platforms respond to the incentives created by these systems. It studies how professional networking behavior balances connectivity against congestion, develops new explanations for how professional networks can amplify inequality and shows how link recommendations can be tuned to mitigate these effects while improving efficiency, and examines the role that social media content moderation policies play in creating and sustaining online communities. Taken together, these chapters provide new methods for measuring the effects of algorithmically mediated systems and new frameworks for understanding how their design shapes social and economic outcomes.

BIOGRAPHY

Chris Hays is a PhD candidate at the MIT Institute for Data, Systems and Society. He is interested in statistical foundations and analyses of strategic behavior in algorithmically mediated systems. He is advised by Manish Raghavan. His PhD is supported by an NDSEG fellowship, and he has won several awards including a Best Paper award at WWW.

COMMITTEE

Manish Raghavan (advisor), Dean Eckles (chair), Jon Kleinberg

EVENT INFORMATION

Hybrid event. To attend virtually, please contact the IDSS Academic Office (idss_academic_office@mit.edu) for connection information.