ICES: Small: Collaborative Research: Robust Preference Aggregation

Grants and Contracts Details

Description

This project explores multi-disciplinary synergistic research at the interface between Behavioral Economics, Computer Science, Mathematics, Political Science, and Psychology. The senior personnel are three Computer Scientists (Judy Goldsmith and Nicholas Mattei, University of Kentucky, and Francesca Rossi, University of Padova, Italy) and a Psychologist (Michel Regenwetter, University of Illinois at Urbana-Champaign). All four scholars have extensive prior experience in interdisciplinary research. The team proposes to link its different perspectives on consensus methods into a novel and more comprehensive analysis of robust aggregation. The project places particular emphasis on the CP-net representation of conditional preferences, and the role of CP-nets in human decision-making behavior. The project consists of three major themes, each with two specific aims. • INDIVIDUAL AND COLLECTIVE CHOICE VIA CP-NETS - The project considers, through laboratory experiments, probabilistic modeling and quantitative statistical data analysis, whether people exhibit preferences in a form that can be modeled by CP-nets. - The team extends ideas from aggregation in Bayesian networks (models of conditional probabilities) to aggregation in CP-nets (models of conditional preferences). The projects investigates measures of closeness of CP-nets; algorithms for aggregating CP-nets, and systems for reaching and supporting approximate consensus, based on individual preferences in the form of CP-nets. • ROBUST AGGREGATION - The team investigates, for rating, ranking, or partial ranking ballots, how statistical inference from noisy data relates to or interacts with strategic manipulability and bribery. Empirically, they consider how each perspective impacts behavioral social choice analyses of, e.g., twelve American Psychological Association election data sets with tens of thousands of voters for five candidates. - The project extends computational social choice on voting systems, and the manipulation thereof, from preferences expressed as ratings, rankings, or subsets to preferences expressed as CP-nets. • BEHAVIORAL SOCIAL CHOICE CASE STUDY ON A LARGE NETFLIX DATA SET - The team uses a publicly released Netflix movie rating database with millions of ratings of tens of thousands of movies by hundreds of thousands of viewers to compare the outcomes under competing aggregation methods using real data sets of various sizes extracted from the database. - The team uses the same Netflix challenge database to investigate and empirically characterize voting scenarios that are especially susceptible or especially resilient to strategic manipulation, as a function of the number of voters and the number of choice options. Intellectual Merit of the Proposed Activity. This project creates new, innovative knowledge by inter-connecting state of the art approaches to decision making research in computer science (artificial intelligence, computational social choice), psychology, political science, and neighboring disciplines. The multi-disciplinary approach specifically crossfertilizes decision science disciplines that are traditionally segregated depending on their focus on individual versus collective choice, normative rational versus descritive behavioral, analytical versus computational approaches, and qualitative versus quantitative approaches. Broader Impacts from the Proposed Activity. The project impacts science and society on several levels. Making election outcomes resilient to probabilistic perturbations in ballot casting, vote counting, and to strategic manipulation is very important to many aspects of a healthy society. This applies across many levels and types of governance, from academic/professional to political elections. More generally, improving processes of consensus formation is relevant to many facets of private, business, and government life. The project connects the disciplines of computational social choice and psychology in new ways by linking the study of computational aspects of voting and manipulation to the study of actual human preferences in a formal laboratory setting. It will also develop a psychologically and statistically valid set of benchmark elections based on the Netflix challenge data, which will enable large-scale analysis of voting patterns. This contributes significantly to the scientific infrastructure for analytic studies of social choice in psychology, political science, statistics, and computer science. Moreover, it will impact both research and education, since graduate students in these disciplines will be able to learn tools and research methods from the other disciplines involved in the project. This will prepare them for a more fruitful use of their techniques and tools in their respective fields, and ultimately prepare them to function in an increasingly multi-disciplinary world. The team aims to disseminate results as broadly as possible, e.g., via symposium proposals on robust aggregation at international conferences in multiple disciplines related to the decision sciences. All four scholars have experience in organizing workshops and symposia on preferences and/or social choice.
StatusFinished
Effective start/end date9/1/128/31/14

Funding

  • National Science Foundation: $89,328.00

Fingerprint

Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.