We are developing methods to estimate both the content and structure of decision makers preferences in a variety of domains, including risky options, climate and energy tradeoffs, and difficult medical decisions.

We are developing methods to estimate both the content and structure of decision-makers’ preferences in a variety of domains, including risky options, climate and energy tradeoffs, and difficult medical decisions. As a generalization of traditional preference analysis, the approach can be used to make recommendations for people who know what they want, uncover complex choice rules, and suggest paths toward clarification for those who are uncertain.

From recommender systems used by technology companies to sell products, to revealed preference studies used by researchers to infer quantities like the value of a statistical life, or voting schemes that determine the trajectory of nations, approaches that elicit and infer preferences from the choices people make assume that decision-makers know what they want. That is true if decision-makers can consistently order the available alternatives, yielding transitive preferences, and are not susceptible to subtle but inconsequential changes in how the alternatives are described or made available (framing effects, context effects, reference dependence).

We leverage recent advances in graph matching and non-linear embeddings, combined with pairwise comparison choice data, to cluster decision-makers based on what they want (the content of their preferences) and whether they know what they want (the structure of their preferences). Across three experiments, including classic studies of risky choice and a two attribute study about state-level electricity generation portfolios, we find significant heterogeneity in both the content and structure of decision-maker preferences. Decision-makers most frequently choose in a way consistent with utility maximization, yet some decision-makers make choices consistent with heuristic rules, while others appear to be uncertain about their preferences.

Featured paper: Are preferences for allocating harm rational? (preprint)