Coarse Preference Elicitation

A planner wants to extract information about an agent’s preference relation, but not necessarily
the entire relation. Formally, a partition of the set of all possible orderings of alternatives (a ‘type
space’) is given, and the planner wants to know to which partition element (‘type’) the agent’s true
preference belongs. We say that a type space is elicitable if there exists a mechanism mapping types
to (possibly random) outcomes in which the agent strictly prefers truth-telling over lying. In the
Savage framework a type space is elicitable if and only if it can be elicited by offering the agent a
list of menus and paying one randomly-chosen choice. When the planner can use objective lotteries,
more type spaces can be elicited.

Joint work with Christopher Chambers and Paul Healy