What Are People’s Actual Utility Functions in Budget Aggregation?

Abstract: Participatory budgeting and budget aggregation mechanisms inherently rely on assumptions about how voters evaluate non-ideal budget allocations. So far, there is little empirical evidence validating which utility models actually capture human preferences. In this work, we empirically study the structure of human utilities in budget aggregation. We introduce a modular, open-source polling framework that elicits participants’ ideal budget allocations and their preferences over carefully constructed pairs of non-ideal alternatives. Using multiple polling algorithms, our experimental design systematically tests both commonly assumed utility models and more general structural properties of preferences. Our results show that standard utility models, including L_1, L_2, and Leontief, fail to consistently explain observed behavior. In contrast, we find strong empirical support for star-shaped and multi-dimensional single-peaked preferences, as well as stronger conditions such as peak-linearity. Beyond these structural properties, our data reveal systematic asymmetries: participants evaluate gains and losses differently and assign unequal importance to different budget categories, which is inconsistent with any symmetric utility model such as L_p distance. These findings have direct implications for the design of budget aggregation mechanisms. While much of the theoretical literature establishes impossibility results for standard distance-based utilities with respect to truthfulness, Pareto efficiency, and proportionality, our evidence indicates that alternative utility models may better reflect actual voter behavior.