An R package for estimating mixtures of Benter models with ranked data
Lighting Talk, 3:15 - 3:35 PM
Ranking data arise in a variety of domains, such as ranked choice voting, psychometrics, and sports. Statistical inference on these data may provide essential insights on the underlying systems (e.g., the political preferences of voters). However, fitting statistical models to rankings is challenging due to the high-dimensional and discrete nature of the underlying data structure. Recent R packages, such as PLMIX [Mollica and Mollica, 2025] and BayesMallows [Sørensen et al., 2020], provide computationally-efficient tools for estimating Plackett-Luce and Mallows ranking models, including in the mixture model setting. However, no such package exists for estimating Benter distributions, which have been demonstrated to be useful when modeling ranked preferences [Gormley and Murphy, 2008]. In this work, we implement an efficient variant of the Expectation-Maximization algorithm for estimating mixtures of Benter data in R. Additionally, we provide tools for model selection and inference, as well as functions to visualize data and parameter estimates. The package is demonstrated on real ranked choice election data. We hope that a tool to fit these models easily and efficiently will lower the barrier of participation for a broader audience and provide a more accessible method for understanding voting or decision patterns in larger bodies of individuals.
 | Pronouns: they/themPortland, OR, USA |