Revealing subgroup structure in ranked data using a Bayesian WAND

TL;DR

This paper introduces a Bayesian nonparametric model called WAND for analyzing ranked data, effectively revealing subgroup structures and assessing ranker reliability, especially when data are incomplete or heterogeneous.

Contribution

It proposes a novel WAND process mixture of Plackett-Luce models that captures subgroup heterogeneity and ranker differences with efficient inference and detailed posterior insights.

Findings

Successfully identifies subgroup structures in simulated data.

Effectively assesses ranker reliability and entity exchangeability.

Provides a flexible framework for incomplete and heterogeneous ranked data.

Abstract

Ranked data arise in many areas of application ranging from the ranking of up-regulated genes for cancer to the ranking of academic statistics journals. Complications can arise when rankers do not report a full ranking of all entities; for example, they might only report their top--$M$ ranked entities after seeing some or all entities. It can also be useful to know whether rankers are equally informative, and whether some entities are effectively judged to be exchangeable. When there is important subgroup structure in the data, summaries such as aggregate (overall) rankings can be misleading. In this paper we propose a flexible Bayesian nonparametric model for identifying heterogeneous structure and ranker reliability in ranked data. The model is a Weighted Adapted Nested Dirichlet (WAND) process mixture of Plackett-Luce models and inference proceeds through a simple and efficient Gibbs sampling scheme for posterior sampling. The richness of information in the posterior distribution allows us to infer many details of the structure both between ranker groups and between entity groups (within ranker groups), in contrast to many other (Bayesian) analyses. We also examine how posterior predictive checks can be used to identify lack of model fit. The methodology is illustrated using several simulation studies and real data examples.