Abstract
Ranked Data arises when judges are asked to order a finite set of items, and it is present in various fields, such as health economics, information retrieval, and election studies. Often, judges only provide partial rankings of the finite set of items, be it by survey design or indifference to certain items after a particular preference level. It is often also true that judges can be subdivided into different clusters based on the both their ranking and social covariates via Mixture Models for ranked data. This project proposes a unified approach at the intersection of both of these realities, namely a Bayesian Mixture of Experts Model for Partially Ranked Data. The proposed model shows encouraging initial applications to election data from the 1997 Irish Presidential Election.
View the pdf in another tab here