CFE 2019: Start Registration
View Submission - CMStatistics
Title: Pairwise likelihood methods for paired comparison models Authors:  Manuela Cattelan - University of Padova (Italy) [presenting]
Abstract: Paired comparisons data are binary or categorical data representing the results of comparisons of a set of items performed two by two. Usually, observations are dependent either because the same person performs multiple paired comparison, or because the same item is involved in more than one comparison. A typical example of the latter instance is a round robin tournament in which each player is involved in different competitions, one against each other player. Hence, models that account for dependence in paired comparison data should specify a multivariate distribution for binary or categorical data whose dimension may be very large. Inferential procedures for this type of models become quickly cumbersome as the number of items involved in the paired comparisons increases. The aim is to present how pairwise likelihood methods can be employed to overcome inferential difficulties in such models. Moreover, the need to specify only bivariate distributions reduces the problems related to the growth of the number of items involved. The proposed methodology will be illustrated through applications to real data sets.