A class of logistic models for analyzing multivariate paired-comparison experiments with and without ties is proposed. These models facilitate fitting of the Bradley-Terry-Luce model or the Davidson model to more than one response variable, and they model the association structure among the multiple response variables. Additional design-related effects, such as order effects within pairs or group treatment effects, may be tested. Parameters of the models may be expressed as linear functions of concomitant variables. The approach is exemplified by analyzing data from two consumer preference experiments.
ASJC Scopus subject areas
- Applied Mathematics