Modelling preferences in the presence of missing data

This talk focuses on models for paired comparisons, which can be used to model ranked and partial ranked data as well as classic paired comparison data- data structures which are common in marketing and food tasting. After briefly introducing the Bardley Tery model, we focus on the need for dependence terms between responses, and introudce the problem of missing data. We present a log-linear model that can be used for both incomplete and complete observed response patterns while still alowing for dependence. The model is fitted to an augmented frequency table in which indicators correspond to whether or not a decision is observed or not. A particular advantage of this approach is that various types of nonresponse mechanisms can also be included and tested as nested models. The method is illustrated by a set of teaching styles data.