Item-specific order effects in paired comparisons

The effect of the order of presentation in paired comparison models is most often assumed to be constant accross the compared items. This assumption is not realistic, especially in sports competitions where players or teams can derive different benefits from playing on their own field in front of their supporters. This work discusses ridge-type estimation of item-specific order effects in paired comparisons using a combination of composite likelihood and empirical Bayes methods. The methodology is illustrated through the analysis of sport tournament data.