Paired comparison models with strength-dependent ties and order effects
By: Mark E. Glickman
Potential Business Impact:
Predicts chess game winners and ties better.
Paired comparison models, such as the Bradley-Terry (1952) model and its variants, are commonly used to measure competitor strength in games and sports. Extensions have been proposed to account for order effects (e.g., home-field advantage) as well as the possibility of a tie as a separate outcome, but such models are rarely adopted in practice due to poor fit with actual data. We propose a novel paired comparison model that accounts not only for ties and order effects, but recognizes two phenomena that are not addressed with commonly used models. First, the probability of a tie may be greater for stronger pairs of competitors. Second, order effects may be more pronounced for stronger competitors. This model is motivated in the context of tournament chess game outcomes. The models are demonstrated on the results of US Chess Open game outcomes from 2006 to 2019, large tournaments consisting of players of wide-ranging strengths.
Similar Papers
Rating competitors in games with strength-dependent tie probabilities
Methodology
Makes game ratings fairer by counting ties.
Pairwise Comparisons without Stochastic Transitivity: Model, Theory and Applications
Machine Learning (Stat)
Better predicts winners without assuming a perfect ranking.
The Bradley-Terry Stochastic Block Model
Methodology
Ranks groups of players by skill level.