Mark Glickman's New Model for Rating Competitors

The Harvard Department of Statistics is thrilled to share that Senior Lecturer Mark Glickman has recently had two papers accepted for publication by peer-reviewed journals in statistics and data science. His work on developing an improved rating system for chess players was also featured in the article “Breaking Chess’s Rating Stalemate” in the Harvard Gazette.

The first paper, "Paired comparison models with strength-dependent ties and order effects," was accepted by Statistical Modelling. This paper develops a novel statistical model for head-to-head games and sports, which reflects the phenomenon that stronger competitors tend to have tied results more often than weaker competitors.  For example, a solved game like checkers will result in a tie 100% of the time if neither player makes mistakes; however, the likelihood of a tie is much lower for weaker players.  This paper provides a model and analysis approach for this previously unexamined phenomenon.

The second paper, "Rating competitors in games with strength-dependent tie probabilities," was accepted by the Journal of Data Science. Building upon the model from the first paper, this paper develops a rating system for chess players that tracks competitors' abilities over time and again accounts for the phenomenon that top players draw games more often.  This work was commissioned by the International Correspondence Chess Federation (ICCF) and is currently in use for rating players in ICCF-organized competition.

Congratulations, Prof. Glickman!