The Predictive Power of Noisy Round-Robin Tournaments

The round-robin tournament format for N players is a scheme that matches players with one another in all possible N(N - 1)/2 pairwise comparisons. A noisy round-robin tournament adds the possibility of upsets, or noise, and hence reduces the power of the tournament to reveal the true ranking of the players. In this article we study theoretically (analytically and by way of computational simulations) the predictive power of noisy round-robin tournaments for three prominent distributions of players’ abilities, as a function of the level of noise and the number of players. At first sight, some of our results (e.g., non-monotonicity as a function of the number of players N, which makes some ranges of N non-optimal) are quite counterintuitive but should be of help to a tournament designer who tries to maximize, or maybe minimize, the probability of the best player winning.