Video and Online Games
Ratings, stats and sample sizes in competitive play.
Start here · Essential guide
How competitive games turn results into ratings, per-match stats and win probabilities, and why noise, sample size and context limit what the numbers can say.
See how best-of-1, 3, 5 and 7 series change the chance of an upset, with worked binomial calculations and the assumptions that make real series less tidy.
How the Elo system turns a rating gap into an expected score, how ratings update after each game, and where the method works well or breaks down.
Compare Glicko and TrueSkill, two rating systems that store how sure they are about each player and handle teams, and see how they differ from plain Elo.
How matchmaking uses ratings to build balanced games, and the trade-offs it faces between match quality, queue time, latency and team composition.
Why unusually good or bad performances in competitive games tend to be followed by more ordinary ones, with a worked example and a simple shrinkage estimate.
Why a few matches say little about a player, with margin-of-error figures for 10, 100 and 1000 games and habits for reading stats more carefully.
Compare single elimination, double elimination, round robin and Swiss formats by games played and how often the strongest team is likely to come out on top.
How live win probability models estimate who is ahead, from pre-match priors to in-game state, with a worked round-score example and notes on calibration.
Related topics