MMR, Elo, and Other Rating Systems
Kind of curious about the algorithms behind the rating systems of different competitive matchmaking games.
Games of interest:
- League of Legends
- Valorant
- CS:GO
- Teamfight Tactics (TFT)
- World of Warcraft PvP
Glossary
- MMR: (hidden) matchmaking rating
- LP: league points
League of Legends
League’s rating system is based on gains and losses of LP. The amount of LP gained and lost depends on:
- the difference between your MMR and the average MMR of all players in the game; and
- win/loss streaks.
League has 9 ranked tiers ranging from Iron to Challenger. Each tier is further divided into 4 divisions. At 100LP in each division, you automatically advance to the next division (with rollover LP, where applicable). At the final division between ranked tiers (e.g. Gold I -> Platinum IV), you play your promotional series (“promos”), a BO5 set of matches to determine if you rank up to the next tier.
Matchmaking is determined by your hidden MMR value, which is solely determined by wins and losses. Your MMR is not affected by dodging games and AFKing, though the above will invoke a LP loss penalty.
As of 2021, League implemented Informed Ranked Matchmaking , a system which also considers the displayed rank in addition to MMR during matchmaking.
CS:GO
CS:GO’s rating system is based on a heavily modified Glicko-2 rating system. The premise of a Glicko-2 system is that each player has two scores: rank, and rank deviation. Rank deviation is the system’s confidence level of a player’s rank. If you have a higher rank deviation score, your rank is expected to change more rapidly. The more games you play, the lower your rank deviation score becomes as the system becomes more confident in your rank. If you take an extended break from the game, your rank deviation score will broaden leading to more pronounced gains and losses.
The Glicko-2 system was originally designed to be replacement for the chess Elo system.