Silver is the equivalent to Diamond and here's how. (Math Inside)
While the Elo system is meant for two-player games, it will also work for two-team games if the teams are static. This is because each team is considered a ‘player’ entity, even though it is comprised of multiple players, and therefore fits within the two-player system. However if teams are randomly selected for every match, they can no longer be tracked using Elo, because a persistent entity no longer exists to fit within the two-player model.
For registered teams (3v3 or 5v5), Riot likely uses a traditional Elo system, because it fits within the above models and works properly within the scope of Elo. However for solo/duo queue, Riot chooses to apply Elo ratings to individual players, and then build a team rating based on the average. After a team wins or loses, the team is disbanded and each player receives a ratings adjustment. This breaks the entire Elo system and leads to completely flawed logic in matchmaking, such as:
Your rating is supposed to reflect individual skill, yet ratings adjustments are not based on you. They are based on randomized teams.
The best and worst players on a team are blindly considered to have contributed equally in a match.
The logic of this system dictates that, if your random team loses, then your personal score accurately deserves the full calculated penalty.
Each player bets their rating against 100% of the penalty, for only a 20% stake in each match. This means 80% of your rating is influenced by external factors, which are randomized before every match. This is not a skill-based system.
Because the system pushes to begin new players at mid-range Elo, using inaccurate random qualifier matches, these players are statistically likely to be overrated.
When an overrated player contributes heavily to a loss, the entire team is penalized, including those possibly underrated. Since new, overrated players are constantly entering the system, based on learning curves, this logically creates a negative feedback loop that some people call ‘Elo Hell’.
Once you get above the Elo range where new players may be placed, external factors in matchmaking become less random. Consistency in matchmaking continues to increase with higher Elo, potentially leading to a positive feedback loop in higher rankings. Some people might call this ‘Elo Heaven’.
The whole point of Elo is to measure the relative skill levels of players in two-player games, not multiple players in random teams. It doesn’t matter whether Riot uses a “proprietary system” or calls it “matchmaking points”. I’ve read many Riot posts about the new system, and while efforts are made to normalize data from streaks and other factors, they are still using an inherently broken Elo system under the hood.
For individual ratings to have any accuracy in team matchmaking there must be some basis on personal attributes. This is the reason why virtually all professional sports track performance metrics on players when calculating odds between team matches. Some games use a “points” system to measure player contribution within a team. Even something overly simplistic as “loss penalties are divided based on deaths” would add some accuracy to the system.
The purpose of this post is to point out the logical fallacies of Riot’s matchmaking system. There are many people who attack or defend the new matchmaking system without rational explanation, providing their isolated experiences as anecdotal evidence. This is not necessary. Based on information released by Riot and numerous players on the forums, the current system behaves exactly as designed. The problem is this design includes many leaps of logic and blind assumptions that are obvious to any programmer or critical thinking, and on top of that, is being falsely represented as being effective skill based matchmaking.
I challenge anyone to explain how this system performs individual skill based matchmaking in any logical way. Please describe how the logic in such a system works and how it represents individual skill levels with any degree of accuracy. (6x6=36)