Boards›Gameplay

It feels like it takes too many games for MMR to reflect skill

You Are Cute XD·10/16/2017, 7:36:00 PM·3 votes·344 views

I am not very good at league. I play a ton, but in low elo the variety of skill levels is so wide that it feels almost impossible to judge where you are, and play with 9 other people of a similar skill level, without playing an absurd amount of games. To me, this is a major problem with using something as simple as elo rating to determine skill levels and match people.

Elo rating only accounts for wins and losses. This is great for individual sports like chess, but in a team game like League, it's much harder to tell individual skill from team wins and losses. The solution to this is that over a large enough sample size the individual performance will be reflected in the overall record. This is not necessarily ideal, since I have no idea what sample size it takes for that to stabilize, but it is what it is and you need a large sample size as a base.

But then League adds in other difficulties. In Silver-Bronze, 6-10% of games have an AFK in it. There is a relatively equal likelihood that either team will have an AFK, so it will come out in the wash eventually. But, it requires a larger sample size in order for that to stabilize. Similarly, low elo is often plagued by smurfs that like to stomp their way through our games. Again, this will eventually be balanced, but also requires a larger sample size of games.

It just seems like the answers to all the problems in matchmaking tend to be that it will stabilize eventually if you play enough games. Given applying an individual rating system to a team game, and any number of variables that can cause the game to be decided regardless of your contribution, it seems like the sample size required to get accurate skill levels is enormous. As a person with a job and other hobbies, it seems difficult to even be able to play enough games to know what my skill is in the first place, unless I play with all my free time.

I think a different system is needed to get an accurate depiction of skill faster. Elo rating is too simple, and may not be the correct thing to use in this scenario. I don't know what is, but it would be nice to see some experimentation with different mathematical techniques/indicators of progress that may be able to stabilize faster. It seems like maybe champion grades were an attempt in that direction, and while there are infinite problems with them, experimenting to try to see other available options seems better than just saying players have to play a ton of games to play with other people at their skill level.

TLDR: in a game with so much variability, Elo rating takes way too long to stabilize.

5 Comments

MysterQ10/16/2017, 7:39:50 PM1 votes

Tldr, MMR where you have one smurf you queue with and suddenly when you soloque you are playing people 2 tiers above you every game.

ARAM ZooKeeper10/16/2017, 7:55:46 PM1 votes

This is an excellent post - I agree 100%. I feel like it does a good job at criticizing the problem with matchmaking, while not triggering higher-ranked players who want to defend their high ranks in the current system.

While a challenger level player could climb through bronze/silver/gold relatively quickly, the current system may require hundreds of games for a silver-skilled player to eventually rank out of bronze. Combine that with the fact that there are much larger skill gaps in the same ranks at lower ELOs, and it means there are more variables to worry about, and that requires countless games to eventually stabilize.

I think the #1 complaint is that players don't feel as if their performance matters, and they want a system that considers their individual performance, at least if the system is not capable of giving them a fair game consistently. As the OP said, especially those of us with jobs/lives... if I only have time for 3 games a night, I'd like my performance and my skill to determine my rank, not my patience and time availability.

Azure Hamster10/16/2017, 8:01:33 PM1 votes

If you've ever built a player-rating system (I have), you know how hard a problem this is for a 5v5 game. A player's skill has much more effect on the outcome of a 1v1 game than a 5v5 game. So for the statistics to be valid, it must take more games for the MMR to adjust in 5v5. Unfortunately, it's just the nature of the beast.

Caitlyn

Eedat10/16/2017, 8:04:48 PM1 votes

There is a relatively equal likelihood that either team will have an AFK, so it will come out in the wash eventually.

Technically there is lower odds of a troll/afker being on your team assuming you aren't doing it yourself. There are 4 other slots on your team and 5 on the enemy team.

Elo rating only accounts for wins and losses. This is great for individual sports like chess, but in a team game like League, it's much harder to tell individual skill from team wins and losses.

There is no good automated way to judge skill level. KDAs and CS numbers can be absolutely meaningless. I've literally had people AFK in base during completely winnable games because they dont want to "ruin my KDA". Same goes with other numbers. If raw numbers start affecting the game, people will play for the numbers and not to win.

The problem with using numbers is that they could encourage a mentality of "this game is lost. I'm 1/2/7 so maybe if I avoid fighting at all costs and eat up as much farm as possible I can squeak out an A+"