wait so decay gets reversed if you play again? how come ive never noticed this, i dont play ranked on my main account very often and ive been decayed 3 or 4 times this season, and ive never gotten the elo back.

Understandable - thanks for the reply.

The system I described is only for normal Elo ratings, which is why we can't discuss the numbers. It's just intended to account for the fact that players who are away for awhile tend to be rusty and then gain their skills back over time.

Ranked Elo decay hasn't changed recently.

thanks for that clarification, it wouldve been pretty cool to get 75-100 elo

Does this apply to Dominion as well? I played it to 1850 elo before taking a break from the game, and when I came back I only played SR. I haven't even tried Dominion because I don't want to jump in where I left off after not playing for so long.

Are you sure that's a good idea? Since we don't care about the actual numbers a skewed normal ELO distribution shouldn't matter as much. So long as the system creates a partial order our matchmaking should not have significant problems in the long term.

When we give people back their ELO instead of making them earn it, in addition to any ELO they lost or gained, we are introducing even more variation into the system. We have higher variance right after they come back[because they're rusty and we are uncertain of the change in skill level] but higher variance in games after they have regained their skills because now we put them at an ELO that isn't the result of playing games against opponents in that area.

It seems like it would make a lot more sense to give the normal ELO reduction and then to let adjustments occur faster [like when you first start playing ranked] since what we are essentially dealing with is an increase in uncertainty and/or a faster skill adjustment time.

I.E. A person who leaves the game and comes back will suffer some amount of lost skills. This amount is testable in aggregate[i think] such that we can model the skill level of a player as E1=E0+Z(t) +e. That is New skill = old skill plus a time factor plus an error(which may also be a function of time but is left out for simplicity).

Since we have uncertainty surrounding their old skill their new skill level is even more uncertain even if we can make a best guess.

In addition we believe that they will regain their skills at some pace. We might generalize this to compare it to the skill advancement of a normal player. I.E. why we have a minimum amount of ELO loss/gain. Such we can model coming back to the game after being away as a player with a larger amount of uncertainty in their skill level.

It seems like this would produce overall better results in terms of reducing the number of "swingy" games than returning their lost ELO to them on top of what they had already won or lost.

I wrote up an idea for a "karma system" (Attitude Rating) designed to avoid certain associated pitfalls, including the problems mentioned in relation to "Prisoner's Island".

Attitude Rating
http://na.leagueoflegends.com/board/....php?t=2371311

I should really make a new thread with a better poll and title, but I'll tl;dr it for you.

Attitude Rating: a rating from -5000 to 5000, given to every player.

• At implementation, everyone gets 0.
• New players start at the 50th percentile, or (if you want to hope that good habits will rub off on them) at the 70th percentile.
• Attitude Rating is hidden.
• However, players who reach high percentiles are given rewards, like badges. People in a high tier will know what tier they're in. Average and below-average players will not know.
• If the greater prevalence of downvotes relative to upvotes becomes a problem, give everyone a small amount of "free" AR per game played.

Voting: After a game, players can elect to give an upvote or a downvote to any player.

• Initially, players may only downvote once per game (or per two games). Reliable downvotes can expand this allotment.
• You cannot vote for someone you queued with or created a Custom Game with.
• Votes do not take effect immediately. They must be ratified by the Attitude Panel.

Here is my proposed voting prompt:
Attitude Panel: A simplified Tribunal-esque system. This helps correct for bias and abuse.

• A player goes to the Attitude Panel when they've received a critical mass of upvotes or downvotes.
• AR is only adjusted after ratification of an upvote or downvote by the Panel.
• If reviewers are scarce, it would be acceptable to assign only 1 or 2 reviewers to each case. (if split 50/50, dismiss.)
• Displays chat transcripts, but not game outcome, KDA, or other information. (Irrelevant.)
• No Riot staff review required, because bans and other serious measures are not at issue.

Three buttons would be presented to reviewers:

[RIGHT]( Courteous, polite, and friendly. )
( Neutral. )
( Rude or insulting.)[/RIGHT]

Matchmaking: The primary match is still an Elo match. But also try to match roughly according to Attitude Rating.

• If possible, match players with others within 500 Attitude Rating. I'd be willing to wait an extra 5 seconds for an increased likelihood of playing with nice people.
• If two or more people queue up together, match them based on their average Attitude Rating.
• Optionally: if a new player (e.g. sub-20) is among them, weight the new player's AR more heavily.
• This should probably only apply to Normal games, not Ranked games.

As to the concern that this will inhibit reform: my intuition is that we would get two main categories of low-AR players.

• The Ragers: They get angry and take it out on their teammates. They would sink to a low AR and stay there. In the current system, they'd just get banned and roll up new accounts (harassing new players in the process). I don't think it's feasible to reform these people by example: they reform if and when they decide to try to keep their emotions in check, or at least regulate their own conduct to prevent it from spilling over. If you want to give them second chances, implement a passive AR-regeneration mechanism that gives out a few AR every dozen games or so. This will give them second chances, at the cost of periodically exposing the next tier above them to their rage.
• The Rough-and-Tumble Crowd: Not ragers, just uncouth. They like to trashtalk. They're probably happier playing with each other than they would be playing with high-AR players, whom they perceive as wimps or crybabies. Everyone is happier if they play with each other. They're likely to downvote genuine Ragers, and drive them off to Rager Island.

Why implement this?

• Complements Tribunal. Tribunal weeds out the worst of the worst. Attitude Rating helps insulate "nice" players from abrasive types, or put less normatively, lets people play with others who share their attitudes as to what is appropriate.
• Compatible with Tribunal: a Tribunal punishment could also inflict a sizable AR penalty.
• Encourages and rewards good play. Spotlights friendly, courteous Summoners. Positive reinforcement is powerful.

Probably unnecessary, and I feel presumptuous adding this, but just in case it matters: I grant Riot the rights to use, modify, sell, and otherwise exploit this idea as they see fit. For that matter, the same goes for every League of Legends idea I have posted or do post on the forums, absent explicit language to the contrary.

I can't talk about what we are doing on the 'carrot' side of things for now. We're working closely with Pendragon to setup multiple experiments that will test the effects of positive behavioral features.

Yes please fix 1200 ELO! Its honestly just a mad rush to get out. I got up to 1288 And now I'm almost below 1200. Nearly every loss is due to the people just starting. No matter how positive and reinforcing you are, they just are not acclimated to ranked play.

Hey Lyte,

I asked this a little bit ago, but I think it may have been passed over...

Is being able to swap order in draft mode an option that's on the table? Few things can mess up a game as much as player 4 saying "get me x" and then not having anyone player 1 wants. Swapping orders would fix this.