Change ELO calculation?

Hi!

Few recent events have caught my attention and I believe fueled the need to change the current ELO system:
1. [Er0]Godde lost his 1st place in the ladder due to few losses (one of them was that he needed to go I believe)
2. kaen GTG in one of the games with lower elo player, which costed him place in 1vs1 top-50.

Now the rationale.

The current ELO rating system makes low-skill vs high skill player matches uneven. High skill player has a lot to lose and almost nothing to win. Besides an accidental loss to lower skill player, a lot of things might go wrong: crash, desync, disconnect, urgent GTG, smurf of high level player, etc. A low skill player on the other hand doesn't practially risk with anything.

Secondly, lots of existing players don't play anymore, however stay in the ratings. I believe some change was made like month or two ago and inactive players don't appear in the ladder, however when they reappear, they still will have the same old rating.


The proposal is following. To change the ELO rating calculation that it only takes into account wins and ignore losses. As before, the amount of elo gained by wining, would still depend on the difference between pre-game elo, however some small changes might still be needed here. To compensate for increased ELO of the winner the ELO of the rest of the gang (basically all other players in the list/ladder) should become "deflated" in order that medium or median ELO stays say around 1500.

The benefits:
1) Single accidental loss to lower ELO player wouldn't harm like it does today. Urgent GTGs, disconnects, loss to smurfs, etc. are also compensated by this change.
3) People will play more:
3a) Less mental blocks, like "a loss will cost me my 1st/top-10/top-50 place".
3b) The more people play(and win) the more they gain and the more the rest of the playerbase gets deflated
4) Inactive part of the playerbase doesn't preserve their place in the rating and gradually goes down the ladder.


Let me know what you think, folks.

What is the purpose of your ranking system and do you have any proof that it achieves this purpose? For example the ELO system makes statements about the proportion of wins you can expect against another player.

Thanks for argument, guys.

Anarchid, sure it's not a vanilla elo. If the name is the only obstacle, then it's solvable :)
GoogleFrog, have you carefully read the post? It has answer(s) to purpose question as well as stating possible benefits. I don't have a math proof that the new system gives an accurate proportion of W/L against a player. The only change I practially propose here is to replace substraction of loser's rating with global deflation, so the proposed system may well retain the W/L ratio at least to some extent. Secondly, I question if knowing of the W/L ratio is really of value. To me boosting amount of matches played and praising active, winning players is much more important than a theoretical statistical result of "!predict"

@Saab, the proposed system will move you down the ladder the longer you don't play (or strictly the more matches are played w/o you).

If Godde plays a couple more games in such a way that he deserves to be #1 then his rating will change to reflect that. Over time small anomalies like players leaving games, etc. even themselves out. I don't see a particular need to change the rating system.

That being said, something else has occured to me: there are at least a few players ranked highly in 1v1 who play the mode very rarely, if at all, but retain their position on the 1v1 list by virtue of playing teams games. I'm not sure if this is desirable.

ivand
It seems like that kind of system would be exploitable. Win trading would be an easy way to get high up in the ladder.

Game quality might degrade as well since trolling could be done without risking to lose (much) Elo, especially by higher Elo players when playing against lower Elo players.

I find my own Elo to usually be around 40 from my average and in rare cases change up to 70 from my average. This isn't really a big change and indicates that the system works fine.

ELO is based on whether you won against another player or not. if the formula is changed, a new name is needed.

Elo Decay could be a way to keep non-players out of the ladder. Something like -3elmos/day after 2 weeks of not doing an actual game.

quote: Few recent events have caught my attention and I believe fueled the need to change the current ELO system: 1. [Er0]Godde lost his 1st place in the ladder due to few losses (one of them was that he needed to go I believe) 2. kaen GTG in one of the games with lower elo player, which costed him place in 1vs1 top-50.

The two points you brought up are no reason to change how Zero-K calculates a player's expected skill. Godde did not play for some time so he got rusty and his elo number now reflects this properly. As for kaen, I couldn't find a game where they 1v1'd a new/much weaker player. Either way, elo taking everything into account is by design - by ZK rules being forced to leave a running game means you lose.

The self-correcting nature of Elo is what makes it great. Sure, a freak loss can make you lose a position in the ladder.. but then if your skill is constant, you will drift back up. Godde will certainly occupy the 1v1 seat once again.

The main problem here is actually how Elo is used to balance team games, which is fubar.

quote: Seems like it's only me, who doesn't like to lose both games and elo. :)

It is true for everyone, but people just understand theres no way around this.

Another problem with your suggested system, by the way, is that it is very easy to exploit.
Its called killswapping. Or, in your case, it would be winswapping. Get a friend to play 1v1 against you repeatedly, with each winning 50% of the time. Since you both win 50% of the time, neither gets a big rating gain penalty and both players can easily inflate their rating to absurd values without ever having to play against anybody else.


As far as i know, all competitive games currently use the elo-type system, where you gain points by winning, lose points by losing, and usually have modifiers for skill difference.

Even the "casual" mode in most of those games usually uses hidden elo for balance, it just doesnt display it.