BTW: To compare an SC-count it's not totally off for an Elo based rating.
Player1 = 9 SC, player2=1 SC => win/loss/draw = 0.9
(Similar to 2-players playing a tournament with 10 games. 9 wins from Player A, 1 win from player B). This would be the expected outcome from Player A having an Elo of 1800 and Player B 1400 and if Player A is not able to score this or a better difference he loose points.

@fasces349 : your answer just shows that you don't know how Elo points are calculated.

A difference of 777 Elo (only the "E" in capital letters, please) points means that the highest ranked player should win 98,87 % of the points scored in games against the least ranked player.

Maybe Bozo is a super champion, I don't know and that's not what I am talking about, but do you really believe that the n°1 player could possibly score in average 98,87 % of the points against the n°2 player (who would then only score 1,13 % of the points) in any discipline wether it is Diplomacy, chess, golf or tennis ?

If you don't believe me here is another example :

You seem to be a real veteran here with your 178 finished games and with your 16 % of victories and 37 % of drawn games !
However, I have only finished 4 games here and I have 579 "Elo" points more than you, which should mean that I should score 96,55 % of the points if you (who would then only score 3,45 % of the points) and I were opposed in games.

Maybe that new scoring system is something wondeful, I don't know, but please understand that all I am trying to say is that the name "Elo" is totally inappropriate for this system because it shows something else than what "Elo" means.

@Retillion: Maybe there is an error in the algorithm. I'm not that tech-savy, maybe you can check if and where the problem is...?

Yes, there is a problem in the algorithm.

The problem is in this line :

$real=$User1->supplyCenterNo/($User1->supplyCenterNo + $User2->supplyCenterNo);

Because of my personal schedule for today, I cannot develop now. I will write a complete answer in some 8 or 10 hours.

Oli,
I would like to check the math also, but I don't understand the programming. Can it be expressed in a math formula?

"I think a draw should be split the way it is in PPSC.

If 2 players work together to stop someone from soloing, an its ends 16-10-6. Why should the 6sc guy get any less? He was equally responsible for making the draw happen and should get an equal share of the pot.

When there is a draw, it is agreed that nobody won, therefore nobody should get a higher share of the pot."

I guess this makes sense based on intended game victory conditions, and now that I think about it, the issue that I see happening currently on large maps such as WWIV (players protecting a 1 or 2 SC country while they are at 30 SCs to be part of the draw) would diminish as well because good players would lose rating by drawing with 1 SC players who are lower ranked and thus the drive would be to reduce the draw count (as it should be in every game).

So winning in WTA obviously gets you 100% (survives in WTA mean nothing).
Draws in either PPSC or WTA get you 100% / (# of players in draw).
Defeats get you 0%.

So my only question then is how you apportion wins/survives in PPSC?

Should winning in PPSC get you 50% of the score, (capped at the victory conditions to prevent players from drawing games out to grab a bunch of SC's at the end to maximize their victory); survives in PPSC divide the remaining 50% or less by their share of the remaining SC's (1/2 * player SC count/(total - victory conditions))? Thus a win in PPSC could be only marginally better than a strong 2nd place at 47%? I haven't checked (though I think it is), but this would still need to be zero sum to fit the modified multiplayer Elo system (or ghost rating system).

Otherwise, do we say survives count for nothing (except DPoints in PPSC and W/D/S/E stats) and use the WTA formula like ghost rating does?

My preference would be to distribute the skill rating the same as the ghost rating (WTA rules) in all cases, and dpoints according to PPSC or WTA, depending on how the game is created.

For PPSC I think it should be awarded based on # of scs even in ELO (drawing dividing equally) because thats what were playing. I prefer PPSC because it incentive the survive>defeat.

@Leif:
You make the error for the rating to compare 1 player against all others.
In fact a 3-palyer draw is all thre players getting a draw against each other and a win against all the rest.
Basically you compare each player one by one against each other player.
So
A = Draw, B = Draw, C=Defeat, D=Defeat
Results in AvsB = Draw, AvsC=Win, AvsD=Win, BvsC=Win, BvsC=Win CvdD=Draw

Wouldn't it be possible to have GR following the nature of the variant in question? In WTA you may have different goals than in PPSC and vice-versa. They're not the same, so a rating system treating them as if they were the same would be wrong, IMHO.
A starting point could be what follows:
1) DRAWN GAMES. Since in those games you can get only 2 possible results ("Drawn" and "Defeated") whether in WTA or in PPSC, the system records this:
- Drawers are equal (draw)
- Defeated are equal
- Each drawer beated each defeated and, obviousely, each defeated was beaten by each drawer.
2) SOLOed GAMES. Things change for each variant:
- WTA: The winner beated all the others (because the 2 possible results are "Win" and "Defeated"). And obviousely each defeated was beaten by the winner. Defeated are equal.
- PPSC. There are 3 possible results: "Win", "Survived" and "Defeated". What looks fair to me is: The winner beated everyone else - Survived are equal, each one was beaten by the winner and each one beated each defeated - Defeated are equal, each one was beaten by any non-defeated.

Thoughts?

______
@Leif, I didn't follow this thread for a while.... has been decided anything for Countries' Stats?

And yes, if it was feasible, what fasces says for PPSC is even better for soloed games. Number of SCs: that's what were playing.

GR rankings already incorporate whether its PPSC or WTA. I keep saying this because its true, I think implementing an algorithm with proven success and popularity would be easier and better then trying to come up with our own.

We can use it as a starting guide and over time make amendments if necessary but I think its really the easiest starting point.

I guess one question we should ask is: Do we want to use the historical data, and make the skill rating consistent with the data? Or do we want create a skill rating which best measures skill, and start over with the games?

If we use historical data, I agree that the system should reflect how the games were played, even if it does not measure skill in the best way. If we don't use historical we can optimize the skill rating to best measure skill. I am fine with either.

But skill in a PPSC game is who get get the most scs, skill in a WTA game is who wins, they are different kinds of games.

I think that this discussion is finally turning interesting as it shifts away from issues of perceived nation imbalance to strength of opposition, which is the real variable worth facotring in. The the real measure of a Diplomacy player is how he (or she, least I be accused of being sexist) stacks up against the top players. It's really not that difficult to trample all over inexperienced noobs and perpetual CD style players. However, if you somehow manage to defeat Devonian in a 1v1 game, then you've truly accomplished something noteworthy, and should be suitably rewarded more than pounding on some guy who signed up yesterday. And if you outperform The Ambassador, Bozo, Cypeg, or Guaroz, in a variant game then you've really earned your wings. On the other hand I think that we need to be extremely cautious about awarding any bounus to players because of a nation's prior performance. Sampling anything less than 100 or more games tells us nothing. But if we truly have maps that have positions that are so imbalanced that they warrant a bonus, then perhaps these maps should be re-drawn. I disagree that they need to be, and aside from the GvI map which should be scuppered as unplayable, no one has presented any evidnce of merit that they really should be. The Egypt analogy from the WW4 map is a myth, as big as the whopper fable that suggests Italy is a weak power on the standard map. There are literally thousands of different ways to play these positions, and blaming a nation's starting position for a game performance failure is counterproductive to learning anything about the map or your own style and skill set. I argue so arduously against this behavior so that we as a community can avaoid making excuses for poor peformances. Everyone has bad games. This is not a bad thing. We can all learn far more from bad games than we can from great victories. Let's not adopt a welfafe mentality here. If you tank a game as Egypt, or Italy then go back and think on your failings, and consider what you might have done differently. Reaseach other strategies. Interview players that like playing these nations. Experiment with the position again, using another approach. The more accurate assessment of nation performance is that some nations are better suited to certain styles of play than others, but there are not necessarily stronger or weaker. .

One of the many differences to the GR is, that GR calculates the expected result as 1player against all others and my first attempt was to split each game in many 1v1. We can give this a try, but this would rate a 1v1 exactly the same as an 34-player worldmap. I'm not sure if this is fair.

As a first test we could use the points-won instead of the SC-count as an easy way to measure success. Equal-points-won = 0.5 = Draw. In a WTA the winner has all the points, so this is easy too. Also this would help with the rating on the Rinimascento-map, as this map has already an alternate scoring system.

I agree with Ruffhaus, if your worried enough that a maps imbalance will effect your ELO rating my simple advice is to not play that map.

"One of the many differences to the GR is, that GR calculates the expected result as 1player against all others and my first attempt was to split each game in many 1v1. We can give this a try, but this would rate a 1v1 exactly the same as an 34-player worldmap. I'm not sure if this is fair."
Not quite, its rates you vs everyone else combined. So say everyone in a game has a GR of 100, since my odds of winning a 1v1 are 1 in 2, I would get only a slight boost in GR from winning, but since my odds of winning a chaos game is 1 in 34, I would get a massive boost for winning that.

But thinking about it you bring up a good point, for a 1v1 GR game to work, every single game would have to be WTA, otherwise if my GR is 500 against a GR of 100 in a 1v1, even if I win a PPSC game 19-15 against you, since its counted as PPSC and my GR is so high and yours is so low, I would actually lose points, because the % of the pot I won is lower then the % of the pot I was expected to get. In fact if my GR is 35 times higher then yours, if you survive with just 1 supply center in a 1v1, I would still lose GR. (That said this would really be one of the best on the site vs one of the worst since based on the current list, the best player on webdiplomacy has a rating of 45 times higher then the worst.

Thinking about it, for GR to work I think smaller maps would have to be WTA for it to work, otherwise in extreme cases (in the unlikely event that Devonian or Bozo faces some unnamed rookie) even if Devonian or Bozo wins, they may end up losing GR.

@fasces, weren't GR and "normalization" 2 separate rating systems? Putting them together (like airborne proposed) would be a third system...

I haven't recommended normalization and have stated that I am against it. What I am talking about it that 2 player variants have to be WTA otherwise GR wont work.

The new rating looks really impressive, but as asked before, it would help to be able to check the formula.

I am glad that I suck enough, not to be on the list, but could it be that the current algo rewards heavy gaming ;-).

Fasces,
"Thinking about it, for GR to work I think smaller maps would have to be WTA for it to work, otherwise in extreme cases (in the unlikely event that Devonian or Bozo faces some unnamed rookie) even if Devonian or Bozo wins, they may end up losing GR."

Absolutely true. But it doesn't have to be against a rookie. Remember that to gain, you need to outperform your expected result. I won't do the math, but if someone has 88% of their opponents rating, and the better player wins 16-18. The better player will lose skill-points. They under-performed against the opponent. This will probably be the case in many of the 1v1 games I have played. I am ok with that though, but from now on I will probably insist on WTA 1v1 games.

@fasces. Ooops, sorry for my inappropriate post. I didn't follow this thread for a while and I missed the point.

well there would have to be some kind of override function, since a good chunk of games have been ppsc and that would reduce the rankings for really good 1v1s

look at the stats screen I did find a bug. Every single 2 or 3 player map has an identical number of PPSC games as they have with # of draws, except for AvF which has an identical number of WTA games as it has draws (and Lepanto which has 0 PPSC games).

I seriously doubt that only 50 Austria v France games have been WTA while over 700 Italy v Germany games have been. I also think the odds of those number of draws being identical to the number of wta/ppsc games for 14 different variants is too low for these numbers to represent what is actually there.

Just a noticed bug in the variants page.

I did some more tests.
If you look at the HoF we have now 3 stats. All are generated by matching each player against all other players in the game individually.
1. SC - Like the last one.
2. DPoints won. This way a draw and SC-based will be valued.
3. Hard coded Won/Draw/Lost. (Eg. A player Won in a 34-player game is matched as Won against all other 34 players. A Draw is matched as drawn against all other drawn and as Won against all Defeated)
You can check the formulas on the Wiki: http://www.vdiplomacy.com/wiki/index.php?title=Rating
And the rating in the HoF: http://www.vdiplomacy.com/halloffame_elo.php

(And I will rename this to something other than Elo later. It's not a priority right now)

"http://www.vdiplomacy.com/halloffame_elo.php"
I can't read PHP, so that just looks like gibberish to me.

I will not post a long message in order to demonstrate that "Elo" rating should be renamed because it seems that it has just been renamed into SCs which is the name that I wanted to propose :)

When the final ways of calculating the various points will be decided, will it please be possible to know what formulas are used ? Indeed, I have no idea what "Points Won" and "# of Matches (1on1)" mean.

By the way, I fully agree with RUFFHAUS_8, when he says that he disagrees with players who think that some countries are weak. For example, I have played one game with Egypt in the World War IV variant and I had a fantastic and easy game !

I wrote simultaneously with Oli's latest message. I now know what the various scores mean.

That's awesome Oli! And I'm always impressed by how fast you get things done. Great job and thanks.