Let me start by saying that I am not against rating systems. Often, though, the implementation of rating systems has unintended consequences. They should encourage the right type of behavior and discourage the wrong type of behavior, but not everyone always manages to predict how they will affect behavior.
One aspect of vRating that I like more than the GhostRating that is used on webDiplomacy is how it treats Civil Disorders and takeovers of them. On webDiplomacy, players who care about their rating will never take over losing positions in games, but in vDiplomacy, it is safer to do so because your vRating won't be hurt as bad if you take over a position.
I do like GhostRating more than vRating, though, because of the following simple fact. Your change in GhostRating after completing a game is ONLY dependent on the following two factors:
- Your result in the game;
- How your rating compares to the other players in the game.
As such, for your GhostRating it doesn't matter if you draw with lower ranked players and eliminate higher ranked players or vice versa. For your vRating, though, it does, and I think it shouldn't.
One consequence of this is what many people in this thread are complaining about: headhunting. If you draw in a game, you get more points if the players eliminated are higher ranked players. However, I do like to point out that a reverse effect exists, too. If you are losing a game, you lose less rating the higher ranked the winner, or the drawers, of the game are. As such, if you are losing a game and care about your rating, it may be a good idea to throw all of your centers to high ranked players in the game. I have no personal experience of this happening, but neither do I have personal experience with the headhunting - I only play few games and they are all anonymous, so I don't think anyone ever knew my identity.
I read that Ruffhaus wrote: "Making matters worse when you lose a game from the top position you lost exponentially more points than the 50th ranked player would." By my understanding of vRating, it is not exponential. In the limit where your vRating goes to infinity, you only lose twice as many points as compared to someone whos rating is dead average compared to the other players in the game. However, if you win or draw a game, your gain in vRating goes exponentially down the higher your vRating already is.
This last fact can have ridiculous consequences in large games. Suppose that there are two WTA Divided States games being played. In one of the games, everyone has a vRating of 1000. In the other game, 49 players have a vRating of 1000 but one player has a vRating of 2000. Both games end in a solo; the player with an initial vRating of 2000 is one of the soloists. According to my calculations, the new vRating of the player with an initial vRating of 1000 will be equal to 3320, which is insanely high and which would make him the #1 of the site by a mile. Some other calculations tell me that the new vRating of the player with an initial vRating of 2000 will be roughly equal to 2450, which is WAY lower.
I have been thinking about solutions to improve vRating. (Disclaimer: I am not a developer, just a math student with some free time, and no one has asked me to do this.)
I think the problem I outlined in the paragraph about the Divided States solo example could be solved if the calculation of the new vRating worked in an 'incremental way'. Currently, the solo of the 1000 vRated player would be interpreted as a full, instant win against 49 opponents. You could also interpret it as a fractional win (similar to how a draw is interpreted) against 49 opponents, then calculate his new rating, then interpret it as a fractional win again, etc. Basically you are approximating an integral. Then his new vRating wouldn't be so ridiculously high, and should never get above what someone with a higher vRating would end up with.
As for implementing a system that does not encourage headhunting, I think that that is hard if you want the ratings to work across all variants. GhostRating, which does not encourage headhunting, works in an easier way than vRating and I do think it is the better rating system for webDiplomacy. I don't think that adopting GhostRating for vDiplomacy would be a good idea, though. In GhostRating, a high ranked player can lose rating in a draw if the draw is too large; he then simply has a worse result than would be expected for someone of his rating. This is a necessary requirement for a rating system that does not encourage headhunting and that works in a somewhat easy way. In vRating, this is never the case. I like GhostRating more on this front too, but I do think that vRating better translates across multiple variants. If you can lose points in a draw when the fraction of players eliminated is too small, then higher rated players who care about their rating would do better in large variants, and lower ranked players would, by symmetry, do better in small variants.
It should be possible to create a system that does not encourage headhunting and works well across all variants, though. I have some ideas in my head that may work but I haven't taken the time to work them out fully.