### Glicko for Bughouse-like game

Posted:

**Mon Oct 22, 2018 2:41 pm**I was wondering what a good rating system for a 2 vs 2 chess variant would be. I was thinking Glicko should be a good starting point so I modified it for this purpose.

However there are a few problems with that: If everyone starts at 1500 and RD 350 or something like that, the RD goes down only very slowly since every opponent has a high RD too so that game doesn't give as much information as it usually would. (this problem gets further increased because in teams of 2 the combined (in whichever way that combination happens, the simplest being to take the sum) RD is even higher)

Is there a good way to fix this?

One obvious improvement is to choose their OTB chess rating as starting point and then lower the starting RD (e.g. to 200). (while chess skill doesn't translate 1:1 to such bughouse like games there at least is a correlation) This does help to some extend but doesn't fully solve the problem. Another idea I had is to fix the ratings for players who played in the first few tournaments to an approximation of their performance and give them a low RD so that they in turn can help other players decrease their RD. I tried to do that by simply evaluating the results of those tournaments again and again however sadly that doesn't converge in my case. (while some players stay stable-ish, some ratings go to plus or minus infinity)

Another feature I would like to have (although this one is of course not necessary at all) is to redistribute rating from players who are not active anymore towards active players. The reason behind that is that I have reason to assume that the playing strength of active players in a fixed rating range should go up. (i.e. rating probably will deflate in some sense) Because of that old ratings would likely be overrated. (that is just a guess though, since I don't even have a working system in the first place I have no concrete data on this) A simple solution for that would be to give active players some fixed amount of rating (for example 20 rating points) per year or so, or alternatively subtract that amount from every inactive players rating. Are there any problems that could occur from doing that?

However there are a few problems with that: If everyone starts at 1500 and RD 350 or something like that, the RD goes down only very slowly since every opponent has a high RD too so that game doesn't give as much information as it usually would. (this problem gets further increased because in teams of 2 the combined (in whichever way that combination happens, the simplest being to take the sum) RD is even higher)

Is there a good way to fix this?

One obvious improvement is to choose their OTB chess rating as starting point and then lower the starting RD (e.g. to 200). (while chess skill doesn't translate 1:1 to such bughouse like games there at least is a correlation) This does help to some extend but doesn't fully solve the problem. Another idea I had is to fix the ratings for players who played in the first few tournaments to an approximation of their performance and give them a low RD so that they in turn can help other players decrease their RD. I tried to do that by simply evaluating the results of those tournaments again and again however sadly that doesn't converge in my case. (while some players stay stable-ish, some ratings go to plus or minus infinity)

Another feature I would like to have (although this one is of course not necessary at all) is to redistribute rating from players who are not active anymore towards active players. The reason behind that is that I have reason to assume that the playing strength of active players in a fixed rating range should go up. (i.e. rating probably will deflate in some sense) Because of that old ratings would likely be overrated. (that is just a guess though, since I don't even have a working system in the first place I have no concrete data on this) A simple solution for that would be to give active players some fixed amount of rating (for example 20 rating points) per year or so, or alternatively subtract that amount from every inactive players rating. Are there any problems that could occur from doing that?