The algorithm I used app is a simple modification of the standard elo formula, which I got from the US Chess Federation's rating estimator[1]. For two player games, the app uses the standard elo formula. For games with more than two players, it simply calculates the expected score as per a two player game and then divides it by (num_players/2). For instance, a two player game with player A and B rated 1200 would result in a expected score of .5 for each player. In a four player game, the pair-wise expected score is calculated by averaging the opponents' ratings and then using a standard elo calculation. This is then divided by (4/2) to give the expected score. For those interested, I wrote a blog post where I go into more detail [2].
Although this is a pretty crude method, I've been pretty happy with the results. I run a Settlers of Catan league where 111 rated games have been played. Here's a graph of all players who have played more than 5 games: http://imgur.com/ZoICwNM
Overall, I think this algorithm produces more accurate ratings than many other multiplayer rating systems, including the one the official online Settlers of Catan site uses (I discuss this in my blog post). This algorithm does break down if there is a really large rating difference (i.e. 500+ points) between one player and the rest, but that's a pretty unlikely scenario.
The source code is available on Github [3]. Pull requests welcome! If anybody wants to create an iOS version, I'd be happy to collaborate and help out in any way. There is also a desktop version I made eons ago for those without Android—unfortunately, I never added in two player functionality [4].
Although this is a pretty crude method, I've been pretty happy with the results. I run a Settlers of Catan league where 111 rated games have been played. Here's a graph of all players who have played more than 5 games: http://imgur.com/ZoICwNM
Here's the rating distribution: http://imgur.com/Rot4zkF
The current ratings for those in the graph above:
1185
1079
1067
1043
1001
991
975
973
937
928
916
912
906
854
Median: 974, Mean: 983
Overall, I think this algorithm produces more accurate ratings than many other multiplayer rating systems, including the one the official online Settlers of Catan site uses (I discuss this in my blog post). This algorithm does break down if there is a really large rating difference (i.e. 500+ points) between one player and the rest, but that's a pretty unlikely scenario.
The source code is available on Github [3]. Pull requests welcome! If anybody wants to create an iOS version, I'd be happy to collaborate and help out in any way. There is also a desktop version I made eons ago for those without Android—unfortunately, I never added in two player functionality [4].
[1] http://www.glicko.net/ratings/approx.pdf [2] http://www.gautamnarula.com/rating/ [3] https://github.com/gnarizzy/gameratingcalculator [4] http://www.mediafire.com/download/3n3ah3lt571ngy2/Game+Ratin...