| |
BGonline.org Forums
win probability for a PR difference and match length
Posted By: niuniuch In Response To: win probability for a PR difference and match length (Robert Andersson)
Date: Thursday, 20 September 2012, at 1:54 p.m.
OK, I have some experience with Elo-type rating systems, so I'll clarify a few things. I know this formula well, it's derived from the standard Elo formula and favours the stronger player in longer matches. And it works only based on the diferrence in ratings. The reason why this formula works in rating systems that use it is that the ratings themselves were calculated based on it. It's not possible that the win % is different for 1950/1850 and 1500/1600. It is however possible that the average PR difference between 1950 and 1850 is smaller (or bigger) than between 1600 and 1500.
Problem is that Elo-type ratings depend on the population that the system is applied to, so there is no way to calculate the rating based on the PR. Let's say that we have 2 servers, which use the exact same rating system, but on one server we have top-class players, and on the other we have mostly casual players. It's quite possible that even the low-rating players from the first server will be better that high-rating players from the second server.
Back to the point: Rating is a relative value, I'm looking for a formula that would use an absolute value - like XGs PR.
I'm aware that GNU produces something called "FIBS", which is an absolute value, although it really has nothing to do with the actual FIBS rating system. I wonder how it is calculated, and if actually the formula from the FIBS system would work for these values.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.