
BGonline.org Forums
Another direction
Posted By: Rick Janowski
Date: Sunday, 5 October 2014, at 7:07 p.m.
In Response To: Wrong direction (Timothy Chow)
There might be a workable way of using win rate to assess longterm Elo’s using the following process:
1. For any player obtain the long term win rate over x matches played. 2. Compute or estimate the average match length over x matches. 3. Calculate the Elo difference commensurate with the win rate and average match length 4. Make original guesstimates of the average Elo rating of all the opponents played weighted against match length and number of rematches, then adding/subtracting the Elo difference. to provide 1st computed estimate of Elo.
This process could be repeated by using the 1st computed estimates of the opponents played instead of the guesstimates, and continued iteratively until the estimate is considered acceptable.
A suitable minimum sample size would be one providing a similar level of confidence as that obtained for the minimum PR computation requirements. This may be in the 100s (guessing), but at least the matches do not require bot analysis. Although this system will tend to be more stable than a conventional Elo calculation with plus and minus swings after each match, the requirement for a large sample size may mean that the Elo is typical of playing performance of 2 or 3 years ago rather than current performance. Moreover, new players would have to wait years before getting a representative and reliable Elo. However, this may sort out Ray’s problem.

BGonline.org Forums is maintained by Stick with WebBBS 5.12.