2) Re. Setting s.d. to twice that used for chess: I tested what total scores the TP formula would be likely to give for various combinations of wins/draws/losses in a tournament and comparing this to the values if W=1/D=0.5/L=0 was used, as in chess. This gave a set of Chess:Tantrix "overall spread of final scores" ratios. These were weighted by the actual frequency of each W/D/L combination in the first tournament and the average ratio found was very close to 2.
3) Re. Setting the weighted (by G) average rating at the World Championship closing date for all tournament players to 1850: This has been changed (23 August) from the original [2 * lobby ranking at the time of the first tournament], which was chosen partly because it was thought that it might be useful to be able to compare lobby rankings and tournament ratings easily, because the overall level of lobby rankings has since increased. The mean and variance can be changed without any effect on the positions of players in the ratings if necessary.
9) b) Re. Keeping SUM (rating * G) pre- and post-tournament the same: As mentioned on the method page, for a single tournament, the average post-tournament rating should in theory be the same as the average pre-tournament rating, measured as the sum over all players of (rating * G for that tournament) - small approximations in Excel's NORMSINV function may stop this being the case. This only makes a big difference if all players' ratings are being iterated. If this results in the overall drift being downwards, then each iteration will reduce the average rating. Hence, if 8) b) is being used, the ratings should be adjusted after each iteration to keep the SUM (rating * G) total the same. In the first tournament, the drift would have been a bit less than one Elo point downwards per player per iteration otherwise.
The date and time are :
Sunday, 06-Jul-2008 06:00:46 GMT
Sunday, 06-Jul-2008 07:00:46 BST (local)
This file was last modified on Monday, 10-Mar-2003 20:03:28 GMT