The date and time are :
Saturday, 17-May-2008 08:20:54 GMT
Saturday, 17-May-2008 09:20:54 BST (local)
* Click the "Back" button on your browser when you want to return to the previous page you were looking at *
How seeding points are calculated:
The most important factor in determining a player's seeding points is their tournament Elo rating.
To get to the seeding points, the Elo rating is adjusted as follows, noting that G is the number of tournament games played after deweighting games played in previous years as described in the main Elo rating rules:
a) If G is less than 36, the player's Elo rating is adjusted down by (250/(36-G))
b) G for the purposes of a) above is reduced to 18 (50% of 36) if less than two tournaments have been completed by the player and to 27 (75% of 36) if exactly two tournaments have been completed, with the proviso that a tournament is only counted as completed if the player has played six or more games in it
c) If no tournament games have been played in or since the last completed calendar year, the seeding points are further reduced by 100 - most players who come back after a long break are out of practice to some extent
d) The seeding points are further reduced by 100 * (6 - G) if a player's overall G is 5 or less - this is because the probability of a rating being very extreme increases dramatically with each 'missing' game if G is 5 or less to the extent that a single linear adjustment like (250/(36-G)) is no longer appropriate
e) If it is higher than the number of seeding points calculated above, then 2 * lobby ranking - 250 becomes the player's seeding point value
f) Teams are seeded on the basis of the average number of seeding points of the players nominated as members of the team when the team was entered
g) For team tournaments only, the lowest seeding point value that can be assigned to a player is 1400 - this has been changed from 1500 to allow better differentiation between the lower teams in the 2003 WTTC (some of which would otherwise have been put into seeding bands on the basis of the rating/ranking of just one player or no players at all) and has not affected any of the top 15 teams in that tournament.
Questions and Answers:
Why is the player's Elo rating treated as the most important factor?
The Elo rating is by far the most accurate measure of performance under tournament conditions, i.e. in serous timed games and against opponents selected at random by the draw for each tournament rather than selected by the player him/herself.
Lobby rankings are based on untimed games, are easier to cheat in, are calculated in a way that makes it easier to get high ratings by playing weaker players and can sometimes be based on large numbers of Robot games.
Master rankings do not cover all players, making them not universal enough to be useful in seeding tournaments, and (as you will know if you have bounced up and down them over like a yo-yo over very short periods) have a calculation method which, while undoubtedly more scientific than the lobby ranking method, measures only very short-term performance except perhaps for those at the extremes of the scale, again inappropriate for seeding tournaments.
In both lobby and master games, players play self-selected opponents - a recent chess experiment proved that this can make a huge difference, massively favouring players who select the people they have whose style favours their own game over those happy to play anyone who asks them to play.
Why adjust the Elo rating down to get the seeding points if it is the most accurate measure?
From the very first game, a player's Elo rating is the best estimate of their strength based on the evidence of the games they have played, i.e. they do not move slowly towards their 'true' values from an arbitrary starting point. Elo ratings based on low numbers of games and/or performance in very few tournaments or only on tournaments played in the distant past have a wide 'confidence interval' around them, ie. though they are likely to be much better estimates than assuming every player starts at an arbitrary level like 500, they are still unlikely to be that accurate, eg. it is much easier to get a very high rating in just one tournament than to keep up that level of performance up over two or more tournaments.
Hence, while the Elo rating itself is not reduced (so that it retains its integrity as the best estimate based on games played so far), seeding points are reduced for those who have not played in many tournaments or who only played in tournaments well over a year ago to take account of this wide confidence interval. Of course, a player's rating from only one tournament may be unluckily low instead of unsustainably high, but the adjustment is always a reduction to give the benefit of the doubt to players who have proved themselves in more tournaments.
Why have the reductions changed for 2002?
The higher the G threshold, the more accurate a rating needs to be before it is considered established, but it is pointless having a threshold so high that hardly any players end up with established ratings, so this value has to be the best compromise available. As many of you know, it was always intended to increase the G threshold when players got the chance to play more tournament games each year, and the change was flagged on the main seeding points page back in January.
Why 36? Well, the increasing number of tournament games an average player can realistically hope to play per year suggested a rise from G=20 to over 30 when I looked at the situation in the middle of the year and it helps if the threshold is a multiple of four so that 50% and 75% of it are both whole numbers. It was originally stated in January that the new threshold would be 36, but one less tournament had been fitted in than was hoped so 32 was used for most of 2002. However, the large entry for the Plate counteracted this so the threshold was raised to 36 at the year end as originally envisaged.
The number of games threshold will continue to be increased gradually when possible. It will be reviewed again in the middle of 2003 once it is clear whether any new tournaments are being run and/or whether the number of games per match has been increased. However, once the threshold has finally reached 50 games, it will not be increased any further.
Can you explain the reductions for players who have played less than three tournaments?
The reductions for players who have played less than three tournaments were made to correct an oft-complained about anomaly which is a particular problem with big knockout tournaments. A player had been able to have an established rating based on performance in only one tournament if they played enough games in that tournament, meaning that a tournament débutant doing extremely well in the World Championship but not winning it had a good chance of ending up with a higher overall rating than the World Champion, even if the latter had played consistently well in tournaments over the longer term.
In fact when I originally analysed this, it became apparent that the chance factor in Tantrix can make it possible to 'fluke' extreme results in two tournaments in a row, but hardly ever three in a row, which is why the rule was written the way it was.
Why do you now require a player to play six games in a tournament for it to count as one of those three tournaments?
After introducing the three tournament rule, it later became clear that the extreme rating anomaly could still arise even under the new rule if someone played a lot of games in the WTC as described above and also played just a match or two in two more tournaments.
This usually means that the player is knocked out early in the additional tournaments and gets a low rating from them, but with a low number of games this may still not be enough to significantly modify the extreme rating from the first tournament. However, if someone got an extreme high rating from the one or two matches played in those other tournament, the problem could be particularly bad.
Previously, this could only happen in the fairly rare cases where a player lost after getting very close to beating a top player in the first round of a tournament or got a big win then defaulted from the rest of the tournament. However in the 2002 WTTC, it was possible to get a very high rating from just one 2-game match and then not have to play any more matches because your team had been knocked out, and this now became a far from rare event.
The new six games per tournament clause, applied retrospectively to give consistency from year to year, should catch the situations referred to above in most cases.
Why adjust by using lobby rankings for some players?
2 * lobby ranking - 250 is not intended to be used in place of Elo ratings in very many cases but is simply designed to ensure that very strong players end up in the top half of the entry list by seeding points, which will mean that they do not have to play the very top seeds in round 1 - a safeguard for both them and the top seeds., Players on high lobby rankings (which can sometimes be based heavily on robot games, games against self-selected opponents, etc), can turn out to be very poor tournament players, hence this factor can never increase a player's seeding points to more than 1750 so that the lobby ranking is not given too much importance. So, while a strong lobby player should end up in the top half of the entry list, they are unlikely to end up as one of the numbered seeds - an uneasy compromise, but hopefully better than nothing.
Why not use master rankings for this adjustment?
The is a simple answer to this - nearly all masters play in tournaments and have an Elo rating - the players with very low Elo ratings and low numbers of tournament games for whom we need a secondary seeding criterion are almost always non-masters. Master rankings could be incorporated into the seeding floor too, but this would create even more work for little gain in terms of fairness.
In addition, a secondary criterion must be calculated in a way that only allows it to take over from the primary criterion in a minority of cases. So, another problem with using master rankings as a secondary seeding criterion would be that their range is very unpredictable and no one seems to know how they are likely to evolve. eg. we could have assumed early in 2002 that master rankings would hardly ever exceed 2000 only to see Matt Peek somehow rise to almost 3000 while everyone else was left nearer 2000 or below.
Why the 1500 floor in team events?
Simply to avoid disadvantaging (in the seeding process) teams from countries without many tournament players who may have to field four strong players and one virtual beginner. 90% of players in the WTTC have between 1600-2000 seeding points and it is possible for a newish player to have only about 800 seeding points without this adjustment, giving their team no chance of getting among the seeds even if the other four players and hence the team as a whole are reasonably strong. Another practical compromise!
Further questions/comments?
I will do my best to answer further questions and welcome comments. Please write to tournaments@tantrix.co.uk.While we do not intend to change the basic principles of the rating system, we are happy to consider adjustments which do not compromose the integrity of past ratings and indeed some of the changes mentioned above have resulted from previous comments about perceived anomalies.
The date and time are :
Saturday, 17-May-2008 08:20:54 GMT
Saturday, 17-May-2008 09:20:54 BST (local)
3024