| |
BGonline.org Forums
More Elo...
Posted By: Henrik Bukkjaer In Response To: A tale of two elo's (Rick Janowski)
Date: Monday, 18 June 2012, at 1:29 p.m.
I have been wondering if Elo type systems could be modified to incorporate prestige related issues. The methodology would be to weight the ranking point swings in accordance with some agreed scale representing the prestige value of the event (in a similar manner as when the ranking point swings vary with match length).
This might deviate from the purist notion of a Elo system but may be a good practical compromise.
Rick, isn't there a rather big correlation between the factors currently affecting Elo ratings, and the ones you might consider when doing a prestige adjustment?
I mean, what events should be the most prestigious to win? I guess it's those with the most fierce opponents (high rated), longest matches (up to 25 points), and lots of participants (ie. rounds)?
Imagine the World Championship being everything we want it to be. Now, say you win (or come 2nd), in that tournament. You will have won over a lot of other top-rated opponents, in long matches* (up to 25 points), and played many rounds (not just 5 nut maybe 8!).
That would give you a lot of Elo rating points, even without the prestige adjustment!
Furthermore, if you want to keep the Elo system a null-sum system, the prestige adjustment would also make it more expensive to loose matches at the big tournaments - which seems counterproductive to what you want to achieve I think. So the prestige points would only apply to the winners? (ie. consist of points being injected into the population, making these tournaments + EV Elo-wise). The latter I suppose is not a problem, but something you'd have to carefully consider!
*) In the DBgF rating system, match lenghts is not adjusted in the same manner as with the FIBS adoption (if I recall correctly). We have a linear adjustment in one of the components for matchlength, instead of a square-root function. Eg. winning a 25 point match over an opponent rating at the same rating as yourself, will give you 12,5 points.
-----
We have a "Rating Systems Committee" in the DBgF, which hasn't really been active for the past year but I think will resurrect within short. In that committee various improvements to the current Elo adaption used by the DBgF has been suggested (as well as a completely new system). All ideas and improvements can be simulated and verified on our match-database, that has a record of all matches played in the DBgF for the last 15 years or so.
Some of the ideas discussed, try to address the known problems: New players entering the population, and inactive players "re-entering" the population. In the standard Elo type systems for backgammon, new players typically enter at the "average" rating (which actually isn't the average rating, because of players leaving the population). In real live tournament play, "new players" however, can be anything from true beginners, to former world champions. One model of fixing this problem, is to add an "accelerator". That's just a bad fix. It adds a lot of volatility in a game that has plenty of that to begin with. And it "buries" the players starting out as true beginners, but quickly improving their game, giving them an even longer path to work their way back to a rating that match their skills. Instead we've loosely discussed a model, where new players will not enter at the same level, but start on a rating based on the tournament and flight they enter. It's not a bulletproof solution, but it will improve things. If different federations across the globe will "merge" their elo systems, I think some sort of differentiated start rating would be a good thing to implement.
Also adding (a small) decay to ratings for inactive players would make a better system. And after some time of inactivity the player would be removed from the official lists, then once he starts playing again, his rating would have been adjusted some points or percentage of points (based on the length of his inactive period and his current rating).
---------------
Daniel writes that different systems doesn't exclude each other. Very true. That's why (in the DBgF) we have introduced 2 (3) different systems! Apart from the Elo ratings (which is the true ranking list in the DBgF, used to seed players and exclude players from different flights), we have added a "Results" system, and on top of the two systems, our Title system (giving the "prestige" dimension that Rick is missing in the Elo system).
Results are given for all approved tournaments, to a predetermined number of players (depending on number of participants and format), and also to certain achievements in the team league and doubles tournaments (eg. winning the Danish team league, or becoming the league top-scorer).
The title-system has been described and discussed here last year, in short it adds titles (IM, GM and SGM) to players achieving impressive feats in either results: Winning the most prestigious tournaments (or getting 3 second places in the same). Achieving high Elo ratings (either a very high peak, or a sustained high rating). Or playing 5 live clocked matches (in big tournaments) within a year at a very low error-rate.
So: An Elo type system for current ratings, and some other system for "titles" and prestige recognition. This goes very much towards the post of Jason Lee in this thread - What is the purpose of a list? With no clear definition of the purpose, how can you then asses if a list "performs well"? Should the list encourage players to play certain events? Should it recognize prestigious titles? Should it reward results or technical strength? Should it be used to predict the outcome of a given match (eg. like Elo, and thus be used to seed players or bar them from certain tournaments or flights)?
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.