Join Bridge Winners
One "Big Idea" - Ratings

I feel like the 10 Big Ideas thread is too cumbersome, as the discussion is all over the place and therefore hard to follow. I am posting this thread as a way to discuss one idea that is near and dear to me. Please confine discussion to this topic. What follows are my thoughts on this. Looking forward to a lively discussion.

The general idea is to create a Rating System for players that varies up and down based on recent performance, rather than the present Rank system that can only move upward with time and is subject to all sorts of problems:

  • General inflation over time
  • Rank is proportional to total points won in lifetime, which may be the result of great skill exhibited over a short time, or low skill exhibitted over a long time.
  • Because of the above, players with long bridge careers eventually 'rise to the level of their incompetance' - they are forced to play in top brackets / strata where they have minial chance of success. This the leads to players dropping out of tournament play.
  • Young but strong players are forced to play in brackets well below their actual ability, although they might prefer to play in the top bracket.

Basing stratification, event qualification, and bracketing on a rating that changes with recent performance would eliminate or reduce these problems. In addition, a rating that reflects performance would also provide players with a better measure of their growth in ability.

The Colorado ratings are a reasonable approach to this problem. However, the algorithm involved is difficult to understand and seemingly does not apply unless each player has multiple partners. I would certainly not object strongly to instituting this as an official ACBL system, but I propose a much simpler system that requires no new data acquisition and is  (I think) easily understood:

  • At any given time, a players Rating would be a rating calculated by a function of points won over the last 12 months (R0), + .9 times his rating the previus year (R1) (or .8, or .7..., depending on how much influence we want previous years to have on the result. So R + R0 + .9*R1. Note that R1 would include >9 times her rating from two years ago, so we have a decaying average that always has some minor influence from the distant past.
  • We calculate R0 as a linear combintion of points won in each pigment category times a multiplier that expresses the difficulty of winning that type of points, and divided by the number events played that pay that type of points. Lets say we use weights of 10 for platinum, 4 for gold, 2 for silver, and 1 for black, red and unpigmented (online) points. We can of course, hold an endless debate over the right multipliers. Thus R0 would equal 10P/NP+4G/NG+2S/NS+(R+B+U)/NRBU. So R0 is a measure of the average points won per outing in each category, weighted by the relative difficulty of winning points in that category. Obviously this suffers from the flaw of not really separating your performance from your partner's, but I think it is close enough, and way better than what we do now. And every bit of information needed to do this calculation is already sitting on the disk drives of ACBL's computers. It also assumes that all points in a category are equally difficult to win, which is clearly a bit of an oversimplification - we could use a real strength-of-field number  as a weight instead, but that would make the whole thing much more complex. My suggestion would be to try this and fine tune if inequities become obvious over time. 

 

Given that all data needed for the scheme I propose already exists, I think it would be very feasible and inexpensive to do a trial of this to see what it looks like.

105 Comments
Getting Comments... loading...
.

Bottom Home Top