Robin's Bridge Attic - Lehman Ratings
There's been lots of discussion about this area. In general, it's very good. But there's room for improvement.
- First of all, just about everyone (I think), agrees that the rating reacts too slowly (that is, the half-life
of results is too long at 10 wks). Speeding this up (say a half-life of six weeks) would help a lot, I think (especially
for new players - see my second point).
- I believe that a modification needs to be made to cope appropriately with new players (the fact that they enter
at 50% and only slowly migrate to their "proper" rating). This is because playing with such a person
- who could be a total novice who has just joined OKB - can have an unreasonably significant effect on a rating.
Personally I avoid playing with new (or newly reset) players in comp mode. This is a shame, because it is not
very welcoming to new players. Please see my separate proposal: Modified Lehman Scheme.
- The current practice of converting IMP scores into percentages tends to note small IMP differences (overtricks)
at the expense of big swings (all swings greater than 7.5 IMPs are considered 100% or 0%). Actually, I can't
see why we need a conversion at all. Why not simply represent the IMP ratings in terms of IMPs?
- A much harder problem to solve is the very slow mixing of results throughout the pool of players. Imagine
one day four experts sit down to play a weekly game with each other for several months. At the same time four
novices do the same thing. During this time they play with nobody else. As time goes on, each of them will end
up with a rating not far from 50. The average for the experts will be a little over 50 and the average for the
novices will be a little under 50. The reason the averages move away from 50 at all is that on any board exactly
half the competition comes from the opponents at the table and half comes from the rest of the same-way pairs on
the board. But this effect is slow (half as slow as when people mix partners and opponents all the time). When
these people begin to play with others in the general pool, their ratings will be artificially close to 50. Now,
this is an extreme example for illustration purposes, but it is not too far off what happens in practice. When
novices join OKB, they tend to find other novices to play with; experts look for other experts.
Documents of interest