The Ratings Committee has produced a major revision to the
rating system, responding to some important weaknesses in the
currently implemented system. As has been known for awhile,
the great influx of scholastic players into the rating pool
has resulted in substantial challenges to the rating system.
One of the particular problems is rating deflation, in which
scholastic players improve more quickly than the rating system
can track. Another problem with the current system involves
calculating ratings for unrated players, especially when
they either have perfect scores (positive or negative), or when
they play against other unrated players. The currently
implemented rating system also does not account for the greater
volatility of player's abilities at the low end of the
rating spectrum. All of these issues have been addressed in our
revision of the rating system. Statistical analyses have
been performed to validate the integrity of the revised
system. A detailed description may be found as a postscript
document on the world-wide web at
http://math.bu.edu/people/mg/rating.system.ps .
We outline briefly some of the important changes to the rating
system. They include
* a new method for determining provisional ratings. In
most cases, the formulas would produce a rating that is
identical to the current system. But when opponents'
ratings are far from a player's pre-event rating, or when
they are widely dispersed, the new formulas appropriately
use information from results against such opponents to
produce an updated rating. Also, the new system uses
age-based imputed ratings to rate games between unrated
players.
* a bonus point mechanism for tracking quickly improving
players. The standard formula for updating ratings
has been adjusted to account for unusually strong performances
in an event. When a player's results exceed the expected
result beyond a certain threshold, the player earns bonus
points. The threshold is a function of the number of rounds
in an event, and only applies if the event is four or more
rounds and if a player competes against no opponent more
than once.
* a "sliding-K" scale to account for more variable player
abilities when players have low ratings or when players have
not played many tournament games. The value of "K" in the
standard rating formula can be thought of as a measure of
uncertainty in a player's pre-event rating. The higher the
value of K, the greater the impact of a tournament result
and the less reliance on the pre-event rating. The revised
rating system therefore uses values of K that are large when a
player is low-rated or when the player has not played in many
USCF-rated games to reflect the greater uncertainty in such
players' abilities.
* a new iterative procedure to rate an event. Rather than
performing a single rating calculation for each player when
rating an event, the revised system performs two calculations
for previously rated players and three for unrated players.
In effect, the rating calculations are performed once to
produce a set of intermediate ratings, and then the rating
calculations are performed a second time using the
intermediate ratings as the opponents' ratings to produce a
final rating. The main benefit to this procedure is that
the results of opponents' games are now incorporated into
a player's rating calculation. If an opponent has a low
pre-event rating but performs well in an event, the revised
rating system will adjust the calculations to recognize that
the opponent is probably better than the pre-event rating
indicated. This is particularly useful as a "feedback"
mechanism to bonus points.
The Ratings Committee will be producing a document that
allows players to approximate their post-event ratings based
on the revised rating system.
Earlier this year, the Policy Board approved a lowering of
the rating floors by 100 points, and set an absolute minimum
floor to 100. The consensus of the ratings committee is to
abolish rating floors entirely (except for an absolute
floor), as the bonus point mechanism in the revised rating
system will prevent rating deflation. Computer simulations
under the revised system, however, have clarified that players
with moderate to high ratings will not likely drop to their
200-point rating floor except in unusual circumstances.
The Ratings Committee is considering the merits of keeping
the USCF title system versus Bill Goichberg's proposed "life rating"
system. At the committee workshop in August, we plan to
discuss the tradeoffs between the two systems, and resolve
which direction to pursue.