The most significant ratings-related event this past year
was the recoding and implementation of the rating system program
by Mike Nolan. Mike and several members of the Ratings Committee
worked closely together to facilitate the process of
moving the rating system to the new computing environment.
One of the important features of the new system is that
large blocks of tournaments can be rerated if mistakes were
discovered. In the past, errors in individual rating
computations were addressed by focusing on that individual's
rating, ignoring that the effect of the error had influence
on other players' ratings. Similarly, events rated
out of chronological order could have an impact on the
resulting ratings. But with the ability to carry
out periodic rerates of events (e.g., rerating them in chronological
order), some of these problems can be solved in a principled
manner.
In the course of implementing the new ratings program, a
few issues were raised and addressed. For example, we
provided Mike Nolan recommendations for handling situations
where duplicate IDs are in the database corresponding to
different ratings. We also had an extensive discussion
about how to rate events with multiple time controls within
the same event (a rare occurrence, but one that needed to be
addressed). The method to which we agreed was to have a TD
divide rounds into two groups; those that were slower than
G/30 and those faster. The former would be rated under the
standard system (only), and the latter would be rated under
the quick chess system. We also made a minor change to the
method of assigning an initial rating to unrated players.
When an unrated player is in an adult membership category but
that the USCF does not have a birthdate, then the ratings
program will now use 1000 as the starting rating rather than 750
(which is what the system had been using). Finally, the discussions
surrounding the new ratings program led to clarifications
and corrections in the ratings systems specifications, which are posted
online at http://math.bu.edu/people/mg/ratings/rating.system.pdf.
The office and the board charged the committee with a few
tasks this year. One of the tasks was to evaluate whether
the USCF can eliminate the half-K rating option. Members of
the Committee who responded unanimously agreed that it would
be fine to abolish the half-K option, reasoning that the
rating system formulas should not depend on choices made by a
tournament organizer. The Committee was also asked about
updating USCF ratings for USCF-rated players competing in
FIDE events. The Committee reminded the office that a method
had been proposed and implemented in 1994, but had only been
used sporadically since then. The Committee updated the
document explaining the methodology, and presented it to the
USCF office. The document can be accessed online at
http://math.bu.edu/people/mg/ratings/fideuscf.pdf.
The Committee was also asked to provide input on a method of rating
blitz games, i.e., games with time controls of G/3 through G/9.
The options were either to construct a new and separate blitz
rating system, or to extend the quick chess rating system to
include time controls down to G/3. While the majority of
committee members who responded thought that a separate
system was more principled, extending the quick chess rating
system to include blitz time controls was deemed the simplest to
implement and manage. A couple of the members suggested that the
value of K in the ratings formulas for blitz games should be
lower than for quick chess games, but the arguments against
fractional-K included increasing the complexity of the algorithm,
and the uncertainty of how to choose the value of K without
extensive analysis.
Each year, the Committee performs data analyses to monitor
changes in the rating system and rating pool. One of our key
analyses is to examine the distribution of ratings of
established players that have played in USCF-rated events
each of the last three years that are between 35-45 years of
age at the start of January 2005, and compare the distribution to the
corresponding distribution based on previous years' analyses.
We focus on this group in particular because, accordingly to
gerontology studies on cognitive learning, we expect them to
have relatively stable abilities. What we found is that the
average rating for this group is around 1770, and that it is 10
points higher than last year's average. This suggests that
the bonus and feedback mechanism in the rating formulas
continues to inflate players' ratings, as we hoped. It is
worth noting, however, that our goal is to reinflate ratings
back to levels in the year 1997, where the average rating for
this cohort was close to 1820. A more complete statistical
summary of our analyses can be found online at
http://math.bu.edu/people/mg/ratings/monitor2005.txt.
We also examined the percentage of established players
with ratings between 1400-2299 in the December 2004 annual
rating list with ratings having last digits of 00. This
analysis is to obtain some understanding of the frequency of
players on their rating floor (such players would have a rating
ending in 00). We discovered that over 6% of these players
had ratings ending in 00. If approximately 1% of established
players have ratings ending in 00 (which is what one might
expect if there were no rating floors), then about 6%-1%=5% of players
are on their rating floor. While the percentage of players on their floor
has been declining since the year 2000, this figure is a slight
increase from that of last year. A graph of the statistical
summaries over the past twelve years can be found online at
http://math.bu.edu/people/mg/ratings/digits00.pdf.
We will continue to monitor the percent of players on their
rating floor and propose rating formula adjustments if the
problem persists.