|
What error rate corresponds to each rating on FIBS? There are fewer than
100 players with ratings over 1900, and some of those are bots or rating
manipulators. I collected some data, and the results surprised me.
When I started playing on FIBS in late 1999, I remember watching many
matches between players rated over 1800. I aspired to achieve that
rating, and felt that when I first did, it was a fluke of the dice. Only
when I finally started to recognize mistake after mistake between
players rated over 1800 did I feel my 1800+ rating was justified. I
don't think I ever broke 2000, and my rating was somewhere over 1900
when I took a long enough break from FIBS (mainly to play for money
elsewhere) for my account to expire.
For the May NEBC final, I played a 15 point match with Walter Trice on
FIBS. I decided to warm up by playing a few matches on FIBS, and then
decided to keep playing until my rating improved from the initial 1500
to 1900, which took about 800 experience points (220 matches).
Here are the average Snowie-style error rates of my human opponents in
matches of 3 points or more, grouped by rating:
1400-1500: 15.25
1500-1600: 13.86
1600-1700: 11.17
1700-1800: 9.55
1800+: 8.13
I have to say that I'm pretty surprised by those results. An 1850 rating
corresponds to an error rate of 8? That's a lot more errors than I
expected. Each 100 rating points corresponds to an improvement of about
1.5 millipoints per move. Extrapolating, someone with an error rate of
3.5 millipoints per move should have a rating of about 2150.
Has the level of play on FIBS dropped recently, or has it been like this
for years, but I didn't notice? Has anyone else conducted similar
studies recently?
Douglas Zare
|
![[Recommended reading]](pics/Star2.gif)
![[GammOnLine forum]](pics/GammOnLine.gif)
![[Long message]](pics/Long.gif)
![[Recent addition]](pics/New.gif)
Recent addition