Ratings Distribution Redux
Submitted by
on Fri, 08/14/2009 at 7:48am.
Okay, so some postings about ratings inflation got me curious, so I redid my quantiles for the chess.com online standard ratings. This is what I got:
max 2789 2900 +111
99% 2189 2291 +102
95% 1874 1970 +96
90% 1723 1806 +83
80% 1547 1617 +70
70% 1426 1486 +60
60% 1326 1383 +57
50% 1236 1288 +52
40% 1158 1200 +42
30% 1077 1118 +41
20% 994 1027 +33
10% 905 927 +22
1% ? 724 +?
min 0 0 +0
The first column is the percentile, the second column is the ratings for that percentile from 1/25/09, the third column is the ratings from 8/14/09, and the fourth column is the increase from January to August. Note that I didn't calculate the 1st percentile in January, so I have no way to figure the difference.
So, we are definitely seeing an increase in the percentiles. What exactly this means is a bit less clear, because we are not talking about a static population here. New people are coming in, and some accounts are being closed. My gut instinct is to go with the median increase (+52) as the ratings inflation, and assume the rest is the better players sucking points out of the weaker players. But I have nothing to back that up with.
The odd thing is the comparison of the averages. In January the average rating was 1345, and now in August it's 1375, an increase of 30 points. But how does that jibe with the distribution above? Over 80% of the players increased their rating by more than the average increase. This is possible, but it would require the bottom 20% being four times as bad as the top 80% is good. The 10th percentile does not seem to be bearing that out. I suppose my hunt and peck through the player list method of calculating the percentiles could be off, but I would expect it to be off by that much.