with no activity nothing changes, so you have to have some level of activity for there to be a operational system. ;o)

i think i agree with meta that any normalization (for your domination elo) should be somehow adjusted to (some) "pool size", but you don't dominate players that don't play. i think such players should be disregarded (until they start playing again, of course).

also, to unite (or mix the ratings) of all the "sub-pools", so that ratings in one geographical area remain comparable to ratings in another, we also need a certain level of activity.

and you and metatron2 who want to compare ratings over eras (despite heavily changed pools) implicitly require continuity - which translates to some minimal level of activity, as far as i can tell.

normally nobody leaves the system, people simply go inactive for shorter or longer periods.

i still think it would be slightly interesting to considering the "rating point economy" of players that enter the system, are active for a couple of rating periods, and then stay inactive for a prolonged period of time, potentially "for ever". just like a lot of things, i'd consider it interesting just to know. :o)

there are a few examples of 2600+ players that have gone inactive for some time (not necessarily the 3 years required to get the inactive flag), and then started playing some games again. hübner is playing in the ecc now, he might have been inactive for a while, and also seirawan had a period with extremely low activity. but single players like these are obviously completely irrelevant.

if there's a pattern for a majority of the now nearly 50 000 inactive players that they played for a limited period of time, lost rating, and then went inactive for a "long time", it might be considered a source of inflation. i'd say it's interesting to have look. :o)

1969-01: 1930.3 (age 38)
1970-01: 1931.4 (age 38)
1971-01: 1930.2 (age 40)
1972-01: 1934.2 (age 37)
1973-01: 1934.2 (age 38)
1974-01: 1937.4 (age 36)
1975-01: 1938.1 (age 36)
1976-01: 1937.6 (age 38)
1977-01: 1939.9 (age 37)
1978-01: 1938.4 (age 39)
1979-01: 1938.8 (age 40)
1980-01: 1940.0 (age 40)

Good question! For one thing, it would mean that on average the rating difference between player #a and player #b should remain approximately constant even as the size of the playing pool changes (where a and b are both sufficiently high ranks as to be in the upper tail of the distribution). Sonas found results consistent with this in his article here, but he wasn't sure why it should be the case: http://www.chessbase.com/newsdetail...

This would be in some ways surprising, since one expects that as the playing pool gets larger, the difference in playing strength between #1 and #100 ought to decrease. This is indeed the case if instead the distribution is Gaussian.

whatthefat, note that sonas didn't check that range. and the difference in elo between #1 and #100 has indeed decreased, compared to when fischer, kasparov/kramnik where miles ahead.

if we want to "safe-guard" us aganinst a couple extreme individuals, one might rather consider the difference between #3 and #100 - tracking that from 1980 to now would be information as well. but the diff between #30 and #100 has <also> been roughly the same from 1990 to 2009 - ca. 60 points - despite a huge increase in number of players; i already checked that.

i did a different "count", too:

year: 2600-2649 * 2650-2699 *
2700-2749 * 2750-2799 * 2800+

1997: 48 * 14 * 05 * 2 * 1
1998: 60 * 17 * 06 * 2 * 1
1999: 53 * 22 * 06 * 3 * 1
2000: 58 * 22 * 06 * 4 * 1
2001: 57 * 23 * 09 * 1 * 2
2002: 58 * 30 * 08 * 2 * 2
2003: ?? * 26 * 13 * 2 * 1
2004: ?? * 32 * 13 * 2 * 1
2005: ?? * 31 * 13 * 4 * 1
2006: ?? * 36 * 16 * 2 * 1
2007: ?? * 40 * 13 * 8 * 0
2008: ?? * 43 * 23 * 6 * 0
2009: ?? * 57 * 23 * 9 * 1

(i did this very quickly, based on the top 100 lists for <july> in the year given - that's the reason why i don't have bothered to find the numbers in the 2600-2649 range after 2002, which was the last year #100 didn't have a higher rating than 2600.)

I have put a link (Elo inflation) on my profile to an excel file (.xls) that will help you to play a bit with the numbers and reach your own conclusions.

You can change the "limit difference" (a1 cell) as you wish but I would advise you to

(i) use only multiples of 5 -- because of the old lists restrictions

(ii) use a larger limit in the last worksheet (100_2years), because it's a larger period. [I think 150%, or limit* square root of 2 should be more or less ok). If the 'inflation' difference between X (1 year) and 1.5X (2 years) is toooo large, than it may be a hint that this limit difference is not the best choice

(iii) use at least 25 or 30 points, in order to exclude the players that are quickly improving (or in decline)

Of course, this is far from perfect, we should have the number of games and so on (the semi-inactive players are a problem etc). But I think this is good enough for the time being.

I hope this file will work for you. If you are unable to download it, just give me a tip, and i will send it to you by e-mail [i'll delete your e-mail from my chessforum later of course]

It is indeed, but I think it helps to set some limits. As I see, 15-20 points of fluctuation in a year is normal (if you exclude those players, you will be excluding 50% of the data), anything beyond 40 or 50 points is a bit 'anomalous', and should not be included in my opinion (no need of a chi-square test for me:)

if we want to "safe-guard" us aganinst a couple extreme individuals, one might rather consider the difference between #3 and #100 - tracking that from 1980 to now would be information as well. but the diff between #30 and #100 has <also> been roughly the same from 1990 to 2009 - ca. 60 points - despite a huge increase in number of players; i already checked that.>

no time to play around with numbers now, but i look forward to comparing these estimates to my own, based on yearly inflation calculations from 1990 to 2009 - i have to limit myself to this range, because prior to 1990 the lists (at least those i have) don't have player ids.