PDA

View Full Version : ELO/Glicko inflation/deflation? (sf ACF newsletter)



Alan Shore
31-10-2006, 11:18 PM
I believe Ian Rogers has the highest rating ever achived by an Australian with 2618 on the January 1999 FIDE list.
Mark Paragua had a rating of 2621 in April 2006.
Of course none of this allows for FIDE rating inflation over the years.

Interesting you make remarks about FIDE inflation. I believe Glicko has the opposite effect though, with ratings deflation being the problem.

From a practical point of view online, ICC has experienced huge inflation with the ELO system, whilst FICS has experienced considerable deflation using Glicko.

The good news is, FICS ratings are far more representative of OTB ratings than ICC.

Duff McKagan
31-10-2006, 11:23 PM
Word to this. Glicko is good, except for when it affects dudes who haven't played in a long time.

Bill Gletsos
31-10-2006, 11:28 PM
Interesting you make remarks about FIDE inflation. I believe Glicko has the opposite effect though, with ratings deflation being the problem.You may believe it but is isnt correct.

Alan Shore
31-10-2006, 11:39 PM
You may believe it but is isnt correct.

Honestly Bill, if you're going to make some kind of claim then back it up. I am telling you a fact based on practical internet systems. I'd be interested to see the stats on mean and median ratings for the present list and the mean/median of the list >7 years ago, excluding juniors, who may improve rapidly, and are a large reason for why Glicko 2 is now used.

You're the man with the stats, so let's see them. I remember you claiming there was no delfation but I'd love to see that data if you exclude juniors.

Do not forget Bill, I am definitely a supporter of Glicko 2 and believe it to be superior to ELO. However, I do not believe it to be perfect and it should be subject to refinements.


Word to this. Glicko is good, except for when it affects dudes who haven't played in a long time.

Absolutely. There are even instances where it does work for inactive players to represent current performance but I'm concerned just how much it affects current players. While I urge anyone gravely affected negatively by Glicko not to panic too much, be aware of the volatility of the system.

Bill Gletsos
01-11-2006, 12:05 AM
Honestly Bill, if you're going to make some kind of claim then back it up.I believe my claim has as much backing it up as your original statement did. ;)

I am telling you a fact based on practical internet systems.You simply made a claim. You didnt mention how such a claim of inflation/deflation was measured.

I'd be interested to see the stats on mean and median ratings for the present list and the mean/median of the list >7 years ago, excluding juniors, who may improve rapidly, and are a large reason for why Glicko 2 is now used.Yes but using means etc can be misleading.
e.g. A club has 10 players with a mean of 2000.
10 new players join the club with a mean of 1600.
The mean rating of the club is now 1800 but that doesnt mean there has been deflation.

You're the man with the stats, so let's see them. I remember you claiming there was no delfation but I'd love to see that data if you exclude juniors.

Do not forget Bill, I am definitely a supporter of Glicko 2 and believe it to be superior to ELO.I'm aware of that.

However, I do not believe it to be perfect and it should be subject to refinements.I agree. Watch the next ACF newsletter.

Alan Shore
01-11-2006, 01:40 AM
I believe my claim has as much backing it up as your original statement did. ;)

Possibly but you're the one with the RD data. :eek:


You simply made a claim. You didnt mention how such a claim of inflation/deflation was measured.

Easy - log onto either server and type 'hbest' (fics) and 'best' (ICC) to see the marked difference. Alternatively you can experience many years on both servers (as I have) and witness the 'best' ratings of various users on FICS and witness the deflation or make a comparison of ICC ratings past and present. The mean ratings are available on both servers with the correct commands and having witnessed means over a sufficient time to establish longitudinal reliability, the effects are plain to see. Feel free to ask FICS admins about deflation issues.


Yes but using means etc can be misleading.
e.g. A club has 10 players with a mean of 2000.
10 new players join the club with a mean of 1600.
The mean rating of the club is now 1800 but that doesnt mean there has been deflation.

Your example does not constitute a random sample - in fact it would be quite an aberration. Similarly, if 10 players 2400 joined it wouldn't lead to the assumption of inflation. The sample size in an online environment such as FICS on the other hand, is huge. Means are generally good measures as you know. Even better though, FICS provides both the means and the standard deviations.


I'm aware of that.
I agree. Watch the next ACF newsletter.

OK.

pax
01-11-2006, 08:39 AM
Easy - log onto either server and type 'hbest' (fics) and 'best' (ICC) to see the marked difference.

Since ICC both has more strong players, and is massively inflated (this is immediately obvious), this tells you precisely zero about deflation of Glicko/FICS.

Apart from this, the ACF implements Glicko in a considerably different manner to FICS (which can substantially affect the comparitive deflation) so the comparison is not valid in any case.

Ian Rout
01-11-2006, 09:33 AM
From a practical point of view online, ICC has experienced huge inflation with the ELO system, whilst FICS has experienced considerable deflation using Glicko.

These effects have to be looked at in context rather than assumed to be intrinsic to the system. Any rating system is naturally deflationary because people join the system, then improve and take rating points away from other players who therefore have lower ratings without a decline in standard. So FICS may well deflate but it may be deflating less than other systems would when applied to the same population.

Itís interesting that the FIDE system seems to inflate whereas it would for the above reason normally deflate, as seen when a fundamentally similar system was used by ACF. I donít know if the reason for FIDE inflating has been fully investigated but a couple of reasons that have been mentioned are that until recently many people joined it when their standard had stabilised, and because of the former rating floor of 2000 so performances that would otherwise have deflated it are removed.

I donít know anything about the ICC system so I donít know why it inflates. Perhaps they start with a larger K factor? It would make commercial sense to have an inflationary system to keep people interested.

Vlad
01-11-2006, 11:27 AM
I believe Glicko has the opposite effect though, with ratings deflation being the problem.


I believe Bill's example is a good illustration of what is happening on FICS. When the server started it looked like a great idea... Free, great spirit, etc. Many people turned up.

But over time good players stopped coming becuase there is no strong enough opposition. It will take half an hour to find anybody to play.

So, if you are measuring the deflation by looking at the best ratings it may not be the best way.

pax
01-11-2006, 12:40 PM
Itís interesting that the FIDE system seems to inflate whereas it would for the above reason normally deflate, as seen when a fundamentally similar system was used by ACF. I donít know if the reason for FIDE inflating has been fully investigated but a couple of reasons that have been mentioned are that until recently many people joined it when their standard had stabilised, and because of the former rating floor of 2000 so performances that would otherwise have deflated it are removed.


The rating floor is undoubtedly why FIDE ratings have been inflationary in the past.

The problem was that for unrated players, performances under the required 2000 (or 2200 previously) rating were ignored. This meant that one or two above average performances led to players getting a FIDE rating sometimes 200-300 higher than was necessarily merited. These players then leak rating points to higher players and so on.

Even now that the floor has been reduced, if you look at the Australians with FIDE ratings in the 2000-2200 range, you will find many who have FIDE rating significantly higher than their ACF rating.

Trent Parker
06-11-2006, 11:29 AM
I believe Bill's example is a good illustration of what is happening on FICS. When the server started it looked like a great idea... Free, great spirit, etc. Many people turned up.

But over time good players stopped coming becuase there is no strong enough opposition. It will take half an hour to find anybody to play.

So, if you are measuring the deflation by looking at the best ratings it may not be the best way.

But if high rated opponents stop coming to FICS their rating would remain static thus not increase or decrease but remain static.

Desmond
06-11-2006, 04:09 PM
But if high rated opponents stop coming to FICS their rating would remain static thus not increase or decrease but remain static.
But they will have taken points from the active players in the pool, so the tendency of ratings in the pool of active players will be downwards.

pax
09-11-2006, 08:53 AM
But they will have taken points from the active players in the pool, so the tendency of ratings in the pool of active players will be downwards.

Only if the players started on a lower rating, improved, and left on a higher rating. If they leave on roughly the same rating they entered, there will be little if any effect on the pool.

Vlad
09-11-2006, 01:49 PM
But if high rated opponents stop coming to FICS their rating would remain static thus not increase or decrease but remain static.

I'm not sure what they do on FICS nowdays because I have not been there for a while. I know that on ICC if you do not play for some time (say a year) your account will be automatically erased. It would be interesting to look at the FICS human ranking and see if any of my old accounts are still there.:)

Vlad
12-11-2006, 11:08 AM
Yes, I have just experimented with FICS. I found my old nick "anu" which I have not used since year 2000. If you use "hrank anu" you will get it but it will not appear if you use "hrank somebody else". I suppose that is because I have not used this account for very long time. That implies that old accounts are not present in the rankings.

"Anu" is approximately top 10 of the current humans playing on FICS. I remember when I played 6 years ago I was never higher than top 20. That suggests that my theory is actually right - strong players left FICS and you do not see them in the rankings.

Garvinator
12-11-2006, 02:25 PM
"Anu" is approximately top 10 of the current humans playing on FICS. I remember when I played 6 years ago I was never higher than top 20. That suggests that my theory is actually right - strong players left FICS and you do not see them in the rankings.or that there is some kind of deflation which you have not experienced since you have played there for over 6 years.

Southpaw Jim
12-11-2006, 06:54 PM
Could it be that old accounts aren't present in the rankings because their RD has become high through inactivity? IIRC, your FICS rating becomes "provisional"/? (?? even?) once your RD is above 80 :hmm:

I might be talking out of my arse though :hmm:

bergil
12-11-2006, 07:21 PM
I might be talking out of my arse though :hmm:
Even if you were, that would hardly make you Robinson Crusoe! ;)