Are Elo ratings going down?

by Walter Wolf
3/2/2019 – Chess fans tend to focus on the FIDE ratings of the top players in the world which are updated monthly. But the Elo system wants to reflect the strength of the chess players of all strengths all over the world. Since its introduction in 1970 it has become immensely popular and important for the chess world. However, it is not without problems. Why, for example, are junior players from Russia or India continuously underrated and how does this affect other players? To answer this and other questions, WALTER WOLF took a close look at the development of some ratings over time. | Photo: Junior players from India, courtesy ChessBase India

ChessBase 15 - Mega package ChessBase 15 - Mega package

Find the right combination! ChessBase 15 program + new Mega Database 2019 with 7.6 million games and more than 70,000 master analyses. Plus ChessBase Magazine (DVD + magazine) and CB Premium membership for 1 year!

More...

Examining ratings of junior players

If you follow chess events all over the world you will often come across players who perform better than their Elo ratings would suggest. Some players indeed appear to be significantly underrated, and how underrated they are often seems to depend on the country they are coming from. This has a number of reasons.

Impact of the lowered-rating floor

Elo rating book coverIn this article I take a look at the impact of the lowered rating-floor (the minimum Elo rating) on Elo ratings in general. This is a topic that usually does not catch much attention even though lowering the rating-floor significantly increases the number of young players with an Elo rating. However, the Elo system does not really seem to be fit to deal with this phenomenon and the question arises whether the rules of calculating the Elo ratings really reflect the playing strength of young players adequately.

The numbers in this article are based on the players files in the FIDE archives that go back to 2001. The data for the time before 2001 is taken from the website OlympBase. The statistics are always based on the January lists.

The numbers given below distinguish between active and inactive players. If a player has not played any rated games for the previous twelve months he is considered to be inactive. If the status is not specifically mentioned below the player is an active player.

The development of the Elo ratings since 1970

FIDE adapted the Elo system in 1970 but it took until 1986 before 5000 players had an Elo rating. Now, the age pyramid of players with an Elo rating grows all over the world.

Table 1: Average values of players with an active Elo rating (January 2019)

In January 2019 India had 82,000 AICF-registered players, 40,514 of them were juniors. In Russia, too, more than half of the active players with an Elo rating are juniors. As far as tournament chess is concerned the USA and China seem to move a bit outside of the Elo system. The US Chess Federation (USCF) has 90,000 members but only 2,500 active players with an Elo rating. England also has a surprisingly low number of players with an Elo rating. Germany has the most active seniors (60+): 3229 seniors have an Elo rating and are active! Together with the inactive players Germany has 25,160 players with an Elo rating, and the average Elo rating of these players is 1882, almost a 100 points more than the average level of the "DWZ" of these players — the "Deutsche Wertungszahl", the national rating of Germany. Spain is the country that has the most players with an Elo rating. However, if you also take the inactive players into account Russia is clear first: 32,730 players in Russia have an Elo rating, and a lot of them are juniors.

The period since Elo ratings were introduced in 1970 can be divided into two phases:

In the first phase from 1970 to about 2000, more and more players had an Elo rating and there was a strong rise in Elo ratings. Among the top 1000 players this rise continued until 2012. It is difficult to say exactly what caused this ratings rise but it is obvious that we have more strong grandmasters today than we had in 1980 (see e.g. countries such as France or China). In 1990 there were no Chinese players in the top 100, today the average rating of the top 100 in January 2019 would drop from 2703 to 2696 if one ignored all Chinese players.

The second phase is marked by lowering the rating floor (minimum Elo) which led to a strong increase of the number of players with an Elo rating, most of them juniors. The Elo ratings of the top 1000 have been rather stable in this time but lower down, the level sinks a trend almost unnoticed by the public. Maybe the focus is too much on the top 100 and the number of GMs and IMs which continues to grow:

Table 2: Statistics of all GMs and IMs since 1985 (active and inactive)

The strong increase of titled players is probably a result of the strong increase of players who have an Elo rating and the fact that there are more tournaments in which you can get title-norms.

So, how about the ratings on the lower Elo levels?

Table 3: Average rating of active players

During the last seven years the top 1000 have established themselves on a high level. But in the top 20,000 to 40,000 the Elo ratings decrease on average, despite the strong overall increase of players with an Elo rating!

Table 4: Number of active players in rating groups

The number of active players who have an Elo rating between 2000 and 2400 also goes down. This is due to the fact that a lot of players in this rating range are listed as inactive and do not appear on this list. All in all about 160,000 players were listed as inactive in 2019, and about 60,000 of these players have not played a rated game in the last five years. In fact, a lot of these players have an Elo rating that has not changed for 10 or 20 years. Many of these inactive players have an Elo rating of more than 2000 because that was the rating-floor when these players were active. But the crucial question is: why do younger players not make up for this lower number of players with a rating between 2000 and 2400 though the total number of active players with an Elo rating has tripled during the last ten years?

In the next two tables the players who have been leaving or entering the rating-pool since February 2012 are not shown in the statistics. Tables 5 and 6 only include those active players from January 2012 who still can be found in the 2019 players list. Whether these players are active or inactive in 2019 is irrelevant. But it is important to understand that the players shown in 2012 in tables 5 and 6 are the same players as in 2019.

You may ask where the balancing influence of K=40 can be seen? Tables 5 and 6 do not show a closed system. The players who have been entering the rating-pool since February 2012 are not in the statistics, but they are influencing the Elo-ratings as some of them have been opponents of the players shown in table 5 and 6.

It can be seen that the youth of course increases their Elo level, but the overall level is slightly declining. The players from India over 20 years have a significant drop in the average rating, though these players are actually in the best chess age. The following chapter tries to explain this rating-decrease as a result of the age structure and of a fast growing amount of rated players in the India Chess Federation.

Lowering the rating-floor to 1000

Note: In 1987 the rating-floor for women was raised from 1800 to 2005

The rating floor (the minimum Elo) was first 2200 but since then has been lowered. This led to more and more junior players with an Elo-rating. Previously players maybe got an Elo rating when they were 20 years old, now they do get an Elo rating when they are 12 years old or even younger.

One may safely assume that every tournament that a junior player plays helps to increase his playing strength. However, the overall level of the young players remains the same after a junior tournament because they all have the same K-factor. But this is an inadequate reflection of their real playing strength.

Players develop particularly rapidly between the age of 10 to 18 but in these years you now only gain Elo points from other players (unlike previously). Some of these rating points come from junior players who quit chess. However, because the playing strength of virtually all junior players increases from the age of 10 to the age of 18 years these junior players mainly gain rating points from adult opponents who "sponsor" the ratings of the juniors. As a result a lot of players ranging from 1000 to 2000 Elo points are underrated which also affected higher Elo levels.

In 2014 the K-factor for juniors was raised to 40. If a junior player (K=40) wins against an adult (K=20) the number of rating points the junior gains is twice as high as the number of rating points the adult loses. The junior player reaches the rating that expresses his playing strength adequately faster but the overall level of the Elo ratings also rises.

Today's K=40 has the greatest impact when junior players play often against adults because they can show that their playing strength is not much ahead of their rating. But K=40 does not fully compensate the lowering impact the juniors have on the rating of the established players because the juniors gain rating points from players whose playing strength is adequately reflected in their Elo rating and who thus cannot fulfil their Elo expectation against underrated players. How junior players who lose Elo points and quit chess affect the ratings is still unclear.

Nihal Sarin

Nihal is one junior player who's not likely to quit | Photo: Amruta Mokal

Arpad Elo (1903-1992) already mentioned the deflationary impact of junior players on the Elo ratings. But if you look at the impact a single game has on the rating of an adult who loses against a junior you see that only a part of the rating loss of the adult is undeserved — depending on how much the junior player is underrated. To illustrate this phenomenon Berthold Plischke gave me the following example:

Junior J (Elo 1600) wins against adult A (also 1600). J wins 40*(1-0.5) = 20 points, A loses 20*(0-0.5) = -10. However, the 10 points A loses are not all undeserved losses: let's say J had a playing strength of 1700. Then A should have lost 20*(0 - 0.36) = - 7.2. Therefore he only lost about 3 points undeservedly.

In the last two years the Indian Chess Federation had 30% more members! Presumably, most of the new members were juniors. Thus the Indian Chess Federation has a lot of junior players who have only few contacts to players from abroad. As a result the percentage of games in which a junior plays against another junior rises while the percentage of games in which a junior plays against an adult goes down. However, the level of the ratings can only change when players with a different K-factor play against each other. In view of the many Indian junior players the Elo level of Indian juniors will probably rise slower than in Germany.

Germany once also had a large increase of the number of junior players. From 1951-1991 the German Chess Federation increased the number of their members from 25,000 to 90,000. But back then the Elo rating was not introduced or the rating floor was too high for most juniors.

Let's assume the following: if 100 players who are 30 years old and have an Elo average of 1800 only play against each other in the ten years to come what Elo level will these players have on average when they are 40? And then you take 100 10-year-olds with an Elo average of 1100 and for the next ten years they play only against each other — what an Elo level will they have when they are 20?

Proposals to calculate the Elo rating

Thanks to Berthold Plischke, responsible for controlling the "DWZ ratings" (the German ratings), and an expert in the field. When calculating the "DWZ", junior players have certain privileges until they are 25 and thus are better integrated into the rating pool. At the age of 25 you should be close to the playing strength you can reach. In 2016, the German Chess Federation introduced another rule to fight the phenomenon of underrated players: if the tournament performance of a player is at least 300 points better than his current DWZ his tournament performance — and not his current DWZ rating — is taken to calculate the new DWZ rating of his opponents.

Measures to raise the Elo ratings

In 1987, FIDE raised the Elo ratings of all women players (apart from Susan Polgar) by 100 points. One should consider whether to do something similar in countries in which the players are said to be underrated. However, such an increase of Elo ratings should be limited to lower ratings and should be implemented by degrees.

Thus, the German Chess Federation considers whether to increase the first DWZ rating of beginners that might be too low to numbers that are a bit below 1000. This would accept that these players might be overrated but it will be good for their future opponents.

Such an increase of the first Elo rating might also be a suitable solution for a country such as India to counter the phenomenon of underrated players.

The K-factor for junior players

To better account for the increasing playing strength of young players one could slightly raise the K-factor of the winner and slightly lower the K-factor of the loser in a game between two juniors. The winner then would gain more rating than the loser would lose.

A similar plus is necessary when adults play against juniors because the overall Elo level of junior players can only rise in games against adults.

Examples:

Now: Junior (K=40) against adult (K=20)

New: Junior (K=35) against adult (K=18 in case of a loss, K=20 in case of a draw, K=22 in case of a win)

Junior against junior (K=32 in case of a loss, K=35 in case of a draw, K=38 in case of a win)             

These are randomly chosen K-values that just show the principle. This rule should also be limited to juniors with Elo<2300. But it is first of all important to do something against the deflation even if this means to adjust the K-factor every year anew.

Summary

More and more players have obtained Elo ratings since the Elo system was implemented in 1970. The low number of players with an Elo rating in the US and in China indicates that there will be even more players with an Elo rating.

Lowering the minimum Elo to 1000 led to a steady stream of low-rated players who joined the rating-pool. The large number of these players is enough to impact the ratings of established players and to lower their ratings. Table 4 shows this dynamic.

Changing the rules to allow juniors to increase their rating level even if they play in junior tournaments could reduce the pressure lasting on the established players.

Links




Walter Wolf lives in Stuttgart, is a passionate chess player and likes to travel.
Discussion and Feedback Join the public discussion or submit your feedback to the editors


Discuss

Rules for reader comments

 
 

Not registered yet? Register

w_wolf w_wolf 4/28/2019 09:52
Hello and thanks for the commentars.
Before I wrote this article I searched in the internet for reasons of underrated players. After i saw that the players-list at ratings.fide.com can be downloaded I made these statistics. Don't know if all numbers are correct as I couldn't found statistics to compare with. When in 2014 the K-Factor for juniors was raised to 40 there have been a great discusssion about this in the internet. In this discussion about the K-Factor some well-known players and mathematicians took part like John Nunn, Barlomiej Macieja, Ken Thompson, Jeff Sonas. The statistics of Jeff Sonas and the whole discussion focused maybe too much on the players above 2400. I welcome the rating-floor of 1000 as I think every tournament-game should be rated to get fair ratings. But so far the Elo-System is based on the idea of young players who are pulling out points from the rating-pool, older players who are giving back points to the rating-pool and a K-Factor of 40 for Juniors to bring it all in balance. But the amount of young players is rising very fast in the last years, as the lowered rating-floor seems to be in company of a chess-boom in open-tournaments. A outstandig example is Sri Lanka: in April 2019 among the 1412 active FIDE-rated players there are 922(!) juniors. It don't need an expert to see that this island is in a elo-deflation. because K=40 has no influence on the elo-niveau when a Junior plays against Junior. as I wrote in the article that is a weakness of the elo-system. I am sure that out there there are many experts and trainers who are aware of the problem of elo-deflation. And I guess most of them see the elo-system as the best system we have so far, it just need to be optimized from time to time according to the circumstances.
w_wolf w_wolf 4/27/2019 06:38
hello Lopak, sorry for the confusion. In the sentence "In Russia, too, more than half of the active players with an Elo rating are juniors" the word "too" should be deleted, as you mentioned 5900 out of 12281 is of course less then 50%. the AICF homepage shows the amount of registered and rated players in India every month. It says that in April 2019 there are 83644 AICF registered players and 30440 FIDE-rated players. To complete this - among these 30440 rated players there are 10370 juniors, and 12038 players with an active status, and 5917 Juniors with an active status.
lopak lopak 3/25/2019 03:14
I don't understand above table 1: "In January 2019 India had 82,000 FIDE-registered players, 40,514 of them were juniors." What I read in the table is: IND total = 12,281; U18: 5,900, so less then 50%. Do I confused?
spaceprobe spaceprobe 3/7/2019 10:27
The Elo system was introduced in 1970 , this holy cow is now 49 years old, it's an inefficient tool to
measure the stenght of chessplayers all over the world, using the parameters of a win,a loss or a draw,
the supreme system of valuing the individual moves is completely ignored.
Practice shows that some high-rated ELO players are underperforming, and some low-rated ELO players
are overperforming, it's really time to drop this crude instrument for ever, and to introduce a system that values
the whole game and the individual moves together.
Jack Nayer Jack Nayer 3/5/2019 01:28
"If you pump a stream of new rating points into a pool, it inflates."

Says who? What does that even mean?

Peer-reviewed statistical artice? Of course not.

Don't tell me what I don't understand. You don't know me.
Masquer Masquer 3/4/2019 08:23
@JackNayer

It is simple. If you pump a stream of new rating points into a pool, it inflates. I explained how this happened historically. That trend may have stalled in the last few years, however.

You can't declare something incoherent babble, just because YOU can't understand it, sorry. That's just bad reasoning.
Michael Jones Michael Jones 3/4/2019 01:58
>England also has a surprisingly low number of players with an Elo rating

This is not in the least surprising, but is because England - like the US, which Aighearach discusses - has its own national rating (grading) system. Almost all local leagues are ECF graded, but the majority are not FIDE rated (although most tournaments in England are now both), so there will be many active players who do not have an Elo rating. There are 126,331 players on the database; that will include many who are inactive, those who have played in the English system although they represent a different country (and no doubt some who are deceased) - but it shows that the 1,781 with an Elo rating is only a fraction of the number of active players in England.
Jack Nayer Jack Nayer 3/3/2019 11:11
Interesting article. Correct analysis.
The problem is rating inflation is easy to solve: provide evidence. There isn’t any. I am not talking about your opinions. Show me statistical peer-reviewed articles. I have been saying this for years. Nobody ever sent me an article.
I’m convinced that there is a case to be made for the deflation of elo rating at the top. Look at the last world championship. These two played the most advanced chess on the planet. How many elo points did Carlsen win?
Not only is there no evidence, the whole inflation fantasy makes no logical sense. There are geographical clusters, yes. But overall, there is Chessbase now, there are databses, trainers, books, there a lot more tournamements, is anyone going to make the point that none of this increased performance?
In 2013, Rowell from Exeter looked at the historical performance of runners. Todays runners run faster. There are, of course, better much shoes today, better tarmacs, better training, better nutrition. So Rowell let people run with old shoes on tracks from before the war. The new generation runs faster. Technical and social progress made that people became intrinsically better. The best chess players the world has ever seen are those with the highest rating today. The rest is fantasy, just incoherent babble.
Masquer Masquer 3/3/2019 02:55
Another long-lasting negative impact that we are stuck with as a result of post 1987 inflation, and mostly due to the 1993 lowering of the Elo floor, is the cheapening and degrading of the GM title. Whereas that title meant something special and highly respectable, after 1993 a lot of players who really only of IM-strength could now qualify as GMs, as result of the rating boost coming from this bottom-feeding system that had been created.

It led to a boom in the number of GMs worldwide, but this was not backed up by substance, or by a tangible increase in actual play, since new rating points were furiously fed into the system at the lower end, trickling their way up. This meant ratings were going up while strength was staying the same! Yes, folks, don't believe the canard that today's players are just better than those of yesterday, meant to appease the ignorant, already fooled, masses. Yes, they're a little better prepared, and certainly not worse than previous generations, but also, let's face it, a whole lot more overrated than those players!
Aighearach Aighearach 3/2/2019 09:27
US players are rated already, though. And the ratings are more stable than using ELO.

FIDE has a history of corruption. The financial culture in the US causes Americans to be highly averse to joining FIDE. The result is that a US tournament that is dual-rated in order to attract international players will result in a lot of Americans who would have a FIDE rating; if they paid the fee. But why would they? Usually the only players who do that are at least FM rating. They might join FIDE if they have to in order to get a title, but even then most are happy with a USCF title. Also, people are more willing to pay to join an organization when required for tournament entry, but they're substantially less interested in paying to join an organization after the tournament just to receive an extra rating; especially an extra rating that only has a chance to move in one or two tournaments per year!

There is no reason to think that because a low number of US players have a FIDE rating, in the future there will be more. This isn't an untapped reservoir of potentially-rated players, this is a pool of already-rated players who already don't find value in FIDE. If in the future there is some alternate international ratings organization, then it will be realistic to expect an increase in US participation.

Remember; when a European tournament organizer pays a bribe to FIDE, it is just some lost money. If an American tournament organizer did that, it would be a felony crime in the US because of the Foreign Corrupt Practices Act. If an American pays a foreign bribe, they go to prison in the US, for breaking US law. This minimizes the cultural mixing and related desire for international rating.

China, OTOH, is investing in chess education because they want the national prestige that Russia used to get from being a chess leader. Memberships are driven more by the coaches than the players. So their numbers will increase substantially in the future.
Masquer Masquer 3/2/2019 09:20
Thanks so much for this article, it is a step in the right direction.
Masquer Masquer 3/2/2019 09:20
The top 100 or so are unaffected by the this phenomenon, so it supports my Elo-points-trickle-to-the-top hypothesis, which also explains why there has been historical INFLATION, and not deflation, at the very top of the Elo scale, but not much elsewhere.

The SuperGM inflation only began in 1987, however, when FIDE decided to raise all women players' ratings by 100 points. Big deal, you might say, when these points would be evenly diluted with the total Elo pool. But that proved to be not the case, and all these new rating points began to work their way up to the top, a sort of trickle up effect.

The FIDE rating system had been very stable from 1970 to 1986, even slightly deflationary, perhaps. After the 1987 infusion of more ratings points, in 1993 the rating floor was lowered from 2200 to 2005, which caused a lot more players to who were also less strong to enter the rating pool. Some of these over-2005 players would qualify for an Elo rating as result of peak performances, meaning they were mostly overrated! This became another source of cheap rating points to the stronger players above them, leading to steady inflation, again trickling to the top but also feeding the whole top 100-1000 as well.

The 1990's were an highly inflationary period for Elo ratings especially for SuperGMs who could now boast Fischer-like ratings without being in the same league as Fischer. Because of this, it was also a very damaging effect because it has misled the new generation of chess fans into grossly discounting players of the previous generation. It has cheapened chess history. Where as a Petrosian or a Tal had been a superGM while rated, say 2650, now anyone under 2700 was not even counted as such and no longer entered the conversation. Long gone were the years of stability, 1970-1986!

The damage to the rating system may be done, but we could dampen some of its effect by educating fans as to the history and evolution of the players ratings.
1