Rating Deflation?

When I returned to chess I was curious how class players of the past compared with those of the present. How does a "B" player from 1975, for instance, compare with a "B" player today?

While reading up on it, I ran across the claim that USCF ratings have noticeably deflated since the nineties. Thus a "B" player from 1995 might have dropped down to a "C" player today, even if he plays as well as he did in 1995.

Was this issue ever settled?

Interesting. I wouldn't be surprised if class players have improved, since chess resources are so much more freely accessible now.

DrCheckevertim: Steven Mitlitzky, A commenter at a NYT article, argues that USCF chess ratings have deflated because:

• Quality chess books, programs and database resources are more accessible than ever, making chess skill easier to develop.
• Many players compete in largely closed rating pools. The strength of the entire group increases but no one's rating goes up because they mostly play each other and trade rating points back and forth.
• "The overwhelming majority of active USCF members are now children. This creates a 'black hole' of rating points for established players. Once the Elo system envisioned a mean rating of 1500 but now two out of three members are rated under 1000. As they rapidly improve, kids suck rating points out of the pool of players with relatively stable playing skills and established ratings."
• "The... most significant deflator of ratings is the unspoken and officially denied USCF policy of keeping ratings low....In an era where most players are children, the USCF doesn't want children's ratings to drop like stones for fear they will get discouraged and drop out."

Tough stuff.

It's an interesting question whether chess ratings should correspond to an absolute level of skill or not.

It's a given that ratings should sort players by their relative strength, but wouldn't it be nice if a rating of 1600, say, represented a certain level of chess strength in the past, present and future?

If the average level of strength in the USCF increased, it would just mean that more players had higher ratings, not that everyone had to play better to keep their current rating.

Caitlyn: It would depend on the math underlying the rating scheme. That's my point.

It would be harder to manage ratings so that they corresponded a level of skill but it could be done or at least attempted.

That seems somewhat to be the intent. As things stand, they've been coping with deflation and inflation since the rating system began and tweaking the system in response to such concerns.

For instance, last May the USCF  made a K-factor change to the rating scheme so that players rated below 2355 can have the ratings change more rapidly.

But it would be very difficult to implement, as there's no clear reference point. How can you evaluate the objective force of a competition player ?

More is known about chess than ever before ,more games can be played and are played  than ever before. Playing chess a while ago was an event that might come a few times a week or even a month depending on where you lived . Now I can play a 100 games in a week. so yes ratings are deflating and a 1600 today uscf would wreck a 1600 of 1970s value. 

The ratings of this sight are extremely skewed deflated as well. players on here show vey high skill and can be rated 1300 to 1500 finding mate in 4 and 5 so on ,either that or there is rampant cheating going on. 

Probably both - rampant cheating IS going on, AND people are improving way over past performances. 

I also believe in rating deflation - people today are generally more aware and know more, as resources are so much more readily available than was the case. 

I also believe in rating deflation - people today are generally more aware and know more, as resources are so much more readily available than was the case.

However, does this mean ratings should deflate? If a 1600 player in 2015 plays much better than a 1600 player in 1975, why isn't that 1600 player rated 1700 instead?

I've read persuasive arguments that ratings did deflate after the schools and USCF started pushing their scholastic programs.

<why isn't that 1600 player rated 1700 instead?>

Because if everybody suddenly knows chess better and trains with more precision and focus because they know what they're doing to a greater extent because of available resources, the chess player at the 20th percentile, the 50th percentile, the 80th percentile etc. today is better than their counterparts back in the day. 

If we assume that by and large, a certain rating score (say 1500) represents a certain chess percentile (say the 40th), then improvement of the entire pool will mean that the same rating (say 1500) represents a greater amount of knowledge in 2015 than it did earlier (say in 1970). 

Hope that explains it. 

it is possible. the internet is strange place where many absurd things take place.

If we assume that by and large, a certain rating score (say 1500) represents a certain chess percentile (say the 40th), then improvement of the entire pool will mean that the same rating (say 1500) represents a greater amount of knowledge in 2015 than it did earlier (say in 1970).

Your scenario makes sense. I believe it's easier these days to acquire chess knowledge and skill. However, I'm not sure that's a full or correct explanation.

The rating deflation mentioned earlier was presumed due to the influx of scholastic players sucking rating points from class players then the scholastic players quit chess and left the class players holding the bag with depressed ratings.

Ultimately ratings are arbitrary numbers signifying nothing more than rankings in a pool. It isn't hard to set up a system which does the ranking, but the trick is to get a system that's stable and has a good feel for human participants. To this end the USCF tweaks the rating system periodically.

So ideally -- it seems to me -- the USCF could tweak the rating system so that a 1600 player was a player with a certain skill level and not a person at a percentile ranking within a pool. However, I suspect it's hugely complex and with ample risks of making things worse.

Ratings predict results within a pool of players, so the strength of players from 40 years ago is irrelevant when discussing ratings today.

Mathematically speaking you are correct. Mathematically it would be fine if the ratings drifted all over the place as long the rank and results maintained the correct relations.

However, humans expect ratings to mean something and it's not an unreasonable expectation. Consequently the USCF tweaks the rating system, with varying degrees of success, to match this expectation.

Lately I'm fascinated by the work of Dr. Ken Regan, an IM and a math Ph.D, who has devoted much of his mathematical skill to chess. He is most famous for his statistical-computer methods for detecting chess cheating.

However, I am most interested in his work on "Intrinsic Performance Rating" by which Dr. Regan uses chess engines and statistical analysis to calculate FIDE chess ratings from game collections.

http://www.cse.buffalo.edu/~regan/papers/pdf/Reg12IPRs.pdf

If his approach works and can be extended to lower rated players, it means that within 5-10 years we can expect commercial programs to calculate reasonably accurate ratings based solely on the PGN scores of our games.

Thanks man! I was really looking for something like that. I was sure someone had this idea before me, there he is, Dr. Ken Regan. Thanks ipcress12

Couldn't agree more, GM Gormally!

I'm pretty sure a tournament 1200 player today is substantially stronger than a 1200 tournament player in the pre-computer era.

I have a few friends who played competitive chess in the 80s, and then stopped. They were rated like 1700+ in UCSF tournaments. They said they don't play anymore because when they go online, they get absolutely demolished by anyone near that level, and often get beaten by 1200s (I was one of those people beating them when I was 1200 here.) 

I'm not sure if this is true at master level etc., but in the  pre-internet era, chess exposure and chess resources were a lot harder to come by if you weren't part of a vibrant chess club. Nowadays, I'm actually surprised if I play a 1300 online and they can't bang out the first 12 moves of the Sicilian Dragon instantly and even the plan for the kingside pawn storm without hesitation. 

In 2009 Mark Glickman didn't think there was inflation: http://gambit.blogs.nytimes.com/2009/01/14/are-todays-top-players-better-than-20-years-ago-not-necessarily/?_r=0

edit: post #3 already had the link

Ok, so an Elo rating is a relative rating to begin with. It only measures a player's skill against his/her competitors. This means that, to begin with, Elo ratings are relative from the very start. Now, in response to what you call "deflation" this is a phenomena that has not been clearly explained or efficiently fixed to date. Prsonally, I think that the ratings have inflated rather than deflated. Isn't it just a little bit odd that there are over 100 players who are 2600+? This is because people have tried to make adjustments to the Elo system to benefit them. For instance, Gary Kasparov moved to have FIDE make a rule where someone could not win a tournament and lose elo. This movement passed and it adds to the inflation problem that we are seeing.

Another plausible theory I've read is that the ratings floor is causing the inflation. This kind of makes sense. If you have people who happen to hit the floor for 1 or 2 games, then drop down to the lower levels again, this will constantly be adding elo to the top players, inflating their scores so long as they continue to play the lowest players. This will, naturally, just keep adding points to the top elo players, which they exchange among each other instead of lower skilled players, thus draining points from everyone else and adding it to their score.

Is there a solution? Honestly, I think that if we just use the Elo system the way it was intended and don't try to change it all the time, then it will not have inflation and/or deflation problems. The beauty of the system is that it is self correcting and never ADDS points when it changes the player ratings. So, honestly, all that the rating floor shows is that the players at the top are better than those below them. It's honestly not real inflation, it's the Elo system correcting itself. Those players at the top would still have probably won those games anyways and the guys who barely crack the rating floor are corrected to their proper rating. 

The only thing that could be adding inflation is the fact that more players are playing these days. However, unless someone else can think of a better way to objectively gauge the skill level of two grandmasters if they've never played each other, then I think we should stick with what we've got.